I originally wanted to get of the legacy IVideoDriver::setRenderTarget altogether,
but that ended up being too much work.
The remaining usage is in "dynamicshadowsrender.cpp".
Here's a comment I wrote about the workaround:
----------------------------------------
Use legacy call when there's single texture without depth texture
This means Irrlicht creates a depth texture for us and binds it to the FBO
This is currently necessary for a working depth buffer in the following cases:
- post-processing disabled, undersampling enabled
(addUpscaling specifies no depth texture)
- post-processing disabled, 3d_mode = sidebyside / topbottom / crossview
(populateSideBySidePipeline specifies no depth texture)
- post-processing disabled, 3d_mode = interlaced
(probably, can't test since it's broken)
(populateInterlacedPipeline specifies no depth texture)
With post-processing disabled, the world is rendered to the TextureBufferOutput
created in the functions listed above, so a depth buffer is needed
(-> this workaround is needed).
With post-processing enabled, only a fullscreen rectangle is rendered to
this TextureBufferOutput, so a depth buffer isn't actually needed.
But: These pipeline steps shouldn't rely on what ends up being rendered to
the TextureBufferOutput they provide, since that may change.
This workaround was added in 1e96403954 /
https://irc.minetest.net/minetest-dev/2022-10-04#i_6021940
This workaround should be replaced by explicitly configuring depth
textures where needed.
----------------------------------------
This should improve compilation speed.
Things changed:
* Prefer forward-declarations in headers.
* Move header-includes out of headers if possible.
* Move some functions definitions out of headers.
* Put some member variables into unique_ptrs (see Client).
