Recover invalid twtxt hashes in subjects

test_twtxtparser.py

@@ -147,6 +147,24 @@ class ParseSubjectTest(unittest.TestCase):
         self.assertEqual(SubjectHash(start_pos=0, end_pos=18, hash="abcd567", url=None),
                          _parse_subject("(  #<  abcd567 > ) more text", 0))
 
+    def test_only_non_twtxt_hash(self):
+        self.assertEqual(SubjectHash(start_pos=0, end_pos=10, hash="abcd567", url=None),
+                         _parse_subject("(#abcd567) more text", 0))
+        self.assertEqual(SubjectHash(start_pos=0, end_pos=14, hash="abcd567", url=None),
+                         _parse_subject("(  #abcd567  ) more text", 0))
+
+    def test_only_non_twtxt_hash_too_short(self):
+        self.assertIsNone(_parse_subject("(#abcd56) more text", 0))
+        self.assertIsNone(_parse_subject("(  #abcd56  ) more text", 0))
+
+    def test_only_non_twtxt_hash_too_long(self):
+        self.assertIsNone(_parse_subject("(#abcd5678) more text", 0))
+        self.assertIsNone(_parse_subject("(  #abcd5678  ) more text", 0))
+
+    def test_only_non_twtxt_hash_with_other_text(self):
+        self.assertIsNone(_parse_subject("(#abcd567 a) more text", 0))
+        self.assertIsNone(_parse_subject("(  #abcd567 a ) more text", 0))
+
     def test_only_url(self):
         self.assertEqual(SubjectHash(start_pos=0, end_pos=25, hash=None, url="https://example.com/"),
                          _parse_subject("(#<https://example.com/>)", 0))

twtxtparser.py

@@ -111,11 +111,36 @@ _TWT_HASH_RE = re.compile(r"^[a-z0-9]{7}$")
 
 def _parse_subject(text, start_pos):
     data, closing_pos = _extract_between(text, start_pos, None, "(", ")")
-    if data is None:
+    if not data:
         return None
     subject_hash_start_pos = _skip_whitespace(text, start_pos + 1)
     if text[subject_hash_start_pos:subject_hash_start_pos + 2] != "#<":
-        return None
+        # This is not a valid twtxt hash tag, let's see if this looks at least
+        # like a non-twtxt hash, because lately this comes up every now and then
+        # in my timeline. Maybe we can recover the technically invalid twt hash
+        # from broken clients and put the twt in the correct conversation. Twt
+        # hashes are always seven chars long, so with the hash sign this makes a
+        # total length of eight.
+        if text[subject_hash_start_pos] != "#" or len(data) < 8:
+            # nope, we're out of luck
+            return None
+
+        hash_end_pos = subject_hash_start_pos + 8
+        hash = text[subject_hash_start_pos + 1:hash_end_pos]
+        if not _TWT_HASH_RE.match(hash):
+            return None
+
+        # there must be no other text in the subject, only optional whitespace
+        if text[hash_end_pos].isspace():
+            hash_end_pos = _skip_whitespace(text, start_pos=hash_end_pos,
+                                                  end_pos=closing_pos)
+        if hash_end_pos + 1 != closing_pos:
+            return None
+        return SubjectHash(start_pos, closing_pos, hash, url=None)
+
+    # Chances are this is a real twtxt hash tag. At least the start looks like
+    # one.
+
     subject_hash = _parse_subject_hash(text, subject_hash_start_pos, closing_pos)
     if subject_hash is None:
         return None