|
| 1 | +# Plan: Project-Based Pytest Execution |
| 2 | + |
| 3 | +## Overview |
| 4 | + |
| 5 | +This plan describes the implementation of **project-based test execution for pytest**, enabling multi-project workspace support where each Python project within a workspace can execute tests using its own Python environment. This builds on top of the project-based discovery work from PR #25760. |
| 6 | + |
| 7 | +## Problem to Solve |
| 8 | + |
| 9 | +In a multi-project workspace (e.g., a monorepo with multiple Python services), users currently cannot: |
| 10 | +- Run tests with the correct Python interpreter for each project |
| 11 | +- Have separate test trees per project in the Test Explorer |
| 12 | +- Properly handle nested projects (parent/child) |
| 13 | + |
| 14 | +## Architecture |
| 15 | + |
| 16 | +### Key Components to Add |
| 17 | + |
| 18 | +| Component | File | Purpose | |
| 19 | +| ----------------------- | -------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------- | |
| 20 | +| **TestProjectRegistry** | [testProjectRegistry.ts](../src/client/testing/testController/common/testProjectRegistry.ts) | Registry that discovers and manages Python projects per workspace | |
| 21 | +| **ProjectAdapter** | [projectAdapter.ts](../src/client/testing/testController/common/projectAdapter.ts) | Interface representing a single Python project with its test infrastructure | |
| 22 | +| **projectUtils** | [projectUtils.ts](../src/client/testing/testController/common/projectUtils.ts) | Utility functions for project ID generation and adapter creation | |
| 23 | + |
| 24 | +### How It Works |
| 25 | + |
| 26 | +``` |
| 27 | +┌─────────────────────────────────────────────────────────────────┐ |
| 28 | +│ VS Code Workspace │ |
| 29 | +│ ┌─────────────────────────────────────────────────────────────┐│ |
| 30 | +│ │ TestController ││ |
| 31 | +│ │ ┌───────────────────────────────────────────────────────┐ ││ |
| 32 | +│ │ │ TestProjectRegistry │ ││ |
| 33 | +│ │ │ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │ ││ |
| 34 | +│ │ │ │ ProjectA │ │ ProjectB │ │ ProjectC │ │ ││ |
| 35 | +│ │ │ │ (Py 3.11) │ │ (Py 3.12) │ │ (Py 3.10) │ │ ││ |
| 36 | +│ │ │ │ ┌─────────┐ │ │ ┌─────────┐ │ │ ┌─────────┐ │ │ ││ |
| 37 | +│ │ │ │ │Discovery│ │ │ │Discovery│ │ │ │Discovery│ │ │ ││ |
| 38 | +│ │ │ │ │Adapter │ │ │ │Adapter │ │ │ │Adapter │ │ │ ││ |
| 39 | +│ │ │ │ ├─────────┤ │ │ ├─────────┤ │ │ ├─────────┤ │ │ ││ |
| 40 | +│ │ │ │ │Execution│ │ │ │Execution│ │ │ │Execution│ │ │ ││ |
| 41 | +│ │ │ │ │Adapter │ │ │ │Adapter │ │ │ │Adapter │ │ │ ││ |
| 42 | +│ │ │ │ └─────────┘ │ │ └─────────┘ │ │ └─────────┘ │ │ ││ |
| 43 | +│ │ │ └─────────────┘ └─────────────┘ └─────────────┘ │ ││ |
| 44 | +│ │ └───────────────────────────────────────────────────────┘ ││ |
| 45 | +│ └─────────────────────────────────────────────────────────────┘│ |
| 46 | +└─────────────────────────────────────────────────────────────────┘ |
| 47 | +``` |
| 48 | + |
| 49 | +### Execution Flow |
| 50 | + |
| 51 | +1. **User runs tests** → `TestRunRequest` with selected `TestItem`s arrives |
| 52 | +2. **Controller** checks if project-based testing is enabled |
| 53 | +3. **Group tests by project** → Tests are sorted by which `ProjectAdapter` they belong to (via URI matching) |
| 54 | +4. **Execute per project** → Each project's `executionAdapter.runTests()` is called with: |
| 55 | + - The project's Python environment |
| 56 | + - `PROJECT_ROOT_PATH` environment variable set to project root |
| 57 | +5. **Results collected** → Each project's `resultResolver` maps results back to test items |
| 58 | + |
| 59 | +### Required Changes by File |
| 60 | + |
| 61 | +#### Controller ([controller.ts](../src/client/testing/testController/controller.ts)) |
| 62 | +- Add `TestProjectRegistry` integration |
| 63 | +- New methods: `discoverForProject()`, `executeTestsForProjects()`, `groupTestItemsByProject()` |
| 64 | +- Debug mode should handle multi-project scenarios by launching multiple debug sessions |
| 65 | + |
| 66 | +#### Pytest Execution Adapter ([pytestExecutionAdapter.ts](../src/client/testing/testController/pytest/pytestExecutionAdapter.ts)) |
| 67 | +- Add `project?: ProjectAdapter` parameter to `runTests()` |
| 68 | +- Set `PROJECT_ROOT_PATH` environment variable when project is provided |
| 69 | +- Use project's Python environment instead of workspace environment |
| 70 | +- Debug launches should use `pythonPath` from project when available |
| 71 | + |
| 72 | +#### Debug Launcher ([debugLauncher.ts](../src/client/testing/common/debugLauncher.ts)) |
| 73 | +- Add optional `pythonPath` to `LaunchOptions` for project-specific interpreter |
| 74 | +- Add optional `debugSessionName` to `LaunchOptions` for session identification |
| 75 | +- Debug sessions should use explicit Python path when provided |
| 76 | +- Use unique session markers to track individual debug sessions (avoids `activeDebugSession` race conditions) |
| 77 | +- Properly dispose event handlers when debugging completes |
| 78 | + |
| 79 | +#### Python Side ([vscode_pytest/__init__.py](../python_files/vscode_pytest/__init__.py)) |
| 80 | +- `get_test_root_path()` should return `PROJECT_ROOT_PATH` env var if set (otherwise cwd) |
| 81 | +- Session node should use project root for test tree structure |
| 82 | + |
| 83 | +## Feature Behavior |
| 84 | + |
| 85 | +### Single Project Workspace |
| 86 | +No change from existing behavior—tests run using the workspace's interpreter. |
| 87 | + |
| 88 | +### Multi-Project Workspace |
| 89 | +- Each project has its own root node in Test Explorer |
| 90 | +- Running tests uses the correct interpreter for each project |
| 91 | +- Results are scoped to the correct project |
| 92 | + |
| 93 | +### Nested Projects |
| 94 | +``` |
| 95 | +workspace/ |
| 96 | +└── parent-project/ |
| 97 | + ├── tests/ |
| 98 | + └── child-project/ |
| 99 | + └── tests/ |
| 100 | +``` |
| 101 | +- Parent project discovery ignores child project via `--ignore` flags |
| 102 | +- Execution receives specific test IDs, so no cross-contamination |
| 103 | + |
| 104 | +### Debug Mode |
| 105 | +- **Single project**: Debug should proceed normally with project interpreter |
| 106 | +- **Multiple projects**: Multiple debug sessions should be launched in parallel—one per project, each using its own interpreter |
| 107 | +- **Session naming**: Each debug session includes the project name (e.g., "Debug Tests: alice (Python 3.11)") |
| 108 | +- **Session isolation**: Each debug session is tracked independently using unique markers, so stopping one session doesn't affect others |
| 109 | + |
| 110 | +### Cancellation Handling |
| 111 | + |
| 112 | +Cancellation is handled at multiple levels to ensure proper cleanup across all parallel project executions: |
| 113 | + |
| 114 | +``` |
| 115 | +┌─────────────────────────────────────────────────────────────────┐ |
| 116 | +│ User Clicks "Stop" │ |
| 117 | +│ │ │ |
| 118 | +│ ▼ │ |
| 119 | +│ CancellationToken fires │ |
| 120 | +│ │ │ |
| 121 | +│ ┌─────────────────┼─────────────────┐ │ |
| 122 | +│ ▼ ▼ ▼ │ |
| 123 | +│ ┌───────────┐ ┌───────────┐ ┌───────────┐ │ |
| 124 | +│ │ Project A │ │ Project B │ │ Project C │ │ |
| 125 | +│ │ Execution │ │ Execution │ │ Execution │ │ |
| 126 | +│ └─────┬─────┘ └─────┬─────┘ └─────┬─────┘ │ |
| 127 | +│ │ │ │ │ |
| 128 | +│ ▼ ▼ ▼ │ |
| 129 | +│ Kill subprocess Kill subprocess Kill subprocess │ |
| 130 | +│ Close pipes Close pipes Close pipes │ |
| 131 | +│ Resolve deferred Resolve deferred Resolve deferred │ |
| 132 | +└─────────────────────────────────────────────────────────────────┘ |
| 133 | +``` |
| 134 | + |
| 135 | +#### Cancellation Levels |
| 136 | + |
| 137 | +1. **Project execution level** ([projectTestExecution.ts](src/client/testing/testController/common/projectTestExecution.ts)) |
| 138 | + - Early exit if cancelled before starting |
| 139 | + - Checks cancellation before starting each project's execution |
| 140 | + - Projects not yet started are skipped gracefully |
| 141 | + |
| 142 | +2. **Execution adapter level** ([pytestExecutionAdapter.ts](src/client/testing/testController/pytest/pytestExecutionAdapter.ts)) |
| 143 | + - `runInstance.token.onCancellationRequested` kills the subprocess |
| 144 | + - Named pipe server is closed via the callback |
| 145 | + - Deferred promises resolve to unblock waiting code |
| 146 | + |
| 147 | +3. **Debug launcher level** ([debugLauncher.ts](src/client/testing/common/debugLauncher.ts)) |
| 148 | + - Token cancellation resolves the deferred and invokes cleanup callback |
| 149 | + - Session termination events are filtered to only react to the correct session |
| 150 | + - Event handlers are disposed when debugging completes |
| 151 | + |
| 152 | +#### Multi-Session Debug Independence |
| 153 | + |
| 154 | +When debugging multiple projects simultaneously, each `launchDebugger()` call must track its own debug session independently. The implementation uses a unique marker system: |
| 155 | + |
| 156 | +```typescript |
| 157 | +// Each debug session gets a unique marker in its configuration |
| 158 | +const sessionMarker = `test-${Date.now()}-${random}`; |
| 159 | +launchArgs[TEST_SESSION_MARKER_KEY] = sessionMarker; |
| 160 | + |
| 161 | +// When sessions start/terminate, we match by marker (not activeDebugSession) |
| 162 | +onDidStartDebugSession((session) => { |
| 163 | + if (session.configuration[TEST_SESSION_MARKER_KEY] === sessionMarker) { |
| 164 | + ourSession = session; // Found our specific session |
| 165 | + } |
| 166 | +}); |
| 167 | +``` |
| 168 | + |
| 169 | +This avoids race conditions where the global `activeDebugSession` could be overwritten by another concurrent session start. |
| 170 | + |
| 171 | +### Legacy Fallback |
| 172 | +When Python Environments API is unavailable, the system falls back to single-workspace adapter mode. |
| 173 | + |
| 174 | +## Files to Change |
| 175 | + |
| 176 | +| Category | Files | |
| 177 | +| ----------------------- | ------------------------------------------------------------------------------------------------------------ | |
| 178 | +| **Core Implementation** | `controller.ts`, `testProjectRegistry.ts`, `projectAdapter.ts`, `projectUtils.ts`, `projectTestExecution.ts` | |
| 179 | +| **Adapters** | `pytestExecutionAdapter.ts`, `pytestDiscoveryAdapter.ts`, `resultResolver.ts` | |
| 180 | +| **Types** | `types.ts` (common), `types.ts` (testController) | |
| 181 | +| **Debug** | `debugLauncher.ts` | |
| 182 | +| **Python** | `vscode_pytest/__init__.py` | |
| 183 | +| **Tests** | `controller.unit.test.ts`, `testProjectRegistry.unit.test.ts`, `projectUtils.unit.test.ts` | |
| 184 | + |
| 185 | +## Testing |
| 186 | + |
| 187 | +### Unit Tests to Add |
| 188 | +- `testProjectRegistry.unit.test.ts` - Registry lifecycle, project discovery, nested projects |
| 189 | +- `controller.unit.test.ts` - Controller integration, debug scenarios, test grouping |
| 190 | +- `projectUtils.unit.test.ts` - Utility functions |
| 191 | + |
| 192 | +### Test Scenarios to Cover |
| 193 | +| Scenario | Coverage | |
| 194 | +| ----------------------------- | --------------------------------------------- | |
| 195 | +| Single project workspace | Unit tests + legacy flows | |
| 196 | +| Multi-project workspace | New controller unit tests | |
| 197 | +| Nested projects | Discovery tests + ignore behavior | |
| 198 | +| Debug mode (single project) | Existing debug tests | |
| 199 | +| Debug mode (multi-project) | Session isolation, independent cancellation | |
| 200 | +| Legacy fallback | Existing controller tests | |
| 201 | +| Test cancellation | Cancellation at all levels (see above) | |
| 202 | + |
| 203 | +## Out of Scope |
| 204 | +- **Unittest support**: Project-based unittest execution will be handled in a separate PR |
| 205 | +- **End-to-end tests**: Manual testing will be required for full validation |
| 206 | +- **Multi-project coverage aggregation**: Deferred to future work |
| 207 | + |
| 208 | +## Expected User Experience |
| 209 | + |
| 210 | +### Debugging Across Multiple Projects |
| 211 | +When debugging tests spanning multiple projects: |
| 212 | +- Multiple debug sessions should be launched simultaneously—one per project |
| 213 | +- Each debug session should use the project's configured Python interpreter |
| 214 | +- All projects' tests should run in debug mode in parallel |
| 215 | +- Users should be able to switch between debug sessions in VS Code's debug panel |
| 216 | +- **Stopping one debug session should NOT affect other running sessions** |
| 217 | +- Each debug session is named with its project (e.g., "Debug Tests: alice (Python 3.11)") |
0 commit comments