1 <chapter id="using-ghc">
2 <title>Using GHC</title>
4 <indexterm><primary>GHC, using</primary></indexterm>
5 <indexterm><primary>using GHC</primary></indexterm>
7 <para>GHC can work in one of three “modes”:</para>
11 <term><cmdsynopsis><command>ghc</command>
12 <arg choice=plain>––interactive</arg>
14 <indexterm><primary>interactive mode</primary>
16 <indexterm><primary>ghci</primary>
19 <para>Interactive mode, which is also available as
20 <command>ghci</command>. Interactive mode is described in
21 more detail in <xref linkend="ghci">.</para>
26 <term><cmdsynopsis><command>ghc</command>
27 <arg choice=plain>––make</arg>
29 <indexterm><primary>make mode</primary>
31 <indexterm><primary><option>––make</option></primary>
34 <para>In this mode, GHC will build a multi-module Haskell
35 program automatically, figuring out dependencies for itself.
36 If you have a straightforward Haskell program, this is likely
37 to be much easier, and faster, than using
38 <command>make</command>.</para>
44 <command>ghc</command>
52 <indexterm><primary><option>-E</option></primary></indexterm>
53 <indexterm><primary><option>-C</option></primary></indexterm>
54 <indexterm><primary><option>-S</option></primary></indexterm>
55 <indexterm><primary><option>-c</option></primary></indexterm>
57 <para>This is the traditional batch-compiler mode, in which
58 GHC can compile source files one at a time, or link objects
59 together into an executable.</para>
65 <title>Options overview</title>
67 <para>GHC's behaviour is controlled by
68 <firstterm>options</firstterm>, which for historical reasons are
69 also sometimes referred to as command-line flags or arguments.
70 Options can be specified in three ways:</para>
73 <title>Command-line arguments</title>
75 <indexterm><primary>structure, command-line</primary></indexterm>
76 <indexterm><primary>command-line</primary><secondary>arguments</secondary></indexterm>
77 <indexterm><primary>arguments</primary><secondary>command-line</secondary></indexterm>
79 <para>An invocation of GHC takes the following form:</para>
85 <para>Command-line arguments are either options or file names.</para>
87 <para>Command-line options begin with <literal>-</literal>.
88 They may <emphasis>not</emphasis> be grouped:
89 <option>-vO</option> is different from <option>-v -O</option>.
90 Options need not precede filenames: e.g., <literal>ghc *.o -o
91 foo</literal>. All options are processed and then applied to
92 all files; you cannot, for example, invoke <literal>ghc -c -O1
93 Foo.hs -O2 Bar.hs</literal> to apply different optimisation
94 levels to the files <filename>Foo.hs</filename> and
95 <filename>Bar.hs</filename>.</para>
98 <Sect2 id="source-file-options">
99 <title>Command line options in source files</title>
101 <indexterm><primary>source-file options</primary></indexterm>
103 <para>Sometimes it is useful to make the connection between a
104 source file and the command-line options it requires quite
105 tight. For instance, if a Haskell source file uses GHC
106 extensions, it will always need to be compiled with the
107 <option>-fglasgow-exts</option> option. Rather than maintaining
108 the list of per-file options in a <filename>Makefile</filename>,
109 it is possible to do this directly in the source file using the
110 <literal>OPTIONS</literal> pragma <indexterm><primary>OPTIONS
111 pragma</primary></indexterm>:</para>
114 {-# OPTIONS -fglasgow-exts #-}
119 <para><literal>OPTIONS</literal> pragmas are only looked for at
120 the top of your source files, upto the first
121 (non-literate,non-empty) line not containing
122 <literal>OPTIONS</literal>. Multiple <literal>OPTIONS</literal>
123 pragmas are recognised. Note that your command shell does not
124 get to the source file options, they are just included literally
125 in the array of command-line arguments the compiler driver
126 maintains internally, so you'll be desperately disappointed if
127 you try to glob etc. inside <literal>OPTIONS</literal>.</para>
129 <para>NOTE: the contents of OPTIONS are prepended to the
130 command-line options, so you <emphasis>do</emphasis> have the
131 ability to override OPTIONS settings via the command
134 <para>It is not recommended to move all the contents of your
135 Makefiles into your source files, but in some circumstances, the
136 <literal>OPTIONS</literal> pragma is the Right Thing. (If you
137 use <option>-keep-hc-file-too</option> and have OPTION flags in
138 your module, the OPTIONS will get put into the generated .hc
143 <title>Setting options in GHCi</title>
145 <para>Options may also be modified from within GHCi, using the
146 <literal>:set</literal> command. See <xref linkend="ghci-set">
147 for more details.</para>
151 <sect1 id="static-dynamic-flags">
152 <title>Static vs. Dynamic options</title>
153 <indexterm><primary>static</primary><secondary>options</secondary>
155 <indexterm><primary>dynamic</primary><secondary>options</secondary>
158 <para>Each of GHC's command line options is classified as either
159 <firstterm>static</firstterm> or <firstterm>dynamic</firstterm>.
160 A static flag may only be specified on the command line, whereas a
161 dynamic flag may also be given in an <literal>OPTIONS</literal>
162 pragma in a source file or set from the GHCi command-line with
163 <literal>:set</literal>.</para>
165 <para>As a rule of thumb, all the language options are dynamic, as
166 are the warning options and the debugging options. The rest are
167 static, with the notable exceptions of <option>-v</option>,
168 <option>-cpp</option>, <option>-fasm</option>,
169 <option>-fvia-C</option>, and <option>-#include</option>.
171 The flag reference tables (<xref linkend="flag-reference">) lists
172 the status of each flag.</para>
175 <sect1 id="file-suffixes">
176 <title>Meaningful file suffixes</title>
178 <indexterm><primary>suffixes, file</primary></indexterm>
179 <indexterm><primary>file suffixes for GHC</primary></indexterm>
181 <para>File names with “meaningful” suffixes (e.g.,
182 <filename>.lhs</filename> or <filename>.o</filename>) cause the
183 “right thing” to happen to those files.</para>
188 <term><filename>.lhs</filename></term>
189 <indexterm><primary><literal>lhs</literal> suffix</primary></indexterm>
191 <para>A “literate Haskell” module.</para>
196 <term><filename>.hs</filename></term>
198 <para>A not-so-literate Haskell module.</para>
203 <term><filename>.hi</filename></term>
205 <para>A Haskell interface file, probably
206 compiler-generated.</para>
211 <term><filename>.hc</filename></term>
213 <para>Intermediate C file produced by the Haskell
219 <term><filename>.c</filename></term>
221 <para>A C file not produced by the Haskell
227 <term><filename>.s</filename></term>
229 <para>An assembly-language source file, usually produced by
235 <term><filename>.o</filename></term>
237 <para>An object file, produced by an assembler.</para>
242 <para>Files with other suffixes (or without suffixes) are passed
243 straight to the linker.</para>
247 <sect1 id="options-help">
248 <title>Help and verbosity options</title>
250 <IndexTerm><Primary>help options</Primary></IndexTerm>
251 <IndexTerm><Primary>verbosity options</Primary></IndexTerm>
255 <term><option>––help</option></term>
256 <term><option>-?</option></term>
257 <indexterm><primary><option>-?</option></primary></indexterm>
258 <indexterm><primary><option>––help</option></primary></indexterm>
260 <para>Cause GHC to spew a long usage message to standard
261 output and then exit.</para>
266 <term><option>-v</option></term>
267 <indexterm><primary><option>-v</option></primary></indexterm>
269 <para>The <option>-v</option> option makes GHC
270 <emphasis>verbose</emphasis>: it reports its version number
271 and shows (on stderr) exactly how it invokes each phase of
272 the compilation system. Moreover, it passes the
273 <option>-v</option> flag to most phases; each reports its
274 version number (and possibly some other information).</para>
276 <para>Please, oh please, use the <option>-v</option> option
277 when reporting bugs! Knowing that you ran the right bits in
278 the right order is always the first thing we want to
284 <term><option>-v</option><replaceable>n</replaceable></term>
285 <indexterm><primary><option>-v</option></primary></indexterm>
287 <para>To provide more control over the compiler's verbosity,
288 the <option>-v</option> flag takes an optional numeric
289 argument. Specifying <option>-v</option> on its own is
290 equivalent to <option>-v3</option>, and the other levels
291 have the following meanings:</para>
295 <term><option>-v0</option></term>
297 <para>Disable all non-essential messages (this is the
303 <term><option>-v1</option></term>
305 <para>Minimal verbosity: print one line per
306 compilation (this is the default when
307 <option>––make</option> or
308 <option>––interactive</option> is on).</para>
313 <term><option>-v2</option></term>
315 <para>Print the name of each compilation phase as it
316 is executed. (equivalent to
317 <option>-dshow-passes</option>).</para>
322 <term><option>-v3</option></term>
324 <para>The same as <option>-v2</option>, except that in
325 addition the full command line (if appropriate) for
326 each compilation phase is also printed.</para>
331 <term><option>-v4</option></term>
333 <para>The same as <option>-v3</option> except that the
334 intermediate program representation after each
335 compilation phase is also printed (excluding
336 preprocessed and C/assembly files).</para>
344 <term><option>––version</option></term>
345 <indexterm><primary><option>––version</option></primary></indexterm>
347 <para>Print a one-line string including GHC's version number.</para>
352 <term><option>––numeric-version</option></term>
353 <indexterm><primary><option>––numeric-version</option></primary></indexterm>
355 <para>Print GHC's numeric version number only.</para>
360 <term><option>––print-libdir</option></term>
361 <indexterm><primary><option>––print-libdir</option></primary></indexterm>
363 <para>Print the path to GHC's library directory. This is
364 the top of the directory tree containing GHC's libraries,
365 interfaces, and include files (usually something like
366 <literal>/usr/local/lib/ghc-5.04</literal> on Unix). This
368 <literal>$libdir</literal><indexterm><primary><literal>libdir</literal></primary>
369 </indexterm>in the package configuration file (see <xref
370 linkend="packages">).</para>
377 <sect1 id="make-mode">
378 <title>Using <command>ghc</command> <option>––make</option></title>
380 <indexterm><primary><option>––make</option></primary>
382 <indexterm><primary>separate compilation</primary>
385 <para>When given the <option>––make</option> option, GHC will
386 build a multi-module Haskell program by following dependencies
387 from a single root module (usually <literal>Main</literal>). For
388 example, if your <literal>Main</literal> module is in a file
389 called <filename>Main.hs</filename>, you could compile and link
390 the program like this:</para>
393 ghc ––make Main.hs
396 <para>The command line may contain any number of source file names
397 or module names; GHC will figure out all the modules in the
398 program by following the imports from these initial modules. It
399 will then attempt to compile each module which is out of date, and
400 finally if there is a <literal>Main</literal> module, the program
401 will also be linked into an executable.</para>
403 <para>The main advantages to using <literal>ghc ––make</literal>
404 over traditional <literal>Makefile</literal>s are:</para>
408 <para>GHC doesn't have to be restarted for each compilation,
409 which means it can cache information between compilations.
410 Compiling a muli-module program with <literal>ghc
411 ––make</literal> can be up to twice as fast as running
412 <literal>ghc</literal> individually on each source
416 <para>You don't have to write a
417 <literal>Makefile</literal>.</para>
419 <indexterm><primary><literal>Makefile</literal>s</primary><secondary>avoiding</secondary>
422 <para>GHC re-calculates the dependencies each time it is
423 invoked, so the dependencies never get out of sync with the
428 <para>Any of the command-line options described in the rest of
429 this chapter can be used with <option>––make</option>, but note
430 that any options you give on the command line will apply to all
431 the source files compiled, so if you want any options to apply to
432 a single source file only, you'll need to use an
433 <literal>OPTIONS</literal> pragma (see <xref
434 linkend="source-file-options">).</para>
436 <para>If the program needs to be linked with additional objects
437 (say, some auxilliary C code), then the object files can be
438 given on the command line and GHC will include them when linking
439 the executable.</para>
441 <para>Note that GHC can only follow dependencies if it has the
442 source file available, so if your program includes a module for
443 which there is no source file, even if you have an object and an
444 interface file for the module, then GHC will complain. The
445 exception to this rule is for package modules, which may or may
446 not have source files.</para>
448 <para>The source files for the program don't all need to be in the
449 same directory; the <option>-i</option> option can be used to add
450 directories to the search path (see <xref
451 linkend="search-path">).</para>
455 <Sect1 id="options-order">
456 <title>GHC without <option>––make</option></title>
458 <para>Without <option>––make</option>, GHC will compile one or
459 more source files given on the command line.</para>
461 <para>The first phase to run is determined by each input-file
462 suffix, and the last phase is determined by a flag. If no
463 relevant flag is present, then go all the way through linking.
464 This table summarises:</para>
468 <colspec align="left">
469 <colspec align="left">
470 <colspec align="left">
471 <colspec align="left">
475 <entry>Phase of the compilation system</entry>
476 <entry>Suffix saying “start here”</entry>
477 <entry>Flag saying “stop after”</entry>
478 <entry>(suffix of) output file</entry>
483 <entry>literate pre-processor</entry>
484 <entry><literal>.lhs</literal></entry>
486 <entry><literal>.hs</literal></entry>
490 <entry>C pre-processor (opt.)
492 <entry><literal>.hs</literal> (with
493 <option>-cpp</option>)</entry>
494 <entry><option>-E</option></entry>
495 <entry><literal>.hspp</literal></entry>
499 <entry>Haskell compiler</entry>
500 <entry><literal>.hs</literal></entry>
501 <entry><option>-C</option>, <option>-S</option></entry>
502 <entry><literal>.hc</literal>, <literal>.s</literal></entry>
506 <entry>C compiler (opt.)</entry>
507 <entry><literal>.hc</literal> or <literal>.c</literal></entry>
508 <entry><option>-S</option></entry>
509 <entry><literal>.s</literal></entry>
513 <entry>assembler</entry>
514 <entry><literal>.s</literal></entry>
515 <entry><option>-c</option></entry>
516 <entry><literal>.o</literal></entry>
520 <entry>linker</entry>
521 <entry><replaceable>other</replaceable></entry>
523 <entry><filename>a.out</filename></entry>
529 <indexterm><primary><option>-C</option></primary></indexterm>
530 <indexterm><primary><option>-E</option></primary></indexterm>
531 <indexterm><primary><option>-S</option></primary></indexterm>
532 <indexterm><primary><option>-c</option></primary></indexterm>
534 <para>Thus, a common invocation would be: <literal>ghc -c
535 Foo.hs</literal></para>
537 <para>Note: What the Haskell compiler proper produces depends on
538 whether a native-code generator<indexterm><primary>native-code
539 generator</primary></indexterm> is used (producing assembly
540 language) or not (producing C). See <xref
541 linkend="options-codegen"> for more details.</para>
543 <para>Note: C pre-processing is optional, the
544 <option>-ccp</option><indexterm><primary><option>-cpp</option></primary>
545 </indexterm>flag turns it on. See <xref
546 linkend="c-pre-processor"> for more details.</para>
548 <para>Note: The option <option>-E</option><IndexTerm><Primary>-E
549 option</Primary></IndexTerm> runs just the pre-processing passes
550 of the compiler, dumping the result in a file. Note that this
551 differs from the previous behaviour of dumping the file to
552 standard output.</para>
557 <sect1 id="options-sanity">
558 <title>Warnings and sanity-checking</title>
560 <indexterm><primary>sanity-checking options</primary></indexterm>
561 <indexterm><primary>warnings</primary></indexterm>
564 <para>GHC has a number of options that select which types of
565 non-fatal error messages, otherwise known as warnings, can be
566 generated during compilation. By default, you get a standard set
567 of warnings which are generally likely to indicate bugs in your
569 <option>-fwarn-overlapping-patterns</option>,
570 <option>-fwarn-deprecations</option>,
571 <option>-fwarn-duplicate-exports</option>,
572 <option>-fwarn-missing-fields</option>, and
573 <option>-fwarn-missing-methods</option>. The following flags are
574 simple ways to select standard “packages” of warnings:
580 <term><option>-W</option>:</term>
582 <IndexTerm><Primary>-W option</Primary></IndexTerm>
583 <para>Provides the standard warnings plus
584 <option>-fwarn-incomplete-patterns</option>,
585 <option>-fwarn-unused-matches</option>,
586 <option>-fwarn-unused-imports</option>,
587 <option>-fwarn-misc</option>, and
588 <option>-fwarn-unused-binds</option>.</para>
593 <term><option>-w</option>:</term>
595 <IndexTerm><Primary><option>-w</option></Primary></IndexTerm>
596 <para>Turns off all warnings, including the standard ones.</para>
601 <term><option>-Wall</option>:</term>
603 <indexterm><primary><option>-Wall</option></primary></indexterm>
604 <para>Turns on all warning options.</para>
609 <term><option>-Werror</option>:</term>
611 <indexterm><primary><option>-Werror</option></primary></indexterm>
612 <para>Makes any warning into a fatal error. Useful so that you don't
613 miss warnings when doing batch compilation. </para>
619 <para>The full set of warning options is described below. To turn
620 off any warning, simply give the corresponding
621 <option>-fno-warn-...</option> option on the command line.</para>
626 <term><option>-fwarn-deprecations</option>:</term>
628 <indexterm><primary><option>-fwarn-deprecations</option></primary>
630 <indexterm><primary>deprecations</primary></indexterm>
631 <para>Causes a warning to be emitted when a deprecated
632 function or type is used. Entities can be marked as
633 deprecated using a pragma, see <xref
634 linkend="deprecated-pragma">.</para>
639 <term><option>-fwarn-duplicate-exports</option>:</term>
641 <indexterm><primary><option>-fwarn-duplicate-exports</option></primary></indexterm>
642 <indexterm><primary>duplicate exports, warning</primary></indexterm>
643 <indexterm><primary>export lists, duplicates</primary></indexterm>
645 <para>Have the compiler warn about duplicate entries in
646 export lists. This is useful information if you maintain
647 large export lists, and want to avoid the continued export
648 of a definition after you've deleted (one) mention of it in
649 the export list.</para>
651 <para>This option is on by default.</para>
656 <term><option>-fwarn-hi-shadowing</option>:</term>
658 <indexterm><primary><option>-fwarn-hi-shadowing</option></primary></indexterm>
659 <indexterm><primary>shadowing</primary>
660 <secondary>interface files</secondary></indexterm>
662 <para>Causes the compiler to emit a warning when a module or
663 interface file in the current directory is shadowing one
664 with the same module name in a library or other
670 <term><option>-fwarn-incomplete-patterns</option>:</term>
672 <indexterm><primary><option>-fwarn-incomplete-patterns</option></primary></indexterm>
673 <indexterm><primary>incomplete patterns, warning</primary></indexterm>
674 <indexterm><primary>patterns, incomplete</primary></indexterm>
676 <para>Similarly for incomplete patterns, the function
677 <function>g</function> below will fail when applied to
678 non-empty lists, so the compiler will emit a warning about
679 this when <option>-fwarn-incomplete-patterns</option> is
686 <para>This option isn't enabled be default because it can be
687 a bit noisy, and it doesn't always indicate a bug in the
688 program. However, it's generally considered good practice
689 to cover all the cases in your functions.</para>
694 <term><option>-fwarn-misc</option>:</term>
695 <indexterm><primary><option>-fwarn-misc</option></primary></indexterm>
697 <para>Turns on warnings for various harmless but untidy
698 things. This currently includes: importing a type with
699 <literal>(..)</literal> when the export is abstract, and
700 listing duplicate class assertions in a qualified type.</para>
705 <term><option>-fwarn-missing-fields</option>:</term>
707 <indexterm><primary><option>-fwarn-missing-fields</option></primary></indexterm>
708 <indexterm><primary>missing fields, warning</primary></indexterm>
709 <indexterm><primary>fields, missing</primary></indexterm>
711 <para>This option is on by default, and warns you whenever
712 the construction of a labelled field constructor isn't
713 complete, missing initializers for one or more fields. While
714 not an error (the missing fields are initialised with
715 bottoms), it is often an indication of a programmer error.</para>
720 <term><option>-fwarn-missing-methods</option>:</term>
722 <indexterm><primary><option>-fwarn-missing-methods</option></primary></indexterm>
723 <indexterm><primary>missing methods, warning</primary></indexterm>
724 <indexterm><primary>methods, missing</primary></indexterm>
726 <para>This option is on by default, and warns you whenever
727 an instance declaration is missing one or more methods, and
728 the corresponding class declaration has no default
729 declaration for them.</para>
730 <para>The warning is suppressed if the method name
731 begins with an underscore. Here's an example where this is useful:
734 _simpleFn :: a -> String
735 complexFn :: a -> a -> String
736 complexFn x y = ... _simpleFn ...
738 The idea is that: (a) users of the class will only call <literal>complexFn</literal>;
739 never <literal>_simpleFn</literal>; and (b)
740 instance declarations can define either <literal>complexFn</literal> or <literal>_simpleFn</literal>.
746 <term><option>-fwarn-missing-signatures</option>:</term>
748 <indexterm><primary><option>-fwarn-missing-signatures</option></primary></indexterm>
749 <indexterm><primary>type signatures, missing</primary></indexterm>
751 <para>If you would like GHC to check that every top-level
752 function/value has a type signature, use the
753 <option>-fwarn-missing-signatures</option> option. This
754 option is off by default.</para>
759 <term><option>-fwarn-name-shadowing</option>:</term>
761 <indexterm><primary><option>-fwarn-name-shadowing</option></primary></indexterm>
762 <indexterm><primary>shadowing, warning</primary></indexterm>
764 <para>This option causes a warning to be emitted whenever an
765 inner-scope value has the same name as an outer-scope value,
766 i.e. the inner value shadows the outer one. This can catch
767 typographical errors that turn into hard-to-find bugs, e.g.,
768 in the inadvertent cyclic definition <literal>let x = ... x
769 ... in</literal>.</para>
771 <para>Consequently, this option does
772 <emphasis>will</emphasis> complain about cyclic recursive
778 <term><option>-fwarn-overlapping-patterns</option>:</term>
779 <indexterm><primary><option>-fwarn-overlapping-patterns</option></primary></indexterm>
780 <indexterm><primary>overlapping patterns, warning</primary></indexterm>
781 <indexterm><primary>patterns, overlapping</primary></indexterm>
783 <para>By default, the compiler will warn you if a set of
784 patterns are overlapping, i.e.,</para>
787 f :: String -> Int
793 <para>where the last pattern match in <Function>f</Function>
794 won't ever be reached, as the second pattern overlaps
795 it. More often than not, redundant patterns is a programmer
796 mistake/error, so this option is enabled by default.</para>
801 <term><option>-fwarn-simple-patterns</option>:</term>
803 <indexterm><primary><option>-fwarn-simple-patterns</option></primary>
805 <para>Causes the compiler to warn about lambda-bound
806 patterns that can fail, eg. <literal>\(x:xs)->...</literal>.
807 Normally, these aren't treated as incomplete patterns by
808 <option>-fwarn-incomplete-patterns</option>.</para>
809 <para>``Lambda-bound patterns'' includes all places where there is a single pattern,
810 including list comprehensions and do-notation. In these cases, a pattern-match
811 failure is quite legitimate, and triggers filtering (list comprehensions) or
812 the monad <literal>fail</literal> operation (monads). For example:
814 f :: [Maybe a] -> [a]
815 f xs = [y | Just y <- xs]
817 Switching on <option>-fwarn-simple-patterns</option> will elicit warnings about
818 these probably-innocent cases, which is why the flag is off by default. </para>
819 <para> The <literal>deriving( Read )</literal> mechanism produces monadic code with
820 pattern matches, so you will also get misleading warnings about the compiler-generated
821 code. (This is arguably a Bad Thing, but it's awkward to fix.)</para>
827 <term><option>-fwarn-type-defaults</option>:</term>
829 <indexterm><primary><option>-fwarn-type-defaults</option></primary></indexterm>
830 <indexterm><primary>defaulting mechanism, warning</primary></indexterm>
831 <para>Have the compiler warn/inform you where in your source
832 the Haskell defaulting mechanism for numeric types kicks
833 in. This is useful information when converting code from a
834 context that assumed one default into one with another,
835 e.g., the `default default' for Haskell 1.4 caused the
836 otherwise unconstrained value <Constant>1</Constant> to be
837 given the type <literal>Int</literal>, whereas Haskell 98
838 defaults it to <literal>Integer</literal>. This may lead to
839 differences in performance and behaviour, hence the
840 usefulness of being non-silent about this.</para>
842 <para>This warning is off by default.</para>
847 <term><option>-fwarn-unused-binds</option>:</term>
849 <indexterm><primary><option>-fwarn-unused-binds</option></primary></indexterm>
850 <indexterm><primary>unused binds, warning</primary></indexterm>
851 <indexterm><primary>binds, unused</primary></indexterm>
852 <para>Report any function definitions (and local bindings)
853 which are unused. For top-level functions, the warning is
854 only given if the binding is not exported.</para>
859 <term><option>-fwarn-unused-imports</option>:</term>
861 <indexterm><primary><option>-fwarn-unused-imports</option></primary></indexterm>
862 <indexterm><primary>unused imports, warning</primary></indexterm>
863 <indexterm><primary>imports, unused</primary></indexterm>
865 <para>Report any objects that are explicitly imported but
871 <term><option>-fwarn-unused-matches</option>:</term>
873 <indexterm><primary><option>-fwarn-unused-matches</option></primary></indexterm>
874 <indexterm><primary>unused matches, warning</primary></indexterm>
875 <indexterm><primary>matches, unused</primary></indexterm>
877 <para>Report all unused variables which arise from pattern
878 matches, including patterns consisting of a single variable.
879 For instance <literal>f x y = []</literal> would report
880 <VarName>x</VarName> and <VarName>y</VarName> as unused. The
881 warning is suppressed if the variable name begins with an underscore, thus:
891 <para>If you're feeling really paranoid, the
892 <option>-dcore-lint</option>
893 option<indexterm><primary><option>-dcore-lint</option></primary></indexterm>
894 is a good choice. It turns on heavyweight intra-pass
895 sanity-checking within GHC. (It checks GHC's sanity, not
902 <sect1 id="options-optimise">
903 <title>Optimisation (code improvement)</title>
905 <indexterm><primary>optimisation</primary></indexterm>
906 <indexterm><primary>improvement, code</primary></indexterm>
908 <para>The <option>-O*</option> options specify convenient
909 “packages” of optimisation flags; the
910 <option>-f*</option> options described later on specify
911 <emphasis>individual</emphasis> optimisations to be turned on/off;
912 the <option>-m*</option> options specify
913 <emphasis>machine-specific</emphasis> optimisations to be turned
916 <sect2 id="optimise-pkgs">
917 <title><option>-O*</option>: convenient “packages” of optimisation flags.</title>
919 <para>There are <emphasis>many</emphasis> options that affect
920 the quality of code produced by GHC. Most people only have a
921 general goal, something like “Compile quickly” or
922 “Make my program run like greased lightning.” The
923 following “packages” of optimisations (or lack
924 thereof) should suffice.</para>
926 <para>Once you choose a <option>-O*</option>
927 “package,” stick with it—don't chop and
928 change. Modules' interfaces <emphasis>will</emphasis> change
929 with a shift to a new <option>-O*</option> option, and you may
930 have to recompile a large chunk of all importing modules before
931 your program can again be run safely (see <XRef
932 LinkEnd="recomp">).</para>
937 <term>No <option>-O*</option>-type option specified:</term>
938 <indexterm><primary>-O* not specified</primary></indexterm>
940 <para>This is taken to mean: “Please compile
941 quickly; I'm not over-bothered about compiled-code
942 quality.” So, for example: <command>ghc -c
943 Foo.hs</command></para>
948 <term><option>-O0</option>:</term>
949 <indexterm><primary><option>-O0</option></primary></indexterm>
951 <para>Means “turn off all optimisation”,
952 reverting to the same settings as if no
953 <option>-O</option> options had been specified. Saying
954 <option>-O0</option> can be useful if
955 eg. <command>make</command> has inserted a
956 <option>-O</option> on the command line already.</para>
961 <term><option>-O</option> or <option>-O1</option>:</term>
962 <indexterm><primary>-O option</primary></indexterm>
963 <indexterm><primary>-O1 option</primary></indexterm>
964 <indexterm><primary>optimise</primary><secondary>normally</secondary></indexterm>
966 <para>Means: “Generate good-quality code without
967 taking too long about it.” Thus, for example:
968 <command>ghc -c -O Main.lhs</command></para>
973 <term><option>-O2</option>:</term>
974 <indexterm><primary>-O2 option</primary></indexterm>
975 <indexterm><primary>optimise</primary><secondary>aggressively</secondary></indexterm>
977 <para>Means: “Apply every non-dangerous
978 optimisation, even if it means significantly longer
979 compile times.”</para>
981 <para>The avoided “dangerous” optimisations
982 are those that can make runtime or space
983 <emphasis>worse</emphasis> if you're unlucky. They are
984 normally turned on or off individually.</para>
986 <para>At the moment, <option>-O2</option> is
987 <emphasis>unlikely</emphasis> to produce better code than
988 <option>-O</option>.</para>
993 <term><option>-Ofile <file></option>:</term>
994 <indexterm><primary>-Ofile <file> option</primary></indexterm>
995 <indexterm><primary>optimising, customised</primary></indexterm>
997 <para>(NOTE: not supported yet in GHC 5.x. Please ask if
998 you're interested in this.)</para>
1000 <para>For those who need <emphasis>absolute</emphasis>
1001 control over <emphasis>exactly</emphasis> what options are
1002 used (e.g., compiler writers, sometimes :-), a list of
1003 options can be put in a file and then slurped in with
1004 <option>-Ofile</option>.</para>
1006 <para>In that file, comments are of the
1007 <literal>#</literal>-to-end-of-line variety; blank
1008 lines and most whitespace is ignored.</para>
1010 <para>Please ask if you are baffled and would like an
1011 example of <option>-Ofile</option>!</para>
1016 <para>We don't use a <option>-O*</option> flag for day-to-day
1017 work. We use <option>-O</option> to get respectable speed;
1018 e.g., when we want to measure something. When we want to go for
1019 broke, we tend to use <option>-O -fvia-C</option> (and we go for
1020 lots of coffee breaks).</para>
1022 <para>The easiest way to see what <option>-O</option> (etc.)
1023 “really mean” is to run with <option>-v</option>,
1024 then stand back in amazement.</para>
1027 <sect2 id="options-f">
1028 <title><option>-f*</option>: platform-independent flags</title>
1030 <indexterm><primary>-f* options (GHC)</primary></indexterm>
1031 <indexterm><primary>-fno-* options (GHC)</primary></indexterm>
1033 <para>These flags turn on and off individual optimisations.
1034 They are normally set via the <option>-O</option> options
1035 described above, and as such, you shouldn't need to set any of
1036 them explicitly (indeed, doing so could lead to unexpected
1037 results). However, there are one or two that may be of
1042 <term><option>-fexcess-precision</option>:</term>
1044 <indexterm><primary><option>-fexcess-precision</option></primary></indexterm>
1045 <para>When this option is given, intermediate floating
1046 point values can have a <emphasis>greater</emphasis>
1047 precision/range than the final type. Generally this is a
1048 good thing, but some programs may rely on the exact
1050 <literal>Float</literal>/<literal>Double</literal> values
1051 and should not use this option for their compilation.</para>
1056 <term><option>-fignore-asserts</option>:</term>
1058 <indexterm><primary><option>-fignore-asserts</option></primary></indexterm>
1059 <para>Causes GHC to ignore uses of the function
1060 <literal>Exception.assert</literal> in source code (in
1061 other words, rewriting <literal>Exception.assert p
1062 e</literal> to <literal>e</literal> (see <xref
1063 linkend="sec-assertions">). This flag is turned on by
1064 <option>-O</option>.
1070 <term><option>-fno-strictness</option></term>
1071 <indexterm><primary><option>-fno-strictness</option></primary>
1074 <para>Turns off the strictness analyser; sometimes it eats
1075 too many cycles.</para>
1080 <term><option>-fno-cpr-analyse</option></term>
1081 <indexterm><primary><option>-fno-cpr-analyse</option></primary>
1084 <para>Turns off the CPR (constructed product result)
1085 analysis; it is somewhat experimental.</para>
1090 <term><option>-funbox-strict-fields</option>:</term>
1092 <indexterm><primary><option>-funbox-strict-fields</option></primary></indexterm>
1093 <indexterm><primary>strict constructor fields</primary></indexterm>
1094 <indexterm><primary>constructor fields, strict</primary></indexterm>
1096 <para>This option causes all constructor fields which are
1097 marked strict (i.e. “!”) to be unboxed or
1098 unpacked if possible. For example:</para>
1101 data T = T !Float !Float
1104 <para>will create a constructor <literal>T</literal>
1105 containing two unboxed floats if the
1106 <option>-funbox-strict-fields</option> flag is given.
1107 This may not always be an optimisation: if the
1108 <Function>T</Function> constructor is scrutinised and the
1109 floats passed to a non-strict function for example, they
1110 will have to be reboxed (this is done automatically by the
1113 <para>This option should only be used in conjunction with
1114 <option>-O</option>, in order to expose unfoldings to the
1115 compiler so the reboxing can be removed as often as
1116 possible. For example:</para>
1120 f (T f1 f2) = f1 + f2
1123 <para>The compiler will avoid reboxing
1124 <Function>f1</Function> and <Function>f2</Function> by
1125 inlining <Function>+</Function> on floats, but only when
1126 <option>-O</option> is on.</para>
1128 <para>Any single-constructor data is eligible for
1129 unpacking; for example</para>
1132 data T = T !(Int,Int)
1135 <para>will store the two <literal>Int</literal>s directly
1136 in the <Function>T</Function> constructor, by flattening
1137 the pair. Multi-level unpacking is also supported:</para>
1141 data S = S !Int !Int
1144 <para>will store two unboxed <literal>Int#</literal>s
1145 directly in the <Function>T</Function> constructor.</para>
1150 <term><option>-funfolding-update-in-place<n></option></term>
1151 <indexterm><primary><option>-funfolding-update-in-place</option></primary></indexterm>
1153 <para>Switches on an experimental "optimisation".
1154 Switching it on makes the compiler a little keener to
1155 inline a function that returns a constructor, if the
1156 context is that of a thunk.
1160 If we inlined plusInt we might get an opportunity to use
1161 update-in-place for the thunk 'x'.</para>
1166 <term><option>-funfolding-creation-threshold<n></option>:</term>
1168 <indexterm><primary><option>-funfolding-creation-threshold</option></primary></indexterm>
1169 <indexterm><primary>inlining, controlling</primary></indexterm>
1170 <indexterm><primary>unfolding, controlling</primary></indexterm>
1172 <para>(Default: 45) Governs the maximum size that GHC will
1173 allow a function unfolding to be. (An unfolding has a
1174 “size” that reflects the cost in terms of
1175 “code bloat” of expanding that unfolding at
1176 at a call site. A bigger function would be assigned a
1177 bigger cost.) </para>
1179 <para> Consequences: (a) nothing larger than this will be
1180 inlined (unless it has an INLINE pragma); (b) nothing
1181 larger than this will be spewed into an interface
1185 <para> Increasing this figure is more likely to result in longer
1186 compile times than faster code. The next option is more
1192 <term><option>-funfolding-use-threshold<n></option>:</term>
1194 <indexterm><primary><option>-funfolding-use-threshold</option></primary></indexterm>
1195 <indexterm><primary>inlining, controlling</primary></indexterm>
1196 <indexterm><primary>unfolding, controlling</primary></indexterm>
1198 <para>(Default: 8) This is the magic cut-off figure for
1199 unfolding: below this size, a function definition will be
1200 unfolded at the call-site, any bigger and it won't. The
1201 size computed for a function depends on two things: the
1202 actual size of the expression minus any discounts that
1203 apply (see <option>-funfolding-con-discount</option>).</para>
1214 <sect1 id="sec-using-concurrent">
1215 <title>Using Concurrent Haskell</title>
1217 <indexterm><primary>Concurrent Haskell—use</primary></indexterm>
1220 GHC supports Concurrent Haskell by default, without requiring a
1221 special option or libraries compiled in a certain way. To get access
1222 to the support libraries for Concurrent Haskell, just import
1223 <literal>Control.Concurrent</literal> (details are in the accompanying
1224 library documentation).</para>
1227 RTS options are provided for modifying the behaviour of the threaded
1228 runtime system. See <XRef LinkEnd="parallel-rts-opts">.
1232 Concurrent Haskell is described in more detail in the documentation
1233 for the <literal>Control.Concurrent</literal> module.
1238 <Sect1 id="sec-using-parallel">
1239 <title>Using Parallel Haskell</title>
1242 <indexterm><primary>Parallel Haskell—use</primary></indexterm>
1246 [You won't be able to execute parallel Haskell programs unless PVM3
1247 (Parallel Virtual Machine, version 3) is installed at your site.]
1251 To compile a Haskell program for parallel execution under PVM, use the
1252 <Option>-parallel</Option> option,<IndexTerm><Primary>-parallel
1253 option</Primary></IndexTerm> both when compiling <Emphasis>and
1254 linking</Emphasis>. You will probably want to <Literal>import
1255 Parallel</Literal> into your Haskell modules.
1259 To run your parallel program, once PVM is going, just invoke it
1260 “as normal”. The main extra RTS option is
1261 <Option>-qp<n></Option>, to say how many PVM
1262 “processors” your program to run on. (For more details of
1263 all relevant RTS options, please see <XRef
1264 LinkEnd="parallel-rts-opts">.)
1268 In truth, running Parallel Haskell programs and getting information
1269 out of them (e.g., parallelism profiles) is a battle with the vagaries of
1270 PVM, detailed in the following sections.
1273 <Sect2 id="pvm-dummies">
1274 <Title>Dummy's guide to using PVM</Title>
1277 <indexterm><primary>PVM, how to use</primary></indexterm>
1278 <indexterm><primary>Parallel Haskell—PVM use</primary></indexterm>
1279 Before you can run a parallel program under PVM, you must set the
1280 required environment variables (PVM's idea, not ours); something like,
1281 probably in your <filename>.cshrc</filename> or equivalent:
1284 setenv PVM_ROOT /wherever/you/put/it
1285 setenv PVM_ARCH `$PVM_ROOT/lib/pvmgetarch`
1286 setenv PVM_DPATH $PVM_ROOT/lib/pvmd
1292 Creating and/or controlling your “parallel machine” is a purely-PVM
1293 business; nothing specific to Parallel Haskell. The following paragraphs
1294 describe how to configure your parallel machine interactively.
1298 If you use parallel Haskell regularly on the same machine configuration it
1299 is a good idea to maintain a file with all machine names and to make the
1300 environment variable PVM_HOST_FILE point to this file. Then you can avoid
1301 the interactive operations described below by just saying
1309 You use the <Command>pvm</Command><IndexTerm><Primary>pvm command</Primary></IndexTerm> command to start PVM on your
1310 machine. You can then do various things to control/monitor your
1311 “parallel machine;” the most useful being:
1317 <ColSpec Align="Left">
1321 <entry><KeyCombo><KeyCap>Control</KeyCap><KeyCap>D</KeyCap></KeyCombo></entry>
1322 <entry>exit <command>pvm</command>, leaving it running</entry>
1326 <entry><command>halt</command></entry>
1327 <entry>kill off this “parallel machine” & exit</entry>
1331 <entry><command>add <host></command></entry>
1332 <entry>add <command><host></command> as a processor</entry>
1336 <entry><command>delete <host></command></entry>
1337 <entry>delete <command><host></command></entry>
1341 <entry><command>reset</command></entry>
1342 <entry>kill what's going, but leave PVM up</entry>
1346 <entry><command>conf</command></entry>
1347 <entry>list the current configuration</entry>
1351 <entry><command>ps</command></entry>
1352 <entry>report processes' status</entry>
1356 <entry><command>pstat <pid></command></entry>
1357 <entry>status of a particular process</entry>
1366 The PVM documentation can tell you much, much more about <command>pvm</command>!
1371 <Sect2 id="par-profiles">
1372 <Title>Parallelism profiles</Title>
1375 <indexterm><primary>parallelism profiles</primary></indexterm>
1376 <indexterm><primary>profiles, parallelism</primary></indexterm>
1377 <indexterm><primary>visualisation tools</primary></indexterm>
1381 With Parallel Haskell programs, we usually don't care about the
1382 results—only with “how parallel” it was! We want pretty pictures.
1386 Parallelism profiles (à la <Command>hbcpp</Command>) can be generated with the
1387 <Option>-qP</Option><IndexTerm><Primary>-qP RTS option (concurrent, parallel)</Primary></IndexTerm> RTS option. The
1388 per-processor profiling info is dumped into files named
1389 <Filename><full-path><program>.gr</Filename>. These are then munged into a PostScript picture,
1390 which you can then display. For example, to run your program
1391 <Filename>a.out</Filename> on 8 processors, then view the parallelism profile, do:
1397 <prompt>$</prompt> ./a.out +RTS -qP -qp8
1398 <prompt>$</prompt> grs2gr *.???.gr > temp.gr # combine the 8 .gr files into one
1399 <prompt>$</prompt> gr2ps -O temp.gr # cvt to .ps; output in temp.ps
1400 <prompt>$</prompt> ghostview -seascape temp.ps # look at it!
1406 The scripts for processing the parallelism profiles are distributed
1407 in <filename>ghc/utils/parallel/</filename>.
1413 <Title>Other useful info about running parallel programs</Title>
1416 The “garbage-collection statistics” RTS options can be useful for
1417 seeing what parallel programs are doing. If you do either
1418 <Option>+RTS -Sstderr</Option><IndexTerm><Primary>-Sstderr RTS option</Primary></IndexTerm> or <Option>+RTS -sstderr</Option>, then
1419 you'll get mutator, garbage-collection, etc., times on standard
1420 error. The standard error of all PE's other than the `main thread'
1421 appears in <filename>/tmp/pvml.nnn</filename>, courtesy of PVM.
1425 Whether doing <option>+RTS -Sstderr</option> or not, a handy way to watch
1426 what's happening overall is: <command>tail -f /tmp/pvml.nnn</command>.
1431 <Sect2 id="parallel-rts-opts">
1432 <title>RTS options for Concurrent/Parallel Haskell
1436 <indexterm><primary>RTS options, concurrent</primary></indexterm>
1437 <indexterm><primary>RTS options, parallel</primary></indexterm>
1438 <indexterm><primary>Concurrent Haskell—RTS options</primary></indexterm>
1439 <indexterm><primary>Parallel Haskell—RTS options</primary></indexterm>
1443 Besides the usual runtime system (RTS) options
1444 (<XRef LinkEnd="runtime-control">), there are a few options particularly
1445 for concurrent/parallel execution.
1452 <Term><Option>-qp<N></Option>:</Term>
1455 <IndexTerm><Primary>-qp<N> RTS option</Primary></IndexTerm>
1456 (PARALLEL ONLY) Use <Literal><N></Literal> PVM processors to run this program;
1462 <term><option>-C[<us>]</option>:</term>
1465 <indexterm><primary>-C<us> RTS option</primary></indexterm> Sets
1466 the context switch interval to <literal><s></literal> seconds.
1467 A context switch will occur at the next heap block allocation after
1468 the timer expires (a heap block allocation occurs every 4k of
1469 allocation). With <option>-C0</option> or <option>-C</option>,
1470 context switches will occur as often as possible (at every heap block
1471 allocation). By default, context switches occur every 20ms
1472 milliseconds. Note that GHC's internal timer ticks every 20ms, and
1473 the context switch timer is always a multiple of this timer, so 20ms
1474 is the maximum granularity available for timed context switches.
1479 <term><option>-q[v]</option>:</term>
1482 <indexterm><primary>-q RTS option</primary></indexterm>
1483 (PARALLEL ONLY) Produce a quasi-parallel profile of thread activity,
1484 in the file <FIlename><program>.qp</FIlename>. In the style of <command>hbcpp</command>, this profile
1485 records the movement of threads between the green (runnable) and red
1486 (blocked) queues. If you specify the verbose suboption (<option>-qv</option>), the
1487 green queue is split into green (for the currently running thread
1488 only) and amber (for other runnable threads). We do not recommend
1489 that you use the verbose suboption if you are planning to use the
1490 <Command>hbcpp</Command> profiling tools or if you are context switching at every heap
1491 check (with <Option>-C</Option>).
1497 <Term><Option>-qt<num></Option>:</Term>
1500 <IndexTerm><Primary>-qt<num> RTS option</Primary></IndexTerm>
1501 (PARALLEL ONLY) Limit the thread pool size, i.e. the number of concurrent
1502 threads per processor to <Literal><num></Literal>. The default is
1503 32. Each thread requires slightly over 1K <Emphasis>words</Emphasis> in
1504 the heap for thread state and stack objects. (For 32-bit machines, this
1505 translates to 4K bytes, and for 64-bit machines, 8K bytes.)
1511 <Term><Option>-d</Option>:</Term>
1514 <IndexTerm><Primary>-d RTS option (parallel)</Primary></IndexTerm>
1515 (PARALLEL ONLY) Turn on debugging. It pops up one xterm (or GDB, or
1516 something…) per PVM processor. We use the standard <Command>debugger</Command>
1517 script that comes with PVM3, but we sometimes meddle with the
1518 <Command>debugger2</Command> script. We include ours in the GHC distribution,
1519 in <Filename>ghc/utils/pvm/</Filename>.
1525 <Term><Option>-qe<num></Option>:</Term>
1528 <IndexTerm><Primary>-qe<num> RTS option
1529 (parallel)</Primary></IndexTerm> (PARALLEL ONLY) Limit the spark pool size
1530 i.e. the number of pending sparks per processor to
1531 <Literal><num></Literal>. The default is 100. A larger number may be
1532 appropriate if your program generates large amounts of parallelism
1538 <Term><Option>-qQ<num></Option>:</Term>
1541 <IndexTerm><Primary>-qQ<num> RTS option (parallel)</Primary></IndexTerm>
1542 (PARALLEL ONLY) Set the size of packets transmitted between processors
1543 to <Literal><num></Literal>. The default is 1024 words. A larger number may be
1544 appropriate if your machine has a high communication cost relative to
1550 <Term><Option>-qh<num></Option>:</Term>
1553 <IndexTerm><Primary>-qh<num> RTS option (parallel)</Primary></IndexTerm>
1554 (PARALLEL ONLY) Select a packing scheme. Set the number of non-root thunks to pack in one packet to
1555 <num>-1 (0 means infinity). By default GUM uses full-subgraph
1556 packing, i.e. the entire subgraph with the requested closure as root is
1557 transmitted (provided it fits into one packet). Choosing a smaller value
1558 reduces the amount of pre-fetching of work done in GUM. This can be
1559 advantageous for improving data locality but it can also worsen the balance
1560 of the load in the system.
1565 <Term><Option>-qg<num></Option>:</Term>
1568 <IndexTerm><Primary>-qg<num> RTS option
1569 (parallel)</Primary></IndexTerm> (PARALLEL ONLY) Select a globalisation
1570 scheme. This option affects the
1571 generation of global addresses when transferring data. Global addresses are
1572 globally unique identifiers required to maintain sharing in the distributed
1573 graph structure. Currently this is a binary option. With <num>=0 full globalisation is used
1574 (default). This means a global address is generated for every closure that
1575 is transmitted. With <num>=1 a thunk-only globalisation scheme is
1576 used, which generated global address only for thunks. The latter case may
1577 lose sharing of data but has a reduced overhead in packing graph structures
1578 and maintaining internal tables of global addresses.
1589 <sect1 id="options-platform">
1590 <title>Platform-specific Flags</title>
1592 <indexterm><primary>-m* options</primary></indexterm>
1593 <indexterm><primary>platform-specific options</primary></indexterm>
1594 <indexterm><primary>machine-specific options</primary></indexterm>
1596 <para>Some flags only make sense for particular target
1602 <term><option>-mv8</option>:</term>
1604 <para>(SPARC machines)<indexterm><primary>-mv8 option (SPARC
1605 only)</primary></indexterm> Means to pass the like-named
1606 option to GCC; it says to use the Version 8 SPARC
1607 instructions, notably integer multiply and divide. The
1608 similiar <option>-m*</option> GCC options for SPARC also
1609 work, actually.</para>
1614 <term><option>-monly-[32]-regs</option>:</term>
1616 <para>(iX86 machines)<indexterm><primary>-monly-N-regs
1617 option (iX86 only)</primary></indexterm> GHC tries to
1618 “steal” four registers from GCC, for performance
1619 reasons; it almost always works. However, when GCC is
1620 compiling some modules with four stolen registers, it will
1621 crash, probably saying:
1624 Foo.hc:533: fixed or forbidden register was spilled.
1625 This may be due to a compiler bug or to impossible asm
1626 statements or clauses.
1629 Just give some registers back with
1630 <option>-monly-N-regs</option>. Try `3' first, then `2'.
1631 If `2' doesn't work, please report the bug to us.</para>
1640 <sect1 id="ext-core">
1641 <title>Generating and compiling External Core Files</title>
1643 <indexterm><primary>intermediate code generation</primary></indexterm>
1645 <para>GHC can dump its optimized intermediate code (said to be in “Core” format)
1646 to a file as a side-effect of compilation. Core files, which are given the suffix
1647 <filename>.hcr</filename>, can be read and processed by non-GHC back-end
1648 tools. The Core format is formally described in <ulink url="http://www.haskell.org/ghc/docs/papers/core.ps.gz"
1649 <citetitle>An External Representation for the GHC Core Language</citetitle></ulink>,
1650 and sample tools (in Haskell)
1651 for manipulating Core files are available in the GHC source distribution
1652 directory <literal>/fptools/ghc/utils/ext-core</literal>.
1653 Note that the format of <literal>.hcr</literal>
1654 files is <emphasis>different</emphasis> (though similar) to the Core output format generated
1655 for debugging purposes (<xref linkend="options-debugging">).</para>
1657 <para>The Core format natively supports notes which you can add to
1658 your source code using the <literal>CORE</literal> pragma (see <xref
1659 linkend="pragmas">).</para>
1664 <term><option>-fext-core</option></term>
1666 <primary><option>-fext-core</option></primary>
1669 <para>Generate <literal>.hcr</literal> files.</para>
1675 <para>GHC can also read in External Core files as source; just give the <literal>.hcr</literal> file on
1676 the command line, instead of the <literal>.hs</literal> or <literal>.lhs</literal> Haskell source.
1677 A current infelicity is that you need to give teh <literal>-fglasgow-exts</literal> flag too, because
1678 ordinary Haskell 98, when translated to External Core, uses things like rank-2 types.</para>
1687 ;;; Local Variables: ***
1689 ;;; sgml-parent-document: ("users_guide.sgml" "book" "chapter") ***