Regenerating the classes from json schema as part of the clan/cli is a bit redundant since we need to vendor the classes by hand anyways
The user can also not practically override the inventory options
I would disable this for now; until we really need it.
We now solely rely on CI checks failing if the classes are out of sync
Some test were referring to the whole source code via ${self} which amde them rebuild on every single commit.
This is not mitigated by introduceing `self.filter { include = [...]; }` allowin to a content addressed subset of the source code in tests.
- generate a flake.lock file for each template by copying the clan-core flake.lock and modifying it
- call nix build with --store for tests inside the sandbox
Many dependencies of clan-cli are currently dynamically loaded via nix-shell on each execution.
This is nice, as it reduces the initial closure size of clan, but the overhead introduced by nix-shell piles up quickly, as some commands shell out many times during their lifetime. For example, when adding a secret git is called 10+ times.
This reduces the time of a test which adds a secret from around 50 seconds to 15 seconds.
- add run_cmd() as an alternative to nix_shell()
- introduce the concept of static dependencies which do not need to go through nix-shell
- static dependencies are defined at build time and included into the wrapper for clan-cli
- add package: clan-cli-full which statically ships all required dependencies
TODO: deprecate nix_shell() in favor of run_cmd()
Currently, important tools from the default shell are lost as soon as select-shell is used to switch to another shell. select-shell itself, for example is missing, which makes it impossible to switch back to another shell.
Also done:
- add @pytest.mark.with_core indicating that a test depends on clan-core and cannot be cached
- separate unit tests into two derivations, ones that can be cached and ones that cannot.
- fix frontend typescript code
