Re: r-cran-rsdmx autopkgtest failing
Hi Sébastien,
On Sat, Feb 13, 2021 at 04:29:59PM +0100, Sébastien Villemot wrote:
> > That's correct but no argument against using an automated tool.
>
> Sure. But there is the risk of forgetting to do the manual task
> associated to what the tool has done automatically. I like to be forced
> to do something manually in order not to forget those steps.
>
> Maybe routine-update could be improved by printing a message about what
> has to be done manually whenever it does something automatically (e.g.
> checking the copyright/licences changes,
Routine-update does *nothing* with copyright/licenses. I have no idea
how to automate this step and its up to the maintainer (no matter
whether an automated tool is used or not).
> checking the policy upgrade
> list, checking the debhelper compat level upgrade list).
I've just pushed this[1].
> > I admit I simply trust lintian to do this job.
>
> In practice there are many aspects of the policy that are not checked
> by lintian.
>
> lintian-brush does some of those upgrades automatically, but only for
> specific pairs of older/newer policy versions, when it can be
> ascertained automatically that compliance is effective.
That's true. But honestly, what policy breaking changes do you expect
in R package upgrades?
> > If the package builds correctly and passes its autopkgtest I assume the
> > compat level bump had no negative effect and should be done.
>
> Again, that is not generally true. It is impossible in practice to
> create an autopkgtest that checks every aspect of a package.
Yes. In R packages we are checking what upstream provides as test. Do
you want to tell me you are testing more in the R packages you are
maintaining. How should this scale for nearly 1000 R packages? The
motivation of writing routine-update was the perception that we are not
doing very well in coping with frequent upstream changes and the Debian
R ecosystem tended to lag behind upstream more and more. Thanks to
routine-update we manage quite OK to be up to date with upstream in
unstable - at least to my perception - and thus detect required new
packages sufficiently well to be in sync with upstream at freeze time.
(It should probably discussed in a separate thread whether we stop
upgrading packages now since we are in soft freeze or whether we
should upload new upstream versions until hard freeze.)
I personally would not manage to upload more than 1000 packages per
year[2] without that level of automatisation. I think expecting the
team will increase by about 10 maintainers sharing that work amongst
each other doing the manual inspection you seem to prefer is not
realistic.
> I value the standardization of packages within a team. However, in
> Debian this is usually achieved by mandating compliance to a team
> policy document.
>
> If you have noticed anything in my R packages that does not comply with
> team practices, please let me know.
My point was not the quality of your packages. I just intended to give
you some hint how to work more efficiently with your packages.
[1] https://salsa.debian.org/science-team/routine-update/-/commit/8af801bd5cf4ff013825969415d34d07ebcb7383
[2] http://blends.debian.net/liststats/uploaders_r-pkg.png
--
http://fam-tille.de
Reply to: