I've written several screeds in the comments here on HN about protobufs being terrible over the past few years. Basically the creators of PB ignored ASN.1 and built a bad version of mid-1980s ASN.1 and DER.

Tag-length-value (TLV) encodings are just overly verbose for no good reason. They are _NOT_ "self-describing", and one does not need everything tagged to support extensibility. Even where one does need tags, tag assignments can be fully automatic and need not be exposed to the module designer. Anyone with a modicum of time spent researching how ASN.1 handles extensibility with non-TLV encoding rules knows these things. The entire arc of ASN.1's evolution over two plus decades was all about extensibility and non-TLV encoding rules!

And yes, ASN.1 started with the same premise as PB, but 40 years ago. Thus it's terribly egregious that PB's designers did not learn any lessons at all from ASN.1!

Near as I can tell PB's designers thought they knew about encodings, but didn't, and near as I can tell they refused to look at ASN.1 and such because of the lack of tooling for ASN.1, but of course there was even less tooling for PB since it hadn't existed.

It's all exasperating.