This MR changes es_format_t::i_extra from a signed int to size_t.
This variable is used to describe the size of the buffer pointed by
p_extra, and is never strictly negative.
Users of this member are also updated in this commit at once. I'm not
sure it's easy to split, except if callers are casted to size_t before
usage in a preparatory change, and then the cast is removed in a cleanup
change.
Some users were already using size_t though, like videotoolbox decoder
or hxxx_helpers.c. The main blocker for splitting is that most clients
not doing that are actually using pointers to i_extra instead of using a
copy, which is much more verbose to fix locally.
Almost No unneeded cast is added to fix signedness mismatch in this
patch. Most notably, only mpeg/ts_psi.c gets a new cast when using
__MIN() with fmt->i_extra, since the dvbpsi_descriptor_t::i_length
parameter is signed.
avformat/demux.c gets a temporary cast which is no-op with a version
libavutil >= 57 but which would trigger the warning before that. It may
be removed when libavutil < 57 is dropped.
The optional part needed by the decoders.
This will restart the decoders if needed, but they should have received no frames yet
or the hardware decoder may be usable again or not (after a sequence header change).
702b20b0e8 has put 'sensible' defaults
into a 'guessed' table for FAAD2.
According to the spec, lots of channels ordering are wrong and the 8th
values does not exists.
There was no guarantee from bs_read() that the result could not return
0xff itself, which would be like the case 0xf but with ar.sar_width and
ar.sar_height uninitialized. This tells the compiler that everything is
fine.
The alternative solution is to patch the vlc_bits.h file adding a macro
for bs_read() doing:
#define bs_read(s, count) (bs_read(s, count) &
(uint32_t)(((uint64_t)1 << (uint64_t)(count+1)) - 1))
Unfortunately, using count from the macro makes is expansion-unsafe and
using this and-operation inside bs_read() is not enough for the compiler
to ensure that there is no more bits that count, despite the function
being inline.
Fixes the following warnings:
In file included from ../../modules/packetizer/mpeg4video.c:43:
In function ‘h26x_get_aspect_ratio’,
inlined from ‘ParseVOL.isra’ at ../../modules/packetizer/mpeg4video.c:403:9:
../../modules/packetizer/h26x_nal_common.h:117:18: warning: ‘ar.sar_width’ may be used uninitialized [-Wmaybe-uninitialized]
117 | *num = ar->sar_width;
| ~~^~~~~~~~~~~
../../modules/packetizer/mpeg4video.c: In function ‘ParseVOL.isra’:
../../modules/packetizer/mpeg4video.c:364:25: note: ‘ar.sar_width’ was declared here
364 | h26x_aspect_ratio_t ar;
| ^~
In function ‘h26x_get_aspect_ratio’,
inlined from ‘ParseVOL.isra’ at ../../modules/packetizer/mpeg4video.c:403:9:
../../modules/packetizer/h26x_nal_common.h:118:18: warning: ‘ar.sar_height’ may be used uninitialized [-Wmaybe-uninitialized]
118 | *den = ar->sar_height;
| ~~^~~~~~~~~~~~
../../modules/packetizer/mpeg4video.c: In function ‘ParseVOL.isra’:
../../modules/packetizer/mpeg4video.c:364:25: note: ‘ar.sar_height’ was declared here
364 | h26x_aspect_ratio_t ar;
| ^~
Alexandre Janniaux
Tristan Matthews
François Cartegnie
Steve Lhomme
Thomas Guillem
Jonas Dittrich