Aller au contenu principal

Backlog Notes

Development

This page collects known issues and edge cases that were identified during development or testing but deferred for later resolution.

Tags — Symlink path mismatch​

Identified: 2026-03-15 Affects: K2 and K3 (core: searchjobstoppable.cpp)

If a tag is created by selecting a folder via a symlink, the path stored in the tag table is the symlink path (e.g. /home/user/link-to-music). However, catalog files are indexed using the real/resolved path (e.g. /media/disk/Music). The tag path comparison in searchFilesInCatalog and searchFilesInDirectory uses QString::contains() on the literal string, so symlinked tag paths never match real catalog paths.

Current comparison logic (searchFilesInCatalog line ~674):

(filePath + "/").contains(queryTag.value(0).toString() + "/")

Suggested fix: Resolve the tag path and/or the file path with QFileInfo::canonicalFilePath() before comparing. Also consider resolving at tag creation time (store the canonical path in the DB) so existing records remain valid after the symlink is removed.

Source path — no retro-normalization of already-stored paths​

Identified: 2026-08-12 Affects: K2 and K3 (core: devicejobstoppable.cpp) Related requirements: SpecValidationRules.md N1–N6

This is a note, not a requirement. Nothing below is authorised work.

N1 normalizes a path only at the moment it is picked or typed. A device whose path was stored with a trailing separator before N1 existed keeps that stored value until the user re-picks or re-edits the path; nothing sweeps the existing device.device_path rows (catalog.catalog_source_path is removed, STI-C19).

Two propagation points bypass the single core rule:

  • core/devicejobstoppable.cpp:351 — device->catalog->sourcePath = device->path; assigns directly instead of going through Catalog::setSourcePath(), so a stale stored path is copied verbatim onto the catalog at update time.
  • qt_widgets/ — K2 never normalizes at all (documented as accepted drift by N7), so a path entered in K2 can re-introduce a trailing separator that K3 then inherits.

Options if this is ever taken up: (a) route the assignment through Catalog::setSourcePath(), (b) add a one-shot normalization pass over existing device/catalog rows, or (c) accept the residual and close it when N7 is closed. A requirement would have to be approved and written into SpecValidationRules.md first.

Explore — directory tree item count is always zero​

Identified: 2026-08-29 Affects: K2 (visible), K3 (value received and ignored) — core: foldertreeloader.cpp Related requirements: SpecExplore.md — see EXP-C10 and the Scope section

This is a note, not a requirement. Nothing below is authorised work.

FolderTreeLoader queries a table or view named filesall at core/foldertreeloader.cpp:54, :62 and :136, and nothing in the repository ever creates it — there is no CREATE TABLE or CREATE VIEW for that name anywhere. Both queries therefore fail silently and FolderNode::fileCount stays at its initialised 0. As a result K2's No of items column (qt_widgets/exploretreemodel.cpp:39 and :175) shows only zeros, and K3 receives the value and ignores it. The count at :136 is also issued once per folder node rather than once per tree.

Deferred deliberately. K2 is in maintenance mode and no UI investment is wanted there. Because K2 and K3 share FolderTreeLoader, making fileCount correct would change K2's visible column from zeros to real numbers — that is a K2 UI change, not a K3 feature, and it is out of scope of the Explore size work approved on 2026-08-29, which deliberately leaves the directory tree untouched.

Separate finding: FolderTreeLoader::loadDirectoryTree (core/foldertreeloader.cpp:54, :62) is the other reader of filesall. It serves DirectoryTreeModel, which is compiled into K2 (qt_widgets/CMakeLists.txt) but instantiated nowhere. That path is dead code and reaches no user; it is equally broken and equally deferred.

Options if this is ever taken up: (a) replace both filesall reads with one grouped aggregate over file and roll the totals up the tree, the shape already required by SpecExplore.md EXP-C6, (b) suppress the count column in K2 and drop FolderNode::fileCount, or (c) create the missing filesall view. Any of them needs an approved requirement first, and (a) and (c) both change what K2 displays.

Search history — raw SQL in the K3 UI layer​

Identified: 2026-08-30 Affects: K3 (qt_quick/appmanager.cpp) Related requirements: SpecSelection.md — see SEL-C5

This is a note, not a requirement. Nothing below is authorised work, and the maintainer has not ruled on it.

AppManager::getSearchHistory() (qt_quick/appmanager.cpp:1357) builds and executes a QSqlQuery directly against the search table and formats the result into the history summary. The UI/core boundary rule is that qt_quick/ files must not contain raw QSqlQuery and must delegate to a core method; this one does not. It predates the Selection work and was found while judging it.

SpecSelection.md SEL-C5 forbids deepening the departure — the device name required by SEL-F2 must come from an existing device accessor, not from a new device join added to this query — but SEL-C5 does not authorise moving the existing query, and this note does not either.

Options if this is ever taken up: (a) move the query and the summary's data retrieval behind a core method on Search or Collection, leaving only the tr() fragments and the joining in AppManager, or (b) accept it as recorded drift. Either way a requirement has to be approved and written first, and (a) touches core/, which needs its own prior approval.

Selection — Open folder context item is always visible in K3​

Identified: 2026-08-30 Affects: K3 (qt_quick/PageSelectionDelegate.qml) Related requirements: SpecSelection.md — Scope section (the context menu's item set is not governed by that spec)

This is a note, not a requirement. Nothing below is authorised work, and the maintainer has not ruled on it.

The Open folder item of the Selection card context menu carries no visible: guard, so it is offered for every device of every type and whether or not a source path exists. The user page Selection.md documents it under Catalog devices only, and as "shown only when a source path is defined". K2 matches the documentation: qt_widgets/mainwindow_tab_filters.cpp:317-323 builds the action inside the Catalog branch only, and only when selectedDevice->path is not empty. K2 also uses the icon document-open-folder where K3 uses document-open.

The code is therefore doing something no requirement authorises and the user documentation contradicts. It was found while judging the Selection work of 2026-08-30 and is outside that work's scope.

Options if this is ever taken up: (a) add the two guards so K3 matches Selection.md and K2, (b) approve a requirement that K3 deliberately offers the item for every device, and correct Selection.md and its translations to match, or (c) leave it recorded. The behaviour must not be enshrined as a requirement merely because it exists.

Backup — K3 run progress does not use the standard message builder​

Identified: 2026-08-31 Affects: K3 (qt_quick/PageBackupForm.qml) Related requirements: SpecBackup.md — see BKP-F9

This is a note, not a requirement. Nothing below is authorised work, and the maintainer has not ruled on it.

BKP-F9 is marked [Implemented] and requires that backup progress and outcome be "reported via the standard status-bar message builder". SpecBackup.md applies to K2 and K3 alike. On the K3 Backup page the running mapping's progress is a ProgressBar plus a label assembled in QML from raw numbers (PageBackupForm.qml:445-473: percent, files done over total, speed, ETA, current file), not a StatusBarMessageBuilder message. BKP-F9 may well be satisfied on the K2 side; only the K3 surface was examined.

Found while judging the activity-panel work of 2026-08-31, and outside that work's scope. Note that SpecOperationQueue.md OPQ-C16 is deliberately scoped to creation, update and search, so it does not reach this screen either.

Options if this is ever taken up: (a) build the K3 backup progress text with the standard builder, keeping the progress bar as an addition rather than a replacement, (b) qualify BKP-F9 per version if the builder is genuinely not the right vehicle for a byte-and-ETA progress bar, or (c) leave it recorded.

Activity panel — duplicated five-second delays​

Identified: 2026-08-31 (revised the same day, after the search timer was removed) Affects: K3 (qt_quick/OperationQueueView.qml, qt_quick/appmanager.cpp) Related requirements: SpecOperationQueue.md — see OPQ-C14, OPQ-F15

This is a note, not a requirement. Nothing below is authorised work, and the maintainer has not ruled on it.

OPQ-C14 requires the panel's linger to be a single named constant defined in one place. It is [Planned] and is not satisfied today. Two hard-coded 5000 values remain:

  • OperationQueueView.qml — the panel's own completionLinger;
  • AppManager::startNextDeviceUpdate() — a QTimer::singleShot(5000, ...) that clears m_deviceUpdateStatusText when the queue empties.

The two race on the same message for the device-update path, which is the exact situation OPQ-C14 exists to prevent. The equivalent timer on the search path was removed when OPQ-F15 was implemented, so search is now clean; the update path is not. Satisfying OPQ-C14 means the update timer goes the same way and the constant gets a single definition.

Keyboard shortcuts — Ctrl+F always targets the Selection filter​

Related requirements: SpecKeyboardShortcuts.md (out of scope there)

This is a note, not a requirement. Nothing below is authorised work, and the maintainer has not ruled on it.

Ctrl+F focuses the Selection page's device filter on every page. The filter is a Kirigami ActionTextField with focusSequence: StandardKey.Find (qt_quick/Main.qml, the deviceSearchField), so the binding is window-wide. The open question is whether Ctrl+F should focus the Search page's own field while Search is open. To be reviewed later.

Collection import — no report after Import (K3)​

Identified: 2026-10-04 (K3 Settings user page) Affects: K3 (qt_quick/PageSettings.qml, AppManager::importDevice, AppManager::updateAllImportsFromSource) Related requirements: SpecCollection.md (#780 update report)

This is a note, not a requirement. Nothing below is authorised work. The maintainer asked on 2026-10-04 to log it as a gap to close.

In K3, Collection Import & Synchronization shows a summary message (catalogs updated, files and size before/after) only after Update (importReportText is set in updateAllImportsFromSource only). After Import, the user gets the progress-panel status but no summary of what was added.

This is not a K2 parity gap: K2 shows neither summary, only a status-bar message for both actions. The K3 update report is itself a K3 addition (#780).

To close it: show, after an Import, a summary of what was added (for example devices, catalogs, files and total size imported), in the same message area as the update report. Needs an approved requirement and its new strings.

Device tree — self-parented default storage​

Identified: 2026-10-04 Affects: K2 and K3 (core: Collection::insertPhysicalStorageGroup, Device::loadSubDeviceList) Related requirements: SpecCollectionOpen.md OPN-O4, OPN-F8, OPN-C7

This is a note, not a requirement. Nothing below is authorised work.

Before the OPN-F8 fix, opening a collection with device 1 present, device 2 absent and no Storage device created a default "Local disk" that was its own parent (device 2, parent 2). The recursive query in Device::loadSubDeviceList (UNION ALL over device_parent_id) never ends on that cycle, so every open of such a collection hangs at start-up. The fix prevents new cases; a collection already damaged keeps the row and still hangs. The Quality check cannot help: the collection hangs before it can run, and check 8 (missing parent) does not see a parent that exists.

Options, each needing its own approved requirement:

  • A Quality check for self-parented or cyclic device_parent_id.
  • Make loadSubDeviceList robust to cycles (UNION instead of UNION ALL, or a depth guard), so a damaged collection at least opens.
  • A manual repair procedure (set the row's device_parent_id to 1).

Duplicated data — further candidates, not decided​

Status: candidates only, recorded 2026-09-28 from the redundancy diagnosis that led to STI-C16, STI-C17, STI-C19 and STI-C20 (SpecStorageIdentity.md). Nothing below is authorised work; each needs its own user decision.

  • search.search_location — Done (user decision 2026-09-29): removed by Database::runMigration_3_0. It was never given a value; reads keep its index as '' AS search_location, and Memory-mode search_history.csv keeps slot 21, written empty.
  • search.search_storage / search.search_catalog — Done (user decision 2026-09-29): removed by Database::runMigration_3_0; the scope is selected_device_ID_list. Reads keep their indexes as '' placeholders, and Memory-mode search_history.csv keeps slots 22–23, written empty. K2's history restore derives its Storage/Catalog labels from the first id in selected_device_ID_list, by the same rule as displaySelectedDeviceName(). K2's migrateExistingSearchDeviceData_2_6 is removed: a collection older than 2.6 must be opened with 2.12 first (SpecVersions.md).
  • file.file_full_path — equals file_folder_path + "/" + file_name.
  • catalog.catalog_file_path — derivable in Memory mode.

K3 user-doc pass (3.0) — K3 features not yet in the user documentation​

Related requirements: see each entry

The English user pages in docs_src/docs/ describe K3 and are kept accurate with every user-facing K3 change, in the same piece of work (maintainer rule, 2026-10-10). No K3 feature waits here for its English documentation. The three entries listed earlier (Esc goes back one step, Selection device filter, Quality check) are documented in Interface.md, Selection.md and Settings.md.

The French and Czech pages (i18n/fr, i18n/cs) are translated later, in the #803 translation phase; until then some of them still hold the K2 text.

Progress report: catalog name in theme colour (K3)​

Identified: 2026-09-30 Affects: K3 (progress/status messages built by StatusBarMessageBuilder) Related requirements: SpecProgressReport.md — Component Breakdown, Part 3 (Device Context)

This is a note, not a requirement. Nothing below is authorised work.

The catalog name in the Device Context part ("Catalog X of Y | CatalogName") is specified as bold, no colour in K2 and K3. It was previously #39b2e5 blue, but K3 renders these messages with Qt Quick Text.StyledText, which ignores <span style='color:…'>, so the blue never showed in K3. The preferred colour is the theme's Kirigami.Theme.linkColor (same colour as the K3 Settings section titles), but core/ is UI-agnostic and K2 has no Kirigami, so a theme colour would need a core-level colour-injection mechanism, which the user declined.

When to take it up: once K2 is discontinued and K3 can implement it without the K2/K3-agnostic core constraint, display the catalog name in Kirigami.Theme.linkColor.

Technical note: Text.StyledText needs <font color="…"> rather than <span style="color:…"> (span colour is ignored). Switching to Text.RichText would render the span but loses eliding.

Collection export — no failure reason for the CSV export​

Identified: 2026-10-10 (SpecCollectionExport.md) Affects: K3 (and K2) — core: Collection::exportAllToMemoryMode, Collection::exportAllCatalogFiles

When a Memory-mode (CSV) export fails, K3 reports only status "Error" (CEX-F11), with no reason. exportAllToMemoryMode always returns true, and exportAllCatalogFiles silently skips (continue) a catalog whose .idx file cannot be opened or whose query fails. A user can end up with an incomplete export and no message.

When to take it up: when a reason is wanted. It needs a core/ change (approval) to return or expose the failure, plus new user-visible strings (per-string approval).

Collection export — folders without files missing after an earlier load​

Identified: 2026-10-10 (SpecCollectionExport.md, CEX-F5 known limitation) Affects: K3, Memory mode, Memory → SQLite export. Core: Catalog::loadFoldersToTable(), Collection::loadAllCatalogFiles()

The K3 export calls loadFoldersToTable() for every catalog, then loadAllCatalogFiles(). If a catalog's file list was loaded earlier in the session (for example by a search), that load created only folders that hold files. loadFoldersToTable() then skips the catalog (folder rows present, load date current), so its folders without files are missing from the exported SQLite file. An export in a fresh session is complete.

Same mechanism: loadCatalogForRootReplace in core/device.cpp (near line 1250), which clears the catalog's folder rows before calling loadFoldersToTable().

Fix: a core/ method that clears and reloads a catalog's folders, used by the export (and possibly by loadCatalogForRootReplace). It needs core approval, which was not given on 2026-10-10.