v0.41.53: restart and update behaviour a session can rely on; link add keeps the language decision; skill command removed #587

Merged

buildagent merged 101 commits from integration-1114 into master 2026-11-14 11:54:08 +01:00

Conversation 0 Commits 101 Files changed 120 +6190 -1247

buildagent commented 2026-11-14 00:43:40 +01:00

Member

Copy link

Release train after v0.41.52: seventeen lanes on top of master, three gate-fix commits and the version bump.

What changes for a user

• Daemon restarts and filters (#584, a customer report). A filter is no longer refused as "the running daemon predates the … filter" after one call met a starting daemon. A call that arrives while a replacement daemon starts waits for it (at most 10 s, never longer than the call's budget) instead of being refused at once; a call that still gives up gets a retryable daemon_restarting with waited_ms and startup_phase in every case.
• Background update (#586, #520). With nothing to update, the update that runs after every daemon start no longer installs and conformance-checks every catalog package again: 43 s → 1.4 s and 584 → 7 complete archive reads on a store with 45 packages. It holds the project's plugin lock only to publish, so approve and link add are no longer refused for minutes; a writer that finds the lock held waits up to 120 s and a refusal names the holder.
• link add (#585). It no longer writes enabled = [] into a new .code-index.toml (which stopped all indexing of the project), and records relationship = "worktree" for another checkout of the same repository.
• Answers while the index settles. search_text, search_symbols, file_outline, read_code and the other tools that do not depend on resolved references answer while references are still being resolved.
• Resolution (#531, step 1). A declaration inside a function body is no longer a target for a reference outside that body (Rust, C#, JavaScript, TypeScript).
• Removed (#571): the code-index skill command. The skill is the MCP resource skill://code-index/SKILL.md.
• Announced (#570): six configuration keys without effect are refused from the next release on; the warning says so.

CI and measurement (#582)

• A red scheduled run of ci.yml or package-ci.yml comments on #582 (job report-red-schedule; report_dry_run input for a dispatched proof).
• An overhead measurement that ends on a saturated runner is repeated once, otherwise NOT MEASURED (contended).
• The quiet-runner check reads the CPU limits of the process's own cgroup hierarchy, so the measurement also runs outside a container.
• Test harness diagnostics for a settle timeout; cargo test no longer waits for cargo deny; downloads retry.

Gate

One full local gate run on the integrated tree cf2d3ee4 (67 steps: every PR job and every nightly-only job of the four workflows except the i686 qemu ABI test). 54 steps passed at once. What it found, all fixed on this branch and re-run:

finding	fix
five tests red in suites no lane had run: pinned ref counts and walker deltas for the new precision-gate decoy fixtures (+2 refs per language, each read at source), the profile vocabulary count, a bare fs::copy of an executable in a new test, a JSON key-order assertion that depends on the build's feature set	6df6e93e
tier-3 baseline: one bind fewer on rust-analyzer	verified at source as a wrong bind the body-local rule removes (mir/pretty.rs:181, a closure parameter f bound to a function f nested in another method); recorded with that reason, d5cbf6b7
no package overhead budget could be measured on a plain host	lane r45, c1dd8d3c

Re-run green after the fixes: fast gates, clippy, the workspace tests (5,701), the C#, JavaScript and TypeScript package legs, the CI script tests and the tier-3 step; after the version bump again the format check, fast gates, script tests, guest gates and tier-3. Six of the seven package overhead budgets pass on a quiet machine; the seventh is Ruby, below. No cost record, corpus baseline or budget was changed; the tier-3 record is the one exception named above.

CI on the PR: four rounds

round	head	result	cause and action
1	6384f9a6	one test red on Linux and Windows	a new assertion counted an info line in the log of a daemon respawned without --log; CI's jobs set CODE_INDEX_LOG=warn. Test fixed in 6fd699c1; the local gate script now exports the jobs' environment.
2	6fd699c1	Windows and pilot green; one test red in cargo test, one in package (typescript)	cargo test: a 60 s settle deadline in a test, while the scheduled nightly ran on the same runner (job 147 min against 82, 104 slow tests against 14). No change. package (typescript): an empty answer right after ready; the log could not say why, so 838cc557 makes the assertion print the daemon's state.
3	838cc557	Windows and pilot green; cargo deny and package (javascript) red	cargo deny: this branch had put cargo install cargo-deny, a build, behind the new 600 s download-retry helper; on a slow runner it timed out three times. Fixed in 8ff34387 with a guard test. package (javascript): the printed state shows the TypeScript package refused at start as extractor.malformed, the 30 s precompile wall-clock budget on a slow runner. Present before this branch; #589.
4	8ff34387	all four workflows green, 44 jobs

Dispatched runs on 8ff34387, one after the other, for the nightly-only jobs:

• ci.yml (run 8726): all 17 jobs green, among them the tier-3 scale run, the grammar rebuild from source, path cost and pool throughput, and the report job.
• package-ci.yml (run 8725): all 69 package and corpus jobs green. The overhead job measured every package after waiting for a quiet runner; six of the seven measured budgets pass (C#, JavaScript, PHP, Python, Rust, TypeScript), Ruby is red at a warm ratio of 1.100 against 1.10 (see below). The report job, in dry-run mode, printed the comment a red scheduled run would post, naming that job and its log.

Known, not changed here

• Ruby overhead budget. The warm ratio of ruby-sinatra measured 1.115, 1.041, 1.116 and 1.063 in four runs on a quiet machine against a limit of 1.10: the verdict depends on the run. On earlier releases ruby-sinatra was red in seven of the eight nightly CI runs inspected for #582; in five of those its cold budget passed, so the warm ratio was what failed. Nothing in this branch touches Ruby's warm path. The other six budgets pass: C# at 1.127 against its limit of 1.15, the rest at 1.06 or lower against 1.10 or 1.20 (PHP and TypeScript, red on the loaded runner in run 8657, measure 1.04 here).
• #586: languages enable, plugin add and languages list still read installed archives more often than needed.
• #589, found by this PR's CI. A precompile that exceeds 30 s of wall-clock time refuses a valid package as extractor.malformed for the daemon's lifetime. On a slow runner this reddens named_script_imports_follow_configuration_exports_and_daemon_restart (two of four rounds here). Fix prepared on a lane (CPU-time budget on Linux and macOS, truthful reason everywhere), not in this branch.
• #588, found while investigating the same test. A worker deadline during a file's first extraction leaves the file empty until it is edited. Reproduced by suspending a worker; not seen in CI. Fix prepared on a lane, not in this branch.
• #570: test timing hooks become debug-only in the next release (lane finished, not in this branch).

🤖 Generated with Claude Code

https://claude.ai/code/session_014XLNPso9cCvzGq5Kt72PSr

Release train after v0.41.52: seventeen lanes on top of master, three gate-fix commits and the version bump. ## What changes for a user - **Daemon restarts and filters (#584, a customer report).** A filter is no longer refused as "the running daemon predates the … filter" after one call met a starting daemon. A call that arrives while a replacement daemon starts waits for it (at most 10 s, never longer than the call's budget) instead of being refused at once; a call that still gives up gets a retryable `daemon_restarting` with `waited_ms` and `startup_phase` in every case. - **Background update (#586, #520).** With nothing to update, the update that runs after every daemon start no longer installs and conformance-checks every catalog package again: 43 s → 1.4 s and 584 → 7 complete archive reads on a store with 45 packages. It holds the project's plugin lock only to publish, so `approve` and `link add` are no longer refused for minutes; a writer that finds the lock held waits up to 120 s and a refusal names the holder. - **`link add` (#585).** It no longer writes `enabled = []` into a new `.code-index.toml` (which stopped all indexing of the project), and records `relationship = "worktree"` for another checkout of the same repository. - **Answers while the index settles.** `search_text`, `search_symbols`, `file_outline`, `read_code` and the other tools that do not depend on resolved references answer while references are still being resolved. - **Resolution (#531, step 1).** A declaration inside a function body is no longer a target for a reference outside that body (Rust, C#, JavaScript, TypeScript). - **Removed (#571):** the `code-index skill` command. The skill is the MCP resource `skill://code-index/SKILL.md`. - **Announced (#570):** six configuration keys without effect are refused from the next release on; the warning says so. ## CI and measurement (#582) - A red scheduled run of `ci.yml` or `package-ci.yml` comments on #582 (job `report-red-schedule`; `report_dry_run` input for a dispatched proof). - An overhead measurement that ends on a saturated runner is repeated once, otherwise `NOT MEASURED (contended)`. - The quiet-runner check reads the CPU limits of the process's own cgroup hierarchy, so the measurement also runs outside a container. - Test harness diagnostics for a settle timeout; `cargo test` no longer waits for `cargo deny`; downloads retry. ## Gate One full local gate run on the integrated tree `cf2d3ee4` (67 steps: every PR job and every nightly-only job of the four workflows except the i686 qemu ABI test). 54 steps passed at once. What it found, all fixed on this branch and re-run: | finding | fix | |---|---| | five tests red in suites no lane had run: pinned ref counts and walker deltas for the new precision-gate decoy fixtures (+2 refs per language, each read at source), the profile vocabulary count, a bare `fs::copy` of an executable in a new test, a JSON key-order assertion that depends on the build's feature set | `6df6e93e` | | tier-3 baseline: one bind fewer on rust-analyzer | verified at source as a wrong bind the body-local rule removes (`mir/pretty.rs:181`, a closure parameter `f` bound to a function `f` nested in another method); recorded with that reason, `d5cbf6b7` | | no package overhead budget could be measured on a plain host | lane r45, `c1dd8d3c` | Re-run green after the fixes: fast gates, clippy, the workspace tests (5,701), the C#, JavaScript and TypeScript package legs, the CI script tests and the tier-3 step; after the version bump again the format check, fast gates, script tests, guest gates and tier-3. Six of the seven package overhead budgets pass on a quiet machine; the seventh is Ruby, below. No cost record, corpus baseline or budget was changed; the tier-3 record is the one exception named above. ### CI on the PR: four rounds | round | head | result | cause and action | |---|---|---|---| | 1 | `6384f9a6` | one test red on Linux and Windows | a new assertion counted an `info` line in the log of a daemon respawned without `--log`; CI's jobs set `CODE_INDEX_LOG=warn`. Test fixed in `6fd699c1`; the local gate script now exports the jobs' environment. | | 2 | `6fd699c1` | Windows and pilot green; one test red in `cargo test`, one in `package (typescript)` | `cargo test`: a 60 s settle deadline in a test, while the scheduled nightly ran on the same runner (job 147 min against 82, 104 slow tests against 14). No change. `package (typescript)`: an empty answer right after `ready`; the log could not say why, so `838cc557` makes the assertion print the daemon's state. | | 3 | `838cc557` | Windows and pilot green; `cargo deny` and `package (javascript)` red | `cargo deny`: this branch had put `cargo install cargo-deny`, a build, behind the new 600 s download-retry helper; on a slow runner it timed out three times. Fixed in `8ff34387` with a guard test. `package (javascript)`: the printed state shows the TypeScript package refused at start as `extractor.malformed`, the 30 s precompile wall-clock budget on a slow runner. Present before this branch; #589. | | 4 | `8ff34387` | all four workflows green, 44 jobs | | Dispatched runs on `8ff34387`, one after the other, for the nightly-only jobs: - `ci.yml` (run 8726): all 17 jobs green, among them the tier-3 scale run, the grammar rebuild from source, path cost and pool throughput, and the report job. - `package-ci.yml` (run 8725): all 69 package and corpus jobs green. The overhead job measured every package after waiting for a quiet runner; six of the seven measured budgets pass (C#, JavaScript, PHP, Python, Rust, TypeScript), Ruby is red at a warm ratio of 1.100 against 1.10 (see below). The report job, in dry-run mode, printed the comment a red scheduled run would post, naming that job and its log. ## Known, not changed here - **Ruby overhead budget.** The warm ratio of `ruby-sinatra` measured 1.115, 1.041, 1.116 and 1.063 in four runs on a quiet machine against a limit of 1.10: the verdict depends on the run. On earlier releases `ruby-sinatra` was red in seven of the eight nightly CI runs inspected for #582; in five of those its cold budget passed, so the warm ratio was what failed. Nothing in this branch touches Ruby's warm path. The other six budgets pass: C# at 1.127 against its limit of 1.15, the rest at 1.06 or lower against 1.10 or 1.20 (PHP and TypeScript, red on the loaded runner in run 8657, measure 1.04 here). - #586: `languages enable`, `plugin add` and `languages list` still read installed archives more often than needed. - **#589, found by this PR's CI.** A precompile that exceeds 30 s of wall-clock time refuses a valid package as `extractor.malformed` for the daemon's lifetime. On a slow runner this reddens `named_script_imports_follow_configuration_exports_and_daemon_restart` (two of four rounds here). Fix prepared on a lane (CPU-time budget on Linux and macOS, truthful reason everywhere), not in this branch. - **#588, found while investigating the same test.** A worker deadline during a file's first extraction leaves the file empty until it is edited. Reproduced by suspending a worker; not seen in CI. Fix prepared on a lane, not in this branch. - #570: test timing hooks become debug-only in the next release (lane finished, not in this branch). 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_014XLNPso9cCvzGq5Kt72PSr

buildagent added 98 commits 2026-11-14 00:43:40 +01:00

Refuse lexical item targets outside their enclosing bodies a899b3d641

Add lexical containment decoys to the precision gate 16844bf09f

Look up lexical targets through indexed parent and reference links ... 0b2f4892bf

Use the language profile for lexical item visibility. Drive the refusal from indexed enclosing functions and their child symbols, then refs.target_id; empty locals reach no reference rows. Use declaration spans instead of parsing signatures. Keep local impl identity for the source-judged #579 fixture: generic V::Value must not bind owner::V::Value.

Register lexical refusal reference stances and C# local functions ... 0e8bb42e51

The refusal is kind blind; type refs alone establish local impl self identity. LexicalScope.cs proves C# local functions are emitted as function rows, so promote the producer stance to Emitted.

Document inherited lexical item visibility defaults 58c5c26a30

Decode and encode the lexical visibility profile field 4fb932e3c5

Preserve methods reached through escaping body-local values ... 0177f06f31

A local type name is lexical, but trait dispatch can reach its methods outside the declaring function. Exercise a Debug implementation returned as an opaque value; admitting methods to lexical refusal makes this fixture fail. Keep receiver and impl identity gaps separate from name containment.

fix(mcp): qualify absence by the registered tool dependency c4b898bddb

docs: distinguish parsed absence from pending resolution 91d2a1fd62

docs: keep settlement contract within the envelope reserve 0ecf930583

test(mcp): use the same text query before and after parsing 71aeb6f51f

State that reference counts stay lower bounds while references resolve ... 58eb2abd42

The tools whose absence no longer waits for resolution still carry
reference-derived counts in their rows; the state semantics and the
envelope reference now say so.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Merge lane r30: absence answers do not wait for reference resolution when the tool does not need it ... cad3f69c88

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Merge lane r29: a body-local declaration is not a target outside its body (#531 step 1) ... a5614e55b5

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

fix: wait for project plugin writers with bounded holder diagnostics b00602b5b7

ci: run tests independently of cargo deny c77c7e7355

ci: label deny failures using tool diagnostics e689c8284b

ci: retry tool and target downloads with a bounded helper e27f2a8bd9

ci: require successful producers before consuming their artifacts 644e49ffb1

ci: report refused idle admission as not measured ca081701f1

test: retain daemon startup stderr and report exit status 556f4e9dc8

ci: use bounded downloads for toolchain bootstrap too 2a44fb1389

ci: guard remaining prebuilt scale and guest consumers c08ea05218

docs: keep envelope explanation within reserved headroom 382691c0f0

test: pin mixed-load ladder to rust-analyzer metadata input 1cefc032bd

test: register plugin lock holder metadata disclosure f7c53afca3

test: exercise extracted deny and measurement workflow commands cde76cdb32

fix: keep MCP runtime available while waiting for plugin writers c07156265b

ci: preserve fail-open scope syntax alongside producer guards d0a61a07f9

test: write startup fixture through the executable fork lock 5ed25995a3

ci: keep a refused measurement red and leave the Windows bootstrap unchanged ... 6ca605041b

A refused runner admission ends the measurement step with its own exit
status and the line NOT MEASURED; green would make an absent measurement
look like a passed one. The steps that read the report still skip.
The download retry helper needs Python, which no step of the Windows
runner uses today, so the Windows bootstrap in ci-windows.yml and
release.yml keeps its direct commands.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Merge lane r32: scheduled and release CI survive network failures and print their state (#582) ... e284dac22b

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Announce once that a plugin writer is waiting, and for whom ... f8e40c50bf

A wait of up to the limit with no output reads as a hang.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Merge lane r31: plugin writers wait for the project lease and a refusal names its holder (#520, #582) ... f56443db96

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Expose startup phases and elapsed time in reconcile detail f959fd603c

Diagnose MCP warm-up and settle deadlines with reply history and child logs 1fd9d0c0d8

Wait for delayed discovery before checking startup timeout diagnostics 3d66257fae

Keep startup phase documentation within envelope headroom d7903184cb

fix(daemon): scope definitive filter capabilities to daemon identity 23f6890932

Merge remote-tracking branch 'origin/master' into integration-1114 816f47dc6d

fix(daemon): retain the daemon diagnostic on typed RPC errors d1c9e5006f

fix(daemon): discard capabilities when the identity token becomes known f0f2dc3671

Keep the envelope reference intact and show child stderr for every failing MCP test ... 38a8c892bf

The lane made room for the startup phase vocabulary by deleting documented
facts from the envelope reference (the per-call refusal, the schema enum note,
the batching advice). The reference is restored unchanged; the phase names are
listed in docs/operations.md and the payload names the phase in plain words.

The daemon takes its census after it is ready, so it no longer records a
startup phase no reader can see. StartupPhase::ALL is the one list the
vocabulary test iterates.

The harness now captures the child's stderr in a file; a failing test prints
its tail on drop, so failures outside the settle wait keep the output they had
when stderr was inherited.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Merge lane r33: a startup that does not settle says which phase it is in (#582) ... b069c13a33

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

test(daemon): count guarded capability probes across a real respawn c9bc79584f

fix(config): preserve automatic languages in incidental manifests ... 095640cdc4

Leave enabled commented in the shared template; init alone renders its detected decision. Audit manifest writers against scratch CLI fixtures and discover new callers of the writing sinks. Reproduction and registry both fail when enabled=[] is restored; the registry also rejects an added unregistered writer. Clarify missing-key precedence and existing package-list writers.

fix(link): infer relationships for checkouts sharing a Git repository ... d7986338f2

Compare canonical per-checkout and common Git directories after resolving the target. Infer worktree only without --relationship; preserve explicit choices. Suppress overlap warnings for nested checkouts using the walker policy. A real git-worktree CLI test catches mutations of inference, explicit-flag precedence and nested warnings.

fix(link): report the full entry removed from the manifest ... ccc2e2e7a9

Print the stored name, path, relationship and config file in the same form as link add. Preserve approval-withdrawal output. The real-worktree CLI test fails when removal is reduced to the former name-only message.

test(config): find manifest writers with a text scan instead of a parser ... 0b19c3dcc5

The writer inventory parsed the workspace with tree-sitter through the builtin
Rust plugin, which added a dev-dependency to the CLI crate and tied the test to
a crate that is to be retired. A line scan finds the same twelve call sites;
the new-writer mutation is still caught.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Merge lane r35: creating the config for a link no longer decides languages (#585) ... 2d8ba04d71

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Merge lane r34: a failed capability probe is no longer remembered as an old daemon (#584) ... ab56843c0d

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

test: normalize syntax-proven C# local functions in ctags oracle 09dcea144a

test: count lexical decoy files and declarations in language fixtures 10510fc576

test: own daemon stderr captures outside observed projects 51a6dfae39

fix(cli): prepare background plugin updates outside the project lease d8a61fa30b

docs: describe background updater lease scope and deferral 0bb6ac2865

fix: wait for an identified starting owner during live recovery f78f7c0191

test: cover concurrent real respawn and document restart refusals cda7fc1687

Merge lane r37: the daemon suite is green again after the lexical decoys and the stderr capture ... 5a0062732a

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

fix(ci): retry contended package overhead measurements within a deadline 5bf489659e

Merge lane r38: a call that finds a starting daemon waits for it; restart refusals carry retryable, the wait and the phase (#584) ... 15eab3e51c

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

fix(ci): accept libtest prefixes on overhead records 5e4ae553d9

Merge lane r36: the background update prepares outside the project lease and publishes under it (#520) ... ff377ee9c4

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Use a CPU saturation proxy for overhead ending load 10827d813f

Stream overhead process output while collecting records f5a3a27dd1

Merge lane r39: an overhead measurement on a saturated runner is repeated, then reported as not measured (#582) ... 77fbf1dfe9

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

fix(plugins): reuse current verdicts and deduplicate automatic admission c2c02fb668

docs(plugins): describe automatic update work counts 8d792589aa

Scope archive snapshots to plugin operations and filter update candidates by manifest 06c97e71ce

Reuse held archive verification through admission and conformance grading 313b59a5ae

ci: report repeated scheduled failures on one tracker issue d007045059

ci: run schedule reporter after every workflow job with dispatch preview 9257740b4f

ci: keep checkout-free reporters bounded to script execution 8295330ccc

ci: read all tracker comments when API pagination is capped 665a3e23b7

test(ci): exclude nightly reporting from per-push grading registry 6189566fd8

ci: accept Forgejo dispatch API string inputs in reporter preview c92476cb40

test(ci): retain push matrix legs when excluding the schedule reporter 37eeb99917

Remove skill CLI installation; serve agent skill only over MCP df50404f63

ci: fold repeated nightly reports per workflow, recognised by their marker ... 678508b30c

The reporter looked for the newest comment of its own author. That needs a
user lookup the job token may not be allowed to make, and it splits a series
whenever the other nightly workflow or a person comments in between: two red
workflows would have posted a new comment every night. A report is now found
by its hidden marker, and the newest report of the same workflow decides.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Merge lane r42: a red scheduled run reports itself on the standing tracker issue (#582) ... 347d598706

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Merge lane r40: a background update with nothing to do verifies nothing again and reads each archive once (#586) ... 968a57eb7a

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

docs: keep the I068 design records as they were written ... 2dc775f089

The lane rewrote two historical records to drop the on-disk skill door. They
describe what that mission built; the removal is recorded where it happened
(this lane's commit and the release notes). The citation gate passes with the
records unchanged.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Merge lane r43: the skill CLI command is removed; the skill is served over MCP only (#571) ... 14c57784eb

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Announce next-release refusal of ineffective configuration keys ... ec418dbba9

(cherry picked from commit 84447ee34d232bc18d8e275300ceb3cf565a29ed)

Match daemon restart diagnostics against the shared removal announcement ... 2ebfd28455

(cherry picked from commit 74fe3fd56cc1623ba11d206ad80314f52312d923)

Keep shared configuration warnings quiet during a failed-start streak ... a3fdfe78a5

(cherry picked from commit 0e2d0e0213f981801962b3640b16e46f6a403364)

fix(daemon): wait through owner publication and pending kill teardown 03327bf801

docs: describe recovery before identity publication and after an unrecorded exit 89ca1e11b6

style: drop a stray blank line left by the split ... 07d2d62517

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Merge lane r41 (keys only): ineffective configuration keys announce their refusal in the next release (#570) ... d8eb2571f5

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

style(daemon): express owner polling as a while-let loop e9cf314d48

Merge lane r44: recovery waits through a starting or dying lock holder and reports restarts as retryable (#584) ... cf2d3ee40d

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Handle cgroup v2 root and ancestor CPU quotas c1dd8d3c40

test: settle the pins and helpers the full gate run found red ... 6df6e93e5f

None of these suites was run by a lane; the full gate run on cf2d3ee4 was
their first run on this tree.

- abi_projection: +2 refs per language (C#, JavaScript, Rust, TypeScript)
  from the precision-gate decoy fixtures added in 16844bf0, each read at
  source (two calls or two instantiations per fixture).
- Walker parity deltas on host_fixtures, same fixtures: C# call-arity
  25 -> 27 (a fact on every call), JavaScript scope-verdict 1 -> 2 and
  TypeScript 6 -> 7 (the instantiation outside the declaring function).
- profile_spec: the vocabulary has 36 fields; body_local_names is the
  twenty-fifth boolean.
- automatic_plugin_update: the two executable copies go through
  install_executable and every child of the suite is spawned through the
  fork lock (exec_copy_registry).
- The restart-refusal render test compared the JSON key ORDER, which is
  insertion order in a workspace build (the CLI enables serde_json's
  preserve_order) and alphabetical when the crate is built alone. It now
  compares the sorted key set.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

Merge lane r45: runner capacity is read from the process's cgroup hierarchy, so the overhead measurement also runs outside a container (#582) ... c0f634b589

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

test(corpus): record the tier-3 baseline after the body-local rule (#531) ... d5cbf6b71a

rust-analyzer: rule.tier1a_unique_own_file 32825 -> 32824, with the three
counts that follow from it (resolved, resolved_call, edges, -1 each).

The one bind, found by diffing the previous release's index against this
tree joined on (path, line, col, kind, occurrence) and read at source:
crates/hir-ty/src/mir/pretty.rs:181 calls `f`, the closure parameter of
`with_block`. It was bound to the function `f` nested in the body of
`place` (line 323), which nothing outside that body can name. Step 1 of
#531 removes exactly this class; py-django did not move.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014XLNPso9cCvzGq5Kt72PSr

test(corpus): name the compared release by commit in the tier-3 bless reason ... 3272a92cb5

The bump script rewrites every spelled release version, which would have
turned the reason into a statement about the wrong release.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014XLNPso9cCvzGq5Kt72PSr

release: v0.41.53 ... 6384f9a658

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014XLNPso9cCvzGq5Kt72PSr

buildagent added 1 commit 2026-11-14 02:36:27 +01:00

test(daemon): pin the log level of the respawned daemon in the probe-count test ... 6fd699c12f

`respawn_cfg_brings_back_an_exited_daemon` counts `info` lines in the
daemon log. The respawned daemon gets no `--log`, so it took
`CODE_INDEX_LOG=warn` from the CI job and wrote none: 0 instead of 1 on
Linux and Windows (PR runs 2616, 2617), 1 on every developer machine.
The body now runs in a child process with the level set, the pattern
`two_existing_clients_wait_for_one_real_respawn` already uses, and the
failure prints the log.

Reproduced with `CODE_INDEX_LOG=warn cargo test` before the change
(same panic) and green after; mutation run: without the pinned level
the test is red with the CI message.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014XLNPso9cCvzGq5Kt72PSr

buildagent referenced this pull request 2026-11-14 05:58:03 +01:00

A package worker deadline during a file's first extraction leaves the file empty for good, and the daemon reports ready #588

buildagent added 1 commit 2026-11-14 06:02:06 +01:00

test: print the daemon's state when a ready daemon answers no callers ... 838cc557bc

PR 587's CI read `left: []` at the first assertion of
`named_script_imports_follow_configuration_exports_and_daemon_restart`
and nothing else. The assertion after each daemon start now prints the
symbols, caller counts, stats and daemon log on a mismatch.

Observed red (worker suspended by exact PID until the product's own
30 s extraction deadline): `parse_errors: 1`, `test/calls.test.js`
`package refused: host.deadline_exceeded (process)`, `refs: 0`, daemon
`ready`. The product defect behind it is tracked separately.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014XLNPso9cCvzGq5Kt72PSr

buildagent added 1 commit 2026-11-14 07:06:26 +01:00

ci: install cargo-deny outside the download-retry helper ... 8ff343872e

The helper ends an attempt after 600 s and starts the next from
scratch. `cargo install cargo-deny --locked --force` is a build; on
PR 587's run 2626 it did not finish in 600 s on a busy runner, three
times, and the job went red after 30 minutes without checking anything.
The line is back to what master runs.

`test_only_downloads_go_through_the_helper` scans every workflow and
allows the helper only in front of rustup and curl downloads.
MUTATION (RUN): wrap the install again. RED:
`[('ci.yml', 'cargo install cargo-deny --locked --force')] != []`.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014XLNPso9cCvzGq5Kt72PSr

buildagent referenced this pull request 2026-11-14 07:07:51 +01:00

A slow host makes the 30 s precompile budget refuse a valid package as extractor.malformed for the daemon's whole lifetime #589

buildagent merged commit 59b8bc3d2d into master 2026-11-14 11:54:08 +01:00

buildagent referenced this pull request from a commit 2026-11-14 11:54:09 +01:00

Merge PR #587: v0.41.53, restart and update behaviour a session can rely on

buildagent referenced this pull request 2026-11-14 12:28:46 +01:00

The nightly runs have been red on 11 of 13 completed runs since 2026-11-06 and nothing reports it: overhead budgets, two intermittent startup failures, a step-guard cascade #582

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels   

code-review

  

correctness

  

dos

  

performance

  

security

  

severity/high

  

severity/low

  

severity/medium

  

tech-debt

  

Kind/Breaking

Breaking change that won't be backward compatible

  

Kind/Bug

Something is not working

  

Kind/Documentation

Documentation changes

  

Kind/Enhancement

Improve existing functionality

  

Kind/Feature

New functionality

  

Kind/Security

This is security issue

  

Kind/Testing

Issue or pull request related to testing

  

Priority

Critical

The priority is critical

  

Priority

High

The priority is high

  

Priority

Low

The priority is low

  

Priority

Medium

The priority is medium

  

Reviewed

Confirmed

Issue has been confirmed

  

Reviewed

Duplicate

This issue or pull request already exists

  

Reviewed

Invalid

Invalid issue

  

Reviewed

Won't Fix

This issue won't be fixed

  

Status

Abandoned

Somebody has started to work on this but abandoned work

  

Status

Blocked

Something is blocking this issue or pull request

  

Status

Need More Info

Feedback is required to reproduce issue or to continue work

No labels

code-review

correctness

dos

performance

security

severity/high

severity/low

severity/medium

tech-debt

Kind/Breaking

Kind/Bug

Kind/Documentation

Kind/Enhancement

Kind/Feature

Kind/Security

Kind/Testing

Priority

Critical

Priority

High

Priority

Low

Priority

Medium

Reviewed

Confirmed

Reviewed

Duplicate

Reviewed

Invalid

Reviewed

Won't Fix

Status

Abandoned

Status

Blocked

Status

Need More Info

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

h-dv/code-index!587

No description provided.