Re: Octave 2.1.57 available for ftp

[Paul Kienzle]

On Mar 12, 2004, at 6:06 PM, John W. Eaton wrote:

> On 12-Mar-2004, Paul Thomas <address@hidden> wrote:
>
> | Has anybody done an inventory of the N-d features available on the 
> other
> | product?
>
> Here is my current list of things that need to be done for Octave that
> may involve fairly basic N-d functionality:
>
>   ...
>   orderfields

I've been puzzling over how to handle orderfields for a while.  This
comes up in mex support, wherein octave uses a map (which is
inherently unordered) and matlab uses a vector of field names,
which new fields added at the end.

The problem is that map cannot guarantee traversal order as you
are adding keys.  In the case of a mex file, it means that if the
programmer is lazy and always does mxSetFieldByNumber
without first doing mxGetFieldNumber, then this could fail
if the user adds new fields.

I don't see how to do fix mex and implement orderfields without
scrapping the use of map, or at a minimum, turning it into a map
from string to cell into a map from string to cell index, and using
a cell array of one dimension more than the dimension of the
structure array to represent the data.  Other than the funny
indexing (e.g., x.a may return a comma separated list), a struct
really is a cell array with string names for one of the axes.

BTW,  are two structs with differing field orders equal or not
equal?

Paul Kienzle
address@hidden