Stop capturing panics and trying to continue execution after them.

[mstoykov]

What?

Drop the following code and it uses:

k6/js/common/util.go

Lines 67 to 87 in 153e0e5

	// RunWithPanicCatching catches panic and converts into an InterruptError error that should abort a script
	func RunWithPanicCatching(logger logrus.FieldLogger, rt *goja.Runtime, fn func() error) (err error) {
	defer func() {
	if r := recover(); r != nil {
	gojaStack := rt.CaptureCallStack(20, nil)
	err = &errext.InterruptError{Reason: fmt.Sprintf("a panic occurred during JS execution: %s", r)}
	// TODO figure out how to use PanicLevel without panicing .. this might require changing
	// the logger we use see
	// https://github.com/sirupsen/logrus/issues/1028
	// https://github.com/sirupsen/logrus/issues/993
	b := new(bytes.Buffer)
	for _, s := range gojaStack {
	s.Write(b)
	}
	logger.Error("panic: ", r, "\n", string(debug.Stack()), "\nGoja stack:\n", b.String())
	}
	}()
	return fn()
	}

Why?

The code was added many years ago when k6 had a bunch of really hard to fix race conditions. And was left as a safeguard.

Unfortunately it has two problems - it is not enough for most cases with multiple goroutines/async. And probably even worse - makes a bunch of potential bugs seem like less of an issue.

We have had at least one user come and ask us to have this be optional as it was problem during extension development.

Also, as far as I am aware it hasn't caught a single issue where k6 continuing has been a good idea for years. Some of the early things were nice, but now it seems like it just gets in the way.

Suggestion:

Just remove the code and it's called site to be rewritten with no defers.