On Tue, Jun 25, 2019 at 10:48:42AM +0200, Miro HronĨok wrote:
On 25. 06. 19 9:50, Pierre-Yves Chibon wrote:
> > > > > > I would say the scoping should happen (e.g. in copr) first,
then we'll
> > > > > > know what the scope (and hence feasibility) of this change
truly is.
> > > > >
> > > > > What will the scoping actually tell us? If it tells us that X
hundred packages
> > > > > need to add BR for python3-test, we'll just do that and
report the list to
> > > > > upstream. We can do that during the mass rebuild.
> > > >
> > > > If there would indeed be X hundred packages affected, would this
move
> > > > still make sense? The python3-test subpackage is even larger than
most
> > > > of the rest of python3 combined, and the vast majority of it is
> > > > irrelevant to other packages. Such usage would indicate to me that
> > > > some work needs to be done within the ecosystem to respect its
official
> > > > status as internal-only. (Perhaps, in that case, a move to python3-
> > > > devel could be considered instead.)
> > >
> > > Good questions. What number of affected packages do you think would be
crucial?
> > > I say X hundred, because I don't anticipate it will go to thousands.
However
> > > with 3000 Python packages, couple hundreds is IMHO not a big deal. Note
that
> > > this is build dependency, not a runtime one.
> >
> > Nevertheless, there has been a great deal of effort recently to shrink
> > buildroots and speed up build times. Having to add BR: python3-test to
> > packages would mean adding a download of ~9MiB and an installation of
> > ~3K files to every single affected package's build time. IMO this
> > needs to be fully scoped before consideration.
>
> Wait a minute, adding a BR python3-test would mean an additional 9MiB ok, but
> currently these files are part of python3-libs which *every packages* get.
> In other words, for the packages that require python3-test the amount of data
> downloaded doesn't change while for all the packages that do *not* require
> python3-test, the amount of data is reduced.
> This sounds like a win to me :)
Unfortunately, this is not correct.
python3-test already is large.
We just move a small bit of python3-libs (test.support) and we move it to
python3-test for consistency.
test.support is 396K installed (for Python 3.8).
Arf, then it's a lesser win than I thought.
Have you considered doing a python3-test-support subpackage? This would allow
the separate package and thus give you the information you're looking for as
well as avoid downloading the entire python3-test for these few files.
Pierre