fix: consolidate updater, desktop UX, runtime bounds, and CI reliability - #1140
Conversation
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5bed10a3dd
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| match lock_scheduler(&scheduler) { | ||
| Ok(scheduler) => cancel_all_jobs(&paths, &jobs, &processes, &scheduler), |
There was a problem hiding this comment.
Close scheduler admission before the cancellation sweep
When a Submit connection has been accepted but has not reached dispatch when ShutdownNow is observed, this one-shot sweep runs before connections.shutdown() drains that request. Because dispatch does not reject submissions after shutdown, the request can register and enqueue a job after cancel_all_jobs returns; JobScheduler::shutdown then drains queued work, allowing an arbitrary child process to run and immediate shutdown to block indefinitely. Disable scheduler admission before draining connections, or repeat cancellation after the drain.
AGENTS.md reference: AGENTS.md:L65-L69
Useful? React with 👍 / 👎.
| } finally { | ||
| setDetailLoading(false); | ||
| if (generation === requestGeneration.current) setDetailLoading(false); |
There was a problem hiding this comment.
Clear detail loading when invalidating a session request
When the user closes the dock while readRuntimeSessionPage is pending, the close effect increments requestGeneration, so this stale finally branch skips setDetailLoading(false). Reopening the dock does not otherwise reset that flag, leaving every session button disabled and the detail-loading UI permanently stuck for the component's lifetime. Reset the loading state when invalidating requests, or ensure the request that set it always releases it.
Useful? React with 👍 / 👎.
|
|
||
| fn main() { | ||
| println!("cargo:rerun-if-env-changed=MEDUSA_BUILD_COMMIT"); | ||
| println!("cargo:rerun-if-changed=../../../.git/HEAD"); |
There was a problem hiding this comment.
Track the branch ref before embedding the desktop revision
When MEDUSA_BUILD_COMMIT is unset and another commit is made on the same checked-out branch, .git/HEAD still contains the same ref: text, so Cargo does not rerun this build script and reuses the previous cargo:rustc-env value. The rebuilt desktop then reports an old currentRevision, causing same-revision detection to fail and offering/reinstalling an update that is already present. Track the resolved ref file as well, including linked-worktree git directories.
Useful? React with 👍 / 👎.
Problem
Medusa's updater, desktop/TUI lifecycle, provider streaming, daemon shutdown, and PR validation path still had reliability and usability gaps. Separate issue work was also leaving slow or ambiguous CI feedback.
What changed
Unsupportedwhen the host rejects Medusa's isolated environment block instead of inheriting secrets or silently downgrading. Document the host capability requirement.Issue mapping
The parent tracking issue #1123 and foundation issue #1128 are already closed. #1129 remains the integrated-validation tracker until merged-branch checks are observed.
Validation
cargo fmt --all -- --checkcargo clippy --workspace --all-targets --all-features --locked -- -D warningscargo test --workspace --all-features --locked(observedEXIT=0)RUSTDOCFLAGS="-D warnings" cargo doc --workspace --all-features --locked --no-depscargo deny check advisories sources(pass; existing yankedchacha20 0.10.1warning)cargo audit(pass; same yanked warning)git diff --checkReview notes
The installed-binary/package restart path still needs release-artifact acceptance on a clean installed environment. Broader profiling/journal architecture and complete TUI corpus/accessibility audit remain tracked in their issue threads. The Windows host used for local validation does not accept the explicit SBOX environment block; the implementation now reports that limitation as a fail-closed unsupported backend, while the authoritative Windows matrix remains the platform proof for successful sandbox execution.