Bug in 3.141592653-2.6-0.999995 (TeX Live 2023) with fontspec and tabularray?

[Karl Berry]

Hi Ross - thanks for pursuing this further.

             // Use summed advances rather than rounded sum
             node_width(node) = advances_fixed;

(For clarity, in the future it would be good to send the actual diff
instead of just the new code. But yes, I see it in XeTeX_ext.c around
line 2100, as you said.)

The question that comes to my mind is, what does LuaTeX do here? 
Luigi (or anyone), when computing the width of a run of
OpenType/TrueType glyphs, does luatex take the width values (reals) from
the font and round each one individually, then sum them up? Or add the
reals and only round the sum?

Since converting to fixed once gives a more accurate result, it doesn't
seem right to convert the individual values.

Overall, I think the thing to find (somehow) is at what low-level point
LuaTeX and XeTeX compute something differently. --thanks, karl.