MSC4224: CBOR Serialization #4224
Conversation
turt2live
added
proposal
There was a problem hiding this comment.
Implementation requirements:
- Client
- Server
turt2live
added
client-server
There was a problem hiding this comment.
As a very small stepping stone in the right direction, I think this MSC does a good job. I would say that the lack of enum integer keys makes this kinda useless to do, as the gain is so small. However, the un-base64ing of binary values results in quite a significant improvement for encrypted messages, and that does make enough of a difference imo to encourage servers to support this MSC.
| ## Potential issues | ||
|
|
||
| At the time of writing, this proposal doesn't provide a integer mapping for | ||
| string keys and static string values, which leaves out encoding efficiency. |
There was a problem hiding this comment.
I suspect this is really going to hurt the numbers on this, as short messages are overwhelmingly made up by key names.
| at the time of writing is too complex to implement and have not yet been | ||
| accepted into the spec, this proposal focuses on CBOR serialization part of it. | ||
|
|
||
| ## Proposal |
There was a problem hiding this comment.
It would be nice to have some example messages to compare JSON vs proposed CBOR sizes and the % reduction. This proposal would have a large benefit on encrypted rooms due to the base64 ciphertexts being un-base64'd. It would have a very small effect on unencrypted data, due to the prolific use of strings as values which don't see any compression benefits.
Rendered
Yes, this is basically #3079 but only CBOR part of it, "one bite at a time" and such.
This MSC is written as an individual FOSS contributor. (Disclosure as by matrix-org/matrix-spec#1700)
Signed-off-by: Alexander Ivanov [email protected]