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="options-language">
143 <title>Language options</title>
145 <indexterm><primary>language</primary><secondary>option</secondary>
147 <indexterm><primary>options</primary><secondary>language</secondary>
149 <indexterm><primary>extensions</primary><secondary>options controlling</secondary>
152 <para> These flags control what variation of the language are
153 permitted. Leaving out all of them gives you standard Haskell
159 <term><option>-fglasgow-exts</option>:</term>
160 <indexterm><primary><option>-fglasgow-exts</option></primary></indexterm>
162 <para>This simultaneously enables all of the extensions to
163 Haskell 98 described in <xref
164 linkend="ghc-language-features">, except where otherwise
170 <term><option>-fno-monomorphism-restriction</option>:</term>
171 <indexterm><primary><option>-fno-monomorphism-restriction</option></primary></indexterm>
173 <para> Switch off the Haskell 98 monomorphism restriction.
174 Independent of the <Option>-fglasgow-exts</Option>
180 <term><option>-fallow-overlapping-instances</option></term>
181 <term><option>-fallow-undecidable-instances</option></term>
182 <term><option>-fcontext-stack</option></term>
183 <indexterm><primary><option>-fallow-overlapping-instances</option></primary></indexterm>
184 <indexterm><primary><option>-fallow-undecidable-instances</option></primary></indexterm>
185 <indexterm><primary><option>-fcontext-stack</option></primary></indexterm>
187 <para> See <XRef LinkEnd="instance-decls">. Only relevant
188 if you also use <option>-fglasgow-exts</option>.</para>
193 <term><option>-fignore-asserts</option>:</term>
194 <indexterm><primary><option>-fignore-asserts</option></primary></indexterm>
196 <para>See <XRef LinkEnd="sec-assertions">. Only relevant if
197 you also use <option>-fglasgow-exts</option>.</Para>
202 <term><option>-finline-phase</option></term>
203 <indexterm><primary><option>-finline-phase</option></primary></indexterm>
205 <para>See <XRef LinkEnd="rewrite-rules">. Only relevant if
206 you also use <Option>-fglasgow-exts</Option>.</para>
211 <term><option>-fgenerics</option></term>
212 <indexterm><primary><option>-fgenerics</option></primary></indexterm>
214 <para>See <XRef LinkEnd="generic-classes">. Independent of
215 <Option>-fglasgow-exts</Option>.</para>
220 <term><option>-fno-implicit-prelude</option></term>
222 <para><indexterm><primary>-fno-implicit-prelude
223 option</primary></indexterm> GHC normally imports
224 <filename>Prelude.hi</filename> files for you. If you'd
225 rather it didn't, then give it a
226 <option>-fno-implicit-prelude</option> option. The idea
227 is that you can then import a Prelude of your own. (But
228 don't call it <literal>Prelude</literal>; the Haskell
229 module namespace is flat, and you must not conflict with
230 any Prelude module.)</para>
232 <para>Even though you have not imported the Prelude, all
233 the built-in syntax still refers to the built-in Haskell
234 Prelude types and values, as specified by the Haskell
235 Report. For example, the type <literal>[Int]</literal>
236 still means <literal>Prelude.[] Int</literal>; tuples
237 continue to refer to the standard Prelude tuples; the
238 translation for list comprehensions continues to use
239 <literal>Prelude.map</literal> etc.</para>
241 <para> With one group of exceptions! You may want to
242 define your own numeric class hierarchy. It completely
243 defeats that purpose if the literal "1" means
244 "<literal>Prelude.fromInteger 1</literal>", which is what
245 the Haskell Report specifies. So the
246 <option>-fno-implicit-prelude</option> flag causes the
247 following pieces of built-in syntax to refer to whatever
248 is in scope, not the Prelude versions:</para>
252 <para>Integer and fractional literals mean
253 "<literal>fromInteger 1</literal>" and
254 "<literal>fromRational 3.2</literal>", not the
255 Prelude-qualified versions; both in expressions and in
260 <para>Negation (e.g. "<literal>- (f x)</literal>")
261 means "<literal>negate (f x)</literal>" (not
262 <literal>Prelude.negate</literal>).</para>
266 <para>In an n+k pattern, the standard Prelude
267 <literal>Ord</literal> class is used for comparison,
268 but the necessary subtraction uses whatever
269 "<literal>(-)</literal>" is in scope (not
270 "<literal>Prelude.(-)</literal>").</para>
280 <Sect1 id="primitives">
281 <Title>Unboxed types and primitive operations
283 <IndexTerm><Primary>PrelGHC module</Primary></IndexTerm>
286 This module defines all the types which are primitive in Glasgow
287 Haskell, and the operations provided for them.
290 <Sect2 id="glasgow-unboxed">
295 <IndexTerm><Primary>Unboxed types (Glasgow extension)</Primary></IndexTerm>
298 <para>Most types in GHC are <firstterm>boxed</firstterm>, which means
299 that values of that type are represented by a pointer to a heap
300 object. The representation of a Haskell <literal>Int</literal>, for
301 example, is a two-word heap object. An <firstterm>unboxed</firstterm>
302 type, however, is represented by the value itself, no pointers or heap
303 allocation are involved.
307 Unboxed types correspond to the “raw machine” types you
308 would use in C: <Literal>Int#</Literal> (long int),
309 <Literal>Double#</Literal> (double), <Literal>Addr#</Literal>
310 (void *), etc. The <Emphasis>primitive operations</Emphasis>
311 (PrimOps) on these types are what you might expect; e.g.,
312 <Literal>(+#)</Literal> is addition on
313 <Literal>Int#</Literal>s, and is the machine-addition that we all
314 know and love—usually one instruction.
318 Primitive (unboxed) types cannot be defined in Haskell, and are
319 therefore built into the language and compiler. Primitive types are
320 always unlifted; that is, a value of a primitive type cannot be
321 bottom. We use the convention that primitive types, values, and
322 operations have a <Literal>#</Literal> suffix.
326 Primitive values are often represented by a simple bit-pattern, such
327 as <Literal>Int#</Literal>, <Literal>Float#</Literal>,
328 <Literal>Double#</Literal>. But this is not necessarily the case:
329 a primitive value might be represented by a pointer to a
330 heap-allocated object. Examples include
331 <Literal>Array#</Literal>, the type of primitive arrays. A
332 primitive array is heap-allocated because it is too big a value to fit
333 in a register, and would be too expensive to copy around; in a sense,
334 it is accidental that it is represented by a pointer. If a pointer
335 represents a primitive value, then it really does point to that value:
336 no unevaluated thunks, no indirections…nothing can be at the
337 other end of the pointer than the primitive value.
341 There are some restrictions on the use of primitive types, the main
342 one being that you can't pass a primitive value to a polymorphic
343 function or store one in a polymorphic data type. This rules out
344 things like <Literal>[Int#]</Literal> (i.e. lists of primitive
345 integers). The reason for this restriction is that polymorphic
346 arguments and constructor fields are assumed to be pointers: if an
347 unboxed integer is stored in one of these, the garbage collector would
348 attempt to follow it, leading to unpredictable space leaks. Or a
349 <Function>seq</Function> operation on the polymorphic component may
350 attempt to dereference the pointer, with disastrous results. Even
351 worse, the unboxed value might be larger than a pointer
352 (<Literal>Double#</Literal> for instance).
356 Nevertheless, A numerically-intensive program using unboxed types can
357 go a <Emphasis>lot</Emphasis> faster than its “standard”
358 counterpart—we saw a threefold speedup on one example.
363 <Sect2 id="unboxed-tuples">
364 <Title>Unboxed Tuples
368 Unboxed tuples aren't really exported by <Literal>PrelGHC</Literal>,
369 they're available by default with <Option>-fglasgow-exts</Option>. An
370 unboxed tuple looks like this:
382 where <Literal>e_1..e_n</Literal> are expressions of any
383 type (primitive or non-primitive). The type of an unboxed tuple looks
388 Unboxed tuples are used for functions that need to return multiple
389 values, but they avoid the heap allocation normally associated with
390 using fully-fledged tuples. When an unboxed tuple is returned, the
391 components are put directly into registers or on the stack; the
392 unboxed tuple itself does not have a composite representation. Many
393 of the primitive operations listed in this section return unboxed
398 There are some pretty stringent restrictions on the use of unboxed tuples:
407 Unboxed tuple types are subject to the same restrictions as
408 other unboxed types; i.e. they may not be stored in polymorphic data
409 structures or passed to polymorphic functions.
416 Unboxed tuples may only be constructed as the direct result of
417 a function, and may only be deconstructed with a <Literal>case</Literal> expression.
418 eg. the following are valid:
422 f x y = (# x+1, y-1 #)
423 g x = case f x x of { (# a, b #) -> a + b }
427 but the following are invalid:
441 No variable can have an unboxed tuple type. This is illegal:
445 f :: (# Int, Int #) -> (# Int, Int #)
450 because <VarName>x</VarName> has an unboxed tuple type.
460 Note: we may relax some of these restrictions in the future.
464 The <Literal>IO</Literal> and <Literal>ST</Literal> monads use unboxed
465 tuples to avoid unnecessary allocation during sequences of operations.
471 <Title>Character and numeric types</Title>
473 <IndexTerm><Primary>character types, primitive</Primary></IndexTerm>
474 <IndexTerm><Primary>numeric types, primitive</Primary></IndexTerm>
475 <IndexTerm><Primary>integer types, primitive</Primary></IndexTerm>
476 <IndexTerm><Primary>floating point types, primitive</Primary></IndexTerm>
478 There are the following obvious primitive types:
492 <IndexTerm><Primary><literal>Char#</literal></Primary></IndexTerm>
493 <IndexTerm><Primary><literal>Int#</literal></Primary></IndexTerm>
494 <IndexTerm><Primary><literal>Word#</literal></Primary></IndexTerm>
495 <IndexTerm><Primary><literal>Addr#</literal></Primary></IndexTerm>
496 <IndexTerm><Primary><literal>Float#</literal></Primary></IndexTerm>
497 <IndexTerm><Primary><literal>Double#</literal></Primary></IndexTerm>
498 <IndexTerm><Primary><literal>Int64#</literal></Primary></IndexTerm>
499 <IndexTerm><Primary><literal>Word64#</literal></Primary></IndexTerm>
502 If you really want to know their exact equivalents in C, see
503 <Filename>ghc/includes/StgTypes.h</Filename> in the GHC source tree.
507 Literals for these types may be written as follows:
516 'a'# a Char#; for weird characters, use e.g. '\o<octal>'#
517 "a"# an Addr# (a `char *'); only characters '\0'..'\255' allowed
520 <IndexTerm><Primary>literals, primitive</Primary></IndexTerm>
521 <IndexTerm><Primary>constants, primitive</Primary></IndexTerm>
522 <IndexTerm><Primary>numbers, primitive</Primary></IndexTerm>
528 <Title>Comparison operations</Title>
531 <IndexTerm><Primary>comparisons, primitive</Primary></IndexTerm>
532 <IndexTerm><Primary>operators, comparison</Primary></IndexTerm>
538 {>,>=,==,/=,<,<=}# :: Int# -> Int# -> Bool
540 {gt,ge,eq,ne,lt,le}Char# :: Char# -> Char# -> Bool
541 -- ditto for Word# and Addr#
544 <IndexTerm><Primary><literal>>#</literal></Primary></IndexTerm>
545 <IndexTerm><Primary><literal>>=#</literal></Primary></IndexTerm>
546 <IndexTerm><Primary><literal>==#</literal></Primary></IndexTerm>
547 <IndexTerm><Primary><literal>/=#</literal></Primary></IndexTerm>
548 <IndexTerm><Primary><literal><#</literal></Primary></IndexTerm>
549 <IndexTerm><Primary><literal><=#</literal></Primary></IndexTerm>
550 <IndexTerm><Primary><literal>gt{Char,Word,Addr}#</literal></Primary></IndexTerm>
551 <IndexTerm><Primary><literal>ge{Char,Word,Addr}#</literal></Primary></IndexTerm>
552 <IndexTerm><Primary><literal>eq{Char,Word,Addr}#</literal></Primary></IndexTerm>
553 <IndexTerm><Primary><literal>ne{Char,Word,Addr}#</literal></Primary></IndexTerm>
554 <IndexTerm><Primary><literal>lt{Char,Word,Addr}#</literal></Primary></IndexTerm>
555 <IndexTerm><Primary><literal>le{Char,Word,Addr}#</literal></Primary></IndexTerm>
561 <Title>Primitive-character operations</Title>
564 <IndexTerm><Primary>characters, primitive operations</Primary></IndexTerm>
565 <IndexTerm><Primary>operators, primitive character</Primary></IndexTerm>
571 ord# :: Char# -> Int#
572 chr# :: Int# -> Char#
575 <IndexTerm><Primary><literal>ord#</literal></Primary></IndexTerm>
576 <IndexTerm><Primary><literal>chr#</literal></Primary></IndexTerm>
582 <Title>Primitive-<Literal>Int</Literal> operations</Title>
585 <IndexTerm><Primary>integers, primitive operations</Primary></IndexTerm>
586 <IndexTerm><Primary>operators, primitive integer</Primary></IndexTerm>
592 {+,-,*,quotInt,remInt,gcdInt}# :: Int# -> Int# -> Int#
593 negateInt# :: Int# -> Int#
595 iShiftL#, iShiftRA#, iShiftRL# :: Int# -> Int# -> Int#
596 -- shift left, right arithmetic, right logical
598 addIntC#, subIntC#, mulIntC# :: Int# -> Int# -> (# Int#, Int# #)
599 -- add, subtract, multiply with carry
602 <IndexTerm><Primary><literal>+#</literal></Primary></IndexTerm>
603 <IndexTerm><Primary><literal>-#</literal></Primary></IndexTerm>
604 <IndexTerm><Primary><literal>*#</literal></Primary></IndexTerm>
605 <IndexTerm><Primary><literal>quotInt#</literal></Primary></IndexTerm>
606 <IndexTerm><Primary><literal>remInt#</literal></Primary></IndexTerm>
607 <IndexTerm><Primary><literal>gcdInt#</literal></Primary></IndexTerm>
608 <IndexTerm><Primary><literal>iShiftL#</literal></Primary></IndexTerm>
609 <IndexTerm><Primary><literal>iShiftRA#</literal></Primary></IndexTerm>
610 <IndexTerm><Primary><literal>iShiftRL#</literal></Primary></IndexTerm>
611 <IndexTerm><Primary><literal>addIntC#</literal></Primary></IndexTerm>
612 <IndexTerm><Primary><literal>subIntC#</literal></Primary></IndexTerm>
613 <IndexTerm><Primary><literal>mulIntC#</literal></Primary></IndexTerm>
614 <IndexTerm><Primary>shift operations, integer</Primary></IndexTerm>
618 <Emphasis>Note:</Emphasis> No error/overflow checking!
624 <Title>Primitive-<Literal>Double</Literal> and <Literal>Float</Literal> operations</Title>
627 <IndexTerm><Primary>floating point numbers, primitive</Primary></IndexTerm>
628 <IndexTerm><Primary>operators, primitive floating point</Primary></IndexTerm>
634 {+,-,*,/}## :: Double# -> Double# -> Double#
635 {<,<=,==,/=,>=,>}## :: Double# -> Double# -> Bool
636 negateDouble# :: Double# -> Double#
637 double2Int# :: Double# -> Int#
638 int2Double# :: Int# -> Double#
640 {plus,minux,times,divide}Float# :: Float# -> Float# -> Float#
641 {gt,ge,eq,ne,lt,le}Float# :: Float# -> Float# -> Bool
642 negateFloat# :: Float# -> Float#
643 float2Int# :: Float# -> Int#
644 int2Float# :: Int# -> Float#
650 <IndexTerm><Primary><literal>+##</literal></Primary></IndexTerm>
651 <IndexTerm><Primary><literal>-##</literal></Primary></IndexTerm>
652 <IndexTerm><Primary><literal>*##</literal></Primary></IndexTerm>
653 <IndexTerm><Primary><literal>/##</literal></Primary></IndexTerm>
654 <IndexTerm><Primary><literal><##</literal></Primary></IndexTerm>
655 <IndexTerm><Primary><literal><=##</literal></Primary></IndexTerm>
656 <IndexTerm><Primary><literal>==##</literal></Primary></IndexTerm>
657 <IndexTerm><Primary><literal>=/##</literal></Primary></IndexTerm>
658 <IndexTerm><Primary><literal>>=##</literal></Primary></IndexTerm>
659 <IndexTerm><Primary><literal>>##</literal></Primary></IndexTerm>
660 <IndexTerm><Primary><literal>negateDouble#</literal></Primary></IndexTerm>
661 <IndexTerm><Primary><literal>double2Int#</literal></Primary></IndexTerm>
662 <IndexTerm><Primary><literal>int2Double#</literal></Primary></IndexTerm>
666 <IndexTerm><Primary><literal>plusFloat#</literal></Primary></IndexTerm>
667 <IndexTerm><Primary><literal>minusFloat#</literal></Primary></IndexTerm>
668 <IndexTerm><Primary><literal>timesFloat#</literal></Primary></IndexTerm>
669 <IndexTerm><Primary><literal>divideFloat#</literal></Primary></IndexTerm>
670 <IndexTerm><Primary><literal>gtFloat#</literal></Primary></IndexTerm>
671 <IndexTerm><Primary><literal>geFloat#</literal></Primary></IndexTerm>
672 <IndexTerm><Primary><literal>eqFloat#</literal></Primary></IndexTerm>
673 <IndexTerm><Primary><literal>neFloat#</literal></Primary></IndexTerm>
674 <IndexTerm><Primary><literal>ltFloat#</literal></Primary></IndexTerm>
675 <IndexTerm><Primary><literal>leFloat#</literal></Primary></IndexTerm>
676 <IndexTerm><Primary><literal>negateFloat#</literal></Primary></IndexTerm>
677 <IndexTerm><Primary><literal>float2Int#</literal></Primary></IndexTerm>
678 <IndexTerm><Primary><literal>int2Float#</literal></Primary></IndexTerm>
682 And a full complement of trigonometric functions:
688 expDouble# :: Double# -> Double#
689 logDouble# :: Double# -> Double#
690 sqrtDouble# :: Double# -> Double#
691 sinDouble# :: Double# -> Double#
692 cosDouble# :: Double# -> Double#
693 tanDouble# :: Double# -> Double#
694 asinDouble# :: Double# -> Double#
695 acosDouble# :: Double# -> Double#
696 atanDouble# :: Double# -> Double#
697 sinhDouble# :: Double# -> Double#
698 coshDouble# :: Double# -> Double#
699 tanhDouble# :: Double# -> Double#
700 powerDouble# :: Double# -> Double# -> Double#
703 <IndexTerm><Primary>trigonometric functions, primitive</Primary></IndexTerm>
707 similarly for <Literal>Float#</Literal>.
711 There are two coercion functions for <Literal>Float#</Literal>/<Literal>Double#</Literal>:
717 float2Double# :: Float# -> Double#
718 double2Float# :: Double# -> Float#
721 <IndexTerm><Primary><literal>float2Double#</literal></Primary></IndexTerm>
722 <IndexTerm><Primary><literal>double2Float#</literal></Primary></IndexTerm>
726 The primitive version of <Function>decodeDouble</Function>
727 (<Function>encodeDouble</Function> is implemented as an external C
734 decodeDouble# :: Double# -> PrelNum.ReturnIntAndGMP
737 <IndexTerm><Primary><literal>encodeDouble#</literal></Primary></IndexTerm>
738 <IndexTerm><Primary><literal>decodeDouble#</literal></Primary></IndexTerm>
742 (And the same for <Literal>Float#</Literal>s.)
747 <Sect2 id="integer-operations">
748 <Title>Operations on/for <Literal>Integers</Literal> (interface to GMP)
752 <IndexTerm><Primary>arbitrary precision integers</Primary></IndexTerm>
753 <IndexTerm><Primary>Integer, operations on</Primary></IndexTerm>
757 We implement <Literal>Integers</Literal> (arbitrary-precision
758 integers) using the GNU multiple-precision (GMP) package (version
763 The data type for <Literal>Integer</Literal> is either a small
764 integer, represented by an <Literal>Int</Literal>, or a large integer
765 represented using the pieces required by GMP's
766 <Literal>MP_INT</Literal> in <Filename>gmp.h</Filename> (see
767 <Filename>gmp.info</Filename> in
768 <Filename>ghc/includes/runtime/gmp</Filename>). It comes out as:
774 data Integer = S# Int# -- small integers
775 | J# Int# ByteArray# -- large integers
778 <IndexTerm><Primary>Integer type</Primary></IndexTerm> The primitive
779 ops to support large <Literal>Integers</Literal> use the
780 “pieces” of the representation, and are as follows:
786 negateInteger# :: Int# -> ByteArray# -> Integer
788 {plus,minus,times}Integer#, gcdInteger#,
789 quotInteger#, remInteger#, divExactInteger#
790 :: Int# -> ByteArray#
791 -> Int# -> ByteArray#
792 -> (# Int#, ByteArray# #)
795 :: Int# -> ByteArray#
796 -> Int# -> ByteArray#
797 -> Int# -- -1 for <; 0 for ==; +1 for >
800 :: Int# -> ByteArray#
802 -> Int# -- -1 for <; 0 for ==; +1 for >
805 :: Int# -> ByteArray#
809 divModInteger#, quotRemInteger#
810 :: Int# -> ByteArray#
811 -> Int# -> ByteArray#
812 -> (# Int#, ByteArray#,
815 integer2Int# :: Int# -> ByteArray# -> Int#
817 int2Integer# :: Int# -> Integer -- NB: no error-checking on these two!
818 word2Integer# :: Word# -> Integer
820 addr2Integer# :: Addr# -> Integer
821 -- the Addr# is taken to be a `char *' string
822 -- to be converted into an Integer.
825 <IndexTerm><Primary><literal>negateInteger#</literal></Primary></IndexTerm>
826 <IndexTerm><Primary><literal>plusInteger#</literal></Primary></IndexTerm>
827 <IndexTerm><Primary><literal>minusInteger#</literal></Primary></IndexTerm>
828 <IndexTerm><Primary><literal>timesInteger#</literal></Primary></IndexTerm>
829 <IndexTerm><Primary><literal>quotInteger#</literal></Primary></IndexTerm>
830 <IndexTerm><Primary><literal>remInteger#</literal></Primary></IndexTerm>
831 <IndexTerm><Primary><literal>gcdInteger#</literal></Primary></IndexTerm>
832 <IndexTerm><Primary><literal>gcdIntegerInt#</literal></Primary></IndexTerm>
833 <IndexTerm><Primary><literal>divExactInteger#</literal></Primary></IndexTerm>
834 <IndexTerm><Primary><literal>cmpInteger#</literal></Primary></IndexTerm>
835 <IndexTerm><Primary><literal>divModInteger#</literal></Primary></IndexTerm>
836 <IndexTerm><Primary><literal>quotRemInteger#</literal></Primary></IndexTerm>
837 <IndexTerm><Primary><literal>integer2Int#</literal></Primary></IndexTerm>
838 <IndexTerm><Primary><literal>int2Integer#</literal></Primary></IndexTerm>
839 <IndexTerm><Primary><literal>word2Integer#</literal></Primary></IndexTerm>
840 <IndexTerm><Primary><literal>addr2Integer#</literal></Primary></IndexTerm>
846 <Title>Words and addresses</Title>
849 <IndexTerm><Primary>word, primitive type</Primary></IndexTerm>
850 <IndexTerm><Primary>address, primitive type</Primary></IndexTerm>
851 <IndexTerm><Primary>unsigned integer, primitive type</Primary></IndexTerm>
852 <IndexTerm><Primary>pointer, primitive type</Primary></IndexTerm>
856 A <Literal>Word#</Literal> is used for bit-twiddling operations.
857 It is the same size as an <Literal>Int#</Literal>, but has no sign
858 nor any arithmetic operations.
861 type Word# -- Same size/etc as Int# but *unsigned*
862 type Addr# -- A pointer from outside the "Haskell world" (from C, probably);
863 -- described under "arrays"
866 <IndexTerm><Primary><literal>Word#</literal></Primary></IndexTerm>
867 <IndexTerm><Primary><literal>Addr#</literal></Primary></IndexTerm>
871 <Literal>Word#</Literal>s and <Literal>Addr#</Literal>s have
872 the usual comparison operations. Other
873 unboxed-<Literal>Word</Literal> ops (bit-twiddling and coercions):
879 {gt,ge,eq,ne,lt,le}Word# :: Word# -> Word# -> Bool
881 and#, or#, xor# :: Word# -> Word# -> Word#
884 quotWord#, remWord# :: Word# -> Word# -> Word#
885 -- word (i.e. unsigned) versions are different from int
886 -- versions, so we have to provide these explicitly.
888 not# :: Word# -> Word#
890 shiftL#, shiftRL# :: Word# -> Int# -> Word#
891 -- shift left, right logical
893 int2Word# :: Int# -> Word# -- just a cast, really
894 word2Int# :: Word# -> Int#
897 <IndexTerm><Primary>bit operations, Word and Addr</Primary></IndexTerm>
898 <IndexTerm><Primary><literal>gtWord#</literal></Primary></IndexTerm>
899 <IndexTerm><Primary><literal>geWord#</literal></Primary></IndexTerm>
900 <IndexTerm><Primary><literal>eqWord#</literal></Primary></IndexTerm>
901 <IndexTerm><Primary><literal>neWord#</literal></Primary></IndexTerm>
902 <IndexTerm><Primary><literal>ltWord#</literal></Primary></IndexTerm>
903 <IndexTerm><Primary><literal>leWord#</literal></Primary></IndexTerm>
904 <IndexTerm><Primary><literal>and#</literal></Primary></IndexTerm>
905 <IndexTerm><Primary><literal>or#</literal></Primary></IndexTerm>
906 <IndexTerm><Primary><literal>xor#</literal></Primary></IndexTerm>
907 <IndexTerm><Primary><literal>not#</literal></Primary></IndexTerm>
908 <IndexTerm><Primary><literal>quotWord#</literal></Primary></IndexTerm>
909 <IndexTerm><Primary><literal>remWord#</literal></Primary></IndexTerm>
910 <IndexTerm><Primary><literal>shiftL#</literal></Primary></IndexTerm>
911 <IndexTerm><Primary><literal>shiftRA#</literal></Primary></IndexTerm>
912 <IndexTerm><Primary><literal>shiftRL#</literal></Primary></IndexTerm>
913 <IndexTerm><Primary><literal>int2Word#</literal></Primary></IndexTerm>
914 <IndexTerm><Primary><literal>word2Int#</literal></Primary></IndexTerm>
918 Unboxed-<Literal>Addr</Literal> ops (C casts, really):
921 {gt,ge,eq,ne,lt,le}Addr# :: Addr# -> Addr# -> Bool
923 int2Addr# :: Int# -> Addr#
924 addr2Int# :: Addr# -> Int#
925 addr2Integer# :: Addr# -> (# Int#, ByteArray# #)
928 <IndexTerm><Primary><literal>gtAddr#</literal></Primary></IndexTerm>
929 <IndexTerm><Primary><literal>geAddr#</literal></Primary></IndexTerm>
930 <IndexTerm><Primary><literal>eqAddr#</literal></Primary></IndexTerm>
931 <IndexTerm><Primary><literal>neAddr#</literal></Primary></IndexTerm>
932 <IndexTerm><Primary><literal>ltAddr#</literal></Primary></IndexTerm>
933 <IndexTerm><Primary><literal>leAddr#</literal></Primary></IndexTerm>
934 <IndexTerm><Primary><literal>int2Addr#</literal></Primary></IndexTerm>
935 <IndexTerm><Primary><literal>addr2Int#</literal></Primary></IndexTerm>
936 <IndexTerm><Primary><literal>addr2Integer#</literal></Primary></IndexTerm>
940 The casts between <Literal>Int#</Literal>,
941 <Literal>Word#</Literal> and <Literal>Addr#</Literal>
942 correspond to null operations at the machine level, but are required
943 to keep the Haskell type checker happy.
947 Operations for indexing off of C pointers
948 (<Literal>Addr#</Literal>s) to snatch values are listed under
949 “arrays”.
955 <Title>Arrays</Title>
958 <IndexTerm><Primary>arrays, primitive</Primary></IndexTerm>
962 The type <Literal>Array# elt</Literal> is the type of primitive,
963 unpointed arrays of values of type <Literal>elt</Literal>.
972 <IndexTerm><Primary><literal>Array#</literal></Primary></IndexTerm>
976 <Literal>Array#</Literal> is more primitive than a Haskell
977 array—indeed, the Haskell <Literal>Array</Literal> interface is
978 implemented using <Literal>Array#</Literal>—in that an
979 <Literal>Array#</Literal> is indexed only by
980 <Literal>Int#</Literal>s, starting at zero. It is also more
981 primitive by virtue of being unboxed. That doesn't mean that it isn't
982 a heap-allocated object—of course, it is. Rather, being unboxed
983 means that it is represented by a pointer to the array itself, and not
984 to a thunk which will evaluate to the array (or to bottom). The
985 components of an <Literal>Array#</Literal> are themselves boxed.
989 The type <Literal>ByteArray#</Literal> is similar to
990 <Literal>Array#</Literal>, except that it contains just a string
991 of (non-pointer) bytes.
1000 <IndexTerm><Primary><literal>ByteArray#</literal></Primary></IndexTerm>
1004 Arrays of these types are useful when a Haskell program wishes to
1005 construct a value to pass to a C procedure. It is also possible to use
1006 them to build (say) arrays of unboxed characters for internal use in a
1007 Haskell program. Given these uses, <Literal>ByteArray#</Literal>
1008 is deliberately a bit vague about the type of its components.
1009 Operations are provided to extract values of type
1010 <Literal>Char#</Literal>, <Literal>Int#</Literal>,
1011 <Literal>Float#</Literal>, <Literal>Double#</Literal>, and
1012 <Literal>Addr#</Literal> from arbitrary offsets within a
1013 <Literal>ByteArray#</Literal>. (For type
1014 <Literal>Foo#</Literal>, the $i$th offset gets you the $i$th
1015 <Literal>Foo#</Literal>, not the <Literal>Foo#</Literal> at
1016 byte-position $i$. Mumble.) (If you want a
1017 <Literal>Word#</Literal>, grab an <Literal>Int#</Literal>,
1022 Lastly, we have static byte-arrays, of type
1023 <Literal>Addr#</Literal> [mentioned previously]. (Remember
1024 the duality between arrays and pointers in C.) Arrays of this types
1025 are represented by a pointer to an array in the world outside Haskell,
1026 so this pointer is not followed by the garbage collector. In other
1027 respects they are just like <Literal>ByteArray#</Literal>. They
1028 are only needed in order to pass values from C to Haskell.
1034 <Title>Reading and writing</Title>
1037 Primitive arrays are linear, and indexed starting at zero.
1041 The size and indices of a <Literal>ByteArray#</Literal>, <Literal>Addr#</Literal>, and
1042 <Literal>MutableByteArray#</Literal> are all in bytes. It's up to the program to
1043 calculate the correct byte offset from the start of the array. This
1044 allows a <Literal>ByteArray#</Literal> to contain a mixture of values of different
1045 type, which is often needed when preparing data for and unpicking
1046 results from C. (Umm…not true of indices…WDP 95/09)
1050 <Emphasis>Should we provide some <Literal>sizeOfDouble#</Literal> constants?</Emphasis>
1054 Out-of-range errors on indexing should be caught by the code which
1055 uses the primitive operation; the primitive operations themselves do
1056 <Emphasis>not</Emphasis> check for out-of-range indexes. The intention is that the
1057 primitive ops compile to one machine instruction or thereabouts.
1061 We use the terms “reading” and “writing” to refer to accessing
1062 <Emphasis>mutable</Emphasis> arrays (see <XRef LinkEnd="sect-mutable">), and
1063 “indexing” to refer to reading a value from an <Emphasis>immutable</Emphasis>
1068 Immutable byte arrays are straightforward to index (all indices in bytes):
1071 indexCharArray# :: ByteArray# -> Int# -> Char#
1072 indexIntArray# :: ByteArray# -> Int# -> Int#
1073 indexAddrArray# :: ByteArray# -> Int# -> Addr#
1074 indexFloatArray# :: ByteArray# -> Int# -> Float#
1075 indexDoubleArray# :: ByteArray# -> Int# -> Double#
1077 indexCharOffAddr# :: Addr# -> Int# -> Char#
1078 indexIntOffAddr# :: Addr# -> Int# -> Int#
1079 indexFloatOffAddr# :: Addr# -> Int# -> Float#
1080 indexDoubleOffAddr# :: Addr# -> Int# -> Double#
1081 indexAddrOffAddr# :: Addr# -> Int# -> Addr#
1082 -- Get an Addr# from an Addr# offset
1085 <IndexTerm><Primary><literal>indexCharArray#</literal></Primary></IndexTerm>
1086 <IndexTerm><Primary><literal>indexIntArray#</literal></Primary></IndexTerm>
1087 <IndexTerm><Primary><literal>indexAddrArray#</literal></Primary></IndexTerm>
1088 <IndexTerm><Primary><literal>indexFloatArray#</literal></Primary></IndexTerm>
1089 <IndexTerm><Primary><literal>indexDoubleArray#</literal></Primary></IndexTerm>
1090 <IndexTerm><Primary><literal>indexCharOffAddr#</literal></Primary></IndexTerm>
1091 <IndexTerm><Primary><literal>indexIntOffAddr#</literal></Primary></IndexTerm>
1092 <IndexTerm><Primary><literal>indexFloatOffAddr#</literal></Primary></IndexTerm>
1093 <IndexTerm><Primary><literal>indexDoubleOffAddr#</literal></Primary></IndexTerm>
1094 <IndexTerm><Primary><literal>indexAddrOffAddr#</literal></Primary></IndexTerm>
1098 The last of these, <Function>indexAddrOffAddr#</Function>, extracts an <Literal>Addr#</Literal> using an offset
1099 from another <Literal>Addr#</Literal>, thereby providing the ability to follow a chain of
1104 Something a bit more interesting goes on when indexing arrays of boxed
1105 objects, because the result is simply the boxed object. So presumably
1106 it should be entered—we never usually return an unevaluated
1107 object! This is a pain: primitive ops aren't supposed to do
1108 complicated things like enter objects. The current solution is to
1109 return a single element unboxed tuple (see <XRef LinkEnd="unboxed-tuples">).
1115 indexArray# :: Array# elt -> Int# -> (# elt #)
1118 <IndexTerm><Primary><literal>indexArray#</literal></Primary></IndexTerm>
1124 <Title>The state type</Title>
1127 <IndexTerm><Primary><literal>state, primitive type</literal></Primary></IndexTerm>
1128 <IndexTerm><Primary><literal>State#</literal></Primary></IndexTerm>
1132 The primitive type <Literal>State#</Literal> represents the state of a state
1133 transformer. It is parameterised on the desired type of state, which
1134 serves to keep states from distinct threads distinct from one another.
1135 But the <Emphasis>only</Emphasis> effect of this parameterisation is in the type
1136 system: all values of type <Literal>State#</Literal> are represented in the same way.
1137 Indeed, they are all represented by nothing at all! The code
1138 generator “knows” to generate no code, and allocate no registers
1139 etc, for primitive states.
1151 The type <Literal>GHC.RealWorld</Literal> is truly opaque: there are no values defined
1152 of this type, and no operations over it. It is “primitive” in that
1153 sense - but it is <Emphasis>not unlifted!</Emphasis> Its only role in life is to be
1154 the type which distinguishes the <Literal>IO</Literal> state transformer.
1168 <Title>State of the world</Title>
1171 A single, primitive, value of type <Literal>State# RealWorld</Literal> is provided.
1177 realWorld# :: State# RealWorld
1180 <IndexTerm><Primary>realWorld# state object</Primary></IndexTerm>
1184 (Note: in the compiler, not a <Literal>PrimOp</Literal>; just a mucho magic
1185 <Literal>Id</Literal>. Exported from <Literal>GHC</Literal>, though).
1190 <Sect2 id="sect-mutable">
1191 <Title>Mutable arrays</Title>
1194 <IndexTerm><Primary>mutable arrays</Primary></IndexTerm>
1195 <IndexTerm><Primary>arrays, mutable</Primary></IndexTerm>
1196 Corresponding to <Literal>Array#</Literal> and <Literal>ByteArray#</Literal>, we have the types of
1197 mutable versions of each. In each case, the representation is a
1198 pointer to a suitable block of (mutable) heap-allocated storage.
1204 type MutableArray# s elt
1205 type MutableByteArray# s
1208 <IndexTerm><Primary><literal>MutableArray#</literal></Primary></IndexTerm>
1209 <IndexTerm><Primary><literal>MutableByteArray#</literal></Primary></IndexTerm>
1213 <Title>Allocation</Title>
1216 <IndexTerm><Primary>mutable arrays, allocation</Primary></IndexTerm>
1217 <IndexTerm><Primary>arrays, allocation</Primary></IndexTerm>
1218 <IndexTerm><Primary>allocation, of mutable arrays</Primary></IndexTerm>
1222 Mutable arrays can be allocated. Only pointer-arrays are initialised;
1223 arrays of non-pointers are filled in by “user code” rather than by
1224 the array-allocation primitive. Reason: only the pointer case has to
1225 worry about GC striking with a partly-initialised array.
1231 newArray# :: Int# -> elt -> State# s -> (# State# s, MutableArray# s elt #)
1233 newCharArray# :: Int# -> State# s -> (# State# s, MutableByteArray# s elt #)
1234 newIntArray# :: Int# -> State# s -> (# State# s, MutableByteArray# s elt #)
1235 newAddrArray# :: Int# -> State# s -> (# State# s, MutableByteArray# s elt #)
1236 newFloatArray# :: Int# -> State# s -> (# State# s, MutableByteArray# s elt #)
1237 newDoubleArray# :: Int# -> State# s -> (# State# s, MutableByteArray# s elt #)
1240 <IndexTerm><Primary><literal>newArray#</literal></Primary></IndexTerm>
1241 <IndexTerm><Primary><literal>newCharArray#</literal></Primary></IndexTerm>
1242 <IndexTerm><Primary><literal>newIntArray#</literal></Primary></IndexTerm>
1243 <IndexTerm><Primary><literal>newAddrArray#</literal></Primary></IndexTerm>
1244 <IndexTerm><Primary><literal>newFloatArray#</literal></Primary></IndexTerm>
1245 <IndexTerm><Primary><literal>newDoubleArray#</literal></Primary></IndexTerm>
1249 The size of a <Literal>ByteArray#</Literal> is given in bytes.
1255 <Title>Reading and writing</Title>
1258 <IndexTerm><Primary>arrays, reading and writing</Primary></IndexTerm>
1264 readArray# :: MutableArray# s elt -> Int# -> State# s -> (# State# s, elt #)
1265 readCharArray# :: MutableByteArray# s -> Int# -> State# s -> (# State# s, Char# #)
1266 readIntArray# :: MutableByteArray# s -> Int# -> State# s -> (# State# s, Int# #)
1267 readAddrArray# :: MutableByteArray# s -> Int# -> State# s -> (# State# s, Addr# #)
1268 readFloatArray# :: MutableByteArray# s -> Int# -> State# s -> (# State# s, Float# #)
1269 readDoubleArray# :: MutableByteArray# s -> Int# -> State# s -> (# State# s, Double# #)
1271 writeArray# :: MutableArray# s elt -> Int# -> elt -> State# s -> State# s
1272 writeCharArray# :: MutableByteArray# s -> Int# -> Char# -> State# s -> State# s
1273 writeIntArray# :: MutableByteArray# s -> Int# -> Int# -> State# s -> State# s
1274 writeAddrArray# :: MutableByteArray# s -> Int# -> Addr# -> State# s -> State# s
1275 writeFloatArray# :: MutableByteArray# s -> Int# -> Float# -> State# s -> State# s
1276 writeDoubleArray# :: MutableByteArray# s -> Int# -> Double# -> State# s -> State# s
1279 <IndexTerm><Primary><literal>readArray#</literal></Primary></IndexTerm>
1280 <IndexTerm><Primary><literal>readCharArray#</literal></Primary></IndexTerm>
1281 <IndexTerm><Primary><literal>readIntArray#</literal></Primary></IndexTerm>
1282 <IndexTerm><Primary><literal>readAddrArray#</literal></Primary></IndexTerm>
1283 <IndexTerm><Primary><literal>readFloatArray#</literal></Primary></IndexTerm>
1284 <IndexTerm><Primary><literal>readDoubleArray#</literal></Primary></IndexTerm>
1285 <IndexTerm><Primary><literal>writeArray#</literal></Primary></IndexTerm>
1286 <IndexTerm><Primary><literal>writeCharArray#</literal></Primary></IndexTerm>
1287 <IndexTerm><Primary><literal>writeIntArray#</literal></Primary></IndexTerm>
1288 <IndexTerm><Primary><literal>writeAddrArray#</literal></Primary></IndexTerm>
1289 <IndexTerm><Primary><literal>writeFloatArray#</literal></Primary></IndexTerm>
1290 <IndexTerm><Primary><literal>writeDoubleArray#</literal></Primary></IndexTerm>
1296 <Title>Equality</Title>
1299 <IndexTerm><Primary>arrays, testing for equality</Primary></IndexTerm>
1303 One can take “equality” of mutable arrays. What is compared is the
1304 <Emphasis>name</Emphasis> or reference to the mutable array, not its contents.
1310 sameMutableArray# :: MutableArray# s elt -> MutableArray# s elt -> Bool
1311 sameMutableByteArray# :: MutableByteArray# s -> MutableByteArray# s -> Bool
1314 <IndexTerm><Primary><literal>sameMutableArray#</literal></Primary></IndexTerm>
1315 <IndexTerm><Primary><literal>sameMutableByteArray#</literal></Primary></IndexTerm>
1321 <Title>Freezing mutable arrays</Title>
1324 <IndexTerm><Primary>arrays, freezing mutable</Primary></IndexTerm>
1325 <IndexTerm><Primary>freezing mutable arrays</Primary></IndexTerm>
1326 <IndexTerm><Primary>mutable arrays, freezing</Primary></IndexTerm>
1330 Only unsafe-freeze has a primitive. (Safe freeze is done directly in Haskell
1331 by copying the array and then using <Function>unsafeFreeze</Function>.)
1337 unsafeFreezeArray# :: MutableArray# s elt -> State# s -> (# State# s, Array# s elt #)
1338 unsafeFreezeByteArray# :: MutableByteArray# s -> State# s -> (# State# s, ByteArray# #)
1341 <IndexTerm><Primary><literal>unsafeFreezeArray#</literal></Primary></IndexTerm>
1342 <IndexTerm><Primary><literal>unsafeFreezeByteArray#</literal></Primary></IndexTerm>
1350 <Title>Synchronizing variables (M-vars)</Title>
1353 <IndexTerm><Primary>synchronising variables (M-vars)</Primary></IndexTerm>
1354 <IndexTerm><Primary>M-Vars</Primary></IndexTerm>
1358 Synchronising variables are the primitive type used to implement
1359 Concurrent Haskell's MVars (see the Concurrent Haskell paper for
1360 the operational behaviour of these operations).
1366 type MVar# s elt -- primitive
1368 newMVar# :: State# s -> (# State# s, MVar# s elt #)
1369 takeMVar# :: SynchVar# s elt -> State# s -> (# State# s, elt #)
1370 putMVar# :: SynchVar# s elt -> State# s -> State# s
1373 <IndexTerm><Primary><literal>SynchVar#</literal></Primary></IndexTerm>
1374 <IndexTerm><Primary><literal>newSynchVar#</literal></Primary></IndexTerm>
1375 <IndexTerm><Primary><literal>takeMVar</literal></Primary></IndexTerm>
1376 <IndexTerm><Primary><literal>putMVar</literal></Primary></IndexTerm>
1383 <Sect1 id="glasgow-ST-monad">
1384 <Title>Primitive state-transformer monad
1388 <IndexTerm><Primary>state transformers (Glasgow extensions)</Primary></IndexTerm>
1389 <IndexTerm><Primary>ST monad (Glasgow extension)</Primary></IndexTerm>
1393 This monad underlies our implementation of arrays, mutable and
1394 immutable, and our implementation of I/O, including “C calls”.
1398 The <Literal>ST</Literal> library, which provides access to the
1399 <Function>ST</Function> monad, is described in <xref
1405 <Sect1 id="glasgow-prim-arrays">
1406 <Title>Primitive arrays, mutable and otherwise
1410 <IndexTerm><Primary>primitive arrays (Glasgow extension)</Primary></IndexTerm>
1411 <IndexTerm><Primary>arrays, primitive (Glasgow extension)</Primary></IndexTerm>
1415 GHC knows about quite a few flavours of Large Swathes of Bytes.
1419 First, GHC distinguishes between primitive arrays of (boxed) Haskell
1420 objects (type <Literal>Array# obj</Literal>) and primitive arrays of bytes (type
1421 <Literal>ByteArray#</Literal>).
1425 Second, it distinguishes between…
1429 <Term>Immutable:</Term>
1432 Arrays that do not change (as with “standard” Haskell arrays); you
1433 can only read from them. Obviously, they do not need the care and
1434 attention of the state-transformer monad.
1439 <Term>Mutable:</Term>
1442 Arrays that may be changed or “mutated.” All the operations on them
1443 live within the state-transformer monad and the updates happen
1444 <Emphasis>in-place</Emphasis>.
1449 <Term>“Static” (in C land):</Term>
1452 A C routine may pass an <Literal>Addr#</Literal> pointer back into Haskell land. There
1453 are then primitive operations with which you may merrily grab values
1454 over in C land, by indexing off the “static” pointer.
1459 <Term>“Stable” pointers:</Term>
1462 If, for some reason, you wish to hand a Haskell pointer (i.e.,
1463 <Emphasis>not</Emphasis> an unboxed value) to a C routine, you first make the
1464 pointer “stable,” so that the garbage collector won't forget that it
1465 exists. That is, GHC provides a safe way to pass Haskell pointers to
1470 Please see <XRef LinkEnd="sec-stable-pointers"> for more details.
1475 <Term>“Foreign objects”:</Term>
1478 A “foreign object” is a safe way to pass an external object (a
1479 C-allocated pointer, say) to Haskell and have Haskell do the Right
1480 Thing when it no longer references the object. So, for example, C
1481 could pass a large bitmap over to Haskell and say “please free this
1482 memory when you're done with it.”
1486 Please see <XRef LinkEnd="sec-ForeignObj"> for more details.
1494 The libraries documentatation gives more details on all these
1495 “primitive array” types and the operations on them.
1501 <Sect1 id="pattern-guards">
1502 <Title>Pattern guards</Title>
1505 <IndexTerm><Primary>Pattern guards (Glasgow extension)</Primary></IndexTerm>
1506 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.)
1510 Suppose we have an abstract data type of finite maps, with a
1514 lookup :: FiniteMap -> Int -> Maybe Int
1517 The lookup returns <Function>Nothing</Function> if the supplied key is not in the domain of the mapping, and <Function>(Just v)</Function> otherwise,
1518 where <VarName>v</VarName> is the value that the key maps to. Now consider the following definition:
1522 clunky env var1 var2 | ok1 && ok2 = val1 + val2
1523 | otherwise = var1 + var2
1525 m1 = lookup env var1
1526 m2 = lookup env var2
1527 ok1 = maybeToBool m1
1528 ok2 = maybeToBool m2
1529 val1 = expectJust m1
1530 val2 = expectJust m2
1534 The auxiliary functions are
1538 maybeToBool :: Maybe a -> Bool
1539 maybeToBool (Just x) = True
1540 maybeToBool Nothing = False
1542 expectJust :: Maybe a -> a
1543 expectJust (Just x) = x
1544 expectJust Nothing = error "Unexpected Nothing"
1548 What is <Function>clunky</Function> doing? The guard <Literal>ok1 &&
1549 ok2</Literal> checks that both lookups succeed, using
1550 <Function>maybeToBool</Function> to convert the <Function>Maybe</Function>
1551 types to booleans. The (lazily evaluated) <Function>expectJust</Function>
1552 calls extract the values from the results of the lookups, and binds the
1553 returned values to <VarName>val1</VarName> and <VarName>val2</VarName>
1554 respectively. If either lookup fails, then clunky takes the
1555 <Literal>otherwise</Literal> case and returns the sum of its arguments.
1559 This is certainly legal Haskell, but it is a tremendously verbose and
1560 un-obvious way to achieve the desired effect. Arguably, a more direct way
1561 to write clunky would be to use case expressions:
1565 clunky env var1 var1 = case lookup env var1 of
1567 Just val1 -> case lookup env var2 of
1569 Just val2 -> val1 + val2
1575 This is a bit shorter, but hardly better. Of course, we can rewrite any set
1576 of pattern-matching, guarded equations as case expressions; that is
1577 precisely what the compiler does when compiling equations! The reason that
1578 Haskell provides guarded equations is because they allow us to write down
1579 the cases we want to consider, one at a time, independently of each other.
1580 This structure is hidden in the case version. Two of the right-hand sides
1581 are really the same (<Function>fail</Function>), and the whole expression
1582 tends to become more and more indented.
1586 Here is how I would write clunky:
1590 clunky env var1 var1
1591 | Just val1 <- lookup env var1
1592 , Just val2 <- lookup env var2
1594 ...other equations for clunky...
1598 The semantics should be clear enough. The qualifers are matched in order.
1599 For a <Literal><-</Literal> qualifier, which I call a pattern guard, the
1600 right hand side is evaluated and matched against the pattern on the left.
1601 If the match fails then the whole guard fails and the next equation is
1602 tried. If it succeeds, then the appropriate binding takes place, and the
1603 next qualifier is matched, in the augmented environment. Unlike list
1604 comprehensions, however, the type of the expression to the right of the
1605 <Literal><-</Literal> is the same as the type of the pattern to its
1606 left. The bindings introduced by pattern guards scope over all the
1607 remaining guard qualifiers, and over the right hand side of the equation.
1611 Just as with list comprehensions, boolean expressions can be freely mixed
1612 with among the pattern guards. For example:
1623 Haskell's current guards therefore emerge as a special case, in which the
1624 qualifier list has just one element, a boolean expression.
1628 <sect1 id="sec-ffi">
1629 <title>The foreign interface</title>
1631 <para>The foreign interface consists of the following components:</para>
1635 <para>The Foreign Function Interface language specification
1636 (included in this manual, in <xref linkend="ffi">).</para>
1640 <para>The <literal>Foreign</literal> module (see <xref
1641 linkend="sec-Foreign">) collects together several interfaces
1642 which are useful in specifying foreign language
1643 interfaces, including the following:</para>
1647 <para>The <literal>ForeignObj</literal> module (see <xref
1648 linkend="sec-ForeignObj">), for managing pointers from
1649 Haskell into the outside world.</para>
1653 <para>The <literal>StablePtr</literal> module (see <xref
1654 linkend="sec-stable-pointers">), for managing pointers
1655 into Haskell from the outside world.</para>
1659 <para>The <literal>CTypes</literal> module (see <xref
1660 linkend="sec-CTypes">) gives Haskell equivalents for the
1661 standard C datatypes, for use in making Haskell bindings
1662 to existing C libraries.</para>
1666 <para>The <literal>CTypesISO</literal> module (see <xref
1667 linkend="sec-CTypesISO">) gives Haskell equivalents for C
1668 types defined by the ISO C standard.</para>
1672 <para>The <literal>Storable</literal> library, for
1673 primitive marshalling of data types between Haskell and
1674 the foreign language.</para>
1681 <para>The following sections also give some hints and tips on the use
1682 of the foreign function interface in GHC.</para>
1684 <Sect2 id="glasgow-foreign-headers">
1685 <Title>Using function headers
1689 <IndexTerm><Primary>C calls, function headers</Primary></IndexTerm>
1693 When generating C (using the <Option>-fvia-C</Option> directive), one can assist the
1694 C compiler in detecting type errors by using the <Command>-#include</Command> directive
1695 to provide <Filename>.h</Filename> files containing function headers.
1707 void initialiseEFS (HsInt size);
1708 HsInt terminateEFS (void);
1709 HsForeignObj emptyEFS(void);
1710 HsForeignObj updateEFS (HsForeignObj a, HsInt i, HsInt x);
1711 HsInt lookupEFS (HsForeignObj a, HsInt i);
1715 <para>The types <literal>HsInt</literal>,
1716 <literal>HsForeignObj</literal> etc. are described in <xref
1717 linkend="sec-mapping-table">.</Para>
1719 <Para>Note that this approach is only
1720 <Emphasis>essential</Emphasis> for returning
1721 <Literal>float</Literal>s (or if <Literal>sizeof(int) !=
1722 sizeof(int *)</Literal> on your architecture) but is a Good
1723 Thing for anyone who cares about writing solid code. You're
1724 crazy not to do it.</Para>
1730 <Sect1 id="multi-param-type-classes">
1731 <Title>Multi-parameter type classes
1735 This section documents GHC's implementation of multi-parameter type
1736 classes. There's lots of background in the paper <ULink
1737 URL="http://research.microsoft.com/~simonpj/multi.ps.gz" >Type
1738 classes: exploring the design space</ULink > (Simon Peyton Jones, Mark
1739 Jones, Erik Meijer).
1743 I'd like to thank people who reported shorcomings in the GHC 3.02
1744 implementation. Our default decisions were all conservative ones, and
1745 the experience of these heroic pioneers has given useful concrete
1746 examples to support several generalisations. (These appear below as
1747 design choices not implemented in 3.02.)
1751 I've discussed these notes with Mark Jones, and I believe that Hugs
1752 will migrate towards the same design choices as I outline here.
1753 Thanks to him, and to many others who have offered very useful
1758 <Title>Types</Title>
1761 There are the following restrictions on the form of a qualified
1768 forall tv1..tvn (c1, ...,cn) => type
1774 (Here, I write the "foralls" explicitly, although the Haskell source
1775 language omits them; in Haskell 1.4, all the free type variables of an
1776 explicit source-language type signature are universally quantified,
1777 except for the class type variables in a class declaration. However,
1778 in GHC, you can give the foralls if you want. See <XRef LinkEnd="universal-quantification">).
1787 <Emphasis>Each universally quantified type variable
1788 <Literal>tvi</Literal> must be mentioned (i.e. appear free) in <Literal>type</Literal></Emphasis>.
1790 The reason for this is that a value with a type that does not obey
1791 this restriction could not be used without introducing
1792 ambiguity. Here, for example, is an illegal type:
1796 forall a. Eq a => Int
1800 When a value with this type was used, the constraint <Literal>Eq tv</Literal>
1801 would be introduced where <Literal>tv</Literal> is a fresh type variable, and
1802 (in the dictionary-translation implementation) the value would be
1803 applied to a dictionary for <Literal>Eq tv</Literal>. The difficulty is that we
1804 can never know which instance of <Literal>Eq</Literal> to use because we never
1805 get any more information about <Literal>tv</Literal>.
1812 <Emphasis>Every constraint <Literal>ci</Literal> must mention at least one of the
1813 universally quantified type variables <Literal>tvi</Literal></Emphasis>.
1815 For example, this type is OK because <Literal>C a b</Literal> mentions the
1816 universally quantified type variable <Literal>b</Literal>:
1820 forall a. C a b => burble
1824 The next type is illegal because the constraint <Literal>Eq b</Literal> does not
1825 mention <Literal>a</Literal>:
1829 forall a. Eq b => burble
1833 The reason for this restriction is milder than the other one. The
1834 excluded types are never useful or necessary (because the offending
1835 context doesn't need to be witnessed at this point; it can be floated
1836 out). Furthermore, floating them out increases sharing. Lastly,
1837 excluding them is a conservative choice; it leaves a patch of
1838 territory free in case we need it later.
1848 These restrictions apply to all types, whether declared in a type signature
1853 Unlike Haskell 1.4, constraints in types do <Emphasis>not</Emphasis> have to be of
1854 the form <Emphasis>(class type-variables)</Emphasis>. Thus, these type signatures
1861 f :: Eq (m a) => [m a] -> [m a]
1868 This choice recovers principal types, a property that Haskell 1.4 does not have.
1874 <Title>Class declarations</Title>
1882 <Emphasis>Multi-parameter type classes are permitted</Emphasis>. For example:
1886 class Collection c a where
1887 union :: c a -> c a -> c a
1898 <Emphasis>The class hierarchy must be acyclic</Emphasis>. However, the definition
1899 of "acyclic" involves only the superclass relationships. For example,
1905 op :: D b => a -> b -> b
1908 class C a => D a where { ... }
1912 Here, <Literal>C</Literal> is a superclass of <Literal>D</Literal>, but it's OK for a
1913 class operation <Literal>op</Literal> of <Literal>C</Literal> to mention <Literal>D</Literal>. (It
1914 would not be OK for <Literal>D</Literal> to be a superclass of <Literal>C</Literal>.)
1921 <Emphasis>There are no restrictions on the context in a class declaration
1922 (which introduces superclasses), except that the class hierarchy must
1923 be acyclic</Emphasis>. So these class declarations are OK:
1927 class Functor (m k) => FiniteMap m k where
1930 class (Monad m, Monad (t m)) => Transform t m where
1931 lift :: m a -> (t m) a
1940 <Emphasis>In the signature of a class operation, every constraint
1941 must mention at least one type variable that is not a class type
1942 variable</Emphasis>.
1948 class Collection c a where
1949 mapC :: Collection c b => (a->b) -> c a -> c b
1953 is OK because the constraint <Literal>(Collection a b)</Literal> mentions
1954 <Literal>b</Literal>, even though it also mentions the class variable
1955 <Literal>a</Literal>. On the other hand:
1960 op :: Eq a => (a,b) -> (a,b)
1964 is not OK because the constraint <Literal>(Eq a)</Literal> mentions on the class
1965 type variable <Literal>a</Literal>, but not <Literal>b</Literal>. However, any such
1966 example is easily fixed by moving the offending context up to the
1971 class Eq a => C a where
1976 A yet more relaxed rule would allow the context of a class-op signature
1977 to mention only class type variables. However, that conflicts with
1978 Rule 1(b) for types above.
1985 <Emphasis>The type of each class operation must mention <Emphasis>all</Emphasis> of
1986 the class type variables</Emphasis>. For example:
1990 class Coll s a where
1992 insert :: s -> a -> s
1996 is not OK, because the type of <Literal>empty</Literal> doesn't mention
1997 <Literal>a</Literal>. This rule is a consequence of Rule 1(a), above, for
1998 types, and has the same motivation.
2000 Sometimes, offending class declarations exhibit misunderstandings. For
2001 example, <Literal>Coll</Literal> might be rewritten
2005 class Coll s a where
2007 insert :: s a -> a -> s a
2011 which makes the connection between the type of a collection of
2012 <Literal>a</Literal>'s (namely <Literal>(s a)</Literal>) and the element type <Literal>a</Literal>.
2013 Occasionally this really doesn't work, in which case you can split the
2021 class CollE s => Coll s a where
2022 insert :: s -> a -> s
2035 <Sect2 id="instance-decls">
2036 <Title>Instance declarations</Title>
2044 <Emphasis>Instance declarations may not overlap</Emphasis>. The two instance
2049 instance context1 => C type1 where ...
2050 instance context2 => C type2 where ...
2054 "overlap" if <Literal>type1</Literal> and <Literal>type2</Literal> unify
2056 However, if you give the command line option
2057 <Option>-fallow-overlapping-instances</Option><IndexTerm><Primary>-fallow-overlapping-instances
2058 option</Primary></IndexTerm> then two overlapping instance declarations are permitted
2066 EITHER <Literal>type1</Literal> and <Literal>type2</Literal> do not unify
2072 OR <Literal>type2</Literal> is a substitution instance of <Literal>type1</Literal>
2073 (but not identical to <Literal>type1</Literal>)
2086 Notice that these rules
2093 make it clear which instance decl to use
2094 (pick the most specific one that matches)
2101 do not mention the contexts <Literal>context1</Literal>, <Literal>context2</Literal>
2102 Reason: you can pick which instance decl
2103 "matches" based on the type.
2110 Regrettably, GHC doesn't guarantee to detect overlapping instance
2111 declarations if they appear in different modules. GHC can "see" the
2112 instance declarations in the transitive closure of all the modules
2113 imported by the one being compiled, so it can "see" all instance decls
2114 when it is compiling <Literal>Main</Literal>. However, it currently chooses not
2115 to look at ones that can't possibly be of use in the module currently
2116 being compiled, in the interests of efficiency. (Perhaps we should
2117 change that decision, at least for <Literal>Main</Literal>.)
2124 <Emphasis>There are no restrictions on the type in an instance
2125 <Emphasis>head</Emphasis>, except that at least one must not be a type variable</Emphasis>.
2126 The instance "head" is the bit after the "=>" in an instance decl. For
2127 example, these are OK:
2131 instance C Int a where ...
2133 instance D (Int, Int) where ...
2135 instance E [[a]] where ...
2139 Note that instance heads <Emphasis>may</Emphasis> contain repeated type variables.
2140 For example, this is OK:
2144 instance Stateful (ST s) (MutVar s) where ...
2148 The "at least one not a type variable" restriction is to ensure that
2149 context reduction terminates: each reduction step removes one type
2150 constructor. For example, the following would make the type checker
2151 loop if it wasn't excluded:
2155 instance C a => C a where ...
2159 There are two situations in which the rule is a bit of a pain. First,
2160 if one allows overlapping instance declarations then it's quite
2161 convenient to have a "default instance" declaration that applies if
2162 something more specific does not:
2171 Second, sometimes you might want to use the following to get the
2172 effect of a "class synonym":
2176 class (C1 a, C2 a, C3 a) => C a where { }
2178 instance (C1 a, C2 a, C3 a) => C a where { }
2182 This allows you to write shorter signatures:
2194 f :: (C1 a, C2 a, C3 a) => ...
2198 I'm on the lookout for a simple rule that preserves decidability while
2199 allowing these idioms. The experimental flag
2200 <Option>-fallow-undecidable-instances</Option><IndexTerm><Primary>-fallow-undecidable-instances
2201 option</Primary></IndexTerm> lifts this restriction, allowing all the types in an
2202 instance head to be type variables.
2209 <Emphasis>Unlike Haskell 1.4, instance heads may use type
2210 synonyms</Emphasis>. As always, using a type synonym is just shorthand for
2211 writing the RHS of the type synonym definition. For example:
2215 type Point = (Int,Int)
2216 instance C Point where ...
2217 instance C [Point] where ...
2221 is legal. However, if you added
2225 instance C (Int,Int) where ...
2229 as well, then the compiler will complain about the overlapping
2230 (actually, identical) instance declarations. As always, type synonyms
2231 must be fully applied. You cannot, for example, write:
2236 instance Monad P where ...
2240 This design decision is independent of all the others, and easily
2241 reversed, but it makes sense to me.
2248 <Emphasis>The types in an instance-declaration <Emphasis>context</Emphasis> must all
2249 be type variables</Emphasis>. Thus
2253 instance C a b => Eq (a,b) where ...
2261 instance C Int b => Foo b where ...
2265 is not OK. Again, the intent here is to make sure that context
2266 reduction terminates.
2268 Voluminous correspondence on the Haskell mailing list has convinced me
2269 that it's worth experimenting with a more liberal rule. If you use
2270 the flag <Option>-fallow-undecidable-instances</Option> can use arbitrary
2271 types in an instance context. Termination is ensured by having a
2272 fixed-depth recursion stack. If you exceed the stack depth you get a
2273 sort of backtrace, and the opportunity to increase the stack depth
2274 with <Option>-fcontext-stack</Option><Emphasis>N</Emphasis>.
2287 <Sect1 id="universal-quantification">
2288 <Title>Explicit universal quantification
2292 GHC now allows you to write explicitly quantified types. GHC's
2293 syntax for this now agrees with Hugs's, namely:
2299 forall a b. (Ord a, Eq b) => a -> b -> a
2305 The context is, of course, optional. You can't use <Literal>forall</Literal> as
2306 a type variable any more!
2310 Haskell type signatures are implicitly quantified. The <Literal>forall</Literal>
2311 allows us to say exactly what this means. For example:
2329 g :: forall b. (b -> b)
2335 The two are treated identically.
2339 <Title>Universally-quantified data type fields
2343 In a <Literal>data</Literal> or <Literal>newtype</Literal> declaration one can quantify
2344 the types of the constructor arguments. Here are several examples:
2350 data T a = T1 (forall b. b -> b -> b) a
2352 data MonadT m = MkMonad { return :: forall a. a -> m a,
2353 bind :: forall a b. m a -> (a -> m b) -> m b
2356 newtype Swizzle = MkSwizzle (Ord a => [a] -> [a])
2362 The constructors now have so-called <Emphasis>rank 2</Emphasis> polymorphic
2363 types, in which there is a for-all in the argument types.:
2369 T1 :: forall a. (forall b. b -> b -> b) -> a -> T a
2370 MkMonad :: forall m. (forall a. a -> m a)
2371 -> (forall a b. m a -> (a -> m b) -> m b)
2373 MkSwizzle :: (Ord a => [a] -> [a]) -> Swizzle
2379 Notice that you don't need to use a <Literal>forall</Literal> if there's an
2380 explicit context. For example in the first argument of the
2381 constructor <Function>MkSwizzle</Function>, an implicit "<Literal>forall a.</Literal>" is
2382 prefixed to the argument type. The implicit <Literal>forall</Literal>
2383 quantifies all type variables that are not already in scope, and are
2384 mentioned in the type quantified over.
2388 As for type signatures, implicit quantification happens for non-overloaded
2389 types too. So if you write this:
2392 data T a = MkT (Either a b) (b -> b)
2395 it's just as if you had written this:
2398 data T a = MkT (forall b. Either a b) (forall b. b -> b)
2401 That is, since the type variable <Literal>b</Literal> isn't in scope, it's
2402 implicitly universally quantified. (Arguably, it would be better
2403 to <Emphasis>require</Emphasis> explicit quantification on constructor arguments
2404 where that is what is wanted. Feedback welcomed.)
2410 <Title>Construction </Title>
2413 You construct values of types <Literal>T1, MonadT, Swizzle</Literal> by applying
2414 the constructor to suitable values, just as usual. For example,
2420 (T1 (\xy->x) 3) :: T Int
2422 (MkSwizzle sort) :: Swizzle
2423 (MkSwizzle reverse) :: Swizzle
2430 MkMonad r b) :: MonadT Maybe
2436 The type of the argument can, as usual, be more general than the type
2437 required, as <Literal>(MkSwizzle reverse)</Literal> shows. (<Function>reverse</Function>
2438 does not need the <Literal>Ord</Literal> constraint.)
2444 <Title>Pattern matching</Title>
2447 When you use pattern matching, the bound variables may now have
2448 polymorphic types. For example:
2454 f :: T a -> a -> (a, Char)
2455 f (T1 f k) x = (f k x, f 'c' 'd')
2457 g :: (Ord a, Ord b) => Swizzle -> [a] -> (a -> b) -> [b]
2458 g (MkSwizzle s) xs f = s (map f (s xs))
2460 h :: MonadT m -> [m a] -> m [a]
2461 h m [] = return m []
2462 h m (x:xs) = bind m x $ \y ->
2463 bind m (h m xs) $ \ys ->
2470 In the function <Function>h</Function> we use the record selectors <Literal>return</Literal>
2471 and <Literal>bind</Literal> to extract the polymorphic bind and return functions
2472 from the <Literal>MonadT</Literal> data structure, rather than using pattern
2477 You cannot pattern-match against an argument that is polymorphic.
2481 newtype TIM s a = TIM (ST s (Maybe a))
2483 runTIM :: (forall s. TIM s a) -> Maybe a
2484 runTIM (TIM m) = runST m
2490 Here the pattern-match fails, because you can't pattern-match against
2491 an argument of type <Literal>(forall s. TIM s a)</Literal>. Instead you
2492 must bind the variable and pattern match in the right hand side:
2495 runTIM :: (forall s. TIM s a) -> Maybe a
2496 runTIM tm = case tm of { TIM m -> runST m }
2499 The <Literal>tm</Literal> on the right hand side is (invisibly) instantiated, like
2500 any polymorphic value at its occurrence site, and now you can pattern-match
2507 <Title>The partial-application restriction</Title>
2510 There is really only one way in which data structures with polymorphic
2511 components might surprise you: you must not partially apply them.
2512 For example, this is illegal:
2518 map MkSwizzle [sort, reverse]
2524 The restriction is this: <Emphasis>every subexpression of the program must
2525 have a type that has no for-alls, except that in a function
2526 application (f e1…en) the partial applications are not subject to
2527 this rule</Emphasis>. The restriction makes type inference feasible.
2531 In the illegal example, the sub-expression <Literal>MkSwizzle</Literal> has the
2532 polymorphic type <Literal>(Ord b => [b] -> [b]) -> Swizzle</Literal> and is not
2533 a sub-expression of an enclosing application. On the other hand, this
2540 map (T1 (\a b -> a)) [1,2,3]
2546 even though it involves a partial application of <Function>T1</Function>, because
2547 the sub-expression <Literal>T1 (\a b -> a)</Literal> has type <Literal>Int -> T
2554 <Title>Type signatures
2558 Once you have data constructors with universally-quantified fields, or
2559 constants such as <Constant>runST</Constant> that have rank-2 types, it isn't long
2560 before you discover that you need more! Consider:
2566 mkTs f x y = [T1 f x, T1 f y]
2572 <Function>mkTs</Function> is a fuction that constructs some values of type
2573 <Literal>T</Literal>, using some pieces passed to it. The trouble is that since
2574 <Literal>f</Literal> is a function argument, Haskell assumes that it is
2575 monomorphic, so we'll get a type error when applying <Function>T1</Function> to
2576 it. This is a rather silly example, but the problem really bites in
2577 practice. Lots of people trip over the fact that you can't make
2578 "wrappers functions" for <Constant>runST</Constant> for exactly the same reason.
2579 In short, it is impossible to build abstractions around functions with
2584 The solution is fairly clear. We provide the ability to give a rank-2
2585 type signature for <Emphasis>ordinary</Emphasis> functions (not only data
2586 constructors), thus:
2592 mkTs :: (forall b. b -> b -> b) -> a -> [T a]
2593 mkTs f x y = [T1 f x, T1 f y]
2599 This type signature tells the compiler to attribute <Literal>f</Literal> with
2600 the polymorphic type <Literal>(forall b. b -> b -> b)</Literal> when type
2601 checking the body of <Function>mkTs</Function>, so now the application of
2602 <Function>T1</Function> is fine.
2606 There are two restrictions:
2615 You can only define a rank 2 type, specified by the following
2620 rank2type ::= [forall tyvars .] [context =>] funty
2621 funty ::= ([forall tyvars .] [context =>] ty) -> funty
2623 ty ::= ...current Haskell monotype syntax...
2627 Informally, the universal quantification must all be right at the beginning,
2628 or at the top level of a function argument.
2635 There is a restriction on the definition of a function whose
2636 type signature is a rank-2 type: the polymorphic arguments must be
2637 matched on the left hand side of the "<Literal>=</Literal>" sign. You can't
2638 define <Function>mkTs</Function> like this:
2642 mkTs :: (forall b. b -> b -> b) -> a -> [T a]
2643 mkTs = \ f x y -> [T1 f x, T1 f y]
2648 The same partial-application rule applies to ordinary functions with
2649 rank-2 types as applied to data constructors.
2662 <Title>Type synonyms and hoisting
2666 GHC also allows you to write a <Literal>forall</Literal> in a type synonym, thus:
2668 type Discard a = forall b. a -> b -> a
2673 However, it is often convenient to use these sort of synonyms at the right hand
2674 end of an arrow, thus:
2676 type Discard a = forall b. a -> b -> a
2678 g :: Int -> Discard Int
2681 Simply expanding the type synonym would give
2683 g :: Int -> (forall b. Int -> b -> Int)
2685 but GHC "hoists" the <Literal>forall</Literal> to give the isomorphic type
2687 g :: forall b. Int -> Int -> b -> Int
2689 In general, the rule is this: <Emphasis>to determine the type specified by any explicit
2690 user-written type (e.g. in a type signature), GHC expands type synonyms and then repeatedly
2691 performs the transformation:</Emphasis>
2693 <Emphasis>type1</Emphasis> -> forall a. <Emphasis>type2</Emphasis>
2695 forall a. <Emphasis>type1</Emphasis> -> <Emphasis>type2</Emphasis>
2697 (In fact, GHC tries to retain as much synonym information as possible for use in
2698 error messages, but that is a usability issue.) This rule applies, of course, whether
2699 or not the <Literal>forall</Literal> comes from a synonym. For example, here is another
2700 valid way to write <Literal>g</Literal>'s type signature:
2702 g :: Int -> Int -> forall b. b -> Int
2709 <Sect1 id="existential-quantification">
2710 <Title>Existentially quantified data constructors
2714 The idea of using existential quantification in data type declarations
2715 was suggested by Laufer (I believe, thought doubtless someone will
2716 correct me), and implemented in Hope+. It's been in Lennart
2717 Augustsson's <Command>hbc</Command> Haskell compiler for several years, and
2718 proved very useful. Here's the idea. Consider the declaration:
2724 data Foo = forall a. MkFoo a (a -> Bool)
2731 The data type <Literal>Foo</Literal> has two constructors with types:
2737 MkFoo :: forall a. a -> (a -> Bool) -> Foo
2744 Notice that the type variable <Literal>a</Literal> in the type of <Function>MkFoo</Function>
2745 does not appear in the data type itself, which is plain <Literal>Foo</Literal>.
2746 For example, the following expression is fine:
2752 [MkFoo 3 even, MkFoo 'c' isUpper] :: [Foo]
2758 Here, <Literal>(MkFoo 3 even)</Literal> packages an integer with a function
2759 <Function>even</Function> that maps an integer to <Literal>Bool</Literal>; and <Function>MkFoo 'c'
2760 isUpper</Function> packages a character with a compatible function. These
2761 two things are each of type <Literal>Foo</Literal> and can be put in a list.
2765 What can we do with a value of type <Literal>Foo</Literal>?. In particular,
2766 what happens when we pattern-match on <Function>MkFoo</Function>?
2772 f (MkFoo val fn) = ???
2778 Since all we know about <Literal>val</Literal> and <Function>fn</Function> is that they
2779 are compatible, the only (useful) thing we can do with them is to
2780 apply <Function>fn</Function> to <Literal>val</Literal> to get a boolean. For example:
2787 f (MkFoo val fn) = fn val
2793 What this allows us to do is to package heterogenous values
2794 together with a bunch of functions that manipulate them, and then treat
2795 that collection of packages in a uniform manner. You can express
2796 quite a bit of object-oriented-like programming this way.
2799 <Sect2 id="existential">
2800 <Title>Why existential?
2804 What has this to do with <Emphasis>existential</Emphasis> quantification?
2805 Simply that <Function>MkFoo</Function> has the (nearly) isomorphic type
2811 MkFoo :: (exists a . (a, a -> Bool)) -> Foo
2817 But Haskell programmers can safely think of the ordinary
2818 <Emphasis>universally</Emphasis> quantified type given above, thereby avoiding
2819 adding a new existential quantification construct.
2825 <Title>Type classes</Title>
2828 An easy extension (implemented in <Command>hbc</Command>) is to allow
2829 arbitrary contexts before the constructor. For example:
2835 data Baz = forall a. Eq a => Baz1 a a
2836 | forall b. Show b => Baz2 b (b -> b)
2842 The two constructors have the types you'd expect:
2848 Baz1 :: forall a. Eq a => a -> a -> Baz
2849 Baz2 :: forall b. Show b => b -> (b -> b) -> Baz
2855 But when pattern matching on <Function>Baz1</Function> the matched values can be compared
2856 for equality, and when pattern matching on <Function>Baz2</Function> the first matched
2857 value can be converted to a string (as well as applying the function to it).
2858 So this program is legal:
2865 f (Baz1 p q) | p == q = "Yes"
2867 f (Baz1 v fn) = show (fn v)
2873 Operationally, in a dictionary-passing implementation, the
2874 constructors <Function>Baz1</Function> and <Function>Baz2</Function> must store the
2875 dictionaries for <Literal>Eq</Literal> and <Literal>Show</Literal> respectively, and
2876 extract it on pattern matching.
2880 Notice the way that the syntax fits smoothly with that used for
2881 universal quantification earlier.
2887 <Title>Restrictions</Title>
2890 There are several restrictions on the ways in which existentially-quantified
2891 constructors can be use.
2900 When pattern matching, each pattern match introduces a new,
2901 distinct, type for each existential type variable. These types cannot
2902 be unified with any other type, nor can they escape from the scope of
2903 the pattern match. For example, these fragments are incorrect:
2911 Here, the type bound by <Function>MkFoo</Function> "escapes", because <Literal>a</Literal>
2912 is the result of <Function>f1</Function>. One way to see why this is wrong is to
2913 ask what type <Function>f1</Function> has:
2917 f1 :: Foo -> a -- Weird!
2921 What is this "<Literal>a</Literal>" in the result type? Clearly we don't mean
2926 f1 :: forall a. Foo -> a -- Wrong!
2930 The original program is just plain wrong. Here's another sort of error
2934 f2 (Baz1 a b) (Baz1 p q) = a==q
2938 It's ok to say <Literal>a==b</Literal> or <Literal>p==q</Literal>, but
2939 <Literal>a==q</Literal> is wrong because it equates the two distinct types arising
2940 from the two <Function>Baz1</Function> constructors.
2948 You can't pattern-match on an existentially quantified
2949 constructor in a <Literal>let</Literal> or <Literal>where</Literal> group of
2950 bindings. So this is illegal:
2954 f3 x = a==b where { Baz1 a b = x }
2958 You can only pattern-match
2959 on an existentially-quantified constructor in a <Literal>case</Literal> expression or
2960 in the patterns of a function definition.
2962 The reason for this restriction is really an implementation one.
2963 Type-checking binding groups is already a nightmare without
2964 existentials complicating the picture. Also an existential pattern
2965 binding at the top level of a module doesn't make sense, because it's
2966 not clear how to prevent the existentially-quantified type "escaping".
2967 So for now, there's a simple-to-state restriction. We'll see how
2975 You can't use existential quantification for <Literal>newtype</Literal>
2976 declarations. So this is illegal:
2980 newtype T = forall a. Ord a => MkT a
2984 Reason: a value of type <Literal>T</Literal> must be represented as a pair
2985 of a dictionary for <Literal>Ord t</Literal> and a value of type <Literal>t</Literal>.
2986 That contradicts the idea that <Literal>newtype</Literal> should have no
2987 concrete representation. You can get just the same efficiency and effect
2988 by using <Literal>data</Literal> instead of <Literal>newtype</Literal>. If there is no
2989 overloading involved, then there is more of a case for allowing
2990 an existentially-quantified <Literal>newtype</Literal>, because the <Literal>data</Literal>
2991 because the <Literal>data</Literal> version does carry an implementation cost,
2992 but single-field existentially quantified constructors aren't much
2993 use. So the simple restriction (no existential stuff on <Literal>newtype</Literal>)
2994 stands, unless there are convincing reasons to change it.
3002 You can't use <Literal>deriving</Literal> to define instances of a
3003 data type with existentially quantified data constructors.
3005 Reason: in most cases it would not make sense. For example:#
3008 data T = forall a. MkT [a] deriving( Eq )
3011 To derive <Literal>Eq</Literal> in the standard way we would need to have equality
3012 between the single component of two <Function>MkT</Function> constructors:
3016 (MkT a) == (MkT b) = ???
3019 But <VarName>a</VarName> and <VarName>b</VarName> have distinct types, and so can't be compared.
3020 It's just about possible to imagine examples in which the derived instance
3021 would make sense, but it seems altogether simpler simply to prohibit such
3022 declarations. Define your own instances!
3034 <Sect1 id="sec-assertions">
3036 <IndexTerm><Primary>Assertions</Primary></IndexTerm>
3040 If you want to make use of assertions in your standard Haskell code, you
3041 could define a function like the following:
3047 assert :: Bool -> a -> a
3048 assert False x = error "assertion failed!"
3055 which works, but gives you back a less than useful error message --
3056 an assertion failed, but which and where?
3060 One way out is to define an extended <Function>assert</Function> function which also
3061 takes a descriptive string to include in the error message and
3062 perhaps combine this with the use of a pre-processor which inserts
3063 the source location where <Function>assert</Function> was used.
3067 Ghc offers a helping hand here, doing all of this for you. For every
3068 use of <Function>assert</Function> in the user's source:
3074 kelvinToC :: Double -> Double
3075 kelvinToC k = assert (k >= 0.0) (k+273.15)
3081 Ghc will rewrite this to also include the source location where the
3088 assert pred val ==> assertError "Main.hs|15" pred val
3094 The rewrite is only performed by the compiler when it spots
3095 applications of <Function>Exception.assert</Function>, so you can still define and
3096 use your own versions of <Function>assert</Function>, should you so wish. If not,
3097 import <Literal>Exception</Literal> to make use <Function>assert</Function> in your code.
3101 To have the compiler ignore uses of assert, use the compiler option
3102 <Option>-fignore-asserts</Option>. <IndexTerm><Primary>-fignore-asserts option</Primary></IndexTerm> That is,
3103 expressions of the form <Literal>assert pred e</Literal> will be rewritten to <Literal>e</Literal>.
3107 Assertion failures can be caught, see the documentation for the
3108 <literal>Exception</literal> library (<xref linkend="sec-Exception">)
3114 <Sect1 id="scoped-type-variables">
3115 <Title>Scoped Type Variables
3119 A <Emphasis>pattern type signature</Emphasis> can introduce a <Emphasis>scoped type
3120 variable</Emphasis>. For example
3126 f (xs::[a]) = ys ++ ys
3135 The pattern <Literal>(xs::[a])</Literal> includes a type signature for <VarName>xs</VarName>.
3136 This brings the type variable <Literal>a</Literal> into scope; it scopes over
3137 all the patterns and right hand sides for this equation for <Function>f</Function>.
3138 In particular, it is in scope at the type signature for <VarName>y</VarName>.
3142 At ordinary type signatures, such as that for <VarName>ys</VarName>, any type variables
3143 mentioned in the type signature <Emphasis>that are not in scope</Emphasis> are
3144 implicitly universally quantified. (If there are no type variables in
3145 scope, all type variables mentioned in the signature are universally
3146 quantified, which is just as in Haskell 98.) In this case, since <VarName>a</VarName>
3147 is in scope, it is not universally quantified, so the type of <VarName>ys</VarName> is
3148 the same as that of <VarName>xs</VarName>. In Haskell 98 it is not possible to declare
3149 a type for <VarName>ys</VarName>; a major benefit of scoped type variables is that
3150 it becomes possible to do so.
3154 Scoped type variables are implemented in both GHC and Hugs. Where the
3155 implementations differ from the specification below, those differences
3160 So much for the basic idea. Here are the details.
3164 <Title>Scope and implicit quantification</Title>
3172 All the type variables mentioned in the patterns for a single
3173 function definition equation, that are not already in scope,
3174 are brought into scope by the patterns. We describe this set as
3175 the <Emphasis>type variables bound by the equation</Emphasis>.
3182 The type variables thus brought into scope may be mentioned
3183 in ordinary type signatures or pattern type signatures anywhere within
3191 In ordinary type signatures, any type variable mentioned in the
3192 signature that is in scope is <Emphasis>not</Emphasis> universally quantified.
3199 Ordinary type signatures do not bring any new type variables
3200 into scope (except in the type signature itself!). So this is illegal:
3209 It's illegal because <VarName>a</VarName> is not in scope in the body of <Function>f</Function>,
3210 so the ordinary signature <Literal>x::a</Literal> is equivalent to <Literal>x::forall a.a</Literal>;
3211 and that is an incorrect typing.
3218 There is no implicit universal quantification on pattern type
3219 signatures, nor may one write an explicit <Literal>forall</Literal> type in a pattern
3220 type signature. The pattern type signature is a monotype.
3228 The type variables in the head of a <Literal>class</Literal> or <Literal>instance</Literal> declaration
3229 scope over the methods defined in the <Literal>where</Literal> part. For example:
3243 (Not implemented in Hugs yet, Dec 98).
3254 <Title>Polymorphism</Title>
3262 Pattern type signatures are completely orthogonal to ordinary, separate
3263 type signatures. The two can be used independently or together. There is
3264 no scoping associated with the names of the type variables in a separate type signature.
3269 f (xs::[b]) = reverse xs
3278 The function must be polymorphic in the type variables
3279 bound by all its equations. Operationally, the type variables bound
3280 by one equation must not:
3287 Be unified with a type (such as <Literal>Int</Literal>, or <Literal>[a]</Literal>).
3293 Be unified with a type variable free in the environment.
3299 Be unified with each other. (They may unify with the type variables
3300 bound by another equation for the same function, of course.)
3307 For example, the following all fail to type check:
3311 f (x::a) (y::b) = [x,y] -- a unifies with b
3313 g (x::a) = x + 1::Int -- a unifies with Int
3315 h x = let k (y::a) = [x,y] -- a is free in the
3316 in k x -- environment
3318 k (x::a) True = ... -- a unifies with Int
3319 k (x::Int) False = ...
3322 w (x::a) = x -- a unifies with [b]
3331 The pattern-bound type variable may, however, be constrained
3332 by the context of the principal type, thus:
3336 f (x::a) (y::a) = x+y*2
3340 gets the inferred type: <Literal>forall a. Num a => a -> a -> a</Literal>.
3351 <Title>Result type signatures</Title>
3359 The result type of a function can be given a signature,
3364 f (x::a) :: [a] = [x,x,x]
3368 The final <Literal>:: [a]</Literal> after all the patterns gives a signature to the
3369 result type. Sometimes this is the only way of naming the type variable
3374 f :: Int -> [a] -> [a]
3375 f n :: ([a] -> [a]) = let g (x::a, y::a) = (y,x)
3376 in \xs -> map g (reverse xs `zip` xs)
3388 Result type signatures are not yet implemented in Hugs.
3394 <Title>Pattern signatures on other constructs</Title>
3402 A pattern type signature can be on an arbitrary sub-pattern, not
3407 f ((x,y)::(a,b)) = (y,x) :: (b,a)
3416 Pattern type signatures, including the result part, can be used
3417 in lambda abstractions:
3421 (\ (x::a, y) :: a -> x)
3425 Type variables bound by these patterns must be polymorphic in
3426 the sense defined above.
3431 f1 (x::c) = f1 x -- ok
3432 f2 = \(x::c) -> f2 x -- not ok
3436 Here, <Function>f1</Function> is OK, but <Function>f2</Function> is not, because <VarName>c</VarName> gets unified
3437 with a type variable free in the environment, in this
3438 case, the type of <Function>f2</Function>, which is in the environment when
3439 the lambda abstraction is checked.
3446 Pattern type signatures, including the result part, can be used
3447 in <Literal>case</Literal> expressions:
3451 case e of { (x::a, y) :: a -> x }
3455 The pattern-bound type variables must, as usual,
3456 be polymorphic in the following sense: each case alternative,
3457 considered as a lambda abstraction, must be polymorphic.
3462 case (True,False) of { (x::a, y) -> x }
3466 Even though the context is that of a pair of booleans,
3467 the alternative itself is polymorphic. Of course, it is
3472 case (True,False) of { (x::Bool, y) -> x }
3481 To avoid ambiguity, the type after the “<Literal>::</Literal>” in a result
3482 pattern signature on a lambda or <Literal>case</Literal> must be atomic (i.e. a single
3483 token or a parenthesised type of some sort). To see why,
3484 consider how one would parse this:
3497 Pattern type signatures that bind new type variables
3498 may not be used in pattern bindings at all.
3503 f x = let (y, z::a) = x in ...
3507 But these are OK, because they do not bind fresh type variables:
3511 f1 x = let (y, z::Int) = x in ...
3512 f2 (x::(Int,a)) = let (y, z::a) = x in ...
3516 However a single variable is considered a degenerate function binding,
3517 rather than a degerate pattern binding, so this is permitted, even
3518 though it binds a type variable:
3522 f :: (b->b) = \(x::b) -> x
3531 Such degnerate function bindings do not fall under the monomorphism
3538 g :: a -> a -> Bool = \x y. x==y
3544 Here <Function>g</Function> has type <Literal>forall a. Eq a => a -> a -> Bool</Literal>, just as if
3545 <Function>g</Function> had a separate type signature. Lacking a type signature, <Function>g</Function>
3546 would get a monomorphic type.
3552 <Title>Existentials</Title>
3560 Pattern type signatures can bind existential type variables.
3565 data T = forall a. MkT [a]
3568 f (MkT [t::a]) = MkT t3
3585 <Sect1 id="pragmas">
3590 GHC supports several pragmas, or instructions to the compiler placed
3591 in the source code. Pragmas don't affect the meaning of the program,
3592 but they might affect the efficiency of the generated code.
3595 <Sect2 id="inline-pragma">
3596 <Title>INLINE pragma
3598 <IndexTerm><Primary>INLINE pragma</Primary></IndexTerm>
3599 <IndexTerm><Primary>pragma, INLINE</Primary></IndexTerm></Title>
3602 GHC (with <Option>-O</Option>, as always) tries to inline (or “unfold”)
3603 functions/values that are “small enough,” thus avoiding the call
3604 overhead and possibly exposing other more-wonderful optimisations.
3608 You will probably see these unfoldings (in Core syntax) in your
3613 Normally, if GHC decides a function is “too expensive” to inline, it
3614 will not do so, nor will it export that unfolding for other modules to
3619 The sledgehammer you can bring to bear is the
3620 <Literal>INLINE</Literal><IndexTerm><Primary>INLINE pragma</Primary></IndexTerm> pragma, used thusly:
3623 key_function :: Int -> String -> (Bool, Double)
3625 #ifdef __GLASGOW_HASKELL__
3626 {-# INLINE key_function #-}
3630 (You don't need to do the C pre-processor carry-on unless you're going
3631 to stick the code through HBC—it doesn't like <Literal>INLINE</Literal> pragmas.)
3635 The major effect of an <Literal>INLINE</Literal> pragma is to declare a function's
3636 “cost” to be very low. The normal unfolding machinery will then be
3637 very keen to inline it.
3641 An <Literal>INLINE</Literal> pragma for a function can be put anywhere its type
3642 signature could be put.
3646 <Literal>INLINE</Literal> pragmas are a particularly good idea for the
3647 <Literal>then</Literal>/<Literal>return</Literal> (or <Literal>bind</Literal>/<Literal>unit</Literal>) functions in a monad.
3648 For example, in GHC's own <Literal>UniqueSupply</Literal> monad code, we have:
3651 #ifdef __GLASGOW_HASKELL__
3652 {-# INLINE thenUs #-}
3653 {-# INLINE returnUs #-}
3661 <Sect2 id="noinline-pragma">
3662 <Title>NOINLINE pragma
3666 <IndexTerm><Primary>NOINLINE pragma</Primary></IndexTerm>
3667 <IndexTerm><Primary>pragma, NOINLINE</Primary></IndexTerm>
3671 The <Literal>NOINLINE</Literal> pragma does exactly what you'd expect: it stops the
3672 named function from being inlined by the compiler. You shouldn't ever
3673 need to do this, unless you're very cautious about code size.
3678 <Sect2 id="specialize-pragma">
3679 <Title>SPECIALIZE pragma
3683 <IndexTerm><Primary>SPECIALIZE pragma</Primary></IndexTerm>
3684 <IndexTerm><Primary>pragma, SPECIALIZE</Primary></IndexTerm>
3685 <IndexTerm><Primary>overloading, death to</Primary></IndexTerm>
3689 (UK spelling also accepted.) For key overloaded functions, you can
3690 create extra versions (NB: more code space) specialised to particular
3691 types. Thus, if you have an overloaded function:
3697 hammeredLookup :: Ord key => [(key, value)] -> key -> value
3703 If it is heavily used on lists with <Literal>Widget</Literal> keys, you could
3704 specialise it as follows:
3707 {-# SPECIALIZE hammeredLookup :: [(Widget, value)] -> Widget -> value #-}
3713 To get very fancy, you can also specify a named function to use for
3714 the specialised value, by adding <Literal>= blah</Literal>, as in:
3717 {-# SPECIALIZE hammeredLookup :: ...as before... = blah #-}
3720 It's <Emphasis>Your Responsibility</Emphasis> to make sure that <Function>blah</Function> really
3721 behaves as a specialised version of <Function>hammeredLookup</Function>!!!
3725 NOTE: the <Literal>=blah</Literal> feature isn't implemented in GHC 4.xx.
3729 An example in which the <Literal>= blah</Literal> form will Win Big:
3732 toDouble :: Real a => a -> Double
3733 toDouble = fromRational . toRational
3735 {-# SPECIALIZE toDouble :: Int -> Double = i2d #-}
3736 i2d (I# i) = D# (int2Double# i) -- uses Glasgow prim-op directly
3739 The <Function>i2d</Function> function is virtually one machine instruction; the
3740 default conversion—via an intermediate <Literal>Rational</Literal>—is obscenely
3741 expensive by comparison.
3745 By using the US spelling, your <Literal>SPECIALIZE</Literal> pragma will work with
3746 HBC, too. Note that HBC doesn't support the <Literal>= blah</Literal> form.
3750 A <Literal>SPECIALIZE</Literal> pragma for a function can be put anywhere its type
3751 signature could be put.
3756 <Sect2 id="specialize-instance-pragma">
3757 <Title>SPECIALIZE instance pragma
3761 <IndexTerm><Primary>SPECIALIZE pragma</Primary></IndexTerm>
3762 <IndexTerm><Primary>overloading, death to</Primary></IndexTerm>
3763 Same idea, except for instance declarations. For example:
3766 instance (Eq a) => Eq (Foo a) where { ... usual stuff ... }
3768 {-# SPECIALIZE instance Eq (Foo [(Int, Bar)] #-}
3771 Compatible with HBC, by the way.
3776 <Sect2 id="line-pragma">
3781 <IndexTerm><Primary>LINE pragma</Primary></IndexTerm>
3782 <IndexTerm><Primary>pragma, LINE</Primary></IndexTerm>
3786 This pragma is similar to C's <Literal>#line</Literal> pragma, and is mainly for use in
3787 automatically generated Haskell code. It lets you specify the line
3788 number and filename of the original code; for example
3794 {-# LINE 42 "Foo.vhs" #-}
3800 if you'd generated the current file from something called <Filename>Foo.vhs</Filename>
3801 and this line corresponds to line 42 in the original. GHC will adjust
3802 its error messages to refer to the line/file named in the <Literal>LINE</Literal>
3809 <Title>RULES pragma</Title>
3812 The RULES pragma lets you specify rewrite rules. It is described in
3813 <XRef LinkEnd="rewrite-rules">.
3820 <Sect1 id="rewrite-rules">
3821 <Title>Rewrite rules
3823 <IndexTerm><Primary>RULES pagma</Primary></IndexTerm>
3824 <IndexTerm><Primary>pragma, RULES</Primary></IndexTerm>
3825 <IndexTerm><Primary>rewrite rules</Primary></IndexTerm></Title>
3828 The programmer can specify rewrite rules as part of the source program
3829 (in a pragma). GHC applies these rewrite rules wherever it can.
3837 "map/map" forall f g xs. map f (map g xs) = map (f.g) xs
3844 <Title>Syntax</Title>
3847 From a syntactic point of view:
3853 Each rule has a name, enclosed in double quotes. The name itself has
3854 no significance at all. It is only used when reporting how many times the rule fired.
3860 There may be zero or more rules in a <Literal>RULES</Literal> pragma.
3866 Layout applies in a <Literal>RULES</Literal> pragma. Currently no new indentation level
3867 is set, so you must lay out your rules starting in the same column as the
3868 enclosing definitions.
3874 Each variable mentioned in a rule must either be in scope (e.g. <Function>map</Function>),
3875 or bound by the <Literal>forall</Literal> (e.g. <Function>f</Function>, <Function>g</Function>, <Function>xs</Function>). The variables bound by
3876 the <Literal>forall</Literal> are called the <Emphasis>pattern</Emphasis> variables. They are separated
3877 by spaces, just like in a type <Literal>forall</Literal>.
3883 A pattern variable may optionally have a type signature.
3884 If the type of the pattern variable is polymorphic, it <Emphasis>must</Emphasis> have a type signature.
3885 For example, here is the <Literal>foldr/build</Literal> rule:
3888 "fold/build" forall k z (g::forall b. (a->b->b) -> b -> b) .
3889 foldr k z (build g) = g k z
3892 Since <Function>g</Function> has a polymorphic type, it must have a type signature.
3899 The left hand side of a rule must consist of a top-level variable applied
3900 to arbitrary expressions. For example, this is <Emphasis>not</Emphasis> OK:
3903 "wrong1" forall e1 e2. case True of { True -> e1; False -> e2 } = e1
3904 "wrong2" forall f. f True = True
3907 In <Literal>"wrong1"</Literal>, the LHS is not an application; in <Literal>"wrong2"</Literal>, the LHS has a pattern variable
3914 A rule does not need to be in the same module as (any of) the
3915 variables it mentions, though of course they need to be in scope.
3921 Rules are automatically exported from a module, just as instance declarations are.
3932 <Title>Semantics</Title>
3935 From a semantic point of view:
3941 Rules are only applied if you use the <Option>-O</Option> flag.
3947 Rules are regarded as left-to-right rewrite rules.
3948 When GHC finds an expression that is a substitution instance of the LHS
3949 of a rule, it replaces the expression by the (appropriately-substituted) RHS.
3950 By "a substitution instance" we mean that the LHS can be made equal to the
3951 expression by substituting for the pattern variables.
3958 The LHS and RHS of a rule are typechecked, and must have the
3966 GHC makes absolutely no attempt to verify that the LHS and RHS
3967 of a rule have the same meaning. That is undecideable in general, and
3968 infeasible in most interesting cases. The responsibility is entirely the programmer's!
3975 GHC makes no attempt to make sure that the rules are confluent or
3976 terminating. For example:
3979 "loop" forall x,y. f x y = f y x
3982 This rule will cause the compiler to go into an infinite loop.
3989 If more than one rule matches a call, GHC will choose one arbitrarily to apply.
3995 GHC currently uses a very simple, syntactic, matching algorithm
3996 for matching a rule LHS with an expression. It seeks a substitution
3997 which makes the LHS and expression syntactically equal modulo alpha
3998 conversion. The pattern (rule), but not the expression, is eta-expanded if
3999 necessary. (Eta-expanding the epression can lead to laziness bugs.)
4000 But not beta conversion (that's called higher-order matching).
4004 Matching is carried out on GHC's intermediate language, which includes
4005 type abstractions and applications. So a rule only matches if the
4006 types match too. See <XRef LinkEnd="rule-spec"> below.
4012 GHC keeps trying to apply the rules as it optimises the program.
4013 For example, consider:
4022 The expression <Literal>s (t xs)</Literal> does not match the rule <Literal>"map/map"</Literal>, but GHC
4023 will substitute for <VarName>s</VarName> and <VarName>t</VarName>, giving an expression which does match.
4024 If <VarName>s</VarName> or <VarName>t</VarName> was (a) used more than once, and (b) large or a redex, then it would
4025 not be substituted, and the rule would not fire.
4032 In the earlier phases of compilation, GHC inlines <Emphasis>nothing
4033 that appears on the LHS of a rule</Emphasis>, because once you have substituted
4034 for something you can't match against it (given the simple minded
4035 matching). So if you write the rule
4038 "map/map" forall f,g. map f . map g = map (f.g)
4041 this <Emphasis>won't</Emphasis> match the expression <Literal>map f (map g xs)</Literal>.
4042 It will only match something written with explicit use of ".".
4043 Well, not quite. It <Emphasis>will</Emphasis> match the expression
4049 where <Function>wibble</Function> is defined:
4052 wibble f g = map f . map g
4055 because <Function>wibble</Function> will be inlined (it's small).
4057 Later on in compilation, GHC starts inlining even things on the
4058 LHS of rules, but still leaves the rules enabled. This inlining
4059 policy is controlled by the per-simplification-pass flag <Option>-finline-phase</Option><Emphasis>n</Emphasis>.
4066 All rules are implicitly exported from the module, and are therefore
4067 in force in any module that imports the module that defined the rule, directly
4068 or indirectly. (That is, if A imports B, which imports C, then C's rules are
4069 in force when compiling A.) The situation is very similar to that for instance
4081 <Title>List fusion</Title>
4084 The RULES mechanism is used to implement fusion (deforestation) of common list functions.
4085 If a "good consumer" consumes an intermediate list constructed by a "good producer", the
4086 intermediate list should be eliminated entirely.
4090 The following are good producers:
4102 Enumerations of <Literal>Int</Literal> and <Literal>Char</Literal> (e.g. <Literal>['a'..'z']</Literal>).
4108 Explicit lists (e.g. <Literal>[True, False]</Literal>)
4114 The cons constructor (e.g <Literal>3:4:[]</Literal>)
4120 <Function>++</Function>
4126 <Function>map</Function>
4132 <Function>filter</Function>
4138 <Function>iterate</Function>, <Function>repeat</Function>
4144 <Function>zip</Function>, <Function>zipWith</Function>
4153 The following are good consumers:
4165 <Function>array</Function> (on its second argument)
4171 <Function>length</Function>
4177 <Function>++</Function> (on its first argument)
4183 <Function>map</Function>
4189 <Function>filter</Function>
4195 <Function>concat</Function>
4201 <Function>unzip</Function>, <Function>unzip2</Function>, <Function>unzip3</Function>, <Function>unzip4</Function>
4207 <Function>zip</Function>, <Function>zipWith</Function> (but on one argument only; if both are good producers, <Function>zip</Function>
4208 will fuse with one but not the other)
4214 <Function>partition</Function>
4220 <Function>head</Function>
4226 <Function>and</Function>, <Function>or</Function>, <Function>any</Function>, <Function>all</Function>
4232 <Function>sequence_</Function>
4238 <Function>msum</Function>
4244 <Function>sortBy</Function>
4253 So, for example, the following should generate no intermediate lists:
4256 array (1,10) [(i,i*i) | i <- map (+ 1) [0..9]]
4262 This list could readily be extended; if there are Prelude functions that you use
4263 a lot which are not included, please tell us.
4267 If you want to write your own good consumers or producers, look at the
4268 Prelude definitions of the above functions to see how to do so.
4273 <Sect2 id="rule-spec">
4274 <Title>Specialisation
4278 Rewrite rules can be used to get the same effect as a feature
4279 present in earlier version of GHC:
4282 {-# SPECIALIZE fromIntegral :: Int8 -> Int16 = int8ToInt16 #-}
4285 This told GHC to use <Function>int8ToInt16</Function> instead of <Function>fromIntegral</Function> whenever
4286 the latter was called with type <Literal>Int8 -> Int16</Literal>. That is, rather than
4287 specialising the original definition of <Function>fromIntegral</Function> the programmer is
4288 promising that it is safe to use <Function>int8ToInt16</Function> instead.
4292 This feature is no longer in GHC. But rewrite rules let you do the
4297 "fromIntegral/Int8/Int16" fromIntegral = int8ToInt16
4301 This slightly odd-looking rule instructs GHC to replace <Function>fromIntegral</Function>
4302 by <Function>int8ToInt16</Function> <Emphasis>whenever the types match</Emphasis>. Speaking more operationally,
4303 GHC adds the type and dictionary applications to get the typed rule
4306 forall (d1::Integral Int8) (d2::Num Int16) .
4307 fromIntegral Int8 Int16 d1 d2 = int8ToInt16
4311 this rule does not need to be in the same file as fromIntegral,
4312 unlike the <Literal>SPECIALISE</Literal> pragmas which currently do (so that they
4313 have an original definition available to specialise).
4319 <Title>Controlling what's going on</Title>
4327 Use <Option>-ddump-rules</Option> to see what transformation rules GHC is using.
4333 Use <Option>-ddump-simpl-stats</Option> to see what rules are being fired.
4334 If you add <Option>-dppr-debug</Option> you get a more detailed listing.
4340 The defintion of (say) <Function>build</Function> in <FileName>PrelBase.lhs</FileName> looks llike this:
4343 build :: forall a. (forall b. (a -> b -> b) -> b -> b) -> [a]
4344 {-# INLINE build #-}
4348 Notice the <Literal>INLINE</Literal>! That prevents <Literal>(:)</Literal> from being inlined when compiling
4349 <Literal>PrelBase</Literal>, so that an importing module will “see” the <Literal>(:)</Literal>, and can
4350 match it on the LHS of a rule. <Literal>INLINE</Literal> prevents any inlining happening
4351 in the RHS of the <Literal>INLINE</Literal> thing. I regret the delicacy of this.
4358 In <Filename>ghc/lib/std/PrelBase.lhs</Filename> look at the rules for <Function>map</Function> to
4359 see how to write rules that will do fusion and yet give an efficient
4360 program even if fusion doesn't happen. More rules in <Filename>PrelList.lhs</Filename>.
4372 <Sect1 id="generic-classes">
4373 <Title>Generic classes</Title>
4376 The ideas behind this extension are described in detail in "Derivable type classes",
4377 Ralf Hinze and Simon Peyton Jones, Haskell Workshop, Montreal Sept 2000, pp94-105.
4378 An example will give the idea:
4386 fromBin :: [Int] -> (a, [Int])
4388 toBin {| Unit |} Unit = []
4389 toBin {| a :+: b |} (Inl x) = 0 : toBin x
4390 toBin {| a :+: b |} (Inr y) = 1 : toBin y
4391 toBin {| a :*: b |} (x :*: y) = toBin x ++ toBin y
4393 fromBin {| Unit |} bs = (Unit, bs)
4394 fromBin {| a :+: b |} (0:bs) = (Inl x, bs') where (x,bs') = fromBin bs
4395 fromBin {| a :+: b |} (1:bs) = (Inr y, bs') where (y,bs') = fromBin bs
4396 fromBin {| a :*: b |} bs = (x :*: y, bs'') where (x,bs' ) = fromBin bs
4397 (y,bs'') = fromBin bs'
4400 This class declaration explains how <Literal>toBin</Literal> and <Literal>fromBin</Literal>
4401 work for arbitrary data types. They do so by giving cases for unit, product, and sum,
4402 which are defined thus in the library module <Literal>Generics</Literal>:
4406 data a :+: b = Inl a | Inr b
4407 data a :*: b = a :*: b
4410 Now you can make a data type into an instance of Bin like this:
4412 instance (Bin a, Bin b) => Bin (a,b)
4413 instance Bin a => Bin [a]
4415 That is, just leave off the "where" clasuse. Of course, you can put in the
4416 where clause and over-ride whichever methods you please.
4420 <Title> Using generics </Title>
4421 <Para>To use generics you need to</para>
4424 <Para>Use the <Option>-fgenerics</Option> flag.</Para>
4427 <Para>Import the module <Literal>Generics</Literal> from the
4428 <Literal>lang</Literal> package. This import brings into
4429 scope the data types <Literal>Unit</Literal>,
4430 <Literal>:*:</Literal>, and <Literal>:+:</Literal>. (You
4431 don't need this import if you don't mention these types
4432 explicitly; for example, if you are simply giving instance
4433 declarations.)</Para>
4438 <Sect2> <Title> Changes wrt the paper </Title>
4440 Note that the type constructors <Literal>:+:</Literal> and <Literal>:*:</Literal>
4441 can be written infix (indeed, you can now use
4442 any operator starting in a colon as an infix type constructor). Also note that
4443 the type constructors are not exactly as in the paper (Unit instead of 1, etc).
4444 Finally, note that the syntax of the type patterns in the class declaration
4445 uses "<Literal>{|</Literal>" and "<Literal>{|</Literal>" brackets; curly braces
4446 alone would ambiguous when they appear on right hand sides (an extension we
4447 anticipate wanting).
4451 <Sect2> <Title>Terminology and restrictions</Title>
4453 Terminology. A "generic default method" in a class declaration
4454 is one that is defined using type patterns as above.
4455 A "polymorphic default method" is a default method defined as in Haskell 98.
4456 A "generic class declaration" is a class declaration with at least one
4457 generic default method.
4465 Alas, we do not yet implement the stuff about constructor names and
4472 A generic class can have only one parameter; you can't have a generic
4473 multi-parameter class.
4479 A default method must be defined entirely using type patterns, or entirely
4480 without. So this is illegal:
4483 op :: a -> (a, Bool)
4484 op {| Unit |} Unit = (Unit, True)
4487 However it is perfectly OK for some methods of a generic class to have
4488 generic default methods and others to have polymorphic default methods.
4494 The type variable(s) in the type pattern for a generic method declaration
4495 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:
4499 op {| p :*: q |} (x :*: y) = op (x :: p)
4507 The type patterns in a generic default method must take one of the forms:
4513 where "a" and "b" are type variables. Furthermore, all the type patterns for
4514 a single type constructor (<Literal>:*:</Literal>, say) must be identical; they
4515 must use the same type variables. So this is illegal:
4519 op {| a :+: b |} (Inl x) = True
4520 op {| p :+: q |} (Inr y) = False
4522 The type patterns must be identical, even in equations for different methods of the class.
4523 So this too is illegal:
4527 op {| a :*: b |} (Inl x) = True
4530 op {| p :*: q |} (Inr y) = False
4532 (The reason for this restriction is that we gather all the equations for a particular type consructor
4533 into a single generic instance declaration.)
4539 A generic method declaration must give a case for each of the three type constructors.
4545 The type for a generic method can be built only from:
4547 <ListItem> <Para> Function arrows </Para> </ListItem>
4548 <ListItem> <Para> Type variables </Para> </ListItem>
4549 <ListItem> <Para> Tuples </Para> </ListItem>
4550 <ListItem> <Para> Arbitrary types not involving type variables </Para> </ListItem>
4552 Here are some example type signatures for generic methods:
4555 op2 :: Bool -> (a,Bool)
4556 op3 :: [Int] -> a -> a
4559 Here, op1, op2, op3 are OK, but op4 is rejected, because it has a type variable
4563 This restriction is an implementation restriction: we just havn't got around to
4564 implementing the necessary bidirectional maps over arbitrary type constructors.
4565 It would be relatively easy to add specific type constructors, such as Maybe and list,
4566 to the ones that are allowed.</para>
4571 In an instance declaration for a generic class, the idea is that the compiler
4572 will fill in the methods for you, based on the generic templates. However it can only
4577 The instance type is simple (a type constructor applied to type variables, as in Haskell 98).
4582 No constructor of the instance type has unboxed fields.
4586 (Of course, these things can only arise if you are already using GHC extensions.)
4587 However, you can still give an instance declarations for types which break these rules,
4588 provided you give explicit code to override any generic default methods.
4596 The option <Option>-ddump-deriv</Option> dumps incomprehensible stuff giving details of
4597 what the compiler does with generic declarations.
4602 <Sect2> <Title> Another example </Title>
4604 Just to finish with, here's another example I rather like:
4608 nCons {| Unit |} _ = 1
4609 nCons {| a :*: b |} _ = 1
4610 nCons {| a :+: b |} _ = nCons (bot::a) + nCons (bot::b)
4613 tag {| Unit |} _ = 1
4614 tag {| a :*: b |} _ = 1
4615 tag {| a :+: b |} (Inl x) = tag x
4616 tag {| a :+: b |} (Inr y) = nCons (bot::a) + tag y
4623 ;;; Local Variables: ***
4625 ;;; sgml-parent-document: ("users_guide.sgml" "book" "chapter" "sect1") ***
projects / ghc-hetmet.git / blob