AutoHotkey Community

If there is an error evaluating an expression, it could set the ErrorLevel. For example x := 1/0 or y := 1<<64 or z := sqrt(-1).
1 = fix point overflow
2 = floating point overflow
3 = imaginary square root
4 = log of negative number…

The error level need not be a bit mask, because the first error usually prevents a meaningful evaluation of the rest of the expression.

It could be shifted up, starting from say 256, reserving 8 bits for the uppermost user function. Of course a construct like f(g(1/0)) indicates a fix point overflow and the user ErrorLevel of f(), but the error level of g() is lost. For that we could have a cumulative error bit: there was a user set ErrorLevel in the course of evaluating the expression.

Another possibility is to leave ErrorLevel alone, but introduce another built in variable: ArithmeticError. It is set to 0 at start of a thread, but it is set and stays set at an error. This way the user can trust the result of a series of calculations if ArithmeticError is still 0 at the end. Otherwise s/he can reset it to 0 and do corrective actions.

I agree that it's best to leave ErrorLevel alone, mainly because changing it would break existing scripts that rely on the fact that most expressions don't change ErrorLevel.

Your idea for implementation seems good. I've added it to the to-do list.

Thanks.