CGAL users discussion list

[Bart Janssens <>]

Hi all,

While working on the integration of CGAL 3D booleans in K-3D, I had
seen some problems where the output mesh was not always the same,
depending on what machine the test was run on. The meshes were always
visually correct, but the order of the edges would change in a
seemingly random way. Until now, I always assumed this was because of
floating point errors in the source mesh, but after having worked
around those, I think I have a test case that proves this behavior may
be caused by CGAL.

Attached is a test case that takes the difference between two slightly
deformed poly-cylinders (see screenshot). Included are the input
meshes and 3 reference output meshes, that resulted from running the
test program here. Compile using:
g++ -frounding-math boolean_consistency_test.cpp -lCGAL -lmpfr -lgmp -o test
Running ./test creates a new output.nef3 that can be diffed against
the reference files. Reference file 1 results from a run inside K-3D
on 32 bit linux, file 2 is the current test program output on 32 bit
linux and output 3 was created on a 64bit linux using the test
program.

Does anyone know what could cause the differences in the outputs? All
tests were run using a homogeneous kernel with Gmpz as number type.

Kind regards,

-- 
Bart

Attachment: boolean_consistency_test.tar.gz
Description: GNU Zip compressed data

Attachment: test_screenshot.png
Description: PNG image