summaryrefslogtreecommitdiffstats
path: root/ext/dv/NOTES
diff options
context:
space:
mode:
authorThomas Vander Stichele <thomas@apestaart.org>2001-12-23 16:42:33 +0000
committerThomas Vander Stichele <thomas@apestaart.org>2001-12-23 16:42:33 +0000
commitb0f5f9a9410efd9ce64dfee67dbc950cbbb2d614 (patch)
tree67bc8451ab25d18d75c410cf8cf97e3cdce2d37c /ext/dv/NOTES
parent1c3d0eb6e8f65df18350536dcd3a8e4e4eee3cdc (diff)
adding dv, raw1934, gnomevfs, rtp
Original commit message from CVS: adding dv, raw1934, gnomevfs, rtp
Diffstat (limited to 'ext/dv/NOTES')
-rw-r--r--ext/dv/NOTES13
1 files changed, 13 insertions, 0 deletions
diff --git a/ext/dv/NOTES b/ext/dv/NOTES
new file mode 100644
index 00000000..84211593
--- /dev/null
+++ b/ext/dv/NOTES
@@ -0,0 +1,13 @@
+Packets come from 1394 480 bytes at a time. This is not a video segment
+length. This causes problems, since a packet boundary crossing a video
+segment can split a video segment if we lose an iso packet. We can
+recover from this, sorta, with significant changes to the parser. We have
+to deal with the idea that a) some macroblocks just don't exist (we have
+zero's for them) and b) when any of the 5 macroblocks doesn't exist, we
+can't do pass 3.
+
+Since things are bitstream-based, we can deal with this, but we have to
+add a layer of code that tries to save time (maybe) by not decoding things
+that don't exist. Not sure how this is gonna work with the parse code
+being based on video segments, and not easily splittable into
+macroblock-level parsing (or is it?).