Slightly more std::filesystem for nix eval

[Ericson2314]

Context

Progress on #9205

Priorities and Process

Add 👍 to pull requests you find important.

The Nix maintainer team uses a GitHub project board to schedule and track reviews.

[bryanhonof]

nitpick(non-blocking): Could maybe even move this inside of the class declaration.

Suggested change

	namespace nix::fs { using namespace std::filesystem; }
	using fs = std::filesystem;

[Ericson2314]

Actually it can't be that, I'll say why below

[bryanhonof]

nitpick(non-blocking): Would prefer to see this wrap the whole file.

[edolstra]

I hadn't noticed this before, but there are a few of these nix::fs namespace declarations in src/nix now. I don't really understand what that's for? Just use std::filesystem so people don't have to wonder what nix::fs is.

[Ericson2314]

That's because of

nix/src/libutil/file-system.hh

Lines 129 to 144 in 26c3fc1

	namespace fs {
	/**
	* ```
	* symlink_exists(p) = std::filesystem::exists(std::filesystem::symlink_status(p))
	* ```
	* Missing convenience analogous to
	* ```
	* std::filesystem::exists(p) = std::filesystem::exists(std::filesystem::status(p))
	* ```
	*/
	inline bool symlink_exists(const std::filesystem::path & path) {
	return std::filesystem::exists(std::filesystem::symlink_status(path));
	}
	} // namespace fs

I hope that with more std::filesystem, most of file-system.hh goes away. However, there are a few things that aren't in there they we need, like this symlink_exists.

I put it in an fs namespace with the idea that if C++ files (not headers!) use fs:: as a shorthand for std::filesystem::, then it will nicely gell with the existing apis.

namespace nix::fs { using namespace std::filesystem; }

I do because if I do

using fs = std::filesystem;

then that fs will shadow, rather than expand, the namespace fs I put in file-system.hh.

I tried a few combinations, and the thing you commented on is the only one that will work (other than declaring symlink_exists in std::filesystem directly, which would be ill-advised.)

[roberth]

clangd reports for an fs::current_path identifier:

function current_path
provided by

→ path

// In namespace std::filesystem
path current_path()

This pattern also seems useful for polyfilling functions from new C++ versions.

[roberth]

Refactor LGTM. Details can be improved if need be.

[xokdvium]

Am I missing something or is this borked with release builds without asserts?

// TODO abstract mkdir perms for Windows
createDir(path.string(), 0777);

// Directory should not already exist
assert(fs::create_directory(path.string()));

Without asserts the call to fs::create_directory will be stripped out. More appropriate would be:

[[maybe_unused]] bool created = fs::create_directory(path.string());
assert(created);

[Ericson2314]

@xokdvium Sorry I missed your message above!