| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
|
|
|
|
The current setting (i.e. pointing into /nix/store) gives weird behavior
with `darwin-rebuild edit` since one cannot edit the readonly store.
|
|
|
|
This patch makes it so common configuration is now located in shared/
and each host basically just becomes a bunch of import statements.
The exception here is host-specific configuration like the `*.linus.onl`
that live inside `hosts/ahmed/`.
Specifically I have:
- moved common configuration `shared/{nixos,home-manager,nix-darwin}`.
- moved `hosts/common.nix` to `shared/nixos/common-{nix,shell}-settings.nix`.
- split `hosts/muhammed/{dev-utils,neovim}` into per-language
configuration.
This patch was done in preparation for the arrival of a new host, which
will need to share a lot of configuration with Muhammed.
|
|
|
|
|
|
In e18423, the builder was accidentally changed to be x86_64 which was
an unrelated experiment. I don't really need another x86_64 builder as I
have ahmed in most cases.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
- Bump perl input to pkgs/trash
- Use {swiftPackages. => }xcbuild for pbv
- Remove cling; was crashing and ) wasn't using it anyways
- LSP tsserver was renamed to ts_ls
|
|
This is a bit of an experiment. With this approach, I should be able to
remotely and asynchronously update this host.
|
|
|
|
I also moved pbv and trash from hosts/muhammed/home/dev-utils into the
new module, since they aren't really development related; I would still
want them on a Darwin system that *wasn't* being used for development.
|
|
|
|
|
|
This commit partially reverts 386d237f. I think this configuration line
was causing errors because my neovim version config didn't actually
support the feature.
|
|
|
|
I am tired of fighting with literally every single tool (neovim, git,
alejandre) to allow tabs inside the Lua code in the big string.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
After attempting to use this configuration on another device, it turns
out that the persistent data directories were being created with the
wrong permissions.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
I thought this restriction was already being enforced since I set
systemd.services."${domain}-source".after = ["network-online.target"]
but that apparently wasn't enough. I had to also ensure the timer, which
triggered the service was ordered after "network-online.target".
|
|
|
|
|
|
This patch sets up a Gleam development environment on Muhammed.
I also took this opportunity to move a whole bunch of
development-related filetype plugins to Muhammed, reducing closure size
on Ahmed.
|
|
|
|
|
|
CGit farms out the handling of the HTTP requests sent by the Git CLI to
another CGI script. This script was failing because of "dubious
ownership". This is a security check run by Git to ensure malicious
repositories on network drives don't get arbitrary code execution. The
problem is: the CGI script was running as root, as that is what the
fcgiwrap systemd service was configured for, but the repository is owned
by the 'git' user.
Since I trust the repositories, I had to patch Git to ignore this mark.
Actually getting the NixOS CGit module to use the patched version of Git
proved rather difficult...
In the future I should probably
a) Make sure fcgiwrap isn't running as root since it directly interacts
with all sorts of untrusted user input.
b) Remove this ugly hack. There's a reason for the security check after
all. Just because it matters _less_ in this case doesn't mean it's
wise to ignore it completely.
|
|
Conjure keeps getting in my way because it pops up when it isn't wanted
and rebinds my keys.
|
