[Text] Unicode and text layout optimizations by Gillibald · Pull Request #21400 · AvaloniaUI/Avalonia

Gillibald · 2026-05-20T11:20:08Z

What does the pull request do?

Performance + correctness pass on the text formatting pipeline. Wrap, measure and metrics paths now read width data from a cached cluster-prefix sum on ShapedBuffer instead of walking glyphs every time, the LineBreakEnumerator stops re-decoding the same codepoint on each rule and now dispatches its rules through direct static calls + selective inlining (no more delegate array), and the Skia GlyphRunImpl builds positions and ink bounds in a single fused pass.

What is the current behavior?

TextLayout re-sums per-glyph advances on every MeasureLength, total-width, and split-driven wrap iteration. SplitTextRuns needs a second pass to total the first half's length.

LineBreakEnumerator rules call state.Next(text).LineBreakClass repeatedly, redoing the trie lookup per rule, and dispatch each of its 42 rules through a BreakUnitDelegate[] — the indirect call blocks JIT inlining across rule boundaries.

GlyphRunImpl walks the glyph array twice with separate currentX accumulators and returns the rented SKRect[] outside of a try/finally.

What is the updated/expected behavior with this PR?

Identical observable behaviour, faster steady-state.

LineBreakEnumerator micro-bench (1024 codepoints, BDN default job, --inProcess, same-session vs upstream/master):

Distribution	master	branch	Speedup
Ascii	174.9 µs	24.84 µs	7.04×
Bmp	190.6 µs	32.81 µs	5.81×
Supplementary	225.4 µs	37.56 µs	6.00×

End-to-end TextLayoutProfile.BuildEmojisWrapped (worst-case WrapWithOverflow + emoji input, same-session pair, BDN default job, --inProcess):

Branch	Mean	Allocated	Gen0	Gen1
master	1.764 ms	600.57 KB	46.8750	15.6250
branch	1.375 ms	570.15 KB	42.9688	15.6250
Δ	−22.0 %	−5.07 %	−8.3 %	0 %

≈ 1.28× end-to-end speedup, no Gen1 regression. New characterisation tests pin wrap and split-runs behaviour so the refactors stay observationally equivalent.

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

ShapedBuffer grows a lazily built cluster-prefix-sum cache (advances + optional start-char offsets), pooled via ArrayPool, shared by reference across Split halves and WithBidiLevel aliases through a private offset-aware constructor.

A fast path skips the start-chars array when every cluster is exactly one char wide. A DEBUG-only _cacheShared guard on the indexer setter catches post-split mutation. GlyphRun.CreateGlyphRunMetrics and the wrap path read TotalGlyphAdvance / MeasureCharactersThatFit in O(1); SplitTextRuns exposes a firstLength out parameter (old signature preserved) so the wrap caller drops its second sweep.

LineBreakEnumerator caches NextClass / PreviousClass in advance, and all LB rules read those fields; its ExecuteRules was rewritten from a BreakUnitDelegate[] loop to direct sequential static calls with goto Done early-exit, and the small single-condition rules are tagged [MethodImpl(AggressiveInlining)] (the attribute was a no-op through the delegate array and only takes effect now). The unused LineBreakPairTable.cs is removed. Skia GlyphRunImpl fuses the two currentX walks into one and moves the pool return into finally.

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

None. Public API surface is unchanged; the new SplitTextRuns overload is internal, and the old signature still exists as a thin wrapper.

Obsoletions / Deprecations

None.

Fixed issues

Copilot

Pull request overview

This WIP PR focuses on improving text layout and Unicode processing performance by caching expensive per-cluster computations, reducing per-iteration overhead in line breaking/wrapping, and adding benchmarks/tests to lock in behavior and allocation characteristics.

Changes:

Added a lazily-built cluster prefix-sum cache to ShapedBuffer and used it to speed up wrapping measurements and glyph-run width metrics.
Optimized LineBreakEnumerator by caching “next” class and combining several pure rules to reduce state accesses.
Added extensive new unit tests and benchmarks for Unicode tries/codepoint enumeration/break enumerators and text wrapping, plus a profiling harness for text layout.

Reviewed changes

Copilot reviewed 20 out of 21 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
tests/Avalonia.Skia.UnitTests/Media/TextFormatting/TextLineTests.cs	Makes bidi bounds assertion tolerant to ULP-level float summation differences.
tests/Avalonia.Skia.UnitTests/Media/TextFormatting/TextFormatterWrapCharacterizationTests.cs	Adds characterization tests to pin wrapping behavior via `TextFormatterImpl.FormatLine`.
tests/Avalonia.Benchmarks/Text/UnicodeTrieBenchmark.cs	Adds benchmarks for raw `UnicodeTrie.Get` throughput across tries/distributions.
tests/Avalonia.Benchmarks/Text/UnicodeBreakEnumeratorBenchmark.cs	Adds benchmarks for line/word/grapheme break enumerator iteration cost.
tests/Avalonia.Benchmarks/Text/TextLayoutProfile.cs	Adds a single-case profiling-oriented benchmark using a shared `TextRunCache`.
tests/Avalonia.Benchmarks/Text/HugeTextLayout.cs	Exposes emoji fixture text for reuse and updates benchmark to use the renamed constant.
tests/Avalonia.Benchmarks/Text/CodepointBenchmark.cs	Adds benchmarks covering `Codepoint.ReadAt`, enumeration, and property access patterns.
tests/Avalonia.Benchmarks/Program.cs	Adds a non-BDN profiling mode (`--profile-textlayout`) that boots real Skia/HarfBuzz services.
tests/Avalonia.Base.UnitTests/Media/TextFormatting/UnicodeTrieTests.cs	Adds direct unit coverage for `UnicodeTrie`/`UnicodeTrieBuilder` branches and round-trips.
tests/Avalonia.Base.UnitTests/Media/TextFormatting/UnicodeEnumeratorAllocationTests.cs	Adds allocation guard tests for break enumerators using `GC.GetAllocatedBytesForCurrentThread`.
tests/Avalonia.Base.UnitTests/Media/TextFormatting/UnicodeDataTests.cs	Adds spot checks for `UnicodeData` public surface (masks/shifts/defaults/known codepoints).
tests/Avalonia.Base.UnitTests/Media/TextFormatting/PropertyValueAliasHelperTests.cs	Adds lightweight round-trip tests for generated alias helper mappings.
tests/Avalonia.Base.UnitTests/Media/TextFormatting/LineBreakEnumeratorTests.cs	Re-enables previously skipped line break test data.
tests/Avalonia.Base.UnitTests/Media/TextFormatting/CodepointTests.cs	Adds comprehensive `Codepoint` unit tests (surrogates, helpers, bracket pairing, etc.).
src/Skia/Avalonia.Skia/GlyphRunImpl.cs	Fuses glyph position and bounds accumulation to reduce redundant passes.
src/Avalonia.Base/Media/TextFormatting/Unicode/LineBreakPairTable.cs	Removes unused historical pair table implementation.
src/Avalonia.Base/Media/TextFormatting/Unicode/LineBreakEnumerator.cs	Adds cached `NextClass`/`PreviousClass` in state and combines several rules for speed.
src/Avalonia.Base/Media/TextFormatting/TextFormatterImpl.cs	Uses `ShapedBuffer` cluster cache for wrap measurement; refactors wrapping loop and `SplitTextRuns` to return split length.
src/Avalonia.Base/Media/TextFormatting/ShapedBuffer.cs	Introduces shared cluster prefix-sum cache (including across splits) and cache-backed measurement helpers.
src/Avalonia.Base/Media/GlyphRun.cs	Uses `ShapedBuffer.TotalGlyphAdvance` to avoid O(n) advance summation in metrics creation.
.gitignore	Ignores `/planning`.

avaloniaui-bot · 2026-05-20T12:39:15Z

You can test this PR using the following package version. 12.1.999-cibuild0065619-alpha. (feed url: https://nuget-feed-all.avaloniaui.net/v3/index.json) [PRBUILDID]

Copilot

Pull request overview

Copilot reviewed 23 out of 24 changed files in this pull request and generated 4 comments.

avaloniaui-bot · 2026-05-21T05:09:10Z

You can test this PR using the following package version. 12.1.999-cibuild0065649-alpha. (feed url: https://nuget-feed-all.avaloniaui.net/v3/index.json) [PRBUILDID]

avaloniaui-bot · 2026-05-21T07:36:15Z

You can test this PR using the following package version. 12.1.999-cibuild0065655-alpha. (feed url: https://nuget-feed-all.avaloniaui.net/v3/index.json) [PRBUILDID]

- ShapedBuffer: lazy, pooled cluster-prefix cache shared across Split; adds TotalGlyphAdvance / MeasureCharactersThatFit. - GlyphRun/TextFormatterImpl/Skia GlyphRunImpl: read the cache, drop duplicate scans; SplitTextRuns now returns firstLength via out. - LineBreakEnumerator: cache Next/PreviousClass per advance; remove unused LineBreakPairTable. - Tests + benchmarks for the new paths and a dotnet-trace harness. ~1.32x faster / -5% alloc on the emoji-wrap micro-benchmark; no observable behaviour or public API change.

Replace the BreakUnitDelegate[] s_rules array dispatch in LineBreakEnumerator.ExecuteRules with a sequence of direct static calls and a `goto Done` early-exit. This removes 42 indirect calls per codepoint and lets the JIT reason across rule boundaries, which in turn makes [MethodImpl(AggressiveInlining)] meaningful — JIT cannot inline through delegate.Invoke, so the attribute was a no-op in the previous shape. Selectively apply AggressiveInlining to the small single-condition rules (LB03, LB04, LB06, LB07, LB08a, LB11–LB15d, LB18, LB20, LB21b, LB22, LB29, LB31) and let the JIT decide on the larger ones to avoid bloating the merged ExecuteRules. The static BreakUnitDelegate[] s_rules array is removed. Benchmarks (BDN default job, --inProcess, N=13–22, rel. StdDev <1.5%): UnicodeBreakEnumeratorBenchmark.LineBreakEnumerator_Sequence Ascii 154.6 µs -> 24.84 µs (6.22x) Bmp 190.2 µs -> 32.81 µs (5.80x) Supplementary 227.6 µs -> 37.56 µs (6.06x) TextLayoutProfile.BuildEmojisWrapped Before (branch, pre-inline) 804.5 µs / 570.15 KB After (branch, post-inline) 547.0 µs / 570.15 KB (-32.0%) vs upstream/master (1061 µs) -> ~1.94x total speedup No allocation change; CPU-only dispatch reshape. All LineBreakEnumerator unit tests pass (5/5).

avaloniaui-bot · 2026-05-27T06:15:40Z

You can test this PR using the following package version. 12.1.999-cibuild0065796-alpha. (feed url: https://nuget-feed-all.avaloniaui.net/v3/index.json) [PRBUILDID]

Copilot

Pull request overview

Copilot reviewed 21 out of 22 changed files in this pull request and generated 2 comments.

avaloniaui-bot · 2026-05-27T07:04:11Z

You can test this PR using the following package version. 12.1.999-cibuild0065798-alpha. (feed url: https://nuget-feed-all.avaloniaui.net/v3/index.json) [PRBUILDID]

…nter Wrap the ArrayPool-rented glyph and cluster-cache arrays in a small PooledArray<T> disposable and expose them through IRef<T>. Split children and WithBidiLevel aliases now Clone() the refs instead of borrowing raw pool arrays, so the backing storage survives until every sibling has been disposed - eliminating the UAF risk that existed when a parent was disposed before its children. Add a per-glyph-holder generation counter (Volatile.Read / Interlocked .Increment). The indexer setter bumps the counter on every write, and EnsureClusterCache compares its recorded generation against the holder's current value, rebuilding on mismatch. This lets us drop the previous "no mutation after Split/WithBidiLevel" contract: mutations performed through any sibling now propagate to the others' caches transparently. Dispose is made idempotent via a _disposed guard so overlapping cache eviction and TextLine teardown only release the IRefs once. Adds ShapedBufferSharedStorageTests covering sibling lifetime, Dispose idempotency, and generation-driven cache invalidation across Split children and WithBidiLevel aliases.

Copilot

Pull request overview

Copilot reviewed 22 out of 23 changed files in this pull request and generated 3 comments.

avaloniaui-bot · 2026-05-27T08:29:01Z

You can test this PR using the following package version. 12.1.999-cibuild0065800-alpha. (feed url: https://nuget-feed-all.avaloniaui.net/v3/index.json) [PRBUILDID]

…Level aliases Add a regression test hook (`ClusterPrefix`) exposing the backing cluster-prefix array reference, plus two tests that mutate a parent buffer before aliasing it and assert the alias reuses the parent's pooled prefix array instead of rebuilding. Guards against forgetting to propagate `_cacheGeneration` to alias buffers, which would silently defeat the cached-split fast path.

avaloniaui-bot · 2026-05-27T10:11:26Z

You can test this PR using the following package version. 12.1.999-cibuild0065816-alpha. (feed url: https://nuget-feed-all.avaloniaui.net/v3/index.json) [PRBUILDID]

MrJul

Impressive numbers, LGTM, let's merge!

avaloniaui-bot · 2026-05-27T15:08:24Z

You can test this PR using the following package version. 12.1.999-cibuild0065832-alpha. (feed url: https://nuget-feed-all.avaloniaui.net/v3/index.json) [PRBUILDID]

* perf(text): O(1) width queries on ShapedBuffer, cached LB-class lookup - ShapedBuffer: lazy, pooled cluster-prefix cache shared across Split; adds TotalGlyphAdvance / MeasureCharactersThatFit. - GlyphRun/TextFormatterImpl/Skia GlyphRunImpl: read the cache, drop duplicate scans; SplitTextRuns now returns firstLength via out. - LineBreakEnumerator: cache Next/PreviousClass per advance; remove unused LineBreakPairTable. - Tests + benchmarks for the new paths and a dotnet-trace harness. ~1.32x faster / -5% alloc on the emoji-wrap micro-benchmark; no observable behaviour or public API change. * perf(TextFormatting): inline LineBreakEnumerator rule dispatch Replace the BreakUnitDelegate[] s_rules array dispatch in LineBreakEnumerator.ExecuteRules with a sequence of direct static calls and a `goto Done` early-exit. This removes 42 indirect calls per codepoint and lets the JIT reason across rule boundaries, which in turn makes [MethodImpl(AggressiveInlining)] meaningful — JIT cannot inline through delegate.Invoke, so the attribute was a no-op in the previous shape. Selectively apply AggressiveInlining to the small single-condition rules (LB03, LB04, LB06, LB07, LB08a, LB11–LB15d, LB18, LB20, LB21b, LB22, LB29, LB31) and let the JIT decide on the larger ones to avoid bloating the merged ExecuteRules. The static BreakUnitDelegate[] s_rules array is removed. Benchmarks (BDN default job, --inProcess, N=13–22, rel. StdDev <1.5%): UnicodeBreakEnumeratorBenchmark.LineBreakEnumerator_Sequence Ascii 154.6 µs -> 24.84 µs (6.22x) Bmp 190.2 µs -> 32.81 µs (5.80x) Supplementary 227.6 µs -> 37.56 µs (6.06x) TextLayoutProfile.BuildEmojisWrapped Before (branch, pre-inline) 804.5 µs / 570.15 KB After (branch, post-inline) 547.0 µs / 570.15 KB (-32.0%) vs upstream/master (1061 µs) -> ~1.94x total speedup No allocation change; CPU-only dispatch reshape. All LineBreakEnumerator unit tests pass (5/5). * Refactor ShapedBuffer to share pool storage via IRef + generation counter Wrap the ArrayPool-rented glyph and cluster-cache arrays in a small PooledArray<T> disposable and expose them through IRef<T>. Split children and WithBidiLevel aliases now Clone() the refs instead of borrowing raw pool arrays, so the backing storage survives until every sibling has been disposed - eliminating the UAF risk that existed when a parent was disposed before its children. Add a per-glyph-holder generation counter (Volatile.Read / Interlocked .Increment). The indexer setter bumps the counter on every write, and EnsureClusterCache compares its recorded generation against the holder's current value, rebuilding on mismatch. This lets us drop the previous "no mutation after Split/WithBidiLevel" contract: mutations performed through any sibling now propagate to the others' caches transparently. Dispose is made idempotent via a _disposed guard so overlapping cache eviction and TextLine teardown only release the IRefs once. Adds ShapedBufferSharedStorageTests covering sibling lifetime, Dispose idempotency, and generation-driven cache invalidation across Split children and WithBidiLevel aliases. * test(ShapedBuffer): cover cluster-cache sharing across Split/WithBidiLevel aliases Add a regression test hook (`ClusterPrefix`) exposing the backing cluster-prefix array reference, plus two tests that mutate a parent buffer before aliasing it and assert the alias reuses the parent's pooled prefix array instead of rebuilding. Guards against forgetting to propagate `_cacheGeneration` to alias buffers, which would silently defeat the cached-split fast path. * Correctly use MathUtilities.LessThanOrClose --------- Co-authored-by: Julien Lebosquain <julien@lebosquain.net>

Copilot AI review requested due to automatic review settings May 20, 2026 11:20

Copilot started reviewing on behalf of Gillibald May 20, 2026 11:20 View session

kerams reviewed May 20, 2026

View reviewed changes

Comment thread src/Avalonia.Base/Media/TextFormatting/Unicode/LineBreakEnumerator.cs Outdated

Copilot AI reviewed May 20, 2026

View reviewed changes

Comment thread src/Avalonia.Base/Media/TextFormatting/TextFormatterImpl.cs Outdated

Comment thread tests/Avalonia.Skia.UnitTests/Media/TextFormatting/TextFormatterWrapCharacterizationTests.cs

MrJul added enhancement area-perf area-textprocessing labels May 20, 2026

Gillibald requested a review from Copilot May 21, 2026 04:36

Gillibald changed the title ~~[WIP] [Text] Unicode and text layout optimizations~~ [Text] Unicode and text layout optimizations May 21, 2026

Copilot started reviewing on behalf of Gillibald May 21, 2026 04:37 View session

Copilot AI reviewed May 21, 2026

View reviewed changes

Gillibald changed the title ~~[Text] Unicode and text layout optimizations~~ [WIP][Text] Unicode and text layout optimizations May 21, 2026

Gillibald changed the title ~~[WIP][Text] Unicode and text layout optimizations~~ [Text] Unicode and text layout optimizations May 21, 2026

Gillibald mentioned this pull request May 21, 2026

[Text] TextCollapsingProperties corrections #21404

Closed

3 tasks

MrJul reviewed May 26, 2026

View reviewed changes

Gillibald force-pushed the unicodeCorrections branch from 8bbb1cb to 3cf6919 Compare May 27, 2026 05:15

Gillibald changed the title ~~[Text] Unicode and text layout optimizations~~ [WIP] [Text] Unicode and text layout optimizations May 27, 2026

Gillibald force-pushed the unicodeCorrections branch from 02923b8 to ec614b7 Compare May 27, 2026 05:38

Gillibald changed the title ~~[WIP] [Text] Unicode and text layout optimizations~~ [Text] Unicode and text layout optimizations May 27, 2026

Gillibald requested a review from Copilot May 27, 2026 06:37

Copilot started reviewing on behalf of Gillibald May 27, 2026 06:38 View session

Copilot AI reviewed May 27, 2026

View reviewed changes

Comment thread src/Avalonia.Base/Media/TextFormatting/ShapedBuffer.cs Outdated

Comment thread src/Avalonia.Base/Media/TextFormatting/ShapedBuffer.cs Outdated

Gillibald requested a review from Copilot May 27, 2026 07:56

Copilot started reviewing on behalf of Gillibald May 27, 2026 07:56 View session

Copilot AI reviewed May 27, 2026

View reviewed changes

Comment thread src/Avalonia.Base/Media/TextFormatting/ShapedBuffer.cs Outdated

Comment thread src/Avalonia.Base/Media/TextFormatting/ShapedBuffer.cs

Comment thread tests/Avalonia.Benchmarks/Text/ShapedBufferOps.cs Outdated

Gillibald changed the title ~~[Text] Unicode and text layout optimizations~~ [WIP] [Text] Unicode and text layout optimizations May 27, 2026

Correctly use MathUtilities.LessThanOrClose

a922e44

Gillibald changed the title ~~[WIP] [Text] Unicode and text layout optimizations~~ [Text] Unicode and text layout optimizations May 27, 2026

Merge branch 'master' into unicodeCorrections

0fe4644

MrJul approved these changes May 27, 2026

View reviewed changes

MrJul enabled auto-merge May 27, 2026 14:42

MrJul added this pull request to the merge queue May 27, 2026

Merged via the queue into AvaloniaUI:master with commit 42bba0b May 27, 2026
10 checks passed

MrJul added the backport-candidate-12.0.x Consider this PR for backporting to 12.0 branch label Jun 23, 2026

MrJul added backported-12.0.x and removed backport-candidate-12.0.x Consider this PR for backporting to 12.0 branch labels Jun 23, 2026

Gillibald deleted the unicodeCorrections branch June 24, 2026 04:00

Uh oh!

Uh oh!

Conversation

Gillibald commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What does the pull request do?

What is the current behavior?

What is the updated/expected behavior with this PR?

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

Checklist

Breaking changes

Obsoletions / Deprecations

Fixed issues

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

avaloniaui-bot commented May 20, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

avaloniaui-bot commented May 21, 2026

Uh oh!

avaloniaui-bot commented May 21, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

avaloniaui-bot commented May 27, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

avaloniaui-bot commented May 27, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

avaloniaui-bot commented May 27, 2026

Uh oh!

avaloniaui-bot commented May 27, 2026

Uh oh!

MrJul left a comment

Choose a reason for hiding this comment

Uh oh!

avaloniaui-bot commented May 27, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

Gillibald commented May 20, 2026 •

edited

Loading