- Generated by
1.17.0
|
Pokered Save Editor 2
Pokemon Red & Blue save file editor - Qt 6 C++/QML
|
The story of how this project was revived. Useful context for understanding why certain decisions were made and why the codebase looks the way it does.
Before this file: the 2019–2020 origin story — three from-scratch rewrites, the JavaScript detour, and the library/DB refactor that shaped today's architecture — lives in origins.md. This file picks up where that one ends (the abandoned March-2020 state). Read origins.md first for the full arc.
The repository had 592 commits, last pushed March 2020. It was mid-refactor — abandoned when its complexity outgrew what a single developer could sustain.
What the code looked like at revival:
None of these were fatal individually. Together they made the project unbuildable in 2026.
Goal: Get the common and db sub-libraries compiling cleanly under Qt 6.
Work done:
Result: common and db compiling clean.
Goal: Fix the remaining two sub-libraries. Hundreds of errors, all following the same patterns — Qt 6 MOC strictness, protected member access via old direct field reads, and non-static method calls against static APIs.
Major patterns fixed throughout:
New API added during this phase: GameCornerDB::getSellPrice() — the call site referenced a sellPrice field that never existed. The class comment already described the policy (sell-back is half buy price), so this was added as a proper method.
Also found and fixed:
Result: All four layers compiling clean.
Goal: App compiled but hung 40+ seconds on startup, then had to be force-killed.
Three independent root causes discovered, all fixed:
Hang 1 — qt_add_qml_module() conflict: CMake had both app.qrc (QML at qrc:/ui/app/...) and qt_add_qml_module() (same files at qrc:/qt/qml/App/...). Generated module init code conflicted with existing QRC paths, hanging QQuickWidget::setSource() inside the MainWindow constructor. Fix: removed qt_add_qml_module() entirely.
Hang 2 — QSurfaceFormat MSAA: boot/boot.cpp set QSurfaceFormat::setDefaultFormat() with setSamples(8). QQuickWidget renders to an offscreen FBO — MSAA on the default format hangs Windows GPU drivers during context creation (40+ second timeout). Fix: removed QSurfaceFormat setup entirely.
Hang 3 — DB static-init deadlock: The main hang. Every DB constructor called load(). In Qt 6, static local init is mutex-guarded. Re-entering inst() from within its own constructor deadlocks. Only two entries triggered this directly — creditdbentry.cpp and gamecornerdbentry.cpp — but the fix was applied to all 21 DB constructors for safety. DB::loadAll() is now the sole caller of each DB's load().
Result: Window appears. App runs.
Goal: Window showed but data fields were blank everywhere. Work through QML bugs screen by screen.
Key fixes:
Still not working after this phase: brg.file.data.dataExpanded still returned undefined in QML — Trainer Card and Pokemon names blank.
This happened twice at different scales — a smaller version first, then a larger sweep.
What happened: During active development, an editor crash caused ~22 QML files and ~15 C++ source files to be truncated mid-line. The truncation was silent — files were present and non-empty but ended in the middle of a statement or function. This caused:
Detection: Python brace-balance checker + check that file ends on a closing }. Repair: Python byte-level append — match exact truncated suffix, append missing bytes, matching the line endings already in the file.
The dataExpanded = undefined bug that persisted through QML testing was caused by this: the binary had been compiled from a truncated bootQmlLinkage.cpp without type registrations. After full truncation repair and rebuild, this chain works correctly.
The signal parameter was never the cause — SaveFile::dataExpandedChanged(SaveFileExpanded*) is correct. See decisions/rejected.md.
The project's AI-assisted development sessions originally captured notes in a folder called ai/. This was reorganized into the current notes/ structure — neutral, readable by anyone, organized by topic rather than session number.
The session-log format was the starting point; this file (and the rest of notes/) represents the synthesized, topic-organized version of everything learned.
Goal: Fix the three known broken areas in the full-keyboard (name editing) UI.
Bugs fixed:
SearchContainer.qml — "Control filter clears everything": reSearch() was calling andNormal() + andControl() together, which ANDs to empty (a char can't be both simultaneously). The real semantic is: within a category axis, multiple Checked items should OR. Rewrote reSearch() with axis-aware logic — type axis (normal/control/picture) and size axis (singleChar/multiChar/variable) each use OR semantics internally; across axes they AND. Also fixed a typo: singleSearchStateState → singleSearchState.
SearchRoot.qml — character clicks don't append + DetailView never shows: Two bugs in one file. (1) onStrChanged: str = top.str had the arrow backwards — it reset the value to the old string instead of propagating the new one up. Fixed to top.str = str. (2) detailView property was defined and received but never forwarded to SearchResults, so searchResults.detailView was always null and hover never fired. Added detailView: top.detailView.
FullKeyboard.qml — DetailView top anchor invalid: anchors.top: pagedPicker.Top had a capital T. pagedPicker.Top is undefined (valid anchor line is lowercase .top). Fixed to pagedPicker.top.
All session 11 changes are pure QML — no rebuild needed.
Goal: The session 10 full repair + session 11 QML fixes put all source in good shape on paper. Session 12 was the actual rebuild attempt — hitting every compile/linker error that had been hiding behind the already-broken state.
All errors traced to one root cause: header truncation from the session 10 incident had also silently dropped private sections from several DB class headers. When those files were left out of the earlier brace-balance repair (which only checked braces, not semantic completeness), their missing content caused clean compile failures during the rebuild.
Truncated private sections found and restored:
Missing friend declarations on ItemDBEntry:
Cross-referencing deep-link code against ItemDBEntry's protected fields revealed three callers that needed friend access: GameCornerDBEntry (sets toGameCorner), PokemonDBEntryEvolution (sets toEvolvePokemon), and PokemonDBEntry (sets toTeachPokemon). Added all three to the friend list.
Linker errors — slots declared but never implemented:
DB::qmlProtect() and DB::qmlHook() were declared as public slots in db.h but had no bodies in db.cpp. MOC generates a dispatch entry for every slot regardless, so the linker always requires them. Added both: qmlProtect cascades to all 26 sub-databases; qmlHook sets the "db" context property on the QML engine context.
App-layer fixes:
Result: All compile and linker errors cleared. Clean rebuild expected to succeed.
Result: All compile and linker errors cleared. Clean rebuild expected to succeed.
After the session-12 rebuild, the whole brg.file.data.dataExpanded.* chain was still undefined — disproving the truncation/qRegisterMetaType theory (the binary was current and the registrations were present).
Real root cause: Q_DECLARE_OPAQUE_POINTER on the QObject chain types. It forces IsPointerToTypeDerivedFromQObject<T*> = false, so Qt stores those QObject pointers as opaque values and QML reads their sub-properties as undefined — and neither qRegisterMetaType nor qmlRegisterUncreatableType overrides it. The giveaway: brg.file worked (FileManagement is fully #included in bridge.h, never opaque) while everything opaque-declared beneath it failed.
Fix: de-opaque the traversed QObject types and #include their full headers at the declaration sites so Qt detects them as QObject pointers. Then (session 13c) the includes were scoped down to only the branches QML actually traverses, because including the whole tree into the everywhere-included savefileexpanded.h dragged db headers into every TU and tanked build time. See decisions/architecture.md → "QML Property-Chain Traversal" and reference/qt-patterns.md.
This unblocked everything: Trainer Card, Bag, Pokédex, Pokémon storage all read and persist; the "Connections: no signal matches" warning flood cleared (the targets were just undefined).
With data flowing, the work shifted to UI bugs, most tracing to the Qt 6 Material 3 control height change (taller TextField/ComboBox than the Qt 5-era hardcoded layouts assumed): clipped number fields (width ignored padding), overlapping trainer-card fields (fixed-offset anchors), item-count centering, Pokémon cell not clickable (the MouseArea had no onClicked), trainer-screen randomize calling the nonexistent randomName(), a null-deref in TilesetPicker. All fixed; see status.md session logs. Remaining work is more of the same Material-height polish plus a few functional gaps (moves dropdown, tileset preview render, hover nickname).
The app is functional: data reads and persists across screens, names/badges/pokédex/items all work, the Pokémon editor opens. Remaining work is UI polish and a handful of small functional bugs — see status.md (current state + open-issues table) and plans/next-steps.md.
1.17.0