Created attachment 267846 [details] [review]
locking patch

Pushing a buffer list locks the payloader's source pad only
when pushing a buffer. If an intervening flush comes through,
this means only the first few buffers may be pushed before
the soure pad becomes flushing. This could cause a spurious
discontinuity, so we lock the (recursive) stream lock on the
source pad to ensure all buffers are pushed atomically.

However, there are two cases where one needs the flush to
break in the middle of a list:

- when the flush is triggered by a seek from downstream (in
this case, since downstream will block till the original
request finishes, it will not process any further buffers
(if it does both in the same thread anyway)).

- when a state change occurs


Not 100% sure about this. It fixes a spurious discontinuity in a test case, but the need for exceptions feels a bit unfortunate.