Make the idle-reaper, PTY input and late-attach tests deterministic - #456
Merged
Merged
Conversation
Three tests raced real timers or thread scheduling and failed under CI load. - resident_idle_reaper_*: the grace timer runs on the background executor, so `run_until_parked` could declare the mailbox idle before the 1 ms timer landed. The timer callback is now a production entry point, `reap_idle_resident`, which the re-adoption test invokes directly with the stale idle stamp (its negative assertion was vacuous when the timer had not fired), and the shutdown test waits on the reaped state via `run_until`. - input_reaches_the_child_process: `GridEmulator::read` sampled the grid and the exit flag under separate locks, so a snapshot could report the exit while missing the output fed just before it. The exit state is now read while the grid lock is still held, which the test's contract requires. - late_terminal_attach_reproduces_the_host_grid: the line discipline hands a sentinel and its `\r\n` to the PTY master as separate writes, so `wait_for` could return with the cursor move still in flight and the attach frame would differ from a host grid read moments later. `wait_for` now also requires the cursor to sit at column 0 below the sentinel.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Three tests that flaked on CI (each passing on rerun), fixed at the source rather than by lengthening sleeps.
resident_idle_reaper_shuts_down_untouched_provider(Windows): the reaper timer runs on smol's executor and only later hops into the host mailbox, sorun_until_parkedcould return before a 1 ms timer fired. The timer body is nowAppState::reap_idle_resident(production entry point); the shutdown test waits on the observable condition with the existingrun_untilseam, and the re-adoption sibling calls the entry point directly, so its negative assertion is no longer vacuous.term::tests::input_reaches_the_child_process(Linux): production fix.GridEmulator::readsampled the grid, dropped the lock, then readexitedfrom the lifecycle lock; if the output bridge fed the last bytes andset_exitedin between, a snapshot could report the exit with a stale grid (the CI panic showed an empty grid withexited == true).readnow takes the lifecycle lock while still holding the grid lock;set_exited/exited()take only the lifecycle lock, so there is no lock-order cycle. Reproduced by widening the window with a 5 ms sleep on a 2-vCPU loaded container (19/20 failures), 20/20 pass with the fix.late_terminal_attach_reproduces_the_host_grid(Linux): the logged failure ("bash" != "sh"title) predates fix: coalesce terminal burst scrollback across short PTY gaps #448 which pinned titles. The residual hole is the line discipline delivering a sentinel and its\r\nas separate reads, so an attach between them sees a different cursor.Session::wait_fornow also requires the host cursor to sit at column 0 below the sentinel row.Loops (macOS release, and Linux
rust:latest--cpus=2with 4 CPU hogs): each test 30–100 iterations, fulltermandtcode_runtimebinaries repeatedly, 0 failures.cargo fmt --all --check,cargo clippy --workspace --all-targets --locked -- -D warnings,cargo test --workspace --lockedgreen.