- Generated by
1.17.0
|
Pokered Save Editor 2
Pokemon Red & Blue save file editor - Qt 6 C++/QML
|
When you see a compiler or runtime error, find it here.
| Error | Fix |
|---|---|
| XxxDB::store private | XxxDB::inst()->getStore() |
| XxxDB::store.size() | XxxDB::inst()->getStoreSize() |
| XxxDB::ind.value(k, nullptr) | XxxDB::inst()->getIndAt(k) |
| XxxDB::ind.value(k) | XxxDB::inst()->getIndAt(k) |
| getIndAt(key, nullptr) (2 args) | getIndAt(key) — 1 arg only |
| Random::rangeExclusive(...) static | Random::inst()->rangeExclusive(...) |
| Random::rangeInclusive(...) static | Random::inst()->rangeInclusive(...) |
| Random::flipCoin() static | Random::inst()->flipCoin() |
| MapsDB::search() static | MapsDB::inst()->search() |
| FontsDB::convertToCode(...) static | FontsDB::inst()->convertToCode(...) |
| FontsDB::convertFromCode(...) static | FontsDB::inst()->convertFromCode(...) |
| Utility::decodeAfterUrl(...) static | Utility::inst()->decodeAfterUrl(...) |
| FontsDB::expandStr(...) static | FontsDB::inst()->expandStr(...) |
| FontSearch::results private | search->getFontCount() / search->fontAt(i) |
| entry->fieldName protected | entry->getFieldName() |
| entry->ind protected | entry->getInd() |
| entry->name protected | entry->getName() |
| entry->readable protected | entry->getReadable() |
| entry->glitch protected | entry->getGlitch() |
| entry->once protected | entry->getOnce() |
| entry->toSprite protected | entry->getToSprite() |
| entry->toTrainer protected | entry->getToTrainer() |
| entry->toItem protected | entry->getToItem() |
| entry->toPokemon protected | entry->getToPokemon() |
| MapDBEntry::ind protected | entry->getInd() |
| MapDBEntry::width/height protected | entry->getWidth() / entry->getHeight() |
| MapDBEntry::warpIn protected | entry->getWarpIn() |
| MapDBEntry::warpOut protected | entry->getWarpOut() |
| MapDBEntry::sprites protected | entry->getSprites() |
| MapDBEntry::signs protected | entry->getSigns() |
| MapDBEntry::toTileset protected | entry->getToTileset() |
| MapDBEntry::toMusic protected | entry->getToMusic() |
| MapDBEntry::toSpriteSet protected | entry->getToSpriteSet() |
| MapDBEntry::glitch/special protected | entry->getGlitch() / entry->getSpecial() |
| pickRandom()->ind | pickRandom()->getInd() |
| *map->width (was optional) | map->getWidth() (plain int, -1 = unset) |
| if(map->width) (optional check) | if(map->getWidth() >= 0) |
| SpriteType::TRAINER | MapDBEntrySprite::SpriteType::TRAINER |
| ConnectDir::EAST etc. | MapDBEntryConnect::ConnectDir::EAST |
| new FontsDB (private ctor) | FontsDB::inst() |
| new Utility (private ctor) | Utility::inst() |
| new NamesDB | NamesPlayer::inst() |
| new NamesPokemonDB | NamesPokemon::inst() |
| new ExamplesPlayer/Pokemon/Rival | ExamplesPlayer::inst() etc. |
| NamesDB::randomName() | Names::inst()->player()->randomExample() |
| NamesPokemonDB::randomName() | Names::inst()->pokemon()->randomExample() |
| Qt::CTRL + Qt::Key_X | Qt::CTRL \| Qt::Key_X (operator+ deleted in Qt 6) |
| QString::SkipEmptyParts | Qt::SkipEmptyParts |
| QString::SplitBehavior::KeepEmptyParts | Qt::KeepEmptyParts |
| Missing DB_AUTOPORT → linker undefined symbol | Add DB_AUTOPORT macro + db_autoport.h include |
| undefined symbol: XxxDB::getSomething() | Declared but not implemented — add body to .cpp |
| area.h:NN: expected unqualified-id at a lone { | Recovery collapsed a block + ate the class-declaration line: area.h had its 10-line Q_DECLARE_OPAQUE_POINTER block + class MapDBEntry; + class SAVEFILE_AUTOPORT Area : public QObject crushed down to a single stray Q_DECLARE_OPAQUE_POINTER(AreaAudio*) directly followed by {. Restored from HEAD. Hunt for this flavor: scan headers for a Q_OBJECT not preceded by any class/struct decl within ~5 lines (a small Python pass), and compare Q_DECLARE_OPAQUE_POINTER counts vs HEAD (but note that phrase also appears in explanatory comments → comment-only count diffs are false alarms). |
| is_complete<T>::value static_assert / "Pointer Meta Types must point to fully-defined types or be declared with Q_DECLARE_OPAQUE_POINTER" (savefile moc) | Recovery dropped the full #includes of QML-traversed property-pointer types from the owning headers, leaving only forward declarations. In the unity moc build that makes the pointee incomplete (and poisons the type's own metatype downstream — the is_complete<Self> errors are secondary). Fix = restore the full #include at the owning header (the documented convention for traversed types; opaque/non-traversed types stay in savefile_autoport.h). Recovery dropped these (all confirmed against HEAD): player.h←playerbasics/playerpokedex/playerpokemon/itemstoragebox; savefileexpanded.h←player/area/world/storage; world.h←worldother; filemanagement.h←savefile; playerpokemon.h←pokemonparty; storage.h←itemstoragebox/playerbasics/pokemonstoragebox; savefile.h←savefileexpanded; mainwindow.cpp←pse-db/db.h. Find dropped includes fast: diff <(grep '#include' f) <(git show HEAD:f | grep '#include'). | | ld.lld: undefined symbol: DB::qmlProtect/qmlHook, FontSearch::clear/keepAnyOf (referenced by moc) | Recovery dropped the method bodies (header still declares them, so moc references them → link error). Recover the real bodies, don't guess. Source priority: working-tree .cpp for structure, then git show <ref>:<path> (HEAD first, then af883fd) — but note both can be truncated, so verify. DB::qmlProtect calls qmlProtect(engine) on all 26 sub-DBs (accessor order in db.cpp); HEAD's copy was cut off at SpritesDB, completed from the full list. DB::qmlHook mirrors Utility::qmlHook → context->setContextProperty("pseDB", const_cast<DB*>(this));. Before adding N sub-calls, confirm each callee is actually defined (grep '::qmlProtect' each .cpp) or you trade 1 link error for N. | | QFileDialog not found | Add Qt6::Widgets to CMakeLists target_link_libraries | | Q_DECLARE_OPAQUE_POINTER parse error | Add #include <QMetaType> before the macro | | convertToCode returns QVector<int>, need QVector<var8> | Cast element-by-element | | GameCornerDB::buyPrice private | GameCornerDB::inst()->getBuyPrice() | | DB constructor static-init deadlock | Remove load() from DB constructor — DB::loadAll() calls it | | './abstracthiddenitemdb.h' file not found (AUTOMOC, from util/hiddencoinsdb.h) | Recovery artifact: a stray truncated hiddencoinsdb.{h,cpp} was placed in pse-db/util/ (where abstracthiddenitemdb.h doesn't exist) and added to CMakeLists. Canonical files are pse-db/hiddencoinsdb.{h,cpp} (next to the abstract base). Removed the src/pse-db/util/hiddencoinsdb.cpp line from db/CMakeLists.txt; the orphaned util/ copies are inert. Lesson: after a bulk recovery, check for duplicate source basenames (find … | sed 's#.*/##' | sort | uniq -d). | | fatal error: pse-db/hiddenitemsdb.h: No such file or directory — builds locally, fails only on Linux CI (GitHub Actions) | Filename CASE mismatch. Windows/macOS default filesystems are case-INSENSITIVE so any include case resolves; Linux CI is case-SENSITIVE. The real file is pse-db/hiddenItemsdb.h (capital I — the one oddly-cased db file; everything else is lowercase), but tst_db_entry_getters2.cpp #included hiddenitemsdb.h. Fix = match the include to the on-disk case. Hunt for all such landmines before pushing: list tracked source whose basename has an uppercase letter (git ls-files '*.h' '*.cpp' | … basename matches [A-Z] → hiddenItemsdb, spriteSet, starterPokemon, tmHm, boot*), then case-sensitively grep for any lowercase include of each. Local builds can NEVER catch this — only Linux CI (or a case-sensitive check) does. | | 'buyPrice' is a private member + reference to non-static member function must be called + expression is not assignable (gamecornerdbentry.cpp) | Recovery glitch in gamecornerdb.h: the private data members QVector<GameCornerDBEntry*> store; and int buyPrice = 0; were mangled into bogus method decls int buyPrice() const; int sellPrice() const; (and the store member + GameCornerDBEntry friend were dropped). Truth is in the .cpp: getBuyPrice() returns buyPrice, getSellPrice() returns buyPrice/2 (no sellPrice member), store used throughout. Restored the two members + friend struct GameCornerDBEntry; (for the inst()->buyPrice = price write in the entry ctor). | | 'toGameCorner' is a protected member of 'ItemDBEntry' | Same deep-link friend pattern as MapDBEntry: the writer must be befriended. Added friend struct GameCornerDBEntry; to ItemDBEntry (for toItem->toGameCorner = this in GameCornerDBEntry::deepLink). | | use of undeclared identifier 'store' / 'ind' (or "did you mean 'strrev'?") in a *db.cpp | Recovery dropped the private container members from the DB header. Restore them in the private: section, e.g. fontsdb.h lost QVector<FontDBEntry*> store; and QHash<QString, FontDBEntry*> ind;. Confirm exact members via the .cpp's getStore()/getInd()/getStoreSize() bodies (and git show HEAD:<header>). NOTE: this is the DB-class store/ind (containers). An entry class's getInd() returns its int ind field — different thing, not this bug. | | no viable conversion from QScopedPointer<const FontSearch,...> to QScopedPointer<FontSearch,...> (fontsdb.cpp search()) | Recovery typo: body built QScopedPointer<const FontSearch, …> but the declared return type is non-const. Drop the stray const. | | QML load fail: Type X unavailable ... Cannot assign to non-existent property "onSomeSignal" | Recovery left a mismatched QML pair — a parent wires onSomeSignal: on a child that no longer declares that signal (an earlier-iteration leftover). Verify where the behavior actually lives now (here: the example toggle/>> moved to FullKeyboard's footer; NameFullEdit no longer emits toggleExample/reUpdateExample), then delete the dead wiring on the child. Don't re-add the signal unless that's truly where it belongs (UX = a design decision). | | Runtime: fonts/pills blank out, go red/empty, names stop saving after interaction (no crash) | The s13f DB::inst()->qmlProtect(engine) call in MainWindow::injectIntoQML() was dropped by recovery (the definition in db.cpp was also dropped — see other row). Without it, QML garbage-collects the shared parentless DB entries (FontDBEntry etc.) mid-session → dangling/null → blank fonts, red/empty picker pills, names stop saving until reboot. Restore DB::inst()->qmlProtect(engine); at the end of injectIntoQML() (needs #include <pse-db/db.h>). Non-deterministic GC = symptom looks random (e.g., only the pills you navigated to). | | Runtime hang (no crash/error) when rendering a name/example containing a variable tile (<player>/<rival>/<pkmn>…), e.g. "next random example" | FontsDB::splice() lost its out.remove(position); (recovery reverted the s13y "replace-not-insert" fix; af883fd had out.remove(ind)). Without it, expandStr re-expands the same tile forever. Re-add out.remove(position); before the insert loop. Verified against af883fd. | | definition of implicitly declared default constructor (Names::Names() etc.) | Recovery dropped the constructor declaration from the header. The .cpp defines X::X() but the .h no longer declares it. Re-add it — for singleton DB classes it goes in a private: section (Names();), per "all DB classes have private constructors". | | use of undeclared identifier 'results' / 'qmlRegister' (mapsearch.cpp) + later out-of-line definition does not match | Recovery truncated the header's tail: mapsearch.h lost getMaps()/mapAt()/qmlProtect()/qmlRegister() and the QVector<MapDBEntry*> results; member, and getMapCount was left non-const vs the .cpp's const. Reconstruct from the .cpp signatures + the parallel fontsearch.h (QML-interface block → public slots: qmlProtect → private slots: qmlRegister → private: QVector<…> results;). NOTE: clang stops at ~20 errors, so the pasted log undercounts — scan the whole .h↔<tt>.cpp for missing decls + const-mismatches in one pass (a small Python regex compare works). |
When you get "incomplete type" or "member access into incomplete type":
| Error | Fix |
|---|---|
| QML: Xxx is not a type / Type Xxx unavailable for a new .qml file | This project resolves QML types via app/app.qrc, not directory scanning — a new .qml file must be added as a <file> entry in app.qrc (and rebuilt) before it resolves as a type, even though debug builds load existing QML from disk. (s13v: TilePreview.qml.) |
| QML: keyboard filter behavior (current, s13z3+) | Filters are single-select radio buttons in a shared ButtonGroup (incl. an All). Backend: FontSearch::keepAnyOf(...) for one category, startOver() for All. (Evolution: tristate → AND s13v → OR/union s13y → radios s13z3 → +All s13z4. a design decision each step.) |
| App freezes/hangs when rendering a lone variable tile (e.g. rival/player name code) | FontsDB::splice infinite loop (s13y): it inserted the expansion but didn't remove the original expandable code, so the code re-expanded forever. Fix: out.removeAt(position) before inserting (replace, not insert). Affects expandStr → font/name preview rendering. |
| Tileset picker: the very last tile can't be hovered/clicked (off-by-one) | FontsDB::fontAt(n) is 1-based (getStoreByVal does ind--), so valid indices are 1..fontCount() inclusive. Bounds checks must be id < 1 \|\| id > fontCount(), NOT id >= fontCount() (that drops the last valid tile). Fixed in TilesetPicker.qml s13z2. |
| QML: property var detailView: null never fires hover | Parent defines detailView but doesn't pass it to child — add detailView: top.detailView in child instantiation. |
| QML: onStrChanged: str = top.str — str never propagates up | Arrow is backwards. str = top.str resets to old value. Should be top.str = str to push new value up. |
| QML: anchors.top: item.Top — anchor has no effect | Capital T in .Top is invalid. Use item.top (lowercase). |
| brg.file.data.dataExpanded = undefined (whole chain) | REAL root cause (session 13): these QObject types were Q_DECLARE_OPAQUE_POINTER'd (in savefile_autoport.h and area/area.h), which forces IsPointerToTypeDerivedFromQObject<T*> = false → Qt stores them as opaque non-QObject values → QML reads undefined. Fix: remove the opaque decl for the QObject type AND fully #include its header wherever its pointer is in a Q_PROPERTY/signal/slot/Q_INVOKABLE. qRegisterMetaType / qmlRegisterUncreatableType do NOT override the opaque decl. |
| TypeError: Property 'X' of object Y is not a function | Y.X is a Q_PROPERTY READ (not Q_INVOKABLE). Remove Q_PROPERTY, add Q_INVOKABLE |
| TypeError: Cannot read property 'player' of undefined | Property chain traversal fails — see dataExpanded = undefined (opaque-pointer) entry above |
| randomName is not a function | Should be randomExample() (from AbstractRandomString). Check if it's Q_PROPERTY instead of Q_INVOKABLE. |
| isNaN / NaN === NaN never triggers | NaN === NaN is always false. Use isNaN(x) |
| unsigned int property blank in TextField | Call .toString() explicitly in Component.onCompleted + Connections handler |
| parent.width throws in ListView delegate | width: parent ? parent.width : 0 |
| Number TextField clips its value (shows 1 char / blank) | Width ignored the field padding. Use width: N * font.pixelSize + leftPadding + rightPadding (N = max digits). Money/Coins avoid this by using width: child.implicitWidth. |
| Stacked Material fields overlap vertically (e.g. Coins on Money) | Layout used anchors.top: prev.top; topMargin: <fixed> tuned for shorter Qt 5 fields. Anchor below instead: anchors.top: prev.bottom; topMargin: 5 — auto-adapts to Qt 6 field height. See qt-patterns "Material 3 control heights". |
| Number in a Row not vertically centered | Don't use anchors.verticalCenter inside a Row (ignored). Use verticalAlignment: TextInput.AlignVCenter on the field (and remove fixed topPadding nudges). |
| randomName' of object NamesPlayer is not a function | Method is randomExample() (from AbstractRandomString). Fixed in NameDisplayMenuNoTileset.qml (s10) AND NameDisplayMenu.qml (s13e). Grep for other .randomName( callers. |
| TilesetPicker.qml Cannot read property 'name' of null | brg.fonts.fontAt(id) can return null even for an in-range id. Guard: var f = brg.fonts.fontAt(id); if(!f) return; ... f.name. |
| Parameter "val" is not declared. Injection ... deprecated | Qt 6 deprecation, not an error (the signal declares val). To silence: onChangeStr: (val) => {...} or function onChangeStr(val) {...}. |
| Cell/sprite click does nothing but a hover-only button works | The cell MouseArea had no onClicked — only the hover button did. Add onClicked to the MouseArea (guard placeholder slots). See PokemonBoxView.qml (s13d). |
| Something "works, then breaks after clicking around; reboot fixes it" (fonts/names blank, dropdowns empty, name stops saving) | QML garbage-collected a parentless C++ object. DB entries (FontDBEntry, move/species/item entries) live in DB QVectors with no QObject parent → QML treats them as JS-owned and GCs them mid-session → dangling pointers. Fix: call DB::inst()->qmlProtect(engine) once at boot (it sets CppOwnership on every entry). It existed but wasn't called — wired into MainWindow::injectIntoQML (s13f). |
| Crash (read access violation at 0xffff…ffff) in a model data() / …At() after clicking an item, esp. after opening+closing its editor | Q_INVOKABLE returned a parentless QObject → QML gave it JavaScriptOwnership → GC freed it → dangling pointer. (Q_PROPERTY returns are safe; Q_INVOKABLE returns are NOT.) Fix: wrap the return in qmlCppOwned() (pse-savefile/qmlownership.h), which sets CppOwnership. Fixed across all 13 savefile …At() methods + the 2 model wrappers (s13g/s13h). Standing rule: any NEW Q_INVOKABLE returning a QObject must be wrapped in qmlCppOwned(). See qt-patterns. REGRESSED + re-fixed 2026-06-08: the recovery corruption had stripped the wrap (and the #include <pse-savefile/qmlownership.h>) from the app-layer model PokemonStorageModel::getBoxMon()/getPartyMon() — re-wrapped both. But the accessor wrap was never the whole story: PokemonBox/PokemonMove are parentless QObjects whose ctors had a commented-out TODO meant to set CppOwnership at construction. Relying only on per-accessor qmlCppOwned() left exposure windows → an intermittent use-after-free: a mon in getCurBox()->pokemon GC'd by QML, then read dangling by either hasChecked() (stack onReset→hasCheckedChanged→checkStateDirty→hasChecked→el->property()) or PokemonStorageModel::data()→PokemonBox::toData() (reads species; stack via a QML binding update). Root-cause fix (realised the TODO): QQmlEngine::setObjectOwnership(this, QQmlEngine::CppOwnership) (static, no engine needed) in the PokemonBox and PokemonMove ctors — PokemonParty and new PokemonBox() both chain the default-arg PokemonBox ctor, so all mons/moves self-protect from birth. Containers still free via deleteLater() (no leak/double-free); accessor wraps kept as redundancy. Same precedent as DB::qmlProtect. Standing rule upgraded: a parentless QObject exposed to QML should take CppOwnership in its constructor, not only at accessors. Rebuild required (C++). |
| Crash in memcpy/memcmp at SaveFile::setData (savefile.cpp) when opening a recent file (stack: openFileRecent → setData) | Null/short buffer deref. FileManagement::readSaveData returns nullptr when QFile::open fails (recent path moved/deleted/locked) and setData then did memcpy(this->data, nullptr, …). Fixes (s14): (1) setData early-returns on a null source; (2) readSaveData rejects files whose file.size() is less than SAV_DATA_SIZE (at-least, not exact — larger files load their first 32 KB), returns nullptr on failure, and records the real technical detail (QFile::errorString() for open failures, the actual byte-count for short files); (3) openFile/openFileRecent/reopenFile route through loadData() which guards null, emits loadError(), and never mutates the live save on failure; (4) startup pruneRecentFiles() drops recent entries that can't be opened ("prune" not "scrub" — "scrub" already means wipeUnusedSpace); (5) a FileError.qml modal (router "fileError", raised from App.qml's Connections on brg.file.onLoadError) shows a plain-English reason centred (lastErrorMessage) with the real one-line detail (lastErrorDetail) small/muted below — no made-up error codes. |
| Save corruption: on every save of a progressed file, the last data byte of Box 6 (0x5A4B) gets clobbered and the bank-2 box checksum is wrong | Off-by-one in recalcChecksums() (savefiletoolset.cpp). It did data[0x5A4B] = getChecksum(0x4000, 0x1A4B); the bank-2 all-boxes checksum actually lives at 0x5A4C over range 0x1A4C (6 boxes × 0x462) — per recalcBoxesChecksums(), the saves/structure.bt map, and real RBY sBank2AllBoxesChecksum. Fix: data[0x5A4C] = getChecksum(0x4000, 0x1A4C). The if(data[0x284C]==0) return guard above it (skip box checksums when boxes were never formatted) is intentional — it mirrors the game. Found + verified by tst_roundtrip 2026-06-07. |
| Crash (access violation) in MapSearch::isType() during map randomization (AreaWarps::randomize → isType("Cave")) | isType() removed entries whose toTileset == nullptr but then fell through and dereferenced that null toTileset on the very next line. Add continue; after the null removal — the sibling notType() already guards this way. Found by tst_verbs (randomizer path) 2026-06-07, fixed in db/.../util/mapsearch.cpp. |
| Crash (access violation 0xc0000005) in Daycare::~Daycare() → QObject::deleteLater when destroying an empty Day Care | pokemon is nullptr when no mon is deposited (load() only allocates it when getByte(0x2CF4) > 0; reset() leaves it null), but the destructor called pokemon->deleteLater() unconditionally → null-this deref. Masked in the running app (a SaveFile is only torn down at process exit, where the crash is swallowed). Fix: guard if(pokemon != nullptr) — matches the guard reset() already has. Found by tst_roundtrip (2026-06-07), fixed in daycare.cpp. |
| Crash (access violation) in HoFPokemon::load() (e.g. via HoFRecord::randomize() / pokemonNew() / new HoFPokemon) | The null check if(saveFile == nullptr) return; came after auto toolset = saveFile->toolset;, so a default-constructed (null-savefile) HoFPokemon dereferenced null. Move the null check before the deref (same bug class as Daycare::~Daycare). Found by tst_randomizer (full randomize) 2026-06-07, fixed in hofpokemon.cpp. |
| Recent-files list keeps 6 entries instead of the intended 5 (cap off-by-one) | FileManagement::processRecentFileChanges() did newList.append(file); if(newList.size() > MAX_RECENT_FILES) break; — appending then breaking on > retains MAX+1. mainwindow's menu already bounded itself at MAX_RECENT_FILES (and its shortcut array is MAX-sized), but RecentFilesModel reads recentFilesCount() directly so the QML list leaked the 6th path. Confirmed the extra slot was not an intentional sentinel before changing. Fix: >= MAX_RECENT_FILES. Found + guarded by tst_filemanagement::recents_capBoundsTheList 2026-06-08, fixed in filemanagement.cpp. |
| Custom ComboBox popup won't scroll — long list clips at screen edge and rubber-bands | The custom popup set implicitHeight: contentItem.implicitHeight and the inner ListView had implicitHeight: contentHeight, so the popup grew to the full list height → ListView height == its content → nothing to flick. Cap it: height: Math.min(contentItem.implicitHeight + 2, 280) (+2 = padding:1 ×2). Matches Qt's default ComboBox popup. Fixed all 7 Select* (s13k). |
| Monochrome SVG icon tinted with MultiEffect{colorization} stays dark/black | colorization scales the tint by the source's luminance; a black SVG (luminance ≈ 0) ⇒ ≈ black regardless of colorizationColor. Push it white first: add brightness: 1.0 before colorization: 1.0; colorizationColor: <color>. Fixed the hover pen in PokemonBoxView.qml (s13k). |
| QML control "doesn't react" — clicking/changing it does nothing, but Connections-driven value sync still works | Handler written as function onActivated() {…} (or onMoved/onClicked/onTextChanged/onTriggered/onCheckedChanged) directly on the control declares an unused method — the signal never calls it. Use property syntax onActivated: {…}. (function onX() is correct ONLY inside Connections{}.) Pervasive in the Pokémon editor, fixed s13l (46 handlers). Tell: working controls used onX:, dead ones used function onX(). |
| Converting a dead function onCheckedChanged() on a CheckBox | Use onToggled: (user-only), not onCheckedChanged: — the latter also fires on the programmatic checked = … sync on open and would run side effects (e.g. makeShiny() rewriting DV bytes) just from viewing data. Save-integrity rule. |
| ⋮ / icon button renders as a tiny sliver | The icon SVG (e.g. ellipsis-v, a tall/narrow glyph) aspect-fits inside icon.width × icon.height. Setting only icon.width: 7 leaves icon.height: 15 (IconButtonSquare default) and the tall SVG fits to a thin sliver. Give it a real square-ish size (icon.width: 16; icon.height: 16). Seen on the Future-Shiny ⋮ (s13r). |
| Material control (CheckBox/Button) keeps a Layout row tall despite Layout.preferredHeight | The control's implicit/minimum height (~40px touch target) floors the layout. Set Layout.minimumHeight: 0 on the control (and a Layout.maximumHeight on the row) to actually shrink it. |
| QML: Cannot assign to non-existent property "topInset" (or similar) after editing a shared component | The component's root type changed and lost a property a caller sets. E.g. NameEdit went TextField→RowLayout (s13s), breaking NameFullEdit which set topInset (s13u). Fix: expose the needed property (property alias topInset: innerField.topInset) or revert the root type. Check every instantiation before changing a shared component's root. NameEdit is used by NameFullEdit (field-only) AND NameDisplay (popup). |
| QML ID not accessible in child file | Qt6 strict ID scoping — pass as property var through hierarchy |
| Detected function "onXxx" — no signal matches | Target object is null/undefined. Fix root cause (property chain), then add null guard: target: cond ? obj : null |
| File truncated mid-line | Check brace balance: text.count('{') - text.count('}'). Rewrite from memory if short. |
| App hangs 40 seconds on startup | Likely MSAA or qt_add_qml_module() conflict. See decisions/architecture.md |
| Window never appears | QQuickWidget::setSource() hung. See decisions/architecture.md |
| Footer button rounded/shadowed in Qt 6.5+ | flat: true + topInset: 0; bottomInset: 0; leftInset: 0; rightInset: 0 |
| IconButtonSquare renders rounded/shadowed | It extended RoundButton via IconButtonRound. Change to Button { flat: true; topInset:0; bottomInset:0; leftInset:0; rightInset:0 } |
| randomName is not a function | Correct method is randomExample() (from AbstractRandomString) — not randomName |
| brg.file.data.dataExpanded = undefined | Superseded (session 13). rebuild + qRegisterMetaType did NOT fix it. Real cause is Q_DECLARE_OPAQUE_POINTER forcing the chain types to be treated as non-QObjects — see the opaque-pointer entry above. |
| Many QML components show blank data after fresh build | Source files may be truncated. Run brace-balance check across all .qml/.cpp/.h. Repair via byte-level Python append. |
| unknown type name 'FontFilter' in fontsdb.h | FontFilter was a private type lost in truncation. Replace with int in private splice() declaration. |
| no member named 'getMapCount'/'isCity'/'notCity' in 'MapSearch' | Moc was generated from old complete header. Add method declarations matching Q_PROPERTY READ specs. |
| DB header: use of undeclared identifier 'store' / 'ind' | Private data members were truncated out of the header. Add QVector<XxxDBEntry*> store; and QHash<QString, XxxDBEntry*> ind; back to the private: section. | | DB header: int buyPrice() const as private method, .cpp uses it as data member | Method declaration was wrong (truncation). Replace int buyPrice() const; int sellPrice() const; with int buyPrice = 0; data member. sellPrice is computed, not stored. | | 'buyPrice' is a private member + expression is not assignable | Both: header declared it as a method, not a field. Fix header first, then add friend struct GameCornerDBEntry; to GameCornerDB. | | 'toGameCorner' is a protected member of 'ItemDBEntry' | Add friend struct GameCornerDBEntry; to ItemDBEntry's friend list (alongside ItemsDB and MapDBEntrySpriteItem). | | 'toEvolvePokemon' is a protected member of 'ItemDBEntry' | Add friend struct PokemonDBEntryEvolution; to ItemDBEntry's friend list. | | 'toTeachPokemon' is a protected member of 'ItemDBEntry' | Add friend struct PokemonDBEntry; to ItemDBEntry's friend list. | | undefined symbol: DB::qmlProtect / DB::qmlHook | Declared as public slots → MOC always generates dispatch entries. Must add bodies in db.cpp even if not yet called. | | definition of implicitly declared default constructor | Header is missing the private constructor declaration. Compiler synthesized one; .cpp's explicit definition then conflicts. Add private: XxxDB(); to the header. | | out-of-line definition of 'getMaps' does not match any declaration | Method implemented in .cpp but not declared in header (truncation). Add matching declaration to header. | | no matching constructor for initialization of 'MapSelectModel' | Default member initializer = new MapSelectModel has no matching ctor. Real init is in the constructor's mem-init-list in .cpp. Change to = nullptr. | | mapData->isOutdoor() — no member named 'isOutdoor' in 'MapDBEntry' | MapDBEntry has no such method. The concept lives on TilesetDBEntry: mapData->getToTileset()->typeAsEnum() == TilesetType::OUTDOOR. | | ignoring return value of function declared with 'nodiscard' on DB::inst() | Call is for side effects (bootstrapping). Use (void)DB::inst(); to suppress. | | ignoring return value on QFile::open() | Qt 6 marked QFile::open() [[nodiscard]]. Wrap: if(!file.open(...)) return; — also improves correctness. |
| Error | Fix | |----—|--—|
| Symptom | Fix |
|---|---|
| Dual-type Pokemon silently loses its 2nd type on level/EV edits | PokemonBox::update() had a bare else type2 = (*record).toType1->ind; after if(resetType && record->toType2), so any update() with resetType=false (maxLevel/maxEVs/resetEVs/reRollEVs/manualLevelChanged) overwrote type2 with type1. Wrap the whole type2 (re)derivation in if(resetType){ … } so a non-reset update leaves type2 alone. (2026-06-08, tst_pokemonbox.) |
| PokemonBox::isCorrected() always false for some single-type species | The DB stores some single types with toType2 == toType1 (not null); update() collapses that to type2=0xFF, but isCorrected demanded type2 == toType2->ind. Treat a record as dual-type only when toType2->ind != toType1->ind; for single types accept 0xFF or type1. (2026-06-08.) |
| TypesModel::data() crashes on the placeholder row 0 with an undeclared role | Row 0 is the "-----" no-type placeholder; it returned for IndRole/NameRole but otherwise fell through to getStore().at(index.row()-1) == at(-1). QML only queries declared roles so the UI never hit it, but it's a latent crash. Fix: return QVariant(); for the placeholder row's other roles (matches PokemonStartersModel's guard). Other select models have no placeholder row / already guard. (2026-06-08, tst_select_models.) |
| Every type combo displays the type one row early (Charizard reads Ghost/Fighting, not Fire/Flying) — data is correct, only the display is wrong | Same "-----" placeholder at model row 0: a type at store index i lives at model row i+1, but TypesModel::valToIndex(val) returned the bare store index i. The combo's currentIndex = valToIndex(type) then selected the row before the real one (Fire→Ghost, Flying→Fighting). Fix: ret = i + 1. The underlying mon DATA was always right (verified by tst_storage loading BaseSAV: party Charizard is type1=20/type2=2). Beware tautological tests: a correctTypes test that compared the result to chz->toType1->ind (the same source the method reads) passed regardless of the bug — assert the literal type ids (Fire=20/Flying=2). valToIndex is only used by the Select-type combos, so the +1 is safe app-wide. (2026-06-17, tst_models/tst_pokemonbox/tst_storage.) |
| MapDBEntryConnect::xAlign() always returns 0 (its math is dead code) | The guard was if(toMap == nullptr \|\| toMap->getWidth()) return 0; — missing the <= 0 that every sibling (yAlign/stripSize/mapPos/window/stripLocation) has. So a non-zero width (every real map) returns 0, and the body only runs when width is 0. Fix: toMap->getWidth() <= 0. Confirmed against the connection-data formulas in gen1-knowledge.md (xAlign mirrors yAlign). Gated: connection-render math for the disabled Maps feature. (2026-06-08, tst_db_entry_getters2.) |
| MapSearch::hasDynamicSpriteSet()/noDynamicSpriteSet() crash on maps with no sprite set | spriteSet is an index where -1 = none (not 0). The guards used !entry->spriteSet (true only for 0), so for a spriteSet == -1 map they fell through and dereferenced the null toSpriteSet → crash. hasSpriteSet/noSpriteSet had the same -1-vs-0 sentinel bug (wrong results, no crash). Fix: test spriteSet < 0 for "none", and guard toSpriteSet == nullptr before isDynamic(). (2026-06-08, tst_mapsearch_predicates. Gated: only the disabled map search/randomizer uses these.) |
| AreaTileset::loadFromData(map,…) crashes on a null map / loads zeros on a real map | The auto tileset = (map == nullptr) ? map->getToTileset() : nullptr; ternary is inverted — both branches wrong (null map → null deref; real map → discards the tileset). Correct: (map == nullptr) ? nullptr : map->getToTileset();. Masked today (loadFromData is the disabled Maps path). (2026-06-08, tst_area_logic.) |
| PokemonBox::newPokemon(Random_Pokedex) can never roll Bulbasaur | The "dexN" keys are 0-based (dex0=Bulbasaur … dex150=Mew; no dex151). It used rangeExclusive(1, pokemonDexCount) → [1,150], missing dex0. Use rangeExclusive(0, pokemonDexCount) → [0,150] (all 151). AreaPokemonWild::randomize() already does this correctly. (2026-06-08.) |
| "Reset EVs" UI action greyed out on an EV'd mon; isMinEvs() true with non-zero EVs | PokemonBox::isMinEvs() used \|\| (true if ANY one of the five stat-exp is 0). "Minimum EVs" means ALL zero -> use && (symmetric with isMaxEVs()'s all-0xFFFF). StatsTab.qml binds enabled: !isMinEvs on Reset EVs, so the \|\| disabled it whenever a single stat-exp was 0. (2026-06-08, tst_pokemonbox.) |
| Pokemon with fewer than 4 moves never reads as "healed" (heal indicator wrong); isPokemonReset() never true | Root cause: PokemonBox::isMaxPP()/isMaxPpUps() looped all 4 slots and counted an EMPTY slot (moveID 0) as "not maxed", which propagates into isHealed(). Fix both to skip empty slots: if(moves[i]->moveID > 0 && !moves[i]->isMaxPP()) (mirrors isMaxedOut()'s existing guard). Then isPokemonReset() simplifies to iterate min(4, toInitial.size()) real moves (empty after), check ppUp==0 (a reset mon has 0 PP-Ups, NOT isMaxPpUps()), and reuse isHealed(). (2026-06-08, tst_pokemonbox.) |
| heap-use-after-free in ItemMarketEntryPlayerItem::_itemWorth() (only ASan/Linux CI caught it; Windows kit can't run ASan) | A market row cached a raw Item*/ItemStorageBox* it does NOT own (they live in the save's item box). ItemMarketEntry::totalWorth() walks the static instancesCombined registry calling cartWorth()→itemWorth()→_itemWorth() on EVERY registered entry — including stale entries left behind by a torn-down ItemMarketModel (the registry is shared across all model instances), or entries whose Item was freed on a box rebuild. Reading the freed Item = UAF. The Windows build reads stale-but-valid memory so it never crashed; Linux ASan (the linux-asan CI job) is the only thing that catches this class (the s13f/g/h QML-GC family too). Fix: hold QPointer<Item>/QPointer<ItemStorageBox> (auto-null on destroy) + guard every deref (if(!toItem) return <default>;). Stale entry → harmless default, not a dangling read. (The destructor already deregisters from the static list, so the entry itself is never read after free.) Reached via ItemMarketModel::moneyLeftover() during a Pokemart screen push; surfaced by the GUI tests under ASan 2026-06-13. Follow-up (tracked): the static instances/instancesCombined registries are cross-instance global state — making them per-model would remove the lingering-entry hazard at the source. |
| MainWindow::setupShortcuts() left the otherShortcuts member empty | auto os = otherShortcuts; COPIED the member, then inserted/connected on the copy — so the member stayed empty (the QShortcuts only survived via their QObject parent this). Harmless at runtime (parent owns them) but a dead member + a destructor cleanup loop that iterated nothing. Fixed while extracting the key sequences into the shared boot/shortcutdefs.h (single source of truth for tst_shortcuts): write into the member directly (auto& os = otherShortcuts;). 2026-06-13. |
| PlayerBasics::badgeCount() always returns 8 (a 0-badge save reports 8) | It was a stub: return maxBadges; — ignoring the actual badges[] flags, contradicting its "how many badges are set" contract. Fix: count the set flags (for i: if(badges[i]) ++count). Only used by tests today (no QML/logic depends on it), but a Q_INVOKABLE that lies. tst_player_basics still passes (BaseSAV has all 8). Found by tst_synthetic_fixtures (new_zeroed) 2026-06-13. |
| Opening a corrupt/garbage save (e.g. all-0xFF) crashes on load (null deref in PokemonStorageSet::loadSpecific, boxes[toBox]) | Storage::load() reads the current-box index from save byte 0x284C (& 0x7F, so 0..127) but never range-checks it before it flows into loadSpecific()/boxAt() which index boxes[] (valid 0..11, two sets of setMaxBoxes=6). A malformed save (all-0xFF → 127 → set-B offset 121) indexes boxes[121] out of bounds → access violation. Fix (graceful degradation, byte-fidelity-safe): clamp if(curBox >= setMaxBoxes*2) curBox = 0; right after the read. A valid save is always in range, so it's a no-op for real saves (roundtrip unchanged); only garbage is defaulted. Found by tst_dirty_fixtures (all_ff) 2026-06-13. |
| Opening the Pokemart screen aborts (Q_ASSERT in QList::at → qFatal) | ItemMarketModel::moneyLeftover()/totalCartWorth()/canAnyCheckout() read itemListCache.at(0) as an aggregate proxy but never guarded the empty cache. The Pokemart QML footer/summary bind these (canAnyCheckout, moneyLeftover, totalCartWorth), and the screen's StackView.push instantiates + evaluates those bindings before the model's pageOpening slot (on the same openNonModal signal) has built the list — so they hit an empty list and abort. Two-part fix: (1) the three accessors return sane empties (moneyStart() / 0 / false); (2) buildList() runs in the ctor so the model is valid from construction (not dependent on navigation slot order). The harness loads the save before the Bridge, so it misses the initial dataExpandedChanged build — which is what exposed it. (2026-06-13, tst_gui_navigation.) |
| QML screen "won't open" / errors on navigation, but all C++ tests pass | A QML load-time error (not visible to non-GUI unit tests, which never instantiate QML). Capture it by launching the app with stderr redirected and reading the qCritical() << "[QML]" / QQmlComponent: Component is not ready lines (Start-Process -RedirectStandardError app.err.log). Two were hit redesigning the Credits screen (2026-06-13): see the next two rows. Lesson: gate main on a QML-load smoke check, not just ctest. | | About.qml:NN: Cannot override FINAL property → Component is not ready → screen never opens | A custom property re-declared a name that a QtQuick.Controls base already defines as FINAL. Here Page already has contentWidth (also contentHeight, contentItem, header, footer, title); declaring readonly property int contentWidth on the Page root made the whole component fail to load. Fix: rename the custom property (→ colWidth). Don't reuse Control/Page property names. (2026-06-13.) Now gated by tst_qml_screens (loads every screen, fails on any QML error/warning). | | TypeError: Property 'X' of object QQuickAnchorLine() is not a function (calling a root function from a delegate) | The root was id: top, and inside delegates that have anchors, bare top resolves to the top anchor line (a QQuickAnchorLine), not the id — so top.someFunc() throws. Avoid id: top (also bottom/left/right/fill). Fix: rename the id (→ root). (2026-06-13, Credits screen.) Now gated by tst_qml_screens. |
Found by the new clang-tidy gate (scripts/lint.*, .clang-tidy). See notes/plans/testing.md → "Static analysis / linting".
| Symptom | Fix |
|---|---|
| PokemonBox::expLevelRangePercent() returns 0.0 everywhere except the very top of a level (progress reads empty) | return curExp / expEnd; divided two var32s in integer math, then widened to the float return — the fraction was always truncated. Fix: static_cast<float>(curExp) / static_cast<float>(expEnd), plus an expEnd == 0 divide-by-zero guard. Regression: tst_pokemonbox now puts exp at the window midpoint and requires ~0.5. (bugprone-integer-division + clang-analyzer-core.DivideZero.) |
| Possible null deref in PokemonBox::update() type2 else-branch | The if(resetType){ if(toType2)… else type2 = (*record).toType1->ind; } else dereferenced toType1 unguarded, though line ~964 guards the same pointer for type1. Fix: else if(record->toType1). Valid records are unaffected (toType1 set); only a record with neither type resolved is now safe. (clang-analyzer-core.NullDereference.) |
| New move slots kept a random PP-Up count instead of 0 (dead store in PokemonMove ctor) | if(move==0){ randomize(); ppUp = 0; } — randomize() assigns a random 0-3 ppUp; the ppUp = 0 meant to reset it but assigned the constructor parameter (shadows the member), a no-op. So new empty-slot moves silently kept a random PP-Up count. Fixed to this->ppUp = 0; (confirmed intended with Twilight 2026-06-22); scoped to construction, so the deliberate randomize-everything actions are unaffected. (clang-analyzer-deadcode.DeadStores.) |
| signed/unsigned char compare in PokemonStorageSet::load/save | i == skipInd compared var8 (unsigned) with svar8 (signed). The skipInd >= 0 guard makes it safe; made explicit with i == static_cast<var8>(skipInd). (bugprone-signed-char-misuse.) |
| int-multiply widened to size_t in WorldScripts::reset() | memset(curScripts, 0, scriptCount * 2) computed the size in int. No overflow (fixed small count) but made explicit: static_cast<size_t>(scriptCount) * 2. (bugprone-implicit-widening-of-multiplication-result.) |
| copy-in-loop / implicit-conversion-in-loop | for(auto x : …) copying each element where a const& suffices (fontpreviewprovider, eventdbentry), and for(const QJsonValue& : QJsonArray) creating a per-iteration temporary (scripts, spriteSet) → for(const auto& …). (performance-for-range-copy, performance-implicit-conversion-in-loop.) |
| PokedexModel::dexSort() switch without default | Added default: break; (order unchanged for any unexpected value). (bugprone-switch-missing-default-case.) |
| ~ItemMarketEntry() calls pure-virtual _whichType() (latent UB) | Suppressed + flagged, not fixed — safe on every real path (cached during the object's life) but UB if destroyed with whichType() never called. A real fix is a lifetime refactor of this UAF-historied area; deferred to Twilight. NOLINT + full note at the call site. (clang-analyzer-cplusplus.PureVirtualCall.) |
| Symptom | Fix |
|---|---|
| GUI test: cannot initialize a parameter of type 'QWidget *' with ... 'QQuickView *' on QTest::keyClicks(view, text) | QTest::keyClicks(const QString&) is a QWidget-only overload — there is no QWindow* version. QQuickView is a QWindow, so type a string by looping the QWindow keyClick(char) overload per character (guiapp.h keyType()). The single-key keyClick(window, Qt::Key) / mouseClick(window, …) DO have QWindow* overloads and are fine. (2026-06-13.) |
| GUI harness can't find the inner appBody StackView via findChildren + className contains("QQuickStackView") (returns null) | A plain StackView { id: appBody } instance has the QML-generated metaobject class name StackView_QMLTYPE_N, NOT QQuickStackView. Match the "StackView" substring instead (covers both the generated name and the C++ class). App.qml's root is App_QMLTYPE_N (no "StackView") and excluded from findChildren anyway, so the only match is appBody. (2026-06-13, guiapp.h.) |
| GUI/QML offscreen test fails on QFontDatabase: Cannot find font directory …/lib/fonts | Benign: Qt 6 ships no fonts and the offscreen platform has no system font dir, so this logs once at first text render (text still lays out via the fallback). Add the substring to the test's isBenign() allowlist — it's an environment artifact, not a QML/app bug. (2026-06-13, guiapp.h + tst_qml_screens.) |
| tst_qml_screens: a detail screen (pokemonDetails/mapDetails) floods null-binding TypeErrors when loaded cold | Detail screens (Screen.homeBtn == false) only ever appear AFTER a parent selection sets their boxData/map property — cold-loading them with that property null is an unreachable state. The smoke test gives them load-only coverage (compile/instantiate clean, no component.errors()); their runtime bindings are verified in the real selection-backed state by the detail-flow GUI test. Don't fail the smoke test on cold-load detail warnings (and don't add reflexive null guards for a state navigation never produces). (2026-06-13.) |
1.17.0