mmg2d segfaults with non SPD metric

[taupalosaurus]

see attached file

segfaultMMG.zip

[taupalosaurus]

I tracked down the bug, the segfault, and the culprit is MMG2D_caltri_ani in src/mmg2d/quality_2d.c. At l.163, the determinant can be <0 and the sqrt gives a nan, which in turns end up with a return value which is nan. The segfault actually happens a bit further in the file, in MMG2D_outqua: l.242/243 a vector index is built from this return value.

I am surprised that we don't check that a metric is actually SPD - I looked but didn't find any function that does that. @corentin-prigent and @dapogny do you know if there's such a function ?

Otherwise, I suggest we check as soon as possible in the code (when reading the file, or right after and in/right after the interface functions).

[corentin-prigent]

Please see this PR.