Comment by netbioserror
2 months ago
Nim. Fantastic choice for modern headless software. Simple obvious type system, preference for immutability and referential transparency. Dynamic collections are by default managed by hidden unique pointers on the stack. So the default RC isn't necessary unless explicitly invoked for a ref type.
Currently solo managing a 30k line data analysis application I built for my company. Easily fits in my head given the obvious pyramidal functional-like structure. Maybe two lines of memory semantics anywhere in the entire thing, and only one module that's OO with a constrained scope. Lots of static data files (style sheets, fonts) slurped up as const strings at compile time. Incredible performance. Invoked by our PHP server backend, so instead of doing parallel or async in the analysis, the server gets that through batch invocation.
Working stupid well for our product, plus I can easily compile binaries that run on ARM and RISC-V chips for our embedded team just by invoking the proper gcc backend.
Replaced an ailing and deliberately obfuscated 20 year old jumble of C and PHP designed to extort an IP settlement from my company. Did it in a year.
Do you have any recommendations for well designed open source Nim projects for someone study to get a feel for the language?
Anything written by Treeform[1] is a good place to start, their libraries make up a big chunk of Nim ecosystem.
1 - https://github.com/treeform/hobby
Honestly hard to say. There are a number of styles of architecting Nim libraries and programs, and almost none match my own. My most particular criticism of the Nim ecosystem is the abuse of macros: There are a number of libraries implementing huge chunks of functionality behind macros such that code paths only appear at compile time and are not reflected in sources. Some libraries constrain macro use, but many are built entirely out of macros. I'd say to avoid looking to those examples.
I think it's safe to say that the proper way to go about using Nim is to use macros only as a last resort. They're powerful and awesome when necessary, but obfuscate the code a lot, and the LSP can't really follow them either. They also make the code feel "fragile" in my opinion. I find that I rarely need to go further than templates, personally.