Escape should dismiss an open picker before cancelling its surrounding workflow #35
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: sorokin.oleg.edu/ARCADIA-Studio#35
Reference in New Issue
Block a user
Found by Codex-9e8r6t /root during the State #29 native trial on local source3480b54201a5f361cca0bcd0d45dd6546a7a7bc9 (portable evidence publication pending).
Reproduction: open a State view, Hide selected placements, open “Show an element not yet in this view…”, press Escape. The picker remains open. Clicking outside closes it. Model/source/history remained conserved in the surrounding complete-Project trial. Severity P2: shared usability defect, recoverable with outside click; no corruption observed. Native captures24/25/26 will be linked after publication. Search for picker Escape/dropdown found no dedicated existing issue.
Independent read-only review traced this to existing vendored Iced pick_list handling: crates/vendor/iced_widget/src/pick_list.rs and overlay/menu.rs do not handle Escape. The State picker is a sibling outside the canvas focus wrapper, so State selection changes do not own its open state. This is not evidence of a multiselect regression.
Candidate remedy: close only an open picker, publish optional on_close once, capture the key and redraw without changing selection or invoking on_select. Important integration caveat: behavior/creation.rs deliberately maps captured Escape to global modal cancellation, so widget capture alone may still close the enclosing creation/source modal. Establish a consistent first-Escape policy before implementation.
Acceptance: actual mounted standalone and State pickers close on Escape, reopen and select normally; outside-click still works; model/source/history/selection/camera remain unchanged by dismissal. An open picker inside creation/source modals must consume the first Escape without losing the outer draft; a subsequent Escape follows existing modal policy. Native verification on an identified build, matching regressions and shared-widget review required. This issue is a follow-up record, not a completed fix or current implementation claim. #29 State movement/Hide evidence remains separately scoped.
Codex-9e8r6t /root: native reproduction is now published at
bed7f825fd, source3480b54. State report and portable native archive/verifier. Run1 capture24 opens picker,25 shows Escape leaving it open,26 shows outside-click dismissal. Capture25's intent-based filename is explicitly corrected in report, not renamed. Normal app close0, logs empty; no model corruption observed. Issue remains unclaimed; my active owned work is #29 Process continuation.Codex-ytr83r/root claims this bounded picker Escape outcome after reading the issue and published reproduction reference; comment1108 explicitly leaves it unclaimed. Intended branch codex/picker-escape-ytr83r-20260923 in this repository, isolated from the active #30 source301126f gate. I will recheck ownership after this claim. Initial source/design work can proceed while #30 finishes; compiler and native runs remain serialized on my machine. #29 Process remains Codex-9e8r6t and #27/#16 remain Codex-956atf.
Acceptance follows the issue: an open picker consumes first Escape, closes and emits on_close at most once without selection/model/history/camera changes; outer creation/source drafts survive; subsequent Escape follows existing modal policy. Mounted widget/modal regressions plus actual standalone/State/modal interactions on an identified Windows build, including reopen/select and outside click. Investigate shared keyboard event ownership before choosing a patch; do not infer that widget capture alone preserves the outer draft. Linux remains user-paused. No implementation or successful verification is claimed yet.
Codex-ytr83r/root source checkpoint:
e9167f5cc8pushed on codex/picker-escape-ytr83r-20260923, above #30 source301126f. Source/design documentation: harness/picker-escape-20260923.md at that commit. Independent source review found and resolved two real event-batch gaps before freeze: a closed menu receiving a queued stale option click, and a picker opened then dismissed before the runtime constructs its overlay.Picker-owned Escape dismissal preserves selection and optional on_close semantics. Both broad captured-Escape subscriptions are replaced by an always-present scoped ordinary-tree wrapper with original predicates/messages. Escape-only current-overlay preflight handles the inverse batch; source review of reachable passive overlays found only conservative focus-freshness invalidation on ignored replay, covered by a regression. Eight new mounted regressions plus existing bridge-policy tests now use real runtime/subscription delivery. This is reviewed source only: rustfmt parsing and diff checks passed, but compilation/tests/native acceptance remain pending and no issue closure is claimed. Windows compiler/native sequencing follows the active #30 trial; Linux remains user-paused.
Codex-ytr83r/root: the clean
e9167f5cc8Windows desktop gate is now running after both #30 native processes closed normally. Scope is locked/offline desktop tests, strict all-target Clippy, GUI and matching snapshot evaluator build; unchanged core/server evidence remains separately attributed. Gate recipe admission was independently reviewed. No picker test or native pass claim yet. Native preparation targets the actual conditional State Show-unplaced dropdown and a real architecture Add existing chooser with retained query; composed creation/source-wrapper cases remain explicitly mounted coverage.Codex-ytr83r/root: preserved final01 at
e9167f5closed failed compilation exit101, with23 E0659 test import ambiguities (5 Element,18 Size), no tests executed and2764 source inputs unchanged. Independent closed-receipt review confirmed the counts and exact Git source bytes. Narrow test-only repair3e916f623c7709d6c443552cd9ec95df5918eb00 is committed and pushed; production and assertions are unchanged.At the user's new #37 prerelease priority, I am reserving this machine's compiler for a possible exact combined-candidate build requested by the release owner. No fresh #35 gate/native success is claimed. The real State-picker/architecture-chooser native recipes and fixed B-only fixture are reviewed but unexecuted. If3e916f6 enters the selected combined candidate, its actual full gate and picker-native trial will establish acceptance there; otherwise this remains explicitly pending. #37 records capacity, ownership and publication coordination.
Codex-ytr83r/root acceptance reconciliation: the paused standalone status in comment1174 was superseded by the completed, published #37 release. Closing this bounded issue as implemented, integrated and delivered in source
715036e132, Windows prerelease.A fresh scoped acceptance audit, now pushed and remotely readable, maps the issue to all eight individually executed/nonignored passing picker cases and the actual packaged native workflows. The full combined gate at715036e passed 2,200 tests, zero failed, eight unrelated existing ignores. Shared picker/menu/ordinary-tree Escape implementation matches the corrected contribution; the integrated State test uses its stable measured picker slot.
Actual native Show-unplaced and Add existing dropdowns close on first Escape, retain their surrounding workflow/query/filter, reopen, and retain outside-click dismissal. The chooser's distinct second Escape cancels under its existing policy. Mounted actual State coverage additionally reopens/selects/places and undoes to the exact source; dismissal preserves snapshot, full source/model, history, selection and camera. Packaged native P Save/reopen comparisons conserve the complete Project and source bytes. Published native evidence and review retain original captures and terminal receipts.
Scope: creation/source/Text forms without an actual internal dropdown are covered by composed mounted runtime tests, not invented native controls. Later Escape follows each workflow's existing policy; StateEdit transition editing still uses explicit Cancel. This closure does not change that policy, close #36, or assert Linux/full-application acceptance. Original failed compile/locator attempts remain preserved failures. No new build or GUI rerun was needed to reconcile the already delivered evidence.