From 12c5d3ca0b135ac5005ce1141100857f9b4efda2 Mon Sep 17 00:00:00 2001
From: Aditya Datta <crazyme07071996@gmail.com>
Date: Sun, 13 Sep 2026 22:17:49 +0530
Subject: [PATCH] fix: decode URL escapes before base64 data URI payloads

---
 .../markitdown/src/markitdown/_uri_utils.py   |  4 +-
 .../tests/test_percent_encoded_data_uri.py    | 62 +++++++++++++++++++
 2 files changed, 65 insertions(+), 1 deletion(-)
 create mode 100644 packages/markitdown/tests/test_percent_encoded_data_uri.py

diff --git a/packages/markitdown/src/markitdown/_uri_utils.py b/packages/markitdown/src/markitdown/_uri_utils.py
index 9deb8ecab..6260a120a 100644
--- a/packages/markitdown/src/markitdown/_uri_utils.py
+++ b/packages/markitdown/src/markitdown/_uri_utils.py
@@ -70,6 +70,8 @@ def parse_data_uri(uri: str) -> Tuple[str | None, Dict[str, str], bytes]:
         elif len(part) > 0:
             attributes[part.lower()] = ""
 
-    content = base64.b64decode(data) if is_base64 else unquote_to_bytes(data)
+    # URL escaping applies to the payload even when it contains base64 text.
+    data_bytes = unquote_to_bytes(data)
+    content = base64.b64decode(data_bytes) if is_base64 else data_bytes
 
     return mime_type, attributes, content
diff --git a/packages/markitdown/tests/test_percent_encoded_data_uri.py b/packages/markitdown/tests/test_percent_encoded_data_uri.py
new file mode 100644
index 000000000..01466299e
--- /dev/null
+++ b/packages/markitdown/tests/test_percent_encoded_data_uri.py
@@ -0,0 +1,62 @@
+import base64
+from urllib.parse import quote
+
+import pytest
+
+from markitdown import MarkItDown, StreamInfo
+from markitdown._uri_utils import parse_data_uri
+
+
+@pytest.mark.parametrize(
+    "payload", [b"Hello", b"\xfb\xff", b"\xff\xff", "Hello, 世界!".encode("utf-8")]
+)
+@pytest.mark.parametrize("lowercase", [False, True])
+def test_percent_encoded_base64_data(payload: bytes, lowercase: bool) -> None:
+    encoded = quote(base64.b64encode(payload).decode("ascii"), safe="")
+    if lowercase:
+        encoded = (
+            encoded.replace("%2B", "%2b").replace("%2F", "%2f").replace("%3D", "%3d")
+        )
+    assert parse_data_uri("data:application/octet-stream;base64," + encoded) == (
+        "application/octet-stream",
+        {},
+        payload,
+    )
+
+
+@pytest.mark.parametrize(
+    "header",
+    [
+        "data:text/plain;base64",
+        "DATA:text/plain;BASE64",
+        "data:;base64",
+        "data:text/plain;charset=utf-8;base64",
+    ],
+)
+def test_percent_encoded_padding_preserves_metadata(header: str) -> None:
+    mime, attributes, payload = parse_data_uri(header + ",SGVsbG8%3D")
+    assert payload == b"Hello"
+    assert mime == (None if header == "data:;base64" else "text/plain")
+    assert attributes == ({"charset": "utf-8"} if "charset" in header else {})
+
+
+@pytest.mark.parametrize("encoded", ["+/8=", "%2B/8=", "+%2F8=", "%2B%2F8%3D"])
+def test_raw_plus_is_not_form_decoded_as_space(encoded: str) -> None:
+    assert (
+        parse_data_uri("data:application/octet-stream;base64," + encoded)[2]
+        == b"\xfb\xff"
+    )
+
+
+def test_non_base64_data_still_decodes_once() -> None:
+    assert parse_data_uri("data:text/plain,%252F+%2C")[2] == b"%2F+,"
+
+
+def test_public_conversion_accepts_percent_encoded_base64() -> None:
+    text = "Hello, 世界!"
+    encoded = quote(base64.b64encode(text.encode("utf-8")).decode("ascii"), safe="")
+    result = MarkItDown().convert_uri(
+        "data:text/plain;charset=utf-8;base64," + encoded,
+        stream_info=StreamInfo(extension=".txt"),
+    )
+    assert result.markdown == text