Observed on #840's Windows CI leg (2026-07-04): probe_clojure_imports_edge (tests/test_grammar_probe_a.c:434, ASSERT(m.imports >= 1)) failed with 0 extracted imports on windows-latest CLANG64, while all POSIX legs passed and 13 PRs merged through the same Windows gate today. The test's own comment marks it expected RED yet it normally passes — so Clojure :require IMPORTS extraction is nondeterministic, most likely an uninitialized-read / UB in the vendored Clojure tree-sitter external scanner that ASan surfaces occasionally on Windows.
Reproduce-first target: run the clojure grammar probe under ASan in a loop (or valgrind on Linux) to catch the uninitialized read in internal/cbm/vendored/grammars/clojure/scanner.c; fix the scanner (root cause) rather than loosening the probe. Until then it is a known intermittent — a Windows rerun clears it. Unrelated to #840 (TS/JS-only diff).
Observed on #840's Windows CI leg (2026-07-04):
probe_clojure_imports_edge(tests/test_grammar_probe_a.c:434,ASSERT(m.imports >= 1)) failed with 0 extracted imports on windows-latest CLANG64, while all POSIX legs passed and 13 PRs merged through the same Windows gate today. The test's own comment marks itexpected REDyet it normally passes — so Clojure:requireIMPORTS extraction is nondeterministic, most likely an uninitialized-read / UB in the vendored Clojure tree-sitter external scanner that ASan surfaces occasionally on Windows.Reproduce-first target: run the clojure grammar probe under ASan in a loop (or valgrind on Linux) to catch the uninitialized read in internal/cbm/vendored/grammars/clojure/scanner.c; fix the scanner (root cause) rather than loosening the probe. Until then it is a known intermittent — a Windows rerun clears it. Unrelated to #840 (TS/JS-only diff).