All good ideas and thoughts. I have been thinking that essentially
separate types per character set is the right thing, but we'll have
plenty of time to talk about it.
One point is that SQL92 assigns a specific character set and collation
sequence to every character string and every column definition; if we
embedded this "type" identification into every string then we would be
replicating the existing Postgres type system one layer down (at least
for argument's sake ;)
There also need to be well defined conversions between character
sets/collations, and some or most combinations will be illegal (e.g.
how do you collate American English against Japanese?). The Postgres
type system can enforce this simply by not providing conversion or
comparison functions for the relevant mixture of types.
- Thomas
--
Thomas Lockhart lockhart@alumni.caltech.edu
South Pasadena, California