Re: Should lambda-align v2 replace v1 or be its own Debian package?
On 14.01.19 17:39, Michael Crusoe wrote:
> [Keeping Hannes on the CC in case he isn't subscribed]
>
> Thanks for the responses, all.
>
Thanks for keeping me in the loop, I am not subscribed to debian-med.
> În lun., 14 ian. 2019 la 17:48, Sascha Steinbiss <satta@debian.org
> <mailto:satta@debian.org>> a scris:
>
> Hi,
>
> just my 2 cents. I introduced Lambda into Debian some time ago just out
> of interest but looking at the uploads, Michael has done more work
> recently...
> So here are just a few comments from me.
> [...]
> >> However, there are no references from other Debian packages to
> >> lambda-align's binaries, so upgrading to v2 without changing the
> package
> >> name won't break any packages, but it will break user scripts.
> >
> > From a plain Debian point of view adding a file debian/NEWS.Debian
> > explaining the issue will show the change in the new version. From a
> > user perspective it might be interesting to have the old version as
> > well.
>
> I would also advise to create a new package for v2 in order not to
> silently break users' pipelines when they upgrade their lambda-align
> package to version 2 (or just do an 'apt upgrade' without checking in
> detail whether lambda is in the -- potentially tens of packages long --
> list of packages to update).
>
>
> That was my thinking.
+1
>
> > If you volunteer to maintain the old version as lambda-align1
> > some users might be happy.
>
> I'd rather use 'lambda-align2' and 'lambda-align'... due to the reasons
> above. For me this is not much different than an ABI change that would
> require bumping a version number.
>
> > PS: In any case I'd stick with the lambda-align package name for the
> > new version. Otherwise we might need to wait for ftpmaster
> > acceptance until after freeze and thus the latest version will
> > not make it into Buster.
>
> IMHO I'd rather not have the newest version in buster (which was
> released just a couple of days ago anyway) than breaking user pipelines.
> If we're lucky then v2 may even make it through NEW until the freeze
> affecting new packages which starts in about a month.
>
> I can maintain lambda-align 1.x for the time being -- Hannes, do you
> still provide updates for the old version (I'm thinking of things like
> ensuring compatibility with gcc updates, C++ standards quirks, etc.)?
>
>
> I think this is a good idea, especially if Hannes will be providing
> updates for the v1.0.x series.
Yes, for the foreseeable future I will make sure that any build failures
with newer compilers are addressed and also try to fix severe issues
like crashes (should there be any!).
> (I consider the v1.9.x series to be release candidates for v2.0)
That's what they were!
> Pursuant to that I split off the v2.0.0 packaging:
> https://salsa.debian.org/med-team/lambda-align2
I have just had a quick look inside and for the 2.x branch I would
recommend not setting:
-DLAMBDA_MMAPPED_DB=ON
or setting it to OFF, because I had problems with high degrees of
parallelisation/vectorisation in combination with MMAP leading to very
severe slowdowns (some strange contention in the OS somewhere).
Also the new cmake script generates man-pages so your packaging doesn't
need to do that anymore (I think in the 1.x release you generated this
yourself).
> Other than the lack of full test suite that we had with v1.0.x [1], I
> think this is ready for sponsorship for the NEW queue.
>
> [1] See https://github.com/seqan/lambda/pull/121 for my (partial)
> attempt at fixing that
>
I will definitely have a look at this this week, I really want the tests
back, too. I just pushed the 2.0.0 tag because some people wanted to
package it for bioconda and I did not want them to start with the old
release.
Best regards,
Hannes
--
Hannes Hauswedell
Scientific staff & PhD candidate
Freie Universität Berlin / Max Planck Institute for Molecular Genetics
address Institut für Informatik
Takustraße 9
Room 019
14195 Berlin
telephone +49 (0)30 838-75241
fax +49 (0)30 838-75218
e-mail hannes.hauswedell@fu-berlin.de
Reply to: