Inconsistencies in @closure semantics

[c42f]

The semantics of variable capture in @closure were originally intended to be the same as julia, but I got this wrong because at the time I didn't understand the intended semantics of the core language which is that the value associated with a name may be changed either within the closure or within the surrounding function and in both of these cases the change must be dynamically reflected in the other scope.

In particular, the value of x in the following is modified outside the closure, followed by returning closure, and there's nothing @closure can know about this:

using FastClosures

function foo()
    x = 10
    c = ()->x
    x = 20
    c()
end

function bar()
    x = 10
    c = @closure ()->x
    x = 20
    c()
end

julia> foo()
20

julia> bar() # @closure breaks this!
10

On the other hand, @closure does quite a bit of work to allow the following pattern to work the same as in the core language:

function baz()
    x = 10
    c = ()->(x = 20)
    c()
    x
end

function qux()
    x = 10
    c = @closure ()->(x = 20)
    c()
    x
end

julia> baz()

20

julia> qux()
20

(Though x+=10 doesn't appear to work... I guess that's a bug. Ugh!)

In hindsight I think this makes the semantics @closure pretty confusing and it would be better to just freeze all bindings inside the closure to fix them to the values they had at closure creation time. See also some discussion in #5 (CC @bramtayl)

Some possible names for such a macro...

@capturevalues # semantically questionable ?
@freeze_captures
@freeze
@rebind
@constbind # questionable
@newbindings
@let # too technical ?

[c42f]

XRefs:

• Scoping bug with anonymous functions? JuliaLang/julia#16727 (comment)
• Explicit capture of variables in a closure. JuliaLang/julia#14959 (comment)
• https://discourse.julialang.org/t/surprising-capture-boxing-behavior-in-closure/20254/3

[c42f]

more possible names...

@localvars
@forcelocal

[bramtayl]

+1 for @freeze_captures

[bramtayl]

Actually just @freeze seems like it would be fine.

[c42f]

@freeze is ok, though it's a bit too much of an analogy for me. I'd prefer something more literal, perhaps @rebind.

[c42f]

Looking back, I can't say I'm entirely happy with any of the names above. @bind might be better?