Re: [HACKERS] Regression tests vs existing users in an installation

OK, here's a completed patch to add checking for naming-rule violations.

I updated regress.sgml to clarify the naming rules (and failed to
resist the temptation to update a lot of other somewhat-obsolete
statements there, too).

Also worth noting is that I added an IsReservedName check to
pg_replication_origin_create(), ie it will no longer allow
user-selected origin names starting with "pg_". This seems
like a good idea considering that we generate internal origin
names that look like that. I have not done anything equivalent
for subscription names, but should we consider doing so?

I touched the TAP tests only to the extent necessary to make them pass
cleanly --- mostly I had to fool with pg_dump/t/010_dump_connstr.pl
because it whines if the restore step emits any warnings. Perhaps
there's another way to address that with less invasive changes to that
test script; but I couldn't think of one other than ignoring warnings,
which didn't seem like a great idea.

This patch doesn't address the issue that we think rolenames.sql is
unsafe to run in "make installcheck" mode. That seems like a separable
problem, and we'd have to adjust that script as shown here anyway.

Barring objections, I'll push this shortly (to HEAD only) and turn on
the enabling switch on one or two of my buildfarm critters.

regards, tom lane