Looking back at this, meta packages should be one of the most convenient
ways to keep track of packages. It gets the job done and should be more
than sufficient for our use case.
In particular, installing all packages guarantees that we won't "miss
something" and have to install it manually, which is more useful than
having a lower package number count.
The main disadvantage is dealing with constant updates to large packages
in a restrictive internet environment, although it's possible to
mitigate this by separating the smaller packages from the larger ones.
Having a meta package was cool and it got the job done, but it's
inconvenient to use when adding and removing a lot of packages.
One alternative I'm looking at is simply keeping track of all the
explicitly installed packages and storing that in a text file. This
makes it easy to keep track of all the installed packages without
introducing downsides, and new machines that don't need certain
packages can simply delete those lines.
Over time compton became unmaintained and a replacement package picom
took its place. After trying out sway for a bit, I realized that it
doesn't need a separate compositor at all like bspwm does, so I might
just switch to it. Note that there is a performance penalty on sway
that I haven't figured out how to solve yet.
evince actually uses less memory than zathura and seems to be more
efficient overall, although it isn't as customizable as zathura and
not as minimal in terms of UI.
Overall, I'd rather just use zathura, which also lets me be more
consistent in my bspwm setup.
Apparently gnome-books and sushi depend on evince, the first of which
is a GUI for djvu/epub files and the second of which lets you preview files
with the spacebar in the file explorer. Ironically, I've never used this
feature until I read about it, and although it seems cool, I don't think
I have a use for it as I've been opening my files normally for years now.
I actually wrote this in 2018 but never committed it. Might as well do
that now. The extensions it installs are uBlock Origin, Vimium, and
HTTPS Everywhere.
I wanted to commit some more stuff for 2020. Better late than never,
right? The most significant change is probably in fish_prompt.fish.
I fixed an edge case where the directory in question could be the
same as the user's username.
Changing between desktop environments is no longer a feature since
it's easier and more convenient to use only one environment, although
it's still certainly possible for the determined user.
Note that plasma-integration is used for KDE theme support without
installing the entire plasma-desktop. The breeze_cursors theme is
included in the breeze package, and xorg-xsetroot is used to change
the default X shaped cursor to a pointer.
Since I figured out how to run both KDE and GTK applications with
their proper settings under any lightweight window manager, and since
I now know how to configure these programs independent of their desktop
environments, this commit removes plasma-desktop, with the possibility
of gnome-shell bring removed in a future commit as well.
The rationale is that using a desktop environment is counter-intuitive
when you already use a window manager. Having multiple desktop environments
to start a session with is certainly amusing, but not at all practical,
especially when you can accomplish anything that needs to be done with
any window manager.
Switching between GNOME, Plasma, and bspwm also caused some inconsistent
X settings that I assume are non-trivial to fix without a proper restart
of the X server.
Simply starting the Plasma desktop environment will create quite a few
files in ~/.config. These files consist of window coordinates and other
non-friendly things, which, IMO, should not belong in the *user's* config
directory.
Regarding other software, GNOME MPV is arguably easier to use and less
buggy than baka-mplayer. Manipulating archives is also much easier with
file-roller than it is with ark. Since I got Audacity to look decent on a
HiDPI display, I no longer have a use for Kwave, which did not have some
of the essential features I used in Audacity anyway.
l3afpad functions much more like a notepad than kwrite, which is what
I'm looking for since I already use Vim. Using Cantata as a mpd client
is simply not the same as using ncmpcpp. zathura and evince are more
than enough for document viewing, hence the removal of Okular.
This is the PKGBUILD I made for a [nice patch][1] that adds border
radius support to bspwm. The patch has not been merged upstream yet,
so this PKGBUILD will have to do in the meantime.
[1]: https://github.com/baskerville/bspwm/pull/856
Since anyone using Tari will probably want the web browsers too,
this change makes sense.
Since all of Tari is now in one PKGBUILD, it is easy to see everything
that will be installed, and it becomes trivial to add or remove packages
as needed.
Updating firefox extensions manually is not ideal. Additionally,
this method is non-trivial to apply if the target system does not
use pacman as the package manager.
Since Tari will work as an independent PKGBUILD now, and since
xeventbind is an individual program not related to Tari, it makes
sense to remove the associated prefix.
Since I don't have to worry about the completeness of individual desktop
environments, this makes it easier for me to tailor the tools I use to
my use cases.
Since the collective system is known as Tari, having individual parts
for the pieces that work together to make the whole is not ideal.
Now that I understand more about how Arch Linux works, if I ever needed
to create a non-Tari operating system, I would just use pacman to install
packages as needed, or modify this PKGBUILD to create a new meta package
specific to that system.
Not all of these packages may be useful to everyone, and that's fine.
Any list I create won't satisfy everyone, but reducing the number of
PKGBUILDs makes it easier to see and change all the packages installed
on the system as a whole.
Due to the nature of how Arch Linux works, most users won't even need my
PKGBUILDs. With the exception of setting up X with a bare-minimum window
manager, installing and configuring most software is as simple as using
pacman to install that software and symlinking any config files you want
to their relevant directories, if any.
There is an open bug that should be resolved before or during the
Arch Linux Bug Day coming up next month (January 2019). Either way,
manually updating the httpie version becomes impractical after some
time; it's best to apply this fix upstream instead.
Since my .vimrc already installs the same version of vim plug,
and since any user can symlink my dotfiles, using a PKGBUILD to
install it is redundant and arguably less portable.
As ironic as this may sound, less user choice means that there's less
room for error. Plus, it makes things conveniently easier for me and
possibly every other user of the software.
Realistically, if I'm installing the tari packages, I more often than
not want everything with it. If I am aiming for a minimalist setup
without X11 or Wayland, I will probably not use the tari packages
anyway.
It turns out that the README for the xeventbind PKGBUILD actually works
very well here. On another note, I've yet to figure out how exactly I
want to document PKGBUILDs.
Since fonts and other desktop-related packages are now handled by
tari-desktop, this is no longer necessary. Note that I also created
tari-web before I figured out HiDPI support for Qt and GTK applications.
I initially created tari-util as a package to be pre-built and installed
in the installation media. This was a while ago, and now that I know
exactly how the installation process and PKGBUILDs work, all the
previous issues I had with makepkg turned out to not be issues at all.
As much as rofi-pass may appear to be useful, it is simply easier to
just use the pass command directly, especially if you have a passphrase
set on the private encryption key you use to unlock passwords.
Now that I know how to use terminal multiplexers, and now that I
understand more about how shells and terminals work, the launch
dependency is no longer necessary.
Additionally, *too many* abbreviations is difficult to use in the long
term, no matter how short the commands are. Since I don't need to launch
programs from the terminal as often anymore (now that I use rofi and
other programs), this change makes sense.
Note that most (if not all) of these programs can run inside any
environment with the same expected behavior, not just Plasma and GNOME,
provided the proper environment variables and config values are set.
xeventbind is a useful program that lets you run any command on
resolution change, similar to how GNOME on Wayland is able to change
scaling on on resolution change, but not exactly.
Since we're still using X, existing programs are not affected; they
have to be restarted instead.
Now that I understand more about how GNU/Linux distributions, display
managers, and X sessions work, it makes sense to separate meta packages
based on environment used.
When you use a display manager, you're just starting X in a fancy way.
Since you can run any program you want on the X server, it is easy to
install multiple desktop environments on the same machine and switch
between them easily, provided your setup is adequately modular (which
it should be by default).
It turns out that kitty has support to change all of its colors
independently from the current terminal with a simple command. This
was the only issue I had with making pywal and kitty work together,
so I'm glad I found it.
Additionally, kitty supports DPI changes immediately (at least with
xrdb). There is no need to detach a session and open a new terminal
since kitty will handle DPI changes automatically, compared to other
terminals like urxvt, which would require a new instance. Even then,
the border padding for urxvt is not adjusted to the new DPI; kitty
is simply the way to go if your monitor setup is non-trivial.
As a side note, the kitty documentation is very good. I highly
recommend reading it if you plan to use kitty (which you should).
Since the community package doesn't include the fish completions found
upstream, we install them here instead. There is already a bug report
about the fish completions, and when it eventually gets fixed, these
steps can be removed from the PKGBUILD.
Bug report: https://bugs.archlinux.org/task/55786
