[FFmpeg-devel] [PATCH]Auto-detection for concat demuxer

Carl Eugen Hoyos cehoyos at ag.or.at
Sun Feb 3 13:02:35 CET 2013

Nicolas George <nicolas.george <at> normalesup.org> writes:

> Le quintidi 15 pluviôse, an CCXXI, Carl Eugen Hoyos a écrit :
> > Sorry, but I both fail to understand how your version 
> > is less security-risky than mine and how misdetection 
> > is possible with my version.
> Security: a script containing "file /path/to/sensible/data" would be
> rejected if it was automatically probed, it would only be accepted if the
> user specifies options, either "-safe 0" or explicitly "-f concat".

But this is only / mostly due to other patches, it is 
not related to 3/3 allow probing, or am I wrong?

> Misdetection: file is a very common word in English, especially when
> talking about computing. A lot of text files can have the word
> file in them, including at the beginning of lines. The string "ffconcat
> version 1.0", on the other hand, is not very common, the only reason a
> file would have it as its very first line would be that it is actually a
> file meant for the concat demuxer.

I believe you missed that my patch will not detect 
(all) files that start with "file " but only files 
that (start with "file " and) actually look like 
concat scripts.

> (Note: this very mail has thrice the "file " string at the beginning of
> lines, which would have it detected as a ffconcat script by your patch. The
> same is true for doc/muxers.texi.)

I don't think my patch would detect your mail.

> > I actually think that it is much easier to edit a real 
> > file that is currently correctly detected by FFmpeg to 
> > a file that is misdetected by your version than to make 
> > it a file that is misdetected with my patch.
> I do not get your point here.

My point is that if misdetections plays any role here (I 
did not claim that) my patch has a significantly lower 
chance of leading to a misdetection than yours.
Or in other words: It is trivial to edit a valid file 
so it is still working with current FFmpeg and FFmpeg 
with my patch, but not with your patch, I don't think 
this is as easy for my patch (it is definitely possible 
but I have not yet found a file for which it actually 
does a misdetection).

I am not opposed to your patches, I just want to 
point out that I believe my probe patch has 

Carl Eugen

More information about the ffmpeg-devel mailing list