[FFmpeg-devel] [PATCH] avcodec/x86/Makefile: Don't build empty files

[Carl Eugen Hoyos]

Am Di., 13. Dez. 2022 um 20:03 Uhr schrieb Andreas Rheinhardt
<andreas.rheinhardt at outlook.com>:
>
> Carl Eugen Hoyos:
> > Am Di., 13. Dez. 2022 um 16:02 Uhr schrieb Andreas Rheinhardt
> > <andreas.rheinhardt at outlook.com>:
> >>
> >> Andreas Rheinhardt:
> >>> Should fix ticket #9909, fixing a regression since
> >>> bfb28b5ce89f3e950214b67ea95b45e3355c2caf.
> >>>
> >>> Thanks to Carl Eugen Hoyos for analyzing the issue.
> >>>
> >>> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt at outlook.com>
> >>> ---
> >>> This would be my solution. What do you think of it?
> >>>
> >>>  libavcodec/x86/Makefile | 7 +++++--
> >>>  1 file changed, 5 insertions(+), 2 deletions(-)
> >>>
> >>> diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
> >>> index 4e448623af..41ca864849 100644
> >>> --- a/libavcodec/x86/Makefile
> >>> +++ b/libavcodec/x86/Makefile
> >>> @@ -140,8 +140,11 @@ X86ASM-OBJS-$(CONFIG_QPELDSP)          += x86/qpeldsp.o                 \
> >>>  X86ASM-OBJS-$(CONFIG_RV34DSP)          += x86/rv34dsp.o
> >>>  X86ASM-OBJS-$(CONFIG_VC1DSP)           += x86/vc1dsp_loopfilter.o       \
> >>>                                            x86/vc1dsp_mc.o
> >>> -X86ASM-OBJS-$(CONFIG_IDCTDSP)          += x86/simple_idct10.o           \
> >>> -                                          x86/simple_idct.o
> >>> +ifdef ARCH_X86_64
> >>> +X86ASM-OBJS-$(CONFIG_IDCTDSP)          += x86/simple_idct10.o
> >>> +else
> >>> +X86ASM-OBJS-$(CONFIG_IDCTDSP)          += x86/simple_idct.o
> >>> +endif
> >>>  X86ASM-OBJS-$(CONFIG_VIDEODSP)         += x86/videodsp.o
> >>>  X86ASM-OBJS-$(CONFIG_VP3DSP)           += x86/vp3dsp.o
> >>>  X86ASM-OBJS-$(CONFIG_VP8DSP)           += x86/vp8dsp.o                  \
> >>
> >> It seems like that there are linkers out there that complain about empty
> >> object files like x86/simple_idct.o ("ranlib: file:
> >> libavcodec/libavcodec.a(simple_idct.o) has no symbols" as reported by
> >> BBB); the above patch is still needed to fix this. I will therefore
> >> apply it tonight (with an updated commit message) unless there are
> >> objections.
> >
> > (Three hours is not a lot)
> >
> > Which toolchain broke?
>
> "Broke" is the wrong word; Xcode emits warnings for empty object files.

It does.
(I did not consider this an issue when I saw it.)

I consider this unmaintainable, an argument that in the past was
used to object significantly more important patches (in cases
where maintenance was imo no problem).

Anyway: No objection.

Carl Eugen