My first impression with porting beta 1 to beta 2

At the time I began writing this it was 2:13am. I happily finished downloading XNA beta 2 at 11pm and have spent the last 3 hours porting my beta 1 code to beta 2. Most of it has been a fairly painless process however there are a few key areas which have been a huge pain in the rear.

To start I had 3 applications and 2 libraries to port:

1) DEMO_FractalViewer (app) - win32 app for testing various fractal terrain generation algorithms
2) DEMO_PlanetViewer (app) - win32 app for testing the procedural generation of planets
3) DEMO_Sandbox (app) - XNA game application for testing my deferred shading engine
4) Engine.Graphics (dll) - deferred shading engine along w/ everything else gfx oriented
5) Engine.Math (dll) - port of XNA's math library to double's (needed for generating planets)

To start I tackled Engine.Math as I figured it would be the simplest to port. It was as the only problem I had was the removal of the A, B, C, and D properties from XNA.Framework.Plane. Not a big deal and it was fixed within a few minutes. I then ported DEMO_Sandbox to the new beta 2 way of doing things which again only took a few minutes. Then the nightmare began...

Since my engine is built around deferred shading it relies heavily on using multiple render targets. I started by changing all of my Texture2D's to RenderTarget2D and all of my GetSurfaceLevel()'s to GetTexture(). My first problem was trying to figure out how to store a pointer to the backbuffer. I decided to call m_device.GetRenderTarget(0) as I assumed that the backbuffer must be a RenderTarget2D since there's no other way to get at it. This quickly exploded with a NullReferenceException as I had been using my pointer to the backbuffer immediately thereafter to get at it's Width and Height for generating the rest of the render targets. I investigated every possibly way I could think of to retrieve the backbuffer but alas it was all for naught. I resorted to using m_device.Viewport.Width and m_device.Viewport.Height and just left my backbuffer pointer null.

The next problem was an InvalidOperationException when calling GetTexture() after the first rendering pass. I was totally stumped with this one. After going through the GraphicsDevice docs for a few minutes I had a hunch I needed to call ResolveRenderTarget(). Conveniently the documentation for ResolveRenderTarget() informs you that the function "Resolves the render target at the specified index". I had kind of figured that out from the function header but I decided I'd give it a shot as I was otherwise clueless. After putting a call to ResolveRenderTarget() immediately after every call to SetRenderTarget() I was pleasantly surprised to see it solved the problem - despite the fact it was extremely annoying to find all of my calls to SetRenderTarget() and put a call to ResolveRenderTarget() afterward as there were quite a few.

This initial success was quickly destroyed by the fact that you cannot call SetRenderTarget(0, target) with target being null. I already knew this from beta 1 and DirectX but since I had no idea how I was supposed to set the render target back to the backbuffer without being able to get a pointer to it I figured I'd give it a shot and see what happens. At this point I said "screw porting the deferred shading" and just commented it out so I could focus on my other 2 applications and then come back to it later.

Porting my other 2 applications was extremely short lived. There was almost nothing that needed to be ported however I was stopped dead by one thing: BeginScene() and EndScene() were removed from GraphicsDevice. Both DEMO_*Viewer applications were standard C# form applications where I was using XNA to render into a panel within a splitter container. I'll be honest I was quite incredulous. I couldn't (and still can't) believe BeginScene() and EndScene() were removed. I searched and searched for a possible solution but came up empty handed. In the end I decided to just comment it out and see what happens. I never made it that far. I get an InvalidOperationException as soon as I try to create an index buffer (after making no changes to how the device is created from beta 1). All of the initialization is done in the constructor of the form after InitializeComponent() is called. At this point I decided I was done. It's getting late and I figure i'll just tell my tale of woe and see if anyone else is having similar problems :p.

So - to sum all of this up:

1) What are ResolveBackBuffer() and ResolveRenderTarget() for? The documentation restates their names and is of absolutely no help. I did read on a thread by someone else that you're supposed to call ResolveRenderTarget() after you're done rendering to it? I am currently calling it immediately after calling SetRenderTarget() and since I havn't been able to get past a null backbuffer I never got to a point where it would matter.

2) How do I set render target 0 back to the back buffer if I cannot get a pointer to it? I noticed getting at the SwapChain has also been removed so definitely stumped with this one atm.

3) VertexElementUsage.PositionTransformed is gone. I am using VertexElementUsage.Position instead and my initial assumption is it shouldn't matter but can anyone confirm that? I havn't been able to test it because of the aforementioned problems.

4) The removal of BeginScene() and EndScene(). Is this the death of all win32 apps that want to use XNA to render in a child control?

5) What are GraphicsDevice.Textures and GraphicsDevice.VertexTextures for? There is 0 documentation on it other than the fact it returns a TextureCollection and I was just curious :p (though i'll admit I havn't tried playing with it yet).

6) If render targets have to be recreated every time the device is reset then why do we have to manually recreate RenderTarget2D's ourselves? It takes a pointer to a GraphicsDevice in its constructor so if it's universal functionality why can't it register with the event and recreate its internal Texture2D itself? I apologise if this is in fact how it works but the documentation provides little information on this other than that it needs to be recreated after the device is lost.

I apologise if any of this is in the documentation and I missed it. I just wanted to post this now because it is an uncontaminated version of my initial 3 hour impression of beta 2 and I figured waiting until I had more time would, for better or for worse, skew it a bit. Unfortunately I havn't gotten to testing the content pipeline yet so i'll write another "first impressions" thread when I get to it. Hopefully this post can help people who are experiencing similar obstacles while porting their own code.

[6853 byte] By [KeithNewton] at [2007-12-27]