Can't length an nx1 matrix

[ghost]

To keep code vector-matrix compatible must do size(v)[1], which works for both. Desirable?

[StefanKarpinski]

I would argue that you shouldn't be calling length on non-vectors. Can you give an example where you need to do this?

[ghost]

My feeling is that as a convention you should not have to remember if
something is an nx1 matrix or a vector in any circumstance.

On Mon, Jul 25, 2011 at 9:44 PM, StefanKarpinski <
reply@reply.github.com>wrote:

I would argue that you shouldn't be calling length on non-vectors. Can you
give an example where you need to do this?

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

[StefanKarpinski]

I'm not sure I quite buy into that. I'd like to see more examples before we commit to this either way. Matlab defines it as

length(a::AbstractArray) = max(size(a))

I suppose we could go with that, but it seems wrong. Especially because we enumerate all of the items in an array when using it as an iterator, and the convention for iterable objects is that length for indexable, iterable objects give the maximum index so that

for x=a
  println(x)
end

will produce the same output as this:

for i=1:length(a)
  println(a[i])
end

That would also work for 1xn and mx1 "vectors". This basically means defining

length(a::AbstractTensor) = numel(a)

[JeffBezanson]

Yes, given linear indexing I suppose length == numel makes sense.

[StefanKarpinski]

Yeah, ok. I'm cool with that.