Deal with technical debt in the kernel

[hannobraun]

Over the last months, I have added quite a bit of unnecessary complexity to the kernel. In some cases, I knew it was unnecessary, but couldn't think of a simpler solution at the time. In other cases, I thought the complexity was just inherent to the problems I was trying to solve. This complexity has turned into a real problem. It hinders every non-trivial change I'm trying to make in the kernel, and caused progress in that area to slow to a crawl.

By now, I have both recognized that a lot of the complexity that I thought necessary is in fact unnecessary, and thought of solutions to reduce it. I have already opened quite a few issues meant to address it, and will open more as I gain new insights. This issue is meant as a central place to track and coordinate these efforts.

Here's the list of issues:

• Clean up object construction code #1249
• Simplify object graph around HalfEdge #1525
• Unify full and partial objects #1570
• Reduce redundancy in object graph through inference and caching #1586
• Consolidate redundant references to Surface #1588
• Consolidate redundant references to GlobalCurve #1605
• Compute surface position on demand instead of storing it #1634
• Consolidate redundant references to Vertex #1643

I will keep this list updated going forward.

These issues are closely related, and it will probably not be practical to address most of them in isolation. This is because the very complexity that they aim to address tends to hinder the necessary cleanup efforts. Most likely, I will try to pick low-hanging fruit from each of them, reducing the overall complexity each time, and making more low-hanging fruit available. This will hopefully allow me to address them in incremental steps, without being forced into too big of a rewrite.

[hannobraun]

#1588 has been addressed. I've updated the list in the issue description.

Finally something that was easier than expected!

[hannobraun]

Added #1605 to the list.

[hannobraun]

#1605 is done! List updated.

[hannobraun]

I've made some progress on #1525. See the update I posted there: #1525 (comment)

[hannobraun]

Added #1634.

[hannobraun]

#1634 and #1525 are done! List updated.

I think the most promising next step is #1586, which I'll be looking into next. Although I'll keep my eyes peeled for more opportunities to simplify things, which might mean more issues being added to the list.

[hannobraun]

I had another idea. Added #1643 to the list.

[hannobraun]

And #1643 is done already. List updated.

This was both surprisingly easy, and surprisingly effective. Almost 300 lines of code removed!

[hannobraun]

And #1586 is also done. List updated.

Very productive day!

[hannobraun]

#1570 is now also done (list has been updated), which takes care of all remaining items on the list!

I have some more ideas on what could be improved that I want to follow up on, but those are just some weak notions that might come to nothing. Most likely, I will at most land some more pull requests that clean up some smaller stuff. I might open more issues and add those to the list, but I don't expect that.

One way or another though, I won't spend much more time on this issue. The improvement so far has been massive, and I think it's time to move on to other opportunities.

[hannobraun]

Okay, it's time to declare victory. I'm done!

I've merged two more pull requests, cleaning up the builder API (#1692, #1693). I've also opened one more issue (#1691) that I could add to the list here, but I decided not to. First, it's not critical and doesn't need to happen now. Second, I'm not sure about the solution, so it makes sense to let that hang around the back of my mind for a bit anyway.

I'm really happy with how this cleanup turned out! In aggregate, it's been a huge improvement. The kernel code has reached a level of simplicity that it hasn't had in many months. And back then, it had fewer features and was much more riddled with bugs, so I think that's a clear win.