BTW, I spent some time adding a cross-check to see if the code was at
least approximately correct for all the queries in the MERGE regression
tests, and couldn't find any failures. I then extended the test to the
other optimizable commands, and couldn't find any problems there either.
My approach was perhaps a bit simple-minded: I just patched
pg_analyze_and_rewrite_fixedparams() to call pg_get_querydef() after
parse_analyze_fixedparams(), then ran the main regression tests. No
crashes. Also had it output as a WARNING together with the
query_string, so that I could eyeball for any discrepancies; I couldn't
find any queries that produce wrong contents, though this was just
manual examination of the resulting noise logs.
I suppose a better approach might be to run the query produced by
pg_get_querydef() again through parse analysis and see if it produces
the same; or better yet, discard the original parsed query and parse the
pg_get_querydef(). I didn't try to do this.
--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/
"No deja de ser humillante para una persona de ingenio saber
que no hay tonto que no le pueda enseñar algo." (Jean B. Say)