Canonical encodings about go HOT 5 CLOSED

Taking a step back, it seems that you want support for sorted encoding of mappings. Is that a fair summary of your request?

from go.

Yep! That's the one thing I've hit in practice that was tricky for getting deterministic output.

I'd almost say that what I want is canonical cbor output, but with the interesting exception that I also need streams, which is technically a break from what the rfc suggests for canonical encoding.

I went ahead and made a prototype that does what I needed, so I'm going full steam ahead with using codec on that project (and it's fantastic! Thank you for all this awesome work!), but I did end up using GenHelperEncoder to do so despite the may-change-without-notice warnings. (We pin dependencies by commit hash in that project, so I'm not worried, and totally accept I might have to nontrivial refactors in the future. :) ) If you want to see the example of how that turned out, it's here and here.

It might also be nice to have a method for encoding maps that'll do keys in an order specified by a []string keys (well, or just expose an interface like the GenHelperEncoder officially where map appends can be called in the desired order directly) -- this would be useful in a situation where someone is implementating an encoding for something that has to match an existing spec that's deterministic order but not sorted.

from go.

K.

I have an upcoming change to support canonical encodings.

It works by configuring an option on the handle i.e. Canonical=true.

Canonical mode is ignored for code generation. It only works for the standard reflection-based runtime introspection. Also, there is a performance hit, as each key must be encoded out-of-band, then all key encodings compared, and then the key-values are encoded.

The good part is that it fits in with the rest of the library, so is supported.
Using "internal", "unsupported" APIs will prevent you from easy upgrade, so is highly discouraged.

The change should be uploaded today. Let me know if this will work for you before I make the update.

Thanks.

from go.

The performance hit caveat makes sense. Even with that, it'd definitely be great to have this in the supported APIs.

Do you think an explicit MapAppend(k, v interface{}) is also exposable? Maybe it's not the most general API, but I think it makes sense for most encoding formats I can think of, and it seems to make sense from the way the internals work right now (though I suppose what I wrote along those lines was cbor specific...). And raising some sort of "unsupported" error an encoding format doesn't support user-chosen map orders seems totally reasonable.

from go.

This is awesome! I look forward to using it in future protocols. :D

from go.