* Adds PostgreSQL search backend.
* Isort nitpicks.
* Fixes PostgreSQL versions incompatibilities.
* Uses Django lru_cache instead of building our own.
* Fixes PostgreSQL search index on some empty vector & query cases.
* Never sets the PostgreSQL search vector to NULL.
* Simplification + removes caching on two fast enough functions.
* Rewrites stale entries deletion to use the ORM.
Closes #2778
Extra ES params are now passed through new OPTIONS key in the
WAGTAILSEARCH_BACKENDS setting.
It's backward compatible: if no OPTIONS key is found and some parameters
still exist, those parameters are used for the ES constructor..
This is the correct lexer for interactive console sessions, according to
<http://pygments.org/docs/lexers/>. This does require command lines to
be prefixed with `$`, otherwise they are interpreted as the output of a
command. It highlights the command nicely, including environment
variables, strings, and comments.
This allows the user to retrieve the scores for each search result:
for page in Page.objects.search("Hello").annotate_score('_score'):
print(page.title, page._score)
* Created Elasticsearch 2 backend
* Added tests for Elasticsearch 2 backend
* Split models up into different indices
pages, images and documents are now in separate indices
* Prefix fields of child models to prevent mapping clashes
* Replaced index_analyzer with analyzer/search_analyzer
index_analyzer has been removed in Elasticsearch 2.0
https://www.elastic.co/guide/en/elasticsearch/reference/current/breaking_20_mapping_changes.html#_analyzer_mappings
There's no indication in Elasticsearch's docs that this wouldn't work on Elasticsearch 1.x. However, we found that the new configuration isn't reliable on Elasticsearch 1.6 and below (causes the test_query_analyzer test to fail randomly).
* Implemented new way of representing content types in search index
Instead of using a long string of model names that is queried using a
"prefix" query, we instead use a multi-value string field and query it
using a simple "match" query.
The only reason why this isn't implemented in the Elasticsearch 1.x
backend yet is backwards compatibility
* Added another child model of SearchTest with clashing field mapping
This checks that the namespacing of fields on child models is working properly (if it doesn't the update_index tests will fail)
* Added tests for get_model_root function
* fixup! Added tests for get_model_root function
* Docs updates for Elasticsearch 2 support
Also tweak examples to use elasticsearch2 backend by default
* Test against Elasticsearch 2 on travis
Indexed.search_fields used to be a tuple. This is incorrect, and it
should have been a list. Changing it to be a list now would be a
backwards incompatible change, as people do
search_fields = Page.search_fields + (
SearchField('body')
)
Adding a tuple to the end of a list causes an error, so this would
cause all old code that used tuples to throw an error. This is not
great.
A new ThisShouldBeAList class, which subclasses list, has been added.
It additionally allows tuples to be added to it, as in the above
behaviour, but will raise a deprecation warning if someone does this.
Old code that uses tuples will continue to work, but raise a deprecation
warning.
See #2310
