Re: [Pyobjc-dev] Build broken (Re: [Pyobjc-checkins] CVS: pyobjc setup.py,1.19,1.20)

[Ronald O. <ous...@ci...>]

On Sunday, Dec 22, 2002, at 20:29 Europe/Amsterdam, Just van Rossum 
wrote:

> Ronald Oussoren wrote:
>
>> Libffi is a library for dynamicly building stackframes and calling
>> existing C functions. One of my earlier checkins today added code
>> that uses libffi to call the superclass implementation of methods and
>> to build the entries in the method tables of Objective-C classes.
>> This allows us to do away with Modules/objc/register.m, a really
>> enormous file. The library is also necessary if you want to add
>> methods to pure Objective-C classes (which is simular to writing
>> Objective-C categories in Python).
>
> [...googling, checking the page at redhat.com...]
> Oooh, way cool! calldll done right! Must Have Python Interface ;-)

+1 on that.

>
>> The main disadvantage of libffi is that it is a 3th party library
>> that is not easily available: The current version is only available
>> as part of the GCC sourcetree.
>
> You wrote in libffi.txt:
>
>> To enable libFFI support:
>> - Build and install libffi 2.x, this is only available from a source
>> distribution of GCC.  Do NOT use the version from sources.redhat.com,
>> it doesn't support MacOS X at all.  The version I used (december
>> 2002) doesn't allow you to build a shared library on MacOSX. This is
>> not a problem, we don't want one anyway.
>
> Do you have a more specific pointer where you got your copy?

I got mine by doing a CVS checkout of the GCC trunk. Just downloading a 
recent copy of GCC will probably work just as well. If you go the CVS 
route, checking out 'gcc/libffi' instead of the entire gcc module saves 
a lot of time.

The configure in libffi will give a fatal error near the end of 
configureing, just ignore the error. Libffi itself is already 
configured by the time you encounter the error; it tries to patch a 
file one level up in the directory tree.

Ronald