fix(agent-tooling): search order, --lite umbrella breakage, Rust test false-positives, export-snapshot (closes #218) - #234
Merged
Conversation
The --help epilog and module docstring showed `codelens search . "query"` (workspace first), the opposite of the actual argparse signature (pattern first, workspace second) — every other umbrella command takes workspace first, so this is an easy trap. Following the wrong example does not error: the workspace path silently becomes the search pattern and returns an empty "ok" result with zero indication anything went wrong. Found while exercising search end-to-end across a polyglot workspace (rs/ts/tsx/js/css/html).
`graph` mode's `truncated` field was set to True whenever the query merely contained a LIMIT clause, regardless of whether any rows were actually cut off — a `LIMIT 50` query matching 1 row reported `truncated: true`, misleading callers into thinking more results existed. Now computed from the real row count vs. limit. Added a regression test covering the under-filled-LIMIT case, which the existing test_truncated_flag test never exercised (it only tested LIMIT-with-enough-matches and no-LIMIT-at-all).
Rust unit tests live inline in the same file as production code inside
`#[cfg(test)] mod tests { #[test] fn ... }`, unlike JS/Python where
tests sit in a separate tests/ directory (already exempted by the
existing test-path patterns). Without this, every Rust test function
and every `mod tests` module false-positives as dead code, since the
test harness invokes them via attribute discovery, not a CALLS edge.
Measured on a real 192-file Rust workspace: this was 56%+ of all
Rust registry_dead findings in the top-100 window (100->69 after the
fix, with `mod tests` blocks also newly exempted). Detects #[test],
#[tokio::test], #[async_std::test] and other `#[x::test]` attribute
variants by checking the source lines immediately above the flagged
symbol.
…snapshot (closes #218) export-snapshot was dropped entirely in the #195 umbrella consolidation with no replacement, so import-snapshot could never find a snapshot file to load ("Snapshot file not found" on every run) — the underlying build_snapshot()/write_snapshot() logic in snapshot_io.py was never deleted, just orphaned with no CLI entry point. Added `deps --check export-snapshot` as a new sub-mode (mirroring import-snapshot's own structure), and verified a full export->import round trip preserves node/edge counts. Also excluded both import-snapshot and export-snapshot from the bare `codelens deps <workspace>` default: they are side-effecting/opt-in actions that always failed with no --input/--output given, so every default `deps` run showed a spurious unrelated error entry. See docs/design/0218-export-snapshot.md for the full design rationale.
Two related bugs found while exercising every umbrella command's --lite output on a real 425-file polyglot workspace: 1. _apply_lite's reducer dispatch table predates the #195 umbrella consolidation and keyed off the OLD leaf command names (smell, dead-code, query, ...). Since umbrella commands always pass their own name (context, audit, security, deps, ...), no branch ever matched, and every umbrella's --lite output silently collapsed to `{"status": "ok"}` with all data dropped. Fixed by unwrapping the `{"s","st","r":[{"_check":name,...}]}` envelope, applying the existing per-check reducers keyed by each item's own `_check` name, then re-wrapping. 2. summary/history are not umbrella-wrapped, so they hit the generic fallback instead — which doesn't know their real field names. summary's own job is "anti-overload prioritized findings" per its --help text, but --lite returned the same multi-thousand-token payload as non-lite (nested top_items/flow_chain per finding were never trimmed). history --lite collapsed to just `{"status", "workspace"}`, dropping snapshots/latest/trends/deltas entirely. Added dedicated reducers for both. Also: _force_utf8_stdio() (the issue #179 Unicode-crash fix) rewraps stdout/stderr with TextIOWrapper(encoding='utf-8') but never set newline='', so every '\n' written on Windows silently became '\r\n' (Python's default write-side os.linesep translation) — affecting every line CodeLens prints on Windows. Fixed by adding newline=''. Caught by an existing test that was failing before this fix.
Result of exercising every umbrella command (search/context/audit/ security/deps/impact/api-map/summary/history/doctor/graph) end-to-end against a real 425-file polyglot workspace (Rust/TS/TSX/JS/CSS/HTML Tauri+React app) plus CodeLens's own Python codebase. Documents: the search pattern/workspace argument-order trap, which command to use for which "replace grep" question, --lite reducer coverage per command, per-language verified coverage (including the confirmed gap: no Rust taint analysis), and a changelog of the 6 fixes made this session with before/after verification evidence.
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
3 tasks
Owner
Author
|
CI failures on this PR (`quality-gate` x4, `Test Suite`) are pre-existing on `main`, not caused by these changes — verified by checking `gh run list --branch main`: the merge commit for PR #233 (last merge before this PR) already shows both jobs failing. Root causes filed separately as #235:
Locally verified (Windows, excluding the two known-flaky/segfaulting test files per the documented baseline methodology): all directly-touched test files pass — `test_export_snapshot.py` (new, 7/7), `test_query_graph.py`, `test_deadcode_engine.py`, `test_command_registry.py`, `test_codelens.py` (81/81), `test_issue195_consolidation.py`. Full command outputs and before/after CLI verification for every fix are in the commit messages and `docs/agent-usage-guide.md`. |
4 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Direct verification pass driven by the goal "make CodeLens genuinely usable as an agent's primary tool, replacing manual grep, across js/css/html/react/py/rs/ts" — exercised every umbrella command end-to-end against a real 425-file polyglot workspace (Rust/TS/TSX/JS/CSS/HTML Tauri+React app) and CodeLens's own Python codebase, fixed everything broken directly (no worker per explicit directive), with before/after verification for each fix.
Fixes
search --helpexample order was backwards — docs showedworkspacebeforepattern, opposite of the real signature. Following it silently returns an empty result.graphmodetruncatedflag was meaningless — alwaysTruewhenever a query had aLIMIT, regardless of whether results were cut off.--litewas completely broken for all 12 umbrella commands — reducer dispatch predates the refactor(commands): consolidate 78 commands → 12 focused commands #195 consolidation, keyed off old leaf command names that never match the new umbrella names. Every umbrella's--litesilently returned{"status": "ok"}with all data dropped.#[test]functions false-flagged as dead code — measured 56%+ of all Rustregistry_deadfindings on a real workspace before the fix.deps --check import-snapshotpermanently broken (closes chore(deps): import-snapshot permanently broken since export-snapshot was dropped #218) —export-snapshotwas dropped in refactor(commands): consolidate 78 commands → 12 focused commands #195 with no replacement; restored asdeps --check export-snapshot, verified full round trip.summary/history--litestill returned bloated/empty output even after fix refactor: backend-focused cleanup — remove hardcoded paths, garbage files, frontend deps #3 (they're not umbrella-enveloped, hit the generic fallback which doesn't know their field names)._force_utf8_stdio()never setnewline='', silently turning every\ninto\r\non Windows.Test plan
pytest tests/— all directly-touched test files pass (search, query_graph, deadcode_engine, export_snapshot [new], command_registry, codelens)codelens scan .unrelated to these changes), documented indocs/agent-usage-guide.mddocs/agent-usage-guide.md"Fixed this session" sectionsync_command_count.py --apply— docs stay in sync, count unchanged (export-snapshot is adepssub-check, not a new top-level command)🤖 Generated with Claude Code — direct fix session (no worker), full details in docs/agent-usage-guide.md