fix: drop variants whose bytes a file holds

A copy under .variants/ exists to hold a content its primary does
not. 29 held bytes another file already carried, under a second or
third identity: msx.rom and MSX.ROM variants equal to
MSX_8020-20bios.rom, naomi2.zip.da79eca4 whose suffix named the
primary's sha1 rather than its own. dedup now removes such a copy,
keeping one when only variants hold the content. The LaserActive
PAC-S10 BIOS lived only under .variants/ while RetroBat names it
without a hash; it moves beside its siblings.

scripts/dedup.py

@@ -129,6 +129,19 @@ def _plan_group_removals(
             name_paths.sort(key=path_priority)
             true_dupes_to_remove.extend(name_paths[1:])
 
+    # A variant exists to hold a content its primary does not. Once another
+    # file holds the same bytes, the copy under .variants/ answers nothing
+    # the index does not already know, whatever its name or suffix: the
+    # resolver finds the content by hash and writes it under the name the
+    # destination asks for.
+    remaining = [p for p in paths if p not in true_dupes_to_remove]
+    variants = [p for p in remaining if ".variants" in Path(p).parts]
+    if variants:
+        holders = [p for p in remaining if p not in variants] or [
+            min(variants, key=path_priority)
+        ]
+        true_dupes_to_remove.extend(p for p in variants if p not in holders)
+
     # Different filenames, same content -need special handling
     unique_names = sorted(by_name.keys())
     if len(unique_names) > 1:

tests/test_dedup.py

@@ -98,6 +98,28 @@ class TrueDuplicates(_Bios):
         self.run_dedup()
         self.assertEqual(self.existing(), {"Sony/PS/scph.bin"})
 
+    def test_a_variant_under_another_name_loses_to_the_file_holding_its_bytes(self):
+        # .variants/scph1001.bin.649895ef held the bytes of scph1001_v20.bin:
+        # the content was answered twice and the index kept one at random.
+        self.write("Sony/PS/scph1001_v20.bin")
+        self.write("Sony/PS/.variants/scph1001.bin.649895ef")
+        self.run_dedup()
+        self.assertEqual(self.existing(), {"Sony/PS/scph1001_v20.bin"})
+
+    def test_two_variants_of_one_content_keep_one(self):
+        # msx.rom.a317e6b4 and MSX.ROM.e998f0c4 differ only by casing and
+        # suffix; nothing outside .variants/ holds the bytes in this tree.
+        self.write("Microsoft/MSX/.variants/msx.rom.a317e6b4")
+        self.write("Microsoft/MSX/.variants/MSX.ROM.e998f0c4")
+        self.run_dedup()
+        self.assertEqual(len(self.existing()), 1)
+
+    def test_a_variant_of_other_bytes_stays(self):
+        self.write("Sony/PS/scph1001.bin", b"PRIMARY")
+        self.write("Sony/PS/.variants/scph1001.bin.649895ef", b"ALTERNATE")
+        self.run_dedup()
+        self.assertEqual(len(self.existing()), 2)
+
     def test_different_content_under_one_name_is_never_touched(self):
         self.write("Sony/PS/a.bin", b"CONTENT ONE")
         self.write("Sega/Saturn/a.bin", b"CONTENT TWO")