[jng]

I'm a bit surprised that Posgres doesn't optimize the COUNT(column) expression to not actually read the value of the column, which is not necessary. Am I missing something?

[nicwolff]

Pg has to read the xmin and xmax anyway to determine if the row is visible to your transaction (for rows without hint bits set, in pages without hint bits set).

[nattaylor]

It does have to read it because NULLs aren't counted

[teddyh]

If the column is NOT NULL, does it still read it?

[avianlyric]

Shouldn’t do. Postgres is usually pretty good at making these types of optimisations.

[zeroimpl]

Same - the row format itself starts with a bitmask listing all null columns. You don’t need to read the column data itself to determine if a column is null.

https://www.postgresql.org/docs/current/storage-page-layout....