Bug 303004 – dev-db/mysql-5.0.90-r1 fails build on pars0sym.lo

• Product: Gentoo Linux
• Status: RESOLVED NEEDINFO

[Description Martin von Gagern 2010-01-31 17:32:21 UTC]

dev-db/mysql-5.0.90-r1 just failed to build for me. Strange thing is, I got an error message from make, but apparently none from the compiler. Will attach full log, but the interesting part is this:

/bin/sh ./libtool --tag=CC   --mode=compile x86_64-pc-linux-gnu-gcc -DHAVE_CONFIG_H -I. -I./include -I./../include    -DDBUG_OFF  -DDBUG_OFF -march=amdfam10 -O2 -ggdb -pipe -DHAVE_ERRNO_AS_DEFINE=1 -fno-strict-aliasing    -DDEBUG_OFF -DUNIV_LINUX -MT pars0sym.lo -MD -MP -MF .deps/pars0sym.Tpo -c -o pars0sym.lo `test -f 'pars/pars0sym.c' || echo './'`pars/pars0sym.c
libtool: compile:  x86_64-pc-linux-gnu-gcc -DHAVE_CONFIG_H -I. -I./include -I./../include -DDBUG_OFF -DDBUG_OFF -march=amdfam10 -O2 -ggdb -pipe -DHAVE_ERRNO_AS_DEFINE=1 -fno-strict-aliasing -DDEBUG_OFF -DUNIV_LINUX -MT pars0sym.lo -MD -MP -MF .deps/pars0sym.Tpo -c pars/pars0sym.c  -fPIC -DPIC -o .libs/pars0sym.o
mv -f .deps/page0cur.Tpo .deps/page0cur.Plo
/bin/sh ./libtool --tag=CC   --mode=compile x86_64-pc-linux-gnu-gcc -DHAVE_CONFIG_H -I. -I./include -I./../include    -DDBUG_OFF  -DDBUG_OFF -march=amdfam10 -O2 -ggdb -pipe -DHAVE_ERRNO_AS_DEFINE=1 -fno-strict-aliasing    -DDEBUG_OFF -DUNIV_LINUX -MT que0que.lo -MD -MP -MF .deps/que0que.Tpo -c -o que0que.lo `test -f 'que/que0que.c' || echo './'`que/que0que.c
libtool: compile:  x86_64-pc-linux-gnu-gcc -DHAVE_CONFIG_H -I. -I./include -I./../include -DDBUG_OFF -DDBUG_OFF -march=amdfam10 -O2 -ggdb -pipe -DHAVE_ERRNO_AS_DEFINE=1 -fno-strict-aliasing -DDEBUG_OFF -DUNIV_LINUX -MT que0que.lo -MD -MP -MF .deps/que0que.Tpo -c que/que0que.c  -fPIC -DPIC -o .libs/que0que.o
libtool: compile:  x86_64-pc-linux-gnu-gcc -DHAVE_CONFIG_H -I. -I./include -I./../include -DDBUG_OFF -DDBUG_OFF -march=amdfam10 -O2 -ggdb -pipe -DHAVE_ERRNO_AS_DEFINE=1 -fno-strict-aliasing -DDEBUG_OFF -DUNIV_LINUX -MT pars0sym.lo -MD -MP -MF .deps/pars0sym.Tpo -c pars/pars0sym.c -o pars0sym.o >/dev/null 2>&1
libtool: compile:  x86_64-pc-linux-gnu-gcc -DHAVE_CONFIG_H -I. -I./include -I./../include -DDBUG_OFF -DDBUG_OFF -march=amdfam10 -O2 -ggdb -pipe -DHAVE_ERRNO_AS_DEFINE=1 -fno-strict-aliasing -DDEBUG_OFF -DUNIV_LINUX -MT pars0pars.lo -MD -MP -MF .deps/pars0pars.Tpo -c pars/pars0pars.c -o pars0pars.o >/dev/null 2>&1
mv -f .deps/pars0opt.Tpo .deps/pars0opt.Plo
/bin/sh ./libtool --tag=CC   --mode=compile x86_64-pc-linux-gnu-gcc -DHAVE_CONFIG_H -I. -I./include -I./../include    -DDBUG_OFF  -DDBUG_OFF -march=amdfam10 -O2 -ggdb -pipe -DHAVE_ERRNO_AS_DEFINE=1 -fno-strict-aliasing    -DDEBUG_OFF -DUNIV_LINUX -MT read0read.lo -MD -MP -MF .deps/read0read.Tpo -c -o read0read.lo `test -f 'read/read0read.c' || echo './'`read/read0read.c
make[3]: *** [pars0sym.lo] Error 1

It seems to be a parallel build issue: repeating the build with MAKEOPTS=-j1 succeeded. Haven't looked at the Makefiles (yet).

[Comment 1 Martin von Gagern 2010-01-31 17:33:56 UTC]

Created attachment 218002 [details]
build log

[Comment 2 Martin von Gagern 2010-01-31 17:35:59 UTC]

Created attachment 218003 [details]
emerge --info

[Comment 3 Robin Johnson 2010-02-02 06:59:30 UTC]

It's not parallel specific in any way.
The file that failed to build there wasn't used in anything else or in any parallel rule conflict.

AND I've been running all of the new stuff w/ -j16 on an 8-core box, else we'd still be waiting.

You'll need to dig harder as to why it failed.

[Comment 4 Martin von Gagern 2010-02-02 08:59:37 UTC]

(In reply to comment #3)
> You'll need to dig harder as to why it failed.

Had a look as well. My best bet is that libtool is to blame for there being no error message: the libtool script compiled the same source repeatedly, and by default seems to suppress all output from the second run.

As to why the second run failed, but only in my first build: iirc I was compiling multiple packages in parallel by passing -j to emerge, so it might have been some kind of interaction with another emerge, probably disk space exhausted or something along those lines. Far from sure, though.

I guess I'd wish for full output on every compiler run. Unfortunately I see no suitable variable in these makefiles which would allow me to pass -no-suppress to all invocations of libtool --mode=compile without passing it to other modes as well, which would probably reject it. If there was a way, I'd add a suitable var to my MAKEOPTS.

[Comment 5 Robin Johnson 2010-03-09 20:28:48 UTC]

I still can't reproduce this :-(