These are no longer used in Zope.

- remove commented-out cruftiness related to a no-longer-supported
  model

that the first thread to launch managed to commit each time, while the
other 3 threads got nothing but conflict errors.  Added some randomization
to boost the odds of success, and to prevent systematic favoring of any
thread.  That appears to work.

Synchronous is what matters for the tests.

Move two class variable definitions into __init__(), because class
variables defined in __slots__ are immutable.  This should still be a
net memory savings, because the a 8-entry dict and a 10-entry dict
both have the same allocated size.

tricks left name thread undefined at runtime.  Repair that.  Plus
add comments, use meaningful variable names, stop pretending we can run
on a system without threads, and don't use "id" (a builtin name) as a
local vrbl name.

caught by pychecker

recent changes:
- the addition of the "file" attribute of the import element
- the removal of the restriction about what sectiontypes components
  can extend

Please make all future changes on the Zope-2_7-branch instead.

checkTimeoutProvokingConflicts:  These shouldn't start a server; setup
already did.  Starting two servers caused intermittent failures on
Windows, due to lockfile contention.

+ If a storage times out between the vote and the finish, the ZEO cache
  could get populated with objects that don't make it to the storage
  server.

  A new flag self._midtxn_disconnect gets reset in tpc_begin() and set
  in notifyDisconnected().  If tpc_finish() discovers this flag set, it
  raises a ClientDisconnected error before calling tpc_finish() on the
  server.

+ In tpc_finish() we re-order the calls so that the server's tpc_finish()
  is called (and must succeed) before we update the ZEO client cache.

+ The storage name is now prepended to the sort key, to ensure a
  unique global sort order if storages are named uniquely.

+ Added new tests for the above (checkTimeoutAfterVote,
  checkTimeoutProvokingConflicts, checkSortKey).

Changes:

 - startup log handler now pays attention to the logging levels of
   the handlers defined within the config file and uses the "lowest"
   level to log messages to stdout during startup.

 - entirely removed warning when the starting user's umask is "too
   permissive".  it wasn't clear that it added any value under normal
   operations.

 - replaced ancient setuid code with code stolen from zdaemon that
   works the same but looks nicer.

Don't try to optimize by checking for disabled and effective levels within the log method.  The logger's level may not be meaningful if this logger is configured with handlers that accept messages at a lower level than that of the logger itself.

usage, python2.1 fixes, and fix one or two bugs (missing import, lacking
check for invalid database type, etc)