Rather than sleeping here, can we instead call getFunctionResult for each message in turn? Although some sleeps still exist in the tests, we should try to avoid wherever possible.

I see the need for this method, but is it really just doing the same thing as checkMessageEquality, just not checking the ID? Could we instead add a bool flag to checkMessageEquality, which is something like checkIDs which is true by default, then pass false when we need to do this sort of check?

Then we could change the parameter on checkExecGraphEquality to also be checkIDs with true default.

The reason we check only these fields is that these are the ones MPI changes, and thus we manually set them in the test.

Any other field we also wanted to check, we'd have to, afaict, manually add ourselves to each message.

But if we don't touch any of the others, is there any harm in checking them? They will all be uninitialised and therefore pass the checks won't they? We'll just be comparing lots of blank strings/ zeros etc.

I.e. can we avoid having to add a new comparison function to check every permutation of edits we might make to a message?

Not really, for instance the executedHost or finishtimestamp fields are set by the scheduler when the chained functions finish (i.e. the world is destroyed).

Will the executed host not be the current host therefore we can check that? We deal with the finish timestamp issue in another tests here by just overriding it: https://github.com/faasm/faabric/blob/master/tests/test/scheduler/test_scheduler.cpp#L663

The reason I'm against having more than one message comparison function is that it's easy to forget to add fields to them, and if we just have one, then it's easier to track.

It's a sort of blacklisting/ whitelisting problem, where rather than saying "we want to check that this subset of fields is the same", we're saying "we only expect these fields to have changed, and all others to be the same". The latter is more strict (and more vebose), but catches problems more frequently as a result.