Problems with High DPI Scaling on Windows

[cercata]

On Windows 10, in the display settings, if we set "scale and layout" to 125%, the performance of plotwidget drops dramatically to the point of not being usable anymore.

Also, the curves are ugly, some are 1 pixel with, others are 2 pixel wide.

Would it be posible to only apply the scaling for icons in the toolbar, texts in the axis, etc, but not for the plot lines ?

In guiqwt we have the same problem with curvewidgets.

[PierreRaybaut]

Thanks for reporting this issue.

I'm currently unable to reproduce the issue.

Would it be possible for you to tell us more about your configuration?
Also, is it happening when you change the scale setting before executing your program or during its execution?

FYI, we are currently doing nothing particular regarding the high DPI screens but we would of course do whatever it takes to avoid any issue with them.

[cercata]

Would it be possible for you to tell us more about your configuration?

Simplifying, it's something like this code, the key I think it is, the number of points in the curve.
With small curves it's ok, but with curves that have hundreds of thousands of points, the difference in performance is very noticeable.

from plotpy.builder import make

curve = make.curve(...)
xaxis = array created by numpy
yaxis = array.array('d', ...)
curve = set_data (xaxis, yaxis)

Also, is it happening when you change the scale setting before executing your program or during its execution?

Both. Also, when it's happening, if we change the "scale and layout" to 100% again on the fly, it goes smooth again.
Well, smooth, it's a lot of points, and it takes some time, but it's 5 times faster.

FYI, we are currently doing nothing particular regarding the high DPI screens but we would of course do whatever it takes to avoid any issue with them.

Yes, it seems related to something done by QT itself, that when this option is active a lot of floating points operations are added for each pixel displayed on screen.

But I think it would be useful if you can reproduce it and understand what exactly is happening, and then report to them. We have been trying to debug plotpy and understand it, bu we are a little overwhelmed.

[cercata]

We have been trying a workaround in qwt/plot_curve.py, if the x axis is strictly crescent, and the increase of x values is not very heterogeneous.
The resulting graph is OK when we have hundreds of thousands of points, and there are some performance improvements, but the selectable points and other things are not reduced, so the performance gains are not notorius, still much worst that when 100% is used.

sparseFactor = 100
def series_to_polyline(xMap, yMap, series, from_, to):
    """
    Convert series data to QPolygon(F) polyline
    """
    xdata = xMap.transform(series.xData()[from_ : to + 1: sparseFactor ])
    ydata = yMap.transform(series.yData()[from_ : to + 1: sparseFactor ])
    return array2d_to_qpolygonf(xdata, ydata)

[PierreRaybaut]

Note that, as a workaround (until this is fixed), you may use the new downsampling feature introduced in DataLab v0.15.

[PierreRaybaut]

DataLab's auto downsampling feature is based on PlotPy's downsampling parameters:

[cercata]

Ohhhh nice, can we use that downsampling dinamically on code ?
Because if we zoom to see details, we would need to disable it

[cercata]

We have also another thing that is making us lose perferomance, we have N curves that all share the same X axis, but are putting that x axis on every curve we create, and and the we add the N created curves to the plotwidget with add_item()

Are we doing that wrong? Is there a way for not duplicating the X-Axis ?

[PierreRaybaut]

The issue will have to be fixed in PythonQwt (that's where it comes from).

[PierreRaybaut]

Ohhhh nice, can we use that downsampling dinamically on code ? Because if we zoom to see details, we would need to disable it

Yes, that's the idea.
See the "Downsample" entry in the context menu when right-clicking on a curve: this enable/disable the downsampling feature.