[MPlayer-dev-eng] [PATCH] Windows DLL support for OS X/Intel

[Valtteri Vuorikoski]

Reimar Döffinger <Reimar.Doeffinger at stud.uni-karlsruhe.de> writes:

> How does this relate to http://bugzilla.mplayerhq.hu/show_bug.cgi?id=514
> ?

Well, that's what I get from only looking at this mailing list's
archives and not the bug tracker etc. Mine does about the same, minus
large dyld/pack()-lossage workarounds, plus more-interoperable
stubs.s. The wrapper.S changes are only required for QTX support,
as far as I can see (my version disables it on Darwin).

> Looks like this is the issue that "caused" all those large patches in
> the URL above *g*

Most likely.

> IMHO requiring anonymous mmap is fine (but make sure to accept both
> MAP_ANON and MAP_ANONYMOUS as libswscale/swscale.c does).
> If some OS does not support it, it should be emulated via a stub in osdep or so.
> Though using FILE_dommap everywhere would be a possibility, too, but it
> should be cleaned up before, e.g. the flags parameter is completely
> ignored...

I think Solaris only supports anonymous mmap() with the /dev/zero idiom,
some OSes support both, Darwin only supports the classic BSD flags=MAP_ANON, fd=-1
idiom. Some sort of "mmap() me some anonymous space" function is probably ok.
I'll see about using FILE_dommap at least in ext.c.

I don't know if using it in ldt_keeper.c as such is a good idea since it's
defined using funky win32 typedefs.

 -v