[Debian-med-packaging] backporting libssw - autopkgtest fails on jessie

[Sascha Steinbiss]

Hi Afif,

thanks for your quick reply.

>> I took the liberty of moving the JDK dependencies back into their
>> original place; please see my mail to Andreas and the list for my
>> reasons. Supported by Lintian’s output, the libssw-java binary
>> package shouldn’t depend on a JRE [1] as there is nothing to ‘run’,
>> and I would assume the also applies to JDKs (as they also contain a
>> JRE). The JDK is indeed only needed in the test case to build a
>> test driver against the libssw-java library.
> 
> Yes, sorry about that. While I was experimenting, I had issues because
> default-jdk-headless isn't available on jessie. That's when I started
> getting confused about why java wasn't a dependency of a java library.



> I see that, while you were undoing the damage I did, you changed
> debian/tests/control in master to be default-jdk when you had
> default-jdk-headless before. I think headless is fine for
> Unstable/Stretch since it's a valid package there.

Sure. I'll adjust that in both unstable and the backport.

>> As far as the run-example autopkgtest is concerned, I have
>> identified some invalid memory accesses in upstream’s code (and
>> also confirmed that in a clean source checkout from upstream’s
>> GitHub repo). This apparently leads to binary data in the reference
>> output, failing the test. What strikes me most is that, looking at
>> your diff, your Jessie test output looks much more correct than the
>> reference output that my sid binary produced. It seems like the new
>> build tries to avoid indels as much as possible; the CIGAR string
>> that I get when asking for SAM output also looks pretty dodgy! Can
>> you reproduce this ‘good’ output on sid? I can’t — on sid I always
>> get garbled results. There might be some fragility involved that’s
>> not obvious to me… I have put in a ticket [2] with upstream, maybe
>> they can help.
> 
> Very interesting. I haven't run it on Sid, though. I did notice that
> the output was very strange. Thanks for reporting the issue upstream.

I managed to fix the problem; apparently the binary built with sid's
default version of GCC produces corrupted CIGAR alignments internally
(all edit operations are Ms). Clang and the default version of GCC used
in Jessie (4.9) produce 'correct' output. That's how the failing tests
came to be in the first place: my reference output (created on sid,
hence broken) did not match the correct alignment output created by the
version built on Jessie, failing there but succeeding on sid, but with
weird results.
I will patch the code to work correctly on sid as well and adjust the
reference output until upstream fixes the problem. I have updated my
upstream issue on GitHub to include my fix.
There might be a slight performance impact as the fix involves reducing
optimization and removing inlining, but it's better than broken output IMHO.

I'll upload the fix when I find some time here. Should I upload the
backport as well?

Cheers
Sascha

> 
> _______________________________________________
> Debian-med-packaging mailing list
> Debian-med-packaging at lists.alioth.debian.org
> http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/debian-med-packaging
> 


-- 
 The Wellcome Trust Sanger Institute is operated by Genome Research 
 Limited, a charity registered in England with number 1021457 and a 
 company registered in England with number 2742969, whose registered 
 office is 215 Euston Road, London, NW1 2BE.