[FFmpeg-devel] OSX 10.5 (Leopard) build problem

[Måns Rullgård]

SciFi <sci-fi at hush.ai> writes:

> Hi,
>
> On Sun, 04 Nov 2007 21:06:47 +0100, Pierre d'Herbemont wrote:
>> On Nov 4, 2007, at 8:33 PM, Steven M. Schultz wrote:
>>> 	It seems that with 10.5 the advice that "-mdynamic-no-pic" is for
>>> 	applications and not shared libraries is being enforced.
>>>
>>> 	So if -mdynamic-no-pic can't be used because of linking problems but
>>> 	the compilation fails if that option is not used what's a person to
>>> 	do?
>> 
>> That's a regression known by Apple in the new ld (ld64).
>> 
>> A work around is to cross compile using the 10.4u SDK, which uses
>> ld_classic.
>> 
>> Pierre.
>
> fwiw on my Leopard partition, I'm busy compiling all the
> prerequisite projects to use the new SDK, and have been seeing
> a number of "new" problems with the new ld, some of which can
> be "googled" for remedies but not all are bypassed yet.  Mostly
> the folks at MacPorts seem to be on top of things, so I figure
> soon enough Apple will know all about these new problems I
> hope.  ;)

Why are Apply making each release more difficult to use?  Are they
running some sort of experiment to see just how loyal their fans are,
or what?

> Presently for Tiger and the libavcodec subtree, I've had to
> patch its Makefile in the following manner to get around some
> compiler problems.  I wonder if this will work for you on
> Leopard?  (it's going to take me quite a while longer to get
> all the prereqs ready for ffmpeg & co. on my Leopard system):
>
> ====-cut-here-====
> Index: libavcodec/Makefile
> ===================================================================
> --- libavcodec/Makefile	(revision 10912)
> +++ libavcodec/Makefile	(working copy)
> @@ -381,6 +381,12 @@
>  OBJS-$(HAVE_VIS)                       += sparc/dsputil_vis.o \
>                                            sparc/simple_idct_vis.o \
>
> +h264.o:	CFLAGS += -fomit-frame-pointer
> +
> +i386/snowdsp_mmx.o:	CFLAGS += -fomit-frame-pointer 
> +
> +i386/dsputil_mmx.o:	CFLAGS += -fomit-frame-pointer -O1 
> +
>  OBJS-$(HAVE_MLIB)                      += mlib/dsputil_mlib.o \
>
>  OBJS-$(ARCH_ALPHA)                     += alpha/dsputil_alpha.o     \
> ====-cut-here-====

The default configuration should already use -fomit-frame-pointer.

> Yes it's the -O1 that _finally_ fixed the issue with me on
> Tiger, any -O higher will introduce the register problem in
> dsputil_mmx.c with Apple's compiler.

Weird.  Usually it's the other way around: register allocation fails
without enough optimisation.

-- 
M?ns Rullg?rd
mans at mansr.com