[Debian-med-packaging] Psortb has errors in Perl testsuite

[Carnë Draug]

On 20 April 2017 at 07:20, Andreas Tille <andreas at fam-tille.de> wrote:
> Hi Carnė,
>
> On Wed, Apr 19, 2017 at 10:12:53PM +0100, Carnė Draug wrote:
>> >> [4] https://rt.cpan.org/Public/Bug/Display.html?id=79106
>> >> [5] https://rt.cpan.org/Public/Dist/Display.html?Name=Algorithm-SVM
>> >> [6] https://rt.cpan.org/Public/Bug/Display.html?id=79754
>> >
>> > [7]
>> > https://anonscm.debian.org/cgit/debian-med/psortb.git/tree/debian/patches/fix_build_Algorithm-SVM.patch
>>
>> To help with this, I have packaged Algorithm::SVM [8], and unbundled
>> libsvm from it with the patches from Fedora.  The changes seem to make
>> autopkgtest happy in sid-amd64 (Algorithm::SVM has a test unit).
>
> I made some cosmetic changes - please pull.  If you add an ITP bug I'd
> sponsor immediately.  Thanks for working on this.
>
>> Note that even if you decide against unbundling Algorithm::SVM out of
>> psortb, psortb is still dependent on it.  The module
>> Bio::Tools::SVMLoc, which is part of psortb, uses Algorithm::SVM and
>> not the bundled version in Bio::Tools::PSort::SVMLoc (I guess the Loc
>> means local to the psortb?).
>
> I'm afraid I do not understand what you want to tell me here.  I
> compared the code of Algorithm::SVM (from your packaging) with
> Bio::Tools::SVMLoc from psortb source and there is quite some diff and a
> separate name space.  I do not intend to change upstream code to a large
> amount to sort this out without even a minor understanding of the
> rationale behind this fork of Algorithm::SVM.
>
> If you think I would need Algorithm::SVM as an extra Build-Depends
> adding libalgorithm-svm-perl did not helped to change anything.
>
>> Can you check if my attempt at packaging Algorithm::SVM works for you
>> or if you still have any issue?
>
> While I can confirm that the packaging of Algorithm::SVM looks fine I
> totally fail to see how this would be helpful for psort.  I'd be happy
> if you simply commit some change to psortb (or send me some git
> format-patch I could apply) to make me better understand in code what
> you want to suggest.
>
> Thanks in any case for your help
>
>       Andreas.
>
>> [8] https://anonscm.debian.org/cgit/pkg-perl/packages/libalgorithm-svm-perl.git/
>
> --
> http://fam-tille.de

You are right.  I now looked at all the files (instead of just a few)
between Algorithm::SVM and Bio::Tools::PSort::SVMLoc, and while some
were pretty much the same I see now others have changed a lot.

But despite having Bio::Tools::PSort::SVMLoc, psortb also uses the
'wild type' Algorithm::SVM:

    $ grep -r 'use Algorithm::SVM' psortb
    psortb/lib/Bio/Tools/SVMLoc.pm:use Algorithm::SVM::DataSet;
    psortb/lib/Bio/Tools/SVMLoc.pm:use Algorithm::SVM;

which means that after fixing whatever issues there are in psortb, it
will still need to be dependent on libalgorithm-svm-perl that I just
packaged.

---

I noticed that there is no `bio-tools-psort-svmloc/bindings.cpp` file,
there is only `bio-tools-psort-svmloc/bindings.h`.  I couldn't find
the definitions of the DataSet and SVM classes declared in bindings.h.
Maybe that is part of the issue?

----

While Algorithm::SVM and Bio::Tools::PSort::SVMLoc have diverged quite
a bit, the changes don't seem major to me.  Since psortb already
requires Algorithm::SVM, since psortb is an active project and
Algorithm::SVM is not, and since psortb developers will understand the
divergence between Algorithm::SVM and Bio::Tools::PSort::SVMLoc (while
it is still small), won't the psortb developers be interested in
sorting the divergence between the two things?

And if SVMLoc really is needed as a separate, it makes more sense to
me that it would made use of Algorithm::SVM instead of their current
approach of copying the code and adding a few more methods.

Carnë