purescript-protobuf

PureScript library and code generator for Google Protocol Buffers version 3.

This library operates on ArrayBuffer, so it will run both in Node.js and in browser environments.

We aim to support binary-encoded protobuf for syntax = "proto3"; descriptor files.

Many syntax = "proto2"; descriptor files will also work, as long as they don't use "proto2" features, especially groups, which we do not support. We also do not support "proto2" extensions.

We do not support services.

We do not support JSON encoding.

In this version, we pass all 651 of the Google conformance tests of binary-wire-format proto3 for Protocol Buffers v21.10. See the conformance/README.md in this repository for details.

We also have our own unit tests, see test/README.md in this repository.

The nix develop environment provides

• The PureScript toolchain: purs, spago, and node.
• The protoc compiler.
• The protoc-gen-purescript executable plugin for protoc on the PATH so that protoc can find it.

$ nix develop

PureScript Protobuf development environment.
libprotoc 3.21.10
purs 0.15.6
node v17.1.0

To build the protoc compiler plugin, run:

    spago -x spago-plugin.dhall build

To compile PureScript .purs files from .proto files, run for example:

    protoc --purescript_out=. google/protobuf/timestamp.proto

We can test out code generation immediately by generating .purs files for any of Google’s built-in “well-known types” in the google.protobuf package namespace. Try the command protoc --purescript_out=. google/protobuf/any.proto or protoc --purescript_out=. google/protobuf/timestamp.proto.

To see all of the .proto definitions included with the Nix PureScript Protobuf installation including the “well-known types,” ls $(nix path-info .#protobuf)/src/google/protobuf/*.proto

If you don't want to use Nix, then install the PureScript toolchain and protoc, and add the executable script bin/protoc-gen-purescript to your PATH.

The code generator will use the package import statement in the .proto file and the base .proto file name as the PureScript module name for that file.

A message in a shapes.proto descriptor file declared as

syntax = "proto3";
package interproc;

message Rectangle {
  double width = 1;
  double height = 2;
}

will export these four names from module Interproc.Shapes in a generated shapes.Interproc.purs file.

1. A message data type.

   newtype Rectangle = Rectangle { width :: Maybe Number, height :: Maybe Number }

   The message data type will also include an __unknown_fields array field for holding received fields which were not in the descriptor .proto file. We can ignore __unknown_fields if we want to.

2. A message maker which constructs a message from a Record with some message fields.

   mkRectangle :: forall r. Record r -> Rectangle

   All message fields are optional, and can be omitted when making a message. There are some extra type constraints, not shown here, which will cause a compiler error if we try to add a field which is not in the message data type.

   If we want the compiler to check that we’ve explicitly supplied all the fields, then we can use the ordinary message data type constructor Rectangle.

3. A message serializer which works with arraybuffer-builder.

   putRectangle :: forall m. MonadEffect m => Rectangle -> PutM m Unit

4. A message deserializer which works with parsing-dataview.

   parseRectangle :: forall m. MonadEffect m => ByteLength -> ParserT DataView m Rectangle

   The message parser needs an argument which tells it the length of the message which it’s about to parse, because “the Protocol Buffer wire format is not self-delimiting.”

In our program, our imports will look something like this. The only module from this package which we will import into our program will be the Protobuf.Library module. We'll import the message modules from the generated .purs files. We'll also import modules for reading and writing ArrayBuffers.

import Protobuf.Library (Bytes(..))
import Interproc.Shapes (Rectangle, mkRectangle, putRectangle, parseRectangle)
import Parsing (runParserT, ParseError, liftMaybe)
import Data.ArrayBuffer.Builder (execPutM)
import Data.ArrayBuffer.DataView (whole)
import Data.ArrayBuffer.ArrayBuffer (byteLength)
import Data.Tuple (Tuple)
import Data.Newtype (unwrap)

This is how we serialize a Rectangle to an ArrayBuffer. We must be in a MonadEffect.

do
    arraybuffer <- execPutM $ putRectangle $ mkRectangle
        { width: Just 3.0
        , height: Just 4.0
        }

Next we’ll deserialize Rectangle from the ArrayBuffer that we just made.

    result :: Either ParseError (Tuple Number Number)
      <- runParserT (whole arraybuffer) $ do

        rectangle :: Rectangle <- parseRectangle (byteLength arraybuffer)

At this point we’ve consumed all of the parser input and constructed our Rectangle message, but we’re not finished parsing. We want to “validate” the Rectangle message to make sure it has all of the fields that we require, because in proto3, all fields are optional.

Fortunately we are already in the ParserT monad, so we can do better than to “validate”: Parse, don't validate.

We will construct a Tuple Number Number with the width and height of the Rectangle. If the width or height are missing from the Rectangle message, then we will fail in the ParserT monad.

For this validation step, pattern matching on the Rectangle message type works well, so we could validate this way:

        case rectangle of
            Rectangle { width: Just width, height: Just height } ->
                pure $ Tuple width height
            _ -> fail "Missing required width or height"

Or we might want to use liftMaybe for more fine-grained validation:

        width <- liftMaybe "Missing required width" (unwrap rectangle).width
        height <- liftMaybe "Missing required height" (unwrap rectangle).height
        pure $ Tuple width height

And now the result is either a parsing error or a fully validated rectangle.

The generated code modules will import modules from this package.

The generated code depends on packages which are all in package-sets.

If the runtime environment is Node.js, then it must be at least v11, because that is the version in which TextDecoder and TextEncoder were added to Globals.

All of the generated message types have instances of Eq, Show, Generic, NewType.

The protobuf repository contains three executable Node.js programs which use code generated by protobuf. Refer to these for further examples of how to use the generated code.

1. The protoc compiler plugin. The code generator imports generated code. Trippy, right? This program literally writes itself.
2. The unit test suite
3. The Google conformance test program

The Protobuf Decoder Explainer shows an example of how to use this library to parse binary Protobuf when we don’t have access to the .proto descriptor schema file and can’t generate message-reading code.

This is how field presence works in our implementation.

A message field will always be Just when the field is present on the wire.

A message field will always be Nothing when the field is not present on the wire, even if it’s a no presence field. If we want to interpret a missing no presence field as a default value then we have the Protobuf.Library.toDefault function for that.

A no presence field will not be serialized on the wire when it is Nothing, or Just the default value.

An explicit presence (optional) field will not be serialized on the wire when it is Nothing. It will be serialized when it is Just the default value.

When the parser encounters an invalid encoding in the Protobuf input stream then it will fail to parse.

When ParserT fails it will return a ParseError String (Position {index::Int,line::Int,column::Int}).

The byte offset at which the parse failure occurred is given by the index.

The path to the Protobuf definition which failed to parse will be included in the ParseError String and delimited by '/', something like "Message1 / string_field_1 / Invalid UTF8 encoding.".

The Protobuf import statement allows Protobuf messages to have fields consisting of Protobuf messages imported from another file, and qualified by the package name in that file. In order to generate the correct PureScript module name qualifier on the types of imported message fields, the code generator must be able to lookup the package name statement in the imported file.

For that reason, we can only use top-level (not nested) message and enum types from a Protobuf import.

The generated PureScript code will usually have module imports which cause the purs compiler to emit redundant import warnings. Sorry. If this causes trouble then the imports can be fixed automatically in a precompiling pass with the command-line tool suggest.

The flake.nix provides a package protoc-gen-purescript so that we can run the .proto → .purs generation step as part of a Nix derivation. Include protoc-gen-purescript and protobuf as nativeBuildInputs. Then protoc --purescript_out=path_to_output file.proto will be runnable in our derivation phases.

(protoc-gen-purescript requires an impure build for the spago2nix step, so you’ll have to grant trust at the prompt or configure a relaxed Nix sandbox.)

The flake.nix provides the Google Protocol Buffers conformance tests as an app. To run the conformance tests right now without installing or cloning anything,

nix run github:xc-jp/purescript-protobuf#conformance

Pull requests welcome.

• justifill package may be useful for message construction.
• morello package may be useful for message validation.
• Third-Party Add-ons for Protocol Buffers Google’s list of Protocol Buffers language implementations.
• A vision for data interchange in Elm Comparison of JSON, ProtoBuf, GraphQL by Evan Czaplicki.