Most of the time you don’t need to compile yourself - which is why OS package managers exist. But sometimes it’s advantageous - E.g. it’s possible for a compil

Most of the time you don’t need to compile yourself - which is why OS package managers exist. But sometimes it’s advantageous - E.g. it’s possible for a compiler to take advantage of special CPU instructions that that are present on your CPU but cannot be guaranteed are present everywhere - resulting in a faster binary. In some special but relatively rare edge cases (e.g. numerical computing / simulations) these speed ups can be significant. But there’s heaps of room for improvement, E.g. most compilations have steps that are embarrassingly parallel - so make has an option “-j” that allows you to use multiple CPU cores - this significantly speeds up the compilation of numpy, the trouble is, it also increases memory consumption quite a bit, so make cannot change its default to use all cores as it would OOM a lot of existing build systems. What what I have seen the vast majority (like > 90%) of automated build systems don’t take advantage of multi core compiles, so that’s why they are dreadfully slow by default. Pythons subsystems are guilty of this, which is why building matplotlib, numpy, scipy, etc from source is very very slow