[debhelper-devel] Bug#895196: debhelper: The failing command on build hangs is no longer in the log

[Niels Thykier]

On Sun, 08 Apr 2018 13:01:15 +0300 Adrian Bunk <bunk at debian.org> wrote:
> Package: debhelper
> Version: 11.1.5
> Severity: important
> 

Hi,

CC'ing the submitter of #768791.  Quoting in full for their convenience.

> https://buildd.debian.org/status/logs.php?pkg=prodigal&arch=armhf
> 
> The root cause of this FTBFS is that building training.c "hangs"
> (6.5 hours build time exceed the 150m timeout, see #894473).
> 
> This bug is about debhelper (likely due to the #768791 change)
> giving much worse logs for that on the buildds:
> 
> debhelper <= 11.1.4 (locally reproduced with 11.1.4):
> 
> ...
> gcc -g -O2 -fstack-protector-strong -Wformat -Werror=format-security -pedantic -Wall -O3 -c -o training.o training.c
> debian/rules:12: recipe for target 'build-arch' failed
> make: *** [build-arch] Terminated
> Makefile:40: recipe for target 'training.o' failed
> make[1]: *** [training.o] Terminated
> Build killed with signal TERM after 150 minutes of inactivity
> 
> 
> debhelper >= 11.1.5 (locally reproduced with 11.1.5, 11.1.6, 11.2):
> 
> ...
> make[1]: Entering directory '/<<PKGBUILDDIR>>'
> gcc -g -O2 -fdebug-prefix-map=/<<PKGBUILDDIR>>=. -fstack-protector-strong -Wformat -Werror=format-security -pedantic -Wall -O3 -c -o node.o node.c
> make[1]: Leaving directory '/<<PKGBUILDDIR>>'
> E: Build killed with signal TERM after 150 minutes of inactivity
> 
> 
> Note how using older debhelper results in exact information
> where the problem is, while with newer debhelper the hanging
> command is not even in the log.
> 
> 
> This can be reproduced locally on amd64:
>   dpkg-buildpackage -B  2>&1 | tee ../LOG.prodigal
> 
> Check whether the training.c compile is already printed before
> the build "hangs" for a few minutes.
> 
> 
> Another example:
>   https://buildd.debian.org/status/logs.php?pkg=mame&arch=amd64
> 
> Despite several attempts noone has been able to reproduce why this
> sometimes FTBFS, and it is not helpful that whatever hangs in the
> build is likely not in the log.
> 
> 
> If this is due to the "make -O" change as I suspect,
> it should IMHO be reverted if this cannot be fixed.
> 
> 

I have very little experience with make -O[...], so I am hoping Goswin
might have some ideas for dealing with this problem (beyond reverting).

Thanks,
~Niels