When decoding blocks, the YUV planes' pitches were computed using the
target video surface size instead of the block based size, resulting in
decoded plane data being overwritten for some video sizes.
Affected videos are LEOS-11102.bik and LEOS-11152.bik from Myst III.
And fix an out of bounds acces when seeking to the end of a video.
Skipping samples is needed even when seeking through silent edits
because a silent stream is queued for those.
Fixes#10219.
All users of BitStream were in fact using a specific, hardcoded variant,
so we can hardcode that variant, removing the need for virtual calls,
and enabling inlining.
A lot of the standard VideoDecoder methods were still treating the
transparency track as part of the video, so methods like getFrameCount
would return double the amount it should be. This refactoring properly
separates the transparency track into a separate field entirely.
The 16-bit DPCM decompressors in SSCI and Urban Runner use a 16-bit
register to store sample data, without any special handling of
overflow. As such, out-of-range samples simply wrap around, rather
than getting clipped.
It is not totally clear if the wrapping behaviour was intentionally
exploited to handle extreme transients, but in any case, videos
like GK2 5280.VMD that generate samples outside the signed 16-bit
range cause a loud pop when using clipping, but play back correctly
when wrapping.
It turns out that at least one video in Starship Titanic, for the
Lift Indicator, has only a single transparency frame in track 2.
The added code, therefore, when it doesn't find an index entry
for the desired frame number, works backwards until it finds a valid
frame (likely frame 0), and then scans forward. If it hits the end
of the video, then it simply uses whatever last frame it last decoded.
This is needed for Starship Titanic, where videos can have a secondary
video track. It was simpler to use the callback as a means to select
one video track each across two decoders than trying to make VideoDecoder
and/or AVIDecoder support decoding from multiple video tracks simultaneously
Starship Titanic in particular needs this, since some of the videos
have extra junk at the end of the file, such as ycursors.avi,
and parsing fails if we don't respect the filesize field
In SSCI, VMD is drawn by a standard CelObjMem wrapped by a
ScreenItem, giving the location of the bitmap memory to the
decoder. The decoder already supports this, but the API was
previously hidden behind the AdvancedVMDDecoder wrapper
(which is more convenient to use than the VMDDecoder class).
We don't support playing videos with multiple edits backwards. Taking the code
path to move to the next edit when playing backwards sets the current edit index
to an invalid value with the video still trying to play. Which results in out of
bounds reads, and ultimately a crash.
This fixes multiple crashes in Myst. Using the key without the chest on
Stoneship, resetting the clock tower puzzle, and using the switch in the trees
in Channelwood.
This was a regression introduced in a59f5db505.
