Use portable C++ Pseudorandom number generator

[Ericson2314]

Avoid calls to rand / srand / random / srandom and other things which are not portable to Windows

Originally posted by @edolstra in #8901 (comment)

[nixos-discourse]

This issue has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/nix-on-windows/1113/109

[RCoeurjoly]

I see that random() is called in two places:

src/libstore/unix/gc.cc:44: Path tempLink = fmt("%1%.tmp-%2%-%3%", link, getpid(), random());
src/libstore/unix/optimise-store.cc:228: Path tempLink = fmt("%1%/.tmp-link-%2%-%3%", realStoreDir, getpid(), random());

Should we create a PRNG before every call?
Or create a PRNG in initNix that somehow gets passed to the functions that need it?

[Ericson2314]

@RCoeurjoly good question! When in doubt, it is better to avoid global state.

I think for the gc one, it would be good to put the PRNG state in IndirectRootStore, and since LocalStore is a subclass, the optimise-store one can use that too.

Thanks for looking into this!

[dottharun]

I see another random() call in - src/libstore/unix/build/derivation-goal.cc:826 too

[Ericson2314]

That can also use a per-store PRNG

[RCoeurjoly]

The problem is that random() in gc.cc:44 is called by makeSymlink(const Path & link, const Path & target), a free function, so there is no way to access a per-store PRNG.

I was thinking of the following:

Initialize the PRNG in initNix, as it is done now with srandom().
Use a class with static members.

Use this class wherever is needed.

Thoughts?

[dottharun]

IndirectRootStore Mix-in seems to have a clear purpose of providing its two methods

• do we have to use it for PRNG just for its structure?
• maybe some other class in libutil/ as seeding needs to done only once?

[Ericson2314]

@RCoeurjoly After #10556 it will not be a free function

@dottharun And the PRNG will be for one the IndirectRootStore methods; the indirect root symlinks not colliding in the directory is part of the spec! :)

[Ericson2314]

After #10556 it will not be a free function

If anyone wants to take this issue, they should feel free to do just redo that part of #10556 in their PR. That is better than waiting for my PR.

[Ericson2314]

#10556 is now merged, so this should be a bit easier.