Attempt different character encodings when previewing a URL. (#11077)

This follows similar logic to BeautifulSoup where we attempt different
character encodings until we find one which works.

changelog.d/11077.bugfix

@@ -0,0 +1 @@
+Fix a long-standing bug when attempting to preview URLs which are in the `windows-1252` character encoding.

synapse/rest/media/v1/preview_url_resource.py

@@ -295,8 +295,7 @@ class PreviewUrlResource(DirectServeJsonResource):
             with open(media_info.filename, "rb") as file:
                 body = file.read()
 
-            encoding = get_html_media_encoding(body, media_info.media_type)
-            tree = decode_body(body, encoding)
+            tree = decode_body(body, media_info.uri, media_info.media_type)
             if tree is not None:
                 # Check if this HTML document points to oEmbed information and
                 # defer to that.
@@ -632,16 +631,19 @@ class PreviewUrlResource(DirectServeJsonResource):
             logger.debug("No media removed from url cache")
 
 
-def get_html_media_encoding(body: bytes, content_type: str) -> str:
+def get_html_media_encodings(body: bytes, content_type: Optional[str]) -> Iterable[str]:
     """
-    Get the encoding of the body based on the (presumably) HTML body or media_type.
+    Get potential encoding of the body based on the (presumably) HTML body or the content-type header.
 
     The precedence used for finding a character encoding is:
 
-    1. meta tag with a charset declared.
+    1. <meta> tag with a charset declared.
     2. The XML document's character encoding attribute.
     3. The Content-Type header.
-    4. Fallback to UTF-8.
+    4. Fallback to utf-8.
+    5. Fallback to windows-1252.
+
+    This roughly follows the algorithm used by BeautifulSoup's bs4.dammit.EncodingDetector.
 
     Args:
         body: The HTML document, as bytes.
@@ -653,36 +655,39 @@ def get_html_media_encoding(body: bytes, content_type: str) -> str:
     # Limit searches to the first 1kb, since it ought to be at the top.
     body_start = body[:1024]
 
-    # Let's try and figure out if it has an encoding set in a meta tag.
+    # Check if it has an encoding set in a meta tag.
     match = _charset_match.search(body_start)
     if match:
-        return match.group(1).decode("ascii")
+        yield match.group(1).decode("ascii")
 
     # TODO Support <meta http-equiv="Content-Type" content="text/html; charset=utf-8"/>
 
-    # If we didn't find a match, see if it an XML document with an encoding.
+    # Check if it has an XML document with an encoding.
     match = _xml_encoding_match.match(body_start)
     if match:
-        return match.group(1).decode("ascii")
+        yield match.group(1).decode("ascii")
 
-    # If we don't find a match, we'll look at the HTTP Content-Type, and
-    # if that doesn't exist, we'll fall back to UTF-8.
+    # Check the HTTP Content-Type header for a character set.
+    if content_type:
         content_match = _content_type_match.match(content_type)
         if content_match:
-        return content_match.group(1)
+            yield content_match.group(1)
 
-    return "utf-8"
+    # Finally, fallback to UTF-8, then windows-1252.
+    yield "utf-8"
+    yield "windows-1252"
 
 
 def decode_body(
-    body: bytes, request_encoding: Optional[str] = None
+    body: bytes, uri: str, content_type: Optional[str] = None
 ) -> Optional["etree.Element"]:
     """
     This uses lxml to parse the HTML document.
 
     Args:
         body: The HTML document, as bytes.
-        request_encoding: The character encoding of the body, as a string.
+        uri: The URI used to download the body.
+        content_type: The Content-Type header.
 
     Returns:
         The parsed HTML body, or None if an error occurred during processed.
@@ -691,32 +696,25 @@ def decode_body(
     if not body:
         return None
 
-    from lxml import etree
-
-    # Create an HTML parser. If this fails, log and return no metadata.
+    for encoding in get_html_media_encodings(body, content_type):
         try:
-        parser = etree.HTMLParser(recover=True, encoding=request_encoding)
-    except LookupError:
-        # blindly consider the encoding as utf-8.
-        parser = etree.HTMLParser(recover=True, encoding="utf-8")
-    except Exception as e:
-        logger.warning("Unable to create HTML parser: %s" % (e,))
+            body_str = body.decode(encoding)
+        except Exception:
+            pass
+        else:
+            break
+    else:
+        logger.warning("Unable to decode HTML body for %s", uri)
         return None
 
-    def _attempt_decode_body(
-        body_attempt: Union[bytes, str]
-    ) -> Optional["etree.Element"]:
+    from lxml import etree
+
+    # Create an HTML parser.
+    parser = etree.HTMLParser(recover=True, encoding="utf-8")
+
     # Attempt to parse the body. Returns None if the body was successfully
     # parsed, but no tree was found.
-        return etree.fromstring(body_attempt, parser)
-
-    # Attempt to parse the body. If this fails, log and return no metadata.
-    try:
-        return _attempt_decode_body(body)
-    except UnicodeDecodeError:
-        # blindly try decoding the body as utf-8, which seems to fix
-        # the charset mismatches on https://google.com
-        return _attempt_decode_body(body.decode("utf-8", "ignore"))
+    return etree.fromstring(body_str, parser)
 
 
 def _calc_og(tree: "etree.Element", media_uri: str) -> Dict[str, Optional[str]]:

tests/test_preview.py

@@ -15,7 +15,7 @@
 from synapse.rest.media.v1.preview_url_resource import (
     _calc_og,
     decode_body,
-    get_html_media_encoding,
+    get_html_media_encodings,
     summarize_paragraphs,
 )
 
@@ -159,7 +159,7 @@ class CalcOgTestCase(unittest.TestCase):
         </html>
         """
 
-        tree = decode_body(html)
+        tree = decode_body(html, "http://example.com/test.html")
         og = _calc_og(tree, "http://example.com/test.html")
 
         self.assertEqual(og, {"og:title": "Foo", "og:description": "Some text."})
@@ -175,7 +175,7 @@ class CalcOgTestCase(unittest.TestCase):
         </html>
         """
 
-        tree = decode_body(html)
+        tree = decode_body(html, "http://example.com/test.html")
         og = _calc_og(tree, "http://example.com/test.html")
 
         self.assertEqual(og, {"og:title": "Foo", "og:description": "Some text."})
@@ -194,7 +194,7 @@ class CalcOgTestCase(unittest.TestCase):
         </html>
         """
 
-        tree = decode_body(html)
+        tree = decode_body(html, "http://example.com/test.html")
         og = _calc_og(tree, "http://example.com/test.html")
 
         self.assertEqual(
@@ -216,7 +216,7 @@ class CalcOgTestCase(unittest.TestCase):
         </html>
         """
 
-        tree = decode_body(html)
+        tree = decode_body(html, "http://example.com/test.html")
         og = _calc_og(tree, "http://example.com/test.html")
 
         self.assertEqual(og, {"og:title": "Foo", "og:description": "Some text."})
@@ -230,7 +230,7 @@ class CalcOgTestCase(unittest.TestCase):
         </html>
         """
 
-        tree = decode_body(html)
+        tree = decode_body(html, "http://example.com/test.html")
         og = _calc_og(tree, "http://example.com/test.html")
 
         self.assertEqual(og, {"og:title": None, "og:description": "Some text."})
@@ -245,7 +245,7 @@ class CalcOgTestCase(unittest.TestCase):
         </html>
         """
 
-        tree = decode_body(html)
+        tree = decode_body(html, "http://example.com/test.html")
         og = _calc_og(tree, "http://example.com/test.html")
 
         self.assertEqual(og, {"og:title": "Title", "og:description": "Some text."})
@@ -260,7 +260,7 @@ class CalcOgTestCase(unittest.TestCase):
         </html>
         """
 
-        tree = decode_body(html)
+        tree = decode_body(html, "http://example.com/test.html")
         og = _calc_og(tree, "http://example.com/test.html")
 
         self.assertEqual(og, {"og:title": None, "og:description": "Some text."})
@@ -268,13 +268,13 @@ class CalcOgTestCase(unittest.TestCase):
     def test_empty(self):
         """Test a body with no data in it."""
         html = b""
-        tree = decode_body(html)
+        tree = decode_body(html, "http://example.com/test.html")
         self.assertIsNone(tree)
 
     def test_no_tree(self):
         """A valid body with no tree in it."""
         html = b"\x00"
-        tree = decode_body(html)
+        tree = decode_body(html, "http://example.com/test.html")
         self.assertIsNone(tree)
 
     def test_invalid_encoding(self):
@@ -287,7 +287,7 @@ class CalcOgTestCase(unittest.TestCase):
         </body>
         </html>
         """
-        tree = decode_body(html, "invalid-encoding")
+        tree = decode_body(html, "http://example.com/test.html", "invalid-encoding")
         og = _calc_og(tree, "http://example.com/test.html")
         self.assertEqual(og, {"og:title": "Foo", "og:description": "Some text."})
 
@@ -302,15 +302,29 @@ class CalcOgTestCase(unittest.TestCase):
         </body>
         </html>
         """
-        tree = decode_body(html)
+        tree = decode_body(html, "http://example.com/test.html")
         og = _calc_og(tree, "http://example.com/test.html")
         self.assertEqual(og, {"og:title": "ÿÿ Foo", "og:description": "Some text."})
 
+    def test_windows_1252(self):
+        """A body which uses windows-1252, but doesn't declare that."""
+        html = b"""
+        <html>
+        <head><title>\xf3</title></head>
+        <body>
+        Some text.
+        </body>
+        </html>
+        """
+        tree = decode_body(html, "http://example.com/test.html")
+        og = _calc_og(tree, "http://example.com/test.html")
+        self.assertEqual(og, {"og:title": "ó", "og:description": "Some text."})
+
 
 class MediaEncodingTestCase(unittest.TestCase):
     def test_meta_charset(self):
         """A character encoding is found via the meta tag."""
-        encoding = get_html_media_encoding(
+        encodings = get_html_media_encodings(
             b"""
         <html>
         <head><meta charset="ascii">
@@ -319,10 +333,10 @@ class MediaEncodingTestCase(unittest.TestCase):
         """,
             "text/html",
         )
-        self.assertEqual(encoding, "ascii")
+        self.assertEqual(list(encodings), ["ascii", "utf-8", "windows-1252"])
 
         # A less well-formed version.
-        encoding = get_html_media_encoding(
+        encodings = get_html_media_encodings(
             b"""
         <html>
         <head>< meta charset = ascii>
@@ -331,11 +345,11 @@ class MediaEncodingTestCase(unittest.TestCase):
         """,
             "text/html",
         )
-        self.assertEqual(encoding, "ascii")
+        self.assertEqual(list(encodings), ["ascii", "utf-8", "windows-1252"])
 
     def test_meta_charset_underscores(self):
         """A character encoding contains underscore."""
-        encoding = get_html_media_encoding(
+        encodings = get_html_media_encodings(
             b"""
         <html>
         <head><meta charset="Shift_JIS">
@@ -344,11 +358,11 @@ class MediaEncodingTestCase(unittest.TestCase):
         """,
             "text/html",
         )
-        self.assertEqual(encoding, "Shift_JIS")
+        self.assertEqual(list(encodings), ["Shift_JIS", "utf-8", "windows-1252"])
 
     def test_xml_encoding(self):
         """A character encoding is found via the meta tag."""
-        encoding = get_html_media_encoding(
+        encodings = get_html_media_encodings(
             b"""
         <?xml version="1.0" encoding="ascii"?>
         <html>
@@ -356,11 +370,11 @@ class MediaEncodingTestCase(unittest.TestCase):
         """,
             "text/html",
         )
-        self.assertEqual(encoding, "ascii")
+        self.assertEqual(list(encodings), ["ascii", "utf-8", "windows-1252"])
 
     def test_meta_xml_encoding(self):
         """Meta tags take precedence over XML encoding."""
-        encoding = get_html_media_encoding(
+        encodings = get_html_media_encodings(
             b"""
         <?xml version="1.0" encoding="ascii"?>
         <html>
@@ -370,7 +384,7 @@ class MediaEncodingTestCase(unittest.TestCase):
         """,
             "text/html",
         )
-        self.assertEqual(encoding, "UTF-16")
+        self.assertEqual(list(encodings), ["UTF-16", "ascii", "utf-8", "windows-1252"])
 
     def test_content_type(self):
         """A character encoding is found via the Content-Type header."""
@@ -384,10 +398,10 @@ class MediaEncodingTestCase(unittest.TestCase):
             'text/html; charset=ascii";',
         )
         for header in headers:
-            encoding = get_html_media_encoding(b"", header)
-            self.assertEqual(encoding, "ascii")
+            encodings = get_html_media_encodings(b"", header)
+            self.assertEqual(list(encodings), ["ascii", "utf-8", "windows-1252"])
 
     def test_fallback(self):
         """A character encoding cannot be found in the body or header."""
-        encoding = get_html_media_encoding(b"", "text/html")
-        self.assertEqual(encoding, "utf-8")
+        encodings = get_html_media_encodings(b"", "text/html")
+        self.assertEqual(list(encodings), ["utf-8", "windows-1252"])