On Tue, 22 Sep 2009 23:25:26 -0400, Daniel Keep
<daniel.keep.li...@gmail.com> wrote:
Steven Schveighoffer wrote:
On Tue, 22 Sep 2009 21:24:10 -0400, Daniel Keep
<daniel.keep.li...@gmail.com> wrote:
Well, if you're not looking at memoization, why is this discussion even
happening?
There are other pure optimizations that can occur without memoization.
In fact I don't even know why memoization is brought up so much with
pure functions, it is an impossible-to-prove optimization. How do you
know memoization helps when you don't know how often the same arguments
will be used?
What other optimisations? The only other one I can think of off the top
of my head is common subexpression elimination. Except that's
effectively caller-side memoisation.
Yes, those optimizations :) They are easy to do and very different than
parameter memoization because you only optimize where you know it makes a
difference (and you don't need to use the heap to do it).
The ONLY way I can see object casting being treated as a pure function
is if all references passed in are immutable and even then only if the
compiler can prove the reference cannot be changed (i.e. the reference
is always kept on the stack AND you assume you never manually delete
it).
How is a thread going to come along and change a local variable in
another thread?
I don't know if you've noticed, but objects tend to be allocated on the
heap, not the stack. :P
The pointer *value* (the actual reference, which is stored on the stack)
is not being changed. That's all dynamic cast cares about. The
referenced data is irrelevant (except for the classinfo, whose access can
be factored out of the function, and which isn't possible to change
without resorting to undefined behavior).
There's nothing to stop another thread from killing the object you're
pointing to and then substituting in a new one. It's pathological, yes,
but this is the sort of thing that, if it happens, you don't have a hope
of debugging. This is why threads completely and utterly suck and I
frequently wish they'd never been invented.
It's not only pathological, but it's undefined. We don't care about
undefined behavior. A thread can come through and trounce on any memory
at any time, this is also not a case we need to concern ourselves with.
-Steve
