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

A transition plan for R data in binary format ?

Le Tue, Aug 06, 2013 at 08:29:14AM +0900, Charles Plessy a écrit :
> While writing this answer, I also read Don's email advocating for Debian to
> take the lead and change the current practice in the R community, that prefers
> to ditribute data as R binary objects in the source packages.  This is
> laudable, but I expect that it will take time, and it needs people who have
> roots in both communities.
> In the current situation, that I describe as "active bitrotting", we do not
> apply the same rules to the packages that enter the archive and the packages
> that are already in, which cause the packages under active development to
> become obsolete each time new dependancies can not enter in Debian.

Dear FTP team and everybody,

how about the following to smoothly solve the problem.

 - We aim at a full resolution within two or three release cycles.  This is
   roughly the time that was needed to solve the problem of binary blobs in the
   Linux kernel.

 - We define precisely what are the criteria for R binary objects to be
   accaptable in Debian, and how to test conformance with these criteria.

 - For the new R packages that do not interact much with the rest of the
   currently distributed packages, we apply these criteria strictly.

 - After reaching consensus between the main stakeholders (FTP team,
   maintainers of R packages in Debian, ...), we contact the main R communities
   (R developers, CRAN, Bioconductor, etc), and ask them to change the way they
   assemble source packages, so that they contain the files used to generate the R
   binary objects.

 - In a transitory period, for the packages that are already in Debian as well
   as the packages needed by them to be upgraded to fresh upstream versions, we
   give the priority to keep the packages up to date, while working in parallel to
   solve the problem as indicated above.

To avoid confusion, I would like to remind again that for the majority of the R
binary objects, loseless conversion with a plain text format is already
possible, and there is no evidence that a more complex source file is needed to
generate them, which I think is enough to make them Free.  The main problem
seems to be that with the binary format, it is difficult to detect the rare
cases where conversion from the source format requires non-trivial commands
that are not in the R source package.

What do you think about this ?


Charles Plessy
Debian Med packaging team,
Tsurumi, Kanagawa, Japan

Reply to: