Re: bytea_output vs make installcheck

Andres Freund <andres(at)anarazel(dot)de> writes:
> On 2017-02-15 09:30:39 -0800, Jeff Janes wrote:
>> I don't really see the cost here.

> Because that means we essentially need to make sure that our tests pass
> with a combination of about ~20-30 behaviour changing gucs, and ~5
> different compilation settings that change output.

Yeah, the problem with addressing this non-systematically is that it'll
never stay fixed for long.

> Alternatively we could ALTER DATABASE a bunch of settings on the
> regression database at the start, but I'm not sure that's nice either,
> because it makes ad-hoc tests with unusual settings harder.

I'd definitely be -1 on that.

I think that it is worth fixing cases where a parameter change leads to
surprising results, like the operator_precedence_warning case just now.
But people should not be surprised when, say, changes in planner
parameters lead to different EXPLAIN output or different row ordering.
If we tried to lock that down it'd be counterproductive for the reason
Andres mentions: sometimes you *want* to see what you get for other
settings.

regards, tom lane