Repository navigation
fix(mcp): coalesce reads across requests and update Lucide - #4392
Conversation
|
ScriptedAlchemy
left a comment
There was a problem hiding this comment.
Reviewed this update against dependency issue #3099 and current dashboard usage. The importer, package version/integrity, and lockfile snapshot consistently update lucide-react 1.60.0 to 1.61.0; the existing ^1.0.0 range and React 19 peers remain compatible. The public package tarball integrity matches the lockfile, and upstream changes are four added circle-chevron icons with no React runtime/API change. TypeScript checking and the production build pass using the actual 1.61.0 package in an isolated checkout. No defect requiring a patch found. Current-head CI was still queued/in progress when checked; this comment does not claim native CI completion.
Edited/Blocked NotificationRenovate will not automatically rebase this PR, because it does not recognize the last commit author and assumes somebody else may have edited the PR. You can manually request rebase by checking the rebase/retry box above. |
Update
lucide-reactfrom 1.60.0 to 1.61.0.Identical graph-backed MCP reads share an in-flight result across connection-specific request identities. The read key excludes only the reserved
__mcp_request_idcorrelation field; graph identity, scope, tool, and semantic arguments remain part of the key. Handler arguments and cancellation controls retain their original identities.Coalescing eligibility now uses the canonical graph-tool operation classification. Non-graph application calls, including configuration, retained memory, profile registry, and Work calls, each reach their own executor with their own authority and cancellation. Owner side effects and graph reads with independent request controls remain excluded. The production concurrency journey still requires two independent managed test runs plus exactly one file-read leader and one follower.
The scheduler shutdown regression waits for the mounted worker's initial text projection and pending work to settle before planting its terminal publication park. A serving seat can become visible before the worker tail finishes; the existing settling barrier prevents that tail from replacing the injected state. The suppression, shutdown, and join assertions retain their original bounds.
Daemon restart fixtures reuse the endpoint and authority-lock wait before opening a stopped daemon's store. The CLI refusal assertion matches that shared helper while retaining the held-lock refusal and successful replacement after release.
Validation:
pnpm install, pinned rustfmt, whitespace checks, and commit hooks passed.//crates/tracedecay-mcp:unit_test,//crates/tracedecay-code-index-runtime:unit_test,//crates/tracedecay:unit_test, and//crates/tracedecay:mcp_suite, including their dependencies.Full CI on
750ec60590458f7da535eb3b472345a8e17b65d9passed all 169 fleet targets, the Linux aggregate, Clippy, the shipped CLI build, repository gates, feature gates, and Dashboard. This includes the cancellation and shutdown regressions that failed on66cd616. After updating master to40065c38e820e23fad8850603b2142a9f2506e68, the next fleet run passed 168/169 targets. Only host journeys failed: 10 fixture tests probed an overlong profile-local Unix socket instead of the registered short endpoint.Head
093ee7474a5b11fd21d1004417066e96e1314bd0applies the shared fixture correction validated in #4350 and #4388: resolve the daemon endpoint from the exact isolated profile, and force an overlong profile in the existing doctor regression. On this branch, the all-host catalog lifecycle and long-profile doctor journeys both passed (2 tests, 19.39s); strict Bazel Clippy for the host target and dependencies, pinned formatting, dependency installation, and commit hooks passed. An earlier combined-filter invocation selected zero tests and is not counted as validation. No production behavior, deadline, or assertion was relaxed. The branch includes current master; fresh full CI on this pushed head must pass before merge.