Google Developers Are Looking At Creating A New libc For LLVM

[ms178]

While others were more friendly to that approach on the list, Rich Felker (creator of musl) wasn't at all, especially as a Google project.

[caligula]

yay, re-inventing the wheel once again, one has to wonder what is wrong with musl libc though? not good enough, how? https://www.youtube.com/watch?v=NMQrnVDf-rY

Musl wastes precious space by supporting dynamic linking. Modern flagship phones have very limited disk space (e.g. 256 to 1024 GB) and might not be able to support all these fancy features.

[xorbe]

ob link: https://xkcd.com/927/

[dreich]

musl is probably too high quality for google.

[microcode]

I think I see what they're doing here: they want to implement parts of the libc in a way which LLVM can choose to inline or co-optimize with the program more aggressively, but they aren't so much looking to replace libc per se.

[brad0]

Anyways, Google is trying hard to skip all GPL code as soon as possible.

I don't blame them. It would be dumb to add GPL code when everything else is not. You don't start going the other way for the wrong reasons.

It would be equally as dumb as having a binary GPU driver in a 99% open source OS.

[wizard69]

I hope they write it in Rust, not in Go

It is a C library why use anything but C?

in any event I have to agree with other level here - what a waste of resources.

[paupav]

Why not write libc dirrectly in assembler? It is used everywhere.

[cl333r]

I hope they write it in Rust, not in Go

In Rust We Trust.

[c117152]

It is a C library why use anything but C?

A case was made for C++ a few times over abstracting certain convoluted hardware details' optimization paths. But since it's almost exclusively relevant only to SIMD which is considered a bad hardware design choice from the get go, it hasn't met consensus.

But other than that, yeah only C and some assembly should be there.

in any event I have to agree with other level here - what a waste of resources.

Not necessarily. Intel just came out with their own subset of CPP. Microsoft has their Microsoft Universal C Runtime. The common denominator is that all three companies are developing new architectures with new features that are alien enough to the C memory model that they need to carefully go over the compilers and stdlib and write it all anew.