Re: Fix misuse use of window_gettupleslot function (src/backend/executor/nodeWindowAgg.c)

Em dom., 5 de out. de 2025 às 13:05, Tom Lane <tgl@sss.pgh.pa.us> escreveu:

Ranier Vilela <ranier.vf@gmail.com> writes:
> Per Coverity.
> CID 1635309: (#1 of 1): Unchecked return value (CHECKED_RETURN)
> 7. check_return: Calling window_gettupleslot without checking return value
> (as is done elsewhere 8 out of 9 times).

Yeah, the security team's Coverity instance just whined about that
too.  But isn't the correct behavior simply "return -1"?

It seems to me a better option.

 

  It looks
to me like a failure should be interpreted as "row doesn't exist,
therefore it's not in frame".

I also believe that the original author did not expect a failure here.

What would be really useful is a test case that reaches this
condition.  That would make it plain what to do.

There is a comment above that indicates that possibly a failure could also be the end of the partition.

v1 patch attached.