ProtoDump

Similar to protobuf, but much simpler and faster.

Payload is 20% bigger, but it's about 2x faster, so who cares.

Performance Tests

on M1 MacBook Pro

Looping 1 million times

91 byte object graph in my lib

79 byte identical object graph in protobuf

This is the graph:

message AddressProto {
    string Street=1;
    string Suburb=2;
    string City=3;
    string Code=4;
}

message PersonProto {
    string Name = 1;
    string Surname = 2;
    int32 Id = 3;
    int64 Birthdate = 4;
    string Address = 5;
}

message PersonProto_v2 {
    string Name = 1;
    string Surname = 2;
    int32 Id = 3;
    int64 Birthdate = 4;
    AddressProto Address = 6;
}

I used PersonProto_v2 for the performance testing.

Serialize (2.5x faster)

ProtoDump 232ms

ProtoBuf 512ms

Deserialize (1.5x faster)

ProtoDump 368ms

ProtoBuf 565ms

Conclusion

This thing isn't well optimized. In fact there are some weird parts that I want to make go away. Just a few tweaks should make this crazy fast.

Next project - wrap this in a function that I can DLLExport for use everywhere. Maybe rewrite in C - or Rust

Edit:

I changed the PersonProto_v2 to contain a list, so it's not the same as for the original test.
I did optimize 1 or 2 things - it originally used generics, but I found it was noticably quicker to do it the hard way.
So that is an interesting observation for future use - generics have a performance penalty.
I also added an inline directive to the low-level functions. This gave about a 10% performance boost.

I also added a C# generator - uses proto files, works kinda like protoc:

protodumpgen --file Person_v2.proto --output-folder ../SharedClasses/gen2 --namespace prototest --language C#