Create child work is broken for activities that are inside entities #4
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: sorokin.oleg.edu/ARCADIA-Studio#4
Reference in New Issue
Block a user
Create child work is broken for activities that are inside entities. You can't create subactivities when the parent activity is inside an entity
Codex-ytr83r /root — taking a bounded diagnosis of this child-work failure on my branch while the current Windows integration is validated in issue5. I will trace the activity-inside-entity case through palette admission, semantic parent/allocation and the creation command, then fix the concrete defect and verify child creation plus Undo/Redo. Initial delegated work is read-only; I will identify affected files before overlapping edits. Activity movement/reparenting remains the separate issue3.
Codex-ytr83r /root — source diagnosis found the failure: Add child creates the correct semantic child, but leaves the performer's direct Allocation on the now-composite parent. Existing leaf-only allocation validation rejects the whole transaction. OA Activity and SA Function share this path; graphical nesting exposes the allocation context.
Chosen bounded fix: preserve the known performer when decomposing work. Move the original direct allocation to the first child in the same atomic command, retaining its identity/metadata; later children inherit only an unambiguous existing performer summary. Unallocated or mixed contexts must not invent a performer. Recompute placement from the adjusted candidate so the composite stays grouped with its performer. Removing allocation would lose the user's context; relaxing the metamodel would broaden semantics unnecessarily.
A private patch and App-level OA/SA first/second/grandchild, grouping and exact Undo/Redo regressions are being prepared. Current issue5 desktop compile inputs remain frozen; no production correction or passing runtime result for issue4 is claimed yet.
Codex-ytr83r /root — independent review found a related role-assigned leaf case, so the correction now transfers all existing target-side Allocation and RoleAssignment relations when creating the first child. Each retains its identity, metadata and artifact references. Role-only, unresolved or ambiguous roles do not gain an invented allocation; later siblings inherit only a unique existing performer summary. This keeps existing leaf constraints intact.
The revised private patch passed a second independent source review with no material finding. I am integrating it in app.rs/tests.rs and starting Windows regression validation. Tests cover OA/SA first child, sibling and grandchild, folded grouping, role contexts and exact Project/SourceBundle Undo/Redo. Native verification remains pending; no completed fix is claimed yet.
Codex-ytr83r /root — fix published at 865483b4c66361a01fe0d9e731859220bafb4136; push and exact branch readback passed. The three focused App regressions passed on unchanged formatted source: OA/SA first child, sibling and grandchild; folded grouping; unallocated/mixed controls; role contexts, artifact references and complete Project/SourceBundle Undo/Redo. Full desktop suite is running separately. Native fixture/checker preparation covers the reported nested workflow and persistence; issue remains open until actual updated-GUI acceptance.
Codex-ytr83r /root — full Windows desktop gate for865483b passed707/707,0failed/ignored/filtered, exit0 at18:18:19.294UTC. All261 captured crate/Cargo/OpenAPI inputs remain unchanged. This includes the three focused child-work cases plus the existing contextual/lifecycle suite. Independent native checker review, warnings-denied compilation and OA/SA fixture admission also passed; GUI build and actual native acceptance remain pending.
Codex-ytr83r /root — actual extracted BF9C2E61 Studio (production865483b) now passed OA first-child, sibling and grandchild creation. Each saved result passed the complete Project and exact SourceBundle oracle; all three Undo/Save and Redo/Save comparisons matched the ORIGINAL accepted checkpoints. Original allocation and both role identities/metadata/artifact references were retained; sibling identity/allocation and predicted nested placement also matched. Full physical two-file copies/hash inventories were captured separately.
Acceptance remains open: actual reopening has NOT completed, and SA native creation is pending. After the accepted grandchild Redo/Save, the Open dialog capture repeatedly failed with
window capture timed out: timed out waiting on channel. Fresh selection/retry and one clean helper restart did not recover screenshots, so native input stopped. The model was already saved; last process read was responding with an empty stderr log. This is a Computer-helper validation gap, not a claimed application failure or reopened-model pass. Exact closed OA evidence is being archived; issue5 retains packaged/integrated release scope.Codex-ytr83r /root — actual OA reopen gap now closed on the same extracted BF9C2E61 GUI/source865483b. Capture recovered, but initial accessibility clicks had only left the native picker pending; Studio Working and ignored Save/Open/close were expected busy guards. Escape produced Open cancelled. A fresh picker using filename keyboard input+Return then explicitly produced Source project opened; CtrlS produced All source files saved. Confirmed snapshot2files/5671bytes passed stage3 with original first/sibling carriers, complete Project/SourceBundle comparison to original grandchild and full physical-inventory equality. The earlier19:49 file-only passes are retained as non-reopen evidence. No deadlock/data loss claim. SA native counterpart still pending; source a57 canvas union is separate and currently running its Windows suite. No Linux/WSL activity.
Codex-ytr83r /root — SA counterpart completed on extracted BF9/source865: actual first child, sibling, grandchild each Save + complete stage1/2/3 model/source check, exact Undo/Save and Redo/Save against original checkpoints, then native filename+Return reopen explicitly Source project opened and CtrlS All source files saved. Reopened stage3/compare both exit0 at20:07:44UTC. Full physical two-file inventories match all7 history/reopen expectations; originalSA checkpoints2665/3323/3827bytes. One transient reopen capture and one missed filename entry recovered after fresh observation; no model error/data loss or process restart. OA reopen confirmation is497; durable evidence archival/independent review follows before issue closure. Separate canvas a57 full desktop752 passed20:06:06; remaining Clippy/build/native are separate. Linux/WSL paused.
Codex-ytr83r /root — closing this bounded child-work defect after source865483b and the complete old865/BF9 OA+SA native continuation. First child, sibling and grandchild each preserve original responsibility identity/metadata/artifact references and pass Save/Undo/Redo against original complete model/source carriers; actual reopen+Save now passes in both layers. Historical OA file-only
1948receipts are explicitly excluded from reopen proof. Independent review found no material evidence issue; all111 continuation payloads,16 closed receipts and9 physical carrier comparisons matched.Durable evidence is normal-pushed in checkpoint 108b1dd: continuation archive, independent review. All209 checkpoint files/9,055,138 bytes matched the Git index and second readback before commit, including every archived log. Newcanvas a57 preserves this implementation and passes752 desktop tests, but its own native acceptance remains separate in #2/#5. No new prerelease or full-goal completion is claimed; #5 still owns remaining package/native acceptance.
Codex-ytr83r referenced this issue2026-09-11 20:42:11 +00:00