Re: py-maturin and version requirement

[maya%NetBSD.org@localhost]

On Thu, Mar 20, 2025 at 06:09:57PM -0400, Greg Troxel wrote:
> Jonathan Perkin <jperkin%pkgsrc.org@localhost> writes:
> 
> > * On 2025-03-20 at 16:38 GMT, Greg Troxel wrote:
> >
> >>  - bin-install is buggy because it should only install the version that
> >>    would have been built
> >
> > Not this one.  pkg_add will choose the best matching version for the
> > supplied match.  If the specified match is wrong and an older version
> > than required is installed then that's a fault of the match.
> >
> > Changing the logic so that the match is overridden to the exact
> > version that is currently supplied by pkgsrc would avoid the problem
> > that you're running into, but it's the wrong approach and would result
> > in a lot of needless builds when the existing binary package would
> > suffice.
> 
> Yeah, I guess that would be like starting to turn regular builds into
> pbulk, so no.
> 
> > Just fix the match to be correct.
> 
> I have a fix ready to commit.  With it I can still build py-rpds-py:
> 
>   => Tool dependency py312-maturin>=1.8.3: found py312-maturin-1.8.3
> 
> The benefit is that anyone who
> 
>   has py-maturin 1.7.8 installed, e.g. from 2024Q4 
> 
>   tries to build devel/py-rpds-py
> 
> will see an attempt to install py-maturin-1.8.3, instead of a mysterious
> build failure, because a dependency requirement that exists is not
> expressed.
> 
> I have no idea which of the 1.8.0-3 series is necessary; upstream's NEWS
> doesn't address this.  I know 1.8.3, which is what is in pkgsrc now,
> works for py-rpds-py.
> 
> This is just a bugfix, but sort of infrastructure.  I'm happy to do it
> now or wait until after branch -- my tree will have it, so it won't
> bother me to wait.  The minus is it's a change, which will trigger
> rebuilds, and all changes have risk -- but it seems safe and it's easy
> to revert if it'sn not.
> 
> Maya?

OK by me