Ever since C++20 introduced coroutine support, I was wondering how this could integrate with Qt. Apparently I wasn’t the only one: before long, QCoro popped up. A really cool library! But it doesn’t use the existing future and promise types in Qt; instead it introduces its own types and mechanisms to support coroutines. I kept wondering why no-one just made QFuture and QPromise compatible – it would certainly be a more lightweight wrapper then.

With a recent project at work being a gargantuan mess of QFuture::then() continuations (ever tried async looping constructs with continuations only? ) I had enough of a reason to finally sit down and implement this myself. The result: https://gitlab.com/pumphaus/qawaitablefuture.

Example

#include <qawaitablefuture/qawaitablefuture.h>

QFuture<QByteArray> fetchUrl(const QUrl &url)
{
    QNetworkAccessManager nam;
    QNetworkRequest request(url);

    QNetworkReply *reply = nam.get(request);

    co_await QtFuture::connect(reply, &QNetworkReply::finished);
    reply->deleteLater();

    if (reply->error()) {
        throw std::runtime_error(reply->errorString().toStdString());
    }
    co_return reply->readAll();
}

It looks a lot like what you’d write with QCoro, but it all fits in a single header and uses native QFuture features to – for example – connect to a signal. It’s really just syntax sugar around QFuture::then(). Well, that, and a bit of effort to propagate cancellation and exceptions. Cancellation propagation works both ways: if you co_await a canceled QFuture, the “outer” QFuture of coroutine will be canceled as well. If you cancelChain() a suspended coroutine-backed QFuture, cancellation will be propagated into the currently awaited QFuture.

What’s especially neat: You can configure where your coroutine will be resumed with co_await continueOn(...). It supports the same arguments as QFuture::then(), so for example:

QFuture<void> SomeClass::someMember()
{
    co_await QAwaitableFuture::continueOn(this);
    co_await someLongRunningProcess();
    // Due to continueOn(this), if "this" is destroyed during someLongRunningProcess(),
    // the coroutine will be destroyed after the suspension point (-> outer QFuture will be canceled)
    // and you won't access a dangling reference here.
    co_return this->frobnicate();
}

QFuture<int> multithreadedProcess()
{
    co_await QAwaitableFuture::continueOn(QtFuture::Launch::Async);

    double result1 = co_await foo();
    // resumes on a free thread in the thread pool
    process(result1);

    double result2 = co_await bar(result1);
    // resumes on a free thread in the thread pool
    double result3 = transmogrify(result2);

    co_return co_await baz(result3);
}

See the docs for QFuture::then() for details.

Also, if you want to check the canceled flag or report progress, you can access the actual QPromise that’s backing the coroutine:

QFuture<int> heavyComputation()
{
    QPromise<int> &promise = co_await QAwaitableFuture::promise();
    promise.setProgressRange(0, 100);

    double result = 0;

    for (int i = 0; i < 100; ++i) {
        promise.setProgressValue(i);
        if (promise.isCanceled()) {
            co_return result;
        }
        frobnicationStep(&result, i);
    }
    co_return result;
} 

Outlook

I’m looking to upstream this. It’s too late for Qt 6.11 (already in feature freeze), but maybe 6.12? There have been some proposals for coroutine support on Qt’s Gerrit already, but none made it past the proof-of-concept stage. Hopefully this one will make it. Let’s see.

Otherwise, just use the single header from the qawaitablefuture repo. It an be included as a git submodule, or you just vendor the header as-is.

Happy hacking!

Caveat: GCC < 13

There was a nasty bug in GCC’s coroutine support: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101367 It affects all GCC versions before 13.0.0 and effectively prevents you from writing co_await foo([&] { ... }); – i.e. you cannot await an expression involving a temporary lambda. You can rewrite this out as auto f = foo([&] { ... }); co_await f; and it will work. But there’s no warning at compile time. As soon as the lambda with captures is a temporary expression inside the co_await, it will crash and burn at runtime. Fixed with GCC13+, but took me a while to figure out why things went haywire on Ubuntu 22.04 (defaults to GCC11).

The second maintenance release of the 25.12 series is with the usual batch of stability fixes and workflow improvements. Highlights of this release include fixes to various monitor issues and refactoring of the monitor dragging mechanism. See the changelog below for more details.

Kdenlive needs your support

Our small team has been working for years to build an intuitive open source video editor that does not track you, does not use your data, and respects your privacy. However, to ensure a proper development requires resources, so please consider a donation if you enjoy using Kdenlive - even small amounts can make a big difference.

For the full changelog continue reading on kdenlive.org.

Transitous is a project that runs a public transport routing service that aspires to work wold-wide. The biggest leaps forward in coverage happened in the beginning, when it was just a matter of finding the right urls to download the schedules from. Most operators provide them in the GTFS format, which is also used in Google Transit and a few other apps.

However, the number of readily available GTFS schedules (so called feeds) that we are not using yet is starting to become quite small. As evident when comparing with Google Transit, there are still a number of feeds that are only privately shared with Google. This is not great from a standpoint of preventing monopolies and also a major problem for free and open-source projects which don’t have the resources to discuss with each operator individually or to even buy access to the data from them. Beyond that case, there is still a suprisingly large number of places in the world that do not publish any schedules in a standardized format, and that is something that we can fix.

Source data comes in many shapes and forms, but the ingredients we’ll definitely need are:

• the lines
• stop times
• stop locations
• and service dates

Sometimes you can find a provider specific API that returns the needed information, or there is open data in a non-standard format. In the worst case, it might be necessary to scrape data out of the HTML of the website.

Some examples:

a stop time from the API of ŽPCG (Railway in Montenegro):

Example 1:

{
    "ArrivalTime" : "15:49:16",
    "DepartureTime" : "15:51:16",
    "stop" : {
        "Latitude" : 42.511829,
        "Longitude" : 19.203468,
        "Name_en" : "Spuž",
        "external_country_id" : 62,
        "external_stop_id" : 31111,
        "local" : 1,
    }
}

It is immediately visible that we get the stop times, in for some reason extreme precision. We also get coordinates for the location, which makes conversion to GTFS much easier. Unfortunately the coordinates from this dataset are not exactly great, and can easily be off by multiple kilometers, but they nevertheless provide a rough estimate that we can improve on by matching them to OpenStreetMap.

The railway-data enthusiasts will also notice that we get a UIC country code and a stop code, which we can concatinate to get a full UIC stop identifier. We can make use of that for OSM matching later on.

Example 2:

<ElementTrasa Ajustari="0" CodStaDest="27888" CodStaOrigine="23428" DenStaDestinatie="Ram. Budieni"
    DenStaOrigine="Târgu Jiu" Km="1207" Lungime="250" OraP="45180" OraS="45300" Rci="R" Rco="R" Restrictie="0"
    Secventa="1" StationareSecunde="0" TipOprire="N" Tonaj="500" VitezaLivret="80"/>

This example is from open data for railways in Romania. Unfortunately this one does not give us coordinates, and the fact that the fields are in abbreviated Romanian doesn’t make it too easy to understand for someone like me who does not speak any vaguely related language. However looking at the numbers, we can figure out that OraP and OraS are seconds and provide the departure and arrival times. However here, the data does not model the times at stops, but the transitions between the stops, so some more reshuffling is necessary.

Example 3:

{
    "ArrivalTimes" : "12:35, 13:37, 14:02, 14:21, 14:52, 14:27, 15:04, 15:50, 16:14, 16:39, 17:08, 17:36, 18:55, 19:03, 19:12, 20:05, 20:33, 21:12, 21:47",
    "Classes" : "2",
    "DepartureTimes" : "12:35, 13:38, 14:03, 14:22, 15:12, 14:42, 15:29, 15:51, 16:15, 16:40, 17:33, 17:37, 18:57, 19:08, 19:13, 20:06, 20:34, 21:13, 21:47",
    "Route" : "Vilnius-Krokuva",
    "RouteStops" : "Vilnius, Kaunas, Kazlų Rūda, Marijampolė, Mockava, Trakiškė/Trakiszki, Suvalkai/Suwałki, Augustavas/Augustów, Dambrava/Dąbrowa Białostocka, Sokulka/Sokółka, Balstogė/Białystok, Balstogė (Žaliakalnio stotis) / Białystok Zielone Wzgórza, Varšuva (Rytinė stotis)/ Warszawa Wschodnia, Varšuva (Centrinė stotis)/ Warszawa Centralna, Varšuva (Vakarinė stotis)/ Warszawa Zachodnia, Opočnas (Pietinė stotis)/Opoczno Południe, Vloščova (Šiaurinė stotis)/Włoszczowa Północ, Mechuvas/Miechów, Krokuva (Pagrindinė stotis)/ Kraków Główny",
    "RunWeekdays" : "1,2,3,4,5,6,7",
    "Spaces" : "WHEELCHAIR, BICYCLE",
    "TrainNumber" : "33/141"
}

This example is from the, to my knowledge, only source of semi-machine-readable information on railway timetables in Lithuania. While json is straightforward to parse, this one for some reason does not use json arrays, but comma-separated lists. We can still work with this of course, until a stop appears whose name contains a comma. Oh well. Once again, no coordinates are provided. While it is tempting to think this should be easier to convert than xml in Romanian, this format has some more hidden fun. If you have been to the area the line from the example operates in, you might already have noticed that the time zone changes between Lithuania and Poland. Unfortunately, there is no notice of that in the data, just a randomly backwards-jumping arrival and departure time.

To work with this, we first need to match the stop names to coordinates, then figure out the time zones from that, and then convert the stop times.

Matching stations to locations

OpenStreetMap is the obvious choice for this. It is fairly easy to query the station locations, but matching the strings to the node in OSM is not trivial. There are often variations in spelling, particularly if the data covers neighbouring countries with different languages. The data may also have latinized names, while the country usually uses a different script and so on.

Since this problem comes up repeatedly, I am slowly improving my rust library (gtfs-generator) for this, so it can hopefully handle most of these cases automatically at some point. It aims to be very customizable, so the matching criteria needs to be supplied by the library user. The following example matches a stop if it has a matching uic_ref tag, which is a strong identifier. If no node has such a matching tag, all nodes in the radius of 20km are considered if their name is either a direct match, an abbreviation of the other spelling, similar enough or has matching words. The matching radius can be overridden in each query, so if nothing is known yet, the first guess can be the middle of the country with a large enough radius. As soon as one station is known, the ones appearing on the same route must be fairly close. The matching quality strongly depends on making a good guess of the distance from the previous stop, as it greatly reduces the risk of similarly named stations being mismatched.

Since OpenStreetMap provides different multi-lingual name tags, the order that these should be considered in needs to be set as well.

The code for matching will look something like this:

let mut matcher = osm::StationMatcherBuilder::new()
    .match_on(osm::MatchRule::FirstMatch(vec![
        osm::MatchRule::CustomTag {
            name: "uic_ref".to_string(),
        },
        osm::MatchRule::Both(
            Box::from(osm::MatchRule::Position {
                max_distance_m_default: 20000,
            }),
            Box::from(osm::MatchRule::FirstMatch(vec![
                osm::MatchRule::Name,
                osm::MatchRule::NameAbbreviation,
                osm::MatchRule::NameSimilar {
                    min_similarity: 0.8,
                },
                osm::MatchRule::NameSubstring,
            ])),
        ),
    ]))
    .name_tag_precedence(
        [
            "name",
            "name:ro",
            "short_name",
            "name:en",
            "alt_name",
            "alt_name:ro",
            "int_name",
        ]
        .into_iter()
        .map(ToString::to_string)
        .collect(),
    )
    .transliteration_lang(TransliterationLanguage::Bg)
    .download_stations(&["RO", "HU", "BG", "MD"])
    .unwrap();
    
    // Parse input

    let station = matcher.find_matching_station(&osm::StationFacts {
        name: Some(name),
        pos: Some(previous_coordinates.unwrap_or((46.13, 24.81))), // If we know nothing yet, bias to the middle of Romania, so we at least don't end up in the wrong country
        max_distance_m: match (previous_coordinates, previous_time, atime) {
            // We have previous coordinates, but nothing for this station. Base limit on max reachable distance at reasonable speed
            (Some(_prev_coords), Some(departure), Some(arrival)) => {
                let travel_seconds = arrival - departure;
                Some(travel_seconds * 70) // m/s
            }
            // No previous location known
            _ => Some(800000),
        },
        values: HashMap::from_iter([("uic_ref".to_string(), uic_ref.clone())]),
    });

For now, until I’m somewhat certain about the API, you’ll need to use the git repository directly to use the OSM matching feature.

Finally writing the GTFS file

After all the ingredients are collected, the actual GTFS conversion should be fairly easy. We now need to sort the data we collected into the main categories of objects represented by GTFS, routes, trips, stops, and stop_times.

Every trip needs to have a corresponding route. The exact distinction between different routes depends on the specific transit system. In the simplest case, if multiple buses operate with the same line number on the same day, they would belong to the same route. Each time the bus operates per day, a new GTFS trip starts.

If the system does not have the concept of routes, every trip simply is its own route. This is for example what Deutsche Bahn does for ICE trains, where each journey has it’s own train number.

The code for building the GTFS data looks somewhat like this:

    let mut gtfs = GtfsGenerator::new();

    // parse input

    gtfs.add_stop(gtfs_structures::Stop {
        id: stop_time.station_code.to_string(),
        name: Some(stop_time.station_name.to_string()),
        latitude: coordinates.as_ref().map(|(lat, _)| *lat),
        longitude: coordinates.as_ref().map(|(_, lon)| *lon),
        ..Default::default()
    })
    .unwrap();

    gtfs.add_stop_time(gtfs_structures::RawStopTime {
        trip_id: trip_id.clone(),
        arrival_time: Some(atime.unwrap_or(dtime)),
        departure_time: Some(dtime),
        stop_id: stop_time.station_code.to_string(),
        stop_sequence: stop_time.sequence,
        ..Default::default()
    })
    .unwrap();
    
    gtfs.write_to("out.gtfs.zip").unwrap();

Validating the result

A new GTFS feed is rarely perfect on the first try. I recommend running it through gtfsclean first. After all obvious issues are fixed (missing fields, broken references), you can use the validator of the French government and the canonical GTFS validator. It is worth using both, as they check for slightly different issues.

Once all critical errors reported by the validators are fixed, you can finally test the result in MOTIS. You can get a precompiled static binary from GitHub Releases. Afterwards create a minimal config file using ./motis config out.gtfs.zip. The API on it’s own is not too useful for testing, so add

server:
  web_folder: /path/to/ui/directory

to the top of the file to make the web interface available.

Make sure to also enable the geocoding option, so you can search for stops.

Now all that’s needed is loading the data and starting the server:

./motis import
./motis server

You should now be able to search for your stops and find routes:

Publishing

Once it is ready, the GTFS feed needs to be uploaded to a location that provides a stable url even if the feed is updated. The webserver should also support the Last-Modified-header, so the feed can be downloaded only when needed. A simple webserver serving a directory like nginx or apache works well here, but something like Nextcloud works equally well if you already have access to an instance of it.

Since the converted dataset needs to be regularly updated, I recommend setting up a CI pipeline for that purpose. The free CI offered by gitlab.com and GitHub is usually good enough for that. I recommend setting up pfaedle in the pipeline, to automatically add the exact path the vehicles take based on routing on OpenStreetMap data.

Once you have a URL, you can add it to Transitous and places where other developers can find it, like the Mobility Database

If you are interested in some examples of datasets generated this way, check out the Mobility Database entries for LTG Link and ŽPCG. You can find a list with some examples of feeds I generate here, including the generator source code based on the gtfs-generator crate.

You can always ask in the Transitous Matrix channel in case you hit any roadblocks with your own GTFS-converter projects.

This is a sequel to my first blog where I’m working on fixing a part of docs.kde.org: PDF generation. The documentation website generation pipeline currently depends on docs-kde-org, and this week I was focused on decoupling it.

Moving away from docs-kde-org

I began by copying the script that builds the PDFs in docs-kde-org to ci-utilities as a temporary step towards independence.

I then dropped the bundled dblatex for the one present in the upstream repos, wired up the new paths and hit build. To no one’s surprise, it failed immediately. The culprit here was the custom made kdestyle latex style. As a temporary measure, I excluded it to get PDFs to generate again, however several (new) locales broke.

So the pipeline now worked, albeit, partially.

(Commit 17bbe090) (Commit 3e8f8dc9)

Trying XeTeX

Most non-ASCII languages (Chinese, Japanese, Korean, Arabic, etc) are not supported for PDF generation because of wonky unicode support in pdfTeX (the existing backend driver that converts LaTeX code to a PDF)

So I switched the backend to XeTeX to improve the font and language handling. Now I’m greeted with a new failure :D

1
xltxtra.sty not found

xltxtra is a package used by dblatex to process LaTeX code with XeTeX. Installing texlive-xltxtra fixed this, and almost everything built (except Turkish)

(Commit 7ede0f94)

The deprecated package rabbit hole

While installing xltxtra, I noticed that it was deprecated and mostly kept for compatibility. So I tried replacing it with modern alternatives (fontspec, realscripts, metalogo).

Back to xltxtra.sty not found

At the point I looked at the upstream dblatex code and found references to xltxtra in a Latex style file and in the XSL templates responsible for generating the latex.

I experimented with local overrides, but the dependency was introduced during the XSLT stage so it isn’t trivial to swap out.

Clearer mental model

A very good outcome from the experiments is that I now understand the pipeline much better

1
DocBook XML -> XSLT Engine -> LaTeX Source -> XeLaTex -> PDF

XSL files are templates for the XSLT engine to convert the docbooks to latex and the problem is probably in one of these templates

Decision with mentor

After discussion, Johnny advised me to

• Use texlive-xltxtra for now
• Open an upstream issue
• And that fixing dblatex itself is outside our scope

So I went ahead and filed the issue https://sourceforge.net/p/dblatex/bugs/134/. Now we keep moving forward while also documenting the technical debt.

Locales experiment

I temporarily re-enabled previously excluded locales (CJK etc) to see what happens. Unfortunately most of them didn’t generate silently. I’m still unsure about this part and the behavior might differ in other projects, so I’ll revisit this later.

(Commit 04b698a5)

kdestyle strikes back

Now I fed the absolute path of kdestyle to dblatex (to maintain PDF output parity) which introduced a new set of errors

• Option clash for graphicx
• conflicting driver options in hyperref (dvips, pdftex)

These seem to originate from the style file and is the next major thing to resolve.

(Commit 7ede0f94)

Improved Git skills

I also had a self inflicted problem this week. My feature branch diverged from master and I merged instead of rebasing. As a result the history got messed up. I fixed it by:

• dropping the merge commit
• rebasing properly
• force pushing the cleaned branch

Lessons learnt. Working with open source projects has really made git less intimidating.

Where things stand

Right now

• we can build with XeTeX
• we rely on texlive-xltxtra
• most locales work (except Turkish of course)
• kdestyle introduces some problems

Also here’s a before vs after of the generated PDFs (the end goal would be to make both of them look identical)

Honestly I don’t feel like I got much accomplished but the understanding gained should make future changes much faster (and safer :D)

Dev logs log 1, log 2, log 3, log 4

Edit (09-02-2026): Added links to relevant commits based on feedback received

Last weekend I attended this years edition of FOSDEM in Brussels again, mostly focusing on KDE and Open Transport topics.

KDE

As usual, KDE had a stand, this time back in the K building and right next to our friends from GNOME. Besides various stickers, t-shirts and hoodies, a few of the rare amigurumi Konqis where also available. And of course demos of our software on laptops, mobile phones, gaming consoles and graphic tables.

Several KDE contributors also appeared in the conference program, such as Aleix’s retrospective on 30 years of KDE and Albert’s talk on Okular.

Itinerary

Meeting many people who have just traveled to FOSDEM is also a great opportunity to gather feedback on Itineray, especially with many disruptions allowing to test various special cases.

• Eurostar’s ticket scanners apparently can’t deal with binary ticket barcodes correctly, however that’s exactly what the standard UIC SSB barcode they use on their Thalys routes is. Their off-spec workaround to base64-encode the content is now preserved by Itinerary.
• With DB’s API being blocked increasingly often from other countries we now regularly end up with data mixed from different sources. That exposed some issues with merging data with a different amount of intermediate stops, e.g. due to ÖBB’s API listing border points there as well.

Open Transport Community

For the fourth time FOSDEM had a dedicated Railways and Open Transport track. It’s great to see this to continue to evolve with now also policymakers and regulators not just attending but being actively involved. And not just at FOSDEM, I’m very happy to see community members being consulted and involved in legislation and standardization processes at the moment that were largely inaccessible to us until not too long ago.

Just in time for FOSDEM we also got a commitment for a venue for the next iteration of the Open Transport Community Conference, in October in Bern, Switzerland, at the SBB headquarter. More on that in a future post.

Transitous

At FOSDEM 2024 Transitous got started. What we have there today goes far beyond what seemed achievable back then, just two years ago. And it’s being put to use, the Transitous website lists a dozen applications built on top of it, and quite a few of the talks in the Railways and Open Transport track at FOSDEM referenced it.

• Jonah showed his work on crowd-sourcing delay information, for filling gaps in officially provided data.
• Adam presented how Bimba got global transport coverage thanks to Transitous.
• David talked about the state of open(ing) Czech public transport data, data that Transitous relies on.
• Lluis showed the work behind Citybikes to collect historic data from bike sharing systems, something potentially very interesting for availability predictions in intermodal routing.

And more

The above doesn’t capture all of FOSDEM of course, as every couple of meters you run into somebody else to talk to, so I also got to discuss new developments on standardization around semantic annotations in emails, OSM indoor mapping or new map vector tile formats, among other things.

Welcome to a new issue of This Week in Plasma!

This week the Plasma team continued polishing up Plasma 6.6 for release in a week and a half. With that being taken care of, a lot of fantastic contributions rolled in on other diverse subjects, adding cool features and improving user interfaces. Check ’em out here:

Notable New Features

Plasma 6.7.0

The Window List widget now supports sorting and shows section headers in its full view, making it easier to navigate windows by virtual desktops, activities, or alphabetically. (Shubham Arora, plasma-desktop MR #3434)

Notable UI Improvements

Plasma 6.6.0

System Settings’ Touchscreen Gestures page now hides itself when there are no touchscreens. This completes the project to hide all inapplicable hardware pages! (Alexander Wilms and Kai Uwe Broulik, KDE Bugzilla #492718 and systemsettings MR #391)

After

Before

The “Enable Bluetooth” switch in the Bluetooth widget no longer randomly displays a blue outline on its handle even when not clearly focused. (Christoph Wolk, KDE Bugzilla #515243)

Plasma 6.7.0

Plasma’s window manager now remembers tiling padding per screen. (Tobias Fella, KDE Bugzilla #488138)

The wallpaper selection dialog now starts in the location you navigated to the last time you used it. (Sangam Pratap Singh, KDE Bugzilla #389554)

Theme previews on System Settings’ cursor settings page now scale better when using massive cursor sizes. (Kai Uwe Broulik, plasma-workspace MR #6244)

After

Before

Update items on Discover’s updates page now have better layout and alignment. (Nate Graham, discover MR #1252)

Completed the project to make the delete buttons on System Settings’ theme chooser pages consistent. (Sam Crawford, plasma-desktop MR #3506, sddm-kcm MR #101, and plymouth-kcm MR #47)

The System Tray icon that Discover uses to represent an in-progress automatic update now looks a lot more like the other update icons, and a lot less like Nextcloud’s icon. (Kai Uwe Broulik, discover MR #1258 and breeze-icons MR #526)

Notable Bug Fixes

Plasma 6.5.6

It’s no longer possible to accidentally close the “Keep display configuration?” confirmation dialog by panic-clicking, unintentionally keeping bad settings instead of reverting them. (Nate Graham, kscreen MR #460)

Fixed a regression in sRGB ICC profile parsing that reduced color accuracy. (Xaver Hugl, KDE Bugzilla #513691)

3rd-party wallpaper plugins that include translations now show that translated text as expected. (Luis Bocanegra, KDE Bugzilla #501400)

Plasma 6.6.0

Fixed multiple significant issues on the lock screen that could be encountered with fingerprint authentication enabled: one that could break fingerprint unlocking, and another that could leave you with an “Unlock” button that did nothing when clicked. (David Edmundson, KDE Bugzilla #506567 and KDE Bugzilla #484363)

Fixed a Plasma crash caused by applying a global theme that includes a malformed layout script. (Marco Martin, KDE Bugzilla #515385)

Panel tooltips no longer inappropriately respect certain window placement policies on Wayland. (Tobias Fella, KDE Bugzilla #514820)

User-created global shortcuts are now always categorized as “Applications”, resolving an issue whereby apps added by choosing an executable using the file picker dialog would be inappropriately categorized as system services and couldn’t be edited or deleted. (Tobias Fella, KDE Bugzilla #513565)

Fixed two issues with recent files and folders in the Kickoff application launcher: now it shows the correct file type icons for items, and no longer sometimes shows a weird duplicate “files” section. (Christoph Wolk, KDE Bugzilla #496179 and KDE Bugzilla #501903)

Spectacle now shows the correct resolution in its tooltip for rectangular region screenshots when using a fractional scale factor on a single screen. (Noah Davis, KDE Bugzilla #488034)

The “Open With” dialog now filters its view properly when opened from a Flatpak app. (David Redondo, KDE Bugzilla #506513)

Keyboard focus no longer gets stuck in the Search widget after its search results appear. (Christoph Wolk, KDE Bugzilla #506505)

Frameworks 6.23

Fixed a complicated issue that could sometimes break automatic KWallet wallet unlocking on login. (Bosco Robinson, KDE Bugzilla #509680)

Fixed a visual regression with certain item lists that made the last one touch the bottom of its view or popup. (Marco Martin, KDE Bugzilla #513459)

Notable in Performance & Technical

Frameworks 6.23

Reduced KRunner’s maximum memory usage while file searching is enabled. (Stefan Brüns, KDE Bugzilla #505838)

How You Can Help

KDE has become important in the world, and your time and contributions have helped us get there. As we grow, we need your support to keep KDE sustainable.

Would you like to help put together this weekly report? Introduce yourself in the Matrix room and join the team!

Beyond that, you can help KDE by directly getting involved in any other projects. Donating time is actually more impactful than donating money. Each contributor makes a huge difference in KDE — you are not a number or a cog in a machine! You don’t have to be a programmer, either; many other opportunities exist.

You can also help out by making a donation! This helps cover operational costs, salaries, travel expenses for contributors, and in general just keep KDE bringing Free Software to the world.

To get a new Plasma feature or a bugfix mentioned here

Push a commit to the relevant merge request on invent.kde.org.

In my first week of Season of KDE, I managed to stop the Lokalize menubar from jumping around by enforcing a strict "Global Skeleton" in the XML files. We gave tabs empty menus just to hold space and keep the layout from shifting. It solved the jumping bug perfectly, but it introduced a new, slightly awkward UX problem.

Because those empty menus were technically just sitting there to hold space, they were still clickable. If you were in the Project Overview tab and clicked "Sync," nothing happened. No dropdown appeared, it was just a clickable menu that did nothing. It didn't crash the app, but it felt unfinished. My mentor, Finley, pointed this out and showed me how Kate handles this by greying out unusable actions. To see if we could replicate that, i built a massive "Complete Menu Action Comparison" matrix to map every action across every tab.

The matrix proved we couldn't just copy Kate. Lokalize menus change drastically depending on context-the Edit menu has 20 actions in the Editor, but only one (Find) in the Project Overview. Showing 19 greyed-out, irrelevant actions would just clutter the ui. Instead, Finley suggested a better approach: what if we grey out the menu headers themselves right on the menubar? Figuring out how to code that became my primary goal for week 3.

While analyzing the menu architecture, we also discussed whether to move all menu definitions into one centralized file, dropping the individual KXMLGUIClient .ui.rc files. I analyzed how other major KDE projects like Kate, Okular, Dolphin, and Ark to see how they handled it.

I found out that every major KDE app keeps a distributed model and merges dynamically. Centralizing Lokalize would force us into manually managing visibility and shortcuts, creating a fragile refactor. I shared this with Finley, and we agreed to stick to the distributed approach.

Also when i was working with .rc files, I found a graveyard of "ghost actions"-XML entries with zero implementation in the C++ backend. I drafted an email to the KDE translators' mailing list asking for direct feedback on which actions they actually use in their workflows so i can decide which actions i need to implement or remove.

KDE Eco is an ongoing initiative taken by the KDE community to support the development and adoption of sustainable Free & Open Source Software worldwide and I am happy to be able to contribute to this mission as part of Season of KDE 2026.

As part of this initiative, KDE aims to measure the energy consumption of its software and eco-certify applications with the Blue Angel ecolabel. To make this possible, KEcoLab a dedicated lab based in Berlin provides remote access to energy-measurement hardware via a GitLab CI/CD pipeline to measure the energy usage of software following the guide documented in KDE Eco Handbook

To measure energy usage of a software we need to prepare three scripts baseline.sh , idle.sh and sus.sh and push them via MR to this repository, SUS or Standard User Scenario script emulates a standard user in a automated manner without human intervention, more on how to prepare SUS can be found here.

Currently, these scripts rely on xdotool to simulate user interactions. However, xdotool does not work on Wayland. Since the KEcoLab computer have recently migrated to Fedora 43, which uses Wayland by default, the existing scripts no longer work. To solve this problem,

I am working with my mentors Joseph, Aakarsh, and Karanjot to port the existing test scripts from xdotool to ydotool and kdotool, which are compatible with Wayland. As part of this effort, I’ve written a guide explaining how to use ydotool and kdotool more easily.

Setup ydotool and kdotool #

To test the scripts you have written it’s recommended to setup locally.

Note: Some features changes and some are removed with updates so it’s recommended to build them from source mentioned here

ydotool #

1. Clone the repository and navigate into it

 git clone -b feat/setup-script https://invent.kde.org/neogg/ydotool.git
 cd ydotool 

2. Run the setup script

chmod +x setup.sh
./setup.sh install

3. Give ydotoold permission to uinput so it can run without root

sudo touch /etc/udev/rules.d/99-uinput.rules
echo 'KERNEL=="uinput", MODE="0660", GROUP="input"' | sudo tee /etc/udev/rules.d/99-uinput.rules > /dev/null
sudo usermod -aG input "$USER"

sudo udevadm control --reload-rules
sudo udevadm trigger

Note: A reboot (or full logout/login) is required for the input group change to take effect.

4. Reboot your computer and run

systemctl --user start ydotoold

Verify that you can use ydotool by ydotool type "Hello" you will see Hello typed if installation was successful.

kdotool #

1. Clone the repository and navigate into it

 git clone -b feat/setup-script https://invent.kde.org/neogg/kdotool.git
 cd kdotool 

2. Run the setup script

chmod +x setup.sh
./setup.sh install

Verify that you can use kdotool by kdotool -h

How to use ydotool and kdotool to perform various actions #

Press keys #

For pressing keys we use ydotool , the syntax for pressing keys is a little confusing when you look at it :

ydotool key KEY_CODE:ACTION

KEY_CODE

Linux uses a file named input-event-codes.h that defines numeric codes for everything an input device can generate in Linux. Simply speaking for every type of keypress there is a numerical representation. For example, the LEFTCTRL key is represented by 29, and the letter T is represented by 20. How to find the code for the key you want to press , You can read the original .h file by nano /usr/include/linux/input-event-codes.h if you are in a linux system ( i hope you are using one). or inspect the file here

I use grep to find the code for the key i want since the header file is too long grep KEY_LEFTCTRL /usr/include/linux/input-event-codes.h and so on.

ACTION

• 1 : key is pressed
• 0 : key is released

So to press and release Ctrl you can use the following command

ydotool key 29:1 29:0

Press key combinations #

Key combinations is the same as above , you just need to keep all the keys pressed. So to press Ctrl+Alt+T which opens up a terminal you can use the following command

ydotool key 29:1 56:1 20:1 29:0 56:0 20:0

This translates to Ctrl(pressed) Alt(pressed) T(pressed) Ctrl(released) Alt(released) T(released)

This will hopefully open up a terminal.

Click ( left, right, double) #

For performing mouse clicks we can use the click command in ydotool , If you are do not like codes unlike me,I am sorry but again we have codes…

ydotool click OPTIONS BUTTON_CODE

OPTIONS

--repeat=N : can be used to press a button N times
--P : can be used to press and release a button as that's what we mostly do

BUTTON_CODE

This is a hexadecimal value that represents different mouse buttons (left, right, middle, etc.).
Yes, it looks ugly at first, but you can always open up my blog.

ydotool click 0xC0 # left click
ydotool click 0xC1 # right click
ydotool click 0xC2 # middle button click (scroll button)

ydotool click 0x40 # left button down (press)
ydotool click 0x80 # left button up (release)

ydotool click 0x41 # right button down (press)
ydotool click 0x81 # right button up (release)

And some combinations This can be useful to select text, drag and drop etc..

ydotool click 0x40 # left down 
ydotool mousemove 300 0 # drag right 
ydotool mousemove 0 200 # drag down 
ydotool click 0x80 # left up

same as above but more reliable with delays to simulate a real user

ydotool click 0x40
ydotool mousemove -D 20 300 0
ydotool mousemove -D 20 0 200
ydotool click 0x80

Type text #

Typing text is pretty simple using ydotool

ydotool type "string-you-want-to-type"

Focus a specific app #

kdotool search --name "Window Title" windowactivate
kdotool search --class appname windowactivate
kdotool search --classname org.kde.app windowactivate ( for kde apps )

Send mouse and keyboard commands to a specific window #

Focus the window using kdotool than execute commands using ydotool since commands are always executed in the active window.

Get current mouse location #

We can use kdotool for getting the current mouse location which also returns the windowID along with x and y coordinate of current mouse location

kdotool getmouselocation

The output :

x:42 y:96 screen:0 window:{e8aeec46-5c45-42cc-9839-8ad2edcb7f4f}

If you only want to extract the x and y value since thats what’s more important you can do that using regex

# Read mouse location
loc="$(kdotool getmouselocation)"

# Extract x and y
x=$(echo "$loc" | sed -n 's/.*x:\([0-9-]*\).*/\1/p')
y=$(echo "$loc" | sed -n 's/.*y:\([0-9-]*\).*/\1/p')

Move mouse to a specific coordinate #

Note : We can use ydotool to move the mouse but that may not give us any visual feedback if you are using it in a VM, that is you cannot see the actual mouse cursor move when the commands are executed but you can use the kdotool getmouselocation to see how the mouse location gets updated.

There is a command in ydotool

ydotool mousemove --absolute -x x-value -y y-value

but that does not work in Wayland properly.

We have an another command that moves the mouse relative to current mouse location so we can use that.

ydotool mousemove -x x-value -y y-value

1. mousemove -9999 -9999 ( moves the mouse to top left 0,0)
2. mousemove x y ( now we move relative to 0,0 so that’s like absolute mousemove)

Move mouse inside a specific window #

Since we cannot directly send mouse events to a window on Wayland, we rely on window geometry that we can get using kdotool and do relative mouse movement using ydotool.

The steps are:

1. Focus the window
2. Get the window geometry
3. Reset the mouse to (0,0)
4. Move the mouse relatively using the window coordinates

# First, focus the window:
kdotool search --class firefox windowactivate

# Get the window geometry:
kdotool search --class firefox getwindowgeometry
# Example output:
# Window {a24d3d34-85f6-4915-821b-54a71a959f6a}
# Position: 340.23748404968967,68.49159882293117
# Geometry: 768x864


#Reset the mouse position to the top-left corner of the screen:
ydotool mousemove -x -9999 -y -9999

# Move the mouse to the Position you got from above command (top left corner of the window )

# Now move the mouse inside the window (top-left corner of the window):
ydotool mousemove -x 120 -y 80

#At this point, the mouse cursor is inside the target window. so you can now move inside the window , but only once than again repeat the above steps once again ( i will think of a script to automate this)

Move mouse to the center of a window #

Moving to the center of a window uses the same idea, but instead of moving to x and y, we add half the width and height.

From the geometry we extract : x=120 y=80 width=1280 height=800

Calculate the center:

center_x = x + width / 2

center_y = y + height / 2

Reset the mouse again:

ydotool mousemove -x -9999 -y -9999

Move the mouse to the center of the window:

ydotool mousemove -x center_x -y center_y

These completes all the basic actions that are required to automate user actions , We may need to combine many of above to do tasks that simulate real user.

In the last month, I have mostly been working on refactoring the Kdenlive keyframes system to make it more powerful. This is part of a NGI Zero Commons grant via NLnet.

Improving existing parameters handling

A first step in preparation for this work was to improve the usability of some effects. In Kdenlive, we support several effect libraries, like MLT, Frei0r and FFmpeg's avfilters. Not all effects expose their parameters in the same way. For example in MLT, we have a rectangle parameter allowing to define a zone in a video frame. Other effects expose several independent parameters for a rectangle, namely x, y, width and height.

Currently, these effects are displayed with a list of sliders for each value:

With my latest changes, these will be handled like a rectangle, which means it can directly be manipulated in the monitor overlay.

Another improvement is the addition of Point parameters which will allow selecting a point in the monitor.

These changes are planned to be in the 26.04 release.

Boosting keyframes

The next improvement being worked on is support of keyframes per parameter. Currently, once you add a keyframe to an effect, it is applied to all parameters, which is sometimes not wanted. This will also allow animating only one parameter while leaving the others parameters fixed. Below is a screenshot of a basic UI created to test the feature.

Kdenlive needs your support

Our small team has been working for years to build an intuitive open source video editor that does not track you, does not use your data, and respects your privacy. However, to ensure a proper development requires resources, so please consider a donation if you enjoy using Kdenlive - even small amounts can make a big difference.