DeusData · allanoepping · Aug 2, 2026
@@ -1249,6 +1249,7 @@ CBMFileResult *cbm_extract_file_ex(const char *source, int source_len, CBMLangua
     // Run extractors: defs + imports use separate walks (unique recursion patterns),
     // then a single unified cursor walk handles the remaining 7 extractors.
     cbm_extract_definitions(&ctx);
+    cbm_extract_embedded_defs(&ctx); // defs inside <cfscript> blocks of CFML tag files
     cbm_extract_imports(&ctx);
     cbm_extract_unified(&ctx);
 

@@ -726,6 +726,8 @@ void cbm_channels_push(CBMChannelArray *arr, CBMArena *a, CBMChannel ch);
 // --- Sub-extractor entry points ---
 
 void cbm_extract_definitions(CBMExtractCtx *ctx);
+void cbm_extract_definitions_body(CBMExtractCtx *ctx); // no Module node (for sub-trees)
+void cbm_extract_embedded_defs(CBMExtractCtx *ctx);    // re-parse embedded script blocks for defs
 void cbm_extract_imports(CBMExtractCtx *ctx);
 void cbm_extract_usages(CBMExtractCtx *ctx);
 void cbm_extract_semantic(CBMExtractCtx *ctx);

@@ -7424,6 +7424,20 @@ static void walk_defs(CBMExtractCtx *ctx, TSNode root, const CBMLangSpec *spec,
     free(s.data);
 }
 
+// Walk the tree for functions/classes/fields and module-level variables, WITHOUT
+// emitting the file-level Module node. Split out so embedded sub-trees (e.g. a
+// CFML <cfscript> block re-parsed with the cfscript grammar) can be walked for
+// definitions without minting a spurious Module per block. See
+// cbm_extract_embedded_defs().
+void cbm_extract_definitions_body(CBMExtractCtx *ctx) {
+    const CBMLangSpec *spec = cbm_lang_spec(ctx->language);
+    if (!spec) {
+        return;
+    }
+    walk_defs(ctx, ctx->root, spec, 0);
+    extract_variables(ctx, ctx->root, spec);
+}
+
 void cbm_extract_definitions(CBMExtractCtx *ctx) {
     const CBMLangSpec *spec = cbm_lang_spec(ctx->language);
     if (!spec) {
@@ -7445,9 +7459,6 @@ void cbm_extract_definitions(CBMExtractCtx *ctx) {
     mod.is_test = ctx->result->is_test_file;
     cbm_defs_push(&ctx->result->defs, a, mod);
 
-    // Walk AST for function/class definitions
-    walk_defs(ctx, ctx->root, spec, 0);
-
-    // Extract module-level variables
-    extract_variables(ctx, ctx->root, spec);
+    // Walk AST for definitions + module-level variables
+    cbm_extract_definitions_body(ctx);
 }
@@ -1476,6 +1476,78 @@ static void parse_embedded_imports(CBMExtractCtx *ctx) {
     }
 }
 
+// Re-parse embedded script blocks flagged extract_definitions in embedded_imports
+// and walk them for DEFINITIONS, not imports. Mirrors parse_embedded_imports but
+// runs cbm_extract_definitions_body on each inner AST. Because the inner tree is
+// parsed from a slice starting at the block, its node rows are block-relative;
+// after extracting, we shift the newly-added definitions' line numbers by the
+// block's start row so they map back to the host file. CBMDefinition carries
+// only line positions (no byte offsets), so a constant row shift is a complete
+// remap. Definitions only — calls/usages inside the block are intentionally not
+// walked here (they would need byte remapping too and could pollute the call
+// graph), so this cannot corrupt existing edges.
+void cbm_extract_embedded_defs(CBMExtractCtx *ctx) {
+    const CBMLangSpec *spec = cbm_lang_spec(ctx->language);
+    if (!spec || !spec->embedded_imports) {
+        return;
+    }
+    for (const CBMEmbeddedLangSpec *e = spec->embedded_imports; e->script_node_type != NULL; e++) {
+        if (!e->extract_definitions) {
+            continue; /* this embedded block is walked for imports only */
+        }
+        const TSLanguage *embedded_lang = cbm_ts_language(e->embedded_language);
+        if (!embedded_lang) {
+            continue; /* embedded grammar not linked in — silently skip */
+        }
+        enum { MAX_EMBEDDED_BLOCKS = 64 };
+        TSNode hits[MAX_EMBEDDED_BLOCKS];
+        int hit_count = 0;
+        embedded_collect_content_nodes(ctx->root, e, hits, &hit_count, MAX_EMBEDDED_BLOCKS);
+        if (hit_count == 0) {
+            continue;
+        }
+        TSParser *parser = ts_parser_new();
+        if (!parser) {
+            continue;
+        }
+        if (!ts_parser_set_language(parser, embedded_lang)) {
+            ts_parser_delete(parser);
+            continue;
+        }
+        for (int i = 0; i < hit_count; i++) {
+            uint32_t s = ts_node_start_byte(hits[i]);
+            uint32_t end = ts_node_end_byte(hits[i]);
+            if (end <= s) {
+                continue;
+            }
+            const char *sub_src = ctx->source + s;
+            uint32_t sub_len = end - s;
+            TSTree *sub_tree = ts_parser_parse_string(parser, NULL, sub_src, sub_len);
+            if (!sub_tree) {
+                continue;
+            }
+            CBMExtractCtx sub_ctx = *ctx;
+            sub_ctx.source = sub_src;
+            sub_ctx.source_len = (int)sub_len;
+            sub_ctx.language = e->embedded_language;
+            sub_ctx.root = ts_tree_root_node(sub_tree);
+
+            int defs_before = ctx->result->defs.count;
+            cbm_extract_definitions_body(&sub_ctx);
+
+            /* Shift block-relative line numbers back to host-file lines. */
+            uint32_t row0 = ts_node_start_point(hits[i]).row;
+            for (int j = defs_before; j < ctx->result->defs.count; j++) {
+                CBMDefinition *d = &ctx->result->defs.items[j];
+                d->start_line += row0;
+                d->end_line += row0;
+            }
+            ts_tree_delete(sub_tree);
+        }
+        ts_parser_delete(parser);
+    }
+}
+
 // --- Namespace / package declaration capture ---
 // Java/Kotlin/C#/PHP put the file's symbols inside a namespace/package whose
 // name is NOT reflected in the path-based QN scheme.  Capturing it lets the

@@ -287,6 +287,16 @@ static const char *cfml_branch_types[] = {
     "cf_if_tag",     "cf_elseif_tag",   "cf_else_tag",      "if_statement",
     "for_statement", "while_statement", "switch_statement", NULL};
 static const char *cfml_module_types[] = {"program", "component_file", NULL};
+// The cfml (HTML-derived) grammar keeps the body of a <cfscript> block as an
+// opaque cf_script_content token — it does NOT parse the script-dialect
+// functions inside. Re-parse that slice with the cfscript grammar (extract_
+// definitions = true) so those functions become real definitions. (Contrast the
+// comment above: embedded <cfscript> functions only "appear as function_
+// declaration" once re-parsed here; in the raw cfml tree they are unparsed text.)
+static const CBMEmbeddedLangSpec cfml_embedded_imports[] = {
+    {"cf_script_tag", "cf_script_content", CBM_LANG_CFSCRIPT, true},
+    {NULL, NULL, 0, false},
+};
 
 // ==================== RUST ====================
 static const char *rust_func_types[] = {"function_item", "function_signature_item",
@@ -861,24 +871,24 @@ static const char *graphql_field_types[] = {"field_definition", "input_value_def
 // so the existing ES import extractor sees real import_statement nodes.
 // Terminator: an entry whose script_node_type is NULL.
 static const CBMEmbeddedLangSpec vue_embedded_imports[] = {
-    {"script_element", "raw_text", CBM_LANG_JAVASCRIPT},
-    {NULL, NULL, 0},
+    {"script_element", "raw_text", CBM_LANG_JAVASCRIPT, false},
+    {NULL, NULL, 0, false},
 };
 static const CBMEmbeddedLangSpec svelte_embedded_imports[] = {
-    {"script_element", "raw_text", CBM_LANG_JAVASCRIPT},
-    {NULL, NULL, 0},
+    {"script_element", "raw_text", CBM_LANG_JAVASCRIPT, false},
+    {NULL, NULL, 0, false},
 };
 static const CBMEmbeddedLangSpec html_embedded_imports[] = {
-    {"script_element", "raw_text", CBM_LANG_JAVASCRIPT},
-    {NULL, NULL, 0},
+    {"script_element", "raw_text", CBM_LANG_JAVASCRIPT, false},
+    {NULL, NULL, 0, false},
 };
 static const CBMEmbeddedLangSpec astro_embedded_imports[] = {
     /* Astro component scripts live in the `---` frontmatter fence, which the
      * grammar keeps as an unparsed frontmatter_js_block. Re-parse that slice
      * with the JS grammar so `import X from './X.astro'` becomes a real edge. */
-    {"frontmatter", "frontmatter_js_block", CBM_LANG_JAVASCRIPT},
-    {"script_element", "raw_text", CBM_LANG_JAVASCRIPT},
-    {NULL, NULL, 0},
+    {"frontmatter", "frontmatter_js_block", CBM_LANG_JAVASCRIPT, false},
+    {"script_element", "raw_text", CBM_LANG_JAVASCRIPT, false},
+    {NULL, NULL, 0, false},
 };
 
 // ==================== VUE ====================
@@ -2064,7 +2074,7 @@ static const CBMLangSpec lang_specs[CBM_LANG_COUNT] = {
     [CBM_LANG_CFML] = {CBM_LANG_CFML, cfml_func_types, empty_types, empty_types, cfml_module_types,
                        cfml_call_types, empty_types, empty_types, cfml_branch_types, empty_types,
                        empty_types, empty_types, NULL, empty_types, NULL, NULL, tree_sitter_cfml,
-                       NULL},
+                       cfml_embedded_imports},
 
     // CBM_LANG_GLEAM
     [CBM_LANG_GLEAM] = {CBM_LANG_GLEAM, gleam_func_types, gleam_class_types, gleam_field_types,

@@ -14,6 +14,13 @@ typedef struct {
     const char *script_node_type;  // e.g. "script_element"
     const char *content_node_type; // e.g. "raw_text"
     CBMLanguage embedded_language; // grammar used to re-parse the content slice
+    // When true, the re-parsed inner AST is also walked for DEFINITIONS, not just
+    // imports — e.g. CFML tag components whose <cfscript> body (cf_script_tag ->
+    // cf_script_content) holds script-dialect functions the HTML-derived cfml
+    // grammar keeps as opaque content. Kept on this small struct (a handful of
+    // instances) rather than as a CBMLangSpec field, which would trip
+    // -Wmissing-field-initializers across every language row.
+    bool extract_definitions;
 } CBMEmbeddedLangSpec;
 
 // CBMLangSpec mirrors Go's lang.LanguageSpec with NULL-terminated string arrays.

@@ -666,6 +666,10 @@ static CBMLanguage detect_file_language(const char *entry_name, const char *abs_
     if (dot && strcmp(dot, ".inc") == 0) {
         lang = cbm_disambiguate_inc(abs_path);
     }
+    /* Special: .cfc components may be script-dialect or tag-dialect (<cfcomponent>) */
+    if (dot && strcmp(dot, ".cfc") == 0) {
+        lang = cbm_disambiguate_cfc(abs_path);
+    }
     /* Special: ObjectScript Studio Export XML (<Export generator="...">) is
      * detected by content; otherwise .xml stays XML. */
     if (lang == CBM_LANG_XML) {

@@ -50,6 +50,12 @@ CBMLanguage cbm_disambiguate_cls(const char *path);
  * On read failure, defaults to CBM_LANG_BITBAKE. */
 CBMLanguage cbm_disambiguate_inc(const char *path);
 
+/* Disambiguate .cfc files by reading the head of the content.
+ * Returns CBM_LANG_CFML if the component is tag-based (a "<cfcomponent" tag, or a
+ * leading '<'), otherwise CBM_LANG_CFSCRIPT for script-dialect components.
+ * On read failure, defaults to CBM_LANG_CFSCRIPT. */
+CBMLanguage cbm_disambiguate_cfc(const char *path);
+
 /* ── Gitignore pattern matching ──────────────────────────────────── */
 
 typedef struct cbm_gitignore cbm_gitignore_t;

@@ -11,6 +11,7 @@
 #include "cbm.h" // CBMLanguage, CBM_LANG_*
 
 #include "foundation/constants.h"
+#include "foundation/compat.h" // cbm_strcasestr
 #include "foundation/compat_fs.h"
 
 enum { LANG_SCAN_PASSES = 2 };
@@ -453,7 +454,10 @@ static const ext_entry_t EXT_TABLE[] = {
     /* Qt QML */
     {".qml", CBM_LANG_QML},
 
-    /* CFML / ColdFusion — .cfc components are script-dialect; .cfm are tag templates */
+    /* CFML / ColdFusion — .cfm are tag templates; .cfc components may be EITHER
+     * script-dialect (component { ... }) or tag-dialect (<cfcomponent> ...). The
+     * table default is script; tag-based .cfc are resolved by content in
+     * cbm_disambiguate_cfc(). */
     {".cfc", CBM_LANG_CFSCRIPT},
     {".cfm", CBM_LANG_CFML},
 
@@ -1122,3 +1126,74 @@ CBMLanguage cbm_disambiguate_inc(const char *path) {
     }
     return CBM_LANG_BITBAKE;
 }
+
+/* Case-insensitive prefix match (portable — no strncasecmp dependency). */
+static bool starts_with_ci(const char *s, const char *prefix) {
+    for (; *prefix; s++, prefix++) {
+        if (tolower((unsigned char)*s) != tolower((unsigned char)*prefix)) {
+            return false;
+        }
+    }
+    return true;
+}
+
+/* Disambiguate .cfc files: a ColdFusion component may be written in the script
+ * dialect ("component { ... }", parsed by the JS-like cfscript grammar) or the
+ * tag dialect ("<cfcomponent> ... <cffunction>", parsed by the HTML-derived cfml
+ * grammar). The extension table defaults to cfscript because that is what modern
+ * Lucee/ACF templates use, but large legacy codebases are predominantly tag-based
+ * and feeding those to the wrong grammar fails wholesale. Routing rules:
+ *   1. A "<cfcomponent" or top-level "<cffunction" tag ⇒ tag dialect. (The latter
+ *      catches "bare" tag components that omit the <cfcomponent> wrapper.) This
+ *      wins regardless of any leading <!---/<cfscript>, so it is checked first.
+ *   2. Otherwise the file is script-dialect content. Find the first significant
+ *      token, skipping whitespace and <!--- ---> comments:
+ *        - a leading "<cfscript>" wrapper is still script content ⇒ cfscript;
+ *        - a different leading tag (e.g. <cfquery> in a bare-tag file) ⇒ cfml;
+ *        - anything else ("component { ... }") ⇒ cfscript.
+ * Defaults to CBM_LANG_CFSCRIPT on any doubt (preserves table behaviour). */
+CBMLanguage cbm_disambiguate_cfc(const char *path) {
+    if (!path) {
+        return CBM_LANG_CFSCRIPT;
+    }
+
+    FILE *f = cbm_fopen(path, "r");
+    if (!f) {
+        return CBM_LANG_CFSCRIPT;
+    }
+
+    /* Read a generous head: tag components can carry a large license/revision
+     * comment block before the <cfcomponent> opener. */
+    char buf[CBM_SZ_16K + SKIP_ONE];
+    size_t n = fread(buf, SKIP_ONE, CBM_SZ_16K, f);
+    buf[n] = '\0';
+    (void)fclose(f);
+
+    /* Rule 1: explicit tag-component markers ⇒ tag dialect. */
+    if (cbm_strcasestr(buf, "<cfcomponent") != NULL || cbm_strcasestr(buf, "<cffunction") != NULL) {
+        return CBM_LANG_CFML;
+    }
+
+    /* Rule 2: locate the first significant token, past whitespace and comments. */
+    const char *p = buf;
+    for (;;) {
+        while (*p && isspace((unsigned char)*p)) {
+            p++;
+        }
+        if (starts_with_ci(p, "<!---")) {
+            const char *end = strstr(p + SLEN("<!---"), "--->");
+            if (!end) {
+                break; /* comment runs past the buffer — treat as no token */
+            }
+            p = end + SLEN("--->");
+            continue;
+        }
+        break;
+    }
+    if (*p == '<') {
+        /* A leading <cfscript> wrapper is script content; any other leading tag
+         * (bare-tag file) is tag content. */
+        return starts_with_ci(p, "<cfscript") ? CBM_LANG_CFSCRIPT : CBM_LANG_CFML;
+    }
+    return CBM_LANG_CFSCRIPT;
+}