This is only a high level overview. For detailed changes, see the git changelog.
The versioning scheme changed with this release. This is the successor to Ogre 1.12 and the porting efforts are comparable with porting from 1.11 to 1.12.
Ogre was put on a dependency diet and now you can get a fully functional build without any additional dependencies. We replaced zlib and libzzip in Main by miniz and use a bundled stb_truetype for Overlays if you dont provide Ogre with freetype.
While miniz provides a seamless replacement, there are visual differences when using stb_truetype due to differences in glyph metrics.
Traditionally Ogre allowed you to place Cameras and Lights in the scene without attaching them to a SceneNode. This resulted in three different positioning APIs for Cameras, Lights and SceneNodes with slight inconsistencies (e.g. Camera::move vs. SceneNode::translate).
Furthermore, this complicated things when e.g. attaching a Camera to a SceneNode as you might have an additional transform "hidden" inside Camera.
Therefore, the use of node-less positioning was deprecated since Ogre 1.10.8. With this release we disable the according API and no longer store any transforms inside Light and Camera so you have to port your code like:
// Ogre 1.12 (warning: ‘setDirection’ is deprecated)
mLight->setDirection(...);
// Ogre 13 (also 1.12 to get rid of warnings)
mSceneMgr->getRootSceneNode()->createChildSceneNode()->attachObject(mLight);
mLight->getParentSceneNode()->setDirection(...);Alternatively, you can set OGRE_NODELESS_POSITIONING=ON in CMake to get back the old behaviour.
All constants are stored in a single byte-buffer instead of being split by
type.
This maps to how Uniform/ Constant buffers work and allows updating a backing hardware buffer with a single memcpy instead of having to explicitly collect the individual floats.
Consequently, physicalIndex is now a byte offset instead of a slot offset.
The high-level APIs based on String names and logicalIndex are not affected by this change.
HighLevelGpuProgramPtr is now an alias for GpuProgramPtr. While the class separation still exists, user code should use both interchangeably.
The only user-visible change is that calls setPreprocessorDefines(...) should be be changed to setParameter("preprocessor_defines", ...).
Furthermore, HighLevelGpuProgramManager is now merely an alias for GpuProgramManager. Internally, this saves us quite some shunting. Most users should not notice - unless you were creating low-level and high-level programs with the same name. This will trigger an error now, as both live in the same resource-namespace.
Historically, Ogre only used only public and protected visibility. While this makes it easy to customise Ogre classes for a specific use, it is not so great when it comes to committing to a stable API. Previously, you could (accidentally) use a inherited class member that we would later change during an internal refactor, causing your code to break. While we could exclude all protected members from our API guarantee, it would be even better if the compiler would check that you only use the stable API - thats exactly what private visibility does. Consequently, most protected members are private now.
shadow_caster_material and shadow_receiver_material are better alternatives available since Ogre 1.6. These not only simplify code, but also allow for efficient caching of generated shaders.
Nevertheless, there is fallback code so you still can use old .material files. Here, a caster/ receiver material will be generated on demand. However, you are highly encouraged to define those explicitly.
- Size of the Particle class was reduced by 12% which results in about 13% faster rendering
- Frustum is not a Renderable any more. Rendering Frusta is now done by the DebugDrawer.
- ShadowTextureListener was factored out of SceneManager::Listener
- SWIG 4.0 is now fully supported
- Flollowing SemVer, incrementing PATCH now guarantees ABI compatibility. Therefore SOVERSION on Unix system now only contains two digits e.g.
libOgreMain.so.13.0instead oflibOgreMain.so.1.12.13.
VET_COLOUR*is now an alias forVET_UBYTE4_NORM. On most RenderSystems VET_COLOUR would be already mapped to VET_UBYTE4_NORM anyway. A notable exception is D3D9, where this changes the channel order from BGRA to RGBA. If you are using custom vertex shaders, you must update them accordingly. For all other cases Ogre will handle this for you. If a buffer conversion is performed on loading, Ogre will issue a warning in the log.- The default transform space of
SceneNode::setDirectionchanged fromTS_LOCALtoTS_PARENTto aid with the node-less positiong transiton. - The
parentWindowHandleandexternalWindowHandleWindow creation parameters are now equivalent on all platforms.- On Win32 the
parentWindowHandlebehaviour is gone and no child window is ever created. - On Linux the
externalWindowHandlebehaviour is gone and an externalGLXDrawablecan be only set via an external context.
- On Win32 the
DisplayString is now merely an alias for String. The UTFString class was replaced by a much smaller (and faster) implementation that is not exposed. For all practical purposes, you can just pass UTF8 encoded strings through String.
Glyph placement has been improved. Previously Ogre only considered glyph width for typesetting. However, this did not take into account that glphs can overlap, as happens with italic fonts. Therefore, there is now an independent property GlyphInfo::advance. Ogre internal code was updated for this change. However, if you directly access the texture atlas, you have to update your code.
An Exception is now thrown, when querying a GlyphInfo of a Font that is not loaded. Make sure to call font->load() before querying.
There is now a RTSS stage for Weighted, blended Order Independent Transparency, which implements the according algorithm by Morgan McGuire. See the transparency Sample for how to integrate it into your pipeline.
The RTSS Normal Map stage, is no longer a lighting stage but only does normal mapping. This allows using it with per-pixel lighting as before, but also enables combining it with the GBuffer stage to enable normal mapping in the GBuffer
The PSSM3 stage now also supports colour shadows in addition to depth shadows. Colour shadows are automatically used for PCT_BYTE texture formats.
The Plugin now supports exporting via a generic SceneNode::saveChildren API. This allows you to dump your dynamically generated Scene to file and later inspect it with ogre-meshviewer, which also got improved .scene support.
Shader interfaces/ subroutines are no longer available. They were cumbersome to use, while adding overhead to all rendersystems.
Constant buffers are now fully supported, so if you write cbuffer Foo in HLSL, you get the data that you declared previously as shared_params Foo in your Ogre material script.
Atomic buffers are no longer supported. They were only partially implemented, while being cumbersome to use and adding overhead to the core loop. SSBOs offer the same functionality, while proving a convenient API.
Separate shader objects are now used by default, if available. This enables an improved shader parser using the GL introspection API and allows mixing GLSL and SPIRV shaders.
The Option "Multi device memory hint" was renamed to "Auto hardware buffer management" as it is generally enables memory recovery on device loss.
This new Plugin enables Ogre to automatically compile GLSL shaders to SPIRV that in turn will be loaded by GL3+. This means you can now easily use the Khronos reference compiler instead of what your driver would do - similar to how things work on D3D11.
The component was removed. The RTSS or just plain shaders in Ogre materials are a better way forward.