ppt - Protocol Platform Tester

I created this to test out how good avro, flexbuffers, cbor, msgpack, and others are at:

Compressing data
How close they come to passing WPT structuredClone tests
How fast they are (and if they make any tradeoff)
And finding out if they have any bugs
And what they support and what not
How they work differently

I haven't written any packer myself so this is an unopinionated test case that test all the packers I could find. I often find the performance results to be somewhat misleading as they only promote themselves, and they only test one specific version of their dependencies and only test in one engine and not the others. I've tried to make this test as fair as possible by using the same data for all packers, and using the same test harness for all packers.

The goal of this was to find a suitable import/export library for things you can store in IndexedDB.

(Click on each ✅ ❌ to see more details)

This is what some libraries are able to structural clone

Conclusions:

core-js serves as a good reference for what can be cloned and could be a good starting point for a library that want to make a spinoff of it.
it dose not however allow you to clone into a different realm, meaning you can't transfer data between a client and a server.
CoreJS also as an unfair performance advantage as it's clones are not serialized to/from binary.
cbor-x seems good but there are some caveats doe...
Use Uint8array instead of buffer 👍
msgpackr
Use Uint8array instead of buffer 👍
FlexBuffers is an version of flatbuffers that is designed to work without a schema, so the data is self describing. (type is stored in the data) but it was not able to clone much web classes. it very much designed to work cross different coding languages. So it's mostly limited to basic number, strings, arrays, and objects.
Use Uint8array instead of buffer 👍
avro It don't seem to be good for arbitrary data. Seems to be designed for schema.
It depends on NodeJS Buffer instead of using Uint8Array/DataView 👎
It depends on NodeJS streams + other subdependencies instead of using plain iterators 👎
bson is a binary format that MongoDB uses.
It depends on NodeJS Buffer instead of using Uint8Array/DataView 👎
ubjson Seems to be abandoned and not updated for a long time.
It depends on NodeJS Buffer instead of using Uint8Array/DataView 👎
structured-clone seems to have been abandoned and not updated for a long time. So it don't support newer types.
It depends on NodeJS Buffer instead of using Uint8Array/DataView 👎

To any implementer packing BigInts:
You should not consider it as 64bit value. think of it more like a arbitrary length Uint8array with any length instead, you need to keep track if it's signed or not more manually.
One solutions to this is reading 8 bits at a time, shift the number and then knowing the length of the buffer. (here is one hex solution)
Another solution is to use a varint where 1 bit in each bytes is a continuation