Pipelining support

[michaelfig]

For proper pipelining, we will need to create a BasePromise class that has the necessary hooks to forward messages from one ERelay to another.

Look at https://github.com/ysmood/yaku/blob/master/src/yaku.core.js for how this can be done minimally, and all we'll require is the ES5-compatible BasePromise constructor function with its prototype.

[michaelfig]

Also, under what circumstances would something be passed as the unresolvedRelay in Agoric/eventual-send#4? IIUC, it's just a placeholder that needs to postpone the operations on the Presence until it is resolved. Would it be better to remove that argument (or make it optional), for this default behaviour?

If that's too magical, another possibility I thought of was to have another unresolvedRelay method that would be called when the presence is resolved, something like TRANSFER_PIPELINE(resolvedPresence) (for lack of a better name). That would give the unresolvedRelay the option of replaying messages on the new presence.

Am I misunderstanding or oversimplifying what the pipelining issue is?

[erights]

Misunderstanding. Messages sent to an unresolved remote promise must be sent towards the decision point --- the vat which will decide what that promise resolves to. Otherwise, we have way too many round trips. See Promise Pipelining chapter of my thesis.

[michaelfig]

Okay, thanks. That helps a lot. I will read your thesis.

The second part of my question is whether we can supply a default unresolvedRelay, if it is not specified? That default would be too round-trippy for remote use, but it might simplify the case where you just want to intercept local eventual-sends while protecting the target object from being exposed with presenceP.then(p => ...manipulate p directly...).

This would be somewhat like a Proxy, but without the full range of operations.

[erights]

Don't know. Worth exploring.

[michaelfig]

Don't know. Worth exploring.

FWIW, I've implemented this default in Agoric/eventual-send#7 and used it in one of the tests. If it turns out to be something that doesn't get accepted as part of the TC39 eventual-send proposal, I can easily remove it.

[erights]

@michaelfig I'm assigning both of us and closing. If I'm missing something, please let me know or reopen.