Split the export model/classification from the session wording, add the no-session, selected-repair and missing-evidence coverage, and update the parity matrix and R4 note. The visit probe now passes the scheme name its fixture reads, asserts the painted theme before capture and re-captures the export states.
8.0 KiB
R4 — visit Session export state
Delivered
The Timeline visit pane's Session section reports the recorded export state as native fact rows:
- Export —
complete,partial,missing,skipped,repair, or the server's exact unrecognised status word. - Reason — the server-recorded diagnostic, verbatim, for a present export that is not a repair; a missing export keeps its diagnostic in the recorded sentence.
- Repair / Repair option — every recorded repair attempt in server order,
labelled
<status> · <n> messages.< diagnostic>, withOpen repaired exportwhere a non-selected attempt has a file. These are plain text: no control and no request. - The existing partial-export sentence and the single
Open <file>action are unchanged.
Completeness is not inferred: it requires status == "complete", positive
messages and html_bytes, no truncation and no diagnostic. A complete status
without that evidence is partial; a legacy session.html with no evidence is
unknown, never complete. Repairs are shown even when the main export is
complete. No repair_session_export request is sent or advertised.
Review fixes (visit 2)
- Removed the analyzer-gate warning at
lib/features/runs/ui/run_detail_visit_session.dart:77(unnecessary_non_null_assertion);flutter analyzeis clean again. - One completeness rule:
_evidenceCompletenow backs bothVisitSessionExport.isCompleteand the evidence-selection check, so selection and display cannot drift. - One status classifier:
_exportStatebacks both the present-export and missing-file fact rows.completewithout evidence is partial, known partial statuses are partial, anything else is unknown with the server's own word. - Split the 449-line session state file by responsibility: session wording and
state assembly (
run_detail_visit_session_state.dart, 109 lines), export model/resolution/classification (run_detail_visit_export.dart, 352 lines), and shared recorded-field reading (run_detail_visit_recorded.dart, 31 lines). The state file re-exports the export file, so the pane and existing tests keep one import. - Added the three coverage gaps the review named: a non-skipped visit with no session has no Export row; a selected repair suppresses the Reason row; an evidence status whose named file is missing keeps the server word and puts the diagnostic in the recorded sentence.
- Extended the existing real-browser visit probe with a complete-export visit that carries a failed repair, and a partial-export visit with a diagnostic.
- Updated
PARITY.mdso the row no longer says "Missing UI".
Theme render fix (visit 3)
The UI review found that PROBE_SCHEME=dark never painted dark: the driver sent
theme=1, while the fixture compares params['theme'] == 'dark'. Every earlier
-dark- capture was therefore light. Fixed in
tool/run_detail_visit_probe.mjs: the driver now passes the scheme name, and the
probe samples the painted page (mean luminance over the screenshot) before any
capture. A dark run must paint below 0.5 mean luminance and a light run above it,
so this mismatch fails loudly instead of silently labelling light captures dark.
A negative check with the old theme=1 URL fails with
PROBE_SCHEME=dark painted the other theme (mean luminance 0.970).
All four dark combinations were re-run and all their captures regenerated: Material/Cupertino × 1280×900/1× and 320×568/2×. Measured mean luminance is 0.084 (Material 1280), 0.140 (Material 320), 0.081 (Cupertino 1280) and 0.116 (Cupertino 320). No test, check or page exception occurred.
Files
Changed:
app/lib/features/runs/ui/run_detail_visit_session.dart— one-line analyzer fix; rendering unchanged.app/lib/features/runs/ui/run_detail_visit_session_state.dart— now session wording/state assembly plus a re-export of the export module.app/test/features/runs/ui/run_detail_visit_session_export_test.dart— 11 tests (8 delivered + 3 review gaps).app/tool/run_detail_visit_probe.dart— two server-shaped export-state visits.app/tool/run_detail_visit_probe.mjs— export-row checks, narrow-layout scrolling, two new captures per host/size, corrected scheme query parameter and a painted-theme luminance gate.app/PARITY.md— session audit/export row updated to Source with the remaining action/publication named.
Added:
app/lib/features/runs/ui/run_detail_visit_export.dart— export state enum, fact/repair/export models, evidence ordering, status classification, completeness rule.app/lib/features/runs/ui/run_detail_visit_recorded.dart— recorded text, count and artifact lookup shared by both modules.app/notes/R4-session-export.md— this note.
Checks run
From app/, Flutter 3.44.6:
| Command | Result |
|---|---|
flutter analyze |
No issues found (was 1 warning, exit 1) |
dart format --output=none --set-exit-if-changed on the 6 changed Dart files |
0 changed, exit 0 |
flutter test test/features/runs/ui |
142 passed |
flutter test test/features/runs |
205 passed |
flutter test |
1247 passed |
flutter build web |
Built build/web, exit 0 |
flutter build web --target tool/run_detail_visit_probe.dart --output build/probe-run-detail-visit |
built, exit 0 |
node tool/run_detail_visit_probe.mjs × 4 host/size combos |
12/12 checks true in each (includes the painted-theme check) |
Browser evidence
The real RunDetailView was driven over the probe's fake get_run gateway with
a fourth and fifth recorded visit. The checks assert, for the audit visit,
Export complete, Repair failed · 0 messages. The repair read failed.,
Repair option Open repaired export, Open session.html, and no Reason row;
for the partial visit, Export partial, Reason The native read was refused.
and the sentence This export is failed; full session evidence is not available. The existing pane/session/gallery checks still pass.
Under /Users/alan/.local/share/pipelines/data/artifacts/110/public/r4-session-export/:
| Host | Viewport | New captures |
|---|---|---|
| Material | 1280×900, dark, 1x | run-detail-visit-material-dark-1x-1280x900-{export-complete-repair,export-partial-reason}.png |
| Material | 320×568, dark, 2x | run-detail-visit-material-dark-2x-320x568-{export-complete-repair,export-partial-reason}.png |
| Cupertino | 1280×900, dark, 1x | run-detail-visit-cupertino-dark-1x-1280x900-{export-complete-repair,export-partial-reason}.png |
| Cupertino | 320×568, dark, 2x | run-detail-visit-cupertino-dark-2x-320x568-{export-complete-repair,export-partial-reason}.png |
Each combo also re-captures the 8 existing states (40 PNGs total). Evidence
JSON: evidence-{material,cupertino}-{1x-1280x900,2x-320x568}.json, 12 true
checks each, including the measured page luminance.
The 1280×900 captures show every row on one screen in true dark. At 320×568 with
2x text the pre-existing 240px minimum tab body leaves a short scroll region: the
probe scrolls to reach and verify the rows (no overflow or exception), and the 2x
captures show the header plus the start of the facts rather than the whole
section. The rows are present and reachable, but the 320 capture is not proof of
legibility at rest; the UI review's scrolled captures cover that. This is the same
narrow-layout constraint recorded in notes/R4-visit.md, not new to this change.
Not verified / preserved
- Web only. Android/iOS builds and device rendering were not run; the source remains portable (Material/Cupertino theme roles and plain text layout only).
- No authenticated server or production run was touched; fixtures only. The preview is not published to 8096.
repair_session_exportremains unwired, and the repair-option label remains inert text by design.- Artifact matching still selects by recorded name only, as it did before this
change; web additionally filters repairs by session/backend/directory and
checks
size_bytes == html_bytes. Inherited, not changed here. - No commit, push, merge or deployment.