* f3 menu text scaling
* Reduce overscaling above 1080p
Restores original scaling for 1440p to try and keep the text size more
sane on high DPI monitors
---------
Co-authored-by: Loki Rautio <lokirautio@gmail.com>
When a new player was whitelisted while they were sitting on the join screen, their next attempt would insta-kick before the world ever loaded, and the retry after that would let them in for roughly 20 seconds before booting them with "connection closed". Two separate bugs were colliding.
The first kick was a stale cancel flag on the client. When the server rejects a join, the "Connecting to host..." screen tears down, and its teardown path fires the cancel callback defensively. That flag would latch on without ever being consumed, so the very first tick of the next join attempt saw it and immediately closed the fresh connection. Clearing the flag when a new join starts prevents this.
The second kick was an orphan on the server. When the first failed join's TCP dropped, its slot got recycled for the next successful join, but the half-built login object from the broken attempt was still in the pending queue. 30 seconds later its "login took too long" timer fired, and the disconnect packet it tried to send was routed to whoever currently held that slot, which was now the new in-world player. It landed on their live socket and kicked them. Telling the game's Socket layer about the TCP drop lets the orphan clean itself up, and refusing to write on an already-closing socket stops any late packet from leaking into the recycled slot.
Before this change, turning VSync off did not actually uncap your frame rate. The game ran in borderless fullscreen, where Windows' desktop compositor owns the display pipeline and applies its own VSync regardless of what the game asks for. So on a 240Hz monitor with a GPU that could draw 800fps, you still only saw 240 frames per second and the other 560 were thrown away. Worse, the frames you did see were up to a refresh cycle stale by the time they hit your eyes, which is input latency you can feel when moving the camera or aiming.
Fullscreen now uses true DXGI exclusive mode, where the compositor is out of the way and the swap chain writes directly to the display. Every frame the GPU produces lands on screen as soon as it is ready, so "VSync off" actually does something. Expect FPS to climb well past your monitor's refresh rate and the mouse to feel noticeably more responsive.
Exclusive fullscreen also displays at your monitor's native resolution with no driver-side scaling or filtering. The backbuffer is grown to match the monitor exactly before the transition, and the display mode is pinned to the backbuffer size so nothing in the output pipeline resamples your pixels on the way to the screen. The result is a crisp 1:1 image with none of the softening or greyish filter that stretched output can introduce.
Yes, this is the mode where screen tearing can happen. Tearing gets a bad reputation but it is a visual tradeoff, not a bug, and many players prefer it over the latency VSync causes. If you want to avoid tearing, just turn VSync on in the settings and the game will cap cleanly to your refresh rate. You now have a real choice between the two instead of "off" quietly being broken.
Under the hood, F11 and the saved fullscreen preference both route through a new ApplyExclusiveFullscreen path that does Microsoft's recommended transition: grow the window to cover the monitor, ResizeTarget with no scaling so the display mode is pinned to the backbuffer size, SetFullscreenState(TRUE), SetColorSpace1(sRGB), then a second ResizeTarget so picky drivers actually apply the mode. Exit forces a real decorated windowed state so F11 cycles cleanly between windowed and exclusive fullscreen. ResizeD3D skips its swap-chain-recreate path while in exclusive mode so it does not fight DXGI for ownership.
The swap chain's RefreshRate changes from a hardcoded 60Hz to 0/0 so DXGI matches the current display mode. Fixes an "input signal out of range" error that could happen on high-refresh monitors after entering fullscreen, where the monitor was being asked to renegotiate timing off of 240Hz down to 60Hz on every launch.
The previous FLIP_DISCARD swap chain configuration was producing broken rendering and a stretched aspect ratio on startup and after window resize, because the closed-source 4J Renderer library holds hidden backbuffer references that prevent ResizeBuffers from succeeding in flip mode. Switches the swap chain to the legacy bitblt DISCARD model with BufferCount=1, which lets the "destroy old, create new" resize path in ResizeD3D work cleanly. In-world rendering and aspect ratio are now correct at launch and across window resizes.
Known regression from this change: screen tearing no longer works when VSync is off. On Windows 10 and 11, bitblt swap chains always go through the DWM compositor, which locks presentation to the monitor refresh rate regardless of the SyncInterval parameter we pass to Present. Every frame the renderer produces above the refresh rate is silently dropped by DWM, which hurts input latency compared to a true uncapped-fps presentation path. The next iteration will reverse-engineer the 4J Renderer struct layout to find where those hidden backbuffer references are stored, release them before ResizeBuffers, and switch back to FLIP_DISCARD with ALLOW_TEARING so real tearing is possible again.
Also removes the dead SwapChainVSyncProxy COM wrapper. The proxy was originally intended to intercept Present calls from the Renderer library for VSync control, but the library hardcodes SyncInterval=1 and does not dispatch Present through the proxy vtable, so it was never actually doing anything useful.
The upstream project (formerly smartcmd/MinecraftConsoles, now MCLCE/MinecraftConsoles) reorganized their in-game credits screen. This pulls in that restructure so our credits accurately reflect who the current and former upstream maintainers are, and points the attribution URL at the right place. codeHusky and mattsumi stay as Project Maintainers, and itsRevela is added alongside them. smartcmd, Patoke, and rtm516 move to a new Former Maintainers section. The contributor count ticks up from 100+ to 120+, and the credit URL at the bottom now reads github.com/MCLCE/MinecraftConsoles with a "(formerly smartcmd/MinecraftConsoles)" line underneath it.
On the README side, only the Star History chart URL was updated from smartcmd to MCLCE. The Nightly client and dedicated server download links stay pointed at itsRevela/LCE-Revelations since our fork has its own release pipeline.
Upstream attribution: d0786f95 by Loki Rautio. Applied as a partial cherry-pick with the two Nightly download URL hunks dropped and itsRevela added to the maintainer list.
The personal repo was renamed from itsRevela/MinecraftConsoles to itsRevela/LCE-Revelations, so this sweeps the rest of the codebase to match. In-game, the credits screen now shows "LCE-Revelations" instead of "MinecraftConsoles" as the project heading. In the README, the Nightly client and dedicated server download links point at the new repo URL, and the Docker image reference is now ghcr.io/itsrevela/lce-revelations-dedicated-server. The dedicated server's generated server.properties file also picks up a new header comment reflecting the rename.
For folks building from source: the CMake project name was renamed, so when you configure the build the generated solution file is now LCE-Revelations.sln instead of MinecraftConsoles.sln. The Nix flake description and the Nightly release uploader script were updated to match, and a historical FourKit port reconnaissance document was removed since that port is already complete.
Also restored the Fluxer server link at the top of the README, which was lost when the repo was fast-forwarded from the upstream that got griefed.
When framerate was uncapped (vsync off, high-end hardware), the controller cursor in the inventory and creative menus moved way too fast. Basically unusable unless you switched to the dpad. The cursor update was tied to how often the screen redraws, so the faster the game ran, the faster the cursor flew.
Now the cursor moves a smaller distance per frame at higher framerates, so the actual on-screen speed stays the same whether you're at 60 FPS or 600 FPS.
While fixing this I also found an old workaround that was rounding the cursor position to whole pixels every frame and nudging it by 1 pixel to keep it from getting stuck. That nudge was pointing the wrong way on the vertical axis, which made up/down movement feel broken once the per-frame distance got small. Removed the rounding and the nudge. The cursor can now hold a fractional position between frames, and the part of the code that actually draws the cursor still snaps it to whole pixels on screen.
Fixes#3
- Protect PlayerList and ServerConnection players vectors with critical
sections; all iterations use copy-on-read snapshots to prevent iterator
invalidation during concurrent join/leave
- Add null check on player bounding box in movement validation to prevent
crash when player is removed mid-tick
- Re-validate socket player pointer immediately before SendData to narrow
the TOCTOU race window on disconnect
- Replace inline disconnect cleanup with a queued system drained on the
main tick thread, eliminating the done_cs -> m_playersCS lock inversion
that caused deadlocks under load
- Disable Windows QuickEdit mode at server startup to prevent console
input selection from freezing the process
- Move chunk priority sort behind ServerConnection::sortPlayersByChunkPriority()
to keep the players vector lock-protected
- Prevent payment item from being consumed when submitting unchanged powers
- Reorder ServerPlayer::openBeacon to send ContainerOpenPacket before
addSlotListener so beacon data packets arrive after the client menu is ready
- Add BeaconMenu::broadcastChanges() to continuously sync levels and powers
to clients, matching the pattern FurnaceMenu already uses
- Initialize UIControl_BeaconEffectButton::m_lastState to prevent stale
heap memory from suppressing Iggy ChangeState calls on menu re-entry
Adds the FourKit .NET 10 plugin host as a second dedicated server
build flavour alongside the existing vanilla server. Both flavours
build from the same source tree, with FourKit gated by the
MINECRAFT_SERVER_FOURKIT_BUILD preprocessor define.
Build layout:
Minecraft.Server vanilla, no plugin support, no .NET dep
Minecraft.Server.FourKit FourKit-enabled, ships with bundled
.NET 10 self-contained runtime in runtime/
and an empty plugins/ folder
Both produce a Minecraft.Server.exe in their own per-target output
dir. The variant identity lives in the directory name, not the
binary name, so either flavour can be shipped as a drop-in.
Native bridge (Minecraft.Server/FourKit*.{cpp,h}):
* FourKitRuntime: hosts CoreCLR via hostfxr's command-line init API
(the runtime-config API does not support self-contained components)
* FourKitBridge: ~50 Fire* event entry points, with inline no-op
stubs for the standalone build so gameplay code can call them
unconditionally
* FourKitNatives: ~80 native callbacks the managed side invokes
for player/world/inventory mutations
* FourKitMappers: type and enum mapping helpers
Managed plugin host (Minecraft.Server.FourKit/):
* Bukkit-style API: Player, World, Block, Inventory, Command,
Listener, EventHandler attribute, ~54 event classes
* PluginLoader with per-plugin AssemblyLoadContext
* FourKitHost as the [UnmanagedCallersOnly] entry point table
* Runtime resolves plugins relative to the host process so they
always live next to Minecraft.Server.exe regardless of where the
managed assembly itself is loaded from
Engine hooks (Minecraft.Client/, Minecraft.World/):
* Player lifecycle (PreLogin, Login, Join, Quit, Kick, Move,
Teleport, Portal, Death) wired into PendingConnection and
PlayerConnection without disturbing the cipher handshake or
identity-token security flow
* Inventory open/click/drop hooks across every container menu type
* Block place/break/grow/burn/spread/from-to hooks across the
full tile family
* Bed enter/leave, sign change, entity damage/death, ender pearl
teleport hooks
Regression fixes preserved while applying donor diffs:
* ServerPlayer::die() retains the LCE-Revelations hardcore branch
(setGameMode(ADVENTURE) + banPlayerForHardcoreDeath) in both the
FourKit and non-FourKit code paths
* ServerLevel::entityAdded() retains the sub-entity ID reassignment
loop required by the client's handleAddMob offset, fixing Ender
Dragon and Wither boss multi-part hit detection
* LivingEntity::travel() retains the raw Player* cast and the
cached frictionTile, both Revelations perf wins that the donor
silently reverted
* ServerLogger.cpp keeps the file-logging code donor stripped
* PlayerList.cpp end portal transition fix and UIScene_EndPoem
bounds-check are intact
Build system:
* Top-level CMakeLists.txt adds the Minecraft.Server.FourKit
subdirectory and pulls in the new shared cmake/ServerTarget.cmake
helper
* Minecraft.Server/cmake/sources/Common.cmake is now location
independent (uses CMAKE_CURRENT_LIST_DIR) so the source list
can be consumed from either server target's CMakeLists.txt
* The seven FourKit*.cpp/h files live in their own
_MINECRAFT_SERVER_COMMON_SERVER_FOURKIT variable so the
standalone target omits them
* configure-time .NET 10 SDK check fails fast with a clear
download link if the SDK is missing
* global.json pins the SDK to 10.0.100 with latestFeature
rollforward
Sample plugin (samples/HelloPlugin/) demonstrates the loader and
the PlayerJoinEvent listener pattern.
CI:
* nightly.yml builds both server flavours, ships
LCE-Revelations-Server-Win64.zip and
LCE-Revelations-Server-Win64-FourKit.zip, attests both, and
updates release notes for the dual-flavour layout
* pull-request.yml pulls in actions/setup-dotnet so the FourKit
publish step works in PR validation
* All zip artifacts and the client zip are renamed from
LCREWindows64 to LCE-Revelations-{Client,Server}-Win64
Documentation:
* COMPILE.md gets a VS 2022 quick start, .NET 10 prereq section,
server flavours explanation, and a troubleshooting section
* docs/FOURKIT_PORT_RECON.md captures the file-by-file recon that
drove the port
* docs/FOURKIT_PARITY.md is the canonical reference for which
events FourKit fires
Docker:
* docker-compose.dedicated-server.yml MC_RUNTIME_DIR default points
at the vanilla CMake output. The FourKit Docker image is
intentionally NOT shipped yet because hosting .NET 10 self
contained inside Wine has not been smoke-tested
Matches the packet size limit used by the plugin-api fork. Our 512KB
limit caused "Connection lost" when their server sent large packets
(e.g. chunk data) that exceeded our cap.
Chunk loading now batches up to 16 nearest-first requests per player per
tick on dedicated server (client stays at 1), improving tick recovery
time after player join.
Reverts the async save system -- the background thread snapshot/compress
path added complexity without measurable benefit. Autosave on Windows64
server now uses the standard synchronous flush like client, in
preparation for a proper async implementation from upstream.
The previous IQNet cleanup in handleRemoveEntity fired on every entity
despawn, which happens both when a player goes out of tracking range
and when they disconnect. This caused players to vanish from the Tab
list whenever they moved beyond render distance.
Introduce two custom payload channels (MC|ForkHello, MC|ForkPLeave) so
the client can distinguish "out of range" from "actually left":
- Server sends MC|ForkHello during login to identify itself as a fork
- Server sends MC|ForkPLeave with the player's gamertag on disconnect
- Client skips IQNet cleanup in handleRemoveEntity on fork servers
- Client cleans up IQNet only when MC|ForkPLeave arrives
Fully backwards-compatible: no existing packet wire formats changed.
Upstream clients ignore the unknown channels, fork clients on upstream
servers fall back to the old entity-tracking-based cleanup.
Rewire the SWF focus chain via Iggy so VSync, Fullscreen, and Render
Distance are reachable with keyboard/gamepad navigation.
Cap render distance slider at 16 chunks. Shift graphics menu layout
up 60px for better centering.
Fix skin preview walk/attack animations running too fast with VSync
off by scaling per-frame increments by delta time relative to 60fps.
The walking and attack animations in the Change Skin menu were
frame-rate-dependent, advancing per render call with no delta time
scaling. With uncapped FPS they ran proportionally too fast.
Add time-based scaling relative to a 60fps baseline. The scale is
computed once per frame (cached for 0.5ms) so multiple skin previews
rendered in the same frame all animate at the correct speed.
StorageManager.SaveSaveData() blocks until the disk write completes.
Calling it from the game thread caused 2.5s freezes. Now it only runs
from TickCoreSystems() on the ServerMain thread, which operates
independently of the game tick loop.
Move StorageManager.AllocateSaveData() back under the save lock (called
on the game thread before releasing). Background thread now only does
pure zlib compression with no StorageManager calls.
Add CommitPendingAsyncSave() to both the game thread tick and
TickCoreSystems() to ensure saves commit during bootstrap, normal
gameplay, and shutdown.
Autosave previously froze the main thread for 2-6 seconds while
compressing the entire save file with zlib. Now the save buffer is
snapshotted under the lock (~18ms), then compression runs on a
background thread. The compressed data is committed to StorageManager
on the next main-thread tick via CommitPendingAsyncSave().
Also skip redundant full chunk saves during autosave on the dedicated
server -- chunks are already persisted by the per-tick trickle save.
Only entity data is flushed, matching Xbox/Orbis behavior.
Added per-step timing to the autosave handler for diagnostics.
The EntityTracker and TrackedEntity classes have O(players^2 * entities)
loops that check IsSameSystem() for split-screen couch co-op visibility
expansion. On the dedicated server, all players are remote so
IsSameSystem() always returns false, making these loops pure overhead.
Skip them entirely when g_Win64DedicatedServer is true. The original
split-screen logic is preserved for game client LAN hosting.
The security gate was closed before LoginPacket was sent in placeNewPlayer,
causing all login setup packets to be buffered behind the cipher handshake.
Under high-latency connections, the flushed data arrived before the player
object was initialized, causing a null pointer crash. The gate now closes
after the login sequence and MC|CKey are sent.
Bypass the 4J RenderManager's hardcoded SyncInterval=1 by calling
the DXGI swap chain directly with Present(0, ALLOW_TEARING) when
VSync is disabled. Falls back to the library's Present on failure.
Relocate VSync/Fullscreen setting flags from bits 18-19 to bits 24-25
to eliminate overlap with the render distance byte (bits 16-23).
Sync the Fullscreen game setting when F11 is pressed so the graphics
menu checkbox stays accurate.
Remove tracked DumpSwf.class (already covered by tools/*.class gitignore).
BedrockFog removal via removeControl with centreScene=true triggered
Flash-side repositioning that didn't account for the tool-added VSync
and Fullscreen checkboxes, creating a gap after CustomSkinAnim and
causing RenderDistance to render behind Gamma.
On Windows64 (single player per client), the console splitscreen
host-check that removed BedrockFog/CustomSkinAnim is unnecessary.
Gate it behind #ifndef _WINDOWS64 so all controls stay visible.
Display hardcore heart textures when a world is in hardcore mode,
matching Java Edition behavior. Hearts switch between normal/hardcore
across all states (poison, wither, flash) and all HUD resolutions.
C++ changes:
- IUIScene_HUD: check isHardcore() and call SetHardcoreMode() each tick
- UIScene_HUD: send hardcore boolean to Flash via Iggy, invalidate
SetHealth dirty check on state change to force heart redraw
- CreateWorldMenu/LoadMenu: lock game mode to Survival when hardcore
- MinecraftServer: gate server.properties hardcore override behind
MINECRAFT_SERVER_BUILD so offline worlds preserve their saved flag
SWF changes (via new Java tools):
- AddHardcoreBitmaps: adds 10 hardcore heart bitmaps to graphics SWFs
- AddHardcoreHearts: adds 10 new frames (15-24) to health sprite
- PatchHudABC: patches HUD ActionScript bytecode with SetHardcore
method and frame offset logic (+14 normal/poison, +6 wither)
Also updates README changelog styling with consistent ### headings.
## Description
Fix issue where typing in a short seed on world creation doesn't save the seed correctly
## Changes
### Previous Behavior
Typing in a seed on the world creation menu that's less than 8 characters long will result in garbage data being saved as the seed. Happens with controller and KBM.
You can see this in-game - if you exit the world options menu and go back in, the seed will show up as boxes □□□.
Weirdly, if you type a seed again, it behaves as expected.
### Root Cause
For some reason, assigning `m_params->seed` to the seed text points it to garbage data, when it's 7 characters or less.
### New Behavior
Seed entry behaves as expected.
### Fix Implementation
- Added `static_cast<wstring>` before assignment to `m_params->seed`.
- Also replaced `(wchar_t *)` with `reinterpret_cast<wchar_t*>` in the functions.
### AI Use Disclosure
No AI was used
Move screenshot capture from Minecraft::tick() (which requires an
active player) to the Windows64 main loop alongside other global
key handlers (F1/F3/F11). F2 now works from the main menu, pause
menu, settings, inventory, crafting, and all other screens. Chat
message still shown when in-game.
Replace the 80% scaled approach with full-size logo bitmaps and
shift the ComponentLogo SWF placements up by 10px (proportionally
scaled for lower resolutions) to avoid occlusion by the load/join
menu. Add ShiftLogo.java tool for adjusting SWF placement offsets.
Replace MenuTitle and MenuTitleSmall bitmaps in skinHDWin.swf and
skinWin.swf with the custom LCRE (Legacy Console Edition Revelations)
logo, scaled to 80% within the original bitmap canvas to avoid
occlusion by the load/join menu.
Add ReplaceLogo.java and ExtractFromArc.java tools for SWF bitmap
replacement and arc file extraction. Keep original arc as .bak.
