allow x[i,j;k] syntax to be parsed

[mlubin]

In JuMP, it would be convenient if we could use the syntax x[i,j;k] within macros. We already take advantage of x{i,j;k}. Any reason not to allow this syntax and make it a compile-time error when used in code instead of a parser error? @IainNZ @joehuchette

[StefanKarpinski]

What should it mean?

[joehuchette]

In JuMP we would use it to add conditionals when creating variables, e.g.

@defVar(m, x[i=1:n, j=1:n; i < j])

So for our purposes we only want that it's valid syntax so we can parse it in a macro.

[mlubin]

It shouldn't mean anything in Julia, but it should be parsed as

Expr(:ref, :x, Expr(:parameters,:k),:i,:j)

This is analogous to x{i,j;k} which is parsed as

Expr(:curly, :x, Expr(:parameters,:k),:i,:j)

and x(i,j;k) which is parsed as

Expr(:call, :x, Expr(:parameters,:k),:i,:j)

[stevengj]

It makes sense to me that getindex should be able to take keyword arguments, which is essentially what this is. i.e. a[args...] should be parsed exactly as getindex(a, args...), including ; and keywords.

[StefanKarpinski]

True, but we had discussed using the v[i;j] syntax at some point.

[mlubin]

If it is used, it will need to be parseable anyway. The meaning of the syntax in Julia doesn't need to be decided here.

[stevengj]

@StefanKarpinski, when was that? What was the proposed meaning? Was that before we had keyword args?

[StefanKarpinski]

It was for "contravariant indexing" – the idea was that v[i;j] would drop leading singleton dimensions instead of trailing singleton dimensions. The idea never went anywhere though.

[mlubin]

@StefanKarpinski, any objections to this?

[StefanKarpinski]

No, I'm ok with it, but it would be good to check with @JeffBezanson – surface syntax is his baby.

[mlubin]

Bump now that 0.4 is open.

[jakebolewski]

closed by #8250