1 <!DOCTYPE Article PUBLIC "-//OASIS//DTD DocBook V3.1//EN">
3 <Article id="building-guide">
7 <Title>Building the Glasgow Functional Programming Tools Suite</Title>
8 <Author><OtherName>The GHC Team</OtherName></Author>
9 <Address><Email>glasgow-haskell-{users,bugs}@haskell.org</Email></Address>
10 <PubDate>November 2001</PubDate>
13 <para>The Glasgow fptools suite is a collection of Functional
14 Programming related tools, including the Glasgow Haskell
15 Compiler (GHC). The source code for the whole suite is kept in
16 a single CVS repository and shares a common build and
17 installation system.</para>
19 <para>This guide is intended for people who want to build or
20 modify programs from the Glasgow <Literal>fptools</Literal>
21 suite (as distinct from those who merely want to
22 <Emphasis>run</Emphasis> them). Installation instructions are
23 now provided in the user guide.</para>
25 <para>The bulk of this guide applies to building on Unix
26 systems; see <XRef LinkEnd="winbuild"> for Windows notes.</para>
32 <sect1 id="sec-getting">
33 <title>Getting the sources</title>
35 <para>You can get your hands on the <literal>fptools</literal>
41 <term><indexterm><primary>Source
42 distributions</primary></indexterm>Source distributions</term>
44 <para>You have a supported platform, but (a) you like
45 the warm fuzzy feeling of compiling things yourself;
46 (b) you want to build something ``extra”—e.g., a
47 set of libraries with strictness-analysis turned off; or
48 (c) you want to hack on GHC yourself.</para>
50 <para>A source distribution contains complete sources for
51 one or more projects in the <literal>fptools</literal>
52 suite. Not only that, but the more awkward
53 machine-independent steps are done for you. For example, if
55 <command>happy</command><indexterm><primary>happy</primary></indexterm>
56 you'll find it convenient that the source distribution
57 contains the result of running <command>happy</command> on
58 the parser specifications. If you don't want to alter the
59 parser then this saves you having to find and install
60 <command>happy</command>. You will still need a working
61 version of GHC (preferably version 4.08+) on your machine in
62 order to compile (most of) the sources, however.</para>
67 <term>The CVS repository.</term>
68 <indexterm><primary>CVS repository</primary>
71 <para>We make releases infrequently. If you want more
72 up-to-the minute (but less tested) source code then you need
73 to get access to our CVS repository.</para>
75 <para>All the <literal>fptools</literal> source code is held
76 in a CVS repository. CVS is a pretty good source-code
77 control system, and best of all it works over the
80 <para>The repository holds source code only. It holds no
81 mechanically generated files at all. So if you check out a
82 source tree from CVS you will need to install every utility
83 so that you can build all the derived files from
86 <para>More information about our CVS repository can be found
87 in <xref linkend="sec-cvs">.</para>
92 <para>If you are going to do any building from sources (either
93 from a source distribution or the CVS repository) then you need to
94 read all of this manual in detail.</para>
98 <title>Using the CVS repository</title>
100 <para>We use <ulink url="http://www.cvshome.org/">CVS</ulink> (Concurrent Version System) to keep track of our
101 sources for various software projects. CVS lets several people
102 work on the same software at the same time, allowing changes to be
103 checked in incrementally. </para>
105 <para>This section is a set of guidelines for how to use our CVS
106 repository, and will probably evolve in time. The main thing to
107 remember is that most mistakes can be undone, but if there's
108 anything you're not sure about feel free to bug the local CVS
109 meister (namely Jeff Lewis
110 <email>jlewis@galconn.com</email>). </para>
112 <sect2 id="cvs-access">
113 <title>Getting access to the CVS Repository</title>
115 <para>You can access the repository in one of two ways:
116 read-only (<xref linkend="cvs-read-only">), or read-write (<xref
117 linkend="cvs-read-write">).</para>
119 <sect3 id="cvs-read-only">
120 <title>Remote Read-only CVS Access</title>
122 <para>Read-only access is available to anyone - there's no
123 need to ask us first. With read-only CVS access you can do
124 anything except commit changes to the repository. You can
125 make changes to your local tree, and still use CVS's merge
126 facility to keep your tree up to date, and you can generate
127 patches using 'cvs diff' in order to send to us for
130 <para>To get read-only access to the repository:</para>
134 <para>Make sure that <application>cvs</application> is
135 installed on your machine.</para>
138 <para>Set your <literal>$CVSROOT</literal> environment variable to
139 <literal>:pserver:anoncvs@glass.cse.ogi.edu:/cvs</literal></para>
142 <para>Run the command</para>
146 <para>The password is simply <literal>cvs</literal>. This
147 sets up a file in your home directory called
148 <literal>.cvspass</literal>, which squirrels away the
149 dummy password, so you only need to do this step once.</para>
153 <para>Now go to <xref linkend="cvs-first">.</para>
158 <sect3 id="cvs-read-write">
159 <title>Remote Read-Write CVS Access</title>
161 <para>We generally supply read-write access to folk doing
162 serious development on some part of the source tree, when
163 going through us would be a pain. If you're developing some
164 feature, or think you have the time and inclination to fix
165 bugs in our sources, feel free to ask for read-write
166 access. There is a certain amount of responsibility that goes
167 with commit privileges; we are more likely to grant you access
168 if you've demonstrated your competence by sending us patches
169 via mail in the past.</para>
171 <para>To get remote read-write CVS access, you need to do the
172 following steps.</para>
176 <para>Make sure that <literal>cvs</literal> and
177 <literal>ssh</literal> are both installed on your
182 <para>Generate a DSA private-key/public-key pair, thus:</para>
186 <para>(<literal>ssh-keygen</literal> comes with
187 <literal>ssh</literal>.) Running <literal>ssh-keygen
188 -d</literal> creates the private and public keys in
189 <literal>$HOME/.ssh/id_dsa</literal> and
190 <literal>$HOME/.ssh/id_dsa.pub</literal> respectively
191 (assuming you accept the standard defaults).</para>
193 <para><literal>ssh-keygen -d</literal> will only work if
194 you have Version 2 <literal>ssh</literal> installed; it
195 will fail harmlessly otherwise. If you only have Version
196 1 you can instead generate an RSA key pair using plain</para>
201 <para>Doing so creates the private and public RSA keys in
202 <literal>$HOME/.ssh/identity</literal> and
203 <literal>$HOME/.ssh/identity.pub</literal>
206 <para>[Deprecated.] Incidentally, you can force a Version
207 2 <literal>ssh</literal> to use the Version 1 protocol by
208 creating <literal>$HOME/config</literal> with the
209 following in it:</para>
217 <para>In both cases, <literal>ssh-keygen</literal> will
218 ask for a <firstterm>passphrase</firstterm>. The
219 passphrase is a password that protects your private key.
220 In response to the 'Enter passphrase' question, you can
224 <para>[Recommended.] Enter a passphrase, which you
225 will quote each time you use CVS.
226 <literal>ssh-agent</literal> makes this entirely
230 <para>[Deprecated.] Just hit return (i.e. use an empty
231 passphrase); then you won't need to quote the
232 passphrase when using CVS. The downside is that
233 anyone who can see into your <literal>.ssh</literal>
234 directory, and thereby get your private key, can mess
235 up the repository. So you must keep the
236 <literal>.ssh</literal> directory with draconian
237 no-access permissions.</para>
243 [Windows users.] The programs <command>ssh-keygen1</command>, <command>ssh1</command>, and <command>cvs</command>,
244 seem to lock up <command>bash</command> entirely if they try to get user input (e.g. if
245 they ask for a password). To solve this, start up <filename>cmd.exe</filename>
246 and run it as follows:
248 c:\tmp> set CYGWIN32=tty
249 c:\tmp> c:/user/local/bin/ssh-keygen1
252 <para>[Windows users.] To protect your
253 <literal>.ssh</literal> from access by anyone else,
254 right-click your <literal>.ssh</literal> directory, and
255 select <literal>Properties</literal>. If you are not on
256 the access control list, add yourself, and give yourself
257 full permissions (the second panel). Remove everyone else
258 from the access control list. Don't leave them there but
259 deny them access, because 'they' may be a list that
261 <para>[March 2003] In fact <command>ssh</command> 3.6.1 now seems to <emphasis>require</emphasis>
262 you to have Unix permissions 600 (read/write for owner only)
263 on the <literal>.ssh/identity</literal> file, else it
264 bombs out. For your local C drive, it seems that <literal>chmod 600 identity</literal> works,
265 but on Windows NT/XP, it doesn't work on a network drive (exact dteails obscure).
266 The solution seems to be to set the CYGWIN environment
267 variable to "<literal>ntsec neta</literal>". The CYGWIN environment variable is discussed
268 in <ulink url="http://cygwin.com/cygwin-ug-net/using-cygwinenv.html">the Cygwin User's Guide</ulink>,
269 and there are more details in <ulink url="http://cygwin.com/faq/faq_4.html#SEC44">the Cygwin FAQ</ulink>.
274 <para>Send a message to to the CVS repository
275 administrator (currently Jeff Lewis
276 <email>jeff@galconn.com</email>), containing:</para>
279 <para>Your desired user-name.</para>
282 <para>Your <literal>.ssh/id_dsa.pub</literal> (or
283 <literal>.ssh/identity.pub</literal>).</para>
286 <para>He will set up your account.</para>
290 <para>Set the following environment variables:</para>
294 <constant>$HOME</constant>: points to your home directory. This is where CVS
295 will look for its <filename>.cvsrc</filename> file.
301 <constant>$CVS_RSH</constant> to <filename>ssh</filename>
303 <para>[Windows users.] Setting your <literal>CVS_RSH</literal> to
304 <literal>ssh</literal> assumes that your CVS client
305 understands how to execute shell script
306 ("#!"s,really), which is what
307 <literal>ssh</literal> is. This may not be the case on
308 Win32 platforms, so in that case set <literal>CVS_RSH</literal> to
309 <literal>ssh1</literal>.</para>
313 <para><literal>$CVSROOT</literal> to
314 <literal>:ext:</literal><replaceable>your-username</replaceable>
315 <literal>@cvs.haskell.org:/home/cvs/root</literal>
316 where <replaceable>your-username</replaceable> is your user name on
317 <literal>cvs.haskell.org</literal>.
319 <para>The <literal>CVSROOT</literal> environment variable will
320 be recorded in the checked-out tree, so you don't need to set
321 this every time. </para>
327 <constant>$CVSEDITOR</constant>: <filename>bin/gnuclient.exe</filename>
328 if you want to use an Emacs buffer for typing in those long commit messages.
334 <constant>$SHELL</constant>: To use bash as the shell in Emacs, you need to
335 set this to point to <filename>bash.exe</filename>.
346 Put the following in <filename>$HOME/.cvsrc</filename>:
357 These are the default options for the specified CVS commands,
358 and represent better defaults than the usual ones. (Feel
359 free to change them.)
363 [Windows users.] Filenames starting with <filename>.</filename> were illegal in
364 the 8.3 DOS filesystem, but that restriction should have
365 been lifted by now (i.e., you're using VFAT or later filesystems.) If
366 you're still having problems creating it, don't worry; <filename>.cvsrc</filename> is entirely
374 <para>[Experts.] Once your account is set up, you can get
375 access from other machines without bothering Jeff, thus:</para>
378 <para>Generate a public/private key pair on the new
382 <para>Use ssh to log in to
383 <literal>cvs.haskell.org</literal>, from your old
387 <para>Add the public key for the new machine to the file
388 <literal>$HOME/ssh/authorized_keys</literal> on
389 <literal>cvs.haskell.org</literal>.
390 (<literal>authorized_keys2</literal>, I think, for Version
394 <para>Make sure that the new version of
395 <literal>authorized_keys</literal> still has 600 file
404 <sect2 id="cvs-first">
405 <title>Checking Out a Source Tree</title>
409 <para>Make sure you set your <literal>CVSROOT</literal>
410 environment variable according to either of the remote
411 methods above. The Approved Way to check out a source tree
412 is as follows:</para>
415 $ cvs checkout fpconfig
418 <para>At this point you have a new directory called
419 <literal>fptools</literal> which contains the basic stuff
420 for the fptools suite, including the configuration files and
421 some other junk. </para>
423 <para>[Windows users.] The following messages appear to be harmless:
425 setsockopt IPTOS_LOWDELAY: Invalid argument
426 setsockopt IPTOS_THROUGHPUT: Invalid argument
431 <para>You can call the fptools directory whatever you like,
432 CVS won't mind: </para>
435 $ mv fptools <replaceable>directory</replaceable>
438 <para> NB: after you've read the CVS manual you might be
439 tempted to try</para>
441 $ cvs checkout -d <replaceable>directory</replaceable> fpconfig
444 <para>instead of checking out <literal>fpconfig</literal>
445 and then renaming it. But this doesn't work, and will
446 result in checking out the entire repository instead of just
447 the <literal>fpconfig</literal> bit.</para>
449 $ cd <replaceable>directory</replaceable>
450 $ cvs checkout ghc hslibs libraries
453 <para>The second command here checks out the relevant
454 modules you want to work on. For a GHC build, for instance,
455 you need at least the <literal>ghc</literal>,
456 <literal>hslibs</literal> and <literal>libraries</literal>
457 modules (for a full list of the projects available, see
458 <xref linkend="projects">).</para>
463 <sect2 id="cvs-committing">
464 <title>Committing Changes</title>
466 <para>This is only if you have read-write access to the
467 repository. For anoncvs users, CVS will issue a "read-only
468 repository" error if you try to commit changes.</para>
472 <para>Build the software, if necessary. Unless you're just
473 working on documentation, you'll probably want to build the
474 software in order to test any changes you make.</para>
478 <para>Make changes. Preferably small ones first.</para>
482 <para>Test them. You can see exactly what changes you've
483 made by using the <literal>cvs diff</literal> command:</para>
487 <para>lists all the changes (using the
488 <literal>diff</literal> command) in and below the current
489 directory. In emacs, <literal>C-c C-v =</literal> runs
490 <literal>cvs diff</literal> on the current buffer and shows
491 you the results.</para>
495 <para>Before checking in a change, you need to update your
502 <para>This pulls in any changes that other people have made,
503 and merges them with yours. If there are any conflicts, CVS
504 will tell you, and you'll have to resolve them before you
505 can check your changes in. The documentation describes what
506 to do in the event of a conflict.</para>
508 <para>It's not always necessary to do a full cvs update
509 before checking in a change, since CVS will always tell you
510 if you try to check in a file that someone else has changed.
511 However, you should still update at regular intervals to
512 avoid making changes that don't work in conjuction with
513 changes that someone else made. Keeping an eye on what goes
514 by on the mailing list can help here.</para>
518 <para>When you're happy that your change isn't going to
519 break anything, check it in. For a one-file change:</para>
522 $ cvs commit <replaceable>filename</replaceable>
525 <para>CVS will then pop up an editor for you to enter a
526 "commit message", this is just a short description
527 of what your change does, and will be kept in the history of
530 <para>If you're using emacs, simply load up the file into a
531 buffer and type <literal>C-x C-q</literal>, and emacs will
532 prompt for a commit message and then check in the file for
535 <para>For a multiple-file change, things are a bit
536 trickier. There are several ways to do this, but this is the
537 way I find easiest. First type the commit message into a
538 temporary file. Then either</para>
541 $ cvs commit -F <replaceable>commit-message</replaceable> <replaceable>file_1</replaceable> .... <replaceable>file_n</replaceable>
544 <para>or, if nothing else has changed in this part of the
548 $ cvs commit -F <replaceable>commit-message</replaceable> <replaceable>directory</replaceable>
551 <para>where <replaceable>directory</replaceable> is a common
552 parent directory for all your changes, and
553 <replaceable>commit-message</replaceable> is the name of the
554 file containing the commit message.</para>
556 <para>Shortly afterwards, you'll get some mail from the
557 relevant mailing list saying which files changed, and giving
558 the commit message. For a multiple-file change, you should
559 still get only <emphasis>one</emphasis> message.</para>
564 <sect2 id="cvs-update">
565 <title>Updating Your Source Tree</title>
567 <para>It can be tempting to cvs update just part of a source
568 tree to bring in some changes that someone else has made, or
569 before committing your own changes. This is NOT RECOMMENDED!
570 Quite often changes in one part of the tree are dependent on
571 changes in another part of the tree (the
572 <literal>mk/*.mk</literal> files are a good example where
573 problems crop up quite often). Having an inconsistent tree is a
574 major cause of headaches. </para>
576 <para>So, to avoid a lot of hassle, follow this recipe for
577 updating your tree: </para>
581 $ cvs update -P 2>&1 | tee log</screen>
583 <para>Look at the log file, and fix any conflicts (denoted by a
584 <quote>C</quote> in the first column). New directories may have
585 appeared in the repository; CVS doesn't check these out by
586 default, so to get new directories you have to explicitly do
588 $ cvs update -d</screen>
589 in each project subdirectory. Don't do this at the top level,
590 because then <emphasis>all</emphasis> the projects will be
593 <para>If you're using multiple build trees, then for every build
594 tree you have pointing at this source tree, you need to update
595 the links in case any new files have appeared: </para>
598 $ cd <replaceable>build-tree</replaceable>
599 $ lndir <replaceable>source-tree</replaceable>
602 <para>Some files might have been removed, so you need to remove
603 the links pointing to these non-existent files:</para>
606 $ find . -xtype l -exec rm '{}' \;
609 <para>To be <emphasis>really</emphasis> safe, you should do
612 <screen>$ gmake all</screen>
614 <para>from the top-level, to update the dependencies and build
615 any changed files. </para>
618 <sect2 id="cvs-tags">
619 <title>GHC Tag Policy</title>
621 <para>If you want to check out a particular version of GHC,
622 you'll need to know how we tag versions in the repository. The
623 policy (as of 4.04) is:</para>
627 <para>The tree is branched before every major release. The
628 branch tag is <literal>ghc-x-xx-branch</literal>, where
629 <literal>x-xx</literal> is the version number of the release
630 with the <literal>'.'</literal> replaced by a
631 <literal>'-'</literal>. For example, the 4.04 release lives
632 on <literal>ghc-4-04-branch</literal>.</para>
636 <para>The release itself is tagged with
637 <literal>ghc-x-xx</literal> (on the branch). eg. 4.06 is
638 called <literal>ghc-4-06</literal>.</para>
642 <para>We didn't always follow these guidelines, so to see
643 what tags there are for previous versions, do <literal>cvs
644 log</literal> on a file that's been around for a while (like
645 <literal>fptools/ghc/README</literal>).</para>
649 <para>So, to check out a fresh GHC 4.06 tree you would
653 $ cvs co -r ghc-4-06 fpconfig
655 $ cvs co -r ghc-4-06 ghc hslibs
659 <sect2 id="cvs-hints">
660 <title>General Hints</title>
664 <para>As a general rule: commit changes in small units,
665 preferably addressing one issue or implementing a single
666 feature. Provide a descriptive log message so that the
667 repository records exactly which changes were required to
668 implement a given feature/fix a bug. I've found this
669 <emphasis>very</emphasis> useful in the past for finding out
670 when a particular bug was introduced: you can just wind back
671 the CVS tree until the bug disappears.</para>
675 <para>Keep the sources at least *buildable* at any given
676 time. No doubt bugs will creep in, but it's quite easy to
677 ensure that any change made at least leaves the tree in a
678 buildable state. We do nightly builds of GHC to keep an eye
679 on what things work/don't work each day and how we're doing
680 in relation to previous verions. This idea is truely wrecked
681 if the compiler won't build in the first place!</para>
685 <para>To check out extra bits into an already-checked-out
686 tree, use the following procedure. Suppose you have a
687 checked-out fptools tree containing just ghc, and you want
688 to add nofib to it:</para>
699 $ cvs update -d nofib
702 <para>(the -d flag tells update to create a new
703 directory). If you just want part of the nofib suite, you
708 $ cvs checkout nofib/spectral
711 <para>This works because <literal>nofib</literal> is a
712 module in its own right, and spectral is a subdirectory of
713 the nofib module. The path argument to checkout must always
714 start with a module name. There's no equivalent form of this
715 command using <literal>update</literal>.</para>
721 <sect1 id="projects">
722 <title>What projects are there?</title>
724 <para>The <literal>fptools</literal> suite consists of several
725 <firstterm>projects</firstterm>, most of which can be downloaded,
726 built and installed individually. Each project corresponds to a
727 subdirectory in the source tree, and if checking out from CVS then
728 each project can be checked out individually by sitting in the top
729 level of your source tree and typing <command>cvs checkout
730 <replaceable>project</replaceable></command>.</para>
732 <para>Here is a list of the projects currently available:</para>
736 <term><literal>ghc</literal></term>
737 <indexterm><primary><literal>ghc</literal></primary>
738 <secondary>project</secondary></indexterm>
740 <para>The <ulink url="http://www.haskell.org/ghc/">Glasgow
741 Haskell Compiler</ulink> (minus libraries). Absolutely
742 required for building GHC.</para>
747 <term><literal>glafp-utils</literal></term>
748 <indexterm><primary><literal>glafp-utils</literal></primary><secondary>project</secondary></indexterm>
750 <para>Utility programs, some of which are used by the
751 build/installation system. Required for pretty much
757 <term><literal>green-card</literal></term>
758 <indexterm><primary><literal>green-card</literal></primary><secondary>project</secondary></indexterm>
761 url="http://www.haskell.org/greencard/">Green Card</ulink>
762 system for generating Haskell foreign function
768 <term><literal>haggis</literal></term>
769 <indexterm><primary><literal>haggis</literal></primary><secondary>project</secondary></indexterm>
772 url="http://www.dcs.gla.ac.uk/fp/software/haggis/">Haggis</ulink>
773 Haskell GUI framework.</para>
778 <term><literal>haddock</literal></term>
779 <indexterm><primary><literal>haddock</literal></primary><secondary>project</secondary></indexterm>
782 url="http://www.haskell.org/haddock/">Haddock</ulink>
783 documentation tool.</para>
788 <term><literal>happy</literal></term>
789 <indexterm><primary><literal>happy</literal></primary><secondary>project</secondary></indexterm>
792 url="http://www.haskell.org/happy/">Happy</ulink> Parser
798 <term><literal>hdirect</literal></term>
799 <indexterm><primary><literal>hdirect</literal></primary><secondary>project</secondary></indexterm>
802 url="http://www.haskell.org/hdirect/">H/Direct</ulink>
803 Haskell interoperability tool.</para>
808 <term><literal>hood</literal></term>
809 <indexterm><primary><literal>hood</literal></primary><secondary>project</secondary></indexterm>
811 <para>The <ulink url="http://www.haskell.org/hood/">Haskell
812 Object Observation Debugger</ulink>.</para>
817 <term><literal>hslibs</literal></term>
818 <indexterm><primary><literal>hslibs</literal></primary><secondary>project</secondary></indexterm>
820 <para>Supplemental libraries for GHC
821 (<emphasis>required</emphasis> for building GHC).</para>
826 <term><literal>libraries</literal></term>
827 <indexterm><primary><literal></literal></primary><secondary>project</secondary></indexterm>
829 <para>Hierarchical Haskell library suite
830 (<emphasis>required</emphasis> for building GHC).</para>
835 <term><literal>mhms</literal></term>
836 <indexterm><primary><literal></literal></primary><secondary>project</secondary></indexterm>
838 <para>The Modular Haskell Metric System.</para>
843 <term><literal>nofib</literal></term>
844 <indexterm><primary><literal>nofib</literal></primary><secondary>project</secondary></indexterm>
846 <para>The NoFib suite: A collection of Haskell programs used
847 primarily for benchmarking.</para>
852 <term><literal>testsuite</literal></term>
853 <indexterm><primary><literal>testsuite</literal></primary><secondary>project</secondary></indexterm>
855 <para>A testing framework, including GHC's regression test
861 <para>So, to build GHC you need at least the
862 <literal>ghc</literal>, <literal>libraries</literal> and
863 <literal>hslibs</literal> projects (a GHC source distribution will
864 already include the bits you need).</para>
867 <sect1 id="sec-build-checks">
868 <title>Things to check before you start</title>
870 <para>Here's a list of things to check before you get
876 <indexterm><primary>Disk space needed</primary></indexterm>
877 <para>Disk space needed: from about 100Mb for a basic GHC
878 build, up to probably 500Mb for a GHC build with everything
879 included (libraries built several different ways,
884 <para>Use an appropriate machine / operating system. <xref
885 linkend="sec-port-info"> lists the supported platforms; if
886 yours isn't amongst these then you can try porting GHC (see
887 <xref linkend="sec-porting-ghc">).</para>
891 <para>Be sure that the “pre-supposed” utilities are
892 installed. <Xref LinkEnd="sec-pre-supposed">
897 <para>If you have any problem when building or installing the
898 Glasgow tools, please check the “known pitfalls” (<Xref
899 LinkEnd="sec-build-pitfalls">). Also check the FAQ for the
900 version you're building, which is part of the User's Guide and
901 available on the <ulink URL="http://www.haskell.org/ghc/" >GHC web
904 <indexterm><primary>bugs</primary><secondary>known</secondary></indexterm>
906 <para>If you feel there is still some shortcoming in our
907 procedure or instructions, please report it.</para>
909 <para>For GHC, please see the <ulink
910 url="http://www.haskell.org/ghc/docs/latest/set/bug-reporting.html">bug-reporting
911 section of the GHC Users' Guide</ulink>, to maximise the
912 usefulness of your report.</para>
914 <indexterm><primary>bugs</primary><secondary>seporting</secondary></indexterm>
915 <para>If in doubt, please send a message to
916 <email>glasgow-haskell-bugs@haskell.org</email>.
917 <indexterm><primary>bugs</primary><secondary>mailing
918 list</secondary></indexterm></para>
923 <sect1 id="sec-port-info">
924 <title>What machines the Glasgow tools run on</title>
926 <indexterm><primary>ports</primary><secondary>GHC</secondary></indexterm>
927 <indexterm><primary>GHC</primary><secondary>ports</secondary></indexterm>
928 <indexterm><primary>platforms</primary><secondary>supported</secondary></indexterm>
930 <para>The main question is whether or not the Haskell compiler
931 (GHC) runs on your platform.</para>
933 <para>A “platform” is a
934 architecture/manufacturer/operating-system combination, such as
935 <literal>sparc-sun-solaris2</literal>. Other common ones are
936 <literal>alpha-dec-osf2</literal>,
937 <literal>hppa1.1-hp-hpux9</literal>,
938 <literal>i386-unknown-linux</literal>,
939 <literal>i386-unknown-solaris2</literal>,
940 <literal>i386-unknown-freebsd</literal>,
941 <literal>i386-unknown-cygwin32</literal>,
942 <literal>m68k-sun-sunos4</literal>,
943 <literal>mips-sgi-irix5</literal>,
944 <literal>sparc-sun-sunos4</literal>,
945 <literal>sparc-sun-solaris2</literal>,
946 <literal>powerpc-ibm-aix</literal>.</para>
948 <para>Some libraries may only work on a limited number of
949 platforms; for example, a sockets library is of no use unless the
950 operating system supports the underlying BSDisms.</para>
953 <title>What platforms the Haskell compiler (GHC) runs on</title>
955 <indexterm><primary>fully-supported platforms</primary></indexterm>
956 <indexterm><primary>native-code generator</primary></indexterm>
957 <indexterm><primary>registerised ports</primary></indexterm>
958 <indexterm><primary>unregisterised ports</primary></indexterm>
960 <para>The GHC hierarchy of Porting Goodness: (a) Best is a
961 native-code generator; (b) next best is a
962 “registerised” port; (c) the bare minimum is an
963 “unregisterised” port.
964 (“Unregisterised” is so terrible that we won't say
965 more about it).</para>
967 <para>We use Sparcs running Solaris 2.7 and x86 boxes running
968 FreeBSD and Linux, so those are the best supported platforms,
969 unsurprisingly.</para>
971 <para>Here's everything that's known about GHC ports. We
972 identify platforms by their “canonical”
973 CPU/Manufacturer/OS triple.</para>
977 <term>alpha-dec-{osf,linux,freebsd,openbsd,netbsd}:</term>
978 <indexterm><primary>alpha-dec-osf</primary></indexterm>
979 <indexterm><primary>alpha-dec-linux</primary></indexterm>
980 <indexterm><primary>alpha-dec-freebsd</primary></indexterm>
981 <indexterm><primary>alpha-dec-openbsd</primary></indexterm>
982 <indexterm><primary>alpha-dec-netbsd</primary></indexterm>
985 <para>The OSF port is currently working (as of GHC version
986 5.02.1) and well supported. The native code generator is
987 currently non-working. Other operating systems will
988 require some minor porting.</para>
993 <term>sparc-sun-sunos4</term>
994 <indexterm><primary>sparc-sun-sunos4</primary></indexterm>
996 <para>Probably works with minor tweaks, hasn't been tested
1002 <term>sparc-sun-solaris2</term>
1003 <indexterm><primary>sparc-sun-solaris2</primary></indexterm>
1005 <para>Fully supported (at least for Solaris 2.7),
1006 including native-code generator.</para>
1011 <term>hppa1.1-hp-hpux (HP-PA boxes running HPUX 9.x)</term>
1012 <indexterm><primary>hppa1.1-hp-hpux</primary></indexterm>
1014 <para>A registerised port is available for version 4.08,
1015 but GHC hasn't been built on that platform since (as far
1016 as we know). No native-code generator.</para>
1021 <term>i386-unknown-linux (PCs running Linux, ELF binary format)</term>
1022 <indexterm><primary>i386-*-linux</primary></indexterm>
1024 <para>GHC works registerised and has a native code
1025 generator. You <Emphasis>must</Emphasis> have GCC 2.7.x
1026 or later. NOTE about <literal>glibc</literal> versions:
1027 GHC binaries built on a system running <literal>glibc
1028 2.0</literal> won't work on a system running
1029 <literal>glibc 2.1</literal>, and vice versa. In general,
1030 don't expect compatibility between
1031 <literal>glibc</literal> versions, even if the shared
1032 library version hasn't changed.</para>
1037 <term>i386-unknown-freebsd (PCs running FreeBSD 2.2 or
1039 <indexterm><primary>i386-unknown-freebsd</primary></indexterm>
1041 <para>GHC works registerised. Pre-built packages are
1042 available in the native package format, so if you just
1043 need binaries you're better off just installing the
1044 package (it might even be on your installation
1050 <term>i386-unknown-openbsd (PCs running OpenBSD)</term>
1051 <indexterm><primary>i386-unknown-openbsd</primary></indexterm>
1053 <para>Supported, with native code generator. Packages are
1054 available through the ports system in the native package
1060 <term>i386-unknown-netbsd (PCs running NetBSD and
1062 <indexterm><primary>i386-unknown-netbsd</primary></indexterm>
1064 <para>Will require some minor porting effort, but should
1065 work registerised.</para>
1070 <term>i386-unknown-mingw32 (PCs running Windows)</term>
1071 <indexterm><primary>i386-unknown-mingw32</primary></indexterm>
1073 <para>Fully supported under Win9x, WinNT, Win2k, and
1074 WinXP. Includes a native code generator. Building from
1075 source requires a recent <ulink
1076 url="http://www.cygwin.com/">Cygwin</ulink> distribution
1077 to be installed.</para>
1082 <term>ia64-unknown-linux</term>
1083 <indexterm><primary>ia64-unknown-linux</primary></indexterm>
1085 <para>GHC currently works unregisterised. A registerised
1086 port is in progress.</para>
1091 <term>mips-sgi-irix5</term>
1092 <indexterm><primary>mips-sgi-irix[5-6]</primary></indexterm>
1094 <para>Port has worked in the past, but hasn't been tested
1095 for some time (and will certainly have rotted in various
1096 ways). As usual, we don't have access to machines and
1097 there hasn't been an overwhelming demand for this port,
1098 but feel free to get in touch.</para>
1103 <term>powerpc-ibm-aix</term>
1104 <indexterm><primary>powerpc-ibm-aix</primary></indexterm>
1106 <para>Port currently doesn't work, needs some minimal
1107 porting effort. As usual, we don't have access to
1108 machines and there hasn't been an overwhelming demand for
1109 this port, but feel free to get in touch.</para>
1114 <term>powerpc-apple-darwin</term>
1115 <indexterm><primary>powerpc-apple-darwin</primary></indexterm>
1117 <para>Supported registerised. No native code
1123 <term>powerpc-apple-linux</term>
1124 <indexterm><primary>powerpc-apple-linux</primary></indexterm>
1126 <para>Not supported (yet).</para>
1131 <para>Various other systems have had GHC ported to them in the
1132 distant past, including various Motorola 68k boxes. The 68k
1133 support still remains, but porting to one of these systems will
1134 certainly be a non-trivial task.</para>
1138 <title>What machines the other tools run on</title>
1140 <para>Unless you hear otherwise, the other tools work if GHC
1146 <sect1 id="sec-pre-supposed">
1147 <title>Installing pre-supposed utilities</title>
1149 <indexterm><primary>pre-supposed utilities</primary></indexterm>
1150 <indexterm><primary>utilities, pre-supposed</primary></indexterm>
1152 <para>Here are the gory details about some utility programs you
1153 may need; <command>perl</command>, <command>gcc</command> and
1154 <command>happy</command> are the only important
1155 ones. (PVM<indexterm><primary>PVM</primary></indexterm> is
1156 important if you're going for Parallel Haskell.) The
1157 <command>configure</command><indexterm><primary>configure</primary></indexterm>
1158 script will tell you if you are missing something.</para>
1164 <indexterm><primary>pre-supposed: GHC</primary></indexterm>
1165 <indexterm><primary>GHC, pre-supposed</primary></indexterm>
1167 <para>GHC is required to build many of the tools, including
1168 GHC itself. If you need to port GHC to your platform
1169 because there isn't a binary distribution of GHC available,
1170 then see <xref linkend="sec-porting-ghc">.</para>
1172 <para>Which version of GHC you need will depend on the
1173 packages you intend to build. GHC itself will normally
1174 build using one of several older versions of itself - check
1175 the announcement or release notes for details.</para>
1181 <indexterm><primary>pre-supposed: Perl</primary></indexterm>
1182 <indexterm><primary>Perl, pre-supposed</primary></indexterm>
1184 <para><emphasis>You have to have Perl to proceed!</emphasis>
1185 Perl version 5 at least is required. GHC has been known to
1186 tickle bugs in Perl, so if you find that Perl crashes when
1187 running GHC try updating (or downgrading) your Perl
1188 installation. Versions of Perl that we use and are known to
1189 be fairly stable are 5.005 and 5.6.1.</para>
1191 <para>For Win32 platforms, you should use the binary
1192 supplied in the InstallShield (copy it to
1193 <filename>/bin</filename>). The Cygwin-supplied Perl seems
1196 <para>Perl should be put somewhere so that it can be invoked
1197 by the <literal>#!</literal> script-invoking
1198 mechanism. The full pathname may need to be less than 32
1199 characters long on some systems.</para>
1204 <term>GNU C (<command>gcc</command>)</term>
1205 <indexterm><primary>pre-supposed: GCC (GNU C
1206 compiler)</primary></indexterm> <indexterm><primary>GCC (GNU C
1207 compiler), pre-supposed</primary></indexterm>
1209 <para>We recommend using GCC version 2.95.2 on all
1210 platforms. Failing that, version 2.7.2 is stable on most
1211 platforms. Earlier versions of GCC can be assumed not to
1212 work, and versions in between 2.7.2 and 2.95.2 (including
1213 <command>egcs</command>) have varying degrees of stability
1214 depending on the platform.</para>
1216 <para>If your GCC dies with “internal error” on
1217 some GHC source file, please let us know, so we can report
1218 it and get things improved. (Exception: on iX86
1219 boxes—you may need to fiddle with GHC's
1220 <option>-monly-N-regs</option> option; see the User's
1226 <term>GNU Make</term>
1227 <indexterm><primary>make</primary><secondary>GNU</secondary>
1230 <para>The fptools build system makes heavy use of features
1231 specific to GNU <command>make</command>, so you must have
1232 this installed in order to build any of the fptools
1239 <indexterm><primary>Happy</primary></indexterm>
1241 <para>Happy is a parser generator tool for Haskell, and is
1242 used to generate GHC's parsers. Happy is written in
1243 Haskell, and is a project in the CVS repository
1244 (<literal>fptools/happy</literal>). It can be built from
1245 source, but bear in mind that you'll need GHC installed in
1246 order to build it. To avoid the chicken/egg problem,
1247 install a binary distribtion of either Happy or GHC to get
1248 started. Happy distributions are available from <ulink
1249 url="http://www.haskell.org/happy/">Happy's Web
1250 Page</ulink>.</para>
1255 <term>Autoconf</term>
1256 <indexterm><primary>pre-supposed: Autoconf</primary></indexterm>
1257 <indexterm><primary>Autoconf, pre-supposed</primary></indexterm>
1259 <para>GNU Autoconf is needed if you intend to build from the
1260 CVS sources, it is <emphasis>not</emphasis> needed if you
1261 just intend to build a standard source distribution.</para>
1263 <para>Autoconf builds the <command>configure</command>
1264 script from <filename>configure.in</filename> and
1265 <filename>aclocal.m4</filename>. If you modify either of
1266 these files, you'll need <command>autoconf</command> to
1267 rebuild <filename>configure</filename>.</para>
1272 <term><command>sed</command></term>
1273 <indexterm><primary>pre-supposed: sed</primary></indexterm>
1274 <indexterm><primary>sed, pre-supposed</primary></indexterm>
1276 <para>You need a working <command>sed</command> if you are
1277 going to build from sources. The build-configuration stuff
1278 needs it. GNU sed version 2.0.4 is no good! It has a bug
1279 in it that is tickled by the build-configuration. 2.0.5 is
1280 OK. Others are probably OK too (assuming we don't create too
1281 elaborate configure scripts.)</para>
1286 <para>One <literal>fptools</literal> project is worth a quick note
1287 at this point, because it is useful for all the others:
1288 <literal>glafp-utils</literal> contains several utilities which
1289 aren't particularly Glasgow-ish, but Occasionally Indispensable.
1290 Like <command>lndir</command> for creating symbolic link
1293 <sect2 id="pre-supposed-gph-tools">
1294 <title>Tools for building parallel GHC (GPH)</title>
1298 <term>PVM version 3:</term>
1299 <indexterm><primary>pre-supposed: PVM3 (Parallel Virtual Machine)</primary></indexterm>
1300 <indexterm><primary>PVM3 (Parallel Virtual Machine), pre-supposed</primary></indexterm>
1302 <para>PVM is the Parallel Virtual Machine on which
1303 Parallel Haskell programs run. (You only need this if you
1304 plan to run Parallel Haskell. Concurent Haskell, which
1305 runs concurrent threads on a uniprocessor doesn't need
1306 it.) Underneath PVM, you can have (for example) a network
1307 of workstations (slow) or a multiprocessor box
1310 <para>The current version of PVM is 3.3.11; we use 3.3.7.
1311 It is readily available on the net; I think I got it from
1312 <literal>research.att.com</literal>, in
1313 <filename>netlib</filename>.</para>
1315 <para>A PVM installation is slightly quirky, but easy to
1316 do. Just follow the <filename>Readme</filename>
1317 instructions.</para>
1322 <term><command>bash</command>:</term>
1323 <indexterm><primary>bash, presupposed (Parallel Haskell only)</primary></indexterm>
1325 <para>Sadly, the <command>gr2ps</command> script, used to
1326 convert “parallelism profiles” to PostScript,
1327 is written in Bash (GNU's Bourne Again shell). This bug
1328 will be fixed (someday).</para>
1334 <sect2 id="pre-supposed-other-tools">
1335 <title>Other useful tools</title>
1340 <indexterm><primary>pre-supposed: flex</primary></indexterm>
1341 <indexterm><primary>flex, pre-supposed</primary></indexterm>
1343 <para>This is a quite-a-bit-better-than-Lex lexer. Used
1344 to build a couple of utilities in
1345 <literal>glafp-utils</literal>. Depending on your
1346 operating system, the supplied <command>lex</command> may
1347 or may not work; you should get the GNU version.</para>
1352 <para>More tools are required if you want to format the documentation
1353 that comes with GHC and other fptools projects. See <xref
1354 linkend="building-docs">.</para>
1358 <sect1 id="sec-building-from-source">
1359 <title>Building from source</title>
1361 <indexterm><primary>Building from source</primary></indexterm>
1362 <indexterm><primary>Source, building from</primary></indexterm>
1364 <para>You've been rash enough to want to build some of the Glasgow
1365 Functional Programming tools (GHC, Happy, nofib, etc.) from
1366 source. You've slurped the source, from the CVS repository or
1367 from a source distribution, and now you're sitting looking at a
1368 huge mound of bits, wondering what to do next.</para>
1370 <para>Gingerly, you type <command>make</command>. Wrong
1373 <para>This rest of this guide is intended for duffers like me, who
1374 aren't really interested in Makefiles and systems configurations,
1375 but who need a mental model of the interlocking pieces so that
1376 they can make them work, extend them consistently when adding new
1377 software, and lay hands on them gently when they don't
1380 <sect2 id="quick-start">
1381 <title>Quick Start</title>
1383 <para>If you are starting from a source distribution, and just
1384 want a completely standard build, then the following should
1387 <screen>$ ./configure
1392 <para>For GHC, this will do a 2-stage bootstrap build of the
1393 compiler, with profiling libraries, and install the
1396 <para>If you want to do anything at all non-standard, or you
1397 want to do some development, read on...</para>
1400 <sect2 id="sec-source-tree">
1401 <title>Your source tree</title>
1403 <para>The source code is held in your <emphasis>source
1404 tree</emphasis>. The root directory of your source tree
1405 <emphasis>must</emphasis> contain the following directories and
1410 <para><filename>Makefile</filename>: the root
1415 <para><filename>mk/</filename>: the directory that contains
1416 the main Makefile code, shared by all the
1417 <literal>fptools</literal> software.</para>
1421 <para><filename>configure.in</filename>,
1422 <filename>config.sub</filename>,
1423 <filename>config.guess</filename>: these files support the
1424 configuration process.</para>
1428 <para><filename>install-sh</filename>.</para>
1432 <para>All the other directories are individual
1433 <emphasis>projects</emphasis> of the <literal>fptools</literal>
1434 system—for example, the Glasgow Haskell Compiler
1435 (<literal>ghc</literal>), the Happy parser generator
1436 (<literal>happy</literal>), the <literal>nofib</literal>
1437 benchmark suite, and so on. You can have zero or more of these.
1438 Needless to say, some of them are needed to build others.</para>
1440 <para>The important thing to remember is that even if you want
1441 only one project (<literal>happy</literal>, say), you must have
1442 a source tree whose root directory contains
1443 <filename>Makefile</filename>, <filename>mk/</filename>,
1444 <filename>configure.in</filename>, and the project(s) you want
1445 (<filename>happy/</filename> in this case). You cannot get by
1446 with just the <filename>happy/</filename> directory.</para>
1450 <title>Build trees</title>
1451 <indexterm><primary>build trees</primary></indexterm>
1452 <indexterm><primary>link trees, for building</primary></indexterm>
1454 <para>If you just want to build the software once on a single
1455 platform, then your source tree can also be your build tree, and
1456 you can skip the rest of this section.</para>
1458 <para>We often want to build multiple versions of our software
1459 for different architectures, or with different options
1460 (e.g. profiling). It's very desirable to share a single copy of
1461 the source code among all these builds.</para>
1463 <para>So for every source tree we have zero or more
1464 <emphasis>build trees</emphasis>. Each build tree is initially
1465 an exact copy of the source tree, except that each file is a
1466 symbolic link to the source file, rather than being a copy of
1467 the source file. There are “standard” Unix
1468 utilities that make such copies, so standard that they go by
1470 <command>lndir</command><indexterm><primary>lndir</primary></indexterm>,
1471 <command>mkshadowdir</command><indexterm><primary>mkshadowdir</primary></indexterm>
1472 are two (If you don't have either, the source distribution
1473 includes sources for the X11
1474 <command>lndir</command>—check out
1475 <filename>fptools/glafp-utils/lndir</filename>). See <Xref
1476 LinkEnd="sec-storysofar"> for a typical invocation.</para>
1478 <para>The build tree does not need to be anywhere near the
1479 source tree in the file system. Indeed, one advantage of
1480 separating the build tree from the source is that the build tree
1481 can be placed in a non-backed-up partition, saving your systems
1482 support people from backing up untold megabytes of
1483 easily-regenerated, and rapidly-changing, gubbins. The golden
1484 rule is that (with a single exception—<XRef
1485 LinkEnd="sec-build-config">) <emphasis>absolutely everything in
1486 the build tree is either a symbolic link to the source tree, or
1487 else is mechanically generated</emphasis>. It should be
1488 perfectly OK for your build tree to vanish overnight; an hour or
1489 two compiling and you're on the road again.</para>
1491 <para>You need to be a bit careful, though, that any new files
1492 you create (if you do any development work) are in the source
1493 tree, not a build tree!</para>
1495 <para>Remember, that the source files in the build tree are
1496 <emphasis>symbolic links</emphasis> to the files in the source
1497 tree. (The build tree soon accumulates lots of built files like
1498 <filename>Foo.o</filename>, as well.) You can
1499 <emphasis>delete</emphasis> a source file from the build tree
1500 without affecting the source tree (though it's an odd thing to
1501 do). On the other hand, if you <emphasis>edit</emphasis> a
1502 source file from the build tree, you'll edit the source-tree
1503 file directly. (You can set up Emacs so that if you edit a
1504 source file from the build tree, Emacs will silently create an
1505 edited copy of the source file in the build tree, leaving the
1506 source file unchanged; but the danger is that you think you've
1507 edited the source file whereas actually all you've done is edit
1508 the build-tree copy. More commonly you do want to edit the
1509 source file.)</para>
1511 <para>Like the source tree, the top level of your build tree
1512 must be (a linked copy of) the root directory of the
1513 <literal>fptools</literal> suite. Inside Makefiles, the root of
1514 your build tree is called
1515 <constant>$(FPTOOLS_TOP)</constant><indexterm><primary>FPTOOLS_TOP</primary></indexterm>.
1516 In the rest of this document path names are relative to
1517 <constant>$(FPTOOLS_TOP)</constant> unless
1518 otherwise stated. For example, the file
1519 <filename>ghc/mk/target.mk</filename> is actually
1520 <filename><constant>$(FPTOOLS_TOP)</constant>/ghc/mk/target.mk</filename>.</para>
1523 <sect2 id="sec-build-config">
1524 <title>Getting the build you want</title>
1526 <para>When you build <literal>fptools</literal> you will be
1527 compiling code on a particular <emphasis>host
1528 platform</emphasis>, to run on a particular <emphasis>target
1529 platform</emphasis> (usually the same as the host
1530 platform)<indexterm><primary>platform</primary></indexterm>.
1531 The difficulty is that there are minor differences between
1532 different platforms; minor, but enough that the code needs to be
1533 a bit different for each. There are some big differences too:
1534 for a different architecture we need to build GHC with a
1535 different native-code generator.</para>
1537 <para>There are also knobs you can turn to control how the
1538 <literal>fptools</literal> software is built. For example, you
1539 might want to build GHC optimised (so that it runs fast) or
1540 unoptimised (so that you can compile it fast after you've
1541 modified it. Or, you might want to compile it with debugging on
1542 (so that extra consistency-checking code gets included) or off.
1545 <para>All of this stuff is called the
1546 <emphasis>configuration</emphasis> of your build. You set the
1547 configuration using a three-step process.</para>
1551 <term>Step 1: get ready for configuration.</term>
1553 <para>NOTE: if you're starting from a source distribution,
1554 rather than CVS sources, you can skip this step.</para>
1556 <para>Change directory to
1557 <constant>$(FPTOOLS_TOP)</constant> and
1559 <command>autoconf</command><indexterm><primary>autoconf</primary></indexterm>
1560 (with no arguments). This GNU program converts
1561 <filename><constant>$(FPTOOLS_TOP)</constant>/configure.in</filename>
1562 to a shell script called
1563 <filename><constant>$(FPTOOLS_TOP)</constant>/configure</filename>.
1566 <para>Some projects, including GHC, have their own
1567 configure script. If there's an
1568 <constant>$(FPTOOLS_TOP)/<project>/configure.in</constant>,
1569 then you need to run <command>autoconf</command> in that
1570 directory too.</para>
1572 <para>Both these steps are completely
1573 platform-independent; they just mean that the
1574 human-written file (<filename>configure.in</filename>) can
1575 be short, although the resulting shell script,
1576 <command>configure</command>, and
1577 <filename>mk/config.h.in</filename>, are long.</para>
1582 <term>Step 2: system configuration.</term>
1584 <para>Runs the newly-created <command>configure</command>
1585 script, thus:</para>
1588 ./configure <optional><parameter>args</parameter></optional>
1591 <para><command>configure</command>'s mission is to scurry
1592 round your computer working out what architecture it has,
1593 what operating system, whether it has the
1594 <Function>vfork</Function> system call, where
1595 <command>yacc</command> is kept, whether
1596 <command>gcc</command> is available, where various obscure
1597 <literal>#include</literal> files are, whether it's a
1598 leap year, and what the systems manager had for lunch. It
1599 communicates these snippets of information in two
1606 <filename>mk/config.mk.in</filename><indexterm><primary>config.mk.in</primary></indexterm>
1608 <filename>mk/config.mk</filename><indexterm><primary>config.mk</primary></indexterm>,
1609 substituting for things between
1610 “<literal>@</literal>” brackets. So,
1611 “<literal>@HaveGcc@</literal>” will be
1612 replaced by “<literal>YES</literal>” or
1613 “<literal>NO</literal>” depending on what
1614 <command>configure</command> finds.
1615 <filename>mk/config.mk</filename> is included by every
1616 Makefile (directly or indirectly), so the
1617 configuration information is thereby communicated to
1618 all Makefiles.</para>
1622 <para> It translates
1623 <filename>mk/config.h.in</filename><indexterm><primary>config.h.in</primary></indexterm>
1625 <filename>mk/config.h</filename><indexterm><primary>config.h</primary></indexterm>.
1626 The latter is <literal>#include</literal>d by
1627 various C programs, which can thereby make use of
1628 configuration information.</para>
1632 <para><command>configure</command> takes some optional
1633 arguments. Use <literal>./configure --help</literal> to
1634 get a list of the available arguments. Here are some of
1635 the ones you might need:</para>
1639 <term><literal>--with-ghc=<parameter>path</parameter></literal></term>
1640 <indexterm><primary><literal>--with-ghc</literal></primary>
1643 <para>Specifies the path to an installed GHC which
1644 you would like to use. This compiler will be used
1645 for compiling GHC-specific code (eg. GHC itself).
1646 This option <emphasis>cannot</emphasis> be specified
1647 using <filename>build.mk</filename> (see later),
1648 because <command>configure</command> needs to
1649 auto-detect the version of GHC you're using. The
1650 default is to look for a compiler named
1651 <literal>ghc</literal> in your path.</para>
1656 <term><literal>--with-hc=<parameter>path</parameter></literal></term>
1657 <indexterm><primary><literal>--with-hc</literal></primary>
1660 <para>Specifies the path to any installed Haskell
1661 compiler. This compiler will be used for compiling
1662 generic Haskell code. The default is to use
1663 <literal>ghc</literal>.</para>
1668 <term><literal>--with-gcc=<parameter>path</parameter></literal></term>
1669 <indexterm><primary><literal>--with-gcc</literal></primary>
1672 <para>Specifies the path to the installed GCC. This
1673 compiler will be used to compile all C files,
1674 <emphasis>except</emphasis> any generated by the
1675 installed Haskell compiler, which will have its own
1676 idea of which C compiler (if any) to use. The
1677 default is to use <literal>gcc</literal>.</para>
1682 <para><command>configure</command> caches the results of
1683 its run in <filename>config.cache</filename>. Quite often
1684 you don't want that; you're running
1685 <command>configure</command> a second time because
1686 something has changed. In that case, simply delete
1687 <filename>config.cache</filename>.</para>
1692 <term>Step 3: build configuration.</term>
1694 <para>Next, you say how this build of
1695 <literal>fptools</literal> is to differ from the standard
1696 defaults by creating a new file
1697 <filename>mk/build.mk</filename><indexterm><primary>build.mk</primary></indexterm>
1698 <emphasis>in the build tree</emphasis>. This file is the
1699 one and only file you edit in the build tree, precisely
1700 because it says how this build differs from the source.
1701 (Just in case your build tree does die, you might want to
1702 keep a private directory of <filename>build.mk</filename>
1703 files, and use a symbolic link in each build tree to point
1704 to the appropriate one.) So
1705 <filename>mk/build.mk</filename> never exists in the
1706 source tree—you create one in each build tree from
1707 the template. We'll discuss what to put in it
1713 <para>And that's it for configuration. Simple, eh?</para>
1715 <para>What do you put in your build-specific configuration file
1716 <filename>mk/build.mk</filename>? <emphasis>For almost all
1717 purposes all you will do is put make variable definitions that
1718 override those in</emphasis>
1719 <filename>mk/config.mk.in</filename>. The whole point of
1720 <filename>mk/config.mk.in</filename>—and its derived
1721 counterpart <filename>mk/config.mk</filename>—is to define
1722 the build configuration. It is heavily commented, as you will
1723 see if you look at it. So generally, what you do is look at
1724 <filename>mk/config.mk.in</filename>, and add definitions in
1725 <filename>mk/build.mk</filename> that override any of the
1726 <filename>config.mk</filename> definitions that you want to
1727 change. (The override occurs because the main boilerplate file,
1728 <filename>mk/boilerplate.mk</filename><indexterm><primary>boilerplate.mk</primary></indexterm>,
1729 includes <filename>build.mk</filename> after
1730 <filename>config.mk</filename>.)</para>
1732 <para>For your convenience, there's a file called <filename>build.mk.sample</filename>
1733 that can serve as a starting point for your <filename>build.mk</filename>.</para>
1735 <para>For example, <filename>config.mk.in</filename> contains
1736 the definition:</para>
1739 GhcHcOpts=-O -Rghc-timing
1742 <para>The accompanying comment explains that this is the list of
1743 flags passed to GHC when building GHC itself. For doing
1744 development, it is wise to add <literal>-DDEBUG</literal>, to
1745 enable debugging code. So you would add the following to
1746 <filename>build.mk</filename>:</para>
1748 <para>or, if you prefer,</para>
1751 GhcHcOpts += -DDEBUG
1754 <para>GNU <command>make</command> allows existing definitions to
1755 have new text appended using the “<literal>+=</literal>”
1756 operator, which is quite a convenient feature.)</para>
1758 <para>If you want to remove the <literal>-O</literal> as well (a
1759 good idea when developing, because the turn-around cycle gets a
1760 lot quicker), you can just override
1761 <literal>GhcLibHcOpts</literal> altogether:</para>
1764 GhcHcOpts=-DDEBUG -Rghc-timing
1767 <para>When reading <filename>config.mk.in</filename>, remember
1768 that anything between “@...@” signs is going to be substituted
1769 by <command>configure</command> later. You
1770 <emphasis>can</emphasis> override the resulting definition if
1771 you want, but you need to be a bit surer what you are doing.
1772 For example, there's a line that says:</para>
1778 <para>This defines the Make variables <constant>YACC</constant>
1779 to the pathname for a <command>yacc</command> that
1780 <command>configure</command> finds somewhere. If you have your
1781 own pet <command>yacc</command> you want to use instead, that's
1782 fine. Just add this line to <filename>mk/build.mk</filename>:</para>
1788 <para>You do not <emphasis>have</emphasis> to have a
1789 <filename>mk/build.mk</filename> file at all; if you don't,
1790 you'll get all the default settings from
1791 <filename>mk/config.mk.in</filename>.</para>
1793 <para>You can also use <filename>build.mk</filename> to override
1794 anything that <command>configure</command> got wrong. One place
1795 where this happens often is with the definition of
1796 <constant>FPTOOLS_TOP_ABS</constant>: this
1797 variable is supposed to be the canonical path to the top of your
1798 source tree, but if your system uses an automounter then the
1799 correct directory is hard to find automatically. If you find
1800 that <command>configure</command> has got it wrong, just put the
1801 correct definition in <filename>build.mk</filename>.</para>
1805 <sect2 id="sec-storysofar">
1806 <title>The story so far</title>
1808 <para>Let's summarise the steps you need to carry to get
1809 yourself a fully-configured build tree from scratch.</para>
1813 <para> Get your source tree from somewhere (CVS repository
1814 or source distribution). Say you call the root directory
1815 <filename>myfptools</filename> (it does not have to be
1816 called <filename>fptools</filename>). Make sure that you
1817 have the essential files (see <XRef
1818 LinkEnd="sec-source-tree">).</para>
1823 <para>(Optional) Use <command>lndir</command> or
1824 <command>mkshadowdir</command> to create a build tree.</para>
1828 $ mkshadowdir . /scratch/joe-bloggs/myfptools-sun4
1831 <para>(N.B. <command>mkshadowdir</command>'s first argument
1832 is taken relative to its second.) You probably want to give
1833 the build tree a name that suggests its main defining
1834 characteristic (in your mind at least), in case you later
1839 <para>Change directory to the build tree. Everything is
1840 going to happen there now.</para>
1843 $ cd /scratch/joe-bloggs/myfptools-sun4
1849 <para>Prepare for system configuration:</para>
1855 <para>(You can skip this step if you are starting from a
1856 source distribution, and you already have
1857 <filename>configure</filename> and
1858 <filename>mk/config.h.in</filename>.)</para>
1860 <para>Some projects, including GHC itself, have their own
1861 configure scripts, so it is necessary to run autoconf again
1862 in the appropriate subdirectories. eg:</para>
1865 $ (cd ghc; autoconf)
1870 <para>Do system configuration:</para>
1876 <para>Don't forget to check whether you need to add any
1877 arguments to <literal>configure</literal>; for example, a
1878 common requirement is to specify which GHC to use with
1879 <option>--with-ghc=<replaceable>ghc</replaceable></option>.</para>
1883 <para>Create the file <filename>mk/build.mk</filename>,
1884 adding definitions for your desired configuration
1893 <para>You can make subsequent changes to
1894 <filename>mk/build.mk</filename> as often as you like. You do
1895 not have to run any further configuration programs to make these
1896 changes take effect. In theory you should, however, say
1897 <command>gmake clean</command>, <command>gmake all</command>,
1898 because configuration option changes could affect
1899 anything—but in practice you are likely to know what's
1904 <title>Making things</title>
1906 <para>At this point you have made yourself a fully-configured
1907 build tree, so you are ready to start building real
1910 <para>The first thing you need to know is that <emphasis>you
1911 must use GNU <command>make</command>, usually called
1912 <command>gmake</command>, not standard Unix
1913 <command>make</command></emphasis>. If you use standard Unix
1914 <command>make</command> you will get all sorts of error messages
1915 (but no damage) because the <literal>fptools</literal>
1916 <command>Makefiles</command> use GNU <command>make</command>'s
1917 facilities extensively.</para>
1919 <para>To just build the whole thing, <command>cd</command> to
1920 the top of your <literal>fptools</literal> tree and type
1921 <command>gmake</command>. This will prepare the tree and build
1922 the various projects in the correct order.</para>
1925 <sect2 id="sec-bootstrapping">
1926 <title>Bootstrapping GHC</title>
1928 <para>GHC requires a 2-stage bootstrap in order to provide
1929 full functionality, including GHCi. By a 2-stage bootstrap, we
1930 mean that the compiler is built once using the installed GHC,
1931 and then again using the compiler built in the first stage. You
1932 can also build a stage 3 compiler, but this normally isn't
1933 necessary except to verify that the stage 2 compiler is working
1936 <para>Note that when doing a bootstrap, the stage 1 compiler
1937 must be built, followed by the runtime system and libraries, and
1938 then the stage 2 compiler. The correct ordering is implemented
1939 by the top-level fptools <filename>Makefile</filename>, so if
1940 you want everything to work automatically it's best to start
1941 <command>make</command> from the top of the tree. When building
1942 GHC, the top-level fptools <filename>Makefile</filename> is set
1943 up to do a 2-stage bootstrap by default (when you say
1944 <command>make</command>). Some other targets it supports
1951 <para>Build everything as normal, including the stage 1
1959 <para>Build the stage 2 compiler only.</para>
1966 <para>Build the stage 3 compiler only.</para>
1971 <term>bootstrap</term> <term>bootstrap2</term>
1973 <para>Build stage 1 followed by stage 2.</para>
1978 <term>bootstrap3</term>
1980 <para>Build stages 1, 2 and 3.</para>
1985 <term>install</term>
1987 <para>Install everything, including the compiler built in
1988 stage 2. To override the stage, say <literal>make install
1989 stage=<replaceable>n</replaceable></literal> where
1990 <replaceable>n</replaceable> is the stage to install.</para>
1995 <para>The top-level <filename>Makefile</filename> also arranges
1996 to do the appropriate <literal>make boot</literal> steps (see
1997 below) before actually building anything.</para>
1999 <para>The <literal>stage1</literal>, <literal>stage2</literal>
2000 and <literal>stage3</literal> targets also work in the
2001 <literal>ghc/compiler</literal> directory, but don't forget that
2002 each stage requires its own <literal>make boot</literal> step:
2003 for example, you must do</para>
2005 <screen>$ make boot stage=2</screen>
2007 <para>before <literal>make stage2</literal> in
2008 <literal>ghc/compiler</literal>.</para>
2011 <sect2 id="sec-standard-targets">
2012 <title>Standard Targets</title>
2013 <indexterm><primary>targets, standard makefile</primary></indexterm>
2014 <indexterm><primary>makefile targets</primary></indexterm>
2016 <para>In any directory you should be able to make the following:</para>
2020 <term><literal>boot</literal></term>
2022 <para>does the one-off preparation required to get ready
2023 for the real work. Notably, it does <command>gmake
2024 depend</command> in all directories that contain programs.
2025 It also builds the necessary tools for compilation to
2028 <para>Invoking the <literal>boot</literal> target
2029 explicitly is not normally necessary. From the top-level
2030 <literal>fptools</literal> directory, invoking
2031 <literal>gmake</literal> causes <literal>gmake boot
2032 all</literal> to be invoked in each of the project
2033 subdirectories, in the order specified by
2034 <literal>$(AllTargets)</literal> in
2035 <literal>config.mk</literal>.</para>
2037 <para>If you're working in a subdirectory somewhere and
2038 need to update the dependencies, <literal>gmake
2039 boot</literal> is a good way to do it.</para>
2044 <term><literal>all</literal></term>
2046 <para>makes all the final target(s) for this Makefile.
2047 Depending on which directory you are in a “final
2048 target” may be an executable program, a library
2049 archive, a shell script, or a Postscript file. Typing
2050 <command>gmake</command> alone is generally the same as
2051 typing <command>gmake all</command>.</para>
2056 <term><literal>install</literal></term>
2058 <para>installs the things built by <literal>all</literal>
2059 (except for the documentation). Where does it install
2060 them? That is specified by
2061 <filename>mk/config.mk.in</filename>; you can override it
2062 in <filename>mk/build.mk</filename>, or by running
2063 <command>configure</command> with command-line arguments
2064 like <literal>--bindir=/home/simonpj/bin</literal>; see
2065 <literal>./configure --help</literal> for the full
2071 <term><literal>install-docs</literal></term>
2073 <para>installs the documentation. Otherwise behaves just
2074 like <literal>install</literal>.</para>
2079 <term><literal>uninstall</literal></term>
2081 <para>reverses the effect of
2082 <literal>install</literal>.</para>
2087 <term><literal>clean</literal></term>
2089 <para>Delete all files from the current directory that are
2090 normally created by building the program. Don't delete
2091 the files that record the configuration, or files
2092 generated by <command>gmake boot</command>. Also preserve
2093 files that could be made by building, but normally aren't
2094 because the distribution comes with them.</para>
2099 <term><literal>distclean</literal></term>
2101 <para>Delete all files from the current directory that are
2102 created by configuring or building the program. If you
2103 have unpacked the source and built the program without
2104 creating any other files, <literal>make
2105 distclean</literal> should leave only the files that were
2106 in the distribution.</para>
2111 <term><literal>mostlyclean</literal></term>
2113 <para>Like <literal>clean</literal>, but may refrain from
2114 deleting a few files that people normally don't want to
2120 <term><literal>maintainer-clean</literal></term>
2122 <para>Delete everything from the current directory that
2123 can be reconstructed with this Makefile. This typically
2124 includes everything deleted by
2125 <literal>distclean</literal>, plus more: C source files
2126 produced by Bison, tags tables, Info files, and so
2129 <para>One exception, however: <literal>make
2130 maintainer-clean</literal> should not delete
2131 <filename>configure</filename> even if
2132 <filename>configure</filename> can be remade using a rule
2133 in the <filename>Makefile</filename>. More generally,
2134 <literal>make maintainer-clean</literal> should not delete
2135 anything that needs to exist in order to run
2136 <filename>configure</filename> and then begin to build the
2142 <term><literal>check</literal></term>
2144 <para>run the test suite.</para>
2149 <para>All of these standard targets automatically recurse into
2150 sub-directories. Certain other standard targets do not:</para>
2154 <term><literal>configure</literal></term>
2156 <para>is only available in the root directory
2157 <constant>$(FPTOOLS_TOP)</constant>; it has
2158 been discussed in <XRef
2159 LinkEnd="sec-build-config">.</para>
2164 <term><literal>depend</literal></term>
2166 <para>make a <filename>.depend</filename> file in each
2167 directory that needs it. This <filename>.depend</filename>
2168 file contains mechanically-generated dependency
2169 information; for example, suppose a directory contains a
2170 Haskell source module <filename>Foo.lhs</filename> which
2171 imports another module <literal>Baz</literal>. Then the
2172 generated <filename>.depend</filename> file will contain
2173 the dependency:</para>
2179 <para>which says that the object file
2180 <filename>Foo.o</filename> depends on the interface file
2181 <filename>Baz.hi</filename> generated by compiling module
2182 <literal>Baz</literal>. The <filename>.depend</filename>
2183 file is automatically included by every Makefile.</para>
2188 <term><literal>binary-dist</literal></term>
2190 <para>make a binary distribution. This is the target we
2191 use to build the binary distributions of GHC and
2197 <term><literal>dist</literal></term>
2199 <para>make a source distribution. Note that this target
2200 does “make distclean” as part of its work;
2201 don't use it if you want to keep what you've built.</para>
2206 <para>Most <filename>Makefile</filename>s have targets other
2207 than these. You can discover them by looking in the
2208 <filename>Makefile</filename> itself.</para>
2212 <title>Using a project from the build tree</title>
2214 <para>If you want to build GHC (say) and just use it direct from
2215 the build tree without doing <literal>make install</literal>
2216 first, you can run the in-place driver script:
2217 <filename>ghc/compiler/ghc-inplace</filename>.</para>
2219 <para> Do <emphasis>NOT</emphasis> use
2220 <filename>ghc/compiler/ghc</filename>, or
2221 <filename>ghc/compiler/ghc-6.xx</filename>, as these are the
2222 scripts intended for installation, and contain hard-wired paths
2223 to the installed libraries, rather than the libraries in the
2226 <para>Happy can similarly be run from the build tree, using
2227 <filename>happy/src/happy-inplace</filename>.</para>
2231 <title>Fast Making</title>
2233 <indexterm><primary>fastmake</primary></indexterm>
2234 <indexterm><primary>dependencies, omitting</primary></indexterm>
2235 <indexterm><primary>FAST, makefile variable</primary></indexterm>
2237 <para>Sometimes the dependencies get in the way: if you've made
2238 a small change to one file, and you're absolutely sure that it
2239 won't affect anything else, but you know that
2240 <command>make</command> is going to rebuild everything anyway,
2241 the following hack may be useful:</para>
2247 <para>This tells the make system to ignore dependencies and just
2248 build what you tell it to. In other words, it's equivalent to
2249 temporarily removing the <filename>.depend</filename> file in
2250 the current directory (where <command>mkdependHS</command> and
2251 friends store their dependency information).</para>
2253 <para>A bit of history: GHC used to come with a
2254 <command>fastmake</command> script that did the above job, but
2255 GNU make provides the features we need to do it without
2256 resorting to a script. Also, we've found that fastmaking is
2257 less useful since the advent of GHC's recompilation checker (see
2258 the User's Guide section on "Separate Compilation").</para>
2262 <sect1 id="sec-makefile-arch">
2263 <title>The <filename>Makefile</filename> architecture</title>
2264 <indexterm><primary>makefile architecture</primary></indexterm>
2266 <para><command>make</command> is great if everything
2267 works—you type <command>gmake install</command> and lo! the
2268 right things get compiled and installed in the right places. Our
2269 goal is to make this happen often, but somehow it often doesn't;
2270 instead some weird error message eventually emerges from the
2271 bowels of a directory you didn't know existed.</para>
2273 <para>The purpose of this section is to give you a road-map to
2274 help you figure out what is going right and what is going
2278 <title>Debugging</title>
2280 <para>Debugging <filename>Makefile</filename>s is something of a
2281 black art, but here's a couple of tricks that we find
2282 particularly useful. The following command allows you to see
2283 the contents of any make variable in the context of the current
2284 <filename>Makefile</filename>:</para>
2286 <screen>$ make show VALUE=HS_SRCS</screen>
2288 <para>where you can replace <literal>HS_SRCS</literal> with the
2289 name of any variable you wish to see the value of.</para>
2291 <para>GNU make has a <option>-d</option> option which generates
2292 a dump of the decision procedure used to arrive at a conclusion
2293 about which files should be recompiled. Sometimes useful for
2294 tracking down problems with superfluous or missing
2295 recompilations.</para>
2299 <title>A small project</title>
2301 <para>To get started, let us look at the
2302 <filename>Makefile</filename> for an imaginary small
2303 <literal>fptools</literal> project, <literal>small</literal>.
2304 Each project in <literal>fptools</literal> has its own directory
2305 in <constant>FPTOOLS_TOP</constant>, so the
2306 <literal>small</literal> project will have its own directory
2307 <constant>FPOOLS_TOP/small/</constant>. Inside the
2308 <filename>small/</filename> directory there will be a
2309 <filename>Makefile</filename>, looking something like
2312 <indexterm><primary>Makefile, minimal</primary></indexterm>
2315 # Makefile for fptools project "small"
2318 include $(TOP)/mk/boilerplate.mk
2320 SRCS = $(wildcard *.lhs) $(wildcard *.c)
2323 include $(TOP)/target.mk
2326 <para>this <filename>Makefile</filename> has three
2331 <para>The first section includes
2334 One of the most important
2335 features of GNU <command>make</command> that we use is the ability for a <filename>Makefile</filename> to
2336 include another named file, very like <command>cpp</command>'s <literal>#include</literal>
2341 a file of “boilerplate” code from the level
2342 above (which in this case will be
2343 <filename><constant>FPTOOLS_TOP</constant>/mk/boilerplate.mk</filename><indexterm><primary>boilerplate.mk</primary></indexterm>).
2344 As its name suggests, <filename>boilerplate.mk</filename>
2345 consists of a large quantity of standard
2346 <filename>Makefile</filename> code. We discuss this
2347 boilerplate in more detail in <XRef LinkEnd="sec-boiler">.
2348 <indexterm><primary>include, directive in
2349 Makefiles</primary></indexterm> <indexterm><primary>Makefile
2350 inclusion</primary></indexterm></para>
2352 <para>Before the <literal>include</literal> statement, you
2353 must define the <command>make</command> variable
2354 <constant>TOP</constant><indexterm><primary>TOP</primary></indexterm>
2355 to be the directory containing the <filename>mk</filename>
2356 directory in which the <filename>boilerplate.mk</filename>
2357 file is. It is <emphasis>not</emphasis> OK to simply say</para>
2360 include ../mk/boilerplate.mk # NO NO NO
2364 <para>Why? Because the <filename>boilerplate.mk</filename>
2365 file needs to know where it is, so that it can, in turn,
2366 <literal>include</literal> other files. (Unfortunately,
2367 when an <literal>include</literal>d file does an
2368 <literal>include</literal>, the filename is treated relative
2369 to the directory in which <command>gmake</command> is being
2370 run, not the directory in which the
2371 <literal>include</literal>d sits.) In general,
2372 <emphasis>every file <filename>foo.mk</filename> assumes
2374 <filename><constant>$(TOP)</constant>/mk/foo.mk</filename>
2375 refers to itself.</emphasis> It is up to the
2376 <filename>Makefile</filename> doing the
2377 <literal>include</literal> to ensure this is the case.</para>
2379 <para>Files intended for inclusion in other
2380 <filename>Makefile</filename>s are written to have the
2381 following property: <emphasis>after
2382 <filename>foo.mk</filename> is <literal>include</literal>d,
2383 it leaves <constant>TOP</constant> containing the same value
2384 as it had just before the <literal>include</literal>
2385 statement</emphasis>. In our example, this invariant
2386 guarantees that the <literal>include</literal> for
2387 <filename>target.mk</filename> will look in the same
2388 directory as that for <filename>boilerplate.mk</filename>.</para>
2392 <para> The second section defines the following standard
2393 <command>make</command> variables:
2394 <constant>SRCS</constant><indexterm><primary>SRCS</primary></indexterm>
2395 (the source files from which is to be built), and
2396 <constant>HS_PROG</constant><indexterm><primary>HS_PROG</primary></indexterm>
2397 (the executable binary to be built). We will discuss in
2398 more detail what the “standard variables” are,
2399 and how they affect what happens, in <XRef
2400 LinkEnd="sec-targets">.</para>
2402 <para>The definition for <constant>SRCS</constant> uses the
2403 useful GNU <command>make</command> construct
2404 <literal>$(wildcard $pat$)</literal><indexterm><primary>wildcard</primary></indexterm>,
2405 which expands to a list of all the files matching the
2406 pattern <literal>pat</literal> in the current directory. In
2407 this example, <constant>SRCS</constant> is set to the list
2408 of all the <filename>.lhs</filename> and
2409 <filename>.c</filename> files in the directory. (Let's
2410 suppose there is one of each, <filename>Foo.lhs</filename>
2411 and <filename>Baz.c</filename>.)</para>
2415 <para>The last section includes a second file of standard
2417 <filename>target.mk</filename><indexterm><primary>target.mk</primary></indexterm>.
2418 It contains the rules that tell <command>gmake</command> how
2419 to make the standard targets (<Xref
2420 LinkEnd="sec-standard-targets">). Why, you ask, can't this
2421 standard code be part of
2422 <filename>boilerplate.mk</filename>? Good question. We
2423 discuss the reason later, in <Xref
2424 LinkEnd="sec-boiler-arch">.</para>
2426 <para>You do not <emphasis>have</emphasis> to
2427 <literal>include</literal> the
2428 <filename>target.mk</filename> file. Instead, you can write
2429 rules of your own for all the standard targets. Usually,
2430 though, you will find quite a big payoff from using the
2431 canned rules in <filename>target.mk</filename>; the price
2432 tag is that you have to understand what canned rules get
2433 enabled, and what they do (<Xref
2434 LinkEnd="sec-targets">).</para>
2438 <para>In our example <filename>Makefile</filename>, most of the
2439 work is done by the two <literal>include</literal>d files. When
2440 you say <command>gmake all</command>, the following things
2445 <para><command>gmake</command> figures out that the object
2446 files are <filename>Foo.o</filename> and
2447 <filename>Baz.o</filename>.</para>
2451 <para>It uses a boilerplate pattern rule to compile
2452 <filename>Foo.lhs</filename> to <filename>Foo.o</filename>
2453 using a Haskell compiler. (Which one? That is set in the
2454 build configuration.)</para>
2458 <para>It uses another standard pattern rule to compile
2459 <filename>Baz.c</filename> to <filename>Baz.o</filename>,
2460 using a C compiler. (Ditto.)</para>
2464 <para>It links the resulting <filename>.o</filename> files
2465 together to make <literal>small</literal>, using the Haskell
2466 compiler to do the link step. (Why not use
2467 <command>ld</command>? Because the Haskell compiler knows
2468 what standard libraries to link in. How did
2469 <command>gmake</command> know to use the Haskell compiler to
2470 do the link, rather than the C compiler? Because we set the
2471 variable <constant>HS_PROG</constant> rather than
2472 <constant>C_PROG</constant>.)</para>
2476 <para>All <filename>Makefile</filename>s should follow the above
2477 three-section format.</para>
2481 <title>A larger project</title>
2483 <para>Larger projects are usually structured into a number of
2484 sub-directories, each of which has its own
2485 <filename>Makefile</filename>. (In very large projects, this
2486 sub-structure might be iterated recursively, though that is
2487 rare.) To give you the idea, here's part of the directory
2488 structure for the (rather large) GHC project:</para>
2498 ...source files for documentation...
2501 ...source files for driver...
2504 parser/...source files for parser...
2505 renamer/...source files for renamer...
2509 <para>The sub-directories <filename>docs</filename>,
2510 <filename>driver</filename>, <filename>compiler</filename>, and
2511 so on, each contains a sub-component of GHC, and each has its
2512 own <filename>Makefile</filename>. There must also be a
2513 <filename>Makefile</filename> in
2514 <filename><constant>$(FPTOOLS_TOP)</constant>/ghc</filename>.
2515 It does most of its work by recursively invoking
2516 <command>gmake</command> on the <filename>Makefile</filename>s
2517 in the sub-directories. We say that
2518 <filename>ghc/Makefile</filename> is a <emphasis>non-leaf
2519 <filename>Makefile</filename></emphasis>, because it does little
2520 except organise its children, while the
2521 <filename>Makefile</filename>s in the sub-directories are all
2522 <emphasis>leaf <filename>Makefile</filename>s</emphasis>. (In
2523 principle the sub-directories might themselves contain a
2524 non-leaf <filename>Makefile</filename> and several
2525 sub-sub-directories, but that does not happen in GHC.)</para>
2527 <para>The <filename>Makefile</filename> in
2528 <filename>ghc/compiler</filename> is considered a leaf
2529 <filename>Makefile</filename> even though the
2530 <filename>ghc/compiler</filename> has sub-directories, because
2531 these sub-directories do not themselves have
2532 <filename>Makefile</filename>s in them. They are just used to
2533 structure the collection of modules that make up GHC, but all
2534 are managed by the single <filename>Makefile</filename> in
2535 <filename>ghc/compiler</filename>.</para>
2537 <para>You will notice that <filename>ghc/</filename> also
2538 contains a directory <filename>ghc/mk/</filename>. It contains
2539 GHC-specific <filename>Makefile</filename> boilerplate code.
2540 More precisely:</para>
2544 <para><filename>ghc/mk/boilerplate.mk</filename> is included
2545 at the top of <filename>ghc/Makefile</filename>, and of all
2546 the leaf <filename>Makefile</filename>s in the
2547 sub-directories. It in turn <literal>include</literal>s the
2548 main boilerplate file
2549 <filename>mk/boilerplate.mk</filename>.</para>
2553 <para><filename>ghc/mk/target.mk</filename> is
2554 <literal>include</literal>d at the bottom of
2555 <filename>ghc/Makefile</filename>, and of all the leaf
2556 <filename>Makefile</filename>s in the sub-directories. It
2557 in turn <literal>include</literal>s the file
2558 <filename>mk/target.mk</filename>.</para>
2562 <para>So these two files are the place to look for GHC-wide
2563 customisation of the standard boilerplate.</para>
2566 <sect2 id="sec-boiler-arch">
2567 <title>Boilerplate architecture</title>
2568 <indexterm><primary>boilerplate architecture</primary></indexterm>
2570 <para>Every <filename>Makefile</filename> includes a
2571 <filename>boilerplate.mk</filename><indexterm><primary>boilerplate.mk</primary></indexterm>
2572 file at the top, and
2573 <filename>target.mk</filename><indexterm><primary>target.mk</primary></indexterm>
2574 file at the bottom. In this section we discuss what is in these
2575 files, and why there have to be two of them. In general:</para>
2579 <para><filename>boilerplate.mk</filename> consists of:</para>
2583 <para><emphasis>Definitions of millions of
2584 <command>make</command> variables</emphasis> that
2585 collectively specify the build configuration. Examples:
2586 <constant>HC_OPTS</constant><indexterm><primary>HC_OPTS</primary></indexterm>,
2587 the options to feed to the Haskell compiler;
2588 <constant>NoFibSubDirs</constant><indexterm><primary>NoFibSubDirs</primary></indexterm>,
2589 the sub-directories to enable within the
2590 <literal>nofib</literal> project;
2591 <constant>GhcWithHc</constant><indexterm><primary>GhcWithHc</primary></indexterm>,
2592 the name of the Haskell compiler to use when compiling
2593 GHC in the <literal>ghc</literal> project.</para>
2597 <para><emphasis>Standard pattern rules</emphasis> that
2598 tell <command>gmake</command> how to construct one file
2599 from another.</para>
2603 <para><filename>boilerplate.mk</filename> needs to be
2604 <literal>include</literal>d at the <emphasis>top</emphasis>
2605 of each <filename>Makefile</filename>, so that the user can
2606 replace the boilerplate definitions or pattern rules by
2607 simply giving a new definition or pattern rule in the
2608 <filename>Makefile</filename>. <command>gmake</command>
2609 simply takes the last definition as the definitive one.</para>
2611 <para>Instead of <emphasis>replacing</emphasis> boilerplate
2612 definitions, it is also quite common to
2613 <emphasis>augment</emphasis> them. For example, a
2614 <filename>Makefile</filename> might say:</para>
2620 <para>thereby adding “<option>-O</option>” to
2622 <constant>SRC_HC_OPTS</constant><indexterm><primary>SRC_HC_OPTS</primary></indexterm>.</para>
2626 <para><filename>target.mk</filename> contains
2627 <command>make</command> rules for the standard targets
2628 described in <Xref LinkEnd="sec-standard-targets">. These
2629 rules are selectively included, depending on the setting of
2630 certain <command>make</command> variables. These variables
2631 are usually set in the middle section of the
2632 <filename>Makefile</filename> between the two
2633 <literal>include</literal>s.</para>
2635 <para><filename>target.mk</filename> must be included at the
2636 end (rather than being part of
2637 <filename>boilerplate.mk</filename>) for several tiresome
2643 <para><command>gmake</command> commits target and
2644 dependency lists earlier than it should. For example,
2645 <FIlename>target.mk</FIlename> has a rule that looks
2649 $(HS_PROG) : $(OBJS)
2650 $(HC) $(LD_OPTS) $< -o $@
2653 <para>If this rule was in
2654 <filename>boilerplate.mk</filename> then
2655 <constant>$(HS_PROG)</constant><indexterm><primary>HS_PROG</primary></indexterm>
2657 <constant>$(OBJS)</constant><indexterm><primary>OBJS</primary></indexterm>
2658 would not have their final values at the moment
2659 <command>gmake</command> encountered the rule. Alas,
2660 <command>gmake</command> takes a snapshot of their
2661 current values, and wires that snapshot into the rule.
2662 (In contrast, the commands executed when the rule
2663 “fires” are only substituted at the moment
2664 of firing.) So, the rule must follow the definitions
2665 given in the <filename>Makefile</filename> itself.</para>
2669 <para>Unlike pattern rules, ordinary rules cannot be
2670 overriden or replaced by subsequent rules for the same
2671 target (at least, not without an error message).
2672 Including ordinary rules in
2673 <filename>boilerplate.mk</filename> would prevent the
2674 user from writing rules for specific targets in specific
2679 <para>There are a couple of other reasons I've
2680 forgotten, but it doesn't matter too much.</para>
2687 <sect2 id="sec-boiler">
2688 <title>The main <filename>mk/boilerplate.mk</filename> file</title>
2689 <indexterm><primary>boilerplate.mk</primary></indexterm>
2691 <para>If you look at
2692 <filename><constant>$(FPTOOLS_TOP)</constant>/mk/boilerplate.mk</filename>
2693 you will find that it consists of the following sections, each
2694 held in a separate file:</para>
2698 <term><filename>config.mk</filename></term>
2699 <indexterm><primary>config.mk</primary></indexterm>
2701 <para>is the build configuration file we discussed at
2702 length in <Xref LinkEnd="sec-build-config">.</para>
2707 <term><filename>paths.mk</filename></term>
2708 <indexterm><primary>paths.mk</primary></indexterm>
2710 <para>defines <command>make</command> variables for
2711 pathnames and file lists. This file contains code for
2712 automatically compiling lists of source files and deriving
2713 lists of object files from those. The results can be
2714 overriden in the <filename>Makefile</filename>, but in
2715 most cases the automatic setup should do the right
2718 <para>The following variables may be set in the
2719 <filename>Makefile</filename> to affect how the automatic
2720 source file search is done:</para>
2724 <term><literal>ALL_DIRS</literal></term>
2725 <indexterm><primary><literal>ALL_DIRS</literal></primary>
2728 <para>Set to a list of directories to search in
2729 addition to the current directory for source
2735 <term><literal>EXCLUDE_SRCS</literal></term>
2736 <indexterm><primary><literal>EXCLUDE_SRCS</literal></primary>
2739 <para>Set to a list of source files (relative to the
2740 current directory) to omit from the automatic
2741 search. The source searching machinery is clever
2742 enough to know that if you exclude a source file
2743 from which other sources are derived, then the
2744 derived sources should also be excluded. For
2745 example, if you set <literal>EXCLUDED_SRCS</literal>
2746 to include <filename>Foo.y</filename>, then
2747 <filename>Foo.hs</filename> will also be
2753 <term><literal>EXTRA_SRCS</literal></term>
2754 <indexterm><primary><literal>EXCLUDE_SRCS</literal></primary>
2757 <para>Set to a list of extra source files (perhaps
2758 in directories not listed in
2759 <literal>ALL_DIRS</literal>) that should be
2765 <para>The results of the automatic source file search are
2766 placed in the following make variables:</para>
2770 <term><literal>SRCS</literal></term>
2771 <indexterm><primary><literal>SRCS</literal></primary></indexterm>
2773 <para>All source files found, sorted and without
2774 duplicates, including those which might not exist
2775 yet but will be derived from other existing sources.
2776 <literal>SRCS</literal> <emphasis>can</emphasis> be
2777 overriden if necessary, in which case the variables
2778 below will follow suit.</para>
2783 <term><literal>HS_SRCS</literal></term>
2784 <indexterm><primary><literal>HS_SRCS</literal></primary></indexterm>
2786 <para>all Haskell source files in the current
2787 directory, including those derived from other source
2788 files (eg. Happy sources also give rise to Haskell
2794 <term><literal>HS_OBJS</literal></term>
2795 <indexterm><primary><literal>HS_OBJS</literal></primary></indexterm>
2797 <para>Object files derived from
2798 <literal>HS_SRCS</literal>.</para>
2803 <term><literal>HS_IFACES</literal></term>
2804 <indexterm><primary><literal>HS_IFACES</literal></primary></indexterm>
2806 <para>Interface files (<literal>.hi</literal> files)
2807 derived from <literal>HS_SRCS</literal>.</para>
2812 <term><literal>C_SRCS</literal></term>
2813 <indexterm><primary><literal>C_SRCS</literal></primary></indexterm>
2815 <para>All C source files found.</para>
2820 <term><literal>C_OBJS</literal></term>
2821 <indexterm><primary><literal>C_OBJS</literal></primary></indexterm>
2823 <para>Object files derived from
2824 <literal>C_SRCS</literal>.</para>
2829 <term><literal>SCRIPT_SRCS</literal></term>
2830 <indexterm><primary><literal>SCRIPT_SRCS</literal></primary></indexterm>
2832 <para>All script source files found
2833 (<literal>.lprl</literal> files).</para>
2838 <term><literal>SCRIPT_OBJS</literal></term>
2839 <indexterm><primary><literal>SCRIPT_OBJS</literal></primary></indexterm>
2841 <para><quote>object</quote> files derived from
2842 <literal>SCRIPT_SRCS</literal>
2843 (<literal>.prl</literal> files).</para>
2848 <term><literal>HSC_SRCS</literal></term>
2849 <indexterm><primary><literal>HSC_SRCS</literal></primary></indexterm>
2851 <para>All <literal>hsc2hs</literal> source files
2852 (<literal>.hsc</literal> files).</para>
2857 <term><literal>HAPPY_SRCS</literal></term>
2858 <indexterm><primary><literal>HAPPY_SRCS</literal></primary></indexterm>
2860 <para>All <literal>happy</literal> source files
2861 (<literal>.y</literal> or <literal>.hy</literal> files).</para>
2866 <term><literal>OBJS</literal></term>
2867 <indexterm><primary>OBJS</primary></indexterm>
2869 <para>the concatenation of
2870 <literal>$(HS_OBJS)</literal>,
2871 <literal>$(C_OBJS)</literal>, and
2872 <literal>$(SCRIPT_OBJS)</literal>.</para>
2877 <para>Any or all of these definitions can easily be
2878 overriden by giving new definitions in your
2879 <filename>Makefile</filename>.</para>
2881 <para>What, exactly, does <filename>paths.mk</filename>
2882 consider a <quote>source file</quote> to be? It's based
2883 on the file's suffix (e.g. <filename>.hs</filename>,
2884 <filename>.lhs</filename>, <filename>.c</filename>,
2885 <filename>.hy</filename>, etc), but this is the kind of
2886 detail that changes, so rather than enumerate the source
2887 suffices here the best thing to do is to look in
2888 <filename>paths.mk</filename>.</para>
2893 <term><filename>opts.mk</filename></term>
2894 <indexterm><primary>opts.mk</primary></indexterm>
2896 <para>defines <command>make</command> variables for option
2897 strings to pass to each program. For example, it defines
2898 <constant>HC_OPTS</constant><indexterm><primary>HC_OPTS</primary></indexterm>,
2899 the option strings to pass to the Haskell compiler. See
2900 <Xref LinkEnd="sec-suffix">.</para>
2905 <term><filename>suffix.mk</filename></term>
2906 <indexterm><primary>suffix.mk</primary></indexterm>
2908 <para>defines standard pattern rules—see <Xref
2909 LinkEnd="sec-suffix">.</para>
2914 <para>Any of the variables and pattern rules defined by the
2915 boilerplate file can easily be overridden in any particular
2916 <filename>Makefile</filename>, because the boilerplate
2917 <literal>include</literal> comes first. Definitions after this
2918 <literal>include</literal> directive simply override the default
2919 ones in <filename>boilerplate.mk</filename>.</para>
2922 <sect2 id="sec-suffix">
2923 <title>Pattern rules and options</title>
2924 <indexterm><primary>Pattern rules</primary></indexterm>
2927 <filename>suffix.mk</filename><indexterm><primary>suffix.mk</primary></indexterm>
2928 defines standard <emphasis>pattern rules</emphasis> that say how
2929 to build one kind of file from another, for example, how to
2930 build a <filename>.o</filename> file from a
2931 <filename>.c</filename> file. (GNU <command>make</command>'s
2932 <emphasis>pattern rules</emphasis> are more powerful and easier
2933 to use than Unix <command>make</command>'s <emphasis>suffix
2934 rules</emphasis>.)</para>
2936 <para>Almost all the rules look something like this:</para>
2941 $(CC) $(CC_OPTS) -c $< -o $@
2944 <para>Here's how to understand the rule. It says that
2945 <emphasis>something</emphasis><filename>.o</filename> (say
2946 <filename>Foo.o</filename>) can be built from
2947 <emphasis>something</emphasis><filename>.c</filename>
2948 (<filename>Foo.c</filename>), by invoking the C compiler (path
2949 name held in <constant>$(CC)</constant>), passing to it
2950 the options <constant>$(CC_OPTS)</constant> and
2951 the rule's dependent file of the rule
2952 <literal>$<</literal> (<filename>Foo.c</filename> in
2953 this case), and putting the result in the rule's target
2954 <literal>$@</literal> (<filename>Foo.o</filename> in this
2957 <para>Every program is held in a <command>make</command>
2958 variable defined in <filename>mk/config.mk</filename>—look
2959 in <filename>mk/config.mk</filename> for the complete list. One
2960 important one is the Haskell compiler, which is called
2961 <constant>$(HC)</constant>.</para>
2963 <para>Every program's options are are held in a
2964 <command>make</command> variables called
2965 <constant><prog>_OPTS</constant>. the
2966 <constant><prog>_OPTS</constant> variables are
2967 defined in <filename>mk/opts.mk</filename>. Almost all of them
2968 are defined like this:</para>
2971 CC_OPTS = $(SRC_CC_OPTS) $(WAY$(_way)_CC_OPTS) $($*_CC_OPTS) $(EXTRA_CC_OPTS)
2974 <para>The four variables from which
2975 <constant>CC_OPTS</constant> is built have the following
2980 <term><constant>SRC_CC_OPTS</constant><indexterm><primary>SRC_CC_OPTS</primary></indexterm>:</term>
2982 <para>options passed to all C compilations.</para>
2987 <term><constant>WAY_<way>_CC_OPTS</constant>:</term>
2989 <para>options passed to C compilations for way
2990 <literal><way></literal>. For example,
2991 <constant>WAY_mp_CC_OPTS</constant>
2992 gives options to pass to the C compiler when compiling way
2993 <literal>mp</literal>. The variable
2994 <constant>WAY_CC_OPTS</constant> holds
2995 options to pass to the C compiler when compiling the
2996 standard way. (<Xref LinkEnd="sec-ways"> dicusses
2997 multi-way compilation.)</para>
3002 <term><constant><module>_CC_OPTS</constant>:</term>
3004 <para>options to pass to the C compiler that are specific
3005 to module <literal><module></literal>. For example,
3006 <constant>SMap_CC_OPTS</constant> gives the
3007 specific options to pass to the C compiler when compiling
3008 <filename>SMap.c</filename>.</para>
3013 <term><constant>EXTRA_CC_OPTS</constant><indexterm><primary>EXTRA_CC_OPTS</primary></indexterm>:</term>
3015 <para>extra options to pass to all C compilations. This
3016 is intended for command line use, thus:</para>
3019 gmake libHS.a EXTRA_CC_OPTS="-v"
3026 <sect2 id="sec-targets">
3027 <title>The main <filename>mk/target.mk</filename> file</title>
3028 <indexterm><primary>target.mk</primary></indexterm>
3030 <para><filename>target.mk</filename> contains canned rules for
3031 all the standard targets described in <Xref
3032 LinkEnd="sec-standard-targets">. It is complicated by the fact
3033 that you don't want all of these rules to be active in every
3034 <filename>Makefile</filename>. Rather than have a plethora of
3035 tiny files which you can include selectively, there is a single
3036 file, <filename>target.mk</filename>, which selectively includes
3037 rules based on whether you have defined certain variables in
3038 your <filename>Makefile</filename>. This section explains what
3039 rules you get, what variables control them, and what the rules
3040 do. Hopefully, you will also get enough of an idea of what is
3041 supposed to happen that you can read and understand any weird
3042 special cases yourself.</para>
3046 <term><constant>HS_PROG</constant><indexterm><primary>HS_PROG</primary></indexterm>.</term>
3048 <para>If <constant>HS_PROG</constant> is defined,
3049 you get rules with the following targets:</para>
3053 <term><filename>HS_PROG</filename><indexterm><primary>HS_PROG</primary></indexterm></term>
3055 <para>itself. This rule links
3056 <constant>$(OBJS)</constant> with the Haskell
3057 runtime system to get an executable called
3058 <constant>$(HS_PROG)</constant>.</para>
3063 <term><literal>install</literal><indexterm><primary>install</primary></indexterm></term>
3066 <constant>$(HS_PROG)</constant> in
3067 <constant>$(bindir)</constant>.</para>
3076 <term><constant>C_PROG</constant><indexterm><primary>C_PROG</primary></indexterm></term>
3078 <para>is similar to <constant>HS_PROG</constant>,
3079 except that the link step links
3080 <constant>$(C_OBJS)</constant> with the C
3081 runtime system.</para>
3086 <term><constant>LIBRARY</constant><indexterm><primary>LIBRARY</primary></indexterm></term>
3088 <para>is similar to <constant>HS_PROG</constant>,
3089 except that it links
3090 <constant>$(LIB_OBJS)</constant> to make the
3091 library archive <constant>$(LIBRARY)</constant>,
3092 and <literal>install</literal> installs it in
3093 <constant>$(libdir)</constant>.</para>
3098 <term><constant>LIB_DATA</constant><indexterm><primary>LIB_DATA</primary></indexterm></term>
3100 <para>…</para>
3105 <term><constant>LIB_EXEC</constant><indexterm><primary>LIB_EXEC</primary></indexterm></term>
3107 <para>…</para>
3112 <term><constant>HS_SRCS</constant><indexterm><primary>HS_SRCS</primary></indexterm>, <constant>C_SRCS</constant><indexterm><primary>C_SRCS</primary></indexterm>.</term>
3114 <para>If <constant>HS_SRCS</constant> is defined
3115 and non-empty, a rule for the target
3116 <literal>depend</literal> is included, which generates
3117 dependency information for Haskell programs. Similarly
3118 for <constant>C_SRCS</constant>.</para>
3123 <para>All of these rules are “double-colon” rules,
3127 install :: $(HS_PROG)
3128 ...how to install it...
3131 <para>GNU <command>make</command> treats double-colon rules as
3132 separate entities. If there are several double-colon rules for
3133 the same target it takes each in turn and fires it if its
3134 dependencies say to do so. This means that you can, for
3135 example, define both <constant>HS_PROG</constant> and
3136 <constant>LIBRARY</constant>, which will generate two rules for
3137 <literal>install</literal>. When you type <command>gmake
3138 install</command> both rules will be fired, and both the program
3139 and the library will be installed, just as you wanted.</para>
3142 <sect2 id="sec-subdirs">
3143 <title>Recursion</title>
3144 <indexterm><primary>recursion, in makefiles</primary></indexterm>
3145 <indexterm><primary>Makefile, recursing into subdirectories</primary></indexterm>
3147 <para>In leaf <filename>Makefile</filename>s the variable
3148 <constant>SUBDIRS</constant><indexterm><primary>SUBDIRS</primary></indexterm>
3149 is undefined. In non-leaf <filename>Makefile</filename>s,
3150 <constant>SUBDIRS</constant> is set to the list of
3151 sub-directories that contain subordinate
3152 <filename>Makefile</filename>s. <emphasis>It is up to you to
3153 set <constant>SUBDIRS</constant> in the
3154 <filename>Makefile</filename>.</emphasis> There is no automation
3155 here—<constant>SUBDIRS</constant> is too important to
3158 <para>When <constant>SUBDIRS</constant> is defined,
3159 <filename>target.mk</filename> includes a rather neat rule for
3160 the standard targets (<Xref LinkEnd="sec-standard-targets"> that
3161 simply invokes <command>make</command> recursively in each of
3162 the sub-directories.</para>
3164 <para><emphasis>These recursive invocations are guaranteed to
3165 occur in the order in which the list of directories is specified
3166 in <constant>SUBDIRS</constant>. </emphasis>This guarantee can
3167 be important. For example, when you say <command>gmake
3168 boot</command> it can be important that the recursive invocation
3169 of <command>make boot</command> is done in one sub-directory
3170 (the include files, say) before another (the source files).
3171 Generally, put the most independent sub-directory first, and the
3172 most dependent last.</para>
3175 <sect2 id="sec-ways">
3176 <title>Way management</title>
3177 <indexterm><primary>way management</primary></indexterm>
3179 <para>We sometimes want to build essentially the same system in
3180 several different “ways”. For example, we want to build GHC's
3181 <literal>Prelude</literal> libraries with and without profiling,
3182 so that there is an appropriately-built library archive to link
3183 with when the user compiles his program. It would be possible
3184 to have a completely separate build tree for each such “way”,
3185 but it would be horribly bureaucratic, especially since often
3186 only parts of the build tree need to be constructed in multiple
3190 <filename>target.mk</filename><indexterm><primary>target.mk</primary></indexterm>
3191 contains some clever magic to allow you to build several
3192 versions of a system; and to control locally how many versions
3193 are built and how they differ. This section explains the
3196 <para>The files for a particular way are distinguished by
3197 munging the suffix. The <quote>normal way</quote> is always
3198 built, and its files have the standard suffices
3199 <filename>.o</filename>, <filename>.hi</filename>, and so on.
3200 In addition, you can build one or more extra ways, each
3201 distinguished by a <emphasis>way tag</emphasis>. The object
3202 files and interface files for one of these extra ways are
3203 distinguished by their suffix. For example, way
3204 <literal>mp</literal> has files
3205 <filename>.mp_o</filename> and
3206 <filename>.mp_hi</filename>. Library archives have their
3207 way tag the other side of the dot, for boring reasons; thus,
3208 <filename>libHS_mp.a</filename>.</para>
3210 <para>A <command>make</command> variable called
3211 <constant>way</constant> holds the current way tag.
3212 <emphasis><constant>way</constant> is only ever set on the
3213 command line of <command>gmake</command></emphasis> (usually in
3214 a recursive invocation of <command>gmake</command> by the
3215 system). It is never set inside a
3216 <filename>Makefile</filename>. So it is a global constant for
3217 any one invocation of <command>gmake</command>. Two other
3218 <command>make</command> variables,
3219 <constant>way_</constant> and
3220 <constant>_way</constant> are immediately derived from
3221 <constant>$(way)</constant> and never altered. If
3222 <constant>way</constant> is not set, then neither are
3223 <constant>way_</constant> and
3224 <constant>_way</constant>, and the invocation of
3225 <command>make</command> will build the <quote>normal
3226 way</quote>. If <constant>way</constant> is set, then the other
3227 two variables are set in sympathy. For example, if
3228 <constant>$(way)</constant> is “<literal>mp</literal>”,
3229 then <constant>way_</constant> is set to
3230 “<literal>mp_</literal>” and
3231 <constant>_way</constant> is set to
3232 “<literal>_mp</literal>”. These three variables are
3233 then used when constructing file names.</para>
3235 <para>So how does <command>make</command> ever get recursively
3236 invoked with <constant>way</constant> set? There are two ways
3237 in which this happens:</para>
3241 <para>For some (but not all) of the standard targets, when
3242 in a leaf sub-directory, <command>make</command> is
3243 recursively invoked for each way tag in
3244 <constant>$(WAYS)</constant>. You set
3245 <constant>WAYS</constant> in the
3246 <filename>Makefile</filename> to the list of way tags you
3247 want these targets built for. The mechanism here is very
3248 much like the recursive invocation of
3249 <command>make</command> in sub-directories (<Xref
3250 LinkEnd="sec-subdirs">). It is up to you to set
3251 <constant>WAYS</constant> in your
3252 <filename>Makefile</filename>; this is how you control what
3253 ways will get built.</para>
3257 <para>For a useful collection of targets (such as
3258 <filename>libHS_mp.a</filename>,
3259 <filename>Foo.mp_o</filename>) there is a rule which
3260 recursively invokes <command>make</command> to make the
3261 specified target, setting the <constant>way</constant>
3262 variable. So if you say <command>gmake
3263 Foo.mp_o</command> you should see a recursive
3264 invocation <command>gmake Foo.mp_o way=mp</command>,
3265 and <emphasis>in this recursive invocation the pattern rule
3266 for compiling a Haskell file into a <filename>.o</filename>
3267 file will match</emphasis>. The key pattern rules (in
3268 <filename>suffix.mk</filename>) look like this:
3272 $(HC) $(HC_OPTS) $< -o $@
3279 <para>You can invoke <command>make</command> with a
3280 particular <literal>way</literal> setting yourself, in order
3281 to build files related to a particular
3282 <literal>way</literal> in the current directory. eg.
3288 will build files for the profiling way only in the current
3295 <title>When the canned rule isn't right</title>
3297 <para>Sometimes the canned rule just doesn't do the right thing.
3298 For example, in the <literal>nofib</literal> suite we want the
3299 link step to print out timing information. The thing to do here
3300 is <emphasis>not</emphasis> to define
3301 <constant>HS_PROG</constant> or
3302 <constant>C_PROG</constant>, and instead define a special
3303 purpose rule in your own <filename>Makefile</filename>. By
3304 using different variable names you will avoid the canned rules
3305 being included, and conflicting with yours.</para>
3309 <sect1 id="building-docs">
3310 <title>Building the documentation</title>
3312 <sect2 id="pre-supposed-doc-tools">
3313 <title>Tools for building the Documentation</title>
3315 <para>The following additional tools are required if you want to
3316 format the documentation that comes with the
3317 <literal>fptools</literal> projects:</para>
3321 <term>DocBook</term>
3322 <indexterm><primary>pre-supposed: DocBook</primary></indexterm>
3323 <indexterm><primary>DocBook, pre-supposed</primary></indexterm>
3325 <para>Much of our documentation is written in SGML, using
3326 the DocBook DTD. Instructions on installing and
3327 configuring the DocBook tools are below.</para>
3333 <indexterm><primary>pre-supposed: TeX</primary></indexterm>
3334 <indexterm><primary>TeX, pre-supposed</primary></indexterm>
3336 <para>A decent TeX distribution is required if you want to
3337 produce printable documentation. We recomment teTeX,
3338 which includes just about everything you need.</para>
3343 <term>Haddock</term>
3344 <indexterm><primary>Haddock</primary>
3347 <para>Haddock is a Haskell documentation tool that we use
3348 for automatically generating documentation from the
3349 library source code. It is an <literal>fptools</literal>
3350 project in itself. To build documentation for the
3351 libraries (<literal>fptools/libraries</literal>) you
3352 should check out and build Haddock in
3353 <literal>fptools/haddock</literal>. Haddock requires GHC
3361 <title>Installing the DocBook tools</title>
3364 <title>Installing the DocBook tools on Linux</title>
3366 <para>If you're on a recent RedHat system (7.0+), you probably
3367 have working DocBook tools already installed. The configure
3368 script should detect your setup and you're away.</para>
3370 <para>If you don't have DocBook tools installed, and you are
3371 using a system that can handle RedHat RPM packages, you can
3372 probably use the <ULink
3373 URL="http://sourceware.cygnus.com/docbook-tools/">Cygnus
3374 DocBook tools</ULink>, which is the most shrink-wrapped SGML
3375 suite that we could find. You need all the RPMs except for
3376 psgml (i.e. <Filename>docbook</Filename>,
3377 <Filename>jade</Filename>, <Filename>jadetex</Filename>,
3378 <Filename>sgmlcommon</Filename> and
3379 <Filename>stylesheets</Filename>). Note that most of these
3380 RPMs are architecture neutral, so are likely to be found in a
3381 <Filename>noarch</Filename> directory. The SuSE RPMs also
3382 work; the RedHat ones <Emphasis>don't</Emphasis> in RedHat 6.2
3383 (7.0 and later should be OK), but they are easy to fix: just
3385 <Filename>/usr/lib/sgml/stylesheets/nwalsh-modular/lib/dblib.dsl</Filename>
3386 to <Filename>/usr/lib/sgml/lib/dblib.dsl</Filename>. </para>
3390 <title>Installing DocBook on FreeBSD</title>
3392 <para>On FreeBSD systems, the easiest way to get DocBook up
3393 and running is to install it from the ports tree or a
3394 pre-compiled package (packages are available from your local
3395 FreeBSD mirror site).</para>
3397 <para>To use the ports tree, do this:
3399 $ cd /usr/ports/textproc/docproj
3402 This installs the FreeBSD documentation project tools, which
3403 includes everything needed to format the GHC
3404 documentation.</para>
3408 <title>Installing from binaries on Windows</title>
3410 <Para>It's a good idea to use Norman Walsh's <ULink
3411 URL="http://nwalsh.com/docbook/dsssl/doc/install.html">installation
3412 notes</ULink> as a guide. You should get version 3.1 of
3413 DocBook, and note that his file <Filename>test.sgm</Filename>
3414 won't work, as it needs version 3.0. You should unpack Jade
3415 into <Filename>\Jade</Filename>, along with the entities,
3416 DocBook into <Filename>\docbook</Filename>, and the DocBook
3417 stylesheets into <Filename>\docbook\stylesheets</Filename> (so
3418 they actually end up in
3419 <Filename>\docbook\stylesheets\docbook</Filename>).</para>
3424 <title>Installing the DocBook tools from source</title>
3429 <para>Install <ULink
3430 URL="http://openjade.sourceforge.net/">OpenJade</ULink>
3431 (Windows binaries are available as well as sources). If you
3432 want DVI, PS, or PDF then install JadeTeX from the
3433 <Filename>dsssl</Filename> subdirectory. (If you get the
3437 ! LaTeX Error: Unknown option implicit=false' for package hyperref'.
3440 your version of <Command>hyperref</Command> is out of date;
3441 download it from CTAN
3442 (<Filename>macros/latex/contrib/supported/hyperref</Filename>),
3443 and make it, ensuring that you have first removed or renamed
3444 your old copy. If you start getting file not found errors
3445 when making the test for <Command>hyperref</Command>, you
3446 can abort at that point and proceed straight to
3447 <Command>make install</Command>, or enter them as
3448 <Filename>../</Filename><Emphasis>filename</Emphasis>.)</para>
3450 <para>Make links from <Filename>virtex</Filename> to
3451 <Filename>jadetex</Filename> and
3452 <Filename>pdfvirtex</Filename> to
3453 <Filename>pdfjadetex</Filename> (otherwise DVI, PostScript
3454 and PDF output will not work). Copy
3455 <Filename>dsssl/*.{dtd,dsl}</Filename> and
3456 <Filename>catalog</Filename> to
3457 <Filename>/usr/[local/]lib/sgml</Filename>.</para>
3461 <title>DocBook and the DocBook stylesheets</title>
3463 <para>Get a Zip of <ULink
3464 URL="http://www.oasis-open.org/docbook/sgml/3.1/index.html">DocBook</ULink>
3465 and install the contents in
3466 <Filename>/usr/[local/]/lib/sgml</Filename>.</para>
3468 <para>Get the <ULink
3469 URL="http://nwalsh.com/docbook/dsssl/">DocBook
3470 stylesheets</ULink> and install in
3471 <Filename>/usr/[local/]lib/sgml/stylesheets</Filename>
3472 (thereby creating a subdirectory docbook). For indexing,
3473 copy or link <Filename>collateindex.pl</Filename> from the
3474 DocBook stylesheets archive in <Filename>bin</Filename> into
3475 a directory on your <Constant>PATH</Constant>.</para>
3477 <para>Download the <ULink
3478 URL="http://www.oasis-open.org/cover/ISOEnts.zip">ISO
3479 entities</ULink> into
3480 <Filename>/usr/[local/]lib/sgml</Filename>.</para>
3486 <title>Configuring the DocBook tools</title>
3488 <Para>Once the DocBook tools are installed, the configure script
3489 will detect them and set up the build system accordingly. If you
3490 have a system that isn't supported, let us know, and we'll try
3495 <title>Remaining problems</title>
3497 <para>If you install from source, you'll get a pile of warnings
3500 <Screen>DTDDECL catalog entries are not supported</Screen>
3502 every time you build anything. These can safely be ignored, but
3503 if you find them tedious you can get rid of them by removing all
3504 the <Constant>DTDDECL</Constant> entries from
3505 <Filename>docbook.cat</Filename>.</para>
3509 <title>Building the documentation</title>
3511 <para>To build documentation in a certain format, you can
3512 say, for example,</para>
3518 <para>to build HTML documentation below the current directory.
3519 The available formats are: <literal>dvi</literal>,
3520 <literal>ps</literal>, <literal>pdf</literal>,
3521 <literal>html</literal>, and <literal>rtf</literal>. Note that
3522 not all documentation can be built in all of these formats: HTML
3523 documentation is generally supported everywhere, and DocBook
3524 documentation might support the other formats (depending on what
3525 other tools you have installed).</para>
3527 <para>All of these targets are recursive; that is, saying
3528 <literal>make html</literal> will make HTML docs for all the
3529 documents recursively below the current directory.</para>
3531 <para>Because there are many different formats that the DocBook
3532 documentation can be generated in, you have to select which ones
3533 you want by setting the <literal>SGMLDocWays</literal> variable
3534 to a list of them. For example, in
3535 <filename>build.mk</filename> you might have a line:</para>
3538 SGMLDocWays = html ps
3541 <para>This will cause the documentation to be built in the requested
3542 formats as part of the main build (the default is not to build
3543 any documentation at all).</para>
3547 <title>Installing the documentation</title>
3549 <para>To install the documentation, use:</para>
3555 <para>This will install the documentation into
3556 <literal>$(datadir)</literal> (which defaults to
3557 <literal>$(prefix)/share</literal>). The exception is HTML
3558 documentation, which goes into
3559 <literal>$(datadir)/html</literal>, to keep things tidy.</para>
3561 <para>Note that unless you set <literal>$(SGMLDocWays)</literal>
3562 to a list of formats, the <literal>install-docs</literal> target
3563 won't do anything for SGML documentation.</para>
3569 <sect1 id="sec-porting-ghc">
3570 <title>Porting GHC</title>
3572 <para>This section describes how to port GHC to a currenly
3573 unsupported platform. There are two distinct
3574 possibilities:</para>
3578 <para>The hardware architecture for your system is already
3579 supported by GHC, but you're running an OS that isn't
3580 supported (or perhaps has been supported in the past, but
3581 currently isn't). This is the easiest type of porting job,
3582 but it still requires some careful bootstrapping. Proceed to
3583 <xref linkend="sec-booting-from-hc">.</para>
3587 <para>Your system's hardware architecture isn't supported by
3588 GHC. This will be a more difficult port (though by comparison
3589 perhaps not as difficult as porting gcc). Proceed to <xref
3590 linkend="unregisterised-porting">.</para>
3594 <sect2 id="sec-booting-from-hc">
3595 <title>Booting/porting from C (<filename>.hc</filename>) files</title>
3597 <indexterm><primary>building GHC from .hc files</primary></indexterm>
3598 <indexterm><primary>booting GHC from .hc files</primary></indexterm>
3599 <indexterm><primary>porting GHC</primary></indexterm>
3601 <para>Bootstrapping GHC on a system without GHC already
3602 installed is achieved by taking the intermediate C files (known
3603 as HC files) from a GHC compilation on a supported system to the
3604 target machine, and compiling them using gcc to get a working
3607 <para><emphasis>NOTE: GHC versions 5.xx and later are
3608 significantly harder to bootstrap from C than earlier versions.
3609 We recommend starting from version 4.08.2 if you need to
3610 bootstrap in this way.</emphasis></para>
3612 <para>HC files are architecture-dependent (but not
3613 OS-dependent), so you have to get a set that were generated on
3614 similar hardware. There may be some supplied on the GHC
3615 download page, otherwise you'll have to compile some up
3616 yourself, or start from <emphasis>unregisterised</emphasis> HC
3617 files - see <xref linkend="unregisterised-porting">.</para>
3619 <para>The following steps should result in a working GHC build
3620 with full libraries:</para>
3624 <para>Unpack the HC files on top of a fresh source tree
3625 (make sure the source tree version matches the version of
3626 the HC files <emphasis>exactly</emphasis>!). This will
3627 place matching <filename>.hc</filename> files next to the
3628 corresponding Haskell source (<filename>.hs</filename> or
3629 <filename>.lhs</filename>) in the compiler subdirectory
3630 <filename>ghc/compiler</filename> and in the libraries
3631 (subdirectories of <filename>hslibs</filename> and
3632 <literal>libraries</literal>).</para>
3636 <para>The actual build process is fully automated by the
3637 <filename>hc-build</filename> script located in the
3638 <filename>distrib</filename> directory. If you eventually
3639 want to install GHC into the directory
3640 <replaceable>dir</replaceable>, the following
3641 command will execute the whole build process (it won't
3642 install yet):</para>
3645 foo% distrib/hc-build --prefix=<replaceable>dir</replaceable>
3647 <indexterm><primary>--hc-build</primary></indexterm>
3649 <para>By default, the installation directory is
3650 <filename>/usr/local</filename>. If that is what you want,
3651 you may omit the argument to <filename>hc-build</filename>.
3652 Generally, any option given to <filename>hc-build</filename>
3653 is passed through to the configuration script
3654 <filename>configure</filename>. If
3655 <filename>hc-build</filename> successfully completes the
3656 build process, you can install the resulting system, as
3666 <sect2 id="unregisterised-porting">
3667 <title>Porting GHC to a new architecture</title>
3669 <para>The first step in porting to a new architecture is to get
3670 an <firstterm>unregisterised</firstterm> build working. An
3671 unregisterised build is one that compiles via vanilla C only.
3672 By contrast, a registerised build uses the following
3673 architecture-specific hacks for speed:</para>
3677 <para>Global register variables: certain abstract machine
3678 <quote>registers</quote> are mapped to real machine
3679 registers, depending on how many machine registers are
3681 <filename>ghc/includes/MachRegs.h</filename>).</para>
3685 <para>Assembly-mangling: when compiling via C, we feed the
3686 assembly generated by gcc though a Perl script known as the
3687 <firstterm>mangler</firstterm> (see
3688 <filename>ghc/driver/mangler/ghc-asm.lprl</filename>). The
3689 mangler rearranges the assembly to support tail-calls and
3690 various other optimisations.</para>
3694 <para>In an unregisterised build, neither of these hacks are
3695 used — the idea is that the C code generated by the
3696 compiler should compile using gcc only. The lack of these
3697 optimisations costs about a factor of two in performance, but
3698 since unregisterised compilation is usually just a step on the
3699 way to a full registerised port, we don't mind too much.</para>
3702 <title>Building an unregisterised port</title>
3704 <para>The first step is to get some unregisterised HC files.
3705 Either (a) download them from the GHC site (if there are
3706 some available for the right version of GHC), or
3707 (b) build them yourself on any machine with a working
3708 GHC. If at all possible this should be a machine with the
3709 same word size as the target.</para>
3711 <para>There is a script available which should automate the
3712 process of doing the 2-stage bootstrap necessary to get the
3713 unregisterised HC files - it's available in <ulink
3714 url="http://cvs.haskell.org/cgi-bin/cvsweb.cgi/fptools/distrib/cross-port"><filename>fptools/distrib/cross-port</filename></ulink>
3717 <para>Now take these unregisterised HC files to the target
3718 platform and bootstrap a compiler from them as per the
3719 instructions in <xref linkend="sec-booting-from-hc">. In
3720 <filename>build.mk</filename>, you need to tell the build
3721 system that the compiler you're building is
3722 (a) unregisterised itself, and (b) builds
3723 unregisterised binaries. This varies depending on the GHC
3724 version you're bootstraping:</para>
3727 # build.mk for GHC 4.08.x
3728 GhcWithRegisterised=NO
3732 # build.mk for GHC 5.xx and 6.x
3733 GhcUnregisterised=YES
3736 <para>Versions 5.xx and 6.x only: use the option
3737 <option>--enable-hc-boot-unregisterised</option> instead of
3738 <option>--enable-hc-boot</option> when running
3739 <filename>./configure</filename>.</para>
3741 <para>The build may not go through cleanly. We've tried to
3742 stick to writing portable code in most parts of the compiler,
3743 so it should compile on any POSIXish system with gcc, but in
3744 our experience most systems differ from the standards in one
3745 way or another. Deal with any problems as they arise - if you
3746 get stuck, ask the experts on
3747 <email>glasgow-haskell-users@haskell.org</email>.</para>
3749 <para>Once you have the unregisterised compiler up and
3750 running, you can use it to start a registerised port. The
3751 following sections describe the various parts of the system
3752 that will need architecture-specific tweaks in order to get a
3753 registerised build going.</para>
3755 <para>Lots of useful information about the innards of GHC is
3756 available in the <ulink
3757 url="http://www.cse.unsw.edu.au/~chak/haskell/ghc/comm/">GHC
3758 Commentary</ulink>, which might be helpful if you run into
3759 some code which needs tweaking for your system.</para>
3763 <title>Porting the RTS</title>
3765 <para>The following files need architecture-specific code for a
3766 registerised build:</para>
3770 <term><filename>ghc/includes/MachRegs.h</filename></term>
3771 <indexterm><primary><filename>MachRegs.h</filename></primary>
3774 <para>Defines the STG-register to machine-register
3775 mapping. You need to know your platform's C calling
3776 convention, and which registers are generally available
3777 for mapping to global register variables. There are
3778 plenty of useful comments in this file.</para>
3782 <term><filename>ghc/includes/TailCalls.h</filename></term>
3783 <indexterm><primary><filename>TailCalls.h</filename></primary>
3786 <para>Macros that cooperate with the mangler (see <xref
3787 linkend="sec-mangler">) to make proper tail-calls
3792 <term><filename>ghc/rts/Adjustor.c</filename></term>
3793 <indexterm><primary><filename>Adjustor.c</filename></primary>
3797 <literal>foreign import "wrapper"</literal>
3799 <literal>foreign export dynamic</literal>).
3800 Not essential for getting GHC bootstrapped, so this file
3801 can be deferred until later if necessary.</para>
3805 <term><filename>ghc/rts/StgCRun.c</filename></term>
3806 <indexterm><primary><filename>StgCRun.c</filename></primary>
3809 <para>The little assembly layer between the C world and
3810 the Haskell world. See the comments and code for the
3811 other architectures in this file for pointers.</para>
3815 <term><filename>ghc/rts/MBlock.h</filename></term>
3816 <term><filename>ghc/rts/MBlock.c</filename></term>
3817 <indexterm><primary><filename>MBlock.h</filename></primary>
3819 <indexterm><primary><filename>MBlock.c</filename></primary>
3822 <para>These files are really OS-specific rather than
3823 architecture-specific. In <filename>MBlock.h</filename>
3824 is specified the absolute location at which the RTS
3825 should try to allocate memory on your platform (try to
3826 find an area which doesn't conflict with code or dynamic
3827 libraries). In <filename>Mblock.c</filename> you might
3828 need to tweak the call to <literal>mmap()</literal> for
3835 <sect3 id="sec-mangler">
3836 <title>The mangler</title>
3838 <para>The mangler is an evil Perl-script that rearranges the
3839 assembly code output from gcc to do two main things:</para>
3843 <para>Remove function prologues and epilogues, and all
3844 movement of the C stack pointer. This is to support
3845 tail-calls: every code block in Haskell code ends in an
3846 explicit jump, so we don't want the C-stack overflowing
3847 while we're jumping around between code blocks.</para>
3850 <para>Move the <firstterm>info table</firstterm> for a
3851 closure next to the entry code for that closure. In
3852 unregisterised code, info tables contain a pointer to the
3853 entry code, but in registerised compilation we arrange
3854 that the info table is shoved right up against the entry
3855 code, and addressed backwards from the entry code pointer
3856 (this saves a word in the info table and an extra
3857 indirection when jumping to the closure entry
3862 <para>The mangler is abstracted to a certain extent over some
3863 architecture-specific things such as the particular assembler
3864 directives used to herald symbols. Take a look at the
3865 definitions for other architectures and use these as a
3866 starting point.</para>
3870 <title>The native code generator</title>
3872 <para>The native code generator isn't essential to getting a
3873 registerised build going, but it's a desirable thing to have
3874 because it can cut compilation times in half. The native code
3875 generator is described in some detail in the <ulink
3876 url="http://www.cse.unsw.edu.au/~chak/haskell/ghc/comm/">GHC
3877 commentary</ulink>.</para>
3883 <para>To support GHCi, you need to port the dynamic linker
3884 (<filename>fptools/ghc/rts/Linker.c</filename>). The linker
3885 currently supports the ELF and PEi386 object file formats - if
3886 your platform uses one of these then you probably don't have
3887 to do anything except fiddle with the
3888 <literal>#ifdef</literal>s at the top of
3889 <filename>Linker.c</filename> to tell it about your OS.</para>
3891 <para>If your system uses a different object file format, then
3892 you have to write a linker — good luck!</para>
3898 <sect1 id="sec-build-pitfalls">
3899 <title>Known pitfalls in building Glasgow Haskell
3901 <indexterm><primary>problems, building</primary></indexterm>
3902 <indexterm><primary>pitfalls, in building</primary></indexterm>
3903 <indexterm><primary>building pitfalls</primary></indexterm></title>
3906 WARNINGS about pitfalls and known “problems”:
3915 One difficulty that comes up from time to time is running out of space
3916 in <literal>TMPDIR</literal>. (It is impossible for the configuration stuff to
3917 compensate for the vagaries of different sysadmin approaches to temp
3919 <indexterm><primary>tmp, running out of space in</primary></indexterm>
3921 The quickest way around it is <command>setenv TMPDIR /usr/tmp</command><indexterm><primary>TMPDIR</primary></indexterm> or
3922 even <command>setenv TMPDIR .</command> (or the equivalent incantation with your shell
3925 The best way around it is to say
3928 export TMPDIR=<dir>
3931 in your <filename>build.mk</filename> file.
3932 Then GHC and the other <literal>fptools</literal> programs will use the appropriate directory
3941 In compiling some support-code bits, e.g., in <filename>ghc/rts/gmp</filename> and even
3942 in <filename>ghc/lib</filename>, you may get a few C-compiler warnings. We think these
3950 When compiling via C, you'll sometimes get “warning: assignment from
3951 incompatible pointer type” out of GCC. Harmless.
3958 Similarly, <command>ar</command>chiving warning messages like the following are not
3962 ar: filename GlaIOMonad__1_2s.o truncated to GlaIOMonad_
3963 ar: filename GlaIOMonad__2_2s.o truncated to GlaIOMonad_
3973 In compiling the compiler proper (in <filename>compiler/</filename>), you <emphasis>may</emphasis>
3974 get an “Out of heap space” error message. These can vary with the
3975 vagaries of different systems, it seems. The solution is simple:
3982 If you're compiling with GHC 4.00 or later, then the
3983 <emphasis>maximum</emphasis> heap size must have been reached. This
3984 is somewhat unlikely, since the maximum is set to 64M by default.
3985 Anyway, you can raise it with the
3986 <option>-optCrts-M<size></option> flag (add this flag to
3987 <constant><module>_HC_OPTS</constant>
3988 <command>make</command> variable in the appropriate
3989 <filename>Makefile</filename>).
3996 For GHC < 4.00, add a suitable <option>-H</option> flag to the <filename>Makefile</filename>, as
4005 and try again: <command>gmake</command>. (see <Xref LinkEnd="sec-suffix"> for information about
4006 <constant><module>_HC_OPTS</constant>.)
4008 Alternatively, just cut to the chase:
4012 % make EXTRA_HC_OPTS=-optCrts-M128M
4021 If you try to compile some Haskell, and you get errors from GCC about
4022 lots of things from <filename>/usr/include/math.h</filename>, then your GCC was
4023 mis-installed. <command>fixincludes</command> wasn't run when it should've been.
4025 As <command>fixincludes</command> is now automagically run as part of GCC installation,
4026 this bug also suggests that you have an old GCC.
4034 You <emphasis>may</emphasis> need to re-<command>ranlib</command><indexterm><primary>ranlib</primary></indexterm> your libraries (on Sun4s).
4038 % cd $(libdir)/ghc-x.xx/sparc-sun-sunos4
4039 % foreach i ( `find . -name '*.a' -print` ) # or other-shell equiv...
4041 ? # or, on some machines: ar s $i
4046 We'd be interested to know if this is still necessary.
4054 GHC's sources go through <command>cpp</command> before being compiled, and <command>cpp</command> varies
4055 a bit from one Unix to another. One particular gotcha is macro calls
4060 SLIT("Hello, world")
4064 Some <command>cpp</command>s treat the comma inside the string as separating two macro
4065 arguments, so you get
4069 :731: macro `SLIT' used with too many (2) args
4073 Alas, <command>cpp</command> doesn't tell you the offending file!
4075 Workaround: don't put weird things in string args to <command>cpp</command> macros.
4086 <Sect1 id="winbuild"><Title>Notes for building under Windows</Title>
4089 This section summarises how to get the utilities you need on your
4090 Win95/98/NT/2000 machine to use CVS and build GHC. Similar notes for
4091 installing and running GHC may be found in the user guide. In general,
4092 Win95/Win98 behave the same, and WinNT/Win2k behave the same.
4093 You should read the GHC installation guide sections on Windows (in the user
4094 guide) before continuing to read these notes.
4098 <sect2 id="cygwin-and-mingw"><Title>Cygwin and MinGW</Title>
4100 <para> The Windows situation for building GHC is rather confusing. This section
4101 tries to clarify, and to establish terminology.</para>
4103 <sect3 id="ghc-mingw"><title>GHC-mingw</title>
4105 <para> <ulink url="http://www.mingw.org">MinGW (Minimalist GNU for Windows)</ulink>
4106 is a collection of header
4107 files and import libraries that allow one to use <command>gcc</command> and produce
4108 native Win32 programs that do not rely on any third-party DLLs. The
4109 current set of tools include GNU Compiler Collection (<command>gcc</command>), GNU Binary
4110 Utilities (Binutils), GNU debugger (Gdb), GNU make, and a assorted
4113 <para>The GHC that we distribute includes, inside the distribution itself, the MinGW <command>gcc</command>,
4114 <command>as</command>, <command>ld</command>, and a bunch of input/output libraries.
4115 GHC compiles Haskell to C (or to
4116 assembly code), and then invokes these MinGW tools to generate an executable binary.
4117 The resulting binaries can run on any Win32 system.
4119 <para> We will call a GHC that targets MinGW in this way <emphasis>GHC-mingw</emphasis>.</para>
4121 <para> The down-side of GHC-mingw is that the MinGW libraries do not support anything like the full
4122 Posix interface. So programs compiled with GHC-mingw cannot import the (Haskell) Posix
4123 library; they have to do
4124 their input output using standard Haskell I/O libraries, or native Win32 bindings.
4128 <sect3 id="ghc-cygwin"><title>GHC-cygwin</title>
4130 <para>There <emphasis>is</emphasis> a way to get the full Posix interface, which is to use Cygwin.
4131 <ulink url="http://www.cygwin.com">Cygwin</ulink> is a complete Unix simulation that runs on Win32.
4132 Cygwin comes with a shell, and all the usual Unix commands: <command>mv</command>, <command>rm</command>,
4133 <command>ls</command>, plus of course <command>gcc</command>, <command>ld</command> and so on.
4134 A C program compiled with the Cygwin <command>gcc</command> certainly can use all of Posix.
4136 <para>So why doesn't GHC use the Cygwin <command>gcc</command> and libraries? Because
4137 Cygwin comes with a DLL <emphasis>that must be linked with every runnable Cygwin-compiled program</emphasis>.
4138 A program compiled by the Cygwin tools cannot run at all unless Cygwin is installed.
4139 If GHC targeted Cygwin, users would have to install Cygwin just to run the Haskell programs
4140 that GHC compiled; and the Cygwin DLL would have to be in the DLL load path.
4141 Worse, Cygwin is a moving target. The name of the main DLL, <literal>cygwin1.dll</literal>
4142 does not change, but the implementation certainly does. Even the interfaces to functions
4143 it exports seem to change occasionally. So programs compiled by GHC might only run with
4144 particular versions of Cygwin. All of this seems very undesirable.
4147 Nevertheless, it is certainly possible to build a version of GHC that targets Cygwin;
4148 we will call that <emphasis>GHC-cygwin</emphasis>. The up-side of GHC-cygwin is
4149 that Haskell programs compiled by GHC-cygwin can import the (Haskell) Posix library.
4153 <sect3><title>HOST_OS vs TARGET_OS</title>
4156 In the source code you'll find various ifdefs looking like:
4158 #ifdef mingw32_HOST_OS
4164 #ifdef mingw32_TARGET_OS
4168 These macros are set by the configure script (via the file config.h).
4169 Which is which? The criterion is this. In the ifdefs in GHC's source code:
4172 The "host" system is the one on which GHC itself will be run.
4175 The "target" system is the one for which the program compiled by GHC will be run.
4178 For a stage-2 compiler, in which GHCi is available, the "host" and "target" systems must be the same.
4179 So then it doesn't really matter whether you use the HOST_OS or TARGET_OS cpp macros.
4184 <sect3><title>Summary</title>
4186 <para>Notice that "GHC-mingw" means "GHC that <emphasis>targets</emphasis> MinGW". It says nothing about
4187 how that GHC was <emphasis>built</emphasis>. It is entirely possible to have a GHC-mingw that was built
4188 by compiling GHC's Haskell sources with a GHC-cygwin, or vice versa.</para>
4190 <para>We distribute only a GHC-mingw built by a GHC-mingw; supporting
4191 GHC-cygwin too is beyond our resources. The GHC we distribute
4192 therefore does not require Cygwin to run, nor do the programs it
4193 compiles require Cygwin.</para>
4195 <para>The instructions that follow describe how to build GHC-mingw. It is
4196 possible to build GHC-cygwin, but it's not a supported route, and the build system might
4199 <para>In your build tree, you build a compiler called <Command>ghc-inplace</Command>. It
4200 uses the <Command>gcc</Command> that you specify using the
4201 <option>--with-gcc</option> flag when you run
4202 <Command>configure</Command> (see below).
4203 The makefiles are careful to use <Command>ghc-inplace</Command> (not <Command>gcc</Command>)
4204 to compile any C files, so that it will in turn invoke the right <Command>gcc</Command> rather that
4205 whatever one happens to be in your path. However, the makefiles do use whatever <Command>ld</Command>
4206 and <Command>ar</Command> happen to be in your path. This is a bit naughty, but (a) they are only
4207 used to glom together .o files into a bigger .o file, or a .a file,
4208 so they don't ever get libraries (which would be bogus; they might be the wrong libraries), and (b)
4209 Cygwin and Mingw use the same .o file format. So its ok.
4214 <Sect2><Title>Installing and configuring Cygwin</Title>
4216 <para>You don't need Cygwin to <emphasis>use</emphasis> GHC,
4217 but you do need it to <emphasis>build</emphasis> GHC.</para>
4219 <para> Install Cygwin from <ulink url="http://www.cygwin.com/">http://www.cygwin.com/</ulink>.
4220 The installation process is straightforward; we install it in <Filename>c:/cygwin</Filename>.
4221 During the installation dialogue, make sure that you select:
4222 <command>cvs</command>, <command>openssh</command>,
4223 <command>autoconf</command>,
4224 <command>binutils</command> (includes ld and (I think) ar),
4225 <command>gcc</command>,
4226 <command>flex</command>,
4227 <command>make</command>.
4230 <para> Now set the following user environment variables:
4233 <listitem><para> Add <filename>c:/cygwin/bin</filename> and <filename>c:/cygwin/usr/bin</filename> to your
4234 <constant>PATH</constant></para></listitem>
4238 Set <constant>MAKE_MODE</constant> to <Literal>UNIX</Literal>. If you
4239 don't do this you get very weird messages when you type
4240 <Command>make</Command>, such as:
4242 /c: /c: No such file or directory
4247 <listitem><para> Set <constant>SHELL</constant> to
4248 <Filename>c:/cygwin/bin/sh</Filename>. When you invoke a shell in Emacs, this
4249 <constant>SHELL</constant> is what you get.
4252 <listitem><para> Set <constant>HOME</constant> to point to your
4253 home directory. This is where, for example,
4254 <command>bash</command> will look for your <filename>.bashrc</filename>
4255 file. Ditto <command>emacs</command> looking for <filename>.emacsrc</filename>
4261 There are a few other things to do:
4265 By default, cygwin provides the command shell <filename>ash</filename>
4266 as <filename>sh.exe</filename>. We have often seen build-system problems that
4267 turn out to be due to bugs in <filename>ash</filename>
4269 and length of command lines). On the other hand <filename>bash</filename> seems
4271 So, in <filename>cygwin/bin</filename>
4272 remove the supplied <filename>sh.exe</filename> (or rename it as <filename>ash.exe</filename>),
4273 and copy <filename>bash.exe</filename> to <filename>sh.exe</filename>.
4274 You'll need to do this in Windows Explorer or the Windows <command>cmd</command> shell, because
4275 you can't rename a running program!
4281 Some script files used in the make system start with "<Command>#!/bin/perl</Command>",
4282 (and similarly for <Command>sh</Command>). Notice the hardwired path!
4283 So you need to ensure that your <Filename>/bin</Filename> directory has the following
4286 <listitem> <para><Command>sh</Command></para></listitem>
4287 <listitem> <para><Command>perl</Command></para></listitem>
4288 <listitem> <para><Command>cat</Command></para></listitem>
4290 All these come in Cygwin's <Filename>bin</Filename> directory, which you probably have
4291 installed as <Filename>c:/cygwin/bin</Filename>. By default Cygwin mounts "<Filename>/</Filename>" as
4292 <Filename>c:/cygwin</Filename>, so if you just take the defaults it'll all work ok.
4293 (You can discover where your Cygwin
4294 root directory <Filename>/</Filename> is by typing <Command>mount</Command>.)
4295 Provided <Filename>/bin</Filename> points to the Cygwin <Filename>bin</Filename>
4296 directory, there's no need to copy anything. If not, copy these binaries from the <filename>cygwin/bin</filename>
4297 directory (after fixing the <filename>sh.exe</filename> stuff mentioned in the previous bullet).
4303 <para>Finally, here are some things to be aware of when using Cygwin:
4305 <listitem> <para>Cygwin doesn't deal well with filenames that include
4306 spaces. "<filename>Program Files</filename>" and "<filename>Local files</filename>" are
4310 <listitem> <para> Cygwin implements a symbolic link as a text file with some
4311 magical text in it. So other programs that don't use Cygwin's
4312 I/O libraries won't recognise such files as symlinks.
4313 In particular, programs compiled by GHC are meant to be runnable
4314 without having Cygwin, so they don't use the Cygwin library, so
4315 they don't recognise symlinks.
4319 Win32 has a <command>find</command> command which is not the same as Cygwin's find.
4320 You will probably discover that the Win32 <command>find</command> appears in your <constant>PATH</constant>
4321 before the Cygwin one, because it's in the <emphasis>system</emphasis> <constant>PATH</constant>
4322 environment variable, whereas you have probably modified the <emphasis>user</emphasis> <constant>PATH</constant>
4323 variable. You can always invoke <command>find</command> with an absolute path, or rename it.
4330 <Sect2><Title>Other things you need to install</Title>
4332 <para>You have to install the following other things to build GHC:
4336 Install an executable GHC, from <ulink url="http://www.haskell.org/ghc">http://www.haskell.org/ghc</ulink>.
4337 This is what you will use to compile GHC. Add it in your
4338 <constant>PATH</constant>: the installer tells you the path element
4339 you need to add upon completion.
4345 Install an executable Happy, from <ulink url="http://www.haskell.org/happy">http://www.haskell.org/happy</ulink>.
4346 Happy is a parser generator used to compile the Haskell grammar. Add it in your
4347 <constant>PATH</constant>.
4353 <para>GHC uses the <emphasis>mingw</emphasis> C compiler to
4354 generate code, so you have to install that (see <xref linkend="cygwin-and-mingw">).
4355 Just pick up a mingw bundle at
4356 <ulink url="http://www.mingw.org/">http://www.mingw.org/</ulink>.
4357 We install it in <filename>c:/mingw</filename>.
4359 <para>Do <emphasis>not</emphasis> add any of the <emphasis>mingw</emphasis> binaries to your path.
4360 They are only going to get used by explicit access (via the --with-gcc flag you
4361 give to <Command>configure</Command> later). If you do add them to your path
4362 you are likely to get into a mess because their names overlap with Cygwin binaries.
4368 <para>We use <command>emacs</command> a lot, so we install that too.
4369 When you are in <filename>fptools/ghc/compiler</filename>, you can use
4370 "<literal>make tags</literal>" to make a TAGS file for emacs. That uses the utility
4371 <filename>fptools/ghc/utils/hasktags/hasktags</filename>, so you need to make that first.
4372 The most convenient way to do this is by going <literal>make boot</literal> in <filename>fptools/ghc</filename>.
4373 The <literal>make tags</literal> command also uses <command>etags</command>, which comes with <command>emacs</command>,
4374 so you will need to add <filename>emacs/bin</filename> to your <literal>PATH</literal>.
4380 <para> Finally, check out a copy of GHC sources from
4381 the CVS repository, following the instructions above (<xref linkend="cvs-access">).
4388 <Sect2><Title>Building GHC</Title>
4391 Now go read the documentation above on building from source (<xref linkend="sec-building-from-source">);
4392 the bullets below only tell
4393 you about Windows-specific wrinkles.</para>
4397 Run <Command>autoconf</Command> both in <filename>fptools</filename>
4398 and in <filename>fptools/ghc</filename>. If you omit the latter step you'll
4399 get an error when you run <filename>./configure</filename>:
4402 creating mk/config.h
4403 mk/config.h is unchanged
4405 running /bin/sh ./configure --cache-file=.././config.cache --srcdir=.
4406 ./configure: ./configure: No such file or directory
4407 configure: error: ./configure failed for ghc
4412 <listitem> <para><command>autoconf</command> seems to create the file <filename>configure</filename>
4413 read-only. So if you need to run autoconf again (which I sometimes do for safety's sake),
4416 /usr/bin/autoconf: cannot create configure: permission denied
4418 Solution: delete <filename>configure</filename> first.
4423 You either need to add <filename>ghc</filename> to your
4424 <constant>PATH</constant> before you invoke
4425 <Command>configure</Command>, or use the <Command>configure</Command>
4426 option <option>--with-ghc=c:/ghc/ghc-some-version/bin/ghc</option>.
4431 If you are paranoid, delete <filename>config.cache</filename> if it exists.
4432 This file occasionally remembers out-of-date configuration information, which
4433 can be really confusing.
4439 After <command>autoconf</command> run <command>./configure</command> in
4440 <filename>fptools/</filename> thus:
4443 ./configure --host=i386-unknown-mingw32 --with-gcc=c:/mingw/bin/gcc
4445 This is the point at which you specify that you are building GHC-mingw
4446 (see <xref linkend="ghc-mingw">). </para>
4448 <para> Both these options are important! It's possible to get into
4449 trouble using the wrong C compiler!</para>
4451 Furthermore, it's <emphasis>very important</emphasis> that you specify a
4452 full MinGW path for <command>gcc</command>, not a Cygwin path, because GHC (which
4453 uses this path to invoke <command>gcc</command>) is a MinGW program and won't
4454 understand a Cygwin path. For example, if you
4455 say <literal>--with-gcc=/mingw/bin/gcc</literal>, it'll be interpreted as
4456 <filename>/cygdrive/c/mingw/bin/gcc</filename>, and GHC will fail the first
4457 time it tries to invoke it. Worse, the failure comes with
4458 no error message whatsoever. GHC simply fails silently when first invoked,
4459 typically leaving you with this:
4461 make[4]: Leaving directory `/cygdrive/e/fptools-stage1/ghc/rts/gmp'
4462 ../../ghc/compiler/ghc-inplace -optc-mno-cygwin -optc-O
4463 -optc-Wall -optc-W -optc-Wstrict-prototypes -optc-Wmissing-prototypes
4464 -optc-Wmissing-declarations -optc-Winline -optc-Waggregate-return
4465 -optc-Wbad-function-cast -optc-Wcast-align -optc-I../includes
4466 -optc-I. -optc-Iparallel -optc-DCOMPILING_RTS
4467 -optc-fomit-frame-pointer -O2 -static
4468 -package-name rts -O -dcore-lint -c Adjustor.c -o Adjustor.o
4469 make[2]: *** [Adjustor.o] Error 1
4470 make[1]: *** [all] Error 1
4471 make[1]: Leaving directory `/cygdrive/e/fptools-stage1/ghc'
4472 make: *** [all] Error 1
4478 If you want to build GHC-cygwin (<xref linkend="ghc-cygwin">)
4479 you'll have to do something more like:
4481 ./configure --with-gcc=...the Cygwin gcc...
4486 <listitem><para> You almost certainly want to set
4490 in your <filename>build.mk</filename> configuration file (see <xref linkend="sec-build-config">).
4491 This tells the build system not to split each library into a myriad of little object files, one
4492 for each function. Doing so reduces binary sizes for statically-linked binaries, but on Windows
4493 it dramatically increases the time taken to build the libraries in the first place.
4497 <listitem><para> Do not attempt to build the documentation.
4498 It needs all kinds of wierd Jade stuff that we haven't worked out for
4499 Win32.</para></listitem>