Petr Belyaev

shouldn't protobuf be listed in the package dependencies? For example, without the protobuf package installed it's impossible to use the Healthcheck as it imports `google.protobuf`.

> Google's `grpcio` also doesn't have `protobuf` as a dependency, only as an optional dependency Why not make it an optional dependency as well, i.e. `grpcio[protobuf]`? It will allow the...

Any progress so far? I was just wondering why options can't be attached to the field metadata? It might be a bit hard to get them out of the dataclass,...

can confirm, to_dict seem to ignore `_serialized_on_wire`

Probably something like this should do: ```diff diff --git a/tests/test_features.py b/tests/test_features.py index f548264..e040f1b 100644 --- a/tests/test_features.py +++ b/tests/test_features.py @@ -197,6 +197,14 @@ def test_optional_flag(): assert Request().parse(b"").flag is None assert Request().parse(b"\n\x00").flag...

I think `_serialized_on_wire` parameter should be moved from `Message` to a subclass of `dataclass.Field`, which should be created for every field, not just messages. In fact, even the doc for...

>You can just check if the field's raw value is PLACEHOLDER hm, should actually work, true. But then why do we even need the `_serialized_on_wire` flag, as every field is...

Turned out it's not that simple with placeholders. Typical test case for the `serialized_on_wire` method is: ``` foo = Foo() assert betterproto.serialized_on_wire(foo.bar) is False ``` Trying to compare `foo.bar` with...

> This library does still support proto 2 is this also true for that version? No, for proto2 there should be a `HasField` method which does exactly what serialized_on_wire does,...

I think field metadata should be easier to maintain and alter if needed than the custom field class. Like the one that's already there. Though I'm not sure if field...