26% slower decoding than Python

[elbaro]

syntax = "proto3";

package App;

message User {
  string name = 1;
  string profile_url = 2;
  repeated int64 follower_ids = 3;
}

message Tweet {
  User author = 4;
  string text = 1;
  int64 created_timestamp = 2;
  repeated string mentions = 5;
}

serialized:

data = b'\n\x11Hi this is a text\x10\xfb\x99K"\x1b\n\x03Who\x12\x14https://example.com/*\x06@trump*\x06@obama'
p = Tweet()

timeit.timeit(
        lambda: (p.Clear(), p.ParseFromString(data)),
        number=int(1e6),
)

Rust (*self = Default::default() + merge) is 26% slower than Python (Clear + ParseFromString). Also Python (upb) is checking for unknown fields while prost does not. Wonder if there is a critical optimization in upb that is missing in prost.

# decoding (secs)
2021                             2023
Python:  14.008996680990094      Python:  0.3742136040236801
Rust  :  0.5708326659951126      Rust  :  0.47216512804152444

pip protobuf 4.23.2
prost 0.11.9

[LucioFranco]

Are you running in release mode?

[elbaro]

Yes. The debug build will be much slower than this.