> The original problem is rare enough

You're wrong to call it rare. Dave Milter, David Engster, and I have
seen it separately. It has occurred because of tasks in Semantic,
Speck, and now NXML. That input-pending-p behaves this way is
apparently surprising to the Elisp developer. It is even more
surprising to the user upon witnessing such a jarring symptom.

> So I tend to agree with Barry that input-pending-p should not run
> timers. Not just based on his particular problem case, but on the
> basis of what ideally input-pending-p should do.

I don't claim the change solves the issue completely. This bug report
would have to stay open because Semantic also calls
accept-process-output while point is on an excursion.

Since Stefan had said input-pending-p behavior seems wrong, I followed
up in the hopes we can both correct a wrong and reduce how common this
user-visible symptom is.