builtins.getFlake: Allow inputs to be overridden #11952
Conversation
This uses the same syntax as flake inputs in flake.nix, e.g.
builtins.getFlake {
url = "github:NixOS/nix/55bc52401966fbffa525c574c14f67b00bc4fb3a";
inputs.nixpkgs.url = "github:NixOS/nixpkgs/c69a9bffbecde46b4b939465422ddc59493d3e4d";
}
Note that currently it's not supported to set `flake = false` or to
use `follows` (because lockFlake() doesn't support that for CLI
overrides), but it could be implemented in the future.
Fixes NixOS#9154.
github-actions
bot
added
the
with-tests
| std::function<void(const InputPath & inputPath, const FlakeInput & input)> recurse; | ||
|
|
||
| recurse = [&](const InputPath & inputPath, const FlakeInput & input) | ||
| { | ||
| if (!input.ref) |
There was a problem hiding this comment.
It's not strictly necessary to use std::function and its inherent indirection for recursive lambdas. Though in this case this shouldn't matter much. With C++23 it's possible to use "deducing this" https://en.cppreference.com/w/cpp/language/member_functions#Explicit_object_member_functions. For pre C++23 one can use the following hack:
auto recurse_impl = [&](auto& self, const InputPath & inputPath, const FlakeInput & input) {
....
recurse_impl(self, ....);
};
auto recurse = [&](const InputPath & inputPath, const FlakeInput & input) {
recurse_impl(recurse_impl, inputPath, input);
};This avoids the virtual call via std::function
There was a problem hiding this comment.
Wow nice, didn't know about "deducing this". Apparently it will be available in gcc 14.
|
This pull request has been mentioned on NixOS Discourse. There might be relevant details there: https://discourse.nixos.org/t/2024-11-27-nix-team-meeting-minutes-198/56691/1 |
edolstra
changed the title
|
Added a release note. |
| flakeRef; | ||
| }) | ||
| : ({ | ||
| auto flakeInput = parseFlakeInput(state, std::nullopt, args[0], pos, {}, {}); |
There was a problem hiding this comment.
parseFlakeInput expects a "trivial" value, but this is not a reasonable expectation for args[0], which is part of the "normal" expression language where evaluation just happens.
(e.g, url = "git+https://" + foo; will fail)
Possible solutions:
- deepSeq
- gets the job done if all is ok, but is too strict: worse error message for errors in invalid (misplaced) attributes.
- force the relevant values here
- code gets out of sync with
parseFlakeInput, and bugs won't be found immediately, because literals and trivial values are common
- code gets out of sync with
- add a flag to
parseFlakeInputto allow it to evaluate when true- don't see much of a problem with this
My preference is with the parseFlakeInput flag.
Such a flag would perhaps also be useful for a feature where an "evaluated" flake can be passed as an input, e.g. getFlake { url = ./tests; inputs.foo.value = self; }, although this has also been requested as a feature for static flake inputs:
| @@ -99,7 +99,7 @@ static std::map<FlakeId, FlakeInput> parseFlakeInputs( | |||
| const std::optional<Path> & baseDir, InputPath lockRootPath); | |||
|
|
|||
| static FlakeInput parseFlakeInput(EvalState & state, | |||
| std::string_view inputName, Value * value, const PosIdx pos, | |||
| std::optional<std::string_view> inputName, Value * value, const PosIdx pos, | |||
There was a problem hiding this comment.
Could inputName be removed? An input shouldn't have to know its name.
I think this might be an issue with the data model, but it seems that we have low hanging fruit, in that we could turn the call with inputName into a separate overload that calls parseFlakeInput(state,value,pos) and postprocesses the return value.
Motivation
Fixes #9154.
This uses the same syntax as flake inputs in flake.nix, e.g.
Note that currently it's not supported to set
flake = falseor touse
follows(because lockFlake() doesn't support that for CLIoverrides), but it could be implemented in the future.
Context
Add 👍 to pull requests you find important.
The Nix maintainer team uses a GitHub project board to schedule and track reviews.