Improve TypedBinding performance by ~29% by StephaneDelcroix · Pull Request #33656 · dotnet/maui

StephaneDelcroix · 2026-01-21T20:36:05Z

Note

Are you waiting for the changes in this PR to be merged?
It would be very helpful if you could test the resulting artifacts from this PR and let us know in a comment if this change resolves your issue. Thank you!

Description

This PR improves TypedBinding and SourceGeneratedBinding performance for value propagation scenarios (binding set once, values updated many times).

Optimizations Applied

Cache Apply() delegate - Avoids lambda allocation on every property change
Cache binding mode - Avoids GetRealizedMode() call on every Apply
Skip Subscribe() loop - Only subscribes to INPC once per binding lifetime
Cache default value - Avoids repeated GetDefaultValue() calls
Skip TryConvert - When TProperty matches property.ReturnType
Cache isTSource - Type check result (source type doesn't change)

Benchmark Results

Before (net10.0 baseline):

Method	Mean	Ratio	Allocated
SetValue	20.77 ns	1.00	40 B
Binding	52.03 ns	2.50	128 B
TypedBinding	47.47 ns	2.28	128 B
SourceGeneratedBinding	45.45 ns	2.19	128 B

After (this PR):

Method	Mean	Ratio	Allocated
SetValue	20.64 ns	1.00	40 B
Binding	52.00 ns	2.52	128 B
TypedBinding	32.90 ns	1.59	64 B
SourceGeneratedBinding	34.10 ns	1.65	64 B

Summary:

Metric	Before	After	Improvement
TypedBinding	47.47 ns (2.28x)	32.90 ns (1.59x)	31% faster
SourceGeneratedBinding	45.45 ns (2.19x)	34.10 ns (1.65x)	25% faster
Memory per operation	128 B	64 B	50% less

New Benchmark

Adds BindingComparisonBenchmarker to Core.Benchmarks for comparing SetValue, Binding, TypedBinding, and SourceGeneratedBinding performance.

cd src/Core/tests/Benchmarks
dotnet run -c Release -- --filter "*BindingComparison*"

Optimizations applied: - Cache Apply() delegate to avoid lambda allocation on every property change - Cache binding mode after first calculation - Skip Subscribe() loop if already subscribed - Cache default value to avoid repeated GetDefaultValue() calls - Skip TryConvert when TProperty matches property.ReturnType - Cache isTSource type check result - Cache dispatcher reference to avoid cast on every call Also adds BindingComparisonBenchmarker for measuring SetValue vs Binding vs TypedBinding vs SourceGeneratedBinding performance. Results (100 value updates after binding once): - SourceGeneratedBinding: 5.05 µs → 3.65 µs (28% faster, 1.57x of SetValue) - TypedBinding: 5.19 µs → 3.80 µs (27% faster, 1.64x of SetValue) - Memory: 12.5 KB → 6.25 KB (50% reduction)

Copilot

Pull request overview

This PR optimizes TypedBinding and SourceGeneratedBinding performance for value propagation scenarios (where bindings are set once and values are updated many times). The optimizations include caching delegates, binding modes, default values, type checks, and dispatchers to reduce allocations and repeated computations.

Changes:

Added new benchmark BindingComparisonBenchmarker to measure binding performance
Implemented 7 performance optimizations in TypedBinding.cs with cached fields
Reported 27-28% performance improvement and 50% memory reduction based on benchmark results

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
src/Core/tests/Benchmarks/Benchmarks/BindingComparisonBenchmarker.cs	New benchmark comparing SetValue, Binding, TypedBinding, and SourceGeneratedBinding performance across 100 iterations
src/Controls/src/Core/TypedBinding.cs	Performance optimizations including cached mode, default values, type checks, dispatcher, Apply delegate, and skip-convert flag with corresponding state management in Unapply

Copilot · 2026-01-21T20:46:14Z

src/Controls/src/Core/TypedBinding.cs

+			// Use cached type check after first apply (source type doesn't change)
+			var isTSource = _isTSource;
+			if (!_hasDefaultValue)
+			{
+				isTSource = sourceObject is TSource;
+				_isTSource = isTSource;
+				if (!isTSource && sourceObject is not null)
+				{
+					BindingDiagnostics.SendBindingFailure(this, "Binding", $"Mismatch between the specified x:DataType ({typeof(TSource)}) and the current binding context ({sourceObject.GetType()}).");
+				}
+			}


The cached type check _isTSource is never reset in Unapply() (line 292-296 resets other cached values but not _isTSource). This causes incorrect behavior when ApplyCore is called with a different source object.

For example, line 520 calls ApplyCore(null, ...) directly without calling Unapply() first. In this case, the condition if (!_hasDefaultValue) on line 311 evaluates to false (since default value was already cached), so the type check is skipped entirely and the stale _isTSource value is used. This will cause incorrect behavior when the source changes from a valid TSource object to null.

Copilot · 2026-01-21T20:46:14Z

src/Controls/src/Core/TypedBinding.cs

+				// Cache dispatcher to avoid cast + property access on every call
+				if (!_dispatcherCached)
+				{
+					_cachedDispatcher = (sender as BindableObject)?.Dispatcher;
+					_dispatcherCached = true;
+				}
+
+				_cachedDispatcher.DispatchIfRequired(_applyAction);


The dispatcher caching implementation is conceptually incorrect. The dispatcher should come from the binding target (as seen in BindingExpression.cs line 731 where it uses the target's dispatcher), but this code caches it from the source object.

In typical binding scenarios, the source is a view model (POCO with INotifyPropertyChanged) and the target is a BindableObject (UI element). Since the source is usually not a BindableObject, the cached dispatcher will be null in most cases. While DispatchIfRequired handles null via EnsureDispatcher, this defeats the optimization purpose since EnsureDispatcher is called on every property change anyway.

Additionally, when the PropertyChangedProxy is resubscribed to a different source object (via the Part setter at line 547-559), the cached dispatcher is never updated, which could lead to using an incorrect dispatcher if the sources have different dispatchers or BindableObject status.

Copilot · 2026-01-21T20:46:14Z

src/Controls/src/Core/TypedBinding.cs

+			_cachedMode = BindingMode.Default;
+			_hasDefaultValue = false;
+			_cachedDefaultValue = null;
+			_skipConvert = false;


The field _isTSource should be reset in Unapply() along with the other cached values. Add _isTSource = false; to properly reset the cached type check result.

Suggested change

_skipConvert = false;

_skipConvert = false;

_isTSource = false;

- Reset _isTSource in Unapply() to handle BindingContext type changes - Remove ineffective dispatcher caching (sender is usually ViewModel, not BindableObject) - Add clarifying comment about dispatcher null handling

StephaneDelcroix

Regarding the review comments:

Comment 1 & 3 (_isTSource not reset): This was already fixed in commit 7a8b3c2. The Unapply() method now includes _isTSource = false; at line 296.

Comment 2 (Dispatcher caching): Marked as outdated - this was removed in the same commit after identifying that the source is typically a ViewModel (not BindableObject), making the cached dispatcher null in most cases.

All three review comments have been addressed in the current code.

Squashed from 26 commits (dev/srozsival/binding-improvements-worker4): - Remove ForSingleNestingLevel factory method - Add new TypedBinding constructor API: TypedBinding(getter, setter, handlersCount, handlers) - Add IPropertyChangeHandler abstraction for handler strategy pattern - Add LegacyPropertyChangeHandler (backward compat with old Tuple-based API) - Add PropertyChangeHandler (new efficient handler using INotifyPropertyChanged) - Unify TypedBinding and TypedBinding2 designs into single TypedBinding class - Update BindingSourceGen to emit new TypedBinding constructor calls - Update CompiledBindingMarkup source generator for new API - Ban old ForSingleNestingLevel usage in BannedSymbols.txt - Add TypedBinding2UnitTests.cs (~2000 lines of new tests) - Update benchmarks to compare TypedBinding variants Note: Rebased onto net11.0 as a squash to avoid conflicts with intermediate states. net11.0 already has the ~29% ApplyCore perf improvements (PR #33656); this PR adds the new flexible handler API on top. Co-authored-by: Simon Rozsival <simon@rozsival.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

> [!NOTE] > Are you waiting for the changes in this PR to be merged? > It would be very helpful if you could [test the resulting artifacts](https://github.com/dotnet/maui/wiki/Testing-PR-Builds) from this PR and let us know in a comment if this change resolves your issue. Thank you! ## Description This PR improves TypedBinding and SourceGeneratedBinding performance for value propagation scenarios (binding set once, values updated many times). ### Optimizations Applied 1. **Cache Apply() delegate** - Avoids lambda allocation on every property change 2. **Cache binding mode** - Avoids `GetRealizedMode()` call on every Apply 3. **Skip Subscribe() loop** - Only subscribes to INPC once per binding lifetime 4. **Cache default value** - Avoids repeated `GetDefaultValue()` calls 5. **Skip TryConvert** - When `TProperty` matches `property.ReturnType` 6. **Cache isTSource** - Type check result (source type doesn't change) ### Benchmark Results **Before (net10.0 baseline):** | Method | Mean | Ratio | Allocated | |----------------------- |---------:|------:|----------:| | SetValue | 20.77 ns | 1.00 | 40 B | | Binding | 52.03 ns | 2.50 | 128 B | | TypedBinding | 47.47 ns | 2.28 | 128 B | | SourceGeneratedBinding | 45.45 ns | 2.19 | 128 B | **After (this PR):** | Method | Mean | Ratio | Allocated | |----------------------- |---------:|------:|----------:| | SetValue | 20.64 ns | 1.00 | 40 B | | Binding | 52.00 ns | 2.52 | 128 B | | TypedBinding | 32.90 ns | 1.59 | 64 B | | SourceGeneratedBinding | 34.10 ns | 1.65 | 64 B | **Summary:** | Metric | Before | After | Improvement | |--------|--------|-------|-------------| | TypedBinding | 47.47 ns (2.28x) | **32.90 ns (1.59x)** | **31% faster** | | SourceGeneratedBinding | 45.45 ns (2.19x) | **34.10 ns (1.65x)** | **25% faster** | | Memory per operation | 128 B | **64 B** | **50% less** | ### New Benchmark Adds `BindingComparisonBenchmarker` to `Core.Benchmarks` for comparing SetValue, Binding, TypedBinding, and SourceGeneratedBinding performance. ```bash cd src/Core/tests/Benchmarks dotnet run -c Release -- --filter "*BindingComparison*" ```

Copilot AI review requested due to automatic review settings January 21, 2026 20:36

StephaneDelcroix added this to the .NET 11.0-preview1 milestone Jan 21, 2026

StephaneDelcroix requested a review from simonrozsival January 21, 2026 20:36

Copilot started reviewing on behalf of StephaneDelcroix January 21, 2026 20:36 View session

StephaneDelcroix added the legacy-area-perf Startup / Runtime performance label Jan 21, 2026

StephaneDelcroix changed the title ~~Improve TypedBinding performance by ~27%~~ [PERF] Improve TypedBinding performance by ~27% Jan 21, 2026

StephaneDelcroix removed the legacy-area-perf Startup / Runtime performance label Jan 21, 2026

Copilot AI reviewed Jan 21, 2026

View reviewed changes

Address code review feedback

c162781

- Reset _isTSource in Unapply() to handle BindingContext type changes - Remove ineffective dispatcher caching (sender is usually ViewModel, not BindableObject) - Add clarifying comment about dispatcher null handling

StephaneDelcroix changed the title ~~[PERF] Improve TypedBinding performance by ~27%~~ Improve TypedBinding performance by ~29% Jan 21, 2026

This was referenced Jan 22, 2026

The Operation will be canceled. The next steps may not contain expected logs. dotnet/dnceng#3008

Open

[android-arm64] The Operation will be canceled. The next steps may not contain expected logs. dotnet/dnceng#6408

Open

StephaneDelcroix commented Jan 22, 2026

View reviewed changes

StephaneDelcroix changed the base branch from net10.0 to main January 22, 2026 12:20

Improve benchmark: let BDN control iterations, use unique values

96f9958

build-analysis bot mentioned this pull request Jan 22, 2026

slow macOS - "##[error]The job running on agent Azure Pipelines 9 ran longer than the maximum time of 60 minutes." dotnet/dnceng#1883

Open

3 tasks

PureWeen modified the milestones: .NET 11.0-preview1, .NET 11.0-preview2 Feb 3, 2026

simonrozsival approved these changes Feb 12, 2026

View reviewed changes

StephaneDelcroix merged commit acc2476 into main Feb 12, 2026
24 of 28 checks passed

StephaneDelcroix deleted the dev/stdelc/typedbindingimprovements branch February 12, 2026 21:46

simonrozsival mentioned this pull request Feb 13, 2026

[net11.0] Improve TypedBinding performance #32382

Open

This was referenced Feb 13, 2026

[repo-status] Daily Repo Status - February 13, 2026 #34044

Closed

[repo-status] 🌟 Daily Repo Status - February 15, 2026 #34061

Closed

[repo-status] 🌟 Daily Repo Status - February 16, 2026 #34063

Closed

lucastitus mentioned this pull request Mar 11, 2026

[Regression] 10.0.50: Compiled Bindings (TypedBinding) does not resubscribe to intermediary values #34428

Closed

github-actions bot locked and limited conversation to collaborators Mar 15, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Improve TypedBinding performance by ~29%#33656

Improve TypedBinding performance by ~29%#33656
StephaneDelcroix merged 3 commits intomainfrom
dev/stdelc/typedbindingimprovements

StephaneDelcroix commented Jan 21, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Jan 21, 2026

Uh oh!

Copilot AI Jan 21, 2026

Uh oh!

Copilot AI Jan 21, 2026

Uh oh!

StephaneDelcroix left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

	_skipConvert = false;
	_skipConvert = false;
	_isTSource = false;

Conversation

StephaneDelcroix commented Jan 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Optimizations Applied

Benchmark Results

New Benchmark

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Jan 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 21, 2026

Choose a reason for hiding this comment

Uh oh!

StephaneDelcroix left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

StephaneDelcroix commented Jan 21, 2026 •

edited

Loading