fix(k8s): address QA round 1 findings

- [Major 1+2] pass_k8s: always re-extract K8s manifests with CBM_LANG_K8S,
  discarding any cached YAML result; pass already-read source buffer to
  handle_k8s_manifest to eliminate the double file read
- [Minor 3] CONTRIBUTING.md: fix extractor path to internal/cbm/extract_k8s.c
  and clarify tree-sitter YAML grammar usage
- [Minor 4] language.c: add comment documenting intentional case-sensitive
  FILENAME_TABLE and case-insensitive cbm_is_kustomize_file() split behaviour
- [Minor 5] test_pipeline.c: add k8s_extract_manifest_multidoc test pinning
  single-document-per-file extraction behaviour
- [Minor 6] pass_infrascan: replace strlen with strnlen(content, 4096) in
  cbm_is_k8s_manifest to bound the scan
- [Minor 7] extract_k8s: add "crds" to is_kustomize_list_key
- [Nit 8] extract_k8s: remove dead second block_node unwrap in
  extract_k8s_scalars metadata descent
- [Nit 10] extract_k8s: use cbm_arena_strdup for def.label "Resource"

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: shanemccarron-maker

CONTRIBUTING.md

@@ -89,7 +89,7 @@ Language support is split between two layers:
 Languages like **Dockerfile**, **docker-compose**, **Kubernetes manifests**, and **Kustomize** do not use tree-sitter grammars. Instead they follow an *infra-pass* pattern:
 
 1. **Detection helpers** in `src/pipeline/pass_infrascan.c` — functions like `cbm_is_dockerfile()`, `cbm_is_k8s_manifest()`, `cbm_is_kustomize_file()` identify files by name and/or content heuristics (e.g., presence of `apiVersion:`).
-2. **Custom extractors** in `src/pipeline/extract_k8s.c` (or `extract_infra.c`) — hand-written parsers that walk the raw YAML/text and populate `CBMFileResult` with imports and definitions.
+2. **Custom extractors** in `internal/cbm/extract_k8s.c` — tree-sitter-based parsers that walk the YAML AST (using the tree-sitter YAML grammar) and populate `CBMFileResult` with imports and definitions.
 3. **Pipeline pass** (`pass_k8s.c`, `pass_infrascan.c`) — calls the extractor and emits graph nodes/edges. K8s manifests emit `Resource` nodes; Kustomize files emit `Module` nodes with `IMPORTS` edges to referenced resource files.
 
 **When adding a new infrastructure language:**

internal/cbm/extract_k8s.c

@@ -57,7 +57,7 @@ static const char *get_scalar_text(CBMArena *a, TSNode node, const char *source)
 static int is_kustomize_list_key(const char *key) {
     return (strcmp(key, "resources") == 0 || strcmp(key, "bases") == 0 ||
             strcmp(key, "patches") == 0 || strcmp(key, "components") == 0 ||
-            strcmp(key, "patchesStrategicMerge") == 0);
+            strcmp(key, "patchesStrategicMerge") == 0 || strcmp(key, "crds") == 0);
 }
 
 // ---------------------------------------------------------------------------
@@ -196,10 +196,8 @@ static void extract_k8s_scalars(CBMExtractCtx *ctx, TSNode mapping, char *kind_b
             }
         } else if (strcmp(key, "metadata") == 0) {
             // Descend into metadata block_mapping to find "name"
+            // val_node is already unwrapped from block_node above.
             TSNode meta_mapping = val_node;
-            if (strcmp(ts_node_type(meta_mapping), "block_node") == 0) {
-                meta_mapping = ts_node_named_child(meta_mapping, 0);
-            }
             if (ts_node_is_null(meta_mapping) ||
                 strcmp(ts_node_type(meta_mapping), "block_mapping") != 0) {
                 continue;
@@ -269,7 +267,7 @@ static void extract_k8s_manifest(CBMExtractCtx *ctx) {
         CBMDefinition def = {0};
         def.name = cbm_arena_strdup(a, def_name);
         def.qualified_name = cbm_arena_sprintf(a, "%s.%s", ctx->module_qn, def_name);
-        def.label = "Resource";
+        def.label = cbm_arena_strdup(a, "Resource");
         def.file_path = ctx->rel_path;
         def.start_line = ts_node_start_point(mapping).row + 1;
         def.end_line = ts_node_end_point(mapping).row + 1;

src/discover/language.c

@@ -279,6 +279,11 @@ static const filename_entry_t FILENAME_TABLE[] = {
     {"meson_options.txt", CBM_LANG_MESON},
     {"kustomization.yaml", CBM_LANG_KUSTOMIZE},
     {"kustomization.yml", CBM_LANG_KUSTOMIZE},
+    /* Note: FILENAME_TABLE uses case-sensitive strcmp, so mixed-case variants
+     * (e.g. "Kustomization.yaml") are not matched here.  They fall through to
+     * CBM_LANG_YAML and are re-classified by cbm_is_kustomize_file() in
+     * pass_k8s.c, which performs a case-insensitive comparison.  This is the
+     * intended behaviour — no additional entries are needed. */
     {".vimrc", CBM_LANG_VIMSCRIPT},
 };
 

src/pipeline/pass_infrascan.c

@@ -207,10 +207,7 @@ bool cbm_is_k8s_manifest(const char *name, const char *content) {
         return false;
     }
     char buf[4097];
-    size_t n = strlen(content);
-    if (n > 4096) {
-        n = 4096;
-    }
+    size_t n = strnlen(content, 4096);
     memcpy(buf, content, n);
     buf[n] = '\0';
     return ci_strstr(buf, "apiVersion:") != NULL;

src/pipeline/pass_k8s.c

@@ -142,24 +142,15 @@ static void handle_kustomize(cbm_pipeline_ctx_t *ctx, const char *path, const ch
 
 /* ── K8s manifest handler ────────────────────────────────────────── */
 
+/* source/src_len are the already-read file bytes (caller retains ownership and
+ * must free after this call returns). */
 static void handle_k8s_manifest(cbm_pipeline_ctx_t *ctx, const char *path, const char *rel_path,
-                                CBMFileResult *result) {
+                                const char *source, int src_len) {
+    (void)path; /* retained for symmetry; source is always provided now */
     int resource_count = 0;
-    CBMFileResult *res = result;
-    bool allocated = false;
-
-    if (!res) {
-        /* Fall back to re-extraction */
-        int src_len = 0;
-        char *source = k8s_read_file(path, &src_len);
-        if (source) {
-            res = cbm_extract_file(source, src_len, CBM_LANG_K8S, ctx->project_name, rel_path,
-                                   CBM_EXTRACT_BUDGET, NULL, NULL);
-            free(source);
-            allocated = true;
-        }
-    }
 
+    CBMFileResult *res = cbm_extract_file(source, src_len, CBM_LANG_K8S, ctx->project_name,
+                                          rel_path, CBM_EXTRACT_BUDGET, NULL, NULL);
     if (!res) {
         return;
     }
@@ -191,9 +182,7 @@ static void handle_k8s_manifest(cbm_pipeline_ctx_t *ctx, const char *path, const
         resource_count++;
     }
 
-    if (allocated) {
-        cbm_free_result(res);
-    }
+    cbm_free_result(res);
 
     cbm_log_info("pass.k8s.manifest", "file", rel_path, "resources", itoa_k8s(resource_count));
 }
@@ -226,38 +215,20 @@ int cbm_pipeline_pass_k8s(cbm_pipeline_ctx_t *ctx, const cbm_file_info_t *files,
             handle_kustomize(ctx, path, rel, cached);
             kustomize_count++;
         } else if (lang == CBM_LANG_YAML || lang == CBM_LANG_K8S) {
-            /* Need source content for cbm_is_k8s_manifest check */
-            if (cached) {
-                /* Use cached result — the file is a k8s manifest if lang is CBM_LANG_K8S,
-                 * or if we check the source. With a cached result available, trust the
-                 * language field set during discovery. */
-                if (lang == CBM_LANG_K8S) {
-                    handle_k8s_manifest(ctx, path, rel, cached);
+            /* Read source once to classify (and reuse for uncached extraction). */
+            int src_len = 0;
+            char *source = k8s_read_file(path, &src_len);
+            if (source) {
+                if (cbm_is_k8s_manifest(base, source)) {
+                    /* Always re-extract with CBM_LANG_K8S regardless of any cached
+                     * result: cached results were produced during the parallel YAML
+                     * pass and contain no "Resource" definitions.  Pass the already-
+                     * read source buffer so handle_k8s_manifest does not re-read. */
+                    (void)cached; /* cached YAML result intentionally discarded */
+                    handle_k8s_manifest(ctx, path, rel, source, src_len);
                     manifest_count++;
-                } else {
-                    /* CBM_LANG_YAML: need source to confirm apiVersion presence */
-                    int src_len = 0;
-                    char *source = k8s_read_file(path, &src_len);
-                    if (source) {
-                        if (cbm_is_k8s_manifest(base, source)) {
-                            handle_k8s_manifest(ctx, path, rel, cached);
-                            manifest_count++;
-                        }
-                        free(source);
-                    }
-                }
-            } else {
-                /* No cached result — read source to classify */
-                int src_len = 0;
-                char *source = k8s_read_file(path, &src_len);
-                if (source) {
-                    if (cbm_is_k8s_manifest(base, source)) {
-                        /* Pass NULL result — handle_k8s_manifest will re-extract */
-                        handle_k8s_manifest(ctx, path, rel, NULL);
-                        manifest_count++;
-                    }
-                    free(source);
                 }
+                free(source);
             }
         }
     }

tests/test_pipeline.c

@@ -4232,6 +4232,45 @@ TEST(k8s_extract_manifest_no_name) {
     PASS();
 }
 
+TEST(k8s_extract_manifest_multidoc) {
+    /* Two-document YAML separated by "---".
+     * extract_k8s_manifest contains a "break" after the first successful push,
+     * so it processes only the first document that has both kind and
+     * metadata.name.  This test pins that behaviour: the first document's
+     * resource must be present and no crash must occur.
+     *
+     * Note: with some tree-sitter YAML grammar versions the root stream may
+     * expose both documents as siblings; the break still fires after the first
+     * successful def push, so defs.count must be exactly 1. */
+    const char *src =
+        "apiVersion: apps/v1\n"
+        "kind: Deployment\n"
+        "metadata:\n"
+        "  name: my-app\n"
+        "---\n"
+        "apiVersion: v1\n"
+        "kind: Service\n"
+        "metadata:\n"
+        "  name: my-svc\n";
+    CBMFileResult *r = cbm_extract_file(src, (int)strlen(src), CBM_LANG_K8S,
+                                        "myproj", "k8s/multi.yaml", 0, NULL, NULL);
+    ASSERT(r != NULL);
+    ASSERT(!r->has_error);
+    /* First document's resource must be present */
+    int found = 0;
+    for (int i = 0; i < r->defs.count; i++) {
+        if (r->defs.items[i].label && strcmp(r->defs.items[i].label, "Resource") == 0 &&
+            r->defs.items[i].name && strcmp(r->defs.items[i].name, "Deployment/my-app") == 0) {
+            found = 1;
+        }
+    }
+    ASSERT(found);
+    /* At least one def, no more than one (only first document processed) */
+    ASSERT(r->defs.count >= 1);
+    cbm_free_result(r);
+    PASS();
+}
+
 /* ── Envscan tests (port of envscan_test.go) ───────────────────── */
 
 /* Helper: write a file inside a temp dir */
@@ -5182,6 +5221,7 @@ SUITE(pipeline) {
     RUN_TEST(k8s_extract_kustomize);
     RUN_TEST(k8s_extract_manifest);
     RUN_TEST(k8s_extract_manifest_no_name);
+    RUN_TEST(k8s_extract_manifest_multidoc);
     /* Env URL scanning */
     RUN_TEST(envscan_dockerfile_env_urls);
     RUN_TEST(envscan_shell_env_urls);