feat: allow access to Self and its public methods from actor initialisers

[ggreif]

This PR accomplishes following improvements:

• allow actor Self { ... Self ... } in the actor initialiser (similarly for actor class)
• ditto for accessing all public methods Self.method from the actor initialiser
• accessing these from under lambdas without making the respective methods undefined when using them

The latter should allow setting up timers: addTimer(b, period, Self.method) from actor initialiser.

---

NOTES:

• calling Self.method from actor initialiser is prohibited by the capability checker
• accessing Self.method from actor initialiser is currently allowed when method is not defined yet, and results in a descriptor (pair)
• comparison (in terms of shared) of the internal methods with external methods is trapping in the interpreter, but should work in the deployed actor.

---

TODO:

• adapt renaming.ml? — nope, I don't have business in IR
• test actor class too
• test clashing Self — there is already self-shadow.mo
• test/run/issue1464ok.mo fails — fixed snafu
• deal with FIXMEs
• shared method equality in the interpreters
• shared call viability checking (in interpreters)
• fix bug: actor self reference oddity #4731
• write issue about the mixed (pair/closure) shared func comparison problem — here: interpreters have trouble comparing shared methods when mixed intern/extern #4732
• resolve the Promise.is_fulfilled problem
• check the "For actors, this may be too permissive" comment in the code

[github-actions]

Comparing from 15a4ac2 to fd53c84:
In terms of gas, no changes are observed in 5 tests.
In terms of size, no changes are observed in 5 tests.

[ggreif]

Bizarre: uncommenting this gives a forward declaration inference error... but we already use debugPrint above!

[ggreif]

@crusso any idea how to explain this?

[crusso]

Is this resolved?

[crusso]

I just played around with it. I expect it's because the call to debugPrint is the last declaration that is causing the issue...

This also produces the error:

import { debugPrint; principalOfActor } = "mo:⛔";

actor class C() = Self {
    debugPrint "So far so good!" 
}

[crusso]

Filed an issue
#4733

[ggreif]

I suspected that it is unrelated. Thanks for checking!

[crusso]

Instead of adding the ~extern optional parameter, why not just add (class_self ... ) to beginning or end of the decs argument, depending on whether this is an actor (at front) or at other object (at end).

[crusso]

(and then rename class_self to obj_self...)

[ggreif]

Well, that would break

$ cat test/run-drun/self-shadow.mo
actor foo {
  public func foo() {};

  flexible func go() : async () {
    let bar = actor bar { public func bar() {} }
  };
}

right? See also the discussion in #1104.

[crusso]

You might not need this if you just extend the decs appropriately,

[crusso]

Is this resolved?

[crusso]

Crazy idea: At the moment, the actor environment maps actor ids to Objects containing a map from field names to values.
Suppose we changed that, so an actor environment just contains a map from field names to promises (i.e. ve_ex).
Then we can update the actor map before we interpret the body. As the promises in ve_ex get defined, so do the ones accessible from the self-reference, a bit like your previous 'increments' but automatic.
I think that might let us implement the equality on mixtures of symbolic and function references (provided we also give the equality operator) access to the environment.

Best attempted in a separate PR I think. Might be a dead-end with too many pervasive changes.

[crusso]

If equality had access to the actor env then we could probably do better here and maybe even get it right (see above)