----------------------------------------------------------------------

In brief, Xlib, Xt and most widget sets have no mutual exclusion for
critical sections.  Any interrupt handler is likely to leave one of
the above libraries in an inconsistent state -- such as all the
appropriate flags not yet set, dangling pointers, in the middle of a
list traversal, etc.  Note that the ANSI C standard points out that
behavior of a signal handler is undefined if the signal handler calls
any function other than signal() itself, so this is not a problem
specific to Xlib and Xt; the POSIX specification mentions other
functions which may be called safely but it may not be assumed that
these functions are called by Xlib or Xt functions.

The only safe way to deal with signals is to set a flag in the
interrupt handler.  This flag later needs to be checked either by a
work procedure or a timeout callback.  It is incorrect to add either
of these in the interrupt handler.  As another note, it is dangerous
to add a work procedure that never finishes.  This effectively
preempts any work procedures previously added and so they will never
be called.  Another option is to open a pipe, tell the event loop
about the read end using XtAppAddInput() and then the signal handler
can write a byte to the write end of the pipe for each signal.
However, this could deadlock your process if the pipe fills up.

Why don't the Intrinsics deal with this problem?  Primarily because it
is supposed to be a portable layer to any hardware and operating
system.   Is that a good enough reason -- I don't think so.

        Note: the article in The X Journal 1:4 and the example in O'Reilly
Volume 6 are in error.