Hello everybody,

So lets say you want to make a game in Go (http://golang.org), but because of the lack of preexisting solutions and because you think it is not what you want to spend your time on, you decide not to do the user interface (that is graphics, sound, input handling) in go, but you would want to use Unity.

Now we have one process running the logic, and another process running the 'viewer'. My idea is to do the IPC over a local TCP-socket since it seems the easiest to implement(?). I would then probably send delta updates of the information on entities which have to be drawn from the logic to the viewer, and input updates form the viewer to the logic.

My main questions are: Has someone already done something like this? And do you think it is a good idea?

I guess the lag from the IPC should be not noticeable, and the amount of data that needs to be transfered is not that high.

BR

Inf

This is how every modern chess and go (the board game, not the language) program works: There is a GUI process that launches the engine and communicates with it via a bidirectional pipe. For chess, the communication happens using one of two protocols: XBoard's protocol or UCI. Go uses GTP. This allows engine programmers to focus on making strong engines, it allows users to switch engines and it allows for much easier testing of the engines.

What kind of game do you have in mind?

The first thing i would want to do is implement a simple go version of the top down shooter demo that comes with unity and then take it from there. So the first thing to do is to think about a protocol to send the position/transformation, animation and input data. Should maybe also include fire and forget events from the game logic like visual and sound effects (explosions etc.).

I would use pipes because they are just as easy to use as TCP sockets (maybe even easier?) and because they are most likely faster. Whether or not the delay matters depends a lot on the game of course, but I guess for a shooter game which is kind of "twitch" by nature, it might just matter.

A pipe is usually something like a kernel handle attached to a small piece of shared memory managed by the kernel. Or, something similar. Writing to and reading from a pipe is something like a memcpy (or on some operating systems, under some conditions, incrementing a page table reference counter and adding a page to a working set).

A TCP socket must at the very least do a memcpy and calculate two checksums, consider a MTU (even if the MTU is the maximum allowable by IP on a local socket, TCP must still honor it and packetize as appropriate) and so some address/port mapping stuff, and possibly go through firewall rules, so conceptually it must be slower.

Usually you will just give a crap, because nobody notices anyway, but when you really need "twitch", it may be worthwile to use something different. A pipe doesn't really have any disadvantages over a socket, other than you can't run the frontend and the logic on different machines, but would one really want to do that anyway?

Memory mapping is yet another thing you could consider. It's supported on pretty much every OS and as fast as you can get (though proper synchronization can be tricky).