#765 CLI takes 2 minutes on 2 hour MKV file [gdb backtrace, reproduce link]

[Hans Ecke]

Hi there-

The command line takes about 2 minutes to process a 8 GB 2 hour movie file. This is a fairly standard MKV produced by HandBrake. I expect mediainfo to be nearly instantaneous for it.

If I split the file into chunks (using mkvmerge's --split option), then each chunk takes a correspondingly smaller amount of time. In other words, mediainfo feels it has to scan the entire file. That is strange since all required information should be contained within the header?

Here is one of the 10 minute chunks: (it takes mediainfo 10-14 seconds to ID it)

http://mesoscopic.mines.edu/~hans/mibug10min.dat [666MB]

and here is a 1 minute chunk:

http://mesoscopic.mines.edu/~hans/mibug1min.dat [75MB]

I compiled mediainfo 0.7.63 from source (fedora 64 bit) including debug info. Then I killed the mediainfo process during those 2 minutes, generated a core dump and a backtrace from that. The backtrace is attached.

I also attach the output of the mediainfo CLI command, both the normal and the "--Full" version.

Is there anything else I can do to help?

Thank you for your very useful tool.

Hans

[Hans Ecke]

I don't know how to attach multiple files at once, so here is one of the outputs....

[Hans Ecke]

... and here is the other one

[Hans Ecke]

and here is strace log for running mediainfo on the original file

[Jerome Martinez]

It is due to the AAC stream, indicated in hte Matroska header, but missing in the file (I was looking for it in the whole stream).
Now I limit the count of total frames I parse before giving up
patch

[Hans Ecke]

Thank you! The files were generated by mkvmerge. I'll ask Moritz Bunkus if this is a possible bug in his tools. (You'd think mkvmerge could fix the headers afterwards if a given stream did not write any packets in the whole file)

[Hans Ecke]

(I mean the original was generated by HandBrake, but the same problem exists for a remux via mkvmerge. And Moritz is very responsive)

[Hans Ecke]

https://trac.bunkus.org/ticket/884

[Jerome Martinez]

FYI, in a future version, I may remove the information about the missing stream (as I do for MPEG-TS)