On Wed, Oct 13, 2021 at 4:43 PM Mark Dilger <mark.dil...@enterprisedb.com> wrote: > The function where the processing occurs is processSQLNamePattern, which is > called by pg_dump, pg_dumpall, and psql. All three callers expect > processSQLNamePattern to append where-clauses to a buffer, not to execute any > sql of its own. I propose that processSQLNamePattern return an error code if > the pattern contains more than three parts, but otherwise insert the database > portion into the buffer as a "pg_catalog.current_database() > OPERATOR(pg_catalog.=) <database>", where <database> is a properly escaped > representation of the database portion. Maybe someday we can change that to > OPERATOR(pg_catalog.~), but for now we lack the sufficient logic for handling > multiple matching database names. (The situation is different for > pg_dumpall, as it's using the normal logic for matching a relation name, not > for matching a database, and we'd still be fine matching that against a > pattern.)
I agree with matching using OPERATOR(pg_catalog.=) but I think it should be an error, not a silently-return-nothing case. > In pg_dumpall, --exclude-database=more.than.one.part would give an error > about too many dotted parts rather than simply trying to exclude the last > "part" and silently ignoring the prefix, which I think is what v13's > pg_dumpall would do. --exclude-database=db?? would work to exclude four > character database names beginning in "db". Those things sound good. -- Robert Haas EDB: http://www.enterprisedb.com
