Re: struct arrays and cs-lists

[Ben Abbott]

On Nov 13, 2012, at 9:42 AM, Jordi Gutiérrez Hermoso wrote:

> On 13 November 2012 09:32, Ben Abbott <address@hidden> wrote:
>> 
>> On Nov 13, 2012, at 8:06 AM, Jordi Gutiérrez Hermoso wrote:
>> 
>>> On 13 November 2012 08:05, Jordi Gutiérrez Hermoso <address@hidden> wrote:
>>>> On 13 November 2012 00:02, John W. Eaton <address@hidden> wrote:
>>>>> On 12-Nov-2012, Jordi Gutiérrez Hermoso wrote:
>>>>> 
>>>>> | I don't really understand why if s is a struct array with field
>>>>> | foo, then "s.foo" is a cs-list. What's the purpose of this? How
>>>>> | can you get something like a cell array instead without
>>>>> | something as roundabout as resize(size(s), {s.foo})? Does it
>>>>> | have to be this way, is there something else we can do instead?
>>>>> 
>>>>> I don't see that we can do anything else if we want to preserve
>>>>> Matlab compatibility.
>>>> 
>>>> Sure, we can't change it, but how is it supposed to be used? Why
>>>> is it like this at all? There must *some* method to the Madlab.
>>> 
>>> By way of contrast, for cell arrays you can avoid cs-lists and get
>>> subcells if you index with () instead of {}. Why is there nothing
>>> equivalent for struct arrays?
>> 
>> For cells indexing with () preserves the cell class just as it does
>> for any data type (including structures). That's equivalent correct?
>> ... So, its not clear to me what you want to do.
> 
> Consider the following:
> 
>    s = struct("foo", {1, "lol"}, "bar", {2, {2}})
> 
> then s.foo is a cs-list with two different types, as is s.bar.
> 
>> Are you looking to create a cell-array from a structure while
>> maintaining the original cell-array's size, or something else.
> 
> Precisely this, I want the equivalent of
> 
>    reshape({s.foo}, size(s))
> 
> which isn't such a mouthful. I would expect something like s("foo") to
> do this, indexing with () and strings, or s.("foo"), but the latter
> syntax is already a synonym for s.foo.
> 
> The point is that I think struct arrays are best interpreted as cell
> arrays where one dimension is indexed with strings instead of
> positive integers. Indeed, the struct2cell and cell2struct functions
> perform the conversion this way, by changing the way that one
> dimension is indexed, whether with strings or positive integers. The
> problem is that there seems to be no way to index struct arrays along
> their string dimension in the same way that you can index cell arrays.
> 
> So, if the language doesn't have what I want, then what's the point of
> s.foo returning cs-lists?

I'm pretty sure the cs-list behavior for structures predates the introduction 
of cell-arrays in Matlab 5.   Meaning I think the language evolved as a 
reaction to a need, rather than a plan.

> How is this supposed to be used? It seems to
> me like the most natural use of struct arrays is to do something like
> a CSV file where the column headers turn into struct array field
> names, but if indexing by those field names gives you cs-lists, then a
> weird indirection appears in the middle.
> 
> So how are struct arrays meant to be used?

Would supporting an empty index to produce an equivalent cell array be useful 
...

        cellarray = s.()

Where the leading dimensions of the cellarray would be equal size(foo) and an 
extra dimension would be added for the fields?

Ben