MediaWiki currently relies heavily on Unicode support to provide support for 300+ languages yet does not require the mbstring PHP extension to function. Instead, we create PHP-only fallbacks if a native support is not available. This creates a few problems:
- These fallbacks are extremely slow. The script in P2734 demonstrates that fallbacks are roughly order of magnitude slower on PHP 5.6. In extreme cases, it can be 100+ times slower, per comment in Fallback.php).
- These fallbacks cover only a few functions. If there's no fallback, either ad-hoc solutions are used in places, or, like in SwiftFileBackend, we just say "mbstring is required".
- This also means that extensions can't expect any consistent Unicode support.
- Won't somebody please think of the children!
Now that we've dramatically increased PHP requirements, we've already cut off a lot of crappy environments so this change will likely not affect too many users.
OS support:
- On Debian-based systems, a simple apt-get install php5 gives you mbstring by default.
- On RPM-based, a separate package is required
- On Windows, people tend to use *AMP all-in-one packages that have mbstring.
Current mbstring usage in core (excluding fallbacks themselves):
mediawiki/includes$ grep -orEh '\bmb_\w+' . | sort | uniq -c 7 mb_check_encoding 6 mb_convert_encoding 12 mb_strlen 4 mb_substr
Some time ago, I committed https://gerrit.wikimedia.org/r/#/c/267309/ to start a discussion, but it went largely unnoticed so I'd like to start a formal RFC.