B E Z I T O P O
This is a bugfix release. Contour drawing occasionally made strange errors. These turned out to be at least five separate bugs:
- When the scene is at least 400 km from the origin, loss of precision may cause a subdivision segment of a triangle with a secondary critical point to be retained along with its reverse. This causes contour tracing through the triangle to get lost.
- Same as above, except that the retained segment is one whose intersection with another segment is very close to T-like.
- Loss of precision or faulty implementation may cause Brent's method to mess up.
- In an acicular triangle with J-shaped contours, contours may be retraced, resulting in tangles. Where this happened depends on whether Brent's or Newton's method was used, but it's happened with both.
- The order of tolerance for rough and smooth contours was reversed, resulting in smooth contours that did not follow the actual contours closely.
Fixing these bugs (all but the last) means that it shouldn't be necessary to translate a scene to the origin in order to get good contours.
I also replaced the geoid file used in the geoidboundary test with a fictitious surface to speed up the test.
Screenshots
These are of the Okular program displaying PostScript files output by convertgeoid.
Part of Independence Park, showing tennis courts (high area in northeast), concrete-covered yard inlets, and a drainage ditch
|
Magnified area outside the surveyed area of Independence Park, but inside its convex hull, which had a tangle after I replaced Brent's with Newton's
|
Download
System requirements
Requires CMake 2.8.12 or later and a C++ compiler that understands string::pop_back.
DragonFly BSD: It works on 4.9.0.
Debian and derivatives: Jessie or later. See /etc/debian_version . Ubuntu Xenial's Debian version is Stretch.
Release build status
- Ubuntu Xenial x86_64: It builds and works well.
- DragonFly BSD 4.9.0 x86_64: It builds and works well.
- Raspbian Jessie: It builds and works well.
Pierre Abbat
NC PLS L-5129