Improve text rendering output

[ensisoft]

Find a way to improve the text rendering output.

The attached screenshot shows multiple problems:

• clipping at the right edge
  • this is ok, it's just the text composition buffer that is too small
• clipping the vertical ascent (for example the bottom part of letter g)
  • line height issue, probably fixable by user setting a larger line height multiplier
• garbled text on multiple occasions

[ensisoft]

The raw text raster from the text rasterizer looks like this

[ensisoft]

A problem related to transferring the text to the screen is

• The text already has alpha blended anti-aliasing built into it with partially transparent pixels
• Sampling from the texture might filter and thus squashes multiple pixels together
• When drawing to the render targets color buffer there's no per pixel level precision to transfer the pixels from the rasterized text to exact pixel locations in the render target by drawing.
  • when the output is a MSAA render target more garbling happens due to aliasing.

Seems that ideally when the rendered output of the text matches exactly the rasterized text (as it should) we'd like to transfer the pixels directly (pixel per pixel) to the render target color buffer.

Theoretically one could resolve the MSAA FBO RT color buffer and then use a compute shader to write to the color buffer texture but this would not work when there's a need to do stencil testing.