allow x[1] for scalar x

[stevengj]

to be consistent with the treatment of 0-dimensional arrays; note that ndims(x)==0 and length(x)==numel(x)==1 for scalar x.

See also the discussion in issue #1829.

[johnmyleswhite]

I now have a small concern about this change: it breaks a kind of symmetry between ref and assign. Typically, if one can ref something, one can assign to it as well. This is no longer the case for scalars after this change:

julia> x = 1
1

julia> x[1]
1

julia> x[1] = 2
no method assign(Int64,Int64,Int64)

To me, this is not evidence against the value of defining ref for scalars. It's just evidence for allowing assign to work as well, but I believe that change has to happen outside of base/number.jl because of the immutability of numbers.

[ViralBShah]

This is not the case, for example with tuples:

julia> a = (5,6,7)
(5,6,7)

julia> a[2]
6

julia> a[2] = 9
no method assign((Int64,Int64,Int64),Int64,Int64)

[johnmyleswhite]

Tuples seem like a special case. Are there are other examples like that?

[ViralBShah]

Wouldn't we have more such examples if we had immutable types? Can't think of others right away, but there may be.

[StefanKarpinski]

Strings too.

[johnmyleswhite]

You're both right: all immutable types have this property, including strings. Request withdrawn.