Same problem with ZIM files, "collection.zim".

I just submitted a patch for this bug: gerrit change I0c53d31e43354a63dcc115b59a961e43bb86e8c4

It still has a few issues though:

• it works only for Latin-1
• Accented characters will be converted to ASCII
• Everything else will be dropped
• If everything is dropped, "collection" + EXT will be used as filename

(In reply to comment #0)

Better would be a meaningful file name title which corresponds to the Wiki
book/collection page title. Non-ASCII character need to be stripped, or
transliterated or url-encoded, and spaces replaced with underscore.

Please don't forget that titles in non-english wiki projects are completely Non-ASCII. So don't drop them :)
What issue do you see when file name contains valid UTF-8 characters?

This is a quick fix. HTTP can only handle ASCII characters by default. RFC 5987 describes a way to handle UTF-8 filenames, but this requires some additional work and will not work in legacy browsers.