From 1271b338f42e4562fbbe9bb8dd23e6cdd3984460 Mon Sep 17 00:00:00 2001
From: Abdessamad Derraz <3028866+Abdess@users.noreply.github.com>
Date: Tue, 3 Nov 2026 18:07:08 +0100
Subject: [PATCH] fix: keep listxml machines larger than a read

---
 scripts/scraper/romset_dat_importer.py | 13 ++++++++++---
 tests/test_romset_recipes.py           | 20 ++++++++++++++++++++
 2 files changed, 30 insertions(+), 3 deletions(-)

diff --git a/scripts/scraper/romset_dat_importer.py b/scripts/scraper/romset_dat_importer.py
index 80e663cd6..55a081af8 100644
--- a/scripts/scraper/romset_dat_importer.py
+++ b/scripts/scraper/romset_dat_importer.py
@@ -200,20 +200,22 @@ def _scan_listxml(
     and those holding one of *targets*.
 
     The document is 311 MB for MAME 0.289, so it is never held whole: the
-    reader keeps a sliding window just large enough to close the element it
-    is in the middle of.
+    window keeps the element still open from its start tag, and grows until
+    that element closes, whatever its size.
     """
     found: dict[str, tuple[str | None, list[dict]]] = {}
     buffer = b""
     while chunk := stream.read(LISTXML_CHUNK):
         buffer += chunk
         position = 0
+        keep_from = None
         while True:
             match = _MACHINE.search(buffer, position)
             if not match:
                 break
             end = buffer.find(b"</machine>", match.end())
             if end == -1:
+                keep_from = match.start()
                 break
             name = match.group(1).decode()
             wanted = selector(name)
@@ -223,7 +225,12 @@ def _scan_listxml(
                     parent = _ROMOF.search(match.group(2).decode())
                     found[name] = (parent.group(1) if parent else None, roms)
             position = end + len(b"</machine>")
-        buffer = buffer[position:] if position else buffer[-(2 * LISTXML_CHUNK) :]
+        if keep_from is None:
+            # No element open: only a start tag cut by the read can follow.
+            start = buffer.rfind(b"<machine", position)
+            tail = max(position, len(buffer) - len(b"<machine"))
+            keep_from = start if start != -1 else tail
+        buffer = buffer[keep_from:]
     return found
 
 
diff --git a/tests/test_romset_recipes.py b/tests/test_romset_recipes.py
index e4d57febf..950f099ad 100644
--- a/tests/test_romset_recipes.py
+++ b/tests/test_romset_recipes.py
@@ -374,6 +374,26 @@ class ListXmlRecipes(unittest.TestCase):
             [m["name"] for m in by_name["parentset.zip"]["members"]], ["shared.rom"]
         )
 
+    def test_a_machine_larger_than_the_window_is_kept(self):
+        """Each machine here spans several reads: the window grows until the
+        element closes instead of cutting its start tag away."""
+        from scripts.scraper import romset_dat_importer
+
+        expected = listxml_entries(self._opener(self._document()), "MAME 0.289", None)
+        saved = romset_dat_importer.LISTXML_CHUNK
+        try:
+            for window in (1, 7, 16, 64):
+                romset_dat_importer.LISTXML_CHUNK = window
+                entries = listxml_entries(
+                    self._opener(self._document()), "MAME 0.289", None
+                )
+                self.assertEqual(entries, expected, window)
+        finally:
+            romset_dat_importer.LISTXML_CHUNK = saved
+        self.assertEqual(
+            [entry["name"] for entry in entries], ["childset.zip", "parentset.zip"]
+        )
+
     def test_an_undumped_only_set_is_dropped(self):
         entries = listxml_entries(self._opener(self._document()), "MAME 0.289", None)
         self.assertNotIn("undumped.zip", {entry["name"] for entry in entries})