Re: faster access private or public?
From: Igor Tandetnik (itandetnik_at_mvps.org)
Date: 09/28/04
- Next message: Arnaud Debaene: "Re: How to get processor serial number?"
- Previous message: Saravana Vijayan.B: "integer pointer values..convert to char"
- In reply to: Sigurd Stenersen: "Re: faster access private or public?"
- Next in thread: Tom Widmer: "Re: faster access private or public?"
- Messages sorted by: [ date ] [ thread ]
Date: Tue, 28 Sep 2004 10:46:19 -0400
"Sigurd Stenersen" <sigurds@utvikling.com> wrote in message
news:OcBuZ6UpEHA.744@TK2MSFTNGP10.phx.gbl
> I guess I have to spell it out...
>
> void g(int i)
> {
> OhFooey->~Foo();
> new (OhFooey) Foo(10);
> }
>
> In this case it *is not* okay for the compiler to assume that private
> variables are not changed when optimizing some other piece of code
> that uses the instance pointed to by OhFooey and that code calls g().
> I believe we all agree on that ?
Right. But this code has well-defined behavior, so I don't quite see how
it is relevant to the question at hand.
> Now, what you're saying seems to be that in this case...
> void g(int i)
> {
> new (OhFooey) Foo(10);
> }
> ...it *is* okay for the compiler to make that same assumption.
Only if the compiler can somehow prove that this code is always called
in a context where it causes undefined behavior. It would take a very
smart compiler to do that, though. You cannot say that just looking at
the code of g() in isolation, since it can be used legally, as in:
OhFooey->~Foo();
g(1);
> I don't understand why you think this is OK, but anyway could you
> tell me how the compiler is going to know what is actually being done
> in this part of the code while optimizing some other piece of code ?
It does not matter what the compiler knows or does not know, or what
reasoning it employs or does not employ. The moment a code that exhibits
undefined behavior is executed, you cannot argue any longer whether the
compiler-generated code is right or wrong. By definition of undefined
behavior, anything the program might do from that point on is correct.
You cannot argue that the code is wrong, because to argue that, you have
to say: I expected the code to behave in such and such a way, but I
observe it behaving differently, therefore it must be wrong. By coding
undefined behavior, you lose the right to claim any expectations as to
the compiler output, so you cannot clear the first hurdle of that
two-pronged test (yes, I had to read a lot of legal documents lately).
--
With best wishes,
Igor Tandetnik
"On two occasions, I have been asked [by members of Parliament], 'Pray,
Mr. Babbage, if you put into the machine wrong figures, will the right
answers come out?' I am not able to rightly apprehend the kind of
confusion of ideas that could provoke such a question." -- Charles
Babbage
- Next message: Arnaud Debaene: "Re: How to get processor serial number?"
- Previous message: Saravana Vijayan.B: "integer pointer values..convert to char"
- In reply to: Sigurd Stenersen: "Re: faster access private or public?"
- Next in thread: Tom Widmer: "Re: faster access private or public?"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
