Changelog
=========

3.0.1 (2013-10-15)
------------------

- Fix BooleanIndex when index inversion occurs as a result of reindexing
  and existing document with the opposite value. Fixes LP #1236354.

3.0 (2013-02-24)
----------------

- Strip white space from name when adding a column or index.

- Forward compatibility for Zope 4 removal of RequestContainer.

- Optimize brain instantiation, by creating underlying record items in a
  single step, instead of creation and three update calls.

3.0b1 (2012-07-19)
------------------

- LP #727981: Fix DateIndex ZMI browsing for dates in the first month of a
  year.

- Unify Unindex and DateIndex search logic (`_apply_index`) adding `not`
  support to DateIndexes.

3.0a2 (2012-04-26)
------------------

- Fixed another issue with preserving score values, when a custom index was
  queried first which was neither ILimitedResultIndex aware nor return scores,
  and a later index was of the default ZCTextIndex type.

3.0a1 (2012-04-22)
------------------

- Expand query report, to cover details on sort indexes, order and limits.

- As part of each progress handler report, also do an automatic transaction
  savepoint, to give the ZODB cache a chance to do garbage collection.

- Added a `threshold` argument to the catalog's `addColumn` and `delColumn`
  methods and used it for a progress handler. Also optimized some of their
  internals.

- Added support for `sort_on` queries with any number of sort indexes and
  differing `sort_order` values. For example:
  `{'foo': 'a', 'sort_on': ('foo', 'bar')}`
  `{'foo': 'a', 'sort_on': ('foo', 'bar'), 'sort_order': ('', 'reverse')}`
  `{'foo': 'a', 'sort_on': ('foo', 'bar', 'baz')}`

- Added support for `not` queries in field and keyword indexes. Both
  restrictions of normal queries and range queries are supported, as well as
  purely exclusive queries. For example:
  `{'foo': {'query': ['a', 'ab'], 'not': 'a'}}`
  `{'foo': {'query': 'a', 'range': 'min', 'not': ['a', 'e', 'f']}}`
  `{'foo': {'not': ['a', 'b']}}`.
  Note that negative filtering on an index still restricts items to those
  having a value in the index. So with 10 documents, 5 of them in the `foo`
  index with a value of `1`, a query for `not 1` will return no items instead
  of the 5 items without a value. You need to index a dummy/default value if
  you want to consider all items for a particular index.

- Updated deprecation warnings to point to Zope 4 instead of 2.14.

2.13.22 (2011-11-17)
--------------------

- Added a new `load_from_path` class method to the `PriorityMap`, which allows
  one to load a plan from a file, instead of a module via an environment var.

2.13.21 (2011-10-20)
--------------------

- Refactored value index logic. Determine value indexes per catalog instead of
  globally. Store value index set in the priority map, so it can be seen in the
  ZMI and stored in the module level storage.

- Added support for using ZCatalog as local utility.
  This feature requires the optional `five.globalrequest` dependency.

2.13.20 (2011-08-23)
--------------------

- Fixed incorrect calculation of batches in the second half of the result set
  in sortResults.

2.13.19 (2011-08-20)
--------------------

- Increase plan precision to 4 digits in its string representation.

2.13.18 (2011-07-29)
--------------------

- In the string representation of a catalog plan, round the times to at most
  two digits after the comma.

2.13.17 (2011-07-29)
--------------------

- Put back the `weightedIntersection` optimization but guard against results
  with values and do the appropriate fallback to the weighted version.

2.13.16 (2011-07-24)
--------------------

- Restored preserving score values from ZCTextIndex indices.
  https://bugs.launchpad.net/zope2/+bug/815469

2.13.15 (2011-06-30)
--------------------

- Fixed undefined variables in BooleanIndex inline migration code.

- Fixed BooleanIndex' items method so the ZMI browse view works.

2.13.14 (2011-05-19)
--------------------

- Fixed addition of two LazyCat's if any of them was already flattened.

- Extend BooleanIndex by making the indexed value variable instead of
  hardcoding it to `True`. The indexed value will determine the smaller set
  automatically and choose its best value. An inline switch is done once the
  indexed value set grows larger than 60% of the total length. 60% was chosen
  to avoid constant switching for indexes that have an almost equal
  distribution of `True/False`.

- Substitute catalog entry in UUIDIndex error message.

2.13.13 (2011-05-04)
--------------------

- Optimize `Catalog.updateMetadata` avoiding a `self.uids` lookup and removing
  inline migration code for converting `self.data` from non-IOBTree types.

- In the path index, don't update data if the value hasn't changed.

2.13.12 (2011-05-02)
--------------------

- Optimize DateRangeIndex for better conflict resolution handling. It always
  starts out with storing an IITreeSet of the value instead of special casing
  storing an int for a single value. The `single value as int` optimization
  should be provided via a separate API to be called periodically outside the
  context of a normal request.

- Replaced `weightedIntersection` and `weightedUnion` calls with their
  non-weighted version, as we didn't pass in weights.

2.13.11 (2011-05-02)
--------------------

- Fix possible TypeError in `sortResults` method if only b_start but not b_size
  has been provided.

- Prevent the new UUIDIndex from acquiring attributes via Acquisition.

2.13.10 (2011-04-21)
--------------------

- Handle `TypeErrors` in the KeywordIndex if an indexed attribute is a method
  with required arguments.

- Added reporting of the intersection time of each index' result with the
  result set of the other indexes and consider this time to be part of each
  index time for prioritizing the index.

- Removed tracking of result length from the query plan. The calculation of the
  length of an intermediate index result can itself be expensive.

2.13.9 (2011-04-10)
-------------------

- Added a floor and ceiling value to the date range index. Values outside the
  specified range will be interpreted the same way as passing `None`, i.e.
  `since the beginning of time` and `until the end of it`. This allows the
  index to apply its optimizations, while objects with values outside this
  range can still be stored in a normal date index, which omits explicitly
  passed in `None` values.

2.13.8 (2011-04-01)
-------------------

- Fixed bug in date range index, which would omit objects exactly matching the
  query term if a resultset was provided.

- Fixed the BooleanIndex to not index objects without the cataloged attribute.

2.13.7 (2011-02-15)
-------------------

- Fixed the `DateIndex._unindex` to be of type `IIBTree` instead of `OIBTree`.
  It stores document ids as keys, which can only be ints.

2.13.6 (2011-02-10)
-------------------

- Remove docstrings from various methods, as they shouldn't be web-publishable.

2.13.5 (2011-02-05)
-------------------

- Fixed test failures introduced in 2.13.4.

2.13.4 (2011-02-05)
-------------------

- Added a new UUIDIndex, based on the common UnIndex. It behaves like a
  FieldIndex, but can only store one document id per value, so there's a 1:1
  mapping from value to document id. An error is logged if a different document
  id is indexed for an already taken value. The internal data structures are
  optimized for this and avoid storing one IITreeSet per value.

- Optimize sorting in presence of batching arguments. If a batch from the end
  of the result set is requested, we internally reverse the sorting order and
  at the end reverse the lazy sequence again. In a sequence with 100 entries,
  if we request the batch with items 80 to 90, we now reverse sort 20 items
  (100 to 80), slice of the first ten items and then reverse them. Before we
  would had to sort the first 90 items and then slice of the last 10.

- If batching arguments are provided, limit the returned lazy sequence to the
  items in the required batch instead of returning leading items falling
  outside of the requested batch.

- Fixed inline `IISet` to `IITreeSet` conversion code inside DateRangeIndex'
  `_insertForwardIndexEntry` method.

2.13.3 (2011-01-01)
-------------------

- Avoid locale-dependent test condition in `test_length_with_filter`.

2.13.2 (2010-12-31)
-------------------

- Preserve `actual_result_count` on flattening nested LazyCat's.

- Preserve the `actual_result_count` on all lazy return values. This allows
  to get proper batching information from catalog results which have been
  restricted by `sort_limit`.

- Made sure `actual_result_count` is available on all lazy classes and falls
  back to `__len__` if not explicitly provided.

- Optimized length calculation of Lazy classes.

2.13.1 (2010-12-25)
-------------------

- Added automatic sorting limit calculation based on batch arguments. If the
  query contains a `b_start` and `b_size` argument and no explicit `sort_limit`
  is provided, the sort limit will be calculated as `b_start + b_size`.

- Avoid pre-allocation of marker items in `LazyMap`.

2.13.0 (2010-12-25)
-------------------

- Fix `LazyMap` to avoid unnecessary function calls.

- Released as separate distribution.
