>Ampersand entities are ASCII in the storage; they just get
>converted to accented letters or Greek or whatever by the browser.
>Are the more recent threads perhaps using the raw characters instead?
>
>(It is equally possible that I am seeing the ghosts of such bugs that
>I and/or my SO have had to fix in other software. It is depressingly
>easy to screw up non-ASCII handling even in quite recent code.)

Hm, yes, that seems plausible... if you've got one API that's counting UTF-8 characters -- and another one that's counting bytes -- and the code is using these counts to index into a flat text file, and expecting them to agree...that seems very much like something that could go sideways quite easily.

It also seems like the sort of thing where a system upgrade could "helpfully" change APIs' behaviors in the scripting language.