-
mu
Soul of a tiny new machine. More thorough tests → More comprehensible and rewrite-friendly software → More resilient society. (by akkartik)
-
Vrmac
Vrmac Graphics, a cross-platform graphics library for .NET. Supports 3D, 2D, and accelerated video playback. Works on Windows 10 and Raspberry Pi4.
-
SurveyJS
Open-Source JSON Form Builder to Create Dynamic Forms Right in Your App. With SurveyJS form UI libraries, you can build and style forms in a fully-integrated drag & drop form builder, render them in your JS app, and store form submission data in any backend, inc. PHP, ASP.NET Core, and Node.js.
-
I've taken it a step further (or three). I'm building a whole computer on the principle[1] (one possible way of looking at it) of avoiding parsing as far as possible.
https://github.com/akkartik/mu
Mu takes the dwm[2] principle of avoid-config-files/modify-sources-to-reconfigure to the limit. The idea is that there are at any given time only 3 languages in the computer:
1. A self-hosted notation for a subset of 32-bit x86 machine code
2. A memory-safe statement-oriented language where most statements map 1:1 to machine-code instructions. Written in level 1 above.
3. A high-level interpreted language.
The vision is to fix 1 and 2 but allow 3 to fork in lots of different folks. If you want a Lisp for your HLL, make a fork. If you want a Python-like syntax, make a fork. But, and this is the important part, in any given computer/repo there is only ever one language. Only one non-trivial parser. (Levels 1 and 2 have extremely uniform syntax.)
As best I can tell, the #1 way to avoid the need to run a fuzzer is to avoid writing parsers. Just say no.
[1] https://lobste.rs/s/to8wpr/configuration_files_are_canary_wa...
[2] https://dwm.suckless.org
> What can it possibly be doing?
Possibly, two things.
1. Indexing the mesh. STL files don't contain meshes, they instead have a triangle soup. Indexed meshes are more efficient for rendering, they save VRAM bandwidth and vertex shaders.
2. Computing normals. STL files have per-triangle normals (can be complete garbage because most software ignores them), for smooth surfaces you want per-vertex normals. Computing them well (like I did there https://github.com/Const-me/Vrmac#3d-gpu-abstraction-layer ) is slow and complicated.
In a project I worked on few years ago, we had persistent performance problems, present for a long time before I showed up. One of the first things I did when I joined the team was do some unsolicited statistical profiling and narrow it down to database interactions. Since we were using a custom-written connector to a graph database, we assumed it's the connector, or the JSON-based API it's using.
The issue got on the back burner, but it bugged me for a long time. I was later writing some performance-intensive number crunching code, and ended up writing a hacky tracing profiler[0] to aid my work. Having that available, I took another swing at our database connector issue...
And it turned out the connector was perfectly fine, its overhead was much lower than the time the DB spent filtering data. The problem was, where I expected a simple operation in our application to do a few DB queries, it did several hundreds of them! Something that, for some reason, wasn't obvious on the statistical profiler, but it stood out on the full trace like a sore thumb.
I cut the run time of most user-facing operations by half by doing a simple tweak to the data connector - it turns out even a tiny unnecessary inefficiency becomes important when it's run a thousand times in a row. But ultimately, we were doing 100x as many queries as we should have, nobody noticed, and fixing it would be a huge architecture rework. We put it on the roadmap, but then the company blew up for unrelated reasons.
I sometimes think that maybe if we tracked and fixed this problem early in the development, our sales would have been better, and the company would have been still alive. For sure, we'd be able to iterate faster.q
--
[0] - https://github.com/TeMPOraL/tracer
