Michael G Schwern <[EMAIL PROTECTED]> writes:
>On Wed, Jul 30, 2003 at 06:34:13AM +0100, Alan Burlison wrote:
>> >https://rt.cpan.org/NoAuth/Bug.html?id=3081
>> >
>> >A little help on this one. I don't have a VC++ compiler and I wouldn't be
>> >much good at fixing a recursive XS build problem on Windows anyway.
>>
>> AFAIK it doesn't work properly on *any* platform. I submitted a patch for
>> this against 5.6.1, but it got dropped as it broke Tk. DEFINE and INC
>> don't get propagated ISTR.
>>
>> http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2001-07/msg00150.html
>
>I don't mean to throw a firebomb here, but Tk is a bit too chummy with
>MakeMaker to use it as a yardstick for backwards compatibility. I'd be
>curious as to what changes could be necessary to Tk to make this work again
>and if it causes problems with any other modules.
All Tk needs is a Makefile variable it can populate with
the -Ixxxx values it needs to compile in this sub directory.
It is currently using INC= 'cos that was what got passed to CCCMD.
What it does not need is its grandparent's Makefile.PL's guess
forced upon it.
So the current MakeMaker scheme which uses PASTHRU_INC is fine.
It would be even better if CCCMD saw
$(INC) $(PASTHRU_INC)
rather than
$(PASTHRU_INC) $(INC)
Now if grandad can sensibly prepend ../ to relative paths and
leave absolute paths alone then Tk need not mess with anything.
(Old MakeMaker used to override INC= on pasthru which Tk
had to stop. Tk used INC as it was used in _lots_ of
targets xs->c, c->i c->o o->so and it was easier
to clobber MM::pasthru than all of those.)
