OGRE (Object-Oriented Graphics Rendering Engine)

Change Log

v0.10.0 (28 Feb 2003)

Changes since 0.9.7:

v0.9.7 (26 January 2003)

Changes since 0.99f:

v0.99f (1 December 2002)

Changes since 0.99e:

v0.99e (25 October 2002)

Changes since 0.99d:

v0.99d (5 September 2002)

Changes since 0.99c:

v0.99c (19 July 2002)

Changes since 0.99b:

Enhancements:

Work-in-progress:

Bugfixes:

v0.99b (9 June 2002)

Changes since 0.98b:

Enhancements:

Bugfixes:

v0.98b (29 March 2002)

Changes since 0.97e:

 

v0.97e (16 Feb 2002)

Changes since 0.97b:

v0.97b (6 Feb 2002)

Changes since 0.96b:

 

v0.96b (6 January 2002)

I've changed this log a little to be more concise regarding changes made between snapshot versions.

Changes since 0.95b:

6 January 2002

Most bugs in shader parsing sorted out. A few shader effects are not supported yet, but most are there and look good. Texture coordinate modifications are missing for the moment, but environment mapping is in (albeit a little different from Q3A's method - I chose to use cameraspace reflection vectors to produce the effect because these are hardware accelerable, whilst Q3A's method of using the vector from camera location to vertex is not), as are all the shafts of light, animated flames etc.

No MD3 models are being rendered so there are a few blank spots in the level where a gargoyle, wall sconce etc should be. I'm not going to implement this for now, I've proved the point I wanted to prove which is that Ogre can handle this type of scene well. It's just a few Q3-specific details which are missing which is no big deal, I never set out to produce a working copy of Q3A.

I will be releasing another snapshot based on this codebase, and also will release binary demos of all the examples.

Next: texture coordinate modifications, extensions to the controller classes. Then particle systems.

5 January 2002

Skydomes and skyboxes are in and working, giving you 3 types of sky to play with. Skyboxes are for static, wrap-around environment textures like space, whilst skydomes are like skyplanes, but curved.

Quake3 shader script conversion is going well, Ogre loads in the .shader files and converts them to Material objects as required, meaning that you can now see flames, shafts of light, TV screens etc inside Ogre. Animated textures are being loaded but not animated yet - I'm going to try and get animation working within a nice generic OO controller framework if I can rather than bash out a single-use approach.

One thing I've noticed now the sky is being displayed is there are 'sparklies' around the curved surfaces visible mostly when there is a light-coloured sky. Damn. It appears the points generated from the bezier patches don't quite match up with the surrounding static geometry, meaning there are small gaps. This is a TODO.

I still need to fix the odd crash point and a few shaders aren't being loaded properly. Also environment mapping needs a bit of work since Quake3 bases them on camera position -> vertex rather than camera eye vector -> normal, which means they look wrong. I should alos render the 'models' in the level i.e. geometry that doesn't have lightmaps like gargoyles.

27 December 2001

Skyplanes are now working. You can now create a plane which is at contant distance from the camera at and orientation for use as a sky. I'll be moving on to skyboxes imminently which will allow full wrap-around skies.

I have also totally revamped the example code that is included with the engine. Instead of the single TestApp project, there is now an Examples folder with multiple projects under this which demo each effect. I have also created an object-oriented framework for the examples too, meaning they are now a lot better structured and hopefully better to understand.

I also intend to release some binary demos soon so people can quickly see what the engine is capable of without having to build from the source. The good news this week is that I just did a Release build of the Quake3 level renderer, and it runs at a very sweet 170fps in windowed mode on my Athlon 1Ghz and GeForce 2. In fullscreen mode at 1024x768 at 32-bit colour it still manages between 50 and 80 fps. I'm very pleased indeed. I would like to tidy up some of the missing textures before I do a binary release so I'll be sorting out the skybox support and other texture effects first.

24 December 2001

Refactored the entity rendering routines in the base SceneManager class to make them easier to reuse in subclasses.

Material class enhanced to contain all the other render settings such as depth buffer parameters, culling modes, texture filtering etc. It is now possible to drive all the rendering settings via the Material class now so that no calls are required to the RenderSystem class. This makes the settings more intuitive to use, and ensures that you can have different settings per object if you want by assigning different materials, and the SceneManager controls the changes to the rendering settings.

Major rewrite of SceneManager::setMaterial and added RenderSystem::_setTextureUnitSettings (used by the former) so that only changes to the material settings are sent to the RenderSystem. This ensures that rendering state changes are minimised and that custom settings are reset only when required. This has improved the framerate across the board by about 50%. A further optimisation which is still to be made is ordering the materials in the render sequence to minimise the render state changes still further, e.g. scheduling 2 materials that have the same layer 0 texture but different layer 1 textures next to each other.

Started adding sky plane and sky box support to the SceneManager class, but this is not yet functional.

Merry Xmas!

4 December 2001

General Bezier Patch support through the PatchSurface class.

Quake3 level loading uses curved surface support to display Q3A's bezier curves. Custom shader scripts still to be done.

IMPORTANT: To compile & run the TestApp demo application you will need to change line 414 of WinMainTest.cpp to point at your Quake3Arena pak0.pk3, since I can't distribute Q3A levels and textures for size and copyright reasons.

11 November 2001

First Quake3 level rendering release. No support for shader scripts or curved surfaces yet.

12 October 2001

Modified the material blending routines to make it more reliable to fall back on multipass rendering whilst preserving all the correct effects. To do this I had to simplify the basic multitexture blending operations which automatically fall back to multipass, and require that the application explicitly specifies a fallback blend multipass operation if it chooses to use the more complex blending modes which are only available under multitexture hardware. I hope by doing this I keep it simple for most cases (people normally use jusy ADD, MODULATE, REPLACE or ALPHABLEND anyway) whilst still allowing complete flexibility for those who like to tinker with the obscure modes.

Decided to make another point release even without the finished Quake3 stuff because there's so much that's useful.

9 October 2001

Object/scene blending operations (as opposed to multiple texture layer blending on the same object) support has been added, allowing new effects like transparency.

I keep getting distracted from my Quake3 level work but I am progressing - the reason why I switched to this subject is that I realised my fallback technique to multipass rendering when not enough multitexturing hardware was available was not correct, and in order to be able to fix it I needed these new blending options. I'm now wondering whether to fix the multipass fallback now I have the tools or to leave it until later. Right now, if you use more texture layers than the hardware supports in a single pass, things won't look right. I'll get to it.

6 October 2001

Ok, remember the problem from 16 September re texture coordinates and confusion over whether (0,0) was the top or bottom left corner of the texture? Well it turns out that once again Direct3D is the odd one out, and 3DS is actually using the more common convention, i.e. (0,0) is the bottom left corner. I have therefore changed the 3ds2oof routine back to how it was before (using 3DS UV corrdinates directly) and am flipping the texture itself in the Y-axis when loading it in the Direct3D texture implementation to compensate (D3DTexture.cpp). I should have checked more thoroughly before changing things to be in line with D3D, I should have known it would probably be the odd one.

I've done this because I want Ogre to follow industry standards (e.g. right-handed coordinates, anticlockwise vertex winding) rather than Direct3D's viewpoint which is usually the opposite, even though it's Direct3D I'm working on at the moment. I do this because it's what everybody will be used to when they come from maths texts or modelling tools, and is the same as the OpenGL approach (so the OpenGL port should be easier than Direct3D). I just compensate inside my Direct3D-specific classes (e.g. by flipping the Z axis, changing the culling mode, or flipping the texture as above).

5 October 2001

Progressive updates are being made to the OgreEngineIndoor module in CVS. Quake3 levels are being loaded and coversion is in progress.

I have also fixed a couple more bugs: the ColourValue class was implementing colours a little too bright and wrapped around to black when pure white was specified. This has been fixed. In addition, the mipmap generation was not generating enough mipmaps under Direct3D because D3DX ignores requests for the number of mipmaps and generates as many as required to reduce to a 1x1 texture. This caused smaller mipmaps to be black and thus textures would wink out at long distances. This has been fixed.

16 September 2001

Slow going due to moving house. However, fixed a problem with texture coordinates as converted from 3DS using 3ds2oof - I just realised that 3DS uses UV coordinates where V=0.0 is at the bottom of the texture map, instead of the top as with everything else! That explains the strange textures on some models. Fixed 3ds2oof to make V=(1.0 - V) during the conversion. Reconvert your models!

29 August 2001

Ogre now supports loading of resources from more flexible locations, including compressed archives (.zip and Quake3 .pk3 files). You can specify a list of search paths and archives which are either general across the whole of Ogre or specific to a resource type (e.g. textures, meshes).

I've also added support for loading TGA images as well, since Quake3 maps use them in places.

I have also dropped support for BMP files. This is because I have had to implement memory-based versions of the image decoders to support loading from archives, and to be frank I couldn't be bothered to implement one for BMP since it's a naff format in my opinion and tied to the Windows platform anyway. Ogre supports JPEG, PNG and TGA which should be plenty of flexibility. Personally I think when you've got PNG (for lossless) and JPEG (for lossy), you don't need anything else.

10 August 2001

Texture blending is now in and working! The new TestApp demonstrates the functionality by loading a base model with a standard texture, then adding a second texture layer and applying an environment map to it. The keys 0-7 change the texture blending mode to show what effects different parameters have.

I'm now moving onto other features; I want to look at Quake3 file loading - there are a couple of things nagging me about managing materials and entities that I'd like to look at but I'd really like to get started on Q3A bsp files now. I also wanted to work on animation, both within models (skeletal animation & bones) and for paths of objects (bezier curves & bsplines), but I just don't have time to do everything I want to! Q3A files first.

5 August 2001

Back from a break for holidays!

The Camera class now allows a SceneNode to override it's position and orientation when the Camera is attached to a SceneNode. This will allow cameras attached to world objects etc.

Started implementing the texture blending operations - the interface is complete, I'm now moving onto the Direct3D implementation. For the moment I'm not worrying about minimising the number of times that changes to blending parameters are made during a frame - I'm still concentrating on minimising the texture changes, which is the main overhead anyway. More optimisation will likely come along with new SceneManager subclasses.

21 July 2001

The Camera class now uses quaternions too. Camera rotation is a lot more concise now (instead of storing & trying to update 3 vectors I just have a single quaternion) and is more predictable and hopefully more flexible for the future. 

You can attach a Camera to a SceneNode (this is meant for use in having cameras tracking other objects in the world) but I have yet to implement this - I intend now that they're both on the same basis that if you choose to attach a camera to a scene node then the SceneNode overrides the camera's own position and direction, and maybe even that they update each other (so telling a camera to look somewhere would update the orientation of it's SceneNode too).

19 July 2001

Updated SceneNode class to use quaternions internally for all orientations / rotations. This should both be faster and give me more flexibility for the future. Also added a lot more manipulation routines the SceneNode class which were sorely lacking before if you wanted to move stuff around. The Camera class still uses matrices internally but I intend to change that so that a Camera depends on the SceneNode it is attached to for positioning rather than doing it itself. To this end I will need to add extra features to the SceneNode class to replicate functions like lookAt(), which is ok since this will be handy for pointing other objects at each other too.

17 July 2001

Fixed an issue with the 3ds2oof utility - it now generates vertex normals based on smooth groups rather than just averaging based on the faces which use the vertex. This removes artefacts along model 'seams' e.g. the point where a texture wraps around on the model.

Reimplemented environment mapping using the new Material::TextureLayer structure. Started work on adding more manipulation features to the SceneNode class (body code still to do).

4 July 2001

Python now has full access to the engine, and it's working great. I've also enabled callbacks into Python from the engine, so you can register a Python class as a listener for frame update events so C++ calls Python every frame. Nice.

Future Python developments include an embedded version of Python in Ogre so you can build a compiled application which still used interpreted scripts for some things - this will be useful for level scripting etc where the script can be kept in the level data. For now though I'm going back to developing the core engine features .

20 June 2001

Spending an awful lot of time exposing all the necessary objects and their interfaces to Python (up till now I have it working but only a few interfaces were exposed). About 90% there now, with a few issues to iron out - I had to extend the compiler heap limit to get it to compile so far since the heavy usage of templates puts a great strain on the compiler memory usage. In a few days I should be in a position to do all my engine testing through Python as well as through a custom C++ application (currently TestApp).

13 June 2001

Added FrameListener and FrameEvent classes in order to represent the frame update callback functions in a more object-oriented way. User applications now extend FrameListener and register their interest in receiving frame events - this is based on the Java model of event listening. Note that a boon of this method is that Boost Python allows you to extend existing C++ classes so this provides a perfect mechanism to have C++ call Python scripts when events occur.

I intend to extend this pattern to other events like collisions when I implement them.

10 June 2001

Changed from using SWIG to Boost Python for Python script integration. I changed because Boost Python supports C++ features much beter (including catching C++ exceptions and turning them into Python exceptions, support for namespaces and overloaded operators), it integrates better with my build cycle (doesn't require a separate generate from interface files and can be built directly into the library), and is more up-to-date with the latest Python libraries. I was finding that SWIG's emphasis on pure C as opposed to C++ was getting in the way, and Boost has definitely addressed this.

Introduced PythonLink class to expose classes to Python.

7 June 2001

Implemented Python integration using SWIG (Simplified Wrapper and Interface Generator). You can now start and initialise Ogre from a Python console and drive it interactively. More work required to expose the full interface to Python however.

31 May 2001

Cut over to CVS - CVS repository is now up and running at SourceForge.

20 May 2001

SceneNode class integrated with Entity / SubEntity classes. Working basic scene graph. Also reimplemented my previous hack using DirectInput as a proper platform-specific user input class - added the OgreUtility module for this purpose.

15 May 2001

SubEntity implementation complete, together with customisable Materials at the SubEntity level. Material usage by SubEntities is now indexed by the SceneManager so that rendering can be grouped by Material such that as few as possible render state changes take place. Note that this does not currently deal with transparency.

13 May 2001

Bugs with the new Mesh loading routines now ironed out.

24 April 2001

Implemented Mesh, SubMesh and MeshManager classes for loading Ogre Object Format files (converted from .3ds using the Ogre 3ds2oof tool). Decided to allow the original 3ds mesh to be broken down into chunks specific per material or grouped as a whole depending on requirements. The 3ds2oof utility makes a goos guess as to what is best, based on the percentage of the vertices each SubMesh uses out of the whole model - the lower the percentage the greater chance it will have a dedicated vertex list rather than sharing the whole since that saves sending unnecessary vertices to the pipeline. However this can be customised with parameters since you may want to use vertex buffers where it's more efficient to store 1 buffer even if you only use parts of it.

1 April 2001

Began overhaul of Mesh loading routines.

30 March 2001

Switched to Doxygen for document generation since ccdoc still doesn't support namespaces. Doxygen is better at cross-referencing anyway and draws nice diagrams.

24 March 2001

Added JPEG loading support, and moved this and existing image loading code (PNG) into the Image class. BMP loading is still supported but only on Windows platforms so this is still residing in the D3DTexture class as a special case.

Improved the handling of non-square textures on hardware that supports only square textures. The image is now stretched into the square area so that the relative texture coordinates are still valid and no black bands appear on the model.

23 March 2001

Multiple viewports per window is implemented and working Ok. Z-Order also applies so picture-in-picture views are possible by placing a smaller viewport on top of another larger one.

11 March 2001

Changed the default mechanism for supplying vertex components (position, normals, texture coords, colours) to the RenderOperation class. They were previously assumed to be part of the same buffer with flags indicating which were present. However this was too restrictive and prevented you using entirely separate memory areas for different components, and adding extra components to vertices on the fly.

Instead the RenderOperation class now takes pointers for each component, and a 'span' value determines the gap between each value. This allows you to use separate buffers if you like, but also use a combined buffer if that suits you (by using the 'span' value to skip over values in between).

6 March 2001

Implemented a 3DStudio conversion tool that works directly from the .3DS files rather than from ASCII exports (as previously) or from a MAX plugin as some others do, which requires a 3DS license. I based my code on an existing open source routine but I have exteded it since it had a few shortcomings. The tool is called '3ds2oof' because it converts 3DS files to a proprietary format, .OOF (Ogre Object File). This is because it's more efficient for me to do extra processing at conversion time like calculating bounding boxes, splitting vertex buffers, strippifying etc (not that I'm doing all this yet, but I can) rather than at load time as I would have to do if I loaded direct from 3DS inside Ogre.

1 March 2001

Over the last 2 weeks I reimplemented my interface to the DirectX rendering pipeline in the new Ogre structure, under the D3DRenderSystem class. Rendering requests are now packaged up in the RenderOperation class, giving me more flexibility on how rendering requests are submitted and handled. Now all the support classes (mesh loading, scene management) are needed.

11 February 2001

I've been picking apart the platform dependencies in Ogre so it'll be more portable. I'm also making it flexible enough to handle any number of rendering windows and viewports within thse windows, and also giving you the choice of an automatic rendering loop (max frame rate) and a render-on-demand interface much like you would need for an application like a modelling tool where you want to free up cycles when things don't move and frae rate isn't the most important thing.

I now have a nice hierarchy crossing the OgreEngine and OgreWin32Platform modules, with functionality broken down into RenderTarget, RenderWindow and Win32RenderWindow classes. Other platforms can be implemented later, just replacing the Win32RenderWindow maybe with XRenderWindow or MacRenderWindow.

29 December 2000

Ogre development starts again after almost a year's break while I finished my degree. Lots of research and design work at this stage to work out how I should structure the object model.