Though ostensibly compatible with typst, there is a show-stopping issue
with the main image.nvim fork which means an error message whenever
there is more than one typst buffer open (e.g. two side-by-side, or
opening a fzf-lua windows with typst previews, and so on).
Until that is fixed, use this fork. Issue here: https://github.com/3rd/image.nvim/pull/280
Dotlink now takes arbitrary arguments which it all simply passes through
to dotter underneath.
Also made the directory changing a little more flexible.
We add an additional 'core' user service (i.e. one that gets loaded
before all others by turnstile) which populates the TURNSTILE_ENV_DIR
with all manner of custom set env vars that are important for other
applications.
Most importantly, this sets up the XDG directory compliance for
applications either managed by turnstile or applications started through
turnstile on my system. So, for example `pass` knows to search for its
database in `XDG_DATA_HOME` and river knows to search for binaries in a
PATH which has been prefixed with my custom user binary location.
Everything is working well again with newer versions of blink-cmp. And
the development velocity is crazy, we are already two major versions
ahead. Some breaking changes but seemingly nothing that my relatively
simple configuration is affected by.
Remove most of the old cruft that was left over from nushell version
0.87 - we are now on version 0.102!
Many of the old options are actually not helpful anymore so let's just
get rid of them entirely (never configured the shell for me too much).
Also there was a lot of 'default' commented code which made it harder to
keep an overview rather than help.
For now just set up a minimal shell experience with vi editing mode, and
the trifecta of startship prompt, zoxide movement and atuin history
enabled.
All additional languages features (LSPs, treesitter parsers, linters and
formatters) are now defined in a single place in 'core/languages'.
This file simply sets up a big table which contains all the enabled
programs and parsers, divided by type. They adhere to the structure
given by the respective plugin.
HACK: We are still cheating a bit currently for treesitter parsers since
I have not had the heart to go through all of them to
activate/deactivate what I could need. Most of them are simply still
loaded, not connected to a specific language. Will have to be sorted out
at some point but it is good enough for now.
Was not installed correctly, and neither set up correctly. Now should be
fully working, with objects targeting Functions, Loops, Conditionals,
Statements and (nushell) Pipelines.
Seems to have since been transferred to nvim-treesitter group ownership.
Probably still worked under the old link, but better to be safe and use
the up-to-date source.
Have gotten LSP, Linters, Treesittesr into one list under core/languages
Missing formatters still, then it's done.
By default sort the exported output by creation date and modify date.
Also use the jq expanded (prettified) output. Will take a little more
space, but ultimately makes it easier to see task changes since each
value is on a single line.
Using external luarocks instead of a luarocks plugin for now. The amount
of dependencies to install is not smaller the other way now (having to
install libreadline-devel) so this should be fine.
I have captured more info in daily log 2025-03-11 on my reasoning.
Suffice it to say: both lazy.nvim 'hererocks' installation, this
external luarocks dep and the luarocks.nvim plugin are brittle in their
own ways.
For now, I have settled on the external dependency as it remains the
simplest way to achieve what I want (image.nvim) pictures. Though it
requires quite a few dependencies overall:
`xbps-install lua51 lua51-devel luarocks-lua51 ImageMagick libmagick-devel`
Since the starting parameters are fairly hardwired for wezterm anyway I
don't see this making much sense with e.g. xterm. However, long-term, we
should rather change the hard-dependency instead of doing it this way.
Instead of finding the notebook at the wikiroot, we directly change the
current working dir to the notebook for the runtime of the zk command.
This has a couple advantages: Any other commands executed will also pick
up the correct directory, and creating notes in a specific directory
also just _works_ by giving a directory in the notes dir even if we are
somewhere else.
Added a new state which should fix the icon spacing issues:
When we have no upcoming events or upcoming events but none today, we
only display the icon and so we do not add any additional spacing.
(This is alt state `event` or `no-event`)
Only if we have an upcoming event today (alt state `event-today`) are we
printing it directly on the status bar and only then should be have
additional spacing. So we have an icon (the same as for event) with the
correct spacing so that whether there is text on the statusbar or not,
we space correctly.
On my current voidlinux system I have no access to elogind and riverwm
will just be restarted when I exit it (since it is watched over by a
user service).
My session is started by the display manager 'greetd' (which in turn
runs tuigreet but that is not important here). So, to log out of the
session - we do a search for the 'greetd' process and if we find it we
search its _child_ processes which are the ones that we can kill without
being a super user.
If we kill the right process here, the session will close (since the
'watched' process exits and greetd just takes over again). This would be
a little easier if we started riverwm through greetd directly but I like
this indirection for stability (automatic service restarts), flexibility
(I can stop the riverwm service independent of my session), and adaption
(we can use Turnstile environment variables between my session
services).
So we just close all the processes that are children of greetd --
perhaps we accidentally catch a process which is _not_ the root session
process run by greetd but we'll kill it nevertheless since all child
processes might stop greetd from taking over (and they would be closed
anyway when ending the session afaik).
I have no linger enabled on my system so far (user processes staying
enabled after log-out) and I do not know how that would interact with
this method.
Running pass-pick (or other gpg requiring software) through river was
not correctly accessing the gpg agent since being managed by runit user
service did not point them to the correct gpg agent socket.
This commit fixes it using keychain as auxiliary software to manage a
session-long gpg agent process which is also exported into the turnstile
environment so other runit services can make use of it.
While there are errors that pop up when inserting text (_each time_), it
is still much preferable to the unbearable slowdowns that happen after a
while when running it in XWayland mode.
Accidentally hitting q was just a little too common for me so now at
least I will have to accidentally hit two keys at once.
Regular q is completely ignored.
So far, we used the external `setxkbmap` tool to map the compose key to
the 'menu' button (AltGr on my keyboard). With this I am able to type
äöü€ß.
Now using the built-in option in riverctl which I just discovered, so we
rely on fewer external programs.
Also setting the keyboard itself to 'us(altgr-intl)' which I _think_ I
had enabled before. If it turns out not, well, can always change back to
us layout with a single line.
Since 612b92d1 we are matching private commits case insensitively, we
should also match default commit search insensitively by default.
This changes the alias to search without caring for the case. We do not
have an equivalent alias for case sensitive search, and I will only add
one if I need the specificity often enough to warrant an extra alias.
Summary logs are changed to a little nicer presentation (from --summary
to 'builtin_log_compact_full_description' template).
Oneline logs have a changed order. Keep the revID as first thing, but
then show bookmarks/tags/head and descriptions. Only afterwards
show email, timesatmp, commitID since I am less interested in them
generally when viewing oneline commits (to get an overview of the last
changes and my current position).
Passing them through as individual arguments on the commandline.
Previously quoting would pass them through as a single argument, making
it not work correctly.
Could switch to using bash arrays for this (to make it more robust
against word-splitting) but a) I prefer the portability of sh for such a
simple script and b) there _should_ never be a time when we pass through
anything that has weird word-splitting issues. Either the location is
passed through or the times.
We load the 'run-help' function to quickly show us documentation for the
command under cursor when we enter vi mode and then hit 'K'.
This mimics the actual vim setup where K will generally show
documentation/hover info/help as well.
The command invocation requires an 'even amount of arguments' which I
don't fully understand but have no time to read into and fix currently,
so it just gets another superfluous 'run-help' tacked on at the end.
The xdg-compliance function would check on shell startup if we have a
file mentioned by 'PYTHONSTARTUP' and create it if not.
However, if there is no env var in the shell calling for this at all it
would still try to create a file at an 'empty' path, so this obviously
won't work.
Added a check that we even have the variable to do work.
swayidle is now also presented as a user service managed by runit on
voidlinux. It comes with the same defaults as before (300 seconds to
lockscreen, 600 seconds to screen dimming and 900 seconds before
suspending).
Additionally the lockscreen script has been updated to correctly tell a
wayland from a non-wayland session without logind being available on the
system, though it still defaults to using loginctl if it finds it.
The service runs as swayidle in the user services directory and can be
confirured using a 'conf' file which would be placed in the 'swayidle'
service directory. Timeouts can be set with `time_to_lockscreen`,
`time_to_screendim` and `time_to_suspend`.
Keeps using the known systemctl commands for voidlinux environments (or
other environments that have access to the `systemctl` command).
For voidlinux it instead uses the commands `poweroff`, `reboot`, `zzz`
and `ZZZ` (which are the same script). Works fairly naively for now and
does no checking if the commands exist or work. Runs required commands
as sudo so ideally the user or group has access to a passwordless
implementation of the commands.
Internally `jnb` runs `jj new -B@` so that it simply always insert a change
directly before the current working copy.
Likewise with `jna` for inserting after current. (Which is still
somewhat helpful since by default `jj new` will create a branch in the
tree if the new change is not the leaf node)
If `vsv` void service manager program is on the system, we also point it
to our own user service directory and make it accessible as `usv` user
service manager.
Fixed quoting so that we can now also search for things which would have
a meaning in jj, e.g. `jlf 'WIP:'` (jj would pick up the ':' and
complain before).
