2 <indexterm><primary>language, GHC</primary></indexterm>
3 <indexterm><primary>extensions, GHC</primary></indexterm>
4 As with all known Haskell systems, GHC implements some extensions to
5 the language. To use them, you'll need to give a <option>-fglasgow-exts</option>
6 <indexterm><primary>-fglasgow-exts option</primary></indexterm> option.
10 Virtually all of the Glasgow extensions serve to give you access to
11 the underlying facilities with which we implement Haskell. Thus, you
12 can get at the Raw Iron, if you are willing to write some non-standard
13 code at a more primitive level. You need not be “stuck” on
14 performance because of the implementation costs of Haskell's
15 “high-level” features—you can always code “under” them. In an extreme case, you can write all your time-critical code in C, and then just glue it together with Haskell!
19 Executive summary of our extensions:
25 <term>Unboxed types and primitive operations:</Term>
27 <para>You can get right down to the raw machine types and
28 operations; included in this are “primitive
29 arrays” (direct access to Big Wads of Bytes). Please
30 see <XRef LinkEnd="glasgow-unboxed"> and following.</para>
35 <term>Type system extensions:</term>
37 <para> GHC supports a large number of extensions to Haskell's
38 type system. Specifically:</para>
42 <term>Multi-parameter type classes:</term>
44 <para><xref LinkEnd="multi-param-type-classes"></para>
49 <term>Functional dependencies:</term>
51 <para><xref LinkEnd="functional-dependencies"></para>
56 <term>Implicit parameters:</term>
58 <para><xref LinkEnd="implicit-parameters"></para>
63 <term>Local universal quantification:</term>
65 <para><xref LinkEnd="universal-quantification"></para>
70 <term>Extistentially quantification in data types:</term>
72 <para><xref LinkEnd="existential-quantification"></para>
77 <term>Scoped type variables:</term>
79 <para>Scoped type variables enable the programmer to
80 supply type signatures for some nested declarations,
81 where this would not be legal in Haskell 98. Details in
82 <xref LinkEnd="scoped-type-variables">.</para>
90 <term>Pattern guards</term>
92 <para>Instead of being a boolean expression, a guard is a list
93 of qualifiers, exactly as in a list comprehension. See <xref
94 LinkEnd="pattern-guards">.</para>
99 <term>Foreign calling:</term>
101 <para>Just what it sounds like. We provide
102 <emphasis>lots</emphasis> of rope that you can dangle around
103 your neck. Please see <xref LinkEnd="ffi">.</para>
110 <para>Pragmas are special instructions to the compiler placed
111 in the source file. The pragmas GHC supports are described in
112 <xref LinkEnd="pragmas">.</para>
117 <term>Rewrite rules:</term>
119 <para>The programmer can specify rewrite rules as part of the
120 source program (in a pragma). GHC applies these rewrite rules
121 wherever it can. Details in <xref
122 LinkEnd="rewrite-rules">.</para>
127 <term>Generic classes:</term>
129 <para>Generic class declarations allow you to define a class
130 whose methods say how to work over an arbitrary data type.
131 Then it's really easy to make any new type into an instance of
132 the class. This generalises the rather ad-hoc "deriving"
133 feature of Haskell 98. Details in <xref
134 LinkEnd="generic-classes">.</para>
140 Before you get too carried away working at the lowest level (e.g.,
141 sloshing <literal>MutableByteArray#</literal>s around your
142 program), you may wish to check if there are libraries that provide a
143 “Haskellised veneer” over the features you want. See
144 <xref linkend="book-hslibs">.
147 <sect1 id="options-language">
148 <title>Language options</title>
150 <indexterm><primary>language</primary><secondary>option</secondary>
152 <indexterm><primary>options</primary><secondary>language</secondary>
154 <indexterm><primary>extensions</primary><secondary>options controlling</secondary>
157 <para> These flags control what variation of the language are
158 permitted. Leaving out all of them gives you standard Haskell
164 <term><option>-fglasgow-exts</option>:</term>
165 <indexterm><primary><option>-fglasgow-exts</option></primary></indexterm>
167 <para>This simultaneously enables all of the extensions to
168 Haskell 98 described in <xref
169 linkend="ghc-language-features">, except where otherwise
175 <term><option>-fno-monomorphism-restriction</option>:</term>
176 <indexterm><primary><option>-fno-monomorphism-restriction</option></primary></indexterm>
178 <para> Switch off the Haskell 98 monomorphism restriction.
179 Independent of the <option>-fglasgow-exts</option>
185 <term><option>-fallow-overlapping-instances</option></term>
186 <term><option>-fallow-undecidable-instances</option></term>
187 <term><option>-fcontext-stack</option></term>
188 <indexterm><primary><option>-fallow-overlapping-instances</option></primary></indexterm>
189 <indexterm><primary><option>-fallow-undecidable-instances</option></primary></indexterm>
190 <indexterm><primary><option>-fcontext-stack</option></primary></indexterm>
192 <para> See <xref LinkEnd="instance-decls">. Only relevant
193 if you also use <option>-fglasgow-exts</option>.</para>
198 <term><option>-finline-phase</option></term>
199 <indexterm><primary><option>-finline-phase</option></primary></indexterm>
201 <para>See <xref LinkEnd="rewrite-rules">. Only relevant if
202 you also use <option>-fglasgow-exts</option>.</para>
207 <term><option>-fgenerics</option></term>
208 <indexterm><primary><option>-fgenerics</option></primary></indexterm>
210 <para>See <xref LinkEnd="generic-classes">. Independent of
211 <option>-fglasgow-exts</option>.</para>
216 <term><option>-fno-implicit-prelude</option></term>
218 <para><indexterm><primary>-fno-implicit-prelude
219 option</primary></indexterm> GHC normally imports
220 <filename>Prelude.hi</filename> files for you. If you'd
221 rather it didn't, then give it a
222 <option>-fno-implicit-prelude</option> option. The idea
223 is that you can then import a Prelude of your own. (But
224 don't call it <literal>Prelude</literal>; the Haskell
225 module namespace is flat, and you must not conflict with
226 any Prelude module.)</para>
228 <para>Even though you have not imported the Prelude, all
229 the built-in syntax still refers to the built-in Haskell
230 Prelude types and values, as specified by the Haskell
231 Report. For example, the type <literal>[Int]</literal>
232 still means <literal>Prelude.[] Int</literal>; tuples
233 continue to refer to the standard Prelude tuples; the
234 translation for list comprehensions continues to use
235 <literal>Prelude.map</literal> etc.</para>
237 <para> With one group of exceptions! You may want to
238 define your own numeric class hierarchy. It completely
239 defeats that purpose if the literal "1" means
240 "<literal>Prelude.fromInteger 1</literal>", which is what
241 the Haskell Report specifies. So the
242 <option>-fno-implicit-prelude</option> flag causes the
243 following pieces of built-in syntax to refer to <emphasis>whatever
244 is in scope</emphasis>, not the Prelude versions:</para>
248 <para>Integer and fractional literals mean
249 "<literal>fromInteger 1</literal>" and
250 "<literal>fromRational 3.2</literal>", not the
251 Prelude-qualified versions; both in expressions and in
256 <para>Negation (e.g. "<literal>- (f x)</literal>")
257 means "<literal>negate (f x)</literal>" (not
258 <literal>Prelude.negate</literal>).</para>
262 <para>In an n+k pattern, the standard Prelude
263 <literal>Ord</literal> class is still used for comparison,
264 but the necessary subtraction uses whatever
265 "<literal>(-)</literal>" is in scope (not
266 "<literal>Prelude.(-)</literal>").</para>
270 <para>Note: Negative literals, such as <literal>-3</literal>, are
271 specified by (a careful reading of) the Haskell Report as
272 meaning <literal>Prelude.negate (Prelude.fromInteger 3)</literal>.
273 However, GHC deviates from this slightly, and treats them as meaning
274 <literal>fromInteger (-3)</literal>. One particular effect of this
275 slightly-non-standard reading is that there is no difficulty with
276 the literal <literal>-2147483648</literal> at type <literal>Int</literal>;
277 it means <literal>fromInteger (-2147483648)</literal>. The strict interpretation
278 would be <literal>negate (fromInteger 2147483648)</literal>,
279 and the call to <literal>fromInteger</literal> would overflow
280 (at type <literal>Int</literal>, remember).
289 <!-- UNBOXED TYPES AND PRIMITIVE OPERATIONS -->
292 <sect1 id="glasgow-ST-monad">
293 <title>Primitive state-transformer monad</title>
296 <indexterm><primary>state transformers (Glasgow extensions)</primary></indexterm>
297 <indexterm><primary>ST monad (Glasgow extension)</primary></indexterm>
301 This monad underlies our implementation of arrays, mutable and
302 immutable, and our implementation of I/O, including “C calls”.
306 The <literal>ST</literal> library, which provides access to the
307 <function>ST</function> monad, is described in <xref
313 <sect1 id="glasgow-prim-arrays">
314 <title>Primitive arrays, mutable and otherwise
318 <indexterm><primary>primitive arrays (Glasgow extension)</primary></indexterm>
319 <indexterm><primary>arrays, primitive (Glasgow extension)</primary></indexterm>
323 GHC knows about quite a few flavours of Large Swathes of Bytes.
327 First, GHC distinguishes between primitive arrays of (boxed) Haskell
328 objects (type <literal>Array# obj</literal>) and primitive arrays of bytes (type
329 <literal>ByteArray#</literal>).
333 Second, it distinguishes between…
337 <term>Immutable:</term>
340 Arrays that do not change (as with “standard” Haskell arrays); you
341 can only read from them. Obviously, they do not need the care and
342 attention of the state-transformer monad.
347 <term>Mutable:</term>
350 Arrays that may be changed or “mutated.” All the operations on them
351 live within the state-transformer monad and the updates happen
352 <emphasis>in-place</emphasis>.
357 <term>“Static” (in C land):</term>
360 A C routine may pass an <literal>Addr#</literal> pointer back into Haskell land. There
361 are then primitive operations with which you may merrily grab values
362 over in C land, by indexing off the “static” pointer.
367 <term>“Stable” pointers:</term>
370 If, for some reason, you wish to hand a Haskell pointer (i.e.,
371 <emphasis>not</emphasis> an unboxed value) to a C routine, you first make the
372 pointer “stable,” so that the garbage collector won't forget that it
373 exists. That is, GHC provides a safe way to pass Haskell pointers to
378 Please see <xref LinkEnd="sec-stable-pointers"> for more details.
383 <term>“Foreign objects”:</term>
386 A “foreign object” is a safe way to pass an external object (a
387 C-allocated pointer, say) to Haskell and have Haskell do the Right
388 Thing when it no longer references the object. So, for example, C
389 could pass a large bitmap over to Haskell and say “please free this
390 memory when you're done with it.”
394 Please see <xref LinkEnd="sec-ForeignObj"> for more details.
402 The libraries documentatation gives more details on all these
403 “primitive array” types and the operations on them.
409 <sect1 id="pattern-guards">
410 <title>Pattern guards</title>
413 <indexterm><primary>Pattern guards (Glasgow extension)</primary></indexterm>
414 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.)
418 Suppose we have an abstract data type of finite maps, with a
422 lookup :: FiniteMap -> Int -> Maybe Int
425 The lookup returns <function>Nothing</function> if the supplied key is not in the domain of the mapping, and <function>(Just v)</function> otherwise,
426 where <VarName>v</VarName> is the value that the key maps to. Now consider the following definition:
430 clunky env var1 var2 | ok1 && ok2 = val1 + val2
431 | otherwise = var1 + var2
442 The auxiliary functions are
446 maybeToBool :: Maybe a -> Bool
447 maybeToBool (Just x) = True
448 maybeToBool Nothing = False
450 expectJust :: Maybe a -> a
451 expectJust (Just x) = x
452 expectJust Nothing = error "Unexpected Nothing"
456 What is <function>clunky</function> doing? The guard <literal>ok1 &&
457 ok2</literal> checks that both lookups succeed, using
458 <function>maybeToBool</function> to convert the <function>Maybe</function>
459 types to booleans. The (lazily evaluated) <function>expectJust</function>
460 calls extract the values from the results of the lookups, and binds the
461 returned values to <VarName>val1</VarName> and <VarName>val2</VarName>
462 respectively. If either lookup fails, then clunky takes the
463 <literal>otherwise</literal> case and returns the sum of its arguments.
467 This is certainly legal Haskell, but it is a tremendously verbose and
468 un-obvious way to achieve the desired effect. Arguably, a more direct way
469 to write clunky would be to use case expressions:
473 clunky env var1 var1 = case lookup env var1 of
475 Just val1 -> case lookup env var2 of
477 Just val2 -> val1 + val2
483 This is a bit shorter, but hardly better. Of course, we can rewrite any set
484 of pattern-matching, guarded equations as case expressions; that is
485 precisely what the compiler does when compiling equations! The reason that
486 Haskell provides guarded equations is because they allow us to write down
487 the cases we want to consider, one at a time, independently of each other.
488 This structure is hidden in the case version. Two of the right-hand sides
489 are really the same (<function>fail</function>), and the whole expression
490 tends to become more and more indented.
494 Here is how I would write clunky:
499 | Just val1 <- lookup env var1
500 , Just val2 <- lookup env var2
502 ...other equations for clunky...
506 The semantics should be clear enough. The qualifers are matched in order.
507 For a <literal><-</literal> qualifier, which I call a pattern guard, the
508 right hand side is evaluated and matched against the pattern on the left.
509 If the match fails then the whole guard fails and the next equation is
510 tried. If it succeeds, then the appropriate binding takes place, and the
511 next qualifier is matched, in the augmented environment. Unlike list
512 comprehensions, however, the type of the expression to the right of the
513 <literal><-</literal> is the same as the type of the pattern to its
514 left. The bindings introduced by pattern guards scope over all the
515 remaining guard qualifiers, and over the right hand side of the equation.
519 Just as with list comprehensions, boolean expressions can be freely mixed
520 with among the pattern guards. For example:
531 Haskell's current guards therefore emerge as a special case, in which the
532 qualifier list has just one element, a boolean expression.
537 <title>The foreign interface</title>
539 <para>The foreign interface consists of the following components:</para>
543 <para>The Foreign Function Interface language specification
544 (included in this manual, in <xref linkend="ffi">).</para>
548 <para>The <literal>Foreign</literal> module (see <xref
549 linkend="sec-Foreign">) collects together several interfaces
550 which are useful in specifying foreign language
551 interfaces, including the following:</para>
555 <para>The <literal>ForeignObj</literal> module (see <xref
556 linkend="sec-ForeignObj">), for managing pointers from
557 Haskell into the outside world.</para>
561 <para>The <literal>StablePtr</literal> module (see <xref
562 linkend="sec-stable-pointers">), for managing pointers
563 into Haskell from the outside world.</para>
567 <para>The <literal>CTypes</literal> module (see <xref
568 linkend="sec-CTypes">) gives Haskell equivalents for the
569 standard C datatypes, for use in making Haskell bindings
570 to existing C libraries.</para>
574 <para>The <literal>CTypesISO</literal> module (see <xref
575 linkend="sec-CTypesISO">) gives Haskell equivalents for C
576 types defined by the ISO C standard.</para>
580 <para>The <literal>Storable</literal> library, for
581 primitive marshalling of data types between Haskell and
582 the foreign language.</para>
589 <para>The following sections also give some hints and tips on the use
590 of the foreign function interface in GHC.</para>
592 <sect2 id="glasgow-foreign-headers">
593 <title>Using function headers
597 <indexterm><primary>C calls, function headers</primary></indexterm>
601 When generating C (using the <option>-fvia-C</option> directive), one can assist the
602 C compiler in detecting type errors by using the <Command>-#include</Command> directive
603 to provide <filename>.h</filename> files containing function headers.
615 void initialiseEFS (HsInt size);
616 HsInt terminateEFS (void);
617 HsForeignObj emptyEFS(void);
618 HsForeignObj updateEFS (HsForeignObj a, HsInt i, HsInt x);
619 HsInt lookupEFS (HsForeignObj a, HsInt i);
623 <para>The types <literal>HsInt</literal>,
624 <literal>HsForeignObj</literal> etc. are described in <xref
625 linkend="sec-mapping-table">.</para>
627 <para>Note that this approach is only
628 <emphasis>essential</emphasis> for returning
629 <literal>float</literal>s (or if <literal>sizeof(int) !=
630 sizeof(int *)</literal> on your architecture) but is a Good
631 Thing for anyone who cares about writing solid code. You're
632 crazy not to do it.</para>
638 <sect1 id="multi-param-type-classes">
639 <title>Multi-parameter type classes
643 This section documents GHC's implementation of multi-parameter type
644 classes. There's lots of background in the paper <ULink
645 URL="http://research.microsoft.com/~simonpj/multi.ps.gz" >Type
646 classes: exploring the design space</ULink > (Simon Peyton Jones, Mark
651 I'd like to thank people who reported shorcomings in the GHC 3.02
652 implementation. Our default decisions were all conservative ones, and
653 the experience of these heroic pioneers has given useful concrete
654 examples to support several generalisations. (These appear below as
655 design choices not implemented in 3.02.)
659 I've discussed these notes with Mark Jones, and I believe that Hugs
660 will migrate towards the same design choices as I outline here.
661 Thanks to him, and to many others who have offered very useful
669 There are the following restrictions on the form of a qualified
676 forall tv1..tvn (c1, ...,cn) => type
682 (Here, I write the "foralls" explicitly, although the Haskell source
683 language omits them; in Haskell 1.4, all the free type variables of an
684 explicit source-language type signature are universally quantified,
685 except for the class type variables in a class declaration. However,
686 in GHC, you can give the foralls if you want. See <xref LinkEnd="universal-quantification">).
695 <emphasis>Each universally quantified type variable
696 <literal>tvi</literal> must be mentioned (i.e. appear free) in <literal>type</literal></emphasis>.
698 The reason for this is that a value with a type that does not obey
699 this restriction could not be used without introducing
700 ambiguity. Here, for example, is an illegal type:
704 forall a. Eq a => Int
708 When a value with this type was used, the constraint <literal>Eq tv</literal>
709 would be introduced where <literal>tv</literal> is a fresh type variable, and
710 (in the dictionary-translation implementation) the value would be
711 applied to a dictionary for <literal>Eq tv</literal>. The difficulty is that we
712 can never know which instance of <literal>Eq</literal> to use because we never
713 get any more information about <literal>tv</literal>.
720 <emphasis>Every constraint <literal>ci</literal> must mention at least one of the
721 universally quantified type variables <literal>tvi</literal></emphasis>.
723 For example, this type is OK because <literal>C a b</literal> mentions the
724 universally quantified type variable <literal>b</literal>:
728 forall a. C a b => burble
732 The next type is illegal because the constraint <literal>Eq b</literal> does not
733 mention <literal>a</literal>:
737 forall a. Eq b => burble
741 The reason for this restriction is milder than the other one. The
742 excluded types are never useful or necessary (because the offending
743 context doesn't need to be witnessed at this point; it can be floated
744 out). Furthermore, floating them out increases sharing. Lastly,
745 excluding them is a conservative choice; it leaves a patch of
746 territory free in case we need it later.
756 These restrictions apply to all types, whether declared in a type signature
761 Unlike Haskell 1.4, constraints in types do <emphasis>not</emphasis> have to be of
762 the form <emphasis>(class type-variables)</emphasis>. Thus, these type signatures
769 f :: Eq (m a) => [m a] -> [m a]
776 This choice recovers principal types, a property that Haskell 1.4 does not have.
782 <title>Class declarations</title>
790 <emphasis>Multi-parameter type classes are permitted</emphasis>. For example:
794 class Collection c a where
795 union :: c a -> c a -> c a
806 <emphasis>The class hierarchy must be acyclic</emphasis>. However, the definition
807 of "acyclic" involves only the superclass relationships. For example,
813 op :: D b => a -> b -> b
816 class C a => D a where { ... }
820 Here, <literal>C</literal> is a superclass of <literal>D</literal>, but it's OK for a
821 class operation <literal>op</literal> of <literal>C</literal> to mention <literal>D</literal>. (It
822 would not be OK for <literal>D</literal> to be a superclass of <literal>C</literal>.)
829 <emphasis>There are no restrictions on the context in a class declaration
830 (which introduces superclasses), except that the class hierarchy must
831 be acyclic</emphasis>. So these class declarations are OK:
835 class Functor (m k) => FiniteMap m k where
838 class (Monad m, Monad (t m)) => Transform t m where
839 lift :: m a -> (t m) a
848 <emphasis>In the signature of a class operation, every constraint
849 must mention at least one type variable that is not a class type
856 class Collection c a where
857 mapC :: Collection c b => (a->b) -> c a -> c b
861 is OK because the constraint <literal>(Collection a b)</literal> mentions
862 <literal>b</literal>, even though it also mentions the class variable
863 <literal>a</literal>. On the other hand:
868 op :: Eq a => (a,b) -> (a,b)
872 is not OK because the constraint <literal>(Eq a)</literal> mentions on the class
873 type variable <literal>a</literal>, but not <literal>b</literal>. However, any such
874 example is easily fixed by moving the offending context up to the
879 class Eq a => C a where
884 A yet more relaxed rule would allow the context of a class-op signature
885 to mention only class type variables. However, that conflicts with
886 Rule 1(b) for types above.
893 <emphasis>The type of each class operation must mention <emphasis>all</emphasis> of
894 the class type variables</emphasis>. For example:
900 insert :: s -> a -> s
904 is not OK, because the type of <literal>empty</literal> doesn't mention
905 <literal>a</literal>. This rule is a consequence of Rule 1(a), above, for
906 types, and has the same motivation.
908 Sometimes, offending class declarations exhibit misunderstandings. For
909 example, <literal>Coll</literal> might be rewritten
915 insert :: s a -> a -> s a
919 which makes the connection between the type of a collection of
920 <literal>a</literal>'s (namely <literal>(s a)</literal>) and the element type <literal>a</literal>.
921 Occasionally this really doesn't work, in which case you can split the
929 class CollE s => Coll s a where
930 insert :: s -> a -> s
943 <sect2 id="instance-decls">
944 <title>Instance declarations</title>
952 <emphasis>Instance declarations may not overlap</emphasis>. The two instance
957 instance context1 => C type1 where ...
958 instance context2 => C type2 where ...
962 "overlap" if <literal>type1</literal> and <literal>type2</literal> unify
964 However, if you give the command line option
965 <option>-fallow-overlapping-instances</option><indexterm><primary>-fallow-overlapping-instances
966 option</primary></indexterm> then two overlapping instance declarations are permitted
974 EITHER <literal>type1</literal> and <literal>type2</literal> do not unify
980 OR <literal>type2</literal> is a substitution instance of <literal>type1</literal>
981 (but not identical to <literal>type1</literal>)
994 Notice that these rules
1001 make it clear which instance decl to use
1002 (pick the most specific one that matches)
1009 do not mention the contexts <literal>context1</literal>, <literal>context2</literal>
1010 Reason: you can pick which instance decl
1011 "matches" based on the type.
1018 Regrettably, GHC doesn't guarantee to detect overlapping instance
1019 declarations if they appear in different modules. GHC can "see" the
1020 instance declarations in the transitive closure of all the modules
1021 imported by the one being compiled, so it can "see" all instance decls
1022 when it is compiling <literal>Main</literal>. However, it currently chooses not
1023 to look at ones that can't possibly be of use in the module currently
1024 being compiled, in the interests of efficiency. (Perhaps we should
1025 change that decision, at least for <literal>Main</literal>.)
1032 <emphasis>There are no restrictions on the type in an instance
1033 <emphasis>head</emphasis>, except that at least one must not be a type variable</emphasis>.
1034 The instance "head" is the bit after the "=>" in an instance decl. For
1035 example, these are OK:
1039 instance C Int a where ...
1041 instance D (Int, Int) where ...
1043 instance E [[a]] where ...
1047 Note that instance heads <emphasis>may</emphasis> contain repeated type variables.
1048 For example, this is OK:
1052 instance Stateful (ST s) (MutVar s) where ...
1056 The "at least one not a type variable" restriction is to ensure that
1057 context reduction terminates: each reduction step removes one type
1058 constructor. For example, the following would make the type checker
1059 loop if it wasn't excluded:
1063 instance C a => C a where ...
1067 There are two situations in which the rule is a bit of a pain. First,
1068 if one allows overlapping instance declarations then it's quite
1069 convenient to have a "default instance" declaration that applies if
1070 something more specific does not:
1079 Second, sometimes you might want to use the following to get the
1080 effect of a "class synonym":
1084 class (C1 a, C2 a, C3 a) => C a where { }
1086 instance (C1 a, C2 a, C3 a) => C a where { }
1090 This allows you to write shorter signatures:
1102 f :: (C1 a, C2 a, C3 a) => ...
1106 I'm on the lookout for a simple rule that preserves decidability while
1107 allowing these idioms. The experimental flag
1108 <option>-fallow-undecidable-instances</option><indexterm><primary>-fallow-undecidable-instances
1109 option</primary></indexterm> lifts this restriction, allowing all the types in an
1110 instance head to be type variables.
1117 <emphasis>Unlike Haskell 1.4, instance heads may use type
1118 synonyms</emphasis>. As always, using a type synonym is just shorthand for
1119 writing the RHS of the type synonym definition. For example:
1123 type Point = (Int,Int)
1124 instance C Point where ...
1125 instance C [Point] where ...
1129 is legal. However, if you added
1133 instance C (Int,Int) where ...
1137 as well, then the compiler will complain about the overlapping
1138 (actually, identical) instance declarations. As always, type synonyms
1139 must be fully applied. You cannot, for example, write:
1144 instance Monad P where ...
1148 This design decision is independent of all the others, and easily
1149 reversed, but it makes sense to me.
1156 <emphasis>The types in an instance-declaration <emphasis>context</emphasis> must all
1157 be type variables</emphasis>. Thus
1161 instance C a b => Eq (a,b) where ...
1169 instance C Int b => Foo b where ...
1173 is not OK. Again, the intent here is to make sure that context
1174 reduction terminates.
1176 Voluminous correspondence on the Haskell mailing list has convinced me
1177 that it's worth experimenting with a more liberal rule. If you use
1178 the flag <option>-fallow-undecidable-instances</option> can use arbitrary
1179 types in an instance context. Termination is ensured by having a
1180 fixed-depth recursion stack. If you exceed the stack depth you get a
1181 sort of backtrace, and the opportunity to increase the stack depth
1182 with <option>-fcontext-stack</option><emphasis>N</emphasis>.
1195 <sect1 id="implicit-parameters">
1196 <title>Implicit parameters
1199 <para> Implicit paramters are implemented as described in
1200 "Implicit parameters: dynamic scoping with static types",
1201 J Lewis, MB Shields, E Meijer, J Launchbury,
1202 27th ACM Symposium on Principles of Programming Languages (POPL'00),
1207 There should be more documentation, but there isn't (yet). Yell if you need it.
1211 <para> You can't have an implicit parameter in the context of a class or instance
1212 declaration. For example, both these declarations are illegal:
1214 class (?x::Int) => C a where ...
1215 instance (?x::a) => Foo [a] where ...
1217 Reason: exactly which implicit parameter you pick up depends on exactly where
1218 you invoke a function. But the ``invocation'' of instance declarations is done
1219 behind the scenes by the compiler, so it's hard to figure out exactly where it is done.
1220 Easiest thing is to outlaw the offending types.</para>
1228 <sect1 id="functional-dependencies">
1229 <title>Functional dependencies
1232 <para> Functional dependencies are implemented as described by Mark Jones
1233 in "Type Classes with Functional Dependencies", Mark P. Jones,
1234 In Proceedings of the 9th European Symposium on Programming,
1235 ESOP 2000, Berlin, Germany, March 2000, Springer-Verlag LNCS 1782.
1239 There should be more documentation, but there isn't (yet). Yell if you need it.
1244 <sect1 id="universal-quantification">
1245 <title>Explicit universal quantification
1249 GHC's type system supports explicit universal quantification in
1250 constructor fields and function arguments. This is useful for things
1251 like defining <literal>runST</literal> from the state-thread world.
1252 GHC's syntax for this now agrees with Hugs's, namely:
1258 forall a b. (Ord a, Eq b) => a -> b -> a
1264 The context is, of course, optional. You can't use <literal>forall</literal> as
1265 a type variable any more!
1269 Haskell type signatures are implicitly quantified. The <literal>forall</literal>
1270 allows us to say exactly what this means. For example:
1288 g :: forall b. (b -> b)
1294 The two are treated identically.
1298 <title>Universally-quantified data type fields
1302 In a <literal>data</literal> or <literal>newtype</literal> declaration one can quantify
1303 the types of the constructor arguments. Here are several examples:
1309 data T a = T1 (forall b. b -> b -> b) a
1311 data MonadT m = MkMonad { return :: forall a. a -> m a,
1312 bind :: forall a b. m a -> (a -> m b) -> m b
1315 newtype Swizzle = MkSwizzle (Ord a => [a] -> [a])
1321 The constructors now have so-called <emphasis>rank 2</emphasis> polymorphic
1322 types, in which there is a for-all in the argument types.:
1328 T1 :: forall a. (forall b. b -> b -> b) -> a -> T a
1329 MkMonad :: forall m. (forall a. a -> m a)
1330 -> (forall a b. m a -> (a -> m b) -> m b)
1332 MkSwizzle :: (Ord a => [a] -> [a]) -> Swizzle
1338 Notice that you don't need to use a <literal>forall</literal> if there's an
1339 explicit context. For example in the first argument of the
1340 constructor <function>MkSwizzle</function>, an implicit "<literal>forall a.</literal>" is
1341 prefixed to the argument type. The implicit <literal>forall</literal>
1342 quantifies all type variables that are not already in scope, and are
1343 mentioned in the type quantified over.
1347 As for type signatures, implicit quantification happens for non-overloaded
1348 types too. So if you write this:
1351 data T a = MkT (Either a b) (b -> b)
1354 it's just as if you had written this:
1357 data T a = MkT (forall b. Either a b) (forall b. b -> b)
1360 That is, since the type variable <literal>b</literal> isn't in scope, it's
1361 implicitly universally quantified. (Arguably, it would be better
1362 to <emphasis>require</emphasis> explicit quantification on constructor arguments
1363 where that is what is wanted. Feedback welcomed.)
1369 <title>Construction </title>
1372 You construct values of types <literal>T1, MonadT, Swizzle</literal> by applying
1373 the constructor to suitable values, just as usual. For example,
1379 (T1 (\xy->x) 3) :: T Int
1381 (MkSwizzle sort) :: Swizzle
1382 (MkSwizzle reverse) :: Swizzle
1389 MkMonad r b) :: MonadT Maybe
1395 The type of the argument can, as usual, be more general than the type
1396 required, as <literal>(MkSwizzle reverse)</literal> shows. (<function>reverse</function>
1397 does not need the <literal>Ord</literal> constraint.)
1403 <title>Pattern matching</title>
1406 When you use pattern matching, the bound variables may now have
1407 polymorphic types. For example:
1413 f :: T a -> a -> (a, Char)
1414 f (T1 f k) x = (f k x, f 'c' 'd')
1416 g :: (Ord a, Ord b) => Swizzle -> [a] -> (a -> b) -> [b]
1417 g (MkSwizzle s) xs f = s (map f (s xs))
1419 h :: MonadT m -> [m a] -> m [a]
1420 h m [] = return m []
1421 h m (x:xs) = bind m x $ \y ->
1422 bind m (h m xs) $ \ys ->
1429 In the function <function>h</function> we use the record selectors <literal>return</literal>
1430 and <literal>bind</literal> to extract the polymorphic bind and return functions
1431 from the <literal>MonadT</literal> data structure, rather than using pattern
1436 You cannot pattern-match against an argument that is polymorphic.
1440 newtype TIM s a = TIM (ST s (Maybe a))
1442 runTIM :: (forall s. TIM s a) -> Maybe a
1443 runTIM (TIM m) = runST m
1449 Here the pattern-match fails, because you can't pattern-match against
1450 an argument of type <literal>(forall s. TIM s a)</literal>. Instead you
1451 must bind the variable and pattern match in the right hand side:
1454 runTIM :: (forall s. TIM s a) -> Maybe a
1455 runTIM tm = case tm of { TIM m -> runST m }
1458 The <literal>tm</literal> on the right hand side is (invisibly) instantiated, like
1459 any polymorphic value at its occurrence site, and now you can pattern-match
1466 <title>The partial-application restriction</title>
1469 There is really only one way in which data structures with polymorphic
1470 components might surprise you: you must not partially apply them.
1471 For example, this is illegal:
1477 map MkSwizzle [sort, reverse]
1483 The restriction is this: <emphasis>every subexpression of the program must
1484 have a type that has no for-alls, except that in a function
1485 application (f e1…en) the partial applications are not subject to
1486 this rule</emphasis>. The restriction makes type inference feasible.
1490 In the illegal example, the sub-expression <literal>MkSwizzle</literal> has the
1491 polymorphic type <literal>(Ord b => [b] -> [b]) -> Swizzle</literal> and is not
1492 a sub-expression of an enclosing application. On the other hand, this
1499 map (T1 (\a b -> a)) [1,2,3]
1505 even though it involves a partial application of <function>T1</function>, because
1506 the sub-expression <literal>T1 (\a b -> a)</literal> has type <literal>Int -> T
1513 <title>Type signatures
1517 Once you have data constructors with universally-quantified fields, or
1518 constants such as <Constant>runST</Constant> that have rank-2 types, it isn't long
1519 before you discover that you need more! Consider:
1525 mkTs f x y = [T1 f x, T1 f y]
1531 <function>mkTs</function> is a fuction that constructs some values of type
1532 <literal>T</literal>, using some pieces passed to it. The trouble is that since
1533 <literal>f</literal> is a function argument, Haskell assumes that it is
1534 monomorphic, so we'll get a type error when applying <function>T1</function> to
1535 it. This is a rather silly example, but the problem really bites in
1536 practice. Lots of people trip over the fact that you can't make
1537 "wrappers functions" for <Constant>runST</Constant> for exactly the same reason.
1538 In short, it is impossible to build abstractions around functions with
1543 The solution is fairly clear. We provide the ability to give a rank-2
1544 type signature for <emphasis>ordinary</emphasis> functions (not only data
1545 constructors), thus:
1551 mkTs :: (forall b. b -> b -> b) -> a -> [T a]
1552 mkTs f x y = [T1 f x, T1 f y]
1558 This type signature tells the compiler to attribute <literal>f</literal> with
1559 the polymorphic type <literal>(forall b. b -> b -> b)</literal> when type
1560 checking the body of <function>mkTs</function>, so now the application of
1561 <function>T1</function> is fine.
1565 There are two restrictions:
1574 You can only define a rank 2 type, specified by the following
1579 rank2type ::= [forall tyvars .] [context =>] funty
1580 funty ::= ([forall tyvars .] [context =>] ty) -> funty
1582 ty ::= ...current Haskell monotype syntax...
1586 Informally, the universal quantification must all be right at the beginning,
1587 or at the top level of a function argument.
1594 There is a restriction on the definition of a function whose
1595 type signature is a rank-2 type: the polymorphic arguments must be
1596 matched on the left hand side of the "<literal>=</literal>" sign. You can't
1597 define <function>mkTs</function> like this:
1601 mkTs :: (forall b. b -> b -> b) -> a -> [T a]
1602 mkTs = \ f x y -> [T1 f x, T1 f y]
1607 The same partial-application rule applies to ordinary functions with
1608 rank-2 types as applied to data constructors.
1621 <title>Type synonyms and hoisting
1625 GHC also allows you to write a <literal>forall</literal> in a type synonym, thus:
1627 type Discard a = forall b. a -> b -> a
1632 However, it is often convenient to use these sort of synonyms at the right hand
1633 end of an arrow, thus:
1635 type Discard a = forall b. a -> b -> a
1637 g :: Int -> Discard Int
1640 Simply expanding the type synonym would give
1642 g :: Int -> (forall b. Int -> b -> Int)
1644 but GHC "hoists" the <literal>forall</literal> to give the isomorphic type
1646 g :: forall b. Int -> Int -> b -> Int
1648 In general, the rule is this: <emphasis>to determine the type specified by any explicit
1649 user-written type (e.g. in a type signature), GHC expands type synonyms and then repeatedly
1650 performs the transformation:</emphasis>
1652 <emphasis>type1</emphasis> -> forall a. <emphasis>type2</emphasis>
1654 forall a. <emphasis>type1</emphasis> -> <emphasis>type2</emphasis>
1656 (In fact, GHC tries to retain as much synonym information as possible for use in
1657 error messages, but that is a usability issue.) This rule applies, of course, whether
1658 or not the <literal>forall</literal> comes from a synonym. For example, here is another
1659 valid way to write <literal>g</literal>'s type signature:
1661 g :: Int -> Int -> forall b. b -> Int
1668 <sect1 id="existential-quantification">
1669 <title>Existentially quantified data constructors
1673 The idea of using existential quantification in data type declarations
1674 was suggested by Laufer (I believe, thought doubtless someone will
1675 correct me), and implemented in Hope+. It's been in Lennart
1676 Augustsson's <Command>hbc</Command> Haskell compiler for several years, and
1677 proved very useful. Here's the idea. Consider the declaration:
1683 data Foo = forall a. MkFoo a (a -> Bool)
1690 The data type <literal>Foo</literal> has two constructors with types:
1696 MkFoo :: forall a. a -> (a -> Bool) -> Foo
1703 Notice that the type variable <literal>a</literal> in the type of <function>MkFoo</function>
1704 does not appear in the data type itself, which is plain <literal>Foo</literal>.
1705 For example, the following expression is fine:
1711 [MkFoo 3 even, MkFoo 'c' isUpper] :: [Foo]
1717 Here, <literal>(MkFoo 3 even)</literal> packages an integer with a function
1718 <function>even</function> that maps an integer to <literal>Bool</literal>; and <function>MkFoo 'c'
1719 isUpper</function> packages a character with a compatible function. These
1720 two things are each of type <literal>Foo</literal> and can be put in a list.
1724 What can we do with a value of type <literal>Foo</literal>?. In particular,
1725 what happens when we pattern-match on <function>MkFoo</function>?
1731 f (MkFoo val fn) = ???
1737 Since all we know about <literal>val</literal> and <function>fn</function> is that they
1738 are compatible, the only (useful) thing we can do with them is to
1739 apply <function>fn</function> to <literal>val</literal> to get a boolean. For example:
1746 f (MkFoo val fn) = fn val
1752 What this allows us to do is to package heterogenous values
1753 together with a bunch of functions that manipulate them, and then treat
1754 that collection of packages in a uniform manner. You can express
1755 quite a bit of object-oriented-like programming this way.
1758 <sect2 id="existential">
1759 <title>Why existential?
1763 What has this to do with <emphasis>existential</emphasis> quantification?
1764 Simply that <function>MkFoo</function> has the (nearly) isomorphic type
1770 MkFoo :: (exists a . (a, a -> Bool)) -> Foo
1776 But Haskell programmers can safely think of the ordinary
1777 <emphasis>universally</emphasis> quantified type given above, thereby avoiding
1778 adding a new existential quantification construct.
1784 <title>Type classes</title>
1787 An easy extension (implemented in <Command>hbc</Command>) is to allow
1788 arbitrary contexts before the constructor. For example:
1794 data Baz = forall a. Eq a => Baz1 a a
1795 | forall b. Show b => Baz2 b (b -> b)
1801 The two constructors have the types you'd expect:
1807 Baz1 :: forall a. Eq a => a -> a -> Baz
1808 Baz2 :: forall b. Show b => b -> (b -> b) -> Baz
1814 But when pattern matching on <function>Baz1</function> the matched values can be compared
1815 for equality, and when pattern matching on <function>Baz2</function> the first matched
1816 value can be converted to a string (as well as applying the function to it).
1817 So this program is legal:
1824 f (Baz1 p q) | p == q = "Yes"
1826 f (Baz1 v fn) = show (fn v)
1832 Operationally, in a dictionary-passing implementation, the
1833 constructors <function>Baz1</function> and <function>Baz2</function> must store the
1834 dictionaries for <literal>Eq</literal> and <literal>Show</literal> respectively, and
1835 extract it on pattern matching.
1839 Notice the way that the syntax fits smoothly with that used for
1840 universal quantification earlier.
1846 <title>Restrictions</title>
1849 There are several restrictions on the ways in which existentially-quantified
1850 constructors can be use.
1859 When pattern matching, each pattern match introduces a new,
1860 distinct, type for each existential type variable. These types cannot
1861 be unified with any other type, nor can they escape from the scope of
1862 the pattern match. For example, these fragments are incorrect:
1870 Here, the type bound by <function>MkFoo</function> "escapes", because <literal>a</literal>
1871 is the result of <function>f1</function>. One way to see why this is wrong is to
1872 ask what type <function>f1</function> has:
1876 f1 :: Foo -> a -- Weird!
1880 What is this "<literal>a</literal>" in the result type? Clearly we don't mean
1885 f1 :: forall a. Foo -> a -- Wrong!
1889 The original program is just plain wrong. Here's another sort of error
1893 f2 (Baz1 a b) (Baz1 p q) = a==q
1897 It's ok to say <literal>a==b</literal> or <literal>p==q</literal>, but
1898 <literal>a==q</literal> is wrong because it equates the two distinct types arising
1899 from the two <function>Baz1</function> constructors.
1907 You can't pattern-match on an existentially quantified
1908 constructor in a <literal>let</literal> or <literal>where</literal> group of
1909 bindings. So this is illegal:
1913 f3 x = a==b where { Baz1 a b = x }
1917 You can only pattern-match
1918 on an existentially-quantified constructor in a <literal>case</literal> expression or
1919 in the patterns of a function definition.
1921 The reason for this restriction is really an implementation one.
1922 Type-checking binding groups is already a nightmare without
1923 existentials complicating the picture. Also an existential pattern
1924 binding at the top level of a module doesn't make sense, because it's
1925 not clear how to prevent the existentially-quantified type "escaping".
1926 So for now, there's a simple-to-state restriction. We'll see how
1934 You can't use existential quantification for <literal>newtype</literal>
1935 declarations. So this is illegal:
1939 newtype T = forall a. Ord a => MkT a
1943 Reason: a value of type <literal>T</literal> must be represented as a pair
1944 of a dictionary for <literal>Ord t</literal> and a value of type <literal>t</literal>.
1945 That contradicts the idea that <literal>newtype</literal> should have no
1946 concrete representation. You can get just the same efficiency and effect
1947 by using <literal>data</literal> instead of <literal>newtype</literal>. If there is no
1948 overloading involved, then there is more of a case for allowing
1949 an existentially-quantified <literal>newtype</literal>, because the <literal>data</literal>
1950 because the <literal>data</literal> version does carry an implementation cost,
1951 but single-field existentially quantified constructors aren't much
1952 use. So the simple restriction (no existential stuff on <literal>newtype</literal>)
1953 stands, unless there are convincing reasons to change it.
1961 You can't use <literal>deriving</literal> to define instances of a
1962 data type with existentially quantified data constructors.
1964 Reason: in most cases it would not make sense. For example:#
1967 data T = forall a. MkT [a] deriving( Eq )
1970 To derive <literal>Eq</literal> in the standard way we would need to have equality
1971 between the single component of two <function>MkT</function> constructors:
1975 (MkT a) == (MkT b) = ???
1978 But <VarName>a</VarName> and <VarName>b</VarName> have distinct types, and so can't be compared.
1979 It's just about possible to imagine examples in which the derived instance
1980 would make sense, but it seems altogether simpler simply to prohibit such
1981 declarations. Define your own instances!
1993 <sect1 id="sec-assertions">
1995 <indexterm><primary>Assertions</primary></indexterm>
1999 If you want to make use of assertions in your standard Haskell code, you
2000 could define a function like the following:
2006 assert :: Bool -> a -> a
2007 assert False x = error "assertion failed!"
2014 which works, but gives you back a less than useful error message --
2015 an assertion failed, but which and where?
2019 One way out is to define an extended <function>assert</function> function which also
2020 takes a descriptive string to include in the error message and
2021 perhaps combine this with the use of a pre-processor which inserts
2022 the source location where <function>assert</function> was used.
2026 Ghc offers a helping hand here, doing all of this for you. For every
2027 use of <function>assert</function> in the user's source:
2033 kelvinToC :: Double -> Double
2034 kelvinToC k = assert (k >= 0.0) (k+273.15)
2040 Ghc will rewrite this to also include the source location where the
2047 assert pred val ==> assertError "Main.hs|15" pred val
2053 The rewrite is only performed by the compiler when it spots
2054 applications of <function>Exception.assert</function>, so you can still define and
2055 use your own versions of <function>assert</function>, should you so wish. If not,
2056 import <literal>Exception</literal> to make use <function>assert</function> in your code.
2060 To have the compiler ignore uses of assert, use the compiler option
2061 <option>-fignore-asserts</option>. <indexterm><primary>-fignore-asserts option</primary></indexterm> That is,
2062 expressions of the form <literal>assert pred e</literal> will be rewritten to <literal>e</literal>.
2066 Assertion failures can be caught, see the documentation for the
2067 <literal>Exception</literal> library (<xref linkend="sec-Exception">)
2073 <sect1 id="scoped-type-variables">
2074 <title>Scoped Type Variables
2078 A <emphasis>pattern type signature</emphasis> can introduce a <emphasis>scoped type
2079 variable</emphasis>. For example
2085 f (xs::[a]) = ys ++ ys
2094 The pattern <literal>(xs::[a])</literal> includes a type signature for <VarName>xs</VarName>.
2095 This brings the type variable <literal>a</literal> into scope; it scopes over
2096 all the patterns and right hand sides for this equation for <function>f</function>.
2097 In particular, it is in scope at the type signature for <VarName>y</VarName>.
2101 At ordinary type signatures, such as that for <VarName>ys</VarName>, any type variables
2102 mentioned in the type signature <emphasis>that are not in scope</emphasis> are
2103 implicitly universally quantified. (If there are no type variables in
2104 scope, all type variables mentioned in the signature are universally
2105 quantified, which is just as in Haskell 98.) In this case, since <VarName>a</VarName>
2106 is in scope, it is not universally quantified, so the type of <VarName>ys</VarName> is
2107 the same as that of <VarName>xs</VarName>. In Haskell 98 it is not possible to declare
2108 a type for <VarName>ys</VarName>; a major benefit of scoped type variables is that
2109 it becomes possible to do so.
2113 Scoped type variables are implemented in both GHC and Hugs. Where the
2114 implementations differ from the specification below, those differences
2119 So much for the basic idea. Here are the details.
2123 <title>Scope and implicit quantification</title>
2131 All the type variables mentioned in the patterns for a single
2132 function definition equation, that are not already in scope,
2133 are brought into scope by the patterns. We describe this set as
2134 the <emphasis>type variables bound by the equation</emphasis>.
2141 The type variables thus brought into scope may be mentioned
2142 in ordinary type signatures or pattern type signatures anywhere within
2150 In ordinary type signatures, any type variable mentioned in the
2151 signature that is in scope is <emphasis>not</emphasis> universally quantified.
2158 Ordinary type signatures do not bring any new type variables
2159 into scope (except in the type signature itself!). So this is illegal:
2168 It's illegal because <VarName>a</VarName> is not in scope in the body of <function>f</function>,
2169 so the ordinary signature <literal>x::a</literal> is equivalent to <literal>x::forall a.a</literal>;
2170 and that is an incorrect typing.
2177 There is no implicit universal quantification on pattern type
2178 signatures, nor may one write an explicit <literal>forall</literal> type in a pattern
2179 type signature. The pattern type signature is a monotype.
2187 The type variables in the head of a <literal>class</literal> or <literal>instance</literal> declaration
2188 scope over the methods defined in the <literal>where</literal> part. For example:
2202 (Not implemented in Hugs yet, Dec 98).
2213 <title>Polymorphism</title>
2221 Pattern type signatures are completely orthogonal to ordinary, separate
2222 type signatures. The two can be used independently or together. There is
2223 no scoping associated with the names of the type variables in a separate type signature.
2228 f (xs::[b]) = reverse xs
2237 The function must be polymorphic in the type variables
2238 bound by all its equations. Operationally, the type variables bound
2239 by one equation must not:
2246 Be unified with a type (such as <literal>Int</literal>, or <literal>[a]</literal>).
2252 Be unified with a type variable free in the environment.
2258 Be unified with each other. (They may unify with the type variables
2259 bound by another equation for the same function, of course.)
2266 For example, the following all fail to type check:
2270 f (x::a) (y::b) = [x,y] -- a unifies with b
2272 g (x::a) = x + 1::Int -- a unifies with Int
2274 h x = let k (y::a) = [x,y] -- a is free in the
2275 in k x -- environment
2277 k (x::a) True = ... -- a unifies with Int
2278 k (x::Int) False = ...
2281 w (x::a) = x -- a unifies with [b]
2290 The pattern-bound type variable may, however, be constrained
2291 by the context of the principal type, thus:
2295 f (x::a) (y::a) = x+y*2
2299 gets the inferred type: <literal>forall a. Num a => a -> a -> a</literal>.
2310 <title>Result type signatures</title>
2318 The result type of a function can be given a signature,
2323 f (x::a) :: [a] = [x,x,x]
2327 The final <literal>:: [a]</literal> after all the patterns gives a signature to the
2328 result type. Sometimes this is the only way of naming the type variable
2333 f :: Int -> [a] -> [a]
2334 f n :: ([a] -> [a]) = let g (x::a, y::a) = (y,x)
2335 in \xs -> map g (reverse xs `zip` xs)
2347 Result type signatures are not yet implemented in Hugs.
2353 <title>Pattern signatures on other constructs</title>
2361 A pattern type signature can be on an arbitrary sub-pattern, not
2366 f ((x,y)::(a,b)) = (y,x) :: (b,a)
2375 Pattern type signatures, including the result part, can be used
2376 in lambda abstractions:
2380 (\ (x::a, y) :: a -> x)
2384 Type variables bound by these patterns must be polymorphic in
2385 the sense defined above.
2390 f1 (x::c) = f1 x -- ok
2391 f2 = \(x::c) -> f2 x -- not ok
2395 Here, <function>f1</function> is OK, but <function>f2</function> is not, because <VarName>c</VarName> gets unified
2396 with a type variable free in the environment, in this
2397 case, the type of <function>f2</function>, which is in the environment when
2398 the lambda abstraction is checked.
2405 Pattern type signatures, including the result part, can be used
2406 in <literal>case</literal> expressions:
2410 case e of { (x::a, y) :: a -> x }
2414 The pattern-bound type variables must, as usual,
2415 be polymorphic in the following sense: each case alternative,
2416 considered as a lambda abstraction, must be polymorphic.
2421 case (True,False) of { (x::a, y) -> x }
2425 Even though the context is that of a pair of booleans,
2426 the alternative itself is polymorphic. Of course, it is
2431 case (True,False) of { (x::Bool, y) -> x }
2440 To avoid ambiguity, the type after the “<literal>::</literal>” in a result
2441 pattern signature on a lambda or <literal>case</literal> must be atomic (i.e. a single
2442 token or a parenthesised type of some sort). To see why,
2443 consider how one would parse this:
2456 Pattern type signatures that bind new type variables
2457 may not be used in pattern bindings at all.
2462 f x = let (y, z::a) = x in ...
2466 But these are OK, because they do not bind fresh type variables:
2470 f1 x = let (y, z::Int) = x in ...
2471 f2 (x::(Int,a)) = let (y, z::a) = x in ...
2475 However a single variable is considered a degenerate function binding,
2476 rather than a degerate pattern binding, so this is permitted, even
2477 though it binds a type variable:
2481 f :: (b->b) = \(x::b) -> x
2490 Such degnerate function bindings do not fall under the monomorphism
2497 g :: a -> a -> Bool = \x y. x==y
2503 Here <function>g</function> has type <literal>forall a. Eq a => a -> a -> Bool</literal>, just as if
2504 <function>g</function> had a separate type signature. Lacking a type signature, <function>g</function>
2505 would get a monomorphic type.
2511 <title>Existentials</title>
2519 Pattern type signatures can bind existential type variables.
2524 data T = forall a. MkT [a]
2527 f (MkT [t::a]) = MkT t3
2544 <sect1 id="pragmas">
2549 GHC supports several pragmas, or instructions to the compiler placed
2550 in the source code. Pragmas don't affect the meaning of the program,
2551 but they might affect the efficiency of the generated code.
2554 <sect2 id="inline-pragma">
2555 <title>INLINE pragma
2557 <indexterm><primary>INLINE pragma</primary></indexterm>
2558 <indexterm><primary>pragma, INLINE</primary></indexterm></title>
2561 GHC (with <option>-O</option>, as always) tries to inline (or “unfold”)
2562 functions/values that are “small enough,” thus avoiding the call
2563 overhead and possibly exposing other more-wonderful optimisations.
2567 You will probably see these unfoldings (in Core syntax) in your
2572 Normally, if GHC decides a function is “too expensive” to inline, it
2573 will not do so, nor will it export that unfolding for other modules to
2578 The sledgehammer you can bring to bear is the
2579 <literal>INLINE</literal><indexterm><primary>INLINE pragma</primary></indexterm> pragma, used thusly:
2582 key_function :: Int -> String -> (Bool, Double)
2584 #ifdef __GLASGOW_HASKELL__
2585 {-# INLINE key_function #-}
2589 (You don't need to do the C pre-processor carry-on unless you're going
2590 to stick the code through HBC—it doesn't like <literal>INLINE</literal> pragmas.)
2594 The major effect of an <literal>INLINE</literal> pragma is to declare a function's
2595 “cost” to be very low. The normal unfolding machinery will then be
2596 very keen to inline it.
2600 An <literal>INLINE</literal> pragma for a function can be put anywhere its type
2601 signature could be put.
2605 <literal>INLINE</literal> pragmas are a particularly good idea for the
2606 <literal>then</literal>/<literal>return</literal> (or <literal>bind</literal>/<literal>unit</literal>) functions in a monad.
2607 For example, in GHC's own <literal>UniqueSupply</literal> monad code, we have:
2610 #ifdef __GLASGOW_HASKELL__
2611 {-# INLINE thenUs #-}
2612 {-# INLINE returnUs #-}
2620 <sect2 id="noinline-pragma">
2621 <title>NOINLINE pragma
2625 <indexterm><primary>NOINLINE pragma</primary></indexterm>
2626 <indexterm><primary>pragma, NOINLINE</primary></indexterm>
2630 The <literal>NOINLINE</literal> pragma does exactly what you'd expect: it stops the
2631 named function from being inlined by the compiler. You shouldn't ever
2632 need to do this, unless you're very cautious about code size.
2637 <sect2 id="specialize-pragma">
2638 <title>SPECIALIZE pragma</title>
2640 <indexterm><primary>SPECIALIZE pragma</primary></indexterm>
2641 <indexterm><primary>pragma, SPECIALIZE</primary></indexterm>
2642 <indexterm><primary>overloading, death to</primary></indexterm>
2644 <para>(UK spelling also accepted.) For key overloaded
2645 functions, you can create extra versions (NB: more code space)
2646 specialised to particular types. Thus, if you have an
2647 overloaded function:</para>
2650 hammeredLookup :: Ord key => [(key, value)] -> key -> value
2653 <para>If it is heavily used on lists with
2654 <literal>Widget</literal> keys, you could specialise it as
2658 {-# SPECIALIZE hammeredLookup :: [(Widget, value)] -> Widget -> value #-}
2661 <para>To get very fancy, you can also specify a named function
2662 to use for the specialised value, as in:</para>
2665 {-# RULES hammeredLookup = blah #-}
2668 <para>where <literal>blah</literal> is an implementation of
2669 <literal>hammerdLookup</literal> written specialy for
2670 <literal>Widget</literal> lookups. It's <emphasis>Your
2671 Responsibility</emphasis> to make sure that
2672 <function>blah</function> really behaves as a specialised
2673 version of <function>hammeredLookup</function>!!!</para>
2675 <para>Note we use the <literal>RULE</literal> pragma here to
2676 indicate that <literal>hammeredLookup</literal> applied at a
2677 certain type should be replaced by <literal>blah</literal>. See
2678 <xref linkend="rules"> for more information on
2679 <literal>RULES</literal>.</para>
2681 <para>An example in which using <literal>RULES</literal> for
2682 specialisation will Win Big:
2685 toDouble :: Real a => a -> Double
2686 toDouble = fromRational . toRational
2688 {-# SPECIALIZE toDouble :: Int -> Double = i2d #-}
2689 i2d (I# i) = D# (int2Double# i) -- uses Glasgow prim-op directly
2692 The <function>i2d</function> function is virtually one machine
2693 instruction; the default conversion—via an intermediate
2694 <literal>Rational</literal>—is obscenely expensive by
2697 <para>A <literal>SPECIALIZE</literal> pragma for a function can
2698 be put anywhere its type signature could be put.</para>
2702 <sect2 id="specialize-instance-pragma">
2703 <title>SPECIALIZE instance pragma
2707 <indexterm><primary>SPECIALIZE pragma</primary></indexterm>
2708 <indexterm><primary>overloading, death to</primary></indexterm>
2709 Same idea, except for instance declarations. For example:
2712 instance (Eq a) => Eq (Foo a) where { ... usual stuff ... }
2714 {-# SPECIALIZE instance Eq (Foo [(Int, Bar)] #-}
2717 Compatible with HBC, by the way.
2722 <sect2 id="line-pragma">
2727 <indexterm><primary>LINE pragma</primary></indexterm>
2728 <indexterm><primary>pragma, LINE</primary></indexterm>
2732 This pragma is similar to C's <literal>#line</literal> pragma, and is mainly for use in
2733 automatically generated Haskell code. It lets you specify the line
2734 number and filename of the original code; for example
2740 {-# LINE 42 "Foo.vhs" #-}
2746 if you'd generated the current file from something called <filename>Foo.vhs</filename>
2747 and this line corresponds to line 42 in the original. GHC will adjust
2748 its error messages to refer to the line/file named in the <literal>LINE</literal>
2755 <title>RULES pragma</title>
2758 The RULES pragma lets you specify rewrite rules. It is described in
2759 <xref LinkEnd="rewrite-rules">.
2766 <sect1 id="rewrite-rules">
2767 <title>Rewrite rules
2769 <indexterm><primary>RULES pagma</primary></indexterm>
2770 <indexterm><primary>pragma, RULES</primary></indexterm>
2771 <indexterm><primary>rewrite rules</primary></indexterm></title>
2774 The programmer can specify rewrite rules as part of the source program
2775 (in a pragma). GHC applies these rewrite rules wherever it can.
2783 "map/map" forall f g xs. map f (map g xs) = map (f.g) xs
2790 <title>Syntax</title>
2793 From a syntactic point of view:
2799 Each rule has a name, enclosed in double quotes. The name itself has
2800 no significance at all. It is only used when reporting how many times the rule fired.
2806 There may be zero or more rules in a <literal>RULES</literal> pragma.
2812 Layout applies in a <literal>RULES</literal> pragma. Currently no new indentation level
2813 is set, so you must lay out your rules starting in the same column as the
2814 enclosing definitions.
2820 Each variable mentioned in a rule must either be in scope (e.g. <function>map</function>),
2821 or bound by the <literal>forall</literal> (e.g. <function>f</function>, <function>g</function>, <function>xs</function>). The variables bound by
2822 the <literal>forall</literal> are called the <emphasis>pattern</emphasis> variables. They are separated
2823 by spaces, just like in a type <literal>forall</literal>.
2829 A pattern variable may optionally have a type signature.
2830 If the type of the pattern variable is polymorphic, it <emphasis>must</emphasis> have a type signature.
2831 For example, here is the <literal>foldr/build</literal> rule:
2834 "fold/build" forall k z (g::forall b. (a->b->b) -> b -> b) .
2835 foldr k z (build g) = g k z
2838 Since <function>g</function> has a polymorphic type, it must have a type signature.
2845 The left hand side of a rule must consist of a top-level variable applied
2846 to arbitrary expressions. For example, this is <emphasis>not</emphasis> OK:
2849 "wrong1" forall e1 e2. case True of { True -> e1; False -> e2 } = e1
2850 "wrong2" forall f. f True = True
2853 In <literal>"wrong1"</literal>, the LHS is not an application; in <literal>"wrong2"</literal>, the LHS has a pattern variable
2860 A rule does not need to be in the same module as (any of) the
2861 variables it mentions, though of course they need to be in scope.
2867 Rules are automatically exported from a module, just as instance declarations are.
2878 <title>Semantics</title>
2881 From a semantic point of view:
2887 Rules are only applied if you use the <option>-O</option> flag.
2893 Rules are regarded as left-to-right rewrite rules.
2894 When GHC finds an expression that is a substitution instance of the LHS
2895 of a rule, it replaces the expression by the (appropriately-substituted) RHS.
2896 By "a substitution instance" we mean that the LHS can be made equal to the
2897 expression by substituting for the pattern variables.
2904 The LHS and RHS of a rule are typechecked, and must have the
2912 GHC makes absolutely no attempt to verify that the LHS and RHS
2913 of a rule have the same meaning. That is undecideable in general, and
2914 infeasible in most interesting cases. The responsibility is entirely the programmer's!
2921 GHC makes no attempt to make sure that the rules are confluent or
2922 terminating. For example:
2925 "loop" forall x,y. f x y = f y x
2928 This rule will cause the compiler to go into an infinite loop.
2935 If more than one rule matches a call, GHC will choose one arbitrarily to apply.
2941 GHC currently uses a very simple, syntactic, matching algorithm
2942 for matching a rule LHS with an expression. It seeks a substitution
2943 which makes the LHS and expression syntactically equal modulo alpha
2944 conversion. The pattern (rule), but not the expression, is eta-expanded if
2945 necessary. (Eta-expanding the epression can lead to laziness bugs.)
2946 But not beta conversion (that's called higher-order matching).
2950 Matching is carried out on GHC's intermediate language, which includes
2951 type abstractions and applications. So a rule only matches if the
2952 types match too. See <xref LinkEnd="rule-spec"> below.
2958 GHC keeps trying to apply the rules as it optimises the program.
2959 For example, consider:
2968 The expression <literal>s (t xs)</literal> does not match the rule <literal>"map/map"</literal>, but GHC
2969 will substitute for <VarName>s</VarName> and <VarName>t</VarName>, giving an expression which does match.
2970 If <VarName>s</VarName> or <VarName>t</VarName> was (a) used more than once, and (b) large or a redex, then it would
2971 not be substituted, and the rule would not fire.
2978 In the earlier phases of compilation, GHC inlines <emphasis>nothing
2979 that appears on the LHS of a rule</emphasis>, because once you have substituted
2980 for something you can't match against it (given the simple minded
2981 matching). So if you write the rule
2984 "map/map" forall f,g. map f . map g = map (f.g)
2987 this <emphasis>won't</emphasis> match the expression <literal>map f (map g xs)</literal>.
2988 It will only match something written with explicit use of ".".
2989 Well, not quite. It <emphasis>will</emphasis> match the expression
2995 where <function>wibble</function> is defined:
2998 wibble f g = map f . map g
3001 because <function>wibble</function> will be inlined (it's small).
3003 Later on in compilation, GHC starts inlining even things on the
3004 LHS of rules, but still leaves the rules enabled. This inlining
3005 policy is controlled by the per-simplification-pass flag <option>-finline-phase</option><emphasis>n</emphasis>.
3012 All rules are implicitly exported from the module, and are therefore
3013 in force in any module that imports the module that defined the rule, directly
3014 or indirectly. (That is, if A imports B, which imports C, then C's rules are
3015 in force when compiling A.) The situation is very similar to that for instance
3027 <title>List fusion</title>
3030 The RULES mechanism is used to implement fusion (deforestation) of common list functions.
3031 If a "good consumer" consumes an intermediate list constructed by a "good producer", the
3032 intermediate list should be eliminated entirely.
3036 The following are good producers:
3048 Enumerations of <literal>Int</literal> and <literal>Char</literal> (e.g. <literal>['a'..'z']</literal>).
3054 Explicit lists (e.g. <literal>[True, False]</literal>)
3060 The cons constructor (e.g <literal>3:4:[]</literal>)
3066 <function>++</function>
3072 <function>map</function>
3078 <function>filter</function>
3084 <function>iterate</function>, <function>repeat</function>
3090 <function>zip</function>, <function>zipWith</function>
3099 The following are good consumers:
3111 <function>array</function> (on its second argument)
3117 <function>length</function>
3123 <function>++</function> (on its first argument)
3129 <function>map</function>
3135 <function>filter</function>
3141 <function>concat</function>
3147 <function>unzip</function>, <function>unzip2</function>, <function>unzip3</function>, <function>unzip4</function>
3153 <function>zip</function>, <function>zipWith</function> (but on one argument only; if both are good producers, <function>zip</function>
3154 will fuse with one but not the other)
3160 <function>partition</function>
3166 <function>head</function>
3172 <function>and</function>, <function>or</function>, <function>any</function>, <function>all</function>
3178 <function>sequence_</function>
3184 <function>msum</function>
3190 <function>sortBy</function>
3199 So, for example, the following should generate no intermediate lists:
3202 array (1,10) [(i,i*i) | i <- map (+ 1) [0..9]]
3208 This list could readily be extended; if there are Prelude functions that you use
3209 a lot which are not included, please tell us.
3213 If you want to write your own good consumers or producers, look at the
3214 Prelude definitions of the above functions to see how to do so.
3219 <sect2 id="rule-spec">
3220 <title>Specialisation
3224 Rewrite rules can be used to get the same effect as a feature
3225 present in earlier version of GHC:
3228 {-# SPECIALIZE fromIntegral :: Int8 -> Int16 = int8ToInt16 #-}
3231 This told GHC to use <function>int8ToInt16</function> instead of <function>fromIntegral</function> whenever
3232 the latter was called with type <literal>Int8 -> Int16</literal>. That is, rather than
3233 specialising the original definition of <function>fromIntegral</function> the programmer is
3234 promising that it is safe to use <function>int8ToInt16</function> instead.
3238 This feature is no longer in GHC. But rewrite rules let you do the
3243 "fromIntegral/Int8/Int16" fromIntegral = int8ToInt16
3247 This slightly odd-looking rule instructs GHC to replace <function>fromIntegral</function>
3248 by <function>int8ToInt16</function> <emphasis>whenever the types match</emphasis>. Speaking more operationally,
3249 GHC adds the type and dictionary applications to get the typed rule
3252 forall (d1::Integral Int8) (d2::Num Int16) .
3253 fromIntegral Int8 Int16 d1 d2 = int8ToInt16
3257 this rule does not need to be in the same file as fromIntegral,
3258 unlike the <literal>SPECIALISE</literal> pragmas which currently do (so that they
3259 have an original definition available to specialise).
3265 <title>Controlling what's going on</title>
3273 Use <option>-ddump-rules</option> to see what transformation rules GHC is using.
3279 Use <option>-ddump-simpl-stats</option> to see what rules are being fired.
3280 If you add <option>-dppr-debug</option> you get a more detailed listing.
3286 The defintion of (say) <function>build</function> in <FileName>PrelBase.lhs</FileName> looks llike this:
3289 build :: forall a. (forall b. (a -> b -> b) -> b -> b) -> [a]
3290 {-# INLINE build #-}
3294 Notice the <literal>INLINE</literal>! That prevents <literal>(:)</literal> from being inlined when compiling
3295 <literal>PrelBase</literal>, so that an importing module will “see” the <literal>(:)</literal>, and can
3296 match it on the LHS of a rule. <literal>INLINE</literal> prevents any inlining happening
3297 in the RHS of the <literal>INLINE</literal> thing. I regret the delicacy of this.
3304 In <filename>ghc/lib/std/PrelBase.lhs</filename> look at the rules for <function>map</function> to
3305 see how to write rules that will do fusion and yet give an efficient
3306 program even if fusion doesn't happen. More rules in <filename>PrelList.lhs</filename>.
3318 <sect1 id="generic-classes">
3319 <title>Generic classes</title>
3322 The ideas behind this extension are described in detail in "Derivable type classes",
3323 Ralf Hinze and Simon Peyton Jones, Haskell Workshop, Montreal Sept 2000, pp94-105.
3324 An example will give the idea:
3332 fromBin :: [Int] -> (a, [Int])
3334 toBin {| Unit |} Unit = []
3335 toBin {| a :+: b |} (Inl x) = 0 : toBin x
3336 toBin {| a :+: b |} (Inr y) = 1 : toBin y
3337 toBin {| a :*: b |} (x :*: y) = toBin x ++ toBin y
3339 fromBin {| Unit |} bs = (Unit, bs)
3340 fromBin {| a :+: b |} (0:bs) = (Inl x, bs') where (x,bs') = fromBin bs
3341 fromBin {| a :+: b |} (1:bs) = (Inr y, bs') where (y,bs') = fromBin bs
3342 fromBin {| a :*: b |} bs = (x :*: y, bs'') where (x,bs' ) = fromBin bs
3343 (y,bs'') = fromBin bs'
3346 This class declaration explains how <literal>toBin</literal> and <literal>fromBin</literal>
3347 work for arbitrary data types. They do so by giving cases for unit, product, and sum,
3348 which are defined thus in the library module <literal>Generics</literal>:
3352 data a :+: b = Inl a | Inr b
3353 data a :*: b = a :*: b
3356 Now you can make a data type into an instance of Bin like this:
3358 instance (Bin a, Bin b) => Bin (a,b)
3359 instance Bin a => Bin [a]
3361 That is, just leave off the "where" clasuse. Of course, you can put in the
3362 where clause and over-ride whichever methods you please.
3366 <title> Using generics </title>
3367 <para>To use generics you need to</para>
3370 <para>Use the <option>-fgenerics</option> flag.</para>
3373 <para>Import the module <literal>Generics</literal> from the
3374 <literal>lang</literal> package. This import brings into
3375 scope the data types <literal>Unit</literal>,
3376 <literal>:*:</literal>, and <literal>:+:</literal>. (You
3377 don't need this import if you don't mention these types
3378 explicitly; for example, if you are simply giving instance
3379 declarations.)</para>
3384 <sect2> <title> Changes wrt the paper </title>
3386 Note that the type constructors <literal>:+:</literal> and <literal>:*:</literal>
3387 can be written infix (indeed, you can now use
3388 any operator starting in a colon as an infix type constructor). Also note that
3389 the type constructors are not exactly as in the paper (Unit instead of 1, etc).
3390 Finally, note that the syntax of the type patterns in the class declaration
3391 uses "<literal>{|</literal>" and "<literal>{|</literal>" brackets; curly braces
3392 alone would ambiguous when they appear on right hand sides (an extension we
3393 anticipate wanting).
3397 <sect2> <title>Terminology and restrictions</title>
3399 Terminology. A "generic default method" in a class declaration
3400 is one that is defined using type patterns as above.
3401 A "polymorphic default method" is a default method defined as in Haskell 98.
3402 A "generic class declaration" is a class declaration with at least one
3403 generic default method.
3411 Alas, we do not yet implement the stuff about constructor names and
3418 A generic class can have only one parameter; you can't have a generic
3419 multi-parameter class.
3425 A default method must be defined entirely using type patterns, or entirely
3426 without. So this is illegal:
3429 op :: a -> (a, Bool)
3430 op {| Unit |} Unit = (Unit, True)
3433 However it is perfectly OK for some methods of a generic class to have
3434 generic default methods and others to have polymorphic default methods.
3440 The type variable(s) in the type pattern for a generic method declaration
3441 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:
3445 op {| p :*: q |} (x :*: y) = op (x :: p)
3453 The type patterns in a generic default method must take one of the forms:
3459 where "a" and "b" are type variables. Furthermore, all the type patterns for
3460 a single type constructor (<literal>:*:</literal>, say) must be identical; they
3461 must use the same type variables. So this is illegal:
3465 op {| a :+: b |} (Inl x) = True
3466 op {| p :+: q |} (Inr y) = False
3468 The type patterns must be identical, even in equations for different methods of the class.
3469 So this too is illegal:
3473 op {| a :*: b |} (Inl x) = True
3476 op {| p :*: q |} (Inr y) = False
3478 (The reason for this restriction is that we gather all the equations for a particular type consructor
3479 into a single generic instance declaration.)
3485 A generic method declaration must give a case for each of the three type constructors.
3491 The type for a generic method can be built only from:
3493 <listitem> <para> Function arrows </para> </listitem>
3494 <listitem> <para> Type variables </para> </listitem>
3495 <listitem> <para> Tuples </para> </listitem>
3496 <listitem> <para> Arbitrary types not involving type variables </para> </listitem>
3498 Here are some example type signatures for generic methods:
3501 op2 :: Bool -> (a,Bool)
3502 op3 :: [Int] -> a -> a
3505 Here, op1, op2, op3 are OK, but op4 is rejected, because it has a type variable
3509 This restriction is an implementation restriction: we just havn't got around to
3510 implementing the necessary bidirectional maps over arbitrary type constructors.
3511 It would be relatively easy to add specific type constructors, such as Maybe and list,
3512 to the ones that are allowed.</para>
3517 In an instance declaration for a generic class, the idea is that the compiler
3518 will fill in the methods for you, based on the generic templates. However it can only
3523 The instance type is simple (a type constructor applied to type variables, as in Haskell 98).
3528 No constructor of the instance type has unboxed fields.
3532 (Of course, these things can only arise if you are already using GHC extensions.)
3533 However, you can still give an instance declarations for types which break these rules,
3534 provided you give explicit code to override any generic default methods.
3542 The option <option>-ddump-deriv</option> dumps incomprehensible stuff giving details of
3543 what the compiler does with generic declarations.
3548 <sect2> <title> Another example </title>
3550 Just to finish with, here's another example I rather like:
3554 nCons {| Unit |} _ = 1
3555 nCons {| a :*: b |} _ = 1
3556 nCons {| a :+: b |} _ = nCons (bot::a) + nCons (bot::b)
3559 tag {| Unit |} _ = 1
3560 tag {| a :*: b |} _ = 1
3561 tag {| a :+: b |} (Inl x) = tag x
3562 tag {| a :+: b |} (Inr y) = nCons (bot::a) + tag y
3569 ;;; Local Variables: ***
3571 ;;; sgml-parent-document: ("users_guide.sgml" "book" "chapter" "sect1") ***