few compiler hitns / warning for version 0.7.4

[Toralf Förster]

I got at an almost stable Gentoo with gcc 4.4.5 :

• QA Notice: The following files contain runtime text relocations
  • Text relocations force the dynamic linker to perform extra
  • work at startup, waste system resources, and may pose a security
  • risk. On some architectures, the code may not even function
  • properly, if at all.
  • For more information, see ​http://hardened.gentoo.org/pic-fix-guide.xml
  • Please include the following list of files in your report:
  • TEXTREL usr/lib/libavutil.so.50.43.0
  • TEXTREL usr/lib/libpostproc.so.51.2.0
  • TEXTREL usr/lib/libavdevice.so.52.5.0
  • TEXTREL usr/lib/libswscale.so.0.14.1
  • TEXTREL usr/lib/libavformat.so.52.110.0
  • TEXTREL usr/lib/libavfilter.so.1.80.0
  • TEXTREL usr/lib/libavcodec.so.52.122.0

• QA Notice: Package may compile
• fine but exhibit random runtime failures.
• libavcodec/wmaprodec.c:819: warning: dereferencing pointer ‘({anonymous})’ does break strict-aliasing rules
• libavfilter/libmpcodecs/vf_spp.c:301: warning: dereferencing pointer ‘({anonymous})’ does break strict-aliasing rules
• ffmpeg.c:1082: warning: dereferencing pointer ‘picture2’ does break strict-aliasing rules

• QA Notice: Package may compile
• fine but exhibit random runtime failures.
• libavcodec/h264.c:3592: warning: array subscript is below array bounds

• QA Notice: Package may compile
• fine but exhibit random runtime failures.
• libavcodec/dca.c:1151: warning: array subscript is above array bounds
• libavcodec/dca.c:1151: warning: array subscript is above array bounds
• libavcodec/dca.c:1151: warning: array subscript is above array bounds

• QA Notice: Package may compile
• fine but exhibit random runtime failures.
• libavcodec/libx264.c:351: warning: the address of ‘val’ will always evaluate as ‘true’

• Please do not file a Gentoo bug and instead report the above QA
• issues directly to the upstream developers of this software.
• Homepage: ​http://ffmpeg.org/

[Toralf Förster]

with new version (0.7.5 and new gcc version 4.5.3) these warnigns still remains:

• QA Notice: Package triggers severe warnings which indicate that it
• may exhibit random runtime failures.
• libavcodec/h264.c:3630:31: warning: array subscript is below array bounds

• QA Notice: Package triggers severe warnings which indicate that it
• may exhibit random runtime failures.
• libavcodec/libx264.c:351:22: warning: the address of ‘val’ will always evaluate as ‘true’

• Please do not file a Gentoo bug and instead report the above QA
• issues directly to the upstream developers of this software.
• Homepage: ​http://ffmpeg.org/

[Carl Eugen Hoyos]

Do you believe any of the notices is valid?
(The notice that val being a char[] and not a char* indicates random runtime failures seems wrong to me.)

[Toralf Förster]

Replying to cehoyos:

Do you believe any of the notices is valid?

What happens if execute_decode_slices() is called with context_count==0 ?

[Carl Eugen Hoyos]

That would require avctx->active_thread_type to have FF_THREAD_SLICE set and avctx->thread_count being 0 at the same time. Afaict, frame_thread_init() in pthreads.c makes this impossible.
Is this not true for win32 threads or do I miss something?

[Toralf Förster]

Replying to cehoyos:

That would require avctx->active_thread_type to have FF_THREAD_SLICE set and avctx->thread_count being 0 at the same time. Afaict, frame_thread_init() in pthreads.c makes this impossible.

Well, maybe that the gcc here under Linux gives a false hit in this case (although I'm not too familar with the code to be sure).

[Toralf Förster]

May a extend this report to these warnings for version 0.7.6) gcc 4.5.3 in the mean while ?) :

Text relocations force the dynamic linker to perform extra

work at startup, waste system resources, and may pose a security
risk. On some architectures, the code may not even function
properly, if at all.
For more information, see ​http://hardened.gentoo.org/pic-fix-guide.xml
Please include the following list of files in your report:

TEXTREL usr/lib/libavutil.so.50.43.0
TEXTREL usr/lib/libpostproc.so.51.2.0
TEXTREL usr/lib/libavdevice.so.52.5.0
TEXTREL usr/lib/libswscale.so.0.14.1
TEXTREL usr/lib/libavformat.so.52.110.0
TEXTREL usr/lib/libavfilter.so.1.80.0
TEXTREL usr/lib/libavcodec.so.52.122.0

[reimar]

PIC code has a major performance impact on some code on 32 bit x86, so I at least don't really care about those TEXTREL warnings (and object to the overly dramatic gentoo claims) unless they also happen on any system other than 32-bit x86.

[reimar]

Also when compiling with --enable-pic only libswscale, libpostproc, libavfilter and libavcodec will contain TEXTRELs.
I don't think that using --enable-pic is a good idea, but complaining about TEXTRELs when not using it is not fair (not that on architectures that have useable PIC support like x86-64 PIC is automatically enabled shared libraries, and 64 bit builds do not contain TEXTRELs).

[Carl Eugen Hoyos]

I close this ticket.

If I am wrong assuming that it is impossible to call execute_decode_slices() with context_count==0, please send a patch to ffmpeg-devel.

Afaict, the TEXTRELs are wontfix, if you disagree, please move the discussion to ffmpeg-devel.

Concerning automatic, static code checks: They have been proven unreliable in the past, and given that it seems simple to crash FFmpeg / libavcodec with zzuf, they are not worth the effort imo.

Also note that all developers see the gcc warnings, so I am unsure they have to be posted here.