* test(engine,fuzzy_matcher): add unit tests for branch coverage; fix use_cache(false) double-borrow
Add targeted unit tests to exercise every reachable branch in the `engine`
and `fuzzy_matcher` modules, measured with cargo-llvm-cov's branch coverage
on nightly. Tests use realistic inputs and assert concrete behaviour:
- engine: empty/offset/byte-range matching ranges in the fuzzy engine, the
Frizbee and typo Arinae build paths, AND/OR empty-term filtering, and
split-engine byte-range char exclusion.
- arinae: typo substitutions and deletions, non-ASCII dispatch, prefilter
rejection paths, and direct kernel tests for the DP guards / band-skip /
dead-row pruning that compute_banding makes unreachable through the API.
- clangd/fzy/skim/util: typo-DP substitution, deletion, gap and length-guard
paths; ASCII/non-ASCII dispatch; single-char and score-only paths; and the
assert_order failure diagnostics.
Fix a latent double-borrow bug: `use_cache(false)` in the clangd, skim and
fzy matchers called `RefCell::replace` on cache cells whose `RefMut` guards
were still alive, panicking on every match. Drop the guards before clearing
the caches so the option works (and is now covered by tests).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Q93ttrw4JoXCjBezV2Skmm
* test(fuzzy_matcher): thread guard paths by isolating callees
Cover branches that are reachable only when the inner helper is invoked
directly with inputs the public matchers can never produce:
- clangd `match_bonus` with `Action::Miss` (callers always pass `Match`) —
asserts the 30-point in-segment-after-miss penalty.
- fzy `internal_to_skim_score(SCORE_MIN)` sentinel mapping; the empty-pattern
slow-path `n == 0` guard; and `fzy_score` driven with a non-subsequence
needle so the position backtrace hits the column-0 fallback.
The branches that remain uncovered are now confirmed structurally
unreachable even via direct callee calls: const-generic monomorphization
artifacts, M-cell `!= SCORE_MIN` checks (an M-cell is never exactly the
sentinel after gap accumulation), a match cell at (i>0, j==0) that is always
SCORE_MIN, and short-circuit operands excluded by upstream invariants.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Q93ttrw4JoXCjBezV2Skmm
* test(fuzzy_matcher/skim): thread arg-reachable guards in skim helpers
build_in_place_bonus's `b.len() > 1` and calculate_score_with_pos's
`op.is_none()` are unreachable through the public matcher (the real caller
never passes an empty choice or an over-wide column range), but they ARE
reachable by calling the private helpers directly with such arguments.
Cover both, leaving only genuinely argument-independent dead branches.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Q93ttrw4JoXCjBezV2Skmm
* chore: misc checks & fixes
* fix: default bench arg
* Potential fix for pull request finding
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
* chore: remove magic number
---------
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
