I don't think reverting to 0.21 is a good idea, looking at https://repology.org/project/gettext/information.

I don't think it's a good idea, looking at https://repology.org/project/gettext/information.

Can you elaborate why it's not a good idea? It's an intermediate fix to unblock (unrelated) PRs to staging, we can re-roll a 0.21.1 bump once someone gets to it.

Can you elaborate why it's not a good idea?

Repology already suggests that more distros are on 0.21.1 than on 0.21, so whether there's an announcement or not, I think this version is well recognized as a new release.

The hash being still sha256-6MNlDh2M7odcTzVWQjgsHfgwWL1aEe6FVcDPJ21kbUU= on my machine in Paris, I think it's probably worth trying ways around the hash mismatch, such as going through Wayback Machine.

If we revert until the hash stabilizes, I can open a PR to patch what was causing problems for the Darwin stdenv update. I was trying to be proactive before landing the actual bump, but blocking staging is no good.

I'll check if fetching directly from https://ftp.gnu.org/pub/gnu/gettext/${pname}-${version}.tar.gz helps. It's likely that some GNU mirrors are faulty.

EDIT: opened #235421.

It definitely seems to be some flaky mirror. Applying #235421 got it to download, but before, fetching was broken continuously.

It was the following derivation:

nix show-derivation /nix/store/2c90gp52cvdprpj314d86qii9qil4fkb-gettext-0.21.1.tar.gz.drv
{
  "/nix/store/2c90gp52cvdprpj314d86qii9qil4fkb-gettext-0.21.1.tar.gz.drv": {
    "args": [],
    "builder": "builtin:fetchurl",
    "env": {
      "builder": "builtin:fetchurl",
      "executable": "",
      "impureEnvVars": "http_proxy https_proxy ftp_proxy all_proxy no_proxy",
      "name": "gettext-0.21.1.tar.gz",
      "out": "/nix/store/y2h7bqjpc4q9g887w8pbwncjrmr4g9sx-gettext-0.21.1.tar.gz",
      "outputHash": "sha256-6MNlDh2M7odcTzVWQjgsHfgwWL1aEe6FVcDPJ21kbUU=",
      "outputHashAlgo": "sha256",
      "outputHashMode": "flat",
      "preferLocalBuild": "1",
      "system": "builtin",
      "unpack": "",
      "url": "https://ftpmirror.gnu.org/gettext/gettext-0.21.1.tar.gz",
      "urls": "https://ftpmirror.gnu.org/gettext/gettext-0.21.1.tar.gz"
    },
    "inputDrvs": {},
    "inputSrcs": [],
    "outputs": {
      "out": {
        "hash": "e8c3650e1d8cee875c4f355642382c1df83058bd5a11ee8555c0cf276d646d45",
        "hashAlgo": "sha256",
        "path": "/nix/store/y2h7bqjpc4q9g887w8pbwncjrmr4g9sx-gettext-0.21.1.tar.gz"
      }
    },
    "system": "builtin"
  }
}

I'm not sure explicitly hardcoding one place over the other is actually a good fix, though.

I wonder if we could do anything in the future to make this more debuggable, see which mirror specifically is buggy?

Closing this issue, this seems to have been one faulty mirror, will open a new discussion about this elsewhere and link from here.

I opened NixOS/nix#8432 to discuss the fetcher problem.