Re: Remove partial-extraction hack for Python modules

[Joerg Sonnenberger <joerg%bec.de@localhost>]

On Wed, Dec 25, 2019 at 08:23:57AM -0500, Greg Troxel wrote:
> Joerg Sonnenberger <joerg%bec.de@localhost> writes:
> 
> > Hello all,
> > we build a number of Python packages separately as they require
> > additional depencies. Those are:
> >   databases/py-gdbm
> >   databases/py-sqlite3
> >   devel/py-curses
> >   devel/py-cursespanel
> >   devel/py-idle
> >   devel/py-readline
> >   textproc/py-cElementTree
> >   textproc/py-expat
> >   x11/py-Tk
> >
> > For historic reasons, those only extract a small part of the Python
> > distfile and hack around the patch failure for non-existing files by
> > ignoring the error status. I think that hack has mostly outlived its
> > usefulness. Recent changes in Python 3.8 for example broke all the
> > packages by requiring additional files to be extracted. The attached
> > patch removes the hack. Downside is that it will require more disk space
> > and associated writes during a build.
> >
> > Joerg
> 
> I'm generally in favor of less magic and less fragile, and don't mind a
> bit of slower to get that.  I wonder if you have any quantitative
> estimate, but I guess it's 9 copies of the unpacked python distfile,
> minus the bits that are needed, so maybe 9*0.98.  Guessing from 'xzcat'
> size on 3.8, that seems like 738 MB of extra write load for a full bulk
> build, about 80 MB per package above.  I'd expect most people building
> more than a few to have a decent-sized tmpfs and workdir on it, so I
> would guess this doesn't really hurt.
> 
> Is that how you see it?

Essentially, yes.

Joerg