Repository navigation
Log a one-time connection's end when it was replaced or disposed - #1483
Conversation
#onOneTimeChanged dropped every move of a runtime that was no longer the current one before it reached the access log. Opening a new link while a phone was confirming the old one therefore left no one-time-denied entry, and a connection the service ended as it was disposed (a VS Code window closing, a quit) left a one-time-connected entry with no one-time-ended. Log first; only publishing state stays limited to the current runtime.
Deploying mouseterm with
|
| Latest commit: |
b11effe
|
| Status: | ✅ Deploy successful! |
| Preview URL: | https://7852b825.mouseterm.pages.dev |
| Branch Preview URL: | https://fix-one-time-access-log-ends.mouseterm.pages.dev |
dormouse-bot
left a comment
dormouse-bot
left a comment
There was a problem hiding this comment.
CI is red only on Argos: argos/storybook-chromium and argos/storybook-webkit each report "1 changed — waiting for your decision". The diff doesn't touch anything Storybook renders. No story imports lib/src/host/remote/service.ts; the remote stories import only types from service-protocol. The same "1 changed" diff is pending on #1479, #1480, and #1482, so it looks like a shared baseline change rather than something this PR caused. It needs a maintainer's accept or reject in Argos (build 2039, build 2040). Every other check is green, apart from the Hosted PR preview deploy jobs, which wait for approval and were not polled.
`PairedPhonesAndAccessLog` built its log in a `parameters` literal, which runs at module load, before the meta's `beforeEach` pins `Date.now` to `STORY_NOW`. The log therefore carried the CI runner's wall-clock time, and every PR showed an Argos diff on `modals-remotecontrolsection--paired-phones-and-access-log`. `pairedDevices` and `accessLogEntries` now take an optional `now`, and the story passes `STORY_NOW`; `OneTimeEndedNetworkNotAllowed`'s refusal time gets the same treatment. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
dormouse-bot
left a comment
dormouse-bot
left a comment
There was a problem hiding this comment.
The only red checks are the Argos ones: argos/storybook-chromium (build 2048) and argos/storybook-webkit (build 2047), each "1 changed". Every other check passed. This branch now carries #1487, which moves the access-log story off the runner's wall clock. Its description expects exactly one more baseline change, and #1487's own Argos builds (2043, 2044) report the same "1 changed". Accepting that diff once in Argos should keep the story stable, and this PR's change doesn't touch anything Storybook renders.
The access log from #1459 missed the end of a one-time connection in two cases.
BurrowService.#onOneTimeChangedreturned early for any runtime that was no longer#oneTime, before calling#logOneTime, and both cases run through a runtime that has just stopped being current:#openOneTimeswaps the new runtime in, then ends the old one. The phone that was confirming gets noone-time-deniedentry.docs/specs/remote-security-model.md→ "Access log" lists a one-time connection's refusal as an entry.dispose()clears#oneTime, then ends the runtime. This happens when a VS Code broker window closes or the app quits, and it leaves aone-time-connectedentry with noone-time-ended.The fix logs before the identity check. Only publishing state to the webviews stays limited to the current runtime. A runtime that is no longer current emits exactly one more state,
ended, sinceend()is synchronous and nothing after it announces.#oneTimeSeenstill holds that runtime's label at that point, because the replacement has not announced anything yet.Four new
service.test.tscases cover the log: a connection and its end, the disposed connection, the replaced confirming phone, and a link no phone confirmed, which logs nothing. The disposed and replaced cases fail onmain. All four pass here, as does the rest oflib/src/host/remote/exceptnative-direct-peer.test.ts. Its four WebRTC cases time out in this sandbox onmainas well, because they need a real network path.This branch also merges #1487's branch, which pins the access-log story's timestamps to the story clock. Without it the Argos check on this PR stays red on that story's wall-clock diff, which has nothing to do with this change. Whichever of the two lands first, the other's copy of that commit becomes a no-op.