How is Tenstorrent h/w more general purpose than NVIDIA chips? TT hardware is only good for matmuls and some elementwise operations, and plain sucks for anything else. Their software is abysmal.
I was recently trying to solve a similar problem but on desktop platforms. I don't want to depend on LaTeX, but I'd like to be able to generate equation images inside a C++ desktop application. I tried to make MathJax run via QuickJS and extract the SVG for rasterization. But I couldn't make MathJax run with QuickJS.
The functionality that I am personally interested in from a binary parsing framework like Kaitai is generating an encoder implementation in addition to a decoder one. In other words, given a description of a binary format, I would like to be able to construct an instance of a class whose memory layout matches the format. For instance, if the format has an int n, then an array `a` of size `n`, and then a double `d`, it would be awesome to be able to construct a corresponding object with fields `n`, `a` and `d` and when I change `n`, then the size of `a` changes accordingly. And then, if I pass a pointer to this object to the decoder, it would be able to parse it correctly, as if the memory representation of the object came from some external buffer.
The only requirement placed on the “moved out” variable is that you should be able to call its destructor. Which means that it has to be in a valid but unspecified state. So it's fine to access such a variable, so long as you don't read its exact state. You can still assign to it, for instance.
I recently decided to give terminal multiplexer a try. I first used tmux for a couple of weeks and then discovered zellij. It felt much more user-friendly, but I don't know how much I am missing out in terms of useful features compared to tmux.