Minecraft

On the heels of our success with Quake II and Angry Birds, we had another opportunity to port a game to Chrome. This time it was Minecraft! I’m not sure if Chrome DevRel reached out to Mojang, or Mojang to them, but either way we were given a copy of the Minecraft Java source, and asked “think you could get this running in the browser?”. Challenge accepted.

This process ended up being fairly similar to the Quake II port. While the source was entirely Java, it was unsurprisingly written almost entirely to desktop APIs – direct filesystem access, raw sockets, old-school glBegin()/glEnd() GL code, and so forth. And as with Quake, “single player” mode involved simply running the server code in-process, with the network layer stubbed out.

Porting was a painstaking process of isolating all these systems, getting everything compiling, watching it fail in various ways, fixing things, and repeating ad infinitum. As before, Stefan did a lot of the heavy lifting on the graphics code, first abstracting the glBegin/End() calls to a layer that produced WebGL-friendly vertex and element buffers to get it running, then painstakingly optimizing the abstractions out.

The resource handling wasn’t as difficult as Quake, simply because Minecraft is almost entirely code, with very few file resources. But the in-process server turned out to be a beast to run in Javascript. We got it working, but almost immediately realized we needed to keep that part on the server for decent performance. Fortunately, the network code was fairly straightforward TCP, so it could be easily ported to websocket frames. Unlike Quake’s UDP nightmare with its own retry mechanisms!

The end result was pretty good. While doubtless less performant than the “native” Java implementation, it ran in Chrome at well over 60 fps with all the effects and a reasonably high draw distance. And it was pretty cool to be able to spin up a world on a server, and have everyone join by just sending them a link – no binary install required, no installing texture packs, etc.

Unfortunately, when it came time to pin down the details of how they might release a product, our contacts at Mojang seemed to lose interest. There were some (valid) concerns about maintenance, and some about running mods, which made less sense to me, as many Minecraft ports couldn’t run Java mods. We never found out much more, and the real issue may have simply been that they were too busy discussing the upcoming Microsoft acquisition.

But I’m still disappointed that we could never announce or release it publicly. The code is probably lying around bit-rotting somewhere at Google now, never to see the light of day. I feel comfortable discussing it at this point, as it’s far in the past now. But I wish we could have at least demonstrated it publicly.