[Ffmpeg-devel] SVN dump

[Rich Felker]

On Mon, Apr 16, 2007 at 03:34:23AM +0200, Michael Niedermayer wrote:
> Hi
> 
> On Sun, Apr 15, 2007 at 09:11:52PM -0400, Rich Felker wrote:
> > On Sun, Apr 15, 2007 at 08:22:59PM -0400, Rich Felker wrote:
> > > > > I don't have Python and it's not terribly portable...
> > > > 
> > > > Any example of relevant platform which is not supported ?
> > > > This list already seems pretty long and contains quite some "exotic" arch:
> > > > http://www.python.org/download/other/
> > > 
> > > Having explicit ports to various obscure systems and having portable
> > > code (i.e. code that does not depend on any nonportable features and
> > > have to be rewritten for each platform) are two very different things.
> > > Just to make you happy I'll try building it again and see what
> > > happens...
> > > 
> > > Results coming soon.
> > 
> > And they're very bad!
> > 
> > I can't figure out how to get the damn thing to stop trying to build
> > all its modules as .so files and link them properly. I tried faking
> > uname to get it to stop assuming I'm using "GNU/Linux", then even
> > passing --without-gcc. No luck.
> 
> hmm, what about a wraper around gcc/linker which fails builing .so? is their
> configure smart enough and falls back to normal objects?
> or what about a gcc warper which simply builds normal objects when asked
> to build .so ...

Nope, it's a crappy build system that doesn't auto-detect anything
really. It just makes assumptions based on uname. And there seems to
be no consideration for a case where shared libs are not available.

> or asking the python devels ...

I asked in the irc channel and got fairly courteous response until the
ppl there realized they didn't know what to. Then they started
accusing me rather than acknowledging that python's build system is
broken. (And FYI I never made this inflammatory comment to them that
the build system is broken; I just asked for help.)

Some Googling turned up Modules/Setup[.local] files, which control
which modules are compiled into the main binary. But I can't figure
out where it decides to compile the ones not listed there as .so
files. I tried hacking the .py file that controls building them, and
the python interpreter started using up 500 megs of core and nearly
crashing my system. Not sure if that was due to my hack or if it's
normal.

> > P.S. There were also minor bugs in the source, missing #ifdefs, etc.,
> > which I corrected without much effort, 
> 
> submit patch to the python devels :)

If I get the rest to work, I will..

Rich