Re: text_pattern_ops and complex regexps

Tom Lane wrote:
> Stephen Frost <sfrost(at)snowman(dot)net> writes:

> > Perhaps this is misguided but I would think that the regexp
> > libraries might have some support for "give me all anchored required
> > text for this regexp" which we could then use in the planner.
>
> I wouldn't see why. It's certainly worth considering to hand the
> pattern to the regex engine and then burrow into the data structure it
> builds; but right now we consider that structure to be entirely private
> to backend/regex/. There's also the problem that we'd have no easy
> way to determine how much the result depends on the current regex flavor
> setting. There are some cases now where regex_fixed_prefix deliberately
> omits possible optimizations because of uncertainty about the flavor.

I think changeable regex flavors turned out to be a bad idea. They can
wreak all sorts of havoc. You change the setting, SIGHUP, and suddenly
your application fails to work as expected. Maybe we should make that
setting PGC_POSTMASTER (or just get rid of it?), and provide was to pass
flags to change the flavor for particular operations (this is easy for
function-based stuff but not so easy for operators). That way it
doesn't intrude in stuff like cached plans and so on.

--
Alvaro Herrera http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.