fix(config): complete project language overrides

README.md

@@ -442,7 +442,11 @@ language.tpl=.ts
 viz_node_limit=0
 ```
 
-The declaration applies everywhere graphify keys a decision on the extension — file classification, extractor dispatch, cross-file resolution, and the AST cache (a file re-parsed under a different language never reuses the old entry). It is read by `graphify extract`, `graphify update`, `watch`, the hooks, the MCP server, and the `/graphify` skill alike. A typo is reported once on stderr and the scan continues with graphify's defaults.
+The declaration controls file classification, extractor selection and language-specific grammar and resolution, and the AST cache: a file re-parsed under a different language never reuses the old entry. Original source paths and physical format conventions, such as Fortran `.F90` preprocessing and C/C++ header/implementation pairing, are preserved. Optional language backends must still be installed. A typo is reported once on stderr and the scan continues with graphify's defaults.
+
+Adding, changing, or removing a language declaration invalidates affected incremental manifest entries even when the source bytes have not changed. `graphify watch` recognizes declared source extensions and root `.graphifyrc` changes, including atomic replacement; configuration changes schedule an AST rebuild and flag semantic work for the next extraction. The same configuration is read by `graphify extract`, `graphify update`, the hooks, the MCP server, and the `/graphify` skill.
+
+Extensionless local JS/TS imports can also resolve declared suffixes. Native suffixes retain precedence within file and directory-index lookup, and file candidates are tried before directory indexes. Python module and package lookups also support declared Python suffixes, while retaining native `.py` precedence.
 
 ---
 

graphify/csharp_dispatch.py

@@ -26,7 +26,7 @@
 """
 from __future__ import annotations
 
-_CSHARP_SUFFIXES = (".cs",)
+from graphify.extractors.csharp import _is_cs_file
 
 DISPATCH_RELATION = "dispatches_to"
 
@@ -44,7 +44,7 @@ def _is_csharp(node: dict | None) -> bool:
     if not node:
         return False
     source_file = node.get("source_file")
-    return bool(source_file) and str(source_file).endswith(_CSHARP_SUFFIXES)
+    return bool(source_file) and _is_cs_file(str(source_file))
 
 
 def _method_label(node: dict) -> str:
@@ -74,11 +74,11 @@ def resolve_csharp_interface_dispatch(
     place, since the call site really does name the interface.
     """
     if not any(
-        str(result.get("source_file", "")).endswith(_CSHARP_SUFFIXES)
+        _is_csharp(result)
         for result in per_file
         if isinstance(result, dict)
     ) and not any(
-        str(n.get("source_file", "")).endswith(_CSHARP_SUFFIXES) for n in all_nodes
+        _is_csharp(n) for n in all_nodes
     ):
         return
 

graphify/detect.py

@@ -20,7 +20,7 @@
     google_workspace_enabled,
 )
 from graphify.paths import GRAPHIFY_OUT, out_path
-from graphify.rcfile import activate_language_overrides, effective_suffix
+from graphify.rcfile import activate_language_overrides, cache_salt, effective_suffix
 
 
 class FileType(str, Enum):
@@ -2227,6 +2227,9 @@ def save_manifest(
     re-queues the file after the failure is fixed, without deleting
     graphify-out/.
     """
+    # Another root may have activated different overrides in this thread.
+    if root is not None:
+        activate_language_overrides(Path(root))
     existing = load_manifest(manifest_path, root=root)
 
     # Index both raw and NFC forms so scan/clear membership survives the
@@ -2347,12 +2350,19 @@ def _normalise_entry(entry):
         mtime, h = hashed[f]
         key = _nfc(f)
         prev = _normalise_entry(existing.get(key, {})) or {}
+        # Hash workers do not inherit this thread's language overrides.
+        current_lang = cache_salt(f)
+        lang_changed = current_lang != prev.get("language")
         if kind in ("ast", "both"):
             ast_h = h
+        elif lang_changed:
+            ast_h = ""
         else:
             ast_h = prev.get("ast_hash", "")
         if kind in ("semantic", "both"):
             sem_h = h
+        elif lang_changed:
+            sem_h = ""
         else:
             # Preserve semantic_hash only when content is unchanged
             sem_h = prev.get("semantic_hash", "") if h == prev.get("ast_hash", "") else ""
@@ -2365,6 +2375,7 @@ def _normalise_entry(entry):
             and mtime == prev.get("mtime")
             and (ast_h == prev.get("ast_hash", "") if kind in ("ast", "both") else True)
             and (sem_h == prev.get("semantic_hash", "") if kind in ("semantic", "both") else True)
+            and not lang_changed
             and not _in_clear_ast(f)
             and not _in_clear(f)
         )
@@ -2374,6 +2385,9 @@ def _normalise_entry(entry):
             "ast_hash": ast_h,
             "semantic_hash": sem_h,
         }
+        if current_lang is not None:
+            # No override keeps legacy rows unchanged.
+            entry["language"] = current_lang
         manifest[key] = entry
     if root is not None:
         # Persist in portable form: forward-slash relative paths. Keys outside
@@ -2504,7 +2518,11 @@ def detect_incremental(
             # the graph drifted from disk (#1859). No stored hash means we
             # cannot verify content — any mtime delta forces a re-extract,
             # and the next save promotes the entry into the dict schema.
-            if isinstance(stored, (int, float)):
+            # A language change invalidates even identical bytes and legacy rows.
+            stored_lang = stored.get("language") if isinstance(stored, dict) else None
+            if stored is not None and cache_salt(f) != stored_lang:
+                changed = True
+            elif isinstance(stored, (int, float)):
                 changed = current_mtime != stored
             elif isinstance(stored, dict):
                 # Normalise legacy {mtime, hash} to new schema