fix: serve a data dir file at its platform path

AmberELEC and EmuELEC declare blueMSX's Machines/<model>/config.ini and
Shared Roms/RS232.ROM with no hash. The name step served msx-emu's
ColecoVision config.ini in 192 machine folders and the 16 KB RS232.ROM,
while data/bluemsx held blueMSX's own file at that very path. A platform
declaration with no hash now reads its destination's tail in the data
directories before the name step. Profile entries keep the foreign-tree
redirect, so Dolphin's dsp_coef.bin dump still outranks the free copy.

Manifests: amberelec and emuelec omit 192 config.ini entries (held in
data/bluemsx only, so the installer has no URL) and serve colecodb.xml,
MSX2+/config.ini, MSXturboR/config.ini and RS232.ROM from blueMSX's bytes.

scripts/common.py

@@ -9,6 +9,7 @@ from __future__ import annotations
 from collections.abc import Callable, Iterable
 from collections import Counter
 
+import filecmp
 import hashlib
 import functools
 import json
@@ -889,6 +890,38 @@ def _resolve_in_data_dirs(
     return None, mismatch
 
 
+def _data_dir_tail(
+    dest_hint: str, data_dir_registry: dict, file_entry: dict, db: dict
+) -> str | None:
+    """The cached data-directory file at the destination's own tail.
+
+    The full destination first, then without its leading segment, never
+    down to the bare name. A shortened tail two caches hold with different
+    bytes proves nothing. The size and Adler-32 the code checks still bound
+    the file, as on every name or path step without a hash.
+    """
+    parts = dest_hint.split("/")
+    for start in range(min(2, len(parts) - 1)):
+        tail = "/".join(parts[start:])
+        hits = [
+            os.path.join(cache_dir, tail)
+            for cache_dir in (
+                dd_entry.get("local_cache", "") for dd_entry in data_dir_registry.values()
+            )
+            if cache_dir and os.path.isfile(os.path.join(cache_dir, tail))
+        ]
+        if start and any(
+            not filecmp.cmp(hits[0], other, shallow=False) for other in hits[1:]
+        ):
+            continue
+        for candidate in hits:
+            if name_match_size_ok(
+                file_entry, os.path.getsize(candidate)
+            ) and adler32_fits(file_entry, candidate, db):
+                return candidate
+    return None
+
+
 # Basename index per data-directory cache. Walking the whole tree again for
 # every unresolved name made resolution O(names x tree) on each pack build.
 _data_dir_index: dict[str, dict[str, list[tuple[int, str]]]] = {}
@@ -1504,6 +1537,27 @@ def resolve_local_file(
                 if status:
                     return path, status
 
+    # 2c. The same tail in a data directory. by_path_suffix indexes bios/
+    # only, so AmberELEC's Machines/MSX - Arabic/config.ini fell through to
+    # the name step and got msx-emu's ColecoVision config.ini, while
+    # data/bluemsx held blueMSX's own at that very path. A platform
+    # declaration is resolved without a hint, and its destination is that
+    # path. The descent and the ambiguity rule are those of the bios/ tails.
+    # A profile's entry is not asked here: it names its owner, the name step
+    # sends it to its data directories when another emulator's copy answers,
+    # and a dump it declares outranks the free one its tree bundles at the
+    # same path (Dolphin's dsp_coef.bin over Sys/GC's).
+    tail_hint = dest_hint or str(file_entry.get("destination") or "")
+    if (
+        tail_hint
+        and data_dir_registry
+        and not has_strong_hash
+        and not _owner_tokens(file_entry)
+    ):
+        hit = _data_dir_tail(tail_hint, data_dir_registry, file_entry, db)
+        if hit:
+            return hit, "path_exact"
+
     # 3. No MD5 = any file with that name or alias (existence check)
     def _size_ok(match_sha1: str) -> bool:
         record = files_db.get(match_sha1, {})

tests/test_foreign_tree_names.py

@@ -93,6 +93,54 @@ class ForeignTreeNames(unittest.TestCase):
         self.assertEqual(status, "data_dir")
         self.assertEqual(Path(path).read_bytes(), b"B" * 64)
 
+    def _bluemsx_cache(self, rel: str, data: bytes) -> dict:
+        cache = self.root / "data" / "bluemsx"
+        (cache / rel).parent.mkdir(parents=True, exist_ok=True)
+        (cache / rel).write_bytes(data)
+        return {"bluemsx": {"local_cache": str(cache)}}
+
+    def test_a_platform_declaration_reads_the_data_directory_at_its_path(self):
+        """AmberELEC declares every blueMSX machine's config.ini with no hash.
+        The name step met msx-emu's ColecoVision config.ini first and wrote it
+        into each machine folder, while data/bluemsx held the machine's own
+        at that very path."""
+        self._hold("bios/Other/msx-emu/Machines/COL - ColecoVision/config.ini", b"C" * 64)
+        registry = self._bluemsx_cache("Machines/MSX - Arabic/config.ini", b"A" * 64)
+        entry = {
+            "name": "config.ini",
+            "destination": "bios/Machines/MSX - Arabic/config.ini",
+        }
+        path, status = resolve_local_file(entry, self.db, data_dir_registry=registry)
+        self.assertEqual(status, "path_exact")
+        self.assertEqual(Path(path).read_bytes(), b"A" * 64)
+
+    def test_a_platform_rs232_rom_reads_the_data_directory_at_its_path(self):
+        self._hold("bios/Microsoft/MSX/RS232.ROM", b"L" * 128)
+        registry = self._bluemsx_cache("Machines/Shared Roms/RS232.ROM", b"S" * 64)
+        entry = {
+            "name": "RS232.ROM",
+            "destination": "bios/Machines/Shared Roms/RS232.ROM",
+        }
+        path, _status = resolve_local_file(entry, self.db, data_dir_registry=registry)
+        self.assertEqual(Path(path).read_bytes(), b"S" * 64)
+
+    def test_a_profile_dump_outranks_the_free_copy_of_its_tree(self):
+        """Dolphin's dsp_coef.bin: the dump in bios/ wins over the free
+        replacement Sys/GC bundles at the same path."""
+        own = self._hold("bios/Nintendo/GameCube/.variants/dsp_coef.bin.c116d867", b"R" * 64)
+        cache = self.root / "data" / "dolphin-sys"
+        (cache / "GC").mkdir(parents=True)
+        (cache / "GC" / "dsp_coef.bin").write_bytes(b"F" * 64)
+        entry = {
+            "name": "dsp_coef.bin", "destination": "GC/dsp_coef.bin",
+            "source_profile": "dolphin",
+        }
+        path, _status = resolve_local_file(
+            entry, self.db, dest_hint="GC/dsp_coef.bin",
+            data_dir_registry={"dolphin-sys": {"local_cache": str(cache)}},
+        )
+        self.assertEqual(path, own)
+
     def test_another_casing_in_another_tree_is_not_the_file(self):
         """reminiscence's music module reunion took skyline's
         zoneinfo/Indian/Reunion, a time-zone file, on the casefold step."""

wiki/verification-modes.md

@@ -363,7 +363,7 @@ evidence-compatible match:
 | 2 | SHA256 | `sha256_exact` | A declared SHA256 identifies a record, again requiring all declarations to agree |
 | 3 | CRC32 plus size | `crc32_exact` | Used only when no stronger hash is declared. Size confirms the weaker checksum when available |
 | 4 | MD5 | `md5_exact` | Direct lookup. An explicitly supported truncated MD5 also needs a compatible name |
-| 5 | Path suffix | `path_exact` or a hash-exact status | Disambiguates regional paths. With any declared hash, the path is accepted only when the content matches it |
+| 5 | Path suffix | `path_exact` or a hash-exact status | Disambiguates regional paths. With any declared hash, the path is accepted only when the content matches it. A platform declaration with no hash then looks for its destination's tail in the data directories: AmberELEC's `Machines/MSX - Arabic/config.ini` is blueMSX's, never the first `config.ini` the name step meets |
 | 6 | Name or alias | `name_exact` | Existence/size resolution only when no content hash is declared. Prefers primary files and supports casefold matching |
 | 7 | Named candidate inspection | `md5_composite_exact`, `md5_exact`, or `hash_mismatch` | Checks composite ZIP MD5 or direct MD5. A matching name with wrong content is surfaced, never treated as exact |
 | 8 | ZIP contents index | `zip_exact` | `zipped_file` with MD5. Searches the inner-ROM index only after name-based resolution fails |