1 <?xml version="1.0" encoding="iso-8859-1"?>
3 <indexterm><primary>language, GHC</primary></indexterm>
4 <indexterm><primary>extensions, GHC</primary></indexterm>
5 As with all known Haskell systems, GHC implements some extensions to
6 the language. They can all be enabled or disabled by commandline flags
7 or language pragmas. By default GHC understands the most recent Haskell
8 version it supports, plus a handful of extensions.
12 Some of the Glasgow extensions serve to give you access to the
13 underlying facilities with which we implement Haskell. Thus, you can
14 get at the Raw Iron, if you are willing to write some non-portable
15 code at a more primitive level. You need not be “stuck”
16 on performance because of the implementation costs of Haskell's
17 “high-level” features—you can always code
18 “under” them. In an extreme case, you can write all your
19 time-critical code in C, and then just glue it together with Haskell!
23 Before you get too carried away working at the lowest level (e.g.,
24 sloshing <literal>MutableByteArray#</literal>s around your
25 program), you may wish to check if there are libraries that provide a
26 “Haskellised veneer” over the features you want. The
27 separate <ulink url="../libraries/index.html">libraries
28 documentation</ulink> describes all the libraries that come with GHC.
31 <!-- LANGUAGE OPTIONS -->
32 <sect1 id="options-language">
33 <title>Language options</title>
35 <indexterm><primary>language</primary><secondary>option</secondary>
37 <indexterm><primary>options</primary><secondary>language</secondary>
39 <indexterm><primary>extensions</primary><secondary>options controlling</secondary>
42 <para>The language option flags control what variation of the language are
45 <para>Language options can be controlled in two ways:
47 <listitem><para>Every language option can switched on by a command-line flag "<option>-X...</option>"
48 (e.g. <option>-XTemplateHaskell</option>), and switched off by the flag "<option>-XNo...</option>";
49 (e.g. <option>-XNoTemplateHaskell</option>).</para></listitem>
51 Language options recognised by Cabal can also be enabled using the <literal>LANGUAGE</literal> pragma,
52 thus <literal>{-# LANGUAGE TemplateHaskell #-}</literal> (see <xref linkend="language-pragma"/>). </para>
54 </itemizedlist></para>
56 <para>The flag <option>-fglasgow-exts</option>
57 <indexterm><primary><option>-fglasgow-exts</option></primary></indexterm>
58 is equivalent to enabling the following extensions:
59 &what_glasgow_exts_does;
60 Enabling these options is the <emphasis>only</emphasis>
61 effect of <option>-fglasgow-exts</option>.
62 We are trying to move away from this portmanteau flag,
63 and towards enabling features individually.</para>
67 <!-- UNBOXED TYPES AND PRIMITIVE OPERATIONS -->
68 <sect1 id="primitives">
69 <title>Unboxed types and primitive operations</title>
71 <para>GHC is built on a raft of primitive data types and operations;
72 "primitive" in the sense that they cannot be defined in Haskell itself.
73 While you really can use this stuff to write fast code,
74 we generally find it a lot less painful, and more satisfying in the
75 long run, to use higher-level language features and libraries. With
76 any luck, the code you write will be optimised to the efficient
77 unboxed version in any case. And if it isn't, we'd like to know
80 <para>All these primitive data types and operations are exported by the
81 library <literal>GHC.Prim</literal>, for which there is
82 <ulink url="&libraryGhcPrimLocation;/GHC-Prim.html">detailed online documentation</ulink>.
83 (This documentation is generated from the file <filename>compiler/prelude/primops.txt.pp</filename>.)
86 If you want to mention any of the primitive data types or operations in your
87 program, you must first import <literal>GHC.Prim</literal> to bring them
88 into scope. Many of them have names ending in "#", and to mention such
89 names you need the <option>-XMagicHash</option> extension (<xref linkend="magic-hash"/>).
92 <para>The primops make extensive use of <link linkend="glasgow-unboxed">unboxed types</link>
93 and <link linkend="unboxed-tuples">unboxed tuples</link>, which
94 we briefly summarise here. </para>
96 <sect2 id="glasgow-unboxed">
101 <indexterm><primary>Unboxed types (Glasgow extension)</primary></indexterm>
104 <para>Most types in GHC are <firstterm>boxed</firstterm>, which means
105 that values of that type are represented by a pointer to a heap
106 object. The representation of a Haskell <literal>Int</literal>, for
107 example, is a two-word heap object. An <firstterm>unboxed</firstterm>
108 type, however, is represented by the value itself, no pointers or heap
109 allocation are involved.
113 Unboxed types correspond to the “raw machine” types you
114 would use in C: <literal>Int#</literal> (long int),
115 <literal>Double#</literal> (double), <literal>Addr#</literal>
116 (void *), etc. The <emphasis>primitive operations</emphasis>
117 (PrimOps) on these types are what you might expect; e.g.,
118 <literal>(+#)</literal> is addition on
119 <literal>Int#</literal>s, and is the machine-addition that we all
120 know and love—usually one instruction.
124 Primitive (unboxed) types cannot be defined in Haskell, and are
125 therefore built into the language and compiler. Primitive types are
126 always unlifted; that is, a value of a primitive type cannot be
127 bottom. We use the convention (but it is only a convention)
128 that primitive types, values, and
129 operations have a <literal>#</literal> suffix (see <xref linkend="magic-hash"/>).
130 For some primitive types we have special syntax for literals, also
131 described in the <link linkend="magic-hash">same section</link>.
135 Primitive values are often represented by a simple bit-pattern, such
136 as <literal>Int#</literal>, <literal>Float#</literal>,
137 <literal>Double#</literal>. But this is not necessarily the case:
138 a primitive value might be represented by a pointer to a
139 heap-allocated object. Examples include
140 <literal>Array#</literal>, the type of primitive arrays. A
141 primitive array is heap-allocated because it is too big a value to fit
142 in a register, and would be too expensive to copy around; in a sense,
143 it is accidental that it is represented by a pointer. If a pointer
144 represents a primitive value, then it really does point to that value:
145 no unevaluated thunks, no indirections…nothing can be at the
146 other end of the pointer than the primitive value.
147 A numerically-intensive program using unboxed types can
148 go a <emphasis>lot</emphasis> faster than its “standard”
149 counterpart—we saw a threefold speedup on one example.
153 There are some restrictions on the use of primitive types:
155 <listitem><para>The main restriction
156 is that you can't pass a primitive value to a polymorphic
157 function or store one in a polymorphic data type. This rules out
158 things like <literal>[Int#]</literal> (i.e. lists of primitive
159 integers). The reason for this restriction is that polymorphic
160 arguments and constructor fields are assumed to be pointers: if an
161 unboxed integer is stored in one of these, the garbage collector would
162 attempt to follow it, leading to unpredictable space leaks. Or a
163 <function>seq</function> operation on the polymorphic component may
164 attempt to dereference the pointer, with disastrous results. Even
165 worse, the unboxed value might be larger than a pointer
166 (<literal>Double#</literal> for instance).
169 <listitem><para> You cannot define a newtype whose representation type
170 (the argument type of the data constructor) is an unboxed type. Thus,
176 <listitem><para> You cannot bind a variable with an unboxed type
177 in a <emphasis>top-level</emphasis> binding.
179 <listitem><para> You cannot bind a variable with an unboxed type
180 in a <emphasis>recursive</emphasis> binding.
182 <listitem><para> You may bind unboxed variables in a (non-recursive,
183 non-top-level) pattern binding, but you must make any such pattern-match
184 strict. For example, rather than:
186 data Foo = Foo Int Int#
188 f x = let (Foo a b, w) = ..rhs.. in ..body..
192 data Foo = Foo Int Int#
194 f x = let !(Foo a b, w) = ..rhs.. in ..body..
196 since <literal>b</literal> has type <literal>Int#</literal>.
204 <sect2 id="unboxed-tuples">
205 <title>Unboxed Tuples
209 Unboxed tuples aren't really exported by <literal>GHC.Exts</literal>,
210 they're available by default with <option>-fglasgow-exts</option>. An
211 unboxed tuple looks like this:
223 where <literal>e_1..e_n</literal> are expressions of any
224 type (primitive or non-primitive). The type of an unboxed tuple looks
229 Unboxed tuples are used for functions that need to return multiple
230 values, but they avoid the heap allocation normally associated with
231 using fully-fledged tuples. When an unboxed tuple is returned, the
232 components are put directly into registers or on the stack; the
233 unboxed tuple itself does not have a composite representation. Many
234 of the primitive operations listed in <literal>primops.txt.pp</literal> return unboxed
236 In particular, the <literal>IO</literal> and <literal>ST</literal> monads use unboxed
237 tuples to avoid unnecessary allocation during sequences of operations.
241 There are some pretty stringent restrictions on the use of unboxed tuples:
246 Values of unboxed tuple types are subject to the same restrictions as
247 other unboxed types; i.e. they may not be stored in polymorphic data
248 structures or passed to polymorphic functions.
255 No variable can have an unboxed tuple type, nor may a constructor or function
256 argument have an unboxed tuple type. The following are all illegal:
260 data Foo = Foo (# Int, Int #)
262 f :: (# Int, Int #) -> (# Int, Int #)
265 g :: (# Int, Int #) -> Int
268 h x = let y = (# x,x #) in ...
275 The typical use of unboxed tuples is simply to return multiple values,
276 binding those multiple results with a <literal>case</literal> expression, thus:
278 f x y = (# x+1, y-1 #)
279 g x = case f x x of { (# a, b #) -> a + b }
281 You can have an unboxed tuple in a pattern binding, thus
283 f x = let (# p,q #) = h x in ..body..
285 If the types of <literal>p</literal> and <literal>q</literal> are not unboxed,
286 the resulting binding is lazy like any other Haskell pattern binding. The
287 above example desugars like this:
289 f x = let t = case h x o f{ (# p,q #) -> (p,q)
294 Indeed, the bindings can even be recursive.
301 <!-- ====================== SYNTACTIC EXTENSIONS ======================= -->
303 <sect1 id="syntax-extns">
304 <title>Syntactic extensions</title>
306 <sect2 id="unicode-syntax">
307 <title>Unicode syntax</title>
309 extension <option>-XUnicodeSyntax</option><indexterm><primary><option>-XUnicodeSyntax</option></primary></indexterm>
310 enables Unicode characters to be used to stand for certain ASCII
311 character sequences. The following alternatives are provided:</para>
314 <tgroup cols="2" align="left" colsep="1" rowsep="1">
318 <entry>Unicode alternative</entry>
319 <entry>Code point</entry>
325 to find the DocBook entities for these characters, find
326 the Unicode code point (e.g. 0x2237), and grep for it in
327 /usr/share/sgml/docbook/xml-dtd-*/ent/* (or equivalent on
328 your system. Some of these Unicode code points don't have
329 equivalent DocBook entities.
334 <entry><literal>::</literal></entry>
335 <entry>::</entry> <!-- no special char, apparently -->
336 <entry>0x2237</entry>
337 <entry>PROPORTION</entry>
342 <entry><literal>=></literal></entry>
343 <entry>⇒</entry>
344 <entry>0x21D2</entry>
345 <entry>RIGHTWARDS DOUBLE ARROW</entry>
350 <entry><literal>forall</literal></entry>
351 <entry>∀</entry>
352 <entry>0x2200</entry>
353 <entry>FOR ALL</entry>
358 <entry><literal>-></literal></entry>
359 <entry>→</entry>
360 <entry>0x2192</entry>
361 <entry>RIGHTWARDS ARROW</entry>
366 <entry><literal><-</literal></entry>
367 <entry>←</entry>
368 <entry>0x2190</entry>
369 <entry>LEFTWARDS ARROW</entry>
376 <entry>↢</entry>
377 <entry>0x2919</entry>
378 <entry>LEFTWARDS ARROW-TAIL</entry>
385 <entry>↣</entry>
386 <entry>0x291A</entry>
387 <entry>RIGHTWARDS ARROW-TAIL</entry>
393 <entry>-<<</entry>
395 <entry>0x291B</entry>
396 <entry>LEFTWARDS DOUBLE ARROW-TAIL</entry>
402 <entry>>>-</entry>
404 <entry>0x291C</entry>
405 <entry>RIGHTWARDS DOUBLE ARROW-TAIL</entry>
412 <entry>★</entry>
413 <entry>0x2605</entry>
414 <entry>BLACK STAR</entry>
422 <sect2 id="magic-hash">
423 <title>The magic hash</title>
424 <para>The language extension <option>-XMagicHash</option> allows "#" as a
425 postfix modifier to identifiers. Thus, "x#" is a valid variable, and "T#" is
426 a valid type constructor or data constructor.</para>
428 <para>The hash sign does not change sematics at all. We tend to use variable
429 names ending in "#" for unboxed values or types (e.g. <literal>Int#</literal>),
430 but there is no requirement to do so; they are just plain ordinary variables.
431 Nor does the <option>-XMagicHash</option> extension bring anything into scope.
432 For example, to bring <literal>Int#</literal> into scope you must
433 import <literal>GHC.Prim</literal> (see <xref linkend="primitives"/>);
434 the <option>-XMagicHash</option> extension
435 then allows you to <emphasis>refer</emphasis> to the <literal>Int#</literal>
436 that is now in scope.</para>
437 <para> The <option>-XMagicHash</option> also enables some new forms of literals (see <xref linkend="glasgow-unboxed"/>):
439 <listitem><para> <literal>'x'#</literal> has type <literal>Char#</literal></para> </listitem>
440 <listitem><para> <literal>"foo"#</literal> has type <literal>Addr#</literal></para> </listitem>
441 <listitem><para> <literal>3#</literal> has type <literal>Int#</literal>. In general,
442 any Haskell integer lexeme followed by a <literal>#</literal> is an <literal>Int#</literal> literal, e.g.
443 <literal>-0x3A#</literal> as well as <literal>32#</literal></para>.</listitem>
444 <listitem><para> <literal>3##</literal> has type <literal>Word#</literal>. In general,
445 any non-negative Haskell integer lexeme followed by <literal>##</literal>
446 is a <literal>Word#</literal>. </para> </listitem>
447 <listitem><para> <literal>3.2#</literal> has type <literal>Float#</literal>.</para> </listitem>
448 <listitem><para> <literal>3.2##</literal> has type <literal>Double#</literal></para> </listitem>
453 <!-- ====================== HIERARCHICAL MODULES ======================= -->
456 <sect2 id="hierarchical-modules">
457 <title>Hierarchical Modules</title>
459 <para>GHC supports a small extension to the syntax of module
460 names: a module name is allowed to contain a dot
461 <literal>‘.’</literal>. This is also known as the
462 “hierarchical module namespace” extension, because
463 it extends the normally flat Haskell module namespace into a
464 more flexible hierarchy of modules.</para>
466 <para>This extension has very little impact on the language
467 itself; modules names are <emphasis>always</emphasis> fully
468 qualified, so you can just think of the fully qualified module
469 name as <quote>the module name</quote>. In particular, this
470 means that the full module name must be given after the
471 <literal>module</literal> keyword at the beginning of the
472 module; for example, the module <literal>A.B.C</literal> must
475 <programlisting>module A.B.C</programlisting>
478 <para>It is a common strategy to use the <literal>as</literal>
479 keyword to save some typing when using qualified names with
480 hierarchical modules. For example:</para>
483 import qualified Control.Monad.ST.Strict as ST
486 <para>For details on how GHC searches for source and interface
487 files in the presence of hierarchical modules, see <xref
488 linkend="search-path"/>.</para>
490 <para>GHC comes with a large collection of libraries arranged
491 hierarchically; see the accompanying <ulink
492 url="../libraries/index.html">library
493 documentation</ulink>. More libraries to install are available
495 url="http://hackage.haskell.org/packages/hackage.html">HackageDB</ulink>.</para>
498 <!-- ====================== PATTERN GUARDS ======================= -->
500 <sect2 id="pattern-guards">
501 <title>Pattern guards</title>
504 <indexterm><primary>Pattern guards (Glasgow extension)</primary></indexterm>
505 The discussion that follows is an abbreviated version of Simon Peyton Jones's original <ulink url="http://research.microsoft.com/~simonpj/Haskell/guards.html">proposal</ulink>. (Note that the proposal was written before pattern guards were implemented, so refers to them as unimplemented.)
509 Suppose we have an abstract data type of finite maps, with a
513 lookup :: FiniteMap -> Int -> Maybe Int
516 The lookup returns <function>Nothing</function> if the supplied key is not in the domain of the mapping, and <function>(Just v)</function> otherwise,
517 where <varname>v</varname> is the value that the key maps to. Now consider the following definition:
521 clunky env var1 var2 | ok1 && ok2 = val1 + val2
522 | otherwise = var1 + var2
533 The auxiliary functions are
537 maybeToBool :: Maybe a -> Bool
538 maybeToBool (Just x) = True
539 maybeToBool Nothing = False
541 expectJust :: Maybe a -> a
542 expectJust (Just x) = x
543 expectJust Nothing = error "Unexpected Nothing"
547 What is <function>clunky</function> doing? The guard <literal>ok1 &&
548 ok2</literal> checks that both lookups succeed, using
549 <function>maybeToBool</function> to convert the <function>Maybe</function>
550 types to booleans. The (lazily evaluated) <function>expectJust</function>
551 calls extract the values from the results of the lookups, and binds the
552 returned values to <varname>val1</varname> and <varname>val2</varname>
553 respectively. If either lookup fails, then clunky takes the
554 <literal>otherwise</literal> case and returns the sum of its arguments.
558 This is certainly legal Haskell, but it is a tremendously verbose and
559 un-obvious way to achieve the desired effect. Arguably, a more direct way
560 to write clunky would be to use case expressions:
564 clunky env var1 var2 = case lookup env var1 of
566 Just val1 -> case lookup env var2 of
568 Just val2 -> val1 + val2
574 This is a bit shorter, but hardly better. Of course, we can rewrite any set
575 of pattern-matching, guarded equations as case expressions; that is
576 precisely what the compiler does when compiling equations! The reason that
577 Haskell provides guarded equations is because they allow us to write down
578 the cases we want to consider, one at a time, independently of each other.
579 This structure is hidden in the case version. Two of the right-hand sides
580 are really the same (<function>fail</function>), and the whole expression
581 tends to become more and more indented.
585 Here is how I would write clunky:
590 | Just val1 <- lookup env var1
591 , Just val2 <- lookup env var2
593 ...other equations for clunky...
597 The semantics should be clear enough. The qualifiers are matched in order.
598 For a <literal><-</literal> qualifier, which I call a pattern guard, the
599 right hand side is evaluated and matched against the pattern on the left.
600 If the match fails then the whole guard fails and the next equation is
601 tried. If it succeeds, then the appropriate binding takes place, and the
602 next qualifier is matched, in the augmented environment. Unlike list
603 comprehensions, however, the type of the expression to the right of the
604 <literal><-</literal> is the same as the type of the pattern to its
605 left. The bindings introduced by pattern guards scope over all the
606 remaining guard qualifiers, and over the right hand side of the equation.
610 Just as with list comprehensions, boolean expressions can be freely mixed
611 with among the pattern guards. For example:
622 Haskell's current guards therefore emerge as a special case, in which the
623 qualifier list has just one element, a boolean expression.
627 <!-- ===================== View patterns =================== -->
629 <sect2 id="view-patterns">
634 View patterns are enabled by the flag <literal>-XViewPatterns</literal>.
635 More information and examples of view patterns can be found on the
636 <ulink url="http://hackage.haskell.org/trac/ghc/wiki/ViewPatterns">Wiki
641 View patterns are somewhat like pattern guards that can be nested inside
642 of other patterns. They are a convenient way of pattern-matching
643 against values of abstract types. For example, in a programming language
644 implementation, we might represent the syntax of the types of the
653 view :: Type -> TypeView
655 -- additional operations for constructing Typ's ...
658 The representation of Typ is held abstract, permitting implementations
659 to use a fancy representation (e.g., hash-consing to manage sharing).
661 Without view patterns, using this signature a little inconvenient:
663 size :: Typ -> Integer
664 size t = case view t of
666 Arrow t1 t2 -> size t1 + size t2
669 It is necessary to iterate the case, rather than using an equational
670 function definition. And the situation is even worse when the matching
671 against <literal>t</literal> is buried deep inside another pattern.
675 View patterns permit calling the view function inside the pattern and
676 matching against the result:
678 size (view -> Unit) = 1
679 size (view -> Arrow t1 t2) = size t1 + size t2
682 That is, we add a new form of pattern, written
683 <replaceable>expression</replaceable> <literal>-></literal>
684 <replaceable>pattern</replaceable> that means "apply the expression to
685 whatever we're trying to match against, and then match the result of
686 that application against the pattern". The expression can be any Haskell
687 expression of function type, and view patterns can be used wherever
692 The semantics of a pattern <literal>(</literal>
693 <replaceable>exp</replaceable> <literal>-></literal>
694 <replaceable>pat</replaceable> <literal>)</literal> are as follows:
700 <para>The variables bound by the view pattern are the variables bound by
701 <replaceable>pat</replaceable>.
705 Any variables in <replaceable>exp</replaceable> are bound occurrences,
706 but variables bound "to the left" in a pattern are in scope. This
707 feature permits, for example, one argument to a function to be used in
708 the view of another argument. For example, the function
709 <literal>clunky</literal> from <xref linkend="pattern-guards" /> can be
710 written using view patterns as follows:
713 clunky env (lookup env -> Just val1) (lookup env -> Just val2) = val1 + val2
714 ...other equations for clunky...
719 More precisely, the scoping rules are:
723 In a single pattern, variables bound by patterns to the left of a view
724 pattern expression are in scope. For example:
726 example :: Maybe ((String -> Integer,Integer), String) -> Bool
727 example Just ((f,_), f -> 4) = True
730 Additionally, in function definitions, variables bound by matching earlier curried
731 arguments may be used in view pattern expressions in later arguments:
733 example :: (String -> Integer) -> String -> Bool
734 example f (f -> 4) = True
736 That is, the scoping is the same as it would be if the curried arguments
737 were collected into a tuple.
743 In mutually recursive bindings, such as <literal>let</literal>,
744 <literal>where</literal>, or the top level, view patterns in one
745 declaration may not mention variables bound by other declarations. That
746 is, each declaration must be self-contained. For example, the following
747 program is not allowed:
753 (For some amplification on this design choice see
754 <ulink url="http://hackage.haskell.org/trac/ghc/ticket/4061">Trac #4061</ulink>.)
763 <listitem><para> Typing: If <replaceable>exp</replaceable> has type
764 <replaceable>T1</replaceable> <literal>-></literal>
765 <replaceable>T2</replaceable> and <replaceable>pat</replaceable> matches
766 a <replaceable>T2</replaceable>, then the whole view pattern matches a
767 <replaceable>T1</replaceable>.
770 <listitem><para> Matching: To the equations in Section 3.17.3 of the
771 <ulink url="http://www.haskell.org/onlinereport/">Haskell 98
772 Report</ulink>, add the following:
774 case v of { (e -> p) -> e1 ; _ -> e2 }
776 case (e v) of { p -> e1 ; _ -> e2 }
778 That is, to match a variable <replaceable>v</replaceable> against a pattern
779 <literal>(</literal> <replaceable>exp</replaceable>
780 <literal>-></literal> <replaceable>pat</replaceable>
781 <literal>)</literal>, evaluate <literal>(</literal>
782 <replaceable>exp</replaceable> <replaceable> v</replaceable>
783 <literal>)</literal> and match the result against
784 <replaceable>pat</replaceable>.
787 <listitem><para> Efficiency: When the same view function is applied in
788 multiple branches of a function definition or a case expression (e.g.,
789 in <literal>size</literal> above), GHC makes an attempt to collect these
790 applications into a single nested case expression, so that the view
791 function is only applied once. Pattern compilation in GHC follows the
792 matrix algorithm described in Chapter 4 of <ulink
793 url="http://research.microsoft.com/~simonpj/Papers/slpj-book-1987/">The
794 Implementation of Functional Programming Languages</ulink>. When the
795 top rows of the first column of a matrix are all view patterns with the
796 "same" expression, these patterns are transformed into a single nested
797 case. This includes, for example, adjacent view patterns that line up
800 f ((view -> A, p1), p2) = e1
801 f ((view -> B, p3), p4) = e2
805 <para> The current notion of when two view pattern expressions are "the
806 same" is very restricted: it is not even full syntactic equality.
807 However, it does include variables, literals, applications, and tuples;
808 e.g., two instances of <literal>view ("hi", "there")</literal> will be
809 collected. However, the current implementation does not compare up to
810 alpha-equivalence, so two instances of <literal>(x, view x ->
811 y)</literal> will not be coalesced.
821 <!-- ===================== n+k patterns =================== -->
823 <sect2 id="n-k-patterns">
824 <title>n+k patterns</title>
825 <indexterm><primary><option>-XNoNPlusKPatterns</option></primary></indexterm>
828 <literal>n+k</literal> pattern support is enabled by default. To disable
829 it, you can use the <option>-XNoNPlusKPatterns</option> flag.
834 <!-- ===================== Recursive do-notation =================== -->
836 <sect2 id="recursive-do-notation">
837 <title>The recursive do-notation
841 The do-notation of Haskell 98 does not allow <emphasis>recursive bindings</emphasis>,
842 that is, the variables bound in a do-expression are visible only in the textually following
843 code block. Compare this to a let-expression, where bound variables are visible in the entire binding
844 group. It turns out that several applications can benefit from recursive bindings in
845 the do-notation. The <option>-XDoRec</option> flag provides the necessary syntactic support.
848 Here is a simple (albeit contrived) example:
850 {-# LANGUAGE DoRec #-}
851 justOnes = do { rec { xs <- Just (1:xs) }
852 ; return (map negate xs) }
854 As you can guess <literal>justOnes</literal> will evaluate to <literal>Just [-1,-1,-1,...</literal>.
857 The background and motivation for recursive do-notation is described in
858 <ulink url="http://sites.google.com/site/leventerkok/">A recursive do for Haskell</ulink>,
859 by Levent Erkok, John Launchbury,
860 Haskell Workshop 2002, pages: 29-37. Pittsburgh, Pennsylvania.
861 The theory behind monadic value recursion is explained further in Erkok's thesis
862 <ulink url="http://sites.google.com/site/leventerkok/erkok-thesis.pdf">Value Recursion in Monadic Computations</ulink>.
863 However, note that GHC uses a different syntax than the one described in these documents.
867 <title>Details of recursive do-notation</title>
869 The recursive do-notation is enabled with the flag <option>-XDoRec</option> or, equivalently,
870 the LANGUAGE pragma <option>DoRec</option>. It introduces the single new keyword "<literal>rec</literal>",
871 which wraps a mutually-recursive group of monadic statements,
872 producing a single statement.
874 <para>Similar to a <literal>let</literal>
875 statement, the variables bound in the <literal>rec</literal> are
876 visible throughout the <literal>rec</literal> group, and below it.
879 do { a <- getChar do { a <- getChar
880 ; let { r1 = f a r2 ; rec { r1 <- f a r2
881 ; r2 = g r1 } ; r2 <- g r1 }
882 ; return (r1 ++ r2) } ; return (r1 ++ r2) }
884 In both cases, <literal>r1</literal> and <literal>r2</literal> are
885 available both throughout the <literal>let</literal> or <literal>rec</literal> block, and
886 in the statements that follow it. The difference is that <literal>let</literal> is non-monadic,
887 while <literal>rec</literal> is monadic. (In Haskell <literal>let</literal> is
888 really <literal>letrec</literal>, of course.)
891 The static and dynamic semantics of <literal>rec</literal> can be described as follows:
895 similar to let-bindings, the <literal>rec</literal> is broken into
896 minimal recursive groups, a process known as <emphasis>segmentation</emphasis>.
899 rec { a <- getChar ===> a <- getChar
900 ; b <- f a c rec { b <- f a c
901 ; c <- f b a ; c <- f b a }
902 ; putChar c } putChar c
904 The details of segmentation are described in Section 3.2 of
905 <ulink url="http://sites.google.com/site/leventerkok/">A recursive do for Haskell</ulink>.
906 Segmentation improves polymorphism, reduces the size of the recursive "knot", and, as the paper
907 describes, also has a semantic effect (unless the monad satisfies the right-shrinking law).
910 Then each resulting <literal>rec</literal> is desugared, using a call to <literal>Control.Monad.Fix.mfix</literal>.
911 For example, the <literal>rec</literal> group in the preceding example is desugared like this:
913 rec { b <- f a c ===> (b,c) <- mfix (\~(b,c) -> do { b <- f a c
914 ; c <- f b a } ; c <- f b a
917 In general, the statment <literal>rec <replaceable>ss</replaceable></literal>
918 is desugared to the statement
920 <replaceable>vs</replaceable> <- mfix (\~<replaceable>vs</replaceable> -> do { <replaceable>ss</replaceable>; return <replaceable>vs</replaceable> })
922 where <replaceable>vs</replaceable> is a tuple of the variables bound by <replaceable>ss</replaceable>.
924 The original <literal>rec</literal> typechecks exactly
925 when the above desugared version would do so. For example, this means that
926 the variables <replaceable>vs</replaceable> are all monomorphic in the statements
927 following the <literal>rec</literal>, because they are bound by a lambda.
930 The <literal>mfix</literal> function is defined in the <literal>MonadFix</literal>
931 class, in <literal>Control.Monad.Fix</literal>, thus:
933 class Monad m => MonadFix m where
934 mfix :: (a -> m a) -> m a
941 Here are some other important points in using the recursive-do notation:
944 It is enabled with the flag <literal>-XDoRec</literal>, which is in turn implied by
945 <literal>-fglasgow-exts</literal>.
949 If recursive bindings are required for a monad,
950 then that monad must be declared an instance of the <literal>MonadFix</literal> class.
954 The following instances of <literal>MonadFix</literal> are automatically provided: List, Maybe, IO.
955 Furthermore, the Control.Monad.ST and Control.Monad.ST.Lazy modules provide the instances of the MonadFix class
956 for Haskell's internal state monad (strict and lazy, respectively).
960 Like <literal>let</literal> and <literal>where</literal> bindings,
961 name shadowing is not allowed within a <literal>rec</literal>;
962 that is, all the names bound in a single <literal>rec</literal> must
963 be distinct (Section 3.3 of the paper).
966 It supports rebindable syntax (see <xref linkend="rebindable-syntax"/>).
972 <sect3 id="mdo-notation"> <title> Mdo-notation (deprecated) </title>
974 <para> GHC used to support the flag <option>-XRecursiveDo</option>,
975 which enabled the keyword <literal>mdo</literal>, precisely as described in
976 <ulink url="http://sites.google.com/site/leventerkok/">A recursive do for Haskell</ulink>,
977 but this is now deprecated. Instead of <literal>mdo { Q; e }</literal>, write
978 <literal>do { rec Q; e }</literal>.
981 Historical note: The old implementation of the mdo-notation (and most
982 of the existing documents) used the name
983 <literal>MonadRec</literal> for the class and the corresponding library.
984 This name is not supported by GHC.
991 <!-- ===================== PARALLEL LIST COMPREHENSIONS =================== -->
993 <sect2 id="parallel-list-comprehensions">
994 <title>Parallel List Comprehensions</title>
995 <indexterm><primary>list comprehensions</primary><secondary>parallel</secondary>
997 <indexterm><primary>parallel list comprehensions</primary>
1000 <para>Parallel list comprehensions are a natural extension to list
1001 comprehensions. List comprehensions can be thought of as a nice
1002 syntax for writing maps and filters. Parallel comprehensions
1003 extend this to include the zipWith family.</para>
1005 <para>A parallel list comprehension has multiple independent
1006 branches of qualifier lists, each separated by a `|' symbol. For
1007 example, the following zips together two lists:</para>
1010 [ (x, y) | x <- xs | y <- ys ]
1013 <para>The behavior of parallel list comprehensions follows that of
1014 zip, in that the resulting list will have the same length as the
1015 shortest branch.</para>
1017 <para>We can define parallel list comprehensions by translation to
1018 regular comprehensions. Here's the basic idea:</para>
1020 <para>Given a parallel comprehension of the form: </para>
1023 [ e | p1 <- e11, p2 <- e12, ...
1024 | q1 <- e21, q2 <- e22, ...
1029 <para>This will be translated to: </para>
1032 [ e | ((p1,p2), (q1,q2), ...) <- zipN [(p1,p2) | p1 <- e11, p2 <- e12, ...]
1033 [(q1,q2) | q1 <- e21, q2 <- e22, ...]
1038 <para>where `zipN' is the appropriate zip for the given number of
1043 <!-- ===================== TRANSFORM LIST COMPREHENSIONS =================== -->
1045 <sect2 id="generalised-list-comprehensions">
1046 <title>Generalised (SQL-Like) List Comprehensions</title>
1047 <indexterm><primary>list comprehensions</primary><secondary>generalised</secondary>
1049 <indexterm><primary>extended list comprehensions</primary>
1051 <indexterm><primary>group</primary></indexterm>
1052 <indexterm><primary>sql</primary></indexterm>
1055 <para>Generalised list comprehensions are a further enhancement to the
1056 list comprehension syntactic sugar to allow operations such as sorting
1057 and grouping which are familiar from SQL. They are fully described in the
1058 paper <ulink url="http://research.microsoft.com/~simonpj/papers/list-comp">
1059 Comprehensive comprehensions: comprehensions with "order by" and "group by"</ulink>,
1060 except that the syntax we use differs slightly from the paper.</para>
1061 <para>The extension is enabled with the flag <option>-XTransformListComp</option>.</para>
1062 <para>Here is an example:
1064 employees = [ ("Simon", "MS", 80)
1065 , ("Erik", "MS", 100)
1066 , ("Phil", "Ed", 40)
1067 , ("Gordon", "Ed", 45)
1068 , ("Paul", "Yale", 60)]
1070 output = [ (the dept, sum salary)
1071 | (name, dept, salary) <- employees
1072 , then group by dept
1073 , then sortWith by (sum salary)
1076 In this example, the list <literal>output</literal> would take on
1080 [("Yale", 60), ("Ed", 85), ("MS", 180)]
1083 <para>There are three new keywords: <literal>group</literal>, <literal>by</literal>, and <literal>using</literal>.
1084 (The function <literal>sortWith</literal> is not a keyword; it is an ordinary
1085 function that is exported by <literal>GHC.Exts</literal>.)</para>
1087 <para>There are five new forms of comprehension qualifier,
1088 all introduced by the (existing) keyword <literal>then</literal>:
1096 This statement requires that <literal>f</literal> have the type <literal>
1097 forall a. [a] -> [a]</literal>. You can see an example of its use in the
1098 motivating example, as this form is used to apply <literal>take 5</literal>.
1109 This form is similar to the previous one, but allows you to create a function
1110 which will be passed as the first argument to f. As a consequence f must have
1111 the type <literal>forall a. (a -> t) -> [a] -> [a]</literal>. As you can see
1112 from the type, this function lets f "project out" some information
1113 from the elements of the list it is transforming.</para>
1115 <para>An example is shown in the opening example, where <literal>sortWith</literal>
1116 is supplied with a function that lets it find out the <literal>sum salary</literal>
1117 for any item in the list comprehension it transforms.</para>
1125 then group by e using f
1128 <para>This is the most general of the grouping-type statements. In this form,
1129 f is required to have type <literal>forall a. (a -> t) -> [a] -> [[a]]</literal>.
1130 As with the <literal>then f by e</literal> case above, the first argument
1131 is a function supplied to f by the compiler which lets it compute e on every
1132 element of the list being transformed. However, unlike the non-grouping case,
1133 f additionally partitions the list into a number of sublists: this means that
1134 at every point after this statement, binders occurring before it in the comprehension
1135 refer to <emphasis>lists</emphasis> of possible values, not single values. To help understand
1136 this, let's look at an example:</para>
1139 -- This works similarly to groupWith in GHC.Exts, but doesn't sort its input first
1140 groupRuns :: Eq b => (a -> b) -> [a] -> [[a]]
1141 groupRuns f = groupBy (\x y -> f x == f y)
1143 output = [ (the x, y)
1144 | x <- ([1..3] ++ [1..2])
1146 , then group by x using groupRuns ]
1149 <para>This results in the variable <literal>output</literal> taking on the value below:</para>
1152 [(1, [4, 5, 6]), (2, [4, 5, 6]), (3, [4, 5, 6]), (1, [4, 5, 6]), (2, [4, 5, 6])]
1155 <para>Note that we have used the <literal>the</literal> function to change the type
1156 of x from a list to its original numeric type. The variable y, in contrast, is left
1157 unchanged from the list form introduced by the grouping.</para>
1167 <para>This form of grouping is essentially the same as the one described above. However,
1168 since no function to use for the grouping has been supplied it will fall back on the
1169 <literal>groupWith</literal> function defined in
1170 <ulink url="&libraryBaseLocation;/GHC-Exts.html"><literal>GHC.Exts</literal></ulink>. This
1171 is the form of the group statement that we made use of in the opening example.</para>
1182 <para>With this form of the group statement, f is required to simply have the type
1183 <literal>forall a. [a] -> [[a]]</literal>, which will be used to group up the
1184 comprehension so far directly. An example of this form is as follows:</para>
1190 , then group using inits]
1193 <para>This will yield a list containing every prefix of the word "hello" written out 5 times:</para>
1196 ["","h","he","hel","hell","hello","helloh","hellohe","hellohel","hellohell","hellohello","hellohelloh",...]
1204 <!-- ===================== REBINDABLE SYNTAX =================== -->
1206 <sect2 id="rebindable-syntax">
1207 <title>Rebindable syntax and the implicit Prelude import</title>
1209 <para><indexterm><primary>-XNoImplicitPrelude
1210 option</primary></indexterm> GHC normally imports
1211 <filename>Prelude.hi</filename> files for you. If you'd
1212 rather it didn't, then give it a
1213 <option>-XNoImplicitPrelude</option> option. The idea is
1214 that you can then import a Prelude of your own. (But don't
1215 call it <literal>Prelude</literal>; the Haskell module
1216 namespace is flat, and you must not conflict with any
1217 Prelude module.)</para>
1219 <para>Suppose you are importing a Prelude of your own
1220 in order to define your own numeric class
1221 hierarchy. It completely defeats that purpose if the
1222 literal "1" means "<literal>Prelude.fromInteger
1223 1</literal>", which is what the Haskell Report specifies.
1224 So the <option>-XRebindableSyntax</option>
1226 the following pieces of built-in syntax to refer to
1227 <emphasis>whatever is in scope</emphasis>, not the Prelude
1231 <para>An integer literal <literal>368</literal> means
1232 "<literal>fromInteger (368::Integer)</literal>", rather than
1233 "<literal>Prelude.fromInteger (368::Integer)</literal>".
1236 <listitem><para>Fractional literals are handed in just the same way,
1237 except that the translation is
1238 <literal>fromRational (3.68::Rational)</literal>.
1241 <listitem><para>The equality test in an overloaded numeric pattern
1242 uses whatever <literal>(==)</literal> is in scope.
1245 <listitem><para>The subtraction operation, and the
1246 greater-than-or-equal test, in <literal>n+k</literal> patterns
1247 use whatever <literal>(-)</literal> and <literal>(>=)</literal> are in scope.
1251 <para>Negation (e.g. "<literal>- (f x)</literal>")
1252 means "<literal>negate (f x)</literal>", both in numeric
1253 patterns, and expressions.
1257 <para>Conditionals (e.g. "<literal>if</literal> e1 <literal>then</literal> e2 <literal>else</literal> e3")
1258 means "<literal>ifThenElse</literal> e1 e2 e3". However <literal>case</literal> expressions are unaffected.
1262 <para>"Do" notation is translated using whatever
1263 functions <literal>(>>=)</literal>,
1264 <literal>(>>)</literal>, and <literal>fail</literal>,
1265 are in scope (not the Prelude
1266 versions). List comprehensions, mdo (<xref linkend="mdo-notation"/>), and parallel array
1267 comprehensions, are unaffected. </para></listitem>
1271 notation (see <xref linkend="arrow-notation"/>)
1272 uses whatever <literal>arr</literal>,
1273 <literal>(>>>)</literal>, <literal>first</literal>,
1274 <literal>app</literal>, <literal>(|||)</literal> and
1275 <literal>loop</literal> functions are in scope. But unlike the
1276 other constructs, the types of these functions must match the
1277 Prelude types very closely. Details are in flux; if you want
1281 <option>-XRebindableSyntax</option> implies <option>-XNoImplicitPrelude</option>.
1284 In all cases (apart from arrow notation), the static semantics should be that of the desugared form,
1285 even if that is a little unexpected. For example, the
1286 static semantics of the literal <literal>368</literal>
1287 is exactly that of <literal>fromInteger (368::Integer)</literal>; it's fine for
1288 <literal>fromInteger</literal> to have any of the types:
1290 fromInteger :: Integer -> Integer
1291 fromInteger :: forall a. Foo a => Integer -> a
1292 fromInteger :: Num a => a -> Integer
1293 fromInteger :: Integer -> Bool -> Bool
1297 <para>Be warned: this is an experimental facility, with
1298 fewer checks than usual. Use <literal>-dcore-lint</literal>
1299 to typecheck the desugared program. If Core Lint is happy
1300 you should be all right.</para>
1304 <sect2 id="postfix-operators">
1305 <title>Postfix operators</title>
1308 The <option>-XPostfixOperators</option> flag enables a small
1309 extension to the syntax of left operator sections, which allows you to
1310 define postfix operators. The extension is this: the left section
1314 is equivalent (from the point of view of both type checking and execution) to the expression
1318 (for any expression <literal>e</literal> and operator <literal>(!)</literal>.
1319 The strict Haskell 98 interpretation is that the section is equivalent to
1323 That is, the operator must be a function of two arguments. GHC allows it to
1324 take only one argument, and that in turn allows you to write the function
1327 <para>The extension does not extend to the left-hand side of function
1328 definitions; you must define such a function in prefix form.</para>
1332 <sect2 id="tuple-sections">
1333 <title>Tuple sections</title>
1336 The <option>-XTupleSections</option> flag enables Python-style partially applied
1337 tuple constructors. For example, the following program
1341 is considered to be an alternative notation for the more unwieldy alternative
1345 You can omit any combination of arguments to the tuple, as in the following
1347 (, "I", , , "Love", , 1337)
1351 \a b c d -> (a, "I", b, c, "Love", d, 1337)
1356 If you have <link linkend="unboxed-tuples">unboxed tuples</link> enabled, tuple sections
1357 will also be available for them, like so
1361 Because there is no unboxed unit tuple, the following expression
1365 continues to stand for the unboxed singleton tuple data constructor.
1370 <sect2 id="disambiguate-fields">
1371 <title>Record field disambiguation</title>
1373 In record construction and record pattern matching
1374 it is entirely unambiguous which field is referred to, even if there are two different
1375 data types in scope with a common field name. For example:
1378 data S = MkS { x :: Int, y :: Bool }
1383 data T = MkT { x :: Int }
1385 ok1 (MkS { x = n }) = n+1 -- Unambiguous
1386 ok2 n = MkT { x = n+1 } -- Unambiguous
1388 bad1 k = k { x = 3 } -- Ambiguous
1389 bad2 k = x k -- Ambiguous
1391 Even though there are two <literal>x</literal>'s in scope,
1392 it is clear that the <literal>x</literal> in the pattern in the
1393 definition of <literal>ok1</literal> can only mean the field
1394 <literal>x</literal> from type <literal>S</literal>. Similarly for
1395 the function <literal>ok2</literal>. However, in the record update
1396 in <literal>bad1</literal> and the record selection in <literal>bad2</literal>
1397 it is not clear which of the two types is intended.
1400 Haskell 98 regards all four as ambiguous, but with the
1401 <option>-XDisambiguateRecordFields</option> flag, GHC will accept
1402 the former two. The rules are precisely the same as those for instance
1403 declarations in Haskell 98, where the method names on the left-hand side
1404 of the method bindings in an instance declaration refer unambiguously
1405 to the method of that class (provided they are in scope at all), even
1406 if there are other variables in scope with the same name.
1407 This reduces the clutter of qualified names when you import two
1408 records from different modules that use the same field name.
1414 Field disambiguation can be combined with punning (see <xref linkend="record-puns"/>). For exampe:
1419 ok3 (MkS { x }) = x+1 -- Uses both disambiguation and punning
1424 With <option>-XDisambiguateRecordFields</option> you can use <emphasis>unqualifed</emphasis>
1425 field names even if the correponding selector is only in scope <emphasis>qualified</emphasis>
1426 For example, assuming the same module <literal>M</literal> as in our earlier example, this is legal:
1429 import qualified M -- Note qualified
1431 ok4 (M.MkS { x = n }) = n+1 -- Unambiguous
1433 Since the constructore <literal>MkS</literal> is only in scope qualified, you must
1434 name it <literal>M.MkS</literal>, but the field <literal>x</literal> does not need
1435 to be qualified even though <literal>M.x</literal> is in scope but <literal>x</literal>
1436 is not. (In effect, it is qualified by the constructor.)
1443 <!-- ===================== Record puns =================== -->
1445 <sect2 id="record-puns">
1450 Record puns are enabled by the flag <literal>-XNamedFieldPuns</literal>.
1454 When using records, it is common to write a pattern that binds a
1455 variable with the same name as a record field, such as:
1458 data C = C {a :: Int}
1464 Record punning permits the variable name to be elided, so one can simply
1471 to mean the same pattern as above. That is, in a record pattern, the
1472 pattern <literal>a</literal> expands into the pattern <literal>a =
1473 a</literal> for the same name <literal>a</literal>.
1480 Record punning can also be used in an expression, writing, for example,
1486 let a = 1 in C {a = a}
1488 The expansion is purely syntactic, so the expanded right-hand side
1489 expression refers to the nearest enclosing variable that is spelled the
1490 same as the field name.
1494 Puns and other patterns can be mixed in the same record:
1496 data C = C {a :: Int, b :: Int}
1497 f (C {a, b = 4}) = a
1502 Puns can be used wherever record patterns occur (e.g. in
1503 <literal>let</literal> bindings or at the top-level).
1507 A pun on a qualified field name is expanded by stripping off the module qualifier.
1514 f (M.C {M.a = a}) = a
1516 (This is useful if the field selector <literal>a</literal> for constructor <literal>M.C</literal>
1517 is only in scope in qualified form.)
1525 <!-- ===================== Record wildcards =================== -->
1527 <sect2 id="record-wildcards">
1528 <title>Record wildcards
1532 Record wildcards are enabled by the flag <literal>-XRecordWildCards</literal>.
1533 This flag implies <literal>-XDisambiguateRecordFields</literal>.
1537 For records with many fields, it can be tiresome to write out each field
1538 individually in a record pattern, as in
1540 data C = C {a :: Int, b :: Int, c :: Int, d :: Int}
1541 f (C {a = 1, b = b, c = c, d = d}) = b + c + d
1546 Record wildcard syntax permits a "<literal>..</literal>" in a record
1547 pattern, where each elided field <literal>f</literal> is replaced by the
1548 pattern <literal>f = f</literal>. For example, the above pattern can be
1551 f (C {a = 1, ..}) = b + c + d
1559 Wildcards can be mixed with other patterns, including puns
1560 (<xref linkend="record-puns"/>); for example, in a pattern <literal>C {a
1561 = 1, b, ..})</literal>. Additionally, record wildcards can be used
1562 wherever record patterns occur, including in <literal>let</literal>
1563 bindings and at the top-level. For example, the top-level binding
1567 defines <literal>b</literal>, <literal>c</literal>, and
1568 <literal>d</literal>.
1572 Record wildcards can also be used in expressions, writing, for example,
1574 let {a = 1; b = 2; c = 3; d = 4} in C {..}
1578 let {a = 1; b = 2; c = 3; d = 4} in C {a=a, b=b, c=c, d=d}
1580 The expansion is purely syntactic, so the record wildcard
1581 expression refers to the nearest enclosing variables that are spelled
1582 the same as the omitted field names.
1586 The "<literal>..</literal>" expands to the missing
1587 <emphasis>in-scope</emphasis> record fields, where "in scope"
1588 includes both unqualified and qualified-only.
1589 Any fields that are not in scope are not filled in. For example
1592 data R = R { a,b,c :: Int }
1594 import qualified M( R(a,b) )
1597 The <literal>{..}</literal> expands to <literal>{M.a=a,M.b=b}</literal>,
1598 omitting <literal>c</literal> since it is not in scope at all.
1605 <!-- ===================== Local fixity declarations =================== -->
1607 <sect2 id="local-fixity-declarations">
1608 <title>Local Fixity Declarations
1611 <para>A careful reading of the Haskell 98 Report reveals that fixity
1612 declarations (<literal>infix</literal>, <literal>infixl</literal>, and
1613 <literal>infixr</literal>) are permitted to appear inside local bindings
1614 such those introduced by <literal>let</literal> and
1615 <literal>where</literal>. However, the Haskell Report does not specify
1616 the semantics of such bindings very precisely.
1619 <para>In GHC, a fixity declaration may accompany a local binding:
1626 and the fixity declaration applies wherever the binding is in scope.
1627 For example, in a <literal>let</literal>, it applies in the right-hand
1628 sides of other <literal>let</literal>-bindings and the body of the
1629 <literal>let</literal>C. Or, in recursive <literal>do</literal>
1630 expressions (<xref linkend="recursive-do-notation"/>), the local fixity
1631 declarations of a <literal>let</literal> statement scope over other
1632 statements in the group, just as the bound name does.
1636 Moreover, a local fixity declaration *must* accompany a local binding of
1637 that name: it is not possible to revise the fixity of name bound
1640 let infixr 9 $ in ...
1643 Because local fixity declarations are technically Haskell 98, no flag is
1644 necessary to enable them.
1648 <sect2 id="package-imports">
1649 <title>Package-qualified imports</title>
1651 <para>With the <option>-XPackageImports</option> flag, GHC allows
1652 import declarations to be qualified by the package name that the
1653 module is intended to be imported from. For example:</para>
1656 import "network" Network.Socket
1659 <para>would import the module <literal>Network.Socket</literal> from
1660 the package <literal>network</literal> (any version). This may
1661 be used to disambiguate an import when the same module is
1662 available from multiple packages, or is present in both the
1663 current package being built and an external package.</para>
1665 <para>Note: you probably don't need to use this feature, it was
1666 added mainly so that we can build backwards-compatible versions of
1667 packages when APIs change. It can lead to fragile dependencies in
1668 the common case: modules occasionally move from one package to
1669 another, rendering any package-qualified imports broken.</para>
1672 <sect2 id="syntax-stolen">
1673 <title>Summary of stolen syntax</title>
1675 <para>Turning on an option that enables special syntax
1676 <emphasis>might</emphasis> cause working Haskell 98 code to fail
1677 to compile, perhaps because it uses a variable name which has
1678 become a reserved word. This section lists the syntax that is
1679 "stolen" by language extensions.
1681 notation and nonterminal names from the Haskell 98 lexical syntax
1682 (see the Haskell 98 Report).
1683 We only list syntax changes here that might affect
1684 existing working programs (i.e. "stolen" syntax). Many of these
1685 extensions will also enable new context-free syntax, but in all
1686 cases programs written to use the new syntax would not be
1687 compilable without the option enabled.</para>
1689 <para>There are two classes of special
1694 <para>New reserved words and symbols: character sequences
1695 which are no longer available for use as identifiers in the
1699 <para>Other special syntax: sequences of characters that have
1700 a different meaning when this particular option is turned
1705 The following syntax is stolen:
1710 <literal>forall</literal>
1711 <indexterm><primary><literal>forall</literal></primary></indexterm>
1714 Stolen (in types) by: <option>-XExplicitForAll</option>, and hence by
1715 <option>-XScopedTypeVariables</option>,
1716 <option>-XLiberalTypeSynonyms</option>,
1717 <option>-XRank2Types</option>,
1718 <option>-XRankNTypes</option>,
1719 <option>-XPolymorphicComponents</option>,
1720 <option>-XExistentialQuantification</option>
1726 <literal>mdo</literal>
1727 <indexterm><primary><literal>mdo</literal></primary></indexterm>
1730 Stolen by: <option>-XRecursiveDo</option>,
1736 <literal>foreign</literal>
1737 <indexterm><primary><literal>foreign</literal></primary></indexterm>
1740 Stolen by: <option>-XForeignFunctionInterface</option>,
1746 <literal>rec</literal>,
1747 <literal>proc</literal>, <literal>-<</literal>,
1748 <literal>>-</literal>, <literal>-<<</literal>,
1749 <literal>>>-</literal>, and <literal>(|</literal>,
1750 <literal>|)</literal> brackets
1751 <indexterm><primary><literal>proc</literal></primary></indexterm>
1754 Stolen by: <option>-XArrows</option>,
1760 <literal>?<replaceable>varid</replaceable></literal>,
1761 <literal>%<replaceable>varid</replaceable></literal>
1762 <indexterm><primary>implicit parameters</primary></indexterm>
1765 Stolen by: <option>-XImplicitParams</option>,
1771 <literal>[|</literal>,
1772 <literal>[e|</literal>, <literal>[p|</literal>,
1773 <literal>[d|</literal>, <literal>[t|</literal>,
1774 <literal>$(</literal>,
1775 <literal>$<replaceable>varid</replaceable></literal>
1776 <indexterm><primary>Template Haskell</primary></indexterm>
1779 Stolen by: <option>-XTemplateHaskell</option>,
1785 <literal>[:<replaceable>varid</replaceable>|</literal>
1786 <indexterm><primary>quasi-quotation</primary></indexterm>
1789 Stolen by: <option>-XQuasiQuotes</option>,
1795 <replaceable>varid</replaceable>{<literal>#</literal>},
1796 <replaceable>char</replaceable><literal>#</literal>,
1797 <replaceable>string</replaceable><literal>#</literal>,
1798 <replaceable>integer</replaceable><literal>#</literal>,
1799 <replaceable>float</replaceable><literal>#</literal>,
1800 <replaceable>float</replaceable><literal>##</literal>,
1801 <literal>(#</literal>, <literal>#)</literal>,
1804 Stolen by: <option>-XMagicHash</option>,
1813 <!-- TYPE SYSTEM EXTENSIONS -->
1814 <sect1 id="data-type-extensions">
1815 <title>Extensions to data types and type synonyms</title>
1817 <sect2 id="nullary-types">
1818 <title>Data types with no constructors</title>
1820 <para>With the <option>-fglasgow-exts</option> flag, GHC lets you declare
1821 a data type with no constructors. For example:</para>
1825 data T a -- T :: * -> *
1828 <para>Syntactically, the declaration lacks the "= constrs" part. The
1829 type can be parameterised over types of any kind, but if the kind is
1830 not <literal>*</literal> then an explicit kind annotation must be used
1831 (see <xref linkend="kinding"/>).</para>
1833 <para>Such data types have only one value, namely bottom.
1834 Nevertheless, they can be useful when defining "phantom types".</para>
1837 <sect2 id="datatype-contexts">
1838 <title>Data type contexts</title>
1840 <para>Haskell allows datatypes to be given contexts, e.g.</para>
1843 data Eq a => Set a = NilSet | ConsSet a (Set a)
1846 <para>give constructors with types:</para>
1850 ConsSet :: Eq a => a -> Set a -> Set a
1853 <para>In GHC this feature is an extension called
1854 <literal>DatatypeContexts</literal>, and on by default.</para>
1857 <sect2 id="infix-tycons">
1858 <title>Infix type constructors, classes, and type variables</title>
1861 GHC allows type constructors, classes, and type variables to be operators, and
1862 to be written infix, very much like expressions. More specifically:
1865 A type constructor or class can be an operator, beginning with a colon; e.g. <literal>:*:</literal>.
1866 The lexical syntax is the same as that for data constructors.
1869 Data type and type-synonym declarations can be written infix, parenthesised
1870 if you want further arguments. E.g.
1872 data a :*: b = Foo a b
1873 type a :+: b = Either a b
1874 class a :=: b where ...
1876 data (a :**: b) x = Baz a b x
1877 type (a :++: b) y = Either (a,b) y
1881 Types, and class constraints, can be written infix. For example
1884 f :: (a :=: b) => a -> b
1888 A type variable can be an (unqualified) operator e.g. <literal>+</literal>.
1889 The lexical syntax is the same as that for variable operators, excluding "(.)",
1890 "(!)", and "(*)". In a binding position, the operator must be
1891 parenthesised. For example:
1893 type T (+) = Int + Int
1897 liftA2 :: Arrow (~>)
1898 => (a -> b -> c) -> (e ~> a) -> (e ~> b) -> (e ~> c)
1904 as for expressions, both for type constructors and type variables; e.g. <literal>Int `Either` Bool</literal>, or
1905 <literal>Int `a` Bool</literal>. Similarly, parentheses work the same; e.g. <literal>(:*:) Int Bool</literal>.
1908 Fixities may be declared for type constructors, or classes, just as for data constructors. However,
1909 one cannot distinguish between the two in a fixity declaration; a fixity declaration
1910 sets the fixity for a data constructor and the corresponding type constructor. For example:
1914 sets the fixity for both type constructor <literal>T</literal> and data constructor <literal>T</literal>,
1915 and similarly for <literal>:*:</literal>.
1916 <literal>Int `a` Bool</literal>.
1919 Function arrow is <literal>infixr</literal> with fixity 0. (This might change; I'm not sure what it should be.)
1926 <sect2 id="type-synonyms">
1927 <title>Liberalised type synonyms</title>
1930 Type synonyms are like macros at the type level, but Haskell 98 imposes many rules
1931 on individual synonym declarations.
1932 With the <option>-XLiberalTypeSynonyms</option> extension,
1933 GHC does validity checking on types <emphasis>only after expanding type synonyms</emphasis>.
1934 That means that GHC can be very much more liberal about type synonyms than Haskell 98.
1937 <listitem> <para>You can write a <literal>forall</literal> (including overloading)
1938 in a type synonym, thus:
1940 type Discard a = forall b. Show b => a -> b -> (a, String)
1945 g :: Discard Int -> (Int,String) -- A rank-2 type
1952 If you also use <option>-XUnboxedTuples</option>,
1953 you can write an unboxed tuple in a type synonym:
1955 type Pr = (# Int, Int #)
1963 You can apply a type synonym to a forall type:
1965 type Foo a = a -> a -> Bool
1967 f :: Foo (forall b. b->b)
1969 After expanding the synonym, <literal>f</literal> has the legal (in GHC) type:
1971 f :: (forall b. b->b) -> (forall b. b->b) -> Bool
1976 You can apply a type synonym to a partially applied type synonym:
1978 type Generic i o = forall x. i x -> o x
1981 foo :: Generic Id []
1983 After expanding the synonym, <literal>foo</literal> has the legal (in GHC) type:
1985 foo :: forall x. x -> [x]
1993 GHC currently does kind checking before expanding synonyms (though even that
1997 After expanding type synonyms, GHC does validity checking on types, looking for
1998 the following mal-formedness which isn't detected simply by kind checking:
2001 Type constructor applied to a type involving for-alls.
2004 Unboxed tuple on left of an arrow.
2007 Partially-applied type synonym.
2011 this will be rejected:
2013 type Pr = (# Int, Int #)
2018 because GHC does not allow unboxed tuples on the left of a function arrow.
2023 <sect2 id="existential-quantification">
2024 <title>Existentially quantified data constructors
2028 The idea of using existential quantification in data type declarations
2029 was suggested by Perry, and implemented in Hope+ (Nigel Perry, <emphasis>The Implementation
2030 of Practical Functional Programming Languages</emphasis>, PhD Thesis, University of
2031 London, 1991). It was later formalised by Laufer and Odersky
2032 (<emphasis>Polymorphic type inference and abstract data types</emphasis>,
2033 TOPLAS, 16(5), pp1411-1430, 1994).
2034 It's been in Lennart
2035 Augustsson's <command>hbc</command> Haskell compiler for several years, and
2036 proved very useful. Here's the idea. Consider the declaration:
2042 data Foo = forall a. MkFoo a (a -> Bool)
2049 The data type <literal>Foo</literal> has two constructors with types:
2055 MkFoo :: forall a. a -> (a -> Bool) -> Foo
2062 Notice that the type variable <literal>a</literal> in the type of <function>MkFoo</function>
2063 does not appear in the data type itself, which is plain <literal>Foo</literal>.
2064 For example, the following expression is fine:
2070 [MkFoo 3 even, MkFoo 'c' isUpper] :: [Foo]
2076 Here, <literal>(MkFoo 3 even)</literal> packages an integer with a function
2077 <function>even</function> that maps an integer to <literal>Bool</literal>; and <function>MkFoo 'c'
2078 isUpper</function> packages a character with a compatible function. These
2079 two things are each of type <literal>Foo</literal> and can be put in a list.
2083 What can we do with a value of type <literal>Foo</literal>?. In particular,
2084 what happens when we pattern-match on <function>MkFoo</function>?
2090 f (MkFoo val fn) = ???
2096 Since all we know about <literal>val</literal> and <function>fn</function> is that they
2097 are compatible, the only (useful) thing we can do with them is to
2098 apply <function>fn</function> to <literal>val</literal> to get a boolean. For example:
2105 f (MkFoo val fn) = fn val
2111 What this allows us to do is to package heterogeneous values
2112 together with a bunch of functions that manipulate them, and then treat
2113 that collection of packages in a uniform manner. You can express
2114 quite a bit of object-oriented-like programming this way.
2117 <sect3 id="existential">
2118 <title>Why existential?
2122 What has this to do with <emphasis>existential</emphasis> quantification?
2123 Simply that <function>MkFoo</function> has the (nearly) isomorphic type
2129 MkFoo :: (exists a . (a, a -> Bool)) -> Foo
2135 But Haskell programmers can safely think of the ordinary
2136 <emphasis>universally</emphasis> quantified type given above, thereby avoiding
2137 adding a new existential quantification construct.
2142 <sect3 id="existential-with-context">
2143 <title>Existentials and type classes</title>
2146 An easy extension is to allow
2147 arbitrary contexts before the constructor. For example:
2153 data Baz = forall a. Eq a => Baz1 a a
2154 | forall b. Show b => Baz2 b (b -> b)
2160 The two constructors have the types you'd expect:
2166 Baz1 :: forall a. Eq a => a -> a -> Baz
2167 Baz2 :: forall b. Show b => b -> (b -> b) -> Baz
2173 But when pattern matching on <function>Baz1</function> the matched values can be compared
2174 for equality, and when pattern matching on <function>Baz2</function> the first matched
2175 value can be converted to a string (as well as applying the function to it).
2176 So this program is legal:
2183 f (Baz1 p q) | p == q = "Yes"
2185 f (Baz2 v fn) = show (fn v)
2191 Operationally, in a dictionary-passing implementation, the
2192 constructors <function>Baz1</function> and <function>Baz2</function> must store the
2193 dictionaries for <literal>Eq</literal> and <literal>Show</literal> respectively, and
2194 extract it on pattern matching.
2199 <sect3 id="existential-records">
2200 <title>Record Constructors</title>
2203 GHC allows existentials to be used with records syntax as well. For example:
2206 data Counter a = forall self. NewCounter
2208 , _inc :: self -> self
2209 , _display :: self -> IO ()
2213 Here <literal>tag</literal> is a public field, with a well-typed selector
2214 function <literal>tag :: Counter a -> a</literal>. The <literal>self</literal>
2215 type is hidden from the outside; any attempt to apply <literal>_this</literal>,
2216 <literal>_inc</literal> or <literal>_display</literal> as functions will raise a
2217 compile-time error. In other words, <emphasis>GHC defines a record selector function
2218 only for fields whose type does not mention the existentially-quantified variables</emphasis>.
2219 (This example used an underscore in the fields for which record selectors
2220 will not be defined, but that is only programming style; GHC ignores them.)
2224 To make use of these hidden fields, we need to create some helper functions:
2227 inc :: Counter a -> Counter a
2228 inc (NewCounter x i d t) = NewCounter
2229 { _this = i x, _inc = i, _display = d, tag = t }
2231 display :: Counter a -> IO ()
2232 display NewCounter{ _this = x, _display = d } = d x
2235 Now we can define counters with different underlying implementations:
2238 counterA :: Counter String
2239 counterA = NewCounter
2240 { _this = 0, _inc = (1+), _display = print, tag = "A" }
2242 counterB :: Counter String
2243 counterB = NewCounter
2244 { _this = "", _inc = ('#':), _display = putStrLn, tag = "B" }
2247 display (inc counterA) -- prints "1"
2248 display (inc (inc counterB)) -- prints "##"
2251 Record update syntax is supported for existentials (and GADTs):
2253 setTag :: Counter a -> a -> Counter a
2254 setTag obj t = obj{ tag = t }
2256 The rule for record update is this: <emphasis>
2257 the types of the updated fields may
2258 mention only the universally-quantified type variables
2259 of the data constructor. For GADTs, the field may mention only types
2260 that appear as a simple type-variable argument in the constructor's result
2261 type</emphasis>. For example:
2263 data T a b where { T1 { f1::a, f2::b, f3::(b,c) } :: T a b } -- c is existential
2264 upd1 t x = t { f1=x } -- OK: upd1 :: T a b -> a' -> T a' b
2265 upd2 t x = t { f3=x } -- BAD (f3's type mentions c, which is
2266 -- existentially quantified)
2268 data G a b where { G1 { g1::a, g2::c } :: G a [c] }
2269 upd3 g x = g { g1=x } -- OK: upd3 :: G a b -> c -> G c b
2270 upd4 g x = g { g2=x } -- BAD (f2's type mentions c, which is not a simple
2271 -- type-variable argument in G1's result type)
2279 <title>Restrictions</title>
2282 There are several restrictions on the ways in which existentially-quantified
2283 constructors can be use.
2292 When pattern matching, each pattern match introduces a new,
2293 distinct, type for each existential type variable. These types cannot
2294 be unified with any other type, nor can they escape from the scope of
2295 the pattern match. For example, these fragments are incorrect:
2303 Here, the type bound by <function>MkFoo</function> "escapes", because <literal>a</literal>
2304 is the result of <function>f1</function>. One way to see why this is wrong is to
2305 ask what type <function>f1</function> has:
2309 f1 :: Foo -> a -- Weird!
2313 What is this "<literal>a</literal>" in the result type? Clearly we don't mean
2318 f1 :: forall a. Foo -> a -- Wrong!
2322 The original program is just plain wrong. Here's another sort of error
2326 f2 (Baz1 a b) (Baz1 p q) = a==q
2330 It's ok to say <literal>a==b</literal> or <literal>p==q</literal>, but
2331 <literal>a==q</literal> is wrong because it equates the two distinct types arising
2332 from the two <function>Baz1</function> constructors.
2340 You can't pattern-match on an existentially quantified
2341 constructor in a <literal>let</literal> or <literal>where</literal> group of
2342 bindings. So this is illegal:
2346 f3 x = a==b where { Baz1 a b = x }
2349 Instead, use a <literal>case</literal> expression:
2352 f3 x = case x of Baz1 a b -> a==b
2355 In general, you can only pattern-match
2356 on an existentially-quantified constructor in a <literal>case</literal> expression or
2357 in the patterns of a function definition.
2359 The reason for this restriction is really an implementation one.
2360 Type-checking binding groups is already a nightmare without
2361 existentials complicating the picture. Also an existential pattern
2362 binding at the top level of a module doesn't make sense, because it's
2363 not clear how to prevent the existentially-quantified type "escaping".
2364 So for now, there's a simple-to-state restriction. We'll see how
2372 You can't use existential quantification for <literal>newtype</literal>
2373 declarations. So this is illegal:
2377 newtype T = forall a. Ord a => MkT a
2381 Reason: a value of type <literal>T</literal> must be represented as a
2382 pair of a dictionary for <literal>Ord t</literal> and a value of type
2383 <literal>t</literal>. That contradicts the idea that
2384 <literal>newtype</literal> should have no concrete representation.
2385 You can get just the same efficiency and effect by using
2386 <literal>data</literal> instead of <literal>newtype</literal>. If
2387 there is no overloading involved, then there is more of a case for
2388 allowing an existentially-quantified <literal>newtype</literal>,
2389 because the <literal>data</literal> version does carry an
2390 implementation cost, but single-field existentially quantified
2391 constructors aren't much use. So the simple restriction (no
2392 existential stuff on <literal>newtype</literal>) stands, unless there
2393 are convincing reasons to change it.
2401 You can't use <literal>deriving</literal> to define instances of a
2402 data type with existentially quantified data constructors.
2404 Reason: in most cases it would not make sense. For example:;
2407 data T = forall a. MkT [a] deriving( Eq )
2410 To derive <literal>Eq</literal> in the standard way we would need to have equality
2411 between the single component of two <function>MkT</function> constructors:
2415 (MkT a) == (MkT b) = ???
2418 But <varname>a</varname> and <varname>b</varname> have distinct types, and so can't be compared.
2419 It's just about possible to imagine examples in which the derived instance
2420 would make sense, but it seems altogether simpler simply to prohibit such
2421 declarations. Define your own instances!
2432 <!-- ====================== Generalised algebraic data types ======================= -->
2434 <sect2 id="gadt-style">
2435 <title>Declaring data types with explicit constructor signatures</title>
2437 <para>GHC allows you to declare an algebraic data type by
2438 giving the type signatures of constructors explicitly. For example:
2442 Just :: a -> Maybe a
2444 The form is called a "GADT-style declaration"
2445 because Generalised Algebraic Data Types, described in <xref linkend="gadt"/>,
2446 can only be declared using this form.</para>
2447 <para>Notice that GADT-style syntax generalises existential types (<xref linkend="existential-quantification"/>).
2448 For example, these two declarations are equivalent:
2450 data Foo = forall a. MkFoo a (a -> Bool)
2451 data Foo' where { MKFoo :: a -> (a->Bool) -> Foo' }
2454 <para>Any data type that can be declared in standard Haskell-98 syntax
2455 can also be declared using GADT-style syntax.
2456 The choice is largely stylistic, but GADT-style declarations differ in one important respect:
2457 they treat class constraints on the data constructors differently.
2458 Specifically, if the constructor is given a type-class context, that
2459 context is made available by pattern matching. For example:
2462 MkSet :: Eq a => [a] -> Set a
2464 makeSet :: Eq a => [a] -> Set a
2465 makeSet xs = MkSet (nub xs)
2467 insert :: a -> Set a -> Set a
2468 insert a (MkSet as) | a `elem` as = MkSet as
2469 | otherwise = MkSet (a:as)
2471 A use of <literal>MkSet</literal> as a constructor (e.g. in the definition of <literal>makeSet</literal>)
2472 gives rise to a <literal>(Eq a)</literal>
2473 constraint, as you would expect. The new feature is that pattern-matching on <literal>MkSet</literal>
2474 (as in the definition of <literal>insert</literal>) makes <emphasis>available</emphasis> an <literal>(Eq a)</literal>
2475 context. In implementation terms, the <literal>MkSet</literal> constructor has a hidden field that stores
2476 the <literal>(Eq a)</literal> dictionary that is passed to <literal>MkSet</literal>; so
2477 when pattern-matching that dictionary becomes available for the right-hand side of the match.
2478 In the example, the equality dictionary is used to satisfy the equality constraint
2479 generated by the call to <literal>elem</literal>, so that the type of
2480 <literal>insert</literal> itself has no <literal>Eq</literal> constraint.
2483 For example, one possible application is to reify dictionaries:
2485 data NumInst a where
2486 MkNumInst :: Num a => NumInst a
2488 intInst :: NumInst Int
2491 plus :: NumInst a -> a -> a -> a
2492 plus MkNumInst p q = p + q
2494 Here, a value of type <literal>NumInst a</literal> is equivalent
2495 to an explicit <literal>(Num a)</literal> dictionary.
2498 All this applies to constructors declared using the syntax of <xref linkend="existential-with-context"/>.
2499 For example, the <literal>NumInst</literal> data type above could equivalently be declared
2503 = Num a => MkNumInst (NumInst a)
2505 Notice that, unlike the situation when declaring an existential, there is
2506 no <literal>forall</literal>, because the <literal>Num</literal> constrains the
2507 data type's universally quantified type variable <literal>a</literal>.
2508 A constructor may have both universal and existential type variables: for example,
2509 the following two declarations are equivalent:
2512 = forall b. (Num a, Eq b) => MkT1 a b
2514 MkT2 :: (Num a, Eq b) => a -> b -> T2 a
2517 <para>All this behaviour contrasts with Haskell 98's peculiar treatment of
2518 contexts on a data type declaration (Section 4.2.1 of the Haskell 98 Report).
2519 In Haskell 98 the definition
2521 data Eq a => Set' a = MkSet' [a]
2523 gives <literal>MkSet'</literal> the same type as <literal>MkSet</literal> above. But instead of
2524 <emphasis>making available</emphasis> an <literal>(Eq a)</literal> constraint, pattern-matching
2525 on <literal>MkSet'</literal> <emphasis>requires</emphasis> an <literal>(Eq a)</literal> constraint!
2526 GHC faithfully implements this behaviour, odd though it is. But for GADT-style declarations,
2527 GHC's behaviour is much more useful, as well as much more intuitive.
2531 The rest of this section gives further details about GADT-style data
2536 The result type of each data constructor must begin with the type constructor being defined.
2537 If the result type of all constructors
2538 has the form <literal>T a1 ... an</literal>, where <literal>a1 ... an</literal>
2539 are distinct type variables, then the data type is <emphasis>ordinary</emphasis>;
2540 otherwise is a <emphasis>generalised</emphasis> data type (<xref linkend="gadt"/>).
2544 As with other type signatures, you can give a single signature for several data constructors.
2545 In this example we give a single signature for <literal>T1</literal> and <literal>T2</literal>:
2554 The type signature of
2555 each constructor is independent, and is implicitly universally quantified as usual.
2556 In particular, the type variable(s) in the "<literal>data T a where</literal>" header
2557 have no scope, and different constructors may have different universally-quantified type variables:
2559 data T a where -- The 'a' has no scope
2560 T1,T2 :: b -> T b -- Means forall b. b -> T b
2561 T3 :: T a -- Means forall a. T a
2566 A constructor signature may mention type class constraints, which can differ for
2567 different constructors. For example, this is fine:
2570 T1 :: Eq b => b -> b -> T b
2571 T2 :: (Show c, Ix c) => c -> [c] -> T c
2573 When patten matching, these constraints are made available to discharge constraints
2574 in the body of the match. For example:
2577 f (T1 x y) | x==y = "yes"
2581 Note that <literal>f</literal> is not overloaded; the <literal>Eq</literal> constraint arising
2582 from the use of <literal>==</literal> is discharged by the pattern match on <literal>T1</literal>
2583 and similarly the <literal>Show</literal> constraint arising from the use of <literal>show</literal>.
2587 Unlike a Haskell-98-style
2588 data type declaration, the type variable(s) in the "<literal>data Set a where</literal>" header
2589 have no scope. Indeed, one can write a kind signature instead:
2591 data Set :: * -> * where ...
2593 or even a mixture of the two:
2595 data Bar a :: (* -> *) -> * where ...
2597 The type variables (if given) may be explicitly kinded, so we could also write the header for <literal>Foo</literal>
2600 data Bar a (b :: * -> *) where ...
2606 You can use strictness annotations, in the obvious places
2607 in the constructor type:
2610 Lit :: !Int -> Term Int
2611 If :: Term Bool -> !(Term a) -> !(Term a) -> Term a
2612 Pair :: Term a -> Term b -> Term (a,b)
2617 You can use a <literal>deriving</literal> clause on a GADT-style data type
2618 declaration. For example, these two declarations are equivalent
2620 data Maybe1 a where {
2621 Nothing1 :: Maybe1 a ;
2622 Just1 :: a -> Maybe1 a
2623 } deriving( Eq, Ord )
2625 data Maybe2 a = Nothing2 | Just2 a
2631 The type signature may have quantified type variables that do not appear
2635 MkFoo :: a -> (a->Bool) -> Foo
2638 Here the type variable <literal>a</literal> does not appear in the result type
2639 of either constructor.
2640 Although it is universally quantified in the type of the constructor, such
2641 a type variable is often called "existential".
2642 Indeed, the above declaration declares precisely the same type as
2643 the <literal>data Foo</literal> in <xref linkend="existential-quantification"/>.
2645 The type may contain a class context too, of course:
2648 MkShowable :: Show a => a -> Showable
2653 You can use record syntax on a GADT-style data type declaration:
2657 Adult :: { name :: String, children :: [Person] } -> Person
2658 Child :: Show a => { name :: !String, funny :: a } -> Person
2660 As usual, for every constructor that has a field <literal>f</literal>, the type of
2661 field <literal>f</literal> must be the same (modulo alpha conversion).
2662 The <literal>Child</literal> constructor above shows that the signature
2663 may have a context, existentially-quantified variables, and strictness annotations,
2664 just as in the non-record case. (NB: the "type" that follows the double-colon
2665 is not really a type, because of the record syntax and strictness annotations.
2666 A "type" of this form can appear only in a constructor signature.)
2670 Record updates are allowed with GADT-style declarations,
2671 only fields that have the following property: the type of the field
2672 mentions no existential type variables.
2676 As in the case of existentials declared using the Haskell-98-like record syntax
2677 (<xref linkend="existential-records"/>),
2678 record-selector functions are generated only for those fields that have well-typed
2680 Here is the example of that section, in GADT-style syntax:
2682 data Counter a where
2683 NewCounter { _this :: self
2684 , _inc :: self -> self
2685 , _display :: self -> IO ()
2690 As before, only one selector function is generated here, that for <literal>tag</literal>.
2691 Nevertheless, you can still use all the field names in pattern matching and record construction.
2693 </itemizedlist></para>
2697 <title>Generalised Algebraic Data Types (GADTs)</title>
2699 <para>Generalised Algebraic Data Types generalise ordinary algebraic data types
2700 by allowing constructors to have richer return types. Here is an example:
2703 Lit :: Int -> Term Int
2704 Succ :: Term Int -> Term Int
2705 IsZero :: Term Int -> Term Bool
2706 If :: Term Bool -> Term a -> Term a -> Term a
2707 Pair :: Term a -> Term b -> Term (a,b)
2709 Notice that the return type of the constructors is not always <literal>Term a</literal>, as is the
2710 case with ordinary data types. This generality allows us to
2711 write a well-typed <literal>eval</literal> function
2712 for these <literal>Terms</literal>:
2716 eval (Succ t) = 1 + eval t
2717 eval (IsZero t) = eval t == 0
2718 eval (If b e1 e2) = if eval b then eval e1 else eval e2
2719 eval (Pair e1 e2) = (eval e1, eval e2)
2721 The key point about GADTs is that <emphasis>pattern matching causes type refinement</emphasis>.
2722 For example, in the right hand side of the equation
2727 the type <literal>a</literal> is refined to <literal>Int</literal>. That's the whole point!
2728 A precise specification of the type rules is beyond what this user manual aspires to,
2729 but the design closely follows that described in
2731 url="http://research.microsoft.com/%7Esimonpj/papers/gadt/">Simple
2732 unification-based type inference for GADTs</ulink>,
2734 The general principle is this: <emphasis>type refinement is only carried out
2735 based on user-supplied type annotations</emphasis>.
2736 So if no type signature is supplied for <literal>eval</literal>, no type refinement happens,
2737 and lots of obscure error messages will
2738 occur. However, the refinement is quite general. For example, if we had:
2740 eval :: Term a -> a -> a
2741 eval (Lit i) j = i+j
2743 the pattern match causes the type <literal>a</literal> to be refined to <literal>Int</literal> (because of the type
2744 of the constructor <literal>Lit</literal>), and that refinement also applies to the type of <literal>j</literal>, and
2745 the result type of the <literal>case</literal> expression. Hence the addition <literal>i+j</literal> is legal.
2748 These and many other examples are given in papers by Hongwei Xi, and
2749 Tim Sheard. There is a longer introduction
2750 <ulink url="http://www.haskell.org/haskellwiki/GADT">on the wiki</ulink>,
2752 <ulink url="http://www.informatik.uni-bonn.de/~ralf/publications/With.pdf">Fun with phantom types</ulink> also has a number of examples. Note that papers
2753 may use different notation to that implemented in GHC.
2756 The rest of this section outlines the extensions to GHC that support GADTs. The extension is enabled with
2757 <option>-XGADTs</option>. The <option>-XGADTs</option> flag also sets <option>-XRelaxedPolyRec</option>.
2760 A GADT can only be declared using GADT-style syntax (<xref linkend="gadt-style"/>);
2761 the old Haskell-98 syntax for data declarations always declares an ordinary data type.
2762 The result type of each constructor must begin with the type constructor being defined,
2763 but for a GADT the arguments to the type constructor can be arbitrary monotypes.
2764 For example, in the <literal>Term</literal> data
2765 type above, the type of each constructor must end with <literal>Term ty</literal>, but
2766 the <literal>ty</literal> need not be a type variable (e.g. the <literal>Lit</literal>
2771 It is permitted to declare an ordinary algebraic data type using GADT-style syntax.
2772 What makes a GADT into a GADT is not the syntax, but rather the presence of data constructors
2773 whose result type is not just <literal>T a b</literal>.
2777 You cannot use a <literal>deriving</literal> clause for a GADT; only for
2778 an ordinary data type.
2782 As mentioned in <xref linkend="gadt-style"/>, record syntax is supported.
2786 Lit { val :: Int } :: Term Int
2787 Succ { num :: Term Int } :: Term Int
2788 Pred { num :: Term Int } :: Term Int
2789 IsZero { arg :: Term Int } :: Term Bool
2790 Pair { arg1 :: Term a
2793 If { cnd :: Term Bool
2798 However, for GADTs there is the following additional constraint:
2799 every constructor that has a field <literal>f</literal> must have
2800 the same result type (modulo alpha conversion)
2801 Hence, in the above example, we cannot merge the <literal>num</literal>
2802 and <literal>arg</literal> fields above into a
2803 single name. Although their field types are both <literal>Term Int</literal>,
2804 their selector functions actually have different types:
2807 num :: Term Int -> Term Int
2808 arg :: Term Bool -> Term Int
2813 When pattern-matching against data constructors drawn from a GADT,
2814 for example in a <literal>case</literal> expression, the following rules apply:
2816 <listitem><para>The type of the scrutinee must be rigid.</para></listitem>
2817 <listitem><para>The type of the entire <literal>case</literal> expression must be rigid.</para></listitem>
2818 <listitem><para>The type of any free variable mentioned in any of
2819 the <literal>case</literal> alternatives must be rigid.</para></listitem>
2821 A type is "rigid" if it is completely known to the compiler at its binding site. The easiest
2822 way to ensure that a variable a rigid type is to give it a type signature.
2823 For more precise details see <ulink url="http://research.microsoft.com/%7Esimonpj/papers/gadt">
2824 Simple unification-based type inference for GADTs
2825 </ulink>. The criteria implemented by GHC are given in the Appendix.
2835 <!-- ====================== End of Generalised algebraic data types ======================= -->
2837 <sect1 id="deriving">
2838 <title>Extensions to the "deriving" mechanism</title>
2840 <sect2 id="deriving-inferred">
2841 <title>Inferred context for deriving clauses</title>
2844 The Haskell Report is vague about exactly when a <literal>deriving</literal> clause is
2847 data T0 f a = MkT0 a deriving( Eq )
2848 data T1 f a = MkT1 (f a) deriving( Eq )
2849 data T2 f a = MkT2 (f (f a)) deriving( Eq )
2851 The natural generated <literal>Eq</literal> code would result in these instance declarations:
2853 instance Eq a => Eq (T0 f a) where ...
2854 instance Eq (f a) => Eq (T1 f a) where ...
2855 instance Eq (f (f a)) => Eq (T2 f a) where ...
2857 The first of these is obviously fine. The second is still fine, although less obviously.
2858 The third is not Haskell 98, and risks losing termination of instances.
2861 GHC takes a conservative position: it accepts the first two, but not the third. The rule is this:
2862 each constraint in the inferred instance context must consist only of type variables,
2863 with no repetitions.
2866 This rule is applied regardless of flags. If you want a more exotic context, you can write
2867 it yourself, using the <link linkend="stand-alone-deriving">standalone deriving mechanism</link>.
2871 <sect2 id="stand-alone-deriving">
2872 <title>Stand-alone deriving declarations</title>
2875 GHC now allows stand-alone <literal>deriving</literal> declarations, enabled by <literal>-XStandaloneDeriving</literal>:
2877 data Foo a = Bar a | Baz String
2879 deriving instance Eq a => Eq (Foo a)
2881 The syntax is identical to that of an ordinary instance declaration apart from (a) the keyword
2882 <literal>deriving</literal>, and (b) the absence of the <literal>where</literal> part.
2883 Note the following points:
2886 You must supply an explicit context (in the example the context is <literal>(Eq a)</literal>),
2887 exactly as you would in an ordinary instance declaration.
2888 (In contrast, in a <literal>deriving</literal> clause
2889 attached to a data type declaration, the context is inferred.)
2893 A <literal>deriving instance</literal> declaration
2894 must obey the same rules concerning form and termination as ordinary instance declarations,
2895 controlled by the same flags; see <xref linkend="instance-decls"/>.
2899 Unlike a <literal>deriving</literal>
2900 declaration attached to a <literal>data</literal> declaration, the instance can be more specific
2901 than the data type (assuming you also use
2902 <literal>-XFlexibleInstances</literal>, <xref linkend="instance-rules"/>). Consider
2905 data Foo a = Bar a | Baz String
2907 deriving instance Eq a => Eq (Foo [a])
2908 deriving instance Eq a => Eq (Foo (Maybe a))
2910 This will generate a derived instance for <literal>(Foo [a])</literal> and <literal>(Foo (Maybe a))</literal>,
2911 but other types such as <literal>(Foo (Int,Bool))</literal> will not be an instance of <literal>Eq</literal>.
2915 Unlike a <literal>deriving</literal>
2916 declaration attached to a <literal>data</literal> declaration,
2917 GHC does not restrict the form of the data type. Instead, GHC simply generates the appropriate
2918 boilerplate code for the specified class, and typechecks it. If there is a type error, it is
2919 your problem. (GHC will show you the offending code if it has a type error.)
2920 The merit of this is that you can derive instances for GADTs and other exotic
2921 data types, providing only that the boilerplate code does indeed typecheck. For example:
2927 deriving instance Show (T a)
2929 In this example, you cannot say <literal>... deriving( Show )</literal> on the
2930 data type declaration for <literal>T</literal>,
2931 because <literal>T</literal> is a GADT, but you <emphasis>can</emphasis> generate
2932 the instance declaration using stand-alone deriving.
2937 <para>The stand-alone syntax is generalised for newtypes in exactly the same
2938 way that ordinary <literal>deriving</literal> clauses are generalised (<xref linkend="newtype-deriving"/>).
2941 newtype Foo a = MkFoo (State Int a)
2943 deriving instance MonadState Int Foo
2945 GHC always treats the <emphasis>last</emphasis> parameter of the instance
2946 (<literal>Foo</literal> in this example) as the type whose instance is being derived.
2948 </itemizedlist></para>
2953 <sect2 id="deriving-typeable">
2954 <title>Deriving clause for extra classes (<literal>Typeable</literal>, <literal>Data</literal>, etc)</title>
2957 Haskell 98 allows the programmer to add "<literal>deriving( Eq, Ord )</literal>" to a data type
2958 declaration, to generate a standard instance declaration for classes specified in the <literal>deriving</literal> clause.
2959 In Haskell 98, the only classes that may appear in the <literal>deriving</literal> clause are the standard
2960 classes <literal>Eq</literal>, <literal>Ord</literal>,
2961 <literal>Enum</literal>, <literal>Ix</literal>, <literal>Bounded</literal>, <literal>Read</literal>, and <literal>Show</literal>.
2964 GHC extends this list with several more classes that may be automatically derived:
2966 <listitem><para> With <option>-XDeriveDataTypeable</option>, you can derive instances of the classes
2967 <literal>Typeable</literal>, and <literal>Data</literal>, defined in the library
2968 modules <literal>Data.Typeable</literal> and <literal>Data.Generics</literal> respectively.
2970 <para>An instance of <literal>Typeable</literal> can only be derived if the
2971 data type has seven or fewer type parameters, all of kind <literal>*</literal>.
2972 The reason for this is that the <literal>Typeable</literal> class is derived using the scheme
2974 <ulink url="http://research.microsoft.com/%7Esimonpj/papers/hmap/gmap2.ps">
2975 Scrap More Boilerplate: Reflection, Zips, and Generalised Casts
2977 (Section 7.4 of the paper describes the multiple <literal>Typeable</literal> classes that
2978 are used, and only <literal>Typeable1</literal> up to
2979 <literal>Typeable7</literal> are provided in the library.)
2980 In other cases, there is nothing to stop the programmer writing a <literal>TypableX</literal>
2981 class, whose kind suits that of the data type constructor, and
2982 then writing the data type instance by hand.
2986 <listitem><para> With <option>-XDeriveFunctor</option>, you can derive instances of
2987 the class <literal>Functor</literal>,
2988 defined in <literal>GHC.Base</literal>.
2991 <listitem><para> With <option>-XDeriveFoldable</option>, you can derive instances of
2992 the class <literal>Foldable</literal>,
2993 defined in <literal>Data.Foldable</literal>.
2996 <listitem><para> With <option>-XDeriveTraversable</option>, you can derive instances of
2997 the class <literal>Traversable</literal>,
2998 defined in <literal>Data.Traversable</literal>.
3001 In each case the appropriate class must be in scope before it
3002 can be mentioned in the <literal>deriving</literal> clause.
3006 <sect2 id="newtype-deriving">
3007 <title>Generalised derived instances for newtypes</title>
3010 When you define an abstract type using <literal>newtype</literal>, you may want
3011 the new type to inherit some instances from its representation. In
3012 Haskell 98, you can inherit instances of <literal>Eq</literal>, <literal>Ord</literal>,
3013 <literal>Enum</literal> and <literal>Bounded</literal> by deriving them, but for any
3014 other classes you have to write an explicit instance declaration. For
3015 example, if you define
3018 newtype Dollars = Dollars Int
3021 and you want to use arithmetic on <literal>Dollars</literal>, you have to
3022 explicitly define an instance of <literal>Num</literal>:
3025 instance Num Dollars where
3026 Dollars a + Dollars b = Dollars (a+b)
3029 All the instance does is apply and remove the <literal>newtype</literal>
3030 constructor. It is particularly galling that, since the constructor
3031 doesn't appear at run-time, this instance declaration defines a
3032 dictionary which is <emphasis>wholly equivalent</emphasis> to the <literal>Int</literal>
3033 dictionary, only slower!
3037 <sect3> <title> Generalising the deriving clause </title>
3039 GHC now permits such instances to be derived instead,
3040 using the flag <option>-XGeneralizedNewtypeDeriving</option>,
3043 newtype Dollars = Dollars Int deriving (Eq,Show,Num)
3046 and the implementation uses the <emphasis>same</emphasis> <literal>Num</literal> dictionary
3047 for <literal>Dollars</literal> as for <literal>Int</literal>. Notionally, the compiler
3048 derives an instance declaration of the form
3051 instance Num Int => Num Dollars
3054 which just adds or removes the <literal>newtype</literal> constructor according to the type.
3058 We can also derive instances of constructor classes in a similar
3059 way. For example, suppose we have implemented state and failure monad
3060 transformers, such that
3063 instance Monad m => Monad (State s m)
3064 instance Monad m => Monad (Failure m)
3066 In Haskell 98, we can define a parsing monad by
3068 type Parser tok m a = State [tok] (Failure m) a
3071 which is automatically a monad thanks to the instance declarations
3072 above. With the extension, we can make the parser type abstract,
3073 without needing to write an instance of class <literal>Monad</literal>, via
3076 newtype Parser tok m a = Parser (State [tok] (Failure m) a)
3079 In this case the derived instance declaration is of the form
3081 instance Monad (State [tok] (Failure m)) => Monad (Parser tok m)
3084 Notice that, since <literal>Monad</literal> is a constructor class, the
3085 instance is a <emphasis>partial application</emphasis> of the new type, not the
3086 entire left hand side. We can imagine that the type declaration is
3087 "eta-converted" to generate the context of the instance
3092 We can even derive instances of multi-parameter classes, provided the
3093 newtype is the last class parameter. In this case, a ``partial
3094 application'' of the class appears in the <literal>deriving</literal>
3095 clause. For example, given the class
3098 class StateMonad s m | m -> s where ...
3099 instance Monad m => StateMonad s (State s m) where ...
3101 then we can derive an instance of <literal>StateMonad</literal> for <literal>Parser</literal>s by
3103 newtype Parser tok m a = Parser (State [tok] (Failure m) a)
3104 deriving (Monad, StateMonad [tok])
3107 The derived instance is obtained by completing the application of the
3108 class to the new type:
3111 instance StateMonad [tok] (State [tok] (Failure m)) =>
3112 StateMonad [tok] (Parser tok m)
3117 As a result of this extension, all derived instances in newtype
3118 declarations are treated uniformly (and implemented just by reusing
3119 the dictionary for the representation type), <emphasis>except</emphasis>
3120 <literal>Show</literal> and <literal>Read</literal>, which really behave differently for
3121 the newtype and its representation.
3125 <sect3> <title> A more precise specification </title>
3127 Derived instance declarations are constructed as follows. Consider the
3128 declaration (after expansion of any type synonyms)
3131 newtype T v1...vn = T' (t vk+1...vn) deriving (c1...cm)
3137 The <literal>ci</literal> are partial applications of
3138 classes of the form <literal>C t1'...tj'</literal>, where the arity of <literal>C</literal>
3139 is exactly <literal>j+1</literal>. That is, <literal>C</literal> lacks exactly one type argument.
3142 The <literal>k</literal> is chosen so that <literal>ci (T v1...vk)</literal> is well-kinded.
3145 The type <literal>t</literal> is an arbitrary type.
3148 The type variables <literal>vk+1...vn</literal> do not occur in <literal>t</literal>,
3149 nor in the <literal>ci</literal>, and
3152 None of the <literal>ci</literal> is <literal>Read</literal>, <literal>Show</literal>,
3153 <literal>Typeable</literal>, or <literal>Data</literal>. These classes
3154 should not "look through" the type or its constructor. You can still
3155 derive these classes for a newtype, but it happens in the usual way, not
3156 via this new mechanism.
3159 Then, for each <literal>ci</literal>, the derived instance
3162 instance ci t => ci (T v1...vk)
3164 As an example which does <emphasis>not</emphasis> work, consider
3166 newtype NonMonad m s = NonMonad (State s m s) deriving Monad
3168 Here we cannot derive the instance
3170 instance Monad (State s m) => Monad (NonMonad m)
3173 because the type variable <literal>s</literal> occurs in <literal>State s m</literal>,
3174 and so cannot be "eta-converted" away. It is a good thing that this
3175 <literal>deriving</literal> clause is rejected, because <literal>NonMonad m</literal> is
3176 not, in fact, a monad --- for the same reason. Try defining
3177 <literal>>>=</literal> with the correct type: you won't be able to.
3181 Notice also that the <emphasis>order</emphasis> of class parameters becomes
3182 important, since we can only derive instances for the last one. If the
3183 <literal>StateMonad</literal> class above were instead defined as
3186 class StateMonad m s | m -> s where ...
3189 then we would not have been able to derive an instance for the
3190 <literal>Parser</literal> type above. We hypothesise that multi-parameter
3191 classes usually have one "main" parameter for which deriving new
3192 instances is most interesting.
3194 <para>Lastly, all of this applies only for classes other than
3195 <literal>Read</literal>, <literal>Show</literal>, <literal>Typeable</literal>,
3196 and <literal>Data</literal>, for which the built-in derivation applies (section
3197 4.3.3. of the Haskell Report).
3198 (For the standard classes <literal>Eq</literal>, <literal>Ord</literal>,
3199 <literal>Ix</literal>, and <literal>Bounded</literal> it is immaterial whether
3200 the standard method is used or the one described here.)
3207 <!-- TYPE SYSTEM EXTENSIONS -->
3208 <sect1 id="type-class-extensions">
3209 <title>Class and instances declarations</title>
3211 <sect2 id="multi-param-type-classes">
3212 <title>Class declarations</title>
3215 This section, and the next one, documents GHC's type-class extensions.
3216 There's lots of background in the paper <ulink
3217 url="http://research.microsoft.com/~simonpj/Papers/type-class-design-space/">Type
3218 classes: exploring the design space</ulink> (Simon Peyton Jones, Mark
3219 Jones, Erik Meijer).
3222 All the extensions are enabled by the <option>-fglasgow-exts</option> flag.
3226 <title>Multi-parameter type classes</title>
3228 Multi-parameter type classes are permitted, with flag <option>-XMultiParamTypeClasses</option>.
3233 class Collection c a where
3234 union :: c a -> c a -> c a
3241 <sect3 id="superclass-rules">
3242 <title>The superclasses of a class declaration</title>
3245 In Haskell 98 the context of a class declaration (which introduces superclasses)
3246 must be simple; that is, each predicate must consist of a class applied to
3247 type variables. The flag <option>-XFlexibleContexts</option>
3248 (<xref linkend="flexible-contexts"/>)
3249 lifts this restriction,
3250 so that the only restriction on the context in a class declaration is
3251 that the class hierarchy must be acyclic. So these class declarations are OK:
3255 class Functor (m k) => FiniteMap m k where
3258 class (Monad m, Monad (t m)) => Transform t m where
3259 lift :: m a -> (t m) a
3265 As in Haskell 98, The class hierarchy must be acyclic. However, the definition
3266 of "acyclic" involves only the superclass relationships. For example,
3272 op :: D b => a -> b -> b
3275 class C a => D a where { ... }
3279 Here, <literal>C</literal> is a superclass of <literal>D</literal>, but it's OK for a
3280 class operation <literal>op</literal> of <literal>C</literal> to mention <literal>D</literal>. (It
3281 would not be OK for <literal>D</literal> to be a superclass of <literal>C</literal>.)
3288 <sect3 id="class-method-types">
3289 <title>Class method types</title>
3292 Haskell 98 prohibits class method types to mention constraints on the
3293 class type variable, thus:
3296 fromList :: [a] -> s a
3297 elem :: Eq a => a -> s a -> Bool
3299 The type of <literal>elem</literal> is illegal in Haskell 98, because it
3300 contains the constraint <literal>Eq a</literal>, constrains only the
3301 class type variable (in this case <literal>a</literal>).
3302 GHC lifts this restriction (flag <option>-XConstrainedClassMethods</option>).
3309 <sect2 id="functional-dependencies">
3310 <title>Functional dependencies
3313 <para> Functional dependencies are implemented as described by Mark Jones
3314 in “<ulink url="http://citeseer.ist.psu.edu/jones00type.html">Type Classes with Functional Dependencies</ulink>”, Mark P. Jones,
3315 In Proceedings of the 9th European Symposium on Programming,
3316 ESOP 2000, Berlin, Germany, March 2000, Springer-Verlag LNCS 1782,
3320 Functional dependencies are introduced by a vertical bar in the syntax of a
3321 class declaration; e.g.
3323 class (Monad m) => MonadState s m | m -> s where ...
3325 class Foo a b c | a b -> c where ...
3327 There should be more documentation, but there isn't (yet). Yell if you need it.
3330 <sect3><title>Rules for functional dependencies </title>
3332 In a class declaration, all of the class type variables must be reachable (in the sense
3333 mentioned in <xref linkend="flexible-contexts"/>)
3334 from the free variables of each method type.
3338 class Coll s a where
3340 insert :: s -> a -> s
3343 is not OK, because the type of <literal>empty</literal> doesn't mention
3344 <literal>a</literal>. Functional dependencies can make the type variable
3347 class Coll s a | s -> a where
3349 insert :: s -> a -> s
3352 Alternatively <literal>Coll</literal> might be rewritten
3355 class Coll s a where
3357 insert :: s a -> a -> s a
3361 which makes the connection between the type of a collection of
3362 <literal>a</literal>'s (namely <literal>(s a)</literal>) and the element type <literal>a</literal>.
3363 Occasionally this really doesn't work, in which case you can split the
3371 class CollE s => Coll s a where
3372 insert :: s -> a -> s
3379 <title>Background on functional dependencies</title>
3381 <para>The following description of the motivation and use of functional dependencies is taken
3382 from the Hugs user manual, reproduced here (with minor changes) by kind
3383 permission of Mark Jones.
3386 Consider the following class, intended as part of a
3387 library for collection types:
3389 class Collects e ce where
3391 insert :: e -> ce -> ce
3392 member :: e -> ce -> Bool
3394 The type variable e used here represents the element type, while ce is the type
3395 of the container itself. Within this framework, we might want to define
3396 instances of this class for lists or characteristic functions (both of which
3397 can be used to represent collections of any equality type), bit sets (which can
3398 be used to represent collections of characters), or hash tables (which can be
3399 used to represent any collection whose elements have a hash function). Omitting
3400 standard implementation details, this would lead to the following declarations:
3402 instance Eq e => Collects e [e] where ...
3403 instance Eq e => Collects e (e -> Bool) where ...
3404 instance Collects Char BitSet where ...
3405 instance (Hashable e, Collects a ce)
3406 => Collects e (Array Int ce) where ...
3408 All this looks quite promising; we have a class and a range of interesting
3409 implementations. Unfortunately, there are some serious problems with the class
3410 declaration. First, the empty function has an ambiguous type:
3412 empty :: Collects e ce => ce
3414 By "ambiguous" we mean that there is a type variable e that appears on the left
3415 of the <literal>=></literal> symbol, but not on the right. The problem with
3416 this is that, according to the theoretical foundations of Haskell overloading,
3417 we cannot guarantee a well-defined semantics for any term with an ambiguous
3421 We can sidestep this specific problem by removing the empty member from the
3422 class declaration. However, although the remaining members, insert and member,
3423 do not have ambiguous types, we still run into problems when we try to use
3424 them. For example, consider the following two functions:
3426 f x y = insert x . insert y
3429 for which GHC infers the following types:
3431 f :: (Collects a c, Collects b c) => a -> b -> c -> c
3432 g :: (Collects Bool c, Collects Char c) => c -> c
3434 Notice that the type for f allows the two parameters x and y to be assigned
3435 different types, even though it attempts to insert each of the two values, one
3436 after the other, into the same collection. If we're trying to model collections
3437 that contain only one type of value, then this is clearly an inaccurate
3438 type. Worse still, the definition for g is accepted, without causing a type
3439 error. As a result, the error in this code will not be flagged at the point
3440 where it appears. Instead, it will show up only when we try to use g, which
3441 might even be in a different module.
3444 <sect4><title>An attempt to use constructor classes</title>
3447 Faced with the problems described above, some Haskell programmers might be
3448 tempted to use something like the following version of the class declaration:
3450 class Collects e c where
3452 insert :: e -> c e -> c e
3453 member :: e -> c e -> Bool
3455 The key difference here is that we abstract over the type constructor c that is
3456 used to form the collection type c e, and not over that collection type itself,
3457 represented by ce in the original class declaration. This avoids the immediate
3458 problems that we mentioned above: empty has type <literal>Collects e c => c
3459 e</literal>, which is not ambiguous.
3462 The function f from the previous section has a more accurate type:
3464 f :: (Collects e c) => e -> e -> c e -> c e
3466 The function g from the previous section is now rejected with a type error as
3467 we would hope because the type of f does not allow the two arguments to have
3469 This, then, is an example of a multiple parameter class that does actually work
3470 quite well in practice, without ambiguity problems.
3471 There is, however, a catch. This version of the Collects class is nowhere near
3472 as general as the original class seemed to be: only one of the four instances
3473 for <literal>Collects</literal>
3474 given above can be used with this version of Collects because only one of
3475 them---the instance for lists---has a collection type that can be written in
3476 the form c e, for some type constructor c, and element type e.
3480 <sect4><title>Adding functional dependencies</title>
3483 To get a more useful version of the Collects class, Hugs provides a mechanism
3484 that allows programmers to specify dependencies between the parameters of a
3485 multiple parameter class (For readers with an interest in theoretical
3486 foundations and previous work: The use of dependency information can be seen
3487 both as a generalization of the proposal for `parametric type classes' that was
3488 put forward by Chen, Hudak, and Odersky, or as a special case of Mark Jones's
3489 later framework for "improvement" of qualified types. The
3490 underlying ideas are also discussed in a more theoretical and abstract setting
3491 in a manuscript [implparam], where they are identified as one point in a
3492 general design space for systems of implicit parameterization.).
3494 To start with an abstract example, consider a declaration such as:
3496 class C a b where ...
3498 which tells us simply that C can be thought of as a binary relation on types
3499 (or type constructors, depending on the kinds of a and b). Extra clauses can be
3500 included in the definition of classes to add information about dependencies
3501 between parameters, as in the following examples:
3503 class D a b | a -> b where ...
3504 class E a b | a -> b, b -> a where ...
3506 The notation <literal>a -> b</literal> used here between the | and where
3507 symbols --- not to be
3508 confused with a function type --- indicates that the a parameter uniquely
3509 determines the b parameter, and might be read as "a determines b." Thus D is
3510 not just a relation, but actually a (partial) function. Similarly, from the two
3511 dependencies that are included in the definition of E, we can see that E
3512 represents a (partial) one-one mapping between types.
3515 More generally, dependencies take the form <literal>x1 ... xn -> y1 ... ym</literal>,
3516 where x1, ..., xn, and y1, ..., yn are type variables with n>0 and
3517 m>=0, meaning that the y parameters are uniquely determined by the x
3518 parameters. Spaces can be used as separators if more than one variable appears
3519 on any single side of a dependency, as in <literal>t -> a b</literal>. Note that a class may be
3520 annotated with multiple dependencies using commas as separators, as in the
3521 definition of E above. Some dependencies that we can write in this notation are
3522 redundant, and will be rejected because they don't serve any useful
3523 purpose, and may instead indicate an error in the program. Examples of
3524 dependencies like this include <literal>a -> a </literal>,
3525 <literal>a -> a a </literal>,
3526 <literal>a -> </literal>, etc. There can also be
3527 some redundancy if multiple dependencies are given, as in
3528 <literal>a->b</literal>,
3529 <literal>b->c </literal>, <literal>a->c </literal>, and
3530 in which some subset implies the remaining dependencies. Examples like this are
3531 not treated as errors. Note that dependencies appear only in class
3532 declarations, and not in any other part of the language. In particular, the
3533 syntax for instance declarations, class constraints, and types is completely
3537 By including dependencies in a class declaration, we provide a mechanism for
3538 the programmer to specify each multiple parameter class more precisely. The
3539 compiler, on the other hand, is responsible for ensuring that the set of
3540 instances that are in scope at any given point in the program is consistent
3541 with any declared dependencies. For example, the following pair of instance
3542 declarations cannot appear together in the same scope because they violate the
3543 dependency for D, even though either one on its own would be acceptable:
3545 instance D Bool Int where ...
3546 instance D Bool Char where ...
3548 Note also that the following declaration is not allowed, even by itself:
3550 instance D [a] b where ...
3552 The problem here is that this instance would allow one particular choice of [a]
3553 to be associated with more than one choice for b, which contradicts the
3554 dependency specified in the definition of D. More generally, this means that,
3555 in any instance of the form:
3557 instance D t s where ...
3559 for some particular types t and s, the only variables that can appear in s are
3560 the ones that appear in t, and hence, if the type t is known, then s will be
3561 uniquely determined.
3564 The benefit of including dependency information is that it allows us to define
3565 more general multiple parameter classes, without ambiguity problems, and with
3566 the benefit of more accurate types. To illustrate this, we return to the
3567 collection class example, and annotate the original definition of <literal>Collects</literal>
3568 with a simple dependency:
3570 class Collects e ce | ce -> e where
3572 insert :: e -> ce -> ce
3573 member :: e -> ce -> Bool
3575 The dependency <literal>ce -> e</literal> here specifies that the type e of elements is uniquely
3576 determined by the type of the collection ce. Note that both parameters of
3577 Collects are of kind *; there are no constructor classes here. Note too that
3578 all of the instances of Collects that we gave earlier can be used
3579 together with this new definition.
3582 What about the ambiguity problems that we encountered with the original
3583 definition? The empty function still has type Collects e ce => ce, but it is no
3584 longer necessary to regard that as an ambiguous type: Although the variable e
3585 does not appear on the right of the => symbol, the dependency for class
3586 Collects tells us that it is uniquely determined by ce, which does appear on
3587 the right of the => symbol. Hence the context in which empty is used can still
3588 give enough information to determine types for both ce and e, without
3589 ambiguity. More generally, we need only regard a type as ambiguous if it
3590 contains a variable on the left of the => that is not uniquely determined
3591 (either directly or indirectly) by the variables on the right.
3594 Dependencies also help to produce more accurate types for user defined
3595 functions, and hence to provide earlier detection of errors, and less cluttered
3596 types for programmers to work with. Recall the previous definition for a
3599 f x y = insert x y = insert x . insert y
3601 for which we originally obtained a type:
3603 f :: (Collects a c, Collects b c) => a -> b -> c -> c
3605 Given the dependency information that we have for Collects, however, we can
3606 deduce that a and b must be equal because they both appear as the second
3607 parameter in a Collects constraint with the same first parameter c. Hence we
3608 can infer a shorter and more accurate type for f:
3610 f :: (Collects a c) => a -> a -> c -> c
3612 In a similar way, the earlier definition of g will now be flagged as a type error.
3615 Although we have given only a few examples here, it should be clear that the
3616 addition of dependency information can help to make multiple parameter classes
3617 more useful in practice, avoiding ambiguity problems, and allowing more general
3618 sets of instance declarations.
3624 <sect2 id="instance-decls">
3625 <title>Instance declarations</title>
3627 <para>An instance declaration has the form
3629 instance ( <replaceable>assertion</replaceable><subscript>1</subscript>, ..., <replaceable>assertion</replaceable><subscript>n</subscript>) => <replaceable>class</replaceable> <replaceable>type</replaceable><subscript>1</subscript> ... <replaceable>type</replaceable><subscript>m</subscript> where ...
3631 The part before the "<literal>=></literal>" is the
3632 <emphasis>context</emphasis>, while the part after the
3633 "<literal>=></literal>" is the <emphasis>head</emphasis> of the instance declaration.
3636 <sect3 id="flexible-instance-head">
3637 <title>Relaxed rules for the instance head</title>
3640 In Haskell 98 the head of an instance declaration
3641 must be of the form <literal>C (T a1 ... an)</literal>, where
3642 <literal>C</literal> is the class, <literal>T</literal> is a data type constructor,
3643 and the <literal>a1 ... an</literal> are distinct type variables.
3644 GHC relaxes these rules in two ways.
3648 The <option>-XFlexibleInstances</option> flag allows the head of the instance
3649 declaration to mention arbitrary nested types.
3650 For example, this becomes a legal instance declaration
3652 instance C (Maybe Int) where ...
3654 See also the <link linkend="instance-overlap">rules on overlap</link>.
3657 With the <option>-XTypeSynonymInstances</option> flag, instance heads may use type
3658 synonyms. As always, using a type synonym is just shorthand for
3659 writing the RHS of the type synonym definition. For example:
3663 type Point = (Int,Int)
3664 instance C Point where ...
3665 instance C [Point] where ...
3669 is legal. However, if you added
3673 instance C (Int,Int) where ...
3677 as well, then the compiler will complain about the overlapping
3678 (actually, identical) instance declarations. As always, type synonyms
3679 must be fully applied. You cannot, for example, write:
3683 instance Monad P where ...
3691 <sect3 id="instance-rules">
3692 <title>Relaxed rules for instance contexts</title>
3694 <para>In Haskell 98, the assertions in the context of the instance declaration
3695 must be of the form <literal>C a</literal> where <literal>a</literal>
3696 is a type variable that occurs in the head.
3700 The <option>-XFlexibleContexts</option> flag relaxes this rule, as well
3701 as the corresponding rule for type signatures (see <xref linkend="flexible-contexts"/>).
3702 With this flag the context of the instance declaration can each consist of arbitrary
3703 (well-kinded) assertions <literal>(C t1 ... tn)</literal> subject only to the
3707 The Paterson Conditions: for each assertion in the context
3709 <listitem><para>No type variable has more occurrences in the assertion than in the head</para></listitem>
3710 <listitem><para>The assertion has fewer constructors and variables (taken together
3711 and counting repetitions) than the head</para></listitem>
3715 <listitem><para>The Coverage Condition. For each functional dependency,
3716 <replaceable>tvs</replaceable><subscript>left</subscript> <literal>-></literal>
3717 <replaceable>tvs</replaceable><subscript>right</subscript>, of the class,
3718 every type variable in
3719 S(<replaceable>tvs</replaceable><subscript>right</subscript>) must appear in
3720 S(<replaceable>tvs</replaceable><subscript>left</subscript>), where S is the
3721 substitution mapping each type variable in the class declaration to the
3722 corresponding type in the instance declaration.
3725 These restrictions ensure that context reduction terminates: each reduction
3726 step makes the problem smaller by at least one
3727 constructor. Both the Paterson Conditions and the Coverage Condition are lifted
3728 if you give the <option>-XUndecidableInstances</option>
3729 flag (<xref linkend="undecidable-instances"/>).
3730 You can find lots of background material about the reason for these
3731 restrictions in the paper <ulink
3732 url="http://research.microsoft.com/%7Esimonpj/papers/fd%2Dchr/">
3733 Understanding functional dependencies via Constraint Handling Rules</ulink>.
3736 For example, these are OK:
3738 instance C Int [a] -- Multiple parameters
3739 instance Eq (S [a]) -- Structured type in head
3741 -- Repeated type variable in head
3742 instance C4 a a => C4 [a] [a]
3743 instance Stateful (ST s) (MutVar s)
3745 -- Head can consist of type variables only
3747 instance (Eq a, Show b) => C2 a b
3749 -- Non-type variables in context
3750 instance Show (s a) => Show (Sized s a)
3751 instance C2 Int a => C3 Bool [a]
3752 instance C2 Int a => C3 [a] b
3756 -- Context assertion no smaller than head
3757 instance C a => C a where ...
3758 -- (C b b) has more more occurrences of b than the head
3759 instance C b b => Foo [b] where ...
3764 The same restrictions apply to instances generated by
3765 <literal>deriving</literal> clauses. Thus the following is accepted:
3767 data MinHeap h a = H a (h a)
3770 because the derived instance
3772 instance (Show a, Show (h a)) => Show (MinHeap h a)
3774 conforms to the above rules.
3778 A useful idiom permitted by the above rules is as follows.
3779 If one allows overlapping instance declarations then it's quite
3780 convenient to have a "default instance" declaration that applies if
3781 something more specific does not:
3789 <sect3 id="undecidable-instances">
3790 <title>Undecidable instances</title>
3793 Sometimes even the rules of <xref linkend="instance-rules"/> are too onerous.
3794 For example, sometimes you might want to use the following to get the
3795 effect of a "class synonym":
3797 class (C1 a, C2 a, C3 a) => C a where { }
3799 instance (C1 a, C2 a, C3 a) => C a where { }
3801 This allows you to write shorter signatures:
3807 f :: (C1 a, C2 a, C3 a) => ...
3809 The restrictions on functional dependencies (<xref
3810 linkend="functional-dependencies"/>) are particularly troublesome.
3811 It is tempting to introduce type variables in the context that do not appear in
3812 the head, something that is excluded by the normal rules. For example:
3814 class HasConverter a b | a -> b where
3817 data Foo a = MkFoo a
3819 instance (HasConverter a b,Show b) => Show (Foo a) where
3820 show (MkFoo value) = show (convert value)
3822 This is dangerous territory, however. Here, for example, is a program that would make the
3827 instance F [a] [[a]]
3828 instance (D c, F a c) => D [a] -- 'c' is not mentioned in the head
3830 Similarly, it can be tempting to lift the coverage condition:
3832 class Mul a b c | a b -> c where
3833 (.*.) :: a -> b -> c
3835 instance Mul Int Int Int where (.*.) = (*)
3836 instance Mul Int Float Float where x .*. y = fromIntegral x * y
3837 instance Mul a b c => Mul a [b] [c] where x .*. v = map (x.*.) v
3839 The third instance declaration does not obey the coverage condition;
3840 and indeed the (somewhat strange) definition:
3842 f = \ b x y -> if b then x .*. [y] else y
3844 makes instance inference go into a loop, because it requires the constraint
3845 <literal>(Mul a [b] b)</literal>.
3848 Nevertheless, GHC allows you to experiment with more liberal rules. If you use
3849 the experimental flag <option>-XUndecidableInstances</option>
3850 <indexterm><primary>-XUndecidableInstances</primary></indexterm>,
3851 both the Paterson Conditions and the Coverage Condition
3852 (described in <xref linkend="instance-rules"/>) are lifted. Termination is ensured by having a
3853 fixed-depth recursion stack. If you exceed the stack depth you get a
3854 sort of backtrace, and the opportunity to increase the stack depth
3855 with <option>-fcontext-stack=</option><emphasis>N</emphasis>.
3861 <sect3 id="instance-overlap">
3862 <title>Overlapping instances</title>
3864 In general, <emphasis>GHC requires that that it be unambiguous which instance
3866 should be used to resolve a type-class constraint</emphasis>. This behaviour
3867 can be modified by two flags: <option>-XOverlappingInstances</option>
3868 <indexterm><primary>-XOverlappingInstances
3869 </primary></indexterm>
3870 and <option>-XIncoherentInstances</option>
3871 <indexterm><primary>-XIncoherentInstances
3872 </primary></indexterm>, as this section discusses. Both these
3873 flags are dynamic flags, and can be set on a per-module basis, using
3874 an <literal>OPTIONS_GHC</literal> pragma if desired (<xref linkend="source-file-options"/>).</para>
3876 When GHC tries to resolve, say, the constraint <literal>C Int Bool</literal>,
3877 it tries to match every instance declaration against the
3879 by instantiating the head of the instance declaration. For example, consider
3882 instance context1 => C Int a where ... -- (A)
3883 instance context2 => C a Bool where ... -- (B)
3884 instance context3 => C Int [a] where ... -- (C)
3885 instance context4 => C Int [Int] where ... -- (D)
3887 The instances (A) and (B) match the constraint <literal>C Int Bool</literal>,
3888 but (C) and (D) do not. When matching, GHC takes
3889 no account of the context of the instance declaration
3890 (<literal>context1</literal> etc).
3891 GHC's default behaviour is that <emphasis>exactly one instance must match the
3892 constraint it is trying to resolve</emphasis>.
3893 It is fine for there to be a <emphasis>potential</emphasis> of overlap (by
3894 including both declarations (A) and (B), say); an error is only reported if a
3895 particular constraint matches more than one.
3899 The <option>-XOverlappingInstances</option> flag instructs GHC to allow
3900 more than one instance to match, provided there is a most specific one. For
3901 example, the constraint <literal>C Int [Int]</literal> matches instances (A),
3902 (C) and (D), but the last is more specific, and hence is chosen. If there is no
3903 most-specific match, the program is rejected.
3906 However, GHC is conservative about committing to an overlapping instance. For example:
3911 Suppose that from the RHS of <literal>f</literal> we get the constraint
3912 <literal>C Int [b]</literal>. But
3913 GHC does not commit to instance (C), because in a particular
3914 call of <literal>f</literal>, <literal>b</literal> might be instantiate
3915 to <literal>Int</literal>, in which case instance (D) would be more specific still.
3916 So GHC rejects the program.
3917 (If you add the flag <option>-XIncoherentInstances</option>,
3918 GHC will instead pick (C), without complaining about
3919 the problem of subsequent instantiations.)
3922 Notice that we gave a type signature to <literal>f</literal>, so GHC had to
3923 <emphasis>check</emphasis> that <literal>f</literal> has the specified type.
3924 Suppose instead we do not give a type signature, asking GHC to <emphasis>infer</emphasis>
3925 it instead. In this case, GHC will refrain from
3926 simplifying the constraint <literal>C Int [b]</literal> (for the same reason
3927 as before) but, rather than rejecting the program, it will infer the type
3929 f :: C Int [b] => [b] -> [b]
3931 That postpones the question of which instance to pick to the
3932 call site for <literal>f</literal>
3933 by which time more is known about the type <literal>b</literal>.
3934 You can write this type signature yourself if you use the
3935 <link linkend="flexible-contexts"><option>-XFlexibleContexts</option></link>
3939 Exactly the same situation can arise in instance declarations themselves. Suppose we have
3943 instance Foo [b] where
3946 and, as before, the constraint <literal>C Int [b]</literal> arises from <literal>f</literal>'s
3947 right hand side. GHC will reject the instance, complaining as before that it does not know how to resolve
3948 the constraint <literal>C Int [b]</literal>, because it matches more than one instance
3949 declaration. The solution is to postpone the choice by adding the constraint to the context
3950 of the instance declaration, thus:
3952 instance C Int [b] => Foo [b] where
3955 (You need <link linkend="instance-rules"><option>-XFlexibleInstances</option></link> to do this.)
3958 Warning: overlapping instances must be used with care. They
3959 can give rise to incoherence (ie different instance choices are made
3960 in different parts of the program) even without <option>-XIncoherentInstances</option>. Consider:
3962 {-# LANGUAGE OverlappingInstances #-}
3965 class MyShow a where
3966 myshow :: a -> String
3968 instance MyShow a => MyShow [a] where
3969 myshow xs = concatMap myshow xs
3971 showHelp :: MyShow a => [a] -> String
3972 showHelp xs = myshow xs
3974 {-# LANGUAGE FlexibleInstances, OverlappingInstances #-}
3980 instance MyShow T where
3981 myshow x = "Used generic instance"
3983 instance MyShow [T] where
3984 myshow xs = "Used more specific instance"
3986 main = do { print (myshow [MkT]); print (showHelp [MkT]) }
3988 In function <literal>showHelp</literal> GHC sees no overlapping
3989 instances, and so uses the <literal>MyShow [a]</literal> instance
3990 without complaint. In the call to <literal>myshow</literal> in <literal>main</literal>,
3991 GHC resolves the <literal>MyShow [T]</literal> constraint using the overlapping
3992 instance declaration in module <literal>Main</literal>. As a result,
3995 "Used more specific instance"
3996 "Used generic instance"
3998 (An alternative possible behaviour, not currently implemented,
3999 would be to reject module <literal>Help</literal>
4000 on the grounds that a later instance declaration might overlap the local one.)
4003 The willingness to be overlapped or incoherent is a property of
4004 the <emphasis>instance declaration</emphasis> itself, controlled by the
4005 presence or otherwise of the <option>-XOverlappingInstances</option>
4006 and <option>-XIncoherentInstances</option> flags when that module is
4007 being defined. Neither flag is required in a module that imports and uses the
4008 instance declaration. Specifically, during the lookup process:
4011 An instance declaration is ignored during the lookup process if (a) a more specific
4012 match is found, and (b) the instance declaration was compiled with
4013 <option>-XOverlappingInstances</option>. The flag setting for the
4014 more-specific instance does not matter.
4017 Suppose an instance declaration does not match the constraint being looked up, but
4018 does unify with it, so that it might match when the constraint is further
4019 instantiated. Usually GHC will regard this as a reason for not committing to
4020 some other constraint. But if the instance declaration was compiled with
4021 <option>-XIncoherentInstances</option>, GHC will skip the "does-it-unify?"
4022 check for that declaration.
4025 These rules make it possible for a library author to design a library that relies on
4026 overlapping instances without the library client having to know.
4029 If an instance declaration is compiled without
4030 <option>-XOverlappingInstances</option>,
4031 then that instance can never be overlapped. This could perhaps be
4032 inconvenient. Perhaps the rule should instead say that the
4033 <emphasis>overlapping</emphasis> instance declaration should be compiled in
4034 this way, rather than the <emphasis>overlapped</emphasis> one. Perhaps overlap
4035 at a usage site should be permitted regardless of how the instance declarations
4036 are compiled, if the <option>-XOverlappingInstances</option> flag is
4037 used at the usage site. (Mind you, the exact usage site can occasionally be
4038 hard to pin down.) We are interested to receive feedback on these points.
4040 <para>The <option>-XIncoherentInstances</option> flag implies the
4041 <option>-XOverlappingInstances</option> flag, but not vice versa.
4049 <sect2 id="overloaded-strings">
4050 <title>Overloaded string literals
4054 GHC supports <emphasis>overloaded string literals</emphasis>. Normally a
4055 string literal has type <literal>String</literal>, but with overloaded string
4056 literals enabled (with <literal>-XOverloadedStrings</literal>)
4057 a string literal has type <literal>(IsString a) => a</literal>.
4060 This means that the usual string syntax can be used, e.g., for packed strings
4061 and other variations of string like types. String literals behave very much
4062 like integer literals, i.e., they can be used in both expressions and patterns.
4063 If used in a pattern the literal with be replaced by an equality test, in the same
4064 way as an integer literal is.
4067 The class <literal>IsString</literal> is defined as:
4069 class IsString a where
4070 fromString :: String -> a
4072 The only predefined instance is the obvious one to make strings work as usual:
4074 instance IsString [Char] where
4077 The class <literal>IsString</literal> is not in scope by default. If you want to mention
4078 it explicitly (for example, to give an instance declaration for it), you can import it
4079 from module <literal>GHC.Exts</literal>.
4082 Haskell's defaulting mechanism is extended to cover string literals, when <option>-XOverloadedStrings</option> is specified.
4086 Each type in a default declaration must be an
4087 instance of <literal>Num</literal> <emphasis>or</emphasis> of <literal>IsString</literal>.
4091 The standard defaulting rule (<ulink url="http://www.haskell.org/onlinereport/decls.html#sect4.3.4">Haskell Report, Section 4.3.4</ulink>)
4092 is extended thus: defaulting applies when all the unresolved constraints involve standard classes
4093 <emphasis>or</emphasis> <literal>IsString</literal>; and at least one is a numeric class
4094 <emphasis>or</emphasis> <literal>IsString</literal>.
4103 import GHC.Exts( IsString(..) )
4105 newtype MyString = MyString String deriving (Eq, Show)
4106 instance IsString MyString where
4107 fromString = MyString
4109 greet :: MyString -> MyString
4110 greet "hello" = "world"
4114 print $ greet "hello"
4115 print $ greet "fool"
4119 Note that deriving <literal>Eq</literal> is necessary for the pattern matching
4120 to work since it gets translated into an equality comparison.
4126 <sect1 id="type-families">
4127 <title>Type families</title>
4130 <firstterm>Indexed type families</firstterm> are a new GHC extension to
4131 facilitate type-level
4132 programming. Type families are a generalisation of <firstterm>associated
4133 data types</firstterm>
4134 (“<ulink url="http://www.cse.unsw.edu.au/~chak/papers/CKPM05.html">Associated
4135 Types with Class</ulink>”, M. Chakravarty, G. Keller, S. Peyton Jones,
4136 and S. Marlow. In Proceedings of “The 32nd Annual ACM SIGPLAN-SIGACT
4137 Symposium on Principles of Programming Languages (POPL'05)”, pages
4138 1-13, ACM Press, 2005) and <firstterm>associated type synonyms</firstterm>
4139 (“<ulink url="http://www.cse.unsw.edu.au/~chak/papers/CKP05.html">Type
4140 Associated Type Synonyms</ulink>”. M. Chakravarty, G. Keller, and
4142 In Proceedings of “The Tenth ACM SIGPLAN International Conference on
4143 Functional Programming”, ACM Press, pages 241-253, 2005). Type families
4144 themselves are described in the paper “<ulink
4145 url="http://www.cse.unsw.edu.au/~chak/papers/SPCS08.html">Type
4146 Checking with Open Type Functions</ulink>”, T. Schrijvers,
4148 M. Chakravarty, and M. Sulzmann, in Proceedings of “ICFP 2008: The
4149 13th ACM SIGPLAN International Conference on Functional
4150 Programming”, ACM Press, pages 51-62, 2008. Type families
4151 essentially provide type-indexed data types and named functions on types,
4152 which are useful for generic programming and highly parameterised library
4153 interfaces as well as interfaces with enhanced static information, much like
4154 dependent types. They might also be regarded as an alternative to functional
4155 dependencies, but provide a more functional style of type-level programming
4156 than the relational style of functional dependencies.
4159 Indexed type families, or type families for short, are type constructors that
4160 represent sets of types. Set members are denoted by supplying the type family
4161 constructor with type parameters, which are called <firstterm>type
4162 indices</firstterm>. The
4163 difference between vanilla parametrised type constructors and family
4164 constructors is much like between parametrically polymorphic functions and
4165 (ad-hoc polymorphic) methods of type classes. Parametric polymorphic functions
4166 behave the same at all type instances, whereas class methods can change their
4167 behaviour in dependence on the class type parameters. Similarly, vanilla type
4168 constructors imply the same data representation for all type instances, but
4169 family constructors can have varying representation types for varying type
4173 Indexed type families come in two flavours: <firstterm>data
4174 families</firstterm> and <firstterm>type synonym
4175 families</firstterm>. They are the indexed family variants of algebraic
4176 data types and type synonyms, respectively. The instances of data families
4177 can be data types and newtypes.
4180 Type families are enabled by the flag <option>-XTypeFamilies</option>.
4181 Additional information on the use of type families in GHC is available on
4182 <ulink url="http://www.haskell.org/haskellwiki/GHC/Indexed_types">the
4183 Haskell wiki page on type families</ulink>.
4186 <sect2 id="data-families">
4187 <title>Data families</title>
4190 Data families appear in two flavours: (1) they can be defined on the
4192 or (2) they can appear inside type classes (in which case they are known as
4193 associated types). The former is the more general variant, as it lacks the
4194 requirement for the type-indexes to coincide with the class
4195 parameters. However, the latter can lead to more clearly structured code and
4196 compiler warnings if some type instances were - possibly accidentally -
4197 omitted. In the following, we always discuss the general toplevel form first
4198 and then cover the additional constraints placed on associated types.
4201 <sect3 id="data-family-declarations">
4202 <title>Data family declarations</title>
4205 Indexed data families are introduced by a signature, such as
4207 data family GMap k :: * -> *
4209 The special <literal>family</literal> distinguishes family from standard
4210 data declarations. The result kind annotation is optional and, as
4211 usual, defaults to <literal>*</literal> if omitted. An example is
4215 Named arguments can also be given explicit kind signatures if needed.
4217 [http://www.haskell.org/ghc/docs/latest/html/users_guide/gadt.html GADT
4218 declarations] named arguments are entirely optional, so that we can
4219 declare <literal>Array</literal> alternatively with
4221 data family Array :: * -> *
4225 <sect4 id="assoc-data-family-decl">
4226 <title>Associated data family declarations</title>
4228 When a data family is declared as part of a type class, we drop
4229 the <literal>family</literal> special. The <literal>GMap</literal>
4230 declaration takes the following form
4232 class GMapKey k where
4233 data GMap k :: * -> *
4236 In contrast to toplevel declarations, named arguments must be used for
4237 all type parameters that are to be used as type-indexes. Moreover,
4238 the argument names must be class parameters. Each class parameter may
4239 only be used at most once per associated type, but some may be omitted
4240 and they may be in an order other than in the class head. Hence, the
4241 following contrived example is admissible:
4250 <sect3 id="data-instance-declarations">
4251 <title>Data instance declarations</title>
4254 Instance declarations of data and newtype families are very similar to
4255 standard data and newtype declarations. The only two differences are
4256 that the keyword <literal>data</literal> or <literal>newtype</literal>
4257 is followed by <literal>instance</literal> and that some or all of the
4258 type arguments can be non-variable types, but may not contain forall
4259 types or type synonym families. However, data families are generally
4260 allowed in type parameters, and type synonyms are allowed as long as
4261 they are fully applied and expand to a type that is itself admissible -
4262 exactly as this is required for occurrences of type synonyms in class
4263 instance parameters. For example, the <literal>Either</literal>
4264 instance for <literal>GMap</literal> is
4266 data instance GMap (Either a b) v = GMapEither (GMap a v) (GMap b v)
4268 In this example, the declaration has only one variant. In general, it
4272 Data and newtype instance declarations are only permitted when an
4273 appropriate family declaration is in scope - just as a class instance declaratoin
4274 requires the class declaration to be visible. Moreover, each instance
4275 declaration has to conform to the kind determined by its family
4276 declaration. This implies that the number of parameters of an instance
4277 declaration matches the arity determined by the kind of the family.
4280 A data family instance declaration can use the full exprssiveness of
4281 ordinary <literal>data</literal> or <literal>newtype</literal> declarations:
4283 <listitem><para> Although, a data family is <emphasis>introduced</emphasis> with
4284 the keyword "<literal>data</literal>", a data family <emphasis>instance</emphasis> can
4285 use either <literal>data</literal> or <literal>newtype</literal>. For example:
4288 data instance T Int = T1 Int | T2 Bool
4289 newtype instance T Char = TC Bool
4292 <listitem><para> A <literal>data instance</literal> can use GADT syntax for the data constructors,
4293 and indeed can define a GADT. For example:
4296 data instance G [a] b where
4297 G1 :: c -> G [Int] b
4301 <listitem><para> You can use a <literal>deriving</literal> clause on a
4302 <literal>data instance</literal> or <literal>newtype instance</literal>
4309 Even if type families are defined as toplevel declarations, functions
4310 that perform different computations for different family instances may still
4311 need to be defined as methods of type classes. In particular, the
4312 following is not possible:
4315 data instance T Int = A
4316 data instance T Char = B
4318 foo A = 1 -- WRONG: These two equations together...
4319 foo B = 2 -- ...will produce a type error.
4321 Instead, you would have to write <literal>foo</literal> as a class operation, thus:
4325 instance Foo Int where
4327 instance Foo Char where
4330 (Given the functionality provided by GADTs (Generalised Algebraic Data
4331 Types), it might seem as if a definition, such as the above, should be
4332 feasible. However, type families are - in contrast to GADTs - are
4333 <emphasis>open;</emphasis> i.e., new instances can always be added,
4335 modules. Supporting pattern matching across different data instances
4336 would require a form of extensible case construct.)
4339 <sect4 id="assoc-data-inst">
4340 <title>Associated data instances</title>
4342 When an associated data family instance is declared within a type
4343 class instance, we drop the <literal>instance</literal> keyword in the
4344 family instance. So, the <literal>Either</literal> instance
4345 for <literal>GMap</literal> becomes:
4347 instance (GMapKey a, GMapKey b) => GMapKey (Either a b) where
4348 data GMap (Either a b) v = GMapEither (GMap a v) (GMap b v)
4351 The most important point about associated family instances is that the
4352 type indexes corresponding to class parameters must be identical to
4353 the type given in the instance head; here this is the first argument
4354 of <literal>GMap</literal>, namely <literal>Either a b</literal>,
4355 which coincides with the only class parameter. Any parameters to the
4356 family constructor that do not correspond to class parameters, need to
4357 be variables in every instance; here this is the
4358 variable <literal>v</literal>.
4361 Instances for an associated family can only appear as part of
4362 instances declarations of the class in which the family was declared -
4363 just as with the equations of the methods of a class. Also in
4364 correspondence to how methods are handled, declarations of associated
4365 types can be omitted in class instances. If an associated family
4366 instance is omitted, the corresponding instance type is not inhabited;
4367 i.e., only diverging expressions, such
4368 as <literal>undefined</literal>, can assume the type.
4372 <sect4 id="scoping-class-params">
4373 <title>Scoping of class parameters</title>
4375 In the case of multi-parameter type classes, the visibility of class
4376 parameters in the right-hand side of associated family instances
4377 depends <emphasis>solely</emphasis> on the parameters of the data
4378 family. As an example, consider the simple class declaration
4383 Only one of the two class parameters is a parameter to the data
4384 family. Hence, the following instance declaration is invalid:
4386 instance C [c] d where
4387 data T [c] = MkT (c, d) -- WRONG!! 'd' is not in scope
4389 Here, the right-hand side of the data instance mentions the type
4390 variable <literal>d</literal> that does not occur in its left-hand
4391 side. We cannot admit such data instances as they would compromise
4396 <sect4 id="family-class-inst">
4397 <title>Type class instances of family instances</title>
4399 Type class instances of instances of data families can be defined as
4400 usual, and in particular data instance declarations can
4401 have <literal>deriving</literal> clauses. For example, we can write
4403 data GMap () v = GMapUnit (Maybe v)
4406 which implicitly defines an instance of the form
4408 instance Show v => Show (GMap () v) where ...
4412 Note that class instances are always for
4413 particular <emphasis>instances</emphasis> of a data family and never
4414 for an entire family as a whole. This is for essentially the same
4415 reasons that we cannot define a toplevel function that performs
4416 pattern matching on the data constructors
4417 of <emphasis>different</emphasis> instances of a single type family.
4418 It would require a form of extensible case construct.
4422 <sect4 id="data-family-overlap">
4423 <title>Overlap of data instances</title>
4425 The instance declarations of a data family used in a single program
4426 may not overlap at all, independent of whether they are associated or
4427 not. In contrast to type class instances, this is not only a matter
4428 of consistency, but one of type safety.
4434 <sect3 id="data-family-import-export">
4435 <title>Import and export</title>
4438 The association of data constructors with type families is more dynamic
4439 than that is the case with standard data and newtype declarations. In
4440 the standard case, the notation <literal>T(..)</literal> in an import or
4441 export list denotes the type constructor and all the data constructors
4442 introduced in its declaration. However, a family declaration never
4443 introduces any data constructors; instead, data constructors are
4444 introduced by family instances. As a result, which data constructors
4445 are associated with a type family depends on the currently visible
4446 instance declarations for that family. Consequently, an import or
4447 export item of the form <literal>T(..)</literal> denotes the family
4448 constructor and all currently visible data constructors - in the case of
4449 an export item, these may be either imported or defined in the current
4450 module. The treatment of import and export items that explicitly list
4451 data constructors, such as <literal>GMap(GMapEither)</literal>, is
4455 <sect4 id="data-family-impexp-assoc">
4456 <title>Associated families</title>
4458 As expected, an import or export item of the
4459 form <literal>C(..)</literal> denotes all of the class' methods and
4460 associated types. However, when associated types are explicitly
4461 listed as subitems of a class, we need some new syntax, as uppercase
4462 identifiers as subitems are usually data constructors, not type
4463 constructors. To clarify that we denote types here, each associated
4464 type name needs to be prefixed by the keyword <literal>type</literal>.
4465 So for example, when explicitly listing the components of
4466 the <literal>GMapKey</literal> class, we write <literal>GMapKey(type
4467 GMap, empty, lookup, insert)</literal>.
4471 <sect4 id="data-family-impexp-examples">
4472 <title>Examples</title>
4474 Assuming our running <literal>GMapKey</literal> class example, let us
4475 look at some export lists and their meaning:
4478 <para><literal>module GMap (GMapKey) where...</literal>: Exports
4479 just the class name.</para>
4482 <para><literal>module GMap (GMapKey(..)) where...</literal>:
4483 Exports the class, the associated type <literal>GMap</literal>
4485 functions <literal>empty</literal>, <literal>lookup</literal>,
4486 and <literal>insert</literal>. None of the data constructors is
4490 <para><literal>module GMap (GMapKey(..), GMap(..))
4491 where...</literal>: As before, but also exports all the data
4492 constructors <literal>GMapInt</literal>,
4493 <literal>GMapChar</literal>,
4494 <literal>GMapUnit</literal>, <literal>GMapPair</literal>,
4495 and <literal>GMapUnit</literal>.</para>
4498 <para><literal>module GMap (GMapKey(empty, lookup, insert),
4499 GMap(..)) where...</literal>: As before.</para>
4502 <para><literal>module GMap (GMapKey, empty, lookup, insert, GMap(..))
4503 where...</literal>: As before.</para>
4508 Finally, you can write <literal>GMapKey(type GMap)</literal> to denote
4509 both the class <literal>GMapKey</literal> as well as its associated
4510 type <literal>GMap</literal>. However, you cannot
4511 write <literal>GMapKey(type GMap(..))</literal> — i.e.,
4512 sub-component specifications cannot be nested. To
4513 specify <literal>GMap</literal>'s data constructors, you have to list
4518 <sect4 id="data-family-impexp-instances">
4519 <title>Instances</title>
4521 Family instances are implicitly exported, just like class instances.
4522 However, this applies only to the heads of instances, not to the data
4523 constructors an instance defines.
4531 <sect2 id="synonym-families">
4532 <title>Synonym families</title>
4535 Type families appear in two flavours: (1) they can be defined on the
4536 toplevel or (2) they can appear inside type classes (in which case they
4537 are known as associated type synonyms). The former is the more general
4538 variant, as it lacks the requirement for the type-indexes to coincide with
4539 the class parameters. However, the latter can lead to more clearly
4540 structured code and compiler warnings if some type instances were -
4541 possibly accidentally - omitted. In the following, we always discuss the
4542 general toplevel form first and then cover the additional constraints
4543 placed on associated types.
4546 <sect3 id="type-family-declarations">
4547 <title>Type family declarations</title>
4550 Indexed type families are introduced by a signature, such as
4552 type family Elem c :: *
4554 The special <literal>family</literal> distinguishes family from standard
4555 type declarations. The result kind annotation is optional and, as
4556 usual, defaults to <literal>*</literal> if omitted. An example is
4560 Parameters can also be given explicit kind signatures if needed. We
4561 call the number of parameters in a type family declaration, the family's
4562 arity, and all applications of a type family must be fully saturated
4563 w.r.t. to that arity. This requirement is unlike ordinary type synonyms
4564 and it implies that the kind of a type family is not sufficient to
4565 determine a family's arity, and hence in general, also insufficient to
4566 determine whether a type family application is well formed. As an
4567 example, consider the following declaration:
4569 type family F a b :: * -> * -- F's arity is 2,
4570 -- although its overall kind is * -> * -> * -> *
4572 Given this declaration the following are examples of well-formed and
4575 F Char [Int] -- OK! Kind: * -> *
4576 F Char [Int] Bool -- OK! Kind: *
4577 F IO Bool -- WRONG: kind mismatch in the first argument
4578 F Bool -- WRONG: unsaturated application
4582 <sect4 id="assoc-type-family-decl">
4583 <title>Associated type family declarations</title>
4585 When a type family is declared as part of a type class, we drop
4586 the <literal>family</literal> special. The <literal>Elem</literal>
4587 declaration takes the following form
4589 class Collects ce where
4593 The argument names of the type family must be class parameters. Each
4594 class parameter may only be used at most once per associated type, but
4595 some may be omitted and they may be in an order other than in the
4596 class head. Hence, the following contrived example is admissible:
4601 These rules are exactly as for associated data families.
4606 <sect3 id="type-instance-declarations">
4607 <title>Type instance declarations</title>
4609 Instance declarations of type families are very similar to standard type
4610 synonym declarations. The only two differences are that the
4611 keyword <literal>type</literal> is followed
4612 by <literal>instance</literal> and that some or all of the type
4613 arguments can be non-variable types, but may not contain forall types or
4614 type synonym families. However, data families are generally allowed, and
4615 type synonyms are allowed as long as they are fully applied and expand
4616 to a type that is admissible - these are the exact same requirements as
4617 for data instances. For example, the <literal>[e]</literal> instance
4618 for <literal>Elem</literal> is
4620 type instance Elem [e] = e
4624 Type family instance declarations are only legitimate when an
4625 appropriate family declaration is in scope - just like class instances
4626 require the class declaration to be visible. Moreover, each instance
4627 declaration has to conform to the kind determined by its family
4628 declaration, and the number of type parameters in an instance
4629 declaration must match the number of type parameters in the family
4630 declaration. Finally, the right-hand side of a type instance must be a
4631 monotype (i.e., it may not include foralls) and after the expansion of
4632 all saturated vanilla type synonyms, no synonyms, except family synonyms
4633 may remain. Here are some examples of admissible and illegal type
4636 type family F a :: *
4637 type instance F [Int] = Int -- OK!
4638 type instance F String = Char -- OK!
4639 type instance F (F a) = a -- WRONG: type parameter mentions a type family
4640 type instance F (forall a. (a, b)) = b -- WRONG: a forall type appears in a type parameter
4641 type instance F Float = forall a.a -- WRONG: right-hand side may not be a forall type
4643 type family G a b :: * -> *
4644 type instance G Int = (,) -- WRONG: must be two type parameters
4645 type instance G Int Char Float = Double -- WRONG: must be two type parameters
4649 <sect4 id="assoc-type-instance">
4650 <title>Associated type instance declarations</title>
4652 When an associated family instance is declared within a type class
4653 instance, we drop the <literal>instance</literal> keyword in the family
4654 instance. So, the <literal>[e]</literal> instance
4655 for <literal>Elem</literal> becomes:
4657 instance (Eq (Elem [e])) => Collects ([e]) where
4661 The most important point about associated family instances is that the
4662 type indexes corresponding to class parameters must be identical to the
4663 type given in the instance head; here this is <literal>[e]</literal>,
4664 which coincides with the only class parameter.
4667 Instances for an associated family can only appear as part of instances
4668 declarations of the class in which the family was declared - just as
4669 with the equations of the methods of a class. Also in correspondence to
4670 how methods are handled, declarations of associated types can be omitted
4671 in class instances. If an associated family instance is omitted, the
4672 corresponding instance type is not inhabited; i.e., only diverging
4673 expressions, such as <literal>undefined</literal>, can assume the type.
4677 <sect4 id="type-family-overlap">
4678 <title>Overlap of type synonym instances</title>
4680 The instance declarations of a type family used in a single program
4681 may only overlap if the right-hand sides of the overlapping instances
4682 coincide for the overlapping types. More formally, two instance
4683 declarations overlap if there is a substitution that makes the
4684 left-hand sides of the instances syntactically the same. Whenever
4685 that is the case, the right-hand sides of the instances must also be
4686 syntactically equal under the same substitution. This condition is
4687 independent of whether the type family is associated or not, and it is
4688 not only a matter of consistency, but one of type safety.
4691 Here are two example to illustrate the condition under which overlap
4694 type instance F (a, Int) = [a]
4695 type instance F (Int, b) = [b] -- overlap permitted
4697 type instance G (a, Int) = [a]
4698 type instance G (Char, a) = [a] -- ILLEGAL overlap, as [Char] /= [Int]
4703 <sect4 id="type-family-decidability">
4704 <title>Decidability of type synonym instances</title>
4706 In order to guarantee that type inference in the presence of type
4707 families decidable, we need to place a number of additional
4708 restrictions on the formation of type instance declarations (c.f.,
4709 Definition 5 (Relaxed Conditions) of “<ulink
4710 url="http://www.cse.unsw.edu.au/~chak/papers/SPCS08.html">Type
4711 Checking with Open Type Functions</ulink>”). Instance
4712 declarations have the general form
4714 type instance F t1 .. tn = t
4716 where we require that for every type family application <literal>(G s1
4717 .. sm)</literal> in <literal>t</literal>,
4720 <para><literal>s1 .. sm</literal> do not contain any type family
4721 constructors,</para>
4724 <para>the total number of symbols (data type constructors and type
4725 variables) in <literal>s1 .. sm</literal> is strictly smaller than
4726 in <literal>t1 .. tn</literal>, and</para>
4729 <para>for every type
4730 variable <literal>a</literal>, <literal>a</literal> occurs
4731 in <literal>s1 .. sm</literal> at most as often as in <literal>t1
4732 .. tn</literal>.</para>
4735 These restrictions are easily verified and ensure termination of type
4736 inference. However, they are not sufficient to guarantee completeness
4737 of type inference in the presence of, so called, ''loopy equalities'',
4738 such as <literal>a ~ [F a]</literal>, where a recursive occurrence of
4739 a type variable is underneath a family application and data
4740 constructor application - see the above mentioned paper for details.
4743 If the option <option>-XUndecidableInstances</option> is passed to the
4744 compiler, the above restrictions are not enforced and it is on the
4745 programmer to ensure termination of the normalisation of type families
4746 during type inference.
4751 <sect3 id-="equality-constraints">
4752 <title>Equality constraints</title>
4754 Type context can include equality constraints of the form <literal>t1 ~
4755 t2</literal>, which denote that the types <literal>t1</literal>
4756 and <literal>t2</literal> need to be the same. In the presence of type
4757 families, whether two types are equal cannot generally be decided
4758 locally. Hence, the contexts of function signatures may include
4759 equality constraints, as in the following example:
4761 sumCollects :: (Collects c1, Collects c2, Elem c1 ~ Elem c2) => c1 -> c2 -> c2
4763 where we require that the element type of <literal>c1</literal>
4764 and <literal>c2</literal> are the same. In general, the
4765 types <literal>t1</literal> and <literal>t2</literal> of an equality
4766 constraint may be arbitrary monotypes; i.e., they may not contain any
4767 quantifiers, independent of whether higher-rank types are otherwise
4771 Equality constraints can also appear in class and instance contexts.
4772 The former enable a simple translation of programs using functional
4773 dependencies into programs using family synonyms instead. The general
4774 idea is to rewrite a class declaration of the form
4776 class C a b | a -> b
4780 class (F a ~ b) => C a b where
4783 That is, we represent every functional dependency (FD) <literal>a1 .. an
4784 -> b</literal> by an FD type family <literal>F a1 .. an</literal> and a
4785 superclass context equality <literal>F a1 .. an ~ b</literal>,
4786 essentially giving a name to the functional dependency. In class
4787 instances, we define the type instances of FD families in accordance
4788 with the class head. Method signatures are not affected by that
4792 NB: Equalities in superclass contexts are not fully implemented in
4797 <sect3 id-="ty-fams-in-instances">
4798 <title>Type families and instance declarations</title>
4799 <para>Type families require us to extend the rules for
4800 the form of instance heads, which are given
4801 in <xref linkend="flexible-instance-head"/>.
4804 <listitem><para>Data type families may appear in an instance head</para></listitem>
4805 <listitem><para>Type synonym families may not appear (at all) in an instance head</para></listitem>
4807 The reason for the latter restriction is that there is no way to check for. Consider
4810 type instance F Bool = Int
4817 Now a constraint <literal>(C (F Bool))</literal> would match both instances.
4818 The situation is especially bad because the type instance for <literal>F Bool</literal>
4819 might be in another module, or even in a module that is not yet written.
4826 <sect1 id="other-type-extensions">
4827 <title>Other type system extensions</title>
4829 <sect2 id="explicit-foralls"><title>Explicit universal quantification (forall)</title>
4831 Haskell type signatures are implicitly quantified. When the language option <option>-XExplicitForAll</option>
4832 is used, the keyword <literal>forall</literal>
4833 allows us to say exactly what this means. For example:
4841 g :: forall b. (b -> b)
4843 The two are treated identically.
4846 Of course <literal>forall</literal> becomes a keyword; you can't use <literal>forall</literal> as
4847 a type variable any more!
4852 <sect2 id="flexible-contexts"><title>The context of a type signature</title>
4854 The <option>-XFlexibleContexts</option> flag lifts the Haskell 98 restriction
4855 that the type-class constraints in a type signature must have the
4856 form <emphasis>(class type-variable)</emphasis> or
4857 <emphasis>(class (type-variable type-variable ...))</emphasis>.
4858 With <option>-XFlexibleContexts</option>
4859 these type signatures are perfectly OK
4862 g :: Ord (T a ()) => ...
4864 The flag <option>-XFlexibleContexts</option> also lifts the corresponding
4865 restriction on class declarations (<xref linkend="superclass-rules"/>) and instance declarations
4866 (<xref linkend="instance-rules"/>).
4870 GHC imposes the following restrictions on the constraints in a type signature.
4874 forall tv1..tvn (c1, ...,cn) => type
4877 (Here, we write the "foralls" explicitly, although the Haskell source
4878 language omits them; in Haskell 98, all the free type variables of an
4879 explicit source-language type signature are universally quantified,
4880 except for the class type variables in a class declaration. However,
4881 in GHC, you can give the foralls if you want. See <xref linkend="explicit-foralls"/>).
4890 <emphasis>Each universally quantified type variable
4891 <literal>tvi</literal> must be reachable from <literal>type</literal></emphasis>.
4893 A type variable <literal>a</literal> is "reachable" if it appears
4894 in the same constraint as either a type variable free in
4895 <literal>type</literal>, or another reachable type variable.
4896 A value with a type that does not obey
4897 this reachability restriction cannot be used without introducing
4898 ambiguity; that is why the type is rejected.
4899 Here, for example, is an illegal type:
4903 forall a. Eq a => Int
4907 When a value with this type was used, the constraint <literal>Eq tv</literal>
4908 would be introduced where <literal>tv</literal> is a fresh type variable, and
4909 (in the dictionary-translation implementation) the value would be
4910 applied to a dictionary for <literal>Eq tv</literal>. The difficulty is that we
4911 can never know which instance of <literal>Eq</literal> to use because we never
4912 get any more information about <literal>tv</literal>.
4916 that the reachability condition is weaker than saying that <literal>a</literal> is
4917 functionally dependent on a type variable free in
4918 <literal>type</literal> (see <xref
4919 linkend="functional-dependencies"/>). The reason for this is there
4920 might be a "hidden" dependency, in a superclass perhaps. So
4921 "reachable" is a conservative approximation to "functionally dependent".
4922 For example, consider:
4924 class C a b | a -> b where ...
4925 class C a b => D a b where ...
4926 f :: forall a b. D a b => a -> a
4928 This is fine, because in fact <literal>a</literal> does functionally determine <literal>b</literal>
4929 but that is not immediately apparent from <literal>f</literal>'s type.
4935 <emphasis>Every constraint <literal>ci</literal> must mention at least one of the
4936 universally quantified type variables <literal>tvi</literal></emphasis>.
4938 For example, this type is OK because <literal>C a b</literal> mentions the
4939 universally quantified type variable <literal>b</literal>:
4943 forall a. C a b => burble
4947 The next type is illegal because the constraint <literal>Eq b</literal> does not
4948 mention <literal>a</literal>:
4952 forall a. Eq b => burble
4956 The reason for this restriction is milder than the other one. The
4957 excluded types are never useful or necessary (because the offending
4958 context doesn't need to be witnessed at this point; it can be floated
4959 out). Furthermore, floating them out increases sharing. Lastly,
4960 excluding them is a conservative choice; it leaves a patch of
4961 territory free in case we need it later.
4972 <sect2 id="implicit-parameters">
4973 <title>Implicit parameters</title>
4975 <para> Implicit parameters are implemented as described in
4976 "Implicit parameters: dynamic scoping with static types",
4977 J Lewis, MB Shields, E Meijer, J Launchbury,
4978 27th ACM Symposium on Principles of Programming Languages (POPL'00),
4982 <para>(Most of the following, still rather incomplete, documentation is
4983 due to Jeff Lewis.)</para>
4985 <para>Implicit parameter support is enabled with the option
4986 <option>-XImplicitParams</option>.</para>
4989 A variable is called <emphasis>dynamically bound</emphasis> when it is bound by the calling
4990 context of a function and <emphasis>statically bound</emphasis> when bound by the callee's
4991 context. In Haskell, all variables are statically bound. Dynamic
4992 binding of variables is a notion that goes back to Lisp, but was later
4993 discarded in more modern incarnations, such as Scheme. Dynamic binding
4994 can be very confusing in an untyped language, and unfortunately, typed
4995 languages, in particular Hindley-Milner typed languages like Haskell,
4996 only support static scoping of variables.
4999 However, by a simple extension to the type class system of Haskell, we
5000 can support dynamic binding. Basically, we express the use of a
5001 dynamically bound variable as a constraint on the type. These
5002 constraints lead to types of the form <literal>(?x::t') => t</literal>, which says "this
5003 function uses a dynamically-bound variable <literal>?x</literal>
5004 of type <literal>t'</literal>". For
5005 example, the following expresses the type of a sort function,
5006 implicitly parameterized by a comparison function named <literal>cmp</literal>.
5008 sort :: (?cmp :: a -> a -> Bool) => [a] -> [a]
5010 The dynamic binding constraints are just a new form of predicate in the type class system.
5013 An implicit parameter occurs in an expression using the special form <literal>?x</literal>,
5014 where <literal>x</literal> is
5015 any valid identifier (e.g. <literal>ord ?x</literal> is a valid expression).
5016 Use of this construct also introduces a new
5017 dynamic-binding constraint in the type of the expression.
5018 For example, the following definition
5019 shows how we can define an implicitly parameterized sort function in
5020 terms of an explicitly parameterized <literal>sortBy</literal> function:
5022 sortBy :: (a -> a -> Bool) -> [a] -> [a]
5024 sort :: (?cmp :: a -> a -> Bool) => [a] -> [a]
5030 <title>Implicit-parameter type constraints</title>
5032 Dynamic binding constraints behave just like other type class
5033 constraints in that they are automatically propagated. Thus, when a
5034 function is used, its implicit parameters are inherited by the
5035 function that called it. For example, our <literal>sort</literal> function might be used
5036 to pick out the least value in a list:
5038 least :: (?cmp :: a -> a -> Bool) => [a] -> a
5039 least xs = head (sort xs)
5041 Without lifting a finger, the <literal>?cmp</literal> parameter is
5042 propagated to become a parameter of <literal>least</literal> as well. With explicit
5043 parameters, the default is that parameters must always be explicit
5044 propagated. With implicit parameters, the default is to always
5048 An implicit-parameter type constraint differs from other type class constraints in the
5049 following way: All uses of a particular implicit parameter must have
5050 the same type. This means that the type of <literal>(?x, ?x)</literal>
5051 is <literal>(?x::a) => (a,a)</literal>, and not
5052 <literal>(?x::a, ?x::b) => (a, b)</literal>, as would be the case for type
5056 <para> You can't have an implicit parameter in the context of a class or instance
5057 declaration. For example, both these declarations are illegal:
5059 class (?x::Int) => C a where ...
5060 instance (?x::a) => Foo [a] where ...
5062 Reason: exactly which implicit parameter you pick up depends on exactly where
5063 you invoke a function. But the ``invocation'' of instance declarations is done
5064 behind the scenes by the compiler, so it's hard to figure out exactly where it is done.
5065 Easiest thing is to outlaw the offending types.</para>
5067 Implicit-parameter constraints do not cause ambiguity. For example, consider:
5069 f :: (?x :: [a]) => Int -> Int
5072 g :: (Read a, Show a) => String -> String
5075 Here, <literal>g</literal> has an ambiguous type, and is rejected, but <literal>f</literal>
5076 is fine. The binding for <literal>?x</literal> at <literal>f</literal>'s call site is
5077 quite unambiguous, and fixes the type <literal>a</literal>.
5082 <title>Implicit-parameter bindings</title>
5085 An implicit parameter is <emphasis>bound</emphasis> using the standard
5086 <literal>let</literal> or <literal>where</literal> binding forms.
5087 For example, we define the <literal>min</literal> function by binding
5088 <literal>cmp</literal>.
5091 min = let ?cmp = (<=) in least
5095 A group of implicit-parameter bindings may occur anywhere a normal group of Haskell
5096 bindings can occur, except at top level. That is, they can occur in a <literal>let</literal>
5097 (including in a list comprehension, or do-notation, or pattern guards),
5098 or a <literal>where</literal> clause.
5099 Note the following points:
5102 An implicit-parameter binding group must be a
5103 collection of simple bindings to implicit-style variables (no
5104 function-style bindings, and no type signatures); these bindings are
5105 neither polymorphic or recursive.
5108 You may not mix implicit-parameter bindings with ordinary bindings in a
5109 single <literal>let</literal>
5110 expression; use two nested <literal>let</literal>s instead.
5111 (In the case of <literal>where</literal> you are stuck, since you can't nest <literal>where</literal> clauses.)
5115 You may put multiple implicit-parameter bindings in a
5116 single binding group; but they are <emphasis>not</emphasis> treated
5117 as a mutually recursive group (as ordinary <literal>let</literal> bindings are).
5118 Instead they are treated as a non-recursive group, simultaneously binding all the implicit
5119 parameter. The bindings are not nested, and may be re-ordered without changing
5120 the meaning of the program.
5121 For example, consider:
5123 f t = let { ?x = t; ?y = ?x+(1::Int) } in ?x + ?y
5125 The use of <literal>?x</literal> in the binding for <literal>?y</literal> does not "see"
5126 the binding for <literal>?x</literal>, so the type of <literal>f</literal> is
5128 f :: (?x::Int) => Int -> Int
5136 <sect3><title>Implicit parameters and polymorphic recursion</title>
5139 Consider these two definitions:
5142 len1 xs = let ?acc = 0 in len_acc1 xs
5145 len_acc1 (x:xs) = let ?acc = ?acc + (1::Int) in len_acc1 xs
5150 len2 xs = let ?acc = 0 in len_acc2 xs
5152 len_acc2 :: (?acc :: Int) => [a] -> Int
5154 len_acc2 (x:xs) = let ?acc = ?acc + (1::Int) in len_acc2 xs
5156 The only difference between the two groups is that in the second group
5157 <literal>len_acc</literal> is given a type signature.
5158 In the former case, <literal>len_acc1</literal> is monomorphic in its own
5159 right-hand side, so the implicit parameter <literal>?acc</literal> is not
5160 passed to the recursive call. In the latter case, because <literal>len_acc2</literal>
5161 has a type signature, the recursive call is made to the
5162 <emphasis>polymorphic</emphasis> version, which takes <literal>?acc</literal>
5163 as an implicit parameter. So we get the following results in GHCi:
5170 Adding a type signature dramatically changes the result! This is a rather
5171 counter-intuitive phenomenon, worth watching out for.
5175 <sect3><title>Implicit parameters and monomorphism</title>
5177 <para>GHC applies the dreaded Monomorphism Restriction (section 4.5.5 of the
5178 Haskell Report) to implicit parameters. For example, consider:
5186 Since the binding for <literal>y</literal> falls under the Monomorphism
5187 Restriction it is not generalised, so the type of <literal>y</literal> is
5188 simply <literal>Int</literal>, not <literal>(?x::Int) => Int</literal>.
5189 Hence, <literal>(f 9)</literal> returns result <literal>9</literal>.
5190 If you add a type signature for <literal>y</literal>, then <literal>y</literal>
5191 will get type <literal>(?x::Int) => Int</literal>, so the occurrence of
5192 <literal>y</literal> in the body of the <literal>let</literal> will see the
5193 inner binding of <literal>?x</literal>, so <literal>(f 9)</literal> will return
5194 <literal>14</literal>.
5199 <!-- ======================= COMMENTED OUT ========================
5201 We intend to remove linear implicit parameters, so I'm at least removing
5202 them from the 6.6 user manual
5204 <sect2 id="linear-implicit-parameters">
5205 <title>Linear implicit parameters</title>
5207 Linear implicit parameters are an idea developed by Koen Claessen,
5208 Mark Shields, and Simon PJ. They address the long-standing
5209 problem that monads seem over-kill for certain sorts of problem, notably:
5212 <listitem> <para> distributing a supply of unique names </para> </listitem>
5213 <listitem> <para> distributing a supply of random numbers </para> </listitem>
5214 <listitem> <para> distributing an oracle (as in QuickCheck) </para> </listitem>
5218 Linear implicit parameters are just like ordinary implicit parameters,
5219 except that they are "linear"; that is, they cannot be copied, and
5220 must be explicitly "split" instead. Linear implicit parameters are
5221 written '<literal>%x</literal>' instead of '<literal>?x</literal>'.
5222 (The '/' in the '%' suggests the split!)
5227 import GHC.Exts( Splittable )
5229 data NameSupply = ...
5231 splitNS :: NameSupply -> (NameSupply, NameSupply)
5232 newName :: NameSupply -> Name
5234 instance Splittable NameSupply where
5238 f :: (%ns :: NameSupply) => Env -> Expr -> Expr
5239 f env (Lam x e) = Lam x' (f env e)
5242 env' = extend env x x'
5243 ...more equations for f...
5245 Notice that the implicit parameter %ns is consumed
5247 <listitem> <para> once by the call to <literal>newName</literal> </para> </listitem>
5248 <listitem> <para> once by the recursive call to <literal>f</literal> </para></listitem>
5252 So the translation done by the type checker makes
5253 the parameter explicit:
5255 f :: NameSupply -> Env -> Expr -> Expr
5256 f ns env (Lam x e) = Lam x' (f ns1 env e)
5258 (ns1,ns2) = splitNS ns
5260 env = extend env x x'
5262 Notice the call to 'split' introduced by the type checker.
5263 How did it know to use 'splitNS'? Because what it really did
5264 was to introduce a call to the overloaded function 'split',
5265 defined by the class <literal>Splittable</literal>:
5267 class Splittable a where
5270 The instance for <literal>Splittable NameSupply</literal> tells GHC how to implement
5271 split for name supplies. But we can simply write
5277 g :: (Splittable a, %ns :: a) => b -> (b,a,a)
5279 The <literal>Splittable</literal> class is built into GHC. It's exported by module
5280 <literal>GHC.Exts</literal>.
5285 <listitem> <para> '<literal>?x</literal>' and '<literal>%x</literal>'
5286 are entirely distinct implicit parameters: you
5287 can use them together and they won't interfere with each other. </para>
5290 <listitem> <para> You can bind linear implicit parameters in 'with' clauses. </para> </listitem>
5292 <listitem> <para>You cannot have implicit parameters (whether linear or not)
5293 in the context of a class or instance declaration. </para></listitem>
5297 <sect3><title>Warnings</title>
5300 The monomorphism restriction is even more important than usual.
5301 Consider the example above:
5303 f :: (%ns :: NameSupply) => Env -> Expr -> Expr
5304 f env (Lam x e) = Lam x' (f env e)
5307 env' = extend env x x'
5309 If we replaced the two occurrences of x' by (newName %ns), which is
5310 usually a harmless thing to do, we get:
5312 f :: (%ns :: NameSupply) => Env -> Expr -> Expr
5313 f env (Lam x e) = Lam (newName %ns) (f env e)
5315 env' = extend env x (newName %ns)
5317 But now the name supply is consumed in <emphasis>three</emphasis> places
5318 (the two calls to newName,and the recursive call to f), so
5319 the result is utterly different. Urk! We don't even have
5323 Well, this is an experimental change. With implicit
5324 parameters we have already lost beta reduction anyway, and
5325 (as John Launchbury puts it) we can't sensibly reason about
5326 Haskell programs without knowing their typing.
5331 <sect3><title>Recursive functions</title>
5332 <para>Linear implicit parameters can be particularly tricky when you have a recursive function
5335 foo :: %x::T => Int -> [Int]
5337 foo n = %x : foo (n-1)
5339 where T is some type in class Splittable.</para>
5341 Do you get a list of all the same T's or all different T's
5342 (assuming that split gives two distinct T's back)?
5344 If you supply the type signature, taking advantage of polymorphic
5345 recursion, you get what you'd probably expect. Here's the
5346 translated term, where the implicit param is made explicit:
5349 foo x n = let (x1,x2) = split x
5350 in x1 : foo x2 (n-1)
5352 But if you don't supply a type signature, GHC uses the Hindley
5353 Milner trick of using a single monomorphic instance of the function
5354 for the recursive calls. That is what makes Hindley Milner type inference
5355 work. So the translation becomes
5359 foom n = x : foom (n-1)
5363 Result: 'x' is not split, and you get a list of identical T's. So the
5364 semantics of the program depends on whether or not foo has a type signature.
5367 You may say that this is a good reason to dislike linear implicit parameters
5368 and you'd be right. That is why they are an experimental feature.
5374 ================ END OF Linear Implicit Parameters commented out -->
5376 <sect2 id="kinding">
5377 <title>Explicitly-kinded quantification</title>
5380 Haskell infers the kind of each type variable. Sometimes it is nice to be able
5381 to give the kind explicitly as (machine-checked) documentation,
5382 just as it is nice to give a type signature for a function. On some occasions,
5383 it is essential to do so. For example, in his paper "Restricted Data Types in Haskell" (Haskell Workshop 1999)
5384 John Hughes had to define the data type:
5386 data Set cxt a = Set [a]
5387 | Unused (cxt a -> ())
5389 The only use for the <literal>Unused</literal> constructor was to force the correct
5390 kind for the type variable <literal>cxt</literal>.
5393 GHC now instead allows you to specify the kind of a type variable directly, wherever
5394 a type variable is explicitly bound, with the flag <option>-XKindSignatures</option>.
5397 This flag enables kind signatures in the following places:
5399 <listitem><para><literal>data</literal> declarations:
5401 data Set (cxt :: * -> *) a = Set [a]
5402 </screen></para></listitem>
5403 <listitem><para><literal>type</literal> declarations:
5405 type T (f :: * -> *) = f Int
5406 </screen></para></listitem>
5407 <listitem><para><literal>class</literal> declarations:
5409 class (Eq a) => C (f :: * -> *) a where ...
5410 </screen></para></listitem>
5411 <listitem><para><literal>forall</literal>'s in type signatures:
5413 f :: forall (cxt :: * -> *). Set cxt Int
5414 </screen></para></listitem>
5419 The parentheses are required. Some of the spaces are required too, to
5420 separate the lexemes. If you write <literal>(f::*->*)</literal> you
5421 will get a parse error, because "<literal>::*->*</literal>" is a
5422 single lexeme in Haskell.
5426 As part of the same extension, you can put kind annotations in types
5429 f :: (Int :: *) -> Int
5430 g :: forall a. a -> (a :: *)
5434 atype ::= '(' ctype '::' kind ')
5436 The parentheses are required.
5441 <sect2 id="universal-quantification">
5442 <title>Arbitrary-rank polymorphism
5446 GHC's type system supports <emphasis>arbitrary-rank</emphasis>
5447 explicit universal quantification in
5449 For example, all the following types are legal:
5451 f1 :: forall a b. a -> b -> a
5452 g1 :: forall a b. (Ord a, Eq b) => a -> b -> a
5454 f2 :: (forall a. a->a) -> Int -> Int
5455 g2 :: (forall a. Eq a => [a] -> a -> Bool) -> Int -> Int
5457 f3 :: ((forall a. a->a) -> Int) -> Bool -> Bool
5459 f4 :: Int -> (forall a. a -> a)
5461 Here, <literal>f1</literal> and <literal>g1</literal> are rank-1 types, and
5462 can be written in standard Haskell (e.g. <literal>f1 :: a->b->a</literal>).
5463 The <literal>forall</literal> makes explicit the universal quantification that
5464 is implicitly added by Haskell.
5467 The functions <literal>f2</literal> and <literal>g2</literal> have rank-2 types;
5468 the <literal>forall</literal> is on the left of a function arrow. As <literal>g2</literal>
5469 shows, the polymorphic type on the left of the function arrow can be overloaded.
5472 The function <literal>f3</literal> has a rank-3 type;
5473 it has rank-2 types on the left of a function arrow.
5476 GHC has three flags to control higher-rank types:
5479 <option>-XPolymorphicComponents</option>: data constructors (only) can have polymorphic argument types.
5482 <option>-XRank2Types</option>: any function (including data constructors) can have a rank-2 type.
5485 <option>-XRankNTypes</option>: any function (including data constructors) can have an arbitrary-rank type.
5486 That is, you can nest <literal>forall</literal>s
5487 arbitrarily deep in function arrows.
5488 In particular, a forall-type (also called a "type scheme"),
5489 including an operational type class context, is legal:
5491 <listitem> <para> On the left or right (see <literal>f4</literal>, for example)
5492 of a function arrow </para> </listitem>
5493 <listitem> <para> As the argument of a constructor, or type of a field, in a data type declaration. For
5494 example, any of the <literal>f1,f2,f3,g1,g2</literal> above would be valid
5495 field type signatures.</para> </listitem>
5496 <listitem> <para> As the type of an implicit parameter </para> </listitem>
5497 <listitem> <para> In a pattern type signature (see <xref linkend="scoped-type-variables"/>) </para> </listitem>
5509 In a <literal>data</literal> or <literal>newtype</literal> declaration one can quantify
5510 the types of the constructor arguments. Here are several examples:
5516 data T a = T1 (forall b. b -> b -> b) a
5518 data MonadT m = MkMonad { return :: forall a. a -> m a,
5519 bind :: forall a b. m a -> (a -> m b) -> m b
5522 newtype Swizzle = MkSwizzle (Ord a => [a] -> [a])
5528 The constructors have rank-2 types:
5534 T1 :: forall a. (forall b. b -> b -> b) -> a -> T a
5535 MkMonad :: forall m. (forall a. a -> m a)
5536 -> (forall a b. m a -> (a -> m b) -> m b)
5538 MkSwizzle :: (Ord a => [a] -> [a]) -> Swizzle
5544 Notice that you don't need to use a <literal>forall</literal> if there's an
5545 explicit context. For example in the first argument of the
5546 constructor <function>MkSwizzle</function>, an implicit "<literal>forall a.</literal>" is
5547 prefixed to the argument type. The implicit <literal>forall</literal>
5548 quantifies all type variables that are not already in scope, and are
5549 mentioned in the type quantified over.
5553 As for type signatures, implicit quantification happens for non-overloaded
5554 types too. So if you write this:
5557 data T a = MkT (Either a b) (b -> b)
5560 it's just as if you had written this:
5563 data T a = MkT (forall b. Either a b) (forall b. b -> b)
5566 That is, since the type variable <literal>b</literal> isn't in scope, it's
5567 implicitly universally quantified. (Arguably, it would be better
5568 to <emphasis>require</emphasis> explicit quantification on constructor arguments
5569 where that is what is wanted. Feedback welcomed.)
5573 You construct values of types <literal>T1, MonadT, Swizzle</literal> by applying
5574 the constructor to suitable values, just as usual. For example,
5585 a3 = MkSwizzle reverse
5588 a4 = let r x = Just x
5595 mkTs :: (forall b. b -> b -> b) -> a -> [T a]
5596 mkTs f x y = [T1 f x, T1 f y]
5602 The type of the argument can, as usual, be more general than the type
5603 required, as <literal>(MkSwizzle reverse)</literal> shows. (<function>reverse</function>
5604 does not need the <literal>Ord</literal> constraint.)
5608 When you use pattern matching, the bound variables may now have
5609 polymorphic types. For example:
5615 f :: T a -> a -> (a, Char)
5616 f (T1 w k) x = (w k x, w 'c' 'd')
5618 g :: (Ord a, Ord b) => Swizzle -> [a] -> (a -> b) -> [b]
5619 g (MkSwizzle s) xs f = s (map f (s xs))
5621 h :: MonadT m -> [m a] -> m [a]
5622 h m [] = return m []
5623 h m (x:xs) = bind m x $ \y ->
5624 bind m (h m xs) $ \ys ->
5631 In the function <function>h</function> we use the record selectors <literal>return</literal>
5632 and <literal>bind</literal> to extract the polymorphic bind and return functions
5633 from the <literal>MonadT</literal> data structure, rather than using pattern
5639 <title>Type inference</title>
5642 In general, type inference for arbitrary-rank types is undecidable.
5643 GHC uses an algorithm proposed by Odersky and Laufer ("Putting type annotations to work", POPL'96)
5644 to get a decidable algorithm by requiring some help from the programmer.
5645 We do not yet have a formal specification of "some help" but the rule is this:
5648 <emphasis>For a lambda-bound or case-bound variable, x, either the programmer
5649 provides an explicit polymorphic type for x, or GHC's type inference will assume
5650 that x's type has no foralls in it</emphasis>.
5653 What does it mean to "provide" an explicit type for x? You can do that by
5654 giving a type signature for x directly, using a pattern type signature
5655 (<xref linkend="scoped-type-variables"/>), thus:
5657 \ f :: (forall a. a->a) -> (f True, f 'c')
5659 Alternatively, you can give a type signature to the enclosing
5660 context, which GHC can "push down" to find the type for the variable:
5662 (\ f -> (f True, f 'c')) :: (forall a. a->a) -> (Bool,Char)
5664 Here the type signature on the expression can be pushed inwards
5665 to give a type signature for f. Similarly, and more commonly,
5666 one can give a type signature for the function itself:
5668 h :: (forall a. a->a) -> (Bool,Char)
5669 h f = (f True, f 'c')
5671 You don't need to give a type signature if the lambda bound variable
5672 is a constructor argument. Here is an example we saw earlier:
5674 f :: T a -> a -> (a, Char)
5675 f (T1 w k) x = (w k x, w 'c' 'd')
5677 Here we do not need to give a type signature to <literal>w</literal>, because
5678 it is an argument of constructor <literal>T1</literal> and that tells GHC all
5685 <sect3 id="implicit-quant">
5686 <title>Implicit quantification</title>
5689 GHC performs implicit quantification as follows. <emphasis>At the top level (only) of
5690 user-written types, if and only if there is no explicit <literal>forall</literal>,
5691 GHC finds all the type variables mentioned in the type that are not already
5692 in scope, and universally quantifies them.</emphasis> For example, the following pairs are
5696 f :: forall a. a -> a
5703 h :: forall b. a -> b -> b
5709 Notice that GHC does <emphasis>not</emphasis> find the innermost possible quantification
5712 f :: (a -> a) -> Int
5714 f :: forall a. (a -> a) -> Int
5716 f :: (forall a. a -> a) -> Int
5719 g :: (Ord a => a -> a) -> Int
5720 -- MEANS the illegal type
5721 g :: forall a. (Ord a => a -> a) -> Int
5723 g :: (forall a. Ord a => a -> a) -> Int
5725 The latter produces an illegal type, which you might think is silly,
5726 but at least the rule is simple. If you want the latter type, you
5727 can write your for-alls explicitly. Indeed, doing so is strongly advised
5734 <sect2 id="impredicative-polymorphism">
5735 <title>Impredicative polymorphism
5737 <para><emphasis>NOTE: the impredicative-polymorphism feature is deprecated in GHC 6.12, and
5738 will be removed or replaced in GHC 6.14.</emphasis></para>
5740 <para>GHC supports <emphasis>impredicative polymorphism</emphasis>,
5741 enabled with <option>-XImpredicativeTypes</option>.
5743 that you can call a polymorphic function at a polymorphic type, and
5744 parameterise data structures over polymorphic types. For example:
5746 f :: Maybe (forall a. [a] -> [a]) -> Maybe ([Int], [Char])
5747 f (Just g) = Just (g [3], g "hello")
5750 Notice here that the <literal>Maybe</literal> type is parameterised by the
5751 <emphasis>polymorphic</emphasis> type <literal>(forall a. [a] ->
5754 <para>The technical details of this extension are described in the paper
5755 <ulink url="http://research.microsoft.com/%7Esimonpj/papers/boxy/">Boxy types:
5756 type inference for higher-rank types and impredicativity</ulink>,
5757 which appeared at ICFP 2006.
5761 <sect2 id="scoped-type-variables">
5762 <title>Lexically scoped type variables
5766 GHC supports <emphasis>lexically scoped type variables</emphasis>, without
5767 which some type signatures are simply impossible to write. For example:
5769 f :: forall a. [a] -> [a]
5775 The type signature for <literal>f</literal> brings the type variable <literal>a</literal> into scope,
5776 because of the explicit <literal>forall</literal> (<xref linkend="decl-type-sigs"/>).
5777 The type variables bound by a <literal>forall</literal> scope over
5778 the entire definition of the accompanying value declaration.
5779 In this example, the type variable <literal>a</literal> scopes over the whole
5780 definition of <literal>f</literal>, including over
5781 the type signature for <varname>ys</varname>.
5782 In Haskell 98 it is not possible to declare
5783 a type for <varname>ys</varname>; a major benefit of scoped type variables is that
5784 it becomes possible to do so.
5786 <para>Lexically-scoped type variables are enabled by
5787 <option>-XScopedTypeVariables</option>. This flag implies <option>-XRelaxedPolyRec</option>.
5789 <para>Note: GHC 6.6 contains substantial changes to the way that scoped type
5790 variables work, compared to earlier releases. Read this section
5794 <title>Overview</title>
5796 <para>The design follows the following principles
5798 <listitem><para>A scoped type variable stands for a type <emphasis>variable</emphasis>, and not for
5799 a <emphasis>type</emphasis>. (This is a change from GHC's earlier
5800 design.)</para></listitem>
5801 <listitem><para>Furthermore, distinct lexical type variables stand for distinct
5802 type variables. This means that every programmer-written type signature
5803 (including one that contains free scoped type variables) denotes a
5804 <emphasis>rigid</emphasis> type; that is, the type is fully known to the type
5805 checker, and no inference is involved.</para></listitem>
5806 <listitem><para>Lexical type variables may be alpha-renamed freely, without
5807 changing the program.</para></listitem>
5811 A <emphasis>lexically scoped type variable</emphasis> can be bound by:
5813 <listitem><para>A declaration type signature (<xref linkend="decl-type-sigs"/>)</para></listitem>
5814 <listitem><para>An expression type signature (<xref linkend="exp-type-sigs"/>)</para></listitem>
5815 <listitem><para>A pattern type signature (<xref linkend="pattern-type-sigs"/>)</para></listitem>
5816 <listitem><para>Class and instance declarations (<xref linkend="cls-inst-scoped-tyvars"/>)</para></listitem>
5820 In Haskell, a programmer-written type signature is implicitly quantified over
5821 its free type variables (<ulink
5822 url="http://www.haskell.org/onlinereport/decls.html#sect4.1.2">Section
5824 of the Haskell Report).
5825 Lexically scoped type variables affect this implicit quantification rules
5826 as follows: any type variable that is in scope is <emphasis>not</emphasis> universally
5827 quantified. For example, if type variable <literal>a</literal> is in scope,
5830 (e :: a -> a) means (e :: a -> a)
5831 (e :: b -> b) means (e :: forall b. b->b)
5832 (e :: a -> b) means (e :: forall b. a->b)
5840 <sect3 id="decl-type-sigs">
5841 <title>Declaration type signatures</title>
5842 <para>A declaration type signature that has <emphasis>explicit</emphasis>
5843 quantification (using <literal>forall</literal>) brings into scope the
5844 explicitly-quantified
5845 type variables, in the definition of the named function. For example:
5847 f :: forall a. [a] -> [a]
5848 f (x:xs) = xs ++ [ x :: a ]
5850 The "<literal>forall a</literal>" brings "<literal>a</literal>" into scope in
5851 the definition of "<literal>f</literal>".
5853 <para>This only happens if:
5855 <listitem><para> The quantification in <literal>f</literal>'s type
5856 signature is explicit. For example:
5859 g (x:xs) = xs ++ [ x :: a ]
5861 This program will be rejected, because "<literal>a</literal>" does not scope
5862 over the definition of "<literal>f</literal>", so "<literal>x::a</literal>"
5863 means "<literal>x::forall a. a</literal>" by Haskell's usual implicit
5864 quantification rules.
5866 <listitem><para> The signature gives a type for a function binding or a bare variable binding,
5867 not a pattern binding.
5870 f1 :: forall a. [a] -> [a]
5871 f1 (x:xs) = xs ++ [ x :: a ] -- OK
5873 f2 :: forall a. [a] -> [a]
5874 f2 = \(x:xs) -> xs ++ [ x :: a ] -- OK
5876 f3 :: forall a. [a] -> [a]
5877 Just f3 = Just (\(x:xs) -> xs ++ [ x :: a ]) -- Not OK!
5879 The binding for <literal>f3</literal> is a pattern binding, and so its type signature
5880 does not bring <literal>a</literal> into scope. However <literal>f1</literal> is a
5881 function binding, and <literal>f2</literal> binds a bare variable; in both cases
5882 the type signature brings <literal>a</literal> into scope.
5888 <sect3 id="exp-type-sigs">
5889 <title>Expression type signatures</title>
5891 <para>An expression type signature that has <emphasis>explicit</emphasis>
5892 quantification (using <literal>forall</literal>) brings into scope the
5893 explicitly-quantified
5894 type variables, in the annotated expression. For example:
5896 f = runST ( (op >>= \(x :: STRef s Int) -> g x) :: forall s. ST s Bool )
5898 Here, the type signature <literal>forall a. ST s Bool</literal> brings the
5899 type variable <literal>s</literal> into scope, in the annotated expression
5900 <literal>(op >>= \(x :: STRef s Int) -> g x)</literal>.
5905 <sect3 id="pattern-type-sigs">
5906 <title>Pattern type signatures</title>
5908 A type signature may occur in any pattern; this is a <emphasis>pattern type
5909 signature</emphasis>.
5912 -- f and g assume that 'a' is already in scope
5913 f = \(x::Int, y::a) -> x
5915 h ((x,y) :: (Int,Bool)) = (y,x)
5917 In the case where all the type variables in the pattern type signature are
5918 already in scope (i.e. bound by the enclosing context), matters are simple: the
5919 signature simply constrains the type of the pattern in the obvious way.
5922 Unlike expression and declaration type signatures, pattern type signatures are not implicitly generalised.
5923 The pattern in a <emphasis>pattern binding</emphasis> may only mention type variables
5924 that are already in scope. For example:
5926 f :: forall a. [a] -> (Int, [a])
5929 (ys::[a], n) = (reverse xs, length xs) -- OK
5930 zs::[a] = xs ++ ys -- OK
5932 Just (v::b) = ... -- Not OK; b is not in scope
5934 Here, the pattern signatures for <literal>ys</literal> and <literal>zs</literal>
5935 are fine, but the one for <literal>v</literal> is not because <literal>b</literal> is
5939 However, in all patterns <emphasis>other</emphasis> than pattern bindings, a pattern
5940 type signature may mention a type variable that is not in scope; in this case,
5941 <emphasis>the signature brings that type variable into scope</emphasis>.
5942 This is particularly important for existential data constructors. For example:
5944 data T = forall a. MkT [a]
5947 k (MkT [t::a]) = MkT t3
5951 Here, the pattern type signature <literal>(t::a)</literal> mentions a lexical type
5952 variable that is not already in scope. Indeed, it <emphasis>cannot</emphasis> already be in scope,
5953 because it is bound by the pattern match. GHC's rule is that in this situation
5954 (and only then), a pattern type signature can mention a type variable that is
5955 not already in scope; the effect is to bring it into scope, standing for the
5956 existentially-bound type variable.
5959 When a pattern type signature binds a type variable in this way, GHC insists that the
5960 type variable is bound to a <emphasis>rigid</emphasis>, or fully-known, type variable.
5961 This means that any user-written type signature always stands for a completely known type.
5964 If all this seems a little odd, we think so too. But we must have
5965 <emphasis>some</emphasis> way to bring such type variables into scope, else we
5966 could not name existentially-bound type variables in subsequent type signatures.
5969 This is (now) the <emphasis>only</emphasis> situation in which a pattern type
5970 signature is allowed to mention a lexical variable that is not already in
5972 For example, both <literal>f</literal> and <literal>g</literal> would be
5973 illegal if <literal>a</literal> was not already in scope.
5979 <!-- ==================== Commented out part about result type signatures
5981 <sect3 id="result-type-sigs">
5982 <title>Result type signatures</title>
5985 The result type of a function, lambda, or case expression alternative can be given a signature, thus:
5988 {- f assumes that 'a' is already in scope -}
5989 f x y :: [a] = [x,y,x]
5991 g = \ x :: [Int] -> [3,4]
5993 h :: forall a. [a] -> a
5997 The final <literal>:: [a]</literal> after the patterns of <literal>f</literal> gives the type of
5998 the result of the function. Similarly, the body of the lambda in the RHS of
5999 <literal>g</literal> is <literal>[Int]</literal>, and the RHS of the case
6000 alternative in <literal>h</literal> is <literal>a</literal>.
6002 <para> A result type signature never brings new type variables into scope.</para>
6004 There are a couple of syntactic wrinkles. First, notice that all three
6005 examples would parse quite differently with parentheses:
6007 {- f assumes that 'a' is already in scope -}
6008 f x (y :: [a]) = [x,y,x]
6010 g = \ (x :: [Int]) -> [3,4]
6012 h :: forall a. [a] -> a
6016 Now the signature is on the <emphasis>pattern</emphasis>; and
6017 <literal>h</literal> would certainly be ill-typed (since the pattern
6018 <literal>(y:ys)</literal> cannot have the type <literal>a</literal>.
6020 Second, to avoid ambiguity, the type after the “<literal>::</literal>” in a result
6021 pattern signature on a lambda or <literal>case</literal> must be atomic (i.e. a single
6022 token or a parenthesised type of some sort). To see why,
6023 consider how one would parse this:
6032 <sect3 id="cls-inst-scoped-tyvars">
6033 <title>Class and instance declarations</title>
6036 The type variables in the head of a <literal>class</literal> or <literal>instance</literal> declaration
6037 scope over the methods defined in the <literal>where</literal> part. For example:
6055 <sect2 id="typing-binds">
6056 <title>Generalised typing of mutually recursive bindings</title>
6059 The Haskell Report specifies that a group of bindings (at top level, or in a
6060 <literal>let</literal> or <literal>where</literal>) should be sorted into
6061 strongly-connected components, and then type-checked in dependency order
6062 (<ulink url="http://www.haskell.org/onlinereport/decls.html#sect4.5.1">Haskell
6063 Report, Section 4.5.1</ulink>).
6064 As each group is type-checked, any binders of the group that
6066 an explicit type signature are put in the type environment with the specified
6068 and all others are monomorphic until the group is generalised
6069 (<ulink url="http://www.haskell.org/onlinereport/decls.html#sect4.5.2">Haskell Report, Section 4.5.2</ulink>).
6072 <para>Following a suggestion of Mark Jones, in his paper
6073 <ulink url="http://citeseer.ist.psu.edu/424440.html">Typing Haskell in
6075 GHC implements a more general scheme. If <option>-XRelaxedPolyRec</option> is
6077 <emphasis>the dependency analysis ignores references to variables that have an explicit
6078 type signature</emphasis>.
6079 As a result of this refined dependency analysis, the dependency groups are smaller, and more bindings will
6080 typecheck. For example, consider:
6082 f :: Eq a => a -> Bool
6083 f x = (x == x) || g True || g "Yes"
6085 g y = (y <= y) || f True
6087 This is rejected by Haskell 98, but under Jones's scheme the definition for
6088 <literal>g</literal> is typechecked first, separately from that for
6089 <literal>f</literal>,
6090 because the reference to <literal>f</literal> in <literal>g</literal>'s right
6091 hand side is ignored by the dependency analysis. Then <literal>g</literal>'s
6092 type is generalised, to get
6094 g :: Ord a => a -> Bool
6096 Now, the definition for <literal>f</literal> is typechecked, with this type for
6097 <literal>g</literal> in the type environment.
6101 The same refined dependency analysis also allows the type signatures of
6102 mutually-recursive functions to have different contexts, something that is illegal in
6103 Haskell 98 (Section 4.5.2, last sentence). With
6104 <option>-XRelaxedPolyRec</option>
6105 GHC only insists that the type signatures of a <emphasis>refined</emphasis> group have identical
6106 type signatures; in practice this means that only variables bound by the same
6107 pattern binding must have the same context. For example, this is fine:
6109 f :: Eq a => a -> Bool
6110 f x = (x == x) || g True
6112 g :: Ord a => a -> Bool
6113 g y = (y <= y) || f True
6118 <sect2 id="mono-local-binds">
6119 <title>Monomorphic local bindings</title>
6121 We are actively thinking of simplifying GHC's type system, by <emphasis>not generalising local bindings</emphasis>.
6122 The rationale is described in the paper
6123 <ulink url="http://research.microsoft.com/~simonpj/papers/constraints/index.htm">Let should not be generalised</ulink>.
6126 The experimental new behaviour is enabled by the flag <option>-XMonoLocalBinds</option>. The effect is
6127 that local (that is, non-top-level) bindings without a type signature are not generalised at all. You can
6128 think of it as an extreme (but much more predictable) version of the Monomorphism Restriction.
6129 If you supply a type signature, then the flag has no effect.
6134 <!-- ==================== End of type system extensions ================= -->
6136 <!-- ====================== TEMPLATE HASKELL ======================= -->
6138 <sect1 id="template-haskell">
6139 <title>Template Haskell</title>
6141 <para>Template Haskell allows you to do compile-time meta-programming in
6144 the main technical innovations is discussed in "<ulink
6145 url="http://research.microsoft.com/~simonpj/papers/meta-haskell/">
6146 Template Meta-programming for Haskell</ulink>" (Proc Haskell Workshop 2002).
6149 There is a Wiki page about
6150 Template Haskell at <ulink url="http://www.haskell.org/haskellwiki/Template_Haskell">
6151 http://www.haskell.org/haskellwiki/Template_Haskell</ulink>, and that is the best place to look for
6155 url="http://www.haskell.org/ghc/docs/latest/html/libraries/index.html">online
6156 Haskell library reference material</ulink>
6157 (look for module <literal>Language.Haskell.TH</literal>).
6158 Many changes to the original design are described in
6159 <ulink url="http://research.microsoft.com/~simonpj/papers/meta-haskell/notes2.ps">
6160 Notes on Template Haskell version 2</ulink>.
6161 Not all of these changes are in GHC, however.
6164 <para> The first example from that paper is set out below (<xref linkend="th-example"/>)
6165 as a worked example to help get you started.
6169 The documentation here describes the realisation of Template Haskell in GHC. It is not detailed enough to
6170 understand Template Haskell; see the <ulink url="http://haskell.org/haskellwiki/Template_Haskell">
6175 <title>Syntax</title>
6177 <para> Template Haskell has the following new syntactic
6178 constructions. You need to use the flag
6179 <option>-XTemplateHaskell</option>
6180 <indexterm><primary><option>-XTemplateHaskell</option></primary>
6181 </indexterm>to switch these syntactic extensions on
6182 (<option>-XTemplateHaskell</option> is no longer implied by
6183 <option>-fglasgow-exts</option>).</para>
6187 A splice is written <literal>$x</literal>, where <literal>x</literal> is an
6188 identifier, or <literal>$(...)</literal>, where the "..." is an arbitrary expression.
6189 There must be no space between the "$" and the identifier or parenthesis. This use
6190 of "$" overrides its meaning as an infix operator, just as "M.x" overrides the meaning
6191 of "." as an infix operator. If you want the infix operator, put spaces around it.
6193 <para> A splice can occur in place of
6195 <listitem><para> an expression; the spliced expression must
6196 have type <literal>Q Exp</literal></para></listitem>
6197 <listitem><para> an type; the spliced expression must
6198 have type <literal>Q Typ</literal></para></listitem>
6199 <listitem><para> a list of top-level declarations; the spliced expression
6200 must have type <literal>Q [Dec]</literal></para></listitem>
6202 Note that pattern splices are not supported.
6203 Inside a splice you can can only call functions defined in imported modules,
6204 not functions defined elsewhere in the same module.</para></listitem>
6207 A expression quotation is written in Oxford brackets, thus:
6209 <listitem><para> <literal>[| ... |]</literal>, or <literal>[e| ... |]</literal>,
6210 where the "..." is an expression;
6211 the quotation has type <literal>Q Exp</literal>.</para></listitem>
6212 <listitem><para> <literal>[d| ... |]</literal>, where the "..." is a list of top-level declarations;
6213 the quotation has type <literal>Q [Dec]</literal>.</para></listitem>
6214 <listitem><para> <literal>[t| ... |]</literal>, where the "..." is a type;
6215 the quotation has type <literal>Q Type</literal>.</para></listitem>
6216 <listitem><para> <literal>[p| ... |]</literal>, where the "..." is a pattern;
6217 the quotation has type <literal>Q Pat</literal>.</para></listitem>
6218 </itemizedlist></para></listitem>
6221 A quasi-quotation can appear in either a pattern context or an
6222 expression context and is also written in Oxford brackets:
6224 <listitem><para> <literal>[<replaceable>varid</replaceable>| ... |]</literal>,
6225 where the "..." is an arbitrary string; a full description of the
6226 quasi-quotation facility is given in <xref linkend="th-quasiquotation"/>.</para></listitem>
6227 </itemizedlist></para></listitem>
6230 A name can be quoted with either one or two prefix single quotes:
6232 <listitem><para> <literal>'f</literal> has type <literal>Name</literal>, and names the function <literal>f</literal>.
6233 Similarly <literal>'C</literal> has type <literal>Name</literal> and names the data constructor <literal>C</literal>.
6234 In general <literal>'</literal><replaceable>thing</replaceable> interprets <replaceable>thing</replaceable> in an expression context.
6236 <listitem><para> <literal>''T</literal> has type <literal>Name</literal>, and names the type constructor <literal>T</literal>.
6237 That is, <literal>''</literal><replaceable>thing</replaceable> interprets <replaceable>thing</replaceable> in a type context.
6240 These <literal>Names</literal> can be used to construct Template Haskell expressions, patterns, declarations etc. They
6241 may also be given as an argument to the <literal>reify</literal> function.
6245 <listitem><para> You may omit the <literal>$(...)</literal> in a top-level declaration splice.
6246 Simply writing an expression (rather than a declaration) implies a splice. For example, you can write
6253 $(deriveStuff 'f) -- Uses the $(...) notation
6257 deriveStuff 'g -- Omits the $(...)
6261 This abbreviation makes top-level declaration slices quieter and less intimidating.
6266 (Compared to the original paper, there are many differences of detail.
6267 The syntax for a declaration splice uses "<literal>$</literal>" not "<literal>splice</literal>".
6268 The type of the enclosed expression must be <literal>Q [Dec]</literal>, not <literal>[Q Dec]</literal>.
6269 Pattern splices and quotations are not implemented.)
6273 <sect2> <title> Using Template Haskell </title>
6277 The data types and monadic constructor functions for Template Haskell are in the library
6278 <literal>Language.Haskell.THSyntax</literal>.
6282 You can only run a function at compile time if it is imported from another module. That is,
6283 you can't define a function in a module, and call it from within a splice in the same module.
6284 (It would make sense to do so, but it's hard to implement.)
6288 You can only run a function at compile time if it is imported
6289 from another module <emphasis>that is not part of a mutually-recursive group of modules
6290 that includes the module currently being compiled</emphasis>. Furthermore, all of the modules of
6291 the mutually-recursive group must be reachable by non-SOURCE imports from the module where the
6292 splice is to be run.</para>
6294 For example, when compiling module A,
6295 you can only run Template Haskell functions imported from B if B does not import A (directly or indirectly).
6296 The reason should be clear: to run B we must compile and run A, but we are currently type-checking A.
6300 The flag <literal>-ddump-splices</literal> shows the expansion of all top-level splices as they happen.
6303 If you are building GHC from source, you need at least a stage-2 bootstrap compiler to
6304 run Template Haskell. A stage-1 compiler will reject the TH constructs. Reason: TH
6305 compiles and runs a program, and then looks at the result. So it's important that
6306 the program it compiles produces results whose representations are identical to
6307 those of the compiler itself.
6311 <para> Template Haskell works in any mode (<literal>--make</literal>, <literal>--interactive</literal>,
6312 or file-at-a-time). There used to be a restriction to the former two, but that restriction
6317 <sect2 id="th-example"> <title> A Template Haskell Worked Example </title>
6318 <para>To help you get over the confidence barrier, try out this skeletal worked example.
6319 First cut and paste the two modules below into "Main.hs" and "Printf.hs":</para>
6326 -- Import our template "pr"
6327 import Printf ( pr )
6329 -- The splice operator $ takes the Haskell source code
6330 -- generated at compile time by "pr" and splices it into
6331 -- the argument of "putStrLn".
6332 main = putStrLn ( $(pr "Hello") )
6338 -- Skeletal printf from the paper.
6339 -- It needs to be in a separate module to the one where
6340 -- you intend to use it.
6342 -- Import some Template Haskell syntax
6343 import Language.Haskell.TH
6345 -- Describe a format string
6346 data Format = D | S | L String
6348 -- Parse a format string. This is left largely to you
6349 -- as we are here interested in building our first ever
6350 -- Template Haskell program and not in building printf.
6351 parse :: String -> [Format]
6354 -- Generate Haskell source code from a parsed representation
6355 -- of the format string. This code will be spliced into
6356 -- the module which calls "pr", at compile time.
6357 gen :: [Format] -> Q Exp
6358 gen [D] = [| \n -> show n |]
6359 gen [S] = [| \s -> s |]
6360 gen [L s] = stringE s
6362 -- Here we generate the Haskell code for the splice
6363 -- from an input format string.
6364 pr :: String -> Q Exp
6365 pr s = gen (parse s)
6368 <para>Now run the compiler (here we are a Cygwin prompt on Windows):
6371 $ ghc --make -XTemplateHaskell main.hs -o main.exe
6374 <para>Run "main.exe" and here is your output:</para>
6384 <title>Using Template Haskell with Profiling</title>
6385 <indexterm><primary>profiling</primary><secondary>with Template Haskell</secondary></indexterm>
6387 <para>Template Haskell relies on GHC's built-in bytecode compiler and
6388 interpreter to run the splice expressions. The bytecode interpreter
6389 runs the compiled expression on top of the same runtime on which GHC
6390 itself is running; this means that the compiled code referred to by
6391 the interpreted expression must be compatible with this runtime, and
6392 in particular this means that object code that is compiled for
6393 profiling <emphasis>cannot</emphasis> be loaded and used by a splice
6394 expression, because profiled object code is only compatible with the
6395 profiling version of the runtime.</para>
6397 <para>This causes difficulties if you have a multi-module program
6398 containing Template Haskell code and you need to compile it for
6399 profiling, because GHC cannot load the profiled object code and use it
6400 when executing the splices. Fortunately GHC provides a workaround.
6401 The basic idea is to compile the program twice:</para>
6405 <para>Compile the program or library first the normal way, without
6406 <option>-prof</option><indexterm><primary><option>-prof</option></primary></indexterm>.</para>
6409 <para>Then compile it again with <option>-prof</option>, and
6410 additionally use <option>-osuf
6411 p_o</option><indexterm><primary><option>-osuf</option></primary></indexterm>
6412 to name the object files differently (you can choose any suffix
6413 that isn't the normal object suffix here). GHC will automatically
6414 load the object files built in the first step when executing splice
6415 expressions. If you omit the <option>-osuf</option> flag when
6416 building with <option>-prof</option> and Template Haskell is used,
6417 GHC will emit an error message. </para>
6422 <sect2 id="th-quasiquotation"> <title> Template Haskell Quasi-quotation </title>
6423 <para>Quasi-quotation allows patterns and expressions to be written using
6424 programmer-defined concrete syntax; the motivation behind the extension and
6425 several examples are documented in
6426 "<ulink url="http://www.eecs.harvard.edu/~mainland/ghc-quasiquoting/">Why It's
6427 Nice to be Quoted: Quasiquoting for Haskell</ulink>" (Proc Haskell Workshop
6428 2007). The example below shows how to write a quasiquoter for a simple
6429 expression language.</para>
6431 Here are the salient features
6434 A quasi-quote has the form
6435 <literal>[<replaceable>quoter</replaceable>| <replaceable>string</replaceable> |]</literal>.
6438 The <replaceable>quoter</replaceable> must be the (unqualified) name of an imported
6439 quoter; it cannot be an arbitrary expression.
6442 The <replaceable>quoter</replaceable> cannot be "<literal>e</literal>",
6443 "<literal>t</literal>", "<literal>d</literal>", or "<literal>p</literal>", since
6444 those overlap with Template Haskell quotations.
6447 There must be no spaces in the token
6448 <literal>[<replaceable>quoter</replaceable>|</literal>.
6451 The quoted <replaceable>string</replaceable>
6452 can be arbitrary, and may contain newlines.
6458 A quasiquote may appear in place of
6460 <listitem><para>An expression</para></listitem>
6461 <listitem><para>A pattern</para></listitem>
6462 <listitem><para>A type</para></listitem>
6463 <listitem><para>A top-level declaration</para></listitem>
6465 (Only the first two are described in the paper.)
6469 A quoter is a value of type <literal>Language.Haskell.TH.Quote.QuasiQuoter</literal>,
6470 which is defined thus:
6472 data QuasiQuoter = QuasiQuoter { quoteExp :: String -> Q Exp,
6473 quotePat :: String -> Q Pat,
6474 quoteType :: String -> Q Type,
6475 quoteDec :: String -> Q [Dec] }
6477 That is, a quoter is a tuple of four parsers, one for each of the contexts
6478 in which a quasi-quote can occur.
6481 A quasi-quote is expanded by applying the appropriate parser to the string
6482 enclosed by the Oxford brackets. The context of the quasi-quote (expression, pattern,
6483 type, declaration) determines which of the parsers is called.
6488 The example below shows quasi-quotation in action. The quoter <literal>expr</literal>
6489 is bound to a value of type <literal>QuasiQuoter</literal> defined in module <literal>Expr</literal>.
6490 The example makes use of an antiquoted
6491 variable <literal>n</literal>, indicated by the syntax <literal>'int:n</literal>
6492 (this syntax for anti-quotation was defined by the parser's
6493 author, <emphasis>not</emphasis> by GHC). This binds <literal>n</literal> to the
6494 integer value argument of the constructor <literal>IntExpr</literal> when
6495 pattern matching. Please see the referenced paper for further details regarding
6496 anti-quotation as well as the description of a technique that uses SYB to
6497 leverage a single parser of type <literal>String -> a</literal> to generate both
6498 an expression parser that returns a value of type <literal>Q Exp</literal> and a
6499 pattern parser that returns a value of type <literal>Q Pat</literal>.
6503 Quasiquoters must obey the same stage restrictions as Template Haskell, e.g., in
6504 the example, <literal>expr</literal> cannot be defined
6505 in <literal>Main.hs</literal> where it is used, but must be imported.
6509 {- ------------- file Main.hs --------------- -}
6515 main = do { print $ eval [expr|1 + 2|]
6517 { [expr|'int:n|] -> print n
6523 {- ------------- file Expr.hs --------------- -}
6526 import qualified Language.Haskell.TH as TH
6527 import Language.Haskell.TH.Quote
6529 data Expr = IntExpr Integer
6530 | AntiIntExpr String
6531 | BinopExpr BinOp Expr Expr
6533 deriving(Show, Typeable, Data)
6539 deriving(Show, Typeable, Data)
6541 eval :: Expr -> Integer
6542 eval (IntExpr n) = n
6543 eval (BinopExpr op x y) = (opToFun op) (eval x) (eval y)
6550 expr = QuasiQuoter { quoteExp = parseExprExp, quotePat = parseExprPat }
6552 -- Parse an Expr, returning its representation as
6553 -- either a Q Exp or a Q Pat. See the referenced paper
6554 -- for how to use SYB to do this by writing a single
6555 -- parser of type String -> Expr instead of two
6556 -- separate parsers.
6558 parseExprExp :: String -> Q Exp
6561 parseExprPat :: String -> Q Pat
6565 <para>Now run the compiler:
6567 $ ghc --make -XQuasiQuotes Main.hs -o main
6571 <para>Run "main" and here is your output:
6582 <!-- ===================== Arrow notation =================== -->
6584 <sect1 id="arrow-notation">
6585 <title>Arrow notation
6588 <para>Arrows are a generalization of monads introduced by John Hughes.
6589 For more details, see
6594 “Generalising Monads to Arrows”,
6595 John Hughes, in <citetitle>Science of Computer Programming</citetitle> 37,
6596 pp67–111, May 2000.
6597 The paper that introduced arrows: a friendly introduction, motivated with
6598 programming examples.
6604 “<ulink url="http://www.soi.city.ac.uk/~ross/papers/notation.html">A New Notation for Arrows</ulink>”,
6605 Ross Paterson, in <citetitle>ICFP</citetitle>, Sep 2001.
6606 Introduced the notation described here.
6612 “<ulink url="http://www.soi.city.ac.uk/~ross/papers/fop.html">Arrows and Computation</ulink>”,
6613 Ross Paterson, in <citetitle>The Fun of Programming</citetitle>,
6620 “<ulink url="http://www.cs.chalmers.se/~rjmh/afp-arrows.pdf">Programming with Arrows</ulink>”,
6621 John Hughes, in <citetitle>5th International Summer School on
6622 Advanced Functional Programming</citetitle>,
6623 <citetitle>Lecture Notes in Computer Science</citetitle> vol. 3622,
6625 This paper includes another introduction to the notation,
6626 with practical examples.
6632 “<ulink url="http://www.haskell.org/ghc/docs/papers/arrow-rules.pdf">Type and Translation Rules for Arrow Notation in GHC</ulink>”,
6633 Ross Paterson and Simon Peyton Jones, September 16, 2004.
6634 A terse enumeration of the formal rules used
6635 (extracted from comments in the source code).
6641 The arrows web page at
6642 <ulink url="http://www.haskell.org/arrows/"><literal>http://www.haskell.org/arrows/</literal></ulink>.
6647 With the <option>-XArrows</option> flag, GHC supports the arrow
6648 notation described in the second of these papers,
6649 translating it using combinators from the
6650 <ulink url="&libraryBaseLocation;/Control-Arrow.html"><literal>Control.Arrow</literal></ulink>
6652 What follows is a brief introduction to the notation;
6653 it won't make much sense unless you've read Hughes's paper.
6656 <para>The extension adds a new kind of expression for defining arrows:
6658 <replaceable>exp</replaceable><superscript>10</superscript> ::= ...
6659 | proc <replaceable>apat</replaceable> -> <replaceable>cmd</replaceable>
6661 where <literal>proc</literal> is a new keyword.
6662 The variables of the pattern are bound in the body of the
6663 <literal>proc</literal>-expression,
6664 which is a new sort of thing called a <firstterm>command</firstterm>.
6665 The syntax of commands is as follows:
6667 <replaceable>cmd</replaceable> ::= <replaceable>exp</replaceable><superscript>10</superscript> -< <replaceable>exp</replaceable>
6668 | <replaceable>exp</replaceable><superscript>10</superscript> -<< <replaceable>exp</replaceable>
6669 | <replaceable>cmd</replaceable><superscript>0</superscript>
6671 with <replaceable>cmd</replaceable><superscript>0</superscript> up to
6672 <replaceable>cmd</replaceable><superscript>9</superscript> defined using
6673 infix operators as for expressions, and
6675 <replaceable>cmd</replaceable><superscript>10</superscript> ::= \ <replaceable>apat</replaceable> ... <replaceable>apat</replaceable> -> <replaceable>cmd</replaceable>
6676 | let <replaceable>decls</replaceable> in <replaceable>cmd</replaceable>
6677 | if <replaceable>exp</replaceable> then <replaceable>cmd</replaceable> else <replaceable>cmd</replaceable>
6678 | case <replaceable>exp</replaceable> of { <replaceable>calts</replaceable> }
6679 | do { <replaceable>cstmt</replaceable> ; ... <replaceable>cstmt</replaceable> ; <replaceable>cmd</replaceable> }
6680 | <replaceable>fcmd</replaceable>
6682 <replaceable>fcmd</replaceable> ::= <replaceable>fcmd</replaceable> <replaceable>aexp</replaceable>
6683 | ( <replaceable>cmd</replaceable> )
6684 | (| <replaceable>aexp</replaceable> <replaceable>cmd</replaceable> ... <replaceable>cmd</replaceable> |)
6686 <replaceable>cstmt</replaceable> ::= let <replaceable>decls</replaceable>
6687 | <replaceable>pat</replaceable> <- <replaceable>cmd</replaceable>
6688 | rec { <replaceable>cstmt</replaceable> ; ... <replaceable>cstmt</replaceable> [;] }
6689 | <replaceable>cmd</replaceable>
6691 where <replaceable>calts</replaceable> are like <replaceable>alts</replaceable>
6692 except that the bodies are commands instead of expressions.
6696 Commands produce values, but (like monadic computations)
6697 may yield more than one value,
6698 or none, and may do other things as well.
6699 For the most part, familiarity with monadic notation is a good guide to
6701 However the values of expressions, even monadic ones,
6702 are determined by the values of the variables they contain;
6703 this is not necessarily the case for commands.
6707 A simple example of the new notation is the expression
6709 proc x -> f -< x+1
6711 We call this a <firstterm>procedure</firstterm> or
6712 <firstterm>arrow abstraction</firstterm>.
6713 As with a lambda expression, the variable <literal>x</literal>
6714 is a new variable bound within the <literal>proc</literal>-expression.
6715 It refers to the input to the arrow.
6716 In the above example, <literal>-<</literal> is not an identifier but an
6717 new reserved symbol used for building commands from an expression of arrow
6718 type and an expression to be fed as input to that arrow.
6719 (The weird look will make more sense later.)
6720 It may be read as analogue of application for arrows.
6721 The above example is equivalent to the Haskell expression
6723 arr (\ x -> x+1) >>> f
6725 That would make no sense if the expression to the left of
6726 <literal>-<</literal> involves the bound variable <literal>x</literal>.
6727 More generally, the expression to the left of <literal>-<</literal>
6728 may not involve any <firstterm>local variable</firstterm>,
6729 i.e. a variable bound in the current arrow abstraction.
6730 For such a situation there is a variant <literal>-<<</literal>, as in
6732 proc x -> f x -<< x+1
6734 which is equivalent to
6736 arr (\ x -> (f x, x+1)) >>> app
6738 so in this case the arrow must belong to the <literal>ArrowApply</literal>
6740 Such an arrow is equivalent to a monad, so if you're using this form
6741 you may find a monadic formulation more convenient.
6745 <title>do-notation for commands</title>
6748 Another form of command is a form of <literal>do</literal>-notation.
6749 For example, you can write
6758 You can read this much like ordinary <literal>do</literal>-notation,
6759 but with commands in place of monadic expressions.
6760 The first line sends the value of <literal>x+1</literal> as an input to
6761 the arrow <literal>f</literal>, and matches its output against
6762 <literal>y</literal>.
6763 In the next line, the output is discarded.
6764 The arrow <function>returnA</function> is defined in the
6765 <ulink url="&libraryBaseLocation;/Control-Arrow.html"><literal>Control.Arrow</literal></ulink>
6766 module as <literal>arr id</literal>.
6767 The above example is treated as an abbreviation for
6769 arr (\ x -> (x, x)) >>>
6770 first (arr (\ x -> x+1) >>> f) >>>
6771 arr (\ (y, x) -> (y, (x, y))) >>>
6772 first (arr (\ y -> 2*y) >>> g) >>>
6774 arr (\ (x, y) -> let z = x+y in ((x, z), z)) >>>
6775 first (arr (\ (x, z) -> x*z) >>> h) >>>
6776 arr (\ (t, z) -> t+z) >>>
6779 Note that variables not used later in the composition are projected out.
6780 After simplification using rewrite rules (see <xref linkend="rewrite-rules"/>)
6782 <ulink url="&libraryBaseLocation;/Control-Arrow.html"><literal>Control.Arrow</literal></ulink>
6783 module, this reduces to
6785 arr (\ x -> (x+1, x)) >>>
6787 arr (\ (y, x) -> (2*y, (x, y))) >>>
6789 arr (\ (_, (x, y)) -> let z = x+y in (x*z, z)) >>>
6791 arr (\ (t, z) -> t+z)
6793 which is what you might have written by hand.
6794 With arrow notation, GHC keeps track of all those tuples of variables for you.
6798 Note that although the above translation suggests that
6799 <literal>let</literal>-bound variables like <literal>z</literal> must be
6800 monomorphic, the actual translation produces Core,
6801 so polymorphic variables are allowed.
6805 It's also possible to have mutually recursive bindings,
6806 using the new <literal>rec</literal> keyword, as in the following example:
6808 counter :: ArrowCircuit a => a Bool Int
6809 counter = proc reset -> do
6810 rec output <- returnA -< if reset then 0 else next
6811 next <- delay 0 -< output+1
6812 returnA -< output
6814 The translation of such forms uses the <function>loop</function> combinator,
6815 so the arrow concerned must belong to the <literal>ArrowLoop</literal> class.
6821 <title>Conditional commands</title>
6824 In the previous example, we used a conditional expression to construct the
6826 Sometimes we want to conditionally execute different commands, as in
6833 which is translated to
6835 arr (\ (x,y) -> if f x y then Left x else Right y) >>>
6836 (arr (\x -> x+1) >>> f) ||| (arr (\y -> y+2) >>> g)
6838 Since the translation uses <function>|||</function>,
6839 the arrow concerned must belong to the <literal>ArrowChoice</literal> class.
6843 There are also <literal>case</literal> commands, like
6849 y <- h -< (x1, x2)
6853 The syntax is the same as for <literal>case</literal> expressions,
6854 except that the bodies of the alternatives are commands rather than expressions.
6855 The translation is similar to that of <literal>if</literal> commands.
6861 <title>Defining your own control structures</title>
6864 As we're seen, arrow notation provides constructs,
6865 modelled on those for expressions,
6866 for sequencing, value recursion and conditionals.
6867 But suitable combinators,
6868 which you can define in ordinary Haskell,
6869 may also be used to build new commands out of existing ones.
6870 The basic idea is that a command defines an arrow from environments to values.
6871 These environments assign values to the free local variables of the command.
6872 Thus combinators that produce arrows from arrows
6873 may also be used to build commands from commands.
6874 For example, the <literal>ArrowChoice</literal> class includes a combinator
6876 ArrowChoice a => (<+>) :: a e c -> a e c -> a e c
6878 so we can use it to build commands:
6880 expr' = proc x -> do
6883 symbol Plus -< ()
6884 y <- term -< ()
6887 symbol Minus -< ()
6888 y <- term -< ()
6891 (The <literal>do</literal> on the first line is needed to prevent the first
6892 <literal><+> ...</literal> from being interpreted as part of the
6893 expression on the previous line.)
6894 This is equivalent to
6896 expr' = (proc x -> returnA -< x)
6897 <+> (proc x -> do
6898 symbol Plus -< ()
6899 y <- term -< ()
6901 <+> (proc x -> do
6902 symbol Minus -< ()
6903 y <- term -< ()
6906 It is essential that this operator be polymorphic in <literal>e</literal>
6907 (representing the environment input to the command
6908 and thence to its subcommands)
6909 and satisfy the corresponding naturality property
6911 arr k >>> (f <+> g) = (arr k >>> f) <+> (arr k >>> g)
6913 at least for strict <literal>k</literal>.
6914 (This should be automatic if you're not using <function>seq</function>.)
6915 This ensures that environments seen by the subcommands are environments
6916 of the whole command,
6917 and also allows the translation to safely trim these environments.
6918 The operator must also not use any variable defined within the current
6923 We could define our own operator
6925 untilA :: ArrowChoice a => a e () -> a e Bool -> a e ()
6926 untilA body cond = proc x ->
6927 b <- cond -< x
6928 if b then returnA -< ()
6931 untilA body cond -< x
6933 and use it in the same way.
6934 Of course this infix syntax only makes sense for binary operators;
6935 there is also a more general syntax involving special brackets:
6939 (|untilA (increment -< x+y) (within 0.5 -< x)|)
6946 <title>Primitive constructs</title>
6949 Some operators will need to pass additional inputs to their subcommands.
6950 For example, in an arrow type supporting exceptions,
6951 the operator that attaches an exception handler will wish to pass the
6952 exception that occurred to the handler.
6953 Such an operator might have a type
6955 handleA :: ... => a e c -> a (e,Ex) c -> a e c
6957 where <literal>Ex</literal> is the type of exceptions handled.
6958 You could then use this with arrow notation by writing a command
6960 body `handleA` \ ex -> handler
6962 so that if an exception is raised in the command <literal>body</literal>,
6963 the variable <literal>ex</literal> is bound to the value of the exception
6964 and the command <literal>handler</literal>,
6965 which typically refers to <literal>ex</literal>, is entered.
6966 Though the syntax here looks like a functional lambda,
6967 we are talking about commands, and something different is going on.
6968 The input to the arrow represented by a command consists of values for
6969 the free local variables in the command, plus a stack of anonymous values.
6970 In all the prior examples, this stack was empty.
6971 In the second argument to <function>handleA</function>,
6972 this stack consists of one value, the value of the exception.
6973 The command form of lambda merely gives this value a name.
6978 the values on the stack are paired to the right of the environment.
6979 So operators like <function>handleA</function> that pass
6980 extra inputs to their subcommands can be designed for use with the notation
6981 by pairing the values with the environment in this way.
6982 More precisely, the type of each argument of the operator (and its result)
6983 should have the form
6985 a (...(e,t1), ... tn) t
6987 where <replaceable>e</replaceable> is a polymorphic variable
6988 (representing the environment)
6989 and <replaceable>ti</replaceable> are the types of the values on the stack,
6990 with <replaceable>t1</replaceable> being the <quote>top</quote>.
6991 The polymorphic variable <replaceable>e</replaceable> must not occur in
6992 <replaceable>a</replaceable>, <replaceable>ti</replaceable> or
6993 <replaceable>t</replaceable>.
6994 However the arrows involved need not be the same.
6995 Here are some more examples of suitable operators:
6997 bracketA :: ... => a e b -> a (e,b) c -> a (e,c) d -> a e d
6998 runReader :: ... => a e c -> a' (e,State) c
6999 runState :: ... => a e c -> a' (e,State) (c,State)
7001 We can supply the extra input required by commands built with the last two
7002 by applying them to ordinary expressions, as in
7006 (|runReader (do { ... })|) s
7008 which adds <literal>s</literal> to the stack of inputs to the command
7009 built using <function>runReader</function>.
7013 The command versions of lambda abstraction and application are analogous to
7014 the expression versions.
7015 In particular, the beta and eta rules describe equivalences of commands.
7016 These three features (operators, lambda abstraction and application)
7017 are the core of the notation; everything else can be built using them,
7018 though the results would be somewhat clumsy.
7019 For example, we could simulate <literal>do</literal>-notation by defining
7021 bind :: Arrow a => a e b -> a (e,b) c -> a e c
7022 u `bind` f = returnA &&& u >>> f
7024 bind_ :: Arrow a => a e b -> a e c -> a e c
7025 u `bind_` f = u `bind` (arr fst >>> f)
7027 We could simulate <literal>if</literal> by defining
7029 cond :: ArrowChoice a => a e b -> a e b -> a (e,Bool) b
7030 cond f g = arr (\ (e,b) -> if b then Left e else Right e) >>> f ||| g
7037 <title>Differences with the paper</title>
7042 <para>Instead of a single form of arrow application (arrow tail) with two
7043 translations, the implementation provides two forms
7044 <quote><literal>-<</literal></quote> (first-order)
7045 and <quote><literal>-<<</literal></quote> (higher-order).
7050 <para>User-defined operators are flagged with banana brackets instead of
7051 a new <literal>form</literal> keyword.
7060 <title>Portability</title>
7063 Although only GHC implements arrow notation directly,
7064 there is also a preprocessor
7066 <ulink url="http://www.haskell.org/arrows/">arrows web page</ulink>)
7067 that translates arrow notation into Haskell 98
7068 for use with other Haskell systems.
7069 You would still want to check arrow programs with GHC;
7070 tracing type errors in the preprocessor output is not easy.
7071 Modules intended for both GHC and the preprocessor must observe some
7072 additional restrictions:
7077 The module must import
7078 <ulink url="&libraryBaseLocation;/Control-Arrow.html"><literal>Control.Arrow</literal></ulink>.
7084 The preprocessor cannot cope with other Haskell extensions.
7085 These would have to go in separate modules.
7091 Because the preprocessor targets Haskell (rather than Core),
7092 <literal>let</literal>-bound variables are monomorphic.
7103 <!-- ==================== BANG PATTERNS ================= -->
7105 <sect1 id="bang-patterns">
7106 <title>Bang patterns
7107 <indexterm><primary>Bang patterns</primary></indexterm>
7109 <para>GHC supports an extension of pattern matching called <emphasis>bang
7110 patterns</emphasis>, written <literal>!<replaceable>pat</replaceable></literal>.
7111 Bang patterns are under consideration for Haskell Prime.
7113 url="http://hackage.haskell.org/trac/haskell-prime/wiki/BangPatterns">Haskell
7114 prime feature description</ulink> contains more discussion and examples
7115 than the material below.
7118 The key change is the addition of a new rule to the
7119 <ulink url="http://haskell.org/onlinereport/exps.html#sect3.17.2">semantics of pattern matching in the Haskell 98 report</ulink>.
7120 Add new bullet 10, saying: Matching the pattern <literal>!</literal><replaceable>pat</replaceable>
7121 against a value <replaceable>v</replaceable> behaves as follows:
7123 <listitem><para>if <replaceable>v</replaceable> is bottom, the match diverges</para></listitem>
7124 <listitem><para>otherwise, <replaceable>pat</replaceable> is matched against <replaceable>v</replaceable> </para></listitem>
7128 Bang patterns are enabled by the flag <option>-XBangPatterns</option>.
7131 <sect2 id="bang-patterns-informal">
7132 <title>Informal description of bang patterns
7135 The main idea is to add a single new production to the syntax of patterns:
7139 Matching an expression <literal>e</literal> against a pattern <literal>!p</literal> is done by first
7140 evaluating <literal>e</literal> (to WHNF) and then matching the result against <literal>p</literal>.
7145 This definition makes <literal>f1</literal> is strict in <literal>x</literal>,
7146 whereas without the bang it would be lazy.
7147 Bang patterns can be nested of course:
7151 Here, <literal>f2</literal> is strict in <literal>x</literal> but not in
7152 <literal>y</literal>.
7153 A bang only really has an effect if it precedes a variable or wild-card pattern:
7158 Here, <literal>f3</literal> and <literal>f4</literal> are identical;
7159 putting a bang before a pattern that
7160 forces evaluation anyway does nothing.
7163 There is one (apparent) exception to this general rule that a bang only
7164 makes a difference when it precedes a variable or wild-card: a bang at the
7165 top level of a <literal>let</literal> or <literal>where</literal>
7166 binding makes the binding strict, regardless of the pattern.
7167 (We say "apparent" exception because the Right Way to think of it is that the bang
7168 at the top of a binding is not part of the <emphasis>pattern</emphasis>; rather it
7169 is part of the syntax of the <emphasis>binding</emphasis>,
7170 creating a "bang-pattern binding".)
7175 is a bang-pattern binding. Operationally, it behaves just like a case expression:
7177 case e of [x,y] -> b
7179 Like a case expression, a bang-pattern binding must be non-recursive, and
7182 However, <emphasis>nested</emphasis> bangs in a pattern binding behave uniformly with all other forms of
7183 pattern matching. For example
7185 let (!x,[y]) = e in b
7187 is equivalent to this:
7189 let { t = case e of (x,[y]) -> x `seq` (x,y)
7194 The binding is lazy, but when either <literal>x</literal> or <literal>y</literal> is
7195 evaluated by <literal>b</literal> the entire pattern is matched, including forcing the
7196 evaluation of <literal>x</literal>.
7199 Bang patterns work in <literal>case</literal> expressions too, of course:
7201 g5 x = let y = f x in body
7202 g6 x = case f x of { y -> body }
7203 g7 x = case f x of { !y -> body }
7205 The functions <literal>g5</literal> and <literal>g6</literal> mean exactly the same thing.
7206 But <literal>g7</literal> evaluates <literal>(f x)</literal>, binds <literal>y</literal> to the
7207 result, and then evaluates <literal>body</literal>.
7212 <sect2 id="bang-patterns-sem">
7213 <title>Syntax and semantics
7217 We add a single new production to the syntax of patterns:
7221 There is one problem with syntactic ambiguity. Consider:
7225 Is this a definition of the infix function "<literal>(!)</literal>",
7226 or of the "<literal>f</literal>" with a bang pattern? GHC resolves this
7227 ambiguity in favour of the latter. If you want to define
7228 <literal>(!)</literal> with bang-patterns enabled, you have to do so using
7233 The semantics of Haskell pattern matching is described in <ulink
7234 url="http://www.haskell.org/onlinereport/exps.html#sect3.17.2">
7235 Section 3.17.2</ulink> of the Haskell Report. To this description add
7236 one extra item 10, saying:
7237 <itemizedlist><listitem><para>Matching
7238 the pattern <literal>!pat</literal> against a value <literal>v</literal> behaves as follows:
7239 <itemizedlist><listitem><para>if <literal>v</literal> is bottom, the match diverges</para></listitem>
7240 <listitem><para>otherwise, <literal>pat</literal> is matched against
7241 <literal>v</literal></para></listitem>
7243 </para></listitem></itemizedlist>
7244 Similarly, in Figure 4 of <ulink url="http://www.haskell.org/onlinereport/exps.html#sect3.17.3">
7245 Section 3.17.3</ulink>, add a new case (t):
7247 case v of { !pat -> e; _ -> e' }
7248 = v `seq` case v of { pat -> e; _ -> e' }
7251 That leaves let expressions, whose translation is given in
7252 <ulink url="http://www.haskell.org/onlinereport/exps.html#sect3.12">Section
7254 of the Haskell Report.
7255 In the translation box, first apply
7256 the following transformation: for each pattern <literal>pi</literal> that is of
7257 form <literal>!qi = ei</literal>, transform it to <literal>(xi,!qi) = ((),ei)</literal>, and and replace <literal>e0</literal>
7258 by <literal>(xi `seq` e0)</literal>. Then, when none of the left-hand-side patterns
7259 have a bang at the top, apply the rules in the existing box.
7261 <para>The effect of the let rule is to force complete matching of the pattern
7262 <literal>qi</literal> before evaluation of the body is begun. The bang is
7263 retained in the translated form in case <literal>qi</literal> is a variable,
7271 The let-binding can be recursive. However, it is much more common for
7272 the let-binding to be non-recursive, in which case the following law holds:
7273 <literal>(let !p = rhs in body)</literal>
7275 <literal>(case rhs of !p -> body)</literal>
7278 A pattern with a bang at the outermost level is not allowed at the top level of
7284 <!-- ==================== ASSERTIONS ================= -->
7286 <sect1 id="assertions">
7288 <indexterm><primary>Assertions</primary></indexterm>
7292 If you want to make use of assertions in your standard Haskell code, you
7293 could define a function like the following:
7299 assert :: Bool -> a -> a
7300 assert False x = error "assertion failed!"
7307 which works, but gives you back a less than useful error message --
7308 an assertion failed, but which and where?
7312 One way out is to define an extended <function>assert</function> function which also
7313 takes a descriptive string to include in the error message and
7314 perhaps combine this with the use of a pre-processor which inserts
7315 the source location where <function>assert</function> was used.
7319 Ghc offers a helping hand here, doing all of this for you. For every
7320 use of <function>assert</function> in the user's source:
7326 kelvinToC :: Double -> Double
7327 kelvinToC k = assert (k >= 0.0) (k+273.15)
7333 Ghc will rewrite this to also include the source location where the
7340 assert pred val ==> assertError "Main.hs|15" pred val
7346 The rewrite is only performed by the compiler when it spots
7347 applications of <function>Control.Exception.assert</function>, so you
7348 can still define and use your own versions of
7349 <function>assert</function>, should you so wish. If not, import
7350 <literal>Control.Exception</literal> to make use
7351 <function>assert</function> in your code.
7355 GHC ignores assertions when optimisation is turned on with the
7356 <option>-O</option><indexterm><primary><option>-O</option></primary></indexterm> flag. That is, expressions of the form
7357 <literal>assert pred e</literal> will be rewritten to
7358 <literal>e</literal>. You can also disable assertions using the
7359 <option>-fignore-asserts</option>
7360 option<indexterm><primary><option>-fignore-asserts</option></primary>
7361 </indexterm>.</para>
7364 Assertion failures can be caught, see the documentation for the
7365 <literal>Control.Exception</literal> library for the details.
7371 <!-- =============================== PRAGMAS =========================== -->
7373 <sect1 id="pragmas">
7374 <title>Pragmas</title>
7376 <indexterm><primary>pragma</primary></indexterm>
7378 <para>GHC supports several pragmas, or instructions to the
7379 compiler placed in the source code. Pragmas don't normally affect
7380 the meaning of the program, but they might affect the efficiency
7381 of the generated code.</para>
7383 <para>Pragmas all take the form
7385 <literal>{-# <replaceable>word</replaceable> ... #-}</literal>
7387 where <replaceable>word</replaceable> indicates the type of
7388 pragma, and is followed optionally by information specific to that
7389 type of pragma. Case is ignored in
7390 <replaceable>word</replaceable>. The various values for
7391 <replaceable>word</replaceable> that GHC understands are described
7392 in the following sections; any pragma encountered with an
7393 unrecognised <replaceable>word</replaceable> is
7394 ignored. The layout rule applies in pragmas, so the closing <literal>#-}</literal>
7395 should start in a column to the right of the opening <literal>{-#</literal>. </para>
7397 <para>Certain pragmas are <emphasis>file-header pragmas</emphasis>:
7401 pragma must precede the <literal>module</literal> keyword in the file.
7404 There can be as many file-header pragmas as you please, and they can be
7405 preceded or followed by comments.
7408 File-header pragmas are read once only, before
7409 pre-processing the file (e.g. with cpp).
7412 The file-header pragmas are: <literal>{-# LANGUAGE #-}</literal>,
7413 <literal>{-# OPTIONS_GHC #-}</literal>, and
7414 <literal>{-# INCLUDE #-}</literal>.
7419 <sect2 id="language-pragma">
7420 <title>LANGUAGE pragma</title>
7422 <indexterm><primary>LANGUAGE</primary><secondary>pragma</secondary></indexterm>
7423 <indexterm><primary>pragma</primary><secondary>LANGUAGE</secondary></indexterm>
7425 <para>The <literal>LANGUAGE</literal> pragma allows language extensions to be enabled
7427 It is the intention that all Haskell compilers support the
7428 <literal>LANGUAGE</literal> pragma with the same syntax, although not
7429 all extensions are supported by all compilers, of
7430 course. The <literal>LANGUAGE</literal> pragma should be used instead
7431 of <literal>OPTIONS_GHC</literal>, if possible.</para>
7433 <para>For example, to enable the FFI and preprocessing with CPP:</para>
7435 <programlisting>{-# LANGUAGE ForeignFunctionInterface, CPP #-}</programlisting>
7437 <para><literal>LANGUAGE</literal> is a file-header pragma (see <xref linkend="pragmas"/>).</para>
7439 <para>Every language extension can also be turned into a command-line flag
7440 by prefixing it with "<literal>-X</literal>"; for example <option>-XForeignFunctionInterface</option>.
7441 (Similarly, all "<literal>-X</literal>" flags can be written as <literal>LANGUAGE</literal> pragmas.
7444 <para>A list of all supported language extensions can be obtained by invoking
7445 <literal>ghc --supported-extensions</literal> (see <xref linkend="modes"/>).</para>
7447 <para>Any extension from the <literal>Extension</literal> type defined in
7449 url="&libraryCabalLocation;/Language-Haskell-Extension.html"><literal>Language.Haskell.Extension</literal></ulink>
7450 may be used. GHC will report an error if any of the requested extensions are not supported.</para>
7454 <sect2 id="options-pragma">
7455 <title>OPTIONS_GHC pragma</title>
7456 <indexterm><primary>OPTIONS_GHC</primary>
7458 <indexterm><primary>pragma</primary><secondary>OPTIONS_GHC</secondary>
7461 <para>The <literal>OPTIONS_GHC</literal> pragma is used to specify
7462 additional options that are given to the compiler when compiling
7463 this source file. See <xref linkend="source-file-options"/> for
7466 <para>Previous versions of GHC accepted <literal>OPTIONS</literal> rather
7467 than <literal>OPTIONS_GHC</literal>, but that is now deprecated.</para>
7470 <para><literal>OPTIONS_GHC</literal> is a file-header pragma (see <xref linkend="pragmas"/>).</para>
7472 <sect2 id="include-pragma">
7473 <title>INCLUDE pragma</title>
7475 <para>The <literal>INCLUDE</literal> used to be necessary for
7476 specifying header files to be included when using the FFI and
7477 compiling via C. It is no longer required for GHC, but is
7478 accepted (and ignored) for compatibility with other
7482 <sect2 id="warning-deprecated-pragma">
7483 <title>WARNING and DEPRECATED pragmas</title>
7484 <indexterm><primary>WARNING</primary></indexterm>
7485 <indexterm><primary>DEPRECATED</primary></indexterm>
7487 <para>The WARNING pragma allows you to attach an arbitrary warning
7488 to a particular function, class, or type.
7489 A DEPRECATED pragma lets you specify that
7490 a particular function, class, or type is deprecated.
7491 There are two ways of using these pragmas.
7495 <para>You can work on an entire module thus:</para>
7497 module Wibble {-# DEPRECATED "Use Wobble instead" #-} where
7502 module Wibble {-# WARNING "This is an unstable interface." #-} where
7505 <para>When you compile any module that import
7506 <literal>Wibble</literal>, GHC will print the specified
7511 <para>You can attach a warning to a function, class, type, or data constructor, with the
7512 following top-level declarations:</para>
7514 {-# DEPRECATED f, C, T "Don't use these" #-}
7515 {-# WARNING unsafePerformIO "This is unsafe; I hope you know what you're doing" #-}
7517 <para>When you compile any module that imports and uses any
7518 of the specified entities, GHC will print the specified
7520 <para> You can only attach to entities declared at top level in the module
7521 being compiled, and you can only use unqualified names in the list of
7522 entities. A capitalised name, such as <literal>T</literal>
7523 refers to <emphasis>either</emphasis> the type constructor <literal>T</literal>
7524 <emphasis>or</emphasis> the data constructor <literal>T</literal>, or both if
7525 both are in scope. If both are in scope, there is currently no way to
7526 specify one without the other (c.f. fixities
7527 <xref linkend="infix-tycons"/>).</para>
7530 Warnings and deprecations are not reported for
7531 (a) uses within the defining module, and
7532 (b) uses in an export list.
7533 The latter reduces spurious complaints within a library
7534 in which one module gathers together and re-exports
7535 the exports of several others.
7537 <para>You can suppress the warnings with the flag
7538 <option>-fno-warn-warnings-deprecations</option>.</para>
7541 <sect2 id="inline-noinline-pragma">
7542 <title>INLINE and NOINLINE pragmas</title>
7544 <para>These pragmas control the inlining of function
7547 <sect3 id="inline-pragma">
7548 <title>INLINE pragma</title>
7549 <indexterm><primary>INLINE</primary></indexterm>
7551 <para>GHC (with <option>-O</option>, as always) tries to
7552 inline (or “unfold”) functions/values that are
7553 “small enough,” thus avoiding the call overhead
7554 and possibly exposing other more-wonderful optimisations.
7555 Normally, if GHC decides a function is “too
7556 expensive” to inline, it will not do so, nor will it
7557 export that unfolding for other modules to use.</para>
7559 <para>The sledgehammer you can bring to bear is the
7560 <literal>INLINE</literal><indexterm><primary>INLINE
7561 pragma</primary></indexterm> pragma, used thusly:</para>
7564 key_function :: Int -> String -> (Bool, Double)
7565 {-# INLINE key_function #-}
7568 <para>The major effect of an <literal>INLINE</literal> pragma
7569 is to declare a function's “cost” to be very low.
7570 The normal unfolding machinery will then be very keen to
7571 inline it. However, an <literal>INLINE</literal> pragma for a
7572 function "<literal>f</literal>" has a number of other effects:
7575 While GHC is keen to inline the function, it does not do so
7576 blindly. For example, if you write
7580 there really isn't any point in inlining <literal>key_function</literal> to get
7582 map (\x -> <replaceable>body</replaceable>) xs
7584 In general, GHC only inlines the function if there is some reason (no matter
7585 how slight) to supose that it is useful to do so.
7589 Moreover, GHC will only inline the function if it is <emphasis>fully applied</emphasis>,
7590 where "fully applied"
7591 means applied to as many arguments as appear (syntactically)
7592 on the LHS of the function
7593 definition. For example:
7595 comp1 :: (b -> c) -> (a -> b) -> a -> c
7596 {-# INLINE comp1 #-}
7597 comp1 f g = \x -> f (g x)
7599 comp2 :: (b -> c) -> (a -> b) -> a -> c
7600 {-# INLINE comp2 #-}
7601 comp2 f g x = f (g x)
7603 The two functions <literal>comp1</literal> and <literal>comp2</literal> have the
7604 same semantics, but <literal>comp1</literal> will be inlined when applied
7605 to <emphasis>two</emphasis> arguments, while <literal>comp2</literal> requires
7606 <emphasis>three</emphasis>. This might make a big difference if you say
7608 map (not `comp1` not) xs
7610 which will optimise better than the corresponding use of `comp2`.
7614 It is useful for GHC to optimise the definition of an
7615 INLINE function <literal>f</literal> just like any other non-INLINE function,
7616 in case the non-inlined version of <literal>f</literal> is
7617 ultimately called. But we don't want to inline
7618 the <emphasis>optimised</emphasis> version
7619 of <literal>f</literal>;
7620 a major reason for INLINE pragmas is to expose functions
7621 in <literal>f</literal>'s RHS that have
7622 rewrite rules, and it's no good if those functions have been optimised
7626 So <emphasis>GHC guarantees to inline precisely the code that you wrote</emphasis>, no more
7627 and no less. It does this by capturing a copy of the definition of the function to use
7628 for inlining (we call this the "inline-RHS"), which it leaves untouched,
7629 while optimising the ordinarly RHS as usual. For externally-visible functions
7630 the inline-RHS (not the optimised RHS) is recorded in the interface file.
7633 An INLINE function is not worker/wrappered by strictness analysis.
7634 It's going to be inlined wholesale instead.
7638 <para>GHC ensures that inlining cannot go on forever: every mutually-recursive
7639 group is cut by one or more <emphasis>loop breakers</emphasis> that is never inlined
7640 (see <ulink url="http://research.microsoft.com/%7Esimonpj/Papers/inlining/index.htm">
7641 Secrets of the GHC inliner, JFP 12(4) July 2002</ulink>).
7642 GHC tries not to select a function with an INLINE pragma as a loop breaker, but
7643 when there is no choice even an INLINE function can be selected, in which case
7644 the INLINE pragma is ignored.
7645 For example, for a self-recursive function, the loop breaker can only be the function
7646 itself, so an INLINE pragma is always ignored.</para>
7648 <para>Syntactically, an <literal>INLINE</literal> pragma for a
7649 function can be put anywhere its type signature could be
7652 <para><literal>INLINE</literal> pragmas are a particularly
7654 <literal>then</literal>/<literal>return</literal> (or
7655 <literal>bind</literal>/<literal>unit</literal>) functions in
7656 a monad. For example, in GHC's own
7657 <literal>UniqueSupply</literal> monad code, we have:</para>
7660 {-# INLINE thenUs #-}
7661 {-# INLINE returnUs #-}
7664 <para>See also the <literal>NOINLINE</literal> (<xref linkend="inlinable-pragma"/>)
7665 and <literal>INLINABLE</literal> (<xref linkend="noinline-pragma"/>)
7668 <para>Note: the HBC compiler doesn't like <literal>INLINE</literal> pragmas,
7669 so if you want your code to be HBC-compatible you'll have to surround
7670 the pragma with C pre-processor directives
7671 <literal>#ifdef __GLASGOW_HASKELL__</literal>...<literal>#endif</literal>.</para>
7675 <sect3 id="inlinable-pragma">
7676 <title>INLINABLE pragma</title>
7678 <para>An <literal>{-# INLINABLE f #-}</literal> pragma on a
7679 function <literal>f</literal> has the following behaviour:
7682 While <literal>INLINE</literal> says "please inline me", the <literal>INLINABLE</literal>
7683 says "feel free to inline me; use your
7684 discretion". In other words the choice is left to GHC, which uses the same
7685 rules as for pragma-free functions. Unlike <literal>INLINE</literal>, that decision is made at
7686 the <emphasis>call site</emphasis>, and
7687 will therefore be affected by the inlining threshold, optimisation level etc.
7690 Like <literal>INLINE</literal>, the <literal>INLINABLE</literal> pragma retains a
7691 copy of the original RHS for
7692 inlining purposes, and persists it in the interface file, regardless of
7693 the size of the RHS.
7697 One way to use <literal>INLINABLE</literal> is in conjunction with
7698 the special function <literal>inline</literal> (<xref linkend="special-ids"/>).
7699 The call <literal>inline f</literal> tries very hard to inline <literal>f</literal>.
7700 To make sure that <literal>f</literal> can be inlined,
7701 it is a good idea to mark the definition
7702 of <literal>f</literal> as <literal>INLINABLE</literal>,
7703 so that GHC guarantees to expose an unfolding regardless of how big it is.
7704 Moreover, by annotating <literal>f</literal> as <literal>INLINABLE</literal>,
7705 you ensure that <literal>f</literal>'s original RHS is inlined, rather than
7706 whatever random optimised version of <literal>f</literal> GHC's optimiser
7711 The <literal>INLINABLE</literal> pragma also works with <literal>SPECIALISE</literal>:
7712 if you mark function <literal>f</literal> as <literal>INLINABLE</literal>, then
7713 you can subsequently <literal>SPECIALISE</literal> in another module
7714 (see <xref linkend="specialize-pragma"/>).</para></listitem>
7717 Unlike <literal>INLINE</literal>, it is OK to use
7718 an <literal>INLINABLE</literal> pragma on a recursive function.
7719 The principal reason do to so to allow later use of <literal>SPECIALISE</literal>
7726 <sect3 id="noinline-pragma">
7727 <title>NOINLINE pragma</title>
7729 <indexterm><primary>NOINLINE</primary></indexterm>
7730 <indexterm><primary>NOTINLINE</primary></indexterm>
7732 <para>The <literal>NOINLINE</literal> pragma does exactly what
7733 you'd expect: it stops the named function from being inlined
7734 by the compiler. You shouldn't ever need to do this, unless
7735 you're very cautious about code size.</para>
7737 <para><literal>NOTINLINE</literal> is a synonym for
7738 <literal>NOINLINE</literal> (<literal>NOINLINE</literal> is
7739 specified by Haskell 98 as the standard way to disable
7740 inlining, so it should be used if you want your code to be
7744 <sect3 id="conlike-pragma">
7745 <title>CONLIKE modifier</title>
7746 <indexterm><primary>CONLIKE</primary></indexterm>
7747 <para>An INLINE or NOINLINE pragma may have a CONLIKE modifier,
7748 which affects matching in RULEs (only). See <xref linkend="conlike"/>.
7752 <sect3 id="phase-control">
7753 <title>Phase control</title>
7755 <para> Sometimes you want to control exactly when in GHC's
7756 pipeline the INLINE pragma is switched on. Inlining happens
7757 only during runs of the <emphasis>simplifier</emphasis>. Each
7758 run of the simplifier has a different <emphasis>phase
7759 number</emphasis>; the phase number decreases towards zero.
7760 If you use <option>-dverbose-core2core</option> you'll see the
7761 sequence of phase numbers for successive runs of the
7762 simplifier. In an INLINE pragma you can optionally specify a
7766 <para>"<literal>INLINE[k] f</literal>" means: do not inline
7767 <literal>f</literal>
7768 until phase <literal>k</literal>, but from phase
7769 <literal>k</literal> onwards be very keen to inline it.
7772 <para>"<literal>INLINE[~k] f</literal>" means: be very keen to inline
7773 <literal>f</literal>
7774 until phase <literal>k</literal>, but from phase
7775 <literal>k</literal> onwards do not inline it.
7778 <para>"<literal>NOINLINE[k] f</literal>" means: do not inline
7779 <literal>f</literal>
7780 until phase <literal>k</literal>, but from phase
7781 <literal>k</literal> onwards be willing to inline it (as if
7782 there was no pragma).
7785 <para>"<literal>NOINLINE[~k] f</literal>" means: be willing to inline
7786 <literal>f</literal>
7787 until phase <literal>k</literal>, but from phase
7788 <literal>k</literal> onwards do not inline it.
7791 The same information is summarised here:
7793 -- Before phase 2 Phase 2 and later
7794 {-# INLINE [2] f #-} -- No Yes
7795 {-# INLINE [~2] f #-} -- Yes No
7796 {-# NOINLINE [2] f #-} -- No Maybe
7797 {-# NOINLINE [~2] f #-} -- Maybe No
7799 {-# INLINE f #-} -- Yes Yes
7800 {-# NOINLINE f #-} -- No No
7802 By "Maybe" we mean that the usual heuristic inlining rules apply (if the
7803 function body is small, or it is applied to interesting-looking arguments etc).
7804 Another way to understand the semantics is this:
7806 <listitem><para>For both INLINE and NOINLINE, the phase number says
7807 when inlining is allowed at all.</para></listitem>
7808 <listitem><para>The INLINE pragma has the additional effect of making the
7809 function body look small, so that when inlining is allowed it is very likely to
7814 <para>The same phase-numbering control is available for RULES
7815 (<xref linkend="rewrite-rules"/>).</para>
7819 <sect2 id="annotation-pragmas">
7820 <title>ANN pragmas</title>
7822 <para>GHC offers the ability to annotate various code constructs with additional
7823 data by using three pragmas. This data can then be inspected at a later date by
7824 using GHC-as-a-library.</para>
7826 <sect3 id="ann-pragma">
7827 <title>Annotating values</title>
7829 <indexterm><primary>ANN</primary></indexterm>
7831 <para>Any expression that has both <literal>Typeable</literal> and <literal>Data</literal> instances may be attached to a top-level value
7832 binding using an <literal>ANN</literal> pragma. In particular, this means you can use <literal>ANN</literal>
7833 to annotate data constructors (e.g. <literal>Just</literal>) as well as normal values (e.g. <literal>take</literal>).
7834 By way of example, to annotate the function <literal>foo</literal> with the annotation <literal>Just "Hello"</literal>
7835 you would do this:</para>
7838 {-# ANN foo (Just "Hello") #-}
7843 A number of restrictions apply to use of annotations:
7845 <listitem><para>The binder being annotated must be at the top level (i.e. no nested binders)</para></listitem>
7846 <listitem><para>The binder being annotated must be declared in the current module</para></listitem>
7847 <listitem><para>The expression you are annotating with must have a type with <literal>Typeable</literal> and <literal>Data</literal> instances</para></listitem>
7848 <listitem><para>The <ulink linkend="using-template-haskell">Template Haskell staging restrictions</ulink> apply to the
7849 expression being annotated with, so for example you cannot run a function from the module being compiled.</para>
7851 <para>To be precise, the annotation <literal>{-# ANN x e #-}</literal> is well staged if and only if <literal>$(e)</literal> would be
7852 (disregarding the usual type restrictions of the splice syntax, and the usual restriction on splicing inside a splice - <literal>$([|1|])</literal> is fine as an annotation, albeit redundant).</para></listitem>
7855 If you feel strongly that any of these restrictions are too onerous, <ulink url="http://hackage.haskell.org/trac/ghc/wiki/MailingListsAndIRC">
7856 please give the GHC team a shout</ulink>.
7859 <para>However, apart from these restrictions, many things are allowed, including expressions which are not fully evaluated!
7860 Annotation expressions will be evaluated by the compiler just like Template Haskell splices are. So, this annotation is fine:</para>
7863 {-# ANN f SillyAnnotation { foo = (id 10) + $([| 20 |]), bar = 'f } #-}
7868 <sect3 id="typeann-pragma">
7869 <title>Annotating types</title>
7871 <indexterm><primary>ANN type</primary></indexterm>
7872 <indexterm><primary>ANN</primary></indexterm>
7874 <para>You can annotate types with the <literal>ANN</literal> pragma by using the <literal>type</literal> keyword. For example:</para>
7877 {-# ANN type Foo (Just "A `Maybe String' annotation") #-}
7882 <sect3 id="modann-pragma">
7883 <title>Annotating modules</title>
7885 <indexterm><primary>ANN module</primary></indexterm>
7886 <indexterm><primary>ANN</primary></indexterm>
7888 <para>You can annotate modules with the <literal>ANN</literal> pragma by using the <literal>module</literal> keyword. For example:</para>
7891 {-# ANN module (Just "A `Maybe String' annotation") #-}
7896 <sect2 id="line-pragma">
7897 <title>LINE pragma</title>
7899 <indexterm><primary>LINE</primary><secondary>pragma</secondary></indexterm>
7900 <indexterm><primary>pragma</primary><secondary>LINE</secondary></indexterm>
7901 <para>This pragma is similar to C's <literal>#line</literal>
7902 pragma, and is mainly for use in automatically generated Haskell
7903 code. It lets you specify the line number and filename of the
7904 original code; for example</para>
7906 <programlisting>{-# LINE 42 "Foo.vhs" #-}</programlisting>
7908 <para>if you'd generated the current file from something called
7909 <filename>Foo.vhs</filename> and this line corresponds to line
7910 42 in the original. GHC will adjust its error messages to refer
7911 to the line/file named in the <literal>LINE</literal>
7916 <title>RULES pragma</title>
7918 <para>The RULES pragma lets you specify rewrite rules. It is
7919 described in <xref linkend="rewrite-rules"/>.</para>
7922 <sect2 id="specialize-pragma">
7923 <title>SPECIALIZE pragma</title>
7925 <indexterm><primary>SPECIALIZE pragma</primary></indexterm>
7926 <indexterm><primary>pragma, SPECIALIZE</primary></indexterm>
7927 <indexterm><primary>overloading, death to</primary></indexterm>
7929 <para>(UK spelling also accepted.) For key overloaded
7930 functions, you can create extra versions (NB: more code space)
7931 specialised to particular types. Thus, if you have an
7932 overloaded function:</para>
7935 hammeredLookup :: Ord key => [(key, value)] -> key -> value
7938 <para>If it is heavily used on lists with
7939 <literal>Widget</literal> keys, you could specialise it as
7943 {-# SPECIALIZE hammeredLookup :: [(Widget, value)] -> Widget -> value #-}
7946 <para>A <literal>SPECIALIZE</literal> pragma for a function can
7947 be put anywhere its type signature could be put.</para>
7949 <para>A <literal>SPECIALIZE</literal> has the effect of generating
7950 (a) a specialised version of the function and (b) a rewrite rule
7951 (see <xref linkend="rewrite-rules"/>) that rewrites a call to the
7952 un-specialised function into a call to the specialised one.</para>
7954 <para>The type in a SPECIALIZE pragma can be any type that is less
7955 polymorphic than the type of the original function. In concrete terms,
7956 if the original function is <literal>f</literal> then the pragma
7958 {-# SPECIALIZE f :: <type> #-}
7960 is valid if and only if the definition
7962 f_spec :: <type>
7965 is valid. Here are some examples (where we only give the type signature
7966 for the original function, not its code):
7968 f :: Eq a => a -> b -> b
7969 {-# SPECIALISE f :: Int -> b -> b #-}
7971 g :: (Eq a, Ix b) => a -> b -> b
7972 {-# SPECIALISE g :: (Eq a) => a -> Int -> Int #-}
7974 h :: Eq a => a -> a -> a
7975 {-# SPECIALISE h :: (Eq a) => [a] -> [a] -> [a] #-}
7977 The last of these examples will generate a
7978 RULE with a somewhat-complex left-hand side (try it yourself), so it might not fire very
7979 well. If you use this kind of specialisation, let us know how well it works.
7982 <sect3 id="specialize-inline">
7983 <title>SPECIALIZE INLINE</title>
7985 <para>A <literal>SPECIALIZE</literal> pragma can optionally be followed with a
7986 <literal>INLINE</literal> or <literal>NOINLINE</literal> pragma, optionally
7987 followed by a phase, as described in <xref linkend="inline-noinline-pragma"/>.
7988 The <literal>INLINE</literal> pragma affects the specialised version of the
7989 function (only), and applies even if the function is recursive. The motivating
7992 -- A GADT for arrays with type-indexed representation
7994 ArrInt :: !Int -> ByteArray# -> Arr Int
7995 ArrPair :: !Int -> Arr e1 -> Arr e2 -> Arr (e1, e2)
7997 (!:) :: Arr e -> Int -> e
7998 {-# SPECIALISE INLINE (!:) :: Arr Int -> Int -> Int #-}
7999 {-# SPECIALISE INLINE (!:) :: Arr (a, b) -> Int -> (a, b) #-}
8000 (ArrInt _ ba) !: (I# i) = I# (indexIntArray# ba i)
8001 (ArrPair _ a1 a2) !: i = (a1 !: i, a2 !: i)
8003 Here, <literal>(!:)</literal> is a recursive function that indexes arrays
8004 of type <literal>Arr e</literal>. Consider a call to <literal>(!:)</literal>
8005 at type <literal>(Int,Int)</literal>. The second specialisation will fire, and
8006 the specialised function will be inlined. It has two calls to
8007 <literal>(!:)</literal>,
8008 both at type <literal>Int</literal>. Both these calls fire the first
8009 specialisation, whose body is also inlined. The result is a type-based
8010 unrolling of the indexing function.</para>
8011 <para>Warning: you can make GHC diverge by using <literal>SPECIALISE INLINE</literal>
8012 on an ordinarily-recursive function.</para>
8015 <sect3><title>SPECIALIZE for imported functions</title>
8018 Generally, you can only give a <literal>SPECIALIZE</literal> pragma
8019 for a function defined in the same module.
8020 However if a function <literal>f</literal> is given an <literal>INLINABLE</literal>
8021 pragma at its definition site, then it can subequently be specialised by
8022 importing modules (see <xref linkend="inlinable-pragma"/>).
8025 module Map( lookup, blah blah ) where
8026 lookup :: Ord key => [(key,a)] -> key -> Maybe a
8028 {-# INLINABLE lookup #-}
8031 import Map( lookup )
8033 data T = T1 | T2 deriving( Eq, Ord )
8034 {-# SPECIALISE lookup :: [(T,a)] -> T -> Maybe a
8036 Here, <literal>lookup</literal> is declared <literal>INLINABLE</literal>, but
8037 it cannot be specialised for type <literal>T</literal> at its definition site,
8038 because that type does not exist yet. Instead a client module can define <literal>T</literal>
8039 and then specialise <literal>lookup</literal> at that type.
8042 Moreover, every module that imports <literal>Client</literal> (or imports a module
8043 that imports <literal>Client</literal>, transitively) will "see", and make use of,
8044 the specialised version of <literal>lookup</literal>. You don't need to put
8045 a <literal>SPECIALIZE</literal> pragma in every module.
8048 Moreover you often don't even need the <literal>SPECIALIZE</literal> pragma in the
8049 first place. When compiling a module M,
8050 GHC's optimiser (with -O) automatically considers each top-level
8051 overloaded function declared in M, and specialises it
8052 for the different types at which it is called in M. The optimiser
8053 <emphasis>also</emphasis> considers each <emphasis>imported</emphasis>
8054 <literal>INLINABLE</literal> overloaded function, and specialises it
8055 for the different types at which it is called in M.
8056 So in our example, it would be enough for <literal>lookup</literal> to
8057 be called at type <literal>T</literal>:
8060 import Map( lookup )
8062 data T = T1 | T2 deriving( Eq, Ord )
8064 findT1 :: [(T,a)] -> Maybe a
8065 findT1 m = lookup m T1 -- A call of lookup at type T
8067 However, sometimes there are no such calls, in which case the
8068 pragma can be useful.
8072 <sect3><title>Obselete SPECIALIZE syntax</title>
8074 <para>Note: In earlier versions of GHC, it was possible to provide your own
8075 specialised function for a given type:
8078 {-# SPECIALIZE hammeredLookup :: [(Int, value)] -> Int -> value = intLookup #-}
8081 This feature has been removed, as it is now subsumed by the
8082 <literal>RULES</literal> pragma (see <xref linkend="rule-spec"/>).</para>
8087 <sect2 id="specialize-instance-pragma">
8088 <title>SPECIALIZE instance pragma
8092 <indexterm><primary>SPECIALIZE pragma</primary></indexterm>
8093 <indexterm><primary>overloading, death to</primary></indexterm>
8094 Same idea, except for instance declarations. For example:
8097 instance (Eq a) => Eq (Foo a) where {
8098 {-# SPECIALIZE instance Eq (Foo [(Int, Bar)]) #-}
8102 The pragma must occur inside the <literal>where</literal> part
8103 of the instance declaration.
8106 Compatible with HBC, by the way, except perhaps in the placement
8112 <sect2 id="unpack-pragma">
8113 <title>UNPACK pragma</title>
8115 <indexterm><primary>UNPACK</primary></indexterm>
8117 <para>The <literal>UNPACK</literal> indicates to the compiler
8118 that it should unpack the contents of a constructor field into
8119 the constructor itself, removing a level of indirection. For
8123 data T = T {-# UNPACK #-} !Float
8124 {-# UNPACK #-} !Float
8127 <para>will create a constructor <literal>T</literal> containing
8128 two unboxed floats. This may not always be an optimisation: if
8129 the <function>T</function> constructor is scrutinised and the
8130 floats passed to a non-strict function for example, they will
8131 have to be reboxed (this is done automatically by the
8134 <para>Unpacking constructor fields should only be used in
8135 conjunction with <option>-O</option>, in order to expose
8136 unfoldings to the compiler so the reboxing can be removed as
8137 often as possible. For example:</para>
8141 f (T f1 f2) = f1 + f2
8144 <para>The compiler will avoid reboxing <function>f1</function>
8145 and <function>f2</function> by inlining <function>+</function>
8146 on floats, but only when <option>-O</option> is on.</para>
8148 <para>Any single-constructor data is eligible for unpacking; for
8152 data T = T {-# UNPACK #-} !(Int,Int)
8155 <para>will store the two <literal>Int</literal>s directly in the
8156 <function>T</function> constructor, by flattening the pair.
8157 Multi-level unpacking is also supported:
8160 data T = T {-# UNPACK #-} !S
8161 data S = S {-# UNPACK #-} !Int {-# UNPACK #-} !Int
8164 will store two unboxed <literal>Int#</literal>s
8165 directly in the <function>T</function> constructor. The
8166 unpacker can see through newtypes, too.</para>
8168 <para>See also the <option>-funbox-strict-fields</option> flag,
8169 which essentially has the effect of adding
8170 <literal>{-# UNPACK #-}</literal> to every strict
8171 constructor field.</para>
8174 <sect2 id="source-pragma">
8175 <title>SOURCE pragma</title>
8177 <indexterm><primary>SOURCE</primary></indexterm>
8178 <para>The <literal>{-# SOURCE #-}</literal> pragma is used only in <literal>import</literal> declarations,
8179 to break a module loop. It is described in detail in <xref linkend="mutual-recursion"/>.
8185 <!-- ======================= REWRITE RULES ======================== -->
8187 <sect1 id="rewrite-rules">
8188 <title>Rewrite rules
8190 <indexterm><primary>RULES pragma</primary></indexterm>
8191 <indexterm><primary>pragma, RULES</primary></indexterm>
8192 <indexterm><primary>rewrite rules</primary></indexterm></title>
8195 The programmer can specify rewrite rules as part of the source program
8201 "map/map" forall f g xs. map f (map g xs) = map (f.g) xs
8206 Use the debug flag <option>-ddump-simpl-stats</option> to see what rules fired.
8207 If you need more information, then <option>-ddump-rule-firings</option> shows you
8208 each individual rule firing and <option>-ddump-rule-rewrites</option> also shows what the code looks like before and after the rewrite.
8212 <title>Syntax</title>
8215 From a syntactic point of view:
8221 There may be zero or more rules in a <literal>RULES</literal> pragma, separated by semicolons (which
8222 may be generated by the layout rule).
8228 The layout rule applies in a pragma.
8229 Currently no new indentation level
8230 is set, so if you put several rules in single RULES pragma and wish to use layout to separate them,
8231 you must lay out the starting in the same column as the enclosing definitions.
8234 "map/map" forall f g xs. map f (map g xs) = map (f.g) xs
8235 "map/append" forall f xs ys. map f (xs ++ ys) = map f xs ++ map f ys
8238 Furthermore, the closing <literal>#-}</literal>
8239 should start in a column to the right of the opening <literal>{-#</literal>.
8245 Each rule has a name, enclosed in double quotes. The name itself has
8246 no significance at all. It is only used when reporting how many times the rule fired.
8252 A rule may optionally have a phase-control number (see <xref linkend="phase-control"/>),
8253 immediately after the name of the rule. Thus:
8256 "map/map" [2] forall f g xs. map f (map g xs) = map (f.g) xs
8259 The "[2]" means that the rule is active in Phase 2 and subsequent phases. The inverse
8260 notation "[~2]" is also accepted, meaning that the rule is active up to, but not including,
8269 Each variable mentioned in a rule must either be in scope (e.g. <function>map</function>),
8270 or bound by the <literal>forall</literal> (e.g. <function>f</function>, <function>g</function>, <function>xs</function>). The variables bound by
8271 the <literal>forall</literal> are called the <emphasis>pattern</emphasis> variables. They are separated
8272 by spaces, just like in a type <literal>forall</literal>.
8278 A pattern variable may optionally have a type signature.
8279 If the type of the pattern variable is polymorphic, it <emphasis>must</emphasis> have a type signature.
8280 For example, here is the <literal>foldr/build</literal> rule:
8283 "fold/build" forall k z (g::forall b. (a->b->b) -> b -> b) .
8284 foldr k z (build g) = g k z
8287 Since <function>g</function> has a polymorphic type, it must have a type signature.
8294 The left hand side of a rule must consist of a top-level variable applied
8295 to arbitrary expressions. For example, this is <emphasis>not</emphasis> OK:
8298 "wrong1" forall e1 e2. case True of { True -> e1; False -> e2 } = e1
8299 "wrong2" forall f. f True = True
8302 In <literal>"wrong1"</literal>, the LHS is not an application; in <literal>"wrong2"</literal>, the LHS has a pattern variable
8309 A rule does not need to be in the same module as (any of) the
8310 variables it mentions, though of course they need to be in scope.
8316 All rules are implicitly exported from the module, and are therefore
8317 in force in any module that imports the module that defined the rule, directly
8318 or indirectly. (That is, if A imports B, which imports C, then C's rules are
8319 in force when compiling A.) The situation is very similar to that for instance
8327 Inside a RULE "<literal>forall</literal>" is treated as a keyword, regardless of
8328 any other flag settings. Furthermore, inside a RULE, the language extension
8329 <option>-XScopedTypeVariables</option> is automatically enabled; see
8330 <xref linkend="scoped-type-variables"/>.
8336 Like other pragmas, RULE pragmas are always checked for scope errors, and
8337 are typechecked. Typechecking means that the LHS and RHS of a rule are typechecked,
8338 and must have the same type. However, rules are only <emphasis>enabled</emphasis>
8339 if the <option>-fenable-rewrite-rules</option> flag is
8340 on (see <xref linkend="rule-semantics"/>).
8349 <sect2 id="rule-semantics">
8350 <title>Semantics</title>
8353 From a semantic point of view:
8358 Rules are enabled (that is, used during optimisation)
8359 by the <option>-fenable-rewrite-rules</option> flag.
8360 This flag is implied by <option>-O</option>, and may be switched
8361 off (as usual) by <option>-fno-enable-rewrite-rules</option>.
8362 (NB: enabling <option>-fenable-rewrite-rules</option> without <option>-O</option>
8363 may not do what you expect, though, because without <option>-O</option> GHC
8364 ignores all optimisation information in interface files;
8365 see <option>-fignore-interface-pragmas</option>, <xref linkend="options-f"/>.)
8366 Note that <option>-fenable-rewrite-rules</option> is an <emphasis>optimisation</emphasis> flag, and
8367 has no effect on parsing or typechecking.
8373 Rules are regarded as left-to-right rewrite rules.
8374 When GHC finds an expression that is a substitution instance of the LHS
8375 of a rule, it replaces the expression by the (appropriately-substituted) RHS.
8376 By "a substitution instance" we mean that the LHS can be made equal to the
8377 expression by substituting for the pattern variables.
8384 GHC makes absolutely no attempt to verify that the LHS and RHS
8385 of a rule have the same meaning. That is undecidable in general, and
8386 infeasible in most interesting cases. The responsibility is entirely the programmer's!
8393 GHC makes no attempt to make sure that the rules are confluent or
8394 terminating. For example:
8397 "loop" forall x y. f x y = f y x
8400 This rule will cause the compiler to go into an infinite loop.
8407 If more than one rule matches a call, GHC will choose one arbitrarily to apply.
8413 GHC currently uses a very simple, syntactic, matching algorithm
8414 for matching a rule LHS with an expression. It seeks a substitution
8415 which makes the LHS and expression syntactically equal modulo alpha
8416 conversion. The pattern (rule), but not the expression, is eta-expanded if
8417 necessary. (Eta-expanding the expression can lead to laziness bugs.)
8418 But not beta conversion (that's called higher-order matching).
8422 Matching is carried out on GHC's intermediate language, which includes
8423 type abstractions and applications. So a rule only matches if the
8424 types match too. See <xref linkend="rule-spec"/> below.
8430 GHC keeps trying to apply the rules as it optimises the program.
8431 For example, consider:
8440 The expression <literal>s (t xs)</literal> does not match the rule <literal>"map/map"</literal>, but GHC
8441 will substitute for <varname>s</varname> and <varname>t</varname>, giving an expression which does match.
8442 If <varname>s</varname> or <varname>t</varname> was (a) used more than once, and (b) large or a redex, then it would
8443 not be substituted, and the rule would not fire.
8453 <sect2 id="conlike">
8454 <title>How rules interact with INLINE/NOINLINE and CONLIKE pragmas</title>
8457 Ordinary inlining happens at the same time as rule rewriting, which may lead to unexpected
8458 results. Consider this (artificial) example
8464 {-# RULES "f" f True = False #-}
8466 Since <literal>f</literal>'s right-hand side is small, it is inlined into <literal>g</literal>,
8471 Now <literal>g</literal> is inlined into <literal>h</literal>, but <literal>f</literal>'s RULE has
8473 If instead GHC had first inlined <literal>g</literal> into <literal>h</literal> then there
8474 would have been a better chance that <literal>f</literal>'s RULE might fire.
8477 The way to get predictable behaviour is to use a NOINLINE
8478 pragma, or an INLINE[<replaceable>phase</replaceable>] pragma, on <literal>f</literal>, to ensure
8479 that it is not inlined until its RULEs have had a chance to fire.
8482 GHC is very cautious about duplicating work. For example, consider
8484 f k z xs = let xs = build g
8485 in ...(foldr k z xs)...sum xs...
8486 {-# RULES "foldr/build" forall k z g. foldr k z (build g) = g k z #-}
8488 Since <literal>xs</literal> is used twice, GHC does not fire the foldr/build rule. Rightly
8489 so, because it might take a lot of work to compute <literal>xs</literal>, which would be
8490 duplicated if the rule fired.
8493 Sometimes, however, this approach is over-cautious, and we <emphasis>do</emphasis> want the
8494 rule to fire, even though doing so would duplicate redex. There is no way that GHC can work out
8495 when this is a good idea, so we provide the CONLIKE pragma to declare it, thus:
8497 {-# INLINE[1] CONLIKE f #-}
8498 f x = <replaceable>blah</replaceable>
8500 CONLIKE is a modifier to an INLINE or NOINLINE pragam. It specifies that an application
8501 of f to one argument (in general, the number of arguments to the left of the '=' sign)
8502 should be considered cheap enough to duplicate, if such a duplication would make rule
8503 fire. (The name "CONLIKE" is short for "constructor-like", because constructors certainly
8504 have such a property.)
8505 The CONLIKE pragam is a modifier to INLINE/NOINLINE because it really only makes sense to match
8506 <literal>f</literal> on the LHS of a rule if you are sure that <literal>f</literal> is
8507 not going to be inlined before the rule has a chance to fire.
8512 <title>List fusion</title>
8515 The RULES mechanism is used to implement fusion (deforestation) of common list functions.
8516 If a "good consumer" consumes an intermediate list constructed by a "good producer", the
8517 intermediate list should be eliminated entirely.
8521 The following are good producers:
8533 Enumerations of <literal>Int</literal> and <literal>Char</literal> (e.g. <literal>['a'..'z']</literal>).
8539 Explicit lists (e.g. <literal>[True, False]</literal>)
8545 The cons constructor (e.g <literal>3:4:[]</literal>)
8551 <function>++</function>
8557 <function>map</function>
8563 <function>take</function>, <function>filter</function>
8569 <function>iterate</function>, <function>repeat</function>
8575 <function>zip</function>, <function>zipWith</function>
8584 The following are good consumers:
8596 <function>array</function> (on its second argument)
8602 <function>++</function> (on its first argument)
8608 <function>foldr</function>
8614 <function>map</function>
8620 <function>take</function>, <function>filter</function>
8626 <function>concat</function>
8632 <function>unzip</function>, <function>unzip2</function>, <function>unzip3</function>, <function>unzip4</function>
8638 <function>zip</function>, <function>zipWith</function> (but on one argument only; if both are good producers, <function>zip</function>
8639 will fuse with one but not the other)
8645 <function>partition</function>
8651 <function>head</function>
8657 <function>and</function>, <function>or</function>, <function>any</function>, <function>all</function>
8663 <function>sequence_</function>
8669 <function>msum</function>
8675 <function>sortBy</function>
8684 So, for example, the following should generate no intermediate lists:
8687 array (1,10) [(i,i*i) | i <- map (+ 1) [0..9]]
8693 This list could readily be extended; if there are Prelude functions that you use
8694 a lot which are not included, please tell us.
8698 If you want to write your own good consumers or producers, look at the
8699 Prelude definitions of the above functions to see how to do so.
8704 <sect2 id="rule-spec">
8705 <title>Specialisation
8709 Rewrite rules can be used to get the same effect as a feature
8710 present in earlier versions of GHC.
8711 For example, suppose that:
8714 genericLookup :: Ord a => Table a b -> a -> b
8715 intLookup :: Table Int b -> Int -> b
8718 where <function>intLookup</function> is an implementation of
8719 <function>genericLookup</function> that works very fast for
8720 keys of type <literal>Int</literal>. You might wish
8721 to tell GHC to use <function>intLookup</function> instead of
8722 <function>genericLookup</function> whenever the latter was called with
8723 type <literal>Table Int b -> Int -> b</literal>.
8724 It used to be possible to write
8727 {-# SPECIALIZE genericLookup :: Table Int b -> Int -> b = intLookup #-}
8730 This feature is no longer in GHC, but rewrite rules let you do the same thing:
8733 {-# RULES "genericLookup/Int" genericLookup = intLookup #-}
8736 This slightly odd-looking rule instructs GHC to replace
8737 <function>genericLookup</function> by <function>intLookup</function>
8738 <emphasis>whenever the types match</emphasis>.
8739 What is more, this rule does not need to be in the same
8740 file as <function>genericLookup</function>, unlike the
8741 <literal>SPECIALIZE</literal> pragmas which currently do (so that they
8742 have an original definition available to specialise).
8745 <para>It is <emphasis>Your Responsibility</emphasis> to make sure that
8746 <function>intLookup</function> really behaves as a specialised version
8747 of <function>genericLookup</function>!!!</para>
8749 <para>An example in which using <literal>RULES</literal> for
8750 specialisation will Win Big:
8753 toDouble :: Real a => a -> Double
8754 toDouble = fromRational . toRational
8756 {-# RULES "toDouble/Int" toDouble = i2d #-}
8757 i2d (I# i) = D# (int2Double# i) -- uses Glasgow prim-op directly
8760 The <function>i2d</function> function is virtually one machine
8761 instruction; the default conversion—via an intermediate
8762 <literal>Rational</literal>—is obscenely expensive by
8768 <sect2 id="controlling-rules">
8769 <title>Controlling what's going on in rewrite rules</title>
8777 Use <option>-ddump-rules</option> to see the rules that are defined
8778 <emphasis>in this module</emphasis>.
8779 This includes rules generated by the specialisation pass, but excludes
8780 rules imported from other modules.
8786 Use <option>-ddump-simpl-stats</option> to see what rules are being fired.
8787 If you add <option>-dppr-debug</option> you get a more detailed listing.
8793 Use <option>-ddump-rule-firings</option> or <option>-ddump-rule-rewrites</option>
8794 to see in great detail what rules are being fired.
8795 If you add <option>-dppr-debug</option> you get a still more detailed listing.
8801 The definition of (say) <function>build</function> in <filename>GHC/Base.lhs</filename> looks like this:
8804 build :: forall a. (forall b. (a -> b -> b) -> b -> b) -> [a]
8805 {-# INLINE build #-}
8809 Notice the <literal>INLINE</literal>! That prevents <literal>(:)</literal> from being inlined when compiling
8810 <literal>PrelBase</literal>, so that an importing module will “see” the <literal>(:)</literal>, and can
8811 match it on the LHS of a rule. <literal>INLINE</literal> prevents any inlining happening
8812 in the RHS of the <literal>INLINE</literal> thing. I regret the delicacy of this.
8819 In <filename>libraries/base/GHC/Base.lhs</filename> look at the rules for <function>map</function> to
8820 see how to write rules that will do fusion and yet give an efficient
8821 program even if fusion doesn't happen. More rules in <filename>GHC/List.lhs</filename>.
8831 <sect2 id="core-pragma">
8832 <title>CORE pragma</title>
8834 <indexterm><primary>CORE pragma</primary></indexterm>
8835 <indexterm><primary>pragma, CORE</primary></indexterm>
8836 <indexterm><primary>core, annotation</primary></indexterm>
8839 The external core format supports <quote>Note</quote> annotations;
8840 the <literal>CORE</literal> pragma gives a way to specify what these
8841 should be in your Haskell source code. Syntactically, core
8842 annotations are attached to expressions and take a Haskell string
8843 literal as an argument. The following function definition shows an
8847 f x = ({-# CORE "foo" #-} show) ({-# CORE "bar" #-} x)
8850 Semantically, this is equivalent to:
8858 However, when external core is generated (via
8859 <option>-fext-core</option>), there will be Notes attached to the
8860 expressions <function>show</function> and <varname>x</varname>.
8861 The core function declaration for <function>f</function> is:
8865 f :: %forall a . GHCziShow.ZCTShow a ->
8866 a -> GHCziBase.ZMZN GHCziBase.Char =
8867 \ @ a (zddShow::GHCziShow.ZCTShow a) (eta::a) ->
8869 %case zddShow %of (tpl::GHCziShow.ZCTShow a)
8871 (tpl1::GHCziBase.Int ->
8873 GHCziBase.ZMZN GHCziBase.Char -> GHCziBase.ZMZN GHCziBase.Cha
8875 (tpl2::a -> GHCziBase.ZMZN GHCziBase.Char)
8876 (tpl3::GHCziBase.ZMZN a ->
8877 GHCziBase.ZMZN GHCziBase.Char -> GHCziBase.ZMZN GHCziBase.Cha
8885 Here, we can see that the function <function>show</function> (which
8886 has been expanded out to a case expression over the Show dictionary)
8887 has a <literal>%note</literal> attached to it, as does the
8888 expression <varname>eta</varname> (which used to be called
8889 <varname>x</varname>).
8896 <sect1 id="special-ids">
8897 <title>Special built-in functions</title>
8898 <para>GHC has a few built-in functions with special behaviour. These
8899 are now described in the module <ulink
8900 url="&libraryGhcPrimLocation;/GHC-Prim.html"><literal>GHC.Prim</literal></ulink>
8901 in the library documentation.
8905 <ulink url="&libraryGhcPrimLocation;/GHC-Prim.html#v%3Ainline"><literal>inline</literal></ulink>
8906 allows control over inlining on a per-call-site basis.
8909 <ulink url="&libraryGhcPrimLocation;/GHC-Prim.html#v%3Alazy"><literal>lazy</literal></ulink>
8910 restrains the strictness analyser.
8913 <ulink url="&libraryGhcPrimLocation;/GHC-Prim.html#v%3AunsafeCoerce%23"><literal>lazy</literal></ulink>
8914 allows you to fool the type checker.
8921 <sect1 id="generic-classes">
8922 <title>Generic classes</title>
8925 The ideas behind this extension are described in detail in "Derivable type classes",
8926 Ralf Hinze and Simon Peyton Jones, Haskell Workshop, Montreal Sept 2000, pp94-105.
8927 An example will give the idea:
8935 fromBin :: [Int] -> (a, [Int])
8937 toBin {| Unit |} Unit = []
8938 toBin {| a :+: b |} (Inl x) = 0 : toBin x
8939 toBin {| a :+: b |} (Inr y) = 1 : toBin y
8940 toBin {| a :*: b |} (x :*: y) = toBin x ++ toBin y
8942 fromBin {| Unit |} bs = (Unit, bs)
8943 fromBin {| a :+: b |} (0:bs) = (Inl x, bs') where (x,bs') = fromBin bs
8944 fromBin {| a :+: b |} (1:bs) = (Inr y, bs') where (y,bs') = fromBin bs
8945 fromBin {| a :*: b |} bs = (x :*: y, bs'') where (x,bs' ) = fromBin bs
8946 (y,bs'') = fromBin bs'
8949 This class declaration explains how <literal>toBin</literal> and <literal>fromBin</literal>
8950 work for arbitrary data types. They do so by giving cases for unit, product, and sum,
8951 which are defined thus in the library module <literal>Generics</literal>:
8955 data a :+: b = Inl a | Inr b
8956 data a :*: b = a :*: b
8959 Now you can make a data type into an instance of Bin like this:
8961 instance (Bin a, Bin b) => Bin (a,b)
8962 instance Bin a => Bin [a]
8964 That is, just leave off the "where" clause. Of course, you can put in the
8965 where clause and over-ride whichever methods you please.
8969 <title> Using generics </title>
8970 <para>To use generics you need to</para>
8973 <para>Use the flags <option>-fglasgow-exts</option> (to enable the extra syntax),
8974 <option>-XGenerics</option> (to generate extra per-data-type code),
8975 and <option>-package lang</option> (to make the <literal>Generics</literal> library
8979 <para>Import the module <literal>Generics</literal> from the
8980 <literal>lang</literal> package. This import brings into
8981 scope the data types <literal>Unit</literal>,
8982 <literal>:*:</literal>, and <literal>:+:</literal>. (You
8983 don't need this import if you don't mention these types
8984 explicitly; for example, if you are simply giving instance
8985 declarations.)</para>
8990 <sect2> <title> Changes wrt the paper </title>
8992 Note that the type constructors <literal>:+:</literal> and <literal>:*:</literal>
8993 can be written infix (indeed, you can now use
8994 any operator starting in a colon as an infix type constructor). Also note that
8995 the type constructors are not exactly as in the paper (Unit instead of 1, etc).
8996 Finally, note that the syntax of the type patterns in the class declaration
8997 uses "<literal>{|</literal>" and "<literal>|}</literal>" brackets; curly braces
8998 alone would ambiguous when they appear on right hand sides (an extension we
8999 anticipate wanting).
9003 <sect2> <title>Terminology and restrictions</title>
9005 Terminology. A "generic default method" in a class declaration
9006 is one that is defined using type patterns as above.
9007 A "polymorphic default method" is a default method defined as in Haskell 98.
9008 A "generic class declaration" is a class declaration with at least one
9009 generic default method.
9017 Alas, we do not yet implement the stuff about constructor names and
9024 A generic class can have only one parameter; you can't have a generic
9025 multi-parameter class.
9031 A default method must be defined entirely using type patterns, or entirely
9032 without. So this is illegal:
9035 op :: a -> (a, Bool)
9036 op {| Unit |} Unit = (Unit, True)
9039 However it is perfectly OK for some methods of a generic class to have
9040 generic default methods and others to have polymorphic default methods.
9046 The type variable(s) in the type pattern for a generic method declaration
9047 scope over the right hand side. So this is legal (note the use of the type variable ``p'' in a type signature on the right hand side:
9051 op {| p :*: q |} (x :*: y) = op (x :: p)
9059 The type patterns in a generic default method must take one of the forms:
9065 where "a" and "b" are type variables. Furthermore, all the type patterns for
9066 a single type constructor (<literal>:*:</literal>, say) must be identical; they
9067 must use the same type variables. So this is illegal:
9071 op {| a :+: b |} (Inl x) = True
9072 op {| p :+: q |} (Inr y) = False
9074 The type patterns must be identical, even in equations for different methods of the class.
9075 So this too is illegal:
9079 op1 {| a :*: b |} (x :*: y) = True
9082 op2 {| p :*: q |} (x :*: y) = False
9084 (The reason for this restriction is that we gather all the equations for a particular type constructor
9085 into a single generic instance declaration.)
9091 A generic method declaration must give a case for each of the three type constructors.
9097 The type for a generic method can be built only from:
9099 <listitem> <para> Function arrows </para> </listitem>
9100 <listitem> <para> Type variables </para> </listitem>
9101 <listitem> <para> Tuples </para> </listitem>
9102 <listitem> <para> Arbitrary types not involving type variables </para> </listitem>
9104 Here are some example type signatures for generic methods:
9107 op2 :: Bool -> (a,Bool)
9108 op3 :: [Int] -> a -> a
9111 Here, op1, op2, op3 are OK, but op4 is rejected, because it has a type variable
9115 This restriction is an implementation restriction: we just haven't got around to
9116 implementing the necessary bidirectional maps over arbitrary type constructors.
9117 It would be relatively easy to add specific type constructors, such as Maybe and list,
9118 to the ones that are allowed.</para>
9123 In an instance declaration for a generic class, the idea is that the compiler
9124 will fill in the methods for you, based on the generic templates. However it can only
9129 The instance type is simple (a type constructor applied to type variables, as in Haskell 98).
9134 No constructor of the instance type has unboxed fields.
9138 (Of course, these things can only arise if you are already using GHC extensions.)
9139 However, you can still give an instance declarations for types which break these rules,
9140 provided you give explicit code to override any generic default methods.
9148 The option <option>-ddump-deriv</option> dumps incomprehensible stuff giving details of
9149 what the compiler does with generic declarations.
9154 <sect2> <title> Another example </title>
9156 Just to finish with, here's another example I rather like:
9160 nCons {| Unit |} _ = 1
9161 nCons {| a :*: b |} _ = 1
9162 nCons {| a :+: b |} _ = nCons (bot::a) + nCons (bot::b)
9165 tag {| Unit |} _ = 1
9166 tag {| a :*: b |} _ = 1
9167 tag {| a :+: b |} (Inl x) = tag x
9168 tag {| a :+: b |} (Inr y) = nCons (bot::a) + tag y
9174 <sect1 id="monomorphism">
9175 <title>Control over monomorphism</title>
9177 <para>GHC supports two flags that control the way in which generalisation is
9178 carried out at let and where bindings.
9182 <title>Switching off the dreaded Monomorphism Restriction</title>
9183 <indexterm><primary><option>-XNoMonomorphismRestriction</option></primary></indexterm>
9185 <para>Haskell's monomorphism restriction (see
9186 <ulink url="http://www.haskell.org/onlinereport/decls.html#sect4.5.5">Section
9188 of the Haskell Report)
9189 can be completely switched off by
9190 <option>-XNoMonomorphismRestriction</option>.
9195 <title>Monomorphic pattern bindings</title>
9196 <indexterm><primary><option>-XNoMonoPatBinds</option></primary></indexterm>
9197 <indexterm><primary><option>-XMonoPatBinds</option></primary></indexterm>
9199 <para> As an experimental change, we are exploring the possibility of
9200 making pattern bindings monomorphic; that is, not generalised at all.
9201 A pattern binding is a binding whose LHS has no function arguments,
9202 and is not a simple variable. For example:
9204 f x = x -- Not a pattern binding
9205 f = \x -> x -- Not a pattern binding
9206 f :: Int -> Int = \x -> x -- Not a pattern binding
9208 (g,h) = e -- A pattern binding
9209 (f) = e -- A pattern binding
9210 [x] = e -- A pattern binding
9212 Experimentally, GHC now makes pattern bindings monomorphic <emphasis>by
9213 default</emphasis>. Use <option>-XNoMonoPatBinds</option> to recover the
9222 ;;; Local Variables: ***
9223 ;;; sgml-parent-document: ("users_guide.xml" "book" "chapter" "sect1") ***
9224 ;;; ispell-local-dictionary: "british" ***