Re: ON CONFLICT DO SELECT (take 3)

On Wed, 11 Feb 2026 at 14:55, Viktor Holmberg <v@viktorh.net> wrote:

Looking more at this, I’m quite sure that the p_is_insert field can just be removed?
See 0002.

Ah, good idea. Well spotted!

This dates back to c1ca3a1, which removed a similar p_is_update field,
but noted that using p_is_insert wasn't particularly pretty.

Going back even further, it looks like p_is_insert and p_is_update
used to be much more widely used, but now we're down to just this one
place in transformAssignedExpr() that reads p_is_insert, and as you
say, it can deduce the same information from the exprKind passed to
it, which is much neater.

Barring objections, I'll push both those shortly.