Re: Build-Depends for CRAN packages
Am Freitag, 21. August 2020, 06:18:21 CEST schrieb Dirk Eddelbuettel:
> On 21 August 2020 at 05:58, Johannes Ranke wrote:
> | Hi all,
> |
> | yesterday I wondered why the Build-Depends of CRAN packages that are
> | uploaded to unstable are adapted to the current R version, as mentioned
> | in the Changelog e.g. of yesterdays upload of mgcv:
> |
> | --------------------
> |
> | Betreff: Accepted mgcv 1.8-32-1 (source) into unstable
> | Datum: Donnerstag, 20. August 2020, 18:06:49 CEST
> | Von: Debian FTP Masters <ftpmaster@ftp-master.debian.org>
> | An: debian-devel-changes@lists.debian.org
> |
> | Format: 1.8
> | Date: Thu, 20 Aug 2020 10:41:33 -0500
> | Source: mgcv
> | Architecture: source
> | Version: 1.8-32-1
> | Distribution: unstable
> | Urgency: medium
> | Maintainer: Dirk Eddelbuettel <edd@debian.org>
> | Changed-By: Dirk Eddelbuettel <edd@debian.org>
> |
> | Changes:
> | mgcv (1.8-32-1) unstable; urgency=medium
> | .
> |
> | * New upstream release
> |
> | .
> |
> | * debian/control: Set Build-Depends: to current R version
> |
> | --------------------
> |
> | I never got the point of why the adaptation of the build dependency is
> | necessary or beneficial for these uploads of CRAN packages to unstable.
> | Shouldn't they build at least on the previous R version, as demonstrated
> | by
> | the CRAN checks on r-oldrel [1]? It seems to me that compatibility with
> | the
> | previous release is implicitly required and -- at least to some degree --
> | guaranteed by these checks.
>
> Debian is independent of CRAN.
Good point. But it would still appear reasonable to me to make use of our
knowledge of the CRAN testing infrastructure when packaging for Debian.
> So when I as a Debian developer upload to _Debian unstable_ we ensure it
> builds with Debian unstable in order to migrate to testing in order to be
> cast one day into a new stable release.
I do not think that this build dependency is necessary for this. In my
experience, CRAN packages build on the previous R version as well. Migration
of binary packages can sufficiently be controlled by the r-api version as far as
I understand.
> That is simply a different set of requirements than we have at CRAN (where I
> also upload a few things as a CRAN package author -- but into a very
> different system).
>
> | [1] https://cran.r-project.org/web/checks/check_flavors.html
>
> While CRAN has Debian flavours for testing, these tend to be the "testing"
> rolling releases so we can't directly rely on them.
Well, the checks on r-oldrel that I am referring to are for Windows and Mac
anyways. So my argument is not specifically strong here.
> I think we are in basic agreement, and having done this here for so long I
> often wondered if we could not have a more formal Debian <-> CRAN link. But
> we don't right now. Hence Debian requirements only in the upload.
>
> That said, it is also possible to systematically patch this. Michael seems
> to cope fine at his PPA with the over 4600 r-cran-* packages for three
> Ubuntu flavours.
This is what I started doing in 2008 [2] for the backports to Debian. I
recently removed this code [3] as I generally compile CRAN packages using the
current R version backported from unstable, so the build dependency is
fulfilled anyways.
I do not want to push this too hard. The discussion already helped my to
understand it better.
Thanks,
Johannes
[2] https://cgit.jrwb.de/r-backports/commit/?id=ea37b5789aec00d47c
[3] https://cgit.jrwb.de/r-backports/commit/?id=b0e1bc37ded73e
Reply to: