The CMake step of the CI runs take about 7 minutes for MacOS, and about 12 minutes for iOS. In contrast, it takes about 1.5 minutes on Windows and 0.25 minutes on Linux. Just watching the builds progressing, it seems to be that the tests for headers/functions/macros is fairly slow (a couple seconds each, maybe).

Possibly we should consider caching some of the results of the header/function/macro checks, e.g. using something like https://github.com/cristianadam/cmake-checks-cache.

Additionally/alternatively, we can see if some of the checks can be grouped together CHECK_INCLUDE_FILES instead of CHECK_INCLUDE_FILE to reduce the total number of checks.

The samples set two different window titles:

In Sample::Setup():

In Sample::Start() (via Sample::SetWindowTitleAndIcon())

Hardly a breaking issue, but one of them should be removed. I would prefer the second, but only because I use a sample as a base for my own projects, and I can overwrite the samples engineParameters_ in my own Setup() function.

ObjectAnimation supports saving to/from XML and JSON. However, when loaded as a resource through BeginLoad (e.g. through the ResourceCache), it will only load it as XML, even if you pass a JSON file. It should instead check the filename extension to deduce whether it is xml or json (or maybe even the content).

Only want to let you know.

I much prefer the fmt library's style of formatting with {} over the traditional printf style formatting with %s, %d, %f, etc. Mainly it's not an issue, as there's already some code I use to add it to Urho, but all of the formatted log macros like URHO3D_LOGDEBUGF use the printf style.

Rather than directly replacing them, an alternative would be something like URHO3D_LOGDEBUGFMT, but that is a really long name. If we're fine with some breaking changes, I would prefer renaming the old one, possibly to something like URHO3D_LOGDEBUGX, and use URHO3D_LOGDEBUGF for the fmt-style formatting.

There is a fairly large amount of special behavior in our build system that is conditional on the CI environment variable. When github executes the workflows, this is set to true, but when running a build locally this is unlikely to be the case.

Is there a reason for this difference?

If not, I would propose that we remove as many of them as feasible, as it should make debugging failed workflow runs simpler, as the logic will be the same between a local machine and the remote machine without having to specifically try to emulate the CI environment.

I think there are several options to improve the QA workflow (which I'm pretty sure just fails each time at present).

1. We can improve the .clang-format file to better reflect Urho's conventions, as well as update it for more modern clang-format (some of the parameters have changed, at least compared to version 14 I have on my laptop).
   • I suspect that at least one commit will need to be made that will apply many style changes throughout the code, but hopefully we can keep that "many" fairly small.
2. We can actually have the QA run push a commit that applies the clang-format results to the code when we have PRs. This means developers do not have to install clang-format themselves, and keeps behavior consistent with the version of clang-format used by the Github Actions runner. The following actions seem to allow it, though there may be other better ones:
   • https://github.com/marketplace/actions/clang-format-lint
   • https://github.com/EndBug/add-and-commit
3. We can enable clang-tidy linting, though I suspect that may also require some work to have it not needlessly flag lots of existing Urho code.

An issue to discuss switching to STL containers for 2.x.

From the project description, we are aiming to be (emphasis added)

Productive: Do not reinvent the wheel. Engine should reduce development time

Switching to the STL means that users have to reinvent the wheel slightly less when looking for code snippets on the internet. But it is also a significant break in compatibility with existing code, as our containers use CamelCase methods, while the STL uses snake_case.

Some questions to start off:

1. Yes or No to switching for 2.x?
2. Which containers (e.g. do we include String)?
3. Which STL (the STL, EASTL, etc.)?
4. Backwards compatibility: Is there anything to aid the user in the transition?
5. What about where the STL is less convenient? E.g. Vector::Contains

https://github.com/recastnavigation/recastnavigation

Currently U3D has AngelScript in the same state as it was in Urho3D 1.8 -- half refactored, half WIP.
1vanK did quite a lot of work on binding generator and this work spanned years.

Since U3D forked from the middle of this work, AS is in weird state here. Bindings are not manual, but they use unfinished generator.

So I think U3D should either abandon AS at some point, rollback AS binding generator and restore manual AS bindings, or cherry-pick AS binding generator updates from Urho3D 1.8-1.9

The meaning of CustomVariantValue::Compare is ambiguous. What does returning true indicate? From the Variant::operator == code (below) we can infer that true means equal, but from the name and documentation within CustomVariantValue it is not obvious whether true means "Equal" or "Less", etc.

At the very least, the documentation should be improved. Alternatively, the function name could be changed to Equals, which would itself clear up the ambiguity.

( This is a copy of issue urho3d/urho3d#2784 )
Some AnimatedSprite2D elements disappear or become weird squares after the OpenGL context is lost. This is most easily reproducible in the web samples (it immediately happens if the window isn't 1024x768), but can also be reproduced by enabling/disabling full screen in OpenGL desktop builds.

The issue does not seem to happen if the AnimationSet2D file uses a spritesheet (it's the case of the Orc in samples 49 and 50).

I set up a self-hosted github actions runner on my mac. The "Scaffolding - cleanup" step of the macos.yml workflow deleted my projects folder on my mac.

      - name: Scaffolding - cleanup
        run: rm -rf ~/{.urho3d,projects}

This should not happen. Fortunately, I had nothing of importance in that folder, just a recently checked out copy of U3D with no important changes. But our workflow should not be creating/deleting these directories in the home folder. The directory should be chosen inside whatever "_work" directory the github runner uses, rather than in the home directory.

That, or we need to provide documentation warning of this possibility. Let's just say that if this had happened on my laptop instead of this new computer I would have been extremely unhappy.

The engine should be able to be included as a submodule in a CMake project. This allows the game developer to use his own build system, or VCS, to manage engine versions. It also gives him the ability to patch any bugs in the engine without major workflow disruptions.

Relevant discussion: https://discourse.urho3d.io/t/embedding-the-engine-as-a-submodule-and-cmake-module/1636
How rbfx does it: https://github.com/rbfx/sample-project/blob/master/CMakeLists.txt#L40

CollisionShape supports being overloaded to allow users to wrap additional btCollisionShapes through UpdateDerivedShape. In working on a MultiSphereHullShape (for btMultiSphereShape), I realized the API could be improved - the UpdateShape function that is called that actually updates the shape_ member is private. This can be worked around by calling SetShapeType twice, but that updates the shape twice rather than once. Instead, I think we should just make UpdateShape protected. Possibly also MarkShapeDirty.

I intend to share the MultiSphereHullShape code as well, so remind me if I forget. Using some geometry nodes with modern Blender, I have a fairly decent workflow for creating them using an edge to mark the position and diameter of each sphere that make up the hull. They have proven fairly useful in generating complex close-fitting collision shapes to detect damage to my game's creatures, while in theory (I have not tested much) remaining more performant than mesh-based alternatives.

The present logo does not work well on a dark background. E.g.

It stands out better with a light stroke applied to it, though I'm not sure that I'd say it looks better. I will leave it to the artists to figure out.

I may be misunderstanding something, but it seems to me that this produces an implicit circular dependency:

The target is tolua++, which has a source file ${CMAKE_CURRENT_BINARY_DIR}/generated/toluabind.c, yet this file is the output of the custom command that uses tolua++ to produce the output (though CMake is not actually told that tolua++ is a dependency of the generated file).

Hi gurus,
I started to learn programming games with the great framework U3D (Urho3D), and in progress of learning the engine I faced some problems. And searched the whole web for the solution, and no luck.

I exported 3D scene made in Blender to .obj, then imported to the U3D Editor.sh. Every thing looks good but the shadows not exactly what I expect to look, but that's not the serious problem I faced. Then saved the file.

The big and serious problem is when imported it to U3D it's look like couldn't find other than one of the material files or the textures exported with it.

Here is the scene I made in Blender:

And here is when imported to the U3D Editor.sh, every thing in place, but the shadows not settled as it in blender:

And eventually here is what the scene look when imported the my app, there is just one material with it's texture image for all the objects in the scene:

And here is the code I used to create the scene and imported materials:
////////////////////////////////////////////////////////////////////////////////
// (4-1) Create sci-fi scene static model
{
scifi = scene_->CreateChild("scifiPlanet");
//scifi->SetPosition(Vector3(-1, 1.6, -3));
scifi->SetPosition(Vector3::ZERO);
scifi->Scale(1);

    StaticModel* object = scifi->CreateComponent<StaticModel>();
    object->SetModel(cache->GetResource<Model>("Models/Sci-fi.mdl"));
    object->SetMaterial(cache->GetResource<Material>("Materials/Corridor_ways_mat.xml"));
    object->SetCastShadows(true);

    RigidBody* scifiBody = scifi->CreateComponent<RigidBody>();
    scifiBody->SetCollisionLayer(2);

    CollisionShape* shape = scifi->CreateComponent<CollisionShape>();
    shape->SetBox(Vector3::ONE);
}

////////////////////////////////////////////////////////////////////////////////

So, where is the wrong here. What mistake I did in the code please?

The audio fails unless the user clicks the page before the game starts. Firefox reports the following warning in the console:

An AudioContext was prevented from starting automatically. It must be created or resumed after a user gesture on the page.

The issue happens on Sample 14: SoundEffects, so I don't think it's just an issue with my code.

Discussed upstream in SDL here: libsdl-org/SDL#6385

This is tied to #12. Follow up on:
https://www.gamedev.net/forums/topic/713287-m1-apple-progress/

It is easy to reach an infinite loop when using the windows cmake scripts. For example, if you call script\cmake_vs2022.bat build -DURHO3D_HOME="C:\Path\To\U3D\Build", you will get such a loop, as cmake_generic.bat only recognizes -D options when they are followed by a space.

The loop

:loop
if not "%~1" == "" (
    if "%~1" == "-D" (
        if "%~2" == "MINGW" if "%~3" == "1" set "OPTS=-G "MinGW Makefiles""
        if "%~2" == "URHO3D_64BIT" if "%~3" == "1" set "arch=-A x64"
        if "%~2" == "URHO3D_64BIT" if "%~3" == "0" set "arch=-A Win32"
        set "BUILD_OPTS=%BUILD_OPTS% -D %~2=%~3"
        shift
        shift
        shift
    )
    if "%~1" == "-VS" (
        set "OPTS=-G "Visual Studio %~2" %arch% %TOOLSET%"
        shift
        shift
    )
    if "%~1" == "-G" (
        set "OPTS=%OPTS% -G %~2"
        shift
        shift
    )
    goto loop
)

should be changed to something like

:loop
if not "%~1" == "" (
    if "%~1" == "-D" (
        if "%~2" == "MINGW" if "%~3" == "1" set "OPTS=-G "MinGW Makefiles""
        if "%~2" == "URHO3D_64BIT" if "%~3" == "1" set "arch=-A x64"
        if "%~2" == "URHO3D_64BIT" if "%~3" == "0" set "arch=-A Win32"
        set "BUILD_OPTS=%BUILD_OPTS% -D %~2=%~3"
        shift
        shift
        shift
    )
    else if "%~1" == "-VS" (
        set "OPTS=-G "Visual Studio %~2" %arch% %TOOLSET%"
        shift
        shift
    )
    else if "%~1" == "-G" (
        set "OPTS=%OPTS% -G %~2"
        shift
        shift
    )
    else (
        set "OPTS=%OPTS% %~1"
        shift
    )
    goto loop
)

to prevent the possibility of endlessly looping on an unrecognized argument.

I recently traced a performance issue in the web build of my game down to doing logging during a character's attack. My build used the default shell (constructed by the UrhoCommon.cmake module from the default emscripten shell.html), which renders the console output both to the java script console and to an element on the page, which it then scrolls to the end of. This triggers a reflow of the text to get the element height, and since the log gets longer and longer, this takes a longer and longer amount of time, dropping frame rate drastically (down to ~3FPS).

The Urho3D shell.html file does not have this problem, as it just prints to the JS console (here).

I think the best solution to this issue is to use the Urho shell file by default in user projects, and not one based on the emscripten default. As a bonus, this shell looks much prettier and is more suited for games (full window, with simply a button to make it full screen) than the default shell (which is also meant to support console apps and such).

Can we add python support to u3d?We can use python as the script of u3d rather than AngelScript because AS is really difficult to learn.I think that could be more convenient to noobs and popularize u3d.
Besides,python should be easier when we want to implement something need multi-thread(like DirectX11,it can implement multi-thread rendering at a high level,of course the efficiency can not compare with Vulkan,but it can also provide us more convenience if u3d support Vulkan in the future)
I think we should keep pace with the times since we separated from urho3d.Maybe we should just abandon AngelScript because the difficulty of AS just like C++,but the performance of it just like lua and the coding efficiency of it is too low to compare with lua.

https://github.com/u3d-community/U3D/actions/runs/3949612987

CMake Error at cmake/Modules/FindDirectX.cmake:86 (message):
  Could not find MinGW system root.  Use MINGW_SYSROOT environment variable
  or build option to specify the location of system root.
Call Stack (most recent call first):
  cmake/Modules/UrhoCommon.cmake:504 (find_package)
  CMakeLists.txt:33 (include)


-- Configuring incomplete, errors occurred!
See also "D:/a/U3D/U3D/build/ci/CMakeFiles/CMakeOutput.log".
script/cmake_mingw.bat "build/ci"  -D URHO3D_EXTRAS=1 -D URHO3D_DOCS=0 -D URHO3D_DATABASE_SQLITE=1 
    -D URHO3D_TESTING=1 -D URHO3D_OPENGL=1 -D URHO3D_LIB_TYPE=SHARED -D URHO3D_DEPLOYMENT_TARGET=generic 
    -D CMAKE_INSTALL_PREFIX=C:/Users/runneradmin/.urho3d/install/win-mingw
Error: Process completed with exit code 1.

Presently we only support mono and stereo audio. SDL seems to support up to 7.1 surround sound, and I don't think it would be that much work to add support for it.

• Opening the sound device would become a little more complicated, as we presently open stereo and fall back to mono if that fails. I think we would just replace this with decreasing the number of channels by 1 until we get to mono or succeed in opening the device with whatever the user requested.
• Mixing would need to be rewritten. Presently we have separate functions for each combination of (from mono|stereo)(to mono|stereo)(interpolated or not). I think we could probably rewrite these into only a couple functions with template parameters specifying these (from channels, to channels, interpolated).
• Panning would need an additional dimension for Front/Back. And while we are at it, we might want to just make it a Vector3 to support full spatial audio in the future. Presently it is only possible to pan a mono source sound, and I think it should remain this way, and stereo/surround sound audio should be handled as is (the user can always separate multi channel audio into mono tracks if they really need the per-channel panning control).

Layout as set up in editor, and saved:

Layout after loading the saved data:

The texts inside the layout in the pictures do not have a fixed width (I haven't set their max and min widths).

Any update to the layout after the load, like manually editing the text in any of the entries in the editor, corrects the spaces, but the vertical layout can remain huge if there is an element, like a button, set with infinite max vertical scale.

A word-wrapped text does not set its own width based on its content, so, when not inside a layout, it becomes a very tall zero-width column until you set a different width for it. This width also won't be saved unless you make it fixed (set min and max width to the same value).

Fixing the text's width also does not cause the spaces to appear in a vertical layout, so it's one of the ways to avoid the problem. However, I find it a bit cumbersome to fix all texts' widths inside a layout, especially when the layout seems to already handle that for you (all elements are stretched to fit horizontally if you don't limit their widths... including word-wrapped texts)

Github is deprecating set-output in their runners: https://github.blog/changelog/2022-10-11-github-actions-deprecating-save-state-and-set-output-commands/

Starting 1st June 2023 workflows using save-state or set-output commands via stdout will fail with an error.

We use this in checking the source checksum in the build runner

Presently this just produces the following warning, but the end-date is listed above.

Warning: The set-output command is deprecated and will be disabled soon. Please upgrade to using Environment Files. For more information see: https://github.blog/changelog/2022-10-11-github-actions-deprecating-save-state-and-set-output-commands/

Since the repository was renamed, do we want to shorten the namespace, macro, and folder names as well?

I vote yes for at least the macro names, I don't have much of an opinion on the rest. Though I think we should keep Urho3D is in the copyright statements in the code/license.

It is impossible to save in Serializable in scripts, all Write functions return a false ( void Save(Serializer& serializer) ). For example, if I run 18_CharacterDemo.as by Urho3DPlayer.exe from batch file and save&load a scene there is also doesn't work saving, the character just rotate to 0 yaw and 0 pitch. If I run CharacterDemo from C++ it works because there are URHO3D_ATTRIBUTE.