[jean-philippe-martin]

• NIO read channel checks on reopen that the generation didn't change
• Fake (in-memory) storage has basic support for generations

This PR makes the NIO code more robust (it'll detect modifications to files while they're being read), and makes the fake storage (FakeStorageRPC) useful for testing a larger number of scenarios.

[coveralls]

Coverage increased (+0.009%) to 81.201% when pulling ea1752c on jean-philippe-martin:jp_generations into b695138 on GoogleCloudPlatform:master.

[coveralls]

Coverage increased (+0.007%) to 81.199% when pulling ea1752c on jean-philippe-martin:jp_generations into b695138 on GoogleCloudPlatform:master.

[garrettjonesgoogle]

@jean-philippe-martin , do you have someone on your own team that is familiar with the NIO implementation that can review your changes on an ongoing basis? I think that would help with your agility, and reviewers in your space might be able to dive more deeply than us. Our team can still sanity check before merging.

[jean-philippe-martin]

@garrettjonesgoogle unfortunately no one in my team is familiar with NIO. I'll see if I can find someone else to help.

[jart]

This is a very important strengthening measure and I can't believe we never got to this sooner. LGTM

[garrettjonesgoogle]

@jean-philippe-martin it looks like conflicts need to be resolved before this can be merged.

[jean-philippe-martin]

This version should resolve the conflict; let's check that the tests still pass.

[garrettjonesgoogle]

LGTM - I'll merge after tests run

[jean-philippe-martin]

Excellent, thank you very much @garrettjonesgoogle ! And happy new year!

[jean-philippe-martin]

The tests failed. I see failures in com.google.cloud.logging.v2.ConfigClientTest, com.google.cloud.datastore.DatastoreTest, and com.google.cloud.pubsub.v1.SubscriberTest. They seem unrelated to the change. Is it possible these tests might report false positives?

[garrettjonesgoogle]

The Travis tests passed. That's good enough for me.

[jean-philippe-martin]

Great, thank you!