This is the mail archive of the
mailing list for the glibc project.
Re: Use gen-libm-test.py to generate ulps table for manual
On Mon, 10 Sep 2018, Dmitry V. Levin wrote:
> On Mon, Sep 10, 2018 at 12:41:45PM +0000, Joseph Myers wrote:
> > Dmitry, do you wish to make any comments on the use of Python in building
> > glibc?
> My concern with making python a requirement to build glibc is that it will
> pull in more packages not required to build glibc. If there is a "minimal"
> subset of python that does not bring indirect dependencies to glibc build,
> and at the same time is sufficient to build glibc, then adding this subset
> of python as a requirement to build glibc shouldn't be a problem.
> I don't know yet whether SUSE or Debian take on the minimal subset pulls
> in extra dependencies, I'll check this when time permits.
Any conclusions there?
I'm not sure it really matters for these purposes what a particular
distribution's package pulls in, if they are happy with what works for
them for the distribution build or bootstrap. The question for us would
be more about what constraints this imposes on distributions in general
(e.g. if it means libffi will be needed for a native build of glibc,
depending on how far the indirect dependencies on ctypes in the Python
standard library go), and whether we are OK with those constraints. The
dependencies of the minimal distribution packages are simply useful
information for us to help in determining those constraints.
Joseph S. Myers