Re: Refs to Conda - gone through the med/bio task list
I have iterated through all the packages that are in our current
distribution that are on the med/bio task list and added references to
conda. What I have learned:
* it was surprisingly rewarding - most packages are in conda, and if
they are, there are typically quite a number of reverse dependencies
listed (a feature dearly missed from our package tracker) which I found
to be very interesting to look at for ideas on how the package is used
and what Debian should possibly also consider packaging
* pymol in conda is directly maintained by the company Schrödinger
itself via the conda channel "schrodinger" - we should think about what
Debian misses to achieve the same distribution of workloads also to
those outside of what we call "main" or "society".
* as few (rare) times that I cared to look deeper, I found that conda
had newer versions of packages due to a relocation of the upstream
source-code repository, typically to git. And I just reported a case
when a package had another version as package2 in a separate repository.
Sigh. I strongly suggest to also report the package version in conda in
our qa page.
* Conda has a series of channels, and sometimes packages appear in
multiple such. "conda-forge" is the most reliable one with the newest
versions - if a package was in there, I referenced to it. "bioconda" is
the one that we were most aware of. And there is also "r". The
bioconductor packages are typically within the "bioconda" channel. I
mentioned "schrodinger" before.
The packages maintained by the med-team alone have already 260
references to conda and another 77 (only started late with this) for
which conda package was in sight and an "NA" was assigned. This is
almost exclusively in bioconda (3 in conda-forge, 1 in r) since most of
R is with the r-pkg-team on salsa. In comparison, we have 280 refs to
bio.tools and 263 additional ones for which bio.tools is NA. For
OMICtools it is a whopping 548 and 23 with NA.
Ideas what to do next:
* add conda refs to med/bio-dev
* fix local mirror of salsa repo to not use ssh - it runs into
fail2ban, I presume.
* fix syntax of all that yaml in d/u/metadata external to the med blend
* have yamllint executed by lintian
* come up with ideas how to use these conda references - QA page was
* blends task pages need to prepare a bit more for the different
channels of conda
* come up with cross-community ((Bio)Conda-Debian) projects
* get inspired by conda's technology to prepare user-installable
In summary, I think the effort should be continued. I feel personally
somewhat responsible for the packages in our bio-* task lists. Beyond
that, we either convince our community that the effort is worth it - or
likely it is not worth it. Without examples how this all can be brought
to some good use, we should possibly invest our energy elsewhere.
On 18.09.19 15:25, Steffen Möller wrote:
I just met Björn from the Conda community over here at the GCB in
Heidelberg. Over some sweet cake we agreed that it would be good to
mutually reference our packages. And we find that it would be mutually
beneficial to develop some training material that references both
conda and Debian packages, allowing the focus to remain on the
biology. The idea is to allow for the smoothest possible transition
from "Jugend forscht!"-/hobby/student side projects to HPC
environments and back, like to explain results to the public. We had
also some early thoughts about how to help automating the exchange of
links with some inter-distributional bots, also involving the
github-reinvented bio.tools. Does anybody of folks reading this attend
their upcoming BioHackathon in Paris?
You are all up for this, right?
I saw Björn do first pull requests towards the mutual links and have
created a bunch on my own now - a* and b* should be complete.
@Andreas, could you please perform your magic on integrating that data
with the udd? And did I do the modification of the blends' task page
Many thanks and greetings,