b-i-b status on Darwin

b-i-b status on Darwin

[Geoff Keating]

To help Zack out (or at least to make sure that the tree doesn't
suddenly break on Darwin), I tried bootstrapping the b-i-b on Darwin.

The first problem I hit appears to be a merge problem,
unwind-dw2-fde-darwin.c doesn't exist on the branch.  I worked around
it by copying the mainline version.  Zack, can you check that no other
files are missing and that no patches got lost?  I'm most worried
about losing testcases.

I restarted the build and tried again, and got:

gcc -c   -g -no-cpp-precomp -DHAVE_DESIGNATED_INITIALIZERS=0 -DIN_GCC
-W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes
-Wtraditional -pedantic -Wno-long-long -fno-common  -DHAVE_CONFIG_H
-I.
-I.
-I/Network/Servers/cauchy/homes/thorin/gkeating/co/basic-improvement/gcc/gcc
-I/Network/Servers/cauchy/homes/thorin/gkeating/co/basic-improvement/gcc/gcc/.
-I/Network/Servers/cauchy/homes/thorin/gkeating/co/basic-improvement/gcc/gcc/config
-I/Network/Servers/cauchy/homes/thorin/gkeating/co/basic-improvement/gcc/gcc/../include
/Network/Servers/cauchy/homes/thorin/gkeating/co/basic-improvement/gcc/gcc/config/darwin-c.c
In file included from
/Network/Servers/cauchy/homes/thorin/gkeating/co/basic-improvement/gcc/gcc/config/darwin-c.c:26:
tm.h:8:29: insn-constants.h: No such file or directory
tm.h:9:25: insn-flags.h: No such file or directory

(sorry it's word-wrapped).  I expect this is some kind of Makefile
dependency problem.  I was building using -j2.

So, I tried again without -j.  This got me into stage2; I'll report
back on the results.

-- 
- Geoffrey Keating <geoffk@geoffk.org>

Re: b-i-b status on Darwin

[Daniel Berlin]

On Saturday, December 14, 2002, at 01:09  AM, Geoff Keating wrote:

>
> To help Zack out (or at least to make sure that the tree doesn't
> suddenly break on Darwin), I tried bootstrapping the b-i-b on Darwin.
>
It won't work, it's missing unwind-dw2-fde-darwin.c. I know because it 
does the same thing on the tree-ssa branch since the last bib merge.

> The first problem I hit appears to be a merge problem,
> unwind-dw2-fde-darwin.c doesn't exist on the branch.  I worked around
> it by copying the mainline version.
If you look, it's *tagged* for a bib merge, it looks like it was never 
committed or something.

Re: b-i-b status on Darwin

[Zack Weinberg]

Geoff Keating <geoffk@geoffk.org> writes:

> To help Zack out (or at least to make sure that the tree doesn't
> suddenly break on Darwin), I tried bootstrapping the b-i-b on Darwin.
>
> The first problem I hit appears to be a merge problem,
> unwind-dw2-fde-darwin.c doesn't exist on the branch.  I worked around
> it by copying the mainline version.  Zack, can you check that no other
> files are missing and that no patches got lost?  I'm most worried
> about losing testcases.

I don't know how this happened.  I generated recursive directory
listings in my 3.3 and 3.4 trees and compared them, this is what I got:

files that should be added to 3.4 -- corrected in my working copy:

gcc/ChangeLog.7
gcc/unwind-dw2-fde-darwin.c

files that were deliberately removed from 3.4:

gcc/config/alpha/vxworks.h
gcc/config/arm/vxarm.h
gcc/config/i386/vxi386.h
gcc/config/i960/t-vxworks960
gcc/config/i960/vx960-coff.h
gcc/config/i960/vx960.h
gcc/config/m68k/t-vxworks68
gcc/config/m68k/vxm68k.h
gcc/config/mips/vxworks.h
gcc/config/rs6000/vxppc.h
gcc/config/sparc/t-vxsparc
gcc/config/sparc/t-vxsparc64
gcc/config/sparc/vxsim.h
gcc/config/sparc/vxsparc64.h
gcc/config/sparc/vxsparc.h
gcc/config/vax/vms.h
gcc/config/vax/xm-vms.h

files only in 3.4 -- I believe all of these are new on the branch, not
removed on the trunk and the deletion didn't propagate:

config/accross.m4
contrib/gthr_supp_vxw_5x.c
gcc/config/rs6000/t-vxworks
gcc/config/rs6000/vxworks.h
gcc/config/sh/libgcc-glibc.ver
gcc/config/t-libunwind
gcc/config/t-vxworks
gcc/config/vxlib.c
gcc/config/vxworks.h
gcc/coretypes.h
gcc/cp/cxxfilt.c
gcc/gcov-dump.c
gcc/gcov-iov.c
gcc/testsuite/gcc.c-torture/execute/shiftopt-1.c
gcc/testsuite/gcc.dg/bf-ms-layout.c
gcc/testsuite/gcc.dg/bf-no-ms-layout.c
gcc/testsuite/gcc.dg/builtins-2.c
gcc/testsuite/gcc.dg/builtins-3.c
gcc/testsuite/gcc.dg/i386-cmov1.c
gcc/testsuite/gcc.dg/i386-cmov2.c
gcc/testsuite/gcc.dg/i386-cmov3.c
gcc/testsuite/gcc.dg/i386-ssefp-1.c
gcc/testsuite/gcc.dg/i386-ssefp-2.c
gcc/testsuite/gcc.dg/sibcall-5.c
gcc/testsuite/gcc.dg/sibcall-6.c
gcc/testsuite/g++.dg/inherit/covariant2.C
gcc/testsuite/g++.dg/inherit/covariant3.C
gcc/testsuite/g++.dg/inherit/covariant4.C
gcc/unwind-libunwind.c

See my next message for the merge plan.

zw

Re: b-i-b status on Darwin

[Geoff Keating]

OK, so the bootstrap completed after working around those two
problems.  Then, while building libstdc++-v3, I got:

/build/basic-improvement/gcc/xgcc -B/build/basic-improvement/gcc/ -B/usr/local/powerpc-apple-darwin6.0/bin/ -B/usr/local/powerpc-apple-darwin6.0/lib/ -isystem /usr/local/powerpc-apple-darwin6.0/include -DHAVE_CONFIG_H -I. -I/Network/Servers/cauchy/homes/thorin/gkeating/co/basic-improvement/gcc/libstdc++-v3/libmath -I.. -O2 -g -O2 -c /Network/Servers/cauchy/homes/thorin/gkeating/co/basic-improvement/gcc/libstdc++-v3/libmath/stubs.c -o stubs.o
/var/tmp//cc45Bz9h.s:535:FATAL:Symbol L_sqrtf$stub already defined.
make[4]: *** [stubs.lo] Error 1
make[3]: *** [all-recursive] Error 1
make[2]: *** [all-recursive-am] Error 2
make[1]: *** [all-target-libstdc++-v3] Error 2
make: *** [bootstrap] Error 2

It looks like the darwin stub-output logic is expecting DECLs to be
shared, but this isn't happening for the case of a builtin which is
later defined, perhaps caused by the changes for better folding of
sqrt operations.  This can be fixed, but not while everything is
frozen.  Once it's fixed, I can try to run the C++ testsuite...

Not to be denied, I at least ran the C testsuite.  The b-i-b showed
one fewer failure, which was good, but 25 fewer passes, which could be
bad.

-- 
- Geoffrey Keating <geoffk@geoffk.org>

Re: b-i-b status on Darwin

[Andreas Tobler]

Geoff Keating wrote:
> OK, so the bootstrap completed after working around those two
> problems.  Then, while building libstdc++-v3, I got:
> 
> /build/basic-improvement/gcc/xgcc -B/build/basic-improvement/gcc/ -B/usr/local/powerpc-apple-darwin6.0/bin/ -B/usr/local/powerpc-apple-darwin6.0/lib/ -isystem /usr/local/powerpc-apple-darwin6.0/include -DHAVE_CONFIG_H -I. -I/Network/Servers/cauchy/homes/thorin/gkeating/co/basic-improvement/gcc/libstdc++-v3/libmath -I.. -O2 -g -O2 -c /Network/Servers/cauchy/homes/thorin/gkeating/co/basic-improvement/gcc/libstdc++-v3/libmath/stubs.c -o stubs.o
> /var/tmp//cc45Bz9h.s:535:FATAL:Symbol L_sqrtf$stub already defined.
> make[4]: *** [stubs.lo] Error 1
> make[3]: *** [all-recursive] Error 1
> make[2]: *** [all-recursive-am] Error 2
> make[1]: *** [all-target-libstdc++-v3] Error 2
> make: *** [bootstrap] Error 2

I see the same on my darwin6.2 box.

> Not to be denied, I at least ran the C testsuite.  The b-i-b showed
> one fewer failure, which was good, but 25 fewer passes, which could be
> bad.

Which dejagnu, expect tcl tripplet do you use? The one from the apple 
gcc3? Or something else?


Here my output for C, it would be about 20 to 30 failures if the test 
environement would be ok.

Andreas

                 === gcc Summary ===

# of expected passes            19671
# of unexpected failures        574
# of unexpected successes       6
# of expected failures          56
# of unresolved testcases       9
# of unsupported tests          150

Re: b-i-b status on Darwin

[Geoff Keating]

> Date: Sat, 14 Dec 2002 15:44:29 +0100
> From: Andreas Tobler <toa@pop.agri.ch>

> Which dejagnu, expect tcl tripplet do you use? The one from the apple 
> gcc3? Or something else?
> 
> 
> Here my output for C, it would be about 20 to 30 failures if the test 
> environement would be ok.

I use the FSF dejagnu, but have a fix for the kernel bug that causes
the spurious failures.

-- 
- Geoffrey Keating <geoffk@geoffk.org>

Re: b-i-b status on Darwin

[Mike Stump]

On Saturday, December 14, 2002, at 06:44 AM, Andreas Tobler wrote:
> Here my output for C, it would be about 20 to 30 failures if the test 
> environement would be ok.

>                 === gcc Summary ===
>
> # of expected passes            19671
> # of unexpected failures        574
> # of unexpected successes       6
> # of expected failures          56
> # of unresolved testcases       9
> # of unsupported tests          150

For comparison:

		=== gcc Summary ===

# of expected passes		20230
# of unexpected failures	26
# of expected failures		62
# of unresolved testcases	9
# of unsupported tests		141
/Volumes/mrs3/net/gcc-darwinO2/gcc/xgcc version 3.3 20021206 
(experimental)

from mainline on a system with a patched kernel (;-() to avoid the 
truncated output bug.

Re: b-i-b status on Darwin

[David Edelsohn]

>>>>> Geoff Keating writes:

Geoff> In file included from
Geoff> /Network/Servers/cauchy/homes/thorin/gkeating/co/basic-improvement/gcc/gcc/config/darwin-c.c:26:
Geoff> tm.h:8:29: insn-constants.h: No such file or directory
Geoff> tm.h:9:25: insn-flags.h: No such file or directory

	This is the same type of dependency problem that occurred for
rs6000-c.c .  The darwin-c.c dependency needs coretypes.h and $(TM_H) added.
See gcc/config/rs6000/t-rs6000-c-rule.

	Devang also mentioned the Darwin stub error.

	The 32x64 cross limitation prevents GCC for PPC64 Linux, GCC for
x86_64, and GCC for AIX as a 32-bit apps from building.

David