That's a good point, it seems the Java code is checking if the entire string is valid UTF-8 while the Go version isn't. I wonder why the behavior of generated protobufs is different between the two.