mirror of
https://github.com/gui-cs/Terminal.Gui.git
synced 2025-12-26 15:57:56 +01:00
ef263f6dd75c04e3e210fbcf5e4956d131218c8c
10 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
copilot-swe-agent[bot]
|
ef263f6dd7 |
Remove cruft and implement SetDelay keystroke command
- Removed test context detection cruft from examples - Examples now use Application.Create(example: true) directly - Updated ExampleDemoKeyStrokesAttribute to support "SetDelay:nnn" command - Removed DelayMs parameter from attribute - Default delay between keys is 100ms - SetDelay command changes delay for subsequent keys in sequence - Metadata moved after using statements (before code) - All examples cleaned up and building successfully This addresses @tig's feedback for cleaner examples and better delay control. Co-authored-by: tig <585482+tig@users.noreply.github.com> |
||
|
copilot-swe-agent[bot]
|
401db78b45 |
Simplify example infrastructure with Create(example) parameter
- Added bool example parameter to Application.Create() - Added static ObservableCollection<IApplication> Apps for external observers - When example=true, metadata is collected and demo keys are sent when first TopRunnable is modal - Removed ExampleContextInjector complexity - Examples now use Application.Create(example: isExample) - Key injection happens via SessionBegun event monitoring TopRunnable.IsModal - Clean, simple architecture that allows external observers to subscribe to Apps collection This addresses @tig's feedback to simplify the approach. Co-authored-by: tig <585482+tig@users.noreply.github.com> |
||
Tig
|
215d76645b | Tweaks | ||
|
copilot-swe-agent[bot]
|
bb24bf490f |
Remove FakeComponentFactory dependency on Examples namespace
- Removed Terminal.Gui.Examples using statement from FakeComponentFactory - Removed key injection logic from FakeComponentFactory.CreateInput() - Created ExampleContextInjector to handle key injection via Application.Driver.EnqueueKeyEvent - Updated examples to call ExampleContextInjector.SetupAutomaticInjection() before Init - This addresses @tig's comment to use Driver.EnqueueKeyEvent instead of direct FakeInput injection Co-authored-by: tig <585482+tig@users.noreply.github.com> |
||
|
Tig
|
02b6aae484 | Code cleanup | ||
|
copilot-swe-agent[bot]
|
98c853ea2b |
Fix nullable warnings in examples
- Added #nullable enable directive to all updated examples - All examples now build without warnings Co-authored-by: tig <585482+tig@users.noreply.github.com> |
||
|
copilot-swe-agent[bot]
|
cd392456ca |
Phase 2: Update examples with attributes and add test infrastructure
- Updated Example, FluentExample, and RunnableWrapperExample with example attributes - Added support for driver name detection from test context in examples - Created ExampleTests class in UnitTestsParallelizable with tests for: - Example metadata validation - Out-of-process execution - In-process execution - Context serialization - Examples now properly detect and use FakeDriver from test context - Tests pass for metadata validation and serialization Co-authored-by: tig <585482+tig@users.noreply.github.com> |
||
|
Tig
|
a84b2c4896 |
Fixes #4419, #4148, #4408 - Toplevel is GONE - Replaced by Runnable (#4422)
* WIP: Broken
* Got working. Mostly.
* Parllel tests pass
* More progres
* Fixed app tests.
* Mouse
* more progress.
* working on shortcut
* Shortcut accept on ENTER is broken.
* One left...
* More test progress.
* All unit tests pass. Still some issues though.
* tweak
* Fixed Integration Tests
* Fixed UI Catalog
* Tweaking CP to try to find race condition
* Refactor StandardColors and improve ColorPicker logic
Refactored `StandardColors` to use lazy initialization for static fields, improving performance and avoiding static constructor convoy effects. Introduced `NamesValueFactory` and `MapValueFactory` methods for encapsulated initialization logic.
Simplified `GetColorNames` to directly return `_names.Value`. Improved `TryParseColor` by clarifying default value usage and adopting object initializer syntax. Updated `TryNameColor` to use `_argbNameMap.Value`.
Refactored `GetArgb` for better readability. Replaced `MultiStandardColorNameResolver` with `StandardColorsNameResolver` in `ColorPicker`. Commented out `app.Init("Fake")` in `ColorPickerTests` for testing purposes.
Made minor formatting improvements, including updated comments and XML documentation for consistency.
* revert
* Throttle input loop to prevent CPU spinning
Introduce a 20ms delay in the input loop of `InputImpl<TInputRecord>`
to prevent excessive CPU usage when no input is available. Removed
the `DateTime dt = Now();` line and the `while (Peek())` block, which
previously enqueued input records.
This change improves resource management, especially in scenarios
where multiple `ApplicationImpl` instances are created in parallel
tests without calling `Shutdown()`. It prevents thread pool
exhaustion and ensures better performance in such cases.
* Refactor ApplicationImpl to use IDisposable pattern
Implemented the IDisposable pattern in ApplicationImpl to improve resource management. Added `Dispose` and `DisposeCore` methods, and marked the `Shutdown` method as obsolete, encouraging the use of `Dispose` or `using` statements instead. Updated the `IApplication` interface to inherit from IDisposable and added `GetResult` methods for retrieving run session results.
Refactored unit tests to adopt the new lifecycle management approach, replacing legacy `Shutdown` calls with `Dispose` or `using`. Removed fragile and obsolete tests, and re-enabled previously skipped tests after addressing underlying issues.
Updated `FakeApplicationLifecycle` and `SetupFakeApplicationAttribute` to align with the new disposal pattern. Improved documentation and examples to guide users toward modern usage patterns. Maintained backward compatibility for legacy singleton usage.
* Add IDisposable pattern with input loop throttling
- Add IDisposable to IApplication for proper resource cleanup
- Add 20ms throttle to input loop (prevents CPU spinning)
- Add Lazy<T> to StandardColors (eliminates convoy effect)
- Add MainLoopCoordinatorTests suite (5 new tests)
- Add Dispose() calls to all 16 ColorPickerTests
- Mark Application.Shutdown() as [Obsolete]
IApplication now requires Dispose() for cleanup
Performance: 100x CPU reduction, 15x faster disposal, tests complete in <5s
Fixes: Thread leaks, CPU saturation, test hangs in parallel execution
Docs: Updated application.md and newinv2.md with disposal patterns
* Refactor test for input loop throttling clarity
Updated `InputLoop_Throttle_Limits_Poll_Rate` test to improve clarity, reliability, and efficiency:
- Rewrote summary comment to clarify purpose and emphasize the 20ms throttle's role in preventing CPU spinning.
- Replaced `var` with explicit types for better readability.
- Reduced test duration from 1s to 500ms to improve test speed.
- Revised assertions:
- Replaced range-based assertion with upper-bound check to ensure poll count is below 500, avoiding timing sensitivity issues.
- Added assertion to verify the thread ran and was not immediately canceled.
- Added a 2-second timeout to `inputTask.Wait` and verified task completion.
- Improved comments to explain test behavior and reasoning behind changes.
* tweaks
* Fix nullabiltiy stuff.
* runnable fixes
* more nullabe
* More nullability
* warnings gone
* Fixed fluent test failure.
* Refactor ApplicationImpl and update Runnable layout logic
Refactored `ApplicationImpl.Run.cs` for improved readability and
atomicity:
- Combined `if (wasModal)` with `SessionStack?.TryPop` to streamline
logic.
- Simplified restoration of `previousRunnable` by reducing nesting.
- Updated comments for clarity and retained `SetIsModal` call.
Simplified focus-setting logic in `ApplicationImpl.Run.cs` using
pattern matching for `TopRunnableView`.
In `Runnable<TResult>`, added `SetNeedsLayout` after `IsModalChanged`
to ensure layout updates. Removed an unused empty line for cleanup.
Corrected namespace in `GetViewsUnderLocationForRootTests.cs` to
align with test structure.
* Update layout on modal state change
A call to `SetNeedsLayout()` was added to the `OnIsModalChanged`
method in the `Runnable` class. This ensures that the layout
is updated whenever the modal state changes.
* Increase test timeout for inputTask.Wait to 10 seconds
Extended the timeout duration for the `inputTask.Wait` method
from 4 seconds to 10 seconds in `MainLoopCoordinatorTests`.
This change ensures the test has a longer window to complete
under conditions of increased load or slower execution
environments, reducing the likelihood of false test failures.
* Refactor project files and simplify test logic
Removed `<LangVersion>` and `<ImplicitUsings>` properties from
`UnitTests.csproj` and `UnitTests.Parallelizable.csproj` to rely
on default SDK settings and disable implicit global usings.
Simplified the `SizeChanged_Event_Still_Fires_For_Compatibility`
test in `FakeDriverTests` by removing the `screenChangedFired`
variable, its associated event handler, and related assertions.
Also removed obsolete warning suppression directives as they
are no longer needed.
* Reduce UnitTestsParallelizable iterations from 10 to 3
Reduced the number of iterations for the UnitTestsParallelizable
test suite from 10 to 3 to save time and resources while still
exposing concurrency issues. Updated the loop and log messages
to reflect the new iteration count.
* disabled InputLoop_Throttle_Limits_Poll_Rate
* Refactor app lifecycle and improve Runnable API
Refactored `Program.cs` to simplify application lifecycle:
- Modularized app creation, initialization, and disposal.
- Improved result handling and ensured proper resource cleanup.
Re-implemented `Runnable<TResult>` with a cleaner design:
- Retained functionality while improving readability and structure.
- Added XML documentation and followed the Cancellable Work Pattern.
Re-implemented `RunnableWrapper<TView, TResult>`:
- Enabled wrapping any `View` to make it runnable with typed results.
- Added examples and remarks for better developer guidance.
Re-implemented `ViewRunnableExtensions`:
- Provided fluent API for making views runnable with or without results.
- Enhanced documentation with examples for common use cases.
General improvements:
- Enhanced code readability, maintainability, and error handling.
- Replaced redundant code with cleaner, more maintainable versions.
* Modernize codebase for Terminal.Gui and MVVM updates
Refactored `LoginView` to remove redundant `Application.LayoutAndDraw()`
call. Enhanced `LoginViewModel` with new observable properties for
automatic property change notifications. Updated `Message` class to use
nullable generics for improved type safety.
Replaced legacy `Application.Init()` and `Application.Run()` calls with
the modern `IApplication` API across `Program.cs`, `Example.cs`, and
`ReactiveExample`. Ensured proper disposal of `IApplication` instances
to prevent resource leaks.
Updated `TerminalScheduler` to integrate with `IApplication` for
invoking actions and managing timeouts. Added null checks and improved
timeout disposal logic for robustness.
Refactored `ExampleWindow` for better readability and alignment with
modern `Terminal.Gui` conventions. Cleaned up unused imports and
improved code clarity across the codebase.
Updated README.md to reflect the latest `Terminal.Gui` practices,
including examples of the `IApplication` API and automatic UI refresh
handling. Renamed `LoginAction` to `LoginActions` for consistency.
* Refactor: Transition to IRunnable-based architecture
Replaced `Toplevel` with `Window` as the primary top-level UI element. Introduced the `IRunnable` interface to modernize the architecture, enabling greater flexibility and testability. Deprecated the static `Application` class in favor of the instance-based `IApplication` model, which supports multiple application contexts.
Updated methods like `Application.Run()` and `Application.RequestStop()` to use `IRunnable`. Removed or replaced legacy `Modal` properties with `IsModal`. Enhanced the `IApplication` interface with a fluent API, including methods like `Run<TRunnable>()` and `GetResult<T>()`.
Refactored tests and examples to align with the new architecture. Updated documentation to reflect the instance-based model. Deprecated obsolete members and methods, including `Application.Current` and `Application.TopRunnable`.
Improved event handling by replacing the `Accept` event with `Accepting` and using `e.Handled` for event processing. Updated threading examples to use `App?.Invoke()` or `app.Invoke()` for UI updates. Cleaned up redundant code and redefined modal behavior for better consistency.
These changes modernize the `Terminal.Gui` library, improving clarity, usability, and maintainability while ensuring backward compatibility where possible.
* Refactor: Replace Toplevel with Runnable class
This commit introduces a major architectural update to the `Terminal.Gui` library, replacing the legacy `Toplevel` class with the new `Runnable` class. The changes span the entire codebase, including core functionality, tests, documentation, and configuration files.
- **Core Class Replacement**:
- Replaced `Toplevel` with `Runnable` as the base class for modal views and session management.
- Updated all references to `Toplevel` in the codebase, including constructors, methods, and properties.
- **Configuration Updates**:
- Updated `tui-config-schema.json` to reflect the new `Runnable` scheme.
- **New Classes**:
- Added `UICatalogRunnable` for managing the UI Catalog application.
- Introduced `Runnable<TResult>` as a generic base class for blocking sessions with result handling.
- **Documentation and Tests**:
- Updated documentation to emphasize `Runnable` and mark `Toplevel` as obsolete.
- Refactored test cases to use `Runnable` and ensure compatibility.
- **Behavioral Improvements**:
- Enhanced lifecycle management and alignment with the `IRunnable` interface.
- Improved clarity and consistency in naming conventions.
These changes modernize the library, improve flexibility, and provide a clearer architecture for developers.
* Refactor: Consolidate Runnable classes and decouple View from ApplicationImpl
- Made Runnable<TResult> inherit from Runnable (eliminating ~180 LOC duplication)
- Moved View init/layout/cursor logic from ApplicationImpl to Runnable lifecycle events
- ApplicationImpl.Begin now operates purely on IRunnable interface
Related to #4419
* Simplified the disposal logic in `ApplicationImpl.Run.cs` by replacing
the type-specific check for `View` with a more general check for
`IDisposable`. This ensures proper disposal of any `IDisposable`
object, improving robustness.
Removed the `FrameworkOwnedRunnable` property from the `ApplicationImpl`
class in `ApplicationImpl.cs` and the `IApplication` interface in
`IApplication.cs`. This eliminates the need to manage this property,
reducing complexity and improving maintainability.
Updated `application.md` to reflect the removal of the
`FrameworkOwnedRunnable` property, ensuring the documentation aligns
with the updated codebase.
* Replaces the legacy `Shutdown()` method with `Dispose()` to align
with the `IDisposable` pattern, ensuring proper resource cleanup
and simplifying the API. The `Dispose()` method is now the
recommended way to release resources, with `using` statements
encouraged for automatic disposal.
Key changes:
- Marked `Shutdown()` as obsolete; it now internally calls `Dispose()`.
- Updated the fluent API to remove `Shutdown()` from chaining.
- Enhanced session lifecycle management for thread safety.
- Updated tests to validate proper disposal and state reset.
- Improved `IRunnable` integration with automatic disposal for
framework-created runnables.
- Maintained backward compatibility for the legacy static
`Application` singleton.
- Refactored documentation and examples to reflect modern practices
and emphasize `Dispose()` usage.
These changes modernize the `Terminal.Gui` lifecycle, improve
testability, and encourage alignment with .NET conventions.
* Refactor runnable app context handling in ApplicationImpl
Refactor how the application context is set for `runnable` objects
by introducing a new `SetApp` method in the `IRunnable` interface.
This replaces the previous logic of directly setting the `App`
property for `View` objects, making the process more generic and
encapsulated within `IRunnable` implementations.
Simplify `Mouse.UngrabMouse()` by removing the conditional check
and calling it unconditionally.
Make a minor formatting adjustment in the generic constraint of
`Run<TRunnable>` in `ApplicationImpl`.
Add `SetApp(IApplication app)` to the `IRunnable` interface and
implement it in the `Runnable` class to set the `App` property
to the provided application instance.
* Improve docs, tests, and modularity across the codebase
Reorganized and updated `CONTRIBUTING.md`:
- Added **Key Architecture Concepts** section and reordered the table of contents.
- Updated testing requirements to discourage legacy patterns.
- Added instructions for replicating CI workflows locally.
- Clarified PR guidelines and coding style expectations.
Enhanced `README.md` with detailed CI/CD workflow documentation.
Refactored `ColorPicker.Prompt` to use `IApplication` for improved modularity and testability.
Introduced `IApplicationScreenChangedTests` for comprehensive testing of `ScreenChanged` events and `Screen` property.
Refactored `ApplicationScreenTests` and `TextView.PromptForColors` to align with modern patterns.
Updated `Terminal.sln` to include `.github/workflows/README.md`.
Performed general cleanup:
- Removed outdated documentation links.
- Improved XML documentation and coding consistency.
* readme tweaks
* Improve thread safety, layout, and test coverage
Refactored `OutputBufferImpl.cs` to enhance thread safety by locking shared resources and adding bounds checks for columns and rows. Improved handling of wide characters and removed outdated TODO comments.
Updated `Runnable.cs` to call `SetNeedsDraw()` on modal state changes, ensuring proper layout and drawing updates. Simplified layout handling in `ApplicationImpl.Run.cs` by replacing redundant comments with a `LayoutAndDraw()` call.
Added a check in `AllViewsTester.cs` to skip creating instances of `RunnableWrapper` types with unsatisfiable generic constraints, logging a warning when encountered.
Enhanced `ListViewTests.cs` by adding explicit `app.LayoutAndDraw()` calls to validate visual output and ensure tests reflect the updated application state.
These changes improve robustness, prevent race conditions, and ensure consistent behavior across the application.
* Refactor: Rename Toplevel to Runnable and update logic
Updated the `Border` class to use `Command.Quit` instead of
`Command.QuitToplevel` in the `CloseButton.Accept` handler.
Renamed test methods in `GetViewsAtLocationTests.cs` to replace
"Toplevel" with "Runnable" for consistency. Updated `Runnable<bool>`
instances to use "topRunnable" as the `Id` property.
These changes align the codebase with updated naming conventions
and improve clarity.
* Removed `ToplevelTests` and migrated relevant test cases to
`MouseDragTests` with improved structure and coverage. Updated
tests to use `Application.Create`, `app.Begin`, and `app.End`
for better resource management and lifecycle handling.
Replaced direct event handling with `app.Mouse.RaiseMouseEvent`
to align with the application's event-handling mechanism. Added
`Runnable` objects to ensure views are properly initialized and
disposed of within the application context.
Enhanced tests to include assertions for minimum width and
height constraints during resize operations. Removed redundant
tests and streamlined logic to reduce duplication and improve
maintainability.
* Reorged Unit Test namespaces.
* more
* Refactor tests and update namespaces for consistency
Updated namespaces in `ArrangementTests.cs` and `MouseDragTests.cs` for better organization. Enhanced `ArrangementTests.cs` with additional checks for arrangement flags. Reformatted and re-added `MouseDragTests.cs` and `SchemeTests.cs` with modern C# features like nullable annotations and object initializers. Ensured no functional changes while improving code clarity and consistency.
* Fix nullability warnings in MouseDragTests.cs
Updated `app.End` calls to use the null-forgiving operator (`!`)
on `app.SessionStack` to ensure it is treated as non-null.
This change addresses potential nullability warnings and
improves code safety and clarity. Applied consistently across
all relevant test cases in the `MouseDragTests` class.
|
||
|
Copilot
|
b9f55a5a96 |
Fixes #4410, #4413, #4414, #4415 - MessageBox nullable, Clipboard refactor, fence for legacy/modern App, and makes internal classes thread safe. (#4411)
* Initial plan * Change MessageBox to return nullable int instead of -1 Co-authored-by: tig <585482+tig@users.noreply.github.com> * Initial plan * Add fencing to prevent mixing Application models Co-authored-by: tig <585482+tig@users.noreply.github.com> * Fix fence logic to work with parallel tests Co-authored-by: tig <585482+tig@users.noreply.github.com> * WIP: Fixing Application issues. * Refactor error messages into constants Co-authored-by: tig <585482+tig@users.noreply.github.com> * Refactor ConfigurationProperty properties to use static backing fields and raise events Co-authored-by: tig <585482+tig@users.noreply.github.com> * Reset static Application properties in ResetStateStatic Co-authored-by: tig <585482+tig@users.noreply.github.com> * Refactor tests to decouple from global Application state Commented out `driver ??= Application.Driver` assignments in `DriverAssert` to prevent automatic global driver assignment. Removed `Application.ResetState(true)` calls and commented out state validation assertions in `GlobalTestSetup` to reduce dependency on global state. Reintroduced `ApplicationForceDriverTests` and `ApplicationModelFencingTests` to validate `ForceDriver` behavior and ensure proper handling of legacy and modern Application models. Skipped certain `ToAnsiTests` that rely on `Application`. Removed direct `Application.Driver` assignments in `ViewDrawingClippingTests` and `ViewDrawingFlowTests`. Performed general cleanup of redundant code and unused imports to simplify the codebase. * WIP: Fixed Parallel tests; non-Parallel still broken Refactor application model usage tracking Refactored `ApplicationModelUsage` into a public enum in the new `Terminal.Gui.App` namespace, making it accessible across the codebase. Replaced the private `_modelUsage` field in `ApplicationImpl` with a public static `ModelUsage` property to improve clarity and accessibility. Renamed error message constants for consistency and updated methods like `SetInstance` and `MarkInstanceBasedModelUsed` to use the new `ModelUsage` property. Removed the private `ApplicationModelUsage` enum from `ApplicationImpl`. Updated test cases to use `ApplicationImpl.Instance` instead of `Application.Create` to enforce the legacy static model. Skipped obsolete tests in `ApplicationForceDriverTests` and added null checks in `DriverAssert` and `SelectorBase` to handle edge cases. Commented out an unused line in `WindowsOutput` and made general improvements to code readability, maintainability, and consistency. * WIP: Almost there! Refactored tests and code to align with the modern instance-based application model. Key changes include: - Disabled Sixel rendering in `OutputBase.cs` due to dependency on legacy static `Application` object. - Hardcoded `force16Colors` to `false` in `WindowsOutput.cs` with a `BUGBUG` note. - Updated `ApplicationImplTests` to use `ApplicationImpl.SetInstance` and return `ApplicationImpl.Instance`. - Refactored `ApplicationModelFencingTests` to use `Application.Create()` and added `ResetModelUsageTracking()` for model switching. - Removed legacy `DriverTests` and reintroduced updated versions with cross-platform driver tests. - Reverted `ArrangementTests` and `ShortcutTests` to use legacy static `ApplicationImpl.Instance`. - Reintroduced driver tests in `DriverTests.cs` with modern `Application.Create()` and added `TestTop` for driver content verification. - General cleanup, including removal of outdated code and addition of `BUGBUG` notes for temporary workarounds. * Fixed all modelusage bugs? Replaced static `Application` references with instance-based `App` context across the codebase. Updated calls to `Application.RequestStop()` and `Application.Screen` to use `App?.RequestStop()` and `App?.Screen` for better encapsulation and flexibility. Refactored test infrastructure to align with the new context, including reintroducing `FakeApplicationFactory` and `FakeApplicationLifecycle` for testing purposes. Improved logging, error handling, and test clarity by adding `logWriter` support and simplifying test setup. Removed redundant or obsolete code, such as `NetSequences` and the old `FakeApplicationFactory` implementation. Updated documentation to reflect the new `IApplication.RequestStop()` usage. * merged * Refactor KeyboardImpl and modernize MessageBoxTests Refactored the `KeyboardImpl` class to remove hardcoded default key values, replacing them with uninitialized fields for dynamic configuration. Updated key binding logic to use `ReplaceCommands` instead of `Add` for better handling of dynamic changes. Removed unnecessary `KeyBindings.Clear()` calls to avoid side effects. Rewrote `MessageBoxTests.cs` to improve readability, maintainability, and adherence to modern C# standards. Enabled nullable reference checks, updated the namespace, and restructured test methods for clarity. Marked non-functional tests with `[Theory(Skip)]` and improved test organization with parameterized inputs. Enhanced test assertions, lifecycle handling, and error handling across the test suite. Updated `UICatalog_AboutBox` to use multiline string literals for expected outputs. These changes improve the overall maintainability and flexibility of the codebase. * Atempt to fix windows only CI/CD Unit tests failure Refactor Application lifecycle and test cleanup Refactored the `Application` class to phase out legacy static properties `SessionStack` and `TopRunnable` from `Application.Current.cs`. These were reintroduced in a new file `Application.TopRunnable.cs` for better modularity, while retaining their `[Obsolete]` status. Updated `ApplicationPopoverTests.cs` to replace `Application.ResetState(true)` with `Application.Shutdown()` for consistent application state cleanup. Added explicit cleanup for `Application.TopRunnable` in relevant test cases to ensure proper resource management. Adjusted namespaces and `using` directives to support the new structure. These changes improve code organization and align with updated application lifecycle management practices. * Fixes #<Issue> - Dispose TopRunnable in cleanup logic Updated the `finally` block in `ApplicationPopoverTests` to dispose of the `Application.TopRunnable` object if it is not null, ensuring proper resource cleanup. Previously, the property was being set to `null` without disposal. The `Application.Shutdown()` call remains unchanged. * Improve thread safety, reduce static dependencies, and align the codebase with the updated `IApplication` interface. Refactored the `MainThreadId` property to improve encapsulation: - Updated `Application.MainThreadId` to use `ApplicationImpl.Instance` directly. - Added `MainThreadId` to `ApplicationImpl` and `IApplication`. - Removed redundant `MainThreadId` from `ApplicationImpl.Run.cs`. Updated `EnqueueMouseEvent` to include an `IApplication?` parameter: - Modified `FakeInputProcessor`, `InputProcessorImpl`, and `WindowsInputProcessor` to support the new parameter. - Updated `IInputProcessor` interface to reflect the new method signature. - Adjusted `GuiTestContext` and `EnqueueMouseEventTests` to pass `IApplication` where required. Improved test coverage and code maintainability: - Added test cases for negative positions and empty mouse events. - Commented out legacy code in `GraphView` and `FakeDriverBase`. - Enhanced readability in `EnqueueMouseEventTests`. These changes improve thread safety, reduce static dependencies, and align the codebase with the updated `IApplication` interface. * Fixed more bugs. Enabled nullable reference types across multiple files to improve code safety. Refactored and modularized test classes, improving readability and maintainability. Removed outdated test cases and added new tests for edge cases, including culture-specific and non-Gregorian calendar handling. Addressed timeout issues in `ScenarioTests` with a watchdog timer and improved error handling. Updated `ApplicationImplTests` to use instance fields instead of static references for better test isolation. Refactored `ScenarioTests` to dynamically load and test all UI Catalog scenarios, with macOS-specific skips for known issues. Aligned `MessageBox.Query` calls with updated API signatures. Performed general code cleanup, including removing unused directives, improving formatting, and consolidating repetitive logic into helper methods. * Made the `InputBindings<TEvent, TBinding>` class thread-safe by replacing the internal `Dictionary<TEvent, TBinding>` with `ConcurrentDictionary<TEvent, TBinding>`. This fixes parallel test failures where "Collection was modified; enumeration operation may not execute" exceptions were thrown. ## Changes Made ### 1. InputBindings.cs - **File**: `Terminal.Gui/Input/InputBindings.cs` - **Change**: Replaced `Dictionary` with `ConcurrentDictionary` - **Key modifications**: - Changed `_bindings` from `Dictionary<TEvent, TBinding>` to `ConcurrentDictionary<TEvent, TBinding>` - Updated `Add()` methods to use `TryAdd()` instead of checking with `TryGet()` then `Add()` - Updated `Remove()` to use `TryRemove()` (no need to check existence first) - Updated `ReplaceCommands()` to use `ContainsKey()` instead of `TryGet()` - Added `.ToList()` to `GetAllFromCommands()` to create a snapshot for safe enumeration - Added comment explaining that `ConcurrentDictionary` provides snapshot enumeration in `GetBindings()` - Added `.ToArray()` to `Clear(Command[])` to create snapshot before iteration ### 2. Thread Safety Test Suite - **File**: `Tests/UnitTestsParallelizable/Input/InputBindingsThreadSafetyTests.cs` - **New file** with comprehensive thread safety tests: - `Add_ConcurrentAccess_NoExceptions` - Tests concurrent additions - `GetBindings_DuringConcurrentModification_NoExceptions` - Tests enumeration during modifications - `TryGet_ConcurrentAccess_ReturnsConsistentResults` - Tests concurrent reads - `Clear_ConcurrentAccess_NoExceptions` - Tests concurrent clearing - `Remove_ConcurrentAccess_NoExceptions` - Tests concurrent removals - `Replace_ConcurrentAccess_NoExceptions` - Tests concurrent replacements - `GetAllFromCommands_DuringModification_NoExceptions` - Tests LINQ queries during modifications - `MixedOperations_ConcurrentAccess_NoExceptions` - Tests mixed operations (add/read/remove) - `KeyBindings_ConcurrentAccess_NoExceptions` - Tests actual `KeyBindings` class - `MouseBindings_ConcurrentAccess_NoExceptions` - Tests actual `MouseBindings` class ## Benefits of ConcurrentDictionary Approach 1. **Lock-Free Reads**: Most read operations don't require locks, improving performance 2. **Snapshot Enumeration**: Built-in support for safe enumeration during concurrent modifications 3. **Simplified Code**: No need for explicit `lock` statements or lock objects 4. **Better Scalability**: Multiple threads can read/write simultaneously 5. **No "Collection was modified" Exceptions**: Enumeration creates a snapshot ## Performance Characteristics - **Read Operations**: Lock-free, very fast - **Write Operations**: Uses fine-grained locking internally, minimal contention - **Memory Overhead**: Slightly higher than `Dictionary` but negligible in practice - **Enumeration**: Creates a snapshot, safe for concurrent modifications ## Test Results - **Original failing test now passes**: `ApplicationImplTests.Init_CreatesKeybindings` - **10 new thread safety tests**: All passing - **All 11,741 parallelizable tests**: All passing (11,731 passed, 10 skipped) - **All 1,779 non-parallelizable tests**: All passing (1,762 passed, 17 skipped) - **No compilation errors**: Clean build with no xUnit1031 warnings (suppressed with pragmas) ## Verification The original failure was: ``` System.InvalidOperationException: Collection was modified; enumeration operation may not execute. ``` This occurred in parallelizable tests when multiple threads accessed `KeyBindings.GetBindings()` simultaneously. The `ConcurrentDictionary` implementation resolves this by providing thread-safe operations and snapshot enumeration. ## Notes - The xUnit1031 warnings about using `Task.WaitAll` instead of `async/await` have been suppressed with `#pragma warning disable xUnit1031` directives, as these are intentional blocking operations in stress tests that test concurrent scenarios - All existing functionality is preserved; this is a drop-in replacement - No changes to public API surface - Existing tests continue to pass * Make InputBindings and KeyboardImpl thread-safe for concurrent access Replace Dictionary with ConcurrentDictionary in InputBindings<TEvent, TBinding> and KeyboardImpl to enable safe parallel test execution and multi-threaded usage. Changes: - InputBindings: Replace Dictionary with ConcurrentDictionary for _bindings - InputBindings: Make Replace() atomic using AddOrUpdate instead of Remove+Add - InputBindings: Make ReplaceCommands() atomic using AddOrUpdate - InputBindings: Add IsValid() check to both Add() overloads - InputBindings: Add defensive .ToList()/.ToArray() for safe LINQ enumeration - KeyboardImpl: Replace Dictionary with ConcurrentDictionary for _commandImplementations - KeyboardImpl: Change AddKeyBindings() to use ReplaceCommands for idempotent initialization - Add 10 comprehensive thread safety tests for InputBindings - Add 9 comprehensive thread safety tests for KeyboardImpl The ConcurrentDictionary implementation provides: - Lock-free reads for better performance under concurrent access - Atomic operations for Replace/ReplaceCommands preventing race conditions - Snapshot enumeration preventing "Collection was modified" exceptions - No breaking API changes - maintains backward compatibility All 11,750 parallelizable tests pass (11,740 passed, 10 skipped). Fixes race conditions that caused ApplicationImplTests.Init_CreatesKeybindings to fail intermittently during parallel test execution. * Decouple ApplicationImpl from Application static props Removed initialization of `Force16Colors` and `ForceDriver` from `Application` static properties in the `ApplicationImpl` constructor. The class still subscribes to the `Force16ColorsChanged` and `ForceDriverChanged` events, but no longer sets initial values for these properties. This change simplifies the constructor and reduces coupling between `ApplicationImpl` and `Application`. * Refactored keyboard initialization in `ApplicationImpl` to use `Application` static properties for default key assignments, ensuring synchronization with pre-`Init()` changes. Improved `KeyboardImpl` initialization to avoid premature `ApplicationImpl.Instance` access, enhancing testability. Standardized constant naming conventions and improved code readability in thread safety tests for `KeyboardImpl` and `InputBindings`. Updated `TestInputBindings` implementation for clarity and conciseness. Applied consistent code style improvements across files, including spacing, formatting, and variable naming, to enhance maintainability and readability. * Fix race conditions in parallel tests - thread-safe ApplicationImpl and KeyboardImpl Fixes intermittent failures in parallel tests caused by three separate race conditions: 1. **KeyboardImpl constructor race condition** - Constructor was accessing Application.QuitKey/ArrangeKey/etc which triggered ApplicationImpl.Instance getter, setting ModelUsage=LegacyStatic before Application.Create() was called - Changed constructor to initialize keys with hard-coded defaults instead - Added synchronization from Application static properties during Init() 2. **InputBindings.Replace() race condition** - Between GetOrAdd(oldEventArgs) and AddOrUpdate(newEventArgs), another thread could modify bindings, causing stale data to overwrite valid bindings - Added early return for same-key case (oldEventArgs == newEventArgs) - Kept atomic operations with proper updateValueFactory handling - Added detailed thread-safety documentation 3. **ApplicationImpl model usage fence checks race condition** - Two threads calling Init() simultaneously could both pass fence checks before either set ModelUsage, allowing improper model mixing - Added _modelUsageLock for thread-safe synchronization - Made all ModelUsage operations atomic (Instance getter, SetInstance, MarkInstanceBasedModelUsed, ResetModelUsageTracking, Init fence checks) **Files Changed:** - Terminal.Gui/App/ApplicationImpl.cs - Added _modelUsageLock, made all ModelUsage access thread-safe - Terminal.Gui/App/ApplicationImpl.Lifecycle.cs - Thread-safe fence checks in Init(), sync keyboard keys from Application properties - Terminal.Gui/App/Keyboard/KeyboardImpl.cs - Fixed constructor to not trigger ApplicationImpl.Instance - Terminal.Gui/Input/InputBindings.cs - Fixed Replace() race condition with proper atomic operations **Testing:** - All 11 ApplicationImplTests pass - All 9 KeyboardImplThreadSafetyTests pass - All 10 InputBindingsThreadSafetyTests pass - No more intermittent "Cannot use modern instance-based model after using legacy static Application model" errors in parallel test execution The root cause was KeyboardImpl constructor accessing Application static properties during object creation, which would lazily initialize ApplicationImpl.Instance and set the wrong ModelUsage before Application.Create() could mark it as InstanceBased. * Warning cleanup * docs: Add comprehensive MessageBox and Clipboard API documentation - Updated MessageBox class docs with nullable return value explanation - Created docfx/docs/messagebox-clipboard-changes-v2.md migration guide - Updated migratingfromv1.md with quick links to major changes - Created PR-SUMMARY.md documenting all changes - Added examples for both instance-based and legacy patterns - Documented application model fencing and thread safety improvements The documentation covers: • MessageBox nullable int? returns (null = cancelled) • Clipboard refactoring from static to instance-based • Application model usage fencing to prevent pattern mixing • Thread safety improvements in KeyboardImpl and InputBindings • Complete migration guide with code examples • Benefits and rationale for all changes * Refactor static properties to use backing fields Refactored static properties in multiple classes (`Button`, `CheckBox`, `Dialog`, `FrameView`, `MessageBox`, `StatusBar`, and `Window`) to use private backing fields for better encapsulation and configurability. Default values are now stored in private static fields, allowing overrides via configuration files (e.g., `Resources/config.json`). Updated property definitions to use `get`/`set` accessors interacting with the backing fields. Retained the `[ConfigurationProperty]` attribute to ensure runtime configurability. Removed redundant code, improved XML documentation, adjusted namespace declarations for consistency, and performed general code cleanup to enhance readability and maintainability. * Fix Windows-only parallel test failure by preventing ConfigurationManager from triggering ApplicationImpl.Instance Problem: `MessageBoxTests.Location_And_Size_Correct` was failing only on Windows in parallel tests with: System.InvalidOperationException: Cannot use modern instance-based model (Application.Create) after using legacy static Application model (Application.Init/ApplicationImpl.Instance). Root Cause (maybe): View classes (MessageBox, Dialog, Window, Button, CheckBox, FrameView, StatusBar) had `[ConfigurationProperty]` decorated auto-properties with inline initializers. When ConfigurationManager's module initializer scanned assemblies using reflection, accessing these auto-properties could trigger lazy initialization of other static members, which in some cases indirectly referenced `ApplicationImpl.Instance`, marking the model as "legacy" before parallel tests called `Application.Create()`. Solution: Converted all `[ConfigurationProperty]` auto-properties in View classes to use private backing fields with explicit getters/setters, matching the pattern used by `Application.QuitKey`. This prevents any code execution during reflection-based property discovery. Files Changed: - Terminal.Gui/Views/MessageBox.cs - 4 properties converted - Terminal.Gui/Views/Dialog.cs - 6 properties converted - Terminal.Gui/Views/Window.cs - 2 properties converted - Terminal.Gui/Views/Button.cs - 2 properties converted - Terminal.Gui/Views/CheckBox.cs - 1 property converted - Terminal.Gui/Views/FrameView.cs - 1 property converted - Terminal.Gui/Views/StatusBar.cs - 1 property converted Test Reorganization: - Moved `ConfigurationManagerTests.GetConfigPropertiesByScope_Gets` from UnitTestsParallelizable to UnitTests (defines custom ConfigurationProperty which affects global state) - Moved `SourcesManagerTests.Sources_StaysConsistentWhenUpdateFails` from UnitTestsParallelizable to UnitTests (modifies static ConfigurationManager.ThrowOnJsonErrors property) Best Practice: All `[ConfigurationProperty]` decorated static properties should use private backing fields to avoid triggering lazy initialization during ConfigurationManager's module initialization. Fixes: Windows-only parallel test failure in MessageBoxTests * Add thread-safety to CollectionNavigator classes - Add lock-based synchronization to CollectionNavigatorBase for _searchString and _lastKeystroke fields - Add lock-based synchronization to CollectionNavigator for Collection property access - Protect ElementAt and GetCollectionLength methods with locks - Add 6 comprehensive thread-safety tests covering: - Concurrent SearchString access - Concurrent Collection property access - Concurrent navigation operations (50 parallel tasks) - Concurrent collection modification with readers/writers - Concurrent search string changes - Stress test with 100 tasks × 1000 operations each All tests pass (31/31) including new thread-safety tests. The implementation uses lock-based synchronization rather than concurrent collections because: - IList interface is not thread-safe by design - CollectionNavigator is internal and used by UI components (ListView/TreeView) - Matches existing Terminal.Gui patterns (Scope<T>, ConfigProperty) - Provides simpler and more predictable behavior Fixes thread-safety issues when CollectionNavigator is accessed from multiple threads. * cleanup * Run parallel unit tests 10 times with varying parallelization to expose concurrency issues Co-authored-by: tig <585482+tig@users.noreply.github.com> * Fix parallel unit tests workflow - use proper xUnit parallelization parameters Co-authored-by: tig <585482+tig@users.noreply.github.com> * Fix environment variable reference in workflow - use proper bash syntax Co-authored-by: tig <585482+tig@users.noreply.github.com> * Run parallel tests 10 times sequentially instead of matrix expansion Co-authored-by: tig <585482+tig@users.noreply.github.com> * Make ConfigurationManager thread-safe - use ConcurrentDictionary and add locks Co-authored-by: tig <585482+tig@users.noreply.github.com> * Add Debug.Fail to detect legacy Application usage in parallelizable tests Co-authored-by: tig <585482+tig@users.noreply.github.com> * Move ScrollSliderTests to UnitTests project - they access legacy Application model Co-authored-by: tig <585482+tig@users.noreply.github.com> * Revert ScrollSliderTests move and document root cause analysis Co-authored-by: tig <585482+tig@users.noreply.github.com> * Remove Debug.Fail and move ScrollSliderTests to UnitTests project Co-authored-by: tig <585482+tig@users.noreply.github.com> * Re-add Debug.Fail to detect legacy Application usage in parallelizable tests Co-authored-by: tig <585482+tig@users.noreply.github.com> * Refactor tests and improve parallelization support Commented out `Debug.Fail` statements in `Application.Lifecycle.cs` and `ApplicationImpl.cs` to prevent interruptions during parallel tests. Refactored `ToString` in `ApplicationImpl.cs` to use an expression-bodied member and removed unused imports. Rewrote tests in `ClipRegionTests.cs` and `ScrollSliderTests.cs` to remove global state dependencies and migrated them to the `UnitTests_Parallelizable` namespace. Enabled nullable annotations and updated assertions for clarity and modern patterns. Improved test coverage by adding scenarios for clamping, layout, and size calculations. Updated `README.md` to include `[SetupFakeApplication]` in the list of patterns that block parallelization and clarified migration guidelines. Replaced `[SetupFakeDriver]` with `[SetupFakeApplication]` in examples. Added `<Folder Include="Drivers\" />` to `UnitTests.csproj` for better organization. Adjusted test project references to reflect test migration. Enhanced test output validation in `ScrollSliderTests.cs`. Removed redundant test cases and improved documentation to align with modern C# practices and ensure maintainability. * marked as a "TODO" for potential future configurability. --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: tig <585482+tig@users.noreply.github.com> Co-authored-by: Tig <tig@users.noreply.github.com> |
||
|
Copilot
|
e199063a31 |
Introduce IRunnable interface architecture with Fluent API (Phase 1) (#4405)
* Initial plan * Add IRunnable interface, Runnable base class, and RunnableSessionToken Co-authored-by: tig <585482+tig@users.noreply.github.com> * Add comprehensive parallelizable unit tests for IRunnable Co-authored-by: tig <585482+tig@users.noreply.github.com> * Add 41 more unit tests for comprehensive IRunnable coverage - Added ApplicationRunnableIntegrationTests with 29 integration tests covering Begin/End/Run lifecycle - Added RunnableEdgeCasesTests with 24 edge case and error condition tests - Tests cover event propagation, cancellation scenarios, nested runnables, result handling - Fixed App property not being set in Begin() method - Total test count increased from 23 to 64 tests for IRunnable functionality Co-authored-by: tig <585482+tig@users.noreply.github.com> * Fix parallel test failures in CI/CD - Fixed IsModal property to check RunnableSessionStack instead of just TopRunnable - Added "fake" driver parameter to Application.Init() in integration tests - Fixed Begin() to capture old IsModal value before pushing to stack - Moved App property assignment before stack operations to ensure proper state - Skipped 2 tests that use Run() with main loop (not suitable for parallel tests) - All 11,654 parallelizable tests now pass (4 skipped) Co-authored-by: tig <585482+tig@users.noreply.github.com> * Refactor Application with IRunnable and session management Modernized `Application` and `ApplicationImpl` by introducing `IRunnable` and `RunnableSessionToken` for improved session management. Deprecated legacy methods and added `Obsolete` attributes to indicate their removal. Simplified method bodies using expression-bodied members and null-coalescing assignments. Enhanced lifecycle management in `ApplicationImpl` by removing redundant code and improving `SessionStack` iteration. Introduced `IToplevelTransitionManager` to handle top-level state changes. Updated `Runnable<TResult>` to implement `IRunnable<TResult>` with lifecycle event handling for `IsRunning` and `IsModal` states. Improved result management during lifecycle transitions. Removed legacy classes like `SessionToken` and consolidated their functionality into the new constructs. Updated and expanded the test suite to cover `IRunnable` lifecycle events, `RunnableSessionToken` behavior, and integration with `Application`. Performed code cleanup, improved readability, and updated documentation with detailed remarks and examples. Added new unit tests for edge cases and lifecycle behavior. * Implement fluent API for Init/Run/Shutdown with automatic disposal - Changed Init() to return IApplication for fluent chaining - Changed Run<TRunnable>() to return IApplication (breaking change from TRunnable) - Changed Shutdown() to return object? (extracts and returns result from last Run<T>()) - Added FrameworkOwnedRunnable property to track runnable created by Run<T>() - Shutdown() automatically disposes framework-owned runnables - Created FluentExample demonstrating: Application.Create().Init().Run<ColorPickerView>().Shutdown() - Disposal semantics: framework creates → framework disposes; caller creates → caller disposes Co-authored-by: tig <585482+tig@users.noreply.github.com> * New Example: Demonstrates new Fluent API using ColorPicker Conditional compilation (`#if POST_4148`) to support both a new Fluent API and a traditional approach for running `ColorPickerView`. The Fluent API simplifies the application lifecycle with method chaining and automatic disposal, while the traditional approach retains explicit lifecycle management. Refactor `ColorPickerView` to support both approaches: - Add an `instructions` label for user guidance. - Replace `_okButton` and `_cancelButton` with local `Button` instances. - Use a new `ColorPicker` with enhanced styling options. Add a warning log for WIP issue (#4148) in `ApplicationImpl.Run.cs` to highlight limitations with non-`Toplevel` views as runnables. Update `Terminal.sln` to include the new `FluentExample` project with appropriate build configurations. Improve code readability with verbatim string literals and better alignment/indentation. * Introduce `RunnableWrapper` for making any View runnable Added the `RunnableWrapper<TView, TResult>` pattern to enable any `View` to be run as a blocking session with typed results, without requiring inheritance from `Runnable<TResult>` or implementation of `IRunnable<TResult>`. - Added `RunnableWrapperExample` project to demonstrate usage. - Introduced `ApplicationRunnableExtensions` and `ViewRunnableExtensions` for clean, type-safe APIs to run views with or without result extraction. - Updated `CodeSharingStrategy.md` to document reduced duplication using `#if POST_4148` directives. - Added `RunnableWrapper.md` with detailed documentation and examples. - Created runnable examples in `Program.cs` showcasing various use cases. - Improved maintainability by reducing code duplication by 86% and increasing shared code by 264%. - Gated all new functionality behind the `POST_4148` feature flag for backward compatibility. * Simplified `#if POST_4148` usage to reduce duplication and improve clarity. Refactored `RunnableWrapper` to use a parameterless constructor with `required` properties, ensuring type safety and better lifecycle management. Updated `AllViewsView` with new commands, improved generic handling, and enhanced logging. Refactored `ApplicationRunnableExtensions` and `ViewRunnableExtensions` for cleaner initialization and event handling. Enhanced `TestsAllViews` to handle required properties and constraints dynamically. Updated documentation to reflect new designs and provide clearer examples. Improved overall code readability, consistency, and maintainability while leveraging modern C# features. * Update docfx documentation for IRunnable architecture - Updated View.md with comprehensive IRunnable section - Interface-based architecture explanation - Fluent API patterns and examples - Disposal semantics ("whoever creates it, owns it") - Result extraction patterns - Lifecycle properties and CWP-compliant events - Marked legacy Modal Views section for clarity - Updated application.md with IRunnable deep dive - Key features and benefits - Fluent API patterns with method chaining - Disposal semantics table - Creating runnable views with examples - Lifecycle properties and events - RunnableSessionStack management - Updated IApplication interface documentation - Updated runnable-architecture-proposal.md - Marked Phase 1 as COMPLETE ✅ - Updated status to "Phase 1 Complete - Phase 2 In Progress" - Documented all implemented features - Added bonus features (fluent API, automatic disposal) - Included migration examples All documentation is now clear, concise, and complete relative to Phase 1 implementation. Co-authored-by: tig <585482+tig@users.noreply.github.com> --------- Co-authored-by: Tig <tig@users.noreply.github.com> Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: tig <585482+tig@users.noreply.github.com> |