On Wed, Jul 2, 2008 at 1:34 PM, Tim Maxwell <timmaxw_at_gmail.com> wrote:
> Hi David,
>
> NULL is built-in because some people, like me, think of NULL as a built-in
> definition and expect it to be there even if we don't include any other
> files. I don't know if that's a good point of view or not. NULL is
> re-defined because it's also found in the system headers.
>
> I think that making NULL hard-coded is a good idea. I think that we should
> keep the changes to TypeCastExpressionNode in case people are casting 0 to a
> pointer type for other reasons. Whatever we do about NULL, though, should be
> moved to the processor module instead of the parser module.
>
> I thought that the dependency finder would mark the macros as depending on
> the *first* definition of "NULL", but I guess it's marking *all* the
> definitions, and that explains why it got included even though --all-headers
> was not set. It doesn't matter.
>
> Regarding subprocess: It's too bad that Python made popen2 deprecated only
> one release after they introduced a replacement. Eh... I guess we can just
> tell people to update their Python distribution. Thanks for the heads-up.
> The generated wrapper files aren't affected, which is why it worked on
> Windows for you.
>
> Ouch. Hashlib is new in 2.5 too... well, I included a hack to make a dummy
> version of hashlib if it wasn't available, so it's all good. (See yacc.py if
> you're curious.)
>
> Tim Maxwell
Thanks Tim! It looks like you committed all of the fixes we discussed
(except for hardcoding NULL) in r48.
I retested on Linux (Python 2.4) and Windows (Python 2.3) and all
tests pass on both platforms. Good work!
Cheers,
David
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe_at_subversion.tigris.org
For additional commands, e-mail: dev-help_at_subversion.tigris.org
Received on 2008-07-02 23:01:29 CEST