>> Urgh ... that means GiST indexing is actually broken, because GiST
>> currently handles multicolumns but not nulls.
Actually, it appears that 7.2 GiST does handle NULLs in columns after
the first one, which I think is enough to avoid the problem Jan
mentioned. The boolean column pg_am.amindexnulls is not really
sufficient to describe this behavior accurately. Looking at current
uses it seems correct to leave it set FALSE for GiST.
In short: false alarm; the 7.2 code is okay as-is, at least on this
particular point.
regards, tom lane