[MERGE][bug 276868] Blacklist pyrex 0.9.4.1

[Martin Pool]

On Thu, Oct 2, 2008 at 8:15 AM, John Arbash Meinel
<john at arbash-meinel.com> wrote:
> It seems that pyrex 0.9.4.1 fails to properly compile the new iter_changes
> code that Robert wrote. The code it generates re-uses a "local" pointer and
> calls "PY_DECREF" after having set it to NULL. (It mixes PY_XDECREF which is
> NULL safe with PY_DECREF which is not.)
>
> This just adds an explicit blacklist for all pyrex compilers <0.9.6.0 for that
> extension.
>
> One alternative is to set "have_pyrex = False" instead, and issue a note that
> we won't generate any of the .c files for old versions of pyrex.
>
> For now, this was the "minimal" change I could do.
>
> Thoughts?

BB: tweak

(But bb doesn't seem to have picked this up yet....)

I'd like the comment next to the code to say more about what the
problem is - just copy that paragraph from your mail.

I think in general we're a bit too tolerant of people running without
compiled extensions and therefore seeing bad performance, and we
should do something about that.  Maybe setup.py should fail unless you
specify --without-compiled-extensions or something.  But that's beyond
the scope of this patch.

-- 
Martin <http://launchpad.net/~mbp/>