treewide: assemble all fetchurlBoot uses in overrides to fetchurl

[oxij]

What?

This kills top-level fetchurlBoot attribute and isolates rest of the uses to stdenv, normal fetchurl and one Darwin reference I don't touch because I can't test it.

Note that this change is a noop for nix-env -qaP --out-path diff. nix-env -qaP --drv-path diff is huge because everything that depends on curl itself and anything curl itself depends on will evaluate to new derivations. I.e. this changes a lot of .drvs but does zero rebuilds.

Stuff referencing pkgsCross works as before, so I declare this not breaking anything. Before rebase I was running on top of this for 3 months or so.

Why?

fetchurlBoot got in the way of clean breaking of dependency cycles for my doCheckByDefault thingy (#39464), I got pissed and (mostly) killed it.

git log

• treewide: assemble all fetchurlBoot uses in overrides to fetchurl itself

  The only outside-curl uses of fetchurlBoot left are stdenv
  and apple-source-releases. The latter one can probably be removed
  too, but I can't test it.

  Pros:

  • Aggregates all behind-the-scenes insanity in a single place.

  Cons:

  • At the cost of 10 more derivations (but 0 new outpaths).

nix-instantiate environment

• Host OS: Linux 4.9, SLNOS 19.03
• Nix: nix-env (Nix) 2.1.3
• Multi-user: yes
• Sandbox: yes
• NIXPKGS_CONFIG:

{
  checkMeta = true;
  doCheckByDefault = true;
}

nix-env -qaP --out-path diffs

• On x86_64-linux: noop
• On aarch64-linux: noop
• On x86_64-darwin: noop

/cc @Ericson2314 @matthewbauer @7c6f434c

[matthewbauer]

Awesome work! This definitely looks good to me.

[grahamc]

I feel this is a degradation in clarity, making it clear this is a super low dependency package. Do you think otherwise?

[oxij]

Well, I see your point, but I think uniformity of package expressions should be preferable to low-level implementation details, and fetchurlBoot is such an implementation detail.

[Ericson2314]

Thanks!! I've been hoping to see this go for awhile now.

[nh2]

@oxij You write:

      # On darwin, libkrb5 needs bootstrap_cmds which would require
      # converting many packages to fetchurl_boot to avoid evaluation cycles.
      gssSupport = !stdenv.isDarwin && !stdenv.hostPlatform.isWindows;

The comment talks only about darwin, but there's also this part about Windows.

Why does hostPlatform.isWindows also disable it?

I'm trying to add some more comments here, so it would be great to know this.

[oxij]

@oxij You write: ``` # On darwin, libkrb5 needs bootstrap_cmds which would require # converting many packages to fetchurl_boot to avoid evaluation cycles. gssSupport = !stdenv.isDarwin && !stdenv.hostPlatform.isWindows; ```

That was introduced in c303047 by @vcunat. (`git log -p -S 'On darwin, libkrb5 needs bootstrap_cmds which would require'`)

The comment talks only about darwin, but there's also this part about Windows. Why does `hostPlatform.isWindows` also disable it?

I took that from the corresponding line of `pkgs/tools/networking/curl/default.nix` to keep the value of that parameter on Windows. Now that I think about it, #66506 should probably do ```diff - # On darwin, libkrb5 needs bootstrap_cmds which would require - # converting many packages to fetchurl_boot to avoid evaluation cycles. - gssSupport = !stdenv.isDarwin && !stdenv.hostPlatform.isWindows; + gssSupport = !stdenv.isDarwin || old.gssSupport; ```

[oxij]

Oops, I meant ```diff + gssSupport = !stdenv.isDarwin && old.gssSupport; ```

[nh2]

@oxij I think old.gssSupport by itself without or is correct, because defaulted bool arguments are missing from old: #66546

Regarding what you said, so you mean that I should change my fix from #66506 from

       gssSupport =
        if stdenv.isDarwin || stdenv.hostPlatform.isWindows
          then false
          else old.gssSupport or true; # `? true` is the default

to

       gssSupport =
        if stdenv.isDarwin
          then false
          # The default of `gssSupport ?` in the curl package; see #66546 for the `or`.
          else old.gssSupport or (!stdenv.hostPlatform.isWindows); 

Is that correct?

[oxij]

Is that correct?

No, my point was to remove duplicated sub-expressions in two definitions of `gssSupport` but, as you note in #66546, that needs a different semantics for `.override (old: ...)`.

[nh2]

@oxij OK. Still, what I suggested there above still is better than what I have now, right?

Using

       gssSupport =
        if stdenv.isDarwin
          then false
          # The default of `gssSupport ?` in the curl package; see #66546 for the `or`.
          else old.gssSupport or (!stdenv.hostPlatform.isWindows); 

would not force gssSupport off for Windows, instead it would use old.gssSupport if it had been set to true for Windows manually by the user.

Do you see what I mean?

The original forcing was added with the intent to be just for Darwin, so I'm trying to make it just for Darwin.