Skip padding data

[alfredoyang]

This is from https://bugzilla.mozilla.org/show_bug.cgi?id=1306236.

Some contents may have extra padding data in the boxes of stbl. For example, stsc, stsz, stsd...etc.
This patch skips these paddings so parser keeps parsing the file.

[rillian]

Is this described in a spec anywhere? I confirmed that stagefright skips everywhere this commit adds skip, but I didn't verify that was all the boxes where it does.

What about tests? Verifying this is especially important for undocumented parsing requirements.

[rillian]

I assume the extra indent so iter releases it's reference to src for the subsequent skip? I don't see an alternate solution besides moving this code into a read_description helper function. That might not be a bad idea for readability anyway.

[alfredoyang]

Yes, it releases the borrow for the following skip.

[rillian]

This will skip additional entries where description_count is short. That's probably fine.

[rillian]

Test failure seems to be an infrastructure problem. I'll go ahead and merge; please file new PRs for followup issues.

[alfredoyang]

I'll add test at another PR.

I can't find any statement about the extra padding in 14496-12; however, there are 'free' and 'skip' box defined in spec. I think it should be a special case.

[rillian]

Ok, thanks.

[alfredoyang]

I check the bitstream again. These paddings are 'skip' boxes actually.
The layout is:

stsd box
|- stsd data
|- skip box
|- skip box

stsc box
|- stsc data
|- skip box

stsz box
|- stsz data
|- skip box

I don't think it is correct behaviour because 'stsd' is not a container box. My dump tool MP4Box can't parse this kind of content correctly neither.

There are 2 ways in my mind currently.

1. Add a function to jump over 'skip' box at the end of read_stsd, read_stsc...etc. Just like this pull.
2. Jump over 'skip' box in check_parser_state(). So all the 'skip' boxes will be omitted in read_stbl.

Any idea?

[rillian]

Hmm. I think this is just broken software. These (stsd, stsc, stsz) are all variable-length boxes. Maybe it leaves some padding while writing the header and then fills in skip boxes later?

Do we need to do more than we did with this PR? Skipping to the end of these boxes, regardless of the contents seems safer than trying to interpret them, when we want compatibility with what stagefright accepts.

[alfredoyang]

Ok, we don't need to do anything for these contents. I'll submit another PR for test later.