An older version of the script did this, but it turned out not to be a good idea. It is possible to generate URLs with the script that nobody else can download. If we store the pre-fetched files, then whoever is running the update script will not realize the problem. I have myself uploaded broken updates this way. Is there some way we can address this problem while also storing the pre-fetched files?

Are the urls working for the same person twice? Or could we just check a second time that the url is valid?

Are the urls working for the same person twice? Or could we just check a second time that the url is valid?

Well, it's not really checking a second time, it's that with Nixpkgs' mirror:// syntax, the URL used by the script is not necessarily the same as the URL used by Nix. If the download is cached in the store, the download is never made by Nix.

Yeah I know that it is currently not loaded a second time. That was the "solution", I thought of.
Any idea why an url is generated in the first run, that does not work? Maybe we can fix it that way.

Any idea why an url is generated in the first run, that does not work? Maybe we can fix it that way.

wget needs the full URL, but fetchurl takes the trailing part of the URL and appends it to mirror://kde/. If wget understood the mirror:// syntax, it would not be a problem; we could use exactly the same URLs for both. We cannot use curl like fetchurl does because it does not know how to recursively download directories (AFAIK).