[cout]

I had an idea the other day to use 64-bit character codes. Then each code can be interpreted as an 8x8 bitmap. Every character would be guaranteed to have a unique bitmap representation. The bitmaps wouldn't bet used for rendering, of course -- but they could be used as a fallback if your font does not define a character. Anyway this would somewhat avoid the problem you describe because two characters that look the same visually would have the same value. Nothing I'll ever implement of course, just a thought experiment.

[JdeBP]

As someone who has worked with 8×8 fonts I can report that you'd have some surprising problems with that idea. Not only would you have problems with there being two distinct forms of letters like "a" and "g", making things over-unique; and not only is it tricky to differentiate forms that actually are not the same, because they are from two different alphabets (especially the 13 extra "mathematical" alphabets); but it's actually quite difficult to make pre-composed forms in that amount of space.

8×8 is a tight squeeze, and 16×16 works a lot better. But that would make your approach vastly more space hungry than a normalization approach using the actual Unicode code points.

* https://github.com/jdebp/unscii/tree/2.1.1f