This seems to be fixing a symptom rather than the problem. Have you investigated why the float difference happen at all? I don't hit this assert.

I don't know if i want to dig into the details of winit & x11, but this value is reported by winit, and it's definitely changing by this small amount.

But besides that, if for whatever reason the DPI changes by a very small amount, what does egui do to handle it?

the mentioned code snipset in my comment shows that it ignores small differences. the warning that I i fixed in this pr does not. So these two code segments behave different on the same logic, or not?

But where does the change come from? Are you storing a Galley between frames (you shouldn't)?

If you create the galley in one frame, and then tesslleate, then I can't see how the pixels_per_point could change.

Have you tried latest master?

But where does the change come from? Are you storing a Galley between frames (you shouldn't)?

Well the Galley comes from from the font:

I honestly dunno how exactly to debug if these Galley are stored between frames, but a GalleyCache sounds like the implementation does that?

I have to say I don't have the insight into egui, the docs for the Galley also only say:
"Needs to be recreated if the underlying font atlas texture changes, which happens under the following conditions: [...]"

(https://docs.rs/egui/latest/egui/struct.Galley.html)

I am not manually recreating my fonts if pixels_per_point change, but the doc sounds like it's doing that internally anyway in crate::text::Fonts::begin_frame

Then bcs of:

The pixels_per_point change is ignored (because of the floating point difference)
=> galley.pixels_per_point

but the line in this pr compares it again self.pixels_per_point which IS changed
(galley.pixels_per_point: 1.16666663 vs. self.pixels_per_point: 1.16666675 (so very very small => galley is not updated in

but the warning is still displayed
)

I'm indeed not using latest master, if you say Galley is never cached between frames, then it's indeed a bug in my application. I am using custom fonts, but i am not sure which font is triggering this issue.

Ah sorry, I finally follow… pixels_per_point_changed uses a 1e-3 epsilon (for whatever reason), so of course we need a similar (or great) epsilon when we are warning. Let me think through this a bit and I'll either merge your PR or come up with something better before next release.

Does

-        let pixels_per_point_changed =
-            (fonts_and_cache.fonts.pixels_per_point - pixels_per_point).abs() > 1e-3;
+        let pixels_per_point_changed = fonts_and_cache.fonts.pixels_per_point != pixels_per_point;

solve your problem?

I really don't see the point of the 1e-3 epsilon there

Yes that works just as good 👍