[jmarshall]

Ah - yep, you are correct in that we should generate the mipmaps as best as we can. I'm sure we can come up with a decent algorithm to do so if it doesn't already exist - it's a one-off operation on load, after all, and for thumbs and the like we could actually request that they're saved with the mipmap info included.

Cheers,
Jonathan

[CapnBry]

While genning mipmaps is the easiest solution code-wise, I'm not sure it is exactly the best solution. I mean they're great for minification when the camera is a variable and arbitrary distance from the texels, but in XBMC's case the size of images in world coordinates is known so we could be a lot smarter about how the images are resampled, and how their memory is managed. Of course, that adds the complexity of XBMC keeping track of texture usage and knowing that an image can be downsized without it affecting other controls.

I'm not sure why I typed all that out, because I'm sure you've already realized that. Allowing for a largeimage to be mipmapped would not be a bad thing.