initPC

🛠 A collection of C++, Python & Rust development-oriented config scripts to quickly init new machines 💻 for my personal use. Dotfiles are in a separate repository here: https://github.com/jan-revay/dotfiles and here: https://github.com/jan-revay/windows_dotfiles

How to use

First run on a new machine

1. cd ~
2. git clone https://github.com/jan-revay/initPC.git
3. cd initPC/
4. git checkout <branch> (optional step)
5. Run the initPC script launcher:
   • ./run_init.sh - on Linux distros or Termux
   • On Windows 11:
     • Install WinGet and run: winget install -e --id Git.Git (TODO automate)
     • Set-ExecutionPolicy unrestricted && cd Windows_10 && .\run_all.ps1 - (with administrator privileges)
     • Windows initPC was moved to its own repo here: https://github.com/jan-revay/initPC_Windows

✔️ Note: Logs will appear in the folder initPC/Logs/. Use cat <logfile> to display the log file with the original VT100 colors.

Applying changes in this repo to your machine

If you have updated the initPC or dotfiles repo, use the:

refresh

command (an alias defined in .bash_aliases) to apply the changes in the repository to your machine (e.g. install added packages). Keep in mind, that the refresh script only adds stuff, it never removes any packages (other than apt-get autoremove) even if those were removed from the initPC repo and it does not reverse already executed changes in the machine configuration (the Nix config will be able to do that, but it is not finished). Removing a package/config option from the initPC script does not affect the configuration of the existing machine after the script is executed for the first time. The refresh script however overwrites the old configuration options with new ones, if a particular config option was modified (e.g. a gsettings option).

OSes and distributions

Currently, the main focus of the initPC script is a well-tuned Ubuntu 24.04 with GNOME resp. CLI, and Windows 11. Some other Debian-based distros are also supported to various degrees, but initPC scripts for other OSes/distros are not maintained as much & the support for them might be completely dropped in the future.

FAQ

1. Is the refresh command idempotent?

Yes, the refresh command should be idempotent. This is also partly verified in CI via GitHub actions (TODO make sure and improve the test).

By idempotency, we mean:

Let:
R_C1 - executing refresh resp. initPC via ./run_init.sh on a commit C1
𝓢 - set of all possible states of an OS image
S ∈ 𝓢 - clean state (state after installing a new OS before running ./run_init.sh)
∘ : R_Ci x 𝓢 -> 𝓢 - application of the `refresh` command to the state of the machine

We want:

R_C1 ∘ (R_C1 ∘ S) = R_C1 ∘ S

<h3> 2. Does the <code>refresh</code> command (on a machine in any state) always produce the same state as running the script on a clean new OS install from scratch? Why not use Nix?</h3>

In general no. Only the Nix package manager can do rollbacks and updates (e.g. removing packages and configs) deterministically. It would be hard to replicate that elsewhere. However, the whole configuration of a new machine takes just around 20 minutes with the Ubuntu initPC scripts. Therefore if the state of an existing machine and the state described in this repo do diverge too much, it should be possible to reinstall the OS and reconfigure the machine from a clean state fairly quickly (TODO after the backup solution is also finished). Hence despite NixOS being more capable in this regard, it is still possible to achieve a similar practical effect using this init script. As Nix is not as widely used as Debian and Ubuntu, I have used Debian package manager as a default rather than Nix. However, I plan to implement the initPC script for Nix as well and I am also considering using Nix more in the future, for now, however, most effort is spent on tuning the Ubuntu version of the setup script.

By deterministic system configuration, we mean:

Let us have 2 commits:
C1 -> C2

And let:
R_Ci - executing refresh resp. initPC via ./run_init.sh on a commit Ci
𝓢 - set of all possible states of an OS image
S ∈ 𝓢 - clean state (state after installing a new OS before running ./run_init.sh)
∘ : R_Ci x 𝓢 -> 𝓢 - application of the `refresh` command to the state of the machine

We want:

R_C2 ∘ (R_C1 ∘ S) = R_C2 ∘ S

3. Why don't I use Ansible?

At the moment it would just add complexity, and unnecessary dependence (e.g. it could break Termux on Android), be less flexible, and would not be as beneficial for a project of this size. Also, I don't have any experience in writing Ansible scripts... However, I might rewrite some parts of the codebase to Ansible scripts later.

4. Why don't I use Chezmoi?

Dotfiles are currently managed via a bare Git repo (home directory) and Stow (system-level configuration files). In the future, I might use Chezmoi or a more sophisticated tool for managing dotfiles & templating them, if the need arises. See: https://www.chezmoi.io/why-use-chezmoi/

Branches

1. devel - development and experiments, might be inconsistent or broken regularly. Consistent, and fully functional changes from the branch devel might be merged into the branch testing. The devel branch is expected to be broken from time to time (e.g. when working on larger changes "per partes" or experimenting) and it might not always be possible to init a machine using it. New changes are usually pushed to the devel branch directly, however, very large changes can have an individual feature branch.
2. testing - shouldn't be broken or inconsistent most of the time. This branch has changes from devel queued to be accepted to the stable branch (or rejected). If a change is rejected from testing it will be dropped via a commit into devel that will be fast-forward merged to the testing branch again.
3. stable - tested, stable, useful, production-ready, and not expected to change more than a few times a year.
4. LTS - debloated, (also tested, stable, useful, production-ready) and not expected to change much in the yearly horizon. Only necessary stuff. Possibly useful for detecting whether bugs in the stable branch are caused by the init script or to be used as a substitute for the stable branch while the stable branch has a critical bug. Debloating is done via additional commits on top of the LTS branch, therefore syncing stable and LTS is done via rebasing to preserve the debloating commits on top. As the LTS branch has additional commits on top, it is tested separately.
5. feature-<name of the feature> - all feature branches should be branched off and merged to devel. Features and bugfixes of testing, stable, or LTS should always go through the devel branch first (following the change workflow below).
6. archived/<branch-name>-<YYYY-MM-DD> - branches archived before a push --force.

LTS, stable, and testing branches are expected to be always in a consistent state so that they can always be used to init a new machine e.g. VM or a bootable partition.

✔️ Note: By stable I mean free of unpredictable behavior and crashes, not as described here: https://medium.com/@gordon.messmer/what-does-stable-mean-4447ac53bac8 (TODO toread)

Change workflow

                    functional &         tested, stable, useful   not changing, debloated,
    impl.            consistent**          & production-ready        retested & stable
O---------> devel ---------------> testing -----------------> stable -----------------> LTS
|             ∧    ff-only merge             ff-only merge                rebase
| impl.       |
|             |
+-----> feature-branch
 large
 change

** "consistent" means, among other things, that all CI tests (implemented via GitHub actions) pass successfully.

Contributions

Bug reports and pull requests are welcome, if a feature is interesting but not useful for me, I will keep it in a separate branch, I might change my mind later :D . Issues and feature requests can be created here: https://github.com/jan-revay/initPC/issues

TODO

1. Merge and deprecate the InitNewPC repo InitPC repo on org GitHub and initAndroid repo (and my_pc_setup repo and all repos of this type I've created).
2. Merge with LogidCfg repo
3. Test the Windows setup script on a VM
4. Create aliases for PowerShelly
5. Try merging the apt, flatpak, and snap install commands
6. Have a look at popOS packages and add the useful ones to other init scripts
7. Design a system for applying the configs on all my machines once they are updated here.
   • implement refresh alias (with CI idempotency tests)
   • add notification to .bashrc if the initPC or dotfiles are not up to date
8. Add more C++ tools from here: https://github.com/cpp-best-practices/cppbestpractices/blob/master/02-Use_the_Tools_Available.md
9. Add Bats automated tests
10. Try adding NixOS
11. Create CI tests on GitHub
12. Todos from the repo
13. Make the core Linux init script Debian-based (i.e. other distros just add stuff to the Debian base init script)
14. Consider replacing the Debian variants (Ubuntu, PopOS...) with a single Ansible script with conditionals.
15. Do some research on whether snap and flatpak packages work in WSL resp. which alternative package manager to use in WSL
16. Consider running the whole ./run_all.sh script as sudo and removing sudo commands from the script.
17. Consider using http://www.bashbooster.net/, https://github.com/bevry/dorothy, https://www.chezmoi.io/ or similar libraries (see: https://www.chezmoi.io/comparison-table/ and https://dotfiles.github.io/utilities/).
18. Format to max 82 char