In ID3 v2.3 compressed frames will have a 4-byte data length indicator
after the frame header to indicate the size of the decompressed data.
This integer is unlikely to be a sync-safe integer for v2.3 tags,
only in v2.4 it's sync-safe.
Reversing the unsynchronisation seems to work slightly differently
for ID3 v2.3 tags and v2.4 tags: v2.3 tags don't have syncsafe frame
sizes in the frame header, so the unsynchronisation is applied to
the whole frame data including all the frame headers. v2.4 frames
have sync-safe sizes, however, so the unsynchronisation only needs
to be applied to the actual frame data, and it seems that's what's
being done as well. So we need to undo the unsynchronisation on a
per-frame basis for v2.4 tags for things to work properly.
Fixes extraction of coverart/images from APIC frames in ID3 v2.4
tags (#588148).
Add unit test for this as well.
Whenever we see a gap, we flush the temporary packets (but not the adapter). If we
had some data temporarily stored it will be outputted (the sound will sound a bit
garbled... but that's how it sounds on MacOSX :)
Reverse-engineered by comparing:
* A rtp hinted file provided by DarwinStreamingServer
* The output procued by DSS for that same file
Also used various streaming sources available on the internet to fine-tune
the code.
The header/codec_data extraction methods are from FFMpeg (LGPL).
Use some of the SDP attributes when they are present to specify the output
dimension and framerate. This allows us to receive jpeg frames larger than
2040 width/height.
Fixes#564437
sizeof("foo") includes the string's NUL-terminator in the size returned,
but we're writing strings here with an explicit size at the beginning
and no NUL-terminator. In most cases using sizeof("foo") as length in
memcpy is not harmful, but it is where the string goes right at the
end of our buffer to write, since we don't allocate space for that
NUL terminator.
These filters use information from previous frames to
generate the current frame and a caps change will make
the effect start from the beginning again.
This produces a water ripple effect on the video input,
based on motion or a rain drop algorithm.
Kindly relicensed to LGPL2+ by Kentaro Fukuchi <fukuchi@megaui.net>.
Fixes bug #588695.
This combines the StreakTV and BaltanTV filters from the
effectv project.
Kindly relicensed to LGPL2+ by Kentaro Fukuchi <fukuchi@megaui.net>.
Fixes bug #588368.
This filter adds a radiation-like motion blur effect
to the video stream.
Kindly relicensed to LGPL2+ by Kentaro Fukuchi <fukuchi@megaui.net>.
Fixes bug #588359.
This filter binarizes input frames and combines them with various
optical pattern.
Kindly relicensed to LGPL2+ by Kentaro Fukuchi <fukuchi@megaui.net>.
Fixes bug #588349.
When blending a source layer with an alpha of 'a' on top of another
destination layer we take the sum of:
* 'a' percent of the source layer
* (100 - 'a') percent of the destination layer (the remainder)
Once buffering has started (with an mdat atom), continue buffering
until moov atom is reached, which handles cases with multiple
mdat atoms. Also keep adapter/offset better in sync with upstream
and fix some debug statements. Fixes#587426.
This reverts commit 5503a59a57.
Reverting this since it causes regressions with a lot of sample files
I have, all of which worked fine with the last -good release (#586891).
Whenever we alloc something based on a user-supplied size, we should
really use g_try_new(), otherwise we can easily be made to abort by
passing a ridiculously large number to us for allocing. Fixes
problems with some fuzzed files.
Check the possibly 64-bit atom size more carefully before casting it
to an int and passing it to gst_pad_pull_range(), otherwise we might
end up pulling 0 bytes, getting an empty buffer as requested and
dereferencing not available data whilst thinking we actually asked
for and got 0x1000000000000 bytes. Similar fix for push mode operation
where neededbytes ends up being 0 bytes, which makes us assert. Fixes
crash with broken or fuzzed file (NB #122378).
Fix the caps to include the depth (instead of width twice) in the caps of
audio/x-raw-int.
Fix negotiation to not only copy the rate/channels of the first structure.
Don't call gst_avi_demux_src_convert() for each single index entry. Not
only do we already have the pointer to the stream context, we also know
the formats we want to convert from and to already, so we may just as
well use optimised conversion routines that bypass some of the checks
and lookups made in gst_avi_demux_src_convert().
