[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: R packages - are we doing the right thing?



Hi,

at first thanks to Steffen for raising this question I was stumbling
upon previously as well.

On Wed, Apr 29, 2015 at 08:06:01AM -0700, Don Armstrong wrote:
> On Wed, 29 Apr 2015, "Steffen Möller" wrote:
> > I would like to have
> > http://cran.r-project.org/web/packages/data.table/index.html
> > for Debian, which is rather fashionable these days, but looking at the
> > dependencies, I refrained from packaging it for Debian Med's R directory.
> > It becomes too much, I thought. And we should not need a sub-blend
> > for R packages :)
> > 
> > To my rescue, Don Armstrong maintains http://debian-r.debian.net/ ,
> > i.e. a repository of auto-created R packages, and I happily found the
> > package there. Now, for anything we want in Debian Med we cannot allow
> > for dependencies to external sources - tcR comes to mind as a reverse
> > depdendency for data.table.
> > 
> > Is anything coming to your minds how we could make use of Don's effort
> > for Debian Med?
> 
> I think we actually should switch to adopting the perl team approach,
> and create a separate sub-team to package R packages in Debian. I've
> been working (very slowly) on doing this, and would love more
> contributors.

I really appreciate this idea and I also had suggested something in the
past in this direction.  The only point is that I personally do not feel
qualified to lead this effort since I have only very rudimentary
knowledge in R.  While packaging R packages is quite simple I can do
this easily but I think some more R skills would not harm for a person
triggering this kind of effort.

I for myself will happily support this and would move all R packages I
had touched in the past into a Debian R team repository.

> I'm currently leaning towards having two classes of packages; those
> which just go directly to debian-r.debian.net, and those which we
> actually upload to debian.
> 
> If we do that, packages can be promoted and demoted easily between the
> two groups simply by changing a single .mrconfig file and doing a real
> upload.

While I for myself have not yet used mr I'd gladly adapt such a
workflow.

> We could then leverage c2d to make the maintenance fairly trivial for
> the packages, but still have the ability to make manual tweaks to
> important packages as necessary.
> 
> > Could we possibly use Don's repository for bits and pieces to upload
> > to the distribution? It would require an overseeable change to c2d to
> > ensure that c2d knows to be in charge over updates despite a presence
> > of the package in the distribution, and maybe we have some extra
> > information for debian/control and debian/copyright, but then we could
> > have a selected few manually monitored packages auto-built for the
> > distribution. Other communities may find their own respective set of
> > packages.
> 
> I think having an overarching single group is probably best, but we
> should certainly have people from all of the different R related
> disciplines involved... I definitely don't want to make anyone's job
> harder.

As far as I see you suggestion I think the work will not be harder if
you are maintaining code in a VCS anyway.  Switching the VCS URL should
be an easy task.

To add to Steffens initial question I also came across previously.  From
my personal view the following reasons are a (strong) pro for keeping
R packages in Debian:

  1. You can Depends / Recommends such packages (which we are heavily
     doing in Debian Med and Debian Science)

  2. You can run a test suite via autopkgtest which guarants a constant
     testing on Debian systems.

IMHO both things are quite important - specifically if we are talking
about scientific software.

Apropos testing: I have some hope that #752609 might have better chances
to get fixed if we create an R team working on this in a dedicated VCS.

Don, thanks a lot for your great suggestion

    Andreas.


-- 
http://fam-tille.de


Reply to: