runCatchingCancellable

Like runCatching but never swallows coroutine cancellation: a CancellationException always propagates, so a cancelled coroutine fails fast instead of being reported as a captured failure. Every other Throwable becomes Result.failure.

Use this, not bare runCatching, in any suspend or coroutine context. Bare runCatching catches CancellationException and converts a cancel into a normal Result, which hides structured-concurrency cancellation — a silent bug, not a safety measure.

But NOT on a long-lived pump — use pumpIn there (#1803)

The name reads as "the cancellation-safe one", and it is, for the case it was built for: a best-effort call with nothing owed behind it. On a pump it is a trap. It discriminates on type, and type cannot separate "my job was cancelled" from "a callee minted a CancellationException and threw it at me" — which is what a consumer writing withTimeout(…) { … } inside Seam.sendTo, Seam.close or Loom.weave does while this job stays alive. Rethrown from a collector, that one cancels the pump instead of failing it: no handler runs, no stack trace is printed, and any report you were about to make never happens. Six instances of that shape were found in this repo, one of them written by an author who had read the issues and cited them in the KDoc of the pump they were breaking. Use pumpIn, which discriminates at runtime with ensureActive() and also guards the upstream half this helper cannot reach at all.