Floating-point math has an answer for everything, even if the question is 1.0/0.0 (INF) or 0.0/0.0 (NaN) or 1.0+Nan (NaN). It even has an answer for "Pop the floating-point value off of the x87 stack even though the stack is empty" although I think that answer is indeterminate.

However some of these answers we don't actually want. And yet, we get them because Chrome, like most Windows software, runs with floating-point exceptions suppressed. This allowed a code-gen bug in PGO builds:

https://connect.microsoft.com/VisualStudio/feedback/details/3102400

to go unnoticed for a while and unfixed for even longer. If we enabled some floating-point exceptions during tests we might also find logic bugs. Or we might find a bunch of benign divide-by-zero errors, but we can always just enable the most useful exceptions.

See this (excellent) blog post for more information.

https://randomascii.wordpress.com/2012/04/21/exceptional-floating-point/