Here's v20, another rebase with some small changes:
- Changed psql tab-completion as suggested by Peter. However, I didn't
use Query_for_list_of_updatables because that includes relation types
that aren't supported by MERGE, so I kept a separate query definition
including only plain and partitioned tables. While at it, fix its
quoting: use of quote_ident() is no longer necessary.
- I changed the MergeWhenClause productions in the grammar to be more
self-contained. In the original, there was too much stuff being done
in its caller production. For example, the production for DELETE was
returning NULL and the caller was creating the struct .. not sure why.
Also, set all the fields in the struct, not just some. This is not
strictly necessary since the struct is zeroed by makeNode anyway, but
this is out standard practice and looks more tidy -- this is what led
me to discover the next point.
- I noticed that node MergeWhenClause had a member 'cols' to store
INSERT columns, but at the same time it was ignoring the targetList
member, which was documented to be used for this. I don't know the
reason for this separate member. I removed 'cols' and made the code
use 'targetList' instead.
- parse_clause.[ch] still contained some changes that were no longer
necessary due to my earlier hacking on parse analysis. Put them back
as they were.
--
Álvaro Herrera Breisgau, Deutschland — https://www.EnterpriseDB.com/