I wrote:
> Yeah, I read that. I think making the backend read these files directly
> makes a ton of sense, so I'm no longer very excited about improving the
> performance or memory consumption of the replace_token stuff.
Actually, wait a second. I grant your point about postgres.bki, but
a whole lot of the replace_token calls in initdb are messing with
the configuration files, which (a) is something I don't see changing,
and (b) pretty much none of that could be pushed back to build time.
So even though the config files are a lot smaller than postgres.bki,
maybe there's still a point in polishing replace_token a bit.
regards, tom lane