Geometry Boost List: Re: [geometry] Using Boost.Geometry with a legacy class hierarchy

Subject: Re: [geometry] Using Boost.Geometry with a legacy class hierarchy
From: Mats Taraldsvik (mats.taraldsvik_at_[hidden])
Date: 2012-02-22 06:53:22

Hi Barend,

On 02/21/2012 08:31 PM, Barend Gehrels wrote:
> Hi Mats,
>
> On 21-2-2012 12:43, Mats Taraldsvik wrote:
>> Hi Barend,
>>
>> On 02/20/2012 07:21 PM, Barend Gehrels wrote:
>>> Hi Mats,
>>>
>>>
>>> On 20-2-2012 9:03, Mats Taraldsvik wrote:
>>>> The object hierarcy is a bit more complex than my current example 
>>>> (also contains 2d, 3d points, curves etc.), and objects are stored 
>>>> in pointer to base/super-containers. Therefore, I can't remove the 
>>>> pointer requirement, I'm afraid. I guess creating separate pointer 
>>>> and non-pointer specializations (e.g. point_type and 
>>>> point_type_ptr) since they are so different, means either a lot of 
>>>> duplicated code, or a very difficult problem...
>>>>
>>>> I was also hoping to overcome the pointer-to-base container issue 
>>>> also, but thought it was best to take one step at a time. :)
>>>
>>>
>>> OK, so the first solution I mailed is still based on pointers. Does 
>>> that work for you?
>>>
>>
>> After re-reading your email, I realise that I misinterpreted it -- 
>> the version you suggested, using references instead of pointers, 
>> should work, since it does not modify my hierarchy.
>>
>> --
>>
>> However, since I the structure is a bit more complex, and I need to 
>> worry about casting from base pointers etc. later, I want to use the 
>> pointer version, as long as it works and behaves correctly.
>>
>> I managed to fix the issue in QRing, without modifying Boost.Geometry. :)
>>
>>
>> Basically, I needed to change the ConstRingIterator :
>> from:
>> typedef RingIteratorImpl<std::vector<QLineString*>::const_iterator, 
>> const QPoint*> ConstRingIterator;
>> to:
>> typedef RingIteratorImpl<std::vector<QLineString*>::const_iterator, 
>> QPoint* const> ConstRingIterator;
>>
>> and then it woked perfectly.
>>
>> However, I'm not sure if the semantics are right -- shouldn't the 
>> QPoint be const as well as the pointer (i.e. const QPoint* const)? 
>> When ConstRingIterator is changed to (const QPoint* const), it does 
>> not compile (perhaps it is because of lacking specializations (I have 
>> defined QPoint* and const QPoint*).
>>
>> I updated my gist with the latest version [1] (a diff userscript is 
>> available for convenience, if you're not using git [2])
>
> OK, great that it compiles.
>
> I agree, the const-pointer does not mean that the pointee cannot be 
> modified so indeed it should be expected as const Qpoint* const (or, 
> in notation I prefer, QPoint const* const).
>
> The following does compile now:
>
> void test(QRing const& ring)
> {
>     boost::range_iterator<QRing const>::type cri = range_begin(ring);
>     (*cri)->x = 5;
>     std::cout << boost::geometry::wkt(*cri) << std::endl;
> }
>
> Which should not compile - it gives the wrong expectations.
>
> The problem with the pointer-version is that QPoint* and QPoint const* 
> has to be specialized separately, and are therefore considered 
> differently in our library. That is not easy to solve. We will discuss 
> that internally.
>

Ok, so even though I define the four "versions" of QPoint (QPoint 
const*const, QPoint* const, QPoint const* and QPoint*), Boost.Geometry 
can't properly handle/distinguish between the QPoint* and QPoint const*?

I guess the only or best available workaround is to iterate through the 
QPoint*, and return references to QPoint?

I hope you can solve this. :) If you want to use the code as an example 
of adapting legacy classes/objects in the documentation, I'll be happy 
to write a few accompanying paragraphs. :)

> Regards, Barend
>
>

Regards,
Mats

>
>
> _______________________________________________
> Geometry mailing list
> Geometry_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/geometry