1 <?xml version="1.0" encoding="iso-8859-1"?>
2 <chapter id="using-ghc">
3 <title>Using GHC</title>
5 <indexterm><primary>GHC, using</primary></indexterm>
6 <indexterm><primary>using GHC</primary></indexterm>
9 <title>Options overview</title>
11 <para>GHC's behaviour is controlled by
12 <firstterm>options</firstterm>, which for historical reasons are
13 also sometimes referred to as command-line flags or arguments.
14 Options can be specified in three ways:</para>
17 <title>command-line arguments</title>
19 <indexterm><primary>structure, command-line</primary></indexterm>
20 <indexterm><primary>command-line</primary><secondary>arguments</secondary></indexterm>
21 <indexterm><primary>arguments</primary><secondary>command-line</secondary></indexterm>
23 <para>An invocation of GHC takes the following form:</para>
29 <para>command-line arguments are either options or file names.</para>
31 <para>command-line options begin with <literal>-</literal>.
32 They may <emphasis>not</emphasis> be grouped:
33 <option>-vO</option> is different from <option>-v -O</option>.
34 Options need not precede filenames: e.g., <literal>ghc *.o -o
35 foo</literal>. All options are processed and then applied to
36 all files; you cannot, for example, invoke <literal>ghc -c -O1
37 Foo.hs -O2 Bar.hs</literal> to apply different optimisation
38 levels to the files <filename>Foo.hs</filename> and
39 <filename>Bar.hs</filename>.</para>
42 <sect2 id="source-file-options">
43 <title>command line options in source files</title>
45 <indexterm><primary>source-file options</primary></indexterm>
47 <para>Sometimes it is useful to make the connection between a
48 source file and the command-line options it requires quite
49 tight. For instance, if a Haskell source file uses GHC
50 extensions, it will always need to be compiled with the
51 <option>-fglasgow-exts</option> option. Rather than maintaining
52 the list of per-file options in a <filename>Makefile</filename>,
53 it is possible to do this directly in the source file using the
54 <literal>OPTIONS</literal> pragma <indexterm><primary>OPTIONS
55 pragma</primary></indexterm>:</para>
58 {-# OPTIONS -fglasgow-exts #-}
63 <para><literal>OPTIONS</literal> pragmas are only looked for at
64 the top of your source files, upto the first
65 (non-literate,non-empty) line not containing
66 <literal>OPTIONS</literal>. Multiple <literal>OPTIONS</literal>
67 pragmas are recognised. Note that your command shell does not
68 get to the source file options, they are just included literally
69 in the array of command-line arguments the compiler driver
70 maintains internally, so you'll be desperately disappointed if
71 you try to glob etc. inside <literal>OPTIONS</literal>.</para>
73 <para>NOTE: the contents of OPTIONS are prepended to the
74 command-line options, so you <emphasis>do</emphasis> have the
75 ability to override OPTIONS settings via the command
78 <para>It is not recommended to move all the contents of your
79 Makefiles into your source files, but in some circumstances, the
80 <literal>OPTIONS</literal> pragma is the Right Thing. (If you
81 use <option>-keep-hc-file-too</option> and have OPTION flags in
82 your module, the OPTIONS will get put into the generated .hc
87 <title>Setting options in GHCi</title>
89 <para>Options may also be modified from within GHCi, using the
90 <literal>:set</literal> command. See <xref linkend="ghci-set"/>
91 for more details.</para>
95 <sect1 id="static-dynamic-flags">
96 <title>Static vs. Dynamic options</title>
97 <indexterm><primary>static</primary><secondary>options</secondary>
99 <indexterm><primary>dynamic</primary><secondary>options</secondary>
102 <para>Each of GHC's command line options is classified as either
103 <firstterm>static</firstterm> or <firstterm>dynamic</firstterm>.
104 A static flag may only be specified on the command line, whereas a
105 dynamic flag may also be given in an <literal>OPTIONS</literal>
106 pragma in a source file or set from the GHCi command-line with
107 <literal>:set</literal>.</para>
109 <para>As a rule of thumb, options which relate to filenames are
110 static, and the rest are dynamic. The flag reference tables (<xref
111 linkend="flag-reference"/>) lists the status of each flag.</para>
114 <sect1 id="file-suffixes">
115 <title>Meaningful file suffixes</title>
117 <indexterm><primary>suffixes, file</primary></indexterm>
118 <indexterm><primary>file suffixes for GHC</primary></indexterm>
120 <para>File names with “meaningful” suffixes (e.g.,
121 <filename>.lhs</filename> or <filename>.o</filename>) cause the
122 “right thing” to happen to those files.</para>
128 <filename>.lhs</filename>
129 <indexterm><primary><literal>lhs</literal> suffix</primary></indexterm>
132 <para>A “literate Haskell” module.</para>
137 <term><filename>.hs</filename></term>
139 <para>A not-so-literate Haskell module.</para>
144 <term><filename>.hi</filename></term>
146 <para>A Haskell interface file, probably
147 compiler-generated.</para>
152 <term><filename>.hc</filename></term>
154 <para>Intermediate C file produced by the Haskell
160 <term><filename>.c</filename></term>
162 <para>A C file not produced by the Haskell
168 <term><filename>.s</filename></term>
170 <para>An assembly-language source file, usually produced by
176 <term><filename>.o</filename></term>
178 <para>An object file, produced by an assembler.</para>
183 <para>Files with other suffixes (or without suffixes) are passed
184 straight to the linker.</para>
189 <title>Modes of operation</title>
191 <para>GHC's behaviour is firstly controlled by a mode flag. Only
192 one of these flags may be given, but it does not necessarily need
193 to be the first option on the command-line. The available modes
199 <cmdsynopsis><command>ghc</command>
200 <arg choice='plain'>––interactive</arg>
202 <indexterm><primary>interactive mode</primary></indexterm>
203 <indexterm><primary>ghci</primary></indexterm>
206 <para>Interactive mode, which is also available as
207 <command>ghci</command>. Interactive mode is described in
208 more detail in <xref linkend="ghci"/>.</para>
214 <cmdsynopsis><command>ghc</command>
215 <arg choice='plain'>––make</arg>
217 <indexterm><primary>make mode</primary></indexterm>
218 <indexterm><primary><option>––make</option></primary></indexterm>
221 <para>In this mode, GHC will build a multi-module Haskell
222 program automatically, figuring out dependencies for itself.
223 If you have a straightforward Haskell program, this is
224 likely to be much easier, and faster, than using
225 <command>make</command>. Make mode is described in <xref
226 linkend="make-mode"/>.</para>
232 <cmdsynopsis><command>ghc</command>
233 <arg choice='plain'>–e</arg> <arg choice='plain'><replaceable>expr</replaceable></arg>
235 <indexterm><primary>eval mode</primary></indexterm>
238 <para>Expression-evaluation mode. This is very similar to
239 interactive mode, except that there is a single expression
240 to evaluate (<replaceable>expr</replaceable>) which is given
241 on the command line. See <xref linkend="eval-mode"/> for
249 <command>ghc</command>
257 <indexterm><primary><option>-E</option></primary></indexterm>
258 <indexterm><primary><option>-C</option></primary></indexterm>
259 <indexterm><primary><option>-S</option></primary></indexterm>
260 <indexterm><primary><option>-c</option></primary></indexterm>
263 <para>This is the traditional batch-compiler mode, in which
264 GHC can compile source files one at a time, or link objects
265 together into an executable. This mode also applies if
266 there is no other mode flag specified on the command line,
267 in which case it means that the specified files should be
268 compiled and then linked to form a program. See <xref
269 linkend="options-order"/>.</para>
276 <command>ghc</command>
277 <arg choice='plain'>–M</arg>
279 <indexterm><primary>dependency-generation mode</primary></indexterm>
282 <para>Dependency-generation mode. In this mode, GHC can be
283 used to generate dependency information suitable for use in
284 a <literal>Makefile</literal>. See <xref
285 linkend="sec-makefile-dependencies"/>.</para>
292 <command>ghc</command>
293 <arg choice='plain'>––mk-dll</arg>
295 <indexterm><primary>dependency-generation mode</primary></indexterm>
298 <para>DLL-creation mode (Windows only). See <xref
299 linkend="win32-dlls-create"/>.</para>
304 <sect2 id="make-mode">
305 <title>Using <command>ghc</command> <option>––make</option></title>
306 <indexterm><primary><option>––make</option></primary></indexterm>
307 <indexterm><primary>separate compilation</primary></indexterm>
309 <para>When given the <option>––make</option> option,
310 GHC will build a multi-module Haskell program by following
311 dependencies from a single root module (usually
312 <literal>Main</literal>). For example, if your
313 <literal>Main</literal> module is in a file called
314 <filename>Main.hs</filename>, you could compile and link the
315 program like this:</para>
318 ghc ––make Main.hs
321 <para>The command line may contain any number of source file
322 names or module names; GHC will figure out all the modules in
323 the program by following the imports from these initial modules.
324 It will then attempt to compile each module which is out of
325 date, and finally if there is a <literal>Main</literal> module,
326 the program will also be linked into an executable.</para>
328 <para>The main advantages to using <literal>ghc
329 ––make</literal> over traditional
330 <literal>Makefile</literal>s are:</para>
334 <para>GHC doesn't have to be restarted for each compilation,
335 which means it can cache information between compilations.
336 Compiling a muli-module program with <literal>ghc
337 ––make</literal> can be up to twice as fast as
338 running <literal>ghc</literal> individually on each source
342 <para>You don't have to write a<literal>Makefile</literal>.</para>
343 <indexterm><primary><literal>Makefile</literal>s</primary><secondary>avoiding</secondary></indexterm>
346 <para>GHC re-calculates the dependencies each time it is
347 invoked, so the dependencies never get out of sync with the
352 <para>Any of the command-line options described in the rest of
353 this chapter can be used with
354 <option>––make</option>, but note that any options
355 you give on the command line will apply to all the source files
356 compiled, so if you want any options to apply to a single source
357 file only, you'll need to use an <literal>OPTIONS</literal>
358 pragma (see <xref linkend="source-file-options"/>).</para>
360 <para>If the program needs to be linked with additional objects
361 (say, some auxilliary C code), then the object files can be
362 given on the command line and GHC will include them when linking
363 the executable.</para>
365 <para>Note that GHC can only follow dependencies if it has the
366 source file available, so if your program includes a module for
367 which there is no source file, even if you have an object and an
368 interface file for the module, then GHC will complain. The
369 exception to this rule is for package modules, which may or may
370 not have source files.</para>
372 <para>The source files for the program don't all need to be in
373 the same directory; the <option>-i</option> option can be used
374 to add directories to the search path (see <xref
375 linkend="search-path"/>).</para>
378 <sect2 id="eval-mode">
379 <title>Expression evaluation mode</title>
381 <para>This mode is very similar to interactive mode, except that
382 there is a single expression to evaluate which is specified on
383 the command line as an argument to the <option>-e</option>
387 ghc -e <replaceable>expr</replaceable>
390 <para>Haskell source files may be named on the command line, and
391 they will be loaded exactly as in interactive mode. The
392 expression is evaluated in the context of the loaded
395 <para>For example, to load and run a Haskell program containing
396 a module <literal>Main</literal>, we might say</para>
399 ghc -e Main.main Main.hs
402 <para>or we can just use this mode to evaluate expressions in
403 the context of the <literal>Prelude</literal>:</para>
406 $ ghc -e "interact (unlines.map reverse.lines)"
412 <sect2 id="options-order">
413 <title>Batch compiler mode</title>
415 <para>In <emphasis>batch mode</emphasis>, GHC will compile one or more source files
416 given on the command line.</para>
418 <para>The first phase to run is determined by each input-file
419 suffix, and the last phase is determined by a flag. If no
420 relevant flag is present, then go all the way through linking.
421 This table summarises:</para>
425 <colspec align="left"/>
426 <colspec align="left"/>
427 <colspec align="left"/>
428 <colspec align="left"/>
432 <entry>Phase of the compilation system</entry>
433 <entry>Suffix saying “start here”</entry>
434 <entry>Flag saying “stop after”</entry>
435 <entry>(suffix of) output file</entry>
440 <entry>literate pre-processor</entry>
441 <entry><literal>.lhs</literal></entry>
443 <entry><literal>.hs</literal></entry>
447 <entry>C pre-processor (opt.) </entry>
448 <entry><literal>.hs</literal> (with
449 <option>-cpp</option>)</entry>
450 <entry><option>-E</option></entry>
451 <entry><literal>.hspp</literal></entry>
455 <entry>Haskell compiler</entry>
456 <entry><literal>.hs</literal></entry>
457 <entry><option>-C</option>, <option>-S</option></entry>
458 <entry><literal>.hc</literal>, <literal>.s</literal></entry>
462 <entry>C compiler (opt.)</entry>
463 <entry><literal>.hc</literal> or <literal>.c</literal></entry>
464 <entry><option>-S</option></entry>
465 <entry><literal>.s</literal></entry>
469 <entry>assembler</entry>
470 <entry><literal>.s</literal></entry>
471 <entry><option>-c</option></entry>
472 <entry><literal>.o</literal></entry>
476 <entry>linker</entry>
477 <entry><replaceable>other</replaceable></entry>
479 <entry><filename>a.out</filename></entry>
485 <indexterm><primary><option>-C</option></primary></indexterm>
486 <indexterm><primary><option>-E</option></primary></indexterm>
487 <indexterm><primary><option>-S</option></primary></indexterm>
488 <indexterm><primary><option>-c</option></primary></indexterm>
490 <para>Thus, a common invocation would be: </para>
493 ghc -c Foo.hs</screen>
495 <para>to compile the Haskell source file
496 <filename>Foo.hs</filename> to an object file
497 <filename>Foo.o</filename>.</para>
499 <para>Note: What the Haskell compiler proper produces depends on
500 whether a native-code generator<indexterm><primary>native-code
501 generator</primary></indexterm> is used (producing assembly
502 language) or not (producing C). See <xref
503 linkend="options-codegen"/> for more details.</para>
505 <para>Note: C pre-processing is optional, the
506 <option>-cpp</option><indexterm><primary><option>-cpp</option></primary></indexterm>
507 flag turns it on. See <xref linkend="c-pre-processor"/> for more
510 <para>Note: The option <option>-E</option><indexterm><primary>-E
511 option</primary></indexterm> runs just the pre-processing passes
512 of the compiler, dumping the result in a file. Note that this
513 differs from the previous behaviour of dumping the file to
514 standard output.</para>
518 <sect1 id="options-help">
519 <title>Help and verbosity options</title>
521 <indexterm><primary>help options</primary></indexterm>
522 <indexterm><primary>verbosity options</primary></indexterm>
527 <option>––help</option>
528 <indexterm><primary><option>––help</option></primary></indexterm>
532 <indexterm><primary><option>-?</option></primary></indexterm>
535 <para>Cause GHC to spew a long usage message to standard
536 output and then exit.</para>
543 <indexterm><primary><option>-v</option></primary></indexterm>
546 <para>The <option>-v</option> option makes GHC
547 <emphasis>verbose</emphasis>: it reports its version number
548 and shows (on stderr) exactly how it invokes each phase of
549 the compilation system. Moreover, it passes the
550 <option>-v</option> flag to most phases; each reports its
551 version number (and possibly some other information).</para>
553 <para>Please, oh please, use the <option>-v</option> option
554 when reporting bugs! Knowing that you ran the right bits in
555 the right order is always the first thing we want to
562 <option>-v</option><replaceable>n</replaceable>
563 <indexterm><primary><option>-v</option></primary></indexterm>
566 <para>To provide more control over the compiler's verbosity,
567 the <option>-v</option> flag takes an optional numeric
568 argument. Specifying <option>-v</option> on its own is
569 equivalent to <option>-v3</option>, and the other levels
570 have the following meanings:</para>
574 <term><option>-v0</option></term>
576 <para>Disable all non-essential messages (this is the
582 <term><option>-v1</option></term>
584 <para>Minimal verbosity: print one line per
585 compilation (this is the default when
586 <option>––make</option> or
587 <option>––interactive</option> is on).</para>
592 <term><option>-v2</option></term>
594 <para>Print the name of each compilation phase as it
595 is executed. (equivalent to
596 <option>-dshow-passes</option>).</para>
601 <term><option>-v3</option></term>
603 <para>The same as <option>-v2</option>, except that in
604 addition the full command line (if appropriate) for
605 each compilation phase is also printed.</para>
610 <term><option>-v4</option></term>
612 <para>The same as <option>-v3</option> except that the
613 intermediate program representation after each
614 compilation phase is also printed (excluding
615 preprocessed and C/assembly files).</para>
625 <indexterm><primary><option>-V</option></primary></indexterm>
628 <option>––version</option>
629 <indexterm><primary><option>––version</option></primary></indexterm>
632 <para>Print a one-line string including GHC's version number.</para>
638 <option>––numeric-version</option>
639 <indexterm><primary><option>––numeric-version</option></primary></indexterm>
642 <para>Print GHC's numeric version number only.</para>
648 <option>––print-libdir</option>
649 <indexterm><primary><option>––print-libdir</option></primary></indexterm>
652 <para>Print the path to GHC's library directory. This is
653 the top of the directory tree containing GHC's libraries,
654 interfaces, and include files (usually something like
655 <literal>/usr/local/lib/ghc-5.04</literal> on Unix). This
657 <literal>$libdir</literal><indexterm><primary><literal>libdir</literal></primary>
658 </indexterm>in the package configuration file (see <xref
659 linkend="packages"/>).</para>
668 <sect1 id="options-sanity">
669 <title>Warnings and sanity-checking</title>
671 <indexterm><primary>sanity-checking options</primary></indexterm>
672 <indexterm><primary>warnings</primary></indexterm>
675 <para>GHC has a number of options that select which types of
676 non-fatal error messages, otherwise known as warnings, can be
677 generated during compilation. By default, you get a standard set
678 of warnings which are generally likely to indicate bugs in your
680 <option>-fwarn-overlapping-patterns</option>,
681 <option>-fwarn-deprecations</option>,
682 <option>-fwarn-duplicate-exports</option>,
683 <option>-fwarn-missing-fields</option>, and
684 <option>-fwarn-missing-methods</option>. The following flags are
685 simple ways to select standard “packages” of warnings:
691 <term><option>-W</option>:</term>
693 <indexterm><primary>-W option</primary></indexterm>
694 <para>Provides the standard warnings plus
695 <option>-fwarn-incomplete-patterns</option>,
696 <option>-fwarn-unused-matches</option>,
697 <option>-fwarn-unused-imports</option>,
698 <option>-fwarn-misc</option>, and
699 <option>-fwarn-unused-binds</option>.</para>
704 <term><option>-w</option>:</term>
706 <indexterm><primary><option>-w</option></primary></indexterm>
707 <para>Turns off all warnings, including the standard ones.</para>
712 <term><option>-Wall</option>:</term>
714 <indexterm><primary><option>-Wall</option></primary></indexterm>
715 <para>Turns on all warning options.</para>
720 <term><option>-Werror</option>:</term>
722 <indexterm><primary><option>-Werror</option></primary></indexterm>
723 <para>Makes any warning into a fatal error. Useful so that you don't
724 miss warnings when doing batch compilation. </para>
730 <para>The full set of warning options is described below. To turn
731 off any warning, simply give the corresponding
732 <option>-fno-warn-...</option> option on the command line.</para>
737 <term><option>-fwarn-deprecations</option>:</term>
739 <indexterm><primary><option>-fwarn-deprecations</option></primary>
741 <indexterm><primary>deprecations</primary></indexterm>
742 <para>Causes a warning to be emitted when a deprecated
743 function or type is used. Entities can be marked as
744 deprecated using a pragma, see <xref
745 linkend="deprecated-pragma"/>.</para>
750 <term><option>-fwarn-duplicate-exports</option>:</term>
752 <indexterm><primary><option>-fwarn-duplicate-exports</option></primary></indexterm>
753 <indexterm><primary>duplicate exports, warning</primary></indexterm>
754 <indexterm><primary>export lists, duplicates</primary></indexterm>
756 <para>Have the compiler warn about duplicate entries in
757 export lists. This is useful information if you maintain
758 large export lists, and want to avoid the continued export
759 of a definition after you've deleted (one) mention of it in
760 the export list.</para>
762 <para>This option is on by default.</para>
767 <term><option>-fwarn-hi-shadowing</option>:</term>
769 <indexterm><primary><option>-fwarn-hi-shadowing</option></primary></indexterm>
770 <indexterm><primary>shadowing</primary>
771 <secondary>interface files</secondary></indexterm>
773 <para>Causes the compiler to emit a warning when a module or
774 interface file in the current directory is shadowing one
775 with the same module name in a library or other
781 <term><option>-fwarn-incomplete-patterns</option>:</term>
783 <indexterm><primary><option>-fwarn-incomplete-patterns</option></primary></indexterm>
784 <indexterm><primary>incomplete patterns, warning</primary></indexterm>
785 <indexterm><primary>patterns, incomplete</primary></indexterm>
787 <para>Similarly for incomplete patterns, the function
788 <function>g</function> below will fail when applied to
789 non-empty lists, so the compiler will emit a warning about
790 this when <option>-fwarn-incomplete-patterns</option> is
797 <para>This option isn't enabled be default because it can be
798 a bit noisy, and it doesn't always indicate a bug in the
799 program. However, it's generally considered good practice
800 to cover all the cases in your functions.</para>
806 <option>-fwarn-misc</option>:
807 <indexterm><primary><option>-fwarn-misc</option></primary></indexterm>
810 <para>Turns on warnings for various harmless but untidy
811 things. This currently includes: importing a type with
812 <literal>(..)</literal> when the export is abstract, and
813 listing duplicate class assertions in a qualified type.</para>
819 <option>-fwarn-missing-fields</option>:
820 <indexterm><primary><option>-fwarn-missing-fields</option></primary></indexterm>
821 <indexterm><primary>missing fields, warning</primary></indexterm>
822 <indexterm><primary>fields, missing</primary></indexterm>
826 <para>This option is on by default, and warns you whenever
827 the construction of a labelled field constructor isn't
828 complete, missing initializers for one or more fields. While
829 not an error (the missing fields are initialised with
830 bottoms), it is often an indication of a programmer error.</para>
835 <term><option>-fwarn-missing-methods</option>:</term>
837 <indexterm><primary><option>-fwarn-missing-methods</option></primary></indexterm>
838 <indexterm><primary>missing methods, warning</primary></indexterm>
839 <indexterm><primary>methods, missing</primary></indexterm>
841 <para>This option is on by default, and warns you whenever
842 an instance declaration is missing one or more methods, and
843 the corresponding class declaration has no default
844 declaration for them.</para>
845 <para>The warning is suppressed if the method name
846 begins with an underscore. Here's an example where this is useful:
849 _simpleFn :: a -> String
850 complexFn :: a -> a -> String
851 complexFn x y = ... _simpleFn ...
853 The idea is that: (a) users of the class will only call <literal>complexFn</literal>;
854 never <literal>_simpleFn</literal>; and (b)
855 instance declarations can define either <literal>complexFn</literal> or <literal>_simpleFn</literal>.
861 <term><option>-fwarn-missing-signatures</option>:</term>
863 <indexterm><primary><option>-fwarn-missing-signatures</option></primary></indexterm>
864 <indexterm><primary>type signatures, missing</primary></indexterm>
866 <para>If you would like GHC to check that every top-level
867 function/value has a type signature, use the
868 <option>-fwarn-missing-signatures</option> option. This
869 option is off by default.</para>
874 <term><option>-fwarn-name-shadowing</option>:</term>
876 <indexterm><primary><option>-fwarn-name-shadowing</option></primary></indexterm>
877 <indexterm><primary>shadowing, warning</primary></indexterm>
879 <para>This option causes a warning to be emitted whenever an
880 inner-scope value has the same name as an outer-scope value,
881 i.e. the inner value shadows the outer one. This can catch
882 typographical errors that turn into hard-to-find bugs, e.g.,
883 in the inadvertent cyclic definition <literal>let x = ... x
884 ... in</literal>.</para>
886 <para>Consequently, this option does
887 <emphasis>will</emphasis> complain about cyclic recursive
894 <option>-fwarn-overlapping-patterns</option>:
895 <indexterm><primary><option>-fwarn-overlapping-patterns</option></primary></indexterm>
896 <indexterm><primary>overlapping patterns, warning</primary></indexterm>
897 <indexterm><primary>patterns, overlapping</primary></indexterm>
900 <para>By default, the compiler will warn you if a set of
901 patterns are overlapping, i.e.,</para>
904 f :: String -> Int
910 <para>where the last pattern match in <function>f</function>
911 won't ever be reached, as the second pattern overlaps
912 it. More often than not, redundant patterns is a programmer
913 mistake/error, so this option is enabled by default.</para>
918 <term><option>-fwarn-simple-patterns</option>:</term>
920 <indexterm><primary><option>-fwarn-simple-patterns</option></primary>
922 <para>Causes the compiler to warn about lambda-bound
923 patterns that can fail, eg. <literal>\(x:xs)->...</literal>.
924 Normally, these aren't treated as incomplete patterns by
925 <option>-fwarn-incomplete-patterns</option>.</para>
926 <para>``Lambda-bound patterns'' includes all places where there is a single pattern,
927 including list comprehensions and do-notation. In these cases, a pattern-match
928 failure is quite legitimate, and triggers filtering (list comprehensions) or
929 the monad <literal>fail</literal> operation (monads). For example:
931 f :: [Maybe a] -> [a]
932 f xs = [y | Just y <- xs]
934 Switching on <option>-fwarn-simple-patterns</option> will elicit warnings about
935 these probably-innocent cases, which is why the flag is off by default. </para>
936 <para> The <literal>deriving( Read )</literal> mechanism produces monadic code with
937 pattern matches, so you will also get misleading warnings about the compiler-generated
938 code. (This is arguably a Bad Thing, but it's awkward to fix.)</para>
944 <term><option>-fwarn-type-defaults</option>:</term>
946 <indexterm><primary><option>-fwarn-type-defaults</option></primary></indexterm>
947 <indexterm><primary>defaulting mechanism, warning</primary></indexterm>
948 <para>Have the compiler warn/inform you where in your source
949 the Haskell defaulting mechanism for numeric types kicks
950 in. This is useful information when converting code from a
951 context that assumed one default into one with another,
952 e.g., the `default default' for Haskell 1.4 caused the
953 otherwise unconstrained value <constant>1</constant> to be
954 given the type <literal>Int</literal>, whereas Haskell 98
955 defaults it to <literal>Integer</literal>. This may lead to
956 differences in performance and behaviour, hence the
957 usefulness of being non-silent about this.</para>
959 <para>This warning is off by default.</para>
964 <term><option>-fwarn-unused-binds</option>:</term>
966 <indexterm><primary><option>-fwarn-unused-binds</option></primary></indexterm>
967 <indexterm><primary>unused binds, warning</primary></indexterm>
968 <indexterm><primary>binds, unused</primary></indexterm>
969 <para>Report any function definitions (and local bindings)
970 which are unused. For top-level functions, the warning is
971 only given if the binding is not exported.</para>
972 <para>A definition is regarded as "used" if (a) it is exported, or (b) it is
973 mentioned in the right hand side of another definition that is used, or (c) the
974 function it defines begins with an underscore. The last case provides a
975 way to suppress unused-binding warnings selectively. </para>
976 <para> Notice that a variable
977 is reported as unused even if it appears in the right-hand side of another
978 unused binding. </para>
983 <term><option>-fwarn-unused-imports</option>:</term>
985 <indexterm><primary><option>-fwarn-unused-imports</option></primary></indexterm>
986 <indexterm><primary>unused imports, warning</primary></indexterm>
987 <indexterm><primary>imports, unused</primary></indexterm>
989 <para>Report any modules that are explicitly imported but
990 never used. However, the form <literal>import M()</literal> is
991 never reported as an unused import, because it is a useful idiom
992 for importing instance declarations, which are anonymous in Haskell.</para>
997 <term><option>-fwarn-unused-matches</option>:</term>
999 <indexterm><primary><option>-fwarn-unused-matches</option></primary></indexterm>
1000 <indexterm><primary>unused matches, warning</primary></indexterm>
1001 <indexterm><primary>matches, unused</primary></indexterm>
1003 <para>Report all unused variables which arise from pattern
1004 matches, including patterns consisting of a single variable.
1005 For instance <literal>f x y = []</literal> would report
1006 <varname>x</varname> and <varname>y</varname> as unused. The
1007 warning is suppressed if the variable name begins with an underscore, thus:
1017 <para>If you're feeling really paranoid, the
1018 <option>-dcore-lint</option>
1019 option<indexterm><primary><option>-dcore-lint</option></primary></indexterm>
1020 is a good choice. It turns on heavyweight intra-pass
1021 sanity-checking within GHC. (It checks GHC's sanity, not
1028 <sect1 id="options-optimise">
1029 <title>Optimisation (code improvement)</title>
1031 <indexterm><primary>optimisation</primary></indexterm>
1032 <indexterm><primary>improvement, code</primary></indexterm>
1034 <para>The <option>-O*</option> options specify convenient
1035 “packages” of optimisation flags; the
1036 <option>-f*</option> options described later on specify
1037 <emphasis>individual</emphasis> optimisations to be turned on/off;
1038 the <option>-m*</option> options specify
1039 <emphasis>machine-specific</emphasis> optimisations to be turned
1042 <sect2 id="optimise-pkgs">
1043 <title><option>-O*</option>: convenient “packages” of optimisation flags.</title>
1045 <para>There are <emphasis>many</emphasis> options that affect
1046 the quality of code produced by GHC. Most people only have a
1047 general goal, something like “Compile quickly” or
1048 “Make my program run like greased lightning.” The
1049 following “packages” of optimisations (or lack
1050 thereof) should suffice.</para>
1052 <para>Note that higher optimisation levels cause more
1053 cross-module optimisation to be performed, which can have an
1054 impact on how much of your program needs to be recompiled when
1055 you change something. This is one reaosn to stick to
1056 no-optimisation when developing code.</para>
1061 <term>No <option>-O*</option>-type option specified:</term>
1062 <indexterm><primary>-O* not specified</primary></indexterm>
1064 <para>This is taken to mean: “Please compile
1065 quickly; I'm not over-bothered about compiled-code
1066 quality.” So, for example: <command>ghc -c
1067 Foo.hs</command></para>
1072 <term><option>-O0</option>:</term>
1073 <indexterm><primary><option>-O0</option></primary></indexterm>
1075 <para>Means “turn off all optimisation”,
1076 reverting to the same settings as if no
1077 <option>-O</option> options had been specified. Saying
1078 <option>-O0</option> can be useful if
1079 eg. <command>make</command> has inserted a
1080 <option>-O</option> on the command line already.</para>
1085 <term><option>-O</option> or <option>-O1</option>:</term>
1086 <indexterm><primary>-O option</primary></indexterm>
1087 <indexterm><primary>-O1 option</primary></indexterm>
1088 <indexterm><primary>optimise</primary><secondary>normally</secondary></indexterm>
1090 <para>Means: “Generate good-quality code without
1091 taking too long about it.” Thus, for example:
1092 <command>ghc -c -O Main.lhs</command></para>
1094 <para><option>-O</option> currently also implies
1095 <option>-fvia-C</option>. This may change in the
1101 <term><option>-O2</option>:</term>
1102 <indexterm><primary>-O2 option</primary></indexterm>
1103 <indexterm><primary>optimise</primary><secondary>aggressively</secondary></indexterm>
1105 <para>Means: “Apply every non-dangerous
1106 optimisation, even if it means significantly longer
1107 compile times.”</para>
1109 <para>The avoided “dangerous” optimisations
1110 are those that can make runtime or space
1111 <emphasis>worse</emphasis> if you're unlucky. They are
1112 normally turned on or off individually.</para>
1114 <para>At the moment, <option>-O2</option> is
1115 <emphasis>unlikely</emphasis> to produce better code than
1116 <option>-O</option>.</para>
1121 <term><option>-Ofile <file></option>:</term>
1122 <indexterm><primary>-Ofile <file> option</primary></indexterm>
1123 <indexterm><primary>optimising, customised</primary></indexterm>
1125 <para>(NOTE: not supported since GHC 4.x. Please ask if
1126 you're interested in this.)</para>
1128 <para>For those who need <emphasis>absolute</emphasis>
1129 control over <emphasis>exactly</emphasis> what options are
1130 used (e.g., compiler writers, sometimes :-), a list of
1131 options can be put in a file and then slurped in with
1132 <option>-Ofile</option>.</para>
1134 <para>In that file, comments are of the
1135 <literal>#</literal>-to-end-of-line variety; blank
1136 lines and most whitespace is ignored.</para>
1138 <para>Please ask if you are baffled and would like an
1139 example of <option>-Ofile</option>!</para>
1144 <para>We don't use a <option>-O*</option> flag for day-to-day
1145 work. We use <option>-O</option> to get respectable speed;
1146 e.g., when we want to measure something. When we want to go for
1147 broke, we tend to use <option>-O2 -fvia-C</option> (and we go for
1148 lots of coffee breaks).</para>
1150 <para>The easiest way to see what <option>-O</option> (etc.)
1151 “really mean” is to run with <option>-v</option>,
1152 then stand back in amazement.</para>
1155 <sect2 id="options-f">
1156 <title><option>-f*</option>: platform-independent flags</title>
1158 <indexterm><primary>-f* options (GHC)</primary></indexterm>
1159 <indexterm><primary>-fno-* options (GHC)</primary></indexterm>
1161 <para>These flags turn on and off individual optimisations.
1162 They are normally set via the <option>-O</option> options
1163 described above, and as such, you shouldn't need to set any of
1164 them explicitly (indeed, doing so could lead to unexpected
1165 results). However, there are one or two that may be of
1170 <term><option>-fexcess-precision</option>:</term>
1172 <indexterm><primary><option>-fexcess-precision</option></primary></indexterm>
1173 <para>When this option is given, intermediate floating
1174 point values can have a <emphasis>greater</emphasis>
1175 precision/range than the final type. Generally this is a
1176 good thing, but some programs may rely on the exact
1178 <literal>Float</literal>/<literal>Double</literal> values
1179 and should not use this option for their compilation.</para>
1184 <term><option>-fignore-asserts</option>:</term>
1186 <indexterm><primary><option>-fignore-asserts</option></primary></indexterm>
1187 <para>Causes GHC to ignore uses of the function
1188 <literal>Exception.assert</literal> in source code (in
1189 other words, rewriting <literal>Exception.assert p
1190 e</literal> to <literal>e</literal> (see <xref
1191 linkend="sec-assertions"/>). This flag is turned on by
1192 <option>-O</option>.
1198 <term><option>-fno-strictness</option></term>
1199 <indexterm><primary><option>-fno-strictness</option></primary>
1202 <para>Turns off the strictness analyser; sometimes it eats
1203 too many cycles.</para>
1208 <term><option>-fno-cpr-analyse</option></term>
1209 <indexterm><primary><option>-fno-cpr-analyse</option></primary>
1212 <para>Turns off the CPR (constructed product result)
1213 analysis; it is somewhat experimental.</para>
1218 <term><option>-funbox-strict-fields</option>:</term>
1220 <indexterm><primary><option>-funbox-strict-fields</option></primary></indexterm>
1221 <indexterm><primary>strict constructor fields</primary></indexterm>
1222 <indexterm><primary>constructor fields, strict</primary></indexterm>
1224 <para>This option causes all constructor fields which are
1225 marked strict (i.e. “!”) to be unboxed or
1226 unpacked if possible. It is equivalent to adding an
1227 <literal>UNPACK</literal> pragma to every strict
1228 constructor field (see <xref
1229 linkend="unpack-pragma"/>).</para>
1231 <para>This option is a bit of a sledgehammer: it might
1232 sometimes make things worse. Selectively unboxing fields
1233 by using <literal>UNPACK</literal> pragmas might be
1239 <term><option>-funfolding-update-in-place<n></option></term>
1240 <indexterm><primary><option>-funfolding-update-in-place</option></primary></indexterm>
1242 <para>Switches on an experimental "optimisation".
1243 Switching it on makes the compiler a little keener to
1244 inline a function that returns a constructor, if the
1245 context is that of a thunk.
1249 If we inlined plusInt we might get an opportunity to use
1250 update-in-place for the thunk 'x'.</para>
1255 <term><option>-funfolding-creation-threshold<n></option>:</term>
1257 <indexterm><primary><option>-funfolding-creation-threshold</option></primary></indexterm>
1258 <indexterm><primary>inlining, controlling</primary></indexterm>
1259 <indexterm><primary>unfolding, controlling</primary></indexterm>
1261 <para>(Default: 45) Governs the maximum size that GHC will
1262 allow a function unfolding to be. (An unfolding has a
1263 “size” that reflects the cost in terms of
1264 “code bloat” of expanding that unfolding at
1265 at a call site. A bigger function would be assigned a
1266 bigger cost.) </para>
1268 <para> Consequences: (a) nothing larger than this will be
1269 inlined (unless it has an INLINE pragma); (b) nothing
1270 larger than this will be spewed into an interface
1274 <para> Increasing this figure is more likely to result in longer
1275 compile times than faster code. The next option is more
1281 <term><option>-funfolding-use-threshold<n></option>:</term>
1283 <indexterm><primary><option>-funfolding-use-threshold</option></primary></indexterm>
1284 <indexterm><primary>inlining, controlling</primary></indexterm>
1285 <indexterm><primary>unfolding, controlling</primary></indexterm>
1287 <para>(Default: 8) This is the magic cut-off figure for
1288 unfolding: below this size, a function definition will be
1289 unfolded at the call-site, any bigger and it won't. The
1290 size computed for a function depends on two things: the
1291 actual size of the expression minus any discounts that
1292 apply (see <option>-funfolding-con-discount</option>).</para>
1303 <sect1 id="sec-using-concurrent">
1304 <title>Using Concurrent Haskell</title>
1306 <indexterm><primary>Concurrent Haskell—use</primary></indexterm>
1309 GHC supports Concurrent Haskell by default, without requiring a
1310 special option or libraries compiled in a certain way. To get access
1311 to the support libraries for Concurrent Haskell, just import
1312 <literal>Control.Concurrent</literal> (details are in the accompanying
1313 library documentation).</para>
1316 RTS options are provided for modifying the behaviour of the threaded
1317 runtime system. See <xref linkend="parallel-rts-opts"/>.
1321 Concurrent Haskell is described in more detail in the documentation
1322 for the <literal>Control.Concurrent</literal> module.
1327 <sect1 id="sec-using-parallel">
1328 <title>Using parallel Haskell</title>
1331 <indexterm><primary>parallel Haskell—use</primary></indexterm>
1335 [You won't be able to execute parallel Haskell programs unless PVM3
1336 (parallel Virtual Machine, version 3) is installed at your site.]
1340 To compile a Haskell program for parallel execution under PVM, use the
1341 <option>-parallel</option> option,<indexterm><primary>-parallel
1342 option</primary></indexterm> both when compiling <emphasis>and
1343 linking</emphasis>. You will probably want to <literal>import
1344 parallel</literal> into your Haskell modules.
1348 To run your parallel program, once PVM is going, just invoke it
1349 “as normal”. The main extra RTS option is
1350 <option>-qp<n></option>, to say how many PVM
1351 “processors” your program to run on. (For more details of
1352 all relevant RTS options, please see <xref
1353 linkend="parallel-rts-opts"/>.)
1357 In truth, running parallel Haskell programs and getting information
1358 out of them (e.g., parallelism profiles) is a battle with the vagaries of
1359 PVM, detailed in the following sections.
1362 <sect2 id="pvm-dummies">
1363 <title>Dummy's guide to using PVM</title>
1366 <indexterm><primary>PVM, how to use</primary></indexterm>
1367 <indexterm><primary>parallel Haskell—PVM use</primary></indexterm>
1368 Before you can run a parallel program under PVM, you must set the
1369 required environment variables (PVM's idea, not ours); something like,
1370 probably in your <filename>.cshrc</filename> or equivalent:
1373 setenv PVM_ROOT /wherever/you/put/it
1374 setenv PVM_ARCH `$PVM_ROOT/lib/pvmgetarch`
1375 setenv PVM_DPATH $PVM_ROOT/lib/pvmd
1381 Creating and/or controlling your “parallel machine” is a purely-PVM
1382 business; nothing specific to parallel Haskell. The following paragraphs
1383 describe how to configure your parallel machine interactively.
1387 If you use parallel Haskell regularly on the same machine configuration it
1388 is a good idea to maintain a file with all machine names and to make the
1389 environment variable PVM_HOST_FILE point to this file. Then you can avoid
1390 the interactive operations described below by just saying
1398 You use the <command>pvm</command><indexterm><primary>pvm command</primary></indexterm> command to start PVM on your
1399 machine. You can then do various things to control/monitor your
1400 “parallel machine;” the most useful being:
1406 <colspec align="left"/>
1410 <entry><keycombo><keycap>Control</keycap><keycap>D</keycap></keycombo></entry>
1411 <entry>exit <command>pvm</command>, leaving it running</entry>
1415 <entry><command>halt</command></entry>
1416 <entry>kill off this “parallel machine” & exit</entry>
1420 <entry><command>add <host></command></entry>
1421 <entry>add <command><host></command> as a processor</entry>
1425 <entry><command>delete <host></command></entry>
1426 <entry>delete <command><host></command></entry>
1430 <entry><command>reset</command></entry>
1431 <entry>kill what's going, but leave PVM up</entry>
1435 <entry><command>conf</command></entry>
1436 <entry>list the current configuration</entry>
1440 <entry><command>ps</command></entry>
1441 <entry>report processes' status</entry>
1445 <entry><command>pstat <pid></command></entry>
1446 <entry>status of a particular process</entry>
1455 The PVM documentation can tell you much, much more about <command>pvm</command>!
1460 <sect2 id="par-profiles">
1461 <title>parallelism profiles</title>
1464 <indexterm><primary>parallelism profiles</primary></indexterm>
1465 <indexterm><primary>profiles, parallelism</primary></indexterm>
1466 <indexterm><primary>visualisation tools</primary></indexterm>
1470 With parallel Haskell programs, we usually don't care about the
1471 results—only with “how parallel” it was! We want pretty pictures.
1475 parallelism profiles (à la <command>hbcpp</command>) can be generated with the
1476 <option>-qP</option><indexterm><primary>-qP RTS option (concurrent, parallel)</primary></indexterm> RTS option. The
1477 per-processor profiling info is dumped into files named
1478 <filename><full-path><program>.gr</filename>. These are then munged into a PostScript picture,
1479 which you can then display. For example, to run your program
1480 <filename>a.out</filename> on 8 processors, then view the parallelism profile, do:
1486 <prompt>$</prompt> ./a.out +RTS -qP -qp8
1487 <prompt>$</prompt> grs2gr *.???.gr > temp.gr # combine the 8 .gr files into one
1488 <prompt>$</prompt> gr2ps -O temp.gr # cvt to .ps; output in temp.ps
1489 <prompt>$</prompt> ghostview -seascape temp.ps # look at it!
1495 The scripts for processing the parallelism profiles are distributed
1496 in <filename>ghc/utils/parallel/</filename>.
1502 <title>Other useful info about running parallel programs</title>
1505 The “garbage-collection statistics” RTS options can be useful for
1506 seeing what parallel programs are doing. If you do either
1507 <option>+RTS -Sstderr</option><indexterm><primary>-Sstderr RTS option</primary></indexterm> or <option>+RTS -sstderr</option>, then
1508 you'll get mutator, garbage-collection, etc., times on standard
1509 error. The standard error of all PE's other than the `main thread'
1510 appears in <filename>/tmp/pvml.nnn</filename>, courtesy of PVM.
1514 Whether doing <option>+RTS -Sstderr</option> or not, a handy way to watch
1515 what's happening overall is: <command>tail -f /tmp/pvml.nnn</command>.
1520 <sect2 id="parallel-rts-opts">
1521 <title>RTS options for Concurrent/parallel Haskell
1525 <indexterm><primary>RTS options, concurrent</primary></indexterm>
1526 <indexterm><primary>RTS options, parallel</primary></indexterm>
1527 <indexterm><primary>Concurrent Haskell—RTS options</primary></indexterm>
1528 <indexterm><primary>parallel Haskell—RTS options</primary></indexterm>
1532 Besides the usual runtime system (RTS) options
1533 (<xref linkend="runtime-control"/>), there are a few options particularly
1534 for concurrent/parallel execution.
1541 <term><option>-qp<N></option>:</term>
1544 <indexterm><primary>-qp<N> RTS option</primary></indexterm>
1545 (paraLLEL ONLY) Use <literal><N></literal> PVM processors to run this program;
1551 <term><option>-C[<us>]</option>:</term>
1554 <indexterm><primary>-C<us> RTS option</primary></indexterm> Sets
1555 the context switch interval to <literal><s></literal> seconds.
1556 A context switch will occur at the next heap block allocation after
1557 the timer expires (a heap block allocation occurs every 4k of
1558 allocation). With <option>-C0</option> or <option>-C</option>,
1559 context switches will occur as often as possible (at every heap block
1560 allocation). By default, context switches occur every 20ms
1561 milliseconds. Note that GHC's internal timer ticks every 20ms, and
1562 the context switch timer is always a multiple of this timer, so 20ms
1563 is the maximum granularity available for timed context switches.
1568 <term><option>-q[v]</option>:</term>
1571 <indexterm><primary>-q RTS option</primary></indexterm>
1572 (paraLLEL ONLY) Produce a quasi-parallel profile of thread activity,
1573 in the file <filename><program>.qp</filename>. In the style of <command>hbcpp</command>, this profile
1574 records the movement of threads between the green (runnable) and red
1575 (blocked) queues. If you specify the verbose suboption (<option>-qv</option>), the
1576 green queue is split into green (for the currently running thread
1577 only) and amber (for other runnable threads). We do not recommend
1578 that you use the verbose suboption if you are planning to use the
1579 <command>hbcpp</command> profiling tools or if you are context switching at every heap
1580 check (with <option>-C</option>).
1586 <term><option>-qt<num></option>:</term>
1589 <indexterm><primary>-qt<num> RTS option</primary></indexterm>
1590 (paraLLEL ONLY) Limit the thread pool size, i.e. the number of concurrent
1591 threads per processor to <literal><num></literal>. The default is
1592 32. Each thread requires slightly over 1K <emphasis>words</emphasis> in
1593 the heap for thread state and stack objects. (For 32-bit machines, this
1594 translates to 4K bytes, and for 64-bit machines, 8K bytes.)
1600 <term><option>-d</option>:</term>
1603 <indexterm><primary>-d RTS option (parallel)</primary></indexterm>
1604 (paraLLEL ONLY) Turn on debugging. It pops up one xterm (or GDB, or
1605 something…) per PVM processor. We use the standard <command>debugger</command>
1606 script that comes with PVM3, but we sometimes meddle with the
1607 <command>debugger2</command> script. We include ours in the GHC distribution,
1608 in <filename>ghc/utils/pvm/</filename>.
1614 <term><option>-qe<num></option>:</term>
1617 <indexterm><primary>-qe<num> RTS option
1618 (parallel)</primary></indexterm> (paraLLEL ONLY) Limit the spark pool size
1619 i.e. the number of pending sparks per processor to
1620 <literal><num></literal>. The default is 100. A larger number may be
1621 appropriate if your program generates large amounts of parallelism
1627 <term><option>-qQ<num></option>:</term>
1630 <indexterm><primary>-qQ<num> RTS option (parallel)</primary></indexterm>
1631 (paraLLEL ONLY) Set the size of packets transmitted between processors
1632 to <literal><num></literal>. The default is 1024 words. A larger number may be
1633 appropriate if your machine has a high communication cost relative to
1639 <term><option>-qh<num></option>:</term>
1642 <indexterm><primary>-qh<num> RTS option (parallel)</primary></indexterm>
1643 (paraLLEL ONLY) Select a packing scheme. Set the number of non-root thunks to pack in one packet to
1644 <num>-1 (0 means infinity). By default GUM uses full-subgraph
1645 packing, i.e. the entire subgraph with the requested closure as root is
1646 transmitted (provided it fits into one packet). Choosing a smaller value
1647 reduces the amount of pre-fetching of work done in GUM. This can be
1648 advantageous for improving data locality but it can also worsen the balance
1649 of the load in the system.
1654 <term><option>-qg<num></option>:</term>
1657 <indexterm><primary>-qg<num> RTS option
1658 (parallel)</primary></indexterm> (paraLLEL ONLY) Select a globalisation
1659 scheme. This option affects the
1660 generation of global addresses when transferring data. Global addresses are
1661 globally unique identifiers required to maintain sharing in the distributed
1662 graph structure. Currently this is a binary option. With <num>=0 full globalisation is used
1663 (default). This means a global address is generated for every closure that
1664 is transmitted. With <num>=1 a thunk-only globalisation scheme is
1665 used, which generated global address only for thunks. The latter case may
1666 lose sharing of data but has a reduced overhead in packing graph structures
1667 and maintaining internal tables of global addresses.
1678 <sect1 id="options-platform">
1679 <title>Platform-specific Flags</title>
1681 <indexterm><primary>-m* options</primary></indexterm>
1682 <indexterm><primary>platform-specific options</primary></indexterm>
1683 <indexterm><primary>machine-specific options</primary></indexterm>
1685 <para>Some flags only make sense for particular target
1691 <term><option>-mv8</option>:</term>
1693 <para>(SPARC machines)<indexterm><primary>-mv8 option (SPARC
1694 only)</primary></indexterm> Means to pass the like-named
1695 option to GCC; it says to use the Version 8 SPARC
1696 instructions, notably integer multiply and divide. The
1697 similiar <option>-m*</option> GCC options for SPARC also
1698 work, actually.</para>
1703 <term><option>-monly-[32]-regs</option>:</term>
1705 <para>(iX86 machines)<indexterm><primary>-monly-N-regs
1706 option (iX86 only)</primary></indexterm> GHC tries to
1707 “steal” four registers from GCC, for performance
1708 reasons; it almost always works. However, when GCC is
1709 compiling some modules with four stolen registers, it will
1710 crash, probably saying:
1713 Foo.hc:533: fixed or forbidden register was spilled.
1714 This may be due to a compiler bug or to impossible asm
1715 statements or clauses.
1718 Just give some registers back with
1719 <option>-monly-N-regs</option>. Try `3' first, then `2'.
1720 If `2' doesn't work, please report the bug to us.</para>
1729 <sect1 id="ext-core">
1730 <title>Generating and compiling External Core Files</title>
1732 <indexterm><primary>intermediate code generation</primary></indexterm>
1734 <para>GHC can dump its optimized intermediate code (said to be in “Core” format)
1735 to a file as a side-effect of compilation. Core files, which are given the suffix
1736 <filename>.hcr</filename>, can be read and processed by non-GHC back-end
1737 tools. The Core format is formally described in <ulink url="http://www.haskell.org/ghc/docs/papers/core.ps.gz">
1738 <citetitle>An External Representation for the GHC Core Language</citetitle></ulink>,
1739 and sample tools (in Haskell)
1740 for manipulating Core files are available in the GHC source distribution
1741 directory <literal>/fptools/ghc/utils/ext-core</literal>.
1742 Note that the format of <literal>.hcr</literal>
1743 files is <emphasis>different</emphasis> (though similar) to the Core output format generated
1744 for debugging purposes (<xref linkend="options-debugging"/>).</para>
1746 <para>The Core format natively supports notes which you can add to
1747 your source code using the <literal>CORE</literal> pragma (see <xref
1748 linkend="pragmas"/>).</para>
1753 <term><option>-fext-core</option></term>
1755 <primary><option>-fext-core</option></primary>
1758 <para>Generate <literal>.hcr</literal> files.</para>
1764 <para>GHC can also read in External Core files as source; just give the <literal>.hcr</literal> file on
1765 the command line, instead of the <literal>.hs</literal> or <literal>.lhs</literal> Haskell source.
1766 A current infelicity is that you need to give teh <literal>-fglasgow-exts</literal> flag too, because
1767 ordinary Haskell 98, when translated to External Core, uses things like rank-2 types.</para>
1776 ;;; Local Variables: ***
1778 ;;; sgml-parent-document: ("users_guide.xml" "book" "chapter") ***