[boost] Re: stride iterators and matricies

From: David Abrahams (dave_at_[hidden])
Date: 2005-06-06 10:10:04

Larry Evans <cppljevans_at_[hidden]> writes:

> On 06/03/2005 02:47 PM, David Abrahams wrote:
>> Neal Becker <ndbecker2_at_[hidden]> writes:
>>>David Abrahams wrote:
> [snip]
>>>>B. What about the undefined behavior that results when you stride off
>>>>   the end of a container?
>>>
>>>Is there any reasonable way to detect this?  If it is feasible without
>>>significantly increasing cost, I'll add it - but offhand I don't know that
>>>it is generally feasible.
>>
>> If it's random access you can store a base iterator and an index.
>                                                              ^^^^^
> Would "offset" be a better word?  An index, i, to me, implies
> something satisfying:
>
>    0 <= i < shape[axis]
>
> where axis satisfies:
>
>    0 <= axis < rank
>
> where rank is the number of dimensions in the array and shape[axis] is
> the length of the axis-th dimension.

Sure, whatever.

> With this defintion, I understand what you mean, i.e. instead of
> the stride iterator incrementing the pointer, it would just store
> the beginning of the object:
>
>   X* px0;
>
> and increment the offset.  Then dereferencing would simply:
>
>   return *(px0+offset);
>
> Is that you're meaning?

Yep.

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com