Re: [Cellar] On the multiplicity of Info elements

[Moritz Bunkus <[e-mail]>]

Hey,

> There are certainly Matroska files that use CRC-32 within Clusters.

True, I'm just debating their value ;) Sure, you can detect that there's
a bit error in that cluster, but what are you going to do? Drop the
whole cluster? OK, you can inform the user about it and she can watch
the appropriate section in the file in order to determine how bad it is.

My current goal for MKVToolNix is to handle CRCs everywhere:

1. mkvmerge will write them for each level 1 element

2. mkvmerge will check existing CRCs and emit warnings about elements
   where the CRC doesn't match the data

3. mkvpropedit will include them in each level 1 element it changes/adds
   as will the GUI's header/chapter/attachment editors

4. mkvinfo will print existing CRCs and emit warnings about elements
   where the CRC doesn't match the data

Unfortunately all of those require changes in libEBML. The ones required
for 1. and 3. are not that bad (handling of masters in which a CRC
element is already present) and are ABI-compatible.

The changes for 2. and 4. are much more involved, though; for one CRC
elements read from storage must be passed through to the application (at
the moment libEBML tries to handle them transparently while mkvinfo
wants to output the element's presence and data); additionally the
verification is currently broken (the value's correct on disk, though)
and I haven't really understood why that is. It's also very slow. Fixing
this most likely requires changing the ABI, and making it fast/use less
memory bandwidth will require even more ABI-breaking changes.

It'll take some time until this has been implemented.

Kind regards,
mosu