Issue: example error parsing problem

[jschaul]

This PR is twofold:

1. bug report
2. question about unit testing

1. bug report

As we deviate from the "happy path" (e.g. here when throwing an error), using mu-grpc-client becomes problematic.

sayHello function:

sayHello :: MonadServer m => HelloRequestMessage -> m HelloReplyMessage
sayHello (HelloRequestMessage nm) = do
  case nm of
    -- in some cases we want to throw an error, here when the name sent is 'Bob'
    "Bob" -> throwError $ ServerError NotFound "Bob not there"
    _ -> pure $ HelloReplyMessage ("hi, " <> nm)

To run this example:

stack exec error-parsing

Output:

The result for saying hello to Alice: "GRpcOk \"hi, Alice\""
The result for saying hello to Bob: GRpcErrorString "not enough bytes"

As can be seen, any error thrown will be falsely interpreted by grpc-client as
"GRpcErrorString". This makes any real-life application with grpcclient
impossible (there are not just happy paths, and not all errors are the
same)

The server side implementation appears correct, as I can poke the server
with the grpcurl tool:

grpcurl -d '{"name": "Bobby"}' -format json -plaintext -proto errorparsing.proto localhost:8070 hello.Service.SayHello
{
  "reply": "hi, Bobby"
}

grpcurl -d '{"name": "Bob"}' -format json -plaintext -proto errorparsing.proto localhost:8070 hello.Service.SayHello
ERROR:
  Code: NotFound
  Message: Bob not there

Related issue: haskell-grpc-native/http2-grpc-haskell#6

Why am I filing this PR here?

Providing easy-to run example code along a github issue is hard, therefore I opted to use a PR instead.

More generally, I would have preferred to make this example a unit test (expecting the reply from Bob to be of a certain type which is not GRpcErrorString "not enough bytes"), however this whole project seems to not contain any unit tests at all.

2. Question about unit tests and production-readyness

Do you plan to set up some tasty or hspec (or something else) unit tests more generally? (why are there no unit tests so far? That makes me doubt whether the mu family of libraries can safely be used in any project other than toy projects. Can you comment on the stability of the libraries outside this particular bug?)

In case unit tests are more generally set up, this PR could be converted to be another unit test, which would allow to more easily catch other parsing issues in the mu family of libraries.