Joe Tsai

At first glance, the generated `.pb.go` looks fine. This suggests that there's a bug in the plumbing of the "UseSizedCache" flag when traversing between generated messages from older version of...

That's great to hear! > (it's actually a bit faster) I'm curious, by how much?

When wrapping a legacy message (i.e., one that doesn't natively support the new `proto.Message` interface), we dynamically create a wrapper for the legacy type. However, the [wrapper type doesn't add...

Changing this issue to be about documentation. The tutorial needs to be updated.

That means that the version of the protobuf toolchain is too old. I'm not sure when the "_opt" feature was introduced, but you can find the latest releases here: https://github.com/protocolbuffers/protobuf/releases...

The devsite tutorial probably still needs to be updated. I'll look into it early next week.

Depending on what we plan to do here, it may or may not be impacted heavily by `vgo`.

Let's keep this issue about migrating the **generated API**. I'll file a separate issue what to do with Go import paths.

The generated messages follow the order specified in the `.proto` files. I don't think the Go generator should re-order them. If there is a benefit to doing so, I argue...

Thanks for the suggestion. Swapping `unknownFields` and `sizeCache` seems valuable since the latter is always 4B wide. Moving them to the end may or may not help depending on what...