Skip to content

GitLab

GHC
GHC
Last edited by Tobias Dammers
Page history New page

ghc old mac

Building GHC on Mac OS X 10.2 in 136 Easy Steps, by Kirsten aged 26 1/52

  1. grab the HEAD off darcs, per Building/GettingTheSources (no problems here)
  2. run autoreconf (this went ok, because I had just upgraded my autoconf in order to build darcs)
  3. run ./configure
  4. oops, I only have happy 1.13, it wants happy 1.15
  5. go to the happy download page. what, no Mac OS X binary?
  6. grab the happy sources
  7. ./configure, make, make install. so far so good
  8. ./configure GHC again
  9. LOL, I need alex 2.0
  10. why isn't this integrated into the GHC build process? also, googling for just "alex" is un-useful. so is googling for "alex lexer" and "alex lexer haskell"
  11. LOL, *still* no Mac OS X binary.
  12. ./configure; make in alex
  13. have disgusting IM conversation with friend while waiting
  14. alex: "you lose at life":
Creating a symbolic link from alex-2.0.1 to alex in /usr/local/bin failed: `/usr/local/bin/alex' already exists
Perhaps remove `/usr/local/bin/alex' manually?
make[2]: *** [install] Error 1
make[1]: *** [install] Error 1
make: *** [install] Error 1
  1. consider a career change
  2. rm /usr/local/bin/alex
  3. sudo rm /usr/local/bin/alex
  4. sudo make me a sandwich
  5. sudo make install
  6. okay, I have alex. yippee.
  7. ./configure in GHC again
  8. seems to have worked. With trembling fingers (or maybe that's just the freezing Southern California weather), type "make".
  9. OH NOEZ:
gcc -O -DTABLES_NEXT_TO_CODE -I. -I../rts    -c mkDerivedConstants.c -o mkDerivedConstants.o
InfoTables.h:314: illegal member declaration, missing name, found `}'
OSThreads.h:135: #error "Threads not supported"
OSThreads.h:141: undefined type, found `OSThreadId'
OSThreads.h:145: illegal external declaration, missing `;' after `OSThreadProcAttr'
OSThreads.h:145: illegal external declaration, missing `;' after `*'
OSThreads.h:147: undefined type, found `OSThreadId'
OSThreads.h:148: undefined type, found `OSThreadProc'
OSThreads.h:153: undefined type, found `Condition'
OSThreads.h:154: undefined type, found `Condition'
OSThreads.h:155: undefined type, found `Condition'
OSThreads.h:156: undefined type, found `Condition'
OSThreads.h:157: undefined type, found `Condition'
OSThreads.h:158: undefined type, found `Mutex'
OSThreads.h:163: undefined type, found `Mutex'
OSThreads.h:164: undefined type, found `Mutex'
OSThreads.h:169: undefined type, found `ThreadLocalKey'
OSThreads.h:170: undefined type, found `ThreadLocalKey'
OSThreads.h:171: undefined type, found `ThreadLocalKey'
Storage.h:211: undefined type, found `Mutex'
Storage.h:212: undefined type, found `Mutex'
../rts/Task.h:88: undefined type, found `OSThreadId'
../rts/Task.h:115: undefined type, found `Condition'
../rts/Task.h:116: undefined type, found `Mutex'
../rts/Task.h:225: illegal function prototype, found `*'
../rts/Task.h:225: illegal function definition, found `)'
../rts/Task.h:235: undefined type, found `ThreadLocalKey'
../rts/Capability.h:74: undefined type, found `Mutex'
../rts/Capability.h:197: undefined type, found `Mutex'
cpp-precomp: warning: errors during smart preprocessing, retrying in basic mode
make[1]: *** [mkDerivedConstants.o] Error 1
make: *** [stage1] Error 1
  1. Cry.
  2. Hmm, how old is my gcc, anyway?
% gcc --version
gcc (GCC) 3.1 20020420 (prerelease)
Copyright (C) 2002 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. [yeah, no shit, Sherlock ]
  1. Will upgrading my gcc break everything? Probably. Should I do it anyway? Probably.
  2. Actually, upgrading gcc sounds about as appealing as a quadruple root canal. Maybe I can use gcc2 (2.95.2) instead.
  3. Aaaaargh, of course I'm not that lucky:
checking for gcc... gcc2
checking for C compiler default output file name... 
configure: error: C compiler cannot create executables
  1. Start downloading tarball for newer gcc from: http://www.opensource.apple.com/darwinsource/August2003GCCUpdate/. Also consider becoming religious so I can pray for it to be actually useful.
Length: 31,674,171 [application/x-tar]

 2% [>                                    ] 895,686       14.36K/s    ETA 34:52
  1. Wish that parents-in-law would get the faster internets at their house.
  2. Eat breakfast.
  3. Installing gcc on a full stomach is probably better anyway (though quite possibly not). Unzip the tarball. Notice that INSTALL is a directory. That's already a bad sign.
  4. Hope that autoconf; ./configure; make; make install will just work, since I don't have the attention span for anything else.
  5. Run make.
  6. For some reason, the configure script thinks that my PPC laptop is an i386. ./configure doesn't do anything. Nor does ./configure --help.
  7. No, actually it thinks I want to build a cross-compiler. Why would I want that?
  8. Apparently the hosts to build for are controlled by the HOSTS environment variable, or maybe it's set somewhere in one of the config files. I'm not sure. Nor am I sure why I have the attention span to grovel through the sources but not to read the documentation.
  9. Try setenv HOSTS ppc
  10. make
  11. Still complains with the same message:
********************************************************************************
* Building Apple GCC 3.3 Compiler(s) (languages = c++,c,objc,objc++) for thins *
* ---------------------------------------------------------------------------- *
* BUILDHOST       = Kirsten-Chevaliers-Computer.local. -- a ppc                *
* HOSTS           = ppc  i386                                                  *
* TARGETS         = ppc  i386                                                  *
* SRCROOT         = /tmp/gcc3/gcc-1495                                         *
* OBJROOT         = /tmp/gcc3/gcc-1495/obj                                     *
* SYMROOT         = /tmp/gcc3/gcc-1495/obj/../sym                              *
* DSTROOT         = /tmp/gcc3/gcc-1495/obj/../dst                              *
* RC_RELEASE      =                                                            *
* CFLAGS          = -g                                                         *
* OPT_OVERRIDE    =                                                            *
* NEXT_ROOT       =                                                            *
* BUILD           = ppc                                                        *
* BOOTSTRAP       = yes                                                        *
* PREFIX          = /usr                                                       *
* DO_SYMLINKS     = no                                                         *
* ENABLE_CHECKING = --disable-checking                                         *
* Default cc      =                                                            *
* Curr. Hdrs.     = 3.1                                                        *
* ---------------------------------------------------------------------------- *
* 12/27/06 10:08:49 PST                                                        *
********************************************************************************


????????????????????????????????????????????????????????????
? The directory /usr/libexec/gcc/darwin/i386 is missing!!! ?
? Please install a compiler that generates code for i386.  ?
????????????????????????????????????????????????????????????

make: *** [build] Error 1
  1. Try export HOSTS
  2. make
  3. Still complains.
  4. For the 65536th time in my life, consider destroying everything Turing-complete and then going out to enjoy the big room with the blue ceilings.
  5. Apparently you can also set the TARGETS environment variable. Try setting it to ppc also.
??????????????????????????????????????????
? Host type i386 should also be a target ?
??????????????????????????????????????????
  1. Well, maybe if I could also set the HOSTS to just ppc, that wouldn't be a problem.
  2. Use grep again (the only IDE I'll ever need) and figure out to edit the GNUmakefile to change the HOSTS = and targets = lines to obliterate any and all traces of i386. There's probably a better way. At this point I don't care.
  3. Start make again. It is, at least, convinced that it only wants to be building a ppc compiler, without any of that crazy i386 nonsense.
  4. Go back to bed.
  5. ...but not *yet*. The hell?
++++++++++++++++++++++++++++++++++++++++++
+ Building libiberty                     +
+ -------------------------------------- +
+ cwd = /tmp/gcc3/gcc-1495/obj/libiberty +
++++++++++++++++++++++++++++++++++++++++++

+ gnumake srcdir=/tmp/gcc3/gcc-1495/libiberty BUILD_PREFIX=ppc- BUILD_PREFIX_1=ppc- 'HOST_CC= cc -arch ppc -no-cpp-precomp' 'CFLA\
GS= -g' 'GCC_CFLAGS=-no-cpp-precomp -g' 'BOOT_CFLAGS=-O2 -g -no-cpp-precomp -mdynamic-no-pic' 'CC=cc -arch ppc -arch ppc -no-cpp-\
precomp -g'
if [x"" != x ] && [! -d pic ]; then \
  mkdir pic; \
else true; fi
touch stamp-picdir
if [x"" != x ]; then \
  cc -arch ppc -arch ppc -no-cpp-precomp -g -c -DHAVE_CONFIG_H -g -I. -I/tmp/gcc3/gcc-1495/libiberty/../include  -W -Wall -Wtradi\
tional -pedantic  /tmp/gcc3/gcc-1495/libiberty/regex.c -o pic/regex.o; \
else true; fi
cc -arch ppc -arch ppc -no-cpp-precomp -g -c -DHAVE_CONFIG_H -g -I. -I/tmp/gcc3/gcc-1495/libiberty/../include  -W -Wall -Wtraditi\
onal -pedantic /tmp/gcc3/gcc-1495/libiberty/regex.c -o regex.o
gnumake[1]: cc: Command not found
gnumake[1]: *** [regex.o] Error 127
+ status=2
+ command set +x
+ set +x

*********************************************
* *** gnumake failed building libiberty *** *
*********************************************
  1. WTF is "libiberty"?
  2. Apparently, not only does it think it should be using cc instead of gcc (which are supposed to be the same anyway, no?), the PATH isn't getting exported correctly, because cc is certainly in /usr, which is in my PATH.
  3. Try setting the CC environment variable to {{{/usr/gcc}}.
  4. make. Doesn't work.
  5. make clean
  6. make. Doesn't work.
  7. Edit the makefile in the libiberty subdir: CC = gcc. Again, there's probably a better way...
  8. make. Doesn't work.
  9. make clean
  10. make. Really go back to bed.
  11. Decide for some reason to get back out of bed and leave the other warm body in it behind in order to start building gcc again. Consider whether I need psychological help.
  12. The problem is that it can't find cc in the PATH and although /usr, where cc is, SHOULD be in the PATH, it's not getting exported properly fsr.
  13. Create a symbolic link to /usr/cc in /usr/bin.
  14. Chainsaw, meet butter.
  15. make. libiberty makes it all the way through this time.
  16. NOW WHAT
cc -arch ppc -no-cpp-precomp -g -c -no-cpp-precomp  -g -no-cpp-precomp -DHAVE_DESIGNATED_INITIALIZERS=0 -DIN_GCC   -W -Wall -Wwri\
te-strings -Wstrict-prototypes -Wmissing-prototypes -Wtraditional -pedantic -Wno-long-long  -mdynamic-no-pic -DHAVE_CONFIG_H -DGE\
NERATOR_FILE    -I. -I. -I/tmp/gcc3/gcc-1495/gcc -I/tmp/gcc3/gcc-1495/gcc/. -I/tmp/gcc3/gcc-1495/gcc/config -I/tmp/gcc3/gcc-1495/\
gcc/../include -I/tmp/gcc3/gcc-1495/gcc/../more-hdrs \
  ./gengtype-lex.c -o gengtype-lex.o
gengtype-lex.c:1: warning: ISO C forbids an empty source file
cc -arch ppc -no-cpp-precomp -g -c -no-cpp-precomp  -g -no-cpp-precomp -DHAVE_DESIGNATED_INITIALIZERS=0 -DIN_GCC   -W -Wall -Wwri\
te-strings -Wstrict-prototypes -Wmissing-prototypes -Wtraditional -pedantic -Wno-long-long  -mdynamic-no-pic -DHAVE_CONFIG_H -DGE\
NERATOR_FILE    -I. -I. -I/tmp/gcc3/gcc-1495/gcc -I/tmp/gcc3/gcc-1495/gcc/. -I/tmp/gcc3/gcc-1495/gcc/config -I/tmp/gcc3/gcc-1495/\
gcc/../include -I/tmp/gcc3/gcc-1495/gcc/../more-hdrs \
  ./gengtype-yacc.c -o gengtype-yacc.o
cc: ./gengtype-yacc.c: No such file or directory
cc: no input files
gnumake[2]: *** [gengtype-yacc.o] Error 1
gnumake[1]: *** [stage1_build] Error 2
+ status=2
+ command set +x
+ set +x

********************************************
* *** gnumake failed building compiler *** *
********************************************
  1. Figure this is probably because I don't have bison and yacc installed. Consider the comical aptness of the term "yak-shaving" here.
  2. Wait a sec, I do have bison and yacc! They're under /usr! Why the hell are they under /usr, anyway, and why isn't /usr in my PATH? Actually, that was probably just because I unzipped a tarball in the wrong place at some point. I guess I'll just put them all in /usr/bin.
  3. make clean so configure can notice that bison and yacc and flex are there now. make 
(cd . && \
 bison  -d -o gengtype-yacc.c /tmp/gcc3/gcc-1495/gcc/gengtype-yacc.y || \
 ( rm -f gengtype-yacc.c && false ) )
bison: /usr/share/bison.simple: No such file or directory
gnumake[2]: *** [gengtype-yacc.c] Error 1
gnumake[1]: *** [stage1_build] Error 2
+ status=2
+ command set +x
+ set +x

********************************************
* *** gnumake failed building compiler *** *
********************************************
  1. WTH is bison.simple?
  2. Download bison.simple off the intorblags and put it in /usr/bin.
  3. Oh, it wants it to be in /usr/share. Well, sure, put it there too.
  4. It's missing another include file:
In file included from /tmp/gcc3/gcc-1495/gcc/config/rs6000/host-darwin.c:25:
/usr/include/sys/mman.h:62:39: mach/shared_memory_server.h: No such file or directory
In file included from /tmp/gcc3/gcc-1495/gcc/config/rs6000/host-darwin.c:25:
/usr/include/sys/mman.h:170: parse error before "sf_mapping_t"
/usr/include/sys/mman.h:171: parse error before "sf_mapping_t"
/tmp/gcc3/gcc-1495/gcc/config/rs6000/host-darwin.c: In function `segv_handler':
/tmp/gcc3/gcc-1495/gcc/config/rs6000/host-darwin.c:66: dereferencing pointer to incomplete type
/tmp/gcc3/gcc-1495/gcc/config/rs6000/host-darwin.c:115: dereferencing pointer to incomplete type
/tmp/gcc3/gcc-1495/gcc/config/rs6000/host-darwin.c:115: dereferencing pointer to incomplete type
gnumake[2]: *** [host-darwin.o] Error 1
gnumake[1]: *** [stage1_build] Error 2
  1. What is shared_memory_server.h and why isn't there an easy way to answer the question "what is it, and to get it can I just download the file somewhere or do I have to finally give in and buy the upgrade to Mac OS 10.4?"
  2. Wait a day. Drive to the Apple store.
  3. Get told that of course I can't exchange money for the goods and services I want, because my four-year-old laptop is too old.
  4. Consider buying a PC.
  5. Post on the GHC mailing list to seek pity (or at least laughs).
  6. Per a resulting suggestion, start downloading the .dmg for gcc 3.3 off http://connect.apple.com/ (which of course isn't google-able because of course you have to create an account and mumble mumble deep web mumble.)
  7. Be thankful that at least this one is only going to take 23 minutes to download instead of 34. Get dressed.
  8. Try to install the gcc .pkg. It says I need the December 2002 Developer Tools update too (which is of course exactly what the person on the mailing list said as well.)
  9. Start the Dec. 2002 update downloading. 6-hour download. Crie.
  10. Be grateful that at least you're not on dialup.
  11. Be grateful that at least you have First World problems rather than the other kind.
  12. About 10 hours later, double-click the installer. Wait 15 minutes.
  13. "The software was successfully installed." Now there's an experience I haven't had in a while.
  14. Double-click the August 2003 GCC updater. Wait 2 minutes.
  15. "The software was successfully installed." I'm on a roll!!
  16. So, where was I two days ago, anyway? Oh yeah, trying to build GHC! That was the original point, wasn't it? I'm not sure anymore.
  17. Of course, GHC is still looking for gcc under /usr (since it's first in your PATH for some reason) and not under /usr/bin where it really lives.
  18. Edit PATH to put /usr at the end. This doesn't work.
  19. Play some symlink games.
  20. Now gcc points to gcc 3.3, but the same error message comes up as before when I run make in ghc:
% make
make -C utils/mkdependC boot
------------------------------------------------------------------------
== make boot -r;
 in /Users/krc/ghc-head/ghc/includes
------------------------------------------------------------------------
gcc -O -DTABLES_NEXT_TO_CODE -I. -I../rts    -c mkDerivedConstants.c -o mkDerivedConstants.o
In file included from Rts.h:144,
                 from mkDerivedConstants.c:23:
OSThreads.h:135:4: #error "Threads not supported"
In file included from Rts.h:144,
                 from mkDerivedConstants.c:23:
OSThreads.h:141: error: parse error before "osThreadId"
OSThreads.h:141: warning: data definition has no type or storage class
OSThreads.h:145: error: parse error before "OSThreadProc"
OSThreads.h:145: warning: data definition has no type or storage class
OSThreads.h:147: error: parse error before '*' token
OSThreads.h:153: error: parse error before '*' token
OSThreads.h:154: error: parse error before '*' token
OSThreads.h:155: error: parse error before '*' token
OSThreads.h:156: error: parse error before '*' token
OSThreads.h:157: error: parse error before '*' token
OSThreads.h:163: error: parse error before '*' token
OSThreads.h:164: error: parse error before '*' token
OSThreads.h:169: error: parse error before '*' token
OSThreads.h:170: error: parse error before '*' token
OSThreads.h:171: error: parse error before '*' token
In file included from Rts.h:160,
                 from mkDerivedConstants.c:23:
Storage.h:211: error: parse error before "sm_mutex"
Storage.h:211: warning: data definition has no type or storage class
Storage.h:212: error: parse error before "atomic_modify_mutvar_mutex"
Storage.h:212: warning: data definition has no type or storage class
In file included from ../rts/Capability.h:27,
                 from mkDerivedConstants.c:28:
../rts/Task.h:88: error: parse error before "OSThreadId"
../rts/Task.h:88: warning: no semicolon at end of struct or union
../rts/Task.h:115: error: parse error before "cond"
../rts/Task.h:115: warning: data definition has no type or storage class
../rts/Task.h:116: error: parse error before "lock"
../rts/Task.h:116: warning: data definition has no type or storage class
../rts/Task.h:155: error: parse error before '}' token
../rts/Task.h:155: warning: data definition has no type or storage class
../rts/Task.h:158: error: parse error before '*' token
../rts/Task.h: In function `isBoundTask':
../rts/Task.h:160: error: `task' undeclared (first use in this function)
../rts/Task.h:160: error: (Each undeclared identifier is reported only once
../rts/Task.h:160: error: for each function it appears in.)
../rts/Task.h: At top level:
../rts/Task.h:166: error: parse error before '*' token
../rts/Task.h:166: warning: data definition has no type or storage class
../rts/Task.h:178: error: parse error before '*' token
../rts/Task.h:178: warning: data definition has no type or storage class
../rts/Task.h:183: error: parse error before '*' token
../rts/Task.h:188: error: parse error before '*' token
../rts/Task.h:194: error: parse error before '*' token
../rts/Task.h:200: error: parse error before '*' token
../rts/Task.h:205: error: parse error before '*' token
../rts/Task.h:209: error: parse error before '*' token
../rts/Task.h:209: warning: data definition has no type or storage class
../rts/Task.h:225: error: parse error before '*' token
../rts/Task.h:225: error: parse error before '*' token
../rts/Task.h:225: error: `OSThreadProcAttr' declared as function returning a function
../rts/Task.h:235: error: parse error before "currentTaskKey"
../rts/Task.h:235: warning: data definition has no type or storage class
../rts/Task.h:246: error: parse error before '*' token
../rts/Task.h:257: error: parse error before '*' token
../rts/Task.h: In function `setMyTask':
../rts/Task.h:260: error: `task' undeclared (first use in this function)
../rts/Task.h: At top level:
../rts/Task.h:270: error: parse error before '*' token
../rts/Task.h: In function `taskEnter':
../rts/Task.h:274: error: `task' undeclared (first use in this function)
In file included from mkDerivedConstants.c:28:
../rts/Capability.h: At top level:
../rts/Capability.h:42: error: parse error before "Task"
../rts/Capability.h:42: warning: no semicolon at end of struct or union
../rts/Capability.h:61: error: parse error before '*' token
../rts/Capability.h:61: warning: data definition has no type or storage class
../rts/Capability.h:71: error: parse error before '*' token
../rts/Capability.h:71: warning: data definition has no type or storage class
../rts/Capability.h:74: error: parse error before "lock"
../rts/Capability.h:74: warning: data definition has no type or storage class
../rts/Capability.h:81: error: parse error before '*' token
../rts/Capability.h:81: warning: data definition has no type or storage class
../rts/Capability.h:82: error: parse error before '*' token
../rts/Capability.h:82: warning: data definition has no type or storage class
../rts/Capability.h:97: error: parse error before '}' token
../rts/Capability.h:174: error: parse error before "Task"
../rts/Capability.h:191: error: parse error before "Task"
../rts/Capability.h:197: error: parse error before '*' token
../rts/Capability.h:220: error: parse error before "Task"
../rts/Capability.h:224: error: parse error before "Task"
../rts/Capability.h: In function `recordMutableCap':
../rts/Capability.h:247: error: dereferencing pointer to incomplete type
../rts/Capability.h:253: error: dereferencing pointer to incomplete type
mkDerivedConstants.c: In function `main':
mkDerivedConstants.c:228: error: dereferencing pointer to incomplete type
mkDerivedConstants.c:228: error: dereferencing pointer to incomplete type
mkDerivedConstants.c:229: error: dereferencing pointer to incomplete type
mkDerivedConstants.c:229: error: dereferencing pointer to incomplete type
mkDerivedConstants.c:231: error: dereferencing pointer to incomplete type
mkDerivedConstants.c:232: error: dereferencing pointer to incomplete type
mkDerivedConstants.c:233: error: dereferencing pointer to incomplete type
mkDerivedConstants.c:233: error: dereferencing pointer to incomplete type
make[1]: *** [mkDerivedConstants.o] Error 1
make: *** [stage1] Error 1
[Kirsten-Chevaliers-Computer:~/ghc-head/ghc] krc%
  1. Go to bed.
  2. No, no, can't rest as long as this problem isn't solved. Maybe this is a terrible idea, but what if I try ./configure --disable-threaded-rts?
  3. make clean
  4. make
  5. OMFG!!!111 (closed) it at least makes it past mkDerivedConstants.c this time
  6. :-((((
------------------------------------------------------------------------
== make boot -r;
 in /Users/krc/ghc-head/ghc/compat
------------------------------------------------------------------------
../utils/mkdependC/mkdependC -f .depend    -I. -Iinclude -I../includes  -- -O -I. -Iinclude -D__\
GHC_PATCHLEVEL__=1 -I../libraries/base/cbits -I../libraries/base/include    -- cbits/directory.c\
 cbits/rawSystem.c cbits/unicode.c
/usr/local/bin/ghc -M -optdep-f -optdep.depend  -osuf o    -H16m -O -I. -Iinclude -Rghc-timing -\
I../libraries -fglasgow-exts -no-recomp Compat/Directory.hs Compat/RawSystem.hs Compat/Unicode.h\
s Distribution/Compat/FilePath.hs Distribution/Compat/ReadP.hs Distribution/Compiler.hs Distribu\
tion/GetOpt.hs Distribution/InstalledPackageInfo.hs Distribution/License.hs Distribution/Package\
.hs Distribution/ParseUtils.hs Distribution/Version.hs Language/Haskell/Extension.hs System/Dire\
ctory/Internals.hs
<<ghc: 8979932 bytes, 3 GCs, 9280/9280 avg/max bytes residency (1 samples), 15M in use, 0.04 INI\
T (0.02 elapsed), 0.17 MUT (1.53 elapsed), 0.03 GC (0.04 elapsed) :ghc>>
make all
/usr/local/bin/ghc -H16m -O -I. -Iinclude -Rghc-timing  -I../libraries -fglasgow-exts -no-recomp\
    -c System/Directory/Internals.hs -o System/Directory/Internals.o  -ohi System/Directory/Inte\
rnals.hi
System/Directory/Internals.hs:1: parse error on input `#'
<<ghc: 5251084 bytes, 3 GCs, 5688/5688 avg/max bytes residency (1 samples), 16M in use, 0.03 INI\
T (0.01 elapsed), 0.06 MUT (0.31 elapsed), 0.02 GC (0.02 elapsed) :ghc>>
make[2]: *** [System/Directory/Internals.o] Error 1
make[1]: *** [boot] Error 2
make: *** [stage1] Error 1
  1. Well, my /usr/local/bin/ghc is really old. 6.0.1, in fact. Maybe I need something newer for bootstrapping. What does the documentation say? Have I ever read it before? If not, why break my streak?
  2. Hmm, says "version 5.04 at least." 6.0.1 > 5.04 (or so I would naively think).
  3. System/Directory/Internals.hs has an OPTIONS pragma. Why am I suspecting that GHC 6.0.1 doesn't know about this pragma? Would the documentation lie to me? (Probably.)
  4. Then again, I no longer have any idea where my version of GHC 6.0.1 came from.
  5. Maybe I'll download a binary of GHC instead.
  6. Oh wait, I CAN'T do that, because I have Mac OS 10.2.
  7. Hey, I might have a functioning other version of GHC sitting around. These ideas just keep getting worse and worse...
./configure --with-ghc=~/GHC/ghc-6.4.2/ghc/compiler/ghc-inplace --disable-threaded-rts
  1. Huh, it got all the way through ./configure even though apparently the path name wasn't getting canonicalized. Nice!
./configure --with-ghc=/Users/krc/GHC/ghc-6.4.2/ghc/compiler/ghc-inplace --disable-threaded-rts
  1. Oh, CRIE, that's not going to work, I never finished building the libraries with the other GHC, and indeed COULDN'T.
  2. Back to square one:
./configure --with-ghc=/usr/local/bin/ghc --disable-threaded-rts
  1. Wonder how I ever got this installed in the first place.
  2. Sleep.
  3. In the morning, look at that Internals.hs file more carefully. Something weird seems to be happening with cpp. For one thing, the file paths in the includes in Internals.hs seem to be wrong, but upon fixing them, cpp dumps in a bunch of lines that begin with things like # 1, which ghc doesn't seem to like.
  4. Start building ghc on a Linux system to compare what that file ends up looking like. (Which should be interesting since that's a webhosting account where autoconf doesn't even seem to be installed. Hope it doesn't matter!)
  5. Okay, well, compat/ builds perfectly on the Linux box. No big shock, I guess. But what does cpp do?
  6. Huh, on the Linux box, that Internals.hs file doesn't seem to even get touched. A configure problem? Or maybe it really is needed on Mac OS but not Linux?
  7. Aw, crap, looking at the Makefile under compat/ tells me that Internals.hs is getting built because I'm bootstrapping with a ghc less than 6.3, whereas on that Linux box, I have ghc 6.4 already, so that's that mystery solved.
  8. Why can't I get a binary ghc for OS 10.2? (Because, I guess, everyone else upgraded already, and I'm the last of the Mohicans.)
  9. So, I suspect Internals.hs may be bitrotted, but that still doesn't solve the mystery of how to get cpp to do the right thing with it.
  10. And thanks to a minimal test case, it's not even those line pragmas that's making ghc throw up.
  11. Go on IRC.
  12. A very nice person, pejo, points out: http://www.haskell.org/ghc/dist/6.4/MacOSX/
  13. Download the mysterious tarball from there and unzip it.
  14. Reconfigure:
./configure --with-ghc=/private/tmp/ghc_d/ghc-bootstrap/bin/ghc --disable-threaded-rts
  1. That went okay. make
  2. Edit the paths in /tmp/ghc_d/ghc-bootstrap/bin/ghc (which is a shell script) to change /opt/local to /tmp/ghc_d/ghc-bootstrap. Four times.
  3. make again
  4. Oh noez:
make -C utils/mkdependC boot
------------------------------------------------------------------------
== make boot -r;
 in /Users/krc/ghc-head/ghc/includes
------------------------------------------------------------------------
../utils/mkdependC/mkdependC -f .depend     -- -O -DTABLES_NEXT_TO_CODE -I. -I../rts    -- mkDerivedConstants.c
------------------------------------------------------------------------
== make boot -r;
 in /Users/krc/ghc-head/ghc/compat
------------------------------------------------------------------------
/bin/sh: test: unknownunknown: integer expression expected
/private/tmp/ghc_d/ghc-bootstrap/bin/ghc -M -optdep-f -optdep.depend  -osuf o -optdep--exclude-module=System.Directory.Internals   -H16m -O -I. -Iinclude -Rghc-timing -I../li\
braries -fglasgow-exts -no-recomp Compat/Directory.hs Compat/RawSystem.hs Compat/Unicode.hs Distribution/Compat/FilePath.hs Distribution/Compat/ReadP.hs Distribution/Compiler\
.hs Distribution/GetOpt.hs Distribution/InstalledPackageInfo.hs Distribution/License.hs Distribution/Package.hs Distribution/ParseUtils.hs Distribution/Version.hs Language/Ha\
skell/Extension.hs tmp.hs
dyld: /private/tmp/ghc_d/ghc-bootstrap/lib/ghc-6.4/ghc-6.4 can't open library: /opt/local/lib/libreadline.5.0.dylib  (No such file or directory, errno = 2)
make[1]: *** [depend] Trace/BPT trap
make: *** [stage1] Error 1
  1. Wonder whether 11:49 AM is too early in the day to start drinking.
  2. Do I have libreadline? Yes, but only version 4. Is that good enough? Maybe. Try some horrible symlink games.
  3. Damn:
dyld: /private/tmp/ghc_d/ghc-bootstrap/lib/ghc-6.4/ghc-6.4 version mismatch for library: /opt/local/lib/libreadline.5.0.dylib (compatibility version of user: 5.0.0 greater th\
an library's version: 4.0.0)
make[1]: *** [depend] Trace/BPT trap
  1. Okay, I'll just go download readline 5.0 instead. Google says that's "trivial".
  2. Several days later after getting distracted by other things, try to actually build readline. Realize it's not actually "trivial".
  3. Buy a PC.