The tests that never got to run

Pressing cancel should stop the work you asked to stop. That gets harder when the work has already started other work, with some of it waiting in a queue. You need the stop request to reach the right places without disrupting everything else.

That was the assignment in Boa, a JavaScript engine written in Rust. Flash-Next worked on parent and child cancellation handles, places where running code could check for cancellation, and the connection between evaluations and their queued jobs. There was a substantial implementation to test.

Some repository test dependencies were unavailable, so the agent built a standalone harness. It reported 18 passing tests there. This is a useful way to keep making progress when the usual test route is blocked, and it had helped in our successful Wasmi run too.

But Boa's submitted repository contained an added test that wouldn't compile. It declared let context and then called methods that needed a mutable borrow. Rust asked for let mut context. The official feature tests stopped before they could run.

The recorded score is 0 of 17, but those aren't seventeen observed behavior failures. We don't know how the implementation would fare after that compiler error was fixed. We do know that the separate harness missed a problem in the submitted tree. For a developer picking up this patch, the first job would be getting the actual repository tests running—not trusting the earlier green result to cover them.