> > Actually it seems like the hard part is not so much the input
> > representation as the output representation --- what should the
> > base-level initcap routine return, to be reasonably efficient for
> > both cases?
>
> I hadn't gotten to trying it out yet, but I can see the output being a
> problem. You can't even really pre-allocate the storage before passing
> it because you don't know the length after case change. You could pass
> back a char* and repalloc to get the varlena header in there but that is
> very messy.
>
> Add to that that the multi-byte case also has to be converted to wide
> characters, so you have text -> char * -> wide chars -> char * -> text
> for the most complex case.
>
> I am starting to think that the simplest case is to keep the single-copy
> version in there for single-byte encodings and not worry about the
> overhead of the multi-byte case.
My new idea is if we pass the length to str_initcap, we can eliminate
the string copy from text to char *. That leaves us with just one extra
string copy from char * to text, which seems acceptable. We still have
the wide char copy but I don't see any easy way to eliminate that
because the multi-byte code is complex and not something we want to
duplicate.
--
Bruce Momjian <bruce@momjian.us>
EnterpriseDB
+ If your life is a hard drive, Christ can be your backup. +
--
Sent via pgsql-patches mailing list (pgsql-patches@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-patches
No comments:
Post a Comment