Make ctx.abort() (for stateless workers) non-experimental.

[kentonv]

All of the concerns in the TODO bullet points had already been addressed:

• We already use TerminateExecution().
• Abort handling has been significantly refactored (some time ago!) closing most ways that code could keep running after abort. One remaining, minor issue is fixed in the first commit in this PR.
• The refactoring of abort handling also ensured that the correct exception propagates.
• All event handler paths were verified as respecting abort.
• The Durable Object version was un-gated a long time ago.

[ask-bonk]

I notice the difference in style: line 1348 uses return ex.clone() (which returns a rejected promise via implicit conversion from kj::Exception to kj::PromiseForResult), while line 1370 uses kj::throwFatalException(ex.clone()). The reason is that at line 1348, we're at the top of the function and the return type is kj::PromiseForResult<Func, Worker::Lock&>, so return ex.clone() works to construct a rejected promise. But inside the .then() callback at line 1370, the return type might be non-promise (e.g., void when the function returns void), so return ex.clone() wouldn't work — kj::throwFatalException is the correct approach. This is sound.

Let me also check the pendingEvent usage in the test to make sure it's properly cleaned up:

Looking at the test: pendingEvent is a local variable, so it gets destroyed when the test function exits. The test calls fixture.drainAndDestroy(kj::mv(request)) at the end, which should handle cleanup properly. The pendingEvent is destroyed before drainAndDestroy is called (since pendingEvent is declared before first and second, and C++ destroys in reverse order — but actually, pendingEvent is destroyed after first and second since it was declared before them... let me re-read).

Actually, looking at the test flow:

1. pendingEvent declared at line 30
2. first promise declared at line 35
3. second promise declared at line 39
4. first.wait() at line 41
5. second.wait() at line 44
6. drainAndDestroy at line 47

After both promises are waited on, pendingEvent is still alive. Then drainAndDestroy is called. Then at scope exit, pendingEvent, second, and first are destroyed in reverse order. This should be fine since the promises have already been resolved/rejected.

This all looks correct. The changes are clean and well-tested. The only potential concern would be the types snapshot, but that's handled by CI (it will fail if the snapshot isn't regenerated). The PR description addresses all the original TODO items that were blocking the feature from going GA.

LGTM

github run

[github-actions]

The generated output of @cloudflare/workers-types matches the snapshot in types/generated-snapshot 🎉

[codspeed-hq]

Merging this PR will not alter performance

✅ 72 untouched benchmarks
⏩ 129 skipped benchmarks¹

---

_{Comparing kenton/enable-ctx-abort (322d59b) with main (5374389)}

Footnotes

1. 129 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩