I've been much more quiet than I want to be or promised to be, sorry for that. I have been very busy with under-the-cover programming and preparations for the (actually very complex) economy system that Citybound will have.
Most of this work requires deep thinking and thus doesn't lend itself well for livestreams and so far it hasn't produced any visible results, that's why I didn't write about it.
Today Michael, however, was so kind to post a really extensive and well-written report of what both of us have been programming on and thinking about behind the scenes. It's very technical, but also gives a very good impression of how Michael and me are working together. And: it even hints at some upcoming fundamental gameplay ideas for Citybound - so I only encourage you to read it!
See you soon,
by Michael Lucas-Smith
First off, internally I upgraded our 2D drawing library which is primarily used for debugging interesting geometric problems when they occur. Unlike the canvas2d that comes with HTML5, this drawing library knows what dots and line segments and rays are. It auto-scales the internal content, while keeping things like rays and text the right size. This has sped up my work in particular a lot.
It was around this time that Anselm started to plot out his ideas for the economy. After a bit of back and forth on the design, Anselm settled on creating an abstracted bigraph of the transportation network. The farther out you zoom the simpler the network becomes. On the last live stream there was a moment where a red line was visible - that was part of the abstract lower level of detail network. This will be used to map out economic information. It's going to be wicked when it's done.
Sometime in this period Anselm started using my straight skeleton code for part of the zoning. It worked out of the box for him, which I was very surprised about. I'm still working on this stuff but more in a later part of this status post. While this was going on I was finally getting to the degenerate test cases for straight skeletons, such as a
The deeper I got in to the degenerate test cases, the more I started to notice that perfectly parallel geometric structures did not have perfectly parallel coordinates. I realised that
gl-matrix boasts speed by way of its design. Internally it tries to use the best representation of vertices it can for maximum efficiency. In this case, it was using Float32Array.
I've implemented a few of these over the years and I didn't look forward to implementing determinants and inverts and matrix multiplication again - so I went for a meta approach. Instead of writing out the expanded versions of those functions by hand and getting it wrong and writing a billion tests to make sure I finally get it right, I made a code generator using static single assignment. Then I wrote code building code for vectors and matrices. The short of it is that on startup, we actually generate our vector and matrix classes and we can generate combined functions that keep variables unfolded and optimised... without having to write it 3 times for each different size of vector.
This new vector and matrix library boringly named
numerics kept everything in 64-bit and numerous bugs I was seeing in my straight skeleton code disappeared. Much to our delight, numerous bugs in Anselm's code disappeared too. This could have been a very nasty bug to track down the farther we got in to the development process, so I'm glad it was caught sooner rather than later.
Somewhere in the mix here we started using more and more ES6 features and simply loving them. Going back to Ecmascript 5 would be pure torture and pain and it's not going to happen. However, whenever we ended up in a debugger at a
for-of the VM would crash out. Ouch, this was really painful. As the earliest adopter of ES6 for-of in our code base I was dealing with it a lot. When Anselm started hitting it too enough was enough. Anselm then integrated the
babel transpiler. What a cool name. It let's us use way more features of ES6 than V8 currently supports and it also fixed up some of our debugging issues in the long run too. It was Anselm's turn to be ahead on ES6 adopting for several weeks (until today in fact!).
Anselm also simplified our module definitions by introducing Object.adopt as a replacement for Object.defineProperties which I had been using extensively. This is a really nice addition but I won't go into its details here. The interesting point of note here is that it doesn't allow Anselm to do auto-complete in Webstorm. LOL!.. so he went on a quest and started to try our github's Atom. I was part of the beta for Atom and at the time it was slow, buggy and lacking features compared to sublime text. It has come a long way since then and I am now using it as my main text editor for Citybound. Anselm is still not happy with his autocomplete situation.
Meanwhile I introduced computed properties so that I could fix up some fundamental issues in the straight skeleton code and for the first time, after putting them in, I was able to run without crash on the random shape generator and all the degenerate tests passed. I have now begun working on open-paths, such as zoning along the side of a road. After that will be support for curves in some manner.
We did some design on how we'd implement [a planning mode for the game] and I can tell you right now, wow, you guys are going to absolutely love what we have in store. If you're still reading this far, feel free to enjoy a bit of hype because planning is now an integral part of the game play and will enrich every part of the design.
Next up Anselm built something really cool on top of my static single assignment meta-programming library, which is called CodeBuilder. He used the babel transpiler parser to parse his method code and then recompile it using the CodeBuilder, which allows even simpler templating of numerical vector or matrix operations. This is seriously cool.
To round this out the latest work we've been touching on is iterators and reducers. Since ES6 gives us a lot more power with iterators and generators and using them is really light-weight on the garbage collection, using them in general is a good idea. But, we wanted more power with them. A few random thoughts several weeks back materialised in to real code and now we're discussing the second iteration of that, where we extend the built in generators to allow chaining of generators for more literate programming in the long term.
In short, a heck of a lot has been going on and I didn't even talk in depth about the stuff Anselm has been implementing on top of all this technology: Economy, road networks, market places, simulation life cycles, demand and of course the (hype) planning mode stuff.
Hopefully this will sate your desire for updates.