> Both Rust and C++ ought to be relegated to high performance cores of larger, "squishier" programs
So, people from Google have explained about this before. The idea you have assumes that if you speed profile the code, 99% of samples land in this function "A()" and so you just re-write that part in C++ and now it's faster, or if you measure allocations you find 99% of RAM was allocated by "B()" and so you just re-write that part in C++ and now it uses less RAM.
Google already did all that low-hanging fruit. When they run the profile it comes back flat. You should rewrite A, B, D, E, F, J, K, L, M, N ... in other words the way to make the system faster is to just write it in C++
Part of this is also scale. If a system I run once week is a little slow, maybe in some sense that costs 40ยข but I don't account for it. At a mid-size non-IT firm maybe a similar performance cost is $1000 per year. Just about worth somebody enquiring if it can be sped up, but not worth arguing about it if the answer is "No". Maybe a mid-size IT firm where you work spends $1k per month on this problem. You could speed it up by rewriting the whole system, this would take some time - how many days work before it's cheaper to leave the problem than pay you to fix it? However, at Google's scale maybe that problem costs them $1M per week. They can justify assigning a whole team to fix that because of scale.
So, people from Google have explained about this before. The idea you have assumes that if you speed profile the code, 99% of samples land in this function "A()" and so you just re-write that part in C++ and now it's faster, or if you measure allocations you find 99% of RAM was allocated by "B()" and so you just re-write that part in C++ and now it uses less RAM.
Google already did all that low-hanging fruit. When they run the profile it comes back flat. You should rewrite A, B, D, E, F, J, K, L, M, N ... in other words the way to make the system faster is to just write it in C++
Part of this is also scale. If a system I run once week is a little slow, maybe in some sense that costs 40ยข but I don't account for it. At a mid-size non-IT firm maybe a similar performance cost is $1000 per year. Just about worth somebody enquiring if it can be sped up, but not worth arguing about it if the answer is "No". Maybe a mid-size IT firm where you work spends $1k per month on this problem. You could speed it up by rewriting the whole system, this would take some time - how many days work before it's cheaper to leave the problem than pay you to fix it? However, at Google's scale maybe that problem costs them $1M per week. They can justify assigning a whole team to fix that because of scale.