Stefano Corsi
2003-05-02 15:13:09 UTC
Dave,
with some efforts I came down to only :) a 50% loss in interpreted mode with
overloaded operators abilited! However the loss is only significative in
nested loops and similar structures that require a lot of iterations.
In "normal" page programming the user experiences no loss of performance.
On the other side, in compiled mode, the loss is 0% and programs execute at
the same speed even during very complex loops, with or without overloaded
operatos abilited.
I'm confident that some work can be done to make the interpreted mode faster
even in complex loops with operator overloading abilited, but I think that a
certain delay (needed to check the function hashtable for a suitable operator
function) cannot be avoided.
So what to do? I think that we should keep the feature, work to make it more
performant and accept a certain delay, only in interpreted mode and only in
special situations.
What do you think?
Stefano
with some efforts I came down to only :) a 50% loss in interpreted mode with
overloaded operators abilited! However the loss is only significative in
nested loops and similar structures that require a lot of iterations.
In "normal" page programming the user experiences no loss of performance.
On the other side, in compiled mode, the loss is 0% and programs execute at
the same speed even during very complex loops, with or without overloaded
operatos abilited.
I'm confident that some work can be done to make the interpreted mode faster
even in complex loops with operator overloading abilited, but I think that a
certain delay (needed to check the function hashtable for a suitable operator
function) cannot be avoided.
So what to do? I think that we should keep the feature, work to make it more
performant and accept a certain delay, only in interpreted mode and only in
special situations.
What do you think?
Stefano