Ion's table timeout code

[pcmoritz]

No description provided.

[stephanie-wang]

Maybe since there are a lot of arguments here and it is hard for a user to parse, we could separate out the ones that only have to do with failures into their own struct argument (this would probably be retry_count, timeout, fail_cb). We could also have a version of object_table_lookup that specifies default values for these arguments.

[stephanie-wang]

Does it make sense to call retry_cb once before exiting this function? It seems like every time we would want to set a table callback, we would also want to call it once initially.

[stephanie-wang]

Is there a way we can make sure that fail_cb won't get called if done_cb has already been called? Also, a way to make sure that if we call done_cb, the callback in outstanding_callbacks will get cleaned up by destroy_table_callback.

We should probably also test for it. :) I think you test whether fail_cb will get called for the object table right now, but not the others. It would be good to also test whether outstanding_callbacks gets updated the way we think it is.

[stephanie-wang]

We should remove the timer here to prevent the failure callback from being called (see comment in state/table.c)

[stephanie-wang]

We should remove the timer here to prevent the failure callback from being called (see comment in state/table.c)

[stephanie-wang]

What happens if we malloc something later on that gets allocated the same pointer?

[pcmoritz]

as you pointed out, one solution is to use the timer_id to distinguish (it is unique)

[stephanie-wang]

I don't think the callback in outstanding_callbacks will get cleaned up.

[stephanie-wang]

Can you add some documentation about what each test is testing for?

[stephanie-wang]

Unfortunately, I think this is really only a test for the object table timeout, not generic timeouts. This goes back to the other comment in state/table.c about making sure we clean up everything properly if done_cb gets called. Currently, the tables seem to do different things once they've called done_cb.

[stephanie-wang]

We should check in all cases that outstanding_cbs is cleaned up properly.