Get rid of formulas objects and implement the formulas as a struct
template taking parameters enabling the formula parts (e.g. distance
and/or azimuth) with apply() static member function. This way the code is
more clear as all temporary values are created directly in the algorithms.
Furthermore there are no uninitialized values and it's not required to
keep a flag indicating if the distance or azimuth can be calculated.
Use the new implementation of formulas in geographic distance and side
strategies.
Alter the unit test.
The optimization is based on the fact that in the most cases the segments
handled in the TurnInfoPolicy are disjoint. For disjoint segments first
the points are rescaled, then the test for disjoint is run, next the
TurnInfoPolicy just returns and the next pair of segments is handled.
Therefore to optimze the get_turns() this commit changes two things:
1. the cart_intersect strategy may return just after the calculation of sides
for the first segment (2 sides calculation instead of 4),
2. the points are rescaled only one time in the intersection_helper,
already rescaled points are passed into the intersection strategy
* step 1: compute difference(g1, g2)
* step 2: compute difference(g2, g1)
* step 3: return the union of the two differences in previous two steps (as opposed to returning
the concatenation of the previous two steps)
Create and return the end iterator in initialize() function the same way how
iterator `it` is, by output function argument passed by reference. Reuse it
inside this function.
In ..._parser::apply() functions reuse the end iterator, do not create it
each time it is needed.
Consistently pass the end iterator by const reference.
A but segment B is much shorter, and IP is also at end-point of segment B.
The arbitrary decision (robust_ra < robust_rb) is changed into another
decision, based on closeness to end-points and on segment length, which leads
to more precise results
distance when the point's longitude does not belong to the longitude range of the
box: instead of computing the distances to both meridian segments of the box, determine
which one among the two box segments the point is closest to, and compute the distance
of the point to the closest segment found
use flatten_iterator to get all rings of a multipolygon as a single range;
separate initializer from num_rings computation; use operator new/operator delete
for dynamic memory allocation of views' container; use automatic deallocator helper
class instead of try/catch block;
