Wait readiness probe - #336
Merged
Merged
Conversation
The reactor backends decided socket wait() completion from cached edge events, which fail in both directions on edge-triggered backends: - False negative: a short read completes without hitting EAGAIN and consumes the readiness edge with no trace; a subsequent wait(read) parks forever although data remains buffered, because EPOLLET/ EV_CLEAR never re-fire without a new transition. - False positive: an edge arriving with no op parked latches the sticky read_ready flag; a speculative read then drains the socket without consuming the flag, and a subsequent wait(read) completes immediately on an empty socket. The select backend latches the same stale flag despite being level-triggered. Decide wait completion by asking the kernel instead: probe with a zero-timeout poll() at initiation (an edge-triggered reactor cannot report a condition that already holds), re-probe when register_op consumes a cached ready flag, and re-probe at event dispatch before completing a parked wait op, keeping it parked on EAGAIN. The probe is side-effect free; in particular it never reads SO_ERROR, which is consume-on-read and belongs to whichever operation observes the failure next. The wait(write) immediate-completion short-circuit is unchanged.
The file mixed three guard patterns: a body-guarded test compiling to an empty function on Windows, two definition regions split by a portable test, and two matching #if blocks in run(). The whole suite registers nothing on Windows, so the guards exist only for compilability; use one scheme: whole definitions guarded, one contiguous POSIX region, one POSIX block in run().
macOS timed out both reactor suites: zero-length datagram sends may be rejected there (nothing queued, so the wait parks), and POLLHUP visibility after the reporting read consumed SO_ERROR is a Linux guarantee. Detect both capabilities at runtime and skip the wait leg when absent instead of hanging; Linux still enforces the full regression witness. The raw poll() check also moves both tests into the POSIX region.
A connect completion decided from SO_ERROR alone reports success for a handshake still in flight: an in-progress connect also reads 0. The check is reachable without a genuine post-handshake event — fresh TCP sockets raise a spurious writable edge at registration, so a loop iteration between open() and connect() latches a stale write_ready that register_op consumes mid-SYN_SENT, and the same stale edge can reach a parked connect through dispatch. Probe writability with a zero-timeout poll() first and report EAGAIN to stay parked; SO_ERROR decides only once the socket is actually writable. The dispatch connect arm re-parks on EAGAIN like every other op instead of completing unconditionally. The regression test holds a connect in SYN_SENT deterministically (backlog-full loopback listener drops the SYN) and drains the ready queue before cancelling so a falsely posted completion cannot be masked as canceled at delivery.
|
An automated preview of the documentation is available at https://336.corosio.prtest3.cppalliance.org/index.html If more commits are pushed to the pull request, the docs will rebuild at the same URL. 2026-08-17 12:43:47 UTC |
|
GCOVR code coverage report https://336.corosio.prtest3.cppalliance.org/gcovr/index.html Build time: 2026-08-17 12:58:32 UTC |
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## develop #336 +/- ##
========================================
Coverage 79.82% 79.82%
========================================
Files 96 96
Lines 5937 5937
Branches 1209 1209
========================================
Hits 4739 4739
Misses 851 851
Partials 347 347 Continue to review full report in Codecov by Harness.
🚀 New features to boost your workflow:
|
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
2 participants
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.
No description provided.