[clon]

The plain utf8 covers the "basic multilingual plane" (x0000-xFFFF), so it will get you very far, actually.

In our app, we finally went for utf8mb4 to allow people to enter emoji. As a side "bonus" you will also soon see some clever people entering their names to stand out, such as "𝙹𝚘𝚑𝚗". Note that this is not "John". It is actually a series of mathematical symbols:

  U+1D679   MATHEMATICAL MONOSPACE CAPITAL J
  U+1D698   MATHEMATICAL MONOSPACE SMALL O
  U+1D691   MATHEMATICAL MONOSPACE SMALL H
  U+1D697   MATHEMATICAL MONOSPACE SMALL N

[capitainenemo]

One fun thing you can do in MariaDB is define a collation based off of the Unicode confusables list. The result of this is that select * from usernames where user = 'John' would still work :)

It also avoids registering both 'John' and 'John'. (changed the o)

Hedgewars user registration does this.

[sam_lowry_]

That's a great feature. I was running a community site once where impersonation became a popular game. So we had dozens of users registering accounts with Cyrillic i,o,a,e etc.

[Ndymium]

This is unfortunate as it makes things unreadable for screen readers. I have seen many tweets from companies and organisations that use these characters to replace funny "fonts" or create emphasis effects in the text, rendering them inaccessible in the process.

In my own service I plan to offer users the ability to define a "stylised" nickname and an alternative text for screen readers, should they be interested in it (like an alt text for images). If Twitter had such a feature, I'd use it because I really like the effects, even if they're cheesy. :)