[FFmpeg-devel] [PATCH] avformat/rtmpcrypt: fix discarded const warning
gajjanag at mit.edu
Sun Oct 11 20:15:59 CEST 2015
On Fri, Sep 18, 2015 at 5:02 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
> On Fri, Sep 18, 2015 at 09:50:29PM +0200, Nicolas George wrote:
>> Le jour du Génie, an CCXXIII, Ganesh Ajjanagadde a écrit :
>> > This patch silences a -Wdiscarded-qualifiers observed with GCC 5.2.
>> > Signed-off-by: Ganesh Ajjanagadde <gajjanagadde at gmail.com>
>> > ---
>> > libavformat/rtmpcrypt.c | 2 +-
>> > 1 file changed, 1 insertion(+), 1 deletion(-)
>> I am not sure this is correct: the buffer is const for a reason, the warning
>> is right. An application would be completely allowed to give a buffer in
>> read-only memory, or to reuse the contents of the buffer afterwards.
>> Actually, I suspect this muxer, if used as first output in the tee muxer,
>> would cause the next outputs to be corrupt.
> IIRC the code is safe, just ugly
> the writing only occurs if handshaked is set, which is only set
> by ff_rtmpe_update_keystream() which is not part of the public
> interface and only called from libavformat/rtmpproto.c
> i assume but did not double check that libavformat/rtmpproto.c
> calls the functions so that writable buffers are used
>> The correct fix would probably be to allocate a new buffer, probably keeping
>> it in the context for performances reasons instead of allocating each time.
> id need to double check but i think the calling code possibly uses
> the written buffer with the expectation that it has been updated
> if that is so then such fix would break it.
Have you checked the code and confirmed that this patch is fine?
> Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
> What does censorship reveal? It reveals fear. -- Julian Assange
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
More information about the ffmpeg-devel