Is this the best approach to this? I feel like since we already run all HTML pages that pretalx serves through a parser anyway, wouldn't it be better to use the parsed information to find URLs to replace?

In any case, we should take this action only on pretalx-generated HTML (and CSS, where needed – we already find URLs in CSS via find_urls. By checking the output for "DOCTYPE html", you a) potentially increase the runtime by a lot, because if that string is not present, the whole potentially very long string will be checked, and b) you'll match e.g. user-uploaded HTML and rewrite it, which we really should not do.

(a) could be alleviated by only checking content[:20] for the substring match, but b) still stands …)

I wonder if it wouldn't be cleaner to ask the user to provide a prefix that we can then supply to the functions here. It could be used with settings.FORCE_SCRIPT_NAME to make static files appear at the correct location (potentially) without laboriously having to rewrite all static files to relative paths. Would have to test if this works correctly, including adding support to EventUrl etc for this, but if that can be made to work, it’d be infinitely prefereable to regex-parsing HTMLs for URLs, imo.