attachment set

Mateusz Loskot — Wed, 28 Sep 2011 17:02:37 GMT

attachment → distance_pythagoras-sqrt-fix.patch

Patch fixing missing sqrt() callin distance_pythagoras.hpp. Also fixes oordinates/points mismatch, trivial.

Mateusz Loskot — Wed, 28 Sep 2011 17:05:57 GMT

Also, please check FIXME comment strategies/agnostic/simplify_douglas_peucker.hpp:148 (r74600)

Mateusz Loskot — Wed, 28 Sep 2011 17:18:23 GMT

The test program is now in the trunk, as l01_multipolygon_simplify.cpp example (r74598)

attachment set

Mateusz Loskot — Wed, 28 Sep 2011 17:19:13 GMT

attachment → l01_multipolygon_simplify.cpp.log

l01_multipolygon_simplify.cpp test run log

Mateusz Loskot — Fri, 30 Sep 2011 11:56:28 GMT

The patch has been proven incorrect as the sqrt is supposed to not to be called where I added it. However, I'm still confused why different strategy is selected in the simplify.

Here is illustration of what strategy is actually called:

#include <boost/geometry.hpp>
#include <boost/geometry/strategies/cartesian/distance_pythagoras.hpp>
#include <boost/geometry/geometries/point_xy.hpp>
using namespace boost::geometry;
int main()
{
  typedef model::d2::point_xy<double> point_xy;
  point_xy p1(0.0, 0.0);
  point_xy p2(5.0, 0.0);
  // 1) This is direct call to Pythagoras algo
  typedef strategy::distance::pythagoras<point_xy, point_xy, double> strategy1_type;
  strategy1_type strategy1;
  strategy1_type ::calculation_type d1 = strategy1.apply(p1, p2);
  // 2) This is what is effectively called by simplify
  typedef strategy::distance::comparable::pythagoras<point_xy, point_xy, double> strategy2_type;
  strategy2_type strategy2;
  strategy2_type::calculation_type d2 = strategy2.apply(p1, p2);
  return 0;
}

The 1) calls all expected implementation stages:

1 -> boost::geometry::strategy::distance::comparable::pythagoras<...>::apply() [Note 1]
2 ->-> boost::geometry::strategy::distance::pythagoras<...>::apply()
3 ->->-> sqrt()

Note 1, with the patch attached above, this will lead to double call to sqrt(), because the step 2 squares the result of step 1 (as expected), so the sqrt in step 1 is redundant and incorrect.

The 2) execution path seems to be missing the step 2 from the call stack above leading to no call to sqrt() at all what yields partial/incorrect result:

1 -> boost::geometry::strategy::distance::comparable::pythagoras<...>::apply() [Note 2]
3 ->->-> sqrt()

Note 2, that's why I hardcoded sqrt call in this function directly, but it is not correct in big picture, as explained in Note 1 above.

I'm having troubles with following the selection of pythagoras vs comparable. I may not understand well what's happening in boost::geometry::strategy::distance::services. So, all my hope in Barend!

anonymous — Thu, 06 Oct 2011 16:44:32 GMT

Is this related #5730 ?

status changed; resolution set

Barend Gehrels — Thu, 06 Oct 2011 17:06:32 GMT

status new → closed
resolution → fixed

(In [74761]) Fix ticket 5954, use strategy directly, not the comparable strategy (unless fixed otherwise)

Boost C++ Libraries: Ticket #5954: distance_pythagoras skips sqrt() step

attachment set

attachment set

status changed; resolution set