fix: credit file readers through the data trees

FileConsumers resolved profile entries without the data directory
registry the builder uses. Dolphin's hashless codehandler.bin then took
Dolphin-MMJR's copy by name, and the site listed Dolphin and Triforce as
its readers while every pack serves data/dolphin-sys. blueMSX's RS232.ROM
and MSX2P.rom were credited to bios/ copies the same way. The registry
is now read by default, as the builder reads it.

With it, bundled_only.py --strict (pipeline step 4d) names
bios/Other/dolphin-mmjr/codehandler.bin: only dolphin-mmjr's bundled
entry, which already carries its sha1 and size, reaches that copy.

scripts/cross_reference.py

@@ -30,6 +30,7 @@ from common import (
     get_mame_clone_map,
     list_emulator_profiles,
     list_registered_platforms,
+    load_data_dir_registry,
     load_database,
     load_emulator_profiles,
     load_platform_config,
@@ -325,9 +326,21 @@ class FileConsumers:
     file by the evidence the pack uses, so a homonym is nobody's file.
     """
 
-    def __init__(self, db: dict, zip_contents: dict | None = None) -> None:
+    def __init__(
+        self, db: dict, zip_contents: dict | None = None,
+        data_dir_registry: dict | None = None,
+    ) -> None:
         self._db = db
         self._zip_contents = zip_contents
+        # The builder resolves with the data trees: Dolphin's codehandler.bin
+        # comes from data/dolphin-sys, and without them the name step gave it
+        # Dolphin-MMJR's copy, which the site then said Dolphin reads.
+        self._data_dir_registry = (
+            data_dir_registry if data_dir_registry is not None
+            else load_data_dir_registry(
+                str(Path(__file__).resolve().parent.parent / DEFAULT_PLATFORMS_DIR)
+            )
+        )
         self._sha1_by_path = {
             path: sha1
             for sha1, entry in db.get("files", {}).items()
@@ -357,7 +370,8 @@ class FileConsumers:
         if owner:
             entry = {**entry, "source_profile": owner}
         path, status = resolve_local_file(
-            entry, self._db, self._zip_contents, dest_hint=dest
+            entry, self._db, self._zip_contents, dest_hint=dest,
+            data_dir_registry=self._data_dir_registry,
         )
         if not path or status in ("not_found", "hash_mismatch"):
             return None

tests/test_file_consumers.py

@@ -83,6 +83,40 @@ class ReadersResolve(unittest.TestCase):
         self.assertNotIn("ioquake3", et)
 
 
+class ReadersFollowTheDataTrees(unittest.TestCase):
+    """Dolphin names codehandler.bin without a hash. The pack takes it from
+    data/dolphin-sys, and the site credited Dolphin and Triforce with the
+    Dolphin-MMJR copy, another revision, because the consumers resolved
+    without the data trees. Reads the repository."""
+
+    def test_dolphin_does_not_read_the_mmjr_copy(self):
+        import os
+
+        from common import load_database, load_emulator_profiles
+
+        if not (REPO_ROOT / "database.json").is_file() or not (
+            REPO_ROOT / "data" / "dolphin-sys" / "codehandler.bin"
+        ).is_file():
+            self.skipTest("database.json or data/dolphin-sys is absent")
+        previous = os.getcwd()
+        os.chdir(REPO_ROOT)
+        try:
+            db = load_database("database.json")
+            profiles = load_emulator_profiles("emulators")
+            consumers = FileConsumers(db)
+            consumers.add_emulators({"dolphin": profiles["dolphin"]})
+        finally:
+            os.chdir(previous)
+        mmjr = next(
+            (sha1 for sha1, entry in db["files"].items()
+             if entry["path"] == "bios/Other/dolphin-mmjr/codehandler.bin"),
+            None,
+        )
+        if mmjr is None:
+            self.skipTest("the Dolphin-MMJR copy is not collected")
+        self.assertNotIn("dolphin", consumers.emulators.get(mmjr, set()))
+
+
 def tearDownModule():
     _TMP.cleanup()