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.

content, but does not match the expected end tag, treat it as character data.
This is mostly useful when script includes string literal that include end
tags.

from 2.5 branch.

      - Collector 386: workaround for hanging FTP connections
        with NcFTP

don't have a corresponding .py file, to prevent tests that import
deleted modules from running using the stale bytecode files.  This has
bitten enough people enough times that it's time it became a standard
part of every test suite runner.  (Zope3 already has it.)

Somebody merge this into the Zope2 trunk please.

refactored much code that conditionally called v.read(). Also, allow the input to a converter to be a unicode string. This is not used by ZPublisher, but is helpful for other code that uses the converters such as OFS.PropertyManager

(revision 1.72 or Catalog.py).

TTW searches are no longer optimized to restore the old
behaviour (user does not fill out any form fields ->
return all hits)

For application related searches we keep the optimization
but it is possible to disable optimization by passing
'optimize=0' as additional parameters to searchResults().

Simplify initZopeSplitter() and remove unnecessary PyErr_Occurred().

Use string macros for objects that are guaranteed to be strings.

Remove unnecessary \ at end of line.

In innermost loop of splitter function, replace ASSIGN() macro with
Py_DECREF() and simple assignment.  The macro was doing more work
than necessary because it called XDECREF on an object that was
guaranteed not to be NULL.

Use less horizontal whitespace in next_word().

(without converting their contents...needs reindexing)

be needed for the ZMI