attrsOf -> lazyAttrsOf

[roberth]

This means we don't support mkIf in the exact places where the type was attrsOf. Almost always, mkIf can be replaced by optionalAttrs.

The benefit is that the code becomes much lazier, improving performance and

• Fixes devShell to eagerly evaluated?  #270

[srid]

Thanks!

I guess I should just use lazyAttrsOf for everything by default? What's the point of keeping the strict version by default in nixpkgs (ie., why not types.strictAttrsOf)?

[roberth]

In Nixpkgs, attrsOf is often "just data" which is consumed wholesale. There it might make sense to still use attrsOf and support mkIf, but I'm not completely convinced.
I think the rule "mkIf for submodules, optionalAttrs for attrsets" works at least as well as the status quo, but I haven't proven it, and everyone will need to be made aware of it, if Nixpkgs does switch.

[roberth]

I've opened a Nixpkgs issue for this topic.

• Always use lazyAttrsOf? Rename attrsOf to strictAttrsOf? NixOS/nixpkgs#295419