On 2020-09-22 20:29, Pavel Stehule wrote: > The result is correct. When I tried to use UNION instead UNION ALL, the > pg crash
I fixed the crash, but UNION [DISTINCT] won't actually work here because row/record types are not hashable. I'm leaving the partial support in, but I'm documenting it as currently not supported.
I think so UNION is a common solution against the cycles. So missing support for this specific case is not a nice thing. How much work is needed for hashing rows. It should not be too much code.
> looks so clause USING in cycle detection is unsupported for DB2 and > Oracle - the examples from these databases doesn't work on PG without > modifications
Yeah, the path clause is actually not necessary from a user's perspective, but it's required for internal bookkeeping. We could perhaps come up with a mechanism to make it invisible coming out of the CTE (maybe give the CTE a target list internally), but that seems like a separate project.
The attached patch fixes the issues you have reported (also the view issue from the other email). I have also moved the whole rewrite support to a new file to not blow up rewriteHandler.c so much.
-- Peter Eisentraut http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
This patch is based on transformation CYCLE and SEARCH clauses to specific expressions - it is in agreement with ANSI SQL
There is not a problem with compilation
Nobody had objections in discussion
There are enough regress tests and documentation
check-world passed
doc build passed
I'll mark this patch as ready for committer
Possible enhancing for this feature (can be done in next steps)
1. support UNION DISTINCT
2. better compatibility with Oracle and DB2 (USING clause can be optional)