get_words, and in calling contexts nothing but a list of wids could
possibly make sense.

I need a break.

indexers.

CAUTION:  I'm sure I don't understand how persistency needs to be spelled.
Is it enough to say just that the base class derives from Persistent, or
does that need to be duplicated (or done instead exclusively) in the
derived classes?

Is there a point to keeping "import ZODB" in the derived-class files?  Is
there a point to keeping it anywhere <wink>?

logic to deal with all cases.  All the tests pass again.

globToWordIds():  This was building a list of words and then throwing
it away without referencing it.  Deleted the code.

If we update a document and reindex it, ZCTextIndex is currently
broken.  The test passes py virtue of calling unindex_object() after
each update, then calling index_object() again.  We need to fix our
code, and then remove the calls to unindex_object() from the test.

XXX This code causes OkapiIndex to fail because it doesn't expect to
have no wordinfo for a wid.  I tried to fix this in CosineIndex, but I
want to Tim think more about it and try to fix OkapiIndex.

This case can arise when the last occurence of a word is removed, or
when a lexicon is shared across multiple indexes.

XXX Not sure this code is correct, but it might be and the tests pass.
If it's wrong, we need more tests.

internals.

testStopWords().  That isn't the real point.  The real point is to
fiddle the scaffolding enough to make it possible to run a white box
test of the Okapi indexer too.  That isn't here yet.

with the docid is greater than 0 too.

and the Okapi indexer.  testZCTextIndex was, and remains, strongly tied
to the cosine indexer.

reinitializing to the global Index even if that wasn't how the object
was constructed.

It prints the the results and a summary of the times for the last 10
of 11 trials.

Otherwise, there's no way to report the total number of matches from
which the nbest were selected.

I changed all the functional tests that I knew how to fix, but not the
mhindex stuff.  All unit tests pass.

Removed QueryParser as a persistent attribute of the ZCTextIndex so that
it doesn't need to be persistent (It stores no state).

Updated tests. Functionally tested in Zope.

indexed (where the bytes are counted before entry into the pipeline,
and the words are counted after the pipeline is done).  To get the
numbers, use the _nbytes and _nwords instance variables directly.

(unless I erred ...).

intersection gimmicks into their own functions with their own test suite.

This turned up two bugs:

1. The mass weighted union gimmick was incorrect when passed a list with
   a single mapping.  In that case, it neglected to multiply the mapping
   by the given weight.

2. The underlying weighted{Intersection, Union} code does something crazy
   if you pass it weights less than 0.  I had vaguely hoped to be able
   to subtract scores by passing 1 and -1 as weights, but this doesn't
   work.  It's hard to say exactly what it does then.  The line
       weightedUnion(IIBTree(), mapping, 1, -2)
   seems to return a mapping with the same keys, but *all* of whose
   values are -2, regardless of the original mapping's values.

www.python.org.

next step is to add queries using ZCTextIndex

A little overzealous in the last checkin.