LLVM-GCC fails to build on MingW (non bootstrap).

[llvmbot]

Bugzilla Link	3552
Resolution	FIXED
Resolved on	Nov 07, 2018 00:22
Version	unspecified
OS	All
Reporter	LLVM Bugzilla Contributor
CC	@asl,@sunfishcode,@tlattner

Extended Description

LLVM-GCC doesn't build on MingW, starting with revision 61207, doing a non-bootstrap build.

• rev 61207 introduces the following error:

make[4]: Entering directory /c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61207/gcc-build/i686-pc-mingw32/libstdc++-v3/include' if [ ! -d "./i686-pc-mingw32/bits/stdc++.h.gch" ]; then \ mkdir -p ./i686-pc-mingw32/bits/stdc++.h.gch; \ fi; \ /c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61207/gcc-build/./gcc/xgcc -shared-libgcc -B/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61207/gcc-build/./gcc -nostdinc++ -L/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61207/gcc -build/i686-pc-mingw32/libstdc++-v3/src -L/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61207/gcc-build/i686-pc-mingw32/libstdc++-v3/src/.libs -L/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61207/gcc-build/i686-pc-mingw32/winsup/ mingw -L/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61207/gcc-build/i686-pc-mingw32/winsup/w32api/lib -isystem /c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-gcc4.2-src/winsup/mingw/include -isystem /c/cygwin/home/jlerouge/buildb ot/llvm-test-fixed-gcc4.2-src/winsup/w32api/include -Bc:/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61207/i686-pc-mingw32/bin/ -Bc:/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61207/i686-pc-mingw32/lib/ -isystem c:/cygwin/home/jler ouge/buildbot/llvm-test-fixed-61207/i686-pc-mingw32/include -isystem c:/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61207/i686-pc-mingw32/sys-include -Winvalid-pch -Wno-deprecated -x c++-header -g -O2 -I/c/cygwin/home/jlerouge/buildb ot/llvm-test-fixed-61207/gcc-build/i686-pc-mingw32/libstdc++-v3/include/i686-pc-mingw32 -I/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61207/gcc-build/i686-pc-mingw32/libstdc++-v3/include -I/c/cygwin/home/jlerouge/buildbot/llvm-test -fixed-gcc4.2-src/libstdc++-v3/libsupc++ -O0 -g /c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-gcc4.2-src/libstdc++-v3/include/precompiled/stdc++.h -o i686-pc-mingw32/bits/stdc++.h.gch/O0g.gch make[4]: *** [i686-pc-mingw32/bits/stdc++.h.gch/O0g.gch] Error 1 make[4]: Leaving directory /c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61207/gcc-build/i686-pc-mingw32/libstdc++-v3/include'
make[3]: *** [all-recursive] Error 1
make[3]: Leaving directory /c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61207/gcc-build/i686-pc-mingw32/libstdc++-v3' make[2]: *** [all] Error 2 make[2]: Leaving directory /c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61207/gcc-build/i686-pc-mingw32/libstdc++-v3'
make[1]: *** [all-target-libstdc++-v3] Error 2
make[1]: Leaving directory `/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61207/gcc-build'

SVN log:

r61207 | dpatel | 2008-12-18 11:18:20 -0800 (Thu, 18 Dec 2008) | 3 lines
Changed paths:
M /llvm-gcc-4.2/trunk/gcc/llvm-debug.cpp

Reuse DIType slot used for foward declaration for real definition.

---

• And then, rev 61215 introduces another error, earlier in the build process:

/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61215/gcc-build/./gcc/xgcc -B/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61215/gcc-build/./gcc/ -L/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61215/gcc-build/i686-pc-mingw32/win
sup/mingw -L/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61215/gcc-build/i686-pc-mingw32/winsup/w32api/lib -isystem /c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-gcc4.2-src/winsup/mingw/include -isystem /c/cygwin/home/jlerouge/bu
ildbot/llvm-test-fixed-gcc4.2-src/winsup/w32api/include -Bc:/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61215/i686-pc-mingw32/bin/ -Bc:/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61215/i686-pc-mingw32/lib/ -isystem c:/cygwin/home/
jlerouge/buildbot/llvm-test-fixed-61215/i686-pc-mingw32/include -isystem c:/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61215/i686-pc-mingw32/sys-include -c -DHAVE_CONFIG_H -O2 -g -O2 -I. -I../../../../llvm-test-fixed-gcc4.2-src/libi
berty/../include -W -Wall -pedantic -Wwrite-strings -Wstrict-prototypes -Wc++-compat ../../../../llvm-test-fixed-gcc4.2-src/libiberty/pex-win32.c -o pex-win32.o
../../../../llvm-test-fixed-gcc4.2-src/libiberty/pex-win32.c: In function 'argv_to_cmdline':
../../../../llvm-test-fixed-gcc4.2-src/libiberty/pex-win32.c:365: warning: request for implicit conversion from 'void *' to 'char *' not permitted in C++
../../../../llvm-test-fixed-gcc4.2-src/libiberty/pex-win32.c: In function 'find_executable':
../../../../llvm-test-fixed-gcc4.2-src/libiberty/pex-win32.c:440: warning: request for implicit conversion from 'void *' to 'char *' not permitted in C++
../../../../llvm-test-fixed-gcc4.2-src/libiberty/pex-win32.c: In function 'win32_spawn':
../../../../llvm-test-fixed-gcc4.2-src/libiberty/pex-win32.c:566: warning: request for implicit conversion from 'void *' to 'char **' not permitted in C++
../../../../llvm-test-fixed-gcc4.2-src/libiberty/pex-win32.c:573: warning: request for implicit conversion from 'void *' to 'char *' not permitted in C++
../../../../llvm-test-fixed-gcc4.2-src/libiberty/pex-win32.c: At top level:
../../../../llvm-test-fixed-gcc4.2-src/libiberty/pex-win32.c:900: internal compiler error: Segmentation fault
Please submit a full bug report,
with preprocessed source if appropriate.
See URL:http://developer.apple.com/bugreporter for instructions.
make[2]: *** [pex-win32.o] Error 1
make[2]: Leaving directory /c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61215/gcc-build/i686-pc-mingw32/libiberty' make[1]: *** [all-target-libiberty] Error 2 make[1]: Leaving directory /c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-61215/gcc-build'
make: *** [all] Error 2

SVN log:

r61215 | djg | 2008-12-18 14:03:42 -0800 (Thu, 18 Dec 2008) | 3 lines
Changed paths:
M /llvm/trunk/lib/Target/X86/X86RegisterInfo.cpp

When emitting instructions that define EFLAGS and the EFLAGS value isn't
used, mark the defs as dead.

---

I believe those two problems are still there in the current 2.5 branch (as of rev 64291).

[tlattner]

These are both blockers for the 2.5 release.

[asl]

Will check

[sunfishcode]

Can someone test if r64316 from trunk fixes the second
issue -- the regression introduced by r61215?
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20090209/073612.html

[llvmbot]

Testing right now... thanks.

[llvmbot]

... still building, I had to apply the patch here also to build LLVM: http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20090209/073573.html

Although there seems to be something weird going on with llvm-config, the one generated on my MingW was empty at first. Removing it and rebuilding fixed the pbm though (I'll investigate later).

[llvmbot]

Ok, r64316 is indeed fixing the pex-win32 error, but I am now hitting a new assertion later on in the build:

/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-64316/gcc-build/./gcc/xgcc -B/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-64316/gcc-build/./gcc/ -L/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-64316/gcc-build/i686-pc-mingw32/win
sup/mingw -L/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-64316/gcc-build/i686-pc-mingw32/winsup/w32api/lib -isystem /c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-gcc4.2-src/winsup/mingw/include -isystem /c/cygwin/home/jlerouge/bu
ildbot/llvm-test-fixed-gcc4.2-src/winsup/w32api/include -Bc:/cygwin/home/jlerouge/buildbot/llvm-test-fixed-64316/i686-pc-mingw32/bin/ -Bc:/cygwin/home/jlerouge/buildbot/llvm-test-fixed-64316/i686-pc-mingw32/lib/ -isystem c:/cygwin/home/
jlerouge/buildbot/llvm-test-fixed-64316/i686-pc-mingw32/include -isystem c:/cygwin/home/jlerouge/buildbot/llvm-test-fixed-64316/i686-pc-mingw32/sys-include -O2 -I../../../llvm-test-fixed-gcc4.2-src/gcc/../winsup/w32api/include -O2 -g -
O2 -DIN_GCC -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -Wold-style-definition -isystem ./include -g -DIN_LIBGCC2 -D__GCC_FLOAT_NOT_NEEDED -I. -I. -I../../../llvm-test-fixed-gcc4.2-src/gcc -I../../../llvm-
test-fixed-gcc4.2-src/gcc/. -I../../../llvm-test-fixed-gcc4.2-src/gcc/../include -I../../../llvm-test-fixed-gcc4.2-src/gcc/../libcpp/include -I../../../llvm-test-fixed-gcc4.2-src/gcc/../libdecnumber -I../libdecnumber -Ic:/cygwin/home/j
lerouge/buildbot/llvm-test-fixed-64316/llvm-build/include -I/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-src/include -DL_mulsc3 -c ../../../llvm-test-fixed-gcc4.2-src/gcc/libgcc2.c -o libgcc/./_mulsc3.o
Assertion failed: StackTop > 0 && "Cannot pop empty stack!", file c:/cygwin/home/jlerouge/buildbot/llvm-test-fixed-src/lib/Target/X86/X86FloatingPoint.cpp, line 553

This application has requested the Runtime to terminate it in an unusual way.
Please contact the application's support team for more information.
../../../llvm-test-fixed-gcc4.2-src/gcc/libgcc2.c:1914: internal compiler error: Aborted
Please submit a full bug report,
with preprocessed source if appropriate.
See URL:http://developer.apple.com/bugreporter for instructions.
make[3]: *** [libgcc/./_mulsc3.o] Error 1
make[3]: Leaving directory /c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-64316/gcc-build/gcc' make[2]: *** [libgcc.a] Error 2 make[2]: Leaving directory /c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-64316/gcc-build/gcc'
make[1]: *** [all-gcc] Error 2
make[1]: Leaving directory `/c/cygwin/home/jlerouge/buildbot/llvm-test-fixed-64316/gcc-build'
make: *** [all] Error 2
++ FailedExit

[llvmbot]

My last try was on the trunk, I will try again on the 2.5 branch to see how far it goes, with Dan's fix.

[llvmbot]

Building on the 2.5 branch went through (minor glitch because my version of make does not support order-only dependencies, but that's an easy upgrade). Dan's fix is the only patch required to get llvm-gcc on MingW !

Any idea about what's causing the assertion on the trunk ?

[llvmbot]

Ok the trunk is working now (it wasn't as of rev 64365). I think Evan fixed if with this one:

---

r64374 | evancheng | 2009-02-12 02:32:17 -0800 (Thu, 12 Feb 2009) | 2 lines
Changed paths:
M /llvm/trunk/lib/CodeGen/VirtRegMap.cpp
A /llvm/trunk/test/CodeGen/X86/2009-02-12-SpillerBug.ll

It's (currently) not safe to keep certain physical registers live across basic blocks, e.g. x86 fp stack registers.

To summarize:

• trunk is good now (I checked in the Alarm thing and the googletest fix).
• 2.5 branch is fixed if Dan's patch (r64316) is applied and googletest is disabled.

Tanya, what do you want to do for googletest in 2.5 ? Can you merge Dan's fix ?

Thanks,
Julien

[asl]

• 2.5 branch is fixed if Dan's patch (r64316) is applied and googletest is
  disabled.
  I'm currently testing mingw32 build with Dan's patch integrated. Also, gtest was completely removed :)

Will post the results as soon as my tests will over

[tlattner]

I've merged 64316 into the branch. I've also merged in Mike's changes for googletest and also the changes Julien made.

I believe that is all that is needed for this bug.

I'll need a final verification and then it can be closed.

[llvmbot]

Ok, I am re-building everything right now. Thanks.

[llvmbot]

Yep, the build went through. Thanks !