That's a very nice change. I especially like that reasons are given when ellipses should not be used. As for writing style, I always try to minimize the usage of parentheses, since too many of them are distracting when reading. Maybe state the reasons with an em dash instead (you could additionally write the reason italic):

Already tracked with T4465.

In D10040#300588, @davidedmundson wrote:

I just removed handling of dynamically updating eisa/serialNumber it doesn't seem to be something that makes sense for it to change at runtime.

Also I don't want to copy the current setEdid pattern, which is broken ATM. Whenever any new client connects it broadcasts a change to every connected client...
I need to follow that up in another patch, possibly by making it static like these two.

I also thought the call to setParentItem would clear the output. But as @broulik says it does not delete automatically. So is this good to go? Or should we make QMLScreen really a Qt Object parent of the outputs?

As already said on IRC and also how Vlad sees it: this optimization like for most of our classes is not worth it.

Note, that m_engine is set in widget.cpp. But it's just a Cpp call to setEngine, which is defined extra. So removing the property macro should be fine.

The old code's output was "2.0" instead of "2.2": integer division first, then cast.

Yes, I know. But my train of thoughts was that the original value was multiplied by 100 when the raw edid data was produced, so it should be divisible by 100 again. But the original value was in general already a float, so this is indeed a bug.

I'm not sure if you could really say that the current code is with a bug, since data[GCM_EDID_OFFSET_GAMMA] should be dividable by 100. I.e. the result is again an integer and nothing gets truncated. But we can still go with the updated code, which looks nicer. Maybe one could add a comment why divide by 100 and add 1.

I do wonder if the division by 100 is right though - and where does the +1 come from?

You use the result var of type float in an addition 10 times. Now the calculation is done with less precision. The git message should state why it's no problem.

What @broulik says. It should be fine with the m_changeCompressor being startet in init, but the git message should note this. Also the git message could be a bit more descriptive on what's happening.

Why push id down? It should stay at the top imo (or alphabetically). Rest is fine.

In T7254#152624, @gladhorn wrote:

Another point to consider is font DPI, which is currently in a different place in system settings, but very related to the scaling.

David told me to commandeer back.

In D11781#294891, @davidedmundson wrote:

@Roman
Check you're ok with the mod, and then I'll ack the rest.

Selections are defined in ICCM: https://tronche.com/gui/x/icccm/sec-2.html#s-2
General introduction to X11 Selections: https://www.uninformativ.de/blog/postings/2017-04-02/0/POSTING-en.html

• Cleanup and remove second updateX11Time call also

Awaits following pr to be complete: https://github.com/swaywm/wlroots/pull/841

• Add comments, constify

@graesslin Thanks for review, but David wanted me to use a cached variable in KWayland. That's why I had to rework the KWin implementation as well. Sorry. Would appreciate another review for this revision.

• Use hint getter

• Add hint getter
• Add since keyword

• Add autotest

• Cleanup: remove hint getter

From a short test: the cursor is always the default cursor on TabBox but not above it. Changing the cursor to some other icon to indicate a click closes the TabBox could be done, but I believe wouldn't look good. Would like to close this task therefore.

Because of risk of regressions abandoned. Pointer constraint deactivation on TabBox invocation was alternatively changed to be done explicitly through f0ba436c724e.