SEC: Optimize _decode_png_prediction regarding memory and speed

pypdf/filters.py

@@ -262,11 +262,11 @@ def _decode_png_prediction(data: bytes, columns: int, row_length: int) -> bytes:
             logger_warning("Image data is not rectangular. Adding padding.", source=__name__)
             data += b"\x00" * (row_length - remainder)
             assert len(data) % row_length == 0
-        output = []
-        previous_row_data = (0,) * row_length
+        output = bytearray()
+        previous_row_data = bytes(row_length)
         bpp = (row_length - 1) // columns  # recomputed locally to not change params
         for row in range(0, len(data), row_length):
-            row_data: list[int] = list(data[row : row + row_length])
+            row_data = bytearray(data[row : row + row_length])
             filter_byte = row_data[0]
 
             if filter_byte == 0:
@@ -315,8 +315,8 @@ def _decode_png_prediction(data: bytes, columns: int, row_length: int) -> bytes:
                 raise PdfReadError(
                     f"Unsupported PNG filter {filter_byte!r}"
                 )  # pragma: no cover
-            previous_row_data = tuple(row_data)
-            output.extend(row_data[1:])
+            previous_row_data = bytes(row_data)
+            output += row_data[1:]
         return bytes(output)
 
     @staticmethod

tests/test_filters.py

@@ -1130,3 +1130,18 @@ def test_lzwdecode__invalid_first_code():
             match=r"^LZW code 258 out of range with empty base at table index 258\.$"
     ):
         LZWDecode.decode(data=lzw_data)
+
+
+@pytest.mark.timeout(5)  # Has been 20 seconds before.
+def test_flatedecode__decode_png_prediction__speed():
+    columns = 4096
+    rows = 120000
+    row_length = columns + 1  # +1 for PNG filter byte
+
+    # Build raw PNG-predicted data: every row starts with filter byte 0 (PNG None)
+    raw = bytearray(rows * row_length)
+    for row in range(rows):
+        raw[row * row_length] = 0
+    data = bytes(raw)
+
+    FlateDecode._decode_png_prediction(data=data, columns=columns, row_length=row_length)