On Mon, 23 Mar 2026 at 19:30, Tender Wang <[email protected]> wrote: > Recently, I encountered an error: could not find memoization table
> The hkeys returned by datum_image_hash() are different, so we can't > find the entry in the hash table. > > In the datum_image_hash(), if typByVal is true, calling > result = hash_bytes((unsigned char *) &value, sizeof(Datum)); > > I think we should use typLen here, not sizeof(Datum). > I tried this way and didn't encounter any errors again. The Datum values should be the same. You can't just compare the lowest attlen bytes of a Datum. It looks to me like the bug is in hash_numeric(). Seems like it has no idea what type it's meant to return. hash_numeric_extended() doesn't seem to be much better. Do you still get the ERROR after patching with the attached? David
numeric.patch
Description: Binary data
