thingverse-init sometimes tries (and fails) to serialize datetimes to JSON

[skrytebane]

Sometimes, when it doesn't fail due to bug https://github.com/NIVANorge/nivacloud/issues/258, thingverse-init fails with a JSON serialization error. This seems to also be somehow nondeterministic even though you would expect it to either always fail or always succeed.

The actual code that fails is in metaflow.py (in pyniva) at the start of the update_thing function. Sometimes thing that is passed in here has start_time and end_time fields with str values, and sometimes they are datetime values.

I've only seen this when debugging, and possibly suspending execution for a while, but I think it's very strange that this should be caused by a timeout somewhere. Some kind of Heisenbug.

[skrytebane]

(It's possible to work around this by providing a custom serializer to json.dumps, so maybe I will do that.)

[skrytebane]

I found this thing in thing.py in pyniva:

    @property
    def start_time(self):
        if self._meta_dict.get("start_time", False):
            if isinstance(self._meta_dict["start_time"], str):
                self._meta_dict["start_time"] = parse(self._meta_dict["start_time"])
            return self._meta_dict["start_time"]
        else:
            return None

    @property
    def end_time(self):
        if self._meta_dict.get("end_time", False):
            if isinstance(self._meta_dict["end_time"], str):
                self._meta_dict["end_time"] = parse(self._meta_dict["end_time"])
            return self._meta_dict["end_time"]
        else:
            return None

This was tripping me up. I couldn't see anywhere in the code where strings were converted to datetimes and why it only happened sometimes. They're being automatically converted when you read them, and then stored. So if you've accessed them at least once, the next time you read them from a Thing object, they'll be datetimes.

I'm pretty sure this is an antipattern...

[skrytebane]

(Anyway, this is why the code is only failing when I look at it in IntelliJ's inspector (debugger), because that's causing these to be called. So this is a bug in pyniva, I guess.)