Re: Proposal for fixing numeric type-resolution issues

> Thomas Lockhart writes:
>
> > Another 7.1 project is to work on alternate languages and character
> > sets, to decouple multibyte and locale from the default SQL_TEXT
> > character set. This will probably bring up issues similar to the
> > numeric problems, and since these character sets will be added as
> > user-defined types it will be important for the backend to understand
> > how to convert them for comparison operations, for example.
>
> Really? I always thought the character set would be some separate entity
> and perhaps an oid reference would be stored with every character string
> and attribute. That would get you around any type conversion as long as
> the functions acting on character types take this "header" field into
> account.

I think that way too. If what Thomas is suggesting is that to make a
user-defined charaset, one need to make everything such as operators,
charset, functions to work with index etc. (like defining new a data
type), that would be too painfull.

> If you want to go the data type way then you'd need to have some sort of
> most general character set to cast to. That could be Unicode but that
> would require that every user-defined character set be a subset of
> Unicode, which is perhaps not a good assumption to make.

Right. But the problem is SQL92 actually requires such a charset
called "SQL_TEXT." For me, the only candidate for SQL_TEX at this
point seems to be "mule internal code." Basically it is a variant of
ISO-2022 and has a capability to adapt to most of charsets defined in
ISO-2022. I think we could expand it so that it could become a
superset even for Unicode. Of course the problem is mule internal code
is a "internal code" and is not widely spread in the world. Even
that's true we could use it for purely internal purpose (for the parse
tree etc.).

> Also, I wonder
> how collations would fit in there. Collations definitely can't be ordered
> at all, so casting can't be done in a controlled fashion.

Hmm... Collations seem to be a different issue. I think there's no
such an idea like "collation casting" in SQL92.
--
Tatsuo Ishii