Function extract_wide_flat_ms

tests/test_extraction.c:4291–4323

Reproduce-first (ms-typescript reallyLargeFile.ts, 2026-07-07): a file * whose root node has hundreds of thousands of FLAT SIBLINGS (580k //// * comment lines in the 3.5 MB fourslash fixture) hung extraction for over * 15 minutes: walk_defs pushed children via index-based ts_node_child(i), * which is O(i) per call in tree-sitter — O(n^2) per wide node (~1.7e11 * iterator steps on the real fil

 * children under the root, plus sparse real defs so the breadth check cannot
 * pass vacuously). Returns elapsed milliseconds; stores the def count. */
static long extract_wide_flat_ms(int n, int *out_defs) {
    const size_t cap = (size_t)n * 24 + (size_t)8192; /* "// wide filler N\n" <= 24 chars */
    char *src = malloc(cap);
    if (!src) {
        return -1;
    }
    size_t off = 0;
    /* CONSTANT def count (10), independent of n: defs that scale WITH n make
     * the fixture superlinear through the separate per-def sibling-scan cost
     * (O(defs x siblings)) — on windows-CLANG64 ASan that pushed the ratio
     * of LINEAR walk code to 43x. Ten spread-out defs keep the breadth check
     * honest while the sibling-scan term stays 10 x n = linear. */
    const int def_stride = n / 10;
    for (int i = 0; i < n; i++) {
        off += (size_t)snprintf(src + off, cap - off, "// wide filler %d\n", i);
        if (i % def_stride == 0) {
            off += (size_t)snprintf(src + off, cap - off, "var wide_a%d = %d;\n", i, i);
        }
    }
    struct timespec a;
    struct timespec b;
    cbm_clock_gettime(CLOCK_MONOTONIC, &a);
    CBMFileResult *r =
        cbm_extract_file(src, (int)off, CBM_LANG_JAVASCRIPT, "proj", "wide.js", 0, NULL, NULL);
    cbm_clock_gettime(CLOCK_MONOTONIC, &b);
    free(src);
    if (!r) {
        return -1;
    }
    *out_defs = r->defs.count;
    cbm_free_result(r);
    return (b.tv_sec - a.tv_sec) * 1000L + (b.tv_nsec - a.tv_nsec) / 1000000L;
}

/* Best-of-N. Timing noise only ever ADDS time, so the minimum of a few runs is
 * the cheapest good estimate of the noise-free cost. A single sample at each