Open millerta opened 6 months ago
Note the macports compilers do not have this issue. The stor test results are the same on mac and linux. Compiled with sonoma on mac laptop: export CC=/opt/local/bin/gcc-mp-12 export CXX=/opt/local/bin/g++-mp-12 export FC=/opt/local/bin/gfortran-mp-12
Issue Example in test/level03/stor_precision_diff
New mac compilers handle floats differently resulting in precision differences for very small values. This caused one of the tests to fail and it was commented out for level01 tests and moved to test/level03/stor_precision_diff
For "degenerate" cases with 4 vertices on circumscribed circle, you can put the diagonal edge of the rectangle on either diagonal and round off may be the deciding factor. Different connection of degenerate case is still correct/Delaunay.
Possible fixes: change test to avoid square, make vertical length shorter than horiz or perhaps there is a flag fix for the mac compiler
NOTE files written without zero coefs are the same: output_mac/out_hex_3x4.uge output_linux/out_hex_3x4.uge
linux and new mac: