[WIP] +RFC fixing arguments parsing and documentation for nix-build and nix-shell #2346
Conversation
They are described as being passed to nix-store --realise later in the documentation so it's fine to add only to synopsis.
|
❤️ |
|
Any reason to attempt to address #2208 in this PR rather than only separating the options and implementing the shell-gc-rooting stuff in another one? With that considered, is this still WIP? |
|
The only reason I thought about #2208 was that it was possible to do it while passing there. It wasn't the initial goal of the PR, and yeah, probably should be handled in another PR. The actual PR, "fixing arguments parsing and documentation" isn't much WIP, but more "tell me what I'm doing wrong". See the "Expectations" and "RFC stdin" sections for other parts where I'm not entirely confident. |
|
I think it would make sense to split out the documentation part from the PR and fast-track it. Additionally, I think it would be better to handle #2208 in a separate PR too. This has been sitting around for a while already and the option housekeeping stuff is pretty useful! |
|
I think such PRs would be eaiser to merge if they come with a bunch of tests :) I agree with @asymmetric on the docs. |
|
I marked this as stale due to inactivity. → More info |
nix-buildandnix-shellare both implemented in the same multi-call binary.The current implementation for arguments parsing makes them share arguments. This can lead to surprising behaviour (or lack thereof) when accidentally using a non-shared parameter.
Abridged list of nonsensical invocations:
nix-build --command "foo" ./release.nixnix-build --impure ./release.nix -A build.x86_64-linuxnix-shell --out-link "foo" ./shell.nixThere are a couple more that could be added.
This PR intends to fix the behaviour by segregating the arguments parsing using an
ifconditional. That is, unless a better method is suggested.Additionally, this PR adds the missing documentation for a couple of options, since while checking for the "owner" of the parameters, I realized that a couple options weren't documented.
Expectations
I am expecting a bit of push-back from "simply removing" the options. That is, it technically breaks backwards-compatibility with (wrong) usage of the 2.x series for nix-build and nix-shell. This has to be kept in mind.
If it is judged better to keep the non-sensical options, it may be preferred to keep them segregated in their respective binaries, turned into no-ops in the other and documented as such.
RFC stdin
I find the
-(stdin) behaviour surprising with regards tonix-shell.nix-shell ./shell.nix!=echo 'import ./shell.nix' | nix-shell -As for
nix-buildit seems the behaviour is fine.I am probably overlooking something with the semantics of it. Though, the way it's implemented it may be better to drop this flag in
nix-shell? The only documentation for-as standard input is for nix-instantiate, and there are no mention of it in neither thenix-buildnor thenix-shellmanuals. (They mentionnix-instantiate.only for--arg | --attr | -A.I may split this PR into a documentation-only one that can be fast-tracked. I may also split this PR furthermore while it's WIP.
Furthermore, this will (still a WIP) try to address #1982 and #2208 by actually implementing the flag.