subsection on let blocks

[StefanKarpinski]

We don't explain let blocks since they're not control structures, per se. However, they're mentioned as introducing new variable scope blocks in [[Variables and Scoping]]. We need to explain them somewhere. @JeffBezanson, can you explain this, optimally with a few example to distinguish from other variables binding constructs — such as plain old assignment? I.e. what's the difference between these three blocks:

let x = 1
  # do something
end

begin
  local x = 1
  # do something
end

for x = [1]
  # do something
end

[JeffBezanson]

The difference is that other scope blocks just control visibility, but let allocates new bindings (where a binding is defined as the location where a variable is stored). This difference is only visible when you have closures, which allow variables to live longer than the blocks that introduced them:

Fs = cell(2)
for i=1:2
    Fs[i] = ()->i
end

Fs[1]()  =>  2
Fs[2]()  =>  2

Fs = cell(2)
for i=1:2
    let x=i
        Fs[i] = ()->x
    end
end

Fs[1]()  =>  1
Fs[2]()  =>  2

[StefanKarpinski]

Awesome, I can include this example in the docs. I wonder if this shouldn't be the behavior of for though, especially in the context of parallel computation. It's a little unorthodox, but what if each iteration of a for loop actually semantically had a different binding. Of course, in most cases it wouldn't make any difference. When is the former behavior ever more useful?

[JeffBezanson]

It's also the case that

let a=b
  code
end

is equivalent to (a->code)(b).

In the case of for loops, the former behavior doesn't seem too useful, but it's much more efficient for common uses, even when closures are involved:

for i=a:b
    map(x->x+i, ...)
    ...
end

Here the closure is only used in the loop, so not only is a new binding not needed, but allocating the closure itself can be lifted out of the loop because you can keep reusing it and just update the binding inside.

[JeffBezanson]

What section of the docs is this going in?

[StefanKarpinski]

Variables and Scoping seems like the only reasonable place, right?

On Aug 31, 2011, at 12:21 PM, JeffBezansonreply@reply.github.com wrote:

What section of the docs is this going in?

Reply to this email directly or view it on GitHub:
#188 (comment)

[JeffBezanson]

Done!