[tiper]

Before: a film whose audio runs more than a second past its last video frame never leaves the player. The EOS is pushed when the video lane drains, the pipeline refuses the remaining audio, its clock stops short of the container's duration, and the player sits on the last frame for good because ended waits for the playhead to come within a second of that duration.

After: the pipeline's own end-of-stream event (type=28, PF_EVENT_TYPE_STR_STATE_UPDATE__ENDOFSTREAM, 0x1c, str=true) received after a picture marks the stream played out alongside the existing arithmetic, so the player closes at the credits. A seek and a new session clear it.

0x1c sits exactly at the point where the event numbering shifts between webOS 4 and 5+, so 28 means the same on both (docs/webos5-port.md §5 is updated to say so).

How: a pipeline_eos flag on Shared, raised in sf_on_event_inner only once seen_frame is set, cleared by reset_session and request_seek; the pump's EOS check becomes a pure stream_ended(pos, duration, pipeline_eos). Two new tests: the callback raises the flag only after a picture and a seek clears it, and the pure check ends a film 3 s short of its duration when the pipeline said so.

Device evidence (LG C1, webOS 6.5.3, a 4K AV1 file whose audio ends 3 s after its last video frame). Before, the clock stopped at 5894 s of 5897 s and the player stayed on the last frame:

EOS pushed at true EOF
smp_cb type=28 num=1 str=true
feed a#184400 sz=2560 fed=5896864000000 reply=e qbytes=7680
loop=60 route=player overlay=none pos=5894s play=0pm ...   (repeating)

With this change, same file, same spot:

EOS pushed at true EOF
smp_cb type=28 num=1 str=true
smp: the pipeline reached end-of-stream
EOS reached: playpos=5894s/5897s → ended
teardown spans: stop total=1.4 ...

cargo test --lib -p plx_media: 1688 passed, 0 failed.