https://bugs.winehq.org/show_bug.cgi?id=38256
--- Comment #5 from Matteo Bruni matteo.mystral@gmail.com --- (In reply to Sebastian Lackner from comment #4)
Would it make sense to add any error checking for invert_matrix_3d_general, or how did native behave before in such odd situations? ^^
Yes, there should be a check there or the BOOL return value from the function removed. Specifically Mesa uses an identity matrix when the source matrix is singular. I should write a test and check what happens on Windows in that case (as unlikely / broken that is). Actually https://msdn.microsoft.com/en-us/library/windows/desktop/bb172390%28v=vs.85%... uses a slightly different definition for the normal matrix. Not sure MSDN is correct, I should test that too.