What happened:
1) Object data is received. Client begins to read the data
2) Client initializes all its children (gob_cmd_update_infant)
3) Children try to attach to parent (yet not added)
4) Parent initializes, is added to the environment
And somewhere in between, Irrlicht wrecks up the attachments due to the missing matrix node.
The solution here is to:
1) Use the same structure as ServerActiveObject
2) Attach all children after the parent is really initialized
The old texture modifier is restored by passing `m_previous_texture_modifier`.
Either copy it manually or let the function parameter do that.
Victims so far:
8e0b80a Apr 2018
eb2bda7 May 2019
* Optimize statbar drawing
The texture name of the statbar is a string passed by value.
That slows down the client and creates litter in the heap
as the content of the string is allocated there. Convert the
offending parameter to a const reference to avoid the
performance hit.
* Optimize texture cache
There is an unnecessary temporary created when the texture
path is being generated. This slows down the cache each time
a new texture is encountered and it needs to be loaded into
the cache. Additionally, the heap litter created by this
unnecessary temporary is particularly troublesome here as
the following code then piles another string (the resulting
full path of the texture) on top of it, followed by the
texture itself, which both are quite long term objects as
they are subsequently inserted into the cache where they can
remain for quite a while (especially if the texture turns
out to be a common one like dirt, grass or stone).
Use std::string.append to get rid of the temporary which
solves both issues (speed and heap fragmentation).
* Optimize animations in client
Each time an animated node is updated, an unnecessary copy of
the texture name is created, littering the heap with lots of
fragments. This can be specifically troublesome when looking
at oceans or large lava lakes as both of these nodes are
usually animated (the lava animation is pretty visible).
Convert the parameter of GenericCAO::updateTextures to a
const reference to get rid of the unnecessary copy.
There is a comment stating "std::string copy is mandatory as
mod can be a class member and there is a swap on those class
members ... do NOT pass by reference", reinforcing the
belief that the unnecessary copy is in fact necessary.
However one of the first things the code of the method does
is to assign the parameter to its class member, creating
another copy. By rearranging the code a little bit this
"another copy" can then be used by the subsequent code,
getting rid of the need to pass the parameter by value and
thus saving that copying effort.
* Optimize chat console history handling
The GUIChatConsole::replaceAndAddToHistory was getting the
line to work on by value which turns out to be unnecessary.
Get rid of that unnecessary copy by converting the parameter
to a const reference.
* Optimize gui texture setting
The code used to set the texture for GUI components was
getting the name of the texture by value, creating
unnecessary performance bottleneck for mods/games with
heavily textured GUIs. Get rid of the bottleneck by passing
the texture name as a const reference.
* Optimize sound playing code in GUIEngine
The GUIEngine's code receives the specification of the sound
to be played by value, which turns out to be most likely a
mistake as the underlying sound manager interface receives
the same thing by reference. Convert the offending parameter
to a const reference to get rid of the rather bulky copying
effort and the associated performance hit.
* Silence CLANG TIDY warnings for unit tests
Change "std::string" to "const std::string &" to avoid an
unnecessary local value copy, silencing the CLANG TIDY
process.
* Optimize formspec handling
The "formspec prepend" parameter was passed to the formspec
handling code by value, creating unnecessary copy of
std::string and slowing down the game if mods add things like
textured backgrounds for the player inventory and/or other
forms. Get rid of that performance bottleneck by converting
the parameter to a const reference.
* Optimize hotbar image handling
The code that sets the background images for the hotbar is
getting the name of the image by value, creating an
unnecessary std::string copying effort. Fix that by
converting the relevant parameters to const references.
* Optimize inventory deserialization
The inventory manager deserialization code gets the
serialized version of the inventory by value, slowing the
server and the client down when there are inventory updates.
This can get particularly troublesome with pipeworks which
adds nodes that can mess around with inventories
automatically or with mods that have mobs with inventories
that actively use them.
* Optimize texture scaling cache
There is an io::path parameter passed by value in the
procedure used to add images converted from textures,
leading to slowdown when the image is not yet created and
the conversion is thus needed. The performance hit is
quite significant as io::path is similar to std::string
so convert the parameter to a const reference to get rid of
it.
* Optimize translation file loader
Use "std::string::append" when calculating the final index
for the translation table to avoid unnecessary temporary
strings. This speeds the translation file loader up
significantly as std::string uses heap allocation which
tends to be rather slow. Additionally, the heap is no
longer being littered by these unnecessary string
temporaries, increasing performance of code that gets
executed after the translation file loader finishes.
* Optimize server map saving
When the directory structure for the world data is created
during server map saving, an unnecessary value passing of
the directory name slows things down. Remove that overhead
by converting the offending parameter to a const reference.
* Improve readability of debug menu by using '|'
* Restore whitespace to separate yaw and cardinal direction
Co-Authored-By: ClobberXD <ClobberXD@gmail.com>
Makes the liquid waving shader per-nodedef like waving leaves/plants,
instead of being applied to all liquids.
Like the waving leaves/plants shaders, the liquid waving shader can
also be applied to meshes and nodeboxes.
Derived from a PR by t0ny2.
This sends the following header to a remote media server:
Referer: minetest://<server_name>:port
This was verified with CTF and the Minetest Public Remove Media
server. If the servername was a plain IPv6 address it will
contain `:` characters and will be encapsulated in `[]` to
be a valid URI.
* Drop the ID mapper, use a big u64 instead. This will permit to resync server ids properly with the manager code
* Modernize some code parts (std::unordered_map, auto)
* generate id on client part on U32_MAX + 1 ids, lower are for server ids
The reverted commit 968ce9af598024ec71e9ffb2d15c3997a13ad754
is suspected (through the use of bisection) of causing network slowdowns.
Revert for now as we are close to release.
Store the rotation in the node as a 4x4 transformation matrix internally (through IDummyTransformationSceneNode), which allows more manipulations without losing precision or having gimbal lock issues.
Network rotation is still transmitted as Eulers, though, not as matrix. But it will stay this way in 5.0.
* Fix a crash on Android with Align2Npot2
glGetString can be NULL. If stored in a string it triggers a SIGSEGV.
Instead do a basic strstr and verify the pointer
* Better Align2Npot2 check (+ performance)
The fallback code shouldn't be needed and is a remnant of the GET
method that old media servers use. Clients using it are likely
to just waste bandwidth and having to download the media again
through the normal transfer from server method. The most reliable
method is to get all missing textures therefore from the server
directly and not spam the remote media server with 404s.
Previously, when the CSM restriction 'LOAD_CLIENT_MODS' was used a
client was still able to add CSM code to 'builtin' to bypass that
restriction, because 'builtin' is not yet verified.
Until server-sent CSM and verifying of 'builtin' are complete, make
'LOAD_CLIENT_MODS' disable the loading of builtin.
Clarify code comments and messages to distinguish between client-side
modding and client-side scripting. 'Scripting' includes 'builtin',
'modding' does not.
Move star draw to before sun glow texture draw and before sun draw,
not currently essential but the logical order. Will be necessary if
a 'no far ground' option is added, to draw stars behind the sun.
At sunset:
Stars first appear slightly later, at the time the sun disappears over the horizon,
this fixes seeing dark stars in front of the sun horizon glow texture.
Stars reach full brightness slightly earlier at time 20000, not so excessively long
after sunset.
The above behaviour is also applied at sunrise, but of course, time-inverted.
In pitch fly mode, you fly to the exact direction you are pointing at, using the forward key. Other move directions are also pitched accordingly.
It allows smoother and more complex movements.
Can be enabled/disabled by L key by default (set keymap_pitchfly in minetest.conf)
Adding support for _NET_WM_PID as defined in Extended Window Manager Hints
Move verbose messaging to setupXorgTopLevelWindow method as Xorg messages should only occur when running in Xorg env.
Irrlicht returns the XDisplay as a void* and XWindow as an unsigned long so reinterpret those as the appropriate type. Also fixed a spaces for tab formating issue
Fix 3 warnings reported by GCC 8.1 of the following type
```src/client/gameui.cpp:191:43: warning: « void* memset(void*, int, size_t) » effacement d'un objet du type non trivial « struct GameUI::Flags »; use assignment or value-initialization instead [-Wclass-memaccess]
memset(&m_flags, 0, sizeof(GameUI::Flags));
```
* Drop EventManager from GameDef & do some client cleanups
* EventManager is only used by Client. Don't expose it on Server & GameDef for nothing
* Drop Client::event() in favor of direct calls to getEventManager
* Cleanup some event put from new + put to put(new)
* MtEvent: add Type(u8) enum
* This will enhance event performance & ensure stricter type
* Drop MtEvent::checkIs (unused)
* clang-tidy reported fixes
* Code style
* Move event_manager.h to the client directory as it's only used by client
Add EventManager unittests + switch to unordered_map as order is not important here
Drop a unused function
* Cleanup sound manager client
* Use some const refs
* Use auto on iterators
* Drop unused parameters
* Move sound_openal.* to client folder
* Move sound.cpp + OnDemandSoundFetcher to client/ folder + reorganize includes properly
* Rename IWritableNodeDefManager to NodeDefManager
* Make INodeDefManager functions const
* Use "const *NodeDefManager" instead of "*INodeDefManager"
* Remove unused INodeDefManager class
* Merge NodeDefManager and CNodeDefManager
* Document NodeDefManager
The CSM HUD PR caused some strange behavior including aborts due to parts of it using some slightly hacky code, the event refactor changing how events are processed and a minor oversight.
* Make InputHandler own the key cache
* Add a helper function InputHandler::cancelPressed to avoid multiple similar calls in game.cpp
* Move RandomInputHandler::step definition into cpp file
Joystick input is a RealInputHandler only usage, make it intelligent and handle the joystick with keyboard direct.
This permits to remove many getters in game which should be owned by RealInputHandler
* Attempt to add registration confirmation
Using SRP auth mechanism, if server sent AUTH_MECHANISM_FIRST_SRP that means the player isn't exist.
Also tell player about the server and chosen username.
Local game has localhost as IP address of the server.
Add RenderingEngine::draw_menu_scene() to draw GUI and clouds background.
aborted -> connection_aborted
* Rewrite information message text
Client::promptConfirmRegister() -> Client::promptConfirmRegistration()
Other enhancements:
* Simplify setStatusText to showStatusText, as it shows the label too (preventing almost every setStatusText to call setStatusTextTime(0)
* Add unittests
Game class is too huge and has too specialization on various subjects, like UI, formspecs, client, renderer. Start to move UI related things to GameUI object and cleanup them
Other improvements:
* updateChat: more performance on error messages by remove string copies
* Initialize all game class members in definition instead of constructor (with nullptr instead of NULL)
* Drop unused Client::show{GameChat,GameHud,Profiler,GameFog}
* Add GameUI unittests
* Fix HP transport + some double <-> float problems
TOCLIENT_HP transport u16 hp as a u8, use u16 HP, this prevent HP over 255 to overflow across network
* Fix more double/float problem in serverpackethandler & remove implicit struct type for TileAnimationParams
* Fix connection unittests container
* Refactor clientevent structure
* Move structure outside of client header
* Create client events on heap not stack, this remove the ClientEvent object copy
* Use clientEventHandler to route events
* Cleanup network headers
* Move peerhandler to a specific header to reduce compilation times
* Move socket.cpp/h to network folder
* More work
* Network code cleanups
* Move socket.{cpp,h} to network folder
* Move Address object to network/address.{cpp,h}
* Move network exceptions to network/networkexceptions.h
* Client: use unique_ptr for Connection
* Server/ClientIface: use shared_ptr for Connection
* Format fixes
* Remove socket.cpp socket.h from clang-format whitelist
* Also fix NetworkPacket code style & make it under clang-format
* Code modernization: subfolders
Modernize various code on subfolders client, network, script, threading, unittests, util
* empty function
* default constructor/destructor
* for range-based loops
* use emplace_back instead of push_back
* C++ STL header style
* Make connection.cpp readable in a pointed place + typo
* Various code style fixes
* Use range based for loops
* Use empty instead of empty objects
* Use C++11 default keyword for trivial constructors and destructors
* Drop some useless casts
* Use emplace_back instead of push_back to improve performance of some vectors push
When minetest is launched, if there was no nameprovided in
configuration or parameters, the game would not show any error in
console. if the --go parameter was also prowided, the game would
exit without an error. This is undesired behavior, so this merged
commit add the missing function that displays the missing error
message in console.
* TileLayer: use shared_ptr for vector framespec
This reduce memory copy of TileLayer from (4 to 16) * FrameSpec where FrameSpec = (sizeof(int) + 3 * sizeof(ptr)) to int + sizeof(ptr)
Callgrind difference
Before: https://lut.im/RGkiJqQb8T/LeQIEXpAuRzfl7gd.png
After: https://lut.im/bcqmwee1xu/cTwtptY5tRuS9lp0.png
* Fix one push_back to use vector::emplace_back & optimize inclusions
Fixes issue #3576
* Clouds now take camera position as 3D, not 2D
* Cloud grid filling extracted to gridFilled method
* Clouds detect whether camera is inside cloud
* Camera in cloud changes fog by overriding sky colors
with cloud color
* Sun, moon and stars can be temporarily disabled
with setBodiesVisible
* Disabling fog also disables all "inside cloud" behaviors
* Irrlicht cleanup: cleanup various object to use RenderingEngine
* CAO doesn't need scenemanager in addToScene
* Camera doesn't need VideoDriver pointer or SceneManager in constructor
* Hud doesn't need driver & scene manager in constructor
* Hud doesn't need scenemanager pointer
* Tile.h doesn't need IrrlichtDevice header (just SMaterial)
* WieldMeshSceneNode: only take scene, we always use scene root node as parent
* Add Device3D class which will contain IrrlichtDevice interface
move getSupportedVideoDrivers to Device3D
Add Device3D singleton & use it in various places
Rename Device3D to Rendering engine & add helper functions to various device pointers
More singleton work
RenderingEngine owns draw_load_screen
move draw functions to RenderingEngine
Reduce IrrlichtDevice exposure and guienvironment
RenderingEngine: Expose get_timer_time() to remove device from guiEngine
Make irrlichtdevice & scene manager less exposed
* Code style fixes
* Move porting::getVideoDriverName, getVideoDriverFriendlyName, getDisplayDensity, getDisplaySize to RenderingEngine
Fix XORG_USED macro -> RenderingEngine + create_engine_device from RenderingEngine constructor directly
* enum paralax => enum parallax
* C++11 patchset 10: continue cleanup on constructors
* Drop obsolete bool MainMenuData::enable_public (setting is called with cURL in server loop)
* More classes cleanup
* More classes cleanup + change NULL tests to boolean tests
* Move KeyList & InputHandler from game.h to client/inputhandler.h
We have a header for inputs, move inputhandler class & related keylist object to it
Also introduce a cpp file for MyEventReceiver::OnEvent function in inputhandler.h because a so huge function doesn't needs to be inlined
* Pass clang-format on inputhandler.{cpp,h} (compatible)
Despite the split of TileSpec into TileDef and TileLayer, the
TileLayer struct is still 66 bytes large, and doesn't fit in
a single cacheline.
I'm moving the color member to cacheline 2, in the hope that it
is less used and the compiler loads all the hot members in a single
cacheline instead. Only color sits now in cacheline 2, all the
other members are in cacheline 1.
Note: is_color is probably rarely set, most nodes will likely
not use hardware coloring, but this may change in the future.
Ideally, this class is shrunk to 64 bytes.
* Fix progressbar for Android
Fixes#5599Fixed#5403
* draw_load_screen: use texturesource
this permits to unify texture loading code
* scale progress bar
* Add gl version check for GL_OES_texture_npot. This fixed the texture on loading screen
* Remove two sanity checks pointed by @celeron55
* sfan5 comments + android ratio fixes
Adds the possibility to colorize item stacks based on their metadata.
In the item/node definition you can specify palette (an image file)
and color (fallback color if the item has no palette or metadata).
Then you can add palette_index to the metadata.
Dropped itemstacks with different colors do not merge.
* Add joystick type detection (with joystick_type setting to override it)
* Fix multiple joysticks from interfering with each other by only reading from one (add joystick_id setting)
* Add support for Xbox controllers
* Some performance optimizations
This is globally removing some memory useless copy
* use a const ref return on std::string Settings::get to prevent data copy on getters which doesn't need to copy it
* pass some stack created strings to static const as they are not modified anywhere
* Camera: return nametags per const ref instead of a list pointer, we only need to read it
* INodeDefManager: getAll should be a result ref writer instead of a return copy
* INodeDefManager: getAlias should return a const std::string ref
* Minimap: unroll a Scolor creation in blitMinimapPixersToImageRadar to prvent many variable construct/destruct which are unneeded (we rewrite the content in the loop)
* CNodeDefManager::updateAliases: prevent a idef getall copy
* Profiler: constness
* rollback_interface: create real_name later, and use const ref
* MapBlockMesh updateFastFaceRow: unroll TileSpec next_tile, which has a cost of 1.8% CPU due to variable allocation/destruction,
* MapBlockMesh updateFastFaceRow: copy next_tile to tile only if it's a different tilespec
* MapBlockMesh updateFastFaceRow: use memcpy to copy next_lights to lights to do it in a single cpu operation
While experimenting with entities I ran into this unresolvable
error where the server is sending some texture that the client
crashes on. The crash prevents the client from ever reconnecting,
resulting in a server that has to use clearobjects.
We shouldn't crash but just ignore the object and move on.
```
0x00000000004dc0de in TextureSource::generateImagePart (this=this@entry=0x7118eb0, part_of_name="[applyfiltersformesh",
baseimg=@0x7fffffffbe98: 0x9f1b010) at /home/sofar/git/minetest/src/client/tile.cpp:1744
1744 u32 xscale = scaleto / dim.Width;
(gdb) bt
#0 0x00000000004dc0de in TextureSource::generateImagePart (this=this@entry=0x7118eb0, part_of_name="[applyfiltersformesh",
baseimg=@0x7fffffffbe98: 0x9f1b010) at /home/sofar/git/minetest/src/client/tile.cpp:1744
```
After reconnecting, the client now can connect without issues
and displays an error message:
```
ERROR[Main]: generateImagePart(): Illegal 0 dimension for part_of_name="[applyfiltersformesh", cancelling.
ERROR[Main]: generateImage(): Failed to generate "[applyfiltersformesh"
ERROR[Main]: Irrlicht: Invalid size of image for OpenGL Texture.
```
- Increase ContentFeatures serialization version
- Color property and palettes for nodes
- paramtype2 = "color", "colored facedir" or "colored wallmounted"
"[sheet:WxH:X,Y" assumes the base image is a tilesheet with W*H tiles
on it and crops to the tile at position X,Y. Basically it works
like "[verticalframe" but in 2D.
For testing, I combined the four default_chest images into one.
Inverts the given channels of the base image.
Mode may contain the characters "r", "g", "b", "a".
Only the channels that are mentioned in the mode string will be inverted.
Since the creation of minetest, it had no window icon on X11.
Now we have one.
The misc/minetest-xorg-icon-128.png file is a rendering of the
misc/minetest.svg file with inkscape, created with something like:
inkscape -z -e misc/minetest-xorg-icon-128.png -w 128 -h 128 misc/minetest.svg
We can remove the function in MtNativeActivity now
as it serves precisely that purpose: to tell irrlicht
that we handled the esc key.
TODO for later:
* Perhaps try to find a more performant container than KeyList
Changes:
* Fix indentation.
* Pass strings by const reference.
* Merge Strfnd and WStrfnd into one class instead of copying them.
* Remove trailing spaces.
* Fix variable names.
* Move to util.
* Other miscellaneous style fixes.
Fix two bugs related to the reconnect feature
introduced by commit
3b50b2766aeb09c9fc0ad0ea07426bb2187df3d7 "Optional reconnect functionality"
1. Set the password to the stored one
Before, we have done the reconnect attempt with a
cleared password, so using the feature would only
work if you had an empty password.
Thanks to @orwell96 for reporting the bug.
2. Reset the reconnect_requested flag after its use
the_game only writes to the reconect_requested flag
if it sets it to true. It never sets it to false.
If the flag is not reset after its use, all "reset"s
to the main menu will look like the server had
requested a reconnect.
The pageflip mode requires a stereo quadbuffer, and a modern graphic
card. Patch tested with NVidia 3D Vision.
The mini-map is not drawn, but that's what is done for topbottom and
sidebyside modes as well.
Also most of the time the user would prefer the HUD to be off. That's
for the user to decide though, and toggle it manually.
Finally, the interocular distance (aka eye separation) is twice as much
as the "3d_paralax_strength" settings. I find this a strange design
decision. I didn't want to chance this though, since it's how the other
3d modes interpret this settings.
- Add warning log level
- Change debug_log_level setting to enumeration string
- Map Irrlicht log events to MT log events
- Encapsulate log_* functions and global variables into a class, Logger
- Unify dstream with standard logging mechanism
- Unify core.debug() with standard core.log() script API
* Rename everything.
* Strip J prefix.
* Change UpperCamelCase functions to lowerCamelCase.
* Remove global (!) semaphore count mutex on OSX.
* Remove semaphore count getter (unused, unsafe, depended on internal
API functions on Windows, and used a hack on OSX).
* Add `Atomic<type>`.
* Make `Thread` handle thread names.
* Add support for C++11 multi-threading.
* Combine pthread and win32 sources.
* Remove `ThreadStarted` (unused, unneeded).
* Move some includes from the headers to the sources.
* Move all of `Event` into its header (allows inlining with no new includes).
* Make `Event` use `Semaphore` (except on Windows).
* Move some porting functions into `Thread`.
* Integrate logging with `Thread`.
* Add threading test.
Use wide_to_utf8 and utf8_to_wide instead of wide_to_narrow and narrow_to_wide at almost all places.
Only exceptions: test functions for narrow conversion, and chat, which is done in a separate commit.
Adds everything needed for SRP (and everything works too),
but still deactivated, as protocol v25 init packets aren't final yet.
Can be activated by changing the LATEST_PROTOCOL_VERSION header to 25
inside networkprotocol.h.
Change types for passed password from wstring to string, which removes converting
back and forth in most cases. Move the narrow_to_wide conversion, where its neccessary,
closer to irrlicht. Remove trailing spaces in guiPasswordChange.cpp. Make parameters for
translatePassword passed as const reference.
Move debug streams to log.cpp|h
Move GUI-related globals to clientlauncher
Move g_settings and g_settings_path to settings.cpp|h
Move g_menuclouds to clouds.cpp|h
Move g_profiler to profiler.cpp|h
* Combine client and server man pages.
* Update unit test options and available databases in man page.
* Add `--worldname` to man page.
* Fix a bunch of places where `"Minetest"` was used directly instead of `PROJECT_NAME`.
* Disable server build by default on all operating systems.
* Make `ENABLE_FREETYPE` not fail if FreeType isn't found.
* Enable LevelDB, Redis, and FreeType detection by default.
* Remove the `VERSION_PATCH_ORIG` hack.
* Add option to search for and use system JSONCPP.
* Remove broken LuaJIT version detection.
* Rename `DISABLE_LUAJIT` to `ENABLE_LUAJIT`.
* Rename `minetest_*` variables in `version.{h,cpp}` to `g_*`.
* Clean up style of CMake files.
