This PR adds two small additions to `LayoutJob`:
- `LayoutJob::format_at_byte` to query the `TextFormat` of the section
covering a given byte index.
- An optimization to `LayoutJob::append` that merges newly appended text
into the previous section when the format matches and there is no
leading space.
It also documents the `LayoutJob::sections` invariant (sections are
ordered and together cover the whole text with no gaps or overlaps) and
adds `LayoutJob::debug_sanity_check`, which verifies this in debug
builds. It is called from `format_at_byte` and from the text layouter.
## Why the `easymarkeditor` snapshot changed
The `append` optimization changes how many sections a `LayoutJob` ends
up with: consecutive runs of identically-formatted text now collapse
into a single section instead of one section per `append` call. The
easymark editor produces many such adjacent same-format sections, so it
is affected.
This matters because text is **shaped per section**: `layout_section`
runs the shaper once per section, so each section is an independent
shaping run. Merging two adjacent sections into one means the text
across the old boundary is now shaped together as a single run, which
enables cross-boundary kerning (and, in principle, ligatures) that
previously did not happen. Additionally, `extra_letter_spacing` is
skipped before the first glyph of a section, so merging removes a "first
glyph" boundary and lets the spacing apply there.
The net effect is sub-pixel glyph position shifts at the former section
boundaries, which is why `easymarkeditor.png` was regenerated. The new
output is the more correct one — the text is now shaped as the author
wrote it, rather than being artificially split at `append` boundaries.
---------
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
<!--
Please read the "Making a PR" section of
[`CONTRIBUTING.md`](https://github.com/emilk/egui/blob/main/CONTRIBUTING.md)
before opening a Pull Request!
* Keep your PR:s small and focused.
* The PR title is what ends up in the changelog, so make it descriptive!
* If applicable, add a screenshot or gif.
* If it is a non-trivial addition, consider adding a demo for it to
`egui_demo_lib`, or a new example.
* Do NOT open PR:s from your `master` branch, as that makes it hard for
maintainers to test and add commits to your PR.
* Remember to run `cargo fmt` and `cargo clippy`.
* Open the PR as a draft until you have self-reviewed it and run
`./scripts/check.sh`.
* When you have addressed a PR comment, mark it as resolved.
Please be patient! I will review your PR, but my time is limited!
-->
* [x] I have followed the instructions in the PR template
At low apha values, premultiplied colors lose precision.
This PR makes the color picker use unmultiplied colors internally.
Before:
https://github.com/user-attachments/assets/4617a355-daa9-4911-86e6-518ac6867014
After:
https://github.com/user-attachments/assets/d9681b01-50d8-418e-b5a5-79b4bd1bbddf
* Related to #56 (Improve text — tracking issue)
## Summary
This PR integrates [harfrust](https://crates.io/crates/harfrust) (a
pure-Rust port of HarfBuzz) into epaint's text layout pipeline,
replacing the character-by-character glyph positioning with proper
OpenType text shaping.
### What this enables
- **GPOS kerning**: most modern fonts only ship kerning in GPOS tables
(not the legacy `kern` table). Pairs like "AV", "VA", "AT" are now
properly tightened.
- **GSUB substitutions**: ligatures (fi, fl), contextual alternates, and
other OpenType features.
- **Combining marks**: diacritics (e.g. ɔ̃) are positioned via anchor
tables instead of being rendered as standalone replacement glyphs.
### Before/After
#### Kerning, etc.
<img width="838" height="726" alt="before_main"
src="https://github.com/user-attachments/assets/f0f26d5f-b117-43a6-b39c-ea40d2e73836"
/>
<img width="838" height="726" alt="after_harfrust"
src="https://github.com/user-attachments/assets/d983e5da-486c-4f39-bd4f-5782a90c6b39"
/>
#### Ligatures
<img width="1117" height="698" alt="before_closeup"
src="https://github.com/user-attachments/assets/7a3b08b4-cf6f-45b7-98ba-07c473cd3b02"
/>
<img width="1117" height="698" alt="after_closeup"
src="https://github.com/user-attachments/assets/6cfc5f21-d32f-4f09-be0c-59c8c553d44f"
/>
### Architecture
The shaping integrates into the existing pipeline without changing the
public API:
1. **`Font::segment_into_runs`** — segments text into contiguous runs by
font face (grapheme-cluster aware, never splits combining sequences)
2. **`FontFace::shape_text`** — calls harfrust to shape each run,
returning glyph IDs + positioned advances/offsets
3. **`layout_shaped_run`** — emits `Glyph` structs from the shaping
output, with NOTDEF fallback to other font faces for missing glyphs
4. **Buffer recycling** — `FontsImpl` pools a `harfrust::UnicodeBuffer`
to avoid per-layout allocations
### Disclaimer
I'm far from being a good Rust programmer. Claude Code did most of the
heavy lifting here. I did my best and used my limited knowledge to avoid
making too many mistakes. If this PR isn't up to quality standards,
please don't hesitate to close it.
## Test plan
- [x] `cargo test -p epaint` — all 18 text tests pass, including 6 new
ones
- [x] `cargo clippy -p epaint --all-features` — clean
- [x] `cargo fmt` — clean
- [ ] Snapshot tests need regeneration (expected: shaping changes glyph
positions)
- New tests added:
- `test_gpos_kerning` — verifies GPOS kerning tightens "AV", "VA", "AT"
pairs
- `test_combining_diacritics` — combining tilde doesn't add extra width
- `test_shaping_basic_latin` — sanity check for Latin text
- `test_shaping_empty_string` — empty input doesn't panic
- `test_shaping_multiple_newlines` — newline splitting works correctly
- `test_shaping_mixed_font_fallback` — Latin + emoji in same string
---------
Co-authored-by: Emil Ernerfeldt <emil.ernerfeldt@gmail.com>
- fix for https://github.com/rerun-io/reality/pull/1075
The galleys row size was calculated by looking at the last glyphs pos_x,
which got changed to be rounded to integers when we added subpixel
binning. This introduced a subtle bug which caused the width of galleys
to be slightly off.
This PR fixes this by looking at the actual cursor position instead,
which is not rounded.
Also added a test to ensure this is correct. Previously, for the second
and last line, the `x` was too close to the `0`.
<img width="48" height="67" alt="image"
src="https://github.com/user-attachments/assets/a69a4cc3-b3f3-4553-ab92-73cb2e7a358c"
/>
---------
Co-authored-by: lucasmerlin <8009393+lucasmerlin@users.noreply.github.com>
* Part of https://github.com/emilk/egui/issues/5605
This changes the default style of egui.
The problem with "expanding" widgets is that they now want to paint
outside their own bounds, which then requires all parent UIs to have
proper margins.
It also means hovered things are no longer properly aligned with every
other widget.
<!--
Please read the "Making a PR" section of
[`CONTRIBUTING.md`](https://github.com/emilk/egui/blob/main/CONTRIBUTING.md)
before opening a Pull Request!
* Keep your PR:s small and focused.
* The PR title is what ends up in the changelog, so make it descriptive!
* If applicable, add a screenshot or gif.
* If it is a non-trivial addition, consider adding a demo for it to
`egui_demo_lib`, or a new example.
* Do NOT open PR:s from your `master` branch, as that makes it hard for
maintainers to test and add commits to your PR.
* Remember to run `cargo fmt` and `cargo clippy`.
* Open the PR as a draft until you have self-reviewed it and run
`./scripts/check.sh`.
* When you have addressed a PR comment, mark it as resolved.
Please be patient! I will review your PR, but my time is limited!
-->
* Closes N/A
* [x] I have followed the instructions in the PR template
I'll probably come back to this and clean it up a bit. This PR
reimplements ab_glyph's functionality on top of Skrifa, a somewhat
lower-level font API that's being used in Chrome now.
Skrifa doesn't perform rasterization itself, so I'm using
[vello_cpu](https://github.com/linebender/vello) from the Linebender
project for rasterization. It's still in its early days, but I believe
it's already quite fast. It also supports color and gradient fills, so
color emoji support will be easier.
Skrifa also supports font hinting, which should make text look a bit
nicer / less blurry.
Here's the current ab_glyph rendering:
<img width="1592" height="1068" alt="image"
src="https://github.com/user-attachments/assets/2385b66e-23f8-4c6e-b8c2-ea90e0eea4e4"
/>
Here's Skrifa *without* hinting--it looks almost identical, but there
are some subpixel differences, probably due to rasterizer behavior:
<img width="1592" height="1068" alt="image"
src="https://github.com/user-attachments/assets/a815f3e9-65ac-4940-bc00-571177bef53d"
/>
Here's Skrifa *with* hinting:
<img width="1592" height="1068" alt="image"
src="https://github.com/user-attachments/assets/d6cc0669-3537-4377-bba9-ed5ef09664db"
/>
Hinting does make the horizontal strokes look a bit bolder, which makes
me wonder once again about increasing the font weight from "light" to
"regular".
---------
Co-authored-by: Emil Ernerfeldt <emil.ernerfeldt@gmail.com>
Sometimes when moving a window, having a tooltip attached to the mouse
pointer, or scrolling a `ScrollArea`, you would see this disturbing
effect:
This is caused by us rounding many visual elements (lines, rectangles,
text, …) to physical pixels in order to keep them sharp. If the
window/tooltip itself is not rounded to a physical pixel, then you can
get this behavior.
So from now on the position of all
areas/windows/tooltips/popups/ScrollArea gets rounded to the closes
pixel.
* Unlocked by https://github.com/emilk/egui/pull/7709
This adds a new mode, `UPDATE_SNAPSHOTS=force`, which will lower the
threshold to zero, overwriting every image that is not _exactly_ the
same.
Most comparisons has a threshold because different GPUs render slightly
differently. However, setting that threshold accurately can be hard.
Sometimes a test will pass locally, but fail on CI. In those cases you
want to force an update of the failing test. You can use
`UPDATE_SNAPSHOTS=force` for that.
And sometimes a small change _should_ update all images, but the change
is so tiny that it falls under the threshold. Still, you want to make a
point of showing that these images have changes. You can use
`UPDATE_SNAPSHOTS=force` for that.
* related #7494
Removes the `deadlock_detection` feature, since we now have a more
primitive panic-after-30s deadlock detection which works well enough and
even detects kinds of deadlocks that the `deadlock_detection` feature
never supported.
This adds a custom Node struct with proper support for egui types
(`Key`, `Modifiers`, `egui::Event`, `Rect`) instead of needing to use
the kittest / accesskit types.
I also changed the `click` function to do a proper mouse move / mouse
down instead of the accesskit click. Also added `accesskit_click` to
trigger the accesskit event. This resulted in some changed snapshots,
since the elements are now hovered.
Also renamed `press_key` to `key_press` for consistency with
`key_down/key_up`.
Also removed the Deref to the AccessKit Node, to make it clearer when to
expect egui and when to expect accesskit types.
* Closes#5705
* [x] I have followed the instructions in the PR template
Today each widget does its own custom layout, which has some drawbacks:
- not very flexible
- you can add an `Image` to `Button` but it will always be shown on the
left side
- you can't add a `Image` to a e.g. a `SelectableLabel`
- a lot of duplicated code
This PR introduces `Atoms` and `AtomLayout` which abstracts over "widget
content" and layout within widgets, so it'd be possible to add images /
text / custom rendering (for e.g. the checkbox) to any widget.
A simple custom button implementation is now as easy as this:
```rs
pub struct ALButton<'a> {
al: AtomicLayout<'a>,
}
impl<'a> ALButton<'a> {
pub fn new(content: impl IntoAtomics) -> Self {
Self { al: content.into_atomics() }
}
}
impl<'a> Widget for ALButton<'a> {
fn ui(mut self, ui: &mut Ui) -> Response {
let response = ui.ctx().read_response(ui.next_auto_id());
let visuals = response.map_or(&ui.style().visuals.widgets.inactive, |response| {
ui.style().interact(&response)
});
self.al.frame = self
.al
.frame
.inner_margin(ui.style().spacing.button_padding)
.fill(visuals.bg_fill)
.stroke(visuals.bg_stroke)
.corner_radius(visuals.corner_radius);
self.al.show(ui)
}
}
```
The initial implementation only does very basic layout, just enough to
be able to implement most current egui widgets, so:
- only horizontal layout
- everything is centered
- a single item may grow/shrink based on the available space
- everything can be contained in a Frame
There is a trait `IntoAtoms` that conveniently allows you to construct
`Atoms` from a tuple
```
ui.button((Image::new("image.png"), "Click me!"))
```
to get a button with image and text.
This PR reimplements three egui widgets based on the new AtomLayout:
- Button
- matches the old button pixel-by-pixel
- Button with image is now [properly
aligned](https://github.com/emilk/egui/pull/5830/files#diff-962ce2c68ab50724b01c6b64c683c4067edd9b79fcdcb39a6071021e33ebe772)
in justified layouts
- selected button style now matches SelecatbleLabel look
- For some reason the DragValue text seems shifted by a pixel almost
everywhere, but I think it's more centered now, yay?
- Checkbox
- basically pixel-perfect but apparently the check mesh is very slightly
different so I had to update the snapshot
- somehow needs a bit more space in some snapshot tests?
- RadioButton
- pixel-perfect
- somehow needs a bit more space in some snapshot tests?
I plan on updating TextEdit based on AtomLayout in a separate PR (so
you could use it to add a icon within the textedit frame).
This is the same change as in #4069 but as this is stale I wanted to
reopen a non stale PR
Modifies ImageLoader's load function to use background threads for the
image decoding work. This avoids blocking the main thread that is
especially noticeable when loading many images at once.
This was modelled after the other loader implementations that also use
threads.
* Closes <https://github.com/emilk/egui/issues/5375>
* [x] I have followed the instructions in the PR template
---------
Co-authored-by: lucasmerlin <lucasmeurer96@gmail.com>
Co-authored-by: Emil Ernerfeldt <emil.ernerfeldt@gmail.com>
The bug was in `Color32::from_rgba_unmultiplied` and by extension
affects:
* `Color32::from_rgba_unmultiplied`
* `hex_color!`
* `HexColor`
* `ColorImage::from_rgba_unmultiplied`
* All images with transparency (png, webp, …)
* `Color32::from_white_alpha`
The bug caused translucent colors to appear too bright.
## More
Color is hard.
When I started out egui I thought "linear space is objectively better,
for everything!" and then I've been slowly walking that back for various
reasons:
* sRGB textures not available everywhere
* gamma-space is more _perceptually_ even, so it makes sense to use for
anti-aliasing
* other applications do everything in gamma space, so that's what people
expect (this PR)
Similarly, pre-multiplied alpha _makes sense_ for blending colors. It
also enables additive colors, which is nice. But it does complicate
things. Especially when mixed with sRGB/gamma (As @karhu [points
out](https://github.com/emilk/egui/pull/5824#issuecomment-2738099254)).
## Related
* Closes https://github.com/emilk/egui/issues/5751
* Closes https://github.com/emilk/egui/issues/5771 ? (probably; hard to
tell without a repro)
* But not https://github.com/emilk/egui/issues/5810
## TODO
* [x] I broke the RGBA u8 color picker. Fix it
---------
Co-authored-by: Andreas Reich <andreas@rerun.io>
I got annoyed by all the slightly different variations of "collect
snapshot results and unwrap them at the end of test" I've written, so I
added a struct to make this nice and simple.
One controversial thing: It panics when dropped. I wanted to ensure
people cannot forget to unwrap the results at the end, and this was the
best thing I could come up with. I don't think this is possible via
clippy lint or something like that.
* [x] I have followed the instructions in the PR template
## Defining what `Rounding` is
This PR defines what `Rounding` means: it is the corner radius of
underlying `RectShape` rectangle. If you use `StrokeKind::Inside`, this
means the rounding is of the outer part of the stroke. Conversely, if
you use `StrokeKind::Outside`, the stroke is outside the rounded
rectangle, so the stroke has an inner radius or `rounding`, and an outer
radius that is larger by `stroke.width`.
This definitions is the same as Figma uses.
## Improving general shape rendering
The rendering of filled shapes (rectangles, circles, paths, bezier) has
been rewritten. Instead of first painting the fill with the stroke on
top, we now paint them as one single mesh with shared vertices at the
border. This has several benefits:
* Less work (faster and with fewer vertices produced)
* No overdraw (nicer rendering of translucent shapes)
* Correct blending of stroke and fill
The logic for rendering thin strokes has also been improved, so that the
width of a stroke of `StrokeKind::Outside` never affects the filled area
(this used to be wrong for thin strokes).
## Improving of rectangle rendering
Rectangles also has specific improvements in how thin rectangles are
painted.
The handling of "Blur width" is also a lot better, and now works for
rectangles with strokes.
There also used to be bugs with specific combinations of corner radius
and stroke width, that are now fixed.
## But why?
With the new `egui::Scene` we end up with a lot of zoomed out shapes,
with sub-pixel strokes. These need to look good! One thing led to
another, and then I became obsessive 😅
## Tessellation Test
In order to investigate the rendering, I created a Tessellation Test in
the `egui_demo_lib`.
[Try it
here](https://egui-pr-preview.github.io/pr/5669-emilkimprove-tessellator)
This is a breaking change, requiring users to think about wether the
stroke is inside/centered/outside the rect.
When in doubt, add `egui::StrokeKind::Inside` to the function call.
Previously, `Harness::run` just called `Harness::step` 3 times. If that
wasn't enough, tests would often call run multiple times so all
animations would finish properly.
Also, I introduced `HarnessBuilder::with_step_dt` to customize with how
big of a dt each frame is called. I set the default to 1.0 / 6.0 (~6fps)
so we don't waste cpu in tests waiting on animations.
`HarnessBuilder::max_steps` allows us to control how many steps
`Harness::run` should run before panicing.
The default is 6, so we run for up to 1.0 logical seconds (six frames at
6 fps), which should be enough to finish most animations.
Turns out a lot of snapshots where rendered before fully shown and had a
light opacity, those are now fixed.
* [x] I have followed the instructions in the PR template
* Part of https://github.com/emilk/egui/issues/4019
`Frame` now includes the width of the stroke as part of its size. From
the new docs:
### `Frame` docs
The total (outer) size of a frame is `content_size + inner_margin +
2*stroke.width + outer_margin`.
Everything within the stroke is filled with the fill color (if any).
```text
+-----------------^-------------------------------------- -+
| | outer_margin |
| +------------v----^------------------------------+ |
| | | stroke width | |
| | +------------v---^---------------------+ | |
| | | | inner_margin | | |
| | | +-----------v----------------+ | | |
| | | | ^ | | | |
| | | | | | | | |
| | | |<------ content_size ------>| | | |
| | | | | | | | |
| | | | v | | | |
| | | +------- content_rect -------+ | | |
| | | | | |
| | +-------------fill_rect ---------------+ | |
| | | |
| +----------------- widget_rect ------------------+ |
| |
+---------------------- outer_rect ------------------------+
```
The four rectangles, from inside to outside, are:
* `content_rect`: the rectangle that is made available to the inner
[`Ui`] or widget.
* `fill_rect`: the rectangle that is filled with the fill color (inside
the stroke, if any).
* `widget_rect`: is the interactive part of the widget (what sense
clicks etc).
* `outer_rect`: what is allocated in the outer [`Ui`], and is what is
returned by [`Response::rect`].
### Notes
This required rewriting a lot of the layout code for `egui::Window`,
which was a massive pain. But now the window margin and stroke width is
properly accounted for everywhere.
