>
> On Sun, Jan 11, 2026 at 04:58:39PM +0500, Kirill Reshke wrote:
> > Maybe, I have stopped some more cases, in v2-0001
>
> Right. It's true that we could be more consistent for all these based on their
> base type, some of them, particularly in the GIN code now, caring about using
> the correct macro. It may be a good occasion to double-check the whole tree
> for similar holes based on unsigned types.
> --
> Michael
Hi Kirill, Roman, and Michael,
While double-checking the tree for similar holes as Michael suggested, I noticed a couple more inconsistencies in
contrib/pageinspect/ginfuncs.cwhere we are using signed macros for unsigned types.
Specifically, in gin_page_opaque_info, we use Int32GetDatum for maxoff:
values[1] = Int32GetDatum(opaq->maxoff);
Since maxoff is of type OffsetNumber (uint16), this seems to be the exact same pattern Kirill is addressing in other
partsof the GIN code. Although it is widened to int32 here, for the sake of consistency, it should probably be using a
16-bitor unsigned macro.
Similarly, in gin_metapage_info, tailFreeSize (which is defined as uint32 in GinMetaPageData) is also converted using
Int32GetDatum:
values[2] = Int32GetDatum(metadata->tailFreeSize);
It might be better to include these cleanups in the next version of the patch to ensure all pageinspect GIN functions
followthe same standard.
Best regards,
Yuan Li(carol)
