Tests fail unless forced to run on single thread

[lolgesten]

Hi! In my code I think I have an issue with renaming of nullable enum values, so I figured I try running your tests and maybe extending them. However I didn't get that far, because I get test fails with the tests that are already there.

Is this something to do with my setup? I tried both postgres 9 and 10.

martin@rexlk:~/dev/diesel-derive-enum/tests$ cargo test --features postgres nullable
    Finished dev [unoptimized + debuginfo] target(s) in 0.0 secs
     Running target/debug/deps/tests-4a7899a1fb646353

running 2 tests
test nullable::nullable_enum_round_trip ... FAILED
test nullable::not_nullable_enum_round_trip ... ok

failures:

---- nullable::nullable_enum_round_trip stdout ----
    thread 'nullable::nullable_enum_round_trip' panicked at 'called `Result::unwrap()` on an `Err` value: DatabaseError(__Unknown, "column \"my_enum\" is of type my_enum but expression is of type pg_temp_2.my_enum")', src/libcore/result.rs:916:5
note: Run with `RUST_BACKTRACE=1` for a backtrace.

failures:
    nullable::nullable_enum_round_trip

test result: FAILED. 1 passed; 1 failed; 0 ignored; 0 measured; 3 filtered out

[lolgesten]

Oh wait. It's stranger than that. It's intermittent. Running the tests over and over, sometimes it succeeds.

[lolgesten]

Could it be that the pg_temp is not isolated per connection and we get concurrency problems between the two tests?

[adwhit]

Hi, yeah there is some kind of problem with isolating the tables in Postgres. I think the problem is that CREATE TYPE ENUM is not local to the (temporary) schema. Rather than bother to understand it properly I just serialized the tests like RUST_TEST_THREADS=1 cargo test --manifest-path=tests/Cargo.toml --features="postgres" (see travis.yml). Still inexplicably fails sometimes but much more rarely.

But I would welcome a proper way to do it!

[adwhit]

To go into more detail, when we obtain a PgConnection, we run conn.execute("SET search_path TO pg_temp;").unwrap();. Now pg_temp is actually an alias to a temporary schema which by my understanding should be connection-local (see https://dba.stackexchange.com/questions/76494/temporary-schema-per-connection). Then we should be able to create types and tables isolated from other connections.

But, for whatever reason, it just don't work the way I expect.

[lolgesten]

Exciting stuff. Thanks! My problem was (obviously) not your package, but my code :D

[adwhit]

Good to know. I'll keep this issue open to remind myself that the testing setup is an unfortunate hack.