for loop surprises

[vtjnash]

In porting code from MATLAB, this one caught me by surprise. Would it be possible (worthwhile) to generate a warning if an inner anonymous function shadows a local array (variable)? The scope of the length(x) call also surprised me, although I can't think of an example where this would matter in real code.

In MATLAB:

x = zeros(10,1)
for i = 1:length(x)
    x(i) = 1;
end

In Julia:

julia> x=zeros(10,1)

julia> for i = 1:length(x)
                  x(i) = 1
              end
in anonymous: x not defined
 in anonymous at no file

julia> for i = 1:10
                  x(i) = 1
              end

julia> x==zeros(10,1)
true

julia> for i = 1:10
                  x[i] = 1
              end

julia> x==ones(10,1)
true

[JeffBezanson]

That's unexpected. I don't believe I intended function definitions to behave differently in this way.

[toivoh]

@JeffBezanson: I take it that the unexpected part is that the function definition creates a new binding x instead of refusing to add a method to the old binding?

I've gradually been feeling more and more that method definitions should probably not be done with =, since it's too easy to confuse them with assignment. (In the case above, and others.) I think that

f(x) := x^2

would make a nice alternate syntax.

Of course, that would be a breaking change of epic proportions. The old syntax could be kept alongside for a pretty long time to ease the transition. I do think that it would allow julia to be more consistent, and possibly open up for more powerful syntax by making the distinction between assignment and method definition cleaner. As a toy example, consider

(x::MyType)[i::Int] := ...

to define a ref method for MyType. (That doesn't generalize to assign, though... )