Greg Stark <stark@mit.edu> writes:
> Huh, I had assumed this was old behaviour. I didn't realize this was
> new with 9.3.
> Considering the thread "pg_get_viewdefs() indentation considered
> harmful" I'm beginning to think this was a regression. It results in
> some dump files being unnecessarily large and the pg_dump consuming
> too much memory and crashing.
> Tom liked my suggestion of doing the indentation modulo 40. There are
> plenty of other suggestions that can work too, giving up on
> indentation after an indent of 40, switching to an indent distance of
> 1 if it's more than 10 levels deep, and so on. I think it doesn't
> really matter which we choose but we have to do something. And given
> this is new behaviour in 9.3 perhaps it should be backported too.
I'm still a bit skeptical about this being a catastrophic problem in
practice ... but anyway, I thought there were two somewhat different
proposals on the table in that thread:
1. Arrange for "x UNION y UNION z ..." to put all the UNION arms at
the same indentation level.
2. Change the indentation rules globally, in one or another fashion
as Greg mentions above, to prevent ruleutils from ever prepending
silly amounts of whitespace.
These are not mutually exclusive, and I think we should do both.
But it seems we're hung up on exactly which flavor of #2 to do.
I would argue that rules such as "reduce the indent step once we're too
far over" don't fix the problem, just postpone it. If we're taking this
complaint seriously at all then there needs to be a guaranteed maximum
indent distance, one way or another. So I'd go for either a modulo-N
rule or a hard limit, with some preference for the modulo-N approach.
Yeah, it does sound silly at first, but I think it'd preserve
readability better than just capping the indent.
regards, tom lane