[Pkg-octave-devel] Bug#706376: Bug#706376: Bug#706376: octave: sparse matrix n*2^16

[Michael D. Godfrey]

On 04/30/2013 01:19 PM, John W. Eaton wrote:
> On 04/30/2013 12:56 PM, Ed Meyer wrote:
>
>> Not only is it desirable to have sparse and full matrices behave 
>> similarly,
>> I believe the user should not need to be aware of which storage format
>> is used so functions like eig() would work for either.
>> The key is to use the C++ class system to have different implementations
>> for each storage format.
>
> I haven't been following this thread closely and I haven't thought 
> much about the details but I have no objection to trying to do a 
> better job with handling numel and dimensions/indices generally.
>
> Is there some way we can get the better behavior in a minimally 
> invasive way?
>
> Even if it requires significant changes, maybe we should consider what 
> the options are anyway.
>
> What changes are needed to make octave_idx_type behave the way you way 
> you want?
>
> jwe
>
It would be good to give some thought to the trade-off of "fixing" the 
current system
against completing the 64bit compiling system.  The 64bit system will be 
useful in
other ways and should be done in any case.  Of course, any improvement 
in the
current system is good, but the full implementation of large matrices 
could be a
64bit system feature.

Jordi seemed to think that 2^63 indicies would work with "no trouble" in 
the 64bit
system...

Michael