'read_char' in src/keyboard.c returns -1 to indicate the end of a
keyboard macro. This case is supposed to be propagated via 
'read_key_sequence' and 'command_loop_2' to 'Fexecute_kbd_macro'.

But 'read_key_sequence' is not the only caller of 'read_char'. It is
also called by 'read-event' and similar lisp functions, and in that
case, the magic C return value -1 is wrongly propagated to the lisp
caller. 

There are at least workarounds for this bug in at least three lisp
modules in the code base, in subr.el, in calc and most recently added
in dbus.el (bug #62018), see the attached patches. These patches are
supposed to resolve the underlying issue, and then remove the
workarounds.