Moving our workspaces to projects still lets them be cloned down as
normal directories but additionally means that clone commands will look
like this

git clone https://git.freshlybakedca.ke/patisserie.git:workspace=projects/packetmix.git

rather than this

git clone https://git.freshlybakedca.ke/patisserie.git:workspace=packetmix.git packetmix

which is nice because it doesn't have the repository name required as a
separate argument

There were various problems with it before - it was only really
half-modified from the tangled.yml workflow we had over on GitHub!

It now works and has been tested to put up a branch with a test push

The footnote.social dev is interested in using our proxy. We're happy to
let them

Speaking of, we haven't run an npins bump in a while... let's do that

This is a nontrivial npins bump, because catppuccin has added a module
that isn't in the version of home-manager we're using so we need to
import it.

There was an update to the record format of spindle which meant adding
new repos was broken. This bump fixes that

Umber is a system that'll be replacing firebrick in the long term. I
need to transfer over the private silverbullet and set up backups on
here for it to fulfil that role...

Previously we were starting swayidle on niri. Unfortunately, this caused
a race condition where niri idle inhibitions were not respected. As niri
gets idle inhibitions from, say, browsers when playing video, this meant
we had to do nasty hacks such as manually systemd-inhibiting via a shell

By moving startup to systemd, we can start swayidle later - and in much
the same way as starting our SSH agent later avoids its race conditions,
starting swayidle later fixes this issue...

deadnix is a package to find unused nix code, we can add it to enforce
that we don't leave let bindings/inputs/etc. around when they are not
needed

We previously had these records, but we lost them when migrating
stuff...

They are internal mirrors of some external mostlyturquoise records

We want to add some patches to our PDS to enable SSO - and those patches
need us to be on a later version. Therefore, let's upgrade!

bluesky-pds is packaged in a rather interesting way
- The vast majority of the code is in a different repo which is normally
fetched with pnpm - we need to twist stuff so that we build it with
nix

This is part of a series of atproto migration patches. Our plan so
far is:
- [x] Set up a tangled.org spindle
- [x] Set up a PDS
- [ ] Pull in upcoming external idp patches for PDS
^ You are here
- [ ] Set up email for our PDS
- [ ] Set up a tangled.org knot
- [ ] Rename tangled.org spindle to spindle.freshly.space

Internally for Freshly we use Tailscale to access hosts, but Collabora
now also use Tailscale to access hosts. For a while, I've been using
'tailscale switch' to move back/forth but this takes some time and
doesn't allow me to use multiple nets at once.

I evaluated what I wanted to use my own personal tailscale for, and it
was the following things:
- https://files.freshly.space (I have a mounted webdav drive which is
only available over tailscale, and the web interface auto-logs-in over
tailscale which is very nice...)
- https://silverbullet.clicks.codes (All of my notes are here. I would
need to switch notetaking app for work stuff if I were to stop using
my own tailnet)
- https://silverbullet.starrysky.fyi (Actually, some work stuff is also
here because I have some things which I am contractually obliged not
to make public - even to my friends. This one does tailscale auth to
check that I am my own tailscale user)
- My own devices (which I could put on to the work tailnet, although I
would either forgo nice device names or manage my own /etc/hosts and I
would need to manage switching back/forth (potentially making any
/etc/hosts editing pretty fraught...))

While some of this could be put on the work tailnet, by no means all of
it could be.

In contrast, I need the work tailnet to access SSH/web interfaces for
several internal services. These can't be used without it, but I only
really need to use them from redhead.

The cleanest solution is to have multiple tailnets at once - and select
which one I'm using such that the collabora tailnet is only used for the
things it's needed for.

Here's how I'm proxying SSH hosts through the tailnet

Host collabora-foo foo
ProxyCommand nc -X 5 -x localhost:1055 %h %p
Hostname <some tailnet hostname>
...

And I'm using the "FoxyProxy" extension in Firefox to do the same for
specific URL matches. I'm hoping this'll be enough to let me neatly
access everything I want to all of the time...

We'd like to move off bluesky's PDS, and the obvious candidate is to set
up our own...

...we've chosen to have our handles *.at.freshlybakedca.ke because they
are technically publicly visible - although we expect to be using custom
domains for most/all accounts on this PDS

...we've chosen to have the pds at pds.freshly.space because, like
files.freshly.space, while *technically* publicly usable (file shares
or accessing our bsky accounts), it is only going to host accounts of
friends and patissiers of freshlybakedca.ke. There is no process to get
an account on our PDS as a general member of the public. It also isn't
generally particularly visible in public UIs. (This criteria also means
that spindle.freshlybakedca.ke should be at spindle.freshly.space, which
we will migrate in the future. Other than that everything we host meets
this criteria)

This is part of a series of atproto migration patches. Our plan so
far is:
- [x] Set up a tangled.org spindle
- [x] Set up a PDS
^ You are here
- [ ] Set up email for our PDS
- [ ] Set up a tangled.org knot
- [ ] Pull in upcoming external idp patches for PDS
- [ ] Rename tangled.org spindle to spindle.freshly.space

While we created the release bookmark fine, we can't push to it unless
we have a non-shallow history...

We just set up a binary cache with nix-serve - let's use it!

We'll keep the old cachix around for now, but it's considered deprecated
:)

Nix-serve is [a cache server](https://github.com/edolstra/nix-serve)
that sets up your nix store to be served as a binary cache. As we're
not going to be using cachix anymore as we are on tangled, we need to do
this to have a cache

I accidentally used the old .sh url when setting this push url. We're
using .org everywhere, let's switch over...

I typoed this command - and unfortunately we can't easily test this
without pushing to main...

We're remote building in CI to get around Tangled's limitations for nix
(and to avoid us having to figure out how to make nixery play nice with
a mounted nix daemon...)

Therefore, we'll need a remoteBuilds key that our Spindle can use

There's a race condition here where teal sometimes looks up midnight
when not connected to tailscale. If it does that, it resolves midnight
on the local network. That would be entirely fine if we weren't just
listening on Tailscale.

Further, that lookup can then get cached, bringing down the spindle even
when tailscale comes up

We trust the local network, let's just allow this route too...

Spindles don't really like it when you delete their state every
time they turn on. It tends to lead to lost logs and repulled nixery
containers. Let's fix that...

We've overloading midnight with our CI - let's stop ourselves from
trying to do more things than midnight supports at once...

On GitHub, we had a workflow which released packetmix when we built
successfully on main - avoiding rebases breaking release builds/etc.

Let's do that again here :)

When we push, josh sometimes seems to pull before it pushes. That causes
our SSH key to be used twice, triggering a double pin entry/etc. with a
security key. That's mildly annoying...

Luckily, we should always be able to pull over http, so we can just use
that for all pulls whether we're nominally using SSH or not!

Tangled has renamed from https://tangled.sh to https://tangled.org

https://bsky.app/profile/tangled.org/post/3lz5dmdtl4s2s

We love that for them - but Josh doesn't properly follow the redirects
for http URLs so we'll have to change them...

Jujutsu's templates for this are for how far ahead/behind the remote
branch is than compared to the local branch vs how far ahead/behind we
are from the remote branch... it makes more sense to have this the other
way around (and clarify in the comment...)

I've attempted to use colors that are fairly close to the original
colors, and use the 8-bit shell colors where possible so these should
change nicely with different terminal themes...

I've not changed colors of anything that looked related to a specific
project (e.g. python logo colors) though I *did* change colors of some
projects that didn't seem to particularly identify with the color picked
(taskwarrior). I'm sure I've done something wrong. Oh well :shrugging: -
maybe we should slim down/split up this p10k file as much as possible...
we don't use most of the things here anyway...

I've not tested this on any theme except latte... perhaps
at://thecoded.prof could take a look at how this looks on dark themes?

It'd be nice to use nix output monitor for builds - there's a project
which automatically uses nix output monitor when nix is called (even
supporting nix-direnv/nixos-rebuild)... this looks probably good for us
to use.

There is [one issue](https://github.com/ners/nix-monitored/issues/3)
of some importance on the nix-monitored tracker which suggests that
nix-monitored incorrectly transforms some 'nix run' invocations...
fortunately, we don't use flakes so this shouldn't be a huge issue

If it is, using

NIX_MONITOR=disable

will temporarily disable nix monitor, allowing you to run whatever you
needed without issue while using nom for everything else

So, we've done some things here...

- We're no longer evaluating homes - which was basically a double-eval
anyway until we get MacOS/etc. up
- We're splitting system evals apart from each other, which will take
longer over all but reduces the peak memory usage from >10GB to ~3GB
- >10GB was unsustainable for midnight ... we were constantly OOMing
when we accidentally triggered CI twice
- ~3GB is very sustainable for midnight :)

ci.nix has different licensing requirements to nilla.nix, namely unfree
packages are not allowed to be built at all in ci.nix. Therefore, it's
good to use it over nilla.nix to avoid accidentally building a package
we cannot distribute

I accidentally merged something in without formatting it locally when
our CI was down - here's that reformat...

We previously didn't use atuin because it was printing garbage
characters in bash... by switching to zsh we seem to have avoided this
problem :)

I've installed zsh, including
- zsh4humans
- powerlevel10k

This is because
- I'm interested installing atuin, and we've had some problems using it
with bash
- I'm interested in adding jujutsu information to my prompt, and zsh
will let me do this much more easily than bash

This isn't an ingredient of its own yet because it doesn't integrate
nicely with, say, catppuccin (or really anything that isn't itself) but
it's a start...

Timestamps currently display only the time, which is... fine, but can
be somewhat annoying to see how long ago something actually was. It's
possible to use timestamp.ago() to show how long ago a thing was as
well. Let's add it to our default timestamp formatter so we can see that
everywhere jujutsu shows us timestamps

We recently reset midnight - and in doing so changed its host key. We
need the new one for remote builds...

- As we've adopted josh, I've needed to type :work (the start of
:workspace) a lot more than I was expecting
- Coded often co-authors stuff with me and I'd like to type his credits
easily
- Speaking of which, I only made ways to add my own co-author credit
but didn't make an easy way to type a blank trailer to credit someone
else...
- And finally, coded also uses the same companies email scheme, and I
occasionally need to type his emails for administrative reasons...

I do Android development - the tools for it aren't in packetmix (see
https://github.com/CollaboraOnline/nix-build-support) but we still need
to persist the data here...

We've been using github CI for a while, let's translate everything to
tangled format so that we can move across!

We need to put our CI in the root of our monorepo as otherwise it won't
run on our tangled spindle...

We've been using https://reuse.software in packetmix for our licensing,
but we haven't made it work in the monorepo and we haven't got sprinkles
under it. Let's change that

We are moving over to tangled and want to have a spindle so that we can
run long builds like PacketMix needs :sweat_smile:

Midnight is our CI server so it makes sense to host the spindle on it,
proxied to the outside world via teal

We're reinstalling midnight! Previously we had the old hardware config
here, but that one didn't support impermanence and was pretty different
to teal's. As part of our reinstall we've repartitioned our disks to
have a much more similar layout...

We're reinstalling midnight, which means these routes will no longer
exist. They've all been migrated anyway...

When we start up headscale, it tries to connect to OIDC. If our OIDC
server is down, there were previously two options:
- We can fail to start the server altogether
- We can fallback to CLI auth and lose the ability to use OIDC until
headscale is restarted

Neither of these are what I want: I want us to start up anyway but not
allow registration until OIDC successfully connects

I've made a patch to do just that! I made it ontop of main, so I've also
had to upgrade headscale to allow use of this patch.

I had some troubles with nilla, which can't properly interpret the
headscale flake (missing rev/shortRev attributes) - therefore I've taken
the nix package definition from the headscale repo and updated it to use
the patch and the correct source manually

Refs: juanfont/headscale#1873

Officially switching over to builtin niri+xwayland satellite integration!
Yay! The update that brought the integration caused issues when launching
X applications due to ordering conflicts since Niri now sets `DISPLAY=:0`

This fixes a regression from tqywkonpmxyulxkyuyusqntrwomtkkxu
(9d7dfaabeb1b2cf9ddb4219a064e75b270b7926a). I accidentally put the
WantedBy in the Unit section rather than the Install section. This left
the unit starting on graphical-session.target rather than niri.service,
which prevented it starting due to the ordering loop...

...oops

Coded added a stopgap solution for walker starting before elephant. We
were still getting pretty bad startup performance, though, so I looked
into it.

It turned out a variety of things were needed to start walker faster:
- Moving to a systemd service was helpful, but only once walker had
started once, otherwise the socket would never connect and walker
wouldn't open at all
- Disabling file indexing by default was necessary for systems with lots
of files in home (for example LibreOffice clones). If you want to add
file indexing back for yourself, I recommend you figure out where your
large files are and use '.ignore' files (.gitignore syntax) to ignore
them
- Elephant needed to be started properly - after niri so WAYLAND_DISPLAY
was set but not in the ordering cycle on graphical-session.target. I
used the same settings as we have for ssh-agent, say

Because of all of this, I also factored walker out into its own file.

This additionally marks

- The first keybind moved outside of the niri.nix file
- The first keybind to be created with Mod rather than ${mod} (= Super)

While it's a bit of a shame to not have all the keybindings in the same
place, I think it's worth it...

Recent updates to Walker (1.0.0) have broken the application. Walker now
uses elephant as a backend service for finding apps, clipboards, and
other things you may want to search but it does not start that service on
startup, we have to do this manually.