cmd/compile: stronger constant propagation around closures

[adonovan]

I notice that the use of a local variable to hold a function (e.g. for brevity) doesn't prevent the compiler from treating the variable as an alias for the function, and dispatching calls statically. However, if the variable is referenced from an inner closure, it is treated as a true variable, resulting in a dynamic call:

func f() any {
	escape := html.EscapeString
	println(escape("foo")) // static call
	return func() {
		println(escape("foo")) // dynamic call
	}
}

The closure may be latent, in the case of a range-over-func:

func f(seq iter.Seq[string]) {
	escape := html.EscapeString
	println(escape("foo")) // static call
	for s := range seq {
		println(escape(s)) // dynamic call
	}
}

This is another facet of #73524 (comment), where closures defeat constant propagation of outer single-assignment variables.

[cherrymui]

cc @dr2chase @golang/compiler

[mknyszek]

CC @JunyangShao

[randall77]

So I think this issue is not actually correct. We do treat the inner use of escape as a constant, and inline the body of html.EscapeString there.

But we do also still put escape into the closure. We just then never use it :) Probably it should be removed.