Technical notes on my code submission of 12 Jan 00
                        Kern Sibbald

General:
- Brian Daniels reported that the master/slave code does not
  properly shutdown the slave. Due to the fact that in 3.7.0,
  the slave code calls do_action(), which was not the case
  in 3.6.2, additional variables needed to be passed from the
  master to the slave (or major changes to apcaction.c were
  needed, which I ruled out at this stage). Consequently,
  the network structure needed to be changed. The original
  implementation, while it worked, did not lead to easy 
  modification. Consequently, I made a major rewrite of the 
  code. Additional coding should be done as noted in my
  todo list as well as the source code of apcnet.c, but it will wait
  until after 3.7.0 is released.        

  Note, there are very few changes to the "core" part of apcupsd, so
  the chances of new bugs are minimal. Of course, the chances of needing
  additional work on the master/slave code is far from minimal given
  the extent of the changes.

Changes submitted this submission:
- Throughout the code, I changed the names of local variables to
  correspond to the global names. This reduces the chances of an
  error in purely mechanical coding.
- apcaction.c, I implemented the send_slaves_urg flag to indicate
  that the slaves should be immediately notified. This consisted
  of setting the flag for an urgent condition (e.g. on batteries)
  and clearing it when off of batteries.
- apcnet.c -- this was a major rewrite at the same time keeping
  the major logic flow.

  I added an error state to the SLAVEINFO packet as well as
  an error count. The former allows printing an error message only
  once per error rather than once every 10 seconds. The latter, in
  future code, will permit us to DOWN any connection with excessive
  errors.  I also set the string length of usermagic[] to the correct
  size rathr than MAXSTRING thus reducing a fair amount of wasted
  memory.

  I modified the state variable remote_state. The new states are 
  described at the top of the file. There is a lot less interaction
  of the states between the master and the slave now.  The only
  case where the slave needs to act on the state of the master is
  in the case of an initial connection or a reconnection.

  In the old code, the master always sent a packet to the slave and
  waited for the slave to repond. However, this was really only necessary
  during the connection phase. Consequently, in the new code, the slave
  responds to the master only on a connection or reconnection.

  write_struct_net(), read_struct_net(), reconnect_slave(), and               
  the contents of prepare_master() are all gone.     

  Previously, during initialization, apcupsd would block for a very
  long period of time waiting for the slaves to come up (5 minutes
  per slave, if I am not mistaken). The new code does not block for 
  either the master or the slaves. It simply logs whether or not
  the connection was successful and continues trying at the interval
  defined by the user NETTIME.

  Also, during an urgent situation as noted above, NETTIME is ignored
  by the master and instead, it sends updates once a second. 
- I increased the time a master waits from 15 to 30 seconds
  before issuing a powerkill. This gives a bit more time to
  the slaves to properly shutdown.
- apcstatus.c: in response to complaints about too much output from
  André, I have reordered the STATUS output so that the items that
  I consider the most important are grouped toward the beginning.
  This was a pure reordering of the code, no functional change was
  made.
- apcupsd.c: I improved the netslave powerkill and prepare_slave()
  error messages now that I understand what was being done.
  I pass the slave pid so that if the slave process exits so does
  apcupsd. I changed the process names so that they more correctly
  reflect what is being done in the process rather than who they
  are talking to.
- During testing, I noticed that I introduced a harmless bug in 
  apccontrol for all the machines. The cancel of the shutdown should
  only be done if the powerfail file was created.  Fixed.
- I eliminated a lot of old out of date stuff in apc_defines.h that
  had to do with master/slaves.
- Updates to the structures in SLAVEINFO, and NETDATA.
- Updated the manual to reflect the changes.