Files
ARCADIA-Studio/harness/release-canvas-native-d61f1f3-20260912.md
T

4.5 KiB
Raw Blame History

Native activation paint correction — 2026-09-12

Bounded Windows acceptance of d61f1f3877: the previously erased activation borders remain continuous and exposed shoulders select their calls, while exact message endpoints retain priority. This is not acceptance of the complete OA/SA release, palette, or the later lifecycle union.

The same source passes all 836 desktop tests, strict all-target Clippy with warnings denied, and the GUI build. All 533 committed inputs remain unchanged. Exact gate evidence.

The staged executable is 113,939,456 bytes, SHA-256 c990fa05772c60666ecf96ab808e6140e57542ad448ddd092b977be0b587136f. The identified trial ran from 23:46:20 to 23:51:20 UTC with a private profile and disposable fixture copies. Its actual process exited normally with code 0. Native manifest, screenshots and source checkpoints retain launch/closure records and the exact staged-build identity.

A fresh independent native-evidence review finds no actionable issue within this scope. It verifies all 43 archived payload hashes and complete saved fixture bytes, and independently checks visible Inspector identities. Its same-coordinate pixel discriminator finds zero missing border rows in each repaired band versus all 21 missing rows per band in the old source83 negative control. This does not extend acceptance to other renderers.

Observed behavior

  • In bound OA nested C1/C2/R2/R1, clicks at the formerly covered outer shoulder (664,440), (664,535) and (664,625) select C1, Inspector alias bound_1040. The selected border is continuous through all three former label bands. Exact C2 receive (668,457), R2 send (669,549) and R1 send (664,641) select bound_1030, bound_1060 and bound_1050 respectively. Labels and arrowheads remain visible. Coordinates refer to the archived 1442×932 window frames.
  • In OA reentrant pressure review, the right outer shoulder (664,440) selects C1 bound_3040; the left inner bar (429,510) selects C2 bound_3030. The reverse call/reply paths and both activation outlines remain visible.
  • In SA nested pressure checks, the former R2-label shoulder (664,535) selects C1 bound_2040, while the exact R2 endpoint (669,549) selects bound_2060. The same repaired outer border and separate inner execution are visible.
  • At a 1100×700 client area (1102×732 outer window), 100% retains the repaired geometry but partly clips the right header. Fit reaches 66%; the outer shoulder (529,363) still selects C1 bound_2040, with its border continuous.
  • Actual Ctrl+S after navigation/selection/resize preserves the complete Project and exact SourceBundle, checked against the untouched bound baseline by the identified core oracle. Four checkpoint files have independent byte hashes.

The GUI trial uses the existing bound fixture: 34 elements, six relations, zero validation errors and two completeness warnings. The preflight and final Save comparison both pass. The oracle's binary identity is recorded separately from the native application's source; it does not imply that the oracle was compiled from d61f1f3.

Scope and remaining limitations

This trial directly retests the defect found in the source83 native trial. It does not repeat the earlier architecture/state Save/Undo focus workflows, change model data, or claim a new native self-call/ALT/PAR/lifecycle-marker result. Those geometry cases have source-specific automated evidence; the forthcoming lifecycle integration still requires combined tests and native acceptance.

Unselected outlines and small message text remain faint, especially at 66% Fit. That is still issue 26. Stronger colors are not part of d61f1f3. Partial header clipping at 100% in the minimum window is recoverable through Fit; this trial does not establish overall diagram ergonomics or readability at the lowest allowed zoom. Linux remains deferred by explicit user instruction.

Reproduction: use the source-identified staged build and archived bound baseline, open bound-nested-calls.arcadia, choose Behavior & scenarios, and repeat the scenario/selection sequence above. The archive retains the exact trial launcher, checkpoint procedure and stage manifest. Original screenshots are unmodified.