Absolute limits and MinimumRange not always honored

[mghie]

Steps to reproduce

Create an empty model with a range-constrained axis but no data, like so:

var pm = new PlotModel();
pm.Axes.Add(new LinearAxis
{
    Position = AxisPosition.Bottom,
    AbsoluteMinimum = -200.0,
    AbsoluteMaximum = -100.0,
    MinimumRange = 10.0,
});
plot1.Model = pm;

I used the WindowsFormsDemo but many of the examples should be usable as well.

Platform: OxyPlot.WinForms on Windows 10 64 bit
.NET version: 4.8

Expected behaviour

I would expect the axis to honour the absolute limits and the minimum range, so any of these visible ranges would be fine:

• [-200.0, -100.0]
• [-200.0, -190.0]
• [-155.0, -145.0]
• [-110.0, -100.0]

Actual behaviour

Instead I get the range [-100.0, -99.0] which violates both the AbsoluteMaximum and the MinimumRange properties:

Cause and proposed fix

The problem is the following code in Axis.cs:

protected virtual void CoerceActualMaxMin()
{
    // Coerce actual minimum
    if (double.IsNaN(this.ActualMinimum) || double.IsInfinity(this.ActualMinimum))
    {
        this.ActualMinimum = 0;
    }

    // Coerce actual maximum
    if (double.IsNaN(this.ActualMaximum) || double.IsInfinity(this.ActualMaximum))
    {
        this.ActualMaximum = 100;
    }

which uses values outside the allowed range. A possible fix would be to change this to:

        this.ActualMinimum = Math.Max(0, this.AbsoluteMinimum);

and

        this.ActualMaximum = Math.Min(100, this.AbsoluteMaximum);

respectively. With these changes I get the initial display range [-110.0, -100.0] which is valid.

[VisualMelon]

Thanks for reporting this. Your suggested solution sounds reasonable to me. Feel free to open a PR making the change if you have time, otherwise I'll sort one out at some point. (may be worth checking the absolutes aren't NaN; not sure if that's supported)

We should consider some tests for this, since it's not something that will occur regularly if there are data on the plot.

[mghie]

Thanks for the reply. I have cloned the repository and will add a few tests as well. It might take me a few days though.

What would in your opinion be the best range for the axis after Reset(), given the absolute limits and the 4 possible ways to scale the axis in my example?

[VisualMelon]

@mghieI think in this unusual scenario - where there is no data-range and the absolute min/max is non-NaN and finite - then I think assuming absolute range is probably the most sensible thing to do, but I should probably mess around with a few options myself and get back to you.

[mghie]

To me it's not at all unusual, but of course it depends on your use case.

I use it for visualization of the last 5 minutes of measured values in testing equipment. The time axis is therefore constrained to [-300.0, 0.0] seconds. There will never be any other time value in those plots, and the user shouldn't be able to zoom or pan outside of this time range. If the equipment is not connected there is no data to be shown.

My preference would be to have the full absolute range visible in this case.

[VisualMelon]

@mghie for your use case, you should probably be setting the Minimum and Maximum in conjunction with the absolute limits. These will be preserved when the user zooms/pans, and are the values used when the axes are reset (when non-NaN).

Your original suggestion certainly improves matters, and would be a welcome change. It's still possible for the series to be 'upside down' in edge cases. That's a more general concern, and isn't necessarily in the scope of a PR to address this. The reality is that when the axis has nothing to go on, it's never going to give everyone what they expect.

[mghie]

If I understand correctly this should also close #21