[Gillibald]

What does the pull request do?

Fixes the long space after an emoji reported in #14011: a fallback run swallowed the space that follows it and shaped that space with the fallback font.

Top layer of a stack; based on #22066.

What is the current behavior?

TextCharacters.CreateShapeableRun extends a fallback run for as long as the fallback font has glyphs, rather than ending it where the default typeface regains coverage. Whitespace was explicitly exempt from the return-to-primary check so a space would not split a fallback run - but practically every font maps U+0020, so the space after fallback text is pulled into the fallback run and rendered with that font's space glyph. In most emoji fonts that is a full em.

Measured with FormattedText at FontSize=16, isolating the space by differencing:

	space advance
plain space	4.45
space immediately after an emoji	16.00 (1em, from the emoji font)

Nothing about it is emoji-specific: any fallback font that maps the space leaks the same way. Characters emoji fonts do not map (U+00A0, U+2060, U+200B) already end the run correctly, which is the same rule seen from the other side.

What is the updated/expected behavior with this PR?

A space next to fallback text measures with the default typeface, the same as anywhere else on the line:

var tf = new Typeface(FontFamily.Default);
double W(string s) => new FormattedText(s, CultureInfo.InvariantCulture,
    FlowDirection.LeftToRight, tf, 16, Brushes.Black).Width;

W("a b") - W("ab");                   // 4.45
W("\U0001F642 b") - W("\U0001F642b"); // 4.45, was 16.00

Or type an emoji followed by a space into a TextBox. This matches browsers, where the primary font wins the space in every position - after an emoji, between two emoji, and between two CJK words (measured in Chrome via canvas measureText for both Arial and Times New Roman).

How was the solution implemented (if it's not obvious)?

• TryGetShapeableLength applies the return-to-primary check to spacing whitespace (Zs) as well, so a fallback run ends at the first cluster the default typeface can render. Only Zs qualifies: control and format codepoints keep their cluster with the fallback run - many fonts map the default-ignorable bidi controls, and a default typeface whose cmap merely has such a mark must not pull it out of the run (TryGetShapeableLength_Does_Not_Reclaim_A_Bidi_Control_As_Whitespace pins this).
• The check is gated on a new defaultCanShapeScript argument. A default typeface that cannot shape the run's script was passed as null and so was no return target at all, which is why Arabic and Hebrew lines never reached the check. It is still not a return target for text - handing clusters back would block a shaping-capable fallback that merely shares its cmap - but it does reclaim the spacing whitespace between the words, which carries no shaping.
• A run of pure whitespace no longer becomes the anti-thrashing bias for the next run. Otherwise the words on either side of a space each resolve their fallback from scratch and can land on different fonts; GetShapeableCharacters_Keeps_The_Previous_Fallback_Across_A_Space pins this with two regional subset fonts.
• TextLineImpl.TryMergeWithLastBounds compares the two coordinates with MathUtilities.AreClose instead of ==. The edges are reached by summing glyph advances along different paths, so abutting bounds can land an ULP apart (43.20703125 vs 43.20703124999999) and one directional span gets reported as two rectangles.

The first commit adds the failing tests.

Test expectations that encoded the old run structure moved with it. Two are worth a reviewer's eye: glyph clusters are relative to each run's own text (the affected helpers now add the run's start via TextTestHelper.GetStartCharIndex, and one helper also indexed its rect list with a per-run glyph index), and the right-to-left trimming rows concatenate run texts in visual run order, so the same characters regroup - "…שתיים one" becomes "… שתיים one".

Gates: Skia unit 445/445, Base unit 3136/3136, Skia render tests 370/370 with no golden image changed.

Checklist

• Added unit tests (if possible)?
• Added XML documentation to any related classes?
• Consider submitting a PR to https://github.com/AvaloniaUI/avalonia-docs with user documentation

Breaking changes

No API changes, but the layout of existing text changes where fallback text and whitespace meet:

• a space adjacent to fallback text measures with the default typeface, so such lines reflow slightly. Widths can go either way - the platform emoji fonts are wider than a typical primary, other fallbacks are narrower.
• those lines are cut into more runs: four emoji separated by spaces now produce 7 text runs instead of 1. Adjacent runs that share a typeface are still re-merged before shaping, so text without fallbacks is unaffected.

Obsoletions / Deprecations

None.

Fixed issues

Fixes #14011

[MrJul]

LGTM!

[avaloniaui-bot]

You can test this PR using the following package version. 12.2.999-cibuild0068937-alpha. (feed url: https://nuget-feed-all.avaloniaui.net/v3/index.json) [PRBUILDID]