Re: numel(foo{:}) - feature or bug ?

[Jordi Gutiérrez Hermoso]

2011/8/5 Sergei Steshenko <address@hidden>:


> Sorry, Jordi, doesn't make sense to me.
>
> The same 2-d example in more detail:
>
> "
> octave:1> foo2d{1,1} = "a"
> foo2d =
> {
>  [1,1] = a
> }
> octave:2> foo2d{:}
> ans = a
> octave:3> sizeof(foo2d)
> ans =  1
> octave:4> numel(foo2d{:})
> ans =  1
> octave:5> foo2d{1,2} = "aa"
> foo2d =
> {
>  [1,1] = a
>  [1,2] = aa
> }
> octave:6> size(foo2d)
> ans =
>
>   1   2
>
> octave:7> numel(foo2d{:})
> ans =  2
> octave:8> foo2d{2,1} = "b"
> foo2d =
> {
>  [1,1] = a
>  [2,1] = b
>  [1,2] = aa
>  [2,2] = [](0x0)
> }
> octave:9> size(foo2d)
> ans =
>
>   2   2
>
> octave:10> foo2d{:}
> ans = a
> ans = b
> ans = aa
> ans = [](0x0)
> octave:11> numel(foo2d{:})

As Ben explained, because foo2d{:} is of type cs-list (NOT CELL), this
is equivalent to

     numel(foo2d{1}, foo2d{2}, foo2d{3}, foo2d{4})

i.e.

    numel("a", "b", "aa", [])

i.e. the number of ways that "a" can be indexed with "b", with "aa",
and with [].

But since size([]) is zero, you can't index with it. So the call above
is zero.

I tried to explain this better with the following changeset, but if it
still doesn't make sense, let me know if there's a better way:

     http://hg.savannah.gnu.org/hgweb/octave/rev/67a5acac8e1b

In order to accomplish what you want, you would have to recurse into
the nested cell structure with numel calls.

> I think 'octave' should have a 'enable_stupid_matlab_bugs' flag.

We have it for some things. It's called --braindead. You can see what
it does if you type "doc braindead".

- Jordi G. H.