Related to #12130, it turns out that computing the line width on a char-by-char basis and as a whole str is different. From the docs of unicode-width, it's these cases where it differs specifically.

In the linter, the LineWidthBuilder determines the width by looking at each char individually. But, in some places the width is computed directly on &str which creates the panic as in #12130. These cases are:

In the formatter,