Potential FPE suppression/hiding in GsfUtils.hpp

[andiwand]

acts/Core/include/Acts/TrackFitting/detail/GsfUtils.hpp

Lines 197 to 200 in c0e65bc

	// If something is not finite here, just leave the weight as it is
	if (std::isfinite(factor)) {
	weights.at(tip) *= factor;
	}

This looks like silent FPE treatment which might either come from the chi2 being too large pushing detR to inf or caused by detR being negative in case the covariance is already broken. Both should in principle not happen but we protect against it and silence any occurrence right now.

In #3449 I observed an FPE at

acts/Core/include/Acts/TrackFitting/detail/GsfUtils.hpp

Line 195 in c0e65bc

const auto factor = std::sqrt(1. / detR) * safeExp(-0.5 * chi2);

which indicates that detR is negative and this should only happen if at least one of the covariances is broken.