Unary length prefix is a solid technique. I would be careful with the performance claims, though. Sometimes what the machine can do is surprising and the perfor

Unary length prefix is a solid technique. I would be careful with the performance claims, though. Sometimes what the machine can do is surprising and the performance of a system that looks like it needs to loop and branch is faster than you expected. The protobuf project, unsurprisingly for a project of its age, has several different varint parsing strategies. One is described at https://github.com/protocolbuffers/protobuf/blob/main/src/google/protobuf/varint_shuffle.h#L59-L76 https://github.com/protocolbuffers/protobuf/blob/main/src/go... and another at https://github.com/protocolbuffers/protobuf/blob/main/src/google/protobuf/parse_context.h#L638 https://github.com/protocolbuffers/protobuf/blob/main/src/go...