2 Turns are generated for a spike, one with operation_blocked the second one with operation_intersection.
relate_linear_linear is not working yet, for now it is only ensured that boundaries will not be checked for those turns, which would result with assert failure.
TODO: opposite collinear and equal, touches c/c and endpoints intersecting a spike + the adaptation of relate(L,L).
It's also consistent with the behavior for Point-like Linestrings.
E.g. some MultiLinestring may have boundary and interior in the same point:
Ls1:
|--------------|
MLs2:
|------------|
|------|
In this case the current version of the algorithm will set BB=0 and BI=0.
Point-size Linestring is a linestring containing exactly 2 equal points.
Add generation of positions for degenerated turns, non-degenerated segments in get_turns_linear_linear.
The code should be polished after the addition of info about the degeneration of segments into Turns.
The intention is to prevent creation of too big files in MinGW.
Move predefined Geometries for relops from test_relate.hpp to predef_relop.hpp
Move test_geometry<> defined in relate.cpp to test_relate.hpp
Enable touches() for other pairs of Geometries.
Implement ready-to-use in the implementations of relate() less comparators of operations.
Change the less comparator for L/A.
Fix: handling of the entry/interior for some turn with MultiID which comes before union for some other turn with different MultiID.
Finish the support for boost::mpl::vector<> as a complex StaticMask.
Also the specializations for Multi Geometries was moved from geometry/multi/algorithms to geometry/algorithms. They're now in one place.
The default strategies are point_in_point for PointLike, winding<> for PointLike/AnyGeometry, relate for L/L, L/A and A/A.
Point/MultiGeometry version now uses point_in_geometry directly which implements the correct behavior - this fixes the wrong handling of MultiLinestrings (boundary point of a Linestring may not be a boundary point of a MultiLinestring which contains it).
Add may_update() funcionality which may be used if it's possible to change the result for the specified matrix fields and overlap dimension.
Initialize flags in no_turns_aa_pred and uncertain_rings_analyser with values calculated using this functionality.
Add manual checks around the parts of areal_areal::apply() where IPs are sorted and analysed.
Rename turns::less_seg_dist_op<> to relate::turns::less<>
It now takes the Comp defining how operations are sorted if segment and distance are equal.
Add 2 different operations comparators for L/L,L/A and A/A.
Add tests.
Add support for:
- single geometries for which no turns were generated
- rings for which no turns were generated
- i/i and u/u turns
TODO: some cases where turns are generated for the intersection of a "real" hole and a "fake" hole (self-touching exterior ring).
Change the ignoring check calculation in get_turn_info_for_endpoint.
Add additional parameters to the second version of collinear_opposite<>::apply() defining if the pk and/or qk are valid (in other words current segment is NOT the last one).
Also make ignoring check for last endpoint in L/A more clear.
In this case no sections were generated for those "empty" Linestrings but if there were some sections already generated they could be marked as last_non_duplicated.
