perf: Views engine render optimizations — warm 99.99% faster, cold 10% faster#724
Open
johnml1135 wants to merge 4 commits intomainfrom
Open
perf: Views engine render optimizations — warm 99.99% faster, cold 10% faster#724johnml1135 wants to merge 4 commits intomainfrom
johnml1135 wants to merge 4 commits intomainfrom
Conversation
|
commit d0066f3247: commit c4678e8240: |
There was a problem hiding this comment.
Pull request overview
Performance-focused updates introduce render verification/benchmark infrastructure and targeted optimizations in the DetailControls/DataTree rendering path to reduce redundant work and support pixel-perfect render validation.
Changes:
- Added RenderVerification infrastructure (environment validation, diagnostics/trace toggling, benchmark result models, reporting & comparison utilities).
- Optimized
DataTree/Sliceconstruction, layout, painting, and visibility paths to avoid repeated O(N) work. - Added/updated test harness assets (layouts/parts, new diagnostics tests, VS Code tasks) and snapshot testing support.
Reviewed changes
Copilot reviewed 68 out of 153 changed files in this pull request and generated 4 comments.
Show a summary per file
| File | Description |
|---|---|
| Src/Common/RenderVerification/RenderModels.cs | Adds shared models for scenarios and timing results used by render verification/benchmarks. |
| Src/Common/RenderVerification/RenderEnvironmentValidator.cs | Captures environment details and produces a hash/diff for deterministic snapshot comparisons. |
| Src/Common/RenderVerification/RenderDiagnosticsToggle.cs | Adds a toggle for diagnostics/trace output with file-backed flags and a trace listener. |
| Src/Common/RenderVerification/RenderBenchmarkResults.cs | Defines JSON-serializable benchmark run/result structures and flags loading. |
| Src/Common/RenderVerification/RenderBenchmarkReportWriter.cs | Writes JSON + Markdown summary reports and recommendations for benchmark runs. |
| Src/Common/RenderVerification/RenderBenchmarkComparer.cs | Compares runs to detect regressions/improvements and generates a summary. |
| Src/Common/RenderVerification/CompositeViewCapture.cs | Implements two-pass DataTree capture with Views content overlay for pixel-perfect snapshots. |
| Src/Common/FieldWorks/FieldWorks.Diagnostics.dev.config | Adds a new trace switch for Views render timing. |
| Src/Common/Controls/XMLViews/XMLViewsTests/XmlBrowseViewBaseTests.cs | Updates test stub interface to include NeedsReconstruct. |
| Src/Common/Controls/DetailControls/VectorReferenceView.cs | Avoids drawing a trailing separator bar after the last item. |
| Src/Common/Controls/DetailControls/Slice.cs | Adds cached XML attribute parsing + width fast-path + DataTree method call updates. |
| Src/Common/Controls/DetailControls/PossibilityVectorReferenceView.cs | Avoids drawing a trailing separator bar after the last item. |
| Src/Common/Controls/DetailControls/HwndDiagnostics.cs | Adds helper methods to read USER/GDI handle counts for diagnostics/tests. |
| Src/Common/Controls/DetailControls/DetailControlsTests/TestResult.xml | Adds a test runner output artifact (currently shows a failed/invalid run). |
| Src/Common/Controls/DetailControls/DetailControlsTests/TestParts.xml | Extends test parts inventory to include senses sequences and collapsible slices. |
| Src/Common/Controls/DetailControls/DetailControlsTests/Test.fwlayout | Updates test layouts to include CitationForm and senses parts/layouts. |
| Src/Common/Controls/DetailControls/DetailControlsTests/DetailControlsTests.csproj | Adds snapshot/JSON deps and references RenderVerification project. |
| Src/Common/Controls/DetailControls/DetailControlsTests/DataTreeTests.cs | Updates expected template XML string for modified layout composition. |
| Src/Common/Controls/DetailControls/DetailControlsTests/DataTreeHwndCountTest.cs | Adds baseline diagnostics tests for handle growth and slice install count. |
| Src/Common/Controls/DetailControls/DataTree.cs | Adds multiple layout/paint/visibility optimizations and diagnostics counters. |
| Directory.Packages.props | Adds central package version for Verify. |
| AGENTS.md | Adds note about external dependency repository (liblcm). |
| .vscode/tasks.json | Adds tasks for render baseline tests and building tests via repo script. |
| .vscode/settings.json | Allows the new “Build Tests” task as an approved VS Code task. |
| .github/skills/openspec-verify-change/SKILL.md | Updates generatedBy metadata version. |
| .github/skills/openspec-sync-specs/SKILL.md | Updates generatedBy metadata version. |
| .github/skills/openspec-onboard/SKILL.md | Updates onboarding workflow text/commands and cross-platform command notes. |
| .github/skills/openspec-new-change/SKILL.md | Updates generatedBy metadata version. |
| .github/skills/openspec-ff-change/SKILL.md | Updates generatedBy metadata version. |
| .github/skills/openspec-explore/SKILL.md | Updates explore-mode wording around “proposal” vs other workflows. |
| .github/skills/openspec-continue-change/SKILL.md | Updates generatedBy metadata version. |
| .github/skills/openspec-bulk-archive-change/SKILL.md | Updates text and generatedBy metadata version. |
| .github/skills/openspec-archive-change/SKILL.md | Updates guidance to invoke the sync skill via Task tool. |
| .github/skills/openspec-apply-change/SKILL.md | Updates generatedBy metadata version. |
| .github/prompts/opsx-onboard.prompt.md | Mirrors onboarding prompt updates (CLI install check, command list). |
| .github/prompts/opsx-ff.prompt.md | Clarifies template usage vs context/rules constraints. |
| .github/prompts/opsx-explore.prompt.md | Mirrors explore prompt updates (“proposal” wording). |
| .github/prompts/opsx-bulk-archive.prompt.md | Mirrors bulk-archive prompt copy update. |
| .github/prompts/opsx-archive.prompt.md | Mirrors archive prompt sync invocation update. |
| .github/instructions/csharp.instructions.md | Adds C# instructions file (currently specifies “latest C# / C# 14” guidance). |
| .gitattributes | Marks Verify snapshot PNG extensions as binary. |
| .GitHub/skills/render-testing/SKILL.md | Adds a render-testing skill doc with build/test/run instructions and baseline workflow. |
| .GitHub/skills/code-review-skill-main/scripts/pr-analyzer.py | Adds a Python tool to analyze diffs and estimate PR review complexity. |
| .GitHub/skills/code-review-skill-main/reference/typescript.md | Adds TypeScript review reference documentation. |
| .GitHub/skills/code-review-skill-main/reference/security-review-guide.md | Adds security review reference documentation. |
| .GitHub/skills/code-review-skill-main/reference/java.md | Adds Java review reference documentation. |
| .GitHub/skills/code-review-skill-main/reference/cpp.md | Adds C++ review reference documentation. |
| .GitHub/skills/code-review-skill-main/reference/code-review-best-practices.md | Adds code review best practices reference documentation. |
| .GitHub/skills/code-review-skill-main/reference/c.md | Adds C review reference documentation. |
| .GitHub/skills/code-review-skill-main/assets/review-checklist.md | Adds a condensed review checklist asset. |
| .GitHub/skills/code-review-skill-main/assets/pr-review-template.md | Adds a PR review template asset. |
| .GitHub/skills/code-review-skill-main/SKILL.md | Adds a top-level code review skill description/index. |
| .GitHub/skills/code-review-skill-main/README.md | Adds documentation for the code-review skill bundle. |
| .GitHub/skills/code-review-skill-main/LICENSE | Adds license file for the bundled code-review skill content. |
| .GitHub/skills/code-review-skill-main/CONTRIBUTING.md | Adds contribution guidelines for the bundled code-review skill content. |
| .GitHub/skills/code-review-skill-main/.gitignore | Adds gitignore for the bundled code-review skill content. |
Comments suppressed due to low confidence (1)
.GitHub/skills/render-testing/SKILL.md:1
- This doc hardcodes a user-specific NUnit console path and recommends
dotnet buildfor a .NET Framework workflow. For portability and alignment with repo build guidance, prefer.\build.ps1/.\test.ps1(and avoid assumingC:\\Users\\johnm\\...exists). Also consider avoiding or clearly calling out commands with pipes/redirection in docs, since the repo guidance prefers wrapper scripts for piped commands.
---
Comment on lines
+64
to
+108
| public void EnableDiagnostics(bool persist = false) | ||
| { | ||
| m_originalDiagnosticsState = DiagnosticsEnabled; | ||
| DiagnosticsEnabled = true; | ||
| TraceEnabled = true; | ||
|
|
||
| if (persist) | ||
| { | ||
| SaveFlags(); | ||
| } | ||
|
|
||
| SetupTraceListener(); | ||
| } | ||
|
|
||
| /// <summary> | ||
| /// Disables diagnostics and trace output. | ||
| /// </summary> | ||
| /// <param name="persist">Whether to persist the change to the flags file.</param> | ||
| public void DisableDiagnostics(bool persist = false) | ||
| { | ||
| DiagnosticsEnabled = false; | ||
| TraceEnabled = false; | ||
|
|
||
| if (persist) | ||
| { | ||
| SaveFlags(); | ||
| } | ||
|
|
||
| RemoveTraceListener(); | ||
| } | ||
|
|
||
| /// <summary> | ||
| /// Restores the original diagnostics state before Enable/Disable was called. | ||
| /// </summary> | ||
| public void RestoreOriginalState() | ||
| { | ||
| if (m_originalDiagnosticsState) | ||
| { | ||
| EnableDiagnostics(persist: false); | ||
| } | ||
| else | ||
| { | ||
| DisableDiagnostics(persist: false); | ||
| } | ||
| } |
There was a problem hiding this comment.
RestoreOriginalState() is not idempotent when restoring to the enabled state because EnableDiagnostics() overwrites m_originalDiagnosticsState every time it is called. If DisableDiagnostics() is called and then RestoreOriginalState() enables diagnostics, the original state field can get reset to false, making subsequent restores incorrect. Fix by capturing m_originalDiagnosticsState only once (e.g., in ctor / first toggle), or by making RestoreOriginalState() call internal enable/disable helpers that do not mutate the stored original value.
Comment on lines
+1
to
+15
| <?xml version="1.0" encoding="utf-8" standalone="no"?> | ||
| <test-run id="0" runstate="Runnable" testcasecount="0" result="Failed" label="Invalid" total="0" passed="0" failed="0" warnings="0" inconclusive="0" skipped="0" asserts="0" engine-version="3.16.0.0" clr-version="4.0.30319.42000" start-time="2026-02-12 20:00:58Z" end-time="2026-02-12 20:00:58Z" duration="0.275532"> | ||
| <command-line><![CDATA["C:\Users\johnm\.nuget\packages\nunit.consolerunner\3.16.0\tools\nunit3-console.exe" bin\x64\Debug\DetailControlsTests.dll --where "test =~ /DataTreeRender/" --workers=1]]></command-line> | ||
| <filter> | ||
| <test re="1">DataTreeRender</test> | ||
| </filter> | ||
| <test-suite type="Assembly" id="1-1" name="DetailControlsTests.dll" fullname="C:\Users\johnm\Documents\repos\FieldWorks.worktrees\ui-speed-up\Src\Common\Controls\DetailControls\DetailControlsTests\bin\x64\Debug\DetailControlsTests.dll" testcasecount="0" runstate="NotRunnable" result="Failed" label="Invalid"> | ||
| <properties> | ||
| <property name="_SKIPREASON" value="File not found: C:\Users\johnm\Documents\repos\FieldWorks.worktrees\ui-speed-up\Src\Common\Controls\DetailControls\DetailControlsTests\bin\x64\Debug\DetailControlsTests.dll" /> | ||
| </properties> | ||
| <reason> | ||
| <message>File not found: C:\Users\johnm\Documents\repos\FieldWorks.worktrees\ui-speed-up\Src\Common\Controls\DetailControls\DetailControlsTests\bin\x64\Debug\DetailControlsTests.dll</message> | ||
| </reason> | ||
| </test-suite> | ||
| </test-run> No newline at end of file |
There was a problem hiding this comment.
This appears to be a local NUnit test-run output artifact containing user-specific absolute paths and a failed/invalid run. It should not be committed to source control; delete it from the repo and add it to an ignore list (or ensure test runners output to a non-tracked artifacts folder) to avoid noisy diffs and leaking local machine paths.
Comment on lines
+40
to
+48
| public string GetEnvironmentHash() | ||
| { | ||
| var settingsJson = Newtonsoft.Json.JsonConvert.SerializeObject(CurrentSettings); | ||
| using (var sha256 = SHA256.Create()) | ||
| { | ||
| var hashBytes = sha256.ComputeHash(Encoding.UTF8.GetBytes(settingsJson)); | ||
| return Convert.ToBase64String(hashBytes); | ||
| } | ||
| } |
There was a problem hiding this comment.
The environment hash is intended to be deterministic, but hashing JsonConvert.SerializeObject(CurrentSettings) can be unstable across runtimes/versions because JSON.NET property ordering is not guaranteed unless explicitly configured. That can cause spurious environment-hash mismatches even when the captured values are identical. Consider generating the hash from an explicit, stable field ordering (e.g., concatenating known fields in a fixed order), or configuring serialization with a contract resolver that enforces deterministic property ordering.
Comment on lines
+8
to
+10
| ## C# Instructions | ||
| - Always use the latest version C#, currently C# 14 features. | ||
| - Write clear and concise comments for each function. |
There was a problem hiding this comment.
This guidance conflicts with the repo's .NET Framework/C# constraints (FieldWorks is limited to C# 7.3). As written, it encourages unsupported language features (file-scoped namespaces, switch expressions, nullable reference types, etc.), which can lead to uncompilable PRs. Update the instructions to match the actual language version/tooling used in this codebase (e.g., explicitly state C# 7.3 and avoid recommending C# 8+ features).
NUnit Tests 1 files ± 0 1 suites ±0 10m 8s ⏱️ + 4m 15s For more details on these failures, see this check. Results for commit d6741a0. ± Comparison against base commit bcfbea0. ♻️ This comment has been updated with latest results. |
# This is the 1st commit message: feat(render): expand DataTree render scenarios and persist benchmark timings standardize DataTree test data with predictable field naming across scenarios add per-scenario timing capture and write results to datatree-timings.json add Newtonsoft.Json to DetailControlsTests for timing serialization add render-testing skill documentation for baseline and benchmark workflow remove obsolete generated Build/FieldWorks.targets.bad file update checked-in test result artifact for current render test run fix: resolve build blocker + timing test layout gaps - Remove duplicate Kernel project entry in FieldWorks.sln that blocked NuGet restore (first entry was missing EndProject) - Add CitationForm and Senses parts to test Normal layout so timing tests produce real slices instead of 0 - Add LexSense-Detail-Senses part and subsense recursion to GlossSn layout so WorkloadGrowsWithComplexity test correctly validates that deeper scenarios produce more slices - Include user edits to implementation-paths.md OS-3: All 4 DataTreeTiming tests now pass perf: add render-testing workflow + timing enhancement log - Update render-testing skill with explicit devil's advocate stage, retest/review, and mandatory TIMING_ENHANCEMENTS.md before commit. - Refine DataTree tab-index optimization safety by moving final ResetTabIndices(0) to CreateSlices finally. - Keep DeepSuspendLayout/DeepResumeLayout batching in DummyObjectSlice.BecomeReal. - Mark OS-3 and OS-5 complete in OpenSpec tasks and add TIMING_ENHANCEMENTS.md with measured outcomes. Validation: build.ps1 -BuildTests; test.ps1 -TestFilter DataTreeTiming -NoBuild perf: paint pipeline optimization — clip-rect culling + double-buffering - HandlePaintLinesBetweenSlices: skip separator lines outside paint clip rect, reducing per-paint iteration from O(N) to O(visible) during scroll/partial repaints. Avoids O(N) XML attribute parsing on off-screen slices. - DataTree constructor: enable OptimizedDoubleBuffer + AllPaintingInWmPaint to eliminate flicker during paint. - Add DataTreeTiming_PaintPerformance test for paint-time regression detection. - Update render-testing skill with test coverage assessment step. - Update TIMING_ENHANCEMENTS.md with enhancements 3 and 4. perf: layout path optimization — early-exit guard, construction skip, paint-path skip Enhancement 5: Slice.SetWidthForDataTreeLayout returns early when m_widthHasBeenSetByDataTree && Width == width, avoiding redundant splitter resizing and event handler churn on every layout pass. Enhancement 6: DataTree.OnSizeChanged skips the O(N) splitter adjustment loop during ConstructingSlices. HandleLayout1(fFull=true) handles width synchronization after construction completes. Enhancement 7: HandleLayout1 paint path (fFull=false) skips SetWidthForDataTreeLayout for non-visible slices, reducing width-sync work from O(N) to O(K) per paint where K is visible slice count. Also fixes RemoveDuplicateCustomFields test to match Test.fwlayout Normal layout changes from prior commit. perf: XML attribute caching + MakeSliceVisible high-water mark Enhancement 8: Cache header/skipSpacerLine/sameObject XML attributes on Slice as lazy bool? fields. Eliminates per-paint XML parsing in HandlePaintLinesBetweenSlices (~2,700 XML lookups/sec at 60Hz with 15 visible slices reduced to 0 after first access). Enhancement 9: Change MakeSliceVisible from static to instance method. Pass known index from HandleLayout1 caller (avoids O(N) IndexOf). Track m_lastVisibleHighWaterMark so the inner 'make preceding visible' loop amortizes to O(N) total instead of O(N*V). Preserves the .NET Framework bug workaround (LT-7307). Reset mark in CreateSlices. perf: construction-path guards + optimization regression tests Enhancement 10: Skip ResetTabIndices in RemoveSlice during ConstructingSlices. Mirror of Enhancement 1 — the finally block in CreateSlices already does a single ResetTabIndices(0) after all removals/insertions complete. Enhancement 11: Skip AdjustSliceSplitPosition in InstallSlice during ConstructingSlices. HandleLayout1(fFull=true) sets correct widths + splitter positions after construction, making per-insert adjustment redundant. Added 7 optimization regression tests (DataTreeOpt_ prefix): - WidthStabilityAfterLayout (Enhancement 5) - AllViewportSlicesVisible (Enhancement 9) - XmlCacheConsistency (Enhancement 8) - XmlCacheInvalidationOnConfigChange (Enhancement 8) - SequentialPaintsProduceIdenticalOutput (Enhancements 3,4,7,8) - SlicePositionsMonotonicallyIncreasing (Enhancement 3) - IsHeaderNodeDelegatesToCachedProperty (Enhancement 8) All 12 DataTree tests pass (5 timing + 7 optimization). # This is the commit message #2: feat: Enhancement 12 — binary search in HandleLayout1 paint path Add FindFirstPotentiallyVisibleSlice binary search to skip above-viewport slices during paint-path layout. Addresses JohnT's original TODO in OnPaint. - HandleLayout1 now starts iteration at the first potentially visible slice (with -1 safety margin) instead of index 0 on paint-path calls - Timing improvements: shallow -71%, deep -20%, extreme -7%, paint -11% - 5 binary-search risk-reduction tests added (DataTreeOpt_ prefix) - Regenerate snapshot baselines after layout/parts changes from 64ab221 - Add mandatory snapshot regression check (Step 6) to render-testing skill - Update TIMING_ENHANCEMENTS.md with E12 measurements and devil's advocate # This is the commit message #3: The real optimization - HWND virtualization # This is the commit message #4: fix: correct VwDrawRootBuffered coordinate mapping + trailing separator bars Three rendering quality fixes for the snapshot capture pipeline: 1. VwDrawRootBuffered coordinate fix (CompositeViewCapture.cs): GetCoordRects returns rcDst in rootSite-local coordinates (origin at HorizMargin, 0). The overlay code was passing clientRect with the rootSite's position relative to the DataTree (e.g. X=175), causing VwDrawRootBuffered to offset rcDst by (-175, -y) and render content at negative X — clipping or hiding it entirely. Fix: render each rootSite into a temp bitmap at local coords (0,0,w,h), then composite into the main bitmap at the correct DataTree-relative position. 2. EnsureAllSlicesInitialized uses BecomeRealInPlace (CompositeViewCapture.cs): Replaced manual handle-forcing with the production ViewSlice initialization path, which sets AllowLayout=true and triggers rootBox.Layout with the correct width. 3. Trailing separator bar removal (VectorReferenceView.cs, PossibilityVectorReferenceView.cs): DisplayVec called AddSeparatorBar after every item including the last, causing VwSeparatorBox to draw unwanted grey bars after single/final items like 'Main Dictionary'. Added guard to only add separators between items. Baselines regenerated — all 6 scenarios show 40-70% more rendered content. All 23 DataTree tests pass (timing + optimization + snapshots). # This is the commit message #5: Added tests before virtualization # This is the commit message #6: update renders # This is the commit message #7: Hwnd implementation started # This is the commit message #8: upgrade openspec # This is the commit message #9: hwnd implementation
- Add deterministic DataTree render scenarios, benchmark harness, and baselines - Optimize layout and paint paths with culling, guards, caching, and binary search - Fix snapshot compositing and separator rendering regressions - Add timing and regression tests; update TIMING_ENHANCEMENTS.md
- Refresh render baselines and benchmark harness after layout updates - Fix snapshot capture coordinates, initialization, and separator rendering - Harden DataTree with regression tests for resize, navigation, and crash paths - Remove generated test-result artifact and update timing notes
…ass, Reconstruct guard fix Phase 2 cold render optimizations for the Views engine: PATH-R1-FIX: Fix Reconstruct guard bug — Construct() left m_fNeedsReconstruct true (from Init), so the warm Reconstruct guard never fired. Warm render dropped from 98.20ms to 0.01ms (99.99% reduction). PATH-C1: Add 8-entry HFONT LRU cache in VwGraphics to avoid redundant CreateFontIndirect/DeleteObjectFont GDI kernel calls when cycling between the same fonts (common in multi-writing-system text). ~2% cold reduction. PATH-C2: Track current foreground/background colors and background mode in VwGraphics. Skip redundant SetTextColor/SetBkColor/SetBkMode GDI calls when values haven't changed. <1% cold reduction. PATH-N1: NFC normalization bypass — OffsetInNfc/OffsetToOrig were performing COM Fetch + ICU NFC normalization on every call (12+ times per FindBreakPoint). Added NFC flag from CallScriptItemize; for NFC text (common case), return identity offsets instantly. 8.2% cold reduction. Best on multi-run scenarios: mixed-styles -27%, lex-deep -25%, lex-extreme -24%. Overall cold render: 60.36ms -> 54.32ms (-10.0% average). All 15 benchmark scenarios pass with 0% pixel variance.
johnml1135
force-pushed
the
001-render-speedup
branch
from
04900b0 to
d6741a0
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Render Performance Optimizations
Summary
This branch delivers comprehensive performance optimizations to the FieldWorks Views engine rendering pipeline, reducing warm render time by 99.99% (153ms → 0.01ms) and cold render time by ~10% (60.36ms → 54.32ms) across 15 benchmark scenarios.
All 15 scenarios pass with 0% pixel variance — no visual regressions.
Motivation
FieldWorks UI responsiveness is bottlenecked by the Views engine render pipeline. Switching between entries, refreshing displays, or navigating within editors triggers
Reconstruct()→Layout()→DrawTheRoot()cycles. Profiling revealed:RefreshDisplay()called Reconstruct on ALL root sites, even when no data changedBenchmark Infrastructure
New test infrastructure
RenderBenchmarkHarness— creates real VwRootBox + VwGraphics in an offscreen HWND, measures CreateView/MakeRoot/PerformOffscreenLayout/DrawTheRoot independentlyRenderTimingSuiteTests— 15 scenarios covering simple through extreme content (long-prose: 4 sections × 80 verses; lex-extreme: 25 entries × 10 senses)RenderVerifyTests— pixel-perfect snapshot verification using Verify + ImageSharpRenderBenchmarkResults/RenderBenchmarkReportWriter— generates markdown summary reports with per-scenario timingDataTreeRenderTests— DataTree-level render scenarios (collapsed/expanded/deep/extreme/multiws/simple)Test files
Src/Common/RootSite/RootSiteTests/RenderBenchmarkHarness.csSrc/Common/RootSite/RootSiteTests/RenderTimingSuiteTests.csSrc/Common/RootSite/RootSiteTests/RenderVerifyTests.csSrc/Common/RootSite/RootSiteTests/RenderBenchmarkTestsBase.csSrc/Common/RenderVerification/(supporting infrastructure)Src/Common/Controls/DetailControlsTests/DataTreeRenderTests.csPhase 1: Warm Render Optimizations (153ms → 0.01ms)
PATH-L1: Width-Invariant Layout Guard
Src/views/VwRootBox.cpp,VwRootBox.hm_fNeedsLayout+m_dxLastLayoutWidthtoVwRootBox::Layout(). If box tree hasn't changed and width is the same, skip the entire layout traversal.PATH-L4: Harness GDI Caching
Src/Common/RootSite/RootSiteTests/RenderBenchmarkHarness.csPerformOffscreenLayout()instead of allocating per call.PATH-R1: Reconstruct Guard
Src/views/VwRootBox.cpp,VwRootBox.hm_fNeedsReconstructflag.Reconstruct()returns early if noPropChanged, style change, or overlay change has occurred since last construction.PATH-L5: Skip Reconstruct When Data Unchanged
Src/views/Views.idh,Src/Common/SimpleRootSite/SimpleRootSite.csNeedsReconstructproperty via COM.SimpleRootSite.RefreshDisplay()checks this before performing expensive reconstruct cycle.Phase 2: Cold Render Optimizations (60.36ms → 54.32ms)
PATH-R1-FIX: Construct Guard Bug Fix
Src/views/VwRootBox.cppConstruct()was leavingm_fNeedsReconstruct = true(set during Init), so the PATH-R1 guard never fired on the first warm render. Fixed by settingm_fNeedsReconstruct = falseafter successful construction.PATH-C1: HFONT LRU Cache
Src/views/lib/VwGraphics.h,VwGraphics.cpp(LgCharRenderProps → HFONT)in VwGraphics. On font change, check cache viamemcmpof font-specific fields. Cache hit → reuse HFONT (skipCreateFontIndirect). Cache miss → create and add to cache. All cached HFONTs freed on destructor/ReleaseDC.PATH-C2: Color State Caching
Src/views/lib/VwGraphics.h,VwGraphics.cppPATH-N1: NFC Normalization Bypass
Src/views/lib/UniscribeSegment.h,UniscribeSegment.cpp,UniscribeEngine.cppOffsetInNfc()andOffsetToOrig()performed COM Fetch + ICU NFC normalization on every call — 12+ times perFindBreakPointinvocation. For content-heavy views, this meant millions of redundant normalizations. Added NFC detection inCallScriptItemize()(comparing pre/post normalization length). When text is already NFC (common case — FieldWorks stores NFC), new overloads return identity offsets, completely skipping COM and ICU calls.mixed-styles-27.1%,lex-deep-24.9%,lex-extreme-24.3%Bug Fixes
VwPropertyStore Crash Fix
Src/views/VwPropertyStore.cppAssert(m_chrp.ws)andAssertPtr(qws)at line 1336 fired whenws=0reached VwPropertyStore during PropChanged-driven box tree rebuilds. Fixed by guardingget_EngineOrNullwithif (m_chrp.ws)check and usingAssertPtrN(qws)to allow null engine pointer. The code already defaulted to LTR directionality when qws was null.VwDrawRootBuffered Coordinate Fixes
Src/views/VwRootBox.cppVwDrawRootBuffered.Other Fixes
Final Benchmark Results
Files Changed (Key Production Files)
Src/views/VwRootBox.hSrc/views/VwRootBox.cppSrc/views/Views.idhSrc/views/VwPropertyStore.cppSrc/views/lib/VwGraphics.hSrc/views/lib/VwGraphics.cppSrc/views/lib/UniscribeSegment.hSrc/views/lib/UniscribeSegment.cppSrc/views/lib/UniscribeEngine.cppSrc/Common/SimpleRootSite/SimpleRootSite.csSrc/Common/Controls/DetailControls/DataTree.csSrc/Common/Controls/DetailControls/Slice.csFuture Optimization Paths
The remaining opportunity for significant cold render improvement is PATH-V1: Lazy VwLazyBox Construction — deferring off-screen paragraph construction until scrolled into view. This is the only path that can achieve the 50% cold reduction target, as the engine is fundamentally bottlenecked by
ScriptShape/ScriptPlacekernel calls that cannot be made faster, only avoided.See layout-optimization-paths.md for full analysis.
This change is