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>Multi-parameter type classes:</Term>
40 GHC's type system supports extended type classes with multiple
41 parameters. Please see <XRef LinkEnd="multi-param-type-classes">.
47 <Term>Local universal quantification:</Term>
50 GHC's type system supports explicit universal quantification in
51 constructor fields and function arguments. This is useful for things
52 like defining <Literal>runST</Literal> from the state-thread world. See <XRef LinkEnd="universal-quantification">.
58 <Term>Extistentially quantification in data types:</Term>
61 Some or all of the type variables in a datatype declaration may be
62 <Emphasis>existentially quantified</Emphasis>. More details in <XRef LinkEnd="existential-quantification">.
68 <Term>Scoped type variables:</Term>
71 Scoped type variables enable the programmer to supply type signatures
72 for some nested declarations, where this would not be legal in Haskell
73 98. Details in <XRef LinkEnd="scoped-type-variables">.
79 <Term>Pattern guards</Term>
82 Instead of being a boolean expression, a guard is a list of qualifiers, exactly as in a list comprehension. See <XRef LinkEnd="pattern-guards">.
88 <Term>Foreign calling:</Term>
91 Just what it sounds like. We provide <Emphasis>lots</Emphasis> of rope that you
92 can dangle around your neck. Please see <XRef LinkEnd="ffi">.
101 Pragmas are special instructions to the compiler placed in the source
102 file. The pragmas GHC supports are described in <XRef LinkEnd="pragmas">.
108 <Term>Rewrite rules:</Term>
111 The programmer can specify rewrite rules as part of the source program
112 (in a pragma). GHC applies these rewrite rules wherever it can.
113 Details in <XRef LinkEnd="rewrite-rules">.
119 <Term>Generic classes:</Term>
122 Generic class declarations allow you to define a class
123 whose methods say how to work over an arbitrary data type.
124 Then it's really easy to make any new type into an instance of
125 the class. This generalises the rather ad-hoc "deriving" feature
127 Details in <XRef LinkEnd="generic-classes">.
135 Before you get too carried away working at the lowest level (e.g.,
136 sloshing <Literal>MutableByteArray#</Literal>s around your
137 program), you may wish to check if there are libraries that provide a
138 “Haskellised veneer” over the features you want. See
139 <xref linkend="book-hslibs">.
142 <Sect1 id="primitives">
143 <Title>Unboxed types and primitive operations
145 <IndexTerm><Primary>PrelGHC module</Primary></IndexTerm>
148 This module defines all the types which are primitive in Glasgow
149 Haskell, and the operations provided for them.
152 <Sect2 id="glasgow-unboxed">
157 <IndexTerm><Primary>Unboxed types (Glasgow extension)</Primary></IndexTerm>
160 <para>Most types in GHC are <firstterm>boxed</firstterm>, which means
161 that values of that type are represented by a pointer to a heap
162 object. The representation of a Haskell <literal>Int</literal>, for
163 example, is a two-word heap object. An <firstterm>unboxed</firstterm>
164 type, however, is represented by the value itself, no pointers or heap
165 allocation are involved.
169 Unboxed types correspond to the “raw machine” types you
170 would use in C: <Literal>Int#</Literal> (long int),
171 <Literal>Double#</Literal> (double), <Literal>Addr#</Literal>
172 (void *), etc. The <Emphasis>primitive operations</Emphasis>
173 (PrimOps) on these types are what you might expect; e.g.,
174 <Literal>(+#)</Literal> is addition on
175 <Literal>Int#</Literal>s, and is the machine-addition that we all
176 know and love—usually one instruction.
180 Primitive (unboxed) types cannot be defined in Haskell, and are
181 therefore built into the language and compiler. Primitive types are
182 always unlifted; that is, a value of a primitive type cannot be
183 bottom. We use the convention that primitive types, values, and
184 operations have a <Literal>#</Literal> suffix.
188 Primitive values are often represented by a simple bit-pattern, such
189 as <Literal>Int#</Literal>, <Literal>Float#</Literal>,
190 <Literal>Double#</Literal>. But this is not necessarily the case:
191 a primitive value might be represented by a pointer to a
192 heap-allocated object. Examples include
193 <Literal>Array#</Literal>, the type of primitive arrays. A
194 primitive array is heap-allocated because it is too big a value to fit
195 in a register, and would be too expensive to copy around; in a sense,
196 it is accidental that it is represented by a pointer. If a pointer
197 represents a primitive value, then it really does point to that value:
198 no unevaluated thunks, no indirections…nothing can be at the
199 other end of the pointer than the primitive value.
203 There are some restrictions on the use of primitive types, the main
204 one being that you can't pass a primitive value to a polymorphic
205 function or store one in a polymorphic data type. This rules out
206 things like <Literal>[Int#]</Literal> (i.e. lists of primitive
207 integers). The reason for this restriction is that polymorphic
208 arguments and constructor fields are assumed to be pointers: if an
209 unboxed integer is stored in one of these, the garbage collector would
210 attempt to follow it, leading to unpredictable space leaks. Or a
211 <Function>seq</Function> operation on the polymorphic component may
212 attempt to dereference the pointer, with disastrous results. Even
213 worse, the unboxed value might be larger than a pointer
214 (<Literal>Double#</Literal> for instance).
218 Nevertheless, A numerically-intensive program using unboxed types can
219 go a <Emphasis>lot</Emphasis> faster than its “standard”
220 counterpart—we saw a threefold speedup on one example.
225 <Sect2 id="unboxed-tuples">
226 <Title>Unboxed Tuples
230 Unboxed tuples aren't really exported by <Literal>PrelGHC</Literal>,
231 they're available by default with <Option>-fglasgow-exts</Option>. An
232 unboxed tuple looks like this:
244 where <Literal>e_1..e_n</Literal> are expressions of any
245 type (primitive or non-primitive). The type of an unboxed tuple looks
250 Unboxed tuples are used for functions that need to return multiple
251 values, but they avoid the heap allocation normally associated with
252 using fully-fledged tuples. When an unboxed tuple is returned, the
253 components are put directly into registers or on the stack; the
254 unboxed tuple itself does not have a composite representation. Many
255 of the primitive operations listed in this section return unboxed
260 There are some pretty stringent restrictions on the use of unboxed tuples:
269 Unboxed tuple types are subject to the same restrictions as
270 other unboxed types; i.e. they may not be stored in polymorphic data
271 structures or passed to polymorphic functions.
278 Unboxed tuples may only be constructed as the direct result of
279 a function, and may only be deconstructed with a <Literal>case</Literal> expression.
280 eg. the following are valid:
284 f x y = (# x+1, y-1 #)
285 g x = case f x x of { (# a, b #) -> a + b }
289 but the following are invalid:
303 No variable can have an unboxed tuple type. This is illegal:
307 f :: (# Int, Int #) -> (# Int, Int #)
312 because <VarName>x</VarName> has an unboxed tuple type.
322 Note: we may relax some of these restrictions in the future.
326 The <Literal>IO</Literal> and <Literal>ST</Literal> monads use unboxed tuples to avoid unnecessary
327 allocation during sequences of operations.
333 <Title>Character and numeric types</Title>
336 <IndexTerm><Primary>character types, primitive</Primary></IndexTerm>
337 <IndexTerm><Primary>numeric types, primitive</Primary></IndexTerm>
338 <IndexTerm><Primary>integer types, primitive</Primary></IndexTerm>
339 <IndexTerm><Primary>floating point types, primitive</Primary></IndexTerm>
340 There are the following obvious primitive types:
356 <IndexTerm><Primary><literal>Char#</literal></Primary></IndexTerm>
357 <IndexTerm><Primary><literal>Int#</literal></Primary></IndexTerm>
358 <IndexTerm><Primary><literal>Word#</literal></Primary></IndexTerm>
359 <IndexTerm><Primary><literal>Addr#</literal></Primary></IndexTerm>
360 <IndexTerm><Primary><literal>Float#</literal></Primary></IndexTerm>
361 <IndexTerm><Primary><literal>Double#</literal></Primary></IndexTerm>
362 <IndexTerm><Primary><literal>Int64#</literal></Primary></IndexTerm>
363 <IndexTerm><Primary><literal>Word64#</literal></Primary></IndexTerm>
367 If you really want to know their exact equivalents in C, see
368 <Filename>ghc/includes/StgTypes.h</Filename> in the GHC source tree.
372 Literals for these types may be written as follows:
381 'a'# a Char#; for weird characters, use e.g. '\o<octal>'#
382 "a"# an Addr# (a `char *'); only characters '\0'..'\255' allowed
385 <IndexTerm><Primary>literals, primitive</Primary></IndexTerm>
386 <IndexTerm><Primary>constants, primitive</Primary></IndexTerm>
387 <IndexTerm><Primary>numbers, primitive</Primary></IndexTerm>
393 <Title>Comparison operations</Title>
396 <IndexTerm><Primary>comparisons, primitive</Primary></IndexTerm>
397 <IndexTerm><Primary>operators, comparison</Primary></IndexTerm>
403 {>,>=,==,/=,<,<=}# :: Int# -> Int# -> Bool
405 {gt,ge,eq,ne,lt,le}Char# :: Char# -> Char# -> Bool
406 -- ditto for Word# and Addr#
409 <IndexTerm><Primary><literal>>#</literal></Primary></IndexTerm>
410 <IndexTerm><Primary><literal>>=#</literal></Primary></IndexTerm>
411 <IndexTerm><Primary><literal>==#</literal></Primary></IndexTerm>
412 <IndexTerm><Primary><literal>/=#</literal></Primary></IndexTerm>
413 <IndexTerm><Primary><literal><#</literal></Primary></IndexTerm>
414 <IndexTerm><Primary><literal><=#</literal></Primary></IndexTerm>
415 <IndexTerm><Primary><literal>gt{Char,Word,Addr}#</literal></Primary></IndexTerm>
416 <IndexTerm><Primary><literal>ge{Char,Word,Addr}#</literal></Primary></IndexTerm>
417 <IndexTerm><Primary><literal>eq{Char,Word,Addr}#</literal></Primary></IndexTerm>
418 <IndexTerm><Primary><literal>ne{Char,Word,Addr}#</literal></Primary></IndexTerm>
419 <IndexTerm><Primary><literal>lt{Char,Word,Addr}#</literal></Primary></IndexTerm>
420 <IndexTerm><Primary><literal>le{Char,Word,Addr}#</literal></Primary></IndexTerm>
426 <Title>Primitive-character operations</Title>
429 <IndexTerm><Primary>characters, primitive operations</Primary></IndexTerm>
430 <IndexTerm><Primary>operators, primitive character</Primary></IndexTerm>
436 ord# :: Char# -> Int#
437 chr# :: Int# -> Char#
440 <IndexTerm><Primary><literal>ord#</literal></Primary></IndexTerm>
441 <IndexTerm><Primary><literal>chr#</literal></Primary></IndexTerm>
447 <Title>Primitive-<Literal>Int</Literal> operations</Title>
450 <IndexTerm><Primary>integers, primitive operations</Primary></IndexTerm>
451 <IndexTerm><Primary>operators, primitive integer</Primary></IndexTerm>
457 {+,-,*,quotInt,remInt,gcdInt}# :: Int# -> Int# -> Int#
458 negateInt# :: Int# -> Int#
460 iShiftL#, iShiftRA#, iShiftRL# :: Int# -> Int# -> Int#
461 -- shift left, right arithmetic, right logical
463 addIntC#, subIntC#, mulIntC# :: Int# -> Int# -> (# Int#, Int# #)
464 -- add, subtract, multiply with carry
467 <IndexTerm><Primary><literal>+#</literal></Primary></IndexTerm>
468 <IndexTerm><Primary><literal>-#</literal></Primary></IndexTerm>
469 <IndexTerm><Primary><literal>*#</literal></Primary></IndexTerm>
470 <IndexTerm><Primary><literal>quotInt#</literal></Primary></IndexTerm>
471 <IndexTerm><Primary><literal>remInt#</literal></Primary></IndexTerm>
472 <IndexTerm><Primary><literal>gcdInt#</literal></Primary></IndexTerm>
473 <IndexTerm><Primary><literal>iShiftL#</literal></Primary></IndexTerm>
474 <IndexTerm><Primary><literal>iShiftRA#</literal></Primary></IndexTerm>
475 <IndexTerm><Primary><literal>iShiftRL#</literal></Primary></IndexTerm>
476 <IndexTerm><Primary><literal>addIntC#</literal></Primary></IndexTerm>
477 <IndexTerm><Primary><literal>subIntC#</literal></Primary></IndexTerm>
478 <IndexTerm><Primary><literal>mulIntC#</literal></Primary></IndexTerm>
479 <IndexTerm><Primary>shift operations, integer</Primary></IndexTerm>
483 <Emphasis>Note:</Emphasis> No error/overflow checking!
489 <Title>Primitive-<Literal>Double</Literal> and <Literal>Float</Literal> operations</Title>
492 <IndexTerm><Primary>floating point numbers, primitive</Primary></IndexTerm>
493 <IndexTerm><Primary>operators, primitive floating point</Primary></IndexTerm>
499 {+,-,*,/}## :: Double# -> Double# -> Double#
500 {<,<=,==,/=,>=,>}## :: Double# -> Double# -> Bool
501 negateDouble# :: Double# -> Double#
502 double2Int# :: Double# -> Int#
503 int2Double# :: Int# -> Double#
505 {plus,minux,times,divide}Float# :: Float# -> Float# -> Float#
506 {gt,ge,eq,ne,lt,le}Float# :: Float# -> Float# -> Bool
507 negateFloat# :: Float# -> Float#
508 float2Int# :: Float# -> Int#
509 int2Float# :: Int# -> Float#
515 <IndexTerm><Primary><literal>+##</literal></Primary></IndexTerm>
516 <IndexTerm><Primary><literal>-##</literal></Primary></IndexTerm>
517 <IndexTerm><Primary><literal>*##</literal></Primary></IndexTerm>
518 <IndexTerm><Primary><literal>/##</literal></Primary></IndexTerm>
519 <IndexTerm><Primary><literal><##</literal></Primary></IndexTerm>
520 <IndexTerm><Primary><literal><=##</literal></Primary></IndexTerm>
521 <IndexTerm><Primary><literal>==##</literal></Primary></IndexTerm>
522 <IndexTerm><Primary><literal>=/##</literal></Primary></IndexTerm>
523 <IndexTerm><Primary><literal>>=##</literal></Primary></IndexTerm>
524 <IndexTerm><Primary><literal>>##</literal></Primary></IndexTerm>
525 <IndexTerm><Primary><literal>negateDouble#</literal></Primary></IndexTerm>
526 <IndexTerm><Primary><literal>double2Int#</literal></Primary></IndexTerm>
527 <IndexTerm><Primary><literal>int2Double#</literal></Primary></IndexTerm>
531 <IndexTerm><Primary><literal>plusFloat#</literal></Primary></IndexTerm>
532 <IndexTerm><Primary><literal>minusFloat#</literal></Primary></IndexTerm>
533 <IndexTerm><Primary><literal>timesFloat#</literal></Primary></IndexTerm>
534 <IndexTerm><Primary><literal>divideFloat#</literal></Primary></IndexTerm>
535 <IndexTerm><Primary><literal>gtFloat#</literal></Primary></IndexTerm>
536 <IndexTerm><Primary><literal>geFloat#</literal></Primary></IndexTerm>
537 <IndexTerm><Primary><literal>eqFloat#</literal></Primary></IndexTerm>
538 <IndexTerm><Primary><literal>neFloat#</literal></Primary></IndexTerm>
539 <IndexTerm><Primary><literal>ltFloat#</literal></Primary></IndexTerm>
540 <IndexTerm><Primary><literal>leFloat#</literal></Primary></IndexTerm>
541 <IndexTerm><Primary><literal>negateFloat#</literal></Primary></IndexTerm>
542 <IndexTerm><Primary><literal>float2Int#</literal></Primary></IndexTerm>
543 <IndexTerm><Primary><literal>int2Float#</literal></Primary></IndexTerm>
547 And a full complement of trigonometric functions:
553 expDouble# :: Double# -> Double#
554 logDouble# :: Double# -> Double#
555 sqrtDouble# :: Double# -> Double#
556 sinDouble# :: Double# -> Double#
557 cosDouble# :: Double# -> Double#
558 tanDouble# :: Double# -> Double#
559 asinDouble# :: Double# -> Double#
560 acosDouble# :: Double# -> Double#
561 atanDouble# :: Double# -> Double#
562 sinhDouble# :: Double# -> Double#
563 coshDouble# :: Double# -> Double#
564 tanhDouble# :: Double# -> Double#
565 powerDouble# :: Double# -> Double# -> Double#
568 <IndexTerm><Primary>trigonometric functions, primitive</Primary></IndexTerm>
572 similarly for <Literal>Float#</Literal>.
576 There are two coercion functions for <Literal>Float#</Literal>/<Literal>Double#</Literal>:
582 float2Double# :: Float# -> Double#
583 double2Float# :: Double# -> Float#
586 <IndexTerm><Primary><literal>float2Double#</literal></Primary></IndexTerm>
587 <IndexTerm><Primary><literal>double2Float#</literal></Primary></IndexTerm>
591 The primitive version of <Function>decodeDouble</Function>
592 (<Function>encodeDouble</Function> is implemented as an external C
599 decodeDouble# :: Double# -> PrelNum.ReturnIntAndGMP
602 <IndexTerm><Primary><literal>encodeDouble#</literal></Primary></IndexTerm>
603 <IndexTerm><Primary><literal>decodeDouble#</literal></Primary></IndexTerm>
607 (And the same for <Literal>Float#</Literal>s.)
612 <Sect2 id="integer-operations">
613 <Title>Operations on/for <Literal>Integers</Literal> (interface to GMP)
617 <IndexTerm><Primary>arbitrary precision integers</Primary></IndexTerm>
618 <IndexTerm><Primary>Integer, operations on</Primary></IndexTerm>
622 We implement <Literal>Integers</Literal> (arbitrary-precision
623 integers) using the GNU multiple-precision (GMP) package (version
628 The data type for <Literal>Integer</Literal> is either a small
629 integer, represented by an <Literal>Int</Literal>, or a large integer
630 represented using the pieces required by GMP's
631 <Literal>MP_INT</Literal> in <Filename>gmp.h</Filename> (see
632 <Filename>gmp.info</Filename> in
633 <Filename>ghc/includes/runtime/gmp</Filename>). It comes out as:
639 data Integer = S# Int# -- small integers
640 | J# Int# ByteArray# -- large integers
643 <IndexTerm><Primary>Integer type</Primary></IndexTerm> The primitive
644 ops to support large <Literal>Integers</Literal> use the
645 “pieces” of the representation, and are as follows:
651 negateInteger# :: Int# -> ByteArray# -> Integer
653 {plus,minus,times}Integer#, gcdInteger#,
654 quotInteger#, remInteger#, divExactInteger#
655 :: Int# -> ByteArray#
656 -> Int# -> ByteArray#
657 -> (# Int#, ByteArray# #)
660 :: Int# -> ByteArray#
661 -> Int# -> ByteArray#
662 -> Int# -- -1 for <; 0 for ==; +1 for >
665 :: Int# -> ByteArray#
667 -> Int# -- -1 for <; 0 for ==; +1 for >
670 :: Int# -> ByteArray#
674 divModInteger#, quotRemInteger#
675 :: Int# -> ByteArray#
676 -> Int# -> ByteArray#
677 -> (# Int#, ByteArray#,
680 integer2Int# :: Int# -> ByteArray# -> Int#
682 int2Integer# :: Int# -> Integer -- NB: no error-checking on these two!
683 word2Integer# :: Word# -> Integer
685 addr2Integer# :: Addr# -> Integer
686 -- the Addr# is taken to be a `char *' string
687 -- to be converted into an Integer.
690 <IndexTerm><Primary><literal>negateInteger#</literal></Primary></IndexTerm>
691 <IndexTerm><Primary><literal>plusInteger#</literal></Primary></IndexTerm>
692 <IndexTerm><Primary><literal>minusInteger#</literal></Primary></IndexTerm>
693 <IndexTerm><Primary><literal>timesInteger#</literal></Primary></IndexTerm>
694 <IndexTerm><Primary><literal>quotInteger#</literal></Primary></IndexTerm>
695 <IndexTerm><Primary><literal>remInteger#</literal></Primary></IndexTerm>
696 <IndexTerm><Primary><literal>gcdInteger#</literal></Primary></IndexTerm>
697 <IndexTerm><Primary><literal>gcdIntegerInt#</literal></Primary></IndexTerm>
698 <IndexTerm><Primary><literal>divExactInteger#</literal></Primary></IndexTerm>
699 <IndexTerm><Primary><literal>cmpInteger#</literal></Primary></IndexTerm>
700 <IndexTerm><Primary><literal>divModInteger#</literal></Primary></IndexTerm>
701 <IndexTerm><Primary><literal>quotRemInteger#</literal></Primary></IndexTerm>
702 <IndexTerm><Primary><literal>integer2Int#</literal></Primary></IndexTerm>
703 <IndexTerm><Primary><literal>int2Integer#</literal></Primary></IndexTerm>
704 <IndexTerm><Primary><literal>word2Integer#</literal></Primary></IndexTerm>
705 <IndexTerm><Primary><literal>addr2Integer#</literal></Primary></IndexTerm>
711 <Title>Words and addresses</Title>
714 <IndexTerm><Primary>word, primitive type</Primary></IndexTerm>
715 <IndexTerm><Primary>address, primitive type</Primary></IndexTerm>
716 <IndexTerm><Primary>unsigned integer, primitive type</Primary></IndexTerm>
717 <IndexTerm><Primary>pointer, primitive type</Primary></IndexTerm>
721 A <Literal>Word#</Literal> is used for bit-twiddling operations.
722 It is the same size as an <Literal>Int#</Literal>, but has no sign
723 nor any arithmetic operations.
726 type Word# -- Same size/etc as Int# but *unsigned*
727 type Addr# -- A pointer from outside the "Haskell world" (from C, probably);
728 -- described under "arrays"
731 <IndexTerm><Primary><literal>Word#</literal></Primary></IndexTerm>
732 <IndexTerm><Primary><literal>Addr#</literal></Primary></IndexTerm>
736 <Literal>Word#</Literal>s and <Literal>Addr#</Literal>s have
737 the usual comparison operations. Other
738 unboxed-<Literal>Word</Literal> ops (bit-twiddling and coercions):
744 {gt,ge,eq,ne,lt,le}Word# :: Word# -> Word# -> Bool
746 and#, or#, xor# :: Word# -> Word# -> Word#
749 quotWord#, remWord# :: Word# -> Word# -> Word#
750 -- word (i.e. unsigned) versions are different from int
751 -- versions, so we have to provide these explicitly.
753 not# :: Word# -> Word#
755 shiftL#, shiftRL# :: Word# -> Int# -> Word#
756 -- shift left, right logical
758 int2Word# :: Int# -> Word# -- just a cast, really
759 word2Int# :: Word# -> Int#
762 <IndexTerm><Primary>bit operations, Word and Addr</Primary></IndexTerm>
763 <IndexTerm><Primary><literal>gtWord#</literal></Primary></IndexTerm>
764 <IndexTerm><Primary><literal>geWord#</literal></Primary></IndexTerm>
765 <IndexTerm><Primary><literal>eqWord#</literal></Primary></IndexTerm>
766 <IndexTerm><Primary><literal>neWord#</literal></Primary></IndexTerm>
767 <IndexTerm><Primary><literal>ltWord#</literal></Primary></IndexTerm>
768 <IndexTerm><Primary><literal>leWord#</literal></Primary></IndexTerm>
769 <IndexTerm><Primary><literal>and#</literal></Primary></IndexTerm>
770 <IndexTerm><Primary><literal>or#</literal></Primary></IndexTerm>
771 <IndexTerm><Primary><literal>xor#</literal></Primary></IndexTerm>
772 <IndexTerm><Primary><literal>not#</literal></Primary></IndexTerm>
773 <IndexTerm><Primary><literal>quotWord#</literal></Primary></IndexTerm>
774 <IndexTerm><Primary><literal>remWord#</literal></Primary></IndexTerm>
775 <IndexTerm><Primary><literal>shiftL#</literal></Primary></IndexTerm>
776 <IndexTerm><Primary><literal>shiftRA#</literal></Primary></IndexTerm>
777 <IndexTerm><Primary><literal>shiftRL#</literal></Primary></IndexTerm>
778 <IndexTerm><Primary><literal>int2Word#</literal></Primary></IndexTerm>
779 <IndexTerm><Primary><literal>word2Int#</literal></Primary></IndexTerm>
783 Unboxed-<Literal>Addr</Literal> ops (C casts, really):
786 {gt,ge,eq,ne,lt,le}Addr# :: Addr# -> Addr# -> Bool
788 int2Addr# :: Int# -> Addr#
789 addr2Int# :: Addr# -> Int#
790 addr2Integer# :: Addr# -> (# Int#, ByteArray# #)
793 <IndexTerm><Primary><literal>gtAddr#</literal></Primary></IndexTerm>
794 <IndexTerm><Primary><literal>geAddr#</literal></Primary></IndexTerm>
795 <IndexTerm><Primary><literal>eqAddr#</literal></Primary></IndexTerm>
796 <IndexTerm><Primary><literal>neAddr#</literal></Primary></IndexTerm>
797 <IndexTerm><Primary><literal>ltAddr#</literal></Primary></IndexTerm>
798 <IndexTerm><Primary><literal>leAddr#</literal></Primary></IndexTerm>
799 <IndexTerm><Primary><literal>int2Addr#</literal></Primary></IndexTerm>
800 <IndexTerm><Primary><literal>addr2Int#</literal></Primary></IndexTerm>
801 <IndexTerm><Primary><literal>addr2Integer#</literal></Primary></IndexTerm>
805 The casts between <Literal>Int#</Literal>,
806 <Literal>Word#</Literal> and <Literal>Addr#</Literal>
807 correspond to null operations at the machine level, but are required
808 to keep the Haskell type checker happy.
812 Operations for indexing off of C pointers
813 (<Literal>Addr#</Literal>s) to snatch values are listed under
814 “arrays”.
820 <Title>Arrays</Title>
823 <IndexTerm><Primary>arrays, primitive</Primary></IndexTerm>
827 The type <Literal>Array# elt</Literal> is the type of primitive,
828 unpointed arrays of values of type <Literal>elt</Literal>.
837 <IndexTerm><Primary><literal>Array#</literal></Primary></IndexTerm>
841 <Literal>Array#</Literal> is more primitive than a Haskell
842 array—indeed, the Haskell <Literal>Array</Literal> interface is
843 implemented using <Literal>Array#</Literal>—in that an
844 <Literal>Array#</Literal> is indexed only by
845 <Literal>Int#</Literal>s, starting at zero. It is also more
846 primitive by virtue of being unboxed. That doesn't mean that it isn't
847 a heap-allocated object—of course, it is. Rather, being unboxed
848 means that it is represented by a pointer to the array itself, and not
849 to a thunk which will evaluate to the array (or to bottom). The
850 components of an <Literal>Array#</Literal> are themselves boxed.
854 The type <Literal>ByteArray#</Literal> is similar to
855 <Literal>Array#</Literal>, except that it contains just a string
856 of (non-pointer) bytes.
865 <IndexTerm><Primary><literal>ByteArray#</literal></Primary></IndexTerm>
869 Arrays of these types are useful when a Haskell program wishes to
870 construct a value to pass to a C procedure. It is also possible to use
871 them to build (say) arrays of unboxed characters for internal use in a
872 Haskell program. Given these uses, <Literal>ByteArray#</Literal>
873 is deliberately a bit vague about the type of its components.
874 Operations are provided to extract values of type
875 <Literal>Char#</Literal>, <Literal>Int#</Literal>,
876 <Literal>Float#</Literal>, <Literal>Double#</Literal>, and
877 <Literal>Addr#</Literal> from arbitrary offsets within a
878 <Literal>ByteArray#</Literal>. (For type
879 <Literal>Foo#</Literal>, the $i$th offset gets you the $i$th
880 <Literal>Foo#</Literal>, not the <Literal>Foo#</Literal> at
881 byte-position $i$. Mumble.) (If you want a
882 <Literal>Word#</Literal>, grab an <Literal>Int#</Literal>,
887 Lastly, we have static byte-arrays, of type
888 <Literal>Addr#</Literal> [mentioned previously]. (Remember
889 the duality between arrays and pointers in C.) Arrays of this types
890 are represented by a pointer to an array in the world outside Haskell,
891 so this pointer is not followed by the garbage collector. In other
892 respects they are just like <Literal>ByteArray#</Literal>. They
893 are only needed in order to pass values from C to Haskell.
899 <Title>Reading and writing</Title>
902 Primitive arrays are linear, and indexed starting at zero.
906 The size and indices of a <Literal>ByteArray#</Literal>, <Literal>Addr#</Literal>, and
907 <Literal>MutableByteArray#</Literal> are all in bytes. It's up to the program to
908 calculate the correct byte offset from the start of the array. This
909 allows a <Literal>ByteArray#</Literal> to contain a mixture of values of different
910 type, which is often needed when preparing data for and unpicking
911 results from C. (Umm…not true of indices…WDP 95/09)
915 <Emphasis>Should we provide some <Literal>sizeOfDouble#</Literal> constants?</Emphasis>
919 Out-of-range errors on indexing should be caught by the code which
920 uses the primitive operation; the primitive operations themselves do
921 <Emphasis>not</Emphasis> check for out-of-range indexes. The intention is that the
922 primitive ops compile to one machine instruction or thereabouts.
926 We use the terms “reading” and “writing” to refer to accessing
927 <Emphasis>mutable</Emphasis> arrays (see <XRef LinkEnd="sect-mutable">), and
928 “indexing” to refer to reading a value from an <Emphasis>immutable</Emphasis>
933 Immutable byte arrays are straightforward to index (all indices in bytes):
936 indexCharArray# :: ByteArray# -> Int# -> Char#
937 indexIntArray# :: ByteArray# -> Int# -> Int#
938 indexAddrArray# :: ByteArray# -> Int# -> Addr#
939 indexFloatArray# :: ByteArray# -> Int# -> Float#
940 indexDoubleArray# :: ByteArray# -> Int# -> Double#
942 indexCharOffAddr# :: Addr# -> Int# -> Char#
943 indexIntOffAddr# :: Addr# -> Int# -> Int#
944 indexFloatOffAddr# :: Addr# -> Int# -> Float#
945 indexDoubleOffAddr# :: Addr# -> Int# -> Double#
946 indexAddrOffAddr# :: Addr# -> Int# -> Addr#
947 -- Get an Addr# from an Addr# offset
950 <IndexTerm><Primary><literal>indexCharArray#</literal></Primary></IndexTerm>
951 <IndexTerm><Primary><literal>indexIntArray#</literal></Primary></IndexTerm>
952 <IndexTerm><Primary><literal>indexAddrArray#</literal></Primary></IndexTerm>
953 <IndexTerm><Primary><literal>indexFloatArray#</literal></Primary></IndexTerm>
954 <IndexTerm><Primary><literal>indexDoubleArray#</literal></Primary></IndexTerm>
955 <IndexTerm><Primary><literal>indexCharOffAddr#</literal></Primary></IndexTerm>
956 <IndexTerm><Primary><literal>indexIntOffAddr#</literal></Primary></IndexTerm>
957 <IndexTerm><Primary><literal>indexFloatOffAddr#</literal></Primary></IndexTerm>
958 <IndexTerm><Primary><literal>indexDoubleOffAddr#</literal></Primary></IndexTerm>
959 <IndexTerm><Primary><literal>indexAddrOffAddr#</literal></Primary></IndexTerm>
963 The last of these, <Function>indexAddrOffAddr#</Function>, extracts an <Literal>Addr#</Literal> using an offset
964 from another <Literal>Addr#</Literal>, thereby providing the ability to follow a chain of
969 Something a bit more interesting goes on when indexing arrays of boxed
970 objects, because the result is simply the boxed object. So presumably
971 it should be entered—we never usually return an unevaluated
972 object! This is a pain: primitive ops aren't supposed to do
973 complicated things like enter objects. The current solution is to
974 return a single element unboxed tuple (see <XRef LinkEnd="unboxed-tuples">).
980 indexArray# :: Array# elt -> Int# -> (# elt #)
983 <IndexTerm><Primary><literal>indexArray#</literal></Primary></IndexTerm>
989 <Title>The state type</Title>
992 <IndexTerm><Primary><literal>state, primitive type</literal></Primary></IndexTerm>
993 <IndexTerm><Primary><literal>State#</literal></Primary></IndexTerm>
997 The primitive type <Literal>State#</Literal> represents the state of a state
998 transformer. It is parameterised on the desired type of state, which
999 serves to keep states from distinct threads distinct from one another.
1000 But the <Emphasis>only</Emphasis> effect of this parameterisation is in the type
1001 system: all values of type <Literal>State#</Literal> are represented in the same way.
1002 Indeed, they are all represented by nothing at all! The code
1003 generator “knows” to generate no code, and allocate no registers
1004 etc, for primitive states.
1016 The type <Literal>GHC.RealWorld</Literal> is truly opaque: there are no values defined
1017 of this type, and no operations over it. It is “primitive” in that
1018 sense - but it is <Emphasis>not unlifted!</Emphasis> Its only role in life is to be
1019 the type which distinguishes the <Literal>IO</Literal> state transformer.
1033 <Title>State of the world</Title>
1036 A single, primitive, value of type <Literal>State# RealWorld</Literal> is provided.
1042 realWorld# :: State# RealWorld
1045 <IndexTerm><Primary>realWorld# state object</Primary></IndexTerm>
1049 (Note: in the compiler, not a <Literal>PrimOp</Literal>; just a mucho magic
1050 <Literal>Id</Literal>. Exported from <Literal>GHC</Literal>, though).
1055 <Sect2 id="sect-mutable">
1056 <Title>Mutable arrays</Title>
1059 <IndexTerm><Primary>mutable arrays</Primary></IndexTerm>
1060 <IndexTerm><Primary>arrays, mutable</Primary></IndexTerm>
1061 Corresponding to <Literal>Array#</Literal> and <Literal>ByteArray#</Literal>, we have the types of
1062 mutable versions of each. In each case, the representation is a
1063 pointer to a suitable block of (mutable) heap-allocated storage.
1069 type MutableArray# s elt
1070 type MutableByteArray# s
1073 <IndexTerm><Primary><literal>MutableArray#</literal></Primary></IndexTerm>
1074 <IndexTerm><Primary><literal>MutableByteArray#</literal></Primary></IndexTerm>
1078 <Title>Allocation</Title>
1081 <IndexTerm><Primary>mutable arrays, allocation</Primary></IndexTerm>
1082 <IndexTerm><Primary>arrays, allocation</Primary></IndexTerm>
1083 <IndexTerm><Primary>allocation, of mutable arrays</Primary></IndexTerm>
1087 Mutable arrays can be allocated. Only pointer-arrays are initialised;
1088 arrays of non-pointers are filled in by “user code” rather than by
1089 the array-allocation primitive. Reason: only the pointer case has to
1090 worry about GC striking with a partly-initialised array.
1096 newArray# :: Int# -> elt -> State# s -> (# State# s, MutableArray# s elt #)
1098 newCharArray# :: Int# -> State# s -> (# State# s, MutableByteArray# s elt #)
1099 newIntArray# :: Int# -> State# s -> (# State# s, MutableByteArray# s elt #)
1100 newAddrArray# :: Int# -> State# s -> (# State# s, MutableByteArray# s elt #)
1101 newFloatArray# :: Int# -> State# s -> (# State# s, MutableByteArray# s elt #)
1102 newDoubleArray# :: Int# -> State# s -> (# State# s, MutableByteArray# s elt #)
1105 <IndexTerm><Primary><literal>newArray#</literal></Primary></IndexTerm>
1106 <IndexTerm><Primary><literal>newCharArray#</literal></Primary></IndexTerm>
1107 <IndexTerm><Primary><literal>newIntArray#</literal></Primary></IndexTerm>
1108 <IndexTerm><Primary><literal>newAddrArray#</literal></Primary></IndexTerm>
1109 <IndexTerm><Primary><literal>newFloatArray#</literal></Primary></IndexTerm>
1110 <IndexTerm><Primary><literal>newDoubleArray#</literal></Primary></IndexTerm>
1114 The size of a <Literal>ByteArray#</Literal> is given in bytes.
1120 <Title>Reading and writing</Title>
1123 <IndexTerm><Primary>arrays, reading and writing</Primary></IndexTerm>
1129 readArray# :: MutableArray# s elt -> Int# -> State# s -> (# State# s, elt #)
1130 readCharArray# :: MutableByteArray# s -> Int# -> State# s -> (# State# s, Char# #)
1131 readIntArray# :: MutableByteArray# s -> Int# -> State# s -> (# State# s, Int# #)
1132 readAddrArray# :: MutableByteArray# s -> Int# -> State# s -> (# State# s, Addr# #)
1133 readFloatArray# :: MutableByteArray# s -> Int# -> State# s -> (# State# s, Float# #)
1134 readDoubleArray# :: MutableByteArray# s -> Int# -> State# s -> (# State# s, Double# #)
1136 writeArray# :: MutableArray# s elt -> Int# -> elt -> State# s -> State# s
1137 writeCharArray# :: MutableByteArray# s -> Int# -> Char# -> State# s -> State# s
1138 writeIntArray# :: MutableByteArray# s -> Int# -> Int# -> State# s -> State# s
1139 writeAddrArray# :: MutableByteArray# s -> Int# -> Addr# -> State# s -> State# s
1140 writeFloatArray# :: MutableByteArray# s -> Int# -> Float# -> State# s -> State# s
1141 writeDoubleArray# :: MutableByteArray# s -> Int# -> Double# -> State# s -> State# s
1144 <IndexTerm><Primary><literal>readArray#</literal></Primary></IndexTerm>
1145 <IndexTerm><Primary><literal>readCharArray#</literal></Primary></IndexTerm>
1146 <IndexTerm><Primary><literal>readIntArray#</literal></Primary></IndexTerm>
1147 <IndexTerm><Primary><literal>readAddrArray#</literal></Primary></IndexTerm>
1148 <IndexTerm><Primary><literal>readFloatArray#</literal></Primary></IndexTerm>
1149 <IndexTerm><Primary><literal>readDoubleArray#</literal></Primary></IndexTerm>
1150 <IndexTerm><Primary><literal>writeArray#</literal></Primary></IndexTerm>
1151 <IndexTerm><Primary><literal>writeCharArray#</literal></Primary></IndexTerm>
1152 <IndexTerm><Primary><literal>writeIntArray#</literal></Primary></IndexTerm>
1153 <IndexTerm><Primary><literal>writeAddrArray#</literal></Primary></IndexTerm>
1154 <IndexTerm><Primary><literal>writeFloatArray#</literal></Primary></IndexTerm>
1155 <IndexTerm><Primary><literal>writeDoubleArray#</literal></Primary></IndexTerm>
1161 <Title>Equality</Title>
1164 <IndexTerm><Primary>arrays, testing for equality</Primary></IndexTerm>
1168 One can take “equality” of mutable arrays. What is compared is the
1169 <Emphasis>name</Emphasis> or reference to the mutable array, not its contents.
1175 sameMutableArray# :: MutableArray# s elt -> MutableArray# s elt -> Bool
1176 sameMutableByteArray# :: MutableByteArray# s -> MutableByteArray# s -> Bool
1179 <IndexTerm><Primary><literal>sameMutableArray#</literal></Primary></IndexTerm>
1180 <IndexTerm><Primary><literal>sameMutableByteArray#</literal></Primary></IndexTerm>
1186 <Title>Freezing mutable arrays</Title>
1189 <IndexTerm><Primary>arrays, freezing mutable</Primary></IndexTerm>
1190 <IndexTerm><Primary>freezing mutable arrays</Primary></IndexTerm>
1191 <IndexTerm><Primary>mutable arrays, freezing</Primary></IndexTerm>
1195 Only unsafe-freeze has a primitive. (Safe freeze is done directly in Haskell
1196 by copying the array and then using <Function>unsafeFreeze</Function>.)
1202 unsafeFreezeArray# :: MutableArray# s elt -> State# s -> (# State# s, Array# s elt #)
1203 unsafeFreezeByteArray# :: MutableByteArray# s -> State# s -> (# State# s, ByteArray# #)
1206 <IndexTerm><Primary><literal>unsafeFreezeArray#</literal></Primary></IndexTerm>
1207 <IndexTerm><Primary><literal>unsafeFreezeByteArray#</literal></Primary></IndexTerm>
1215 <Title>Synchronizing variables (M-vars)</Title>
1218 <IndexTerm><Primary>synchronising variables (M-vars)</Primary></IndexTerm>
1219 <IndexTerm><Primary>M-Vars</Primary></IndexTerm>
1223 Synchronising variables are the primitive type used to implement
1224 Concurrent Haskell's MVars (see the Concurrent Haskell paper for
1225 the operational behaviour of these operations).
1231 type MVar# s elt -- primitive
1233 newMVar# :: State# s -> (# State# s, MVar# s elt #)
1234 takeMVar# :: SynchVar# s elt -> State# s -> (# State# s, elt #)
1235 putMVar# :: SynchVar# s elt -> State# s -> State# s
1238 <IndexTerm><Primary><literal>SynchVar#</literal></Primary></IndexTerm>
1239 <IndexTerm><Primary><literal>newSynchVar#</literal></Primary></IndexTerm>
1240 <IndexTerm><Primary><literal>takeMVar</literal></Primary></IndexTerm>
1241 <IndexTerm><Primary><literal>putMVar</literal></Primary></IndexTerm>
1248 <Sect1 id="glasgow-ST-monad">
1249 <Title>Primitive state-transformer monad
1253 <IndexTerm><Primary>state transformers (Glasgow extensions)</Primary></IndexTerm>
1254 <IndexTerm><Primary>ST monad (Glasgow extension)</Primary></IndexTerm>
1258 This monad underlies our implementation of arrays, mutable and
1259 immutable, and our implementation of I/O, including “C calls”.
1263 The <Literal>ST</Literal> library, which provides access to the
1264 <Function>ST</Function> monad, is described in <xref
1270 <Sect1 id="glasgow-prim-arrays">
1271 <Title>Primitive arrays, mutable and otherwise
1275 <IndexTerm><Primary>primitive arrays (Glasgow extension)</Primary></IndexTerm>
1276 <IndexTerm><Primary>arrays, primitive (Glasgow extension)</Primary></IndexTerm>
1280 GHC knows about quite a few flavours of Large Swathes of Bytes.
1284 First, GHC distinguishes between primitive arrays of (boxed) Haskell
1285 objects (type <Literal>Array# obj</Literal>) and primitive arrays of bytes (type
1286 <Literal>ByteArray#</Literal>).
1290 Second, it distinguishes between…
1294 <Term>Immutable:</Term>
1297 Arrays that do not change (as with “standard” Haskell arrays); you
1298 can only read from them. Obviously, they do not need the care and
1299 attention of the state-transformer monad.
1304 <Term>Mutable:</Term>
1307 Arrays that may be changed or “mutated.” All the operations on them
1308 live within the state-transformer monad and the updates happen
1309 <Emphasis>in-place</Emphasis>.
1314 <Term>“Static” (in C land):</Term>
1317 A C routine may pass an <Literal>Addr#</Literal> pointer back into Haskell land. There
1318 are then primitive operations with which you may merrily grab values
1319 over in C land, by indexing off the “static” pointer.
1324 <Term>“Stable” pointers:</Term>
1327 If, for some reason, you wish to hand a Haskell pointer (i.e.,
1328 <Emphasis>not</Emphasis> an unboxed value) to a C routine, you first make the
1329 pointer “stable,” so that the garbage collector won't forget that it
1330 exists. That is, GHC provides a safe way to pass Haskell pointers to
1335 Please see <XRef LinkEnd="sec-stable-pointers"> for more details.
1340 <Term>“Foreign objects”:</Term>
1343 A “foreign object” is a safe way to pass an external object (a
1344 C-allocated pointer, say) to Haskell and have Haskell do the Right
1345 Thing when it no longer references the object. So, for example, C
1346 could pass a large bitmap over to Haskell and say “please free this
1347 memory when you're done with it.”
1351 Please see <XRef LinkEnd="sec-ForeignObj"> for more details.
1359 The libraries documentatation gives more details on all these
1360 “primitive array” types and the operations on them.
1366 <Sect1 id="pattern-guards">
1367 <Title>Pattern guards</Title>
1370 <IndexTerm><Primary>Pattern guards (Glasgow extension)</Primary></IndexTerm>
1371 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.)
1375 Suppose we have an abstract data type of finite maps, with a
1379 lookup :: FiniteMap -> Int -> Maybe Int
1382 The lookup returns <Function>Nothing</Function> if the supplied key is not in the domain of the mapping, and <Function>(Just v)</Function> otherwise,
1383 where <VarName>v</VarName> is the value that the key maps to. Now consider the following definition:
1387 clunky env var1 var2 | ok1 && ok2 = val1 + val2
1388 | otherwise = var1 + var2
1390 m1 = lookup env var1
1391 m2 = lookup env var2
1392 ok1 = maybeToBool m1
1393 ok2 = maybeToBool m2
1394 val1 = expectJust m1
1395 val2 = expectJust m2
1399 The auxiliary functions are
1403 maybeToBool :: Maybe a -> Bool
1404 maybeToBool (Just x) = True
1405 maybeToBool Nothing = False
1407 expectJust :: Maybe a -> a
1408 expectJust (Just x) = x
1409 expectJust Nothing = error "Unexpected Nothing"
1413 What is <Function>clunky</Function> doing? The guard <Literal>ok1 &&
1414 ok2</Literal> checks that both lookups succeed, using
1415 <Function>maybeToBool</Function> to convert the <Function>Maybe</Function>
1416 types to booleans. The (lazily evaluated) <Function>expectJust</Function>
1417 calls extract the values from the results of the lookups, and binds the
1418 returned values to <VarName>val1</VarName> and <VarName>val2</VarName>
1419 respectively. If either lookup fails, then clunky takes the
1420 <Literal>otherwise</Literal> case and returns the sum of its arguments.
1424 This is certainly legal Haskell, but it is a tremendously verbose and
1425 un-obvious way to achieve the desired effect. Arguably, a more direct way
1426 to write clunky would be to use case expressions:
1430 clunky env var1 var1 = case lookup env var1 of
1432 Just val1 -> case lookup env var2 of
1434 Just val2 -> val1 + val2
1440 This is a bit shorter, but hardly better. Of course, we can rewrite any set
1441 of pattern-matching, guarded equations as case expressions; that is
1442 precisely what the compiler does when compiling equations! The reason that
1443 Haskell provides guarded equations is because they allow us to write down
1444 the cases we want to consider, one at a time, independently of each other.
1445 This structure is hidden in the case version. Two of the right-hand sides
1446 are really the same (<Function>fail</Function>), and the whole expression
1447 tends to become more and more indented.
1451 Here is how I would write clunky:
1455 clunky env var1 var1
1456 | Just val1 <- lookup env var1
1457 , Just val2 <- lookup env var2
1459 ...other equations for clunky...
1463 The semantics should be clear enough. The qualifers are matched in order.
1464 For a <Literal><-</Literal> qualifier, which I call a pattern guard, the
1465 right hand side is evaluated and matched against the pattern on the left.
1466 If the match fails then the whole guard fails and the next equation is
1467 tried. If it succeeds, then the appropriate binding takes place, and the
1468 next qualifier is matched, in the augmented environment. Unlike list
1469 comprehensions, however, the type of the expression to the right of the
1470 <Literal><-</Literal> is the same as the type of the pattern to its
1471 left. The bindings introduced by pattern guards scope over all the
1472 remaining guard qualifiers, and over the right hand side of the equation.
1476 Just as with list comprehensions, boolean expressions can be freely mixed
1477 with among the pattern guards. For example:
1488 Haskell's current guards therefore emerge as a special case, in which the
1489 qualifier list has just one element, a boolean expression.
1493 <sect1 id="sec-ffi">
1494 <title>The foreign interface</title>
1496 <para>The foreign interface consists of the following components:</para>
1500 <para>The Foreign Function Interface language specification
1501 (included in this manual, in <xref linkend="ffi">).</para>
1505 <para>The <literal>Foreign</literal> module (see <xref
1506 linkend="sec-Foreign">) collects together several interfaces
1507 which are useful in specifying foreign language
1508 interfaces, including the following:</para>
1512 <para>The <literal>ForeignObj</literal> module (see <xref
1513 linkend="sec-ForeignObj">), for managing pointers from
1514 Haskell into the outside world.</para>
1518 <para>The <literal>StablePtr</literal> module (see <xref
1519 linkend="sec-stable-pointers">), for managing pointers
1520 into Haskell from the outside world.</para>
1524 <para>The <literal>CTypes</literal> module (see <xref
1525 linkend="sec-CTypes">) gives Haskell equivalents for the
1526 standard C datatypes, for use in making Haskell bindings
1527 to existing C libraries.</para>
1531 <para>The <literal>CTypesISO</literal> module (see <xref
1532 linkend="sec-CTypesISO">) gives Haskell equivalents for C
1533 types defined by the ISO C standard.</para>
1537 <para>The <literal>Storable</literal> library, for
1538 primitive marshalling of data types between Haskell and
1539 the foreign language.</para>
1546 <para>The following sections also give some hints and tips on the use
1547 of the foreign function interface in GHC.</para>
1549 <Sect2 id="glasgow-foreign-headers">
1550 <Title>Using function headers
1554 <IndexTerm><Primary>C calls, function headers</Primary></IndexTerm>
1558 When generating C (using the <Option>-fvia-C</Option> directive), one can assist the
1559 C compiler in detecting type errors by using the <Command>-#include</Command> directive
1560 to provide <Filename>.h</Filename> files containing function headers.
1572 void initialiseEFS (HsInt size);
1573 HsInt terminateEFS (void);
1574 HsForeignObj emptyEFS(void);
1575 HsForeignObj updateEFS (HsForeignObj a, HsInt i, HsInt x);
1576 HsInt lookupEFS (HsForeignObj a, HsInt i);
1580 <para>The types <literal>HsInt</literal>,
1581 <literal>HsForeignObj</literal> etc. are described in <xref
1582 linkend="sec-mapping-table">.</Para>
1584 <Para>Note that this approach is only
1585 <Emphasis>essential</Emphasis> for returning
1586 <Literal>float</Literal>s (or if <Literal>sizeof(int) !=
1587 sizeof(int *)</Literal> on your architecture) but is a Good
1588 Thing for anyone who cares about writing solid code. You're
1589 crazy not to do it.</Para>
1595 <Sect1 id="multi-param-type-classes">
1596 <Title>Multi-parameter type classes
1600 This section documents GHC's implementation of multi-parameter type
1601 classes. There's lots of background in the paper <ULink
1602 URL="http://research.microsoft.com/~simonpj/multi.ps.gz" >Type
1603 classes: exploring the design space</ULink > (Simon Peyton Jones, Mark
1604 Jones, Erik Meijer).
1608 I'd like to thank people who reported shorcomings in the GHC 3.02
1609 implementation. Our default decisions were all conservative ones, and
1610 the experience of these heroic pioneers has given useful concrete
1611 examples to support several generalisations. (These appear below as
1612 design choices not implemented in 3.02.)
1616 I've discussed these notes with Mark Jones, and I believe that Hugs
1617 will migrate towards the same design choices as I outline here.
1618 Thanks to him, and to many others who have offered very useful
1623 <Title>Types</Title>
1626 There are the following restrictions on the form of a qualified
1633 forall tv1..tvn (c1, ...,cn) => type
1639 (Here, I write the "foralls" explicitly, although the Haskell source
1640 language omits them; in Haskell 1.4, all the free type variables of an
1641 explicit source-language type signature are universally quantified,
1642 except for the class type variables in a class declaration. However,
1643 in GHC, you can give the foralls if you want. See <XRef LinkEnd="universal-quantification">).
1652 <Emphasis>Each universally quantified type variable
1653 <Literal>tvi</Literal> must be mentioned (i.e. appear free) in <Literal>type</Literal></Emphasis>.
1655 The reason for this is that a value with a type that does not obey
1656 this restriction could not be used without introducing
1657 ambiguity. Here, for example, is an illegal type:
1661 forall a. Eq a => Int
1665 When a value with this type was used, the constraint <Literal>Eq tv</Literal>
1666 would be introduced where <Literal>tv</Literal> is a fresh type variable, and
1667 (in the dictionary-translation implementation) the value would be
1668 applied to a dictionary for <Literal>Eq tv</Literal>. The difficulty is that we
1669 can never know which instance of <Literal>Eq</Literal> to use because we never
1670 get any more information about <Literal>tv</Literal>.
1677 <Emphasis>Every constraint <Literal>ci</Literal> must mention at least one of the
1678 universally quantified type variables <Literal>tvi</Literal></Emphasis>.
1680 For example, this type is OK because <Literal>C a b</Literal> mentions the
1681 universally quantified type variable <Literal>b</Literal>:
1685 forall a. C a b => burble
1689 The next type is illegal because the constraint <Literal>Eq b</Literal> does not
1690 mention <Literal>a</Literal>:
1694 forall a. Eq b => burble
1698 The reason for this restriction is milder than the other one. The
1699 excluded types are never useful or necessary (because the offending
1700 context doesn't need to be witnessed at this point; it can be floated
1701 out). Furthermore, floating them out increases sharing. Lastly,
1702 excluding them is a conservative choice; it leaves a patch of
1703 territory free in case we need it later.
1713 These restrictions apply to all types, whether declared in a type signature
1718 Unlike Haskell 1.4, constraints in types do <Emphasis>not</Emphasis> have to be of
1719 the form <Emphasis>(class type-variables)</Emphasis>. Thus, these type signatures
1726 f :: Eq (m a) => [m a] -> [m a]
1733 This choice recovers principal types, a property that Haskell 1.4 does not have.
1739 <Title>Class declarations</Title>
1747 <Emphasis>Multi-parameter type classes are permitted</Emphasis>. For example:
1751 class Collection c a where
1752 union :: c a -> c a -> c a
1763 <Emphasis>The class hierarchy must be acyclic</Emphasis>. However, the definition
1764 of "acyclic" involves only the superclass relationships. For example,
1770 op :: D b => a -> b -> b
1773 class C a => D a where { ... }
1777 Here, <Literal>C</Literal> is a superclass of <Literal>D</Literal>, but it's OK for a
1778 class operation <Literal>op</Literal> of <Literal>C</Literal> to mention <Literal>D</Literal>. (It
1779 would not be OK for <Literal>D</Literal> to be a superclass of <Literal>C</Literal>.)
1786 <Emphasis>There are no restrictions on the context in a class declaration
1787 (which introduces superclasses), except that the class hierarchy must
1788 be acyclic</Emphasis>. So these class declarations are OK:
1792 class Functor (m k) => FiniteMap m k where
1795 class (Monad m, Monad (t m)) => Transform t m where
1796 lift :: m a -> (t m) a
1805 <Emphasis>In the signature of a class operation, every constraint
1806 must mention at least one type variable that is not a class type
1807 variable</Emphasis>.
1813 class Collection c a where
1814 mapC :: Collection c b => (a->b) -> c a -> c b
1818 is OK because the constraint <Literal>(Collection a b)</Literal> mentions
1819 <Literal>b</Literal>, even though it also mentions the class variable
1820 <Literal>a</Literal>. On the other hand:
1825 op :: Eq a => (a,b) -> (a,b)
1829 is not OK because the constraint <Literal>(Eq a)</Literal> mentions on the class
1830 type variable <Literal>a</Literal>, but not <Literal>b</Literal>. However, any such
1831 example is easily fixed by moving the offending context up to the
1836 class Eq a => C a where
1841 A yet more relaxed rule would allow the context of a class-op signature
1842 to mention only class type variables. However, that conflicts with
1843 Rule 1(b) for types above.
1850 <Emphasis>The type of each class operation must mention <Emphasis>all</Emphasis> of
1851 the class type variables</Emphasis>. For example:
1855 class Coll s a where
1857 insert :: s -> a -> s
1861 is not OK, because the type of <Literal>empty</Literal> doesn't mention
1862 <Literal>a</Literal>. This rule is a consequence of Rule 1(a), above, for
1863 types, and has the same motivation.
1865 Sometimes, offending class declarations exhibit misunderstandings. For
1866 example, <Literal>Coll</Literal> might be rewritten
1870 class Coll s a where
1872 insert :: s a -> a -> s a
1876 which makes the connection between the type of a collection of
1877 <Literal>a</Literal>'s (namely <Literal>(s a)</Literal>) and the element type <Literal>a</Literal>.
1878 Occasionally this really doesn't work, in which case you can split the
1886 class CollE s => Coll s a where
1887 insert :: s -> a -> s
1901 <Title>Instance declarations</Title>
1909 <Emphasis>Instance declarations may not overlap</Emphasis>. The two instance
1914 instance context1 => C type1 where ...
1915 instance context2 => C type2 where ...
1919 "overlap" if <Literal>type1</Literal> and <Literal>type2</Literal> unify
1921 However, if you give the command line option
1922 <Option>-fallow-overlapping-instances</Option><IndexTerm><Primary>-fallow-overlapping-instances
1923 option</Primary></IndexTerm> then two overlapping instance declarations are permitted
1931 EITHER <Literal>type1</Literal> and <Literal>type2</Literal> do not unify
1937 OR <Literal>type2</Literal> is a substitution instance of <Literal>type1</Literal>
1938 (but not identical to <Literal>type1</Literal>)
1951 Notice that these rules
1958 make it clear which instance decl to use
1959 (pick the most specific one that matches)
1966 do not mention the contexts <Literal>context1</Literal>, <Literal>context2</Literal>
1967 Reason: you can pick which instance decl
1968 "matches" based on the type.
1975 Regrettably, GHC doesn't guarantee to detect overlapping instance
1976 declarations if they appear in different modules. GHC can "see" the
1977 instance declarations in the transitive closure of all the modules
1978 imported by the one being compiled, so it can "see" all instance decls
1979 when it is compiling <Literal>Main</Literal>. However, it currently chooses not
1980 to look at ones that can't possibly be of use in the module currently
1981 being compiled, in the interests of efficiency. (Perhaps we should
1982 change that decision, at least for <Literal>Main</Literal>.)
1989 <Emphasis>There are no restrictions on the type in an instance
1990 <Emphasis>head</Emphasis>, except that at least one must not be a type variable</Emphasis>.
1991 The instance "head" is the bit after the "=>" in an instance decl. For
1992 example, these are OK:
1996 instance C Int a where ...
1998 instance D (Int, Int) where ...
2000 instance E [[a]] where ...
2004 Note that instance heads <Emphasis>may</Emphasis> contain repeated type variables.
2005 For example, this is OK:
2009 instance Stateful (ST s) (MutVar s) where ...
2013 The "at least one not a type variable" restriction is to ensure that
2014 context reduction terminates: each reduction step removes one type
2015 constructor. For example, the following would make the type checker
2016 loop if it wasn't excluded:
2020 instance C a => C a where ...
2024 There are two situations in which the rule is a bit of a pain. First,
2025 if one allows overlapping instance declarations then it's quite
2026 convenient to have a "default instance" declaration that applies if
2027 something more specific does not:
2036 Second, sometimes you might want to use the following to get the
2037 effect of a "class synonym":
2041 class (C1 a, C2 a, C3 a) => C a where { }
2043 instance (C1 a, C2 a, C3 a) => C a where { }
2047 This allows you to write shorter signatures:
2059 f :: (C1 a, C2 a, C3 a) => ...
2063 I'm on the lookout for a simple rule that preserves decidability while
2064 allowing these idioms. The experimental flag
2065 <Option>-fallow-undecidable-instances</Option><IndexTerm><Primary>-fallow-undecidable-instances
2066 option</Primary></IndexTerm> lifts this restriction, allowing all the types in an
2067 instance head to be type variables.
2074 <Emphasis>Unlike Haskell 1.4, instance heads may use type
2075 synonyms</Emphasis>. As always, using a type synonym is just shorthand for
2076 writing the RHS of the type synonym definition. For example:
2080 type Point = (Int,Int)
2081 instance C Point where ...
2082 instance C [Point] where ...
2086 is legal. However, if you added
2090 instance C (Int,Int) where ...
2094 as well, then the compiler will complain about the overlapping
2095 (actually, identical) instance declarations. As always, type synonyms
2096 must be fully applied. You cannot, for example, write:
2101 instance Monad P where ...
2105 This design decision is independent of all the others, and easily
2106 reversed, but it makes sense to me.
2113 <Emphasis>The types in an instance-declaration <Emphasis>context</Emphasis> must all
2114 be type variables</Emphasis>. Thus
2118 instance C a b => Eq (a,b) where ...
2126 instance C Int b => Foo b where ...
2130 is not OK. Again, the intent here is to make sure that context
2131 reduction terminates.
2133 Voluminous correspondence on the Haskell mailing list has convinced me
2134 that it's worth experimenting with a more liberal rule. If you use
2135 the flag <Option>-fallow-undecidable-instances</Option> can use arbitrary
2136 types in an instance context. Termination is ensured by having a
2137 fixed-depth recursion stack. If you exceed the stack depth you get a
2138 sort of backtrace, and the opportunity to increase the stack depth
2139 with <Option>-fcontext-stack</Option><Emphasis>N</Emphasis>.
2152 <Sect1 id="universal-quantification">
2153 <Title>Explicit universal quantification
2157 GHC now allows you to write explicitly quantified types. GHC's
2158 syntax for this now agrees with Hugs's, namely:
2164 forall a b. (Ord a, Eq b) => a -> b -> a
2170 The context is, of course, optional. You can't use <Literal>forall</Literal> as
2171 a type variable any more!
2175 Haskell type signatures are implicitly quantified. The <Literal>forall</Literal>
2176 allows us to say exactly what this means. For example:
2194 g :: forall b. (b -> b)
2200 The two are treated identically.
2204 <Title>Universally-quantified data type fields
2208 In a <Literal>data</Literal> or <Literal>newtype</Literal> declaration one can quantify
2209 the types of the constructor arguments. Here are several examples:
2215 data T a = T1 (forall b. b -> b -> b) a
2217 data MonadT m = MkMonad { return :: forall a. a -> m a,
2218 bind :: forall a b. m a -> (a -> m b) -> m b
2221 newtype Swizzle = MkSwizzle (Ord a => [a] -> [a])
2227 The constructors now have so-called <Emphasis>rank 2</Emphasis> polymorphic
2228 types, in which there is a for-all in the argument types.:
2234 T1 :: forall a. (forall b. b -> b -> b) -> a -> T a
2235 MkMonad :: forall m. (forall a. a -> m a)
2236 -> (forall a b. m a -> (a -> m b) -> m b)
2238 MkSwizzle :: (Ord a => [a] -> [a]) -> Swizzle
2244 Notice that you don't need to use a <Literal>forall</Literal> if there's an
2245 explicit context. For example in the first argument of the
2246 constructor <Function>MkSwizzle</Function>, an implicit "<Literal>forall a.</Literal>" is
2247 prefixed to the argument type. The implicit <Literal>forall</Literal>
2248 quantifies all type variables that are not already in scope, and are
2249 mentioned in the type quantified over.
2253 As for type signatures, implicit quantification happens for non-overloaded
2254 types too. So if you write this:
2257 data T a = MkT (Either a b) (b -> b)
2260 it's just as if you had written this:
2263 data T a = MkT (forall b. Either a b) (forall b. b -> b)
2266 That is, since the type variable <Literal>b</Literal> isn't in scope, it's
2267 implicitly universally quantified. (Arguably, it would be better
2268 to <Emphasis>require</Emphasis> explicit quantification on constructor arguments
2269 where that is what is wanted. Feedback welcomed.)
2275 <Title>Construction </Title>
2278 You construct values of types <Literal>T1, MonadT, Swizzle</Literal> by applying
2279 the constructor to suitable values, just as usual. For example,
2285 (T1 (\xy->x) 3) :: T Int
2287 (MkSwizzle sort) :: Swizzle
2288 (MkSwizzle reverse) :: Swizzle
2295 MkMonad r b) :: MonadT Maybe
2301 The type of the argument can, as usual, be more general than the type
2302 required, as <Literal>(MkSwizzle reverse)</Literal> shows. (<Function>reverse</Function>
2303 does not need the <Literal>Ord</Literal> constraint.)
2309 <Title>Pattern matching</Title>
2312 When you use pattern matching, the bound variables may now have
2313 polymorphic types. For example:
2319 f :: T a -> a -> (a, Char)
2320 f (T1 f k) x = (f k x, f 'c' 'd')
2322 g :: (Ord a, Ord b) => Swizzle -> [a] -> (a -> b) -> [b]
2323 g (MkSwizzle s) xs f = s (map f (s xs))
2325 h :: MonadT m -> [m a] -> m [a]
2326 h m [] = return m []
2327 h m (x:xs) = bind m x $ \y ->
2328 bind m (h m xs) $ \ys ->
2335 In the function <Function>h</Function> we use the record selectors <Literal>return</Literal>
2336 and <Literal>bind</Literal> to extract the polymorphic bind and return functions
2337 from the <Literal>MonadT</Literal> data structure, rather than using pattern
2342 You cannot pattern-match against an argument that is polymorphic.
2346 newtype TIM s a = TIM (ST s (Maybe a))
2348 runTIM :: (forall s. TIM s a) -> Maybe a
2349 runTIM (TIM m) = runST m
2355 Here the pattern-match fails, because you can't pattern-match against
2356 an argument of type <Literal>(forall s. TIM s a)</Literal>. Instead you
2357 must bind the variable and pattern match in the right hand side:
2360 runTIM :: (forall s. TIM s a) -> Maybe a
2361 runTIM tm = case tm of { TIM m -> runST m }
2364 The <Literal>tm</Literal> on the right hand side is (invisibly) instantiated, like
2365 any polymorphic value at its occurrence site, and now you can pattern-match
2372 <Title>The partial-application restriction</Title>
2375 There is really only one way in which data structures with polymorphic
2376 components might surprise you: you must not partially apply them.
2377 For example, this is illegal:
2383 map MkSwizzle [sort, reverse]
2389 The restriction is this: <Emphasis>every subexpression of the program must
2390 have a type that has no for-alls, except that in a function
2391 application (f e1…en) the partial applications are not subject to
2392 this rule</Emphasis>. The restriction makes type inference feasible.
2396 In the illegal example, the sub-expression <Literal>MkSwizzle</Literal> has the
2397 polymorphic type <Literal>(Ord b => [b] -> [b]) -> Swizzle</Literal> and is not
2398 a sub-expression of an enclosing application. On the other hand, this
2405 map (T1 (\a b -> a)) [1,2,3]
2411 even though it involves a partial application of <Function>T1</Function>, because
2412 the sub-expression <Literal>T1 (\a b -> a)</Literal> has type <Literal>Int -> T
2419 <Title>Type signatures
2423 Once you have data constructors with universally-quantified fields, or
2424 constants such as <Constant>runST</Constant> that have rank-2 types, it isn't long
2425 before you discover that you need more! Consider:
2431 mkTs f x y = [T1 f x, T1 f y]
2437 <Function>mkTs</Function> is a fuction that constructs some values of type
2438 <Literal>T</Literal>, using some pieces passed to it. The trouble is that since
2439 <Literal>f</Literal> is a function argument, Haskell assumes that it is
2440 monomorphic, so we'll get a type error when applying <Function>T1</Function> to
2441 it. This is a rather silly example, but the problem really bites in
2442 practice. Lots of people trip over the fact that you can't make
2443 "wrappers functions" for <Constant>runST</Constant> for exactly the same reason.
2444 In short, it is impossible to build abstractions around functions with
2449 The solution is fairly clear. We provide the ability to give a rank-2
2450 type signature for <Emphasis>ordinary</Emphasis> functions (not only data
2451 constructors), thus:
2457 mkTs :: (forall b. b -> b -> b) -> a -> [T a]
2458 mkTs f x y = [T1 f x, T1 f y]
2464 This type signature tells the compiler to attribute <Literal>f</Literal> with
2465 the polymorphic type <Literal>(forall b. b -> b -> b)</Literal> when type
2466 checking the body of <Function>mkTs</Function>, so now the application of
2467 <Function>T1</Function> is fine.
2471 There are two restrictions:
2480 You can only define a rank 2 type, specified by the following
2485 rank2type ::= [forall tyvars .] [context =>] funty
2486 funty ::= ([forall tyvars .] [context =>] ty) -> funty
2488 ty ::= ...current Haskell monotype syntax...
2492 Informally, the universal quantification must all be right at the beginning,
2493 or at the top level of a function argument.
2500 There is a restriction on the definition of a function whose
2501 type signature is a rank-2 type: the polymorphic arguments must be
2502 matched on the left hand side of the "<Literal>=</Literal>" sign. You can't
2503 define <Function>mkTs</Function> like this:
2507 mkTs :: (forall b. b -> b -> b) -> a -> [T a]
2508 mkTs = \ f x y -> [T1 f x, T1 f y]
2513 The same partial-application rule applies to ordinary functions with
2514 rank-2 types as applied to data constructors.
2527 <Title>Type synonyms and hoisting
2531 GHC also allows you to write a <Literal>forall</Literal> in a type synonym, thus:
2533 type Discard a = forall b. a -> b -> a
2538 However, it is often convenient to use these sort of synonyms at the right hand
2539 end of an arrow, thus:
2541 type Discard a = forall b. a -> b -> a
2543 g :: Int -> Discard Int
2546 Simply expanding the type synonym would give
2548 g :: Int -> (forall b. Int -> b -> Int)
2550 but GHC "hoists" the <Literal>forall</Literal> to give the isomorphic type
2552 g :: forall b. Int -> Int -> b -> Int
2554 In general, the rule is this: <Emphasis>to determine the type specified by any explicit
2555 user-written type (e.g. in a type signature), GHC expands type synonyms and then repeatedly
2556 performs the transformation:</Emphasis>
2558 <Emphasis>type1</Emphasis> -> forall a. <Emphasis>type2</Emphasis>
2560 forall a. <Emphasis>type1</Emphasis> -> <Emphasis>type2</Emphasis>
2562 (In fact, GHC tries to retain as much synonym information as possible for use in
2563 error messages, but that is a usability issue.) This rule applies, of course, whether
2564 or not the <Literal>forall</Literal> comes from a synonym. For example, here is another
2565 valid way to write <Literal>g</Literal>'s type signature:
2567 g :: Int -> Int -> forall b. b -> Int
2574 <Sect1 id="existential-quantification">
2575 <Title>Existentially quantified data constructors
2579 The idea of using existential quantification in data type declarations
2580 was suggested by Laufer (I believe, thought doubtless someone will
2581 correct me), and implemented in Hope+. It's been in Lennart
2582 Augustsson's <Command>hbc</Command> Haskell compiler for several years, and
2583 proved very useful. Here's the idea. Consider the declaration:
2589 data Foo = forall a. MkFoo a (a -> Bool)
2596 The data type <Literal>Foo</Literal> has two constructors with types:
2602 MkFoo :: forall a. a -> (a -> Bool) -> Foo
2609 Notice that the type variable <Literal>a</Literal> in the type of <Function>MkFoo</Function>
2610 does not appear in the data type itself, which is plain <Literal>Foo</Literal>.
2611 For example, the following expression is fine:
2617 [MkFoo 3 even, MkFoo 'c' isUpper] :: [Foo]
2623 Here, <Literal>(MkFoo 3 even)</Literal> packages an integer with a function
2624 <Function>even</Function> that maps an integer to <Literal>Bool</Literal>; and <Function>MkFoo 'c'
2625 isUpper</Function> packages a character with a compatible function. These
2626 two things are each of type <Literal>Foo</Literal> and can be put in a list.
2630 What can we do with a value of type <Literal>Foo</Literal>?. In particular,
2631 what happens when we pattern-match on <Function>MkFoo</Function>?
2637 f (MkFoo val fn) = ???
2643 Since all we know about <Literal>val</Literal> and <Function>fn</Function> is that they
2644 are compatible, the only (useful) thing we can do with them is to
2645 apply <Function>fn</Function> to <Literal>val</Literal> to get a boolean. For example:
2652 f (MkFoo val fn) = fn val
2658 What this allows us to do is to package heterogenous values
2659 together with a bunch of functions that manipulate them, and then treat
2660 that collection of packages in a uniform manner. You can express
2661 quite a bit of object-oriented-like programming this way.
2664 <Sect2 id="existential">
2665 <Title>Why existential?
2669 What has this to do with <Emphasis>existential</Emphasis> quantification?
2670 Simply that <Function>MkFoo</Function> has the (nearly) isomorphic type
2676 MkFoo :: (exists a . (a, a -> Bool)) -> Foo
2682 But Haskell programmers can safely think of the ordinary
2683 <Emphasis>universally</Emphasis> quantified type given above, thereby avoiding
2684 adding a new existential quantification construct.
2690 <Title>Type classes</Title>
2693 An easy extension (implemented in <Command>hbc</Command>) is to allow
2694 arbitrary contexts before the constructor. For example:
2700 data Baz = forall a. Eq a => Baz1 a a
2701 | forall b. Show b => Baz2 b (b -> b)
2707 The two constructors have the types you'd expect:
2713 Baz1 :: forall a. Eq a => a -> a -> Baz
2714 Baz2 :: forall b. Show b => b -> (b -> b) -> Baz
2720 But when pattern matching on <Function>Baz1</Function> the matched values can be compared
2721 for equality, and when pattern matching on <Function>Baz2</Function> the first matched
2722 value can be converted to a string (as well as applying the function to it).
2723 So this program is legal:
2730 f (Baz1 p q) | p == q = "Yes"
2732 f (Baz1 v fn) = show (fn v)
2738 Operationally, in a dictionary-passing implementation, the
2739 constructors <Function>Baz1</Function> and <Function>Baz2</Function> must store the
2740 dictionaries for <Literal>Eq</Literal> and <Literal>Show</Literal> respectively, and
2741 extract it on pattern matching.
2745 Notice the way that the syntax fits smoothly with that used for
2746 universal quantification earlier.
2752 <Title>Restrictions</Title>
2755 There are several restrictions on the ways in which existentially-quantified
2756 constructors can be use.
2765 When pattern matching, each pattern match introduces a new,
2766 distinct, type for each existential type variable. These types cannot
2767 be unified with any other type, nor can they escape from the scope of
2768 the pattern match. For example, these fragments are incorrect:
2776 Here, the type bound by <Function>MkFoo</Function> "escapes", because <Literal>a</Literal>
2777 is the result of <Function>f1</Function>. One way to see why this is wrong is to
2778 ask what type <Function>f1</Function> has:
2782 f1 :: Foo -> a -- Weird!
2786 What is this "<Literal>a</Literal>" in the result type? Clearly we don't mean
2791 f1 :: forall a. Foo -> a -- Wrong!
2795 The original program is just plain wrong. Here's another sort of error
2799 f2 (Baz1 a b) (Baz1 p q) = a==q
2803 It's ok to say <Literal>a==b</Literal> or <Literal>p==q</Literal>, but
2804 <Literal>a==q</Literal> is wrong because it equates the two distinct types arising
2805 from the two <Function>Baz1</Function> constructors.
2813 You can't pattern-match on an existentially quantified
2814 constructor in a <Literal>let</Literal> or <Literal>where</Literal> group of
2815 bindings. So this is illegal:
2819 f3 x = a==b where { Baz1 a b = x }
2823 You can only pattern-match
2824 on an existentially-quantified constructor in a <Literal>case</Literal> expression or
2825 in the patterns of a function definition.
2827 The reason for this restriction is really an implementation one.
2828 Type-checking binding groups is already a nightmare without
2829 existentials complicating the picture. Also an existential pattern
2830 binding at the top level of a module doesn't make sense, because it's
2831 not clear how to prevent the existentially-quantified type "escaping".
2832 So for now, there's a simple-to-state restriction. We'll see how
2840 You can't use existential quantification for <Literal>newtype</Literal>
2841 declarations. So this is illegal:
2845 newtype T = forall a. Ord a => MkT a
2849 Reason: a value of type <Literal>T</Literal> must be represented as a pair
2850 of a dictionary for <Literal>Ord t</Literal> and a value of type <Literal>t</Literal>.
2851 That contradicts the idea that <Literal>newtype</Literal> should have no
2852 concrete representation. You can get just the same efficiency and effect
2853 by using <Literal>data</Literal> instead of <Literal>newtype</Literal>. If there is no
2854 overloading involved, then there is more of a case for allowing
2855 an existentially-quantified <Literal>newtype</Literal>, because the <Literal>data</Literal>
2856 because the <Literal>data</Literal> version does carry an implementation cost,
2857 but single-field existentially quantified constructors aren't much
2858 use. So the simple restriction (no existential stuff on <Literal>newtype</Literal>)
2859 stands, unless there are convincing reasons to change it.
2867 You can't use <Literal>deriving</Literal> to define instances of a
2868 data type with existentially quantified data constructors.
2870 Reason: in most cases it would not make sense. For example:#
2873 data T = forall a. MkT [a] deriving( Eq )
2876 To derive <Literal>Eq</Literal> in the standard way we would need to have equality
2877 between the single component of two <Function>MkT</Function> constructors:
2881 (MkT a) == (MkT b) = ???
2884 But <VarName>a</VarName> and <VarName>b</VarName> have distinct types, and so can't be compared.
2885 It's just about possible to imagine examples in which the derived instance
2886 would make sense, but it seems altogether simpler simply to prohibit such
2887 declarations. Define your own instances!
2899 <Sect1 id="sec-assertions">
2901 <IndexTerm><Primary>Assertions</Primary></IndexTerm>
2905 If you want to make use of assertions in your standard Haskell code, you
2906 could define a function like the following:
2912 assert :: Bool -> a -> a
2913 assert False x = error "assertion failed!"
2920 which works, but gives you back a less than useful error message --
2921 an assertion failed, but which and where?
2925 One way out is to define an extended <Function>assert</Function> function which also
2926 takes a descriptive string to include in the error message and
2927 perhaps combine this with the use of a pre-processor which inserts
2928 the source location where <Function>assert</Function> was used.
2932 Ghc offers a helping hand here, doing all of this for you. For every
2933 use of <Function>assert</Function> in the user's source:
2939 kelvinToC :: Double -> Double
2940 kelvinToC k = assert (k >= 0.0) (k+273.15)
2946 Ghc will rewrite this to also include the source location where the
2953 assert pred val ==> assertError "Main.hs|15" pred val
2959 The rewrite is only performed by the compiler when it spots
2960 applications of <Function>Exception.assert</Function>, so you can still define and
2961 use your own versions of <Function>assert</Function>, should you so wish. If not,
2962 import <Literal>Exception</Literal> to make use <Function>assert</Function> in your code.
2966 To have the compiler ignore uses of assert, use the compiler option
2967 <Option>-fignore-asserts</Option>. <IndexTerm><Primary>-fignore-asserts option</Primary></IndexTerm> That is,
2968 expressions of the form <Literal>assert pred e</Literal> will be rewritten to <Literal>e</Literal>.
2972 Assertion failures can be caught, see the documentation for the
2973 <literal>Exception</literal> library (<xref linkend="sec-Exception">)
2979 <Sect1 id="scoped-type-variables">
2980 <Title>Scoped Type Variables
2984 A <Emphasis>pattern type signature</Emphasis> can introduce a <Emphasis>scoped type
2985 variable</Emphasis>. For example
2991 f (xs::[a]) = ys ++ ys
3000 The pattern <Literal>(xs::[a])</Literal> includes a type signature for <VarName>xs</VarName>.
3001 This brings the type variable <Literal>a</Literal> into scope; it scopes over
3002 all the patterns and right hand sides for this equation for <Function>f</Function>.
3003 In particular, it is in scope at the type signature for <VarName>y</VarName>.
3007 At ordinary type signatures, such as that for <VarName>ys</VarName>, any type variables
3008 mentioned in the type signature <Emphasis>that are not in scope</Emphasis> are
3009 implicitly universally quantified. (If there are no type variables in
3010 scope, all type variables mentioned in the signature are universally
3011 quantified, which is just as in Haskell 98.) In this case, since <VarName>a</VarName>
3012 is in scope, it is not universally quantified, so the type of <VarName>ys</VarName> is
3013 the same as that of <VarName>xs</VarName>. In Haskell 98 it is not possible to declare
3014 a type for <VarName>ys</VarName>; a major benefit of scoped type variables is that
3015 it becomes possible to do so.
3019 Scoped type variables are implemented in both GHC and Hugs. Where the
3020 implementations differ from the specification below, those differences
3025 So much for the basic idea. Here are the details.
3029 <Title>Scope and implicit quantification</Title>
3037 All the type variables mentioned in the patterns for a single
3038 function definition equation, that are not already in scope,
3039 are brought into scope by the patterns. We describe this set as
3040 the <Emphasis>type variables bound by the equation</Emphasis>.
3047 The type variables thus brought into scope may be mentioned
3048 in ordinary type signatures or pattern type signatures anywhere within
3056 In ordinary type signatures, any type variable mentioned in the
3057 signature that is in scope is <Emphasis>not</Emphasis> universally quantified.
3064 Ordinary type signatures do not bring any new type variables
3065 into scope (except in the type signature itself!). So this is illegal:
3074 It's illegal because <VarName>a</VarName> is not in scope in the body of <Function>f</Function>,
3075 so the ordinary signature <Literal>x::a</Literal> is equivalent to <Literal>x::forall a.a</Literal>;
3076 and that is an incorrect typing.
3083 There is no implicit universal quantification on pattern type
3084 signatures, nor may one write an explicit <Literal>forall</Literal> type in a pattern
3085 type signature. The pattern type signature is a monotype.
3093 The type variables in the head of a <Literal>class</Literal> or <Literal>instance</Literal> declaration
3094 scope over the methods defined in the <Literal>where</Literal> part. For example:
3108 (Not implemented in Hugs yet, Dec 98).
3119 <Title>Polymorphism</Title>
3127 Pattern type signatures are completely orthogonal to ordinary, separate
3128 type signatures. The two can be used independently or together. There is
3129 no scoping associated with the names of the type variables in a separate type signature.
3134 f (xs::[b]) = reverse xs
3143 The function must be polymorphic in the type variables
3144 bound by all its equations. Operationally, the type variables bound
3145 by one equation must not:
3152 Be unified with a type (such as <Literal>Int</Literal>, or <Literal>[a]</Literal>).
3158 Be unified with a type variable free in the environment.
3164 Be unified with each other. (They may unify with the type variables
3165 bound by another equation for the same function, of course.)
3172 For example, the following all fail to type check:
3176 f (x::a) (y::b) = [x,y] -- a unifies with b
3178 g (x::a) = x + 1::Int -- a unifies with Int
3180 h x = let k (y::a) = [x,y] -- a is free in the
3181 in k x -- environment
3183 k (x::a) True = ... -- a unifies with Int
3184 k (x::Int) False = ...
3187 w (x::a) = x -- a unifies with [b]
3196 The pattern-bound type variable may, however, be constrained
3197 by the context of the principal type, thus:
3201 f (x::a) (y::a) = x+y*2
3205 gets the inferred type: <Literal>forall a. Num a => a -> a -> a</Literal>.
3216 <Title>Result type signatures</Title>
3224 The result type of a function can be given a signature,
3229 f (x::a) :: [a] = [x,x,x]
3233 The final <Literal>:: [a]</Literal> after all the patterns gives a signature to the
3234 result type. Sometimes this is the only way of naming the type variable
3239 f :: Int -> [a] -> [a]
3240 f n :: ([a] -> [a]) = let g (x::a, y::a) = (y,x)
3241 in \xs -> map g (reverse xs `zip` xs)
3253 Result type signatures are not yet implemented in Hugs.
3259 <Title>Pattern signatures on other constructs</Title>
3267 A pattern type signature can be on an arbitrary sub-pattern, not
3272 f ((x,y)::(a,b)) = (y,x) :: (b,a)
3281 Pattern type signatures, including the result part, can be used
3282 in lambda abstractions:
3286 (\ (x::a, y) :: a -> x)
3290 Type variables bound by these patterns must be polymorphic in
3291 the sense defined above.
3296 f1 (x::c) = f1 x -- ok
3297 f2 = \(x::c) -> f2 x -- not ok
3301 Here, <Function>f1</Function> is OK, but <Function>f2</Function> is not, because <VarName>c</VarName> gets unified
3302 with a type variable free in the environment, in this
3303 case, the type of <Function>f2</Function>, which is in the environment when
3304 the lambda abstraction is checked.
3311 Pattern type signatures, including the result part, can be used
3312 in <Literal>case</Literal> expressions:
3316 case e of { (x::a, y) :: a -> x }
3320 The pattern-bound type variables must, as usual,
3321 be polymorphic in the following sense: each case alternative,
3322 considered as a lambda abstraction, must be polymorphic.
3327 case (True,False) of { (x::a, y) -> x }
3331 Even though the context is that of a pair of booleans,
3332 the alternative itself is polymorphic. Of course, it is
3337 case (True,False) of { (x::Bool, y) -> x }
3346 To avoid ambiguity, the type after the “<Literal>::</Literal>” in a result
3347 pattern signature on a lambda or <Literal>case</Literal> must be atomic (i.e. a single
3348 token or a parenthesised type of some sort). To see why,
3349 consider how one would parse this:
3362 Pattern type signatures that bind new type variables
3363 may not be used in pattern bindings at all.
3368 f x = let (y, z::a) = x in ...
3372 But these are OK, because they do not bind fresh type variables:
3376 f1 x = let (y, z::Int) = x in ...
3377 f2 (x::(Int,a)) = let (y, z::a) = x in ...
3381 However a single variable is considered a degenerate function binding,
3382 rather than a degerate pattern binding, so this is permitted, even
3383 though it binds a type variable:
3387 f :: (b->b) = \(x::b) -> x
3396 Such degnerate function bindings do not fall under the monomorphism
3403 g :: a -> a -> Bool = \x y. x==y
3409 Here <Function>g</Function> has type <Literal>forall a. Eq a => a -> a -> Bool</Literal>, just as if
3410 <Function>g</Function> had a separate type signature. Lacking a type signature, <Function>g</Function>
3411 would get a monomorphic type.
3417 <Title>Existentials</Title>
3425 Pattern type signatures can bind existential type variables.
3430 data T = forall a. MkT [a]
3433 f (MkT [t::a]) = MkT t3
3450 <Sect1 id="pragmas">
3455 GHC supports several pragmas, or instructions to the compiler placed
3456 in the source code. Pragmas don't affect the meaning of the program,
3457 but they might affect the efficiency of the generated code.
3460 <Sect2 id="inline-pragma">
3461 <Title>INLINE pragma
3463 <IndexTerm><Primary>INLINE pragma</Primary></IndexTerm>
3464 <IndexTerm><Primary>pragma, INLINE</Primary></IndexTerm></Title>
3467 GHC (with <Option>-O</Option>, as always) tries to inline (or “unfold”)
3468 functions/values that are “small enough,” thus avoiding the call
3469 overhead and possibly exposing other more-wonderful optimisations.
3473 You will probably see these unfoldings (in Core syntax) in your
3478 Normally, if GHC decides a function is “too expensive” to inline, it
3479 will not do so, nor will it export that unfolding for other modules to
3484 The sledgehammer you can bring to bear is the
3485 <Literal>INLINE</Literal><IndexTerm><Primary>INLINE pragma</Primary></IndexTerm> pragma, used thusly:
3488 key_function :: Int -> String -> (Bool, Double)
3490 #ifdef __GLASGOW_HASKELL__
3491 {-# INLINE key_function #-}
3495 (You don't need to do the C pre-processor carry-on unless you're going
3496 to stick the code through HBC—it doesn't like <Literal>INLINE</Literal> pragmas.)
3500 The major effect of an <Literal>INLINE</Literal> pragma is to declare a function's
3501 “cost” to be very low. The normal unfolding machinery will then be
3502 very keen to inline it.
3506 An <Literal>INLINE</Literal> pragma for a function can be put anywhere its type
3507 signature could be put.
3511 <Literal>INLINE</Literal> pragmas are a particularly good idea for the
3512 <Literal>then</Literal>/<Literal>return</Literal> (or <Literal>bind</Literal>/<Literal>unit</Literal>) functions in a monad.
3513 For example, in GHC's own <Literal>UniqueSupply</Literal> monad code, we have:
3516 #ifdef __GLASGOW_HASKELL__
3517 {-# INLINE thenUs #-}
3518 {-# INLINE returnUs #-}
3526 <Sect2 id="noinline-pragma">
3527 <Title>NOINLINE pragma
3531 <IndexTerm><Primary>NOINLINE pragma</Primary></IndexTerm>
3532 <IndexTerm><Primary>pragma, NOINLINE</Primary></IndexTerm>
3536 The <Literal>NOINLINE</Literal> pragma does exactly what you'd expect: it stops the
3537 named function from being inlined by the compiler. You shouldn't ever
3538 need to do this, unless you're very cautious about code size.
3543 <Sect2 id="specialize-pragma">
3544 <Title>SPECIALIZE pragma
3548 <IndexTerm><Primary>SPECIALIZE pragma</Primary></IndexTerm>
3549 <IndexTerm><Primary>pragma, SPECIALIZE</Primary></IndexTerm>
3550 <IndexTerm><Primary>overloading, death to</Primary></IndexTerm>
3554 (UK spelling also accepted.) For key overloaded functions, you can
3555 create extra versions (NB: more code space) specialised to particular
3556 types. Thus, if you have an overloaded function:
3562 hammeredLookup :: Ord key => [(key, value)] -> key -> value
3568 If it is heavily used on lists with <Literal>Widget</Literal> keys, you could
3569 specialise it as follows:
3572 {-# SPECIALIZE hammeredLookup :: [(Widget, value)] -> Widget -> value #-}
3578 To get very fancy, you can also specify a named function to use for
3579 the specialised value, by adding <Literal>= blah</Literal>, as in:
3582 {-# SPECIALIZE hammeredLookup :: ...as before... = blah #-}
3585 It's <Emphasis>Your Responsibility</Emphasis> to make sure that <Function>blah</Function> really
3586 behaves as a specialised version of <Function>hammeredLookup</Function>!!!
3590 NOTE: the <Literal>=blah</Literal> feature isn't implemented in GHC 4.xx.
3594 An example in which the <Literal>= blah</Literal> form will Win Big:
3597 toDouble :: Real a => a -> Double
3598 toDouble = fromRational . toRational
3600 {-# SPECIALIZE toDouble :: Int -> Double = i2d #-}
3601 i2d (I# i) = D# (int2Double# i) -- uses Glasgow prim-op directly
3604 The <Function>i2d</Function> function is virtually one machine instruction; the
3605 default conversion—via an intermediate <Literal>Rational</Literal>—is obscenely
3606 expensive by comparison.
3610 By using the US spelling, your <Literal>SPECIALIZE</Literal> pragma will work with
3611 HBC, too. Note that HBC doesn't support the <Literal>= blah</Literal> form.
3615 A <Literal>SPECIALIZE</Literal> pragma for a function can be put anywhere its type
3616 signature could be put.
3621 <Sect2 id="specialize-instance-pragma">
3622 <Title>SPECIALIZE instance pragma
3626 <IndexTerm><Primary>SPECIALIZE pragma</Primary></IndexTerm>
3627 <IndexTerm><Primary>overloading, death to</Primary></IndexTerm>
3628 Same idea, except for instance declarations. For example:
3631 instance (Eq a) => Eq (Foo a) where { ... usual stuff ... }
3633 {-# SPECIALIZE instance Eq (Foo [(Int, Bar)] #-}
3636 Compatible with HBC, by the way.
3641 <Sect2 id="line-pragma">
3646 <IndexTerm><Primary>LINE pragma</Primary></IndexTerm>
3647 <IndexTerm><Primary>pragma, LINE</Primary></IndexTerm>
3651 This pragma is similar to C's <Literal>#line</Literal> pragma, and is mainly for use in
3652 automatically generated Haskell code. It lets you specify the line
3653 number and filename of the original code; for example
3659 {-# LINE 42 "Foo.vhs" #-}
3665 if you'd generated the current file from something called <Filename>Foo.vhs</Filename>
3666 and this line corresponds to line 42 in the original. GHC will adjust
3667 its error messages to refer to the line/file named in the <Literal>LINE</Literal>
3674 <Title>RULES pragma</Title>
3677 The RULES pragma lets you specify rewrite rules. It is described in
3678 <XRef LinkEnd="rewrite-rules">.
3685 <Sect1 id="rewrite-rules">
3686 <Title>Rewrite rules
3688 <IndexTerm><Primary>RULES pagma</Primary></IndexTerm>
3689 <IndexTerm><Primary>pragma, RULES</Primary></IndexTerm>
3690 <IndexTerm><Primary>rewrite rules</Primary></IndexTerm></Title>
3693 The programmer can specify rewrite rules as part of the source program
3694 (in a pragma). GHC applies these rewrite rules wherever it can.
3702 "map/map" forall f g xs. map f (map g xs) = map (f.g) xs
3709 <Title>Syntax</Title>
3712 From a syntactic point of view:
3718 Each rule has a name, enclosed in double quotes. The name itself has
3719 no significance at all. It is only used when reporting how many times the rule fired.
3725 There may be zero or more rules in a <Literal>RULES</Literal> pragma.
3731 Layout applies in a <Literal>RULES</Literal> pragma. Currently no new indentation level
3732 is set, so you must lay out your rules starting in the same column as the
3733 enclosing definitions.
3739 Each variable mentioned in a rule must either be in scope (e.g. <Function>map</Function>),
3740 or bound by the <Literal>forall</Literal> (e.g. <Function>f</Function>, <Function>g</Function>, <Function>xs</Function>). The variables bound by
3741 the <Literal>forall</Literal> are called the <Emphasis>pattern</Emphasis> variables. They are separated
3742 by spaces, just like in a type <Literal>forall</Literal>.
3748 A pattern variable may optionally have a type signature.
3749 If the type of the pattern variable is polymorphic, it <Emphasis>must</Emphasis> have a type signature.
3750 For example, here is the <Literal>foldr/build</Literal> rule:
3753 "fold/build" forall k z (g::forall b. (a->b->b) -> b -> b) .
3754 foldr k z (build g) = g k z
3757 Since <Function>g</Function> has a polymorphic type, it must have a type signature.
3764 The left hand side of a rule must consist of a top-level variable applied
3765 to arbitrary expressions. For example, this is <Emphasis>not</Emphasis> OK:
3768 "wrong1" forall e1 e2. case True of { True -> e1; False -> e2 } = e1
3769 "wrong2" forall f. f True = True
3772 In <Literal>"wrong1"</Literal>, the LHS is not an application; in <Literal>"wrong2"</Literal>, the LHS has a pattern variable
3779 A rule does not need to be in the same module as (any of) the
3780 variables it mentions, though of course they need to be in scope.
3786 Rules are automatically exported from a module, just as instance declarations are.
3797 <Title>Semantics</Title>
3800 From a semantic point of view:
3806 Rules are only applied if you use the <Option>-O</Option> flag.
3812 Rules are regarded as left-to-right rewrite rules.
3813 When GHC finds an expression that is a substitution instance of the LHS
3814 of a rule, it replaces the expression by the (appropriately-substituted) RHS.
3815 By "a substitution instance" we mean that the LHS can be made equal to the
3816 expression by substituting for the pattern variables.
3823 The LHS and RHS of a rule are typechecked, and must have the
3831 GHC makes absolutely no attempt to verify that the LHS and RHS
3832 of a rule have the same meaning. That is undecideable in general, and
3833 infeasible in most interesting cases. The responsibility is entirely the programmer's!
3840 GHC makes no attempt to make sure that the rules are confluent or
3841 terminating. For example:
3844 "loop" forall x,y. f x y = f y x
3847 This rule will cause the compiler to go into an infinite loop.
3854 If more than one rule matches a call, GHC will choose one arbitrarily to apply.
3860 GHC currently uses a very simple, syntactic, matching algorithm
3861 for matching a rule LHS with an expression. It seeks a substitution
3862 which makes the LHS and expression syntactically equal modulo alpha
3863 conversion. The pattern (rule), but not the expression, is eta-expanded if
3864 necessary. (Eta-expanding the epression can lead to laziness bugs.)
3865 But not beta conversion (that's called higher-order matching).
3869 Matching is carried out on GHC's intermediate language, which includes
3870 type abstractions and applications. So a rule only matches if the
3871 types match too. See <XRef LinkEnd="rule-spec"> below.
3877 GHC keeps trying to apply the rules as it optimises the program.
3878 For example, consider:
3887 The expression <Literal>s (t xs)</Literal> does not match the rule <Literal>"map/map"</Literal>, but GHC
3888 will substitute for <VarName>s</VarName> and <VarName>t</VarName>, giving an expression which does match.
3889 If <VarName>s</VarName> or <VarName>t</VarName> was (a) used more than once, and (b) large or a redex, then it would
3890 not be substituted, and the rule would not fire.
3897 In the earlier phases of compilation, GHC inlines <Emphasis>nothing
3898 that appears on the LHS of a rule</Emphasis>, because once you have substituted
3899 for something you can't match against it (given the simple minded
3900 matching). So if you write the rule
3903 "map/map" forall f,g. map f . map g = map (f.g)
3906 this <Emphasis>won't</Emphasis> match the expression <Literal>map f (map g xs)</Literal>.
3907 It will only match something written with explicit use of ".".
3908 Well, not quite. It <Emphasis>will</Emphasis> match the expression
3914 where <Function>wibble</Function> is defined:
3917 wibble f g = map f . map g
3920 because <Function>wibble</Function> will be inlined (it's small).
3922 Later on in compilation, GHC starts inlining even things on the
3923 LHS of rules, but still leaves the rules enabled. This inlining
3924 policy is controlled by the per-simplification-pass flag <Option>-finline-phase</Option><Emphasis>n</Emphasis>.
3931 All rules are implicitly exported from the module, and are therefore
3932 in force in any module that imports the module that defined the rule, directly
3933 or indirectly. (That is, if A imports B, which imports C, then C's rules are
3934 in force when compiling A.) The situation is very similar to that for instance
3946 <Title>List fusion</Title>
3949 The RULES mechanism is used to implement fusion (deforestation) of common list functions.
3950 If a "good consumer" consumes an intermediate list constructed by a "good producer", the
3951 intermediate list should be eliminated entirely.
3955 The following are good producers:
3967 Enumerations of <Literal>Int</Literal> and <Literal>Char</Literal> (e.g. <Literal>['a'..'z']</Literal>).
3973 Explicit lists (e.g. <Literal>[True, False]</Literal>)
3979 The cons constructor (e.g <Literal>3:4:[]</Literal>)
3985 <Function>++</Function>
3991 <Function>map</Function>
3997 <Function>filter</Function>
4003 <Function>iterate</Function>, <Function>repeat</Function>
4009 <Function>zip</Function>, <Function>zipWith</Function>
4018 The following are good consumers:
4030 <Function>array</Function> (on its second argument)
4036 <Function>length</Function>
4042 <Function>++</Function> (on its first argument)
4048 <Function>map</Function>
4054 <Function>filter</Function>
4060 <Function>concat</Function>
4066 <Function>unzip</Function>, <Function>unzip2</Function>, <Function>unzip3</Function>, <Function>unzip4</Function>
4072 <Function>zip</Function>, <Function>zipWith</Function> (but on one argument only; if both are good producers, <Function>zip</Function>
4073 will fuse with one but not the other)
4079 <Function>partition</Function>
4085 <Function>head</Function>
4091 <Function>and</Function>, <Function>or</Function>, <Function>any</Function>, <Function>all</Function>
4097 <Function>sequence_</Function>
4103 <Function>msum</Function>
4109 <Function>sortBy</Function>
4118 So, for example, the following should generate no intermediate lists:
4121 array (1,10) [(i,i*i) | i <- map (+ 1) [0..9]]
4127 This list could readily be extended; if there are Prelude functions that you use
4128 a lot which are not included, please tell us.
4132 If you want to write your own good consumers or producers, look at the
4133 Prelude definitions of the above functions to see how to do so.
4138 <Sect2 id="rule-spec">
4139 <Title>Specialisation
4143 Rewrite rules can be used to get the same effect as a feature
4144 present in earlier version of GHC:
4147 {-# SPECIALIZE fromIntegral :: Int8 -> Int16 = int8ToInt16 #-}
4150 This told GHC to use <Function>int8ToInt16</Function> instead of <Function>fromIntegral</Function> whenever
4151 the latter was called with type <Literal>Int8 -> Int16</Literal>. That is, rather than
4152 specialising the original definition of <Function>fromIntegral</Function> the programmer is
4153 promising that it is safe to use <Function>int8ToInt16</Function> instead.
4157 This feature is no longer in GHC. But rewrite rules let you do the
4162 "fromIntegral/Int8/Int16" fromIntegral = int8ToInt16
4166 This slightly odd-looking rule instructs GHC to replace <Function>fromIntegral</Function>
4167 by <Function>int8ToInt16</Function> <Emphasis>whenever the types match</Emphasis>. Speaking more operationally,
4168 GHC adds the type and dictionary applications to get the typed rule
4171 forall (d1::Integral Int8) (d2::Num Int16) .
4172 fromIntegral Int8 Int16 d1 d2 = int8ToInt16
4176 this rule does not need to be in the same file as fromIntegral,
4177 unlike the <Literal>SPECIALISE</Literal> pragmas which currently do (so that they
4178 have an original definition available to specialise).
4184 <Title>Controlling what's going on</Title>
4192 Use <Option>-ddump-rules</Option> to see what transformation rules GHC is using.
4198 Use <Option>-ddump-simpl-stats</Option> to see what rules are being fired.
4199 If you add <Option>-dppr-debug</Option> you get a more detailed listing.
4205 The defintion of (say) <Function>build</Function> in <FileName>PrelBase.lhs</FileName> looks llike this:
4208 build :: forall a. (forall b. (a -> b -> b) -> b -> b) -> [a]
4209 {-# INLINE build #-}
4213 Notice the <Literal>INLINE</Literal>! That prevents <Literal>(:)</Literal> from being inlined when compiling
4214 <Literal>PrelBase</Literal>, so that an importing module will “see” the <Literal>(:)</Literal>, and can
4215 match it on the LHS of a rule. <Literal>INLINE</Literal> prevents any inlining happening
4216 in the RHS of the <Literal>INLINE</Literal> thing. I regret the delicacy of this.
4223 In <Filename>ghc/lib/std/PrelBase.lhs</Filename> look at the rules for <Function>map</Function> to
4224 see how to write rules that will do fusion and yet give an efficient
4225 program even if fusion doesn't happen. More rules in <Filename>PrelList.lhs</Filename>.
4237 <Sect1 id="generic-classes">
4238 <Title>Generic classes</Title>
4241 The ideas behind this extension are described in detail in "Derivable type classes",
4242 Ralf Hinze and Simon Peyton Jones, Haskell Workshop, Montreal Sept 2000, pp94-105.
4243 An example will give the idea:
4249 fromBin :: [Int] -> (a, [Int])
4251 toBin {| Unit |} Unit = []
4252 toBin {| a :+: b |} (Inl x) = 0 : toBin x
4253 toBin {| a :+: b |} (Inr y) = 1 : toBin y
4254 toBin {| a :*: b |} (x :*: y) = toBin x ++ toBin y
4256 fromBin {| Unit |} bs = (Unit, bs)
4257 fromBin {| a :+: b |} (0:bs) = (Inl x, bs') where (x,bs') = fromBin bs
4258 fromBin {| a :+: b |} (1:bs) = (Inr y, bs') where (y,bs') = fromBin bs
4259 fromBin {| a :*: b |} bs = (x :*: y, bs'') where (x,bs' ) = fromBin bs
4260 (y,bs'') = fromBin bs'
4263 This class declaration explains how <Literal>toBin</Literal> and <Literal>fromBin</Literal>
4264 work for arbitrary data types. They do so by giving cases for unit, product, and sum,
4265 which are defined thus:
4269 data a :+: b = Inl a | Inr b
4270 data a :*: b = a :*: b
4273 Now you can make a data type into an instance of Bin like this:
4275 instance (Bin a, Bin b) => Bin (a,b)
4276 instance Bin a => Bin [a]
4278 That is, just leave off the "where" clasuse. Of course, you can put in the
4279 where clause and over-ride whichever methods you please.
4282 <Sect2> <Title> Using generics </Title>
4285 To use generics you need to
4289 Use the <Option>-fgenerics</Option> flag.
4294 Import the module <Literal>Generics</Literal> from the <Literal>lang</Literal> package.
4295 This import brings into scope the data types <Literal>Unit</Literal>, <Literal>:*:</Literal>,
4296 and <Literal>:+:</Literal>. (You don't need this import if you don't mention these
4297 types explicitly; for example, if you are simply giving instance declarations.)
4302 <Sect2> <Title> Changes wrt the paper </Title>
4304 Note that the type constructors <Literal>:+:</Literal> and <Literal>:*:</Literal>
4305 can be written infix (indeed, you can now use
4306 any operator starting in a colon as an infix type constructor). Also note that
4307 the type constructors are not exactly as in the paper (Unit instead of 1, etc).
4308 Finally, note that the syntax of the type patterns in the class declaration
4309 uses "<Literal>{|</Literal>" and "<Literal>{|</Literal>" brackets; curly braces
4310 alone would ambiguous when they appear on right hand sides (an extension we
4311 anticipate wanting).
4315 <Sect2> <Title>Terminology and restrictions</Title>
4317 Terminology. A "generic default method" in a class declaration
4318 is one that is defined using type patterns as above.
4319 A "polymorphic default method" is a default method defined as in Haskell 98.
4320 A "generic class declaration" is a class declaration with at least one
4321 generic default method.
4329 Alas, we do not yet implement the stuff about constructor names and
4336 A generic class can have only one parameter; you can't have a generic
4337 multi-parameter class.
4343 A default method must be defined entirely using type patterns, or entirely
4344 without. So this is illegal:
4347 op :: a -> (a, Bool)
4348 op {| Unit |} Unit = (Unit, True)
4351 However it is perfectly OK for some methods of a generic class to have
4352 generic default methods and others to have polymorphic default methods.
4358 The type variable(s) in the type pattern for a generic method declaration
4359 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:
4363 op {| p :*: q |} (x :*: y) = op (x :: p)
4371 The type patterns in a generic default method must take one of the forms:
4377 where "a" and "b" are type variables. Furthermore, all the type patterns for
4378 a single type constructor (<Literal>:*:</Literal>, say) must be identical; they
4379 must use the same type variables. So this is illegal:
4383 op {| a :+: b |} (Inl x) = True
4384 op {| p :+: q |} (Inr y) = False
4386 The type patterns must be identical, even in equations for different methods of the class.
4387 So this too is illegal:
4391 op {| a :*: b |} (Inl x) = True
4394 op {| p :*: q |} (Inr y) = False
4396 (The reason for this restriction is that we gather all the equations for a particular type consructor
4397 into a single generic instance declaration.)
4403 A generic method declaration must give a case for each of the three type constructors.
4409 In an instance declaration for a generic class, the idea is that the compiler
4410 will fill in the methods for you, based on the generic templates. However it can only
4415 The instance type is simple (a type constructor applied to type variables, as in Haskell 98).
4420 No constructor of the instance type has unboxed fields.
4424 (Of course, these things can only arise if you are already using GHC extensions.)
4425 However, you can still give an instance declarations for types which break these rules,
4426 provided you give explicit code to override any generic default methods.
4434 The option <Option>-ddump-deriv</Option> dumps incomprehensible stuff giving details of
4435 what the compiler does with generic declarations.
4440 <Sect2> <Title> Another example </Title>
4442 Just to finish with, here's another example I rather like:
4446 nCons {| Unit |} _ = 1
4447 nCons {| a :*: b |} _ = 1
4448 nCons {| a :+: b |} _ = nCons (bot::a) + nCons (bot::b)
4451 tag {| Unit |} _ = 1
4452 tag {| a :*: b |} _ = 1
4453 tag {| a :+: b |} (Inl x) = tag x
4454 tag {| a :+: b |} (Inr y) = nCons (bot::a) + tag y
4461 ;;; Local Variables: ***
4463 ;;; sgml-parent-document: ("users_guide.sgml" "book" "chapter" "sect1") ***