ScopedValues: add ScopedThunk

[kpamnany]

This enables taking a "snapshot" of scoped values and using that snapshot later.

[adienes]

bikeshed: ScopedThunk ? Functor has always seemed both loaded and ambiguous to me

[JeffBezanson]

We seem to be missing with(f, ::Scope). Then this could be implemented as

function inscope(f)
    s = current_scope()
    ()->with(f, s)
end

[Keno]

ScopedThunk seems better to me.

[kpamnany]

I should have mentioned, that this was already added to ScopedValues.jl and is being upstreamed after some iteration. This interface was defined after some debate with @vchuravy.

I would have no objection to ScopedThunk instead of ScopedFunctor, but I do think that this use matches a rather well understood definition of a functor, and it has already been released in the package with this name. We could still change it if people insist...?

Re: @JeffBezanson's suggestion, using Scopes in user code was strongly discouraged by @vchuravy, and a with form that takes a Scope opens the door to that.

So, waiting on input from Valentin.

[Keno]

definition of a functor

What definition?

[kpamnany]

definition of a functor

What definition?

The C++ one.

Wikipedia:

In some languages, particularly C++, function objects are often called functors...
... the function is really a façade for a full object, carrying its own state.

[Keno]

Yikes. We do not use that terminology in Julia. (Please feel free to use this as a citation to remove the incorrect information from the julia section in the wikipedia page ;) ).

[Keno]

Although apparently we do in the manual... Am I just living in the twilight zone? We should not be.

[kpamnany]

I don't see a problem with using a reasonably well-defined term pragmatically, even if category theorists complain.

[Keno]

It's just not what it does, in either definition - it's a type of closure that closes over the dynamic scope.

[Keno]

Although apparently we do in the manual

Ok, fixed that ;). I think for this pr, either ScopedThunk or ScopeClosure or ScopedClosure would be fine, but I'm vetoing Functor ;).

[kpamnany]

I think the test failure is unrelated?

[kpamnany]

@KristofferC can we backport this to 1.12?