I'm missing some important parts that I would argue any (binary)format needs.

  - a magic header
  - a version number
  - a crc or data corruption check

Additionally I would argue one would be better off writing a custom text parser instead of parsing a binary format in this case, this is similar to the debate about unixlike config vs windows regedit. I would prefer something other then json but sill readable as txt.

Even the json example provided at the bottom can be minified from 418 characters to 166 by replacing the field names with single characters and removing the spaces. Almost all of the savings in this format come from not including the the field names and having a position dependent layout. You can choose to have delimiters or arrange for a byte to indicate the type(+size) for example the following string encodes the example data almost (85bytes vs 80bytes) as efficient but is still readable and supports utf-8 interpretation.

  123456789;LeroyJenkins;60;alliance;p,100,200,300;i,999,1,1;i,45,100,0;a,s,120;a,a,45;

You may optimize it further by allowing recurring entries and allowing assumed values from a defined default and only sending delta's can be dependent on the type of data you are expecting.

  { "itemId": 999, "quantity": 1, "isSoulbound": false }
  i,999,1,0

could become:

  default = { "itemId": 0, "quantity": 1, "isSoulbound": false }
  { "itemId": 999}
  i,999

I also want to be a real developer so i just GZIP the JSON to make it binary.

Schemas save you space, but then you lose the ability to understand the data without the schema. That's where JSON has always been handy despite its inefficiencies.

Of course you can get the same kind of thing in binary. I wrote a drop-in binary JSON replacement because it's easy to write a binary one that's twice as fast as simjson and yyjson. The important thing is to never give the drop-in replacement any extras that break roundtrip compatibility.

Was a fun little project to write, and quite useful for me: https://github.com/kstenerud/bonjson

Cool. And the next enlightening step would be to generate custom parser code from message schemas. So that when de-serializer reads a next field, it does not read what name and type it will be, it expects and fails if it's not. Safer and more compiler friendly.

One of my favourite yak shaving adventures in a previous job was writing a parse in place UBJSON decoder for ~1MB of data on a device with about as much free memory. Fast (enough) access by key, binary search with a few shortcuts for the mostly numeric payload. It built an index of maybe 30 bytes to speed things along, but also to let it keep working on the tail of the old message whilst the new one was overwriting it's start.

Should the system design have required all this of a device with 4MB of memory? Probably not. But it worked, and I had a great time