Avalonia

Commit Graph

Author	SHA1	Message	Date
Benedikt Stebner	acc1494308	Consolidate the Unicode segmentation trie and add UAX-29 sentence segmentation (#21875 ) * Pack the segmentation properties into one Unicode trie Grapheme, indic-conjunct, word, line and sentence break classes now share a single SegmentationTrie, together with the Emoji, Emoji_Presentation and Default_Ignorable_Code_Point flags that rode in the grapheme word. The UnicodeDataTrie word keeps general category, script and script extensions. - SegmentationTrieGenerator replaces GraphemeBreakClassTrieGenerator and emits Segmentation.trie.cs; UnicodeData.trie.cs is regenerated without the break fields it no longer carries. - SentenceBreakClass data and Codepoint.SentenceBreakClass expose UAX-29 sentence classes to callers. - WordSegment carries the source text slice alongside the code-unit and code-point readouts, matching the shape newer segment types use; the existing constructor and members stay for compatibility. Data regenerated against UCD 17.0.0. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * Add UAX-29 sentence segmentation SentenceBreakEnumerator walks sentence boundaries over a UTF-16 span (rules SB1-SB11 and SB998), yielding SentenceSegment values of offset plus text slice. Sentence is the one segmentation granularity accessibility clients and IME navigation ask for that had no implementation, so callers had to approximate it with punctuation heuristics. The conformance suite runs against the Unicode SentenceBreakTest.txt data. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * Make sentence segmentation linear in the text length Every rule that needed left-hand context rescanned the text backwards from the current position, so a run of spaces or closing brackets was walked once per character in it. Enumerating 40k spaces took 5.5 s; 40k closing brackets took 8.1 s, and both grew quadratically. - Fold the left-hand context into the enumerator as it advances: the last two significant classes for SB6/SB7, and a stage counter over the (STerm \| ATerm) Close* Sp* (Sep \| CR \| LF)? grammar for SB8-SB11. - Memoize the SB8 lookahead. A scan answers every start position it passed over, so the Close* Sp* run after an ATerm is scanned once, not once per position. - Classify with bitmasks, matching LineBreakEnumerator. Both pathological inputs now finish in 0.1 ms at 40k characters. Output is unchanged: the UCD SentenceBreakTest.txt suite passes, and 1.4M random strings over the break classes segment identically to the previous implementation. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * Read the word segment length from the property valid for every segment WordSegment.Text is empty for segments built through the code-unit constructor that shipped in 12.1, so Text.Length holds only for segments that come from WordBreakEnumerator. Length is set by both constructors. - Measure the ellipsis break position from Length again. - Say on the Text property itself, where it is read, that it can be empty. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * Benchmark word and sentence segmentation across text shapes The break enumerator benchmark covered three enumerators over random codepoint distributions, which measures trie lookup cost but not the per-character cost of the rules themselves. Long runs of one class are where a segmentation algorithm degrades, and nothing measured them. - Add the sentence enumerator to UnicodeBreakEnumeratorBenchmark. - Add UnicodeSegmentationScalingBenchmark: prose, Han, ZWJ emoji, and runs of spaces, closing punctuation, terminator-plus-spaces and regional indicators, each at two lengths so the cost per character is comparable between them. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * Document the sentence break classes and trim the break unit SentenceBreakClass is public API with nothing but the UCD aliases to go on, so a caller reading Codepoint.SentenceBreakClass has to consult UAX #29 to learn what a value means or which rule reacts to it. - Describe the type and every member in terms of what the sentence rules do with it, keeping the UCD alias as a trailing comment. - Give SentenceBreakUnit a primary constructor and drop its Codepoint property, which nothing reads now that the class is stored directly. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * Clean up documentation for SentenceBreakClass Removed remarks from the SentenceBreakClass documentation. --------- Co-authored-by: Claude Opus 5 <noreply@anthropic.com>	1 week ago
Benedikt Stebner	79ff4ff73e	[Text] Glyph Outline Part 5/13 - Shaped buffer glyph indices (#21501 ) * Fix glyph metrics; expose glyph bounding box GlyphMetrics previously stored the advance in Width/Height, leaving no field for the actual ink bounding box. Correct the contract: - Expand GlyphMetrics with AdvanceWidth / AdvanceHeight (and the bitmap / vertical-layout members XOffset / YOffset / VerticalOriginX/Y, reserved for later population). - Add GlyfTable.TryGetGlyphBounds: an allocation-free, header-only read of the glyph's control-point bounding box (xMin/yMin/xMax/yMax). Empty glyphs return true with zero bounds; out-of-range / short data return false. Composite glyphs use their header bbox without recursion. - TryGetGlyphMetrics (single + batch) now sources advances into AdvanceWidth / AdvanceHeight and the bounding box into Width / Height / XBearing / YBearing via TryGetGlyphBounds. Side bearings fall back to hmtx/vmtx when the glyph has no outline. The previously latent API had one test relying on the old behaviour (Width == advance); it's updated to assert AdvanceWidth. New tests cover TryGetGlyphBounds (range, empty, letter, descriptor parity) and the corrected metrics (advance placement, ink width distinct from advance, empty-glyph advance-without-ink, batch/single parity). * Add glyph-bounds benchmark GlyphBoundsBenchmark (Avalonia.Benchmarks): a BenchmarkDotNet comparison of the table-based bounds read against Skia's SKFont.GetGlyphWidths, in kernel (font pre-created) and per-run variants across 1 / 16 / 256 glyphs, with MemoryDiagnoser. * Add allocation-free batch glyph-bounds path Adds a bounds-only batch reader so ink-bounds computation (the GlyphRunImpl use case, which already has advances from shaping) doesn't pay for advance lookup or per-glyph indirection: - GlyphBounds: internal value type carrying the control-point box. - GlyfTable.GetGlyphBounds / GlyphTypeface.TryGetGlyphBounds: fetch the glyf and loca spans once per batch and read offsets + headers directly via BinaryPrimitives — no per-glyph ReadOnlyMemory.Span conversion, no intermediate Memory.Slice, no nested call chain. LocaTable exposes RawData / IsShortFormat for this. - Batch TryGetGlyphMetrics refactored onto the same span-cached reader. Measured against Skia's SKFont.GetGlyphWidths (bounds-only, in-process BenchmarkDotNet): the table path is faster at every run length (0.06× / 0.32× / 0.66× at 1 / 16 / 256 glyphs) and allocation-free, where Skia allocates 104 B/run for the SKFont. The earlier advance+bounds path via TryGetGlyphMetrics was ~2.2× slower than Skia at 256 glyphs; the bounds-only span-cached path is ~1.5× faster. The benchmark is updated to compare bounds-only on both sides. * Restore System.Buffers.Binary using in GlyfTable The glyf bounds readers (TryGetGlyphBounds / GetGlyphBounds) read the header int16s via BinaryPrimitives. The base pr2 commit trimmed this using as unused there, but it is needed once pr2b adds the bounds code — without it GlyfTable does not compile on this branch. * Clamp glyph bounding-box extents to non-negative GlyphBounds.Width/Height and both TryGetGlyphMetrics overloads computed the ink extent as (xMax - xMin) / (yMax - yMin). A malformed glyf header where xMax < xMin (or yMax < yMin) produced a negative value that, when narrowed to the ushort GlyphMetrics.Width/Height, wrapped to a huge positive extent instead of a safe zero. Clamp the extent to non-negative in GlyphBounds (the single source of truth for both the single and batch metric paths) and route the single TryGetGlyphMetrics path through GlyphBounds so both behave identically. For int16 coordinates the maximum extent is 65535 = ushort.MaxValue, so the clamped value always fits without overflow. Adds GlyphBoundsTests. * Allocate batch glyph-metric buffers only when their source is present Batch TryGetGlyphMetrics unconditionally sized the hMetrics, vMetrics and bounds temporary buffers to glyphIds.Length, even when the font has no hmtx, no vmtx, or no glyf table. For CFF/CFF2 fonts (no glyf) the bounds buffer is never read, and most fonts have no vmtx, so a large batch on such a font heap-allocated a per-glyph array that was pure waste. Size each buffer to zero — a free, empty stackalloc — when its source table is absent, so it only materialises (and only spills to the heap past 256 glyphs) when it will actually be filled and read. * Note the halved loca offsets in the batch bounds reader The short loca format stores every offset divided by two, which is why the batch reader doubles what it reads. LocaTable's single-glyph path already carries that note; the inline decode in the batch path did not. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * Split the batch metrics loop on the glyf branch instead of per glyph The glyf check does not vary across the run, so it belongs above the loop rather than inside it. Splitting it also lets the bounds buffer live only in the branch that reads it, which drops the zero-length allocation that CFF / CFF2 fonts used to make just to satisfy the shared loop. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * Add vertical glyph advance API to GlyphTypeface Mirror the existing TryGetHorizontalGlyphAdvance / ...Advances pair with vertical equivalents. Until now a vertical-layout caller (CJK, Mongolian) had no advance-only path on GlyphTypeface and had to go through TryGetGlyphMetrics, which after PR2b reads the glyf bounding box per glyph — pure waste when only the advance is needed. VerticalMetricsTable already exposes the per-glyph and batch readers internally (used by TryGetGlyphMetrics(batch)); this is a thin wrapper over them. Tests cover the Latin (no vmtx → false) and CJK (vmtx → positive advances) cases and batch/single parity using Inter and MiSans-Normal. Audit note: HorizontalMetricsTable and VerticalMetricsTable batch readers already span-cache (each fetches _data.Span once per call, reusing a BigEndianBinaryReader across glyphs). No metrics-table refactor is needed — the speculative Part 2 of the plan collapses to nothing. * Expose contiguous glyph indices span on ShapedBuffer Consumers that need a ReadOnlySpan<ushort> of glyph IDs (the upcoming GlyphTypeface.TryGetGlyphBounds batch path, SKFont.GetGlyphWidths, atlas lookups...) previously had to allocate a parallel ushort[] and project GlyphInfo.GlyphIndex into it. Carry that array on the ShapedBuffer instead: - ShapedBuffer rents a parallel ushort[] from ArrayPool alongside the existing GlyphInfo[] rental, exposed as `public ReadOnlySpan<ushort> GlyphIndices`. - The indexer setter syncs both, so HarfBuzz shaping and post-shape mutators (InterWordJustification rewrites the same GlyphIndex) keep the parallel view in lockstep with no caller change. - Split / WithBidiLevel slice the indices alongside the GlyphInfos at the same offsets — both share the parent's pooled arrays. - Dispose returns both rentals. Consumer wiring: - GlyphRunImpl: copies ShapedBuffer.GlyphIndices once via CopyTo when the source is a ShapedBuffer, dropping the per-glyph extract from the position loop. - TextFormatterImpl.CreateEmptyTextLine: switched to the pooled public ctor + indexer (was passing a heap-allocated GlyphInfo[] to the internal slice ctor, which now requires a parallel ushort slice). Tests cover indexer sync, overwrite, ascending split alignment, empty-leading split, and dispose clearing the view. --------- Co-authored-by: Claude Opus 5 <noreply@anthropic.com>	2 weeks ago
Benedikt Stebner	aee3f68551	TextCollapsingProperties bidirectional-text (BiDi) correctness fixes (#21446 ) * Fix BiDi correctness across TextCollapsingProperties implementations Audit and fix logical-vs-visual ordering bugs in every text-collapsing implementation and the helpers they share, so trimming always operates in logical order (the consumer re-runs the BiDi reorderer in FinalizeLine, so pre-applying visual order would be reordered a second time). - LogicalTextRunEnumerator returned the first run on every MoveNext when _indexedTextRuns was null. - TextLeadingPrefixCharacterEllipsis: walked visual-order runs while splitting at logical offsets, guarded the wrong field in the ctor, hardcoded LeftToRight for the symbol, appended the suffix in reverse-logical order, and mis-tracked the width budget across multi-run lines. - TextPathSegmentEllipsis: crashed via Split(0) in the fallback path, and measured RTL segments as ~0 width (it assumed ascending cluster ids). - ShapedTextRun.TryMeasureCharacters/Backwards returned visual-order counts for RTL runs. ShapedBuffer gains logical-order, cluster-cache-backed width helpers (GetCharRangeWidth, FindLeading/FindTrailingCharCountWithinWidth) that are safe in the one-char-per-cluster simple mode (null _clusterStartChars). Adds a BiDi characterization test suite and documents the logical-order Collapse contract. * Align text width comparisons to epsilon-tolerant MathUtilities Several width-fit comparisons in the measure/collapse path used raw operators with no epsilon, while the cluster-cache hot path (MeasureCharactersThatFit) and the ShapedTextRun branch of MeasureLength already went through MathUtilities. Replace the raw float comparisons with their epsilon-aware equivalents (<= -> LessThanOrClose, < -> LessThan, > -> GreaterThan) in ShapedBuffer.FindLeading/FindTrailingCharCountWithinWidth, TextPathSegmentEllipsis, and TextLeadingPrefixCharacterEllipsis. Also unify the DrawableTextRun branch of TextFormatterImpl.MeasureLength with the ShapedTextRun branch (>= -> GreaterThan) so the "whole run fits" boundary is identical for both run types: a run fits when width <= remaining (tolerant) and overflows only when strictly greater. Integer char-offset/index comparisons and the availableWidth <= 0 positivity guards are intentionally left as raw comparisons. * Cleanup some comments * Fix Split(0) crash in TextLeadingPrefixCharacterEllipsis suffix collection When a logical-tail run fits entirely within the remaining suffix budget, TryMeasureCharactersBackwards returns suffixCount == run.Length, so Split(run.Length - suffixCount) evaluated to Split(0), which throws ArgumentOutOfRangeException. This was reachable on multi-run lines whose post-split tail contains a small later run that fully fits. Guard the split on splitAt > 0 and use the whole run as-is when it fully fits (dropping it only when suffixCount == 0) — the same boundary pattern already applied to TextPathSegmentEllipsis. Add a multi-run regression test. * Remove duplicate ShapedBuffer.MeasureCharactersThatFit MeasureCharactersThatFit and FindLeadingCharCountWithinWidth ran the identical cluster-prefix binary search; the only functional difference was the unused `out widthConsumed`. Drop the former and repoint its callers (TextFormatterImpl.MeasureLength and the benchmark) at FindLeadingCharCountWithinWidth, which also unifies the availableWidth<=0 guard. The one test that asserted the consumed width now recovers it via GetCharRangeWidth(0, fit). Also tidy GetCharRangeWidth's simple-mode bounds to use Math.Clamp.	3 months ago
Benedikt Stebner	42bba0b859	[Text] Unicode and text layout optimizations (#21400 ) * 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>	4 months ago
Benedikt Stebner	f41f97ee26	Implement TextRunCache (#21030 )	5 months ago
Benedikt Stebner	e8424283fe	Universal GlyphTypeface implementation (#19852 ) * Introduce a universal IGlyphTypeface implementation that does not rely on any platform implementation * Revert changes * Fix Android * Make the test happy * Fix build * Update baseline * Fix naming * Fix headless * Move interfaces to dedicated files Make GlyphTypeface.GlyphCount an integer * Fix GlyphCount * Make IGlyphTypeface NotClientImplementable * Make sure we cache platform typefaces by their desired name, style, weight and stretch * Update baseline * Only use IGlyphTypeface * Fix Android * Try to clear the buffer before we encode somethimg * Add needed test font * Add more unit tests * Reduce allocations * Remove Direct2D1 test files * More tests * More complete table implementations * More adjustments * Use batch APIs * Handle invalid timestamps * Update baseline * Introduce a CharacterToGlyphMap struct for faster access * Remove AggressiveInlining * Remove AggressiveInlining * Make the head table optional for legacy fonts * Remove Load method. Fix TextBlockTests * Fix nullables * Remove redundant folder * Update Api baseline * revert diff helper changes * revert changes * Use bare minimum font for Headless platform and introduce a test font manager that uses the Inter font for testing. * Add missing font file for Headless platform --------- Co-authored-by: Gillibald <stebner@avaloniaui.net> Co-authored-by: Julien Lebosquain <julien@lebosquain.net>	8 months ago
Max Katz	4aa6035ec3	Reuse text and geometry related headless mocks in Avalonia unit tests and benchmarks	3 years ago
Nikita Tsukanov	0f4ed2a4b9	Updated tests to use IDispatcherImpl	4 years ago
Julien Lebosquain	5ffd961742	Perf: improved text wrapping	4 years ago
Julien Lebosquain	7a1f74a3d3	Benchmarks: option to use Skia for text layout	4 years ago
Julien Lebosquain	10a3b79d12	Perf: various misc text layout optimizations	4 years ago
Julien Lebosquain	89a78f557b	Don't keep the text layout buffers around if they're too large	4 years ago
Julien Lebosquain	fd0720fc56	Updated text layout benchmark with trimming/wrapping	4 years ago
Sergey Mikolaitis	56e94e424c	[Text] Add textblocks virtualization benchmark, rename field	4 years ago
Sergey Mikolaitis	39374451dd	[Text] Add pooling to shapedBuffer. works only with dispose.	4 years ago
Sergey Mikolaitis	2565538fd3	[Text] improve benchmarks. add BuildManySmallTexts benchmark	4 years ago
Sergey Mikolaitis	84206e93dd	[Text] Optimize grapheme and add performance benchmark	4 years ago

17 Commits (main)