fix(savegame): rebuild shroud grid on load to clear fog over own base

GeneralsMD/Code/GameEngine/Source/GameLogic/Object/PartitionManager.cpp

@@ -4773,7 +4773,63 @@ void PartitionManager::xfer( Xfer *xfer )
 // ------------------------------------------------------------------------------------------------
 void PartitionManager::loadPostProcess()
 {
+	// GeneralsX @bugfix coolswood 22/07/2026 Fix fog-of-war after save/load.
+	// After a save load, the per-cell shroud counters and the pending undo-queue are
+	// restored from the file, but nothing makes objects re-apply their looks. When the
+	// restored undo-queue drains (~m_unlookPersistDuration frames later), removeLooker
+	// fires at every queued position. Static structures never re-look, so their cells
+	// drop to FOGGED: the player's own base becomes gray "previously seen" ghosts that
+	// can no longer be selected. The restored m_partitionLastLook and the cell counters
+	// also disagree (counters reflect looks that no live object owns), so any later
+	// handleShroud() would double-count. The fix mirrors the established
+	// TerrainLogic::setActiveBoundary rebuild: drain the queue, snapshot the explored
+	// (FOGGED) and permanently-revealed (CLEAR) cells, tear the cell grid down to the
+	// all-shrouded default, then make every object re-look fresh onto the clean grid.
+	// After this the undo-queue is empty, every live object owns its looker, and static
+	// structures stay CLEAR permanently.
+	ShroudStatusStoreRestore partitionStore;
+
+	// Flush the restored undo-queue so lingering undoes do not fire after the rebuild.
+	processEntirePendingUndoShroudRevealQueue();
+
+	// Remember cells that are explored-but-not-visible (FOGGED).
+	storeFoggedCells(partitionStore, TRUE);
+
+	// Release ghost-object partition data before the cell grid is torn down.
+	TheGhostObjectManager->releasePartitionData();
+
+	// Invalidate every object's sighting state so the later unlook()/unshroud() inside
+	// handleShroud() are no-ops instead of queueing double-undoes or double-counting.
+	for (Object *obj = TheGameLogic->getFirstObject(); obj; obj = obj->getNextObject())
+		obj->friend_prepareForMapBoundaryAdjust();
+
+	// Remember cells that are CLEAR (permanently revealed) after the queue drain.
+	storeFoggedCells(partitionStore, FALSE);
+
+	// Rebuild the cell grid: shutdown deletes m_cells, init re-allocates every cell at
+	// the fully-shrouded default (m_currentShroud = 1).
+	reset();
+	init();
+
+	// Restore permanently-revealed cells with extra lookers.
+	restoreFoggedCells(partitionStore, FALSE);
+
+	// Prevent new ghost objects from being created while the existing ones are restored.
+	TheGhostObjectManager->lockGhostObjects(TRUE);
+
+	// Re-register and re-look every object onto the clean grid. handleShroud() now does
+	// unlook (no-op, SightingInfo is invalid) + look (fresh doShroudReveal), so each
+	// object owns exactly one looker and the undo-queue stays empty.
+	for (Object *obj = TheGameLogic->getFirstObject(); obj; obj = obj->getNextObject())
+		obj->friend_notifyOfNewMapBoundary();
+
+	// Restore explored-but-not-visible (FOGGED) cells and ghost-object partition data.
+	restoreFoggedCells(partitionStore, TRUE);
+	TheGhostObjectManager->restorePartitionData();
+	TheGhostObjectManager->lockGhostObjects(FALSE);
 
+	// Redraw the display/radar from the rebuilt cell state.
+	refreshShroudForLocalPlayer();
 }
 
 //-----------------------------------------------------------------------------

GeneralsMD/Code/GameEngine/Source/GameLogic/System/GameLogic.cpp

@@ -2226,7 +2226,15 @@ void GameLogic::tryStartNewGame( Bool loadingSaveGame )
 	if( loadingSaveGame == FALSE )
 		ThePlayerList->newMap();
 
-	if ( isChallengeCampaign )
+	// GeneralsX @bugfix coolswood 22/07/2026 Fix fog-of-war after save/load in Generals Challenge.
+	// The challenge-mode relationship setup must not run when loading a save game: the
+	// player relationships were already serialized into the save via CHUNK_Players (which
+	// loads before CHUNK_GameLogic). Running this block on load overwrites the restored
+	// ALLY relationships (e.g. between the human player slot and their chosen general) with
+	// fresh ENEMIES derived from the map's "ThePlayer" placeholder. That causes every
+	// structure of the local general to stop revealing shroud for the human player, so the
+	// player's own base becomes covered by fog of war within ~30 frames of loading.
+	if( loadingSaveGame == FALSE && isChallengeCampaign )
 	{
 		// Establish local player relationships with other teams as
 		// they're set up for "ThePlayer" in challenge mode map.

docs/WORKLOG/2026-07-DIARY.md

@@ -1,5 +1,23 @@
 # July 2026
 
+## 22/07/2026
+### Fix Fog-of-War Over Own Base After Save/Load (Generals Challenge)
+- **Symptom**: In Generals Challenge, loading a save caused fog of war to creep over the player's own base within ~5-10 seconds. The player's own buildings became gray "previously seen" ghosts that could no longer be selected or controlled.
+- **Disproved hypothesis (player relationships)**: First suspected a lost ALLY relationship between the human player slot and their chosen general after load. Diagnostics showed the relationship was already ENEMY at save time (by design — the alliance is held in team-relations, evaluated before player-relations), and brute-forcing all players to mutual ALLIES on load did **not** fix it. This proved the fog was independent of relationships.
+- **Root cause (partition/shroud load path)**: After a save load, `PartitionManager` restores the per-cell shroud counters (`m_currentShroud`) and the pending undo-queue from the file, but nothing makes objects re-apply their lookers — `PartitionManager::loadPostProcess()` was an empty stub. When the restored undo-queue drained (~`m_unlookPersistDuration` = 30 frames later), `removeLooker` fired at each queued position. Static structures never re-look, so their cells dropped to FOGGED: the player's own base became gray, unselectable ghosts. The restored `m_partitionLastLook` and the cell counters also disagreed, so any later `handleShroud()` would double-count.
+- **Fix** (`PartitionManager.cpp::loadPostProcess`): Mirrored the established `TerrainLogic::setActiveBoundary` rebuild pattern — drain the queue, snapshot explored (FOGGED) and permanently-revealed (CLEAR) cells, release ghost-object partition data, invalidate every object's sighting state (`friend_prepareForMapBoundaryAdjust`), tear the cell grid down (`reset()`+`init()` to all-shrouded default), restore permanently-revealed cells, re-look every object (`friend_notifyOfNewMapBoundary` → each object owns exactly one live looker), then restore explored cells + ghost objects + `refreshShroudForLocalPlayer()`. After this the undo-queue is empty, each object owns its looker, and static structures stay CLEAR permanently.
+- **Defensive guard** (`GameLogic.cpp::tryStartNewGame`): Guarded the challenge-mode relationship-setup block with `loadingSaveGame == FALSE` so it cannot overwrite serialized relationships on load.
+
+## 22/07/2026
+### Fix "Failed to initialize WSI" Launch Regression (macOS)
+- **Symptom**: Game stopped launching after ~18 July with `ERROR: TheDisplay->init() threw exception: Failed to initialize WSI.` (crash before main menu). The generic message hid two distinct, layered root causes.
+- **Root cause #1 — DXVK compiled without SDL3 WSI backend**: DXVK's `meson.build` auto-detects the WSI backend via `dependency('SDL3')`/`('SDL2')` and only defines `-DDXVK_WSI_SDL3` (registering the runtime "SDL3" bootstrap) when that pkg-config dependency resolves. The game *unconditionally* sets `DXVK_WSI_DRIVER=SDL3` at runtime (`GeneralsMD/Code/Main/SDL3Main.cpp:288`). The FetchContent SDL3 build (`cmake/sdl3.cmake`) does NOT emit a `.pc` file, so Meson only sees a system SDL3. Without one (and with Homebrew `sdl2-compat` installed), Meson silently compiled DXVK with **only** the SDL2 WSI — so the SDL3 bootstrap was absent and `wsi::init()` threw. The regression was triggered by a clean `meson --reconfigure` losing a previously-available system SDL3.
+- **Root cause #2 — d3d8 rpath breaks in flat deploys**: Meson bakes `@loader_path/../d3d9` into `libdxvk_d3d8.0.dylib` (matches the `src/d3d8 -> src/d3d9` build layout). Flat runtime dirs (deploy dir, `.app` bundle `Libraries/`) place d3d8 and d3d9 side-by-side, so `../d3d9` points at a non-existent subdir and `dlopen(libdxvk_d3d8)` fails with `Library not loaded: libdxvk_d3d9.0.dylib`. This was previously masked because the main binary's rpath included `build/macos-vulkan`, where the `dxvk_d3d8_install` custom target placed a flat d3d9 copy — but only when that ALL target ran (a `--target z_generals` rebuild skips it).
+- **Fix #1** (`cmake/dx8.cmake`): Added a configure-time guard that checks the Homebrew pkg-config (`/opt/homebrew/bin/pkg-config`, the one Meson's ExternalProject subprocess actually invokes — not the vcpkg-shimmed `PKG_CONFIG_EXECUTABLE`) for SDL3 and fails fast with an actionable `brew install sdl3` message instead of producing a broken DXVK.
+- **Fix #2** (`cmake/dx8.cmake` + `cmake/fixup_d3d8_rpath.cmake.in` + deploy/bundle scripts): Added an idempotent `@loader_path` rpath to d3d8 after the build-dir copy step (via a generated CMake script to avoid ninja shell-escaping issues; re-signs adhoc because `install_name_tool` invalidates the signature), and mirrored the same guard in `deploy-macos-{zh,generals}.sh` and `bundle-macos-{zh,generals}.sh` as belt-and-suspenders for non-`dxvk_d3d8_install` build paths.
+- **Verification**: Clean `rm -rf build/macos-vulkan` + rebuild + deploy; game now reaches main menu (`Shell::push() filename='Menus/MainMenu.wnd'`) with `Vulkan: Found vkGetInstanceProcAddr` + `Platform WSI` + `VK_EXT_metal_surface` + `Driver : MoltenVK 0.2.2209` — matching the working 18 July log signature.
+- **Prereq update**: Documented `brew install sdl3` as required in `build-macos-zh.sh` header.
+
 ## 17/07/2026
 ### Fix Non-Deterministic Math in WWMath Matrix Classes
 - Traced the cross-platform desync at frame 100 to the `TrainCabUngarrisonable` object's translation matrix diverging between macOS (ARM64) and Linux (x86_64).