Re: View definition changes after reloading pg_dump export

On 9/1/22 09:08, Tom Lane wrote:
> Wesley Schwengle <wesley.schwengle@xxllnc.nl> writes:
>> There is a view that we create and it uses a function and the view
>> definition changes between runs. I'm not sure why this is happening,
>> does someone know?
> The core reason for the discrepancy is that the parser inserts
> implicit coercion steps into your initial, not-very-consistently-typed
> query; but the view-dumping logic (ruleutils.c) tends to display
> those steps as explicit coercions.  Formally speaking it shouldn't
> do that, because the parser may act differently depending on whether
> a cast is explicitly present, as indeed you see here.  But if we don't
> write the coercions explicitly then there is a risk of the reloaded view
> being interpreted differently than it was before (perhaps because there
> are more or different overloaded functions/operators available at reload
> time).  So the dumping logic is kind of between a rock and a hard place.
> Our experience has been that printing the coercions explicitly causes
> fewer problems than not doing so, so that's what it does.
>
>             regards, tom lane
>
>

"foo.baz::character varying::text" seems pretty odd.


-- 
Angular momentum makes the world go 'round.