------_extPart_001_01C646CE.4DE9DE54
Content-Type: text/plain;
	charsetso-8859-1"
Content-Transfer-Encoding: quoted-printable

>Instead, partition both your vectorspaces as direct sums of smaller
>vectorspaces: then the big matrix will be made of many small matrices, and
>each small matrix may be elided if equal to zero; pick the partition such
>that you get a good space saving (DON'T try to find the optimal
>space-saving). My first guess would be to try direct sums of sqrt(N)
>components of dimension sqrt(N) each.

I never tried this approach, but I see no reason why I can assume that my call-graph would have a block-like structure that would be ideal for the compression technique you suggest. I fear that either too much space will be wasted on storing zeros when the dimension of the sub-matrices are choosen to be sqrt(n) or log(n) or whatever, and if it is choosen to be smaller to have a better fit, this becomes virtually the same as storing each element separately. But I could be wrong since I never tried this.
Have you experience with this?

>> Anyway, can you explain why you consider a reference object as a
>> "mis-feature"?
>
>All sorts of opinions can be invented and propagated if there are enough
>people willing to believe in them and a social network that is appropriate
>for sustaining a feedback loop of cognitive consonance.

Hypes do not interess me, an improvement of Ruby's capabilities does.

Greetings,
Geert.

------_extPart_001_01C646CE.4DE9DE54--