Set inNixShell

[ncfavier]

nix-shell passes a inNixShell = true argument when evaluating a file, so we should replicate this behaviour. There is another mechanism for signaling that we are in a nix shell, namely the IN_NIX_SHELL environment variable (which both nix-shell and nix-direnv also set), but this is by nature not a good solution since it leaks into e.g. further nix-build invocations in nix shells.

This only concerns non-flake setups, of course.

ping @roberth @sternenseemann

[roberth]

EDIT: nvm the "migitation" - it's actually pointless

It did require adjustment in some shell expressions, where Nix can't trivially detect that the "auto-arg" is not accepted.
You could limit this to shell.nix files to limit the impact of that drawback, but doing so introduces complexity and a bit of risk on its own.
I'd accept this PR as is, and only implement that partial mitigation if it proves necessary.

[ncfavier]

Limiting this to shell.nix would pretty much defeat the purpose, which is to combine shell.nix into default.nix. Maybe that is a bad idea though, and in my use case I ended up writing a shell.nix.

I guess you're referring to NixOS/nix#5543? That was closed inconclusively; I'll try to replicate the problem there later.

[ncfavier]

So the problem in NixOS/nix#5543 was that inNixShell was previously simply added to autoArgs, but the heuristic for inserting autoArgs into function arguments was too eager, so an expression like

{ ... }@args: ({ foo, bar }: qux) args

would be given inNixShell and fail. This was fixed in NixOS/nix#6664 by modifying the heuristic specifically for the internally-generated inNixShell to only insert it for functions that explicitly take an inNixShell argument (e.g. { inNixShell, ... }: … but not { ... }: …). Of course it's undecidable whether an expression will crash when given an inNixShell argument, so heuristics are the best we can do.

This PR adds --arg inNixShell true to use_nix calls that do not use -p. So, while this would indeed fail for expressions such as the above (because --arg arguments are not subject to the modified heuristic), we are unlikely to run into the breakage mentioned in NixOS/nix#5543 which was specifically about wrapped nixpkgs expressions (which get evaluated when using -p).

Come to think of it, since nix-direnv calls nix print-dev-env maybe the proper fix is to make nix print-dev-env replicate the inNixShell logic of nix-shell.

[bbenne10]

I have to give this some thought and investigate the linked context before I make a call.

Based on what I am seeing so far though, I think I agree that print-dev-env should send inNixShell and I believe this is a reasonable compromise in the short term.

We are just coming up for air after a pretty hectic weekend of winter storms here - sorry for not getting back to you faster.