branch *damn CVS*

       with utf-8 encoded strings.

tearDown().  One result is that all tests here except the first died
on Windows with a winlock.error; this is probably unique to lock semantics
on Windows, and doesn't show up at all unless you've installed a modern
Berkeley DB on Windows (unless you do that, these tests don't even try
to run on Windows).

for that the passed-in err[0] may be a string, and when it was died
with an AttributeError in excname().  The verbose_on_error branch does
cater to err[0] being a string, so copied that logic into the showAll
branch.

This prevents problems during test debugging where a logfile isn't
generated and a module might be half-imported.

second of wall-clock time.  That's plenty, except when, e.g., a
burst of network activity takes the CPU away for a second.  Jeremy saw
it fail once yesterday when running Zope-2_6-branch pre-release tests,
and that was unfortunate timing <wink>.  I found that running it in a loop
eventually caused a spurious failure whenever I ran it too.

So, in just this test, boosted the timeout by a factor of 60.  This should
make it astronomically unlikely to fail.  Also sped the test by removing
unneeded .keys() and .sort() calls.  Also switched from using Python's
assert stmt to unittest's assertEqual() facility.

cycle is left behind that includes a SafeMapping, and SafeMappings are not
aware of GC, therefore the cycle can't be collected.  Iterator is also in the
cycle, however, and it does not need a strong reference to the Context object,
since the Context refers to the Iterator.  So Iterators now refer to Context
through a weak reference.

importing AccessControl.User specifically will now function correctly.

reason "import Zope" is often required in unit tests.

from a dict with a float value.

Mapping_update():  This squashed errors raised by PyObject_SetItem(),
changing them to a "not a 2-tuple" TypeError instead.  Unclear why, since
2-tupleness was checked earlier.  Fixed by backporting part of the
ZODB4 BTree construction code, which doesn't have this problem.  Bonus:
it's faster now too.

Make sure test_MultiMapping doesn't rely on the order in which dict
entries are printed.

I'm hesitant to see this fix really works, except that I wrote a very
minimal test suite that covers all the cases I'm aware of.  The
previous checkin caused numberic operations when the Missing object
was on the right-hand side.

with new Zope3 interfaces

Fixes crash reported by Tres Seaver.  Missing.Value offered to coerce
itself to unicode.  unicode_compare() blew up because it expected that
the other argument had been converted to unicode.

XXX But why does this implement coercion at all?

  - Do what the docstring says:

    If -i is specified or no action is specified on the command line, a
    "shell" interpreting actions typed interactively is started.

Port multiple selection property fix from 2.6 branch. Also undid a previous fix which prevented multiple select properties from being added TTW.

addressed, so I just improved encodeExpansion()'s compression behavior.

specifically set.

selection" properties better (ie. default to an empty list, not empty string)

- normalize whitespace

  it more readable
- remove unused instance variable

PageTemplateFile.__init__() does some weird magic.
Without this, the sniffer tests failed on Windows.