On Mon, 23 Nov 2020 at 16:58:48 +0100, Vincent Lefevre wrote:
> On 2020-11-23 10:21:32 -0500, Sandro Tosi wrote:
> > > But now it depends on both python3.8:any and python3.9:any.
> >
> > and this will continue to be the case for as long as we have 2
> > supported python3 versions.
>
> Shouldn't python3.8:any be removed after the transition is complete?
Eventually yes, but the transition *isn't* complete.
The sequence goes like this:
* initial state: only old version is supported
* add new supported version
* rebuild multi-version packages so they work with both
* change default to new version
* rebuild single-version packages for new version <-- [we are here]
* remove old supported version
* rebuild multi-version packages so they only support the new version
* final state: only new version is supported
where "multi-version packages" are things like numpy that can be built
for more than one Python version at a time (using python3-all-dev), and
"single-version packages" are things like python3-ldb that are only built
for the default version (using python3-dev).
> I'm wondering why these two dependencies are there in the first place
> (if this is because of f2py3.8 and f2py3.9, they could be regarded as
> optional, suggesting to use f2py3 instead, and if some package needs
> a specific version of python, then it would already depend on it).
Maybe they can be relaxed to Recommends or Suggests or something, if
"I installed this package while python3.9 was the default, and I can't
run f2py3.8" would be wishlist/wontfix rather than RC.
smcv
