Expose finalPackages
#68
Conversation
roberth
force-pushed
the
expose-finalPackages
branch
from
54f67c8
to
5708b2b
Compare
flake-module.nix
Outdated
| type = types.raw; | ||
| readOnly = true; | ||
| internal = true; | ||
| default = lib.composeManyExtensions [ |
There was a problem hiding this comment.
The difference is the mkOverride priority that is assigned to the definition. In case of a readOnly = true option, it doesn't really matter which one you use, because only one definition is allowed anyway, making override priorities irrelevant.
I think placing definitions in config is a good habit though, so I'll change this.
There was a problem hiding this comment.
I think placing definitions in
configis a good habit though, so I'll change this.
Cool; it would also minimize the diff in this PR, I imagine, since all config stuff lives in haskell-project.nix.
There was a problem hiding this comment.
To move it back, I'd have to change haskell-project.nix to a module instead of a function to a config value. Basically this https://github.com/srid/haskell-flake/pull/63/files#r1100250488
I guess why not
roberth
force-pushed
the
expose-finalPackages
branch
from
5708b2b
to
c93938e
Compare
runtest.sh
Outdated
| nix --option sandbox false \ | ||
| build --override-input haskell-flake path:${FLAKE} \ | ||
| -L --show-trace \ | ||
| github:hercules-ci/flake.parts-website#checks.x86_64-linux.linkcheck |
There was a problem hiding this comment.
I'd remove -L --show-trace since they are not used in the commands above (or is there a reason for their inclusion, aside from debugging purposes?). Also, must sandbox be disabled for this build?
There was a problem hiding this comment.
The sandbox was disabled in an earlier command, so I assumed that you run this in a setup where that's necessary. This build works fine in the sandbox, but I've only ever built it in a nice NixOS-based sandbox.
The verbosity is helpful when this runs in CI. You'll only look at it when it fails, and then you want all the info. Module system errors tend to need --show-trace, at least until we have
There was a problem hiding this comment.
A proper CI solves this problem in the UI, but admittedly consuming an unpinned input like the site here isn't possible yet in Hercules.
There was a problem hiding this comment.
Anyway, I've changed it so that logs only appear when there's a problem. Still not great if the builder hangs, but that's unlikely to be a problem here.
1000 lines should generally be enough to show the whole log. If more, you'll be iterating the problem anyway; it doesn't log that much and extra log lines are proportional to the problems found.
flake-module.nix
Outdated
| @@ -71,7 +71,7 @@ in | |||
| }; | |||
| }; | |||
| }; | |||
| projectSubmodule = types.submodule (args@{ name, config, lib, ... }: | |||
| projectSubmodule = types.submoduleWith { specialArgs = { inherit pkgs self; }; modules = [ ./haskell-project.nix (args@{ name, config, lib, ... }: | |||
There was a problem hiding this comment.
All this happening on a single line is ugly, but I think almost all of the things that would otherwise be reindented below can move to the haskell-project.nix module later. Doing that here would cause a large diff (again).
Keep all project config in haskell-project.nix
- `-e` is not recognized by macOS's zsh's echo command - Use current system when building the check
|
|
||
| finalPackages = mkOption { | ||
| type = types.attrsOf raw; | ||
| readOnly = true; |
There was a problem hiding this comment.
It looks like readonly has no effect, because I can still set finalPackages.foo = ./.; (even finalPackages = ./.;) in test/flake.nix?
Or does it just silently ignore user values?
#52
Access to
finalPackageswas useful for some troubleshooting.This also adds the docs to CI and adds a script to preview the docs locally.