I don’t think that even the languages are the problem, it’s the toolchain. While certainly if you went back to C or whatever, you can design more performant systems, I think the problem overall stems from modern toolchains being kinda ridiculous. It is entirely common in any language to load in massive libraries that suck up 100’s of mb of RAM (if not gigs) to get a slightly nicer function to lowercase text or something.
The other confounding factor is “write once, run anywhere” which in practice means that there is a lot of shared code and such that does nothing on your machine. The most obvious example being Electron. Pretty much all of the Electron apps I use on the reg (which are mostly just Discord and slack) are conceptually simple apps that have analogues that used to run on a few hundred mbs of storage and 10’s of mb of RAM.
Oh, one other sidetone - how many CPUs are wasting cycles on things that no one wants, like extremely complex ad-tracking/data mining/etc.
I know why this is the case, and ease of development does enable us to have software that we probably otherwise wouldn’t, but this is a thing that I think is a real blight on modern computing, and I think it’s solvable. I mean, probably the dumbest idea, but improving translation layers to run platform-native code can be vastly improved. Especially in a world where we have generative AI, there has to be a way to say “hey, I’ve got this javascript function, I need this to work in kotlin, swift, c++, etc.”
LLVM is ironically a very slow compiler back-end, whose popularity has contributed to a general slow-down in compilation speed across the whole industry (it’s even slow at doing debug builds for fast iteration).
Doesn’t really matter if the compiler is slow if the result is optimized and fast 🤷 Rust compiles slower than C, but that’s because C has no safeguards (excluding static typing). Very often the wasted CPU cycles are on the end of the user, not the developer.
I don’t think that even the languages are the problem, it’s the toolchain. While certainly if you went back to C or whatever, you can design more performant systems, I think the problem overall stems from modern toolchains being kinda ridiculous. It is entirely common in any language to load in massive libraries that suck up 100’s of mb of RAM (if not gigs) to get a slightly nicer function to lowercase text or something.
The other confounding factor is “write once, run anywhere” which in practice means that there is a lot of shared code and such that does nothing on your machine. The most obvious example being Electron. Pretty much all of the Electron apps I use on the reg (which are mostly just Discord and slack) are conceptually simple apps that have analogues that used to run on a few hundred mbs of storage and 10’s of mb of RAM.
Oh, one other sidetone - how many CPUs are wasting cycles on things that no one wants, like extremely complex ad-tracking/data mining/etc.
I know why this is the case, and ease of development does enable us to have software that we probably otherwise wouldn’t, but this is a thing that I think is a real blight on modern computing, and I think it’s solvable. I mean, probably the dumbest idea, but improving translation layers to run platform-native code can be vastly improved. Especially in a world where we have generative AI, there has to be a way to say “hey, I’ve got this javascript function, I need this to work in kotlin, swift, c++, etc.”
LLVM will save us.
Maybe…
LLVM is ironically a very slow compiler back-end, whose popularity has contributed to a general slow-down in compilation speed across the whole industry (it’s even slow at doing debug builds for fast iteration).
WASM has some promise though
Doesn’t really matter if the compiler is slow if the result is optimized and fast 🤷 Rust compiles slower than C, but that’s because C has no safeguards (excluding static typing). Very often the wasted CPU cycles are on the end of the user, not the developer.