is it possible to solve your problem through the apply option? I mean you can set the defualt to be {}, use apply instead

apply = cfg: lib.recursiveUpdate {<default>} cfg;

I'm a little lost. Which line would I inject this:

out = mapAttrs (name: mkDefault) config.default

Also doesn't mapAttrs a function of 2 variables?

I'm a little lost. Which line would I inject this:

out = mapAttrs (name: mkDefault) config.default

Also isn't mapAttrs a function of 2 variables?

Cool man. Thanks!

What is really nice is that I can use ({...}: {user = {b = 3;};}) and it overrides. Which provides a nice API for end users (i.e. not having to use mkForce).

So this partially makes sense. But why didn't having a user config with ({...}: {user = {b = lib.mkForce 3;};}) work before? It should have a really high priority.

Without your fix, does the config.default have... no priority? What was the failure mechanism before the fix?

In reply to @infinisil:matrix.org
Before, config.user and config.default had the same priority. The conflict happened one level above

Finally, you have effectively

{
  out = lib.mkMerge [
    config.default
    config.user
  ];
}

Finally, you have effectively

{
  out = lib.mkMerge [
      { a = 1; b = 2; }
      { b = 3; }
  ];
}

Both of which are without priority, so it recurses into the attrs and conflicts because two bs