Get proper file creation (birth) time on Linux

File.stat gives me a pretty useless value for ctime. As documented it is “the last time the file or the inode was changed“.
Using stat via bash gives me a correct creation time via the Birth field.

I understand why that’s the case, as Elixir just calls out to :file.read_file_info and the birth field is a somewhat recent thing in Linux and I could not find a way to read it yet from Erlang.

Does anyone know of a way to get that field without shelling out to bash? Is there maybe a package that has implemented a NIF for that?

Or do I need to write my own NIF?
If so, Rust seems to get this one right: Metadata in std::fs - Rust

1 Like

I don’t know a particular NIF for you; but I’m wondering about the “shelling out to bash part”. I don’t know the harm in just using System.cmd

    {result, _exit} = System.cmd("stat", ["/home/cam/hello"])

    %{"Birth" => birth} =
      result
      |> String.split("\n")
      |> Enum.reject(&(&1 == ""))
      |> Map.new(fn line ->
        [key, value] =
          line
          |> String.trim()
          |> String.split(":", parts: 2)

        {key, value |> String.trim()}
      end)

    assert "2025-09-06 10:12:29.200959233 -0400" == birth

Thanks. But I need to get this not just for a single file, but as part of a file scanner.
Performance is key. I haven’t benchmarked yet, but I doubt actually calling stat + parsing is a viable option for me.

Huh, apparently they added a new syscall (statx) with this information. It’s also safe for 2038.

Does Erlang use that call underneath? If not, maybe it should going forward?

It does not, it calls fstat. It looks like on Windows the ctime is the same as Rust’s created.

1 Like

Yeah, I wouldn’t call 2017 new, but I guess new enough that it’s not in all standard libraries yet.
Python also doesn’t have it, but ruby, node, rust and zig do.

Anyway, I went ahead with the plan and created a small rustler NIF:

Now I just need to figure out how to publish this to hex such that a user doesn’t need to compile the thing?
Looks like rustler_precompiled is the way to go.

2 Likes

I was quite surprised Python didn’t have it; they usually have everything. Apparently there has been an open issue/PR for several years now. Strangely it seems to have become active again recently.

A nif seems like the way to go but maybe also consider opening an Erlang/OTP issue as this seems like something that should be updated to me. At least before 2038 :slight_smile:

1 Like

Package is published: file_statx | Hex :tada:

(Setting up rustler_precompiled in CI took me so long, I hate setting up CI stuff…)

Done :+1:

1 Like

The issue for future reference (if only the WWW could have had backlinks).