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_GHC</literal> pragma <indexterm><primary>OPTIONS_GHC
55 pragma</primary></indexterm>:</para>
58 {-# OPTIONS_GHC -fglasgow-exts #-}
63 <para><literal>OPTIONS_GHC</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_GHC</literal>. Multiple <literal>OPTIONS_GHC</literal>
67 pragmas are recognised. Do not put comments before, or on the same line
68 as, the <literal>OPTIONS_GHC</literal> pragma.</para>
70 <para>Note that your command shell does not
71 get to the source file options, they are just included literally
72 in the array of command-line arguments the compiler
73 maintains internally, so you'll be desperately disappointed if
74 you try to glob etc. inside <literal>OPTIONS_GHC</literal>.</para>
76 <para>NOTE: the contents of OPTIONS_GHC are prepended to the
77 command-line options, so you <emphasis>do</emphasis> have the
78 ability to override OPTIONS_GHC settings via the command
81 <para>It is not recommended to move all the contents of your
82 Makefiles into your source files, but in some circumstances, the
83 <literal>OPTIONS_GHC</literal> pragma is the Right Thing. (If you
84 use <option>-keep-hc-file-too</option> and have OPTION flags in
85 your module, the OPTIONS_GHC will get put into the generated .hc
90 <title>Setting options in GHCi</title>
92 <para>Options may also be modified from within GHCi, using the
93 <literal>:set</literal> command. See <xref linkend="ghci-set"/>
94 for more details.</para>
98 <sect1 id="static-dynamic-flags">
99 <title>Static, Dynamic, and Mode options</title>
100 <indexterm><primary>static</primary><secondary>options</secondary>
102 <indexterm><primary>dynamic</primary><secondary>options</secondary>
104 <indexterm><primary>mode</primary><secondary>options</secondary>
107 <para>Each of GHC's command line options is classified as either
108 <firstterm>static</firstterm> or <firstterm>dynamic</firstterm> or
109 <firstterm>mode</firstterm>:</para>
113 <term>Mode flags</term>
115 <para>For example, <option>--make</option> or <option>-E</option>.
116 There may be only a single mode flag on the command line. The
117 available modes are listed in <xref linkend="modes"/>.</para>
121 <term>Dynamic Flags</term>
123 <para>Most non-mode flags fall into this category. A dynamic flag
124 may be used on the command line, in a
125 <literal>GHC_OPTIONS</literal> pragma in a source file, or set
126 using <literal>:set</literal> in GHCi.</para>
130 <term>Static Flags</term>
132 <para>A few flags are "static", which means they can only be used on
133 the command-line, and remain in force over the entire GHC/GHCi
139 <para>The flag reference tables (<xref
140 linkend="flag-reference"/>) lists the status of each flag.</para>
142 <para>There are a few flags that are static except that they can
143 also be used with GHCi's <literal>:set</literal> command; these
144 are listed as “static/<literal>:set</literal>” in the
148 <sect1 id="file-suffixes">
149 <title>Meaningful file suffixes</title>
151 <indexterm><primary>suffixes, file</primary></indexterm>
152 <indexterm><primary>file suffixes for GHC</primary></indexterm>
154 <para>File names with “meaningful” suffixes (e.g.,
155 <filename>.lhs</filename> or <filename>.o</filename>) cause the
156 “right thing” to happen to those files.</para>
161 <term><filename>.hs</filename></term>
163 <para>A Haskell module.</para>
169 <filename>.lhs</filename>
170 <indexterm><primary><literal>lhs</literal> suffix</primary></indexterm>
173 <para>A “literate Haskell” module.</para>
178 <term><filename>.hi</filename></term>
180 <para>A Haskell interface file, probably
181 compiler-generated.</para>
186 <term><filename>.hc</filename></term>
188 <para>Intermediate C file produced by the Haskell
194 <term><filename>.c</filename></term>
196 <para>A C file not produced by the Haskell
202 <term><filename>.s</filename></term>
204 <para>An assembly-language source file, usually produced by
210 <term><filename>.o</filename></term>
212 <para>An object file, produced by an assembler.</para>
217 <para>Files with other suffixes (or without suffixes) are passed
218 straight to the linker.</para>
223 <title>Modes of operation</title>
225 <para>GHC's behaviour is firstly controlled by a mode flag. Only
226 one of these flags may be given, but it does not necessarily need
227 to be the first option on the command-line. The available modes
233 <cmdsynopsis><command>ghc --interactive</command>
235 <indexterm><primary>interactive mode</primary></indexterm>
236 <indexterm><primary>ghci</primary></indexterm>
239 <para>Interactive mode, which is also available as
240 <command>ghci</command>. Interactive mode is described in
241 more detail in <xref linkend="ghci"/>.</para>
247 <cmdsynopsis><command>ghc --make</command>
249 <indexterm><primary>make mode</primary></indexterm>
250 <indexterm><primary><option>––make</option></primary></indexterm>
253 <para>In this mode, GHC will build a multi-module Haskell
254 program automatically, figuring out dependencies for itself.
255 If you have a straightforward Haskell program, this is
256 likely to be much easier, and faster, than using
257 <command>make</command>. Make mode is described in <xref
258 linkend="make-mode"/>.</para>
264 <cmdsynopsis><command>ghc -e</command>
265 <arg choice='plain'><replaceable>expr</replaceable></arg>
267 <indexterm><primary>eval mode</primary></indexterm>
270 <para>Expression-evaluation mode. This is very similar to
271 interactive mode, except that there is a single expression
272 to evaluate (<replaceable>expr</replaceable>) which is given
273 on the command line. See <xref linkend="eval-mode"/> for
281 <command>ghc -E</command>
282 <command>ghc -c</command>
283 <command>ghc -S</command>
284 <command>ghc -c</command>
286 <indexterm><primary><option>-E</option></primary></indexterm>
287 <indexterm><primary><option>-C</option></primary></indexterm>
288 <indexterm><primary><option>-S</option></primary></indexterm>
289 <indexterm><primary><option>-c</option></primary></indexterm>
292 <para>This is the traditional batch-compiler mode, in which
293 GHC can compile source files one at a time, or link objects
294 together into an executable. This mode also applies if
295 there is no other mode flag specified on the command line,
296 in which case it means that the specified files should be
297 compiled and then linked to form a program. See <xref
298 linkend="options-order"/>.</para>
305 <command>ghc -M</command>
307 <indexterm><primary>dependency-generation mode</primary></indexterm>
310 <para>Dependency-generation mode. In this mode, GHC can be
311 used to generate dependency information suitable for use in
312 a <literal>Makefile</literal>. See <xref
313 linkend="makefile-dependencies"/>.</para>
320 <command>ghc --mk-dll</command>
322 <indexterm><primary>DLL-creation mode</primary></indexterm>
325 <para>DLL-creation mode (Windows only). See <xref
326 linkend="win32-dlls-create"/>.</para>
333 <command>ghc --help</command> <command>ghc -?</command>
335 <indexterm><primary><option>––help</option></primary></indexterm>
338 <para>Cause GHC to spew a long usage message to standard
339 output and then exit.</para>
346 <command>ghc --version</command>
347 <command>ghc -V</command>
349 <indexterm><primary><option>-V</option></primary></indexterm>
350 <indexterm><primary><option>––version</option></primary></indexterm>
353 <para>Print a one-line string including GHC's version number.</para>
360 <command>ghc --numeric-version</command>
362 <indexterm><primary><option>––numeric-version</option></primary></indexterm>
365 <para>Print GHC's numeric version number only.</para>
372 <command>ghc --print-libdir</command>
374 <indexterm><primary><option>––print-libdir</option></primary></indexterm>
377 <para>Print the path to GHC's library directory. This is
378 the top of the directory tree containing GHC's libraries,
379 interfaces, and include files (usually something like
380 <literal>/usr/local/lib/ghc-5.04</literal> on Unix). This
382 <literal>$libdir</literal><indexterm><primary><literal>libdir</literal></primary>
383 </indexterm>in the package configuration file (see <xref
384 linkend="packages"/>).</para>
390 <sect2 id="make-mode">
391 <title>Using <command>ghc</command> <option>––make</option></title>
392 <indexterm><primary><option>––make</option></primary></indexterm>
393 <indexterm><primary>separate compilation</primary></indexterm>
395 <para>When given the <option>––make</option> option,
396 GHC will build a multi-module Haskell program by following
397 dependencies from a single root module (usually
398 <literal>Main</literal>). For example, if your
399 <literal>Main</literal> module is in a file called
400 <filename>Main.hs</filename>, you could compile and link the
401 program like this:</para>
404 ghc ––make Main.hs
407 <para>The command line may contain any number of source file
408 names or module names; GHC will figure out all the modules in
409 the program by following the imports from these initial modules.
410 It will then attempt to compile each module which is out of
411 date, and finally if there is a <literal>Main</literal> module,
412 the program will also be linked into an executable.</para>
414 <para>The main advantages to using <literal>ghc
415 ––make</literal> over traditional
416 <literal>Makefile</literal>s are:</para>
420 <para>GHC doesn't have to be restarted for each compilation,
421 which means it can cache information between compilations.
422 Compiling a multi-module program with <literal>ghc
423 ––make</literal> can be up to twice as fast as
424 running <literal>ghc</literal> individually on each source
428 <para>You don't have to write a <literal>Makefile</literal>.</para>
429 <indexterm><primary><literal>Makefile</literal>s</primary><secondary>avoiding</secondary></indexterm>
432 <para>GHC re-calculates the dependencies each time it is
433 invoked, so the dependencies never get out of sync with the
438 <para>Any of the command-line options described in the rest of
439 this chapter can be used with
440 <option>––make</option>, but note that any options
441 you give on the command line will apply to all the source files
442 compiled, so if you want any options to apply to a single source
443 file only, you'll need to use an <literal>OPTIONS_GHC</literal>
444 pragma (see <xref linkend="source-file-options"/>).</para>
446 <para>If the program needs to be linked with additional objects
447 (say, some auxiliary C code), then the object files can be
448 given on the command line and GHC will include them when linking
449 the executable.</para>
451 <para>Note that GHC can only follow dependencies if it has the
452 source file available, so if your program includes a module for
453 which there is no source file, even if you have an object and an
454 interface file for the module, then GHC will complain. The
455 exception to this rule is for package modules, which may or may
456 not have source files.</para>
458 <para>The source files for the program don't all need to be in
459 the same directory; the <option>-i</option> option can be used
460 to add directories to the search path (see <xref
461 linkend="search-path"/>).</para>
464 <sect2 id="eval-mode">
465 <title>Expression evaluation mode</title>
467 <para>This mode is very similar to interactive mode, except that
468 there is a single expression to evaluate which is specified on
469 the command line as an argument to the <option>-e</option>
473 ghc -e <replaceable>expr</replaceable>
476 <para>Haskell source files may be named on the command line, and
477 they will be loaded exactly as in interactive mode. The
478 expression is evaluated in the context of the loaded
481 <para>For example, to load and run a Haskell program containing
482 a module <literal>Main</literal>, we might say</para>
485 ghc -e Main.main Main.hs
488 <para>or we can just use this mode to evaluate expressions in
489 the context of the <literal>Prelude</literal>:</para>
492 $ ghc -e "interact (unlines.map reverse.lines)"
498 <sect2 id="options-order">
499 <title>Batch compiler mode</title>
501 <para>In <emphasis>batch mode</emphasis>, GHC will compile one or more source files
502 given on the command line.</para>
504 <para>The first phase to run is determined by each input-file
505 suffix, and the last phase is determined by a flag. If no
506 relevant flag is present, then go all the way through linking.
507 This table summarises:</para>
511 <colspec align="left"/>
512 <colspec align="left"/>
513 <colspec align="left"/>
514 <colspec align="left"/>
518 <entry>Phase of the compilation system</entry>
519 <entry>Suffix saying “start here”</entry>
520 <entry>Flag saying “stop after”</entry>
521 <entry>(suffix of) output file</entry>
526 <entry>literate pre-processor</entry>
527 <entry><literal>.lhs</literal></entry>
529 <entry><literal>.hs</literal></entry>
533 <entry>C pre-processor (opt.) </entry>
534 <entry><literal>.hs</literal> (with
535 <option>-cpp</option>)</entry>
536 <entry><option>-E</option></entry>
537 <entry><literal>.hspp</literal></entry>
541 <entry>Haskell compiler</entry>
542 <entry><literal>.hs</literal></entry>
543 <entry><option>-C</option>, <option>-S</option></entry>
544 <entry><literal>.hc</literal>, <literal>.s</literal></entry>
548 <entry>C compiler (opt.)</entry>
549 <entry><literal>.hc</literal> or <literal>.c</literal></entry>
550 <entry><option>-S</option></entry>
551 <entry><literal>.s</literal></entry>
555 <entry>assembler</entry>
556 <entry><literal>.s</literal></entry>
557 <entry><option>-c</option></entry>
558 <entry><literal>.o</literal></entry>
562 <entry>linker</entry>
563 <entry><replaceable>other</replaceable></entry>
565 <entry><filename>a.out</filename></entry>
571 <indexterm><primary><option>-C</option></primary></indexterm>
572 <indexterm><primary><option>-E</option></primary></indexterm>
573 <indexterm><primary><option>-S</option></primary></indexterm>
574 <indexterm><primary><option>-c</option></primary></indexterm>
576 <para>Thus, a common invocation would be: </para>
579 ghc -c Foo.hs</screen>
581 <para>to compile the Haskell source file
582 <filename>Foo.hs</filename> to an object file
583 <filename>Foo.o</filename>.</para>
585 <para>Note: What the Haskell compiler proper produces depends on
586 whether a native-code generator<indexterm><primary>native-code
587 generator</primary></indexterm> is used (producing assembly
588 language) or not (producing C). See <xref
589 linkend="options-codegen"/> for more details.</para>
591 <para>Note: C pre-processing is optional, the
592 <option>-cpp</option><indexterm><primary><option>-cpp</option></primary></indexterm>
593 flag turns it on. See <xref linkend="c-pre-processor"/> for more
596 <para>Note: The option <option>-E</option><indexterm><primary>-E
597 option</primary></indexterm> runs just the pre-processing passes
598 of the compiler, dumping the result in a file. Note that this
599 differs from the previous behaviour of dumping the file to
600 standard output.</para>
602 <sect3 id="overriding-suffixes">
603 <title>Overriding the default behaviour for a file</title>
605 <para>As described above, the way in which a file is processed by GHC
606 depends on its suffix. This behaviour can be overriden using the
607 <option>-x</option> option:</para>
611 <term><option>-x</option> <replaceable>suffix</replaceable>
612 <indexterm><primary><option>-x</option></primary>
615 <para>Causes all files following this option on the command
616 line to be processed as if they had the suffix
617 <replaceable>suffix</replaceable>. For example, to compile a
618 Haskell module in the file <literal>M.my-hs</literal>,
619 use <literal>ghc -c -x hs M.my-hs</literal>.</para>
628 <sect1 id="options-help">
629 <title>Help and verbosity options</title>
631 <indexterm><primary>help options</primary></indexterm>
632 <indexterm><primary>verbosity options</primary></indexterm>
634 <para>See also the <option>--help</option>, <option>--version</option>, <option>--numeric-version</option>,
635 and <option>--print-libdir</option> modes in <xref linkend="modes"/>.</para>
640 <indexterm><primary><option>-n</option></primary></indexterm>
643 <para>Does a dry-run, i.e. GHC goes through all the motions
644 of compiling as normal, but does not actually run any
645 external commands.</para>
652 <indexterm><primary><option>-v</option></primary></indexterm>
655 <para>The <option>-v</option> option makes GHC
656 <emphasis>verbose</emphasis>: it reports its version number
657 and shows (on stderr) exactly how it invokes each phase of
658 the compilation system. Moreover, it passes the
659 <option>-v</option> flag to most phases; each reports its
660 version number (and possibly some other information).</para>
662 <para>Please, oh please, use the <option>-v</option> option
663 when reporting bugs! Knowing that you ran the right bits in
664 the right order is always the first thing we want to
671 <option>-v</option><replaceable>n</replaceable>
672 <indexterm><primary><option>-v</option></primary></indexterm>
675 <para>To provide more control over the compiler's verbosity,
676 the <option>-v</option> flag takes an optional numeric
677 argument. Specifying <option>-v</option> on its own is
678 equivalent to <option>-v3</option>, and the other levels
679 have the following meanings:</para>
683 <term><option>-v0</option></term>
685 <para>Disable all non-essential messages (this is the
691 <term><option>-v1</option></term>
693 <para>Minimal verbosity: print one line per
694 compilation (this is the default when
695 <option>––make</option> or
696 <option>––interactive</option> is on).</para>
701 <term><option>-v2</option></term>
703 <para>Print the name of each compilation phase as it
704 is executed. (equivalent to
705 <option>-dshow-passes</option>).</para>
710 <term><option>-v3</option></term>
712 <para>The same as <option>-v2</option>, except that in
713 addition the full command line (if appropriate) for
714 each compilation phase is also printed.</para>
719 <term><option>-v4</option></term>
721 <para>The same as <option>-v3</option> except that the
722 intermediate program representation after each
723 compilation phase is also printed (excluding
724 preprocessed and C/assembly files).</para>
732 <term><option>-ferror-spans</option>
733 <indexterm><primary><option>-ferror-spans</option></primary>
737 <para>Causes GHC to emit the full source span of the
738 syntactic entity relating to an error message. Normally, GHC
739 emits the source location of the start of the syntactic
742 <para>For example:</para>
744 <screen>test.hs:3:6: parse error on input `where'</screen>
746 <para>becomes:</para>
748 <screen>test296.hs:3:6-10: parse error on input `where'</screen>
750 <para>And multi-line spans are possible too:</para>
752 <screen>test.hs:(5,4)-(6,7):
753 Conflicting definitions for `a'
754 Bound at: test.hs:5:4
756 In the binding group for: a, b, a</screen>
758 <para>Note that line numbers start counting at one, but
759 column numbers start at zero. This choice was made to
760 follow existing convention (i.e. this is how Emacs does
766 <term><option>-H</option><replaceable>size</replaceable>
767 <indexterm><primary><option>-H</option></primary></indexterm>
770 <para>Set the minimum size of the heap to
771 <replaceable>size</replaceable>.
772 This option is equivalent to
773 <literal>+RTS -H<replaceable>size</replaceable></literal>,
774 see <xref linkend="rts-options-gc" />.
780 <term><option>-Rghc-timing</option>
781 <indexterm><primary><option>-Rghc-timing</option></primary></indexterm>
784 <para>Prints a one-line summary of timing statistics for the
785 GHC run. This option is equivalent to
786 <literal>+RTS -tstderr</literal>, see <xref
787 linkend="rts-options-gc" />.
796 <sect1 id="options-sanity">
797 <title>Warnings and sanity-checking</title>
799 <indexterm><primary>sanity-checking options</primary></indexterm>
800 <indexterm><primary>warnings</primary></indexterm>
803 <para>GHC has a number of options that select which types of
804 non-fatal error messages, otherwise known as warnings, can be
805 generated during compilation. By default, you get a standard set
806 of warnings which are generally likely to indicate bugs in your
808 <option>-fwarn-overlapping-patterns</option>,
809 <option>-fwarn-deprecations</option>,
810 <option>-fwarn-duplicate-exports</option>,
811 <option>-fwarn-missing-fields</option>, and
812 <option>-fwarn-missing-methods</option>. The following flags are
813 simple ways to select standard “packages” of warnings:
819 <term><option>-W</option>:</term>
821 <indexterm><primary>-W option</primary></indexterm>
822 <para>Provides the standard warnings plus
823 <option>-fwarn-incomplete-patterns</option>,
824 <option>-fwarn-unused-matches</option>,
825 <option>-fwarn-unused-imports</option>, and
826 <option>-fwarn-unused-binds</option>.</para>
831 <term><option>-w</option>:</term>
833 <indexterm><primary><option>-w</option></primary></indexterm>
834 <para>Turns off all warnings, including the standard ones.</para>
839 <term><option>-Wall</option>:</term>
841 <indexterm><primary><option>-Wall</option></primary></indexterm>
842 <para>Turns on all warning options that indicate potentially
843 suspicious code. The warnings that are
844 <emphasis>not</emphasis> enabled by <option>-Wall</option>
847 <listitem><option>-fwarn-simple-patterns</option></listitem>
848 <listitem><option>-fwarn-tabs</option></listitem>
849 <listitem><option>-fwarn-incomplete-record-updates</option></listitem>
850 <listitem><option>-fwarn-monomorphism-restriction</option></listitem>
856 <term><option>-Werror</option>:</term>
858 <indexterm><primary><option>-Werror</option></primary></indexterm>
859 <para>Makes any warning into a fatal error. Useful so that you don't
860 miss warnings when doing batch compilation. </para>
866 <para>The full set of warning options is described below. To turn
867 off any warning, simply give the corresponding
868 <option>-fno-warn-...</option> option on the command line.</para>
873 <term><option>-fwarn-deprecations</option>:</term>
875 <indexterm><primary><option>-fwarn-deprecations</option></primary>
877 <indexterm><primary>deprecations</primary></indexterm>
878 <para>Causes a warning to be emitted when a deprecated
879 function or type is used. Entities can be marked as
880 deprecated using a pragma, see <xref
881 linkend="deprecated-pragma"/>.</para>
886 <term><option>-fwarn-duplicate-exports</option>:</term>
888 <indexterm><primary><option>-fwarn-duplicate-exports</option></primary></indexterm>
889 <indexterm><primary>duplicate exports, warning</primary></indexterm>
890 <indexterm><primary>export lists, duplicates</primary></indexterm>
892 <para>Have the compiler warn about duplicate entries in
893 export lists. This is useful information if you maintain
894 large export lists, and want to avoid the continued export
895 of a definition after you've deleted (one) mention of it in
896 the export list.</para>
898 <para>This option is on by default.</para>
903 <term><option>-fwarn-hi-shadowing</option>:</term>
905 <indexterm><primary><option>-fwarn-hi-shadowing</option></primary></indexterm>
906 <indexterm><primary>shadowing</primary>
907 <secondary>interface files</secondary></indexterm>
909 <para>Causes the compiler to emit a warning when a module or
910 interface file in the current directory is shadowing one
911 with the same module name in a library or other
917 <term><option>-fwarn-incomplete-patterns</option>:</term>
919 <indexterm><primary><option>-fwarn-incomplete-patterns</option></primary></indexterm>
920 <indexterm><primary>incomplete patterns, warning</primary></indexterm>
921 <indexterm><primary>patterns, incomplete</primary></indexterm>
923 <para>Similarly for incomplete patterns, the function
924 <function>g</function> below will fail when applied to
925 non-empty lists, so the compiler will emit a warning about
926 this when <option>-fwarn-incomplete-patterns</option> is
933 <para>This option isn't enabled be default because it can be
934 a bit noisy, and it doesn't always indicate a bug in the
935 program. However, it's generally considered good practice
936 to cover all the cases in your functions.</para>
941 <term><option>-fwarn-incomplete-record-updates</option>:</term>
943 <indexterm><primary><option>-fwarn-incomplete-record-updates</option></primary></indexterm>
944 <indexterm><primary>incomplete record updates, warning</primary></indexterm>
945 <indexterm><primary>record updates, incomplete</primary></indexterm>
948 <function>f</function> below will fail when applied to
949 <literal>Bar</literal>, so the compiler will emit a warning about
950 this when <option>-fwarn-incomplete-record-updates</option> is
954 data Foo = Foo { x :: Int }
958 f foo = foo { x = 6 }
961 <para>This option isn't enabled be default because it can be
962 very noisy, and it often doesn't indicate a bug in the
969 <option>-fwarn-missing-fields</option>:
970 <indexterm><primary><option>-fwarn-missing-fields</option></primary></indexterm>
971 <indexterm><primary>missing fields, warning</primary></indexterm>
972 <indexterm><primary>fields, missing</primary></indexterm>
976 <para>This option is on by default, and warns you whenever
977 the construction of a labelled field constructor isn't
978 complete, missing initializers for one or more fields. While
979 not an error (the missing fields are initialised with
980 bottoms), it is often an indication of a programmer error.</para>
985 <term><option>-fwarn-missing-methods</option>:</term>
987 <indexterm><primary><option>-fwarn-missing-methods</option></primary></indexterm>
988 <indexterm><primary>missing methods, warning</primary></indexterm>
989 <indexterm><primary>methods, missing</primary></indexterm>
991 <para>This option is on by default, and warns you whenever
992 an instance declaration is missing one or more methods, and
993 the corresponding class declaration has no default
994 declaration for them.</para>
995 <para>The warning is suppressed if the method name
996 begins with an underscore. Here's an example where this is useful:
999 _simpleFn :: a -> String
1000 complexFn :: a -> a -> String
1001 complexFn x y = ... _simpleFn ...
1003 The idea is that: (a) users of the class will only call <literal>complexFn</literal>;
1004 never <literal>_simpleFn</literal>; and (b)
1005 instance declarations can define either <literal>complexFn</literal> or <literal>_simpleFn</literal>.
1011 <term><option>-fwarn-missing-signatures</option>:</term>
1013 <indexterm><primary><option>-fwarn-missing-signatures</option></primary></indexterm>
1014 <indexterm><primary>type signatures, missing</primary></indexterm>
1016 <para>If you would like GHC to check that every top-level
1017 function/value has a type signature, use the
1018 <option>-fwarn-missing-signatures</option> option. As part of
1019 the warning GHC also reports the inferred type. The
1020 option is off by default.</para>
1025 <term><option>-fwarn-name-shadowing</option>:</term>
1027 <indexterm><primary><option>-fwarn-name-shadowing</option></primary></indexterm>
1028 <indexterm><primary>shadowing, warning</primary></indexterm>
1030 <para>This option causes a warning to be emitted whenever an
1031 inner-scope value has the same name as an outer-scope value,
1032 i.e. the inner value shadows the outer one. This can catch
1033 typographical errors that turn into hard-to-find bugs, e.g.,
1034 in the inadvertent cyclic definition <literal>let x = ... x
1035 ... in</literal>.</para>
1037 <para>Consequently, this option
1038 <emphasis>will</emphasis> complain about cyclic recursive
1044 <term><option>-fwarn-orphans</option>:</term>
1046 <indexterm><primary><option>-fwarn-orphans</option></primary></indexterm>
1047 <indexterm><primary>orphan instances, warning</primary></indexterm>
1048 <indexterm><primary>orphan rules, warning</primary></indexterm>
1050 <para>This option causes a warning to be emitted whenever the
1051 module contains an "orphan" instance declaration or rewrite rule.
1052 An instance declartion is an orphan if it appears in a module in
1053 which neither the class nor the type being instanced are declared
1054 in the same module. A rule is an orphan if it is a rule for a
1055 function declared in another module. A module containing any
1056 orphans is called an orphan module.</para>
1057 <para>The trouble with orphans is that GHC must pro-actively read the interface
1058 files for all orphan modules, just in case their instances or rules
1059 play a role, whether or not the module's interface would otherwise
1060 be of any use. Other things being equal, avoid orphan modules.</para>
1066 <option>-fwarn-overlapping-patterns</option>:
1067 <indexterm><primary><option>-fwarn-overlapping-patterns</option></primary></indexterm>
1068 <indexterm><primary>overlapping patterns, warning</primary></indexterm>
1069 <indexterm><primary>patterns, overlapping</primary></indexterm>
1072 <para>By default, the compiler will warn you if a set of
1073 patterns are overlapping, i.e.,</para>
1076 f :: String -> Int
1082 <para>where the last pattern match in <function>f</function>
1083 won't ever be reached, as the second pattern overlaps
1084 it. More often than not, redundant patterns is a programmer
1085 mistake/error, so this option is enabled by default.</para>
1090 <term><option>-fwarn-simple-patterns</option>:</term>
1092 <indexterm><primary><option>-fwarn-simple-patterns</option></primary>
1094 <para>Causes the compiler to warn about lambda-bound
1095 patterns that can fail, eg. <literal>\(x:xs)->...</literal>.
1096 Normally, these aren't treated as incomplete patterns by
1097 <option>-fwarn-incomplete-patterns</option>.</para>
1098 <para>``Lambda-bound patterns'' includes all places where there is a single pattern,
1099 including list comprehensions and do-notation. In these cases, a pattern-match
1100 failure is quite legitimate, and triggers filtering (list comprehensions) or
1101 the monad <literal>fail</literal> operation (monads). For example:
1103 f :: [Maybe a] -> [a]
1104 f xs = [y | Just y <- xs]
1106 Switching on <option>-fwarn-simple-patterns</option> will elicit warnings about
1107 these probably-innocent cases, which is why the flag is off by default. </para>
1108 <para> The <literal>deriving( Read )</literal> mechanism produces monadic code with
1109 pattern matches, so you will also get misleading warnings about the compiler-generated
1110 code. (This is arguably a Bad Thing, but it's awkward to fix.)</para>
1116 <term><option>-fwarn-tabs</option>:</term>
1118 <indexterm><primary><option>-fwarn-tabs</option></primary></indexterm>
1119 <indexterm><primary>tabs, warning</primary></indexterm>
1120 <para>Have the compiler warn if there are tabs in your source
1123 <para>This warning is off by default.</para>
1128 <term><option>-fwarn-type-defaults</option>:</term>
1130 <indexterm><primary><option>-fwarn-type-defaults</option></primary></indexterm>
1131 <indexterm><primary>defaulting mechanism, warning</primary></indexterm>
1132 <para>Have the compiler warn/inform you where in your source
1133 the Haskell defaulting mechanism for numeric types kicks
1134 in. This is useful information when converting code from a
1135 context that assumed one default into one with another,
1136 e.g., the `default default' for Haskell 1.4 caused the
1137 otherwise unconstrained value <constant>1</constant> to be
1138 given the type <literal>Int</literal>, whereas Haskell 98
1139 defaults it to <literal>Integer</literal>. This may lead to
1140 differences in performance and behaviour, hence the
1141 usefulness of being non-silent about this.</para>
1143 <para>This warning is off by default.</para>
1148 <term><option>-fwarn-monomorphism-restriction</option>:</term>
1150 <indexterm><primary><option>-fwarn-monomorphism-restriction</option></primary></indexterm>
1151 <indexterm><primary>monomorphism restriction, warning</primary></indexterm>
1152 <para>Have the compiler warn/inform you where in your source
1153 the Haskell Monomorphism Restriction is applied. If applied silently
1154 the MR can give rise to unexpected behaviour, so it can be helpful
1155 to have an explicit warning that it is being applied.</para>
1157 <para>This warning is off by default.</para>
1162 <term><option>-fwarn-unused-binds</option>:</term>
1164 <indexterm><primary><option>-fwarn-unused-binds</option></primary></indexterm>
1165 <indexterm><primary>unused binds, warning</primary></indexterm>
1166 <indexterm><primary>binds, unused</primary></indexterm>
1167 <para>Report any function definitions (and local bindings)
1168 which are unused. For top-level functions, the warning is
1169 only given if the binding is not exported.</para>
1170 <para>A definition is regarded as "used" if (a) it is exported, or (b) it is
1171 mentioned in the right hand side of another definition that is used, or (c) the
1172 function it defines begins with an underscore. The last case provides a
1173 way to suppress unused-binding warnings selectively. </para>
1174 <para> Notice that a variable
1175 is reported as unused even if it appears in the right-hand side of another
1176 unused binding. </para>
1181 <term><option>-fwarn-unused-imports</option>:</term>
1183 <indexterm><primary><option>-fwarn-unused-imports</option></primary></indexterm>
1184 <indexterm><primary>unused imports, warning</primary></indexterm>
1185 <indexterm><primary>imports, unused</primary></indexterm>
1187 <para>Report any modules that are explicitly imported but
1188 never used. However, the form <literal>import M()</literal> is
1189 never reported as an unused import, because it is a useful idiom
1190 for importing instance declarations, which are anonymous in Haskell.</para>
1195 <term><option>-fwarn-unused-matches</option>:</term>
1197 <indexterm><primary><option>-fwarn-unused-matches</option></primary></indexterm>
1198 <indexterm><primary>unused matches, warning</primary></indexterm>
1199 <indexterm><primary>matches, unused</primary></indexterm>
1201 <para>Report all unused variables which arise from pattern
1202 matches, including patterns consisting of a single variable.
1203 For instance <literal>f x y = []</literal> would report
1204 <varname>x</varname> and <varname>y</varname> as unused. The
1205 warning is suppressed if the variable name begins with an underscore, thus:
1215 <para>If you're feeling really paranoid, the
1216 <option>-dcore-lint</option>
1217 option<indexterm><primary><option>-dcore-lint</option></primary></indexterm>
1218 is a good choice. It turns on heavyweight intra-pass
1219 sanity-checking within GHC. (It checks GHC's sanity, not
1226 <sect1 id="options-optimise">
1227 <title>Optimisation (code improvement)</title>
1229 <indexterm><primary>optimisation</primary></indexterm>
1230 <indexterm><primary>improvement, code</primary></indexterm>
1232 <para>The <option>-O*</option> options specify convenient
1233 “packages” of optimisation flags; the
1234 <option>-f*</option> options described later on specify
1235 <emphasis>individual</emphasis> optimisations to be turned on/off;
1236 the <option>-m*</option> options specify
1237 <emphasis>machine-specific</emphasis> optimisations to be turned
1240 <sect2 id="optimise-pkgs">
1241 <title><option>-O*</option>: convenient “packages” of optimisation flags.</title>
1243 <para>There are <emphasis>many</emphasis> options that affect
1244 the quality of code produced by GHC. Most people only have a
1245 general goal, something like “Compile quickly” or
1246 “Make my program run like greased lightning.” The
1247 following “packages” of optimisations (or lack
1248 thereof) should suffice.</para>
1250 <para>Note that higher optimisation levels cause more
1251 cross-module optimisation to be performed, which can have an
1252 impact on how much of your program needs to be recompiled when
1253 you change something. This is one reaosn to stick to
1254 no-optimisation when developing code.</para>
1260 No <option>-O*</option>-type option specified:
1261 <indexterm><primary>-O* not specified</primary></indexterm>
1264 <para>This is taken to mean: “Please compile
1265 quickly; I'm not over-bothered about compiled-code
1266 quality.” So, for example: <command>ghc -c
1267 Foo.hs</command></para>
1273 <option>-O0</option>:
1274 <indexterm><primary><option>-O0</option></primary></indexterm>
1277 <para>Means “turn off all optimisation”,
1278 reverting to the same settings as if no
1279 <option>-O</option> options had been specified. Saying
1280 <option>-O0</option> can be useful if
1281 eg. <command>make</command> has inserted a
1282 <option>-O</option> on the command line already.</para>
1288 <option>-O</option> or <option>-O1</option>:
1289 <indexterm><primary>-O option</primary></indexterm>
1290 <indexterm><primary>-O1 option</primary></indexterm>
1291 <indexterm><primary>optimise</primary><secondary>normally</secondary></indexterm>
1294 <para>Means: “Generate good-quality code without
1295 taking too long about it.” Thus, for example:
1296 <command>ghc -c -O Main.lhs</command></para>
1302 <option>-O2</option>:
1303 <indexterm><primary>-O2 option</primary></indexterm>
1304 <indexterm><primary>optimise</primary><secondary>aggressively</secondary></indexterm>
1307 <para>Means: “Apply every non-dangerous
1308 optimisation, even if it means significantly longer
1309 compile times.”</para>
1311 <para>The avoided “dangerous” optimisations
1312 are those that can make runtime or space
1313 <emphasis>worse</emphasis> if you're unlucky. They are
1314 normally turned on or off individually.</para>
1316 <para>At the moment, <option>-O2</option> is
1317 <emphasis>unlikely</emphasis> to produce better code than
1318 <option>-O</option>.</para>
1324 <option>-Ofile <file></option>:
1325 <indexterm><primary>-Ofile <file> option</primary></indexterm>
1326 <indexterm><primary>optimising, customised</primary></indexterm>
1329 <para>(NOTE: not supported since GHC 4.x. Please ask if
1330 you're interested in this.)</para>
1332 <para>For those who need <emphasis>absolute</emphasis>
1333 control over <emphasis>exactly</emphasis> what options are
1334 used (e.g., compiler writers, sometimes :-), a list of
1335 options can be put in a file and then slurped in with
1336 <option>-Ofile</option>.</para>
1338 <para>In that file, comments are of the
1339 <literal>#</literal>-to-end-of-line variety; blank
1340 lines and most whitespace is ignored.</para>
1342 <para>Please ask if you are baffled and would like an
1343 example of <option>-Ofile</option>!</para>
1348 <para>We don't use a <option>-O*</option> flag for day-to-day
1349 work. We use <option>-O</option> to get respectable speed;
1350 e.g., when we want to measure something. When we want to go for
1351 broke, we tend to use <option>-O2 -fvia-C</option> (and we go for
1352 lots of coffee breaks).</para>
1354 <para>The easiest way to see what <option>-O</option> (etc.)
1355 “really mean” is to run with <option>-v</option>,
1356 then stand back in amazement.</para>
1359 <sect2 id="options-f">
1360 <title><option>-f*</option>: platform-independent flags</title>
1362 <indexterm><primary>-f* options (GHC)</primary></indexterm>
1363 <indexterm><primary>-fno-* options (GHC)</primary></indexterm>
1365 <para>These flags turn on and off individual optimisations.
1366 They are normally set via the <option>-O</option> options
1367 described above, and as such, you shouldn't need to set any of
1368 them explicitly (indeed, doing so could lead to unexpected
1369 results). However, there are one or two that may be of
1374 <term><option>-fexcess-precision</option>:</term>
1376 <indexterm><primary><option>-fexcess-precision</option></primary></indexterm>
1377 <para>When this option is given, intermediate floating
1378 point values can have a <emphasis>greater</emphasis>
1379 precision/range than the final type. Generally this is a
1380 good thing, but some programs may rely on the exact
1382 <literal>Float</literal>/<literal>Double</literal> values
1383 and should not use this option for their compilation.</para>
1388 <term><option>-fignore-asserts</option>:</term>
1390 <indexterm><primary><option>-fignore-asserts</option></primary></indexterm>
1391 <para>Causes GHC to ignore uses of the function
1392 <literal>Exception.assert</literal> in source code (in
1393 other words, rewriting <literal>Exception.assert p
1394 e</literal> to <literal>e</literal> (see <xref
1395 linkend="assertions"/>). This flag is turned on by
1396 <option>-O</option>.
1403 <option>-fno-cse</option>
1404 <indexterm><primary><option>-fno-cse</option></primary></indexterm>
1407 <para>Turns off the common-sub-expression elimination optimisation.
1408 Can be useful if you have some <literal>unsafePerformIO</literal>
1409 expressions that you don't want commoned-up.</para>
1415 <option>-fno-strictness</option>
1416 <indexterm><primary><option>-fno-strictness</option></primary></indexterm>
1419 <para>Turns off the strictness analyser; sometimes it eats
1420 too many cycles.</para>
1426 <option>-fno-full-laziness</option>
1427 <indexterm><primary><option>-fno-full-laziness</option></primary></indexterm>
1430 <para>Turns off the full laziness optimisation (also known as
1431 let-floating). Full laziness increases sharing, which can lead
1432 to increased memory residency.</para>
1434 <para>NOTE: GHC doesn't implement complete full-laziness.
1435 When optimisation in on, and
1436 <option>-fno-full-laziness</option> is not given, some
1437 transformations that increase sharing are performed, such
1438 as extracting repeated computations from a loop. These
1439 are the same transformations that a fully lazy
1440 implementation would do, the difference is that GHC
1441 doesn't consistently apply full-laziness, so don't rely on
1448 <option>-fno-state-hack</option>
1449 <indexterm><primary><option>-fno-state-hack</option></primary></indexterm>
1452 <para>Turn off the "state hack" whereby any lambda with a
1453 <literal>State#</literal> token as argument is considered to be
1454 single-entry, hence it is considered OK to inline things inside
1455 it. This can improve performance of IO and ST monad code, but it
1456 runs the risk of reducing sharing.</para>
1462 <option>-funbox-strict-fields</option>:
1463 <indexterm><primary><option>-funbox-strict-fields</option></primary></indexterm>
1464 <indexterm><primary>strict constructor fields</primary></indexterm>
1465 <indexterm><primary>constructor fields, strict</primary></indexterm>
1468 <para>This option causes all constructor fields which are
1469 marked strict (i.e. “!”) to be unboxed or
1470 unpacked if possible. It is equivalent to adding an
1471 <literal>UNPACK</literal> pragma to every strict
1472 constructor field (see <xref
1473 linkend="unpack-pragma"/>).</para>
1475 <para>This option is a bit of a sledgehammer: it might
1476 sometimes make things worse. Selectively unboxing fields
1477 by using <literal>UNPACK</literal> pragmas might be
1484 <option>-funfolding-update-in-place<n></option>
1485 <indexterm><primary><option>-funfolding-update-in-place</option></primary></indexterm>
1488 <para>Switches on an experimental "optimisation".
1489 Switching it on makes the compiler a little keener to
1490 inline a function that returns a constructor, if the
1491 context is that of a thunk.
1495 If we inlined plusInt we might get an opportunity to use
1496 update-in-place for the thunk 'x'.</para>
1502 <option>-funfolding-creation-threshold<n></option>:
1503 <indexterm><primary><option>-funfolding-creation-threshold</option></primary></indexterm>
1504 <indexterm><primary>inlining, controlling</primary></indexterm>
1505 <indexterm><primary>unfolding, controlling</primary></indexterm>
1508 <para>(Default: 45) Governs the maximum size that GHC will
1509 allow a function unfolding to be. (An unfolding has a
1510 “size” that reflects the cost in terms of
1511 “code bloat” of expanding that unfolding at
1512 at a call site. A bigger function would be assigned a
1513 bigger cost.) </para>
1515 <para> Consequences: (a) nothing larger than this will be
1516 inlined (unless it has an INLINE pragma); (b) nothing
1517 larger than this will be spewed into an interface
1521 <para> Increasing this figure is more likely to result in longer
1522 compile times than faster code. The next option is more
1528 <term><option>-funfolding-use-threshold<n></option>:</term>
1530 <indexterm><primary><option>-funfolding-use-threshold</option></primary></indexterm>
1531 <indexterm><primary>inlining, controlling</primary></indexterm>
1532 <indexterm><primary>unfolding, controlling</primary></indexterm>
1534 <para>(Default: 8) This is the magic cut-off figure for
1535 unfolding: below this size, a function definition will be
1536 unfolded at the call-site, any bigger and it won't. The
1537 size computed for a function depends on two things: the
1538 actual size of the expression minus any discounts that
1539 apply (see <option>-funfolding-con-discount</option>).</para>
1550 <sect1 id="using-concurrent">
1551 <title>Using Concurrent Haskell</title>
1552 <indexterm><primary>Concurrent Haskell</primary><secondary>using</secondary></indexterm>
1554 <para>GHC supports Concurrent Haskell by default, without requiring a
1555 special option or libraries compiled in a certain way. To get access to
1556 the support libraries for Concurrent Haskell, just import
1558 url="../libraries/base/Control-Concurrent.html"><literal>Control.Concurrent</literal></ulink>. More information on Concurrent Haskell is provided in the documentation for that module.</para>
1560 <para>The following RTS option(s) affect the behaviour of Concurrent
1561 Haskell programs:<indexterm><primary>RTS options, concurrent</primary></indexterm></para>
1565 <term><option>-C<replaceable>s</replaceable></option></term>
1567 <para><indexterm><primary><option>-C<replaceable>s</replaceable></option></primary><secondary>RTS option</secondary></indexterm>
1568 Sets the context switch interval to <replaceable>s</replaceable>
1569 seconds. A context switch will occur at the next heap block
1570 allocation after the timer expires (a heap block allocation occurs
1571 every 4k of allocation). With <option>-C0</option> or
1572 <option>-C</option>, context switches will occur as often as
1573 possible (at every heap block allocation). By default, context
1574 switches occur every 20ms.</para>
1580 <sect1 id="using-smp">
1581 <title>Using SMP parallelism</title>
1582 <indexterm><primary>parallelism</primary>
1584 <indexterm><primary>SMP</primary>
1587 <para>GHC supports running Haskell programs in parallel on an SMP
1588 (symmetric multiprocessor).</para>
1590 <para>There's a fine distinction between
1591 <emphasis>concurrency</emphasis> and <emphasis>parallelism</emphasis>:
1592 parallelism is all about making your program run
1593 <emphasis>faster</emphasis> by making use of multiple processors
1594 simultaneously. Concurrency, on the other hand, is a means of
1595 abstraction: it is a convenient way to structure a program that must
1596 respond to multiple asynchronous events.</para>
1598 <para>However, the two terms are certainly related. By making use of
1599 multiple CPUs it is possible to run concurrent threads in parallel,
1600 and this is exactly what GHC's SMP parallelism support does. But it
1601 is also possible to obtain performance improvements with parallelism
1602 on programs that do not use concurrency. This section describes how to
1603 use GHC to compile and run parallel programs, in <xref
1604 linkend="lang-parallel" /> we desribe the language features that affect
1607 <sect2 id="parallel-options">
1608 <title>Options to enable SMP parallelism</title>
1610 <para>In order to make use of multiple CPUs, your program must be
1611 linked with the <option>-threaded</option> option (see <xref
1612 linkend="options-linker" />). Then, to run a program on multiple
1613 CPUs, use the RTS <option>-N</option> option:</para>
1617 <term><option>-N<replaceable>x</replaceable></option></term>
1619 <para><indexterm><primary><option>-N<replaceable>x</replaceable></option></primary><secondary>RTS option</secondary></indexterm>
1620 Use <replaceable>x</replaceable> simultaneous threads when
1621 running the program. Normally <replaceable>x</replaceable>
1622 should be chosen to match the number of CPU cores on the machine.
1623 There is no means (currently) by which this value may vary after
1624 the program has started.</para>
1626 <para>For example, on a dual-core machine we would probably use
1627 <literal>+RTS -N2 -RTS</literal>.</para>
1629 <para>Whether hyperthreading cores should be counted or not is an
1630 open question; please feel free to experiment and let us know what
1631 results you find.</para>
1638 <title>Hints for using SMP parallelism</title>
1640 <para>Add the <literal>-sstderr</literal> RTS option when
1641 running the program to see timing stats, which will help to tell you
1642 whether your program got faster by using more CPUs or not. If the user
1643 time is greater than
1644 the elapsed time, then the program used more than one CPU. You should
1645 also run the program without <literal>-N</literal> for comparison.</para>
1647 <para>GHC's parallelism support is new and experimental. It may make your
1648 program go faster, or it might slow it down - either way, we'd be
1649 interested to hear from you.</para>
1651 <para>One significant limitation with the current implementation is that
1652 the garbage collector is still single-threaded, and all execution must
1653 stop when GC takes place. This can be a significant bottleneck in a
1654 parallel program, especially if your program does a lot of GC. If this
1655 happens to you, then try reducing the cost of GC by tweaking the GC
1656 settings (<xref linkend="rts-options-gc" />): enlarging the heap or the
1657 allocation area size is a good start.</para>
1661 <sect1 id="options-platform">
1662 <title>Platform-specific Flags</title>
1664 <indexterm><primary>-m* options</primary></indexterm>
1665 <indexterm><primary>platform-specific options</primary></indexterm>
1666 <indexterm><primary>machine-specific options</primary></indexterm>
1668 <para>Some flags only make sense for particular target
1674 <term><option>-monly-[32]-regs</option>:</term>
1676 <para>(iX86 machines)<indexterm><primary>-monly-N-regs
1677 option (iX86 only)</primary></indexterm> GHC tries to
1678 “steal” four registers from GCC, for performance
1679 reasons; it almost always works. However, when GCC is
1680 compiling some modules with four stolen registers, it will
1681 crash, probably saying:
1684 Foo.hc:533: fixed or forbidden register was spilled.
1685 This may be due to a compiler bug or to impossible asm
1686 statements or clauses.
1689 Just give some registers back with
1690 <option>-monly-N-regs</option>. Try `3' first, then `2'.
1691 If `2' doesn't work, please report the bug to us.</para>
1700 <sect1 id="ext-core">
1701 <title>Generating and compiling External Core Files</title>
1703 <indexterm><primary>intermediate code generation</primary></indexterm>
1705 <para>GHC can dump its optimized intermediate code (said to be in “Core” format)
1706 to a file as a side-effect of compilation. Core files, which are given the suffix
1707 <filename>.hcr</filename>, can be read and processed by non-GHC back-end
1708 tools. The Core format is formally described in <ulink url="http://www.haskell.org/ghc/docs/papers/core.ps.gz">
1709 <citetitle>An External Representation for the GHC Core Language</citetitle></ulink>,
1710 and sample tools (in Haskell)
1711 for manipulating Core files are available in the GHC source distribution
1712 directory <literal>/fptools/ghc/utils/ext-core</literal>.
1713 Note that the format of <literal>.hcr</literal>
1714 files is <emphasis>different</emphasis> (though similar) to the Core output format generated
1715 for debugging purposes (<xref linkend="options-debugging"/>).</para>
1717 <para>The Core format natively supports notes which you can add to
1718 your source code using the <literal>CORE</literal> pragma (see <xref
1719 linkend="pragmas"/>).</para>
1725 <option>-fext-core</option>
1726 <indexterm><primary><option>-fext-core</option></primary></indexterm>
1729 <para>Generate <literal>.hcr</literal> files.</para>
1735 <para>GHC can also read in External Core files as source; just give the <literal>.hcr</literal> file on
1736 the command line, instead of the <literal>.hs</literal> or <literal>.lhs</literal> Haskell source.
1737 A current infelicity is that you need to give the <literal>-fglasgow-exts</literal> flag too, because
1738 ordinary Haskell 98, when translated to External Core, uses things like rank-2 types.</para>
1747 ;;; Local Variables: ***
1749 ;;; sgml-parent-document: ("users_guide.xml" "book" "chapter") ***