Add a fchownat(2) wrapper

[jmmv]

No description provided.

[jmmv]

Per the TODO, do you think chown should be removed?

The way I'd validate that is: add a trivial test for fchownat to ensure we reference this function somewhere and then ask CI to validate all platforms. If that succeeds, I can later try to see if this is sufficient for older macOS versions, which still exist in Travis, to prevent the issue we saw with utimes.

I can do this before the merge; just let me know if there are other considerations.

[asomers]

A successful Travis run is insufficient for removing a binding. We need to make sure that the newer alternative is supported on all Nix platforms, including those that aren't tested in CI, like NetBSD and OpenBSD. Though in this case, I don't think there's much to be gained by removing chown. Since it's required by POSIX, OSes won't ever drop it from their C libraries. And it doesn't cost anything to keep around. Do you agree?

[jmmv]

I don't mind keeping chown, but then would the same rationale apply for lchown and we should add it? Why doesn't the same apply for other traditional functions like utimes?

In fact, I think that if the OS exposes such functions, they should be exposed in nix as well. nix shouldn't be in the business of calling "traditional but ugly" functions obsolete... because all the new-ish counterparts are not always available :(

[asomers]

I don't mind keeping chown, but then would the same rationale apply for lchown and we should add it? Why doesn't the same apply for other traditional functions like utimes?

In fact, I think that if the OS exposes such functions, they should be exposed in nix as well. nix shouldn't be in the business of calling "traditional but ugly" functions obsolete... because all the new-ish counterparts are not always available :(

The same rationale would apply to lchown only if we already provide it. I don't want to withdraw chown because consumers may already be using it. lchown, OTOH, is unnecessary unless somebody ports Nix to a weird platform that doesn't have fchownat.

[jmmv]

I've added tests to ensure these functions exists in all CI platforms, and also confirmed in my project that fchownat exists for the older macOS that I need. So no need to add lchown from my point of view, which is what I wanted to confirm.

[asomers]

This LGTM now. But I think you should remove the mention of deprecating chown from your commit message. I don't think we should ever remove chown.

[jmmv]

Updated the commit message and removed the original contents of the PR message, which I think would be added to the history as well.

Also rebased on top of the truncate merge to avoid CHANGELOG conflicts.

[asomers]

bors r+

[bors]

Build succeeded

• continuous-integration/travis-ci/push
• buildbot/nix-rust/nix amd64_fbsd11
• buildbot/nix-rust/nix i386_fbsd11