intersection produces invalid polygon (is_valid fails) with (nearly) duplicated point

[rgreenblatt]

#include <boost/geometry.hpp>

#include <iostream>

int main() {
  using Point = boost::geometry::model::d2::point_xy<double>;
  using Polygon = boost::geometry::model::polygon<Point>;

  Polygon l{{
      {0.33333327000590729, 0.33333344706803125},
      {0.50967067548140821, 0},
      {0, 0},
      {0, 0.93199284523078707},
      {0.33333327000590729, 0.33333344706803125},
  }};
  Polygon r{{
      {0, 0.8554426406197928},
      {0, 1},
      {1, 0},
      {0.54614530500375813, 0},
      {0, 0.8554426406197928},
  }};
  assert(boost::geometry::is_valid(l));
  assert(boost::geometry::is_valid(r));
  std::vector<Polygon> out;
  boost::geometry::intersection(l, r, out);
  for (const Polygon &p : out) {
    std::cout << boost::geometry::wkt(p) << std::endl;
    assert(boost::geometry::is_valid(p)); // fails!
  }
}

Fails with:

POLYGON((0.333333 0.333333,0 0.855443,0 0.931993,0.333333 0.333333,0.333333 0.333333))
a.out: temp.cpp:29: int main(): Assertion `boost::geometry::is_valid(p)' failed.
[1]    2691824 abort (core dumped)  ./a.out

This reproduces with 1.75 and the develop branch of this repo. Note that the 2 points which are very close to (1/3, 1/3) as output by wkt are not equal - they are just very close.

[tinko92]

I can no longer reproduce this with 1.79 (still core dumps with 1.78), so maybe it is fixed.