Both the parsers for commit messages and the code comments are broken in odd ways.
http://www.mediawiki.org/wiki/Special:Code/MediaWiki/16967#code-comments is a fairly good example of two issues.
mysql> select cc_text from code_comment where cc_rev_id = 16967\G
- 1. row *******
cc_text: The line "if($vCode==$wgContLang->getCode()) continue; // we don't want default variant" doesn't work correctly on zhwiki (and other wikis with $wgLanguageCode = 'zh'), where the language code 'zh' is also used as a variant name (which means 'no conversion'; technically it's a variant). Omitting zh variant causes /zh/Page_Name urls (when $wgArticlePath = '/$2/$1') not to be purged.
- 2. row *******
cc_text: Filed to bugzilla as [[bugzilla:24027|bug 24027]].
- 3. row *******
cc_text: <nowiki>bug 24027</nowiki> will link to bugzilla :)
3 rows in set (0.02 sec)
... shows the raw text. The parser doesn't listen to <nowiki> and it doesn't properly expand interwiki links like "bugzilla:".
http://www.mediawiki.org/wiki/Special:Code/MediaWiki/67090 demonstrates a problem with the commit message parser. The raw text is "The reason for this change is given in http://lists.wikimedia.org/pipermail/wikitech-l/2010-April/047617.html." However the trailing ">" is turned into > inexplicably.
Some parser tests (or a change in the overall implementation of CodeReview's message parsing) would be good.