[Debian-med-packaging] Please verify whether pyhml has done the expected number of bootstraps which is not guaranteed when using MPI

[Andreas Tille]

Hi Manolo,

On Tue, Jan 31, 2017 at 06:34:45PM +0100, Manolo Gouy wrote:
> Good suggestion. I see the point.
> 
> You should find with seaview_4.6.1.3.tar.gz a source code that will
> recompute the # of bootstrap replicates after phyml has run,
> so that should accept to find a few more replicates than asked.

Thanks a lot for your very quick response (I wished each upstream author
would be that fast ;-)).  Usually I would simple upload your new version
as is but currently we are in freeze for a new Debian release and there
are no new upstream versions allowed.  I can only fix Release Critical
bugs.  I consider this issue critical and would like to fix it via a
patch against version 4.6.1.2.

I tried to extract the code snippet that fixes the said issue and I
think it is just this:

  https://anonscm.debian.org/cgit/debian-med/seaview.git/tree/debian/patches/evaluate_phyml_bootstrap_number.patch

but it would be nice if you could review this and check whether I missed
something.
 
> Thanks a lot for your care about seaview.

You are welcome and thanks for the very good cooperation

      Andreas. 
 
> > Le 31 janv. 2017 à 16:45, Andreas Tille <tille at debian.org> a écrit :
> > 
> > Hi Manolo,
> > 
> > as you know from my last mail I created a wrapper inside the Debian
> > package of phyml which uses MPI if available.  This resulted in quite
> > some gain of speed.  However, we realised that it also might cause a
> > problem since phyml "decides" itself how many bootstrap steps it takes
> > depending from the amount of processors can be used.  I have described
> > the issue in Debian bug #853738[1] and fixed the bug in a way which
> > allows the user to call seaview like:
> > 
> >   PHYMLCPUS=8 seaview *.fna
> > 
> > when intending to do for instance 1000 bootstrap steps.  This works
> > since 1000%8=0.  However, for other numbers of CPUs for instance when
> > calling
> > 
> >   PHYMLCPUS=16 seaview *.fna
> > 
> > phyml would do
> > 
> >   (1000 / 16 + 1) * 16 = 1008
> > 
> > bootstrap processes.  From phyml perspective this does not matter but
> > the effect in seaview is that if 1000 bootstraps are considered 100%
> > some values might end up beeing 101% which is nonsense.
> > 
> > So I wonder whether you could please consider parsing the phyml output
> > for changed numbers of bootstraps to adjust the base of the calculations
> > that are based on the effective number of bootstraps.
> > 
> > Kind regards
> > 
> >      Andreas.
> > 
> > [1] https://bugs.debian.org/853738
> > 
> > -- 
> > http://fam-tille.de
> 
> 

-- 
http://fam-tille.de