[FFmpeg-devel] New patch for mpegts.c

[Michael Niedermayer]

On Tue, Oct 16, 2012 at 02:33:01PM +0100, JULIAN GARDNER wrote:
> 
> 
> 
> 
> 
> ----- Original Message -----
> > From: Michael Niedermayer <michaelni at gmx.at>
> > To: FFmpeg development discussions and patches <ffmpeg-devel at ffmpeg.org>
> > Cc: 
> > Sent: Tuesday, 16 October 2012, 15:05
> > Subject: Re: [FFmpeg-devel] New patch for mpegts.c
> > 
> > On Tue, Oct 16, 2012 at 01:31:35PM +0100, JULIAN GARDNER wrote:
> >> 
> >> 
> >>  Whilst working on some new additions to MPEGTS support i found that the 
> > current code did not take into account the "Table Version Number", 
> > which caused FFMPEG to process every table that was in the TS. After adding the 
> > new code and validating against a 3 minute TS the counts on tables processed 
> > went down from 1360 to 3.
> > 
> > can you provide a testcase that shows the improvment?
> 
> Remove the check against the version number and run the code with debug and you will see hundreds of messags saying it has processed the table, but with the same version number, put the check back in and rerun and you will 3 messages, well that is it the version number does not change

well, this way we would test different things having a different TS
stream as input and this may lead to different conclusions.

The debug messages can also be suppresed with quite a bit less code
than this patch. So being more verbose my question really is about
a testcase that shows some improvment beyond commenting a debug
message out, like showing a speed gain.


[...]
> >>  +
> >>  +static int parse_section_header(SectionHeader *h,
> >>  +                                const uint8_t **pp, const uint8_t *p_end);
> >>  +
> >>  +typedef void SectionCallback(MpegTSFilter *f, const uint8_t *buf, const 
> > uint8_t *buf_end, SectionHeader *h);
> >>   
> >>   typedef void SetServiceCallback(void *opaque, int ret);
> >>   
> >>  @@ -76,11 +87,14 @@ struct MpegTSFilter {
> >>       int pid;
> >>       int es_id;
> >>       int last_cc; /* last cc code (-1 if first packet) */
> >>  +    uint8_t last_version;
> >>       enum MpegTSFilterType type;
> >>       union {
> >>           MpegTSPESFilter pes_filter;
> >>           MpegTSSectionFilter section_filter;
> >>       } u;
> >>  +    const char *name;
> >>  +    uint8_t tid;
> >>   };
> >>   
> >>   #define MAX_PIDS_PER_PROGRAM 64
> >>  @@ -318,18 +332,45 @@ static void write_section_data(AVFormatContext *s, 
> > MpegTSFilter *tss1,
> >>               }else
> >>                   crc_valid = 2;
> >>           }
> >>  -        if (crc_valid)
> >>  -            tss->section_cb(tss1, tss->section_buf, 
> > tss->section_h_size);
> >>  +        if (crc_valid) {
> >>  +            const uint8_t *p_end;
> >>  +            const uint8_t *p;
> >>  +            SectionHeader h;
> >>  +
> >>  +            av_dlog(ts->stream, "%s: len %i\n", 
> > tss1->name, tss->section_h_size);
> >>  +            hex_dump_debug(ts->stream, tss->section_buf, 
> > tss->section_h_size);
> >>  +
> >>  +            p = tss->section_buf;
> >>  +            p_end = p + tss->section_h_size - 4;
> >>  +
> >>  +            if (parse_section_header(&h, &p, p_end) < 0)
> >>  +                return;
> >>  +
> >>  +            if (h.tid != tss1->tid)
> >>  +                return;
> > 
> > changing tid and parse_section_header handling belongs to 2 seperate
> > patches (and why are these changes needed at all?)
> > 
> I have moved similar code that is used in all callbacks so it is done in this routine, if we go one to add more tables this will function for them as well.
> 
> > 
> >>  +
> >>  +            // Check Version Numbers
> >>  +            if (tss1->last_version==h.version)
> >>  +                return;
> >>  +            tss1->last_version = h.version;
> > 
> > the last_version should only be updated after successfull parsing of
> > the data
> 
> There is no error return from the section callback, so there is no way to know if it is a valid decode!

If iam not missing anything then what your patch does can be
implemented by adding a if(version != last) into the existing code.
and a last=version at the end of the existing code. Which does consider
the successfull parsing.

Also it is uncertain if such version checking itself is safe with real
TS streams "out there". At the least it is missing seek handling and
needs to be tested with concatenated streams that have matching
versions but different content. (Theres a 1 in 32 change that 2 random
concatenated streams will have matching versions, we cant just fail
in 1 out of 32 cases)


> 
> > also how does this interact when things are split into multiple
> > sections?
> > 
> 
> Well as the original code DOES NOT do anything about multiple sections, this and the unpatched code will FAIL. I will look into adding the extra code to process multiple tables. In my view the PAT/PMT will really only evey have 1 section, but it is possible that the SDT could be split over multiple ones. I will find a test stream with this and add code to process this, but

> this will change this decoder fundamentally.

please make sure every change is selfcontaind in a seperate patch and
neccessary for the support of multiple sections.

Thanks

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

There seems to be only one solution to NIH syndrom, ... a shooting squad
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20121016/bb303715/attachment.asc>