>things like glyph ordering, bidi resolution, cursor movement, hit‑testing, line breaking, and font fallback
I think part of the problem is that things like this are often just barely working on the one set of test data that often just barely covers the fundamental requirement in its most simplified "happy path" form to begin with. LTR in this case.
And then the coders move on to get the next objective barely working, but nobody ever goes back far enough to reinforce the weak points in the fundamental structure before it's a far more challenging task. And then if done it's still too challenging to go all the way back.
I think part of the problem is that things like this are often just barely working on the one set of test data that often just barely covers the fundamental requirement in its most simplified "happy path" form to begin with. LTR in this case.
And then the coders move on to get the next objective barely working, but nobody ever goes back far enough to reinforce the weak points in the fundamental structure before it's a far more challenging task. And then if done it's still too challenging to go all the way back.