Randomness in built-in .sort

[Bill Baxter]

On Tue, Jan 6, 2009 at 6:15 AM, dsimcha <dsimcha at yahoo.com> wrote:
> == Quote from Bill Baxter (wbaxter at gmail.com)'s article
>> Actually, a function to sort multiple arrays in parallel was exactly
>> what I was implementing using .sort.  So that doesn't sound like a
>> limitation to me at all.   :-)
>> --bb
>
> Am I (and possibly you) the only one(s) who think that sorting multiple arrays in
> parallel should be standard library functionality?

I use this all the time in NumPy in the form of "argsort" which
returns a list of indices giving the sort order.  That can then be
used as to index other arrays (thereby permuting them to the sorted
order).
order = npy.argsort(keys)
sortedA = A[order]
sortedB = B[order]

> The standard rebuttal might be
> "use arrays of structs instead of parallel arrays".
> This is a good idea in some
> situations, but for others, parallel arrays are just plain better.

Yeh, that's just a silly argument.  Sometimes a column-oriented
organization is more suitable than a row-oriented one.
Sort can be made to work on column-oriented data, so there's no reason
not to make it so.

> Furthermore, with D's handling of variadic functions, generalizing any sort to handle parallel
> arrays is easy.

Yeh, this works pretty nicely.

--bb