2 <indexterm><primary>language, GHC</primary></indexterm>
3 <indexterm><primary>extensions, GHC</primary></indexterm>
4 As with all known Haskell systems, GHC implements some extensions to
5 the language. To use them, you'll need to give a <option>-fglasgow-exts</option>
6 <indexterm><primary>-fglasgow-exts option</primary></indexterm> option.
10 Virtually all of the Glasgow extensions serve to give you access to
11 the underlying facilities with which we implement Haskell. Thus, you
12 can get at the Raw Iron, if you are willing to write some non-standard
13 code at a more primitive level. You need not be “stuck” on
14 performance because of the implementation costs of Haskell's
15 “high-level” features—you can always code “under” them. In an extreme case, you can write all your time-critical code in C, and then just glue it together with Haskell!
19 Executive summary of our extensions:
25 <term>Unboxed types and primitive operations:</Term>
27 <para>You can get right down to the raw machine types and
28 operations; included in this are “primitive
29 arrays” (direct access to Big Wads of Bytes). Please
30 see <XRef LinkEnd="glasgow-unboxed"> and following.</para>
35 <term>Type system extensions:</term>
37 <para> GHC supports a large number of extensions to Haskell's
38 type system. Specifically:</para>
42 <term>Multi-parameter type classes:</term>
44 <para><xref LinkEnd="multi-param-type-classes"></para>
49 <term>Functional dependencies:</term>
51 <para><xref LinkEnd="functional-dependencies"></para>
56 <term>Implicit parameters:</term>
58 <para><xref LinkEnd="implicit-parameters"></para>
63 <term>Local universal quantification:</term>
65 <para><xref LinkEnd="universal-quantification"></para>
70 <term>Extistentially quantification in data types:</term>
72 <para><xref LinkEnd="existential-quantification"></para>
77 <term>Scoped type variables:</term>
79 <para>Scoped type variables enable the programmer to
80 supply type signatures for some nested declarations,
81 where this would not be legal in Haskell 98. Details in
82 <xref LinkEnd="scoped-type-variables">.</para>
90 <term>Pattern guards</term>
92 <para>Instead of being a boolean expression, a guard is a list
93 of qualifiers, exactly as in a list comprehension. See <xref
94 LinkEnd="pattern-guards">.</para>
99 <term>Data types with no constructors</term>
101 <para>See <xref LinkEnd="nullary-types">.</para>
106 <term>Parallel list comprehensions</term>
108 <para>An extension to the list comprehension syntax to support
109 <literal>zipWith</literal>-like functionality. See <xref
110 linkend="parallel-list-comprehensions">.</para>
115 <term>Foreign calling:</term>
117 <para>Just what it sounds like. We provide
118 <emphasis>lots</emphasis> of rope that you can dangle around
119 your neck. Please see <xref LinkEnd="ffi">.</para>
126 <para>Pragmas are special instructions to the compiler placed
127 in the source file. The pragmas GHC supports are described in
128 <xref LinkEnd="pragmas">.</para>
133 <term>Rewrite rules:</term>
135 <para>The programmer can specify rewrite rules as part of the
136 source program (in a pragma). GHC applies these rewrite rules
137 wherever it can. Details in <xref
138 LinkEnd="rewrite-rules">.</para>
143 <term>Generic classes:</term>
145 <para>Generic class declarations allow you to define a class
146 whose methods say how to work over an arbitrary data type.
147 Then it's really easy to make any new type into an instance of
148 the class. This generalises the rather ad-hoc "deriving"
149 feature of Haskell 98. Details in <xref
150 LinkEnd="generic-classes">.</para>
156 Before you get too carried away working at the lowest level (e.g.,
157 sloshing <literal>MutableByteArray#</literal>s around your
158 program), you may wish to check if there are libraries that provide a
159 “Haskellised veneer” over the features you want. See
160 <xref linkend="book-hslibs">.
163 <sect1 id="options-language">
164 <title>Language options</title>
166 <indexterm><primary>language</primary><secondary>option</secondary>
168 <indexterm><primary>options</primary><secondary>language</secondary>
170 <indexterm><primary>extensions</primary><secondary>options controlling</secondary>
173 <para> These flags control what variation of the language are
174 permitted. Leaving out all of them gives you standard Haskell
180 <term><option>-fglasgow-exts</option>:</term>
181 <indexterm><primary><option>-fglasgow-exts</option></primary></indexterm>
183 <para>This simultaneously enables all of the extensions to
184 Haskell 98 described in <xref
185 linkend="ghc-language-features">, except where otherwise
191 <term><option>-fno-monomorphism-restriction</option>:</term>
192 <indexterm><primary><option>-fno-monomorphism-restriction</option></primary></indexterm>
194 <para> Switch off the Haskell 98 monomorphism restriction.
195 Independent of the <option>-fglasgow-exts</option>
201 <term><option>-fallow-overlapping-instances</option></term>
202 <term><option>-fallow-undecidable-instances</option></term>
203 <term><option>-fcontext-stack</option></term>
204 <indexterm><primary><option>-fallow-overlapping-instances</option></primary></indexterm>
205 <indexterm><primary><option>-fallow-undecidable-instances</option></primary></indexterm>
206 <indexterm><primary><option>-fcontext-stack</option></primary></indexterm>
208 <para> See <xref LinkEnd="instance-decls">. Only relevant
209 if you also use <option>-fglasgow-exts</option>.</para>
214 <term><option>-finline-phase</option></term>
215 <indexterm><primary><option>-finline-phase</option></primary></indexterm>
217 <para>See <xref LinkEnd="rewrite-rules">. Only relevant if
218 you also use <option>-fglasgow-exts</option>.</para>
223 <term><option>-fgenerics</option></term>
224 <indexterm><primary><option>-fgenerics</option></primary></indexterm>
226 <para>See <xref LinkEnd="generic-classes">. Independent of
227 <option>-fglasgow-exts</option>.</para>
232 <term><option>-fno-implicit-prelude</option></term>
234 <para><indexterm><primary>-fno-implicit-prelude
235 option</primary></indexterm> GHC normally imports
236 <filename>Prelude.hi</filename> files for you. If you'd
237 rather it didn't, then give it a
238 <option>-fno-implicit-prelude</option> option. The idea
239 is that you can then import a Prelude of your own. (But
240 don't call it <literal>Prelude</literal>; the Haskell
241 module namespace is flat, and you must not conflict with
242 any Prelude module.)</para>
244 <para>Even though you have not imported the Prelude, all
245 the built-in syntax still refers to the built-in Haskell
246 Prelude types and values, as specified by the Haskell
247 Report. For example, the type <literal>[Int]</literal>
248 still means <literal>Prelude.[] Int</literal>; tuples
249 continue to refer to the standard Prelude tuples; the
250 translation for list comprehensions continues to use
251 <literal>Prelude.map</literal> etc.</para>
253 <para> With one group of exceptions! You may want to
254 define your own numeric class hierarchy. It completely
255 defeats that purpose if the literal "1" means
256 "<literal>Prelude.fromInteger 1</literal>", which is what
257 the Haskell Report specifies. So the
258 <option>-fno-implicit-prelude</option> flag causes the
259 following pieces of built-in syntax to refer to <emphasis>whatever
260 is in scope</emphasis>, not the Prelude versions:</para>
264 <para>Integer and fractional literals mean
265 "<literal>fromInteger 1</literal>" and
266 "<literal>fromRational 3.2</literal>", not the
267 Prelude-qualified versions; both in expressions and in
272 <para>Negation (e.g. "<literal>- (f x)</literal>")
273 means "<literal>negate (f x)</literal>" (not
274 <literal>Prelude.negate</literal>).</para>
278 <para>In an n+k pattern, the standard Prelude
279 <literal>Ord</literal> class is still used for comparison,
280 but the necessary subtraction uses whatever
281 "<literal>(-)</literal>" is in scope (not
282 "<literal>Prelude.(-)</literal>").</para>
286 <para>Note: Negative literals, such as <literal>-3</literal>, are
287 specified by (a careful reading of) the Haskell Report as
288 meaning <literal>Prelude.negate (Prelude.fromInteger 3)</literal>.
289 However, GHC deviates from this slightly, and treats them as meaning
290 <literal>fromInteger (-3)</literal>. One particular effect of this
291 slightly-non-standard reading is that there is no difficulty with
292 the literal <literal>-2147483648</literal> at type <literal>Int</literal>;
293 it means <literal>fromInteger (-2147483648)</literal>. The strict interpretation
294 would be <literal>negate (fromInteger 2147483648)</literal>,
295 and the call to <literal>fromInteger</literal> would overflow
296 (at type <literal>Int</literal>, remember).
305 <!-- UNBOXED TYPES AND PRIMITIVE OPERATIONS -->
308 <sect1 id="glasgow-ST-monad">
309 <title>Primitive state-transformer monad</title>
312 <indexterm><primary>state transformers (Glasgow extensions)</primary></indexterm>
313 <indexterm><primary>ST monad (Glasgow extension)</primary></indexterm>
317 This monad underlies our implementation of arrays, mutable and
318 immutable, and our implementation of I/O, including “C calls”.
322 The <literal>ST</literal> library, which provides access to the
323 <function>ST</function> monad, is described in <xref
329 <sect1 id="glasgow-prim-arrays">
330 <title>Primitive arrays, mutable and otherwise
334 <indexterm><primary>primitive arrays (Glasgow extension)</primary></indexterm>
335 <indexterm><primary>arrays, primitive (Glasgow extension)</primary></indexterm>
339 GHC knows about quite a few flavours of Large Swathes of Bytes.
343 First, GHC distinguishes between primitive arrays of (boxed) Haskell
344 objects (type <literal>Array# obj</literal>) and primitive arrays of bytes (type
345 <literal>ByteArray#</literal>).
349 Second, it distinguishes between…
353 <term>Immutable:</term>
356 Arrays that do not change (as with “standard” Haskell arrays); you
357 can only read from them. Obviously, they do not need the care and
358 attention of the state-transformer monad.
363 <term>Mutable:</term>
366 Arrays that may be changed or “mutated.” All the operations on them
367 live within the state-transformer monad and the updates happen
368 <emphasis>in-place</emphasis>.
373 <term>“Static” (in C land):</term>
376 A C routine may pass an <literal>Addr#</literal> pointer back into Haskell land. There
377 are then primitive operations with which you may merrily grab values
378 over in C land, by indexing off the “static” pointer.
383 <term>“Stable” pointers:</term>
386 If, for some reason, you wish to hand a Haskell pointer (i.e.,
387 <emphasis>not</emphasis> an unboxed value) to a C routine, you first make the
388 pointer “stable,” so that the garbage collector won't forget that it
389 exists. That is, GHC provides a safe way to pass Haskell pointers to
394 Please see <xref LinkEnd="sec-stable-pointers"> for more details.
399 <term>“Foreign objects”:</term>
402 A “foreign object” is a safe way to pass an external object (a
403 C-allocated pointer, say) to Haskell and have Haskell do the Right
404 Thing when it no longer references the object. So, for example, C
405 could pass a large bitmap over to Haskell and say “please free this
406 memory when you're done with it.”
410 Please see <xref LinkEnd="sec-ForeignObj"> for more details.
418 The libraries documentatation gives more details on all these
419 “primitive array” types and the operations on them.
425 <sect1 id="nullary-types">
426 <title>Data types with no constructors</title>
428 <para>With the <option>-fglasgow-exts</option> flag, GHC lets you declare
429 a data type with no constructors. For example:</para>
432 data T a -- T :: * -> *
434 <para>Syntactically, the declaration lacks the "= constrs" part. The
435 type can be parameterised, but only over ordinary types, of kind *; since
436 Haskell does not have kind signatures, you cannot parameterise over higher-kinded
439 <para>Such data types have only one value, namely bottom.
440 Nevertheless, they can be useful when defining "phantom types".</para>
443 <sect1 id="pattern-guards">
444 <title>Pattern guards</title>
447 <indexterm><primary>Pattern guards (Glasgow extension)</primary></indexterm>
448 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.)
452 Suppose we have an abstract data type of finite maps, with a
456 lookup :: FiniteMap -> Int -> Maybe Int
459 The lookup returns <function>Nothing</function> if the supplied key is not in the domain of the mapping, and <function>(Just v)</function> otherwise,
460 where <VarName>v</VarName> is the value that the key maps to. Now consider the following definition:
464 clunky env var1 var2 | ok1 && ok2 = val1 + val2
465 | otherwise = var1 + var2
476 The auxiliary functions are
480 maybeToBool :: Maybe a -> Bool
481 maybeToBool (Just x) = True
482 maybeToBool Nothing = False
484 expectJust :: Maybe a -> a
485 expectJust (Just x) = x
486 expectJust Nothing = error "Unexpected Nothing"
490 What is <function>clunky</function> doing? The guard <literal>ok1 &&
491 ok2</literal> checks that both lookups succeed, using
492 <function>maybeToBool</function> to convert the <function>Maybe</function>
493 types to booleans. The (lazily evaluated) <function>expectJust</function>
494 calls extract the values from the results of the lookups, and binds the
495 returned values to <VarName>val1</VarName> and <VarName>val2</VarName>
496 respectively. If either lookup fails, then clunky takes the
497 <literal>otherwise</literal> case and returns the sum of its arguments.
501 This is certainly legal Haskell, but it is a tremendously verbose and
502 un-obvious way to achieve the desired effect. Arguably, a more direct way
503 to write clunky would be to use case expressions:
507 clunky env var1 var1 = case lookup env var1 of
509 Just val1 -> case lookup env var2 of
511 Just val2 -> val1 + val2
517 This is a bit shorter, but hardly better. Of course, we can rewrite any set
518 of pattern-matching, guarded equations as case expressions; that is
519 precisely what the compiler does when compiling equations! The reason that
520 Haskell provides guarded equations is because they allow us to write down
521 the cases we want to consider, one at a time, independently of each other.
522 This structure is hidden in the case version. Two of the right-hand sides
523 are really the same (<function>fail</function>), and the whole expression
524 tends to become more and more indented.
528 Here is how I would write clunky:
533 | Just val1 <- lookup env var1
534 , Just val2 <- lookup env var2
536 ...other equations for clunky...
540 The semantics should be clear enough. The qualifers are matched in order.
541 For a <literal><-</literal> qualifier, which I call a pattern guard, the
542 right hand side is evaluated and matched against the pattern on the left.
543 If the match fails then the whole guard fails and the next equation is
544 tried. If it succeeds, then the appropriate binding takes place, and the
545 next qualifier is matched, in the augmented environment. Unlike list
546 comprehensions, however, the type of the expression to the right of the
547 <literal><-</literal> is the same as the type of the pattern to its
548 left. The bindings introduced by pattern guards scope over all the
549 remaining guard qualifiers, and over the right hand side of the equation.
553 Just as with list comprehensions, boolean expressions can be freely mixed
554 with among the pattern guards. For example:
565 Haskell's current guards therefore emerge as a special case, in which the
566 qualifier list has just one element, a boolean expression.
570 <sect1 id="parallel-list-comprehensions">
571 <title>Parallel List Comprehensions</title>
572 <indexterm><primary>list comprehensions</primary><secondary>parallel</secondary>
574 <indexterm><primary>parallel list comprehensions</primary>
577 <para>Parallel list comprehensions are a natural extension to list
578 comprehensions. List comprehensions can be thought of as a nice
579 syntax for writing maps and filters. Parallel comprehensions
580 extend this to include the zipWith family.</para>
582 <para>A parallel list comprehension has multiple independent
583 branches of qualifier lists, each separated by a `|' symbol. For
584 example, the following zips together two lists:</para>
587 [ (x, y) | x <- xs | y <- ys ]
590 <para>The behavior of parallel list comprehensions follows that of
591 zip, in that the resulting list will have the same length as the
592 shortest branch.</para>
594 <para>We can define parallel list comprehensions by translation to
595 regular comprehensions. Here's the basic idea:</para>
597 <para>Given a parallel comprehension of the form: </para>
600 [ e | p1 <- e11, p2 <- e12, ...
601 | q1 <- e21, q2 <- e22, ...
606 <para>This will be translated to: </para>
609 [ e | ((p1,p2), (q1,q2), ...) <- zipN [(p1,p2) | p1 <- e11, p2 <- e12, ...]
610 [(q1,q2) | q1 <- e21, q2 <- e22, ...]
615 <para>where `zipN' is the appropriate zip for the given number of
620 <sect1 id="multi-param-type-classes">
621 <title>Multi-parameter type classes
625 This section documents GHC's implementation of multi-parameter type
626 classes. There's lots of background in the paper <ULink
627 URL="http://research.microsoft.com/~simonpj/multi.ps.gz" >Type
628 classes: exploring the design space</ULink > (Simon Peyton Jones, Mark
633 I'd like to thank people who reported shorcomings in the GHC 3.02
634 implementation. Our default decisions were all conservative ones, and
635 the experience of these heroic pioneers has given useful concrete
636 examples to support several generalisations. (These appear below as
637 design choices not implemented in 3.02.)
641 I've discussed these notes with Mark Jones, and I believe that Hugs
642 will migrate towards the same design choices as I outline here.
643 Thanks to him, and to many others who have offered very useful
651 There are the following restrictions on the form of a qualified
658 forall tv1..tvn (c1, ...,cn) => type
664 (Here, I write the "foralls" explicitly, although the Haskell source
665 language omits them; in Haskell 1.4, all the free type variables of an
666 explicit source-language type signature are universally quantified,
667 except for the class type variables in a class declaration. However,
668 in GHC, you can give the foralls if you want. See <xref LinkEnd="universal-quantification">).
677 <emphasis>Each universally quantified type variable
678 <literal>tvi</literal> must be mentioned (i.e. appear free) in <literal>type</literal></emphasis>.
680 The reason for this is that a value with a type that does not obey
681 this restriction could not be used without introducing
682 ambiguity. Here, for example, is an illegal type:
686 forall a. Eq a => Int
690 When a value with this type was used, the constraint <literal>Eq tv</literal>
691 would be introduced where <literal>tv</literal> is a fresh type variable, and
692 (in the dictionary-translation implementation) the value would be
693 applied to a dictionary for <literal>Eq tv</literal>. The difficulty is that we
694 can never know which instance of <literal>Eq</literal> to use because we never
695 get any more information about <literal>tv</literal>.
702 <emphasis>Every constraint <literal>ci</literal> must mention at least one of the
703 universally quantified type variables <literal>tvi</literal></emphasis>.
705 For example, this type is OK because <literal>C a b</literal> mentions the
706 universally quantified type variable <literal>b</literal>:
710 forall a. C a b => burble
714 The next type is illegal because the constraint <literal>Eq b</literal> does not
715 mention <literal>a</literal>:
719 forall a. Eq b => burble
723 The reason for this restriction is milder than the other one. The
724 excluded types are never useful or necessary (because the offending
725 context doesn't need to be witnessed at this point; it can be floated
726 out). Furthermore, floating them out increases sharing. Lastly,
727 excluding them is a conservative choice; it leaves a patch of
728 territory free in case we need it later.
738 These restrictions apply to all types, whether declared in a type signature
743 Unlike Haskell 1.4, constraints in types do <emphasis>not</emphasis> have to be of
744 the form <emphasis>(class type-variables)</emphasis>. Thus, these type signatures
751 f :: Eq (m a) => [m a] -> [m a]
758 This choice recovers principal types, a property that Haskell 1.4 does not have.
764 <title>Class declarations</title>
772 <emphasis>Multi-parameter type classes are permitted</emphasis>. For example:
776 class Collection c a where
777 union :: c a -> c a -> c a
788 <emphasis>The class hierarchy must be acyclic</emphasis>. However, the definition
789 of "acyclic" involves only the superclass relationships. For example,
795 op :: D b => a -> b -> b
798 class C a => D a where { ... }
802 Here, <literal>C</literal> is a superclass of <literal>D</literal>, but it's OK for a
803 class operation <literal>op</literal> of <literal>C</literal> to mention <literal>D</literal>. (It
804 would not be OK for <literal>D</literal> to be a superclass of <literal>C</literal>.)
811 <emphasis>There are no restrictions on the context in a class declaration
812 (which introduces superclasses), except that the class hierarchy must
813 be acyclic</emphasis>. So these class declarations are OK:
817 class Functor (m k) => FiniteMap m k where
820 class (Monad m, Monad (t m)) => Transform t m where
821 lift :: m a -> (t m) a
830 <emphasis>In the signature of a class operation, every constraint
831 must mention at least one type variable that is not a class type
838 class Collection c a where
839 mapC :: Collection c b => (a->b) -> c a -> c b
843 is OK because the constraint <literal>(Collection a b)</literal> mentions
844 <literal>b</literal>, even though it also mentions the class variable
845 <literal>a</literal>. On the other hand:
850 op :: Eq a => (a,b) -> (a,b)
854 is not OK because the constraint <literal>(Eq a)</literal> mentions on the class
855 type variable <literal>a</literal>, but not <literal>b</literal>. However, any such
856 example is easily fixed by moving the offending context up to the
861 class Eq a => C a where
866 A yet more relaxed rule would allow the context of a class-op signature
867 to mention only class type variables. However, that conflicts with
868 Rule 1(b) for types above.
875 <emphasis>The type of each class operation must mention <emphasis>all</emphasis> of
876 the class type variables</emphasis>. For example:
882 insert :: s -> a -> s
886 is not OK, because the type of <literal>empty</literal> doesn't mention
887 <literal>a</literal>. This rule is a consequence of Rule 1(a), above, for
888 types, and has the same motivation.
890 Sometimes, offending class declarations exhibit misunderstandings. For
891 example, <literal>Coll</literal> might be rewritten
897 insert :: s a -> a -> s a
901 which makes the connection between the type of a collection of
902 <literal>a</literal>'s (namely <literal>(s a)</literal>) and the element type <literal>a</literal>.
903 Occasionally this really doesn't work, in which case you can split the
911 class CollE s => Coll s a where
912 insert :: s -> a -> s
925 <sect2 id="instance-decls">
926 <title>Instance declarations</title>
934 <emphasis>Instance declarations may not overlap</emphasis>. The two instance
939 instance context1 => C type1 where ...
940 instance context2 => C type2 where ...
944 "overlap" if <literal>type1</literal> and <literal>type2</literal> unify
946 However, if you give the command line option
947 <option>-fallow-overlapping-instances</option><indexterm><primary>-fallow-overlapping-instances
948 option</primary></indexterm> then two overlapping instance declarations are permitted
956 EITHER <literal>type1</literal> and <literal>type2</literal> do not unify
962 OR <literal>type2</literal> is a substitution instance of <literal>type1</literal>
963 (but not identical to <literal>type1</literal>)
976 Notice that these rules
983 make it clear which instance decl to use
984 (pick the most specific one that matches)
991 do not mention the contexts <literal>context1</literal>, <literal>context2</literal>
992 Reason: you can pick which instance decl
993 "matches" based on the type.
1000 Regrettably, GHC doesn't guarantee to detect overlapping instance
1001 declarations if they appear in different modules. GHC can "see" the
1002 instance declarations in the transitive closure of all the modules
1003 imported by the one being compiled, so it can "see" all instance decls
1004 when it is compiling <literal>Main</literal>. However, it currently chooses not
1005 to look at ones that can't possibly be of use in the module currently
1006 being compiled, in the interests of efficiency. (Perhaps we should
1007 change that decision, at least for <literal>Main</literal>.)
1014 <emphasis>There are no restrictions on the type in an instance
1015 <emphasis>head</emphasis>, except that at least one must not be a type variable</emphasis>.
1016 The instance "head" is the bit after the "=>" in an instance decl. For
1017 example, these are OK:
1021 instance C Int a where ...
1023 instance D (Int, Int) where ...
1025 instance E [[a]] where ...
1029 Note that instance heads <emphasis>may</emphasis> contain repeated type variables.
1030 For example, this is OK:
1034 instance Stateful (ST s) (MutVar s) where ...
1038 The "at least one not a type variable" restriction is to ensure that
1039 context reduction terminates: each reduction step removes one type
1040 constructor. For example, the following would make the type checker
1041 loop if it wasn't excluded:
1045 instance C a => C a where ...
1049 There are two situations in which the rule is a bit of a pain. First,
1050 if one allows overlapping instance declarations then it's quite
1051 convenient to have a "default instance" declaration that applies if
1052 something more specific does not:
1061 Second, sometimes you might want to use the following to get the
1062 effect of a "class synonym":
1066 class (C1 a, C2 a, C3 a) => C a where { }
1068 instance (C1 a, C2 a, C3 a) => C a where { }
1072 This allows you to write shorter signatures:
1084 f :: (C1 a, C2 a, C3 a) => ...
1088 I'm on the lookout for a simple rule that preserves decidability while
1089 allowing these idioms. The experimental flag
1090 <option>-fallow-undecidable-instances</option><indexterm><primary>-fallow-undecidable-instances
1091 option</primary></indexterm> lifts this restriction, allowing all the types in an
1092 instance head to be type variables.
1099 <emphasis>Unlike Haskell 1.4, instance heads may use type
1100 synonyms</emphasis>. As always, using a type synonym is just shorthand for
1101 writing the RHS of the type synonym definition. For example:
1105 type Point = (Int,Int)
1106 instance C Point where ...
1107 instance C [Point] where ...
1111 is legal. However, if you added
1115 instance C (Int,Int) where ...
1119 as well, then the compiler will complain about the overlapping
1120 (actually, identical) instance declarations. As always, type synonyms
1121 must be fully applied. You cannot, for example, write:
1126 instance Monad P where ...
1130 This design decision is independent of all the others, and easily
1131 reversed, but it makes sense to me.
1138 <emphasis>The types in an instance-declaration <emphasis>context</emphasis> must all
1139 be type variables</emphasis>. Thus
1143 instance C a b => Eq (a,b) where ...
1151 instance C Int b => Foo b where ...
1155 is not OK. Again, the intent here is to make sure that context
1156 reduction terminates.
1158 Voluminous correspondence on the Haskell mailing list has convinced me
1159 that it's worth experimenting with a more liberal rule. If you use
1160 the flag <option>-fallow-undecidable-instances</option> can use arbitrary
1161 types in an instance context. Termination is ensured by having a
1162 fixed-depth recursion stack. If you exceed the stack depth you get a
1163 sort of backtrace, and the opportunity to increase the stack depth
1164 with <option>-fcontext-stack</option><emphasis>N</emphasis>.
1177 <sect1 id="implicit-parameters">
1178 <title>Implicit parameters
1181 <para> Implicit paramters are implemented as described in
1182 "Implicit parameters: dynamic scoping with static types",
1183 J Lewis, MB Shields, E Meijer, J Launchbury,
1184 27th ACM Symposium on Principles of Programming Languages (POPL'00),
1189 There should be more documentation, but there isn't (yet). Yell if you need it.
1193 <para> You can't have an implicit parameter in the context of a class or instance
1194 declaration. For example, both these declarations are illegal:
1196 class (?x::Int) => C a where ...
1197 instance (?x::a) => Foo [a] where ...
1199 Reason: exactly which implicit parameter you pick up depends on exactly where
1200 you invoke a function. But the ``invocation'' of instance declarations is done
1201 behind the scenes by the compiler, so it's hard to figure out exactly where it is done.
1202 Easiest thing is to outlaw the offending types.</para>
1210 <sect1 id="functional-dependencies">
1211 <title>Functional dependencies
1214 <para> Functional dependencies are implemented as described by Mark Jones
1215 in "Type Classes with Functional Dependencies", Mark P. Jones,
1216 In Proceedings of the 9th European Symposium on Programming,
1217 ESOP 2000, Berlin, Germany, March 2000, Springer-Verlag LNCS 1782.
1221 There should be more documentation, but there isn't (yet). Yell if you need it.
1226 <sect1 id="universal-quantification">
1227 <title>Explicit universal quantification
1231 GHC's type system supports explicit universal quantification in
1232 constructor fields and function arguments. This is useful for things
1233 like defining <literal>runST</literal> from the state-thread world.
1234 GHC's syntax for this now agrees with Hugs's, namely:
1240 forall a b. (Ord a, Eq b) => a -> b -> a
1246 The context is, of course, optional. You can't use <literal>forall</literal> as
1247 a type variable any more!
1251 Haskell type signatures are implicitly quantified. The <literal>forall</literal>
1252 allows us to say exactly what this means. For example:
1270 g :: forall b. (b -> b)
1276 The two are treated identically.
1280 <title>Universally-quantified data type fields
1284 In a <literal>data</literal> or <literal>newtype</literal> declaration one can quantify
1285 the types of the constructor arguments. Here are several examples:
1291 data T a = T1 (forall b. b -> b -> b) a
1293 data MonadT m = MkMonad { return :: forall a. a -> m a,
1294 bind :: forall a b. m a -> (a -> m b) -> m b
1297 newtype Swizzle = MkSwizzle (Ord a => [a] -> [a])
1303 The constructors now have so-called <emphasis>rank 2</emphasis> polymorphic
1304 types, in which there is a for-all in the argument types.:
1310 T1 :: forall a. (forall b. b -> b -> b) -> a -> T a
1311 MkMonad :: forall m. (forall a. a -> m a)
1312 -> (forall a b. m a -> (a -> m b) -> m b)
1314 MkSwizzle :: (Ord a => [a] -> [a]) -> Swizzle
1320 Notice that you don't need to use a <literal>forall</literal> if there's an
1321 explicit context. For example in the first argument of the
1322 constructor <function>MkSwizzle</function>, an implicit "<literal>forall a.</literal>" is
1323 prefixed to the argument type. The implicit <literal>forall</literal>
1324 quantifies all type variables that are not already in scope, and are
1325 mentioned in the type quantified over.
1329 As for type signatures, implicit quantification happens for non-overloaded
1330 types too. So if you write this:
1333 data T a = MkT (Either a b) (b -> b)
1336 it's just as if you had written this:
1339 data T a = MkT (forall b. Either a b) (forall b. b -> b)
1342 That is, since the type variable <literal>b</literal> isn't in scope, it's
1343 implicitly universally quantified. (Arguably, it would be better
1344 to <emphasis>require</emphasis> explicit quantification on constructor arguments
1345 where that is what is wanted. Feedback welcomed.)
1351 <title>Construction </title>
1354 You construct values of types <literal>T1, MonadT, Swizzle</literal> by applying
1355 the constructor to suitable values, just as usual. For example,
1361 (T1 (\xy->x) 3) :: T Int
1363 (MkSwizzle sort) :: Swizzle
1364 (MkSwizzle reverse) :: Swizzle
1371 MkMonad r b) :: MonadT Maybe
1377 The type of the argument can, as usual, be more general than the type
1378 required, as <literal>(MkSwizzle reverse)</literal> shows. (<function>reverse</function>
1379 does not need the <literal>Ord</literal> constraint.)
1385 <title>Pattern matching</title>
1388 When you use pattern matching, the bound variables may now have
1389 polymorphic types. For example:
1395 f :: T a -> a -> (a, Char)
1396 f (T1 f k) x = (f k x, f 'c' 'd')
1398 g :: (Ord a, Ord b) => Swizzle -> [a] -> (a -> b) -> [b]
1399 g (MkSwizzle s) xs f = s (map f (s xs))
1401 h :: MonadT m -> [m a] -> m [a]
1402 h m [] = return m []
1403 h m (x:xs) = bind m x $ \y ->
1404 bind m (h m xs) $ \ys ->
1411 In the function <function>h</function> we use the record selectors <literal>return</literal>
1412 and <literal>bind</literal> to extract the polymorphic bind and return functions
1413 from the <literal>MonadT</literal> data structure, rather than using pattern
1418 You cannot pattern-match against an argument that is polymorphic.
1422 newtype TIM s a = TIM (ST s (Maybe a))
1424 runTIM :: (forall s. TIM s a) -> Maybe a
1425 runTIM (TIM m) = runST m
1431 Here the pattern-match fails, because you can't pattern-match against
1432 an argument of type <literal>(forall s. TIM s a)</literal>. Instead you
1433 must bind the variable and pattern match in the right hand side:
1436 runTIM :: (forall s. TIM s a) -> Maybe a
1437 runTIM tm = case tm of { TIM m -> runST m }
1440 The <literal>tm</literal> on the right hand side is (invisibly) instantiated, like
1441 any polymorphic value at its occurrence site, and now you can pattern-match
1448 <title>The partial-application restriction</title>
1451 There is really only one way in which data structures with polymorphic
1452 components might surprise you: you must not partially apply them.
1453 For example, this is illegal:
1459 map MkSwizzle [sort, reverse]
1465 The restriction is this: <emphasis>every subexpression of the program must
1466 have a type that has no for-alls, except that in a function
1467 application (f e1…en) the partial applications are not subject to
1468 this rule</emphasis>. The restriction makes type inference feasible.
1472 In the illegal example, the sub-expression <literal>MkSwizzle</literal> has the
1473 polymorphic type <literal>(Ord b => [b] -> [b]) -> Swizzle</literal> and is not
1474 a sub-expression of an enclosing application. On the other hand, this
1481 map (T1 (\a b -> a)) [1,2,3]
1487 even though it involves a partial application of <function>T1</function>, because
1488 the sub-expression <literal>T1 (\a b -> a)</literal> has type <literal>Int -> T
1495 <title>Type signatures
1499 Once you have data constructors with universally-quantified fields, or
1500 constants such as <Constant>runST</Constant> that have rank-2 types, it isn't long
1501 before you discover that you need more! Consider:
1507 mkTs f x y = [T1 f x, T1 f y]
1513 <function>mkTs</function> is a fuction that constructs some values of type
1514 <literal>T</literal>, using some pieces passed to it. The trouble is that since
1515 <literal>f</literal> is a function argument, Haskell assumes that it is
1516 monomorphic, so we'll get a type error when applying <function>T1</function> to
1517 it. This is a rather silly example, but the problem really bites in
1518 practice. Lots of people trip over the fact that you can't make
1519 "wrappers functions" for <Constant>runST</Constant> for exactly the same reason.
1520 In short, it is impossible to build abstractions around functions with
1525 The solution is fairly clear. We provide the ability to give a rank-2
1526 type signature for <emphasis>ordinary</emphasis> functions (not only data
1527 constructors), thus:
1533 mkTs :: (forall b. b -> b -> b) -> a -> [T a]
1534 mkTs f x y = [T1 f x, T1 f y]
1540 This type signature tells the compiler to attribute <literal>f</literal> with
1541 the polymorphic type <literal>(forall b. b -> b -> b)</literal> when type
1542 checking the body of <function>mkTs</function>, so now the application of
1543 <function>T1</function> is fine.
1547 There are two restrictions:
1556 You can only define a rank 2 type, specified by the following
1561 rank2type ::= [forall tyvars .] [context =>] funty
1562 funty ::= ([forall tyvars .] [context =>] ty) -> funty
1564 ty ::= ...current Haskell monotype syntax...
1568 Informally, the universal quantification must all be right at the beginning,
1569 or at the top level of a function argument.
1576 There is a restriction on the definition of a function whose
1577 type signature is a rank-2 type: the polymorphic arguments must be
1578 matched on the left hand side of the "<literal>=</literal>" sign. You can't
1579 define <function>mkTs</function> like this:
1583 mkTs :: (forall b. b -> b -> b) -> a -> [T a]
1584 mkTs = \ f x y -> [T1 f x, T1 f y]
1589 The same partial-application rule applies to ordinary functions with
1590 rank-2 types as applied to data constructors.
1603 <title>Type synonyms and hoisting
1607 GHC also allows you to write a <literal>forall</literal> in a type synonym, thus:
1609 type Discard a = forall b. a -> b -> a
1614 However, it is often convenient to use these sort of synonyms at the right hand
1615 end of an arrow, thus:
1617 type Discard a = forall b. a -> b -> a
1619 g :: Int -> Discard Int
1622 Simply expanding the type synonym would give
1624 g :: Int -> (forall b. Int -> b -> Int)
1626 but GHC "hoists" the <literal>forall</literal> to give the isomorphic type
1628 g :: forall b. Int -> Int -> b -> Int
1630 In general, the rule is this: <emphasis>to determine the type specified by any explicit
1631 user-written type (e.g. in a type signature), GHC expands type synonyms and then repeatedly
1632 performs the transformation:</emphasis>
1634 <emphasis>type1</emphasis> -> forall a. <emphasis>type2</emphasis>
1636 forall a. <emphasis>type1</emphasis> -> <emphasis>type2</emphasis>
1638 (In fact, GHC tries to retain as much synonym information as possible for use in
1639 error messages, but that is a usability issue.) This rule applies, of course, whether
1640 or not the <literal>forall</literal> comes from a synonym. For example, here is another
1641 valid way to write <literal>g</literal>'s type signature:
1643 g :: Int -> Int -> forall b. b -> Int
1650 <sect1 id="existential-quantification">
1651 <title>Existentially quantified data constructors
1655 The idea of using existential quantification in data type declarations
1656 was suggested by Laufer (I believe, thought doubtless someone will
1657 correct me), and implemented in Hope+. It's been in Lennart
1658 Augustsson's <Command>hbc</Command> Haskell compiler for several years, and
1659 proved very useful. Here's the idea. Consider the declaration:
1665 data Foo = forall a. MkFoo a (a -> Bool)
1672 The data type <literal>Foo</literal> has two constructors with types:
1678 MkFoo :: forall a. a -> (a -> Bool) -> Foo
1685 Notice that the type variable <literal>a</literal> in the type of <function>MkFoo</function>
1686 does not appear in the data type itself, which is plain <literal>Foo</literal>.
1687 For example, the following expression is fine:
1693 [MkFoo 3 even, MkFoo 'c' isUpper] :: [Foo]
1699 Here, <literal>(MkFoo 3 even)</literal> packages an integer with a function
1700 <function>even</function> that maps an integer to <literal>Bool</literal>; and <function>MkFoo 'c'
1701 isUpper</function> packages a character with a compatible function. These
1702 two things are each of type <literal>Foo</literal> and can be put in a list.
1706 What can we do with a value of type <literal>Foo</literal>?. In particular,
1707 what happens when we pattern-match on <function>MkFoo</function>?
1713 f (MkFoo val fn) = ???
1719 Since all we know about <literal>val</literal> and <function>fn</function> is that they
1720 are compatible, the only (useful) thing we can do with them is to
1721 apply <function>fn</function> to <literal>val</literal> to get a boolean. For example:
1728 f (MkFoo val fn) = fn val
1734 What this allows us to do is to package heterogenous values
1735 together with a bunch of functions that manipulate them, and then treat
1736 that collection of packages in a uniform manner. You can express
1737 quite a bit of object-oriented-like programming this way.
1740 <sect2 id="existential">
1741 <title>Why existential?
1745 What has this to do with <emphasis>existential</emphasis> quantification?
1746 Simply that <function>MkFoo</function> has the (nearly) isomorphic type
1752 MkFoo :: (exists a . (a, a -> Bool)) -> Foo
1758 But Haskell programmers can safely think of the ordinary
1759 <emphasis>universally</emphasis> quantified type given above, thereby avoiding
1760 adding a new existential quantification construct.
1766 <title>Type classes</title>
1769 An easy extension (implemented in <Command>hbc</Command>) is to allow
1770 arbitrary contexts before the constructor. For example:
1776 data Baz = forall a. Eq a => Baz1 a a
1777 | forall b. Show b => Baz2 b (b -> b)
1783 The two constructors have the types you'd expect:
1789 Baz1 :: forall a. Eq a => a -> a -> Baz
1790 Baz2 :: forall b. Show b => b -> (b -> b) -> Baz
1796 But when pattern matching on <function>Baz1</function> the matched values can be compared
1797 for equality, and when pattern matching on <function>Baz2</function> the first matched
1798 value can be converted to a string (as well as applying the function to it).
1799 So this program is legal:
1806 f (Baz1 p q) | p == q = "Yes"
1808 f (Baz1 v fn) = show (fn v)
1814 Operationally, in a dictionary-passing implementation, the
1815 constructors <function>Baz1</function> and <function>Baz2</function> must store the
1816 dictionaries for <literal>Eq</literal> and <literal>Show</literal> respectively, and
1817 extract it on pattern matching.
1821 Notice the way that the syntax fits smoothly with that used for
1822 universal quantification earlier.
1828 <title>Restrictions</title>
1831 There are several restrictions on the ways in which existentially-quantified
1832 constructors can be use.
1841 When pattern matching, each pattern match introduces a new,
1842 distinct, type for each existential type variable. These types cannot
1843 be unified with any other type, nor can they escape from the scope of
1844 the pattern match. For example, these fragments are incorrect:
1852 Here, the type bound by <function>MkFoo</function> "escapes", because <literal>a</literal>
1853 is the result of <function>f1</function>. One way to see why this is wrong is to
1854 ask what type <function>f1</function> has:
1858 f1 :: Foo -> a -- Weird!
1862 What is this "<literal>a</literal>" in the result type? Clearly we don't mean
1867 f1 :: forall a. Foo -> a -- Wrong!
1871 The original program is just plain wrong. Here's another sort of error
1875 f2 (Baz1 a b) (Baz1 p q) = a==q
1879 It's ok to say <literal>a==b</literal> or <literal>p==q</literal>, but
1880 <literal>a==q</literal> is wrong because it equates the two distinct types arising
1881 from the two <function>Baz1</function> constructors.
1889 You can't pattern-match on an existentially quantified
1890 constructor in a <literal>let</literal> or <literal>where</literal> group of
1891 bindings. So this is illegal:
1895 f3 x = a==b where { Baz1 a b = x }
1899 You can only pattern-match
1900 on an existentially-quantified constructor in a <literal>case</literal> expression or
1901 in the patterns of a function definition.
1903 The reason for this restriction is really an implementation one.
1904 Type-checking binding groups is already a nightmare without
1905 existentials complicating the picture. Also an existential pattern
1906 binding at the top level of a module doesn't make sense, because it's
1907 not clear how to prevent the existentially-quantified type "escaping".
1908 So for now, there's a simple-to-state restriction. We'll see how
1916 You can't use existential quantification for <literal>newtype</literal>
1917 declarations. So this is illegal:
1921 newtype T = forall a. Ord a => MkT a
1925 Reason: a value of type <literal>T</literal> must be represented as a pair
1926 of a dictionary for <literal>Ord t</literal> and a value of type <literal>t</literal>.
1927 That contradicts the idea that <literal>newtype</literal> should have no
1928 concrete representation. You can get just the same efficiency and effect
1929 by using <literal>data</literal> instead of <literal>newtype</literal>. If there is no
1930 overloading involved, then there is more of a case for allowing
1931 an existentially-quantified <literal>newtype</literal>, because the <literal>data</literal>
1932 because the <literal>data</literal> version does carry an implementation cost,
1933 but single-field existentially quantified constructors aren't much
1934 use. So the simple restriction (no existential stuff on <literal>newtype</literal>)
1935 stands, unless there are convincing reasons to change it.
1943 You can't use <literal>deriving</literal> to define instances of a
1944 data type with existentially quantified data constructors.
1946 Reason: in most cases it would not make sense. For example:#
1949 data T = forall a. MkT [a] deriving( Eq )
1952 To derive <literal>Eq</literal> in the standard way we would need to have equality
1953 between the single component of two <function>MkT</function> constructors:
1957 (MkT a) == (MkT b) = ???
1960 But <VarName>a</VarName> and <VarName>b</VarName> have distinct types, and so can't be compared.
1961 It's just about possible to imagine examples in which the derived instance
1962 would make sense, but it seems altogether simpler simply to prohibit such
1963 declarations. Define your own instances!
1975 <sect1 id="sec-assertions">
1977 <indexterm><primary>Assertions</primary></indexterm>
1981 If you want to make use of assertions in your standard Haskell code, you
1982 could define a function like the following:
1988 assert :: Bool -> a -> a
1989 assert False x = error "assertion failed!"
1996 which works, but gives you back a less than useful error message --
1997 an assertion failed, but which and where?
2001 One way out is to define an extended <function>assert</function> function which also
2002 takes a descriptive string to include in the error message and
2003 perhaps combine this with the use of a pre-processor which inserts
2004 the source location where <function>assert</function> was used.
2008 Ghc offers a helping hand here, doing all of this for you. For every
2009 use of <function>assert</function> in the user's source:
2015 kelvinToC :: Double -> Double
2016 kelvinToC k = assert (k >= 0.0) (k+273.15)
2022 Ghc will rewrite this to also include the source location where the
2029 assert pred val ==> assertError "Main.hs|15" pred val
2035 The rewrite is only performed by the compiler when it spots
2036 applications of <function>Exception.assert</function>, so you can still define and
2037 use your own versions of <function>assert</function>, should you so wish. If not,
2038 import <literal>Exception</literal> to make use <function>assert</function> in your code.
2042 To have the compiler ignore uses of assert, use the compiler option
2043 <option>-fignore-asserts</option>. <indexterm><primary>-fignore-asserts option</primary></indexterm> That is,
2044 expressions of the form <literal>assert pred e</literal> will be rewritten to <literal>e</literal>.
2048 Assertion failures can be caught, see the documentation for the
2049 <literal>Exception</literal> library (<xref linkend="sec-Exception">)
2055 <sect1 id="scoped-type-variables">
2056 <title>Scoped Type Variables
2060 A <emphasis>pattern type signature</emphasis> can introduce a <emphasis>scoped type
2061 variable</emphasis>. For example
2067 f (xs::[a]) = ys ++ ys
2076 The pattern <literal>(xs::[a])</literal> includes a type signature for <VarName>xs</VarName>.
2077 This brings the type variable <literal>a</literal> into scope; it scopes over
2078 all the patterns and right hand sides for this equation for <function>f</function>.
2079 In particular, it is in scope at the type signature for <VarName>y</VarName>.
2083 Pattern type signatures are completely orthogonal to ordinary, separate
2084 type signatures. The two can be used independently or together.
2085 At ordinary type signatures, such as that for <VarName>ys</VarName>, any type variables
2086 mentioned in the type signature <emphasis>that are not in scope</emphasis> are
2087 implicitly universally quantified. (If there are no type variables in
2088 scope, all type variables mentioned in the signature are universally
2089 quantified, which is just as in Haskell 98.) In this case, since <VarName>a</VarName>
2090 is in scope, it is not universally quantified, so the type of <VarName>ys</VarName> is
2091 the same as that of <VarName>xs</VarName>. In Haskell 98 it is not possible to declare
2092 a type for <VarName>ys</VarName>; a major benefit of scoped type variables is that
2093 it becomes possible to do so.
2097 Scoped type variables are implemented in both GHC and Hugs. Where the
2098 implementations differ from the specification below, those differences
2103 So much for the basic idea. Here are the details.
2107 <title>What a pattern type signature means</title>
2109 A type variable brought into scope by a pattern type signature is simply
2110 the name for a type. The restriction they express is that all occurrences
2111 of the same name mean the same type. For example:
2113 f :: [Int] -> Int -> Int
2114 f (xs::[a]) (y::a) = (head xs + y) :: a
2116 The pattern type signatures on the left hand side of
2117 <literal>f</literal> express the fact that <literal>xs</literal>
2118 must be a list of things of some type <literal>a</literal>; and that <literal>y</literal>
2119 must have this same type. The type signature on the expression <literal>(head xs)</literal>
2120 specifies that this expression must have the same type <literal>a</literal>.
2121 <emphasis>There is no requirement that the type named by "<literal>a</literal>" is
2122 in fact a type variable</emphasis>. Indeed, in this case, the type named by "<literal>a</literal>" is
2123 <literal>Int</literal>. (This is a slight liberalisation from the original rather complex
2124 rules, which specified that a pattern-bound type variable should be universally quantified.)
2125 For example, all of these are legal:</para>
2128 t (x::a) (y::a) = x+y*2
2130 f (x::a) (y::b) = [x,y] -- a unifies with b
2132 g (x::a) = x + 1::Int -- a unifies with Int
2134 h x = let k (y::a) = [x,y] -- a is free in the
2135 in k x -- environment
2137 k (x::a) True = ... -- a unifies with Int
2138 k (x::Int) False = ...
2141 w (x::a) = x -- a unifies with [b]
2147 <title>Scope and implicit quantification</title>
2155 All the type variables mentioned in a pattern,
2156 that are not already in scope,
2157 are brought into scope by the pattern. We describe this set as
2158 the <emphasis>type variables bound by the pattern</emphasis>.
2161 f (x::a) = let g (y::(a,b)) = fst y
2165 The pattern <literal>(x::a)</literal> brings the type variable
2166 <literal>a</literal> into scope, as well as the term
2167 variable <literal>x</literal>. The pattern <literal>(y::(a,b))</literal>
2168 contains an occurrence of the already-in-scope type variable <literal>a</literal>,
2169 and brings into scope the type variable <literal>b</literal>.
2175 The type variables thus brought into scope may be mentioned
2176 in ordinary type signatures or pattern type signatures anywhere within
2184 In ordinary type signatures, any type variable mentioned in the
2185 signature that is in scope is <emphasis>not</emphasis> universally quantified.
2193 Ordinary type signatures do not bring any new type variables
2194 into scope (except in the type signature itself!). So this is illegal:
2201 It's illegal because <VarName>a</VarName> is not in scope in the body of <function>f</function>,
2202 so the ordinary signature <literal>x::a</literal> is equivalent to <literal>x::forall a.a</literal>;
2203 and that is an incorrect typing.
2210 There is no implicit universal quantification on pattern type
2211 signatures, nor may one write an explicit <literal>forall</literal> type in a pattern
2212 type signature. The pattern type signature is a monotype.
2220 The type variables in the head of a <literal>class</literal> or <literal>instance</literal> declaration
2221 scope over the methods defined in the <literal>where</literal> part. For example:
2235 (Not implemented in Hugs yet, Dec 98).
2246 <title>Result type signatures</title>
2254 The result type of a function can be given a signature,
2259 f (x::a) :: [a] = [x,x,x]
2263 The final <literal>:: [a]</literal> after all the patterns gives a signature to the
2264 result type. Sometimes this is the only way of naming the type variable
2269 f :: Int -> [a] -> [a]
2270 f n :: ([a] -> [a]) = let g (x::a, y::a) = (y,x)
2271 in \xs -> map g (reverse xs `zip` xs)
2283 Result type signatures are not yet implemented in Hugs.
2289 <title>Where a pattern type signature can occur</title>
2292 A pattern type signature can occur in any pattern, but there
2293 are restrictions on pattern bindings:
2298 A pattern type signature can be on an arbitrary sub-pattern, not
2303 f ((x,y)::(a,b)) = (y,x) :: (b,a)
2312 Pattern type signatures, including the result part, can be used
2313 in lambda abstractions:
2316 (\ (x::a, y) :: a -> x)
2323 Pattern type signatures, including the result part, can be used
2324 in <literal>case</literal> expressions:
2328 case e of { (x::a, y) :: a -> x }
2336 To avoid ambiguity, the type after the “<literal>::</literal>” in a result
2337 pattern signature on a lambda or <literal>case</literal> must be atomic (i.e. a single
2338 token or a parenthesised type of some sort). To see why,
2339 consider how one would parse this:
2353 Pattern type signatures can bind existential type variables.
2358 data T = forall a. MkT [a]
2361 f (MkT [t::a]) = MkT t3
2374 Pattern type signatures that bind new type variables
2375 may not be used in pattern bindings at all.
2380 f x = let (y, z::a) = x in ...
2384 But these are OK, because they do not bind fresh type variables:
2388 f1 x = let (y, z::Int) = x in ...
2389 f2 (x::(Int,a)) = let (y, z::a) = x in ...
2393 However a single variable is considered a degenerate function binding,
2394 rather than a degerate pattern binding, so this is permitted, even
2395 though it binds a type variable:
2399 f :: (b->b) = \(x::b) -> x
2408 Such degnerate function bindings do not fall under the monomorphism
2415 g :: a -> a -> Bool = \x y. x==y
2421 Here <function>g</function> has type <literal>forall a. Eq a => a -> a -> Bool</literal>, just as if
2422 <function>g</function> had a separate type signature. Lacking a type signature, <function>g</function>
2423 would get a monomorphic type.
2431 <sect1 id="pragmas">
2432 <title>Pragmas</title>
2434 <indexterm><primary>pragma</primary></indexterm>
2436 <para>GHC supports several pragmas, or instructions to the
2437 compiler placed in the source code. Pragmas don't normally affect
2438 the meaning of the program, but they might affect the efficiency
2439 of the generated code.</para>
2441 <para>Pragmas all take the form
2443 <literal>{-# <replaceable>word</replaceable> ... #-}</literal>
2445 where <replaceable>word</replaceable> indicates the type of
2446 pragma, and is followed optionally by information specific to that
2447 type of pragma. Case is ignored in
2448 <replaceable>word</replaceable>. The various values for
2449 <replaceable>word</replaceable> that GHC understands are described
2450 in the following sections; any pragma encountered with an
2451 unrecognised <replaceable>word</replaceable> is (silently)
2454 <sect2 id="inline-pragma">
2455 <title>INLINE pragma
2457 <indexterm><primary>INLINE pragma</primary></indexterm>
2458 <indexterm><primary>pragma, INLINE</primary></indexterm></title>
2461 GHC (with <option>-O</option>, as always) tries to inline (or “unfold”)
2462 functions/values that are “small enough,” thus avoiding the call
2463 overhead and possibly exposing other more-wonderful optimisations.
2467 You will probably see these unfoldings (in Core syntax) in your
2472 Normally, if GHC decides a function is “too expensive” to inline, it
2473 will not do so, nor will it export that unfolding for other modules to
2478 The sledgehammer you can bring to bear is the
2479 <literal>INLINE</literal><indexterm><primary>INLINE pragma</primary></indexterm> pragma, used thusly:
2482 key_function :: Int -> String -> (Bool, Double)
2484 #ifdef __GLASGOW_HASKELL__
2485 {-# INLINE key_function #-}
2489 (You don't need to do the C pre-processor carry-on unless you're going
2490 to stick the code through HBC—it doesn't like <literal>INLINE</literal> pragmas.)
2494 The major effect of an <literal>INLINE</literal> pragma is to declare a function's
2495 “cost” to be very low. The normal unfolding machinery will then be
2496 very keen to inline it.
2500 An <literal>INLINE</literal> pragma for a function can be put anywhere its type
2501 signature could be put.
2505 <literal>INLINE</literal> pragmas are a particularly good idea for the
2506 <literal>then</literal>/<literal>return</literal> (or <literal>bind</literal>/<literal>unit</literal>) functions in a monad.
2507 For example, in GHC's own <literal>UniqueSupply</literal> monad code, we have:
2510 #ifdef __GLASGOW_HASKELL__
2511 {-# INLINE thenUs #-}
2512 {-# INLINE returnUs #-}
2520 <sect2 id="noinline-pragma">
2521 <title>NOINLINE pragma
2524 <indexterm><primary>NOINLINE pragma</primary></indexterm>
2525 <indexterm><primary>pragma</primary><secondary>NOINLINE</secondary></indexterm>
2526 <indexterm><primary>NOTINLINE pragma</primary></indexterm>
2527 <indexterm><primary>pragma</primary><secondary>NOTINLINE</secondary></indexterm>
2530 The <literal>NOINLINE</literal> pragma does exactly what you'd expect:
2531 it stops the named function from being inlined by the compiler. You
2532 shouldn't ever need to do this, unless you're very cautious about code
2536 <para><literal>NOTINLINE</literal> is a synonym for
2537 <literal>NOINLINE</literal> (<literal>NOTINLINE</literal> is specified
2538 by Haskell 98 as the standard way to disable inlining, so it should be
2539 used if you want your code to be portable).</para>
2543 <sect2 id="specialize-pragma">
2544 <title>SPECIALIZE pragma</title>
2546 <indexterm><primary>SPECIALIZE pragma</primary></indexterm>
2547 <indexterm><primary>pragma, SPECIALIZE</primary></indexterm>
2548 <indexterm><primary>overloading, death to</primary></indexterm>
2550 <para>(UK spelling also accepted.) For key overloaded
2551 functions, you can create extra versions (NB: more code space)
2552 specialised to particular types. Thus, if you have an
2553 overloaded function:</para>
2556 hammeredLookup :: Ord key => [(key, value)] -> key -> value
2559 <para>If it is heavily used on lists with
2560 <literal>Widget</literal> keys, you could specialise it as
2564 {-# SPECIALIZE hammeredLookup :: [(Widget, value)] -> Widget -> value #-}
2567 <para>To get very fancy, you can also specify a named function
2568 to use for the specialised value, as in:</para>
2571 {-# RULES hammeredLookup = blah #-}
2574 <para>where <literal>blah</literal> is an implementation of
2575 <literal>hammerdLookup</literal> written specialy for
2576 <literal>Widget</literal> lookups. It's <emphasis>Your
2577 Responsibility</emphasis> to make sure that
2578 <function>blah</function> really behaves as a specialised
2579 version of <function>hammeredLookup</function>!!!</para>
2581 <para>Note we use the <literal>RULE</literal> pragma here to
2582 indicate that <literal>hammeredLookup</literal> applied at a
2583 certain type should be replaced by <literal>blah</literal>. See
2584 <xref linkend="rules"> for more information on
2585 <literal>RULES</literal>.</para>
2587 <para>An example in which using <literal>RULES</literal> for
2588 specialisation will Win Big:
2591 toDouble :: Real a => a -> Double
2592 toDouble = fromRational . toRational
2594 {-# SPECIALIZE toDouble :: Int -> Double = i2d #-}
2595 i2d (I# i) = D# (int2Double# i) -- uses Glasgow prim-op directly
2598 The <function>i2d</function> function is virtually one machine
2599 instruction; the default conversion—via an intermediate
2600 <literal>Rational</literal>—is obscenely expensive by
2603 <para>A <literal>SPECIALIZE</literal> pragma for a function can
2604 be put anywhere its type signature could be put.</para>
2608 <sect2 id="specialize-instance-pragma">
2609 <title>SPECIALIZE instance pragma
2613 <indexterm><primary>SPECIALIZE pragma</primary></indexterm>
2614 <indexterm><primary>overloading, death to</primary></indexterm>
2615 Same idea, except for instance declarations. For example:
2618 instance (Eq a) => Eq (Foo a) where { ... usual stuff ... }
2620 {-# SPECIALIZE instance Eq (Foo [(Int, Bar)] #-}
2623 Compatible with HBC, by the way.
2628 <sect2 id="line-pragma">
2633 <indexterm><primary>LINE pragma</primary></indexterm>
2634 <indexterm><primary>pragma, LINE</primary></indexterm>
2638 This pragma is similar to C's <literal>#line</literal> pragma, and is mainly for use in
2639 automatically generated Haskell code. It lets you specify the line
2640 number and filename of the original code; for example
2646 {-# LINE 42 "Foo.vhs" #-}
2652 if you'd generated the current file from something called <filename>Foo.vhs</filename>
2653 and this line corresponds to line 42 in the original. GHC will adjust
2654 its error messages to refer to the line/file named in the <literal>LINE</literal>
2661 <title>RULES pragma</title>
2664 The RULES pragma lets you specify rewrite rules. It is described in
2665 <xref LinkEnd="rewrite-rules">.
2670 <sect2 id="deprecated-pragma">
2671 <title>DEPRECATED pragma</title>
2674 The DEPRECATED pragma lets you specify that a particular function, class, or type, is deprecated.
2675 There are two forms.
2679 You can deprecate an entire module thus:</para>
2681 module Wibble {-# DEPRECATED "Use Wobble instead" #-} where
2685 When you compile any module that import <literal>Wibble</literal>, GHC will print
2686 the specified message.</para>
2691 You can deprecate a function, class, or type, with the following top-level declaration:
2694 {-# DEPRECATED f, C, T "Don't use these" #-}
2697 When you compile any module that imports and uses any of the specifed entities,
2698 GHC will print the specified message.
2702 <para>You can suppress the warnings with the flag <option>-fno-warn-deprecations</option>.</para>
2708 <sect1 id="rewrite-rules">
2709 <title>Rewrite rules
2711 <indexterm><primary>RULES pagma</primary></indexterm>
2712 <indexterm><primary>pragma, RULES</primary></indexterm>
2713 <indexterm><primary>rewrite rules</primary></indexterm></title>
2716 The programmer can specify rewrite rules as part of the source program
2717 (in a pragma). GHC applies these rewrite rules wherever it can.
2725 "map/map" forall f g xs. map f (map g xs) = map (f.g) xs
2732 <title>Syntax</title>
2735 From a syntactic point of view:
2741 Each rule has a name, enclosed in double quotes. The name itself has
2742 no significance at all. It is only used when reporting how many times the rule fired.
2748 There may be zero or more rules in a <literal>RULES</literal> pragma.
2754 Layout applies in a <literal>RULES</literal> pragma. Currently no new indentation level
2755 is set, so you must lay out your rules starting in the same column as the
2756 enclosing definitions.
2762 Each variable mentioned in a rule must either be in scope (e.g. <function>map</function>),
2763 or bound by the <literal>forall</literal> (e.g. <function>f</function>, <function>g</function>, <function>xs</function>). The variables bound by
2764 the <literal>forall</literal> are called the <emphasis>pattern</emphasis> variables. They are separated
2765 by spaces, just like in a type <literal>forall</literal>.
2771 A pattern variable may optionally have a type signature.
2772 If the type of the pattern variable is polymorphic, it <emphasis>must</emphasis> have a type signature.
2773 For example, here is the <literal>foldr/build</literal> rule:
2776 "fold/build" forall k z (g::forall b. (a->b->b) -> b -> b) .
2777 foldr k z (build g) = g k z
2780 Since <function>g</function> has a polymorphic type, it must have a type signature.
2787 The left hand side of a rule must consist of a top-level variable applied
2788 to arbitrary expressions. For example, this is <emphasis>not</emphasis> OK:
2791 "wrong1" forall e1 e2. case True of { True -> e1; False -> e2 } = e1
2792 "wrong2" forall f. f True = True
2795 In <literal>"wrong1"</literal>, the LHS is not an application; in <literal>"wrong2"</literal>, the LHS has a pattern variable
2802 A rule does not need to be in the same module as (any of) the
2803 variables it mentions, though of course they need to be in scope.
2809 Rules are automatically exported from a module, just as instance declarations are.
2820 <title>Semantics</title>
2823 From a semantic point of view:
2829 Rules are only applied if you use the <option>-O</option> flag.
2835 Rules are regarded as left-to-right rewrite rules.
2836 When GHC finds an expression that is a substitution instance of the LHS
2837 of a rule, it replaces the expression by the (appropriately-substituted) RHS.
2838 By "a substitution instance" we mean that the LHS can be made equal to the
2839 expression by substituting for the pattern variables.
2846 The LHS and RHS of a rule are typechecked, and must have the
2854 GHC makes absolutely no attempt to verify that the LHS and RHS
2855 of a rule have the same meaning. That is undecideable in general, and
2856 infeasible in most interesting cases. The responsibility is entirely the programmer's!
2863 GHC makes no attempt to make sure that the rules are confluent or
2864 terminating. For example:
2867 "loop" forall x,y. f x y = f y x
2870 This rule will cause the compiler to go into an infinite loop.
2877 If more than one rule matches a call, GHC will choose one arbitrarily to apply.
2883 GHC currently uses a very simple, syntactic, matching algorithm
2884 for matching a rule LHS with an expression. It seeks a substitution
2885 which makes the LHS and expression syntactically equal modulo alpha
2886 conversion. The pattern (rule), but not the expression, is eta-expanded if
2887 necessary. (Eta-expanding the epression can lead to laziness bugs.)
2888 But not beta conversion (that's called higher-order matching).
2892 Matching is carried out on GHC's intermediate language, which includes
2893 type abstractions and applications. So a rule only matches if the
2894 types match too. See <xref LinkEnd="rule-spec"> below.
2900 GHC keeps trying to apply the rules as it optimises the program.
2901 For example, consider:
2910 The expression <literal>s (t xs)</literal> does not match the rule <literal>"map/map"</literal>, but GHC
2911 will substitute for <VarName>s</VarName> and <VarName>t</VarName>, giving an expression which does match.
2912 If <VarName>s</VarName> or <VarName>t</VarName> was (a) used more than once, and (b) large or a redex, then it would
2913 not be substituted, and the rule would not fire.
2920 In the earlier phases of compilation, GHC inlines <emphasis>nothing
2921 that appears on the LHS of a rule</emphasis>, because once you have substituted
2922 for something you can't match against it (given the simple minded
2923 matching). So if you write the rule
2926 "map/map" forall f,g. map f . map g = map (f.g)
2929 this <emphasis>won't</emphasis> match the expression <literal>map f (map g xs)</literal>.
2930 It will only match something written with explicit use of ".".
2931 Well, not quite. It <emphasis>will</emphasis> match the expression
2937 where <function>wibble</function> is defined:
2940 wibble f g = map f . map g
2943 because <function>wibble</function> will be inlined (it's small).
2945 Later on in compilation, GHC starts inlining even things on the
2946 LHS of rules, but still leaves the rules enabled. This inlining
2947 policy is controlled by the per-simplification-pass flag <option>-finline-phase</option><emphasis>n</emphasis>.
2954 All rules are implicitly exported from the module, and are therefore
2955 in force in any module that imports the module that defined the rule, directly
2956 or indirectly. (That is, if A imports B, which imports C, then C's rules are
2957 in force when compiling A.) The situation is very similar to that for instance
2969 <title>List fusion</title>
2972 The RULES mechanism is used to implement fusion (deforestation) of common list functions.
2973 If a "good consumer" consumes an intermediate list constructed by a "good producer", the
2974 intermediate list should be eliminated entirely.
2978 The following are good producers:
2990 Enumerations of <literal>Int</literal> and <literal>Char</literal> (e.g. <literal>['a'..'z']</literal>).
2996 Explicit lists (e.g. <literal>[True, False]</literal>)
3002 The cons constructor (e.g <literal>3:4:[]</literal>)
3008 <function>++</function>
3014 <function>map</function>
3020 <function>filter</function>
3026 <function>iterate</function>, <function>repeat</function>
3032 <function>zip</function>, <function>zipWith</function>
3041 The following are good consumers:
3053 <function>array</function> (on its second argument)
3059 <function>length</function>
3065 <function>++</function> (on its first argument)
3071 <function>map</function>
3077 <function>filter</function>
3083 <function>concat</function>
3089 <function>unzip</function>, <function>unzip2</function>, <function>unzip3</function>, <function>unzip4</function>
3095 <function>zip</function>, <function>zipWith</function> (but on one argument only; if both are good producers, <function>zip</function>
3096 will fuse with one but not the other)
3102 <function>partition</function>
3108 <function>head</function>
3114 <function>and</function>, <function>or</function>, <function>any</function>, <function>all</function>
3120 <function>sequence_</function>
3126 <function>msum</function>
3132 <function>sortBy</function>
3141 So, for example, the following should generate no intermediate lists:
3144 array (1,10) [(i,i*i) | i <- map (+ 1) [0..9]]
3150 This list could readily be extended; if there are Prelude functions that you use
3151 a lot which are not included, please tell us.
3155 If you want to write your own good consumers or producers, look at the
3156 Prelude definitions of the above functions to see how to do so.
3161 <sect2 id="rule-spec">
3162 <title>Specialisation
3166 Rewrite rules can be used to get the same effect as a feature
3167 present in earlier version of GHC:
3170 {-# SPECIALIZE fromIntegral :: Int8 -> Int16 = int8ToInt16 #-}
3173 This told GHC to use <function>int8ToInt16</function> instead of <function>fromIntegral</function> whenever
3174 the latter was called with type <literal>Int8 -> Int16</literal>. That is, rather than
3175 specialising the original definition of <function>fromIntegral</function> the programmer is
3176 promising that it is safe to use <function>int8ToInt16</function> instead.
3180 This feature is no longer in GHC. But rewrite rules let you do the
3185 "fromIntegral/Int8/Int16" fromIntegral = int8ToInt16
3189 This slightly odd-looking rule instructs GHC to replace <function>fromIntegral</function>
3190 by <function>int8ToInt16</function> <emphasis>whenever the types match</emphasis>. Speaking more operationally,
3191 GHC adds the type and dictionary applications to get the typed rule
3194 forall (d1::Integral Int8) (d2::Num Int16) .
3195 fromIntegral Int8 Int16 d1 d2 = int8ToInt16
3199 this rule does not need to be in the same file as fromIntegral,
3200 unlike the <literal>SPECIALISE</literal> pragmas which currently do (so that they
3201 have an original definition available to specialise).
3207 <title>Controlling what's going on</title>
3215 Use <option>-ddump-rules</option> to see what transformation rules GHC is using.
3221 Use <option>-ddump-simpl-stats</option> to see what rules are being fired.
3222 If you add <option>-dppr-debug</option> you get a more detailed listing.
3228 The defintion of (say) <function>build</function> in <FileName>PrelBase.lhs</FileName> looks llike this:
3231 build :: forall a. (forall b. (a -> b -> b) -> b -> b) -> [a]
3232 {-# INLINE build #-}
3236 Notice the <literal>INLINE</literal>! That prevents <literal>(:)</literal> from being inlined when compiling
3237 <literal>PrelBase</literal>, so that an importing module will “see” the <literal>(:)</literal>, and can
3238 match it on the LHS of a rule. <literal>INLINE</literal> prevents any inlining happening
3239 in the RHS of the <literal>INLINE</literal> thing. I regret the delicacy of this.
3246 In <filename>ghc/lib/std/PrelBase.lhs</filename> look at the rules for <function>map</function> to
3247 see how to write rules that will do fusion and yet give an efficient
3248 program even if fusion doesn't happen. More rules in <filename>PrelList.lhs</filename>.
3260 <sect1 id="generic-classes">
3261 <title>Generic classes</title>
3264 The ideas behind this extension are described in detail in "Derivable type classes",
3265 Ralf Hinze and Simon Peyton Jones, Haskell Workshop, Montreal Sept 2000, pp94-105.
3266 An example will give the idea:
3274 fromBin :: [Int] -> (a, [Int])
3276 toBin {| Unit |} Unit = []
3277 toBin {| a :+: b |} (Inl x) = 0 : toBin x
3278 toBin {| a :+: b |} (Inr y) = 1 : toBin y
3279 toBin {| a :*: b |} (x :*: y) = toBin x ++ toBin y
3281 fromBin {| Unit |} bs = (Unit, bs)
3282 fromBin {| a :+: b |} (0:bs) = (Inl x, bs') where (x,bs') = fromBin bs
3283 fromBin {| a :+: b |} (1:bs) = (Inr y, bs') where (y,bs') = fromBin bs
3284 fromBin {| a :*: b |} bs = (x :*: y, bs'') where (x,bs' ) = fromBin bs
3285 (y,bs'') = fromBin bs'
3288 This class declaration explains how <literal>toBin</literal> and <literal>fromBin</literal>
3289 work for arbitrary data types. They do so by giving cases for unit, product, and sum,
3290 which are defined thus in the library module <literal>Generics</literal>:
3294 data a :+: b = Inl a | Inr b
3295 data a :*: b = a :*: b
3298 Now you can make a data type into an instance of Bin like this:
3300 instance (Bin a, Bin b) => Bin (a,b)
3301 instance Bin a => Bin [a]
3303 That is, just leave off the "where" clasuse. Of course, you can put in the
3304 where clause and over-ride whichever methods you please.
3308 <title> Using generics </title>
3309 <para>To use generics you need to</para>
3312 <para>Use the flags <option>-fglasgow-exts</option> (to enable the extra syntax),
3313 <option>-fgenerics</option> (to generate extra per-data-type code),
3314 and <option>-package lang</option> (to make the <literal>Generics</literal> library
3318 <para>Import the module <literal>Generics</literal> from the
3319 <literal>lang</literal> package. This import brings into
3320 scope the data types <literal>Unit</literal>,
3321 <literal>:*:</literal>, and <literal>:+:</literal>. (You
3322 don't need this import if you don't mention these types
3323 explicitly; for example, if you are simply giving instance
3324 declarations.)</para>
3329 <sect2> <title> Changes wrt the paper </title>
3331 Note that the type constructors <literal>:+:</literal> and <literal>:*:</literal>
3332 can be written infix (indeed, you can now use
3333 any operator starting in a colon as an infix type constructor). Also note that
3334 the type constructors are not exactly as in the paper (Unit instead of 1, etc).
3335 Finally, note that the syntax of the type patterns in the class declaration
3336 uses "<literal>{|</literal>" and "<literal>|}</literal>" brackets; curly braces
3337 alone would ambiguous when they appear on right hand sides (an extension we
3338 anticipate wanting).
3342 <sect2> <title>Terminology and restrictions</title>
3344 Terminology. A "generic default method" in a class declaration
3345 is one that is defined using type patterns as above.
3346 A "polymorphic default method" is a default method defined as in Haskell 98.
3347 A "generic class declaration" is a class declaration with at least one
3348 generic default method.
3356 Alas, we do not yet implement the stuff about constructor names and
3363 A generic class can have only one parameter; you can't have a generic
3364 multi-parameter class.
3370 A default method must be defined entirely using type patterns, or entirely
3371 without. So this is illegal:
3374 op :: a -> (a, Bool)
3375 op {| Unit |} Unit = (Unit, True)
3378 However it is perfectly OK for some methods of a generic class to have
3379 generic default methods and others to have polymorphic default methods.
3385 The type variable(s) in the type pattern for a generic method declaration
3386 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:
3390 op {| p :*: q |} (x :*: y) = op (x :: p)
3398 The type patterns in a generic default method must take one of the forms:
3404 where "a" and "b" are type variables. Furthermore, all the type patterns for
3405 a single type constructor (<literal>:*:</literal>, say) must be identical; they
3406 must use the same type variables. So this is illegal:
3410 op {| a :+: b |} (Inl x) = True
3411 op {| p :+: q |} (Inr y) = False
3413 The type patterns must be identical, even in equations for different methods of the class.
3414 So this too is illegal:
3418 op1 {| a :*: b |} (x :*: y) = True
3421 op2 {| p :*: q |} (x :*: y) = False
3423 (The reason for this restriction is that we gather all the equations for a particular type consructor
3424 into a single generic instance declaration.)
3430 A generic method declaration must give a case for each of the three type constructors.
3436 The type for a generic method can be built only from:
3438 <listitem> <para> Function arrows </para> </listitem>
3439 <listitem> <para> Type variables </para> </listitem>
3440 <listitem> <para> Tuples </para> </listitem>
3441 <listitem> <para> Arbitrary types not involving type variables </para> </listitem>
3443 Here are some example type signatures for generic methods:
3446 op2 :: Bool -> (a,Bool)
3447 op3 :: [Int] -> a -> a
3450 Here, op1, op2, op3 are OK, but op4 is rejected, because it has a type variable
3454 This restriction is an implementation restriction: we just havn't got around to
3455 implementing the necessary bidirectional maps over arbitrary type constructors.
3456 It would be relatively easy to add specific type constructors, such as Maybe and list,
3457 to the ones that are allowed.</para>
3462 In an instance declaration for a generic class, the idea is that the compiler
3463 will fill in the methods for you, based on the generic templates. However it can only
3468 The instance type is simple (a type constructor applied to type variables, as in Haskell 98).
3473 No constructor of the instance type has unboxed fields.
3477 (Of course, these things can only arise if you are already using GHC extensions.)
3478 However, you can still give an instance declarations for types which break these rules,
3479 provided you give explicit code to override any generic default methods.
3487 The option <option>-ddump-deriv</option> dumps incomprehensible stuff giving details of
3488 what the compiler does with generic declarations.
3493 <sect2> <title> Another example </title>
3495 Just to finish with, here's another example I rather like:
3499 nCons {| Unit |} _ = 1
3500 nCons {| a :*: b |} _ = 1
3501 nCons {| a :+: b |} _ = nCons (bot::a) + nCons (bot::b)
3504 tag {| Unit |} _ = 1
3505 tag {| a :*: b |} _ = 1
3506 tag {| a :+: b |} (Inl x) = tag x
3507 tag {| a :+: b |} (Inr y) = nCons (bot::a) + tag y
3514 ;;; Local Variables: ***
3516 ;;; sgml-parent-document: ("users_guide.sgml" "book" "chapter" "sect1") ***