A partial implementation of Protocol Buffers in Idris

Protocol buffers are a serializable format for structured data that is ubiquitous at Google. Idris is a language where types are first class values that is useful for theorem proving and metaprogramming. This package contains a partial implementation of protocol buffers in the Idris language. The goal is to demonstrate how Idris is capable of advanced metaprogramming that simplifies the implementation of protocol buffers.

This is not an official Google product. This project is a personal 20% project.

Protocol buffers are a way to describe a schema for serializable data. The schema is called a protocol buffer “message”. A message describes the kind of data that is to be serialized, and comprises a set of fields. Each field can be a primitive type, an enum, or another message. In this way protocol buffers allow the description of arbitrary data types. Below is shown a protocol message describing a phone number, including an enum definition for the type of a phone number (mobile, home or work).

enum PhoneType {
  MOBILE = 0;
  HOME = 1;
  WORK = 2;
}
message PhoneNumber {
  required string number = 1;
  optional PhoneType type = 2 [default = HOME];
}
message Person {
  required string name = 1;
  required int32 id = 2;
  optional string email = 3;
  repeated PhoneNumber phone = 4;
}

In most languages, protocol buffers are turned into native data types by code generation. E.g. in C++, the generated code for the Person class will be a C++ class with methods like set_name, name and more complex method to set nested messages like phone. The code is generated by a compiler that takes protocol message descriptions like the one above, parses them and emits generated C++ code.

However in Idris, we can avoid generating code by considering a type of message descriptors (so the Person message description would be an value with this type). This is the MessageDescriptor class. In the protocol buffers documentation this is just called a Descriptor but here it’s useful to be more consistent in naming. The unique property of Idris is that we can define an inductive data type InterpMessage which has MessageDescriptor for its index space, so that InterpMessage Person is the type of messages of type Person. Inductive data types are similar to C++ templates but with the advantage their arguments can be any value and that value need not even be known at compile time.

In Idris we can write generic, type safe, code for working with protocol buffers. In particular we have code for serialization and deserialization to and from text form. Serialization looks like

printToTextFormat : InterpMessage d -> String

Note that d is an implicit argument. If we were to expand the above declaration to include the type of d (which Idris is able to infer), it would be

printToTextFormat : {d : MessageDescriptor} -> InterpMessage d -> String

Note that this function is generic in that d ranges over the space of message descriptors, and for each d, the type of the second argument is a message of type InterpMessage d, that is a protocol message in the format given by the descriptor d. Deserialization can also be done generically.

About this Repo

The main purpose of the repo is to demonstrate and explore some interesting applications of Idris. Most of the interesting applications are in the Test directory which contains an example descriptor Person along with some examples of creating an instance of InterpMessage Person and some methods to inspect properties.

The Person descriptor is given by

PhoneType : EnumDescriptor
PhoneType = MkEnumDescriptor [
  MkEnumValueDescriptor "MOBILE" 0,
  MkEnumValueDescriptor "HOME" 1,
  MkEnumValueDescriptor "WORK" 5
]

PhoneNumber : MessageDescriptor
PhoneNumber = MkMessageDescriptor [
  MkFieldDescriptor Required PBString "number",
  MkFieldDescriptor Optional (PBEnum PhoneType) "type"
]

Person : MessageDescriptor
Person = MkMessageDescriptor [
  MkFieldDescriptor Required PBString "name",
  MkFieldDescriptor Required PBInt32 "id",
  MkFieldDescriptor Optional PBString "email",
  MkFieldDescriptor Repeated (PBMessage PhoneNumber) "phone"
]

A example of an element of InterpMessage Person is

John : InterpMessage Person
John = MkMessage [
  "John Doe",
  1234,
  Just "jdoe@example.com",
  [
    MkMessage ["123-456-7890", Just 1],
    MkMessage ["987-654-3210", Nothing]
  ]
]

Note that the MkMessage constructor accepts a heterogeneous list. This is not an HVect but a data type we construct in Protobuf.idr whose constructors are Nil and (::), but whose elements have types corresponding to the types of the fields given by the message descriptor.

Installing

Install the Lightyear package.

This can be done by cloning the repo by running this command from your home directory:

git clone https://github.com/ziman/lightyear

and installing it with

cd lightyear
idris --install lightyear.ipkg

Clone this repo and run the tests.

Clone this repo by running the following from your home directory

git clone https://github.com/google/idris-protobuf

The Test directory contains interesting examples that you might want to modify and re-run. To run the tests run this command from this repo

cd idris-protobuf
idris --testpkg protobuf.ipkg

Experiment in the REPL.

In order to load the repl with this package, first install it with the command

idris --install protobuf.ipkg

then load the Idris REPL with this package along with the test utils, with the command

idris -p lightyear Protobuf.idr Test/Utils.idr

While in the REPL you can explore the package, e.g.

*Test/Utils *Protobuf> printToTextFormat Test.Utils.John
"name: \"John Doe\"\nid: 1234\nemail: \"jdoe@example.com\"\nphone: {\n  number: \"123-456-7890\"\n  type: HOME\n}\nphone: {\n  number: \"987-654-3210\"\n}\n" : String
*Test/Utils *Protobuf> parseFromTextFormat {d=Test.Utils.Person} "name: \"Kester Tong\" id: 1234"
Right (MkMessage ["Kester Tong", 1234, Nothing, []]) : Either String