[Python-Dev] LC_NUMERIC and C libraries

Martin v. L÷wis martin@v.loewis.de
20 Jul 2003 11:50:31 +0200

Christian Reis <kiko@async.com.br> writes:

> > a) it is unlikely that patches are accepted from anybody but
> >    the author of the code, and
> > b) it is unlikely that patches are implemented that provide huge
> >    chunks of the C library.
> I'm afraid I didn't quite understand these two points:
> a) Do you mean to say that the patch should be sent *to* the original
> author of the locale code? The original author of the code that *calls*
> atof/strtod?

No. I said that the original author should send us the patches; we
will only accept them from that author.

> If the latter I can try and get Alex Larsson to submit the code. Is
> written permission from the glib team, relicensing the code, not
> acceptable enough, though?

We would have to discuss this in the PSF. In general, we want the PSF
to be owner of all contributed code, see


We had bad experience with contributions by non-authors in the past,
and had to back out changes that we later found we had no right to

> b) I'm unsure as to how we should proceed without offering alternative
> versions of strtod/formatd (which is what the pystrtod.c file includes);
> AFAICS, the current situation is *caused* by the libc versions not being
> LC_NUMERIC-safe. Do you see an alternative?

It might well be unimplementable. However, if you can find
platform-specific routines that you can wrap with a small wrapper,
falling back to strtod if such routines are not available, that might
be a way out. For example, on glibc, you could use __strtod_l.