Use agent instead of atom to implement mock connection

[ChrisBlom]

Uses an agent to manage the state of the mock connection,
see #2

[vvvvalvalval]

Can't merge this right now because it may leave some concurrency issues unaddressed. In particular, the proposed change does not let you read your own writes with (datomic.api/db conn), and I'm not sure that's okay (stay tuned to this StackOverflow question).

Even if it is okay, the need for a proper implementation of sync will probably become more necessary.

I'll keep you posted as I make progress.

[ChrisBlom]

According to the docs transact returns a completed future, so that would suggest that
after (d/transact conn data) finished, the data is transacted, and your writes should be in (d/db conn) , but it would be good if someone from Datomic can confirm this.

I see now that my implementation is not correct, i've implemened transact-async not transact,
i'll fix this. I this the concurrency issue you mean?

[ChrisBlom]

I noticed another problem: the future is delivered before the fn that updates the agent returns, so
potentially the agent state is not updated yet once the future is delivered, do you now a clean way to solve this?

[vvvvalvalval]

I noticed another problem: the future is delivered before the fn that updates the agent returns, so
potentially the agent state is not updated yet once the future is delivered, do you now a clean way to solve this?

@ChrisBlom it is not another problem, it is the same problem :) I had not noticed the issue with transact's implementation.

[vvvvalvalval]

@ChrisBlom thanks, I'd appreciate if you could add a little test for the RYW semantics

[vvvvalvalval]

And we still need a proper implementation of (d/sync conn t), if you want to give it a shot the best strategy I can think of is adding a watch for to the agent that looks up the basis-t (then self-removes)

[ChrisBlom]

should use identical? here

[vvvvalvalval]

@ChrisBlom yes, it will be equivalent and faster since the deliver-tx-res field will not be collection-typed.

[ChrisBlom]

I've implemented (d/sync conn t) and added a test for d/transact-async

[ChrisBlom]

the implementation is not correct, (d/sync conn t) does not work when the db already has t, please ignore for now