This is a clean revert. The original commit (a6fa6a5) added ref = addRef() captures in two lambda closures to prevent the owning ReadableStream (and thus the controller) from being garbage collected while a read promise was pending. The revert removes those captures.

The PR description explains the rationale: it conflicts with an internal floated patch and will be re-landed after resolving that conflict.

The key question is whether this revert reintroduces a use-after-free risk. The original commit message says "In some edge cases, the this can be collected while the promise is still pending." The controller captures this in the lambdas, and owner is a bare kj::Maybe<ReadableStream&> (non-owning reference). Without the addRef() capture, if the ReadableStream JS object gets GC'd while the kj::Promise is in flight, the controller could be destroyed and the lambda would dereference a dangling this.

However, this is the author reverting their own commit with a clear explanation that it's temporary (conflicts with an internal patch, will be re-resolved). The code existed without the addRef() captures before the original commit landed (just yesterday), so this is restoring the prior state temporarily.

LGTM

github run