Fix #612

[julienmalik]

Works for me under MemorySanitizer

[julienmalik]

Following @boxerab remark in #612 , I'm testing this fix.
It does avoid the uninitialized read.

I don't have much idea about the actual consequences, and about why this test was here in the first place.

So this needs to be validated by the experts :)
ping @detonin @mayeut

[julienmalik]

The test for bpno > 0 dates back to the early days, and appeared in

commit 2ad6a9770a867ddc3ecc31f61a248152b7c6eeac
Author: Sebastien Lugan <[email protected]>
Date:   Thu Nov 27 10:10:17 2003 +0000

    Initial revision

The available history does not go further in the past.

[boxerab]

Looks like a 13 year old bug :) Surprising.

[mayeut]

My knowledge doesn't go that far... @detonin, any idea ?

[detonin]

I've started investigated this bypass and restart problem and created a branch for this: https://github.com/uclouvain/openjpeg/tree/fix-bypass-restart
The whole function t1_encode_cblk would actually need a major refactoring.
For now, the RESTART mode seems to be fixed and the BYPASS mode is behaving better but not lossless yet.
The branch still needs refactoring before merging into master. I'll continue working on it.
ping #612 #674

[detonin]

More specifically, part of the problem comes from these lines:
https://github.com/uclouvain/openjpeg/blob/fix-bypass-restart/src/lib/openjp2/t1.c#L1689-L1697
that have been introduced in March 2007 (surprising indeed):
5a3c1ff

The FF-bytes management is something quite tricky in JPEG 2000 and the additional arithmetic coder termination brought by the BYPASS and RESTART modes can interfere with the way FF-bytes are usually managed (i.e. removed if they are the last bytes of a coding pass).

[rouault]

Fixed per #949