Learning a language is, to me, about grinding. Continously exposing yourself.

Ich lerne Deutsch. Oder, ich versuche Deutsch zu lernen.

I try to expose my self to the language via YouTube (thx Nils for the tip about 7 gegen Wild), but also news papers and just chatting with people. I’d say the biggest hurdle is that people find English easier than having me try to find and reorder the words, so practice at full speed is hard to find.

I guess I do the same for people trying to learn Swedish, and i really shouldn’t.

If you have tips for how to expose myself more to German – spoken or written – please drop a comment here or join the conversation at mastodon.

A few days ago Volker Krause posted this blog about the Nextcloud conference - a very interesting read.

One of the topics is the VFS (Virtual Filesystem-) API for the Linux desktop. Indeed that is a topic for us at ownCloud as well, and I like to share our perspective on it, discussing it in the scope of the free desktop.

The topic is very important, as “syncing” of data from and to cloud storages has changed over time. From having all files mirrored from client to server and vice versa, it has now shifted to keep all files in the cloud, and have them as so called placeholders on the desktop. That means that most files on the client appear with size zero to save space, but the complete filesystem structure is available.

If a user starts to interact with such a dehydrated file, the content is of the file is downloaded transparently utilizing the cloud system client, for example ownClouds desktop client. The same happens when an application accesses such a file. As a result, the placeholders look and behave like the normal filesystem we are used to.

On Windows and on MacOSX, the problem is kind of solved. Both have added APIs to their OS that can be used to implement the access of data on the cloud.

On Linux, we do not have this kind of API yet. That means that it is close to impossible to implement this user experience. Volker already said that desktop environment specific solutions probably do not scale, which I agree with.

At ownCloud we have looked into the implementation of a specific FUSE file system. That should certainly be possible, and is probably a part of the solution, but is considerable effort because of the asynchronous nature of the topic. Given that the market share of Linux desktop systems is pretty small it is not attractive for companies to invest a lot into a Linux only system. Here the power of community could make a difference again.

It would be best if we as open source community would come up with a shared solution as a free desktop standard, that might be oriented on one of the existing APIs, maybe the MacOSX File Provider API: A library and little framework that the linux desktop environments can work with abstracting the VFS.

While collaborating on that, all data clouds could implement the bindings to their storage. With that, the extra implementation efforts for the Linux solution hopefully wouldnt be dramatic any more.

Let’s call this system openVFS as a work title. How can we evolve it? I’d like to invite all interested parties to discuss in this temporary Github repo to collect ideas and opinions. There is also a little experimental code.

for what seems an eternity i've been running linux as my daily driver and evangilising to anyone that would listen.  however working as a carpenter and having a young family didn't leave much energy, let alone time to contribute.  but over the course of the last year slowly my contributions have been increasing.  as we age our bodies aren't capable of the physical things that you used to be easy and this combined with some other events have allowed me to able fulfil a long held ambition and contribute in a meaningful way to KDE and neon.  anyway enough self-reflection from this antopidean.

thanks to the patience of @jriddell, @sitter and @sgmoore i'm very happy to be helping get neon out the door, so to speak.  after the big plasma 6 push that started in March, i've been improving my ruby skills  trying to get neon's tooling into the best possible shape.  as qt6 started to filter down from unstable, it became apparent that compilation time saving hack of building qt5 and kde frameworks5 once for unstable and copying into the stable and user archives wasn't scaling any longer and couldn't be applied to qt6 and kf6.  after changing unstable to track the latest qt6.6 beta's at the request of the plasma devs, stable and user started building their very own qt5 and kf5.  besides a few late nights and much swearing at versioning problems it all went swimmingly. \0/

lately my energies have been split between trying to get more apps with a decicated kf6 branch into the experimental overlay and trying to keep up with the rapid rate of apps in unstable whose master branch has gone qt6/kf6 only.  as sgmoore pointed out porting pim6 alone has been quite the process.  with a raft of new red/failed job build's in unstable and qt6.6 beta4 just released, it looks like the pace isn't going to slow down for the forseeable future.  anyhow i'll sign off with the obligatory screenshot. cheery bye ;]

 

This is a longer one … it took long too.

I would apologise, but it is one of those things where it hurt me more than it will hurt you to read this extremely abbreviated version.

Setting up sudoedit¶

There is no two ways about it: sudoedit is the smart way to edit text files as root.

In short what sudoedit does is to:

1. sudo,
2. copy the file you want to edit into a temporary location,
3. edit that file with sudo $VISUAL, or failing that sudo $EDITOR, and
4. when you save and exit the editor, replace the original file with the temporary copy.

To set up Helix as the default text editor in the console, and KWrite when X11 or Wayland are available (both also for sudoedit), I did the following (in Fish shell):

set --universal --export EDITOR helix
set --universal --export VISUAL kwrite

Borg backup¶

Before I started messing with the Btrfs RAID, I decided to set up the backups.

I decided to stick with Borg, but to simplify setting up and running, used the Borgmatic wrapper. I have to say I am very pleased with it.

First, I installed it with yay borgmatic.

I took my sweet time to set it up, but most of that was spent reading the documentation and the extensive (over 900 lines) config file template.

But ultimately my /etc/borgmatic/config.yaml consist of less than 30 actual lines and, including all the reading, I was done in an afternoon. The interesting bits I share below.

Excludes¶

Some things are just not worth backing up and just waste resources. So I added the following:

exclude_from:
    - /etc/borgmatic/excludes

exclude_caches: true

exclude_if_present:
    - .nobackup

Then I slapped .nobackup into directories in my home that I typically do not want backed up like downloads, video, games and some (non-hidden) temporary folders.

The exclude_caches bit relies on people following the Cache Directory Tagging Specification, which from what I can tell, is a pretty rare occasion.

In fact, locate CACHEDIR.TAG on my system shows only three such files – I hope this improves as I install more things:

/home/hook/.cache/borg/CACHEDIR.TAG
/home/hook/.cache/fontconfig/CACHEDIR.TAG
/home/hook/.cargo/registry/CACHEDIR.TAG

## Temporary and backup files

*~
*.tmp
*.swp

## Temporary and cache directories

**/tmp
**/temp
**/cache
**/Cache
**/.cache
**/.Cache
**.cache
**.Cache
**.ccache

## Big files

*.iso
*.img

## Top level directories to always exclude

/dev
/etc/mtab
/media
/mnt
/proc
/run
/sys
/tmp
/var/cache
/var/tmp

## Local applications to be excluded (for more use `.nobackup`)

/home/*/.local/share/akonadi
/home/*/.local/share/baloo
/home/*/.local/share/lutris
/home/*/.local/share/Trash
/home/*/.local/share/Steam
/home/*/.kde4/share/apps/ktorrent
/home/*/.config/chromium
/home/*/.wine

I have to agree with the author of the CacheDir spec, caches are aplenty and it is hard to RegExp them all. If only everyone just put their cache in ~/.cache …

Relying on .nobackup is a new approach I tried, but let us see how it works out for me. I am cautiously optimistic, as it is much easier to just touch .nobackup than it is to sudoedit /etc/borgmatic/config.yaml, enter the password and copy paste the folder.

Checks¶

Backups are only worth anything if you can restore from them. Borg does offer checks, and Borgmatic offers a way to easily fine-tune which checks to run and how often.

With the below settings Borgmatic:

• every day when it creates a backup, also checks the integrity of the repository and the archives
• once a month, tries to extract the last backup
• every three months checks the integrity of all data

checks:
    - name: repository
      frequency: always
    - name: archives
      frequency: always
    - name: extract
      frequency: 1 month
    - name: data
      frequency: 3 months

Restrict access¶

Of course I encrypt my backups.

To further limit access to the backup server, accessing the backup server is only possible with an SSH key. Furthermore, the ~/.ssh/authorized_keys on the server restricts access only to the specific backup repository and only allows the borg serve command to be ran:

command="cd {$path_to_borg_repos}; borg serve --restrict-to-path {$path_to_borg_repos}/{$repo_of_this_machine}",no-pty,no-agent-forwarding,no-port-forwarding,no-X11-forwarding,no-user-rc {$ssh_key}

I already had Borg running on the backup server, so I merely needed to add another line like the above to the server and set up Borgmatic on Leza.

Automate it all¶

At the end I enabled systemctl enable --now borgmatic.timer to have systemd handle the “cronjob” on a daily basis.

Borgmatic does the heavy lifting of figuring what exactly (if anything) it needs to do that day, so that was super-simple.

Btrfs RAID¶

Well, RAID-ifying my Btrfs was quite a ride … and I have no-one to blame but myself¹ for the issues I had.

I also have to thank everyone on #btrfs IRC channel (esp. balrog, dicot, kepstin, multicore, opty, specing, Zygo); dalto from the EndeavourOS forum, and above all TJ from KDE’s Matrix channel for their help, helping me to dig myself out of the mess I made.

Below, I will document things as they should have been done, and add what I did wrong and how we fixed it in a expandable sub-section.

Do note that “what I should have done” is just me applying some hindsight to my errors, it may still have holes. Also the “what I did” parts are partially re-created from memory and omit a lot of trial and error of trying to fix stuff.

Add a new device¶

As I described in the my base install blog post, the partitioning of my Goodram PX500 SSD was (roughly):

• 1 GB – ESP
• 990 GB – LUKS + Btrfs
• 9 GB – LUKS + swap

The Goodram PX500 has 953,86 GiB of space.

But Samsung 970 Evo Plus has 931,51 GiB of space.

Although both drives are marketed as 1 TB (= 931,32 GiB), there was a difference of 20 GiB between them. And unfortunately in the wrong direction, so just the system partition on the Goodram SSD was larger than the whole Samsung SSD.

After plugging in the new SSD, what I should have done was just:

• create a 1 GiB fat32 partition at the beginning,
• (create a 10 GiB LUKS + swap partition at the end),
• create a LUKS partition of what-ever remains on the newly added drive.

… and simply not care that the (to-be) Btrfs partition is not the same size as on the old one.

If fact, I could have also just skipped making a swap partition on the Samsung SSD. If/when one of the drives dies, I would replace it with a new one, and could just create swap on the new drive anyway.

Was that what I did?

Of course bloody not! 🤦‍♂️

# Configuration file automatically written by the Calamares system installer
# (This file is written once at install time and should be safe to edit.)
# Enables support for LUKS full disk encryption with single sign on from GRUB.

# force installing /etc/crypttab even if hostonly="no", install the keyfile
install_items+=" /etc/crypttab /crypto_keyfile.bin "
# enable automatic resume from swap
add_device+=" /dev/disk/by-uuid/2d90af35-7e6a-40f8-8353-f20433d0f994 "

omit_dracutmodules+=" network cifs nfs brltty "
compress="zstd"

force_drivers+=" amdgpu "
add_dracutmodules+=" plymouth "

add_dracutmodules+=" resume "

# GRUB boot loader configuration

GRUB_DEFAULT='0'
GRUB_TIMEOUT='5'
GRUB_DISTRIBUTOR='EndeavourOS'
GRUB_CMDLINE_LINUX_DEFAULT='nowatchdog nvme_load=YES rd.luks.uuid=1a45a072-e9ed-4416-ac7e-04b69f11a9cc rd.luks.uuid=c82fca05-59d3-4595-969b-c1c4124d8559 rd.luks.uuid=2d90af35-7e6a-40f8-8353-f20433d0f994 rd.luks.uuid=2e91342f-3d19-4f75-a9a6-fc3f9798cb30 resume=/dev/mapper/luks-2d90af35-7e6a-40f8-8353-f20433d0f994 loglevel=3 splash quiet'
GRUB_CMDLINE_LINUX=""

# Preload both GPT and MBR modules so that they are not missed
GRUB_PRELOAD_MODULES="part_gpt part_msdos"

# Uncomment to enable booting from LUKS encrypted devices
GRUB_ENABLE_CRYPTODISK=y

# Set to 'countdown' or 'hidden' to change timeout behavior,
# press ESC key to display menu.
GRUB_TIMEOUT_STYLE=menu

# Uncomment to use basic console
GRUB_TERMINAL_INPUT=console

# Uncomment to disable graphical terminal
#GRUB_TERMINAL_OUTPUT=console

# The resolution used on graphical terminal
# note that you can use only modes which your graphic card supports via VBE
# you can see them in real GRUB with the command `videoinfo'
GRUB_GFXMODE=auto

# Uncomment to allow the kernel use the same resolution used by grub
GRUB_GFXPAYLOAD_LINUX=keep

# Uncomment if you want GRUB to pass to the Linux kernel the old parameter
# format "root=/dev/xxx" instead of "root=/dev/disk/by-uuid/xxx"
#GRUB_DISABLE_LINUX_UUID=true

# Uncomment to disable generation of recovery mode menu entries
GRUB_DISABLE_RECOVERY='true'

# Uncomment and set to the desired menu colors.  Used by normal and wallpaper
# modes only.  Entries specified as foreground/background.
#GRUB_COLOR_NORMAL="light-blue/black"
#GRUB_COLOR_HIGHLIGHT="light-cyan/blue"

# Uncomment one of them for the gfx desired, a image background or a gfxtheme
GRUB_BACKGROUND='/usr/share/endeavouros/splash.png'
#GRUB_THEME="/path/to/gfxtheme"

# Uncomment to get a beep at GRUB start
#GRUB_INIT_TUNE="480 440 1"

# Uncomment to make GRUB remember the last selection. This requires
# setting 'GRUB_DEFAULT=saved' above.
#GRUB_SAVEDEFAULT=true

# Uncomment to disable submenus in boot menu
GRUB_DISABLE_SUBMENU='false'

# Probing for other operating systems is disabled for security reasons. Read
# documentation on GRUB_DISABLE_OS_PROBER, if still want to enable this
# functionality install os-prober and uncomment to detect and include other
# operating systems.
#GRUB_DISABLE_OS_PROBER=false

Automate decryption¶

Having four LUKS-encrypted partitions also means needing to decrypt all of them.

To make things easier, I added the same key that nvme1n1p2 uses also to all the other three partitions:

cryptsetup luksAddKey /dev/nvme1n1p3 /crypto_keyfile.bin    # new swap @ Goodram
cryptsetup luksAddKey /dev/nvme0n1p3 /crypto_keyfile.bin    # btrfs @ Samsung
cryptsetup luksAddKey /dev/nvme0n1p2 /crypto_keyfile.bin    # swap @ Samsung

And then I added them also to /etc/crypttab:

# <name>                                    <device>                                        <password>              <options>

### Goodram

## root in RAID1
luks-1a45a072-e9ed-4416-ac7e-04b69f11a9cc   UUID=1a45a072-e9ed-4416-ac7e-04b69f11a9cc       /crypto_keyfile.bin     luks

## swap
luks-2d90af35-7e6a-40f8-8353-f20433d0f994   UUID=2d90af35-7e6a-40f8-8353-f20433d0f994       /crypto_keyfile.bin     luks


### Samsung

## root in RAID1
luks-c82fca05-59d3-4595-969b-c1c4124d8559   UUID=c82fca05-59d3-4595-969b-c1c4124d8559       /crypto_keyfile.bin     luks

## swap
luks-2e91342f-3d19-4f75-a9a6-fc3f9798cb30   UUID=2e91342f-3d19-4f75-a9a6-fc3f9798cb30       /crypto_keyfile.bin     luks

Even after this, I still need to enter LUKS password thrice:

• before GRUB to unlock the Goodram SSD’s root partition,
• before GRUB to unlock the Samsung SSD’s root partition,
• during systemd to unlock all four partitions.

If I could shorten this down to just once, it would be even nicer. But that is as far as I managed to get so far. Happy to hear suggestions, of course!

Add new drive to Btrfs to make RAID1¶

On the new Samsung SSD the nvme1n1p2 partition is LUKS + Btrfs, but when adding a new device to Btrfs RAID with btrfs device add, it expects the partition to be without a file system.

This was a(nother self-inflicted) problem.

I could probably avoid this if I did it in CLI – and perhaps even in KDE Partition Manager, if I spent more time with it.

But now I had to deal with it.

Initially I planned to simply use --force the `btrfs device, but was quickly told by the Btrfs gurus, that there was a much safer option:

So I used wipefs to hide the file system:

wipefs --all /dev/disk/by-uuid/a19847bc-d137-4443-9cd5-9f311a5d8636

Then I had to add the device to the same Btrfs mount point:

btrfs device add /dev/mapper/luks-c82fca05-59d3-4595-969b-c1c4124d8559 /

And finally convert the two devices into Btrfs RAID1³ with:

btrfs balance start -mconvert=raid1,soft /
btrfs balance start -dconvert=raid1,soft /

At the end of all this my /etc/fstab looks like:

# /etc/fstab: static file system information.
# Use 'blkid' to print the universally unique identifier for a device; this may
# be used with UUID= as a more robust way to name devices that works even if
# disks are added and removed. See fstab(5).
#
# <file system>                                        <mount point>    <type>  <options>                      <dump> <pass>

## ESP @ Goodram
UUID=B33A-4C29                                          /boot/efi        vfat   noatime                              0 2

## ESP backup @ Samsung
UUID=44D2-04AD                                          /mnt/backup_efi  vfat   noatime                              0 2

## btrfs @ Goodram (in RAID1 with Samsung)
/dev/mapper/luks-1a45a072-e9ed-4416-ac7e-04b69f11a9cc   /                btrfs  subvol=/@,noatime,compress=zstd      0 0
/dev/mapper/luks-1a45a072-e9ed-4416-ac7e-04b69f11a9cc   /home            btrfs  subvol=/@home,noatime,compress=zstd  0 0
/dev/mapper/luks-1a45a072-e9ed-4416-ac7e-04b69f11a9cc   /var/cache       btrfs  subvol=/@cache,noatime,compress=zstd 0 0
/dev/mapper/luks-1a45a072-e9ed-4416-ac7e-04b69f11a9cc   /var/log         btrfs  subvol=/@log,noatime,compress=zstd   0 0

## swap @ Goodram
/dev/mapper/luks-2d90af35-7e6a-40f8-8353-f20433d0f994   swap             swap   defaults                             0 0

## swap @ Samsung
/dev/mapper/luks-2e91342f-3d19-4f75-a9a6-fc3f9798cb30   swap             swap   defaults                             0 0

## tmpfs
tmpfs                                                   /tmp             tmpfs  noatime,mode=1777                    0 0

And with that my Btrfs RAID1 was basically done. 😌

There were some smart things to do still …

Automate Btrfs maintenance¶

According to the Btrfs documentation:

[Btrfs scrub is an] online filesystem checking tool. Reads all the data and metadata on the filesystem and uses checksums and the duplicate copies from RAID storage to identify and repair any corrupt data.

Which is one of the main reasons I embarked on this convoluted set-up adventure 😅

After consulting the Arch Wiki: Btrfs and the gurus on #btrfs IRC channel, it turns out I only needed to enable systemctl enable btrfs-scrub@-.timer.

The wiki says that @- equals the / mount point, @home equals /home mount point, etc., which suggests one should scrub each of the subvolumes / mount points.

But it turns out (at least the way I have things up) scrubbing / (i.e. @-) is perfectly enough, as it scrubs the whole device(s) anyway.

Re-introduce the “reserve tank”¶

Since I was resizing the original Btrfs partition, I wanted to re-introduce the “reserve tank”.

In my case I started with 0 Bytes of Device slack, as sudo btrfs filesystem usage / -T shows:

Overall:
    Device size:                   1.78TiB
    Device allocated:            120.06GiB
    Device unallocated:            1.67TiB
    Device missing:                  0.00B
    Device slack:                    0.00B
    Used:                        114.63GiB
    Free (estimated):            864.45GiB      (min: 854.45GiB)
    Free (statfs, df):           862.56GiB
    Data ratio:                       2.00
    Metadata ratio:                   2.00
    Global reserve:              160.89MiB      (used: 0.00B)
    Multiple profiles:                  no

                                                         Data     Metadata System
Id Path                                                  RAID1    RAID1    RAID1    Unallocated Total     Slack
-- ----------------------------------------------------- -------- -------- -------- ----------- --------- -----
 1 /dev/mapper/luks-1a45a072-e9ed-4416-ac7e-04b69f11a9cc 58.00GiB  2.00GiB 32.00MiB   864.52GiB 924.55GiB 0.00B
 2 /dev/mapper/luks-c82fca05-59d3-4595-969b-c1c4124d8559 58.00GiB  2.00GiB 32.00MiB   860.74GiB 920.77GiB 0.00B
-- ----------------------------------------------------- -------- -------- -------- ----------- --------- -----
   Total                                                 58.00GiB  2.00GiB 32.00MiB     1.67TiB   1.78TiB 0.00B
   Used                                                  56.18GiB  1.14GiB 16.00KiB

To add some slack / “reserve tank” to the Btrfs file system, I had to run:

sudo btrfs filesystem resize 1:-10G /
sudo btrfs filesystem resize 2:-10G /

The first command reduced the file system on the device ID 1 by 10 GiB, the second one reduced it on device ID 2.

As a result, I ended up 20 GiB of Device slack, 10 GiB on each drive, as sudo btrfs filesystem usage / -T shows:

Overall:
    Device size:                   1.78TiB
    Device allocated:            120.06GiB
    Device unallocated:            1.67TiB
    Device missing:                  0.00B
    Device slack:                 20.00GiB
    Used:                        114.63GiB
    Free (estimated):            854.45GiB      (min: 854.45GiB)
    Free (statfs, df):           852.56GiB
    Data ratio:                       2.00
    Metadata ratio:                   2.00
    Global reserve:              160.89MiB      (used: 0.00B)
    Multiple profiles:                  no

                                                         Data     Metadata System
Id Path                                                  RAID1    RAID1    RAID1    Unallocated Total     Slack
-- ----------------------------------------------------- -------- -------- -------- ----------- --------- --------
 1 /dev/mapper/luks-1a45a072-e9ed-4416-ac7e-04b69f11a9cc 58.00GiB  2.00GiB 32.00MiB   854.52GiB 914.55GiB 10.00GiB
 2 /dev/mapper/luks-c82fca05-59d3-4595-969b-c1c4124d8559 58.00GiB  2.00GiB 32.00MiB   850.74GiB 910.77GiB 10.00GiB
-- ----------------------------------------------------- -------- -------- -------- ----------- --------- --------
   Total                                                 58.00GiB  2.00GiB 32.00MiB     1.67TiB   1.78TiB 20.00GiB
   Used                                                  56.18GiB  1.14GiB 16.00KiB

How to restore from a failed drive¶

This is more a note to future self.

When one of the drives dies:

1. turn off laptop
2. physically remove the faulty drive
3. turn laptop back on and during boot mount the remaining drive as “degraded”
4. buy new drive (use laptop normally in the meantime)
5. when it arrives, turn off laptop
6. put in replacement drive
7. turn laptop back on and run btrfs replace

That is assuming you do not have a spare at hand. If you have it, just skip steps 3-5.

Create fallback ESP¶

So, with that I should be well equipped for when one of the drives dies.

But wait! There is an important part missing!

I cannot boot if ESP is also dead.

Remember the /mnt/backup_efi? Now it is time to make use of it.

Making sure the backup ESP includes everything, takes just a simple:

rsync --archive --delete /boot/efi/ /mnt/backup_efi

And to make sure this happens regularly enough, I decide to create a systemd service that triggers rsync every time I reboot or shutdown my computer.

For that I put into /etc/systemd/system/sync-efi.service the following:

[Unit]
Description=Sync EFI partitions
DefaultDependencies=no
Before=shutdown.target

[Service]
Type=oneshot
ExecStart=/usr/bin/rsync --archive --delete /boot/efi/ /mnt/backup_efi
TimeoutStartSec=0

[Install]
WantedBy=shutdown.target

Of course, the service unit should be enabled too:

systemctl enable sync-efi.service

hook out → well that was a rollercoaster ride

Recently, I’ve stumbled across some behavior of C++ lambda captures that has, initially, made absolutely no sense to me. Apparently, I wasn’t alone with this, because it has resulted in a memory leak in QtFuture::whenAll() and QtFuture::whenAny() (now fixed; more on that further down).

I find the corner cases of C++ quite interesting, so I wanted to share this. Luckily, we can discuss this without getting knee-deep into the internals of QtFuture. So, without further ado:

Time for an example

Consider this (godbolt):

#include <iostream>
#include <functional>
#include <memory>
#include <cassert>
#include <vector>

struct Job
{
    template<class T>
    Job(T &&func) : func(std::forward<T>(func)) {}

    void run() { func(); hasRun = true; }

    std::function<void()> func;
    bool hasRun = false;
};

std::vector<Job> jobs;

template<class T>
void enqueueJob(T &&func)
{
    jobs.emplace_back([func=std::forward<T>(func)]() mutable {
        std::cout << "Starting job..." << std::endl;
        // Move func to ensure that it is destroyed after running
        auto fn = std::move(func);
        fn();
        std::cout << "Job finished." << std::endl;
    });
}

int main()
{
    struct Data {};
    std::weak_ptr<Data> observer;
    {
        auto context = std::make_shared<Data>();
        observer = context;
        enqueueJob([context] {
            std::cout << "Running..." << std::endl;
        });
    }
    for (auto &job : jobs) {
        job.run();
    }
    assert((observer.use_count() == 0) 
                && "There's still shared data left!");
}

Output:

Starting job...
Running...
Job finished.

The code is fairly straight forward. There’s a list of jobs to which we can be append with enqueueJob(). enqueueJob() wraps the passed callable with some debug output and ensures that it is destroyed after calling it. The Job objects themselves are kept around a little longer; we can imagine doing something with them, even though the jobs have already been run.
In main(), we enqueue a job that captures some shared state Data, run all jobs, and finally assert that the shared Data has been destroyed. So far, so good.

Now you might have some issues with the code. Apart from the structure, which, arguably, is a little forced, you might think “context is never modified, so it should be const!”. And you’re right, that would be better. So let’s change it (godbolt):

--- old
+++ new
@@ -34,7 +34,7 @@
     struct Data {};
     std::weak_ptr<Data> observer;
     {
-        auto context = std::make_shared<Data>();
+        const auto context = std::make_shared<Data>();
         observer = context;
         enqueueJob([context] {
             std::cout << "Running..." << std::endl;

Looks like a trivial change, right? But when we run it, the assertion fails now!

int main(): Assertion `(observer.use_count() == 0) && "There's still shared data left!"' failed.

How can this be? We’ve just declared a variable const that isn’t even used once! This does not seem to make any sense.
But it gets better: we can fix this by adding what looks like a no-op (godbolt):

--- old
+++ new
@@ -34,9 +34,9 @@
     struct Data {};
     std::weak_ptr<Data> observer;
     {
-        auto context = std::make_shared<Data>();
+        const auto context = std::make_shared<Data>();
         observer = context;
-        enqueueJob([context] {
+        enqueueJob([context=context] {
             std::cout << "Running..." << std::endl;
         });
     }

Wait, what? We just have to tell the compiler that we really want to capture context by the name context – and then it will correctly destroy the shared data? Would this be an application for the really keyword? Whatever it is, it works; you can check it on godbolt yourself.

When I first stumbled across this behavior, I just couldn’t wrap my head around it. I was about to think “compiler bug”, as unlikely as that may be. But GCC and Clang both behave like this, so it’s pretty much guaranteed not to be a compiler bug.

So, after combing through the interwebs, I’ve found this StackOverflow answer that gives the right hint: [context] is not the same as [context=context]! The latter drops cv qualifiers while the former does not! Quoting cppreference.com:

Those data members that correspond to captures without initializers are direct-initialized when the lambda-expression is evaluated. Those that correspond to captures with initializers are initialized as the initializer requires (could be copy- or direct-initialization). If an array is captured, array elements are direct-initialized in increasing index order. The order in which the data members are initialized is the order in which they are declared (which is unspecified).

https://en.cppreference.com/w/cpp/language/lambda

So [context] will direct-initialize the corresponding data member, whereas [context=context] (in this case) does copy-initialization! In terms of code this means:

• [context] is equivalent to decltype(context) captured_context{context};, i.e. const std::shared_ptr<Data> captured_context{context};
• [context=context] is equivalent to auto capture_context = context;, i.e. std::shared_ptr<Data> captured_context = context;

Good, so writing [context=context] actually drops the const qualifier on the captured variable! Thus, for the lambda, it is equivalent to not having written it in the first place and using direct-initialization.

But why does this even matter? Why do we leak references to the shared_ptr<Data> if the captured variable is const? We only ever std::move() or std::forward() the lambda, right up to the place where we invoke it. After that, it goes out of scope, and all captures should be destroyed as well. Right?

Nearly. Let’s think about the compiler generates for us when we write a lambda. For the direct-initialization capture (i.e. [context]() {}), the compiler roughly generates something like this:

struct lambda
{
    const std::shared_ptr<Data> context;
    // ...
};

This is what we want to to std::move() around. But it contains a const data member, and that cannot be moved from (it’s const after all)! So even with std::move(), there’s still a part of the lambda that lingers, keeping a reference to context. In the example above, the lingering part is in func, the capture of the wrapper lambda created in enqueueJob(). We move from func to ensure that all captures are destroyed when the it goes out of scope. But for the const std::shared_ptr<Data> context, which is hidden inside func, this does not work. It keeps holding the reference. The wrapper lambda itself would have to be destroyed for the reference count to drop to zero.
However, we keep the already-finished jobs around, so this never happens. The assertion fails.

How does this matter for Qt?

QtFuture::whenAll() and whenAny() create a shared_ptr to a Context struct and capture that in two lambdas used as continuations on a QFuture. Upon completion, the Context stores a reference to the QFuture. Similar to what we have seen above, continuations attached to QFuture are also wrapped by another lambda before being stored. When invoked, the “inner” lambda is supposed to be destroyed, while the outer (wrapper) one is kept alive.

In contrast to our example, the QFuture situation had created an actual memory leak, though (QTBUG-116731): The “inner” continuation references the Context, which references the QFuture, which again references the continuation lambda, referencing the Context. The “inner” continuation could not be std::move()d and destroyed after invocation, because the std::shared_ptr data member was const. This had created a reference cycle, leaking memory. I’ve also cooked this more complex case down to a small example (godbolt).

The patch for all of this is very small. As in the example, it simply consists of making the capture [context=context]. It’s included in the upcoming Qt 6.6.0.

Bottom line

I seriously didn’t expect there to be these differences in initialization of by-value lambda captures. Why doesn’t [context] alone also do direct- or copy-initialization, i.e. be exactly the same as [context=context]? That would be the sane thing to do, I think. I guess there is some reasoning for this; but I couldn’t find it (yet). It probably also doesn’t make a difference in the vast majority of cases.

In any case, I liked hunting this one down and getting to know another one of those dark corners of the C++ spec. So it’s not all bad .

Today I was doing some experiments with qmllint hoping it would help us make QML code more robust.

I created a very simple test which is basically a single QML file that creates an instance of an object I've created from C++.

But when running qmllint via the all_qmllint  target it tells me

Warning: Main.qml:14:9: No type found for property "model". This may be due to a missing import statement or incomplete qmltypes files. [missing-type]
        model: null
        ^^^^^
Warning: Main.qml:14:16: Cannot assign literal of type null to QAbstractItemModel [incompatible-type]
        model: null
               ^^^^

Which is a relatively confusing error, since it first says that it doesn't know what the model property is, but then says "the model property is an QAbstractItemModel and you can't assign null to it"

Here the full code https://bugreports.qt.io/secure/attachment/146411/untitled1.zip in case you want to fully reproduce but first some samples of what i think it's important

QML FILE

import QtQuick
import QtQuick.Window

import untitled1 // This is the name of my import

Window {
    // things     

    ObjectWithModel {
        model: null
    }
}
 

HEADER FILE (there's nothing interesting in the cpp file)

#pragma once

#include <QtQmlIntegration>
#include <QAbstractItemModel>
#include <QObject>

class ObjectWithModel : public QObject {
    Q_OBJECT
    QML_ELEMENT  
  
    Q_PROPERTY(QAbstractItemModel* model READ model WRITE setModel NOTIFY modelChanged)

public:
    explicit ObjectWithModel(QObject* parent = nullptr);  

    AbstractItemModel* model() const;
    void setModel(QAbstractItemModel* model);

signals:
    void modelChanged();

private:
    QAbstractItemModel* mModel  = nullptr;
};

CMAKE FILE

cmake_minimum_required(VERSION 3.16)
project(untitled1 VERSION 0.1 LANGUAGES CXX)
set(CMAKE_CXX_STANDARD_REQUIRED ON)
find_package(Qt6 6.4 REQUIRED COMPONENTS Quick)
qt_standard_project_setup()

qt_add_executable(appuntitled1 main.cpp)

qt_add_qml_module(appuntitled1
    URI untitled1 VERSION 1.0
    QML_FILES Main.qml
    SOURCES ObjectWithModel.h ObjectWithModel.cpp
)

target_link_libraries(appuntitled1 PRIVATE Qt6::Quick)  

As you can see it's quite simple and, as far as I know, using the recommended way of setting up a QML module when using a standalone app.

But maybe I am holding it wrong?

I might be busy early next month, so I’m posting this a few days early so I get it out of the way! I managed to do a lot of big stuff this month, and pretty happy with my pace. I still have way too many open MRs though, I really need to get that sorted.

Sorry about the shoddiness of some of the screenshots. We are the midst of our Qt6 transition, and sometimes my Breeze is broken and fell back to a built-in Qt theme. I promise it won’t look that ugly in a couple of months!

Plasma

I redid the Accessibility KCM to make it look a bit nicer, by using the newer sidebar view we use in other KCMs. This still needs some time in the oven, though.

The kaccess daemon now reloads it’s config files properly, causing odd behavior like the screen reader never turning off.

Tokodon

The Send button in the composer now changes depending on why you opened it.. This is an easy way to confirm you’re resending, editing and so on.

I implemented a lot of UX improvements for the profile page. It’s way harder to mess up the timeline state by clicking through tabs too quickly. Oh yeah, and that’s fixed for the notification page filters too.

The settings are overhauled and using the new CategorizedSettings component which will make it easier to add more. This has already made space for granular per-account notification controls!

High character count posters rejoice, as the status composer is now usable for you! This will also appear in 23.08, so you don’t have to wait until the next major release.

The alignment of the top row of buttons in posts is ever so slightly fixed now so it looks prettier, and has better clickable areas.

Before	After

I ported the whole application to Qt6 declarative type registration, and other niceties. This doesn’t mean anything for users, but Tokodon should be a bit faster.

If you were ever frustrated with logging into Tokodon, fear not as in the next release the entire process is redone. I rewrote the entire UX to be way more user friendly, less buggy and it supports some cool features like a integrated authorization flow and registration!

Tokodon will now show you a visible warning and explain why your account did not log in, instead of kicking you back to the login page like before:

Finally, a few media attachment improvements such as media attachments being blacked out if the blurhash is missing (which happens, apparently) and an “Alt” tag that shows up in the top right if the image has alt text. Saves me a click or two, and especially useful for video.

NeoChat

I’m attempting to fix the lack of formatting when re-editing messages. It won’t be finished this month, but I’m close!

Two event source dialog changes, including it not showing any data and only showing the option if you have developer tools enabled.

The error message when your device encryption keys are somehow different than what’s currently in your database is now clearer, and tells you to log back in.

PlasmaTube

The sidebar is reorganized so more pages are separated as you might expect. There’s still some work to be done here.

Added support for passing a video URL via the commandline.

Made sure PlasmaTube inhbits sleep like other well-behaving video applications when fullscreen.

Kirigami

Finally merged the Navigation Tab Bar page for Kirigami Gallery! It’s a simple example of this component that we use quite often on mobile.

I changed the fullscreen image viewer used in NeoChat, Tokodon and more to stop listening to swipe events with the mouse and stop wrapping key navigation. For application developers, make sure you set the focus properly when opening it so key navigation works.

I fixed the FormArrow bug in Qt6 where it would point the wrong direction and thanks to Ivan Tkachenko for pointing out that we could use the existing enum in Qt. All consumers of this component have already been ported.

The CategorizedSettings component got some fixes as well, including the ability to filter out invisible actions (useful for hiding certain pages on other platforms, e.g. Android.) and fixing the stuck checked state. There’s still a lot of work to do on this component, but it’s a start!

KCoreAddons

I added a QML singleton for grabbing the applications’ KAboutData instead of it being reimplemented for every single QtQuick application. I have no idea why we didn’t do this before!

import QtQuick
import org.kde.kirigamiaddons.formcard as FormCard
import org.kde.coreaddons

FormCard.AboutPage {
    aboutData: AboutData
}

Qt

We are trying to adopt qmlformat in KDE. I spend a unreasonable amount of time fixing formatting, so it would be nice to have it automatically happen like we already use clang-format for with C++. I have managed to make some really good headway here, and squash lots of little nagging bugs we have hit. These have not been merged into Qt yet, but I hope we can get them reviewed soon. (If you have approver rights, I would appreciate if you took a look!)

I fixed a bug where qmlformat would indent call expressions twice causing weird indentation like this:

onTestChanged: {
fooBar(test, {
        // Testing
        "foo": "bar"
        });
}

qmlformat shouldn’t insert newlines in empty components and objects which we use in a lot of QML code. Normally qmlformat would format them like this, which wastes space and looks kinda ugly:

QtObject {
}

Oh yeah, and object declarations in arrays should look less insane.

If you use spaces to delineate groups of import statements, qmlformat should now try to preserve that instead of destroying it.

And two more small things: fixing the command line arguments not overriding anything and fixing the QML coding conventions documentation.

I had the hankering for tinkering the KDE application style. The default style by KDE, Breeze, is pretty nice as is, but there are small things I'd like to modify.

There's Klassy which is quite customizable and fun, but I don't really need all of the settings it has.

Then there's Kvantum which uses SVG files to create a theme, but they don't follow KDE colorschemes. And I dislike working with SVG files.

Both are brilliant for their usecases, but I wanted just Breeze with few changes.

Fork time!

So, I did what one has to do, forked Breeze and renamed everything Breeze related to Zephyr. I chose Zephyr because it was synonym for Breeze in Thesaurus lol. Also, it makes sure it's last in the list of the application styles, so people don't accidentally confuse it to Breeze.

Here's link to the repository: https://codeberg.org/akselmo/Zephyr

Installation help is also there, but feel free to make issue and/or merge requests for adding stuff like what packages one has to install for their distro.

Unfortunately due to the massive size of the Breeze Gitlab repo, I didn't want to flood Codeberg with the whole history. So, some of the history got lost. I have mentioned it in the readme file though.

After renaming all the things, the whole thing built and installed surprisingly easily.

I then implemented following:

• Black outline setting, so the default outline has a black one around it.
  • Why? Idk looks cool. Not really other reason.
  • Yes, it can be disabled.
• Traffic color icons in window deco
  • I am allergic to Apple but the traffic light concept just makes sense to me.
  • Also can be enabled or disabled
• Customizable style frame and window deco outline colors
  • You can completely change the frame colors.
  • You can also make them invisible! No outlines, no frames! Fun!
• Slightly rounder windows and buttons
  • At some point I will make a setting for these too, but now they're applied when the thing is built
• Fitting Plasma style if you use the defaults Zephyr offers (mostly black outlines)
  • The plasma theme buttons do not match the application style in roundness, yet.
  • I am lazy and avoid working with SVG files as long as I can

Why

For fun! For learning! And I wanted to make something that is super close to Breeze (hell, it is Breeze, just few mods), but still has it's own charm and how I like seeing my desktop.

It also can work as a great test bench for others who want to see if they can modify application style.

Just rename anything Zephyr to YourForkNameHere and have fun. But it's probably better to fork the original Breeze project :)

Also, when making my own things for Breeze, it's nice to just implement them in something similar but different name so I can test the changes for longer period of time. And if I like the changes I can maybe show them to upstream.

In future, I will make it work with Plasma 6 (unless i feel lazy). Probably will have to fork Breeze then again and apply my changes. Hopefully it's not too big of a change.

Also, I will be working on the actual Breeze in future too! I hope to implement separator colors for the Plasma colorscheme, so basically you can change the color of all frames and outlines and whatnot. This kinda helped me to figure how that works as well!

All in all, good project, I keep tinkering with it and it helps me understand the Breeze styling and Qt in general more.

Revontuli and Zephyr

My colorscheme Revontuli works really well together with Zephyr. So, feel free to give them a go!

Thanks for reading as usual!

On Thursday and Friday evenings, I went to the Matrix Community Summit at C-Base in Berlin with Tobias. It was the occasion to meet a few other Matrix developers particularly the Nheko developer, MTRNord and a few other devs whom I only knew by nickname. It was great even though I could only spend a few hours there. Tobias stayed longer and will be able to blog more about the event.

During the weekend, instead of going to the Matrix summit, I participated to the KDE Promo sprint with Paul, Aniqa, Niccolo, Volker, Joseph. Aron also joined us via video call on Saturday. This event was also in Berlin at the KDAB officem which we are very thankful for hosting us.

This sprint was the perfect occasion to move forward with many of our pending tasks. I mainly worked on web-related projects as I tried to work on a few items on my large todo list.

We now have an updated donation page, which includes the new donnorbox widget. Donnorboy is now our preferred way to make recurring donations and recurring donations are vital to the success of KDE. Check it out!

With Paul, we also looked at the next KDE For-pages. Two of them are now done and we will publish them in the coming weeks. There are plans for a few more and if you want to get involved there, this is the phabricator task to follow.

I also updated the KDE For Kids with the help of Aniqa. It now features the book Ada & Zangemann from Matthias Kirschner and Sandra Brandstätter that sensibilise kids to Free Software. Let me know if you have other books suggestions for kids around Free Software and KDE that we can include on our websites.

This was only a short version of all the things we did during this sprint, I will let the others blog about what they did. More blog posts will certainly pop up on planet.kde.org soon.

The sprint would have been only possible thanks to the generous donation from our users, so consider making a donation today! Your donation also helps to pay for the cost of hosting conferences, server infrastructure, and maintain KDE software.

The type system of OPC UA permits the creation of complex and nested data types. With the merge of the generic struct decoding and encoding feature, the Qt OPC UA module has greatly improved the comfort of handling such types. But for large projects with lots of custom data types, its QVariant based interface might still feel a bit too complicated.

Continue reading Qt OPC UA – Data Type Code Generation at basysKom GmbH.