ShimCall error handling is full of dragons

[Kubuxu]

We should not be using numerical exit codes there

lotus/chain/vm/runtime.go

Lines 111 to 148 in a081dc7

	func (rt *Runtime) shimCall(f func() interface{}) (rval []byte, aerr aerrors.ActorError) {
	defer func() {
	if r := recover(); r != nil {
	if ar, ok := r.(aerrors.ActorError); ok {
	log.Warnf("VM.Call failure: %+v", ar)
	aerr = ar
	return
	}
	//log.Desugar().WithOptions(zap.AddStacktrace(zapcore.ErrorLevel)).
	//Sugar().Errorf("spec actors failure: %s", r)
	log.Errorf("spec actors failure: %s", r)
	aerr = aerrors.Newf(1, "spec actors failure: %s", r)
	}
	}()
	ret := f()
	if !rt.callerValidated {
	rt.Abortf(exitcode.SysErrorIllegalActor, "Caller MUST be validated during method execution")
	}
	switch ret := ret.(type) {
	case []byte:
	return ret, nil
	case *adt.EmptyValue:
	return nil, nil
	case cbg.CBORMarshaler:
	buf := new(bytes.Buffer)
	if err := ret.MarshalCBOR(buf); err != nil {
	return nil, aerrors.Absorb(err, 2, "failed to marshal response to cbor")
	}
	return buf.Bytes(), nil
	case nil:
	return nil, nil
	default:
	return nil, aerrors.New(3, "could not determine type for response from call")
	}
	}

[arajasek]

@Kubuxu Duplicate of / subsumed by #3015 maybe?

[raulk]

@alanshaw is tackling this one, in the context of the test failures he found. @alanshaw, mind elaborating on what you're doing here?

[raulk]

@Kubuxu also, mind quickly listing the dragons here (besides the numerical exit code use)?

[Kubuxu]

Yeah, it is mostly the numerical exit codes that are completely wrong.
We will also need to figure out an exit code for the non-actor panic.

[alanshaw]

I'm hoping that #3697 is an ok solution for one of these numerical exit codes 🤞

[magik6k]

No longer relevant in the FVM world