PG Bug reporting form <noreply@postgresql.org> writes:
> In the following test case, the `EXECUTE` statement will fail with an error
> `result of range union would not be contiguous`. The final SELECT query
> should return 1 row as there is only one value `1` in t1.c0 and only one
> value `1` in t3.c0, however, the query returns 0 rows.
I don't see a bug here particularly. If you check the contents of
table t3 at the end of the test, you will see
c0 | c1
----+------------------------
2 | [-761818403,793731612)
(1 row)
in the "wrong" case. This shows that the prepared insert advanced
the sequence underlying "c0 bigserial" before failing, whereas
in the default plan_cache_mode the error occurred before touching
the sequence. This discrepancy isn't a bug. It occurs because
in the default mode the planner will attempt to constant-fold
that messy range expression during planning, and thus it will
hit the range union failure before anything is done to the
sequence. In the generic case the failure occurs in the executor,
and the sequence has already been advanced.
regards, tom lane
