`nix-build` / `nix-instantiate` will silently ignore unknown arguments being passed with `--arg` #865

New issue

Closed

opened 2025-06-21 22:37:47 +00:00 by raito · 6 comments

raito commented

2025-06-21 22:37:47 +00:00

Owner

Describe the bug

Imagine you have: { }: true as a file.

You do: nix-build --arg something true on that file and you expect to get an error because you are passing an argument that doesn't exist.

But, you do not.

Steps To Reproduce

{ }: true in default.nix
nix-instantiate --arg something true --eval -f default.nix
See true being returned

Expected behavior

nix-repl> ({ }: true) { a = 5; }
error:
       … from call site
         at «string»:1:1:
            1| ({ }: true) { a = 5; }
             | ^

       error: function 'anonymous lambda' called with unexpected argument 'a'
       at «string»:1:2:
            1| ({ }: true) { a = 5; }
             |  ^

`nix --version` output

Lix 2.94.0-dev.

Happens also in CppNix.

Additional context

This caused a silent bug in the functional1 test suite with non-UTF8 inodes, I passed the wrong argument nonUtf8Inode instead of nonUtf8Inodes and generated NO non-UTF8 inodes.

Tough luck.

## Describe the bug Imagine you have: `{ }: true` as a file. You do: `nix-build --arg something true` on that file and you expect to get an error because you are passing an argument that doesn't exist. But, you do not. ## Steps To Reproduce 1. `{ }: true` in `default.nix` 2. `nix-instantiate --arg something true --eval -f default.nix` 3. See `true` being returned ## Expected behavior ``` nix-repl> ({ }: true) { a = 5; } error: … from call site at «string»:1:1: 1| ({ }: true) { a = 5; } | ^ error: function 'anonymous lambda' called with unexpected argument 'a' at «string»:1:2: 1| ({ }: true) { a = 5; } | ^ ``` ## `nix --version` output Lix 2.94.0-dev. Happens also in CppNix. ## Additional context This caused a silent bug in the functional1 test suite with non-UTF8 inodes, I passed the wrong argument `nonUtf8Inode` instead of `nonUtf8Inodes` and generated _NO_ non-UTF8 inodes. Tough luck.

raito added the

bug

label

2025-06-21 22:37:47 +00:00

lix-bot commented

2025-06-21 22:58:31 +00:00

Member

This issue was mentioned on Gerrit on the following CLs:

commit message in cl/3407 ("libstore/binary-cache: fix catching JSON exceptions")

This issue was mentioned on Gerrit on the following CLs: * commit message in [cl/3407](https://gerrit.lix.systems/c/lix/+/3407) ("libstore/binary-cache: fix catching JSON exceptions")

jade commented

2025-06-22 02:53:22 +00:00

Owner

This is because of autocall, which is really jank. I think if we wanted to fix this, we could track what was autocalled and what with and then generate a diagnostic if an input was never used. It's plausible that we can't make this a hard error as autocall is at any level: lix tries to call every function it runs into with the given arg in a kind of sketchy way, so it's very possible that a lot of people are relying on this unfortunate design.

👍 1

helle commented

2025-06-22 06:49:38 +00:00

Member

There is however the question, do we want people to rely on this design? Can we come up with a timeline along which we can eventually turn that diagnostic into a hard error, even if it is going to sometimes painful as we only know it at the far end of eval that we have not consumed all the args.

That doesn't sound like it is unreasonable to emit a diagnostic for in nearly every case, unless people have come up not just relying on this behavior by accident, but as intentional design (ie, they would get spammed with warnings from that). There is however the question, do we want people to rely on this design? Can we come up with a timeline along which we can eventually turn that diagnostic into a hard error, even if it is going to sometimes painful as we only know it at the far end of eval that we have not consumed all the args.

jade commented

2025-06-22 08:33:45 +00:00

Owner

I don't think this is the first bug report about autocall either, pennae thinks it's evil and it's been on their shit list forever. The last one I'm thinking of is relating to --arg and --expr not liking each other iirc