Rick Richardson

I, too, am interested in being able to make updates (new copies) of the hamt while other threads are reading the data structure. Looking at the implementation, it doesn't appear...

Here is a basic repro/test-project : When I get a sec, I can try to instrument the Deserialization functions for `Endpoints` figure out where it's failing. https://github.com/rrichardson/bincode-k8s-repro/blob/main/src/main.rs The error is...

I thought it was just a specific Endpoint example, but now I think it's everything in bincode.

Looking at the implementation of Deserialize, and looking at the bincode docs. I am willing to say that the two implementations are at odds with one another. serialize/deserialize_struct uses a...

Hmm.. in looking for a compact alternative to bincode, I tried MessagePack, and got exactly the same result :( MessagePack is self-describing, so I figured it'd support `map` and `deserialize_identifier`.

That makes sense. It is my understanding that MessagePack does persist field names. So maybe the rmp implementation should be updated to support the map approach. In either case, it...

It looks like https://github.com/serde-rs/serde/issues/2012 would give a path forward, that is actually pretty elegant. `skip_field` and `skip_serializing*` just becomes a suggestion, and it lets the encoding implementation decide whether or...

> However even that won't help you with either bincode or rmp currently, because both of those don't override the default `skip_field` and thus ignore it. Perhaps you can ask...

I think that makes sense. I think that this is a fairly pervasive problem in the serde ecosystem, I can test it with a modified bincode, and, if it works...

That seems to be the case. If I were on Linux I could provide some more in-depth analysis.. (strace, etc) I'm on a mac, so I'm researching how to figure...