Proposed fix for https://github.com/earwig/mwparserfromhell/issues/197

mwparserfromhell/parser/tokenizer.py

@@ -393,7 +393,8 @@ class Tokenizer:
             # scheme since it was just parsed as text:
             for chunk in reversed(self._textbuffer):
                 for char in reversed(chunk):
-                    if char.isspace() or char in self.MARKERS:
+                    # stop at the first non-word character
+                    if re.fullmatch(r"\W", char):
                         raise StopIteration()
                     if char not in valid:
                         raise BadRoute()

tests/tokenizer/external_links.mwtest

@@ -478,3 +478,17 @@ name:   brackets_scheme_title_but_no_url
 label:  brackets around a scheme, colon, and slashes, with a title
 input:  "[http:// Example]"
 output: [Text(text="[http:// Example]")]
+
+---
+
+name:   url_preceded_by_non_word_character
+label:  non-word character immediately before a valid URL
+input:  "svn+ssh://server.domain.com:/reponame"
+output: [Text(text="svn+"), ExternalLinkOpen(brackets=False), Text(text="ssh://server.domain.com:/reponame"), ExternalLinkClose()]
+
+---
+
+name:   url_preceded_by_underscore
+label:  underscore immediately before a valid URL
+input:  "svn_ssh://server.domain.com:/reponame"
+output: [Text(text="svn_ssh://server.domain.com:/reponame")]