fix: handle Windows drive letters in get_parquet_indexer_cls

vini-fda · vini-fda · commit bcb13c9d55f6 · 2026-05-26T13:38:24.000-03:00
diff --git a/src/litdata/utilities/parquet.py b/src/litdata/utilities/parquet.py
@@ -326,7 +326,12 @@ def get_parquet_indexer_cls(
 
     obj = parse.urlparse(dir_path)
 
-    if obj.scheme in ("local", ""):
+    # On Windows, paths like "C:\\Users\\..." parse with scheme='c'. A single-letter
+    # scheme is never a valid URI scheme (RFC 3986 requires >=2 chars) so treat it
+    # as a Windows drive letter and dispatch to LocalParquetDir.
+    is_windows_drive = len(obj.scheme) == 1 and obj.scheme.isalpha()
+
+    if obj.scheme in ("local", "") or is_windows_drive:
         return LocalParquetDir(*args)
     if obj.scheme in ("gs", "s3"):
         return CloudParquetDir(*args)
