Andrew Dunstan <andrew@dunslane.net> writes:
> On 12/30/2014 09:20 AM, Tom Lane wrote:
>> In one light this is certainly a bug fix, but in another it's just
>> definitional instability.
>>
>> If we'd gotten a field bug report we might well have chosen to back-patch,
>> though, and perhaps your client's complaint counts as that.
> I got caught by this with ON_ERROR_ROLLBACK on 9.3 just this afternoon
> before remembering this thread. So there's a field report :-)
> +0.75 for backpatching (It's hard to imagine someone relying on the bad
> behaviour, but you never know).
It seems like there's a consensus in favor of back-patching this change,
so I'll go ahead and do that.
regards, tom lane