The current frontend/backend split assumes that the frontend can use the postscript name, the matrix, and the screen font setting to uniquely identify a font to the backend.

If the backend has additional properties in a font, these need to be computed from the properties -gui gives. Since -gui often does a bunch of trivial font changes (eg. flipping fonts vertically, or switching between screen and printer fonts), and since all other properties need to be preserved when doing these changes, the backends must recompute their properties exactly. This is fragile, has caused problems in the past, and probably still has latent issues (eg. round-off problems are a risk).

It also means that backends can't provide fonts that don't fit in the "name/matrix/screen font" scheme, like fonts created from a block of data, or multiple-master fonts.

The internal font system and the frontend/backend split should be reworked to make it easier for backends to keep additional properties for fonts. This would make the system less fragile, and would allow us to support stuff like fonts from memory and multiple-master fonts without resorting to horrible hacks.