On 3/4/15 12:20 PM, Marko Tiikkaja wrote:
> On 3/4/15 6:19 PM, I wrote:
>> On 3/4/15 5:26 PM, Tom Lane wrote:
>>> It was considered to be a bug fix; more, given the few complaints about
>>> the clearly-broken old behavior, we thought it was a fix that would
>>> affect
>>> few people, and them positively.
>>
>> Yeah, but these things usually go the other way. "This has been broken
>> for 10 years but nobody noticed, so we're not going to fix this"
>
> Sorry, that should have said "we're not going to fix this *in the back
> branches*".
I tend to agree. I was not in favor of backpatching, but other people
were in favor of it, and no one spoke up against it.
That said, if I understand your test case correctly, this would
basically be an argument against any semantic corrections in stable
releases, since user code could expect to continue to work with the
previous incorrect value.
You can always test the server version number, and you'll have to for
the next major release anyway.