# Starting a run: labelled entry points ## Change - Pipeline detail: the icon-only play button in the top bar is now a labelled `FilledButton.icon` "Run" (same key, `runPipelineActionKey`). In a narrow bar it moves into the More menu with its label, via the existing `SegmentedOverflowAction`. - Runs: an extended "New run" button (`runs-new-run`) floats over the list's bottom-right corner while the list is the visible view (phone list, or the desktop list pane). It sits above the shell's bottom bar: the button's bottom is lifted by the body's bottom padding, the same inset the list clears. While it is shown, the list's bottom padding also reserves its band (`_newRunBand` in `lib/features/runs/screen.dart`), so the last row should rest above the button (no test covers this; ui-review to confirm). It opens a searchable picker of active pipelines (archived ones are withheld), reads the chosen pipeline's latest version, then opens the same start dialog the Run button opens. - The start flow moved out of `RunPipelineAction` into `lib/features/pipelines/ui/run_pipeline_flow.dart` (`openRunPipelineFlow`), so both entry points share one submit, pending-intent and accept path. - The pending-intent store on Runs uses the same endpoint scope as Pipelines (`lib/application/pending_run_scope.dart`). - Walkthrough step 3 now reads: "Press Run on a workflow, or New run on the Runs tab, to start one." ## Fix for ui-review rejection (visit 1) - Defect: on a phone the "New run" button sat beneath the bottom navigation (semantics rect y=928-984 at 430x900), so it was hidden and a tap reached Settings. - Cause: the button was the last child of a column under the list, and the shell's body extends under the bar (`extendBody`), so the bottom of the column was under the bar. - Fix: the button is now an overlay positioned above the body's bottom inset, and the list reserves the band. Desktop uses the same overlay. - Tried first and rejected: wrapping the list in a `Column` with the button as a second child (in both an `Expanded` and an outer form). It fixed the phone position but broke row selection on desktop (`list_detail_consistency_test` "selected rows" at 640 wide: row 1 kept its selected flag after row 2 was opened). The overlay keeps the list as a direct child and passes that test. ## Files - Source: `lib/features/pipelines/ui/run_pipeline_action.dart`, `lib/features/pipelines/ui/run_pipeline_flow.dart` (new), `lib/features/runs/screen.dart`, `lib/features/runs/ui/new_run_button.dart` (new), `lib/features/runs/ui/new_run_flow.dart` (new), `lib/features/runs/ui/new_run_picker.dart` (new), `lib/application/pending_run_scope.dart` (new), `lib/application/pipelines_destination.dart`, `lib/application/runs_destination.dart`, `lib/features/walkthrough/walkthrough_steps.dart`. - Tests: `test/features/runs/ui/new_run_test.dart` (new), `test/features/pipelines/ui/run_pipeline_launch_test.dart` (one case added), `test/features/runs/ui/runs_list_refresh_test.dart` (finder scoped to the list rows; its expectations are unchanged). ## Regression coverage for the rejection - `new_run_test.dart` "New run clears the bottom bar and opens the picker on a phone": the screen sits in a scaffold with `extendBody` and a NavigationBar, as the shell does. It asserts the button's bottom is at or above the bar's top, and that a tap on the button opens the picker. Before the fix it failed (button bottom 884, bar top 820). ## Focused results (visit 2) Command: `tool/memwatch/full.sh` with 19 test files (the new-run and launch tests plus every test file that builds RunsScreen, the Pipelines Run action, the Pipelines screen, Settings, or the walkthrough). - 119 passed, 1 failed. Verdict named only the intended paths. - Failing: `test/application/run_start_wiring_test.dart` "an accepted launch opens the returned run and re-reads the runs list exactly once". It also fails on untouched `main` (checked in a temporary main worktree, since removed): it times out waiting for `get_pipeline` after tapping `pipeline-row-1`. - `list_detail_consistency_test.dart`: passes (it failed with the rejected `Column` layout; see Fix above). - `new_run_test.dart`: 5 passed, including the phone bottom-bar case. - `tool/analyze.sh`: clean (exit 0). ## Affected UI states (for ui-review to re-capture; not captured here) 1. Pipeline detail, wide bar: labelled "Run" next to the list and more icons. 2. Pipeline detail, narrow phone bar: Run should move into More, labelled. 3. Runs list, phone (430x900): "New run" floats above the bottom bar, clear of the last row at rest. This is the rejected state; it must be re-captured. 4. Runs list, desktop: "New run" floats over the list pane's bottom-right. 5. New run picker: default, searching, no match, no active pipelines, error+retry. 6. Start dialog reached from the picker. 7. Walkthrough step 3 copy. Before captures: `/tmp/ui-review/pipe-light-430x900/01-pipeline-detail.png` (play icon unlabelled) and `/tmp/ui-review/base-light-430x900/03-runs.png` (no New run). ## Limitations - No UI was built or captured in this stage (per the stage instructions); rendering is unverified. The phone position is verified only by the widget test's layout geometry under a scaffold that mirrors the shell. - `_newRunBand` is a fixed reservation (16 + 56 + 16). If the extended button grows with text scale, the last row may need more room. Measuring it would need a size callback; not done. - The Runs picker reads only the current endpoint scope. Pending intents stored under legacy endpoints are not offered by New run (Pipelines still offers them). - Not run on web, a device or the Fold.