> Is this asymmetry a bug? I realize these event trigger functions are > typically written in C, but I'd like to keep this idea contained to a > plpgsql function and cover as many reversible changes as I can.
Sadly, you cannot obtain all the necessary details using only plpgsql. The only way is to compile a C extension. There's a very rough, incomplete skeleton to show how to do this in src/test/modules/test_ddl_deparse. A much more complete approach, which was not completed, was used to produce JSON output from the C structures -- see https://www.postgresql.org/message-id/20150409161419.GC4369@alvh.no-ip.org
Ah thank you, test_ddl_deparse gives me a good idea what's going on now. Very interesting thread, thanks for your work on this.
Keep in mind ALTER TABLE is a very sophisticated; you can do things like ALTER TABLE one ADD COLUMN two INT, DROP COLUMN three, ADD COLUMN four text; (i.e. do several things in one command) so if you were to use a simplistic approach, you might end up missing things.
This is a support feature for some users so they are encouraged to write down scripts. The idea being the tool would spit one out and they'll be advised to use it as a starting point for the actual post-hoc down script, so catching CREATE TABLE/INDEX is a good start. As this ddl deparsing feature evolves in the future, I can add more complex cases.
