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.

ZCTextIndex has grown a new argument with a default value that can be
used to specify an Index class to use.  The default is OkapiIndex.Index.

There is a little kludge to make the test succeed.
testZCTestIndex.IndexTests uses the Index.Index tests instead of
OkapiIndex.Index.  Tim will probably fix this.

names.

Re-order imports so that all Zope imports go together and are separate
from all the ZCTextIndex imports.

Reformat _apply_index() doc string to use std Python style, which is
one-line summary followed by paragraphs of text that start at the same
offset as the function name.

Do comparison of None using is instead of ==.

Some additional plug-in index APIs were added to ZCTextIndex and support APIs added to Index and Lexicon.

_apply_index does not use NBest since ZCatalog has an incompatible strategy for finding the top results. NBest might be abstracted from this product for general consumption in application code.

inside the while loop either.

first byte -- we always find the end of a particular encoded number by
searching for the next byte with the high bit set.  This simplifies
the encoding and gives us more space for small encodings: 128 values
can now be encoded in 1 byte, and 16K in 2 bytes.

_indexedSearch():

    Simplify logic that called _apply_index() for each index in the
    catalog.  The if statement under the comment "Optimization" had
    identical code on either branch.  Perhaps the odd indentation made
    this confusing.  Regardless, remove the conditional.

    Change computation of normalized scores to multiply first, then
    divide.  Use literal 100. to make sure mult and div are floating
    point ops.

searchResults():

    Simplify logic at beginning of searchResults().  The first two
    conditionals depended on kw, so organize the logic to make that
    clearer.

    Write helper method to find "sort-on" and "sort-index" instead of
    duplicating code in searchResults().

    For case were results are sorted, simplify construction of the
    final LazyCat and make it more efficient to boot.  Instead of use
    a list comprehension and a reduce + lambda to construct list and
    length of contained lists, do it with one explicit for loop that
    constructs both values.

        Note: I did detailed timing stats on three ways to compute the
        length of a sequence of sequences.  reduce + lambda was the
        slowest.  For short lists, an explicit for loop is fastest.
        For long lists, reduce(operater.add, map(len, list)) is
        fastest.  The explicit for loop is big win here, because we've
        got to walk over the elements anyway to undo the Schwarzian
        transform.

Sundry:

Use getattr() with default value of None in preference to hasattr()
followed by getattr().  This gets the same result with half the work.

Changes for consistent and frequent use of whitespace.

Use types.StringType and isinstance() to test for strings.