Thanks for link. But it would be nice to know where this inaccuracy could be seen. Fonts are equally sharp as in 1280-mode. This applies to mouse cursor too. Can you really pinpoint which app and in which situation you see less sharp elements with scaled resolution?
And I know what a blurry non-native LCD looks like. This a completely different thing.
I don't think fonts are equally as sharp. I don't have a Retina machine in front of me, but this is how I understand the scaling process:
(1) User chooses perceptual resolution. For example, 1680x1050 (on a 13" Retina Pro, the highest "scaled" resolution).
(2) Display is rendered at double the perceptual resolution. Continuing the example, all of the content and UI elements are doubled in pixel dimension to be scaled to 3360x2100. For elements like text and vector graphics that don't have pixel dimensions, those elements are drawn to twice their equivalent pixel size, to have the same perceptual size at 3360x2100 as they would have in a 1680x1050 workspace. For bitmaps, they are scaled using some method of interpolation to double their pixel dimensions. If an application is "retina-aware" and has high-resolution resources, those resources will be used for interpolation instead of the lower-resolution resources.
(3) The resulting 3360x2100 display image is scaled down to the panel's native resolution using some type of downsampling algorithm (so on a 13" Retina Pro, 2560x1600).
For text to appear just as sharp at a Retina scaled resolution as at the "Best for Retina" setting, the text would have to be rendered
after the other display elements were upsampled and downsampled. In my steps above, that would have to occur as a step four, rather than at step two. Otherwise, the clarity of the text will suffer somewhat as a result of the downsampling that occurs at step three (since the image isn't being downsampled at an even integer). In other words, as I understand the process, the text will be rasterized during step two, and thus will suffer from the inherent inaccuracy of downsampling to a non-integer resolution during step three.
Do you understand the process to work differently?