What I have is data with two different characters for "start quote"
and "end quote". In my case it's '[' and ']', but it could be
anything from "smart quotes", to parentheses, to brackets, braces, ^/$
in regexps, etc. I think this isn't too unreasonable a feature to
have to make copy more functional when importing data that is
difficult to transform properly beforehand (in my case is about half a
terabyte of log files, which takes hours and hours, just to cat, let
alone reparse and dump into COPY).
I think regexps would be going too far. One simple approach which
wouldn't involve any grammar changes would be to allow the quote
parameter to be 2 chars long instead of one, and treat the second as the
closing quote char (which would otherwise default to the first char). An
argument against that would be that it would preclude us from in future
allowing multi-char quote marks. I'm not sure if that matters quite so
much as it would for the delimiter parameter.
Now, in my case I can just say "cat file | tr '' '""' | psql -f
import.sql", but then I lose the ability for psql to do anything smart
like using mmap (I'm making assumptions that it does anything smart
like that, but even if it doesn't now, it could some day).
Well, you also earn a "Useless use of cat" award ;-)
And no, we don't mmap the file, but we do do some smart buffering stuff,
thanks to Greenplum.
---------------------------(end of broadcast)---------------------------
TIP 2: Don't 'kill -9' the postmaster