Getting such a mode to work fully seems like it'd be quite a deep set of changes to V8, since you'd want it to happen in all paths and all tiers of optimization. I don't think this is something we'd want to do.

I doubt that the devtools part is much more tractable, but I'm happy to leave that label for them to answer.

Note that x86* can trap floating-point exceptions that by default are masked. This includes invalid operations (NaN), division by zero (infinity), overflow (infinity), underflow (denormal), and inexact results (precision loss).

On Linux you can control these using sigfpe(), on Windows you can use _controlfp() and signal().

Note that these are exceptions at the native code level (actually at the machine level), so correlating them to JavaScript code is the harder part. That said I assume V8 has ways of handling some of these low-level exceptions, so it might not be all that hard to get it working as desired.

[*] I think other CPU architectures support this too, because these exceptions are defined by the IEEE-754 floating-point standard.

Alexey, can we make embedder to play with sigfpe when debugger is active?