Jay Pipes wrote: > Monty Taylor wrote: >> Hey all, >> >> Similar to the charsets conversation, we were also talking about >> collations. Currently, of course, we do these per-column, with defaults >> available per table and per schema. How useful is this really? What are >> people doing who are supporting multi-languages in their systems. Are >> they _actually_ storing the multi-language data in different columns >> with different collations? > > Never seen this ever. Typically, I see a default charset and collation > for either a schema (most common) and sometimes for a table. Never > different collations per column. > >> On the other hand, if there was a per-server collation set by locale, >> would that be sufficient? > > Nah, a schema collation is probably a good thing. Think of a hosted > environment where different schemas for different websites around the > world are set differently.
I tell you -- these hosted people start to drive me batty... :) >> I'm asking for two reasons. One, targeting the cloud rather than the >> enterprise, I'm wondering how many people are wanting their database to >> handle multi-language sorting directly. > > Again, hosted solutions will need per-schema. > >> The second is that in the quest >> to remove our internally implemented charset/collation situation, the >> other libs that do this don't seem to have a good way to do per-method >> collation picking. So if we don't actually need to have the collation >> switch during the lifetime of the process, we can use system locale >> processing. If we _do_ there are other, less attractive options... >> although we'll do what we have to do I suppose. > > Perhaps the question to ask is, do we want the ability to do > cross-schema collation switches? For example, a query which joins > across multiple schemas with different collations. I'm pretty confident > this is not a common occurrence at all. If we could impose a limit that > queries across multiple schemas on a single server would be constrained > to the server locale, then there wouldn't be a need to consider > cross-schema collation switching, which would solve the problem, no? I wish. The problem with the collation libraries I've looked at so far that don't require utf-16 like libicu does (like system libc or glib) is that they except locale, and thus collation to be global. So any switching within the lifespan of our process is a global operation. Anybody know of a locale/collation lib that allows me to pass in a locale as a param but also lets me play with utf8 natively? _______________________________________________ Mailing list: https://launchpad.net/~drizzle-discuss Post to : [email protected] Unsubscribe : https://launchpad.net/~drizzle-discuss More help : https://help.launchpad.net/ListHelp
