[Pythonmac-SIG] PyXML install problems

[Ronald Oussoren]

On 5-apr-2006, at 18:18, Christopher Barker wrote:

> Bob Ippolito wrote:
>
>> I'll see about creating a repository
>> for universal packages sometime over the next few days (weekend is my
>> best guess).
>
> Did you get a chance to take a look at my re-structuring of the
> pythonmac repository site? Anything else I can do to help?
>
> While we are on the topic, I'm looking at building a couple packages
> that require external libs (PIL, matplotlib). In the past, I've  
> tried to
>   build them as static libs and link them to the extensions to make it
> all self contained. I see two issues with this now;
>
> 1) How do I make Universal static lib?

Basically use 'cc -arch i386 -arch ppc -isysroot /Developer/SDKs/ 
10.4u.sdk' as
the compiler and hope that configure doesn't check for endianness.  
(There may
be spelling errors in that command-line, I haven't checked my build  
script).

I have a version of PIL with tiff and jpeg support (Universal of  
course).

>
> 2) We'll end up using the same libs over and over in multiple  
> packages:
> Both PIL and matplotlib require libpng and libfreetype, for  
> instance. I
> expect we'll see other overlapping libs.
>
> a) Is there any problem with two different extensions being loaded  
> that
> are linked to the same static lib, particularly if they are different
> versions?

That shouldn't be a problem until someone starts playing with dlopen  
flags,
and that can't happen at the moment because dynload_next doesn't look at
those :-).

>
> b) If not, then I guess it's just a waste-of-memory issue, which
> probably isn't a big deal, but seems wasteful. I do think that any
> packages put on pythonmac.org should only depend on other packages
> available there. Perhaps we could make a "libs" package with a  
> bunch of
> these common libs.
>
> If we put them in the python tree, would the linker find them? Would
> Py2App find them?

I don't know about the latter but the former shouldn't be a problem,
you'll just have to make sure that the right linker arguments are used.

Ronald