Summary
Test/CI/docs/maintainability findings. No single item is critical, but together they explain the repeated source-only-tag history (v0.4.0 macOS link, v0.6.3 lockfile, v0.7.0 authctl --version, v0.7.1 fix-forward) and make the next planner bug harder to review.
Findings
- Dirty tree is fmt-only but release-blocking: 4 files, 34+/12- (
engine.rs:146,1292,4634, catalog_publication.rs:148, p2_delta_refresh.rs:18,58,103,272, shared.rs:590). TESTING.md:252,388 and docs/RELEASING.md:12 require a clean tree — evidence/tags taken now verify nothing. cargo fmt --all + commit.
- God files:
query_tvf.rs 5385 lines / 217 fns; engine.rs 5272 / 163 (single impl Engine from txn → 6 query families query_range:2435, query_latest:2724, query_aggregate:2975, query_window:3528+3791, query_rollup:3972+4085 + PromQL helpers :5063-5248). Triplicated vtab layer (metrics 1320 + logs 1440 + traces 1580 + shadow stores) enforced by convention, not types. Global-registry concurrency concentrated in shared.rs (1116 lines, WriterGate:218, SharedEngine:452, static mutexes) with only inline gate tests (:772-889).
- Test gaps: no
crates/timeless-ext/tests/ and no crates/timeless-codec/tests/ (only inline #[cfg(test)]); TVF planner (decode_args, best_index_args*, metric_candidates) has matcher unit tests but no malformed-argv/bitmask red-team; codec-8 CLP fallback has no dedicated test; tools/demogen (4 of last 8 commits) has no test dir; --include-ignored real-extension contracts are easy to skip (green-but-vacuous run); 5 allow(dead_code) suppressions hide dead-code signal.
- CI hole: no auto-triggers (manual dispatch only,
TESTING.md:424-429); query-contracts.yml runs only harness tests, not workspace/servers/fmt/clippy/shell suites; production-gate.yml misses root fmt/clippy/doc, oracle/sql-equivalents, cli/correctness/crash/dbhealth, libsql-check, embedded example. Split workspaces (root exclude=["servers"], default-members excludes dbhealth-ext, detached tools/*) mean bare cargo test --workspace covers ~3 of ~15 testable crates. No clippy.toml/rustfmt.toml/MSRV pin (toolchain drift already visible as fmt churn).
- Docs drift:
QUERY_RELEASE_REPORT.md:3-32 still says v0.5.0 is latest (stale since 0.7.7/aa258fa); ARTIFACTS.md:11-22 and COMPATIBILITY.md:143-146 stuck at v0.7.6; TESTING.md:52-64 Python inventory omits demogen/screencast.py + 7 bench/*.py so its own test -z "$(git ls-files '*.py')" gate can't pass; dist/ holds only 0.3.0 tarballs. HEAD is 8 past v0.7.7 with growing Unreleased (WAL profile) while AGENTS.md pre-tag checklist is unsatisfied. (WAL docs themselves — GUIDE.md:674-687 vs CHANGELOG.md:13 — are in sync, good.)
Suggested fix
- Commit/revert fmt churn; keep
fmt --check green.
- Refresh
ARTIFACTS/COMPATIBILITY to v0.7.7/unreleased; label QUERY_RELEASE_REPORT archived or refresh; fix TESTING.md Python inventory.
- Extend
production-gate.yml with root fmt + clippy + harness contracts + oracle validate + one correctness.sh smoke (no auto-triggers needed).
- Extract
promql.rs from engine.rs:5063-5248 and query_tvf::planning from :638-870 with malformed-input regression tests — gives timeless-ext its first real tests/ candidate.
Source
Full code review @ bacdf6f, tests/CI/docs track.
Summary
Test/CI/docs/maintainability findings. No single item is critical, but together they explain the repeated source-only-tag history (
v0.4.0macOS link,v0.6.3lockfile,v0.7.0authctl--version,v0.7.1fix-forward) and make the next planner bug harder to review.Findings
engine.rs:146,1292,4634,catalog_publication.rs:148,p2_delta_refresh.rs:18,58,103,272,shared.rs:590).TESTING.md:252,388anddocs/RELEASING.md:12require a clean tree — evidence/tags taken now verify nothing.cargo fmt --all+ commit.query_tvf.rs5385 lines / 217 fns;engine.rs5272 / 163 (singleimpl Enginefrom txn → 6 query familiesquery_range:2435,query_latest:2724,query_aggregate:2975,query_window:3528+3791,query_rollup:3972+4085+ PromQL helpers:5063-5248). Triplicated vtab layer (metrics 1320+logs 1440+traces 1580+ shadow stores) enforced by convention, not types. Global-registry concurrency concentrated inshared.rs(1116 lines,WriterGate:218,SharedEngine:452, static mutexes) with only inline gate tests (:772-889).crates/timeless-ext/tests/and nocrates/timeless-codec/tests/(only inline#[cfg(test)]); TVF planner (decode_args,best_index_args*,metric_candidates) has matcher unit tests but no malformed-argv/bitmask red-team; codec-8 CLP fallback has no dedicated test;tools/demogen(4 of last 8 commits) has no test dir;--include-ignoredreal-extension contracts are easy to skip (green-but-vacuous run); 5allow(dead_code)suppressions hide dead-code signal.TESTING.md:424-429);query-contracts.ymlruns only harness tests, not workspace/servers/fmt/clippy/shell suites;production-gate.ymlmisses root fmt/clippy/doc, oracle/sql-equivalents, cli/correctness/crash/dbhealth, libsql-check, embedded example. Split workspaces (rootexclude=["servers"],default-membersexcludesdbhealth-ext, detachedtools/*) mean barecargo test --workspacecovers ~3 of ~15 testable crates. Noclippy.toml/rustfmt.toml/MSRV pin (toolchain drift already visible as fmt churn).QUERY_RELEASE_REPORT.md:3-32still saysv0.5.0is latest (stale since0.7.7/aa258fa);ARTIFACTS.md:11-22andCOMPATIBILITY.md:143-146stuck atv0.7.6;TESTING.md:52-64Python inventory omitsdemogen/screencast.py+ 7bench/*.pyso its owntest -z "$(git ls-files '*.py')"gate can't pass;dist/holds only0.3.0tarballs. HEAD is 8 pastv0.7.7with growingUnreleased(WAL profile) whileAGENTS.mdpre-tag checklist is unsatisfied. (WAL docs themselves —GUIDE.md:674-687vsCHANGELOG.md:13— are in sync, good.)Suggested fix
fmt --checkgreen.ARTIFACTS/COMPATIBILITYtov0.7.7/unreleased; labelQUERY_RELEASE_REPORTarchived or refresh; fixTESTING.mdPython inventory.production-gate.ymlwith root fmt + clippy + harness contracts + oracle validate + onecorrectness.shsmoke (no auto-triggers needed).promql.rsfromengine.rs:5063-5248andquery_tvf::planningfrom:638-870with malformed-input regression tests — givestimeless-extits first realtests/candidate.Source
Full code review @
bacdf6f, tests/CI/docs track.