Disable Server Protobuf Fixes

[RobertBColton]

This pull request attempts to address issues like #1723 which stem from #1292. Old protocol buffer's compiler can not even handle the RPC syntax. Since we aren't requiring GRPC yet in the default LGM/ENIGMA, we should not run protoc on the server proto.

What I've done here is readd my filter-out that fundies had removed in #1292. I also split the server RPC interface's compiler buffers out into a new compiler.proto file since emake does depend on that (just not the RPC stuff).

I tested locally myself to make clean with and without CLI_ENABLE_SERVER=TRUE defined. When it's not passed, server.proto is not passed to protoc at all, and no codegen is produced. However, emake does continue to work and compiler.proto is built. When server mode is enabled, then server.proto is passed to protoc, and the codegen is generated. I tested the emake server against the latest RGM release executable. Also, in both cases, LateralGM was able to communicate with compileEGMf. So this should be good to go once it passes.

[codecov]

Codecov Report

Merging #1724 into master will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master    #1724   +/-   ##
=======================================
  Coverage   24.31%   24.31%           
=======================================
  Files         166      166           
  Lines       16603    16603           
=======================================
  Hits         4037     4037           
  Misses      12566    12566

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 55fde1e...132278f. Read the comment docs.

[JoshDreamland]

One comment; make sure you have a plan.

[JoshDreamland]

Where is the request message used if not in the RPC? I don't think there's a reason to separate request/response messages from the service definition. What's the point of building the request if there's nowhere to send it?

[RobertBColton]

The RPC stubs just call out to the compileEGMf functions. Also it doesn't matter, you said to use the Protocol Buffer interfaces directly for new stuff. That means in order to compile the compiler you'd need GRPC, but we don't want that. Emake could be built without server support but still do command line syntax checking.

[JoshDreamland]

The new compile method doesn't accept a CompileRequest—it takes a Project proto and converts it to a GameData struct. To my knowledge, only the RPC service would accept a CompileRequest.

[RobertBColton]

I am sorry, I was talking about the new logging methods.

enigma-dev/CommandLine/emake/Server.cpp

Line 140 in 4d62bf0

SyntaxError GetSyntaxError(syntax_error* err) {

enigma-dev/CommandLine/emake/EnigmaCallbacks.cpp

Line 53 in 4d62bf0

void CallBack::ProcessOutput() {

You are correct. Regardless, ideally, we'd just add the wrapper accepting SyntaxCheckRequest in compileEGMf, then have the old version call the new one. So again, we want to use these buffers in the compiler now, but don't want to make the compiler depend on GRPC by default.

[JoshDreamland]

Okay, let's not get too carried away, here. Remember, you can't just pass the protos themselves across DLL bounds, anyway; you have to serialize them, first. I think we have emake just calling external methods accepting and returning protos all willy-nilly, and that works because emake and ENIGMA are built by the same compiler at the same time. But RGM and LGM cannot use that interface. They have to use an overload that accepts byte[] (or char*) and performs the serialization in the C++.

If that's where you want to use the request/response messages, then okay. I'd just as soon accept (string, SomeSyntaxCheckMetadataProto) (as bytes, not C++ types) and then use optional string and SomeSyntaxCheckMetadataProto as fields in the Request proto.

And before you ask, if you want to use threads with the non-RPC compiler API, you will want to use a ring buffer of string of serialized bytes, and a counting semaphore. Otherwise, just use a single static string.

[RobertBColton]

Ok, that all sounds about right and I had expected as much. I don't know why I don't think I actually considered the DLL bounds until you said that much though, but that would be extremely useful. The same should be done for libEGM, then people could redistribute it. Hell, they could make a GMS extension using libEGM to load GMKs. The more third party tools that use our stuff, the more developers we will have that want to help maintain our stuff.