Dear debian-gcc,
When building current stable's gcc-4.9 I encountered bug 769218 and I
was also able to get a successful build on the same machine.
System:
Dual Xeon L5640 (24 threads) with 32 GB RAM, and all disk-based
filesystems residing on a magnetic hdd. The host system contains debian
jessie. The pbuilder base.tgz contains jessie.
On this system, so far I have observed:
Attempt 1
working directory on a disk-based file system
Build method: apt-get build-dep gcc-4.9, apt-get source gcc-4.9, debuild
-us -uc -b
Result: Successful build
Attempt 2
Build method: apt-get source gcc-4.9, pbuilder build *.dsc
Result: Successful build
Attempt 3
/var/cache/pbuilder/build mounted on 16 GB tmpfs, APTCACHEHARDLINK=no in
.pbuilderrc
Build method: apt-get source gcc-4.9, pbuilder build *.dsc
Result: Fails with exactly the CRC error reported in #769218.
Attempt 4
/dev/shm on a 16 GB tmpfs, /tmp on a 16 GB tmpfs
Build method: cd /dev/shm/gcc, apt-get build-dep gcc-4.9, apt-get source
gcc-4.9, debuild -us -uc -b
Result: Fails with a different CRC error at the same point.
Given that the failures happen specifically when the build can use 24
threads and (almost) everything is in RAM, I suspect there is some race
condition that does not show up when the build or its temporary files
have to go through a relatively slow disk. Another explanation would be
that the tmpfs is too small, but the successful build does not seem to
need that much disk space either.
If the debian-gcc people want to find out more, I would be willing to do
additional tests or otherwise help to find the cause. However, I can
imagine that it would not be worth the effort anymore for gcc-4.9 at
this point in time. Please give some indication of how serious you
consider the situation that gcc in debian stable seems to have a FTBFS
bug that occurs under specific (I think mostly build-host-related)
circumstances.
Best regards,
Frans van der Have.
