There are 3 independent patches associated to one thread and one CF entry.
*** About toast table v3:
Patch applies cleanly, compiles, works for me.
ISTM that the he query should be unambiguous: pg_catalog.pg_class instead
of pg_class, add an alias (eg c), use c.FIELD to access an attribute. In
its current form "pg_class" could resolve to another table depending on
the search path.
C style is broken. On "if () {", brace must be on next line. On "1 !=
PQntuples(result)", I would exchange operands.
PQclear must be called on the main path.
If the table name contains a ", the result looks awkward:
For table: "public.foo"bla"
I'm wondering whether some escaping should be done. Well, it is not done
for other simular entries, so probably this is bad but okay:-)
There are no tests:-(
*** About toast index v3
Patch applies cleanly, compiles, works for me.
There are no tests:-(
*** About the next one, v4
Patch applies cleanly, compiles. Not sure how to test it.
"switch (*PQgetvalue(result, i, 2))": I understand that relkind is a must
admit I do not like this style much, an intermediate variable would
improve readability. Also, a simple if instead of a swich might be more
appropriate, and be closer to the previous implementation.
There are no tests:-(
--
Fabien.
