You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
extractors: smoke-matrix tests across 9 language extractors
From fleet code review 2026-06-01 (`inadequate-tests`, WARN). Each of
`extract_lua.zig`, `extract_nix.zig`, `extract_bash.zig`,
`extract_haskell.zig`, `extract_idris.zig`, `extract_nim.zig`,
`extract_lean.zig`, `extract_text.zig`, and `extract_log.zig` had a
single happy-path test for ~100-200 lines of tree-sitter wrapper code.
A regression that skipped processing on any non-happy path would not
be caught.
Added a smoke matrix per extractor (lengths vary by language semantics):
- All 9: "returns no symbols on empty source" + "handles UTF-8 content
without crashing" — proves the early-empty and binary-safety paths.
- Code extractors (lua, nix, bash, haskell, idris, nim, lean): also
"declaration with no preceding comment (doc_comment null)" and
"attaches preceding comment as doc_comment" — proves the doc-comment
attachment is conditioned on actual comment presence, not always-on
or always-off.
- extract_lua additionally: "`local function`", "method definitions
(`function obj:method`)", and "`--[[ ]]` block comment" — Lua-specific
syntactic surface that the original happy-path missed.
- extract_text / extract_log: "emits at least one symbol for non-empty
source" — the structured-paragraph / structured-line extractors don't
have doc_comment semantics.
Net: 36 new tests across 9 files (1 → 4-7 per extractor). All 52
test binaries pass.
0 commit comments