Re: Bug#871514: clamav: FTBFS on mips64el
On 2017-08-08 20:41, Sebastian Andrzej Siewior wrote:
> On 2017-08-08 20:34:37 [+0200], To submit@bugs.debian.org wrote:
> …
> > returned (the important part):
> > |LibClamAV debug: parseEmailBody() rc 1 infect 0
> > |LibClamAV debug: parseEmailBody() returning 3
> …
> > The exp build passed with gcc-6_6.4.0-1 [0]. Is there an easy way to
> > downgrade the compiler on eller/porterbox? Or could a porter double
> > check this please?
>
> on eller in a buster chroot:
> |LibClamAV debug: parseEmailBody() rc 1 infect 0
> |LibClamAV debug: parseEmailBody() returning 1
>
> further suggestions?
I got a quick look. It's indeed a regression introduced by GCC 7. It can
be workarounded by building the file with -O0, but already appears with
-O1 optimization.
I got a quick look with gdb and it seems that loading either the rc
(enum) or infect (bool variable to test it against 0, the load is done
with the ld instruction instead of the lw instruction. It means garbage
from another local variable is loaded into the high 32 bits, which
causes the comparison against 0 to be false instead of true.
Aurelien
--
Aurelien Jarno GPG: 4096R/1DDD8C9B
aurelien@aurel32.net http://www.aurel32.net
Reply to: