[TimWhiting]

Three fixes to std/async, one commit each, each with a test. They cover the same three bugs as #933, #934 and #935; two of them take a different approach, for comparison.

firstof is decided by its first strand (alternative to #933)

An exception in a firstof strand was lost (a timeout returned Nothing), and if every strand raised, firstof aborted with "impossible: empty result list". The result started as Completed([]), which an exception never replaced, and a raising strand did not cancel the others.

Here the result starts as a new Pending variant, and the first strand to finish decides, with its value or its exception, and cancels the others; a canceled strand decides only if no other finishes. interleaved keeps its rules. #933 reuses Canceled as the initial state instead.

Test: test/async/firstof-first.kk

Interleaved strands resume within their cancelation scope (the same change as #934)

A strand that has awaited is resumed by the interleaving's loop, which ran outside the interleaving's cancelation scope, so a timeout in such a strand canceled the caller's scope.

Test: test/async/interleave-scope.kk

A canceled channel receive leaves the channel (alternative to #935)

A canceled receive stayed in the channel's list of listeners, so the next value went to it: the value was lost, or an ended strand was resumed a second time.

Here each listener carries a live flag (a ref, which gives it an identity); its dispose function clears the flag and drops the listeners that are no longer live. #935 threads an integer id through instead. await-noexn becomes public, as receive needs a dispose function without exn.

Test: test/async/channel-cancel.kk

🤖 Generated with Claude Code

[TimWhiting]

This is just another reference point. I ran into this at the same time or just prior to @pigoz, and I used AI to review both my fixes and his approach, and come up with something maybe a bit better than either individually. Still worth understanding the explanations in his PR descriptions and his tests thoroughly, and determining the right fixes. Thanks for your contributions @pigoz!