Re: [Monotone-devel] Re: using empty() instead of size()

[Ulf Ochsenfahrt]

Bruce Stephens wrote:
> Ulf Ochsenfahrt <address@hidden> writes:
>
> [...]
>
> > If there is an invalid pointer in the non-empty list, the program
> > can crash.
>
> It can crash, but this is presumably in the realm of undefined
> behaviour, so arbitrary things are permitted to happen.
>
> > If the compiler optimizes (i.e. removes) the list traversal, the
> > program doesn't crash. Thus, the optimization would change the
> > runtime behavior of the program.
>
> Probably, but I think that's a permitted change.

What about inifinite loops? The linked list could also loop back to itself.

> I'd guess a bug report along those lines ("your optimizer caused my
> program not to crash!") would be unlikely to be given a high priority.

Compiler writers are usually very conservative. I would guess that 'the 
optimizer makes the program behave differently' is a serious enough issue.

> An implementation with an O(1) implementation of size() would
> presumably also not crash at that point either.

True. But the compiler can't know that. The compiler only sees the code, 
it doesn't know what it's supposed to do.

Cheers,

-- Ulf