[FFmpeg-devel] MJPEG decoder crash!!!

[Pavel Pavlov]

> On Thu, Jul 22, 2010 at 07:58:40PM -0400, Pavel Pavlov wrote:
> > Hello list, I reported a bug, I actually took time to try to find out what was
> wrong, I posted a bug report and it was simply closed with a very informative
> message saying: .
> > Here's the link to the closed bug report: http://bit.ly/dmUG0V
...
> you read bugreports.html and provide the information that it lists because if
> you dont your bugreport is technically invalid. If its closed or not as invalid
> depends on peoples mood.
> And their mood worsens if you randomly close other bugreports like 2113
> and add unrelated comments to random issues.
> 

 In my organization the way it works is that I cannot close a bug report if it was reported by somebody else. I mark it as fixed and only the person who submitted the report can close it. I always need to remind them to leave some more feedback or to close the bug report. The moment where there was a working solution provided I closed my bug report, because that's something I like when other ppl do and I'm used to doing it this way. I did it purely out of positive intentions, sorry if it was something that somebody didn't like or hated, just like I hate when somebody closed my bug report.

cehoyos closed it initially with "I don't' give a fuck, it works for me" attitude, without any explanation and more importantly without even telling how he made sure it works for him. My request was totally ignored when I asked how did he check if it works for him.
I understand that probably not many people will actually do a test using straight libavcodec api and the bug would happily live in the code till that day when one of the ffmpeg-devs has the same problem; so I had to go other way around, tried to find out how to map that c api usage to ffmpeg command line. Only then It was accepted as not a false positive.


> That said when people provide an analysis of a crash without gdb output
> thats a bad sign, generally meaning the analysis is of low quality or outright
> wrong.


Not all people use gdb. I don't use it, and prefer to avoid it if possible. 
This time I posted a command line, I uploaded a file and I said that it happens with latest svn ffmpeg build as of july 17. I also wrote that there is no sensible output from ffmpeg cli.
I don't know how helpful gdb output would be in that particular case: it crashes inside GET_VLC which is a huge function-like macro. With ms compiler it shows the line of GET_VLC as the offending one and it's quite useless, plus there are many macro used that generate stack variables that would be mentioned in output and would look quite strange to the code itself (obviously, that code would be very familiar to the author, as he probably spent lots of time debugging it)
Therefore, I consider that I did more or less OK with this bug report, and I don't give a fuck if cehoyos closed it purely out of ignorance and if this bug will ever be addressed. All I can say, from my debug session is that many functions in mjpeg decoder just keep reading way past the end of bit buffer up until it tries to read past page boundary were windows terminates the process.

Sorry if you consider my posts trolling, next time I'll make my bug reports perfectly according to the guidelines and I'll be using gdb.