On Sun, Jun 23, 2019 at 10:26:47PM -0400, Robert Treat wrote:
>On Sun, Jun 23, 2019 at 1:25 PM Peter Eisentraut
><peter.eisentraut@2ndquadrant.com> wrote:
>>
>> On 2019-04-12 19:52, Robert Treat wrote:
>> > It is clear to me that the docs are wrong, but I don't see anything
>> > inherently incorrect about the code itself. Do you have suggestions
>> > for how you would like to see the code comments improved?
>>
>> The question is perhaps whether we want to document that non-matching
>> data types do work. It happens to work now, but do we always want to
>> guarantee that? There is talk of a binary mode for example.
>>
>
>Whether we *want* to document that it works, documenting that it
>doesn't work when it does can't be the right answer. If you want to
>couch the language to leave the door open that we may not support this
>the same way in the future I wouldn't be opposed to that, but at this
>point we will have three releases with the current behavior in
>production, so if we decide to change the behavior, it is likely going
>to break certain use cases. That may be ok, but I'd expect a
>documentation update to accompany a change that would cause such a
>breaking change.
>
I agree with that. We have this behavior for quite a bit of time, and
while technically we could change the behavior in the future (using the
"not supported" statement), IMO that'd be pretty annoying move. I always
despised systems that "fix" bugs by documenting that it does not work, and
this is a bit similar.
FWIW I don't quite see why supporting binary mode would change this?
Surely we can't just enable binary mode blindly, there need to be some
sort of checks (alignment, type sizes, ...) with fallback to text mode.
And perhaps support only for built-in types.
regards
--
Tomas Vondra http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services