feat(internal): new prepareAttenuator leveraging callbacks
#7586
Conversation
michaelfig
force-pushed
the
mfig-exo-attenuator
branch
from
0f8ac37
to
6f460fc
Compare
michaelfig
force-pushed
the
mfig-exo-attenuator
branch
from
eea8945
to
a9a0975
Compare
|
Hi @gibson042, I'm merging this now to unblock other work. However, if you still want to give a detailed review, please leave it here and I will address comments and implement changes in a new PR. |
michaelfig
added
the
automerge:no-update
michaelfig
force-pushed
the
mfig-exo-attenuator
branch
from
e81f268
to
884fa3b
Compare
| * @param {unknown} [opts.target] The target for any methods that | ||
| * weren't specified in `opts.overrides`. |
There was a problem hiding this comment.
What is the use case for making target optional (e.g., makeAttenuator({ target: undefined, overrides }))?
There was a problem hiding this comment.
If you want to compose an object entirely out of callbacks, you don't need a default target.
There was a problem hiding this comment.
Should we encourage use of makeAttenuator for composing an object from callbacks? I would expect not, as opposed to e.g.
const makeInstance = zone.exoClass(tag, interfaceGuard, initEmpty, { ...callbacks });There was a problem hiding this comment.
(Aside from legacy...)
Now that we have attenuators, when would we ever use callbacks? AFAICT, attenuators are just better. If so, perhaps we should try to phase out callbacks over time.
There was a problem hiding this comment.
I don't see how attenuators supplant callbacks. Thanks to callbacks you can compose a new object of a given shape from different existing objects. They are complimentary.
| // @ts-expect-error deliberately attenuated out of existence | ||
| t.throws(() => atE.m3(), { message: /not a function/ }); | ||
| await t.throwsAsync(() => atE.m4(), { message: /target has no method "m4"/ }); |
There was a problem hiding this comment.
Should these errors be more similar? Can they be?
There was a problem hiding this comment.
The first is an engine-specific TypeError for which not a function is the terse message you get when atE.m3 is not defined (in Node.js for this test).
The second is a rejection by the HandledPromise shim that contains much more specific information, such as the methods that do exist, and the type of target. We could change this one, but its format is pretty deeply entrenched in Agoric SDK unit tests.
closes: Needs issue
Description
Adds a mechanism to define an exo class presenting the same external API, but backed per-instance by passable callback descriptors.
Security Considerations
Improves encapsulation; instead of marshalling raw callback descriptors (which expose their unattenuated targets), an exo ocap can be conveyed.
Scaling Considerations
Documentation Considerations
Testing Considerations