nix: add a comment on the many nixpkgs-with-cuda instances

diff --git a/.devops/nix/nixpkgs-instances.nix b/.devops/nix/nixpkgs-instances.nix
index 6e9872b28c8fb1a42c392ade8e9244186c49c6fd..4a2f81c4bfd044835a89bbfca09e039172dee354 100644 (file)
--- a/.devops/nix/nixpkgs-instances.nix
+++ b/.devops/nix/nixpkgs-instances.nix
@@ -7,6 +7,18 @@
     { system, ... }:
     {
       _module.args = {
+        # Note: bringing up https://zimbatm.com/notes/1000-instances-of-nixpkgs
+        # again, the below creates several nixpkgs instances which the
+        # flake-centric CLI will be forced to evaluate e.g. on `nix flake show`.
+        #
+        # This is currently "slow" and "expensive", on a certain scale.
+        # This also isn't "right" in that this hinders dependency injection at
+        # the level of flake inputs. This might get removed in the foreseeable
+        # future.
+        #
+        # Note that you can use these expressions without Nix
+        # (`pkgs.callPackage ./devops/nix/scope.nix { }` is the entry point).
+
         pkgsCuda = import inputs.nixpkgs {
           inherit system;
           # Ensure dependencies use CUDA consistently (e.g. that openmpi, ucc,