5.9 KiB
Narrow Rename Details/Preview native trial preparation
Status: PREPARATION ONLY, pending reviewed test-only correction. No helper check, gate admission, stage, fixture copy, launch, compiler or GUI input was executed. Original440 trial/receipts/fixtures remain unchanged. Only new .local helper/config/handoff files were written.
Files: .local/rename-scroll-native-bebf9f5.py and .local/rename-scroll-native-bebf9f5.config.json. The helper uses explicit require checks throughout and rejects Python -O. Its PENDING_FINAL_SOURCE=True blocks every command; clearing this alone is not sufficient because the configured actual bebf preview receipt failed. A future revision needs the exact reviewed next source pin, fresh stem names, actual gate-preparation helper/config hashes and all fresh closed results/admissions. Preserve this preparation and original failed receipts when producing that revision. No admission/pass is inferred from expected membership.
Source of fixture: the six original navigation-palette-20260913-01/source-baseline files, bound against source-example-preparation.json, rather than the mutated source-work. Target declaration is assess_report, UUID00000000-0000-4000-8000-000000000403, at sa/reporting.arcadia line11. Use a fresh unaccepted proposal such as assess_report_trial_preview. This differs from the already-renamed alias shown in original043, so compare the same semantic target/file/hunk and actual measured visual position, not old pixel coordinates.
I viewed original JPEG043,044,045 through view_image. Frame043 shows sa/reporting.arcadia file1/2, hunk1/3, rows1-7/7, scrolled to contextline10 and changed declarationline11;044 actually displays Full details;045 returns to that same logical page at line8/top. Both source and screenshots support the failure description. The new trial must capture before/details/return before any new wheel input.
The original440 trial closed normally at2026-09-13T10:20:14.983254Z, PID6772, exit0. The new actual bebf preview gate closed exit101 at10:27:06.170099Z with4passed/1failed, zero ignored,581unchanged inputs. Its panic at preview_scroll_tests.rs458 occurs BEFORE the Details transition:114.09999 report versus114 painted coordinate. Original inputs/log/result are now additionally pinned in the config. They remain failed evidence, not a native failure or a completed correction test. Parent/Mencius owns exact-report versus rounded-painted assertion repair.
Design and operations after a reviewed final revision:
checkrequires seven actual closed list/preview/rename/navigation/source-navigation/full/strict receipts and their real admission files. It checks exact commands, source/status/closure, helper/config/raw hashes, list1099 and exact focused/full outcomes/manual exceptions, same actual test executable, and the separate GUI Cargo JSON/result/environment/preserved executable. Missing, pending or failed records refuse before output mutation.- It reconstructs581 gate and615 GUI inputs from exact Git blobs and the source worktree, with exactly the three reviewed desktop paths changed from440. No core oracle or old pass is reused. GUI is one actual unoptimized x64 PE; libunwind.dll must match the prior preserved runtime and actual recorded compiler-directory copy, with matching recorded Clang identity. No server executable is required.
preparecreates only NEW.local/native-qa/rename-scroll-stage-bebf9f5-20260913-01and.local/native-qa/rename-scroll-bebf9f5-20260913-01paths. Stage has GUI+libunwind plus manifest; trial has byte-checked baseline/work copies and preparation. No huge executable is copied until actual admission succeeds. Every receipt uses exclusive create; no existing trial or stage is overwritten.launchrepeats admission and stage/fixture readback, creates a private profile/AppData/temp/cwd and minimal runtime PATH, records exact command/environment/PID, launches one owned normal GUI and waits for its actual exit. Root performs normal UI Open of the new source-work/project.arcadia; no launch argument or old profile is assumed. It records actual closure/log hashes/stage readback/disk equality. Abnormal or interrupted supervision records an error without inventing normal closure or terminating unrelated processes.checkpoint --label cancel-saverequires the owned launch and pinned stage/preparation/helper/config. It snapshots current complete work files before comparing every byte and file membership against the six-file baseline. A mismatch is preserved and fails. It does not execute a semantic oracle or assert transient model/history equivalence. Closure independently repeats complete disk equality.
Smallest itinerary: open the copied baseline; select the exact assess_report declaration; F2 and preview the unaccepted proposal; move its actual preview to a visible nonzero position in file1/hunk1; capture; show Details and capture its mounted text; return and capture the identical file/hunk/rows and visual offset BEFORE any rescroll. Then exercise explicit file and hunk/page navigation plus changed file filter, confirming deliberate top reset on each changed destination, and a repeated unchanged filter/visibility control. Cancel, Save all, wait for visible saved status, and take a disk-equality checkpoint. Close normally (Alt+F4); retain actual owned-process closure and zero/nonzero log facts. Do not click Rename alias or Apply. Any accidental accepted/editing action invalidates this Cancel-only comparison and needs separately authorized semantic evidence.
Review obligations: parent should inspect current helper source and the eventual exact source/stem/config diff before any command. AST syntax and absence of assert statements were checked without importing/executing the helper. Live behavior, supervisor mechanics, post-compile GUI schema and final membership have NOT been executed by this preparation. No full prior native trial, scenario fixtures, dual-server build, package configuration or e124 oracle dependency proof was copied into this narrow helper.