armv7 builds leaking into armel
I just filed a couple of bugs about packages whose build systems are
building for armv7 on armel[1][2].
While doing so I also noticed that openmpi also seems to have similar
problems but that case is more complex and I don't have the time/energy
to investigate it right now. We have a patch in raspbian[3] but it needs
some work to figure out what should be done in Debian.
The two issues mentioned above were using armv7 cflags unconditionally
on all arm architctures but i've also come across ones in the past that
use armv7 cflags only if the system they are being built on is armv7
[4]. Since Debian recently switched to armv7 autobuilders for armel it
is likely that Debian has latent issues of this type.
For the most part armv7 binaries can be detected using readelf. It does
suffer from some false positives (runtime cpu detection etc) and false
negatives (jit compilers etc) but it should be sufficient to discover
most issues.
In Raspbian we run all packages we build through a check that unpacks
them and runs readelf[5]. I try to push issues I detect this way back up
to Debian but I sometimes forget and there are some less clear cut cases
that I don't have the time to triage (for example with neon related ones
there may or may not be runtime detection, in raspbian I just tend to
disable neon completely but that may not be appropriate for Debian). I
don't think Debian armel should be relying on Raspbian to find these issues.
Anyone here who cares about armel fancy putting something together to
check the armel archive for armv7 binaries? or are we just going to
leave them arround until some user gets the illegal instruction error?
[1] http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=770782
[2] http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=770793
[3] http://debdiffs.raspbian.org/main/o/openmpi/openmpi_1.6.5-8+rpi1.debdiff
[4] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=719355
[5] https://lists.debian.org/debian-arm/2013/05/msg00154.html
Reply to: