Now that GIN can index null items, its amindexnulls flag in pg_am really
ought to be set true. However, looking at the usage of that column,
I wonder whether we shouldn't just delete it instead. The only present
usage is a rather convoluted test in CLUSTER to see if the index is
safely clusterable --- and AFAICS that test has been obsoleted anyhow
by the later addition of pg_am.amclusterable. Anyone against
simplifying matters by getting rid of pg_am.amindexnulls?
regards, tom lane