natsukium/felis-fcast
Watch
1
Code Issues 10 Pull requests Projects Releases 1 Packages Wiki Activity Actions

fix(cast): end a recording with the exited session's last screen #19

Merged
natsukium merged 1 commit from fix/record-final-screen into main 2026-10-15 01:31:56 +09:00
Conversation 0 Commits 1 Files changed 3 +73
natsukium commented 2026-10-15 01:28:49 +09:00
Owner
Copy link

Fixes the flaky program-exit recording: output lost between pulls at PTY EOF, and an exit before the subscription never seen. record re-attaches to the exited session and writes its last screen.

Fixes the flaky program-exit recording: output lost between pulls at PTY EOF, and an exit before the subscription never seen. record re-attaches to the exited session and writes its last screen.
fix(cast): end a recording with the exited session's last screen
All checks were successful
ci / cargo fmt / clippy / test / deny (pull_request) Successful in 53s
Details
ci / cargo fuzz smoke (pull_request) Successful in 1m59s
Details
ci / cargo fmt / clippy / test / deny (push) Successful in 49s
Details
ci / cargo fuzz smoke (push) Successful in 1m58s
Details
20d94b957f
 
A program that prints and exits at once was sometimes recorded without
its output, or not seen to exit at all (the recording ran to
--max-duration). Both showed up as a flaky integration test under load,
roughly one run in ten of the release test binary.

The causes are in how felis ends a session. The daemon sends a
pull-paced subscriber its next cycle only in answer to a pull, and at
PTY EOF it pushes SessionExited and drops its subscribers, so output
that landed between two of the recorder's pulls is never sent. And a
program that exits before the recorder's subscription lands sends that
subscription no SessionExited at all; the attach reports the exit in
SessionInfo.exited instead.

record now treats an attach that reports an exited session as the
program exiting, and on the exit attaches again (not live_only, so it
lands on the screen the session keeps through its post-exit grace) and
writes the size and burst that attach opens with as the last units. The
extra attach is bounded by the follow-up timeout and is skipped silently
when it fails, leaving the recording as before. Asking felis to flush a
final cycle at EOF would fix it at the source but needs a felis release;
this works against the daemon as it is.
natsukium deleted branch fix/record-final-screen 2026-10-15 01:31:56 +09:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
natsukium/felis-fcast!19
No description provided.