Plot bounds broken when all y values are the same

[nknotts]

Describe the bug
When all y values are identically the same, the calculated y bounds will be considered invalid and the default y-range of [0, 1] will be used instead. The same bug applies to x as well.

To Reproduce
I've created an example application here:

https://github.com/nknotts/egui/blob/added-plot-bounds-example/examples/plot_bounds/src/main.rs

The gist is

            let x_range = 0..1000;
            let single2: PlotPoints = x_range
                .clone()
                .map(|i| {
                    let x = i as f64 * 0.01;
                    [x, 2.0]
                })
                .collect();

            ui.label("All values == 2, completely missing");
            Plot::new("plot3")
                .allow_drag(false)
                .allow_zoom(false)
                .allow_scroll(false)
                .height(row_height)
                .show(ui, |plot_ui| plot_ui.line(Line::new(single2)));

Where determining the bounds, there is

egui/crates/egui/src/widgets/plot/mod.rs

Line 831 in a1f6f68

let auto_x = !bounds_modified.x && (!min_auto_bounds.is_valid_x() || auto_bounds.x);

        let auto_x = !bounds_modified.x && (!min_auto_bounds.is_valid_x() || auto_bounds.x);
        let auto_y = !bounds_modified.y && (!min_auto_bounds.is_valid_y() || auto_bounds.y);

Ultimately there is a call to

egui/crates/egui/src/widgets/plot/transform.rs

Line 52 in a1f6f68

self.is_finite() && self.width() > 0.0 && self.height() > 0.0

    pub fn is_valid(&self) -> bool {
        self.is_finite() && self.width() > 0.0 && self.height() > 0.0
    }

where, in this case, the height will be == 0.0 since all values are the same. Since the y bounds are not valid, the default bounds of [0, 1] are used.

Expected behavior
An appropriate bound to be calculated for this case.

Screenshots

Desktop (please complete the following information):

• OS: macOS
• Version 12.6

Additional context

[JumpyLionnn]

I have this issue as well, any updates?

[emilk]

The bounds will be zero-height, which probably leads to a divide by zero somewhere. Basically it is hard to pick a zoom on an axis if it is infinitely narrow. No zoom will make it fill the available width.

The same problem should happen if all X values are the same, or if there is just a single point in a plot (all X and all Y values are all the same).

It should be fairly easy to fix if someone wants to take a stab at it.

Related issue:

• "Double-click to reset view" for single-point plots doesn't work #3808

[bonsairobo]

@emilk Maybe instead of this block in PlotTransform::new:

        // Make sure they are not empty.
        if !bounds.is_valid_x() {
            bounds.set_x(&PlotBounds::new_symmetrical(1.0));
        }
        if !bounds.is_valid_y() {
            bounds.set_y(&PlotBounds::new_symmetrical(1.0));
        }

it could be:

        // Make sure they are not empty.
        if bounds.is_finite_x() {
            if bounds.width() == 0.0 {
                bounds.expand_x(0.001);
            }
        } else {
            bounds.set_x(&PlotBounds::new_symmetrical(1.0));
        }
        if bounds.is_finite_y() {
            if bounds.height() == 0.0 {
                bounds.expand_y(0.001);
            }
        } else {
            bounds.set_y(&PlotBounds::new_symmetrical(1.0));
        }