[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#972436: RM: mapsembler2 [armel armhf i386 hurd-i386 kfreebsd-i386] -- ANAIS; package broken on 32bits archs and non-maintained upstream



Package: ftp.debian.org
Severity: normal

Greetings,

I spent some time on mapsembler2 to put it into adequate shape,
so that it might make it to Testing.  This involved some
investigatons of autopkgtest regressions on 32 bits
architectures, at least armhf and i386, as reported by testing
excuses migrations[0].

[0] https://tracker.debian.org/pkg/mapsembler2

The initial segmentation fault encountered by the test allowed
me to locate a missing check of memory allocation, and change
some default program parameters to values more suitable for
32 bits architectures, when running on such arch[1].

[1] https://salsa.debian.org/med-team/mapsembler2/-/commit/73d9dae0c0c876a415ba804e83602bb984a20ac0

I then had to track a second segmentation fault down to a highly
optimized hashing function, which looks compatible with all
combinations of 32, 64 bits, little and big endian.  However one
of the steps of this hashing function is triggering the fault,
and I'm not comfortable enough with the program to identify
whether it is the value of the hash key which is at fault, or
the logic of the hashing function in itself.  Looking at the
change log[2] of the program, I even think my change could have
undone a fix for that crash, or even for another crash:

> mapsembler2_2.2.4: fixes k-mer counting problems. Use by default 4GB memory

[2] https://colibread.inria.fr/software/mapsembler2/

There is always the possibility that a simple knob has changed
at some point, such as an architecture specific code path, but
if so, I failed to identify it.  Bringing the issue to upstream
is not an option here unfortunately, as the program is not
maintained since 2014[2].

Due to the above reasons, I think mapsembler2 is probably
not suitable for 32 bits variants of Debian unfortunately.

Kind Regards,
-- 
Étienne Mollier <etienne.mollier@mailoo.org>
Old rsa/3072: 5ab1 4edf 63bb ccff 8b54  2fa9 59da 56fe fff3 882d
New rsa/4096: 8f91 b227 c7d6 f2b1 948c  8236 793c f67e 8f0d 11da
Sent from /dev/pts/6, please excuse my verbosity.

Attachment: signature.asc
Description: PGP signature


Reply to: