Looks like there's a panic caused by GoogleCloudPlatform/terraform-google-conversion#709

The tests are passing for both me & @c2thorn locally. This is because the results of convert are not fully deterministic.

We use map[string]Asset to store assets internally; Golang maps are not ordered. So when we pull out the list of assets, we order them by Asset Name - for example, //storage.googleapis.com/placeholder-c2WD8F2q (the last bit is generated by terraform-google-conversion).

In test runs, we then "normalize" assets for comparison, which replaces placeholder-.* with placeholder-foobar so that we can have passing tests even with randomly-generated asset names. This obscures the fact that the assets were first sorted by their random name (which means that the results are not deterministic even though they're sorted by name.)

It's odd that it seems to pass consistently for me locally though... :-/

Okay, so if I am just running the one test, it always gets the same values for the random placeholder strings. This is almost certainly because we use rand.Intn to generate the placeholder values, which relies on Golang's pseudo-random builtins, which always start with the same randomness seed.

So, the full test suite is probably failing because it's preceded by a bunch of randIntn calls, resulting in output that happens to be different than what we get just running that test. We're lucky we caught this now!

On the plus side, this means that if we can find a thread-safe way to reset the seed before each test, we could make all tests deterministic with regard to their generated strings. If that doesn't work we'll have to find some other solution.

@melinath Moved the test failure to its own issue, skipped the test and rebased.