Opened 5 years ago
Closed 5 years ago
#8254 closed Bugs (fixed)
Faulty intersection calculation
Reported by: | Owned by: | Barend Gehrels | |
---|---|---|---|
Milestone: | To Be Determined | Component: | geometry |
Version: | Boost 1.53.0 | Severity: | Showstopper |
Keywords: | Cc: |
Description
I encountered a (supposed) bug in the boost::geometry::intersection functionality.
The calculation of the intersection of a tiny triangle lying in a considerably bigger trapezoid where one edge of the triangle lies (within floating point precision) on one edge of the trapezoid results in an intersection which consists of the original trapezoid with one additional node on one of its sides.
The geometry is given in a C test routine attached to this ticked. The actual call of the intersection functionality is conducted in a C++ wrapper routine, which is also attached. The bug occurred on Mountain Lion with the llvm-gcc-4.2 compiler.
Attachments (1)
Change History (3)
Changed 5 years ago by
Attachment: | IntersectionBugFiles.zip added |
---|
comment:1 Changed 5 years ago by
comment:2 Changed 5 years ago by
Resolution: | → fixed |
---|---|
Status: | new → closed |
Fixed. The underlying sections (used for performance) were too small in rare cases. Made them a tiny bit larger, such that intersections are not missed by this. By change this was also necessary for one of my own testcases, for multi-point buffers.
Fixed in trunk, will be released in Boost 1.54
I reproduce this. Using MSVC the output is correct, according to the input. Using gcc (I'm using 4.4.5) I get similar output as you give.
I will have a closer look, but I will have to debug differently because it's not occuring in MSVC.