Andres Freund <andres@2ndquadrant.com> writes:
> On 2015-02-26 17:45:26 -0500, Tom Lane wrote:
>> With all due respect, that's utterly wrong. I have looked at every single
>> MemoryContextReset call in the codebase, and as far as I can see the
>> *only* one that is in an error path is elog.c:336, which I'm quite certain
>> is wrong anyway (the other reset of ErrorContext in that file uses
>> MemoryContextResetAndDeleteChildren, this one should too).
> Sure, in the pg codebase. But there definitely are extensions using
> memory contexts. And there's code that has to work across several
> branches. Backpatching alone imo presents a risk; there's nothing to
> warn you three years down the road that the MemoryContextReset() you
> just backpatched doesn't work the same in the backbranches.
> If the changes breaks some code it's likely actually a good thing:
> Because, as you say, using MemoryContextReset() will likely be the wrong
> thing, and they'll want to fix it for the existing branches as well.
Is that likely to happen? I doubt it. They'll just mutter under their
breath about useless API breakage and move on.
Basically, this is a judgment call, and I disagree with your judgment.
I think changing the behavior of MemoryContextReset is exactly what we
want to have happen. (And I've got to say that I'm losing patience with
backwards-compatibility arguments taken to this degree. We might as
well stop development entirely.)
regards, tom lane