![\"Directed](\"https://tudorr.ro/svg/python-numpy-deps-light.svg\")
![\"Directed](\"https://tudorr.ro/svg/python-numpy-deps-dark.svg\")
This article considers OS package managers, but much of it\ncan also apply to programming language package managers, like npm\nand Cargo.
\nIn the pursuit of getting out of my tech comfort zone, I have been\nworking on my “zen garden”: a small, cheap virtual private server\nrunning FreeBSD that runs some services I self-host.\nI am mainly a Linux person, so being confronted with all the things\nthat I take for granted on Linux which work differently —\nor not at all! — on FreeBSD lets me learn more about what happens\nunder the hood on both platforms.
\nOne highlight of the newly released FreeBSD 15.0 is that\nyou can now let the package manager manage the operating system.
\n“ (side note: \n“Look at this kid, I used to install Slackware from floppies!”\n)” says the Linux user.
\nEven Windows has its own OS package manager, what does\nFreeBSD do? Well, it uses, or used to use, the simplest package manager: tar.
Well, maybe not just tar, but any archive file tool.
\nWe want to be able to install packages on top of a “thing”,\nand maybe to remove them at some point. An archive file\nbundles files together and stores some attributes, like file name,\naccess bits, directory structure etc. It’s no wonder that more developed\npackage managers like rpm and dpkg are based on archives,\nas this is effectively the starting point.
To install a “package”, we can just extract an archive in / with the -x flag (eXtract):
$ tar -C / -xjvf base.txz\n(The -j flag tells tar that the archive is compressed with xz.)
Uninstalling is interesting. You can get a list of all the files\nin an archive like so, with the -t flag (lisT?):
$ tar -tjf base.txz\n./\n./COPYRIGHT\n./bin/\n./bin/[\n./bin/cat\n./bin/chflags\n./bin/chio\n./bin/chmod\n./bin/cp\n./bin/cpuset\n...\nRemember the file attributes I just mentioned? You can see them\nwith the -v flag:
$ tar -tjvf base.txz\ndrwxr-xr-x 0 root wheel 0 Nov 28 03:51 ./\n-r--r--r-- 0 root wheel 6070 Nov 28 03:51 ./COPYRIGHT\ndrwxr-xr-x 0 root wheel 0 Nov 28 03:43 ./bin/\n-r-xr-xr-x 0 root wheel 11736 Nov 28 03:42 ./bin/[\n-r-xr-xr-x 0 root wheel 13808 Nov 28 03:42 ./bin/cat\n...\nlrwxr-xr-x 0 root wheel 0 Nov 28 03:43 ./sbin/nologin -> ../usr/sbin/nologin\n-r-sr-xr-x 0 root wheel 26312 Nov 28 03:43 ./usr/bin/login\n...\nhr-xr-xr-x 0 root wheel 0 Nov 28 03:45 ./usr/bin/ssh link to ./usr/bin/slogin\n...\nBesides the access flags and owner information of each file, there are also some\ninteresting attributes. ./sbin/nologin is a symbolic link, ./usr/bin/login\nhas the SETUID flag, and ./usr/bin/ssh is a hard link to\n./usr/bin/slogin. Depending on the archive file format,\nthere can also be extended attributes and flags, but that’s\na discussion for another time…
So, we have something we can pipe to xargs rm\nto remove the files from the system. We can see that the output\nfrom tar -t also contains directories. We need to filter them out\nbefore passing them to rm, because removing a directory with plain rm would lead to an error. And we cannot use xargs rm -r because\nthat could delete more things than we want! We get:
$ # remove files\n$ cd / && tar -jtf base.txz | grep -v '/$' | xargs rm\n$ # remove empty directories\n$ cd / && tar -jtf base.txz | grep '/$' | sort -r | xargs rmdir\nThis is more or less how the (side note: \nFreeBSD, OpenBSD, NetBSD.\n) are distributed — in a couple tar archives. More or less — we\nwill get to that later.
\nA couple of questions immediately come to mind:
\nThese are table stakes for the package managers most people are used to, like\nrpm, dpkg, apk, pacman etc. The first question can be answered easily as\nlong as we have the file list of the old archive. You remove the old files, and\nthen you extract the new archive.
And if we have the file list, then we can also answer the other questions. If we keep a database of files installed by each package,\nwe can manage some packages.
\nIt is time for some disruptive innovation. Let’s store a database\nwith all the files that make up each package. Given the name of\nan installed package, we should get a list of all files that\nare part of that package.
\nThere are many ways of storing such a database. For example, RPM uses the\nBerkeley DB format, and pkg (from FreeBSD) keeps tables in an SQLite 3 database. I will use (side note: \nI don’t use Arch, btw.) as an example, owing to its text-based simplicity.
\nPacman creates a directory for each installed package in /var/lib/pacman/local:
$ ls /var/lib/pacman/local/\nacl-2.3.2-1\nALPM_DB_VERSION\narchlinux-keyring-20251116-1\nattr-2.5.2-1\naudit-4.1.2-1\nautoconf-2.72-1\nautomake-1.18.1-1\navahi-1:0.9rc2-1\nbase-3-2\nbase-devel-1-2\n...\nThere is also a regular text file named (side note: \nAlpm stands for “Arch Linux Package Management”, according\nto pacman(8).\n) that holds the database format version. On my system, this is equal\nto 9.
Let’s see what’s inside one of these directories:
\n$ ls /var/lib/pacman/local/pacman-7.1.0.r7.gb9f7d4a-1/\ndesc\nfiles\nmtree\nThree files: desc, files, and mtree.\nIn the aptly named text file files\neach line is a path to a file originating from this package:
$ head /var/lib/pacman/local/pacman-7.1.0.r7.gb9f7d4a-1/files\n%FILES%\netc/\netc/makepkg.conf\netc/makepkg.conf.d/\netc/makepkg.conf.d/fortran.conf\netc/makepkg.conf.d/rust.conf\netc/makepkg.d/\netc/pacman.conf\nusr/\nusr/bin/\n...\nNow we have a clear idea of which package contains what, which\nallows us to answer a multitude of questions. We can now avoid conflicts,\ncorrectly upgrade or downgrade packages, and check which package\nmanages a file. Which package offers the file /usr/bin/bash?
$ grep -lr usr/bin/bash /var/lib/pacman/local/ | cut -d'/' -f1\nbash-5.3.9-1\n...\nOn FreeBSD, we can leverage pkg’s SQLite 3 database:
\n# pkg shell\nSQLite version 3.50.4 2025-07-30 19:33:53\nEnter \".help\" for usage hints.\nsqlite> select packages.name from files\n ...> left join packages on files.package_id = packages.id\n ...> where files.path = '/usr/local/etc/rsync/rsyncd.conf.sample';\nrsync\nWe can also check whether a package has been incorrectly applied, or whether\nfiles from the package are for some reason missing on the host. You might\nremember from a couple paragraphs up that pacman also stores an mtree file in\nits database. But what is it?
$ file mtree\nmtree: gzip compressed data, from Unix, original size modulo 2^32 29540\nHmm, let’s use zcat to uncompress it and pipe the result to cat:
$ zcat mtree | head\n#mtree\n/set type=file uid=0 gid=0 mode=644\n./.BUILDINFO time=1765404175.0 size=5292 sha256digest=c1e0872d1c7200038790e6e90c54ea58be4f9321eda478dde48af7f367c6c926\n./.INSTALL time=1765404175.0 size=454 sha256digest=a041703891b00fc7c2109d004ac548de8e3f684a910887d7d463adcc4984548d\n./.PKGINFO time=1765404175.0 size=633 sha256digest=766e91bb5b8d47f9b93d1f6367002e9bb28de43280f6ffa5b56814bd42566a14\n./etc time=1765404175.0 mode=755 type=dir\n./etc/bash.bash_logout time=1765404175.0 size=28 sha256digest=025bccfb374a3edce0ff8154d990689f30976b78f7a932dc9a6fcef81821811e\n./etc/bash.bashrc time=1765404175.0 size=733 sha256digest=563e03eb4b40edfcc778f04efa2b4913bcdc6929c73d2bd4c07e1f799b98721a\n./etc/skel time=1765404175.0 mode=755 type=dir\n./etc/skel/.bash_logout time=1765404175.0 size=21 sha256digest=4330edf340394d0dae50afb04ac2a621f106fe67fb634ec81c4bfb98be2a1eb5\nOh, this is also a file list, but enhanced with hashes, access flags\nand ownership information. In fact, this listing is to be used by\nmtree, a tool which can either create such a file from a file\nhierarchy or compare a file hierarchy against a file listing of this kind.\nPerfect for checking that our system is consistent after installing packages.
In theory, with only archive files and mtree, we now have enough tools to\ninstall independent packages that don’t depend on each other. Or packages that\nhave such simple dependency relations that the user can handle them,\nlike (side note: \nOn FreeBSD, the only required set is base.txz. You can throw on top\nother optional sets that don’t have any dependency other than base, such as\nbase-dbg for debugging symbols and src for the OS source code.\n). But if we\nwant to be able to automatically install and uninstall dependencies, we need to\nadd another puzzle piece: requirements.
Here is what happens when I try to install NumPy with pacman:
\n$ sudo pacman -S python-numpy\nresolving dependencies...\nlooking for conflicting packages...\n\nPackage (6) New Version Net Change\n\nextra/blas 3.12.1-2 0.74 MiB\nextra/cblas 3.12.1-2 0.34 MiB\nextra/lapack 3.12.1-2 15.06 MiB\ncore/mpdecimal 4.0.1-1 0.33 MiB\ncore/python 3.13.11-1 67.66 MiB\nextra/python-numpy 2.4.0-1 46.42 MiB\n(-S stands for sync. Pacman works with the sync database when running\nwith -S, i.e. information about all remote packages that is synchronised on\nyour system. This database can be found in /var/lib/pacman/sync/.)
The dependency tree of python-numpy up to depth 2 looks something like this:
python-numpy\ncblas\nblasglibclapac\nblasgcc-libsglibcpython\nbzip2expatgdbmlibffilibnsllibxcryptopensslzlibtzdatampdecimalIn the pacman run from above, pacman computed the complete dependency graph, and\nkept only the nodes corresponding to packages that are not already installed.\npython-numpy obviously needs Python because it’s a Python library.\nIt also needs Blas,\nthe de-facto standard in matrix computations.\n(Blas stands for Basic Linear Algebra Subprograms.)
This dependency resolution works very well when installing packages!\nBut what happens if I want to uninstall a package that depends on\nother packages?
\n$ sudo pacman -R python-numpy\nchecking dependencies...\n\nPackage (1) Old Version Net Change\n\npython-numpy 2.4.0-1 -46.42 MiB\nWait, where did Python and Blas go?\nWhy doesn’t pacman remove, them, too? I don’t need them any more!
\n (side note: But also with DNF, Zypper, APT, pkg…) you need to explicitly mention that you\nwant to remove packages recursively using the -s flag:
$ sudo pacman -Rs python-numpy\nPackage (6) Old Version Net Change\n\nblas 3.12.1-2 -0.74 MiB\ncblas 3.12.1-2 -0.34 MiB\nlapack 3.12.1-2 -15.06 MiB\nmpdecimal 4.0.1-1 -0.33 MiB\npython 3.13.11-1 -67.66 MiB\npython-numpy 2.4.0-1 -46.42 MiB\n...\nAh, just what I wanted! And it figured out that I don’t need\nPython and those weird sciency libraries.
\nWell, how does pacman know that you need a package or not?\nThe answer lies in the local package database. You remember\nfor each package whether it was installed explicitly by the user,\nor implicitly. Explicit means that I spelled the name of the package\nwhen I ran pacman, like I did with python-numpy in the example above.\nImplicit means that I didn’t spell it, but I agreed to install it after\npacman figured out that I need it. Some package managers call implicit installation\nautomatic installation. Pacman sets\nan internal %REASON% flag to 1 if the package was installed implicitly.
$ cat /var/lib/pacman/local/cblas-3.12.1-2/desc\n%NAME%\ncblas\n\n%VERSION%\n3.12.1-2\n\n...\n\n%REASON%\n1\n...\nSay I want to keep cblas because I happen to use it for some\nscientific programming project of mine, but I still want\nto remove Numpy and whatever else it needs that I don’t.\nI can first tell pacman to mark cblas as an explicitly-installed\npackage — even though I initially installed it implicitly — and\nthen I can remove Numpy with the command from above.
$ sudo pacman -D --asexplicit cblas\ncblas: install reason has been set to 'explicitly installed'\n$ sudo pacman -Rs python-numpy\nPackage (4) Old Version Net Change\n\nlapack 3.12.1-2 -15.06 MiB\nmpdecimal 4.0.1-1 -0.33 MiB\npython 3.13.11-1 -67.66 MiB\npython-numpy 2.4.0-1 -46.42 MiB\n...\n(pacman -D is a special mode for modifying fields in the local package database.)
Works as intended!
\nOK, so now we can install and remove packages with dependencies,\nand we can determine whether packages are needed or not when\nwe want to free up some disk space. But what about the following:\nWhat if a package needs something that is available\nin more than one package? Say I want to install Steam to play\nvideo games. Steam for some reason needs a Vulkan driver to be present,\nmaybe it uses it for rendering. But there are multiple Vulkan drivers\nout there, made by different graphics card manufacturers.\nWhat dependencies does Steam have?
\n$ sudo pacman -Si steam\nRepository : multilib\nName : steam\nVersion : 1.0.0.85-1\nDescription : Valve's digital software delivery system\nArchitecture : x86_64\nURL : https://steampowered.com/\nLicenses : LicenseRef-steam-subscriber-agreement\nGroups : None\nProvides : None\nDepends On : bash coreutils curl dbus desktop-file-utils diffutils freetype2 gcc-libs gdk-pixbuf2 glibc hicolor-icon-theme libxcrypt\n libxcrypt-compat libxkbcommon-x11 lsb-release lsof nss python ttf-font usbutils vulkan-driver vulkan-icd-loader xdg-user-dirs\n xorg-xrandr xz zenity lib32-alsa-plugins lib32-fontconfig lib32-gcc-libs lib32-glibc lib32-libgl lib32-libgpg-error lib32-libnm\n lib32-libva lib32-libx11 lib32-libxcrypt lib32-libxcrypt-compat lib32-libxinerama lib32-libxss lib32-nss lib32-pipewire\n lib32-systemd lib32-vulkan-driver lib32-vulkan-icd-loader\n...\n(-i stands for “info”. You can find information about an already-installed\npackage with -Qi. pacman -Q queries the local database.)
That’s a lot of dependencies! Can you spot the Vulkan driver that Steam\nwants?
\n... ttf-font usbutils vulkan-driver vulkan-icd-loader xdg-user-dirs ...\nAh, there it is! It’s just that, vulkan-driver? But I thought there\nare multiple of them?
There are, indeed, multiple. The vulkan-driver package is not real!
We solve this problem by introducing fake packages that do not\ncontain anything: they are either installed, or not installed,\nbut otherwise they don’t do anything. Other names include\n“virtual packages” and “capabilities”. They act kind of like\ngeneral-purpose booleans in the package manager’s local database.\nOther packages can then assert these booleans to be true\nfor their dependencies to be satisfied.\n“Real” packages can claim that they provide\nthe fake package: the package upholds a guarantee; if you install it,\nthe boolean will be set, the requirement will be satisfied.
\nLet’s continue with the Vulkan driver example. As of December 2025,\nthere are ten packages providing a Vulkan driver in the Arch Linux\nrepositories. There is one for Intel GPUs, (side note: One provided by Nvidia and one made by the Mesa project,\nwhich is called Nouveau. Clever.), one for use in virtual machines etc.\nThe package containing the Steam client lists vulkan-driver as\na requirement. This is the fake package: there is no package\nnamed vulkan-driver you can download from the Arch Linux repos!\nBut if your computer for example has an Intel GPU, and you already have\nworking graphics, chances are that you installed the right Vulkan\ndriver together with the rest of the graphics stack — likely\nas an implicit dependency.
But if you did not, and you want to install Steam, pacman will\nask you nicely: which of the following Vulkan drivers do you need?
\nLet’s ask pacman some information about the vulkan-driver package\nI have on my system:
$ sudo pacman -Q vulkan-driver\nvulkan-intel 1:25.3.2-1\nSo I asked for vulkan-driver and I got vulkan-intel back!\nLet’s ask for more information:
$ sudo pacman -Qi vulkan-driver\nName : vulkan-intel\nVersion : 1:25.3.2-1\nDescription : Open-source Vulkan driver for Intel GPUs\nArchitecture : x86_64\nURL : https://www.mesa3d.org/\nLicenses : MIT AND BSD-3-Clause AND SGI-B-2.0\nGroups : None\nProvides : vulkan-driver\n...\nA-ha! Provides: vulkan-driver, that’s the line!\nYou can think of this line as telling pacman\n“mark in the local database that there is a Vulkan driver installed”.
But wait, it can get trickier: programs are often linked against dynamic libraries.\nThey need files ending in .so to work. We can find out which using ldd:
$ ldd $(which bash)\nlinux-vdso.so.1 (0x00007f167b946000)\nlibreadline.so.8 => /usr/lib/libreadline.so.8 (0x00007f167b7ad000)\nlibc.so.6 => /usr/lib/libc.so.6 (0x00007f167b59b000)\nlibncursesw.so.6 => /usr/lib/libncursesw.so.6 (0x00007f167b52c000)\n/lib64/ld-linux-x86-64.so.2 => /usr/lib64/ld-linux-x86-64.so.2 (0x00007f167b948000)\nSo, bash needs (side note: With the exception\nof linux-vdso.so.1, that one is special. It doesn’t even exist\non disk. You can read more about it from its dedicated man page,\nvdso(7).) of these files to be present in a known\ndirectory in the system — /usr/lib — or it will not work at all. Sounds\nlike the Vulkan driver case! I know for example that libreadline.so.8\ncomes from the readline package, but the name of the exact file the program\nrequires is libreadline.so.8. Maybe there are packages that provide multiple\nlibrary files at once, do I then need to remember the name of the package\nproviding each .so file if I want to make my own package? What if the name changes,\nor if the package is split, or if some packages merge into one? All my program\ncares about is having the right .so present.
Package capabilities to the rescue, again! On many platforms\n— Arch Linux, dpkg-based distros, rpm-based distros, Alpine Linux etc. —\npackages containing .so libraries advertise them as capabilities.
Say I need to use some binary I downloaded off the internet, but it complains\nthat it needs libasound.so.2. Let’s find out whether pacman has it for us:
$ sudo pacman -Ss libasound.so\nextra/alsa-lib 1.2.14-2\n An alternative implementation of Linux sound support\nGreat! What does its Provides line look like?
$ sudo pacman -Si alsa-lib | grep Provides\nProvides : libasound.so=2-64 libatopology.so=2-64\nWe can see that it has a version attached. That is important,\nbecause an older version of libasound.so may not be compatible with the\nbinary I downloaded off the internet. This can be the case when a function’s prototype\nchanges, for example when its return type is different in a later version. Such incompatibilities are marked\nby incrementing the number in the filename. This is called a “soname bump”.
Coming back to my random binary example, we want to make sure that we\ninstall a compatible libasound.so. Thanks to the version information,\nwe can be specific when asking pacman to install it:
$ sudo pacman -S libasound.so=2\nresolving dependencies...\nlooking for conflicting packages...\n\nPackage (3) New Version Net Change Download Size\n\nextra/alsa-topology-conf 1.2.5.1-4 0.33 MiB 0.01 MiB\nextra/alsa-ucm-conf 1.2.14-2 0.54 MiB 0.11 MiB\nextra/alsa-lib 1.2.14-2 1.68 MiB 0.48 MiB\n...\n(side note: Red Hat Enterprise Linux,\nFedora, CentOS, CentOS Stream, Alma Linux, Rocky Linux etc.)\nmake heavy use of package capabilities. These distros use DNF and RPM to manage\npackages. I will show some examples from Alma Linux 9.
\nExample 1, a Python library:
\n$ sudo dnf repoquery --provides python3-numpy\nlibnpymath-static = 1:1.23.5-1.el9\nlibnpymath-static(x86-64) = 1:1.23.5-1.el9\nnumpy = 1:1.23.5-1.el9\nnumpy(x86-64) = 1:1.23.5-1.el9\npython-numpy = 1:1.23.5-1.el9\npython3-numpy = 1:1.23.5-1.el9\npython3-numpy(x86-64) = 1:1.23.5-1.el9\npython3.9-numpy = 1:1.23.5-1.el9\npython3.9dist(numpy) = 1.23.5\npython3dist(numpy) = 1.23.5\nHere we have capabilities for the python library itself in different forms, with and without the\nCPU architecture, and capabilities with the Python interpreter’s version.
\nExample 2, a C library, without headers:
\n$ sudo dnf repoquery --provides readline\nlibhistory.so.8\nlibhistory.so.8()(64bit)\nlibreadline.so.8\nlibreadline.so.8()(64bit)\nreadline = 8.1-4.el9\nreadline(x86-32) = 8.1-4.el9\nreadline(x86-64) = 8.1-4.el9\nHere we have two .so files with architecture information. The package also provides\nfiles for running 32-bit executables.
Example 3, the development files of the same C library:
\n$ sudo dnf repoquery --provides readline-devel\npkgconfig(readline) = 8.1\nreadline-devel = 8.1-4.el9\nreadline-devel(x86-32) = 8.1-4.el9\nreadline-devel(x86-64) = 8.1-4.el9\nThis package provides database files for pkg-config,\na tool which finds the right directories\nfor the header (.h) and library files (.so) when building C and C++ code\nand generates the right compiler flags to use them.
Example 4, DNF itself:
\n$ sudo dnf repoquery --provides dnf\ndnf = 4.14.0-31.el9.alma.1\ndnf-command(alias)\ndnf-command(autoremove)\ndnf-command(check-update)\ndnf-command(clean)\ndnf-command(distro-sync)\ndnf-command(downgrade)\n...\nDNF can be extended with plug-ins, which can, among other things, extend DNF with new subcommands.\nSay you found a command on a forum to fix the problem you’re having, and it uses dnf diff:
$ dnf diff\nNo such command: diff. Please use /usr/bin/dnf --help\nIt could be a DNF plugin command, try: \"dnf install 'dnf-command(diff)'\"\nYou run the suggested command, and:
\n$ sudo dnf install 'dnf-command(diff)'\n...\nInstalling:\ndnf-plugin-diff noarch 2.0-1.el9 epel 21 k\n...\nSo easy, and you don’t need to know the name of the package at all! It could be named\nsomething like dnf-plugin-magic-diff-command or something (it is not).
Example 5, Perl modules:
\n$ sudo dnf repoquery --provides perl-IO\nperl(IO) = 1.43\nperl(IO::Dir) = 1.41\nperl(IO::File) = 1.41\nperl(IO::Handle) = 1.42\nperl(IO::Pipe) = 1.41\nperl(IO::Pipe::End)\nperl(IO::Poll) = 1.41\nperl(IO::Seekable) = 1.41\nperl(IO::Select) = 1.42\nperl(IO::Socket) = 1.43\nperl(IO::Socket::INET) = 1.41\nperl(IO::Socket::UNIX) = 1.42\nperl-IO = 1.43-481.1.el9_6\nperl-IO(x86-64) = 1.43-481.1.el9_6\nJust like with .so files, you don’t have to know the name of\nthe package providing the modules.
And the list can go on… You can read more about the various\ncapabilities used in these distros in the Fedora Packaging Guidelines.
\nOK, now, one more feature that would be very convenient in a package manager…\nYou know how some programs also provide shell completion definitions? I love\nshell completion; I want them for as many programs as possible. But there are\nalso multiple shells around. It would be nice if the shell completions\nwould just install themselves automatically. It would also be nice if, when\nI install a different shell, all the completion definitions for the myriads\nof command line programs I use would also be installed automatically for that\nshell.
\nThe only package manager I know of that supports this is apk,\nfrom (side note: Probably one of the most freeloaded Linux distros. I wonder how much critical infrastructure\nat big name companies runs on it…). Apk is also\nused on Chimera Linux.
\nAn apk package can have a field named “install-if”, in order to define\nauto-install rules. An example from helix-fish-completion:
$ sudo apk info --install-if helix-fish-completion\nhelix-fish-completion-25.07.1-r2 has auto-install rule:\nhelix=25.07.1-r2\nfish\nWhen all the packages in that list will be installed,\nhelix-fish-completion will also be installed automatically.
If I now install Helix, helix-bash-completion will\nalso be installed, because I use bash:
$ sudo apk add helix\n(1/2) Installing helix (25.07.1-r2)\n(2/3) Installing helix-bash-completion (25.07.1-r2)\n...\nLater, if I want to install the Fish shell,\nhelix-fish-completion will be installed automatically.
$ sudo apk add fish\n(1/4) Installing fish (4.0.2-r0)\n(2/4) Installing curl-fish-completion (8.17.0-r1)\n(3/4) Installing fish-doc (4.0.2-r0)\n(4/4) Installing helix-fish-completion (25.07.1-r2)\n...\nThe same thing (side note: \nThis stems from the way apk works. In apk, the add and del\ncommands modify the “world file” at /etc/apk/world.\nThis is a regular text file where each line contains the name of a package\nthat was installed explicitly. apk add adds a line, apk del removes a line.\nAfter the file is modified, apk computes the dependency graph of all the files mentioned in that file —\ntaking into account the auto-install rules — compares it against the graph of all packages that are\nactually installed, and installs and uninstalls packages accordingly to make the second graph match the first.\n): if a package\nwas installed implicitly, and the auto-install rule\ndoesn’t hold any more, then the package will be removed:
$ sudo apk del helix\n(1/3) Purging helix-bash-completion (25.07.1-r2)\n(2/3) Purging helix-fish-completion (25.07.1-r2)\n(3/3) Purging helix (25.07.1-r2)\nGiving shell completion as an example was a bit silly, I must admit.\nShell completion files are very small files; it’s not the end\nof the world if you have them around. Maybe it is the end\nof the world for some people, so it’s good to have this feature!
\nThis feature shines when you want an environment where\nyou do not wish to have man pages and documentation.\nOn Alpine Linux, documentation is split into separate\npackages with names ending with -doc. These packages\nhave auto-install rules:
$ apk info --install-if zfs-doc\nzfs-doc-2.4.0-r0 has auto-install rule:\ndocs\nzfs=2.4.0-r0\nSo, if you want to get rid of all documentation, you just\nuninstall the docs package! Conversely, if you want documentation,\nyou install the docs package and the -doc packages corresponding\nto your installed packages will be magically installed.\nAnd whenever you will install a new package that has documentation\navailable, its corresponding -doc package will be installed with it.\nSo neat and elegant!
OK, but I started this post with FreeBSD not being managed with a package\nmanager before. How does that work?
\nSo FreeBSD has only a handful of packages: base, kernel-dbg, base-dbg, ports etc.\nOnly base is required, and all the other packages depend on base only.\nSo dependency resolution is really simple, it fits in your head.
But what about package upgrades? And, what about configuration files? Your\nOS comes with some default configs. What happens when\nyou modify a config, and a new version of the OS comes with new defaults?
\nThat’s so 1997, I hear you say. Haven’t you heard about (side note: \nLike /etc/ssh/sshd_config and\n/etc/ssh/sshd_config.d/*. )? And in 2025 we have these things\ncalled “container images”, we upgrade our OSes atomically. OK, a package\ninstalls a config and I override it. So what? When there’s a new package\nversion, I don’t apply it on top, I just reinstall the whole OS! Why would I\neven care…
There used to be a time when sysadmins used to care about bandwidth,\nstorage and electricity and they didn’t reinstall their entire OS for each\npackage update — figuratively, that is pretty much what happens when you rebuild\na container image. As for why config files weren’t split in defaults and overrides,\nbeats me. I am too young to know that.
\nBut OK, this is an actual problem. You have the old default, the new default, and\nwhatever the user modified in, say, /etc/resolv.conf, or /etc/passwd. What do you do?
RPM and (side note: \nSee Pacnew and Pacsave.\n), for example, just save the new version with a new extension and\nkeep your changes untouched. Hopefully you will see the warning on the screen\nnext to the other hundreds of scrolling lines and you will check\nwhether the file needs any manual intervention.
\nOn FreeBSD, there is a tool named etcupdate that you are supposed\nto run after OS upgrades. But you usually don’t run it manually, unless you\ninstalled the OS from source or something weird like that. The conventional\nroute is to run freebsd-update which will take care of\ndownloading the new archives, extracting them in root, removing old files and so\non. Hey, that looks like a package manager!
Indeed it does, albeit a purpose-built one. It was made to install new versions of that couple\nof OS archives, rollback from an upgrade, and nothing else, basically.\nAnd it does some fancy magic to only download binary diff files to save space and bandwidth, neat.\nBut I assume that it must be pretty hard to maintain it, because the effort that goes into this tool\ndoesn’t translate into improvements for pkg, which was used — until now — only to install third-party software.\nAnd, you have a problem if you want a slim OS install. What if you don’t need everything that comes\nin the base package? On Linux distros you can install a small set of packages and be done with it.\nOn FreeBSD, you had to recompile the OS from source for that! That’s a bit of a waste, and it requires you\nto understand how the build system works before you even try to obtain a smaller install. That’s quite the overhead.\nWith the new pkg-based installation workflow, if I don’t need the ZFS filesystem for example, I can just\ndo pkg delete FreeBSD-zfs and it’s gone.
How did we end up here? I didn’t live through the good old days when Linux just\nbecame a thing and the BSD world split into three, and I did not read any UNIX\nhistory book on this subject either — if it exists.\nBut I can speculate that the Linux world, being broken into multiple, seemingly independent projects\nthat (side note: Not literally, I hope that there is no\nKPI reporting with the conversion rate of Plasma users to Gnome.), felt the need\nfor a program that allows users to mix and match software before the BSD world needed.\nUnlike Linux distros, BSD operating systems are one whole product, developed by one team per operating system,\nso they have easier control of the software developed under their umbrella. They built their own\ncustom tools — etcupdate, freebsd-update etc. — and those worked well, then they created\nimproved tools like pkg. To me it seems like a good, natural progression.
If there is anything to be learned from this, is that package managers started from the basic requirement\nof installing (and removing!) software and slowly added smart features on the go.\nThere you have it: an overview of how package managers can be helpful!
","date_published":"Thu, 25 Dec 2025 18:52:11 GMT"},{"id":"https://tudorr.ro/blog/2025-03-29-faster-zypper/","url":"https://tudorr.ro/blog/2025-03-29-faster-zypper/","title":"How to Speed Up Package Installations on OpenSUSE","content_html":"Currently, this article applies only to openSUSE Tumbleweed.
\n(side note: The default package manager on OpenSUSE and\nSUSE Linux Enterprise.) just added support for parallel\ndownloads, which can make large upgrades become\ntwice as fast. Here I will write about some more tweaks I\ndiscovered that make Zypper even quicker.
\nMake sure you are using the newest Zypper and libzypp:
\n$ sudo zypper update libzypp zypper\nInstall the openSUSE-repos package to use CDN-based repositories:
$ sudo zypper install openSUSE-repos\nThis applies multiple tweaks in one, see below for a detailed explanation.
\nEnable the experimental single RPM transaction mode:
\n$ echo \"export ZYPP_SINGLE_RPMTRANS=1\" > /etc/profile.d/my-zypp-settings.sh\nIf you run Zypper with sudo: let sudo pass the values\nof ZYPP_PCK_PRELOAD and ZYPP_SINGLE_RPMTRANS.
Run sudo visudo -f /etc/sudoers.d/zypper-vars to safely\nedit the file and write the following:
Defaults env_keep += \"ZYPP_PCK_PRELOAD ZYPP_SINGLE_RPMTRANS\"\nLog out and log back in for environment variable changes to take effect.
\nYou should not skip this step if you preferred to not turn on the experimental\nsingle RPM transaction mode from above. Step 2 also does some environment\nvariable changes behind the scenes.
\nopenSUSE-reposInstalling this package replaces the default Zypper repo definitions with ones\nderived from openSUSE’s Repository Index Service, which is\na facility provided by Zypper to (side note: \nAccording to the OpenSUSE Wiki page about Repository Index Service,\nthe protocol is an extension of Novell Update,\npart of the Zenworks Linux Management “client-server protocol”.\nZenworks Linux Management is a configuration management product that\nallows one to “embrace and extend” Linux in their enterprise environment.\nThese might all sound unrelated, but openSUSE is partly developed by SUSE,\nwhich is a company that used to be part of the Attachmate Group,\nwhich contained Novell, and this group was purchased by Micro Focus,\nwhich is now named OpenText. You can see part of these companies’\nhistories in this inoffensive feature! (Source.)\n).\nThese repositories are basically the same as the default ones, the only difference\nbeing that they are served from a CDN.\nFrom the post that announces Zypper’s parallel download abilities,\nwe learn that this package also enables both parallel downloads and the new\nfile download backend, which reduces overhead.
\nAfter installing this package, you need to log out and log back in\nbecause parallel downloads are enabled by means of installing\na snippet in /etc/profile.d/ which toggles the right environment variable.
I haven’t looked into it yet, but Zypper for some reason runs a separate\nRPM transaction for every package it installs. This can quickly add overhead\nwhen you install a package that needs many dependencies, or when upgrading\nmany packages at once. For a couple of years already, Zypper has been offering\nan experimental mode in which all package installations done by a command\nwould be done in a single transaction, toggled by the ZYPP_SINGLE_RPMTRANS\nenvironment variable.
Be aware that some package installations can fail in this mode!\nPackages are not always tested for installing with this experimental feature turned on.\nIf this happens, run Zypper like so:
\n$ sudo ZYPP_SINGLE_RPMTRANS=0 zypper ... # the rest of your command\nAnother catch is that YaST might not be able to do some package management operations any more,\nas highlighted in the announcement of this feature.
\nsudo configWhen run normally, sudo provides only a limited set of\nenvironment variables to the program you want to run,\neven when those variables are set system-wide in /etc/profile.d.\nYou can run sudo with the -i flag to run the command in a login shell,\nwhich will load files from /etc/profile.d, or you can configure\nsudo to pass the environment variables we use to tweak Zypper.\nI prefer the latter, because it brings fewer changes to the way\nsudo runs commands.
If you just want to see how to do it without the backstory,\nclick here if you’re running Ansible ≥ 2.18.1,\nor click here if your Ansible version is older.
\nI didn’t write much about it yet, but a couple of months ago I scrapped my NixOS\nhomelab setup in favour of a FreeBSD VM. I’m enjoying it so far — it’s the\nexact opposite of NixOS in my opinion. Instead of playing with abstractions\nstacked into each other like a matryoshka doll, I have to make my own little\nabstractions to suit my own, simpler use cases.
\nOne of these abstractions is managing jails. They are comparable to Linux\nnamespaces: you can use them to make something akin to (side note: \nThere have been various technologies called “containers” throughout\nhistory. The one we know and love the most today, OCI Containers\n— or “Docker” containers — are a standardised way of\npackaging, distributing, and running software. It’s not just about running\nprograms isolated.\n) — little isolated systems that live on the same host.\nIn the FreeBSD fan world, a “jail” can often mean some kind of container.
\nAnsible modules made for FreeBSD usually have a jail argument, for example to\nset rc arguments in a jail, or to install a package with pkg. The service\nmodule, however, does not. What if you want to manage services inside such a jail?
Luckily, I discovered a method that lets you do just\nthat, without reinventing the wheel! This method works only if you run\nAnsible ≥ 2.18.1. If you don’t, I also discovered a hack that works around\nthis version requirement.
\nWhy that version specifically?\nBecause it contains this bugfix.
\n- name: Start SSHD in myjail\n ansible.builtin.service:\n arguments: \"-j myjail\"\n name: sshd\n state: started\nWhen service is called with the -j flag, it will work inside a jail;\nsee the man page.
Here, whatever is given to arguments is going to be then given to the service command,\nwhich Ansible calls behind the scenes. Ansible does something like this:
if ! service -j myjail onestatus sshd; then # check if service is already running\n service -j myjail onestart sshd # if not, start it\nfi\nThis can be seen in Ansible’s source code.
\nIf you want to enable a service, do not use the enabled parameter!\nInstead, call the sysrc module first:
- name: Enable SSHD in myjail\n community.general.sysrc:\n jail: myjail\n state: present\n name: sshd_enable\n value: \"YES\"\n\n- name: Start SSHD in myjail\n ansible.builtin.service:\n arguments: \"-j myjail\"\n name: sshd\n state: started\nThe reason for this is that there is a bug in the way Ansible sets the required rc variable.\nAnsible will run sysrc, but without the -j parameter which lets it work in a jail\n(sysrc(8)).
- name: Start SSHD in myjail\n ansible.builtin.service:\n name: \"-j myjail sshd\"\n state: started\nSimilarly, if you want to enable a service before starting it, you can use sysrc:
- name: Enable SSHD in myjail\n community.general.sysrc:\n jail: myjail\n state: present\n name: sshd_enable\n value: \"YES\"\n\n- name: Start SSHD in myjail\n ansible.builtin.service:\n name: \"-j myjail sshd\"\n state: started\nWhy is this needed? Because of a bug, which was fixed in version 2.18.1.\nAnsible would add the value of the arguments parameter after the name\nof the service when calling service, which is not correct. Behind the scenes,\nAnsible would run:
# This is wrong!\nif ! service onestatus sshd -j myjail; then # check if service is already running\n service onestart sshd -j myjail # if not, start it\nfi\nHere, the -j myjail argument would be passed to the init script, in this case\n/etc/rc.d/sshd. This is not what we want! The -j argument is supposed to be\ngiven to the service tool, as you can see in the man page.
![\"Picture](\"https://tudorr.ro/static/pictures/blackforest1.avif\")
It’s been a while since I posted on my blog! I’ve been wanting to write some\narticles, but I did not have any fleshed-out ideas, hence the break. I keep\nseeing the same advice from bloggers and other writers on the Internet: Just\nstart writing!
What to write about, though? Anything!
The first thought I had when I opened my text editor to write this post was\nthat it’s been a while since I wrote something, might as well write about what\nhappened in the meantime!
\nSince I wrote my last blog post, I’ve been busy with finishing my bachelor’s\ndegree, summer holidays, and starting my master’s. I’m thankful for my thesis\nsupervisor, who gave me plenty of freedom, which led to a thesis about Ken\nThompson’s “Trusting Trust” attack and reproducible builds.\nI attacked the (then) latest version of the Go compiler to target Go’s own\nreproducibility checker took, gorebuild. Ironically enough,\nthere are no official, (side note: Or at least hashed.) gorebuild binaries, so targetting it with such an attack is\nactually practical: the “official” way to install gorebuild is to just compile\nit from source. My attack changed the compiler to make it detect when\nit’s compiling gorebuild to then change its logic to lie to the user that the\nattacked compiler actually matches the official hashes as published on the Go\nwebsite. It is on my list to write a nice post about this attack and challenges\nI encountered. If you find this interesting, you might be interested in reading\nmy thesis here. There are some extras in there too, such as evilgen\n— (side note: \nIt generates source code that can regenerate itself. The source code is in the repo, together with some examples.\n) — and some code examples.
After I was done with the academic year, I went on a five week-long holiday\nto Romania and Germany. Of those five weeks, three were spent without my laptop,\nwhich really gave my mind some rest after all the studying. Visited family,\nsaw some nature, hiked; great holiday all-around.
\nAnother thing that I felt was lacking is that my blogging setup did not\nfully reflect the way I think, which made writing feel not that great. I took\ninspiration from other websites, pondered a bit, and I realised that what I\nneeded all this time were sidenotes. You can see them in action on this page!\nI wanted to have a way to mention all the tid-bits and fun facts and little\ndetails that I like to mention when I have a conversation, because these are\nall things that can make a discussion even more interesting.
\nNormally one might use footnotes for that, but I find them quite distracting on\nwebsites, actually. On paper you can switch you gaze really quickly between the\nbody text and the footnotes, but on screens — especially little ones — that is\nnot the case, in my opinion1. A web page has an arbitrary length, of which\nonly (side note: The viewport.) is\nactually displayed to your eyes. I find it really easy to lose the position I’m\nat in the text when I look at a footnote on the web — it’s not like I can just\nleave my finger there and then take a quick look at the bottom of the screen.\nThere are ways to improve this, like having the footnote indicator be a link\nthat makes your browser scroll enough to bring it on the screen. Some\ncleverer implementations even have a link that sends you back right at the end of\nthe footnote. But then the line containing the footnote will be stuck to the top\nof your browser window, which is also very annoying!
\nBut this still doesn’t fix my main gripe I have with footnotes on\nthe web: they require you to move the page around! It’s annoying, it’s confusing and dizzying\neven, and this is the web after all; we can make good use of the differences\nbetween the two media. One of them is that desktop screens are wide: there is\nplenty of space for some notes that go along with the text. On phones, one may use CSS\nto make the sidenotes inline. I actually made them interactive on my website for a reason:\nphones are pretty small, and can only display a limited amount of text when using a decent\nfont size. I wanted to have the body text have priority when using this space, and let\nthe reader optionally reveal the notes by tapping on the associated text.\nI figured this may not be the most discoverable UI, so I just put a note at the\ntop of the page that is only visible when using a narrow screen.
\nThe sidenotes also render perfectly fine in reader mode, and\n (side note: I haven’t tested that, yet.).\nMy main inspiration for my implementation was this article by Koos Looijesteijn.\nI implemented the sidenotes as a JSX component, and I used Tailwindcss for the styling.\nMy website is completely statically-generated, there is no JSX on your browser right now,\ndon’t worry! You can find the source code for this component here.
\nIf you have any suggestions, praise, (constructive) criticism or just want to say something\nabout my sidenotes, I’d be very happy to read about it!\nYou may send me an e-mail or a message on Mastodon, check the home page for links!
\nNext to all of that, I started a new study at the university: the Computer Security master’s degree.\nIt’s about low-level systems security! It’s currently going well, I really like it, I feel like I’m already learning a ton.\nI’m excited for the courses that come next! I really want to write more here, I’ll see how that goes: the study\nis keeping me rather busy. Till next time!
\nSee? ↩
\nYarr is, according to its GitHub page, yet another rss reader
.\nBut I think this description doesn’t do it justice.
I started using a feed reader less than half a year ago, to easily keep track\nof new content I enjoy on the internet without depending on social media.\nWeb feeds are, in 2024 terms, old technology: according to the RSS Advisory Board,\nRSS, one of the main web feed formats available, has been published in 1999, and it’s not\neven the first web syndication format ever made.
\nI’m a Gen Z-er: I was a toddler when RSS reached its 2.0 milestone,\nand when Apple popularised RSS for podcasts by adding support in iTunes.\nI was in middle school when Google Reader was shut down. When I started using the internet\nby myself, it was already dominated by social media, which is still the dominant\nsource of information for many people around me, younger and older alike.
\nGetting all my information from social media got really tiring; I was using Reddit mostly,\nand a carefully curated set of information-oriented Instagram profiles — so nice to read pictures of text! I didn’t really like wasting my time doom-scrolling\ninstead of doing other things that could generate some creative output — like\nwriting on this blog, maybe — or\ndoing some fun coding. Or maybe I could tidy up my kitchen, read a book and actually learn\nsome interesting, useful things laid out in a long-form structure. Heck, even getting bored\nis pleasant, it makes your brain think!
\nI ditched Instagram altogether, and then I also uninstalled the Reddit client I was using on my phone.\nI still read some subreddits from time to time, but now it requires me to manually type in\nthe subreddit I want to read, scroll a bit, and then become very frustrated with the new web UI —\nwhich is totally not intentionally bad to make you download the mobile app.\nNot only that, but I also change the URL to old.reddit.com when needed (even harder when doing it on my phone…).\nAnd then I only read without ever logging in, which is why using Old Reddit is a requirement.\nYou can easily see that it’s not the best reading experience for the web.
With this in mind, I remembered RSS and feeds are a thing, so I figured I should set up a reader.\nI’ve never used one before, but the concept sounded promising: I can subscribe to blogs and news outlets, directly!
\nI had three requirements for choosing one:
\nFeeder is a pretty Android app, it implements the newest Material 3 design language, and it’s generally very pleasant to use.\nIt definitely satisfies the first requirement. And it knows how to extract the content from a web page, very nice!\nUnfortunately, it’s only for your phone. If you want to change your reader app, it does support importing and exporting OPML files; that is pretty sweet,\nbut I don’t want to be tied to my phone to read the news and the blogs I want to follow.
\nMiniflux is what I’ve been using up until I discovered Yarr, actually. It’s wonderful software: it’s a web application, very easy to self-host, and it’s quite featureful.\nIt has a (side note: \nSome might say “arcane”.\n), stays out of your way, no useless spacing, readable, accessible, it’s just great.\nLooks similar to Lobsters and sourcehut, I really like this style. It even has the required manifests to make it a Progressive Web App,\nmaking it look like an app on your phone. It’s just an all-around pleasant experience to use Miniflux.\nIt’s very easy to self-host: being written in Go, the program is just a single statically-linked executable that you can drop on a server.
\nI have one big gripe with Miniflux, though: it only supports PostgreSQL, though very understandably so.\nIt’s much easier to support a single configuration, a very respectable one in this case.\nMiniflux also has a commercial offering, so it shouldn’t come as a surprise that it’s in its best interest to have a well-tested and reliable storage layer.
\nIt’s just that I prefer having as many services as possible use SQLite when it comes to my self-hosted services. One of the main downsides\nof self-hosting is that you have to take care of your own services — who would’ve thought! I really, really don’t want to take care of a PostgreSQL\ncluster, even if it’s a single-node one. You might argue that it’s just one node, what’s to be taken care of? Identity maps, backups, authentication…\nI can just use SQLite, have all my data in one file (per service), you cannot access it remotely, so that’s covered,\nbackups are a (side note: \nsqlite3 db.sqlite \".backup db.sqlite.bak\"\n), and the only authentication and authorisation is in the file system permissions.\nI don’t even have to think about what needs to be taken care of, there isn’t really anything I need to worry about!\nOf course, serious usage should make me worried about the throughput, write blocking, cache sizes, journal modes and so on. But at my scale, that’s not really an issue.\nIt reduces the cognitive complexity, let’s put it this way.
The less I need to think about things when self-hosting
\n→ The easier it is to self-host
\n→ The more satisfied I am
\n→ The more I am encouraged to keep on doing it!
It’s the same as: the cleaner and well-stocked my kitchen is, the more I am encouraged to cook and eat healthy.\nThe sharper my knives, the more I enjoy cutting onions.\nHence, sorry PostgreSQL, but I don’t want you here. If this is not a problem for you, then I strongly encourage you to give Miniflux a try.
\nWhich brings me to…
\n![\"A](\"https://tudorr.ro/static/pictures/yarr.avif\")
Yarr is, in some regards, even more minimalistic than Miniflux.\nIt doesn’t provide fancy integrations, it doesn’t support multiple users (it does support authentication, though).\nIt doesn’t do much, really.
\nBut what it does, it does well. And it has a couple of niceties, such as the possibility to change\nthe auto-refresh interval — or disable it altogether —\nor to switch between the predefined colour palettes.\nThere are also some keyboard shortcuts, and the ability to save (star ⭐️) articles for later.\nIt has a small search box, too.
\nWhat I find to be the best feature, though, is the UI: it’s simple, yet it looks good.\nAnd on a large enough screen, it has this three-column layout, with the feeds, articles, and\narticle view columns, really nice for reading the news. Yet, if your screen — or browser window — is smaller, it\nwill switch to a one-column layout. I find it clean and practical, and I like its choice of\nicons. It also has a couple dynamic elements, just enough to enhance it and not to be defined by it.\nFor example, when you manually refresh the feeds, it shows the progress in the bottom-left corner,\ntogether with a little animated spinner. And when you add a new feed, in the event that it finds multiple\nfeeds for the same websites, it will present the choices in the same “Add feed” modal.\nIt’s completely a single-page application, unlike Miniflux, which is the opposite: Miniflux\nis a completely server-side application. And I’m usually against that, but in the case of Yarr,\nthere isn’t any “routing”, there aren’t different screens, it’s just one large feed-reading window,\nand the dynamic elements only enhance the interaction.
\nI have even more appreciation for the developer for publishing this text file\nin the repo with the rationale behind Yarr and the sources of inspiration for the user interface.\nIt would be great if more people did this, to get a small glimpse of what they were thinking\nof when they started a project. The message of the commit that added this file is let’s be honest
, nice touch.
Yarr is a bit stuck in time: the author claims that the software is complete,\nthat it fulfils all their expectations, and that no pull requests with new features will be merged.\nI think this is good for a feed reader: it does what it says on the tin, without burdening\nanyone with maintaining the program, and without burdening users with upgrades, config changes and the like.\nThere isn’t much to be done, it just works; I applaud the lack of feature creep.\nContrast this to Miniflux, which is an actively maintained and updated project, with many features (not a bad thing!).
\nYarr can also be used as a desktop application: it supports being launched from the system tray, of all things.\nThis is a feature that I will never use, yet the author wanted to use their program this way,\nso they implemented it. For self-hosting, it supports only a single user account (or no account),\nprotected with a username and password.
\nAnd when it comes to storage, everything is stored in a single SQLite database.\nThere are no other data folders kept around, just that (side note: \nThis might be a con if you care about cross-compiling the binary, or if anything else breaks because of cgo. Use of cgo is required,\nbecause SQLite is written in C, and is embedded in programs directly by linking to its shared library.
\nRead: cgo is not Go.\n).\nI love it so much, it’s one of my favourite programs I use on a daily basis.\nOpen-source, useful, easy to self-host, easy to use, works with a mouse, works with a keyboard, works with a touchscreen. I don’t have to think about it.
And whenever I’m thinking of it, it’s always out of gratefulness. This is the most I can wish\nfrom software, in my opinion.
\nAll in all, I think switching to feed readers has brought a net improvement in my quality of life, when it\ncomes to consuming information. If Yarr sounds compelling to you, I suggest you give it a try.\nYou don’t have to self-host it, you can just grab a binary release from GitHub and run it locally.
\nHacker News Daily: This is a feed containing the ten highest-rating articles on Hacker News which have not appeared on previous daily digests. I read HN from time to time, but the volume of submissions it receives is too high for me,\nthus I prefer visiting the website directly.
\nMy favourite link aggregator is Lobsters, yet I don’t have it in my feed reader, because I prefer just browsing\nthe website directly.
\nVarious blogs I enjoy reading. At the time of writing this article, in no particular order:
\nI only follow xkcd. I haven’t found other comics I would enjoy reading, yet.
\nI use my feed reader to read both Dutch news — I live here, and it’s very good language immersion —\nand Romanian news, to keep an eye on what’s going on there.
\nDutch:
\nAs you can see, I mostly read news from the public service.\nFor investigative journalism and more detailed stories,\nI prefer purchasing magazines and newspapers. It’s always nice to read from paper!
\nRomanian:
\nYouTube offers feeds for all their channels, which I use to subscribe to them without doing it from YouTube itself.\nYou should be able to just paste a YouTube channel in any reader, and it should just pick it up.
\nCurrently, I only follow Ars Technica and Teletext (not to be confused with the NOS Teletext).
\nTeletext is, according to their website, the newest platform dedicated to pop culture in Romania
.\nI’m more of a technical person, I don’t have all the cultural knowledge Teletext’s authors do,\nyet I really love reading their articles, precisely because it’s all new and different to me.
As I have highlighted above, podcasts are actually based on RSS, so nothing stops you from putting them in your feed reader.\nI personally use dedicated apps for podcasts like GNOME Podcasts on my computer, and AntennaPod on Android.
\nI also have a recipe website in there which I really like, RecipeTin Eats.\nOf course the best way to cook is to train your intuition and just make something you’d fancy with the ingredients you have,\nbut it’s always good to get some inspiration from time to time, which is the reason why I have it in my feed reader.
","date_published":"Thu, 04 Apr 2024 12:22:58 GMT"},{"id":"https://tudorr.ro/blog/2024-01-04-easy-spell-checking/","url":"https://tudorr.ro/blog/2024-01-04-easy-spell-checking/","title":"Hacky Spell Checking for Blog Posts","content_html":"While writing another post for this blog, I came up\nwith a dirty hack to quickly check the spelling and grammar of my text.
\nI use Helix as my text editor of choice, also when authoring blog posts.\nIt does, however, come with a major shortcoming: no support for spell checking whatsoever!\nOr maybe it does support spell checking: you can configure custom language servers,\nwhich should allow me to use LTEX as an\n (abbreviation: Language Server Protocol)\ninterface to LanguageTool,\na pretty decent spell and grammar checker.\nJens Getreu wrote an article about it, among a couple other things.
\nBut that’s too much work, and I’m very prone to yak shaving!
\nSet the contenteditable attribute on your content.\nNow, your browser will check your content!
![\"Screenshot](\"https://tudorr.ro/static/pictures/easy-spell-checking.avif\")
You can exploit this even further by installing extensions to alter the\nspell checking behaviour of your browser, such as\nthe LanguageTool browser extension.
\nTime it took to set this up: 30 seconds.
\nAs you can see from the picture above, my website has a “staging mode”.\nI can use a flag — which is descriptively called release —\nto show content only when the website is in one of the two modes: staging and release.\nTo “enable” spell checking, I only had to change this line:
- <main class=\"p-2 text-lg\">\n+ <main class=\"p-2 text-lg\" {{ if !release }}contenteditable{{ /if }}>\nI gave myself a good pat on the back after discovering this!
","date_published":"Mon, 04 Mar 2024 14:27:26 GMT"},{"id":"https://tudorr.ro/blog/technical/2023/10/12/modern-shell-tools/","url":"https://tudorr.ro/blog/technical/2023/10/12/modern-shell-tools/","title":"Postmodern Command Line Tools","content_html":"The standard UNIX command line tools are still enjoying a multi-decade-long reign,\nwhether we’re talking about the bare-bones echo and cat, the familiar ls,\ndate and (side note: I can’t be the only one who’s still using cal, right?),\nthe weirdly-named text processing utilities tr, sed, cut, grep and\ntheir main nemesis, awk, or the ubiquitous text editors, vi and the 30+ year old Vi IMproved.
Here are a couple “postmodern” CLI and TUI tools I like to use. They are mainly\nalternatives to the well-established tools that try to take on trivial problems,\nlike string processing, finding files, looking for strings in files and so on.
\nThese tools have a couple things in common, such as:
\nI don’t know about you but I cannot remember for the love of what’s holy find’s flags.\nIt also grinds my gears heavily that it doesn’t automatically ignore files declared in .gitignore,\nalthough find was made when git’s creator was a toddler.
fd, when supplied with an argument, does what you probably want to do most of the time\nwhen looking for files in a directory: look for any files that contain a pattern in the filename.\nAnd it does that with pretty colours! It respects LS_COLORS even, so the colours have meaning!
Another nicety is that it uses Rust’s regex library.\nGNU Find offers many regex types, which can be quite confusing. Run find -regextype help to see them all.
Example usage, finding all Markdown files in the blog’s source code repo:
\n$ fd .md$\n404.md\nREADME.md\n_posts/2021-01-24-the-wayland-experience.md\n_posts/2023-08-29-modern-shell-tools.md\n_posts/2023-09-11-low-tech.md\nabout.md\nindex.md\ntechnical.md\nwriting.md\nYou can find it in your package manager probably: rust-fd-find in Fedora, Ubuntu, and EPEL.\nRepology.
sd is a lesser-known postmodern alternative to sed. It simplifies greatly the most common\nuse case for sed — to replace and extract data — and also uses (side note: No more sed -E!).\nThe examples also show nice ways to use it together with fd.
Example usage:
\n$ # -f i stands for `flags: case insensitive’\n$ sd -f i emacs VIM file.txt\n$ # sd does by default in-place substitution, so this will print nothing.\n$ # we can do, however, the following\n$ sd -f i -p emacs VIM file.txt\nThe original VIM did not have Lisp in it. The lower level language, the non-interpreted language—was PDP-10 Assembler.\nThe interpreter we wrote in that actually wasn’t written for VIM, it was written for TECO.\nText fragment taken from here.
\nThis one is also available in the repositories of some popular distros: rust-sd for Fedora and Ubuntu (but no EPEL).\nRepology.
Currently my text editor of choice. It’s a modal text editor, like {neo,}vim,\nbut it sacrificies extensibility — no plugin support! — for natively implementing\nvarious niceties, such as:
\nw and you are presented window manipulation functions,It’s also heavily inspired by Kakoune in the way commands are used:\nfirst you press the key that selects the text object you operate on, and (side note: Unlike Vim, which does it in reverse.).\nSo, if I want to change the next word, I’d press wc. After pressing w, the selection is also highlighted.
This one is not as widely-available in distro repositories (yet), but it’s easy to install\nif you happen to have the Rust toolset installed. Repology.
\nI like to describe Nushell as “what if you take Powershell’s object oriented everything approach, but you apply it to UNIX”.\nInstead of using objects, Nushell operates on tables made of records reminescent of JSON’s data types.\nYou have mostly lists, key-value hash tables, numbers, strings, and booleans, plus a couple specific data types like\ndate, filesize, durations, but also ranges and closures.
\nWhat I really like about Nu is that it strives to make all of this cross-platform.\nThe idea is that you learn once the Nu language and commands, and you use them on all these systems in a portable manner.\nSo, no platform-specific ls flags for example. Of course, this also means that Nu tries to avoid platform-specific commands in the first place.
Besides coming with a very nice command language, it comes with many other niceties,\nsuch as a help system that has syntax highlighted examples and example outputs (for example, type help ls),\na command search mode (after typing a pipe, you can press F1 and a menu pops up with a command search and small help overview),\nsyntax highlighted commands as you type, and quite decent completion.
There aren’t many completion definitions for all the other “normal” commands you may use,\nbut you can easily declare completers in your config, such as using Fish as a completer.\nFish comes with stellar completion support, so now Nu does too!
\nMany string processing (and not only) tools can feel obsolete when using Nu, which\ncomes with its own typed functions and function programming-esque constructs.\nThe sd utility named above, for example, is built in, in the form of str replace.\nWe can run help str replace and we get a nice coloured, and syntax highlighted, help\npage right in the terminal. All these help pages are also on the website.
Go ahead and try it out!
","date_published":"Thu, 12 Oct 2023 18:23:29 GMT"},{"id":"https://tudorr.ro/blog/technical/2021/01/26/the-wayland-experience/","url":"https://tudorr.ro/blog/technical/2021/01/26/the-wayland-experience/","title":"Should You Write a Wayland Compositor?","content_html":"Wayland is the “new” (13 years old already) display server technology on Linux, which is supposed to replace\nthe antiquated X11. It promises better security, performance, portability, everything, compared to X11, and\nit sure does deliver, provided that you’re not using unsupported graphics cards.\nYou can watch this talk / rant about X11 to get an idea about how bad it is.
\nSome power users also haven’t switched to Wayland because their window manager doesn’t have a\nWayland equivalent (XMonad, Awesome, Bspwm and the others, all having their unique feature-set). Some may wonder “hmm maybe I can port it myself”. I wrote this post to make you (re)consider that.
\nOne of the main ideas of Wayland is that it’s merely a specialised IPC protocol, and the communication is\nstrictly between the clients (applications on your screen mainly) and the server.
\nThe server is what we are interested in now. Unlike X11, where the server is X.Org and the window manager\nis just an X11 client with special privileges, on Wayland the window manager is also a server and a compositor. In fact, the correct terminology is “Wayland compositor”. This piece of software has the\ntask of doing everything it wants with the clients, which is usually among the lines of showing them\non the screen and giving them inputs events from your keyboards, mice, touchscreens etc. In fact, no one\nstops you from leaving window management as an API.
\nOn my laptop, I prefer using GNOME under Wayland, because it has the most stable, fully-featured and “just works”\nexperience there is. However, I wanted a better way of organising windows on my screen than leaving them\nshuffled around. Tiling was also not good, because every time you launch a new window, the others get\nshrunk to fit, which is not good on a small laptop display.
\nLuckily I discovered PaperWM. It’s perfect for laptops, instead of\nshrinking your windows, it just renders them off-screen. To switch between one or the other you can just\nflick three fingers on the touchpad. It’s great. It’s also “glossy” and polished and has great UI and animations.
\nIt has some disadvantages though, mainly concerning its speed. I find it sluggish, and it’s no surprise that\nlike all other GNOME Shell extensions, it’s a JavaScript behemoth. PaperWM still uses the GNOME Tweener\nframework for its animations, which is entirely written in JS. Because of that, it needs to communicate with the main\nGNOME compositor process on each operation. And because we’re talking animations, said operations happen\nfor every frame. That means there is JS executed for every frame, 60 times a second. It’s horrible!
\nBTW nowadays GNOME uses animations implemented in C (since GNOME 3.34), the “GNOME runs JS on every frame” statement is for the most part false today.
\nBecause of the inefficiency of PaperWM, it also means that it drains my battery quickly, which is bad for a laptop!
\nSo, because I am a student with a lot of free time, together with Alex we figured we shall develop a scrollable tiling (PaperWM-like) Wayland compositor. It’s called Cardboard (get it??). It also has the nicety of\nbeing controlled and configured by a remote control program, like bspwm.\nNeither Alex nor I had experience with any kind of Wayland development. However, I wrote an X11 window manager,\nsome XRandR querying utilities and a window rule daemon. Alex has experience\nwith graphics programming and C++ devilry.
\nOur “tech stack” was wlroots, which is an exceptional Wayland compositor library (apart from its lack of documentation). The description from the README is spot on:
\n\n\nAbout 50,000 lines of code you were going to write anyway.
\n
I (because I wrote most of the Wayland-interfacing code) realised soon that there is still a lot of boilerplate involved in writing a compositor, much more than for an X11 window manager,\nnamely setting up your own rendering code, registering and storing input devices and screens in your own data structures, passing input events to windows, calculating bounds for bars\nand other overlays (courtesy of layer-shell) and others. X11 handles all of this for you,\nthe window manager just reacts to events regarding window inputs to establish its behaviour. With Wayland, you handle everything, even with wlroots. The upside is that if you don’t like the way X11 does something\n(which is a given), not only that you can do it in your own way on Wayland, you are required to do so.
Because we weren’t really prepared for what writing a compositor involved, we thought that it must be approached like a “normal” program: split code into modules, each with their own responsibilities, call wlroots to do its thing, the usual stuff.\nWe are writing a program in our way and wlroots lets us “interface with Wayland”. Or so we thought.
\nWe were as careful as possible to separate responsibilities into different structures / classes, yet we ended up with most functions taking a pointer to Server as their first parameter.\nServer is a singleton holding all the other “sub-modules” in it. It represents the compositor itself. The reason most functions need Server is that everything is related to everything,\nnot because of a mistake in structuring the code, but by design. This is what a compositor requires, the problem of writing a compositor is somewhat complex because it has a great deal variables ranging\nfrom input events, drawing on the screen, creating behaviours that the user leverages, reacting to application events. All of them can affect the other, you can not separate the thing into modules.\nThe best you can do is separate the code in different files and folders based on some criteria, like grouping data structures with related routines.
Excerpt from Seat.h:
/// Hides the \\a view from the screen without unmapping. Happens when a Workspace is deactivated.\nvoid hide_view(Server& server, View& view);\n/// Gives keyboard focus to a plain surface (OR xwayland usually)\nvoid focus_surface(struct wlr_surface* surface);\n/**\n * \\brief Sets the focus state on \\a view. Auto-scrolls the Workspace if it's tiled.\n *\n * If \\a view is null, the previously focused view will be unfocused and no other view will be focused.\n */\nvoid focus_view(Server& server, OptionalRef<View> view, bool condense_workspace = false);\n/// Marks the layer as receiving keyboard focus from this seat.\nvoid focus_layer(Server& server, struct wlr_layer_surface_v1* layer);\n/**\n * \\brief Focus the most recently focused view on \\a column.\n *\n * \\param column - must be from this workspace\n */\nvoid focus_column(Server& server, Workspace::Column& column);\n/// Removes the \\a view from the focus stack.\nvoid remove_from_focus_stack(View& view);\n\nvoid begin_move(Server& server, View& view);\nvoid begin_resize(Server& server, View& view, uint32_t edges);\nvoid begin_workspace_scroll(Server& server, Workspace& workspace);\nvoid process_cursor_motion(Server& server, uint32_t time = 0);\nvoid process_cursor_move(Server&, GrabState::Move move_data);\nvoid process_cursor_resize(Server&, GrabState::Resize resize_data);\nvoid process_swipe_begin(Server& server, uint32_t fingers);\nvoid process_swipe_update(Server& server, uint32_t fingers, double dx, double dy);\nvoid process_swipe_end(Server& server);\nvoid end_interactive(Server& server);\nvoid end_touchpad_swipe(Server& server);\n\n/// Updates the scroll of the workspace during three-finger swipe, taking in account speed and friction.\nvoid update_swipe(Server& server);\n\n/// Returns true if the \\a view is currently in a grab operation.\nbool is_grabbing(View& view);\n\n/// Returns the workspace under the cursor.\nOptionalRef<Workspace> get_focused_workspace(Server& server);\n\n/// Moves the focus to a different workspace, if the workspace is already on a monitor, it focuses that monitor\nvoid focus(Server& server, Workspace& workspace); // TODO: yikes, passing Server*\nAs an example, let’s take damage tracking. For starters, because the compositor is also tasked with rendering the content and displaying it on the screen, we have some rendering code that runs on every frame.\nDamage tracking is tracking which areas of the screen have changed in an amount of time. An example would\nbe typing a letter in the terminal. The place where the cursor is changes to the letter you typed, and the cursor advances.\nIf there is no change, the frame is not rendered, as it would look\nexactly the same as the previous one, which would be a waste of processor time.\nThis way, instead of re-rendering everything 60 times a second (I assume that you use a common display), we can render and paint as little as possible to account for the changed region.\nYou can read an introduction to damage tracking written by one of the main developers of wlroots.
\nI have just implemented the most basic form of damage tracking: do not render the frame if nothing on the screen changes.\nIt doesn’t track the damage itself, just that it exists.\nTo do this, I first added a wlr_output_damage object to my Output structure:
struct Output {\n struct wlr_output* wlr_output;\n struct wlr_output_damage* wlr_output_damage;\n+ struct wlr_box usable_area;\n ...\nThis structure tracks the damage accumulated per-output, as rendering is also per-output (this means that you can use screens of different refresh rates, yay!). However, to make this initial attempt at\ndamage tracking easier, I decided to trigger rendering for all attached outputs. I added a set_dirty() function to the OutputManager class that does just that:
void OutputManager::set_dirty()\n{\n for (auto& output : outputs) {\n wlr_output_damage_add_whole(output.wlr_output_damage);\n }\n}\nThis marks every output as entirely damaged, and as such triggers a render.
\nWith this function set into place, I had to identify when does the “screen” change, namely when a window “commits” (changes its contents) and when a window is moved. The window move is one example\nof the way everything is related to everything in the compositor. Before damage tracking, the View::move() (we call windows “views”) method just changed the x and y fields of the View structures.\nNow, a move must call a method of OutputManager, so we need to give that as a parameter. This is almost like giving Server as a parameter, as OutputManager is a singleton inside Server.
-void View::move(int x_, int y_)\n+void View::move(OutputManager& output_manager, int x_, int y_)\n{\n x = x_;\n y = y_;\n+ output_manager.set_dirty();\n}\nThat’s when it hit me that wlroots is more of a framework and the compositor is one of its modules. Thinking that wlroots is an “interface to Wayland” is plain wrong, as the Wayland server is the\nprogram that I am writing. The next refactor is going to make the Server instance global…
Now that we have a wlr_output_damage object in place and set_dirty() calls where they’re needed, we only need to call the render function when wlr_output_damage tells us instead of\nevery 1/60 seconds:
@@ -46,11 +46,12 @@ void register_output(Server& server, Output&& output_)\n server.output_manager->outputs.emplace_back(output_);\n auto& output = server.output_manager->outputs.back();\n output.wlr_output->data = &output;\n+ output.wlr_output_damage = wlr_output_damage_create(output.wlr_output);\n\n register_handlers(server,\n &output,\n {\n- { &output.wlr_output->events.frame, Output::frame_handler },\n+ { &output.wlr_output_damage->events.frame, Output::frame_handler },\n { &output.wlr_output->events.present, Output::present_handler },\n { &output.wlr_output->events.commit, Output::commit_handler },\n { &output.wlr_output->events.mode, Output::mode_handler },\nThis is not complete code for a basic damage tracking implementation with wlroots. You can see the whole commit here.
\nThis is an example of how wlroots provides yet another “module” that we can use in the grand scheme of the compositor. wlr_output_damage accumulates damaged rectangles in time and even turns these\nnumerous small rectangles into a big one as an optimisation. It also calls the frame handler when it’s needed, and this is not only just when something on the screen changed, but also when the underlying “backend”\nof the compositor changes. The simplest situation is when the compositor starts: it needs to render an initial frame so the screen isn’t pitch black.
All in all, I do not recommend writing your own compositor if you only want some gimmicky user interface.\nIn the X world there is a WM for every single way of window tiling plus a couple more. It doesn’t work like\nthat with Wayland, you will spend more time implementing basic compositor duties than\non your compositor’s unique features. Instead, if I were to rewrite Cardboard, I would rather do it\nas a Wayfire plugin or maybe as a KWin script. However, I think Wayfire is more “enthusiast-friendly”, as it uses protocols from\nwlroots such as layer-shell (for panels, overlays and backgrounds), gamma-control (for Redshift), screencopy (for screenshots) and others,\nallowing people to write tools that are not specific to the compositor.
Nevertheless, if you want to do it for the learning experience, I definitely recommend writing\na “full-fledged” compositor with wlroots, learning from other compositors (Sway,\ncage, Wayfire, hikari and others; cage is the simplest, hikari second simplest) and from their creators on\nIRC (#sway-devel on Freenode), they are very kind and knowledgeable.
There is also a discussion about introducing a high-level scene API in wlroots. Maybe when it will arrive, I will change my opinion.
\nRecommended lectures:
\nAlso, I suggest you write the compositor in either C, C++ or Zig (with zig-wlroots or just doing your thing, Zig is C compatible).\nSee this article on why the Rust bindings failed.
","date_published":"Tue, 26 Jan 2021 08:59:23 GMT"}]}