Add Galley::intrinsic_size and use it in AtomLayout

[lucasmerlin]

• part of Experiment with using cross_intrinsic_size bookkeeping in Placer #5762
• also allows me to simplify sizing logic in egui_flex

[github-actions]

Preview available at https://egui-pr-preview.github.io/pr/7146-lucasatoms-preferred-size
Note that it might take a couple seconds for the update to show up after the preview_build workflow has completed.

[lucasmerlin]

I've changed the approach and now calculate the desired_size as the galley is laid out. This should improve performance (since the calculation is now cached) and makes it more accurate by having access to the Paragraphs.
Another benefit is that this will also work for truncated text.

[lucasmerlin]

I ran the demo and button benchmarks, the change was "within noise threshold" as criterion likes to say.

[emilk]

Why only if idx == 0 ?

[emilk]

Did you mean to (also) use first_row_min_height here?

[lucasmerlin]

I did that first but got wrong results. This matches what happens in rows_from_paragraphs.

In this if

        if paragraph.glyphs.is_empty() {
            rows.push(PlacedRow {
                pos: Pos2::ZERO,
                row: Arc::new(Row {
                    section_index_at_start: paragraph.section_index_at_start,
                    glyphs: vec![],
                    visuals: Default::default(),
                    size: vec2(0.0, paragraph.empty_paragraph_height),
                    ends_with_newline: !is_last_paragraph,
                }),
            });
        } else {

the empty row is always placed at (0, 0), so it only has an effect on the total rect if it's the first and only row.

This seems like a bug, I'll try to fix it in a separate PR. first_row_min_height is also not used in the empty case (which is probably also a bug).

If I remove the if, the test_split_paragraph test fails since it causes a empty Paragraph to be added at the end of the first split Galley.

[lucasmerlin]

I renamed the desired and preferred size names I had in AtomLayout and in this PR to intrinsic size since I think this is the best name (we use desired size a lot in other places and preferred size is very similar to desired size) and for consistency with response.intrinsic_size.

[lucasmerlin]

Also made a small test ui to show that this works:

Screen.Recording.2025-07-08.at.17.49.51.mov

    egui::Window::new("Test").show(ctx, |ui| {
            ui.set_min_size(ui.available_size());
            for wrapping in [
                // TextWrapMode::Extend,
                TextWrapMode::Wrap,
                TextWrapMode::Truncate,
            ] {
                ui.style_mut().wrap_mode = Some(wrapping);
                let response = ui.button(format!(
                    "Hello world this is a long text that should {wrapping:?}"
                ));

                ui.ctx().debug_painter().debug_rect(
                    Rect::from_min_size(
                        response.rect.min,
                        response.intrinsic_size.unwrap_or_default(),
                    ),
                    Color32::RED,
                    "",
                );
            }
        });

[emilk]

We should add a test for this in

egui/crates/epaint/src/text/fonts.rs

Line 1079 in f46926a

fn jobs() -> Vec<LayoutJob> {

[emilk]

I saw your comment in #7146 (comment) - let's do that in a follow-up PR then :)

[emilk]

Is this if-statement even needed? Seems like the logic below should be robust to handling this regardless? In fact, more robust, since the code below takes first_row_min_height into account

[lucasmerlin]

Yes, otherwise the test_split_paragraphs test fails. (I explained why in this comment in case you didn't see it #7146 (comment))

I'll try to fix this in a separate PR.

[emilk]

Nice, thanks!