Repository navigation
Human checks for recently merged features #63
Description
Activity
New human checks from the 2026-10-04 merges:
- Batch rename (#85 Batch rename: rename photos as units, with tokens, preview and undo #96, issue Batch rename: keep pairs and all sidecars together, add metadata tokens, preview and undo #85): select a RAW+JPEG pair and a photo with
.xmp/.rrdatasidecars, right-click → Rename, try tokens like{YYYY}{MM}{DD}_{sequence}and{rating}, check the preview, rename, then use Undo from the toast. All sidecars and both pair members should follow. - Denoise cancel (#68 Denoise: cancel backend jobs by ID #97, issue Denoise: interrupt backend jobs and identify their events/results #68): start a BM3D or AI denoise on a large raw, press Cancel. The app should stay responsive, no result should appear later, and no
.partialfiles should be left behind. - Reference View (#80 Editor: Reference View, pin a read-only photo beside the current edit #94, issue Editor: pin a reference image beside the current edit #80): Shift+R, pick a reference from the filmstrip, zoom and pan it independently.
- Lightroom import report (#54 Lightroom import: lens profile, exposure unit flag, dead keys, nested looks, import report #95, issue Lightroom import: quick fixes (lens profile, exposure unit, dead keys, nested looks, import report) #54): import a photo with an XMP that has masks or a custom white balance; a toast or the report should list what didn't transfer.
- Batch rename (#85 Batch rename: rename photos as units, with tokens, preview and undo #96, issue Batch rename: keep pairs and all sidecars together, add metadata tokens, preview and undo #85): select a RAW+JPEG pair and a photo with
AI RAW denoise (#91, draft PR #129) needs a human check before merge. Automated dialog tests use a mock Tauri bridge; normal RAW/DNG export and native model/cache tests pass locally.
- On a real Sony RAW, check Best/Fast and independent Sharpen (on by default), first-use model download/progress/errors, More methods and the before/after view at fit, phone-sized view and 100%. Judge detail, plastic appearance, dark bokeh and smooth gradients; no sharpening preference is proven at normal viewing size.
- Cancel during model preparation/inference, close and reopen on another photo; no stale previews or completion should appear. Check batch cancellation preserves finished files and existing DNGs.
- Save/reopen the denoised DNG, confirm crop/orientation/WB/ISO/camera metadata and subsequent interactive editing. Run a larger batch and another camera; unsupported sensors should give a clear error with More methods available.
- Open Quick noise reduction (basic): existing nonzero sidecar slider values remain unchanged. Check subsection collapse, focus mode, normal panel navigation and translated layout.
- Check macOS/Windows, including Apple RAW 9 under More methods. Model inference/provider probes and release verification here are Linux only.
Private screenshots, photo derivatives and timings remain on the maintainer's machine. OpenVINO/FP32 acceleration prototypes are excluded from this PR; their precision differences require separate image-quality review. Focus-aware denoise remains a later follow-up.
#26 Adobe lossy DNG decoder follow-up: on the local samples in
samples/issue-26/, open the two actual pink/white reproductions in RapidRoom and compare framing, colours and exposure with the private before/after contacts and Adobe previews. The corrected renders restore natural colours but Adobe profile/exposure parity is not promised. Also open the supplied Sony A7R V Lightroom DNG, whose full render is unchanged. Linux automated exports/conversion and the 60-render corpus pass; native desktop interaction and macOS/Windows remain untested. No photographs are attached.#75 / PR105 RAW thumbnail follow-up: on local Sony A7C II/A7R V samples and the Adobe/no-preview DNG fixtures, check library/Card previews and medium-thumbnail quality, rotated portraits, edited/default-only sidecars, real crop framing, AI patches, and rapid navigation/cache invalidation. Edits must remain visible; camera JPEG previews may differ from developed RAW. Linux native probes cover 72 cold/warm cases and the 60-export corpus is pixel-identical, but running desktop interaction and macOS/Windows/Android remain untested. Full-image DNG fallback retains substantial memory cost (~2.9GiB in the 61MP QA process); verify normal responsiveness. Originals and contacts remain local; none attached.
Remaining human desktop check for #131 / merged #132 (
621b6de2):- Open an OM System/Olympus ORF repeatedly in Library and Editor, then export it with several adjustment sets. Confirm normal white balance, consistent colours across repeated opens/exports, and that the embedded thumbnail agrees with the developed image.
Linux automated verification passes: 30 independent fresh-process OM-1 exports and the full 60-image corpus are pixel-exact; all 72 thumbnail/proxy probes pass. Interactive desktop use and Windows/macOS remain unchecked.
PR #125 merged as
f56d79f8. Open Terminal Here: verify the context menu opens the selected existing folder, including spaces; configured TERMINAL and fallback behavior; packaged AppImage; other-platform behavior where available. Flatpak should show its explicit unsupported error. No embedded terminal was added.PR #124 merged as
11d047b9. Assistant plugin/skill: verify actual Claude Code and Codex setup/commands on a feature checkout with #102, folder bootstrap, tutor behavior and source lookup. Local bootstrap tests use a stub assistant and do not validate a live MCP session. Main has no MCP listener yet.PR #109 merged as
2ef8a88a. Compact sliders: turn Compact on in Settings > General, then compare Comfortable and Compact in the running desktop WebView; check real sidebar resizing, all slider consumers, translations, keyboard focus/editing and physical touch. Automated browser fixtures cover complete ControlsPanel/MasksPanel and interaction handlers, with evidence committed under rapidroom/validation/compact-sliders/.PR #127 merged as
1179229d. Sony camera aspect crops: open the4:3 file, compare the camera frame, expand/reset to7008×4672, reopen saved crop and explicit full-frame sidecars, and check3:2 unchanged. Josh approved merging on2026-10-04; automated16 private exports verify exact crop/reset/sidecar precedence. Actual1:1/16:9 files and desktop interaction remain untested.PR #100 merged as
4f5fa204. Josh approved this merge: “ok on 100 and 126”.Lensfun corrections: in the native app, choose/detect Sony/Tamron/Sigma E-mount profiles, including a full-frame lens on an APS-C body; check crop/warp, lens paste and batch auto-lens. Open and save an older edited photo to confirm legacy lens values retain its appearance. Rendering merge was approved; remaining native UI/quality checks are not claimed as complete.
Interpolation remains linear; centre offsets and current TCA behavior remain unchanged; 100% vignetting still applies 80% of the gain. No physical lens/print quality judgment or cross-platform native UI execution is claimed.
PR #126 merged as
c8b58802. Josh approved this merge: “ok on 100 and 126”.Export recipes: test native preset selection/dropdowns, user-versus-built-in precedence and saved sharpening options. Confirm the three Instagram dimensions and padding in the UI. Phone/upload appearance and physical print results require human checks. MCP selection/listing remains a follow-up after #102; #121 remains open.
The pad/resize correction may move a derived edge by one pixel. Enabled output sharpening intentionally changes detail and can affect colour near clipping; no claim of noise-free output, Adobe equivalence or print quality is made. Output sharpening stays off by default.
Native automation added for #134 in draft PR #136. On Arch Linux / WebKitGTK / Intel LNL / private Weston, it verifies real thumbnail and GPU-processed preview loading, Exposure/Contrast edits, exact Undo preview restoration, enabling Compact in General settings and persisting it, full-size GUI JPEG export, and clean exit with zero frontend/native errors. Screenshots capture the actual app on the private compositor.
This reduces the untested native flow for #109, but its human checks remain: physical touch and keyboard/focus behavior on the actual desktop, resizer gestures, remaining slider/mask consumers and translation judgments. The automation sends DOM mouse/keyboard events and tests one fixed CC0 photo; it does not establish physical input or all-platform behavior. No human item here is marked done.
PR #129 merged as
5c79e076. Josh approved this merge: “129, I will take your recco”.AI RAW denoise: in the native app compare Best/Fast and Sharpen on/off on real noisy RAWs, judge colour/detail/halos at normal and100% views, exercise cancellation/batch reuse and stale-selection handling. Validate first-use model download/offline errors. Legacy noise controls must still be reachable under More methods.
OpenVINO and promoted FP32 models are separate follow-ups; this merge retains the original F16 CPU path and Sharpen-on default approved by Josh. No claim of Lightroom AI equivalence or universally improved detail is made. Unsupported CFA/camera normalization cases report explicit errors.
PR #112 merged as
d1725b29. Josh approved this merge: “approve 112, with a note that we need to do a full import calibration before we can use the feature”.Exercise single-image and folder Lightroom XMP imports in the native app, inspect masks and warnings, test unsupported AI-mask reporting, and verify geometry on rotated/cropped inputs. Full global and mask calibration is required before real use; track #31 and #133.
Josh approval is conditional on the calibration note. Lightroom global adjustments and masks remain uncalibrated and not ready for real use until #31 and #133 are complete. Import does not imply equivalent Lightroom tone, profile, white balance, local contrast or AI denoise. Private reference samples remain local.
#119 terminal panel follow-up for the UI checklist (PR #139)
The automated native release scenario passed all 15 steps, including real Claude Code/Codex from built-in and external launcher paths, strict CSP, docking/resize/collapse, exact UI/preview edits and quit cleanup. Evidence:
rapidroom/validation/native-ui/terminal-evidence.json.Additional manual checks for the installed app: physical typing and Ctrl+Z/Escape focus in the terminal; drag docking and resize with normal pointer input; restart with saved tabs and shell/font preferences (fresh shells, no restored transcript or automatic assistant); external Start Claude/Codex in the actual desktop terminal while the panel is closed. The automated external terminal executable is an owned launcher stub, so desktop focus was not measured. macOS/Windows and packaged AppImage runtime remain untested. Packaging/MCP enablement policy is separate under #135; this PR does not install or publish.
Things that passed every automated check but still need a person in the running app. Tick them off, or open a bug.
BM3D denoise strength (#27 BM3D denoise: raising the strength no longer adds noise back #61): on a real noisy raw, compare 50% and 80% strength with the previous build. Is it too smooth or plasticky now that less detail is blended back?
White balance picker (#22 White balance picker samples the original linear image #60): pick on a neutral patch several times. Is the result stable, and unchanged after editing temperature/tint and picking again? Do picks land on the right spot after rotate, flip, crop and transform?
Culling auto-advance (#21 Culling: auto-advance after rating, and an "exactly N stars" filter #56): turn on Settings → General → Auto-Advance After Rating. Does rating with 0–5 move to the next photo in the library and the editor? Does the "exactly N stars" filter work?
Read-only Card mode (#25 Read-only Card mode #50): open a memory card (or a folder marked as a card). Do edits stay in memory and nothing gets written to the card, with a clear notice shown?
In-camera ratings (#20 Show in-camera star ratings in the library (harvest from csiroqa, explicit-clear from masmoriya) #48): do a camera-rated Sony ARW's stars show, and do a rating you set or clear in RapidRoom win over the camera's?
Lightroom import (#53 Import Lightroom XMP sidecars (harvest of upstream PR 1465 by dimafa) #62): right-click a photo or folder with Lightroom XMP sidecars and import. Do crop, basics, HSL and curves come across, and do ratings, labels and keywords transfer?
Keyboard focus (Headless export: a failed export exits with status 0 #29 of markallisongit's fork, in RapidRoom since 0.1.0): done (checked 2026-10-02).
Lightroom catalog collections (Lightroom catalog migration: preview and import collections with folder relinking #76, PR #76 Import Lightroom collections from a catalog, with preview and folder relinking #103): import a real current and older catalog; compare ordinary/set counts and smart-collection exclusions with Lightroom. Preview/relink a moved drive and confirm the catalog/original contents and mtimes stay unchanged; refusal while Lightroom has its lock open. In the app, add roots to Folders and re-import a moved/renamed group without duplicate groups. Check Windows/macOS paths and Android bundled SQLite. NFC/NFD filename normalization has no fallback.
Lightroom Lua-table presets (Presets: import legacy Lua-table Lightroom .lrtemplate files without executing Lua #78, PR #78 Presets: import Lua-table .lrtemplate files without executing Lua #107): import real
.lrtemplatefiles, including Looks/local corrections, and inspect values plus unsupported-field console warnings. Check escaped/localized names. Verify photos change only on explicit application, and confirm Windows/macOS imports.Pick/reject culling (Culling: pick/reject flags independent of star ratings #32, PR #32 Pick and reject flags, separate from star ratings (harvest from lalibertemarc) #99): check P/X/U on single photos and selections in grid, filmstrip, culling and editor; Shift+P opens Presets. Check the five flag filters and translated layout, badges/dimming/buttons, and Delete rejected confirmation/trash/virtual-copy warning. Round-trip rejects with real Lightroom/darktable XMP. In Card mode, refused flags or stars must return to their saved display; rapid repeated actions must keep the last successful state.
Presets sidebar (Develop left sidebar: presets with safe hover previews, favorites and folders #82, PR #82 Develop left sidebar: presets with safe hover previews, favorites and folders #108): check grid/list layouts and migration of existing presets; hover and leave previews, modal navigation and undo history; confirm previews never save photo edits. Apply presets explicitly and check the result. Verify favorite/folder drag ordering after restart, keyboard focus and collapsed sidebar behavior, including Card mode.
Catalog photo-edit migration (Lightroom catalog migration: preview develop settings and virtual copies before apply #77, PR #77 Preview catalog photo edits and virtual copies before import #128): use a real closed/copied catalog, including paths moved from Windows or another computer. Review per-photo edits, ratings, virtual copies and unsupported fields before selecting a subset. Cancel/relink/preview must not write photo data. Confirm saved edits, masks, EXIF, keywords and cleared ratings are preserved by default, and edit/rating replacement choices work independently. Repeat import and restart to check virtual-copy identity and visibility. Check portrait crop/straighten/rotation and compare appearance with Lightroom; unsupported profile/local/AI settings remain explicit. Verify desktop interaction and macOS/Windows. Local calibration covered only a synthetic catalog derived from a private sidecar, not a real catalog.
Color tool eyes (Color tools: independent visibility toggles for Color Grading and Mixer #84, PR #84 Independent visibility toggles for Color Grading and Color Mixer #106): check Color Grading and Color Mixer eyes separately on global and mask panels, keyboard focus and accessible pressed labels. Verify parent Color bypass, old edits with missing keys, reset/paste/undo and saved state after restart. Local RAW probes and native tests pass; desktop interaction remains unchecked.
Lights Out (Viewing: Lightroom-style Lights Out in Library and Develop #83, PR #83 Lights Out viewing in Library and Editor #104): cycle L/Shift+L in library/editor, Escape to normal, verify default folder-tree shortcut N and preserved custom bindings. Test rapid Black/off/on and fullscreen restoration from both windowed and already-fullscreen states. In Reference View, photos stay visible and controls disappear. Check editing/crop/masks, light/dark themes and platform behavior. Automated transition tests use a mock window.
HSL pipette (#98, merged e85f200)
Local sampler/gesture tests and 60 default renders pass; automated tests do not establish native desktop behavior.