Slincing behaviour

[Ali Çehreli]

On 11/11/2011 01:42 PM, Steven Schveighoffer wrote:
> On Fri, 11 Nov 2011 16:10:12 -0500, Simon <s.d.hammett at gmail.com> wrote:
>
>> On 11/11/2011 19:04, Steven Schveighoffer wrote:
>>> On Fri, 11 Nov 2011 14:01:42 -0500, Steven Schveighoffer
>>> <schveiguy at yahoo.com> wrote:
>>>
>>>> There should be no bounds error in any case, an empty slice is valid.
>>>
>>> By "in any case" I meant in either debug or release mode.
>>>
>>> -Steve
>>
>> even when you index beyond the bounds of the slice?
>>
>> you may not actually be reading memory because it's zero length, but
>> it's still logically invalid; you've gone outside the valid range.
>
> You are not reading beyond the valid range. A zero-length slice is
> perfectly legal to point at the end of an array or other slice. Reading
> any data from a zero-length slice will cause an out-of-bounds error in
> debug mode, because it has no elements.
>
>> in vc9, if you increment an iterator beyond the valid range you get a
>> debug assert. that's caught quite a few bugs where I work when we
>> upgraded to vc9.
>
> I think you are misunderstanding what the $ actually means.
>
> It's the equivalent in C++ iterators to x.end.
>
> The pair of iterators x.end, x.end is a valid range. Going *beyond*
> x.end would be illegal. But iterating *to* x.end is legal (which would
> be the equivalent of [$..$] range), and you will not be able to convince
> me that vc9 doesn't allow it.
>
> -Steve

How about Jonathan's this comment: "It wouldn't surprise me if arr[500 
.. 500] worked exactly the same way. Because the array is empty, it 
doesn't really matter  what values you gave it."

I think Simon is objecting to 500..500 being accepted (if at all). I 
agree that $..$ is correct.

Ali