Re: [boost] [Review] GGL review starts today, November 5th

Subject: Re: [boost] [Review] GGL review starts today, November 5th
From: Barend Gehrels (barend_at_[hidden])
Date: 2009-11-22 13:44:06

Hi Pierre,

> [...] and that something simplier could even work : instead of a variant, a tuple of back_inserter could do the trick very nicely. 
>
> vector<point2D> myPoints;
> vector<line2D> myLines;
> vector<polygon2D> myPolygons;
>
> typedef tuple<vector<point2D>, vector<line2D>, vector<polygon2D>  > AFakeTupleType;
>
> intersection_inserter<AFakeTupleType>(poly1, poly2, make_tuple(back_inserter(myPoints), back_inserter(myLines), back_inserter(myPolygons));
>
> Please note there is no object of type AFakeTupleType constructed. This type is just used to drive the algorithm.
>
> The output can be dispatched to different vectors, and the input is still 1 object. Then, the user can simply check the lenght of each result to check if there was an intersection.
>
> This is less mathematics oriented, but still completely usable, and more simplier, because of the redirection of the result to the
> selected vectors.
>
> You already use tuple, so maybe references wrapper in AFakeTupleType would be needed to make the distinction with your other tuple.
>
> What do you think of that? Do you see any drawbacks?
>   
I think it will work. It would certainly solve the performance issue, 
you get all three collections back in one calculation.

Potential drawbacks (or, better stated, issues):
- it is more complex; I think (personally) that most users want only the 
intersected polygon back. To always require the complexity described 
here might be a bit too much
- but that is no real problem, because we can create another function 
for it, or even automatically detecting what is happening here, in the 
tag dispatching
- the AFakeTupleType (tuple of vectors) does not match the make_tuple 
(tuple of output iterators), as you said
- but also that is no real issue (actually, it is the same as it is 
working now, the OutputType and the output iterator)

I think it is feasable and a very good idea; but not by default (because 
of the complexity involved, also for the library user). But if people 
want to get all these three different things back, yes, I think this is 
the best option discussed until now. Does the trick very nicely, as you 
said, indeed. Thanks!

Regards, Barend