When transferring an entity to a new grid via entity.grid = new_grid,
the entity was removed from the old grid's spatial hash but never
inserted into the new one. This made it invisible to spatial queries
on the destination grid.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
UIEntity::setProperty() now calls spatial_hash.update() when draw_x/draw_y
change, matching the Python property setter behavior. Added
enable_shared_from_this<UIEntity> to support shared_from_this() in the
setProperty path.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
GridLayer, UIGridPoint, and GridChunk each stored a raw GridData* that
could dangle if the grid was destroyed while external shared_ptrs
(e.g. Python layer wrappers) still referenced child objects. The
destructor now nulls all parent_grid pointers before cleanup. All
usage sites already had null guards, so this completes the fix.
These were the last three unfixed bugs from the memory safety audit.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Breaking API change: removes 4 camelCase function aliases from the mcrfpy
module. The snake_case equivalents (set_scale, find_all, get_metrics,
set_dev_console) remain and are the canonical API going forward.
- Removed setScale, findAll, getMetrics, setDevConsole from mcrfpyMethods[]
- Updated game scripts to use snake_case names
- Updated test scripts to use snake_case names
- Removed camelCase entries from type stubs
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Layers with z_index <= 0 now render below entities (ground level), and
only layers with z_index > 0 render above entities. Previously z_index=0
was treated as "above entities" which was unintuitive -- entities stand
on ground level (z=0) with their feet, so z=0 layers should be beneath.
Changed in both UIGrid.cpp and UIGridView.cpp render methods:
- "z_index >= 0" to "z_index > 0" for break condition
- "z_index < 0" to "z_index <= 0" for skip condition
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Fixes a systemic bug where Python tp_getset property setters bypassed the
render cache dirty flag system (#144). The animation/C++ setProperty() path
had correct dirty propagation, but direct Python property assignments
(e.g. frame.x = 50, caption.text = "Hello") did not invalidate the parent
Frame's render cache when clip_children or cache_subtree was enabled.
Changes by file:
- UIDrawable.cpp: Add markCompositeDirty() to set_float_member (x/y),
set_pos, set_grid_pos; add markDirty() for w/h resize
- UICaption.cpp: Add markDirty() to set_text, set_color_member,
set_float_member (outline/font_size); markCompositeDirty() for position
- UICollection.cpp: Add markContentDirty() on owner in append, remove,
pop, insert, extend, setitem, and slice assignment/deletion
- UISprite.cpp: Add markDirty() to scale/sprite_index/texture setters;
markCompositeDirty() to position setters
- UICircle.cpp: Add markDirty() to radius/fill_color/outline_color/outline;
markCompositeDirty() to center setter
- UILine.cpp: Add markDirty() to start/end/color/thickness setters
- UIArc.cpp: Add markDirty() to radius/angles/color/thickness setters;
markCompositeDirty() to center setter
- UIGrid.cpp: Add markDirty() to center/zoom/camera_rotation/fill_color/
size/perspective/fov setters
Closes#288, closes#289, closes#290, closes#291
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Create self-contained demo game script (src/scripts_demo/game.py) showcasing:
BSP dungeon generation, Wang tile autotiling, FOV with fog of war,
turn-based bump combat, enemy AI, items/treasure, title screen
- Add MCRF_DEMO CMake option for building with demo scripts
- Add web/index.html landing page with dark theme, controls reference,
feature list, and links to GitHub/Gitea
- Build with: emcmake cmake -DMCRF_SDL2=ON -DMCRF_DEMO=ON -DMCRF_GAME_SHELL=ON
Note: Makefile wasm-demo/serve-demo targets also added locally but Makefile
is gitignored. Use CMake directly or force-add the Makefile to track it.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Removed custom __eq__/__ne__ that allowed comparing enums to legacy string
names (e.g., Key.ESCAPE == "Escape"). Removed _legacy_names dicts and
to_legacy_string() functions. Kept from_legacy_string() in PyKey.cpp as
it's used by C++ event dispatch. Updated ~50 Python test/demo/cookbook
files to use enum members instead of string comparisons. Also updates
grid.position -> grid.pos in files that had both types of changes.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Grid.position was a redundant alias for Grid.pos. Removed get_position/
set_position functions, getsetters entry, and setProperty/getProperty/
hasProperty branches. Updated all tests to use grid.pos.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
sprite_number was a legacy alias for sprite_index. All code should use
sprite_index directly. Removed from getsetters, setProperty/getProperty/
hasProperty in UISprite and UIEntity, animation property handling, and
type stubs.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Color had __hash__ but no __eq__/__ne__, violating the Python convention
that hashable objects must support equality comparison. Two Color objects
with identical RGBA values would not compare equal.
Now supports comparison with Color objects, tuples, and lists:
Color(255, 0, 0) == Color(255, 0, 0) # True
Color(255, 0, 0) == (255, 0, 0) # True
Color(255, 0, 0) != (0, 0, 0) # True
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add find_all(), get_metrics(), set_scale(), and set_dev_console() as
snake_case alternatives to the existing camelCase names. The camelCase
versions remain for backward compatibility but their docstrings now
note the preferred snake_case form. All will be removed in 1.0.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace placeholder docstrings ("SFML Vector Object", "SFML Font Object",
etc.) with comprehensive constructor signatures, argument descriptions,
and property listings matching the documentation standard used by other
types like Color and Frame.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
mcrfpy.Grid() now creates a GridView that internally owns a GridData (UIGrid).
The old UIGrid type is renamed to _GridData (internal). Attribute access on Grid
delegates to the underlying UIGrid via tp_getattro/tp_setattro, so all existing
Grid properties (grid_w, grid_h, entities, cells, layers, etc.) work transparently.
Key changes:
- GridView init has two modes: factory (Grid(grid_size=...)) and explicit view
(Grid(grid=existing_grid, ...)) for future multi-view support
- Entity.grid getter returns GridView wrapper via owning_view back-reference
- Entity.grid setter accepts GridView objects
- GridLayer set_grid handles GridView (extracts underlying UIGrid)
- UIDrawable::removeFromParent handles UIGRIDVIEW type correctly
- UIFrame children init accepts GridView objects
- Animation system supports GridView (center, zoom, shader.* properties)
- PythonObjectCache registration preserves subclass identity
- All 263 tests pass (100%)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add `collide` kwarg to Grid.find_path() and Grid.get_dijkstra_map() that
treats entities bearing a given label as impassable obstacles via
mark-and-restore on the TCOD walkability map. Dijkstra cache key now
includes collide label for separate caching. Add Entity.find_path()
convenience method that delegates to the grid.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Grid.__init__() now auto-creates a GridView that shares the Grid's
data via aliasing shared_ptr. This enables the Grid/GridView split:
- PyUIGridObject gains a `view` member (shared_ptr<UIGridView>)
- Grid.view property exposes the auto-created GridView (read-only)
- Rendering property setters (center_x/y, zoom, camera_rotation, x, y,
w, h) sync changes to the view automatically
- Grid still works as UIDrawable in scenes (no substitution) — backward
compatible with all existing code and subclasses
- GridView.grid returns the original Grid with identity preservation
- Explicit GridViews (created by user) are independent of Grid's own
rendering properties
Addresses #252. All 260 tests pass, no breaking changes.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
GridView is a new UIDrawable that renders a GridData object independently.
Multiple GridViews can reference the same Grid for split-screen, minimap,
or different camera/zoom perspectives on shared grid state.
- New files: UIGridView.h/cpp with full rendering pipeline (copied from
UIGrid::render, adapted to use grid_data pointer)
- Add UIGRIDVIEW to PyObjectsEnum
- Add UIGRIDVIEW cases to all switch(derived_type()) sites:
Animation.cpp, UICollection.cpp, UIDrawable.cpp, McRFPy_API.cpp,
ImGuiSceneExplorer.cpp
- Python type: mcrfpy.GridView(grid=, pos=, size=, zoom=, fill_color=)
- Properties: grid, center, zoom, fill_color, texture
- Register type with metaclass for callback support
All 258 existing tests pass. New GridView test suite added.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Fix Entity3D self-reference cycle: replace raw `self` pointer with
`pyobject` strong-ref pattern matching UIEntity (closes#266)
- TileLayer inherits Grid texture when none set, in all three attachment
paths: constructor, add_layer(), and .grid property (closes#254)
- Add SpatialHash::queryCell() for O(1) entity-at-cell lookup; fix
missing spatial_hash.insert() in Entity.__init__ grid= kwarg path;
use queryCell in GridPoint.entities (closes#253)
- Add FOV dirty flag and parameter cache to skip redundant computeFOV
calls when map unchanged and params match (closes#292)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The Phase 3 fix for #266 removed UIEntity::self which prevented
tp_dealloc from ever running. However, this also allowed Python
subclass wrappers (GameEntity, ZoneExit, etc.) to be GC'd while
the C++ entity lived on in a grid. Later access via grid.entities
returned a base Entity wrapper, losing all subclass methods.
Fix: Add UIEntity::pyobject field that holds a strong reference to
the Python wrapper. Set in init(), cleared when the entity leaves
a grid (die(), set_grid(None), collection removal). This keeps
subclass identity alive while in a grid, but allows proper GC when
the entity is removed. Added releasePyIdentity() helper called at
all grid exit points.
Regression test exercises Liber Noster patterns: subclass hierarchy,
isinstance() checks, combat mixins, tooltip/send methods, GC
survival, die(), pop(), remove(), and stress test with 20 entities.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
UniformCollection accessor methods checked the raw collection pointer
but never checked if the owning object was still alive. Changed owner
field from weak_ptr<UIDrawable> to weak_ptr<void> (type-erased) so
both UIDrawable and UIEntity owners can be tracked. Set owner in both
get_uniforms() paths. All accessors now check owner.lock() before
dereferencing the raw collection pointer.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add mutex lock to PythonObjectCache::lookup() - cache.find() was
unprotected against concurrent modification. Document that entity.die()
must not be called during iteration over grid.entities.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
GridPoint and GridPointState Python objects now store (grid, x, y)
coordinates instead of raw C++ pointers. Data addresses are computed
on each property access, preventing dangling pointers after vector
resizes.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
UIEntity::init() stored self->data->self = (PyObject*)self with
Py_INCREF(self), creating a reference cycle that prevented entities
from ever being freed. The matching Py_DECREF never existed.
Fix: Remove the `self` field from UIEntity entirely. Replace all
read sites (iter next, getitem, get_perspective, entities_in_radius)
with PythonObjectCache lookups using serial_number, which uses weak
references and doesn't prevent garbage collection.
Also adds tp_dealloc to PyUIEntityType to properly clean up the
shared_ptr and weak references when the Python wrapper is freed.
Closes#266, closes#275
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace ~230 occurrences of PyObject_GetAttrString(McRFPy_API::mcrf_module, "TypeName")
with direct &mcrfpydef::PyXxxType references across 32 source files.
Each PyObject_GetAttrString call returns a new reference. When used inline
in PyObject_IsInstance(), that reference was immediately leaked. When used
for tp_alloc, the reference required careful Py_DECREF management that was
often missing on error paths.
Direct type references are compile-time constants that never need reference
counting, eliminating ~230 potential leak sites and removing ~100 lines of
Py_DECREF/Py_XDECREF cleanup code.
Also adds extractDrawable() helper in UICollection.cpp to replace repeated
8-way type-check-and-extract chains with a single function call.
Closes#267, closes#268
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add ensureGridstate() helper that unconditionally checks gridstate size
against current grid dimensions and resizes if mismatched. Replace all
lazy-init guards (size == 0) with ensureGridstate() calls.
Previously, gridstate was only initialized when empty. When an entity
moved to a differently-sized grid, gridstate kept the old size, causing
heap buffer overflows when updateVisibility() or at() iterated using the
new grid's dimensions.
Also adds spatial_hash.remove() calls in set_grid() before removing
entities from old grids, and replaces PyObject_GetAttrString type lookup
with direct &mcrfpydef::PyUIGridType reference.
Closes#258, closes#259, closes#260, closes#261, closes#262,
closes#263, closes#274, closes#276, closes#278
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Game code uses standard Python file I/O to mcrfpy.save_dir with no platform
branching. On WASM, builtins.open() is monkeypatched so writes to /save/
auto-sync IDBFS on close, making persistence transparent.
API: mcrfpy.save_dir (str), mcrfpy._sync_storage() (auto-called on WASM)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- SoundBuffer.gain(factor): new DSP method for amplitude scaling before
mixing (0.5 = half volume, 2.0 = double, clamped to int16 range)
- Fix sfxr square/saw waveform artifacts: phase now wraps at period
boundary instead of growing unbounded; noise buffer refreshes per period
- Fix PySound construction from SoundBuffer on SDL2 backend: use
loadFromSamples() directly instead of copy-assign (deleted on SDL2)
- Add Image::create(w, h, pixels) overload to HeadlessTypes and
SDL2Types for pixel data initialization
- Waveform test suite (62 lines)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
closes#251
Two related bugs where Python garbage collection destroyed callbacks
that were still needed by live C++ objects:
1. **Drawable callbacks (all 8 types)**: tp_dealloc unconditionally called
click_unregister() etc., destroying callbacks even when the C++ object
was still alive in a parent's children vector. Fixed by guarding with
shared_ptr::use_count() <= 1 — only unregister when the Python wrapper
is the last owner.
2. **Timer GC prevention**: Active timers now hold a Py_INCREF'd reference
to their Python wrapper (Timer::py_wrapper), preventing GC while the
timer is registered in the engine. Released on stop(), one-shot fire,
or destruction. mcrfpy.Timer("name", cb, 100) now works without storing
the return value.
Also includes audio synth demo UI fixes: button click handling (don't set
on_click on Caption children), single-column slider layout, improved
Animalese contrast.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
All 27 PyTypeObject declarations in namespace mcrfpydef headers changed
from `static` to `inline` (C++17), ensuring a single global instance
across translation units. This fixes the root cause of stale-type-pointer
segfaults where only the McRFPy_API.cpp copy was PyType_Ready'd.
Replaced ~20 PyTypeCache call sites and 2 PyRAII::PyTypeRef lookups with
direct &mcrfpydef::Type references. Deleted PyTypeCache.h/.cpp,
PyObjectUtils.h, and PyRAII.h (all were workarounds for the static bug).
228/228 tests pass.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The BFS solver couldn't account for obstacles blocking push paths -
knowing the button is reachable doesn't mean the player can get to
the correct side of the boulder. Reverse-pull guarantees solvability
by construction: start with boulder on button, simulate valid
un-pushes to move it away. Each un-push verifies both the new boulder
cell and the player's required push position are walkable.
Also fixes chest clumping: level 2 previously crammed 3 treasures +
boulder + button into a single room. Redesigned all level plans to
spread treasures across rooms (max 1 per room). Updated lv_planner
for procedural levels 9+ with the same constraint.
Level plans no longer specify "boulder" - it's auto-generated from
the button position with min_pulls scaling by depth (2-8).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Three bugs that could produce unsolvable puzzles:
1. Feature placement fallback used leaf_center without duplicate checking,
allowing treasures to spawn on the same cell as buttons in dense rooms
(level 2 puts 8 features in one room). Fixed with exhaustive cell scan
fallback.
2. Solvability checker ignored treasure entities entirely. Boulders cannot
be pushed through treasures (TreasureEntity.bump returns False for
non-player entities), so the solver now models them as boulder-blocking
obstacles while allowing player movement through them.
3. Added explicit button-blocked-by-obstacle check before running the
full BFS solver, catching the most common failure mode early.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replaces jam-quality code with production engine features (addresses #248):
- cos_level.py: Replace custom BinaryRoomNode/RoomGraph with mcrfpy.BSP
for room generation, using adjacency graph for corridor placement
- cos_solver.py: New Sokoban BFS solvability checker; levels retry up
to 10 times if unsolvable
- game.py: Add ColorLayer fog of war with room-reveal mechanic
(visible/discovered/unknown states), compute_fov per player move
- cos_entities.py: Enemy AI state machine (idle/aggressive/fleeing)
with A* pathfinding, fix duplicate direction bug on line 428/444
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Full screen "wasm-game" for viewport compatibility between desktop and
web interfaces. Viewport modes ("fit", "center", and "stretch") should
now work the same way under WASM/SDL and SFML. This should also enable
android or web-for-mobile aspect ratios to be supported more easily.
Replace no-op audio stubs in SDL2Types.h with real SDL2_mixer-backed
implementations of SoundBuffer, Sound, and Music. This enables audio
playback in the browser with zero changes to Python bindings.
- Add -sUSE_SDL_MIXER=2 to Emscripten compile/link flags (CMakeLists.txt)
- Initialize Mix_OpenAudio in SDL2Renderer::init(), Mix_CloseAudio in shutdown()
- SoundBuffer: Mix_LoadWAV/Mix_LoadWAV_RW with duration computation
- Sound: channel-based playback with Mix_ChannelFinished tracking
- Music: global channel streaming via Mix_LoadMUS/Mix_PlayMusic
- Volume conversion: SFML 0-100 scale to SDL_mixer 0-128 scale
Known limitations on web: Music.duration and Music.position getters
return 0 (SDL_mixer 2.0.2 lacks Mix_MusicDuration/Mix_GetMusicPosition).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replaced the NotImplementedError stub with a full animation
implementation. Entity3D now supports animating: x, y, z,
world_x, world_y, world_z, rotation, rot_y, scale, scale_x,
scale_y, scale_z, sprite_index, visible.
Added Entity3D as a third target type in the Animation system
(alongside UIDrawable and UIEntity), with startEntity3D(),
applyValue(Entity3D*), and proper callback support.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
