fix(cv): retain preview geometry while refreshing last-good pages
This commit is contained in:
@@ -48,3 +48,7 @@ The first independent full review found three P2 acceptance gaps: location was c
|
||||
Fresh full Release backend1227/1227 (40 added cases, zero removed from the preceding1187 report) and frontend981/981 across107 suites passed. Source-bound frontend process before/after hashes match. Two letter browser cases passed after corrections; the unchanged native oracle passed all four letter fixtures/14pages with zero differences. Actual MariaDB source-bound transaction cases and final aggregate Windows/Linux runs remain pending. Re-review of the corrections found no remaining material defect; review did not substitute for execution.
|
||||
|
||||
Actual MariaDB source-boundary successor59/59 passed (`provider-final-03`), including three real relational transaction/held-source cases with a stub inference service. Source hashes match and owned cleanup is confirmed. The intermediate `provider-final-02` preparation failed before starting a service or executing tests and remains preserved. Real API authorization checks also passed (missing owned interaction404; free account403). The seven-scenario responsive browser case additionally passed five targeted WCAG A/AA scans with zero violations; incomplete/manual qualifications remain open and final native reports retain their scan details. Its preceding exact-name locator timeout is a fixture failure retained separately: MUI exposes the select's label together with its selected value.
|
||||
|
||||
## Last-good preview scroll regression
|
||||
|
||||
The aggregate `final-04` run exposed a real regression in both original preview-presentation recovery cases: a held refresh changed scrollTop by7px against the unchanged1px tolerance. The new page selector unmounted while `previewReady` was false, shortening the toolbar and expanding the retained preview viewport. The narrow correction keeps the selector mounted when last-good multi-page HTML exists and disables it until ready. Both original EN/light and NB/dark scenarios now pass, including held/error/retry, retained iframe identity/content and scroll checks (`preview-scroll-successor-01.json`); focused editor50/50 and a fresh static build pass. Independent review found no material issue. No oracle tolerance or original assertion changed. Full frontend successor and fresh aggregate/PDF runs remain pending; failed/interrupted `final-04` is retained.
|
||||
|
||||
Reference in New Issue
Block a user