Improve URL previews for sites with only Twitter card information. (#13056)

Pull out `twitter:` meta tags when generating a preview and
use it to augment any `og:` meta tags.

Prefers Open Graph information over Twitter card information.

changelog.d/13056.feature

@@ -0,0 +1 @@
+Improve URL previews for sites which only provide Twitter Card metadata, e.g. LWN.net.

synapse/rest/media/v1/preview_html.py

@@ -15,7 +15,16 @@ import codecs
 import itertools
 import logging
 import re
-from typing import TYPE_CHECKING, Dict, Generator, Iterable, Optional, Set, Union
+from typing import (
+    TYPE_CHECKING,
+    Callable,
+    Dict,
+    Generator,
+    Iterable,
+    Optional,
+    Set,
+    Union,
+)
 
 if TYPE_CHECKING:
     from lxml import etree
@@ -146,6 +155,70 @@ def decode_body(
     return etree.fromstring(body, parser)
 
 
+def _get_meta_tags(
+    tree: "etree.Element",
+    property: str,
+    prefix: str,
+    property_mapper: Optional[Callable[[str], Optional[str]]] = None,
+) -> Dict[str, Optional[str]]:
+    """
+    Search for meta tags prefixed with a particular string.
+
+    Args:
+        tree: The parsed HTML document.
+        property: The name of the property which contains the tag name, e.g.
+            "property" for Open Graph.
+        prefix: The prefix on the property to search for, e.g. "og" for Open Graph.
+        property_mapper: An optional callable to map the property to the Open Graph
+            form. Can return None for a key to ignore that key.
+
+    Returns:
+        A map of tag name to value.
+    """
+    results: Dict[str, Optional[str]] = {}
+    for tag in tree.xpath(
+        f"//*/meta[starts-with(@{property}, '{prefix}:')][@content][not(@content='')]"
+    ):
+        # if we've got more than 50 tags, someone is taking the piss
+        if len(results) >= 50:
+            logger.warning(
+                "Skipping parsing of Open Graph for page with too many '%s:' tags",
+                prefix,
+            )
+            return {}
+
+        key = tag.attrib[property]
+        if property_mapper:
+            key = property_mapper(key)
+            # None is a special value used to ignore a value.
+            if key is None:
+                continue
+
+        results[key] = tag.attrib["content"]
+
+    return results
+
+
+def _map_twitter_to_open_graph(key: str) -> Optional[str]:
+    """
+    Map a Twitter card property to the analogous Open Graph property.
+
+    Args:
+        key: The Twitter card property (starts with "twitter:").
+
+    Returns:
+        The Open Graph property (starts with "og:") or None to have this property
+        be ignored.
+    """
+    # Twitter card properties with no analogous Open Graph property.
+    if key == "twitter:card" or key == "twitter:creator":
+        return None
+    if key == "twitter:site":
+        return "og:site_name"
+    # Otherwise, swap twitter to og.
+    return "og" + key[7:]
+
+
 def parse_html_to_open_graph(tree: "etree.Element") -> Dict[str, Optional[str]]:
     """
     Parse the HTML document into an Open Graph response.
@@ -160,10 +233,8 @@ def parse_html_to_open_graph(tree: "etree.Element") -> Dict[str, Optional[str]]:
         The Open Graph response as a dictionary.
     """
 
-    # if we see any image URLs in the OG response, then spider them
-    # (although the client could choose to do this by asking for previews of those
-    # URLs to avoid DoSing the server)
-
+    # Search for Open Graph (og:) meta tags, e.g.:
+    #
     # "og:type"         : "video",
     # "og:url"          : "https://www.youtube.com/watch?v=LXDBoHyjmtw",
     # "og:site_name"    : "YouTube",
@@ -176,19 +247,11 @@ def parse_html_to_open_graph(tree: "etree.Element") -> Dict[str, Optional[str]]:
     # "og:video:height" : "720",
     # "og:video:secure_url": "https://www.youtube.com/v/LXDBoHyjmtw?version=3",
 
-    og: Dict[str, Optional[str]] = {}
-    for tag in tree.xpath(
-        "//*/meta[starts-with(@property, 'og:')][@content][not(@content='')]"
-    ):
-        # if we've got more than 50 tags, someone is taking the piss
-        if len(og) >= 50:
-            logger.warning("Skipping OG for page with too many 'og:' tags")
-            return {}
-
-        og[tag.attrib["property"]] = tag.attrib["content"]
-
-    # TODO: grab article: meta tags too, e.g.:
+    og = _get_meta_tags(tree, "property", "og")
 
+    # TODO: Search for properties specific to the different Open Graph types,
+    # such as article: meta tags, e.g.:
+    #
     # "article:publisher" : "https://www.facebook.com/thethudonline" />
     # "article:author" content="https://www.facebook.com/thethudonline" />
     # "article:tag" content="baby" />
@@ -196,6 +259,21 @@ def parse_html_to_open_graph(tree: "etree.Element") -> Dict[str, Optional[str]]:
     # "article:published_time" content="2016-03-31T19:58:24+00:00" />
     # "article:modified_time" content="2016-04-01T18:31:53+00:00" />
 
+    # Search for Twitter Card (twitter:) meta tags, e.g.:
+    #
+    # "twitter:site"    : "@matrixdotorg"
+    # "twitter:creator" : "@matrixdotorg"
+    #
+    # Twitter cards tags also duplicate Open Graph tags.
+    #
+    # See https://developer.twitter.com/en/docs/twitter-for-websites/cards/guides/getting-started
+    twitter = _get_meta_tags(tree, "name", "twitter", _map_twitter_to_open_graph)
+    # Merge the Twitter values with the Open Graph values, but do not overwrite
+    # information from Open Graph tags.
+    for key, value in twitter.items():
+        if key not in og:
+            og[key] = value
+
     if "og:title" not in og:
         # Attempt to find a title from the title tag, or the biggest header on the page.
         title = tree.xpath("((//title)[1] | (//h1)[1] | (//h2)[1] | (//h3)[1])/text()")

tests/rest/media/v1/test_html_preview.py

@@ -370,6 +370,47 @@ class OpenGraphFromHtmlTestCase(unittest.TestCase):
         og = parse_html_to_open_graph(tree)
         self.assertEqual(og, {"og:title": "ó", "og:description": "Some text."})
 
+    def test_twitter_tag(self) -> None:
+        """Twitter card tags should be used if nothing else is available."""
+        html = b"""
+        <html>
+        <meta name="twitter:card" content="summary">
+        <meta name="twitter:description" content="Description">
+        <meta name="twitter:site" content="@matrixdotorg">
+        </html>
+        """
+        tree = decode_body(html, "http://example.com/test.html")
+        og = parse_html_to_open_graph(tree)
+        self.assertEqual(
+            og,
+            {
+                "og:title": None,
+                "og:description": "Description",
+                "og:site_name": "@matrixdotorg",
+            },
+        )
+
+        # But they shouldn't override Open Graph values.
+        html = b"""
+        <html>
+        <meta name="twitter:card" content="summary">
+        <meta name="twitter:description" content="Description">
+        <meta property="og:description" content="Real Description">
+        <meta name="twitter:site" content="@matrixdotorg">
+        <meta property="og:site_name" content="matrix.org">
+        </html>
+        """
+        tree = decode_body(html, "http://example.com/test.html")
+        og = parse_html_to_open_graph(tree)
+        self.assertEqual(
+            og,
+            {
+                "og:title": None,
+                "og:description": "Real Description",
+                "og:site_name": "matrix.org",
+            },
+        )
+
 
 class MediaEncodingTestCase(unittest.TestCase):
     def test_meta_charset(self) -> None: