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:
26 <Term>Unboxed types and primitive operations:</Term>
29 You can get right down to the raw machine types and operations;
30 included in this are “primitive arrays” (direct access to Big Wads
31 of Bytes). Please see <XRef LinkEnd="glasgow-unboxed"> and following.
37 <Term>Type system extensions:</Term>
39 <Para> GHC supports a large number of extensions to Haskell's type
45 <Term>Multi-parameter type classes:</Term>
48 <XRef LinkEnd="multi-param-type-classes">
54 <Term>Functional dependencies:</Term>
57 <XRef LinkEnd="functional-dependencies">
63 <Term>Implicit parameters:</Term>
66 <XRef LinkEnd="implicit parameters">
72 <Term>Local universal quantification:</Term>
75 <XRef LinkEnd="universal-quantification">
81 <Term>Extistentially quantification in data types:</Term>
84 <XRef LinkEnd="existential-quantification">
90 <Term>Scoped type variables:</Term>
93 Scoped type variables enable the programmer to supply type signatures
94 for some nested declarations, where this would not be legal in Haskell
95 98. Details in <XRef LinkEnd="scoped-type-variables">.
103 <Term>Pattern guards</Term>
106 Instead of being a boolean expression, a guard is a list of qualifiers, exactly as in a list comprehension. See <XRef LinkEnd="pattern-guards">.
112 <Term>Foreign calling:</Term>
115 Just what it sounds like. We provide <Emphasis>lots</Emphasis> of rope that you
116 can dangle around your neck. Please see <XRef LinkEnd="ffi">.
125 Pragmas are special instructions to the compiler placed in the source
126 file. The pragmas GHC supports are described in <XRef LinkEnd="pragmas">.
132 <Term>Rewrite rules:</Term>
135 The programmer can specify rewrite rules as part of the source program
136 (in a pragma). GHC applies these rewrite rules wherever it can.
137 Details in <XRef LinkEnd="rewrite-rules">.
143 <Term>Generic classes:</Term>
146 Generic class declarations allow you to define a class
147 whose methods say how to work over an arbitrary data type.
148 Then it's really easy to make any new type into an instance of
149 the class. This generalises the rather ad-hoc "deriving" feature
151 Details in <XRef LinkEnd="generic-classes">.
159 Before you get too carried away working at the lowest level (e.g.,
160 sloshing <Literal>MutableByteArray#</Literal>s around your
161 program), you may wish to check if there are libraries that provide a
162 “Haskellised veneer” over the features you want. See
163 <xref linkend="book-hslibs">.
166 <sect1 id="options-language">
167 <title>Language options</title>
169 <indexterm><primary>language</primary><secondary>option</secondary>
171 <indexterm><primary>options</primary><secondary>language</secondary>
173 <indexterm><primary>extensions</primary><secondary>options controlling</secondary>
176 <para> These flags control what variation of the language are
177 permitted. Leaving out all of them gives you standard Haskell
183 <term><option>-fglasgow-exts</option>:</term>
184 <indexterm><primary><option>-fglasgow-exts</option></primary></indexterm>
186 <para>This simultaneously enables all of the extensions to
187 Haskell 98 described in <xref
188 linkend="ghc-language-features">, except where otherwise
194 <term><option>-fno-monomorphism-restriction</option>:</term>
195 <indexterm><primary><option>-fno-monomorphism-restriction</option></primary></indexterm>
197 <para> Switch off the Haskell 98 monomorphism restriction.
198 Independent of the <Option>-fglasgow-exts</Option>
204 <term><option>-fallow-overlapping-instances</option></term>
205 <term><option>-fallow-undecidable-instances</option></term>
206 <term><option>-fcontext-stack</option></term>
207 <indexterm><primary><option>-fallow-overlapping-instances</option></primary></indexterm>
208 <indexterm><primary><option>-fallow-undecidable-instances</option></primary></indexterm>
209 <indexterm><primary><option>-fcontext-stack</option></primary></indexterm>
211 <para> See <XRef LinkEnd="instance-decls">. Only relevant
212 if you also use <option>-fglasgow-exts</option>.</para>
217 <term><option>-fignore-asserts</option>:</term>
218 <indexterm><primary><option>-fignore-asserts</option></primary></indexterm>
220 <para>See <XRef LinkEnd="sec-assertions">. Only relevant if
221 you also use <option>-fglasgow-exts</option>.</Para>
226 <term><option>-finline-phase</option></term>
227 <indexterm><primary><option>-finline-phase</option></primary></indexterm>
229 <para>See <XRef LinkEnd="rewrite-rules">. Only relevant if
230 you also use <Option>-fglasgow-exts</Option>.</para>
235 <term><option>-fgenerics</option></term>
236 <indexterm><primary><option>-fgenerics</option></primary></indexterm>
238 <para>See <XRef LinkEnd="generic-classes">. Independent of
239 <Option>-fglasgow-exts</Option>.</para>
244 <term><option>-fno-implicit-prelude</option></term>
246 <para><indexterm><primary>-fno-implicit-prelude
247 option</primary></indexterm> GHC normally imports
248 <filename>Prelude.hi</filename> files for you. If you'd
249 rather it didn't, then give it a
250 <option>-fno-implicit-prelude</option> option. The idea
251 is that you can then import a Prelude of your own. (But
252 don't call it <literal>Prelude</literal>; the Haskell
253 module namespace is flat, and you must not conflict with
254 any Prelude module.)</para>
256 <para>Even though you have not imported the Prelude, all
257 the built-in syntax still refers to the built-in Haskell
258 Prelude types and values, as specified by the Haskell
259 Report. For example, the type <literal>[Int]</literal>
260 still means <literal>Prelude.[] Int</literal>; tuples
261 continue to refer to the standard Prelude tuples; the
262 translation for list comprehensions continues to use
263 <literal>Prelude.map</literal> etc.</para>
265 <para> With one group of exceptions! You may want to
266 define your own numeric class hierarchy. It completely
267 defeats that purpose if the literal "1" means
268 "<literal>Prelude.fromInteger 1</literal>", which is what
269 the Haskell Report specifies. So the
270 <option>-fno-implicit-prelude</option> flag causes the
271 following pieces of built-in syntax to refer to whatever
272 is in scope, not the Prelude versions:</para>
276 <para>Integer and fractional literals mean
277 "<literal>fromInteger 1</literal>" and
278 "<literal>fromRational 3.2</literal>", not the
279 Prelude-qualified versions; both in expressions and in
284 <para>Negation (e.g. "<literal>- (f x)</literal>")
285 means "<literal>negate (f x)</literal>" (not
286 <literal>Prelude.negate</literal>).</para>
290 <para>In an n+k pattern, the standard Prelude
291 <literal>Ord</literal> class is used for comparison,
292 but the necessary subtraction uses whatever
293 "<literal>(-)</literal>" is in scope (not
294 "<literal>Prelude.(-)</literal>").</para>
304 <Sect1 id="primitives">
305 <Title>Unboxed types and primitive operations
307 <IndexTerm><Primary>PrelGHC module</Primary></IndexTerm>
310 This module defines all the types which are primitive in Glasgow
311 Haskell, and the operations provided for them.
314 <Sect2 id="glasgow-unboxed">
319 <IndexTerm><Primary>Unboxed types (Glasgow extension)</Primary></IndexTerm>
322 <para>Most types in GHC are <firstterm>boxed</firstterm>, which means
323 that values of that type are represented by a pointer to a heap
324 object. The representation of a Haskell <literal>Int</literal>, for
325 example, is a two-word heap object. An <firstterm>unboxed</firstterm>
326 type, however, is represented by the value itself, no pointers or heap
327 allocation are involved.
331 Unboxed types correspond to the “raw machine” types you
332 would use in C: <Literal>Int#</Literal> (long int),
333 <Literal>Double#</Literal> (double), <Literal>Addr#</Literal>
334 (void *), etc. The <Emphasis>primitive operations</Emphasis>
335 (PrimOps) on these types are what you might expect; e.g.,
336 <Literal>(+#)</Literal> is addition on
337 <Literal>Int#</Literal>s, and is the machine-addition that we all
338 know and love—usually one instruction.
342 Primitive (unboxed) types cannot be defined in Haskell, and are
343 therefore built into the language and compiler. Primitive types are
344 always unlifted; that is, a value of a primitive type cannot be
345 bottom. We use the convention that primitive types, values, and
346 operations have a <Literal>#</Literal> suffix.
350 Primitive values are often represented by a simple bit-pattern, such
351 as <Literal>Int#</Literal>, <Literal>Float#</Literal>,
352 <Literal>Double#</Literal>. But this is not necessarily the case:
353 a primitive value might be represented by a pointer to a
354 heap-allocated object. Examples include
355 <Literal>Array#</Literal>, the type of primitive arrays. A
356 primitive array is heap-allocated because it is too big a value to fit
357 in a register, and would be too expensive to copy around; in a sense,
358 it is accidental that it is represented by a pointer. If a pointer
359 represents a primitive value, then it really does point to that value:
360 no unevaluated thunks, no indirections…nothing can be at the
361 other end of the pointer than the primitive value.
365 There are some restrictions on the use of primitive types, the main
366 one being that you can't pass a primitive value to a polymorphic
367 function or store one in a polymorphic data type. This rules out
368 things like <Literal>[Int#]</Literal> (i.e. lists of primitive
369 integers). The reason for this restriction is that polymorphic
370 arguments and constructor fields are assumed to be pointers: if an
371 unboxed integer is stored in one of these, the garbage collector would
372 attempt to follow it, leading to unpredictable space leaks. Or a
373 <Function>seq</Function> operation on the polymorphic component may
374 attempt to dereference the pointer, with disastrous results. Even
375 worse, the unboxed value might be larger than a pointer
376 (<Literal>Double#</Literal> for instance).
380 Nevertheless, A numerically-intensive program using unboxed types can
381 go a <Emphasis>lot</Emphasis> faster than its “standard”
382 counterpart—we saw a threefold speedup on one example.
387 <Sect2 id="unboxed-tuples">
388 <Title>Unboxed Tuples
392 Unboxed tuples aren't really exported by <Literal>PrelGHC</Literal>,
393 they're available by default with <Option>-fglasgow-exts</Option>. An
394 unboxed tuple looks like this:
406 where <Literal>e_1..e_n</Literal> are expressions of any
407 type (primitive or non-primitive). The type of an unboxed tuple looks
412 Unboxed tuples are used for functions that need to return multiple
413 values, but they avoid the heap allocation normally associated with
414 using fully-fledged tuples. When an unboxed tuple is returned, the
415 components are put directly into registers or on the stack; the
416 unboxed tuple itself does not have a composite representation. Many
417 of the primitive operations listed in this section return unboxed
422 There are some pretty stringent restrictions on the use of unboxed tuples:
431 Unboxed tuple types are subject to the same restrictions as
432 other unboxed types; i.e. they may not be stored in polymorphic data
433 structures or passed to polymorphic functions.
440 Unboxed tuples may only be constructed as the direct result of
441 a function, and may only be deconstructed with a <Literal>case</Literal> expression.
442 eg. the following are valid:
446 f x y = (# x+1, y-1 #)
447 g x = case f x x of { (# a, b #) -> a + b }
451 but the following are invalid:
465 No variable can have an unboxed tuple type. This is illegal:
469 f :: (# Int, Int #) -> (# Int, Int #)
474 because <VarName>x</VarName> has an unboxed tuple type.
484 Note: we may relax some of these restrictions in the future.
488 The <Literal>IO</Literal> and <Literal>ST</Literal> monads use unboxed
489 tuples to avoid unnecessary allocation during sequences of operations.
495 <Title>Character and numeric types</Title>
497 <IndexTerm><Primary>character types, primitive</Primary></IndexTerm>
498 <IndexTerm><Primary>numeric types, primitive</Primary></IndexTerm>
499 <IndexTerm><Primary>integer types, primitive</Primary></IndexTerm>
500 <IndexTerm><Primary>floating point types, primitive</Primary></IndexTerm>
502 There are the following obvious primitive types:
516 <IndexTerm><Primary><literal>Char#</literal></Primary></IndexTerm>
517 <IndexTerm><Primary><literal>Int#</literal></Primary></IndexTerm>
518 <IndexTerm><Primary><literal>Word#</literal></Primary></IndexTerm>
519 <IndexTerm><Primary><literal>Addr#</literal></Primary></IndexTerm>
520 <IndexTerm><Primary><literal>Float#</literal></Primary></IndexTerm>
521 <IndexTerm><Primary><literal>Double#</literal></Primary></IndexTerm>
522 <IndexTerm><Primary><literal>Int64#</literal></Primary></IndexTerm>
523 <IndexTerm><Primary><literal>Word64#</literal></Primary></IndexTerm>
526 If you really want to know their exact equivalents in C, see
527 <Filename>ghc/includes/StgTypes.h</Filename> in the GHC source tree.
531 Literals for these types may be written as follows:
540 'a'# a Char#; for weird characters, use e.g. '\o<octal>'#
541 "a"# an Addr# (a `char *'); only characters '\0'..'\255' allowed
544 <IndexTerm><Primary>literals, primitive</Primary></IndexTerm>
545 <IndexTerm><Primary>constants, primitive</Primary></IndexTerm>
546 <IndexTerm><Primary>numbers, primitive</Primary></IndexTerm>
552 <Title>Comparison operations</Title>
555 <IndexTerm><Primary>comparisons, primitive</Primary></IndexTerm>
556 <IndexTerm><Primary>operators, comparison</Primary></IndexTerm>
562 {>,>=,==,/=,<,<=}# :: Int# -> Int# -> Bool
564 {gt,ge,eq,ne,lt,le}Char# :: Char# -> Char# -> Bool
565 -- ditto for Word# and Addr#
568 <IndexTerm><Primary><literal>>#</literal></Primary></IndexTerm>
569 <IndexTerm><Primary><literal>>=#</literal></Primary></IndexTerm>
570 <IndexTerm><Primary><literal>==#</literal></Primary></IndexTerm>
571 <IndexTerm><Primary><literal>/=#</literal></Primary></IndexTerm>
572 <IndexTerm><Primary><literal><#</literal></Primary></IndexTerm>
573 <IndexTerm><Primary><literal><=#</literal></Primary></IndexTerm>
574 <IndexTerm><Primary><literal>gt{Char,Word,Addr}#</literal></Primary></IndexTerm>
575 <IndexTerm><Primary><literal>ge{Char,Word,Addr}#</literal></Primary></IndexTerm>
576 <IndexTerm><Primary><literal>eq{Char,Word,Addr}#</literal></Primary></IndexTerm>
577 <IndexTerm><Primary><literal>ne{Char,Word,Addr}#</literal></Primary></IndexTerm>
578 <IndexTerm><Primary><literal>lt{Char,Word,Addr}#</literal></Primary></IndexTerm>
579 <IndexTerm><Primary><literal>le{Char,Word,Addr}#</literal></Primary></IndexTerm>
585 <Title>Primitive-character operations</Title>
588 <IndexTerm><Primary>characters, primitive operations</Primary></IndexTerm>
589 <IndexTerm><Primary>operators, primitive character</Primary></IndexTerm>
595 ord# :: Char# -> Int#
596 chr# :: Int# -> Char#
599 <IndexTerm><Primary><literal>ord#</literal></Primary></IndexTerm>
600 <IndexTerm><Primary><literal>chr#</literal></Primary></IndexTerm>
606 <Title>Primitive-<Literal>Int</Literal> operations</Title>
609 <IndexTerm><Primary>integers, primitive operations</Primary></IndexTerm>
610 <IndexTerm><Primary>operators, primitive integer</Primary></IndexTerm>
616 {+,-,*,quotInt,remInt,gcdInt}# :: Int# -> Int# -> Int#
617 negateInt# :: Int# -> Int#
619 iShiftL#, iShiftRA#, iShiftRL# :: Int# -> Int# -> Int#
620 -- shift left, right arithmetic, right logical
622 addIntC#, subIntC#, mulIntC# :: Int# -> Int# -> (# Int#, Int# #)
623 -- add, subtract, multiply with carry
626 <IndexTerm><Primary><literal>+#</literal></Primary></IndexTerm>
627 <IndexTerm><Primary><literal>-#</literal></Primary></IndexTerm>
628 <IndexTerm><Primary><literal>*#</literal></Primary></IndexTerm>
629 <IndexTerm><Primary><literal>quotInt#</literal></Primary></IndexTerm>
630 <IndexTerm><Primary><literal>remInt#</literal></Primary></IndexTerm>
631 <IndexTerm><Primary><literal>gcdInt#</literal></Primary></IndexTerm>
632 <IndexTerm><Primary><literal>iShiftL#</literal></Primary></IndexTerm>
633 <IndexTerm><Primary><literal>iShiftRA#</literal></Primary></IndexTerm>
634 <IndexTerm><Primary><literal>iShiftRL#</literal></Primary></IndexTerm>
635 <IndexTerm><Primary><literal>addIntC#</literal></Primary></IndexTerm>
636 <IndexTerm><Primary><literal>subIntC#</literal></Primary></IndexTerm>
637 <IndexTerm><Primary><literal>mulIntC#</literal></Primary></IndexTerm>
638 <IndexTerm><Primary>shift operations, integer</Primary></IndexTerm>
642 <Emphasis>Note:</Emphasis> No error/overflow checking!
648 <Title>Primitive-<Literal>Double</Literal> and <Literal>Float</Literal> operations</Title>
651 <IndexTerm><Primary>floating point numbers, primitive</Primary></IndexTerm>
652 <IndexTerm><Primary>operators, primitive floating point</Primary></IndexTerm>
658 {+,-,*,/}## :: Double# -> Double# -> Double#
659 {<,<=,==,/=,>=,>}## :: Double# -> Double# -> Bool
660 negateDouble# :: Double# -> Double#
661 double2Int# :: Double# -> Int#
662 int2Double# :: Int# -> Double#
664 {plus,minux,times,divide}Float# :: Float# -> Float# -> Float#
665 {gt,ge,eq,ne,lt,le}Float# :: Float# -> Float# -> Bool
666 negateFloat# :: Float# -> Float#
667 float2Int# :: Float# -> Int#
668 int2Float# :: Int# -> Float#
674 <IndexTerm><Primary><literal>+##</literal></Primary></IndexTerm>
675 <IndexTerm><Primary><literal>-##</literal></Primary></IndexTerm>
676 <IndexTerm><Primary><literal>*##</literal></Primary></IndexTerm>
677 <IndexTerm><Primary><literal>/##</literal></Primary></IndexTerm>
678 <IndexTerm><Primary><literal><##</literal></Primary></IndexTerm>
679 <IndexTerm><Primary><literal><=##</literal></Primary></IndexTerm>
680 <IndexTerm><Primary><literal>==##</literal></Primary></IndexTerm>
681 <IndexTerm><Primary><literal>=/##</literal></Primary></IndexTerm>
682 <IndexTerm><Primary><literal>>=##</literal></Primary></IndexTerm>
683 <IndexTerm><Primary><literal>>##</literal></Primary></IndexTerm>
684 <IndexTerm><Primary><literal>negateDouble#</literal></Primary></IndexTerm>
685 <IndexTerm><Primary><literal>double2Int#</literal></Primary></IndexTerm>
686 <IndexTerm><Primary><literal>int2Double#</literal></Primary></IndexTerm>
690 <IndexTerm><Primary><literal>plusFloat#</literal></Primary></IndexTerm>
691 <IndexTerm><Primary><literal>minusFloat#</literal></Primary></IndexTerm>
692 <IndexTerm><Primary><literal>timesFloat#</literal></Primary></IndexTerm>
693 <IndexTerm><Primary><literal>divideFloat#</literal></Primary></IndexTerm>
694 <IndexTerm><Primary><literal>gtFloat#</literal></Primary></IndexTerm>
695 <IndexTerm><Primary><literal>geFloat#</literal></Primary></IndexTerm>
696 <IndexTerm><Primary><literal>eqFloat#</literal></Primary></IndexTerm>
697 <IndexTerm><Primary><literal>neFloat#</literal></Primary></IndexTerm>
698 <IndexTerm><Primary><literal>ltFloat#</literal></Primary></IndexTerm>
699 <IndexTerm><Primary><literal>leFloat#</literal></Primary></IndexTerm>
700 <IndexTerm><Primary><literal>negateFloat#</literal></Primary></IndexTerm>
701 <IndexTerm><Primary><literal>float2Int#</literal></Primary></IndexTerm>
702 <IndexTerm><Primary><literal>int2Float#</literal></Primary></IndexTerm>
706 And a full complement of trigonometric functions:
712 expDouble# :: Double# -> Double#
713 logDouble# :: Double# -> Double#
714 sqrtDouble# :: Double# -> Double#
715 sinDouble# :: Double# -> Double#
716 cosDouble# :: Double# -> Double#
717 tanDouble# :: Double# -> Double#
718 asinDouble# :: Double# -> Double#
719 acosDouble# :: Double# -> Double#
720 atanDouble# :: Double# -> Double#
721 sinhDouble# :: Double# -> Double#
722 coshDouble# :: Double# -> Double#
723 tanhDouble# :: Double# -> Double#
724 powerDouble# :: Double# -> Double# -> Double#
727 <IndexTerm><Primary>trigonometric functions, primitive</Primary></IndexTerm>
731 similarly for <Literal>Float#</Literal>.
735 There are two coercion functions for <Literal>Float#</Literal>/<Literal>Double#</Literal>:
741 float2Double# :: Float# -> Double#
742 double2Float# :: Double# -> Float#
745 <IndexTerm><Primary><literal>float2Double#</literal></Primary></IndexTerm>
746 <IndexTerm><Primary><literal>double2Float#</literal></Primary></IndexTerm>
750 The primitive version of <Function>decodeDouble</Function>
751 (<Function>encodeDouble</Function> is implemented as an external C
758 decodeDouble# :: Double# -> PrelNum.ReturnIntAndGMP
761 <IndexTerm><Primary><literal>encodeDouble#</literal></Primary></IndexTerm>
762 <IndexTerm><Primary><literal>decodeDouble#</literal></Primary></IndexTerm>
766 (And the same for <Literal>Float#</Literal>s.)
771 <Sect2 id="integer-operations">
772 <Title>Operations on/for <Literal>Integers</Literal> (interface to GMP)
776 <IndexTerm><Primary>arbitrary precision integers</Primary></IndexTerm>
777 <IndexTerm><Primary>Integer, operations on</Primary></IndexTerm>
781 We implement <Literal>Integers</Literal> (arbitrary-precision
782 integers) using the GNU multiple-precision (GMP) package (version
787 The data type for <Literal>Integer</Literal> is either a small
788 integer, represented by an <Literal>Int</Literal>, or a large integer
789 represented using the pieces required by GMP's
790 <Literal>MP_INT</Literal> in <Filename>gmp.h</Filename> (see
791 <Filename>gmp.info</Filename> in
792 <Filename>ghc/includes/runtime/gmp</Filename>). It comes out as:
798 data Integer = S# Int# -- small integers
799 | J# Int# ByteArray# -- large integers
802 <IndexTerm><Primary>Integer type</Primary></IndexTerm> The primitive
803 ops to support large <Literal>Integers</Literal> use the
804 “pieces” of the representation, and are as follows:
810 negateInteger# :: Int# -> ByteArray# -> Integer
812 {plus,minus,times}Integer#, gcdInteger#,
813 quotInteger#, remInteger#, divExactInteger#
814 :: Int# -> ByteArray#
815 -> Int# -> ByteArray#
816 -> (# Int#, ByteArray# #)
819 :: Int# -> ByteArray#
820 -> Int# -> ByteArray#
821 -> Int# -- -1 for <; 0 for ==; +1 for >
824 :: Int# -> ByteArray#
826 -> Int# -- -1 for <; 0 for ==; +1 for >
829 :: Int# -> ByteArray#
833 divModInteger#, quotRemInteger#
834 :: Int# -> ByteArray#
835 -> Int# -> ByteArray#
836 -> (# Int#, ByteArray#,
839 integer2Int# :: Int# -> ByteArray# -> Int#
841 int2Integer# :: Int# -> Integer -- NB: no error-checking on these two!
842 word2Integer# :: Word# -> Integer
844 addr2Integer# :: Addr# -> Integer
845 -- the Addr# is taken to be a `char *' string
846 -- to be converted into an Integer.
849 <IndexTerm><Primary><literal>negateInteger#</literal></Primary></IndexTerm>
850 <IndexTerm><Primary><literal>plusInteger#</literal></Primary></IndexTerm>
851 <IndexTerm><Primary><literal>minusInteger#</literal></Primary></IndexTerm>
852 <IndexTerm><Primary><literal>timesInteger#</literal></Primary></IndexTerm>
853 <IndexTerm><Primary><literal>quotInteger#</literal></Primary></IndexTerm>
854 <IndexTerm><Primary><literal>remInteger#</literal></Primary></IndexTerm>
855 <IndexTerm><Primary><literal>gcdInteger#</literal></Primary></IndexTerm>
856 <IndexTerm><Primary><literal>gcdIntegerInt#</literal></Primary></IndexTerm>
857 <IndexTerm><Primary><literal>divExactInteger#</literal></Primary></IndexTerm>
858 <IndexTerm><Primary><literal>cmpInteger#</literal></Primary></IndexTerm>
859 <IndexTerm><Primary><literal>divModInteger#</literal></Primary></IndexTerm>
860 <IndexTerm><Primary><literal>quotRemInteger#</literal></Primary></IndexTerm>
861 <IndexTerm><Primary><literal>integer2Int#</literal></Primary></IndexTerm>
862 <IndexTerm><Primary><literal>int2Integer#</literal></Primary></IndexTerm>
863 <IndexTerm><Primary><literal>word2Integer#</literal></Primary></IndexTerm>
864 <IndexTerm><Primary><literal>addr2Integer#</literal></Primary></IndexTerm>
870 <Title>Words and addresses</Title>
873 <IndexTerm><Primary>word, primitive type</Primary></IndexTerm>
874 <IndexTerm><Primary>address, primitive type</Primary></IndexTerm>
875 <IndexTerm><Primary>unsigned integer, primitive type</Primary></IndexTerm>
876 <IndexTerm><Primary>pointer, primitive type</Primary></IndexTerm>
880 A <Literal>Word#</Literal> is used for bit-twiddling operations.
881 It is the same size as an <Literal>Int#</Literal>, but has no sign
882 nor any arithmetic operations.
885 type Word# -- Same size/etc as Int# but *unsigned*
886 type Addr# -- A pointer from outside the "Haskell world" (from C, probably);
887 -- described under "arrays"
890 <IndexTerm><Primary><literal>Word#</literal></Primary></IndexTerm>
891 <IndexTerm><Primary><literal>Addr#</literal></Primary></IndexTerm>
895 <Literal>Word#</Literal>s and <Literal>Addr#</Literal>s have
896 the usual comparison operations. Other
897 unboxed-<Literal>Word</Literal> ops (bit-twiddling and coercions):
903 {gt,ge,eq,ne,lt,le}Word# :: Word# -> Word# -> Bool
905 and#, or#, xor# :: Word# -> Word# -> Word#
908 quotWord#, remWord# :: Word# -> Word# -> Word#
909 -- word (i.e. unsigned) versions are different from int
910 -- versions, so we have to provide these explicitly.
912 not# :: Word# -> Word#
914 shiftL#, shiftRL# :: Word# -> Int# -> Word#
915 -- shift left, right logical
917 int2Word# :: Int# -> Word# -- just a cast, really
918 word2Int# :: Word# -> Int#
921 <IndexTerm><Primary>bit operations, Word and Addr</Primary></IndexTerm>
922 <IndexTerm><Primary><literal>gtWord#</literal></Primary></IndexTerm>
923 <IndexTerm><Primary><literal>geWord#</literal></Primary></IndexTerm>
924 <IndexTerm><Primary><literal>eqWord#</literal></Primary></IndexTerm>
925 <IndexTerm><Primary><literal>neWord#</literal></Primary></IndexTerm>
926 <IndexTerm><Primary><literal>ltWord#</literal></Primary></IndexTerm>
927 <IndexTerm><Primary><literal>leWord#</literal></Primary></IndexTerm>
928 <IndexTerm><Primary><literal>and#</literal></Primary></IndexTerm>
929 <IndexTerm><Primary><literal>or#</literal></Primary></IndexTerm>
930 <IndexTerm><Primary><literal>xor#</literal></Primary></IndexTerm>
931 <IndexTerm><Primary><literal>not#</literal></Primary></IndexTerm>
932 <IndexTerm><Primary><literal>quotWord#</literal></Primary></IndexTerm>
933 <IndexTerm><Primary><literal>remWord#</literal></Primary></IndexTerm>
934 <IndexTerm><Primary><literal>shiftL#</literal></Primary></IndexTerm>
935 <IndexTerm><Primary><literal>shiftRA#</literal></Primary></IndexTerm>
936 <IndexTerm><Primary><literal>shiftRL#</literal></Primary></IndexTerm>
937 <IndexTerm><Primary><literal>int2Word#</literal></Primary></IndexTerm>
938 <IndexTerm><Primary><literal>word2Int#</literal></Primary></IndexTerm>
942 Unboxed-<Literal>Addr</Literal> ops (C casts, really):
945 {gt,ge,eq,ne,lt,le}Addr# :: Addr# -> Addr# -> Bool
947 int2Addr# :: Int# -> Addr#
948 addr2Int# :: Addr# -> Int#
949 addr2Integer# :: Addr# -> (# Int#, ByteArray# #)
952 <IndexTerm><Primary><literal>gtAddr#</literal></Primary></IndexTerm>
953 <IndexTerm><Primary><literal>geAddr#</literal></Primary></IndexTerm>
954 <IndexTerm><Primary><literal>eqAddr#</literal></Primary></IndexTerm>
955 <IndexTerm><Primary><literal>neAddr#</literal></Primary></IndexTerm>
956 <IndexTerm><Primary><literal>ltAddr#</literal></Primary></IndexTerm>
957 <IndexTerm><Primary><literal>leAddr#</literal></Primary></IndexTerm>
958 <IndexTerm><Primary><literal>int2Addr#</literal></Primary></IndexTerm>
959 <IndexTerm><Primary><literal>addr2Int#</literal></Primary></IndexTerm>
960 <IndexTerm><Primary><literal>addr2Integer#</literal></Primary></IndexTerm>
964 The casts between <Literal>Int#</Literal>,
965 <Literal>Word#</Literal> and <Literal>Addr#</Literal>
966 correspond to null operations at the machine level, but are required
967 to keep the Haskell type checker happy.
971 Operations for indexing off of C pointers
972 (<Literal>Addr#</Literal>s) to snatch values are listed under
973 “arrays”.
979 <Title>Arrays</Title>
982 <IndexTerm><Primary>arrays, primitive</Primary></IndexTerm>
986 The type <Literal>Array# elt</Literal> is the type of primitive,
987 unpointed arrays of values of type <Literal>elt</Literal>.
996 <IndexTerm><Primary><literal>Array#</literal></Primary></IndexTerm>
1000 <Literal>Array#</Literal> is more primitive than a Haskell
1001 array—indeed, the Haskell <Literal>Array</Literal> interface is
1002 implemented using <Literal>Array#</Literal>—in that an
1003 <Literal>Array#</Literal> is indexed only by
1004 <Literal>Int#</Literal>s, starting at zero. It is also more
1005 primitive by virtue of being unboxed. That doesn't mean that it isn't
1006 a heap-allocated object—of course, it is. Rather, being unboxed
1007 means that it is represented by a pointer to the array itself, and not
1008 to a thunk which will evaluate to the array (or to bottom). The
1009 components of an <Literal>Array#</Literal> are themselves boxed.
1013 The type <Literal>ByteArray#</Literal> is similar to
1014 <Literal>Array#</Literal>, except that it contains just a string
1015 of (non-pointer) bytes.
1024 <IndexTerm><Primary><literal>ByteArray#</literal></Primary></IndexTerm>
1028 Arrays of these types are useful when a Haskell program wishes to
1029 construct a value to pass to a C procedure. It is also possible to use
1030 them to build (say) arrays of unboxed characters for internal use in a
1031 Haskell program. Given these uses, <Literal>ByteArray#</Literal>
1032 is deliberately a bit vague about the type of its components.
1033 Operations are provided to extract values of type
1034 <Literal>Char#</Literal>, <Literal>Int#</Literal>,
1035 <Literal>Float#</Literal>, <Literal>Double#</Literal>, and
1036 <Literal>Addr#</Literal> from arbitrary offsets within a
1037 <Literal>ByteArray#</Literal>. (For type
1038 <Literal>Foo#</Literal>, the $i$th offset gets you the $i$th
1039 <Literal>Foo#</Literal>, not the <Literal>Foo#</Literal> at
1040 byte-position $i$. Mumble.) (If you want a
1041 <Literal>Word#</Literal>, grab an <Literal>Int#</Literal>,
1046 Lastly, we have static byte-arrays, of type
1047 <Literal>Addr#</Literal> [mentioned previously]. (Remember
1048 the duality between arrays and pointers in C.) Arrays of this types
1049 are represented by a pointer to an array in the world outside Haskell,
1050 so this pointer is not followed by the garbage collector. In other
1051 respects they are just like <Literal>ByteArray#</Literal>. They
1052 are only needed in order to pass values from C to Haskell.
1058 <Title>Reading and writing</Title>
1061 Primitive arrays are linear, and indexed starting at zero.
1065 The size and indices of a <Literal>ByteArray#</Literal>, <Literal>Addr#</Literal>, and
1066 <Literal>MutableByteArray#</Literal> are all in bytes. It's up to the program to
1067 calculate the correct byte offset from the start of the array. This
1068 allows a <Literal>ByteArray#</Literal> to contain a mixture of values of different
1069 type, which is often needed when preparing data for and unpicking
1070 results from C. (Umm…not true of indices…WDP 95/09)
1074 <Emphasis>Should we provide some <Literal>sizeOfDouble#</Literal> constants?</Emphasis>
1078 Out-of-range errors on indexing should be caught by the code which
1079 uses the primitive operation; the primitive operations themselves do
1080 <Emphasis>not</Emphasis> check for out-of-range indexes. The intention is that the
1081 primitive ops compile to one machine instruction or thereabouts.
1085 We use the terms “reading” and “writing” to refer to accessing
1086 <Emphasis>mutable</Emphasis> arrays (see <XRef LinkEnd="sect-mutable">), and
1087 “indexing” to refer to reading a value from an <Emphasis>immutable</Emphasis>
1092 Immutable byte arrays are straightforward to index (all indices in bytes):
1095 indexCharArray# :: ByteArray# -> Int# -> Char#
1096 indexIntArray# :: ByteArray# -> Int# -> Int#
1097 indexAddrArray# :: ByteArray# -> Int# -> Addr#
1098 indexFloatArray# :: ByteArray# -> Int# -> Float#
1099 indexDoubleArray# :: ByteArray# -> Int# -> Double#
1101 indexCharOffAddr# :: Addr# -> Int# -> Char#
1102 indexIntOffAddr# :: Addr# -> Int# -> Int#
1103 indexFloatOffAddr# :: Addr# -> Int# -> Float#
1104 indexDoubleOffAddr# :: Addr# -> Int# -> Double#
1105 indexAddrOffAddr# :: Addr# -> Int# -> Addr#
1106 -- Get an Addr# from an Addr# offset
1109 <IndexTerm><Primary><literal>indexCharArray#</literal></Primary></IndexTerm>
1110 <IndexTerm><Primary><literal>indexIntArray#</literal></Primary></IndexTerm>
1111 <IndexTerm><Primary><literal>indexAddrArray#</literal></Primary></IndexTerm>
1112 <IndexTerm><Primary><literal>indexFloatArray#</literal></Primary></IndexTerm>
1113 <IndexTerm><Primary><literal>indexDoubleArray#</literal></Primary></IndexTerm>
1114 <IndexTerm><Primary><literal>indexCharOffAddr#</literal></Primary></IndexTerm>
1115 <IndexTerm><Primary><literal>indexIntOffAddr#</literal></Primary></IndexTerm>
1116 <IndexTerm><Primary><literal>indexFloatOffAddr#</literal></Primary></IndexTerm>
1117 <IndexTerm><Primary><literal>indexDoubleOffAddr#</literal></Primary></IndexTerm>
1118 <IndexTerm><Primary><literal>indexAddrOffAddr#</literal></Primary></IndexTerm>
1122 The last of these, <Function>indexAddrOffAddr#</Function>, extracts an <Literal>Addr#</Literal> using an offset
1123 from another <Literal>Addr#</Literal>, thereby providing the ability to follow a chain of
1128 Something a bit more interesting goes on when indexing arrays of boxed
1129 objects, because the result is simply the boxed object. So presumably
1130 it should be entered—we never usually return an unevaluated
1131 object! This is a pain: primitive ops aren't supposed to do
1132 complicated things like enter objects. The current solution is to
1133 return a single element unboxed tuple (see <XRef LinkEnd="unboxed-tuples">).
1139 indexArray# :: Array# elt -> Int# -> (# elt #)
1142 <IndexTerm><Primary><literal>indexArray#</literal></Primary></IndexTerm>
1148 <Title>The state type</Title>
1151 <IndexTerm><Primary><literal>state, primitive type</literal></Primary></IndexTerm>
1152 <IndexTerm><Primary><literal>State#</literal></Primary></IndexTerm>
1156 The primitive type <Literal>State#</Literal> represents the state of a state
1157 transformer. It is parameterised on the desired type of state, which
1158 serves to keep states from distinct threads distinct from one another.
1159 But the <Emphasis>only</Emphasis> effect of this parameterisation is in the type
1160 system: all values of type <Literal>State#</Literal> are represented in the same way.
1161 Indeed, they are all represented by nothing at all! The code
1162 generator “knows” to generate no code, and allocate no registers
1163 etc, for primitive states.
1175 The type <Literal>GHC.RealWorld</Literal> is truly opaque: there are no values defined
1176 of this type, and no operations over it. It is “primitive” in that
1177 sense - but it is <Emphasis>not unlifted!</Emphasis> Its only role in life is to be
1178 the type which distinguishes the <Literal>IO</Literal> state transformer.
1192 <Title>State of the world</Title>
1195 A single, primitive, value of type <Literal>State# RealWorld</Literal> is provided.
1201 realWorld# :: State# RealWorld
1204 <IndexTerm><Primary>realWorld# state object</Primary></IndexTerm>
1208 (Note: in the compiler, not a <Literal>PrimOp</Literal>; just a mucho magic
1209 <Literal>Id</Literal>. Exported from <Literal>GHC</Literal>, though).
1214 <Sect2 id="sect-mutable">
1215 <Title>Mutable arrays</Title>
1218 <IndexTerm><Primary>mutable arrays</Primary></IndexTerm>
1219 <IndexTerm><Primary>arrays, mutable</Primary></IndexTerm>
1220 Corresponding to <Literal>Array#</Literal> and <Literal>ByteArray#</Literal>, we have the types of
1221 mutable versions of each. In each case, the representation is a
1222 pointer to a suitable block of (mutable) heap-allocated storage.
1228 type MutableArray# s elt
1229 type MutableByteArray# s
1232 <IndexTerm><Primary><literal>MutableArray#</literal></Primary></IndexTerm>
1233 <IndexTerm><Primary><literal>MutableByteArray#</literal></Primary></IndexTerm>
1237 <Title>Allocation</Title>
1240 <IndexTerm><Primary>mutable arrays, allocation</Primary></IndexTerm>
1241 <IndexTerm><Primary>arrays, allocation</Primary></IndexTerm>
1242 <IndexTerm><Primary>allocation, of mutable arrays</Primary></IndexTerm>
1246 Mutable arrays can be allocated. Only pointer-arrays are initialised;
1247 arrays of non-pointers are filled in by “user code” rather than by
1248 the array-allocation primitive. Reason: only the pointer case has to
1249 worry about GC striking with a partly-initialised array.
1255 newArray# :: Int# -> elt -> State# s -> (# State# s, MutableArray# s elt #)
1257 newCharArray# :: Int# -> State# s -> (# State# s, MutableByteArray# s elt #)
1258 newIntArray# :: Int# -> State# s -> (# State# s, MutableByteArray# s elt #)
1259 newAddrArray# :: Int# -> State# s -> (# State# s, MutableByteArray# s elt #)
1260 newFloatArray# :: Int# -> State# s -> (# State# s, MutableByteArray# s elt #)
1261 newDoubleArray# :: Int# -> State# s -> (# State# s, MutableByteArray# s elt #)
1264 <IndexTerm><Primary><literal>newArray#</literal></Primary></IndexTerm>
1265 <IndexTerm><Primary><literal>newCharArray#</literal></Primary></IndexTerm>
1266 <IndexTerm><Primary><literal>newIntArray#</literal></Primary></IndexTerm>
1267 <IndexTerm><Primary><literal>newAddrArray#</literal></Primary></IndexTerm>
1268 <IndexTerm><Primary><literal>newFloatArray#</literal></Primary></IndexTerm>
1269 <IndexTerm><Primary><literal>newDoubleArray#</literal></Primary></IndexTerm>
1273 The size of a <Literal>ByteArray#</Literal> is given in bytes.
1279 <Title>Reading and writing</Title>
1282 <IndexTerm><Primary>arrays, reading and writing</Primary></IndexTerm>
1288 readArray# :: MutableArray# s elt -> Int# -> State# s -> (# State# s, elt #)
1289 readCharArray# :: MutableByteArray# s -> Int# -> State# s -> (# State# s, Char# #)
1290 readIntArray# :: MutableByteArray# s -> Int# -> State# s -> (# State# s, Int# #)
1291 readAddrArray# :: MutableByteArray# s -> Int# -> State# s -> (# State# s, Addr# #)
1292 readFloatArray# :: MutableByteArray# s -> Int# -> State# s -> (# State# s, Float# #)
1293 readDoubleArray# :: MutableByteArray# s -> Int# -> State# s -> (# State# s, Double# #)
1295 writeArray# :: MutableArray# s elt -> Int# -> elt -> State# s -> State# s
1296 writeCharArray# :: MutableByteArray# s -> Int# -> Char# -> State# s -> State# s
1297 writeIntArray# :: MutableByteArray# s -> Int# -> Int# -> State# s -> State# s
1298 writeAddrArray# :: MutableByteArray# s -> Int# -> Addr# -> State# s -> State# s
1299 writeFloatArray# :: MutableByteArray# s -> Int# -> Float# -> State# s -> State# s
1300 writeDoubleArray# :: MutableByteArray# s -> Int# -> Double# -> State# s -> State# s
1303 <IndexTerm><Primary><literal>readArray#</literal></Primary></IndexTerm>
1304 <IndexTerm><Primary><literal>readCharArray#</literal></Primary></IndexTerm>
1305 <IndexTerm><Primary><literal>readIntArray#</literal></Primary></IndexTerm>
1306 <IndexTerm><Primary><literal>readAddrArray#</literal></Primary></IndexTerm>
1307 <IndexTerm><Primary><literal>readFloatArray#</literal></Primary></IndexTerm>
1308 <IndexTerm><Primary><literal>readDoubleArray#</literal></Primary></IndexTerm>
1309 <IndexTerm><Primary><literal>writeArray#</literal></Primary></IndexTerm>
1310 <IndexTerm><Primary><literal>writeCharArray#</literal></Primary></IndexTerm>
1311 <IndexTerm><Primary><literal>writeIntArray#</literal></Primary></IndexTerm>
1312 <IndexTerm><Primary><literal>writeAddrArray#</literal></Primary></IndexTerm>
1313 <IndexTerm><Primary><literal>writeFloatArray#</literal></Primary></IndexTerm>
1314 <IndexTerm><Primary><literal>writeDoubleArray#</literal></Primary></IndexTerm>
1320 <Title>Equality</Title>
1323 <IndexTerm><Primary>arrays, testing for equality</Primary></IndexTerm>
1327 One can take “equality” of mutable arrays. What is compared is the
1328 <Emphasis>name</Emphasis> or reference to the mutable array, not its contents.
1334 sameMutableArray# :: MutableArray# s elt -> MutableArray# s elt -> Bool
1335 sameMutableByteArray# :: MutableByteArray# s -> MutableByteArray# s -> Bool
1338 <IndexTerm><Primary><literal>sameMutableArray#</literal></Primary></IndexTerm>
1339 <IndexTerm><Primary><literal>sameMutableByteArray#</literal></Primary></IndexTerm>
1345 <Title>Freezing mutable arrays</Title>
1348 <IndexTerm><Primary>arrays, freezing mutable</Primary></IndexTerm>
1349 <IndexTerm><Primary>freezing mutable arrays</Primary></IndexTerm>
1350 <IndexTerm><Primary>mutable arrays, freezing</Primary></IndexTerm>
1354 Only unsafe-freeze has a primitive. (Safe freeze is done directly in Haskell
1355 by copying the array and then using <Function>unsafeFreeze</Function>.)
1361 unsafeFreezeArray# :: MutableArray# s elt -> State# s -> (# State# s, Array# s elt #)
1362 unsafeFreezeByteArray# :: MutableByteArray# s -> State# s -> (# State# s, ByteArray# #)
1365 <IndexTerm><Primary><literal>unsafeFreezeArray#</literal></Primary></IndexTerm>
1366 <IndexTerm><Primary><literal>unsafeFreezeByteArray#</literal></Primary></IndexTerm>
1374 <Title>Synchronizing variables (M-vars)</Title>
1377 <IndexTerm><Primary>synchronising variables (M-vars)</Primary></IndexTerm>
1378 <IndexTerm><Primary>M-Vars</Primary></IndexTerm>
1382 Synchronising variables are the primitive type used to implement
1383 Concurrent Haskell's MVars (see the Concurrent Haskell paper for
1384 the operational behaviour of these operations).
1390 type MVar# s elt -- primitive
1392 newMVar# :: State# s -> (# State# s, MVar# s elt #)
1393 takeMVar# :: SynchVar# s elt -> State# s -> (# State# s, elt #)
1394 putMVar# :: SynchVar# s elt -> State# s -> State# s
1397 <IndexTerm><Primary><literal>SynchVar#</literal></Primary></IndexTerm>
1398 <IndexTerm><Primary><literal>newSynchVar#</literal></Primary></IndexTerm>
1399 <IndexTerm><Primary><literal>takeMVar</literal></Primary></IndexTerm>
1400 <IndexTerm><Primary><literal>putMVar</literal></Primary></IndexTerm>
1407 <Sect1 id="glasgow-ST-monad">
1408 <Title>Primitive state-transformer monad
1412 <IndexTerm><Primary>state transformers (Glasgow extensions)</Primary></IndexTerm>
1413 <IndexTerm><Primary>ST monad (Glasgow extension)</Primary></IndexTerm>
1417 This monad underlies our implementation of arrays, mutable and
1418 immutable, and our implementation of I/O, including “C calls”.
1422 The <Literal>ST</Literal> library, which provides access to the
1423 <Function>ST</Function> monad, is described in <xref
1429 <Sect1 id="glasgow-prim-arrays">
1430 <Title>Primitive arrays, mutable and otherwise
1434 <IndexTerm><Primary>primitive arrays (Glasgow extension)</Primary></IndexTerm>
1435 <IndexTerm><Primary>arrays, primitive (Glasgow extension)</Primary></IndexTerm>
1439 GHC knows about quite a few flavours of Large Swathes of Bytes.
1443 First, GHC distinguishes between primitive arrays of (boxed) Haskell
1444 objects (type <Literal>Array# obj</Literal>) and primitive arrays of bytes (type
1445 <Literal>ByteArray#</Literal>).
1449 Second, it distinguishes between…
1453 <Term>Immutable:</Term>
1456 Arrays that do not change (as with “standard” Haskell arrays); you
1457 can only read from them. Obviously, they do not need the care and
1458 attention of the state-transformer monad.
1463 <Term>Mutable:</Term>
1466 Arrays that may be changed or “mutated.” All the operations on them
1467 live within the state-transformer monad and the updates happen
1468 <Emphasis>in-place</Emphasis>.
1473 <Term>“Static” (in C land):</Term>
1476 A C routine may pass an <Literal>Addr#</Literal> pointer back into Haskell land. There
1477 are then primitive operations with which you may merrily grab values
1478 over in C land, by indexing off the “static” pointer.
1483 <Term>“Stable” pointers:</Term>
1486 If, for some reason, you wish to hand a Haskell pointer (i.e.,
1487 <Emphasis>not</Emphasis> an unboxed value) to a C routine, you first make the
1488 pointer “stable,” so that the garbage collector won't forget that it
1489 exists. That is, GHC provides a safe way to pass Haskell pointers to
1494 Please see <XRef LinkEnd="sec-stable-pointers"> for more details.
1499 <Term>“Foreign objects”:</Term>
1502 A “foreign object” is a safe way to pass an external object (a
1503 C-allocated pointer, say) to Haskell and have Haskell do the Right
1504 Thing when it no longer references the object. So, for example, C
1505 could pass a large bitmap over to Haskell and say “please free this
1506 memory when you're done with it.”
1510 Please see <XRef LinkEnd="sec-ForeignObj"> for more details.
1518 The libraries documentatation gives more details on all these
1519 “primitive array” types and the operations on them.
1525 <Sect1 id="pattern-guards">
1526 <Title>Pattern guards</Title>
1529 <IndexTerm><Primary>Pattern guards (Glasgow extension)</Primary></IndexTerm>
1530 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.)
1534 Suppose we have an abstract data type of finite maps, with a
1538 lookup :: FiniteMap -> Int -> Maybe Int
1541 The lookup returns <Function>Nothing</Function> if the supplied key is not in the domain of the mapping, and <Function>(Just v)</Function> otherwise,
1542 where <VarName>v</VarName> is the value that the key maps to. Now consider the following definition:
1546 clunky env var1 var2 | ok1 && ok2 = val1 + val2
1547 | otherwise = var1 + var2
1549 m1 = lookup env var1
1550 m2 = lookup env var2
1551 ok1 = maybeToBool m1
1552 ok2 = maybeToBool m2
1553 val1 = expectJust m1
1554 val2 = expectJust m2
1558 The auxiliary functions are
1562 maybeToBool :: Maybe a -> Bool
1563 maybeToBool (Just x) = True
1564 maybeToBool Nothing = False
1566 expectJust :: Maybe a -> a
1567 expectJust (Just x) = x
1568 expectJust Nothing = error "Unexpected Nothing"
1572 What is <Function>clunky</Function> doing? The guard <Literal>ok1 &&
1573 ok2</Literal> checks that both lookups succeed, using
1574 <Function>maybeToBool</Function> to convert the <Function>Maybe</Function>
1575 types to booleans. The (lazily evaluated) <Function>expectJust</Function>
1576 calls extract the values from the results of the lookups, and binds the
1577 returned values to <VarName>val1</VarName> and <VarName>val2</VarName>
1578 respectively. If either lookup fails, then clunky takes the
1579 <Literal>otherwise</Literal> case and returns the sum of its arguments.
1583 This is certainly legal Haskell, but it is a tremendously verbose and
1584 un-obvious way to achieve the desired effect. Arguably, a more direct way
1585 to write clunky would be to use case expressions:
1589 clunky env var1 var1 = case lookup env var1 of
1591 Just val1 -> case lookup env var2 of
1593 Just val2 -> val1 + val2
1599 This is a bit shorter, but hardly better. Of course, we can rewrite any set
1600 of pattern-matching, guarded equations as case expressions; that is
1601 precisely what the compiler does when compiling equations! The reason that
1602 Haskell provides guarded equations is because they allow us to write down
1603 the cases we want to consider, one at a time, independently of each other.
1604 This structure is hidden in the case version. Two of the right-hand sides
1605 are really the same (<Function>fail</Function>), and the whole expression
1606 tends to become more and more indented.
1610 Here is how I would write clunky:
1614 clunky env var1 var1
1615 | Just val1 <- lookup env var1
1616 , Just val2 <- lookup env var2
1618 ...other equations for clunky...
1622 The semantics should be clear enough. The qualifers are matched in order.
1623 For a <Literal><-</Literal> qualifier, which I call a pattern guard, the
1624 right hand side is evaluated and matched against the pattern on the left.
1625 If the match fails then the whole guard fails and the next equation is
1626 tried. If it succeeds, then the appropriate binding takes place, and the
1627 next qualifier is matched, in the augmented environment. Unlike list
1628 comprehensions, however, the type of the expression to the right of the
1629 <Literal><-</Literal> is the same as the type of the pattern to its
1630 left. The bindings introduced by pattern guards scope over all the
1631 remaining guard qualifiers, and over the right hand side of the equation.
1635 Just as with list comprehensions, boolean expressions can be freely mixed
1636 with among the pattern guards. For example:
1647 Haskell's current guards therefore emerge as a special case, in which the
1648 qualifier list has just one element, a boolean expression.
1652 <sect1 id="sec-ffi">
1653 <title>The foreign interface</title>
1655 <para>The foreign interface consists of the following components:</para>
1659 <para>The Foreign Function Interface language specification
1660 (included in this manual, in <xref linkend="ffi">).</para>
1664 <para>The <literal>Foreign</literal> module (see <xref
1665 linkend="sec-Foreign">) collects together several interfaces
1666 which are useful in specifying foreign language
1667 interfaces, including the following:</para>
1671 <para>The <literal>ForeignObj</literal> module (see <xref
1672 linkend="sec-ForeignObj">), for managing pointers from
1673 Haskell into the outside world.</para>
1677 <para>The <literal>StablePtr</literal> module (see <xref
1678 linkend="sec-stable-pointers">), for managing pointers
1679 into Haskell from the outside world.</para>
1683 <para>The <literal>CTypes</literal> module (see <xref
1684 linkend="sec-CTypes">) gives Haskell equivalents for the
1685 standard C datatypes, for use in making Haskell bindings
1686 to existing C libraries.</para>
1690 <para>The <literal>CTypesISO</literal> module (see <xref
1691 linkend="sec-CTypesISO">) gives Haskell equivalents for C
1692 types defined by the ISO C standard.</para>
1696 <para>The <literal>Storable</literal> library, for
1697 primitive marshalling of data types between Haskell and
1698 the foreign language.</para>
1705 <para>The following sections also give some hints and tips on the use
1706 of the foreign function interface in GHC.</para>
1708 <Sect2 id="glasgow-foreign-headers">
1709 <Title>Using function headers
1713 <IndexTerm><Primary>C calls, function headers</Primary></IndexTerm>
1717 When generating C (using the <Option>-fvia-C</Option> directive), one can assist the
1718 C compiler in detecting type errors by using the <Command>-#include</Command> directive
1719 to provide <Filename>.h</Filename> files containing function headers.
1731 void initialiseEFS (HsInt size);
1732 HsInt terminateEFS (void);
1733 HsForeignObj emptyEFS(void);
1734 HsForeignObj updateEFS (HsForeignObj a, HsInt i, HsInt x);
1735 HsInt lookupEFS (HsForeignObj a, HsInt i);
1739 <para>The types <literal>HsInt</literal>,
1740 <literal>HsForeignObj</literal> etc. are described in <xref
1741 linkend="sec-mapping-table">.</Para>
1743 <Para>Note that this approach is only
1744 <Emphasis>essential</Emphasis> for returning
1745 <Literal>float</Literal>s (or if <Literal>sizeof(int) !=
1746 sizeof(int *)</Literal> on your architecture) but is a Good
1747 Thing for anyone who cares about writing solid code. You're
1748 crazy not to do it.</Para>
1754 <Sect1 id="multi-param-type-classes">
1755 <Title>Multi-parameter type classes
1759 This section documents GHC's implementation of multi-parameter type
1760 classes. There's lots of background in the paper <ULink
1761 URL="http://research.microsoft.com/~simonpj/multi.ps.gz" >Type
1762 classes: exploring the design space</ULink > (Simon Peyton Jones, Mark
1763 Jones, Erik Meijer).
1767 I'd like to thank people who reported shorcomings in the GHC 3.02
1768 implementation. Our default decisions were all conservative ones, and
1769 the experience of these heroic pioneers has given useful concrete
1770 examples to support several generalisations. (These appear below as
1771 design choices not implemented in 3.02.)
1775 I've discussed these notes with Mark Jones, and I believe that Hugs
1776 will migrate towards the same design choices as I outline here.
1777 Thanks to him, and to many others who have offered very useful
1782 <Title>Types</Title>
1785 There are the following restrictions on the form of a qualified
1792 forall tv1..tvn (c1, ...,cn) => type
1798 (Here, I write the "foralls" explicitly, although the Haskell source
1799 language omits them; in Haskell 1.4, all the free type variables of an
1800 explicit source-language type signature are universally quantified,
1801 except for the class type variables in a class declaration. However,
1802 in GHC, you can give the foralls if you want. See <XRef LinkEnd="universal-quantification">).
1811 <Emphasis>Each universally quantified type variable
1812 <Literal>tvi</Literal> must be mentioned (i.e. appear free) in <Literal>type</Literal></Emphasis>.
1814 The reason for this is that a value with a type that does not obey
1815 this restriction could not be used without introducing
1816 ambiguity. Here, for example, is an illegal type:
1820 forall a. Eq a => Int
1824 When a value with this type was used, the constraint <Literal>Eq tv</Literal>
1825 would be introduced where <Literal>tv</Literal> is a fresh type variable, and
1826 (in the dictionary-translation implementation) the value would be
1827 applied to a dictionary for <Literal>Eq tv</Literal>. The difficulty is that we
1828 can never know which instance of <Literal>Eq</Literal> to use because we never
1829 get any more information about <Literal>tv</Literal>.
1836 <Emphasis>Every constraint <Literal>ci</Literal> must mention at least one of the
1837 universally quantified type variables <Literal>tvi</Literal></Emphasis>.
1839 For example, this type is OK because <Literal>C a b</Literal> mentions the
1840 universally quantified type variable <Literal>b</Literal>:
1844 forall a. C a b => burble
1848 The next type is illegal because the constraint <Literal>Eq b</Literal> does not
1849 mention <Literal>a</Literal>:
1853 forall a. Eq b => burble
1857 The reason for this restriction is milder than the other one. The
1858 excluded types are never useful or necessary (because the offending
1859 context doesn't need to be witnessed at this point; it can be floated
1860 out). Furthermore, floating them out increases sharing. Lastly,
1861 excluding them is a conservative choice; it leaves a patch of
1862 territory free in case we need it later.
1872 These restrictions apply to all types, whether declared in a type signature
1877 Unlike Haskell 1.4, constraints in types do <Emphasis>not</Emphasis> have to be of
1878 the form <Emphasis>(class type-variables)</Emphasis>. Thus, these type signatures
1885 f :: Eq (m a) => [m a] -> [m a]
1892 This choice recovers principal types, a property that Haskell 1.4 does not have.
1898 <Title>Class declarations</Title>
1906 <Emphasis>Multi-parameter type classes are permitted</Emphasis>. For example:
1910 class Collection c a where
1911 union :: c a -> c a -> c a
1922 <Emphasis>The class hierarchy must be acyclic</Emphasis>. However, the definition
1923 of "acyclic" involves only the superclass relationships. For example,
1929 op :: D b => a -> b -> b
1932 class C a => D a where { ... }
1936 Here, <Literal>C</Literal> is a superclass of <Literal>D</Literal>, but it's OK for a
1937 class operation <Literal>op</Literal> of <Literal>C</Literal> to mention <Literal>D</Literal>. (It
1938 would not be OK for <Literal>D</Literal> to be a superclass of <Literal>C</Literal>.)
1945 <Emphasis>There are no restrictions on the context in a class declaration
1946 (which introduces superclasses), except that the class hierarchy must
1947 be acyclic</Emphasis>. So these class declarations are OK:
1951 class Functor (m k) => FiniteMap m k where
1954 class (Monad m, Monad (t m)) => Transform t m where
1955 lift :: m a -> (t m) a
1964 <Emphasis>In the signature of a class operation, every constraint
1965 must mention at least one type variable that is not a class type
1966 variable</Emphasis>.
1972 class Collection c a where
1973 mapC :: Collection c b => (a->b) -> c a -> c b
1977 is OK because the constraint <Literal>(Collection a b)</Literal> mentions
1978 <Literal>b</Literal>, even though it also mentions the class variable
1979 <Literal>a</Literal>. On the other hand:
1984 op :: Eq a => (a,b) -> (a,b)
1988 is not OK because the constraint <Literal>(Eq a)</Literal> mentions on the class
1989 type variable <Literal>a</Literal>, but not <Literal>b</Literal>. However, any such
1990 example is easily fixed by moving the offending context up to the
1995 class Eq a => C a where
2000 A yet more relaxed rule would allow the context of a class-op signature
2001 to mention only class type variables. However, that conflicts with
2002 Rule 1(b) for types above.
2009 <Emphasis>The type of each class operation must mention <Emphasis>all</Emphasis> of
2010 the class type variables</Emphasis>. For example:
2014 class Coll s a where
2016 insert :: s -> a -> s
2020 is not OK, because the type of <Literal>empty</Literal> doesn't mention
2021 <Literal>a</Literal>. This rule is a consequence of Rule 1(a), above, for
2022 types, and has the same motivation.
2024 Sometimes, offending class declarations exhibit misunderstandings. For
2025 example, <Literal>Coll</Literal> might be rewritten
2029 class Coll s a where
2031 insert :: s a -> a -> s a
2035 which makes the connection between the type of a collection of
2036 <Literal>a</Literal>'s (namely <Literal>(s a)</Literal>) and the element type <Literal>a</Literal>.
2037 Occasionally this really doesn't work, in which case you can split the
2045 class CollE s => Coll s a where
2046 insert :: s -> a -> s
2059 <Sect2 id="instance-decls">
2060 <Title>Instance declarations</Title>
2068 <Emphasis>Instance declarations may not overlap</Emphasis>. The two instance
2073 instance context1 => C type1 where ...
2074 instance context2 => C type2 where ...
2078 "overlap" if <Literal>type1</Literal> and <Literal>type2</Literal> unify
2080 However, if you give the command line option
2081 <Option>-fallow-overlapping-instances</Option><IndexTerm><Primary>-fallow-overlapping-instances
2082 option</Primary></IndexTerm> then two overlapping instance declarations are permitted
2090 EITHER <Literal>type1</Literal> and <Literal>type2</Literal> do not unify
2096 OR <Literal>type2</Literal> is a substitution instance of <Literal>type1</Literal>
2097 (but not identical to <Literal>type1</Literal>)
2110 Notice that these rules
2117 make it clear which instance decl to use
2118 (pick the most specific one that matches)
2125 do not mention the contexts <Literal>context1</Literal>, <Literal>context2</Literal>
2126 Reason: you can pick which instance decl
2127 "matches" based on the type.
2134 Regrettably, GHC doesn't guarantee to detect overlapping instance
2135 declarations if they appear in different modules. GHC can "see" the
2136 instance declarations in the transitive closure of all the modules
2137 imported by the one being compiled, so it can "see" all instance decls
2138 when it is compiling <Literal>Main</Literal>. However, it currently chooses not
2139 to look at ones that can't possibly be of use in the module currently
2140 being compiled, in the interests of efficiency. (Perhaps we should
2141 change that decision, at least for <Literal>Main</Literal>.)
2148 <Emphasis>There are no restrictions on the type in an instance
2149 <Emphasis>head</Emphasis>, except that at least one must not be a type variable</Emphasis>.
2150 The instance "head" is the bit after the "=>" in an instance decl. For
2151 example, these are OK:
2155 instance C Int a where ...
2157 instance D (Int, Int) where ...
2159 instance E [[a]] where ...
2163 Note that instance heads <Emphasis>may</Emphasis> contain repeated type variables.
2164 For example, this is OK:
2168 instance Stateful (ST s) (MutVar s) where ...
2172 The "at least one not a type variable" restriction is to ensure that
2173 context reduction terminates: each reduction step removes one type
2174 constructor. For example, the following would make the type checker
2175 loop if it wasn't excluded:
2179 instance C a => C a where ...
2183 There are two situations in which the rule is a bit of a pain. First,
2184 if one allows overlapping instance declarations then it's quite
2185 convenient to have a "default instance" declaration that applies if
2186 something more specific does not:
2195 Second, sometimes you might want to use the following to get the
2196 effect of a "class synonym":
2200 class (C1 a, C2 a, C3 a) => C a where { }
2202 instance (C1 a, C2 a, C3 a) => C a where { }
2206 This allows you to write shorter signatures:
2218 f :: (C1 a, C2 a, C3 a) => ...
2222 I'm on the lookout for a simple rule that preserves decidability while
2223 allowing these idioms. The experimental flag
2224 <Option>-fallow-undecidable-instances</Option><IndexTerm><Primary>-fallow-undecidable-instances
2225 option</Primary></IndexTerm> lifts this restriction, allowing all the types in an
2226 instance head to be type variables.
2233 <Emphasis>Unlike Haskell 1.4, instance heads may use type
2234 synonyms</Emphasis>. As always, using a type synonym is just shorthand for
2235 writing the RHS of the type synonym definition. For example:
2239 type Point = (Int,Int)
2240 instance C Point where ...
2241 instance C [Point] where ...
2245 is legal. However, if you added
2249 instance C (Int,Int) where ...
2253 as well, then the compiler will complain about the overlapping
2254 (actually, identical) instance declarations. As always, type synonyms
2255 must be fully applied. You cannot, for example, write:
2260 instance Monad P where ...
2264 This design decision is independent of all the others, and easily
2265 reversed, but it makes sense to me.
2272 <Emphasis>The types in an instance-declaration <Emphasis>context</Emphasis> must all
2273 be type variables</Emphasis>. Thus
2277 instance C a b => Eq (a,b) where ...
2285 instance C Int b => Foo b where ...
2289 is not OK. Again, the intent here is to make sure that context
2290 reduction terminates.
2292 Voluminous correspondence on the Haskell mailing list has convinced me
2293 that it's worth experimenting with a more liberal rule. If you use
2294 the flag <Option>-fallow-undecidable-instances</Option> can use arbitrary
2295 types in an instance context. Termination is ensured by having a
2296 fixed-depth recursion stack. If you exceed the stack depth you get a
2297 sort of backtrace, and the opportunity to increase the stack depth
2298 with <Option>-fcontext-stack</Option><Emphasis>N</Emphasis>.
2311 <Sect1 id="implicit-parameters">
2312 <Title>Implicit parameters
2315 <Para> Implicit paramters are implemented as described in
2316 "Implicit parameters: dynamic scoping with static types",
2317 J Lewis, MB Shields, E Meijer, J Launchbury,
2318 27th ACM Symposium on Principles of Programming Languages (POPL'00),
2323 There should be more documentation, but there isn't (yet). Yell if you need it.
2327 <Para> You can't have an implicit parameter in the context of a class or instance
2328 declaration. For example, both these declarations are illegal:
2330 class (?x::Int) => C a where ...
2331 instance (?x::a) => Foo [a] where ...
2333 Reason: exactly which implicit parameter you pick up depends on exactly where
2334 you invoke a function. But the ``invocation'' of instance declarations is done
2335 behind the scenes by the compiler, so it's hard to figure out exactly where it is done.
2336 Easiest thing is to outlaw the offending types.
2344 <Sect1 id="functional-dependencies">
2345 <Title>Functional dependencies
2348 <Para> Functional dependencies are implemented as described by Mark Jones
2349 in "Type Classes with Functional Dependencies", Mark P. Jones,
2350 In Proceedings of the 9th European Symposium on Programming,
2351 ESOP 2000, Berlin, Germany, March 2000, Springer-Verlag LNCS 1782.
2355 There should be more documentation, but there isn't (yet). Yell if you need it.
2360 <Sect1 id="universal-quantification">
2361 <Title>Explicit universal quantification
2365 GHC's type system supports explicit universal quantification in
2366 constructor fields and function arguments. This is useful for things
2367 like defining <Literal>runST</Literal> from the state-thread world.
2368 GHC's syntax for this now agrees with Hugs's, namely:
2374 forall a b. (Ord a, Eq b) => a -> b -> a
2380 The context is, of course, optional. You can't use <Literal>forall</Literal> as
2381 a type variable any more!
2385 Haskell type signatures are implicitly quantified. The <Literal>forall</Literal>
2386 allows us to say exactly what this means. For example:
2404 g :: forall b. (b -> b)
2410 The two are treated identically.
2414 <Title>Universally-quantified data type fields
2418 In a <Literal>data</Literal> or <Literal>newtype</Literal> declaration one can quantify
2419 the types of the constructor arguments. Here are several examples:
2425 data T a = T1 (forall b. b -> b -> b) a
2427 data MonadT m = MkMonad { return :: forall a. a -> m a,
2428 bind :: forall a b. m a -> (a -> m b) -> m b
2431 newtype Swizzle = MkSwizzle (Ord a => [a] -> [a])
2437 The constructors now have so-called <Emphasis>rank 2</Emphasis> polymorphic
2438 types, in which there is a for-all in the argument types.:
2444 T1 :: forall a. (forall b. b -> b -> b) -> a -> T a
2445 MkMonad :: forall m. (forall a. a -> m a)
2446 -> (forall a b. m a -> (a -> m b) -> m b)
2448 MkSwizzle :: (Ord a => [a] -> [a]) -> Swizzle
2454 Notice that you don't need to use a <Literal>forall</Literal> if there's an
2455 explicit context. For example in the first argument of the
2456 constructor <Function>MkSwizzle</Function>, an implicit "<Literal>forall a.</Literal>" is
2457 prefixed to the argument type. The implicit <Literal>forall</Literal>
2458 quantifies all type variables that are not already in scope, and are
2459 mentioned in the type quantified over.
2463 As for type signatures, implicit quantification happens for non-overloaded
2464 types too. So if you write this:
2467 data T a = MkT (Either a b) (b -> b)
2470 it's just as if you had written this:
2473 data T a = MkT (forall b. Either a b) (forall b. b -> b)
2476 That is, since the type variable <Literal>b</Literal> isn't in scope, it's
2477 implicitly universally quantified. (Arguably, it would be better
2478 to <Emphasis>require</Emphasis> explicit quantification on constructor arguments
2479 where that is what is wanted. Feedback welcomed.)
2485 <Title>Construction </Title>
2488 You construct values of types <Literal>T1, MonadT, Swizzle</Literal> by applying
2489 the constructor to suitable values, just as usual. For example,
2495 (T1 (\xy->x) 3) :: T Int
2497 (MkSwizzle sort) :: Swizzle
2498 (MkSwizzle reverse) :: Swizzle
2505 MkMonad r b) :: MonadT Maybe
2511 The type of the argument can, as usual, be more general than the type
2512 required, as <Literal>(MkSwizzle reverse)</Literal> shows. (<Function>reverse</Function>
2513 does not need the <Literal>Ord</Literal> constraint.)
2519 <Title>Pattern matching</Title>
2522 When you use pattern matching, the bound variables may now have
2523 polymorphic types. For example:
2529 f :: T a -> a -> (a, Char)
2530 f (T1 f k) x = (f k x, f 'c' 'd')
2532 g :: (Ord a, Ord b) => Swizzle -> [a] -> (a -> b) -> [b]
2533 g (MkSwizzle s) xs f = s (map f (s xs))
2535 h :: MonadT m -> [m a] -> m [a]
2536 h m [] = return m []
2537 h m (x:xs) = bind m x $ \y ->
2538 bind m (h m xs) $ \ys ->
2545 In the function <Function>h</Function> we use the record selectors <Literal>return</Literal>
2546 and <Literal>bind</Literal> to extract the polymorphic bind and return functions
2547 from the <Literal>MonadT</Literal> data structure, rather than using pattern
2552 You cannot pattern-match against an argument that is polymorphic.
2556 newtype TIM s a = TIM (ST s (Maybe a))
2558 runTIM :: (forall s. TIM s a) -> Maybe a
2559 runTIM (TIM m) = runST m
2565 Here the pattern-match fails, because you can't pattern-match against
2566 an argument of type <Literal>(forall s. TIM s a)</Literal>. Instead you
2567 must bind the variable and pattern match in the right hand side:
2570 runTIM :: (forall s. TIM s a) -> Maybe a
2571 runTIM tm = case tm of { TIM m -> runST m }
2574 The <Literal>tm</Literal> on the right hand side is (invisibly) instantiated, like
2575 any polymorphic value at its occurrence site, and now you can pattern-match
2582 <Title>The partial-application restriction</Title>
2585 There is really only one way in which data structures with polymorphic
2586 components might surprise you: you must not partially apply them.
2587 For example, this is illegal:
2593 map MkSwizzle [sort, reverse]
2599 The restriction is this: <Emphasis>every subexpression of the program must
2600 have a type that has no for-alls, except that in a function
2601 application (f e1…en) the partial applications are not subject to
2602 this rule</Emphasis>. The restriction makes type inference feasible.
2606 In the illegal example, the sub-expression <Literal>MkSwizzle</Literal> has the
2607 polymorphic type <Literal>(Ord b => [b] -> [b]) -> Swizzle</Literal> and is not
2608 a sub-expression of an enclosing application. On the other hand, this
2615 map (T1 (\a b -> a)) [1,2,3]
2621 even though it involves a partial application of <Function>T1</Function>, because
2622 the sub-expression <Literal>T1 (\a b -> a)</Literal> has type <Literal>Int -> T
2629 <Title>Type signatures
2633 Once you have data constructors with universally-quantified fields, or
2634 constants such as <Constant>runST</Constant> that have rank-2 types, it isn't long
2635 before you discover that you need more! Consider:
2641 mkTs f x y = [T1 f x, T1 f y]
2647 <Function>mkTs</Function> is a fuction that constructs some values of type
2648 <Literal>T</Literal>, using some pieces passed to it. The trouble is that since
2649 <Literal>f</Literal> is a function argument, Haskell assumes that it is
2650 monomorphic, so we'll get a type error when applying <Function>T1</Function> to
2651 it. This is a rather silly example, but the problem really bites in
2652 practice. Lots of people trip over the fact that you can't make
2653 "wrappers functions" for <Constant>runST</Constant> for exactly the same reason.
2654 In short, it is impossible to build abstractions around functions with
2659 The solution is fairly clear. We provide the ability to give a rank-2
2660 type signature for <Emphasis>ordinary</Emphasis> functions (not only data
2661 constructors), thus:
2667 mkTs :: (forall b. b -> b -> b) -> a -> [T a]
2668 mkTs f x y = [T1 f x, T1 f y]
2674 This type signature tells the compiler to attribute <Literal>f</Literal> with
2675 the polymorphic type <Literal>(forall b. b -> b -> b)</Literal> when type
2676 checking the body of <Function>mkTs</Function>, so now the application of
2677 <Function>T1</Function> is fine.
2681 There are two restrictions:
2690 You can only define a rank 2 type, specified by the following
2695 rank2type ::= [forall tyvars .] [context =>] funty
2696 funty ::= ([forall tyvars .] [context =>] ty) -> funty
2698 ty ::= ...current Haskell monotype syntax...
2702 Informally, the universal quantification must all be right at the beginning,
2703 or at the top level of a function argument.
2710 There is a restriction on the definition of a function whose
2711 type signature is a rank-2 type: the polymorphic arguments must be
2712 matched on the left hand side of the "<Literal>=</Literal>" sign. You can't
2713 define <Function>mkTs</Function> like this:
2717 mkTs :: (forall b. b -> b -> b) -> a -> [T a]
2718 mkTs = \ f x y -> [T1 f x, T1 f y]
2723 The same partial-application rule applies to ordinary functions with
2724 rank-2 types as applied to data constructors.
2737 <Title>Type synonyms and hoisting
2741 GHC also allows you to write a <Literal>forall</Literal> in a type synonym, thus:
2743 type Discard a = forall b. a -> b -> a
2748 However, it is often convenient to use these sort of synonyms at the right hand
2749 end of an arrow, thus:
2751 type Discard a = forall b. a -> b -> a
2753 g :: Int -> Discard Int
2756 Simply expanding the type synonym would give
2758 g :: Int -> (forall b. Int -> b -> Int)
2760 but GHC "hoists" the <Literal>forall</Literal> to give the isomorphic type
2762 g :: forall b. Int -> Int -> b -> Int
2764 In general, the rule is this: <Emphasis>to determine the type specified by any explicit
2765 user-written type (e.g. in a type signature), GHC expands type synonyms and then repeatedly
2766 performs the transformation:</Emphasis>
2768 <Emphasis>type1</Emphasis> -> forall a. <Emphasis>type2</Emphasis>
2770 forall a. <Emphasis>type1</Emphasis> -> <Emphasis>type2</Emphasis>
2772 (In fact, GHC tries to retain as much synonym information as possible for use in
2773 error messages, but that is a usability issue.) This rule applies, of course, whether
2774 or not the <Literal>forall</Literal> comes from a synonym. For example, here is another
2775 valid way to write <Literal>g</Literal>'s type signature:
2777 g :: Int -> Int -> forall b. b -> Int
2784 <Sect1 id="existential-quantification">
2785 <Title>Existentially quantified data constructors
2789 The idea of using existential quantification in data type declarations
2790 was suggested by Laufer (I believe, thought doubtless someone will
2791 correct me), and implemented in Hope+. It's been in Lennart
2792 Augustsson's <Command>hbc</Command> Haskell compiler for several years, and
2793 proved very useful. Here's the idea. Consider the declaration:
2799 data Foo = forall a. MkFoo a (a -> Bool)
2806 The data type <Literal>Foo</Literal> has two constructors with types:
2812 MkFoo :: forall a. a -> (a -> Bool) -> Foo
2819 Notice that the type variable <Literal>a</Literal> in the type of <Function>MkFoo</Function>
2820 does not appear in the data type itself, which is plain <Literal>Foo</Literal>.
2821 For example, the following expression is fine:
2827 [MkFoo 3 even, MkFoo 'c' isUpper] :: [Foo]
2833 Here, <Literal>(MkFoo 3 even)</Literal> packages an integer with a function
2834 <Function>even</Function> that maps an integer to <Literal>Bool</Literal>; and <Function>MkFoo 'c'
2835 isUpper</Function> packages a character with a compatible function. These
2836 two things are each of type <Literal>Foo</Literal> and can be put in a list.
2840 What can we do with a value of type <Literal>Foo</Literal>?. In particular,
2841 what happens when we pattern-match on <Function>MkFoo</Function>?
2847 f (MkFoo val fn) = ???
2853 Since all we know about <Literal>val</Literal> and <Function>fn</Function> is that they
2854 are compatible, the only (useful) thing we can do with them is to
2855 apply <Function>fn</Function> to <Literal>val</Literal> to get a boolean. For example:
2862 f (MkFoo val fn) = fn val
2868 What this allows us to do is to package heterogenous values
2869 together with a bunch of functions that manipulate them, and then treat
2870 that collection of packages in a uniform manner. You can express
2871 quite a bit of object-oriented-like programming this way.
2874 <Sect2 id="existential">
2875 <Title>Why existential?
2879 What has this to do with <Emphasis>existential</Emphasis> quantification?
2880 Simply that <Function>MkFoo</Function> has the (nearly) isomorphic type
2886 MkFoo :: (exists a . (a, a -> Bool)) -> Foo
2892 But Haskell programmers can safely think of the ordinary
2893 <Emphasis>universally</Emphasis> quantified type given above, thereby avoiding
2894 adding a new existential quantification construct.
2900 <Title>Type classes</Title>
2903 An easy extension (implemented in <Command>hbc</Command>) is to allow
2904 arbitrary contexts before the constructor. For example:
2910 data Baz = forall a. Eq a => Baz1 a a
2911 | forall b. Show b => Baz2 b (b -> b)
2917 The two constructors have the types you'd expect:
2923 Baz1 :: forall a. Eq a => a -> a -> Baz
2924 Baz2 :: forall b. Show b => b -> (b -> b) -> Baz
2930 But when pattern matching on <Function>Baz1</Function> the matched values can be compared
2931 for equality, and when pattern matching on <Function>Baz2</Function> the first matched
2932 value can be converted to a string (as well as applying the function to it).
2933 So this program is legal:
2940 f (Baz1 p q) | p == q = "Yes"
2942 f (Baz1 v fn) = show (fn v)
2948 Operationally, in a dictionary-passing implementation, the
2949 constructors <Function>Baz1</Function> and <Function>Baz2</Function> must store the
2950 dictionaries for <Literal>Eq</Literal> and <Literal>Show</Literal> respectively, and
2951 extract it on pattern matching.
2955 Notice the way that the syntax fits smoothly with that used for
2956 universal quantification earlier.
2962 <Title>Restrictions</Title>
2965 There are several restrictions on the ways in which existentially-quantified
2966 constructors can be use.
2975 When pattern matching, each pattern match introduces a new,
2976 distinct, type for each existential type variable. These types cannot
2977 be unified with any other type, nor can they escape from the scope of
2978 the pattern match. For example, these fragments are incorrect:
2986 Here, the type bound by <Function>MkFoo</Function> "escapes", because <Literal>a</Literal>
2987 is the result of <Function>f1</Function>. One way to see why this is wrong is to
2988 ask what type <Function>f1</Function> has:
2992 f1 :: Foo -> a -- Weird!
2996 What is this "<Literal>a</Literal>" in the result type? Clearly we don't mean
3001 f1 :: forall a. Foo -> a -- Wrong!
3005 The original program is just plain wrong. Here's another sort of error
3009 f2 (Baz1 a b) (Baz1 p q) = a==q
3013 It's ok to say <Literal>a==b</Literal> or <Literal>p==q</Literal>, but
3014 <Literal>a==q</Literal> is wrong because it equates the two distinct types arising
3015 from the two <Function>Baz1</Function> constructors.
3023 You can't pattern-match on an existentially quantified
3024 constructor in a <Literal>let</Literal> or <Literal>where</Literal> group of
3025 bindings. So this is illegal:
3029 f3 x = a==b where { Baz1 a b = x }
3033 You can only pattern-match
3034 on an existentially-quantified constructor in a <Literal>case</Literal> expression or
3035 in the patterns of a function definition.
3037 The reason for this restriction is really an implementation one.
3038 Type-checking binding groups is already a nightmare without
3039 existentials complicating the picture. Also an existential pattern
3040 binding at the top level of a module doesn't make sense, because it's
3041 not clear how to prevent the existentially-quantified type "escaping".
3042 So for now, there's a simple-to-state restriction. We'll see how
3050 You can't use existential quantification for <Literal>newtype</Literal>
3051 declarations. So this is illegal:
3055 newtype T = forall a. Ord a => MkT a
3059 Reason: a value of type <Literal>T</Literal> must be represented as a pair
3060 of a dictionary for <Literal>Ord t</Literal> and a value of type <Literal>t</Literal>.
3061 That contradicts the idea that <Literal>newtype</Literal> should have no
3062 concrete representation. You can get just the same efficiency and effect
3063 by using <Literal>data</Literal> instead of <Literal>newtype</Literal>. If there is no
3064 overloading involved, then there is more of a case for allowing
3065 an existentially-quantified <Literal>newtype</Literal>, because the <Literal>data</Literal>
3066 because the <Literal>data</Literal> version does carry an implementation cost,
3067 but single-field existentially quantified constructors aren't much
3068 use. So the simple restriction (no existential stuff on <Literal>newtype</Literal>)
3069 stands, unless there are convincing reasons to change it.
3077 You can't use <Literal>deriving</Literal> to define instances of a
3078 data type with existentially quantified data constructors.
3080 Reason: in most cases it would not make sense. For example:#
3083 data T = forall a. MkT [a] deriving( Eq )
3086 To derive <Literal>Eq</Literal> in the standard way we would need to have equality
3087 between the single component of two <Function>MkT</Function> constructors:
3091 (MkT a) == (MkT b) = ???
3094 But <VarName>a</VarName> and <VarName>b</VarName> have distinct types, and so can't be compared.
3095 It's just about possible to imagine examples in which the derived instance
3096 would make sense, but it seems altogether simpler simply to prohibit such
3097 declarations. Define your own instances!
3109 <Sect1 id="sec-assertions">
3111 <IndexTerm><Primary>Assertions</Primary></IndexTerm>
3115 If you want to make use of assertions in your standard Haskell code, you
3116 could define a function like the following:
3122 assert :: Bool -> a -> a
3123 assert False x = error "assertion failed!"
3130 which works, but gives you back a less than useful error message --
3131 an assertion failed, but which and where?
3135 One way out is to define an extended <Function>assert</Function> function which also
3136 takes a descriptive string to include in the error message and
3137 perhaps combine this with the use of a pre-processor which inserts
3138 the source location where <Function>assert</Function> was used.
3142 Ghc offers a helping hand here, doing all of this for you. For every
3143 use of <Function>assert</Function> in the user's source:
3149 kelvinToC :: Double -> Double
3150 kelvinToC k = assert (k >= 0.0) (k+273.15)
3156 Ghc will rewrite this to also include the source location where the
3163 assert pred val ==> assertError "Main.hs|15" pred val
3169 The rewrite is only performed by the compiler when it spots
3170 applications of <Function>Exception.assert</Function>, so you can still define and
3171 use your own versions of <Function>assert</Function>, should you so wish. If not,
3172 import <Literal>Exception</Literal> to make use <Function>assert</Function> in your code.
3176 To have the compiler ignore uses of assert, use the compiler option
3177 <Option>-fignore-asserts</Option>. <IndexTerm><Primary>-fignore-asserts option</Primary></IndexTerm> That is,
3178 expressions of the form <Literal>assert pred e</Literal> will be rewritten to <Literal>e</Literal>.
3182 Assertion failures can be caught, see the documentation for the
3183 <literal>Exception</literal> library (<xref linkend="sec-Exception">)
3189 <Sect1 id="scoped-type-variables">
3190 <Title>Scoped Type Variables
3194 A <Emphasis>pattern type signature</Emphasis> can introduce a <Emphasis>scoped type
3195 variable</Emphasis>. For example
3201 f (xs::[a]) = ys ++ ys
3210 The pattern <Literal>(xs::[a])</Literal> includes a type signature for <VarName>xs</VarName>.
3211 This brings the type variable <Literal>a</Literal> into scope; it scopes over
3212 all the patterns and right hand sides for this equation for <Function>f</Function>.
3213 In particular, it is in scope at the type signature for <VarName>y</VarName>.
3217 At ordinary type signatures, such as that for <VarName>ys</VarName>, any type variables
3218 mentioned in the type signature <Emphasis>that are not in scope</Emphasis> are
3219 implicitly universally quantified. (If there are no type variables in
3220 scope, all type variables mentioned in the signature are universally
3221 quantified, which is just as in Haskell 98.) In this case, since <VarName>a</VarName>
3222 is in scope, it is not universally quantified, so the type of <VarName>ys</VarName> is
3223 the same as that of <VarName>xs</VarName>. In Haskell 98 it is not possible to declare
3224 a type for <VarName>ys</VarName>; a major benefit of scoped type variables is that
3225 it becomes possible to do so.
3229 Scoped type variables are implemented in both GHC and Hugs. Where the
3230 implementations differ from the specification below, those differences
3235 So much for the basic idea. Here are the details.
3239 <Title>Scope and implicit quantification</Title>
3247 All the type variables mentioned in the patterns for a single
3248 function definition equation, that are not already in scope,
3249 are brought into scope by the patterns. We describe this set as
3250 the <Emphasis>type variables bound by the equation</Emphasis>.
3257 The type variables thus brought into scope may be mentioned
3258 in ordinary type signatures or pattern type signatures anywhere within
3266 In ordinary type signatures, any type variable mentioned in the
3267 signature that is in scope is <Emphasis>not</Emphasis> universally quantified.
3274 Ordinary type signatures do not bring any new type variables
3275 into scope (except in the type signature itself!). So this is illegal:
3284 It's illegal because <VarName>a</VarName> is not in scope in the body of <Function>f</Function>,
3285 so the ordinary signature <Literal>x::a</Literal> is equivalent to <Literal>x::forall a.a</Literal>;
3286 and that is an incorrect typing.
3293 There is no implicit universal quantification on pattern type
3294 signatures, nor may one write an explicit <Literal>forall</Literal> type in a pattern
3295 type signature. The pattern type signature is a monotype.
3303 The type variables in the head of a <Literal>class</Literal> or <Literal>instance</Literal> declaration
3304 scope over the methods defined in the <Literal>where</Literal> part. For example:
3318 (Not implemented in Hugs yet, Dec 98).
3329 <Title>Polymorphism</Title>
3337 Pattern type signatures are completely orthogonal to ordinary, separate
3338 type signatures. The two can be used independently or together. There is
3339 no scoping associated with the names of the type variables in a separate type signature.
3344 f (xs::[b]) = reverse xs
3353 The function must be polymorphic in the type variables
3354 bound by all its equations. Operationally, the type variables bound
3355 by one equation must not:
3362 Be unified with a type (such as <Literal>Int</Literal>, or <Literal>[a]</Literal>).
3368 Be unified with a type variable free in the environment.
3374 Be unified with each other. (They may unify with the type variables
3375 bound by another equation for the same function, of course.)
3382 For example, the following all fail to type check:
3386 f (x::a) (y::b) = [x,y] -- a unifies with b
3388 g (x::a) = x + 1::Int -- a unifies with Int
3390 h x = let k (y::a) = [x,y] -- a is free in the
3391 in k x -- environment
3393 k (x::a) True = ... -- a unifies with Int
3394 k (x::Int) False = ...
3397 w (x::a) = x -- a unifies with [b]
3406 The pattern-bound type variable may, however, be constrained
3407 by the context of the principal type, thus:
3411 f (x::a) (y::a) = x+y*2
3415 gets the inferred type: <Literal>forall a. Num a => a -> a -> a</Literal>.
3426 <Title>Result type signatures</Title>
3434 The result type of a function can be given a signature,
3439 f (x::a) :: [a] = [x,x,x]
3443 The final <Literal>:: [a]</Literal> after all the patterns gives a signature to the
3444 result type. Sometimes this is the only way of naming the type variable
3449 f :: Int -> [a] -> [a]
3450 f n :: ([a] -> [a]) = let g (x::a, y::a) = (y,x)
3451 in \xs -> map g (reverse xs `zip` xs)
3463 Result type signatures are not yet implemented in Hugs.
3469 <Title>Pattern signatures on other constructs</Title>
3477 A pattern type signature can be on an arbitrary sub-pattern, not
3482 f ((x,y)::(a,b)) = (y,x) :: (b,a)
3491 Pattern type signatures, including the result part, can be used
3492 in lambda abstractions:
3496 (\ (x::a, y) :: a -> x)
3500 Type variables bound by these patterns must be polymorphic in
3501 the sense defined above.
3506 f1 (x::c) = f1 x -- ok
3507 f2 = \(x::c) -> f2 x -- not ok
3511 Here, <Function>f1</Function> is OK, but <Function>f2</Function> is not, because <VarName>c</VarName> gets unified
3512 with a type variable free in the environment, in this
3513 case, the type of <Function>f2</Function>, which is in the environment when
3514 the lambda abstraction is checked.
3521 Pattern type signatures, including the result part, can be used
3522 in <Literal>case</Literal> expressions:
3526 case e of { (x::a, y) :: a -> x }
3530 The pattern-bound type variables must, as usual,
3531 be polymorphic in the following sense: each case alternative,
3532 considered as a lambda abstraction, must be polymorphic.
3537 case (True,False) of { (x::a, y) -> x }
3541 Even though the context is that of a pair of booleans,
3542 the alternative itself is polymorphic. Of course, it is
3547 case (True,False) of { (x::Bool, y) -> x }
3556 To avoid ambiguity, the type after the “<Literal>::</Literal>” in a result
3557 pattern signature on a lambda or <Literal>case</Literal> must be atomic (i.e. a single
3558 token or a parenthesised type of some sort). To see why,
3559 consider how one would parse this:
3572 Pattern type signatures that bind new type variables
3573 may not be used in pattern bindings at all.
3578 f x = let (y, z::a) = x in ...
3582 But these are OK, because they do not bind fresh type variables:
3586 f1 x = let (y, z::Int) = x in ...
3587 f2 (x::(Int,a)) = let (y, z::a) = x in ...
3591 However a single variable is considered a degenerate function binding,
3592 rather than a degerate pattern binding, so this is permitted, even
3593 though it binds a type variable:
3597 f :: (b->b) = \(x::b) -> x
3606 Such degnerate function bindings do not fall under the monomorphism
3613 g :: a -> a -> Bool = \x y. x==y
3619 Here <Function>g</Function> has type <Literal>forall a. Eq a => a -> a -> Bool</Literal>, just as if
3620 <Function>g</Function> had a separate type signature. Lacking a type signature, <Function>g</Function>
3621 would get a monomorphic type.
3627 <Title>Existentials</Title>
3635 Pattern type signatures can bind existential type variables.
3640 data T = forall a. MkT [a]
3643 f (MkT [t::a]) = MkT t3
3660 <Sect1 id="pragmas">
3665 GHC supports several pragmas, or instructions to the compiler placed
3666 in the source code. Pragmas don't affect the meaning of the program,
3667 but they might affect the efficiency of the generated code.
3670 <Sect2 id="inline-pragma">
3671 <Title>INLINE pragma
3673 <IndexTerm><Primary>INLINE pragma</Primary></IndexTerm>
3674 <IndexTerm><Primary>pragma, INLINE</Primary></IndexTerm></Title>
3677 GHC (with <Option>-O</Option>, as always) tries to inline (or “unfold”)
3678 functions/values that are “small enough,” thus avoiding the call
3679 overhead and possibly exposing other more-wonderful optimisations.
3683 You will probably see these unfoldings (in Core syntax) in your
3688 Normally, if GHC decides a function is “too expensive” to inline, it
3689 will not do so, nor will it export that unfolding for other modules to
3694 The sledgehammer you can bring to bear is the
3695 <Literal>INLINE</Literal><IndexTerm><Primary>INLINE pragma</Primary></IndexTerm> pragma, used thusly:
3698 key_function :: Int -> String -> (Bool, Double)
3700 #ifdef __GLASGOW_HASKELL__
3701 {-# INLINE key_function #-}
3705 (You don't need to do the C pre-processor carry-on unless you're going
3706 to stick the code through HBC—it doesn't like <Literal>INLINE</Literal> pragmas.)
3710 The major effect of an <Literal>INLINE</Literal> pragma is to declare a function's
3711 “cost” to be very low. The normal unfolding machinery will then be
3712 very keen to inline it.
3716 An <Literal>INLINE</Literal> pragma for a function can be put anywhere its type
3717 signature could be put.
3721 <Literal>INLINE</Literal> pragmas are a particularly good idea for the
3722 <Literal>then</Literal>/<Literal>return</Literal> (or <Literal>bind</Literal>/<Literal>unit</Literal>) functions in a monad.
3723 For example, in GHC's own <Literal>UniqueSupply</Literal> monad code, we have:
3726 #ifdef __GLASGOW_HASKELL__
3727 {-# INLINE thenUs #-}
3728 {-# INLINE returnUs #-}
3736 <Sect2 id="noinline-pragma">
3737 <Title>NOINLINE pragma
3741 <IndexTerm><Primary>NOINLINE pragma</Primary></IndexTerm>
3742 <IndexTerm><Primary>pragma, NOINLINE</Primary></IndexTerm>
3746 The <Literal>NOINLINE</Literal> pragma does exactly what you'd expect: it stops the
3747 named function from being inlined by the compiler. You shouldn't ever
3748 need to do this, unless you're very cautious about code size.
3753 <Sect2 id="specialize-pragma">
3754 <Title>SPECIALIZE pragma
3758 <IndexTerm><Primary>SPECIALIZE pragma</Primary></IndexTerm>
3759 <IndexTerm><Primary>pragma, SPECIALIZE</Primary></IndexTerm>
3760 <IndexTerm><Primary>overloading, death to</Primary></IndexTerm>
3764 (UK spelling also accepted.) For key overloaded functions, you can
3765 create extra versions (NB: more code space) specialised to particular
3766 types. Thus, if you have an overloaded function:
3772 hammeredLookup :: Ord key => [(key, value)] -> key -> value
3778 If it is heavily used on lists with <Literal>Widget</Literal> keys, you could
3779 specialise it as follows:
3782 {-# SPECIALIZE hammeredLookup :: [(Widget, value)] -> Widget -> value #-}
3788 To get very fancy, you can also specify a named function to use for
3789 the specialised value, by adding <Literal>= blah</Literal>, as in:
3792 {-# SPECIALIZE hammeredLookup :: ...as before... = blah #-}
3795 It's <Emphasis>Your Responsibility</Emphasis> to make sure that <Function>blah</Function> really
3796 behaves as a specialised version of <Function>hammeredLookup</Function>!!!
3800 NOTE: the <Literal>=blah</Literal> feature isn't implemented in GHC 4.xx.
3804 An example in which the <Literal>= blah</Literal> form will Win Big:
3807 toDouble :: Real a => a -> Double
3808 toDouble = fromRational . toRational
3810 {-# SPECIALIZE toDouble :: Int -> Double = i2d #-}
3811 i2d (I# i) = D# (int2Double# i) -- uses Glasgow prim-op directly
3814 The <Function>i2d</Function> function is virtually one machine instruction; the
3815 default conversion—via an intermediate <Literal>Rational</Literal>—is obscenely
3816 expensive by comparison.
3820 By using the US spelling, your <Literal>SPECIALIZE</Literal> pragma will work with
3821 HBC, too. Note that HBC doesn't support the <Literal>= blah</Literal> form.
3825 A <Literal>SPECIALIZE</Literal> pragma for a function can be put anywhere its type
3826 signature could be put.
3831 <Sect2 id="specialize-instance-pragma">
3832 <Title>SPECIALIZE instance pragma
3836 <IndexTerm><Primary>SPECIALIZE pragma</Primary></IndexTerm>
3837 <IndexTerm><Primary>overloading, death to</Primary></IndexTerm>
3838 Same idea, except for instance declarations. For example:
3841 instance (Eq a) => Eq (Foo a) where { ... usual stuff ... }
3843 {-# SPECIALIZE instance Eq (Foo [(Int, Bar)] #-}
3846 Compatible with HBC, by the way.
3851 <Sect2 id="line-pragma">
3856 <IndexTerm><Primary>LINE pragma</Primary></IndexTerm>
3857 <IndexTerm><Primary>pragma, LINE</Primary></IndexTerm>
3861 This pragma is similar to C's <Literal>#line</Literal> pragma, and is mainly for use in
3862 automatically generated Haskell code. It lets you specify the line
3863 number and filename of the original code; for example
3869 {-# LINE 42 "Foo.vhs" #-}
3875 if you'd generated the current file from something called <Filename>Foo.vhs</Filename>
3876 and this line corresponds to line 42 in the original. GHC will adjust
3877 its error messages to refer to the line/file named in the <Literal>LINE</Literal>
3884 <Title>RULES pragma</Title>
3887 The RULES pragma lets you specify rewrite rules. It is described in
3888 <XRef LinkEnd="rewrite-rules">.
3895 <Sect1 id="rewrite-rules">
3896 <Title>Rewrite rules
3898 <IndexTerm><Primary>RULES pagma</Primary></IndexTerm>
3899 <IndexTerm><Primary>pragma, RULES</Primary></IndexTerm>
3900 <IndexTerm><Primary>rewrite rules</Primary></IndexTerm></Title>
3903 The programmer can specify rewrite rules as part of the source program
3904 (in a pragma). GHC applies these rewrite rules wherever it can.
3912 "map/map" forall f g xs. map f (map g xs) = map (f.g) xs
3919 <Title>Syntax</Title>
3922 From a syntactic point of view:
3928 Each rule has a name, enclosed in double quotes. The name itself has
3929 no significance at all. It is only used when reporting how many times the rule fired.
3935 There may be zero or more rules in a <Literal>RULES</Literal> pragma.
3941 Layout applies in a <Literal>RULES</Literal> pragma. Currently no new indentation level
3942 is set, so you must lay out your rules starting in the same column as the
3943 enclosing definitions.
3949 Each variable mentioned in a rule must either be in scope (e.g. <Function>map</Function>),
3950 or bound by the <Literal>forall</Literal> (e.g. <Function>f</Function>, <Function>g</Function>, <Function>xs</Function>). The variables bound by
3951 the <Literal>forall</Literal> are called the <Emphasis>pattern</Emphasis> variables. They are separated
3952 by spaces, just like in a type <Literal>forall</Literal>.
3958 A pattern variable may optionally have a type signature.
3959 If the type of the pattern variable is polymorphic, it <Emphasis>must</Emphasis> have a type signature.
3960 For example, here is the <Literal>foldr/build</Literal> rule:
3963 "fold/build" forall k z (g::forall b. (a->b->b) -> b -> b) .
3964 foldr k z (build g) = g k z
3967 Since <Function>g</Function> has a polymorphic type, it must have a type signature.
3974 The left hand side of a rule must consist of a top-level variable applied
3975 to arbitrary expressions. For example, this is <Emphasis>not</Emphasis> OK:
3978 "wrong1" forall e1 e2. case True of { True -> e1; False -> e2 } = e1
3979 "wrong2" forall f. f True = True
3982 In <Literal>"wrong1"</Literal>, the LHS is not an application; in <Literal>"wrong2"</Literal>, the LHS has a pattern variable
3989 A rule does not need to be in the same module as (any of) the
3990 variables it mentions, though of course they need to be in scope.
3996 Rules are automatically exported from a module, just as instance declarations are.
4007 <Title>Semantics</Title>
4010 From a semantic point of view:
4016 Rules are only applied if you use the <Option>-O</Option> flag.
4022 Rules are regarded as left-to-right rewrite rules.
4023 When GHC finds an expression that is a substitution instance of the LHS
4024 of a rule, it replaces the expression by the (appropriately-substituted) RHS.
4025 By "a substitution instance" we mean that the LHS can be made equal to the
4026 expression by substituting for the pattern variables.
4033 The LHS and RHS of a rule are typechecked, and must have the
4041 GHC makes absolutely no attempt to verify that the LHS and RHS
4042 of a rule have the same meaning. That is undecideable in general, and
4043 infeasible in most interesting cases. The responsibility is entirely the programmer's!
4050 GHC makes no attempt to make sure that the rules are confluent or
4051 terminating. For example:
4054 "loop" forall x,y. f x y = f y x
4057 This rule will cause the compiler to go into an infinite loop.
4064 If more than one rule matches a call, GHC will choose one arbitrarily to apply.
4070 GHC currently uses a very simple, syntactic, matching algorithm
4071 for matching a rule LHS with an expression. It seeks a substitution
4072 which makes the LHS and expression syntactically equal modulo alpha
4073 conversion. The pattern (rule), but not the expression, is eta-expanded if
4074 necessary. (Eta-expanding the epression can lead to laziness bugs.)
4075 But not beta conversion (that's called higher-order matching).
4079 Matching is carried out on GHC's intermediate language, which includes
4080 type abstractions and applications. So a rule only matches if the
4081 types match too. See <XRef LinkEnd="rule-spec"> below.
4087 GHC keeps trying to apply the rules as it optimises the program.
4088 For example, consider:
4097 The expression <Literal>s (t xs)</Literal> does not match the rule <Literal>"map/map"</Literal>, but GHC
4098 will substitute for <VarName>s</VarName> and <VarName>t</VarName>, giving an expression which does match.
4099 If <VarName>s</VarName> or <VarName>t</VarName> was (a) used more than once, and (b) large or a redex, then it would
4100 not be substituted, and the rule would not fire.
4107 In the earlier phases of compilation, GHC inlines <Emphasis>nothing
4108 that appears on the LHS of a rule</Emphasis>, because once you have substituted
4109 for something you can't match against it (given the simple minded
4110 matching). So if you write the rule
4113 "map/map" forall f,g. map f . map g = map (f.g)
4116 this <Emphasis>won't</Emphasis> match the expression <Literal>map f (map g xs)</Literal>.
4117 It will only match something written with explicit use of ".".
4118 Well, not quite. It <Emphasis>will</Emphasis> match the expression
4124 where <Function>wibble</Function> is defined:
4127 wibble f g = map f . map g
4130 because <Function>wibble</Function> will be inlined (it's small).
4132 Later on in compilation, GHC starts inlining even things on the
4133 LHS of rules, but still leaves the rules enabled. This inlining
4134 policy is controlled by the per-simplification-pass flag <Option>-finline-phase</Option><Emphasis>n</Emphasis>.
4141 All rules are implicitly exported from the module, and are therefore
4142 in force in any module that imports the module that defined the rule, directly
4143 or indirectly. (That is, if A imports B, which imports C, then C's rules are
4144 in force when compiling A.) The situation is very similar to that for instance
4156 <Title>List fusion</Title>
4159 The RULES mechanism is used to implement fusion (deforestation) of common list functions.
4160 If a "good consumer" consumes an intermediate list constructed by a "good producer", the
4161 intermediate list should be eliminated entirely.
4165 The following are good producers:
4177 Enumerations of <Literal>Int</Literal> and <Literal>Char</Literal> (e.g. <Literal>['a'..'z']</Literal>).
4183 Explicit lists (e.g. <Literal>[True, False]</Literal>)
4189 The cons constructor (e.g <Literal>3:4:[]</Literal>)
4195 <Function>++</Function>
4201 <Function>map</Function>
4207 <Function>filter</Function>
4213 <Function>iterate</Function>, <Function>repeat</Function>
4219 <Function>zip</Function>, <Function>zipWith</Function>
4228 The following are good consumers:
4240 <Function>array</Function> (on its second argument)
4246 <Function>length</Function>
4252 <Function>++</Function> (on its first argument)
4258 <Function>map</Function>
4264 <Function>filter</Function>
4270 <Function>concat</Function>
4276 <Function>unzip</Function>, <Function>unzip2</Function>, <Function>unzip3</Function>, <Function>unzip4</Function>
4282 <Function>zip</Function>, <Function>zipWith</Function> (but on one argument only; if both are good producers, <Function>zip</Function>
4283 will fuse with one but not the other)
4289 <Function>partition</Function>
4295 <Function>head</Function>
4301 <Function>and</Function>, <Function>or</Function>, <Function>any</Function>, <Function>all</Function>
4307 <Function>sequence_</Function>
4313 <Function>msum</Function>
4319 <Function>sortBy</Function>
4328 So, for example, the following should generate no intermediate lists:
4331 array (1,10) [(i,i*i) | i <- map (+ 1) [0..9]]
4337 This list could readily be extended; if there are Prelude functions that you use
4338 a lot which are not included, please tell us.
4342 If you want to write your own good consumers or producers, look at the
4343 Prelude definitions of the above functions to see how to do so.
4348 <Sect2 id="rule-spec">
4349 <Title>Specialisation
4353 Rewrite rules can be used to get the same effect as a feature
4354 present in earlier version of GHC:
4357 {-# SPECIALIZE fromIntegral :: Int8 -> Int16 = int8ToInt16 #-}
4360 This told GHC to use <Function>int8ToInt16</Function> instead of <Function>fromIntegral</Function> whenever
4361 the latter was called with type <Literal>Int8 -> Int16</Literal>. That is, rather than
4362 specialising the original definition of <Function>fromIntegral</Function> the programmer is
4363 promising that it is safe to use <Function>int8ToInt16</Function> instead.
4367 This feature is no longer in GHC. But rewrite rules let you do the
4372 "fromIntegral/Int8/Int16" fromIntegral = int8ToInt16
4376 This slightly odd-looking rule instructs GHC to replace <Function>fromIntegral</Function>
4377 by <Function>int8ToInt16</Function> <Emphasis>whenever the types match</Emphasis>. Speaking more operationally,
4378 GHC adds the type and dictionary applications to get the typed rule
4381 forall (d1::Integral Int8) (d2::Num Int16) .
4382 fromIntegral Int8 Int16 d1 d2 = int8ToInt16
4386 this rule does not need to be in the same file as fromIntegral,
4387 unlike the <Literal>SPECIALISE</Literal> pragmas which currently do (so that they
4388 have an original definition available to specialise).
4394 <Title>Controlling what's going on</Title>
4402 Use <Option>-ddump-rules</Option> to see what transformation rules GHC is using.
4408 Use <Option>-ddump-simpl-stats</Option> to see what rules are being fired.
4409 If you add <Option>-dppr-debug</Option> you get a more detailed listing.
4415 The defintion of (say) <Function>build</Function> in <FileName>PrelBase.lhs</FileName> looks llike this:
4418 build :: forall a. (forall b. (a -> b -> b) -> b -> b) -> [a]
4419 {-# INLINE build #-}
4423 Notice the <Literal>INLINE</Literal>! That prevents <Literal>(:)</Literal> from being inlined when compiling
4424 <Literal>PrelBase</Literal>, so that an importing module will “see” the <Literal>(:)</Literal>, and can
4425 match it on the LHS of a rule. <Literal>INLINE</Literal> prevents any inlining happening
4426 in the RHS of the <Literal>INLINE</Literal> thing. I regret the delicacy of this.
4433 In <Filename>ghc/lib/std/PrelBase.lhs</Filename> look at the rules for <Function>map</Function> to
4434 see how to write rules that will do fusion and yet give an efficient
4435 program even if fusion doesn't happen. More rules in <Filename>PrelList.lhs</Filename>.
4447 <Sect1 id="generic-classes">
4448 <Title>Generic classes</Title>
4451 The ideas behind this extension are described in detail in "Derivable type classes",
4452 Ralf Hinze and Simon Peyton Jones, Haskell Workshop, Montreal Sept 2000, pp94-105.
4453 An example will give the idea:
4461 fromBin :: [Int] -> (a, [Int])
4463 toBin {| Unit |} Unit = []
4464 toBin {| a :+: b |} (Inl x) = 0 : toBin x
4465 toBin {| a :+: b |} (Inr y) = 1 : toBin y
4466 toBin {| a :*: b |} (x :*: y) = toBin x ++ toBin y
4468 fromBin {| Unit |} bs = (Unit, bs)
4469 fromBin {| a :+: b |} (0:bs) = (Inl x, bs') where (x,bs') = fromBin bs
4470 fromBin {| a :+: b |} (1:bs) = (Inr y, bs') where (y,bs') = fromBin bs
4471 fromBin {| a :*: b |} bs = (x :*: y, bs'') where (x,bs' ) = fromBin bs
4472 (y,bs'') = fromBin bs'
4475 This class declaration explains how <Literal>toBin</Literal> and <Literal>fromBin</Literal>
4476 work for arbitrary data types. They do so by giving cases for unit, product, and sum,
4477 which are defined thus in the library module <Literal>Generics</Literal>:
4481 data a :+: b = Inl a | Inr b
4482 data a :*: b = a :*: b
4485 Now you can make a data type into an instance of Bin like this:
4487 instance (Bin a, Bin b) => Bin (a,b)
4488 instance Bin a => Bin [a]
4490 That is, just leave off the "where" clasuse. Of course, you can put in the
4491 where clause and over-ride whichever methods you please.
4495 <Title> Using generics </Title>
4496 <Para>To use generics you need to</para>
4499 <Para>Use the <Option>-fgenerics</Option> flag.</Para>
4502 <Para>Import the module <Literal>Generics</Literal> from the
4503 <Literal>lang</Literal> package. This import brings into
4504 scope the data types <Literal>Unit</Literal>,
4505 <Literal>:*:</Literal>, and <Literal>:+:</Literal>. (You
4506 don't need this import if you don't mention these types
4507 explicitly; for example, if you are simply giving instance
4508 declarations.)</Para>
4513 <Sect2> <Title> Changes wrt the paper </Title>
4515 Note that the type constructors <Literal>:+:</Literal> and <Literal>:*:</Literal>
4516 can be written infix (indeed, you can now use
4517 any operator starting in a colon as an infix type constructor). Also note that
4518 the type constructors are not exactly as in the paper (Unit instead of 1, etc).
4519 Finally, note that the syntax of the type patterns in the class declaration
4520 uses "<Literal>{|</Literal>" and "<Literal>{|</Literal>" brackets; curly braces
4521 alone would ambiguous when they appear on right hand sides (an extension we
4522 anticipate wanting).
4526 <Sect2> <Title>Terminology and restrictions</Title>
4528 Terminology. A "generic default method" in a class declaration
4529 is one that is defined using type patterns as above.
4530 A "polymorphic default method" is a default method defined as in Haskell 98.
4531 A "generic class declaration" is a class declaration with at least one
4532 generic default method.
4540 Alas, we do not yet implement the stuff about constructor names and
4547 A generic class can have only one parameter; you can't have a generic
4548 multi-parameter class.
4554 A default method must be defined entirely using type patterns, or entirely
4555 without. So this is illegal:
4558 op :: a -> (a, Bool)
4559 op {| Unit |} Unit = (Unit, True)
4562 However it is perfectly OK for some methods of a generic class to have
4563 generic default methods and others to have polymorphic default methods.
4569 The type variable(s) in the type pattern for a generic method declaration
4570 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:
4574 op {| p :*: q |} (x :*: y) = op (x :: p)
4582 The type patterns in a generic default method must take one of the forms:
4588 where "a" and "b" are type variables. Furthermore, all the type patterns for
4589 a single type constructor (<Literal>:*:</Literal>, say) must be identical; they
4590 must use the same type variables. So this is illegal:
4594 op {| a :+: b |} (Inl x) = True
4595 op {| p :+: q |} (Inr y) = False
4597 The type patterns must be identical, even in equations for different methods of the class.
4598 So this too is illegal:
4602 op {| a :*: b |} (Inl x) = True
4605 op {| p :*: q |} (Inr y) = False
4607 (The reason for this restriction is that we gather all the equations for a particular type consructor
4608 into a single generic instance declaration.)
4614 A generic method declaration must give a case for each of the three type constructors.
4620 The type for a generic method can be built only from:
4622 <ListItem> <Para> Function arrows </Para> </ListItem>
4623 <ListItem> <Para> Type variables </Para> </ListItem>
4624 <ListItem> <Para> Tuples </Para> </ListItem>
4625 <ListItem> <Para> Arbitrary types not involving type variables </Para> </ListItem>
4627 Here are some example type signatures for generic methods:
4630 op2 :: Bool -> (a,Bool)
4631 op3 :: [Int] -> a -> a
4634 Here, op1, op2, op3 are OK, but op4 is rejected, because it has a type variable
4638 This restriction is an implementation restriction: we just havn't got around to
4639 implementing the necessary bidirectional maps over arbitrary type constructors.
4640 It would be relatively easy to add specific type constructors, such as Maybe and list,
4641 to the ones that are allowed.</para>
4646 In an instance declaration for a generic class, the idea is that the compiler
4647 will fill in the methods for you, based on the generic templates. However it can only
4652 The instance type is simple (a type constructor applied to type variables, as in Haskell 98).
4657 No constructor of the instance type has unboxed fields.
4661 (Of course, these things can only arise if you are already using GHC extensions.)
4662 However, you can still give an instance declarations for types which break these rules,
4663 provided you give explicit code to override any generic default methods.
4671 The option <Option>-ddump-deriv</Option> dumps incomprehensible stuff giving details of
4672 what the compiler does with generic declarations.
4677 <Sect2> <Title> Another example </Title>
4679 Just to finish with, here's another example I rather like:
4683 nCons {| Unit |} _ = 1
4684 nCons {| a :*: b |} _ = 1
4685 nCons {| a :+: b |} _ = nCons (bot::a) + nCons (bot::b)
4688 tag {| Unit |} _ = 1
4689 tag {| a :*: b |} _ = 1
4690 tag {| a :+: b |} (Inl x) = tag x
4691 tag {| a :+: b |} (Inr y) = nCons (bot::a) + tag y
4698 ;;; Local Variables: ***
4700 ;;; sgml-parent-document: ("users_guide.sgml" "book" "chapter" "sect1") ***