finish up lexical warnings

omitting parked=>1 should be a warning, not an error?

demo/tail-f
  see File::Tail or GNU textutils for implementation examples

add timeout for var watchers?

find owner for Win32 port

perhaps some help with distributing SIGCHLD?

documentation
  how to design new event types (in perl)
  how to design new event types (in C)??

mjtg@cus.cam.ac.uk signal discussion:
  The API rsignal overrides the current signal handler.  If the signal
  watcher is cancelled (or otherwise disabled) then rsignal is reset to
  SIG_DFL.  Maybe the original handler
  should be restored.  On the other hand, assignment to %SIG will mess
  up Event watchers so maybe Event should refuse to watch a signal if a
  handler is already installed.

optimization
  update io event-mask without invalidating cache
  determine IntervalEpsilon dynamically & make available from perl
  use fancy profiler tricks to figure out how to boost performance
    queue events from both sides?
  try to hoist system calls from the inner loop

there seems to be some sort of bug in the shutdown phase
  Attempt to free unreferenced scalar during global destruction.
  Use of uninitialized value during global destruction.
  Explicit blessing to '' (assuming package main) during global
  destruction.

rethink
  pe_check_recovery & ENTER/LEAVE points

increase the number of priority levels?
  ASYNC    0    1    2    3    4    5    6    7    8    9   IDLE
                         HIGH           NORM

loop
  StarvePrio?

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

Move process.pm to external module?

accept "%f minutes" & "%f hours" for all time specifications?


On Wed, Aug 30, 2000 at 02:50:20PM -0400, Joshua N Pritikin wrote:
> Wow!  I am reading about the Linux linked list implementation.  In
> comparison to Linux, Event wastes an unnecessary pointer on every link.
> I'd better fix this!  :-)

False alarm?  Linux is able to make the assumption that the next & prev
pointers are always the same offset from the top of the structure.  This
assumption is true for pe_watcher.all, pe_event.peer, pe_event.que, and
pe_qcallback.ring.  However, pe_timeable.ring is stored at lots of
different offsets.  Hm hm.  I guess some bytes are wasted but otherwise
Event will need two different kinds of linked lists...