On Tue, Mar 29, 2016 at 2:45 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
Joshua Ma <josh@benchling.com> writes: > This might not be a common case, but we're using pg_dump in a testing > environment to check migrations - 1) we initialize the db from HEAD, > pg_dump it, 2) we initialize the db from migration_base.sql, apply > migrations, pg_dump it, and 3) compare the two dumps to verify that our > migrations are correct wrt schema.
> However, we're seeing pg_restore transforming our check constraints with > different casting.
It's not really different. What you're seeing is pg_dump (or actually ruleutils.c) choosing to dump some implicit casts explicitly to ensure that the expression is parsed the same way next time. It might be overly conservative to do so, but we've found that erring in this direction tends to avoid breakage when the result is loaded into another server version; it's a bit like the intentional overparenthesization.
Why don't we just use ruleutils.c to generate \d results so that what we end up showing is canonical?