consider implementing a compiler in js again

[rom1504]

support .json , only objective is performance improvement

[Saiv46]

Bump

The only problem is re-implementing read/write functions and optimization of produced code (can be solved with Closure Compiler, but that's huge overhead).

[rom1504]

Reimplementing what ?
The point of this issue is to implement a json to js compiler as if the code was manually written.
That is optimal

[rom1504]

And not sure what you mean by "bump".
Do you want to work on it ?

[Saiv46]

I'm already working on that, trying to solve multiple issues at once (#87, #69, #63, #49, #28).
Messages like "bump" are used to bump threads up and make it more visible.

I'm re-implementing interpreter as type classes, to compile read/write/size functions to AST later.

[rom1504]

putting some background in this issue for anyone falling on it :

• node-protodef is an interpreter
• js is not amazing at polymorphic operation due to absence of types
• writing the serialization/deserialization manually results in 100x faster performance (I tried this)
• simple consequence => compilation would be much faster.
  Ways to do this :
• completely new protocol definition language with compilation to various language https://github.com/ProtoDef-io/protodefc . Issue: need completely new integration in js and rewrite of all protocol definition. Was started at https://github.com/ProtoDef-io/node-protodefc but not quite there + no benchmark was done yet
• just write a current json format to manual-like compile in js : would make integration easy and would solve the performance issue

this is this last point that this issue is about.
I believe this could make all users of protodef much faster (including node-minecraft-protocol)

[Karang]

putting some background in this issue for anyone falling on it :

* node-protodef is an interpreter

* js is not amazing at polymorphic operation due to absence of types

* writing the serialization/deserialization manually results in 100x faster performance (I tried this)

* simple consequence => compilation would be much faster.
  Ways to do this :

* completely new protocol definition language with compilation to various language https://github.com/ProtoDef-io/protodefc . Issue: need completely new integration in js and rewrite of all protocol definition. Was started at https://github.com/ProtoDef-io/node-protodefc but not quite there + no benchmark was done yet

* just write a current json format to manual-like compile in js : would make integration easy and would solve the performance issue

this is this last point that this issue is about.
I believe this could make all users of protodef much faster (including node-minecraft-protocol)

This is interesting. What you suggest is to make a node module that could take a protocol.json with the current specification and generate a .js file that contains an optimized ser/deser code ?

You said you tried writting it manually, is there an example (code) of what you did by hand ?

Also, what benchmark do you have to test the performance of the compiled vs interpreted protodef ?

[rom1504]

yeah. That js file could then either be saved and committed or simply be evaluated using eval

#65 has some examples but basically the idea of writing manually is just to write naive code to read buffers using https://nodejs.org/api/buffer.html
what node-protodef implements is an interpreter on top of buffer API

for benchmarks, see https://github.com/ProtoDef-io/node-protodef/pull/65/files#diff-95f0a310f289230d56c3a4949c17963eR85

see also all of this https://github.com/ProtoDef-io/node-protodef/tree/master/benchmark

I did not implement manually everything (that would be crazy long !), but on simple examples it's much faster to use manual code.

[rom1504]

now done