Redirect loop after editing a question title ending with no-break space, question mark

Question

I edited the title of the question Quelle est la signification de l'expression "une fille indigne" ? Comment l'utilise-t-on au quotidien ? from the moderator dashboard. Original title:

Une fille indigne -- Qu'est que c'est? Comment on le utilise dans le conversation quotidien?

New title:

Une fille indigne — est-ce une allusion ?

Now the page is caught in an infinite reditect loop from https://french.stackexchange.com/questions/10871/une-fille-indigne-%e2%80%94-est-ce-une-allusion%c2%a0 to itself.

Other posts exhibit the same behavior: https://french.stackexchange.com/posts/8962/revisions, https://french.stackexchange.com/posts/10870/revisions. On the other hand, Comment dire « go take a flying leap » en français ? and What do French tennis announcers say when the game is tied at 15 or 30 ?, which have similarly constructed titles but were edited earlier, are fine. It looks like titles ending with U+00A0 NO-BREAK-SPACE followed by U+003F QUESTION MARK used to be silently rewritten to U+0020 SPACE followed by U+003F QUESTION MARK, and now they are no longer mangled. It's nice not to have typographical errors silently introduced by the system, but it would be nicer if that didn't prevent the page from loading.

(Note that an unbreakable space — it should be a thin space but most people type U+00A0 or have software that does it for them — before a question mark is normal typography in France. All that English text with no space before the question mark looks weird to us.)

Answer 1

This is another occurrence of web-servers and web-frameworks struggling with utf-8 encoded urls; for previous episodes, please see here and here. Well, it looks like it doesn't like these urls either (it comes in as though it were truncated without the whitespace/punctuation pair, hence it tries to redirect to a canonical url). We recently enabled utf-8 urls network-wide, which is why it has suddently appeared. I have disabled this again while I investigate.

Incidentally, you can't repro this on most sites: we have a flag that changes how whitespace-followed-by-punctuation is handled in question titles - and it is only enabled on language sites.

But: I have access to the low-level http request, so I'm working on a way of completely ignoring all the framework binding layers, so that we just read the raw data directly. When I have this deployed, I will re-enable the utf-8 url support.