Following [0], I did a broader analysis of some dubious or nonsensical
like/unlike combinations in the pg_dump tests.
This includes
1) Remove useless entries from "unlike" lists. Runs that are not
listed in "like" don't need to be excluded in "unlike".
2) Ensure there is always a "like" list, even if it is empty. This
makes the test more self-documenting.
3) Use predefined lists such as %full_runs where appropriate, instead
of listing all runs separately.
I also added code that checks 1 and 2 automatically and issues a message
for violations. (This is currently done with "diag". We could also
make it an error.)
The results are in the attached patch.
[0]:
https://www.postgresql.org/message-id/3ddf79f2-8b7b-a093-11d2-5c739bc64f86@eisentraut.org