[teamleaderleo]

Base red fix. #34250 added a hanging makePage to the test fake FakeChromiumHosting using AsyncStream<Void>.makeStream(). The concurrency gate refuses an unbounded AsyncStream buffer (FakeChromiumHosting.swift:44: unbounded AsyncStream buffer (give it a cap and an overflow policy)), so cmux-next fast gates and cmux-next checks have been red on feat-cmux-next since f3a4bab (base runs 41725135574 and 41727385244).

This gives the stream a one-element .bufferingNewest(1) buffer. Nothing ever yields to it: the stream only ends when the task is cancelled. So behavior is unchanged.

Observed / Not run

• Observed: the gate's own rule regex no longer matches the line. python3 scripts/cmux-next/prepush-gate.py reports "no new violations".
• Not run: the Swift test aScriptCallReturnsNilWhenTheChromiumStartNeverEnds. The change only adds a buffer policy to a stream that never receives an element.

Changelog

none

Agent session: agent session

🤖 Generated with Claude Code

---

^{Open in [code]smith · Need help on this PR? Tag @codesmith-bot with what you need.}

---

Summary by cubic

Fixes the red cmux-next gates caused by the unbounded AsyncStream buffer introduced in #34250. The hanging fake Chromium start's stream now gets a one-element .bufferingNewest(1) buffer; since nothing ever yields to it, behavior is unchanged.

• Observed: the gate's rule regex no longer matches the line, and prepush-gate.py reports no violations.
• Not run: the Swift test for the hanging start; the change only adds a buffer policy to a stream that never receives an element.

^{Written for commit 3380e72. Summary will update on new commits.}

[github-actions]

cmux status _{(one comment, edited in place; each line is updated by the job that owns it)}

Review 3380e72: HQ feed card · gallery before/after runs · checks

App of this head (app.zip, kept 1 day)

gh workflow run reload-build.yml -R manaflow-ai/cmux --ref feat-cmux-next -f tag=pr34436 -f ref=lane/perf-limit-fake-chromium-stream-cap -f platform=macos -f backend_mode=local

[teamleaderleo]

Review (claude-sonnet-5-5): SHIP at 77faa0b

Review (subagent claude-sonnet-5-5): SHIP at 77faa0b

Checked: the new call has an explicit policy, so none of the three alternatives of the "unbounded AsyncStream buffer" regex in scripts/cmux-next/check-concurrency.sh match (the makeStream branch only matches empty or of:-only parentheses). makeStream(of:bufferingPolicy:) has defaulted parameters, so AsyncStream.makeStream(bufferingPolicy: .bufferingNewest(1)) infers and compiles. Nothing yields, so the stream still never produces a value and the for-await loop ends only on cancellation, then CancellationError is thrown as before; the continuation stays alive via withExtendedLifetime.

Follow-ups: none

[teamleaderleo]

Closing: #34430 (ceb35d2) landed the same bounded stream in FakeChromiumHosting.swift, and this branch's tree now matches base. The #34250 base red is fixed there.