(Footnote on parsing the thread into individual comments: simply using signatures is probably best. Even when a user forgets to sign, a bot comes along and signs for them. One could imagine a "smarter" parsing which also detects unsigned comments (which the bot has missed or hasn't gotten to yet), using indentation, but that would be confounded by the following use of indentation for a block quote: https://en.m.wikipedia.org/wiki/Help_talk:Diff#What_is_the_diff_algorithm_used_in_Wikipedia%2C_and_can_we_improve_it%3F )

Note: this is a specific case of whitespace handling, which I have laid out in detail because I can specifically suggest the fix.

Someone with experience would be much better than me, but, if someone were to point me to the exact section of code which is doing the things we don't like, perhaps I could propose some specific changes. (My CV: some experience with regex; zero experience with MediaWiki, diff, or data comparison.)