7.2 KiB
Compact diagram controls and contextual help
Branch-specific work for issue 39, based on published preview source 715036e132677dd2bc2fb37ed316d93dc9987aec.
This record does not accept the complete first release or the other diagram defects.
Problem and choice
The user's screenshot showed two permanent Traceability explanation paragraphs and a separate generic gesture-hint row. Direct Windows inspection of the released water-pressure example reproduced these. At a 1100×700 client size (1102×732 captured window), the Traceability canvas began around window y=470 and ended around y=692, leaving about 220 pixels for modeling. At the normal 1440×900 client size its top was around y=417. The release was opened from the existing downloaded package; only an in-memory projection switch was made and Undo restored the clean model before closure.
The candidate moves general instructions into contextual Help, reusing the offline guide. Architecture, State and Process selection bars use a stable compact row. Allocation retains its full-width destination slot; a concise Arrange control sits beside the existing view-content and Help controls. Traceability retains its displayed-line count and existing actions. Active drawing/assignment feedback appears beside the operation outcome in the persistent status bar, with full text on hover. Operation errors must never be replaced by generic guidance.
Help opens as an overlay and closes with Close or Escape. F1 opens the relevant diagram topic. The workspace, selected group, pending source/property/behavior drafts and camera remain owned by their existing state. Background user commands are rejected before specialized canvas/reference dispatch. Passive results retain their established validity checks. The editor remains the first child of a stable Stack while Help opens and closes.
Rejected alternatives: ordinary navigation to the Guide page cleared selection and could reject Help while a behavior draft was pending; placing feedback beside the entire selection toolbar exhausted narrow-window width; unconditional status substitution hid operation failures. Independent source review identified these before acceptance. Hover-only help was rejected because the complete instructions need a keyboard-accessible route.
Editing workspace UX review is the development/review contract requested by the user. It covers persistent prose, actionable feedback, context preservation and independent native interaction checks.
Verification status
Clean source f83068757af885572f6b3f932d07233dd1b94b7e passed cargo build -p oa-desktop --locked on Windows gnullvm. Its candidate executable SHA-256 is c1a9b196ec03e86f049fa2aadb2b86bfcaa3f0fbae582a80ee39ee8adf0758e3e. The independent source review records the corrected source findings, including passive viewport updates while Help is open.
Independent native review of that executable recovered approximately 61 pixels in normal Traceability, 114 pixels in minimum Traceability and 58 pixels in minimum Allocation. Help resizes, pointer-scrolls, selects the State topic, and closes through Close/Escape. The review also reproduced two acceptance failures: PageDown cannot scroll Help, and F1/Escape preserves a property or Process-creation draft but loses typing focus/caret continuation. These findings require correction and another native check; model-state unit tests alone do not cover internal widget focus. The idle State-canvas gesture sentence is also removed in the successor; active/pending and hovered-item feedback retain their stable slot.
The initial focused test run passed nine and failed one: the allocation test fixture used the default Overview projection before asserting allocation ownership. A separate ten-case mounted geometry run passed nine and failed the destination's existing 400-pixel minimum-width check. Sharing its row with the verbose Arrange action reduced that slot to 328 pixels at minimum size. The successor fixes the fixture explicitly and restores the full destination row without weakening its width check. The original failed logs are preserved in the native/test evidence. These diagnostic runs began before the successor edits and are not frozen-source acceptance of those edits.
Frozen successor 5657a996ed2cefb1fa84ef967b655eaf8f9f6817 passed formatting, strict all-target desktop Clippy and the Windows GUI build. The full desktop run passed 1,538 tests, ignored seven and failed one new mounted Help test before Help opened: its synthetic multi-character KeyPressed event inserted only the first character, as the actual Iced text input specifies. The final correction sends individual character events and retains the exact-caret, scrolling and IME assertions. The existing allocation width and ownership checks and new modifier/pointer cancellation checks passed. See the preserved successor evidence and logs; the failed run is not reported as a passing suite.
Independent native review of that exact executable confirmed keyboard Help scrolling, precise Inspector and Process-draft caret continuation after Help, resize, retained State tool state and complete allocation-error tooltips. Minimum Allocation recovers approximately 64 pixels with the full destination slot; minimum Traceability recovers 114 pixels. The reviewer found no new blocker within that scope.
Final source 6e7c4ddaf0775198384a8c4882809ea8e14b2976 passes formatting, strict desktop/all-target Clippy, GUI build and 125 focused Help/allocation/guide tests (zero failures/ignores). The corrected mounted caret test now executes its complete assertions. Unchanged evidence from the earlier full run is retained at its original source; the remaining 1,421 desktop cases were not rerun at the final source. The initial final GUI link failed when the disk filled; after the user expanded the disk, the same-source rebuild passed. The failed log remains preserved.
The final native report directly verifies cancellation status, guide wording, native group selection through context menus, primary Inspector changes, long-name controls at minimum size, exact caret return after both Help dismissal routes, and Apply/Undo/Redo/Undo recovery. Independent source review found no blocker in the final small delta. The watcher records exit 0 and empty stdout/stderr; all 37 copied model files still match the original after closure. An early transient OSK/modifier obstruction occurred before model loading and is recorded separately from app behavior. No blocking defect was found within this contribution's review scope. Shared integration and release admission remain in issue 39; full product completion is not claimed.
Required native checks: minimum and normal windows; Traceability and Allocation space comparison; multi-selection with long names; Help/F1/Close/Escape with property and behavior drafts; editor focus and canvas scroll/camera restoration; active feedback plus operation failure; State and Process controls; subsequent edit/Undo/Redo and source preservation. Linux remains deferred under the recorded user instruction.