11.04.2010 20:40, Joel E. Denny wrote: > On Sun, 11 Apr 2010, Bruno Haible wrote: > >> The code in bison-2.4.2/src/output.c looks fine. > > Thanks for confirming. > >> Can you help reproducing or debugging it, with just 2 executables: bison >> and m4? (I.e. give a "how to reproduce" that does not involve bison's >> testsuite? I'm not inclined to look into a 3 MB large machine generated >> shell script.) > > Because I do not have access to the affected platform, I can only explain > the actions that the failing test group performs. Petr, can you please > confirm that the following instructions produce the problem for you? > Recall that there may be a race condition, so run bison many times if it > does not produce the problem at first. Also, please remind us exactly > which platform exhibits the problem.
I saw the problem on x86_64 and x86 and didn't check any other platform. The platform is "fedora build system". This is RHEL 5 with mock chroot where Fedora rawhide is installed. Unfortunately I was unsuccessful reproducing it outside the actual build system, even mimicking these conditions down to kernel version. I don't know whether some part of userspace of host can influence what happens in mock, but that's the only part of the whole setup that I didn't try to mach 1:1 to build system. The instructions that you gave indeed do reproduce the problem. The build log, should you be interested, is here: http://koji.fedoraproject.org/koji/getfile?taskID=2110920&name=build.log Here's hoping that it's accessible without some sort of registration. The build log will be wiped out automatically in near future, but it should hopefully stay up a week or so. Relevant quote: > + echo =xxx================================================================= > + ./tests/bison /builddir/build/SOURCES/input.y > /builddir/build/SOURCES/input.y: fatal error: too many arguments for @output > directive in skeleton > + : > -xxx----------------------------------------------------------------- > /usr/bin/m4:/builddir/build/SOURCES/./skel.c:175: ERROR: copying inserted > file: Broken pipe > + echo -xxx----------------------------------------------------------------- > /usr/bin/m4: write error > + make check > make check-recursive PM P.S. I seriously wonder whether it's worth it to hunt this esoteric bug. (Especially given that it was confirmed that the code is written correctly.) As I said, it's essentially impossible to reproduce outside that very specific setup, meaning user impact low, and it's not like this is a major breaker.
signature.asc
Description: OpenPGP digital signature
