Protobuf in R: Streaming multiple messages


Protocol buffers (protobuf) are a method developed by Google (1) for serializing structured data that is both language and platform neutral.

The structure of the information is defined using a protocol buffer message type in a .proto file. The protobuf program then generates source code from this description for generating or parsing stream of bytes that represent the structured data. Unfortunately, R is not one of the languages supported.

Instead, the RProtobuf (3) R package provides a convenient wrapper around the C++ version of the protobuf interface which can be used to read and write protobuf messages.

The RProtobuf package provide functions for writing a single message to a binary file. However, it does not provide any functionality to write multiple messages (in fact only the java version of protobuf provides this functionality). The standard way to including multiple protobuf messages in one binary file is to precede the message by its byte size in varint format (Variable Length Integer) (6). This is how functions parseDelimitedFrom and writeDelimitedTo in the java implementation work. A python implementation of this procedure is given at (2).

I was unable to find any R implementations of the varint encoders and decoders required to convert the message size to the right format so I ended up simplifying and rewriting the python versions available at (5) and (4) .


First we need a .proto file with the definition of the structure of the messages. For example, in the following code block I show the example in the protobuf tutorial (1). This code is provided in the ./Code/example.proto in the repository 7.

syntax = "proto2";

package example;

message Person {
  required string name = 1;
  required int32 id = 2;
  optional string email = 3;

  enum PhoneType {
    MOBILE = 0;
    HOME = 1;
    WORK = 2;

  message PhoneNumber {
    required string number = 1;
    optional PhoneType type = 2 [default = HOME];

  repeated PhoneNumber phones = 4;

message AddressBook {
  repeated Person people = 1;

Load Protobuf Definition

In R we first need to load the .proto file using the readProtoFiles function.


Then we can create two messages that we would like to write to a binary file.

p1 = new(example.Person, id = 1, name = 'Dirk')
p1$email = ""
p2 = new(example.Person, id = 2)
p2$name = "Mary"
name: "Dirk"
id: 1
email: ""
name: "Mary"
id: 2

Write Protobuf To binary file

The function VarintBytes32 given in the ./Code/ folder is used to encode the message length using the varint format. The length is written to the binary file with the funciton writeBin while the actual message is written using the serialize function.

fname = "./Code/binproto.protobuf"
fileo = file(fname, open = "wb", raw = TRUE)
mysize = VarintBytes32(p1$bytesize())
writeBin(mysize, fileo)

mysize2 = VarintBytes32(p2$bytesize())
writeBin(mysize2, fileo)

There should now be a file called binproto.protobuf on the Code folder with the binary content.

system("ls -l ./Code")
total 24
-rw-r--r--  1 fred  staff   33 Apr  8 20:38 binproto.protobuf
-rw-r--r--  1 fred  staff  409 Apr  8 20:29 example.proto
-rw-r--r--  1 fred  staff  911 Apr  8 20:13 varint.R
system("cat ./Code/binproto.protobuf", intern=TRUE)
[1] "\027"
[2] "\004Dirk\020\001\032\\b"
[3] "\004Mary\020\002"

Read Protobuf file

In order to read the binary protobuf file it is necessary to first read the binary file to memory.

fileloc = "Code/binproto.protobuf"
alldata = readBin(fileloc, "raw", file.size(fileloc))
 [1] 17 0a 04 44 69 72 6b 10 01 1a 0d 74 65 73 74 40 74 65 73 74 2e 63 6f 6d 08
[26] 0a 04 4d 61 72 79 10 02

Then the function DecodeVarint32 is used to decode the length of the message as well as the start position. With this information it is possible to parse the protobuf using the read function. This is repeated for every message.

n = 1
pos = DecodeVarint32(alldata,n)
clen = pos[[1]]
n = pos[[2]]
nend = n + clen - 1
tmp = alldata[n:nend]
p1_read = example.Person$read(tmp)
n = clen + n
pos = DecodeVarint32(alldata,n)
clen = pos[[1]]
n = pos[[2]]
nend = n + clen - 1
tmp = alldata[n:nend]
p2_read = example.Person$read(tmp)


name: "Dirk"
id: 1
email: ""

name: "Mary"
id: 2


Fred Gruber
Fred Gruber
Senior Principal Scientist

My research interests include causal inference, Bayesian networks, causal discovery, machine learning.