same results regardless of case.

(1) Change the mapping from zLOG levels to logging levels to use
    custom intermediary levels 15 and 5 for BLATHER and TRACE, rather
    than using a confusing skewed mapping.

(2) In the ZConfig datatype definition for a logging level, added
    'blather' and 'trace' to the level names, added 'warning' as an
    alias for 'warn', change 'all' to mean 1, and add 'notset' to mean
    0.  The semantics of NOTSET are very different than those of 1;
    setting a logger's level to NOTSET searches for a parent logger
    with a nonzero level.  The root of all loggers is initialized with
    WARN as its level, so setting the level to NOTSET effectively sets
    it to WARN rather than to the most inclusive level!

(3) In the schema, change the default level for handlers to notset, so
    they use their logger's level; the default level for the logger is
    set to info, corresponding to the old default for
    STUPID_LOG_SEVERITY.

properly, including that use of the these stream names uses a delayed
lookup of the streams).

configuration files easier to read.

is piping it to a program that reads logging message from stdin.

exist (no guarantees though, oh well).

it is already absolute; there may be a good reason.

Add an <eventlog> section to the runner schema.
Some comment and docstring tweakage.

types.

Add socket-name to runner conf template.
Check that the port isn't in use already.

acts more like what happens when you commit new object revisions using
the higher level interface.

- make it easier for an admin to configure logging levels
  (by making the defaults more sane)
- refactor the tests a little, in preparation for more tests

- remove methods that aren't needed
- rename "resolved" to "instance" for readability

- de-tabify everything

_deadline to None, to prevent reporting the same timeout more than
once.  Added a helpful comment too. :-)

intent according to the comment.

- Get rid of _lock() and _unlock(); _lock() is inlined, and so is
  _unlock(), after factoring out some common code at the end of
  tpc_begin() and tpc_end().

- In the refactored _unlock() code, only call self.timeout.end() if
  self.locked was actually set.

In the TimeoutThread class:

- Add some comments.

- Add some assertions.

- Get rid of the stop() method; we're a daemon thread so we'll be
  killed anyway; close_server() is only used from the test suite.

- Switch from using a lock + an event to a condition variable.  Be
  religious about doing stuff only while holding the lock.

- Inline the timeout() function; it shouldn't reacquire the lock
  anyway.

--Guido & Jeremy

earlier in the loop.  Basically, once we've got a tid from the
transaction record, we check it against our stop tid, and we check the
status looking for a 'c' flag.

the middle of a transaction.  This test isn't totally helpful in its
own right, but run it with logging and you'll see the bogus warning
(before the fix is applied).

Fixed to correctly munge sys.path.

that we can get distutils install to munge the #! line to insert the
proper Python executable file name.  (/usr/local/bin/python would work
too, but may not be the Python version that you intended to use, so
IMO it's better to require that you use an explicit Python interpreter
when invoking the scripts before they are installed.