There's an issue with

Traceback (most recent call last):  File "/home/travis/build/matplotlib/matplotlib/lib/matplotlib/backends/backend_qt5cairo.py", line 25, in paintEvent    surface = cairo.ImageSurface(cairo.FORMAT_ARGB32, width, height)TypeError: integer argument expected, got floatFatal Python error: AbortedCurrent thread 0x00007f065e8d2700 (most recent call first):  File "/home/travis/build/matplotlib/matplotlib/lib/matplotlib/backends/backend_qt5.py", line 1044 in mainloop  File "/home/travis/build/matplotlib/matplotlib/lib/matplotlib/backend_bases.py", line 3395 in show  File "/home/travis/build/matplotlib/matplotlib/lib/matplotlib/cbook/deprecation.py", line 398 in wrapper  File "/home/travis/build/matplotlib/matplotlib/lib/matplotlib/pyplot.py", line 266 in show  File "<string>", line 58 in <module>

While I will try to fix this in isolation, we must check if other places also need an integer conversion. Also, should we cast to int or round to int?

Probably do the same as what the qt version does?

It turns out Qt now has an enum to control how it rounds to pixels (https://doc.qt.io/qt-5/qt.html#HighDpiScaleFactorRoundingPolicy-enum).

You can test this behavior by setting theQT_SCALE_FACTOR=1.5 ENV in linux and running

importmatplotlib.pyplotaspltimportnumpyasnpth=np.linspace(0,2*np.pi,512)fig,ax=plt.subplots()ax.plot(th,np.sin(th))ax.plot(th,np.cos(th))ax.set_title(fig.canvas._dpi_ratio)plt.show()

Without this patch it rounds down via truncating for non-integer ratios. I think what is going on is that we are getting the integer version of the dpi scaling and render the figure at (N, M) size. However, the widgets are seeing the fractional part and actually taking up (N1.2, M1.2) pixels on screen. When we paint the rendered figure Qt helpfully re-samples to the actual number of pixels and our anti-aliasing interacts really badly with that.

I don‘t have the time to look into this in the next couple of days. Anybody can take over and push to this PR if necessary.

arghh.sorry@QuLogic

Owee, I'm MrMeeseeks, Look at me.

There seem to be a conflict, please backport manually. Here are approximate instructions:

1. Checkout backport branch and update it.

$ git checkout v3.2.x$ git pull

2. Cherry pick the first parent branch of the this PR on top of the older branch:

$ git cherry-pick -m1 77f23da286def97e4f05f4c59f2d5c4f0bfe7af2

3. You will likely have some merge/cherry-pick conflict here, fix them and commit:

$ git commit -am 'Backport PR #15656: Support fractional HiDpi scaling with Qt backends'

4. Push to a named branch :

git push YOURFORK v3.2.x:auto-backport-of-pr-15656-on-v3.2.x

5. Create a PR against branch v3.2.x, I would have named this PR:

"Backport PR#15656 on branch v3.2.x"

And apply the correct labels and milestones.

Congratulation you did some good work ! Hopefully your backport PR will be tested by the continuous integration and merged soon!

If these instruction are inaccurate, feel free tosuggest an improvement.

@timhoffm Actually, is the setDevicePixelRatioF shim ever used? AFAICT there's onlysetDevicePixelRatio, which directly takes a qreal (https://doc.qt.io/qt-5/qpixmap.html#setDevicePixelRatio), but no such thing assetDevicePixelRatioF. (On the getter side there's indeed bothdevicePixelRatio anddevicePixelRatioFhttps://doc.qt.io/qt-5/qpaintdevice.html#devicePixelRatio.)

@anntzer You are probably right. I probably did the getter first and assumed that the setter follows the same pattern.

td;dr: Additional findings on the getterdevicePixelRatio()

Digging a bit deeper: It seems thatQPaintDevice initially only hadint QPaintDevice::devicePixelRatio() but grewqreal QPaintDevice::devicePixelRatioF()in Qt 5.6. Note thatQPaintDevice itself does not support setting the pixel ratio.

However, the derived classesQImage andQPixmap seemed to always have had floating point support viaqreal devicePixelRatio() andvoid setDevicePixelRatio(qreal scaleFactor). (at least since Qt 5.5 - The Qt archive does not seem to have docs for older versions). I find that a little surprising in Qt - I would not have expected that they overload the API in the subclasses.

I was about to propose that we could replace_devicePixelRatioF(obj) byobj.devicePixelRatio(). That would work forQPixmap andQImage. But we also use_devicePixelRatioF inFigureCanvasQt, which inherits-> QWidget -> QPaintDevice but does not overloaddevicePixelRatio for qreal.

So indeed we could usedevicePixelRatio() directly on allQPixmap,QImage and move the contents of_devicePixelRatioF to it's only needed usage inFigureCanvasQt._dpi_ratio.