Vik Fearing <vik.fearing@2ndquadrant.com> writes:
> On 27/11/2018 01:13, Stephen Frost wrote:
>> Parallel safe functions should be marked as such. Immutable functions
>> should be marked as such. We should not assume that one implies the
>> other, nor should we operate as if they do.
> Yes we should! Unless you can produce a case where an immutable
> function is not parallel safe.
As far as that goes, I agree with the idea of adding an oprsanity test
that shows any built-in functions that are immutable but not parallel
safe, on the grounds Stephen mentioned: it's probably a mistake, and
if it isn't, we can add that function to the expected output.
I'm way less inclined to buy into the idea that it MUST be wrong, though.
Immutability is a promise about result stability and lack of side effects,
but it is not a promise about implementation details. There could be an
implementation reason not to run something in a parallel worker. Off the
top of my head, a possible example is "it's written in plfoo which hasn't
yet been made to work correctly in parallel workers".
regards, tom lane