Adrian Klaver <adrian.klaver@xxxxxxxxxxx> writes: > On 2/17/20 7:01 AM, Jeremy Finzel wrote: >> I'm still not sure why a rebuild of this index would be required, >> technically speaking. But perhaps in any case the docs should have >> something to the effect that expression indexes may require rebuild >> under specific circumstances? > [ it is already ] Yeah. In principle, we wouldn't need to rebuild the indexes in this case, since there's no semantic difference between a value sourced from a varchar(N) column and a varchar(some-other-N) column. In general, though, ALTER COLUMN TYPE doesn't know whether that's true; and there are definitely binary-compatible cases where it *does* matter. (An example is that coercing integer to OID is allowed without a table rewrite, but an index rebuild is needed because the sort order is different.) So right now, any index mentioning the altered column has to be rebuilt. Maybe someday we'll figure out how to do better. I'm kind of wondering whether it wouldn't be safe to assume that changes that only change the typmod and not the type OID don't require index rebuilds. regards, tom lane
