Implement serde's is_human_readable to return false

[aszkid]

Currently, the Serializer and Deserializer pairs in rmp-serde do not implement is_human_readable. As a consequence of this, types that support a more compact representation serialize and deserialize inefficiently: for example, IPv4 addresses should take something like 4 bytes, but currently use an annoying 17 bytes, as they serialize as regular strings (my qualm stems from serializing IpNet structs).

Has this issue been discussed before? While I do think this is a reasonable request to make, it is nevertheless a breaking change (mentioned on the serde docs—previously serialized types that support a compact representation will fail at deserialization).

I am currently using a forked repo implementing this change; merging would be extremely easy from a technical point of view (an effective 2 lines of code addded), but there is a discussion to be had.

[daboross]

I'm in full support of this.

If msgpack-rust is against making breaking changes like this, maybe it could be done as an added option, like StructMapSerializer?

[Taywee]

I'm pretty interested in this. I'm hitting a couple snags where my msgpack format that is supposed to be compact is bigger than it should be, and I end up having to wrap in a newtype to get a compact representation where I want it instead of just being able to use a type with a human readable and non-human-readable context. Uuid is an example of this, which serializes in rmp-serde to the full 36-character hyphenated representation instead of the 16 byte array.

I could see a serializer config option working to avoid a breaking change, or a feature flag, but it would make sense at some point for this to be default behavior.

[kornelski]

Yes, this should definitely be behind a flag.

It's not as simple as adding fn is_human_readable to existing implementations, since the flag doesn't seem to be passed down reliably, but I was unable to find where exactly it was missing.

[joshtriplett]

#257 (together with the serde fix in serde-rs/serde#1888 ) should fix this.