Corentin Jabot:
> Unicode advise not doing that at all
> http://www.unicode.org/faq/private_use.html#pua10

That paragraph talks about operating system. I'm proposing
application-level API.

Steve Downey:> ICU, as a reference, only allows customizing the database
at build time,
> not at runtime. They do call out custom properties for the PUA, though.
>
> http://userguide.icu-project.org/strings/properties#TOC-Customization

Hmm, and what if I want to work with different sets of PUA? They talk
about modifying existing file, not adding new one.

JeanHeyd Meneide:
> Because the unicode character database properties have pervasive uses in
> Unicode (staring with Normalization, and branching out from there), it
> would be incredibly hard to properly provide all of the C++ algorithms a
> chance to override the database used. You'd have to make sure every single
> algorithm can receive the database;

Yes, exactly. If algorithm messes with UCD and doesn't receive const
std::unicode::character_database&, it's broken.