Skip to content

Examples

These examples use the generated benchmark.v2 types. Generate them with pixi run generate if tests/generated/ is missing. The schemas and oracle bytes behind those types are listed on Test data.

Encode and decode a Message

examples/encode_message.mojo:

from benchmark.v2 import Message

def main() raises:
    var msg = Message()
    msg.f_bool = True
    msg.f_int32 = 150
    msg.f_string = "hi"
    var buf = msg.encode()
    print("encoded", len(buf), "bytes")
    var again = Message.decode(buf)
    print("round-trip", again.f_int32, again.f_string)

pixi run mojo run -I src -I tests/generated examples/encode_message.mojo

A fuller value (f_bool=true, f_int32=150, f_int64=1, f_float64=1.5, f_string="hi") matches official protoc --encode byte for byte:

08 01 10 96 01 18 01 21 00 00 00 00 00 00 f8 3f 2a 02 68 69
Bytes Meaning
08 01 field 1, varint, true
10 96 01 field 2, varint, 150
18 01 field 3, varint, 1
21 + 8 bytes field 4, I64, IEEE-754 1.5
2a 02 68 69 field 5, LEN 2, hi

Zero, false, and empty strings are omitted. That is proto3 implicit presence.

Empty present submessage

A set-but-empty nested message is not omitted. Document.meta is field 3.

from benchmark.v2 import Document, DocumentMeta

def main() raises:
    var doc = Document()
    doc.id = "x"
    doc.status = 1
    doc.meta = DocumentMeta()
    var buf = doc.encode()
    # 0a 01 78   id = "x"
    # 10 01      status = 1
    # 1a 00      meta present, length 0

Official Python emits the same 1a 00. A second LEN record for field 3 merges into the existing DocumentMeta; it does not replace it.

Packed repeated double

Telemetry.values encodes as one packed LEN (field 4, tag 22).

from benchmark.v2 import Telemetry

def main() raises:
    var tel = Telemetry()
    tel.source = "s"
    tel.ts = 9
    tel.values.append(1.0)
    tel.values.append(2.0)
    var buf = tel.encode()
    var got = Telemetry.decode(buf)

The decoder also accepts unpacked I64 records for the same field. Both forms are legal proto3. Generated encode always writes packed.

Interop with protoc

printf 'f_bool: true f_int32: 150 f_string: "hi"\n' \
  | python3 tests_interop/encode_ref.py benchmark.v2.Message \
  > message.bin

python3 tests_interop/decode_ref.py benchmark.v2.Message < message.bin

Those scripts call protoc --encode and protoc --decode.

No length prefix

The pipe is raw protobuf bytes. There is no length prefix. The official conformance runner uses a 4-byte little-endian length prefix; do not mix the two.