Legacy nix-shell with a path prefers shell.nix over default.nix

[K900]

Describe the bug

~/test
❯ cat shell.nix
throw "i'm shell.nix"

~/test
❯ cat default.nix
throw "i'm default.nix"

~/test
❯ nix shell nixpkgs#nixVersions.nix_2_24 --command nix-shell . --attr hello
error:
       … while calling the 'throw' builtin
         at /home/k900/test/shell.nix:1:1:
            1| throw "i'm shell.nix"
             | ^
            2|

       error: i'm shell.nix

~/test
❯ nix shell nixpkgs#nixVersions.nix_2_18 --command nix-shell . --attr hello
error:
       … while calling the 'throw' builtin

         at /home/k900/test/default.nix:1:1:

            1| throw "i'm default.nix"
             | ^
            2|

       error: i'm default.nix

This also breaks a common pattern of nix-shell path/to/nixpkgs -A foo.

Steps To Reproduce

1. Get a nixpkgs checkout
2. nix-shell . -A hello
3. See error

Expected behavior

A shell.

nix-env --version output

nix-env (Nix) 2.24.9

Additional context

2.18 is not affected, neither is Lix, may bisect later.

Priorities

Add 👍 to issues you find important.

[K900]

Never mind, it's an intentional change: cfe3ee3

• nix-shell: look up shell.nix when argument is a directory #11057

Whether that was a good idea or not is to be discussed, but this is not a bug.

[roberth]

Hi @K900, don't hesitate to ping me next time

to be discussed

Do you think it should still look for default.nix when -A/--attr is passed?
I think that would still solve the problem in issues such as #496, without breaking this use case.

[emilazy]

It seems sort of odd to me for the meaning of . in nix-shell . … to change if an -A follows.

[roberth]

I agree it seems like an ad hoc extra rule on the face of it, but we could think of it as only changing the shell rule:

nix-shell resolves to shell.nix when no attribute is specified.

It's not great, I have to admit, but it's still a manageable rule, and least it wouldn't regress a significant Nixpkgs use case.

• Development shell with a pinned nixfmt nixpkgs#322512 (comment)

[emilazy]

FWIW in discussion with @K900 and @alyssais on Matrix I think we considered that it might be reasonable to simply re‐export the entire package set from Nixpkgs’ shell.nix, which is kind of gross but IIRC we ultimately decided might be nicer than changing Nix behaviour here. Sadly it was a little while ago now so I don’t have a pointer to the discussion or recollection or the trade‐offs we perceived.

[roberth]

@emilazy I just came up with the same thoughts! NixOS/nixpkgs#353240

[K900]

Yeah, I think the nixpkgs change is the way to go now. At least that way all the Nix versions work as intended.

[roberth]

Complicating the rules to accomodate this usage of nix-shell would make its behavior incomprehensible, and probably cause even more bug reports than those that led to #11057
Closing again.