Buddha Buck wrote:
> > I think the better way is to take Buddha's idea (soon to be RFC'ed, by
> > the sounds of it), and make current arrays a little more flexible. It
> > sounds like we'll just have to add an extra dimension somehow, and then
> > "arrays of compact arrays" will simply be "arrays with some compact
> > elements". This sounds more flexible and less bloated to me.
>
> Thanks for the vote of confidence... The RFC was sent to the RFC
> editor this afternoon (about 7 hours ago as I type), but it doesn't
> look processed yet.
>
I'm writing one too! I'll take a look at your RFC when it appears and see if
we can create a consolidated one.
Basically, I want to be able to create a list ref of list refs, specify a
data type (eg 'int') and attributes of ':compact' or ':sparse', and have the
resulting structure stored efficiently. The n-dim indexing (the notation for
which we still need to nut out before we have a strong RFC, I think) then is
just syntactic sugar for list ref tree indexing (and can be used for any
tree of lists, not just those with a basic data type and the :compact
attribute).
After we sort out notation for creating and indexing compact and sparse
arrays, we need to look back at RFCs <76 (v1): Builtin: reduce>, <81 (v2):
Lazily evaluated list generation functions>, and <82 (v2): Apply operators
component-wise in a list context> to review what changes are required to
make them play nice with n-dim tensors.
