| Symbolic math solver & compiler. Only 5000 lines of code but the complexity is insane. The engine deal with a a subset of math equations in quantum field theory. But the deal is that I have to make it compile several optimized math code in several situations. I would be nice to treat those math ops objectively but it will be too slow. Inmutate will also be slow. So inplace modification is everywhere in the code. Also the problem is multidimensional. Mathematicaly, I am working with tensors, matrices, scalars. Different treatment for each of them in different compile strategy. Programmatically, different situations sometimes need me to modify only one line of code, sometimes need to modify a block of code. Sometimes a block need to be coherent with the module it sits in, sometimes the module depends on states of those blocks. Everything is entangled and there is no obvious way to do it right. Applying clean code strategy like "do not repeat" or "keep operation on same level" or "do not make side effect" all suck. Compiling the largest part will then take 1 hr, but now I happy with my messy code base that can compile in 1 min. Because of this, debugging is so much easier and I can just verify arbitrary line of resulting code on the fly. I have been rewritten the whole thing 6 times in 1 year to try which programming decision is better also extending functionalities. Still a mess. |