Unreleased

Empty for now.

1.3.2 (2025-11-27)

Changes

  • Multi-value filter support in shell (#922): All shell search commands (SEARCH, EXPLAIN, BUCKET, BENCH, REVERSE) now support multi-value filters using parameter repetition (TYPE street TYPE city) or pipe separator (TYPE street|city).
  • Housenumber filtering with multi-value filters (#921): Fixed housenumber type filtering logic to properly handle multi-value filters (e.g., type=housenumber&type=street) in both search() and reverse() functions.

1.3.1 (2025-11-16)

Changes

  • Multi-value filters improvements (#914, #915): The multi-value filter separator is now configurable via FILTERS_MULTI_VALUE_SEPARATOR parameter (default: ' '). Set to None to disable separator parsing (multiple parameters still work). Additionally, the Python API (search(), reverse()) now accepts lists directly for multi-value filters (e.g., type=["street", "city"]), while the HTTP layer handles string parsing. Single string values remain supported for backward compatibility. This improves code clarity and type safety while maintaining full HTTP API compatibility.
  • Temporary setuptools pin (#913): Added setuptools < 81 constraint to avoid deprecation warnings from pkg_resources. This temporary fix will be removed in the next major version when migrating away from pkg_resources.

1.3.0 (2025-11-02)

Breaking changes

  • Update redis to 6.4.0 and hiredis to 3.3.0 for Redis 8 compatibility
  • Minimum Redis version requirement increased from 5.0 to 7.2 (required by redis-py 6.x)
  • hiredis is now an optional dependency for performance optimization. Install with pip install addok[perf] to enable it. Addok will work without it but with slightly reduced Redis performance.

New features

  • Multi-value filters with OR logic: Filters now support multiple values using the + separator (e.g., ?type=street+city). This allows searching for documents matching any of the specified values. Multi-value filters work on both /search/ and /reverse/ endpoints.
  • Add MAX_FILTER_VALUES configuration option to control the maximum number of values allowed in a multi-value filter (default: 10)

Changes

  • Add Python 3.14 support
  • Modernize Python packaging to use pyproject.toml (PEP 621)
  • Development dependencies moved to optional [dev] group: use pip install -e .[dev] for development setup
  • Enable multiprocessing on macOS: Use spawn context instead of fork to ensure proper Redis connection handling and avoid fork-safety issues on macOS. This makes macOS behavior consistent with production Linux environments.
  • Improved filter strategy for common tokens: When searching with only common tokens (high frequency) and filters, the system now intelligently compares filter size with token frequency to choose the most efficient strategy (Redis intersection vs manual scan). This can significantly improve performance when using filters that are more selective than the search tokens (e.g., searching "la" with type=locality or region-based filters).

1.2.1 (2025-08-26)

  • Make results more deterministic (#899)

1.2.0 (2025-06-01)

  • Add Python 3.12 and 3.13 support
  • Update falcon to 4.0.2
  • Update multiple small dependencies
  • Some minor code improvements
  • Drop Cython support
  • Drop Python 3.7 and 3.8 support (end of life since 27 Jun 2023 and 7 Oct 2024)

1.1.2 (2024-01-26)

  • Improve values handling in index and deindex methods (#798)

1.1.1

  • Make addok ngrams work again on macOS
  • Disable multiprocessing on macOS (affect only ngrams generation)
  • Update redis to 4.5.4 and hiredis to 2.2.2

1.1.0

  • Drop Python 3.6 support (end of life since 23 Dec 2021)
  • Added /health endpoint to monitor Addok (#750)

1.1.0-rc2

  • Added load_csv_file batch loader
  • Fixed type=housenumber also returning other results in some cases (#478)
  • Fixed ordering of housenumbers with non alpha-num chars (#656)
  • Added ID_FIELD to control which field is used as document _id
  • config.SYNONYMS_PATH is now config.SYNONYMS_PATHS and is a list to allow multiple files
  • Fixed non unique id across multiple docker sharing same Redis instance (#607)
  • Added more variants for lat and lon params and better control their values (#592)
  • Better ordering of candidates in case of autocomplete (#494)
  • By default, use more common chars when building fuzzy variants
  • Added python >= 3.8 compat
  • Restore legacy scoring algorithm (#746): the new experimental scoring must be activated manually, replacing addok.helpers.results.score_by_ngram_distance with addok.helpers.results.score_by_str_distance in SEARCH_RESULT_PROCESSORS_PYPATHS

1.1.0-rc1

  • Faster new scoring algorithm (#431)
  • Upgraded Falcon to 1.4.1
  • autocomplete and fuzzy are not adding any more their collectors automagically, instead they are now hard coded in the default config; if you haven't changed RESULTS_COLLECTORS_PYPATHS in your local config this should not impact you, otherwise, see "Upgrading" below.
  • Added a slow queries logger (see config for usage)

Upgrading to 1.1.0-rc1

If you have changed RESULTS_COLLECTORS_PYPATHS in your local config file, make sure to add manually fuzzy and autocomplete ones. Check the config doc for an example.

1.0.3

  • make it work with python >= 3.8
  • upgrade python deps

1.0.2

  • allow to connect to Redis through unix socket
  • fix reverse not honouring extra housenumber fields
  • fix default BATCH_WORKERS values failing on systems with only one CPU

1.0.1

  • Upgraded Falcon to 1.2.0
  • Fix bug when search request is empty but lat and lon are given and valid
  • Handle filters when doing a manual scan with Lua
  • Allow to configure Redis password from config

1.0.0

The 1.0.0 has been a big rewrite, with main features:

  • split in plugins
  • allow for external storage of documents (in SQLite, PostgreSQL, etc.)
  • use LUA scripting for performances
  • less RAM consumption
  • replaced Flask by Falcon for performances
  • housenumbers are not indexed anymore (to gain RAM), they are only matched in result postprocessing

It contains many breaking changes. Best option when possible is to restart from scratch (see the tutorial) and reindex everything.

Breaking changes

  • PROCESSORS, INDEXERS, etc. have been renamed to PROCESSORS_PYPATHS, INDEXERS_PYPATHS, etc.
  • HOUSENUMBERS_PROCESSORS have been removed
  • config must now be loaded by from addok.config import config
  • removed DEINDEXERS, now INDEXERS must point to python classes having both index and deindex methods
  • endpoints API changed
  • by default, documents are now stored in a separate Redis database
  • the key "id" is not required anymore in the loaded data and as such has been removed from the geojson Feature root.

Minor changes

  • index multi values in filters
  • add a "reset" command to reset all data (indexes and documents)
  • added quote parameter for CSV endpoints (now in addok-csv plugin)
  • addok now tries to read config from /etc/addok/addok.conf as fallback
  • SMALL_BUCKET_LIMIT is now a setting

Also check the new FAQ section in the documentation.

0.5.0

  • Expose housenumber parent name in result geojson
  • add support for housenumber payload (#134)
  • Fix clean_query being too much greedy for "cs" (#125
  • also accept long for longitude
  • replace "s/s" in French preprocessing
  • fix autocomplete querystring casting to boolean
  • Always add housenumber in label candidates if set (#120)
  • make CSVView more hackable by plugins ([#116][https://github.com/addok/addok/issues/116))

0.4.0

  • fix filters not taken into account in manual scan (#105)
  • added experimental list support for document values
  • Added MIN_EDGE_NGRAMS and MAX_EDGE_NGRAMS settings (#102)
  • documented MAKE_LABELS setting
  • Allow to pass functions as PROCESSORS, instead of path
  • remove raw housenumbers returned in result properties
  • do not consider filter if column is empty, in csv (#109)
  • allow to pass lat and lon to define columns to be used for geo preference, in csv (#110)
  • replace "s/" by "sur" in French preprocessing (#107)
  • fix server failing when document was missing importance value
  • refuse to load if ADDOK_CONFIG_MODULE is given but not found
  • allow to set ADDOK_CONFIG_MODULE with command line parameter --config
  • mention request parameters in geojson (#113)

0.3.1

  • fix single character wrongly glued to housenumber (#99)

0.3.0

  • use housenumber id as result id, when given (#38)
  • shell: warn when requested id does not exist (#75)
  • print filters in debug mode
  • added filters to CSV endpoint (#67)
  • also accept lng as parameter (#88)
  • add /get/ endpoint (#87)
  • display distance in meters (not kilometers)
  • add distance in single /reverse/ call
  • workaround python badly sniffing csv file with only one column (#90)
  • add housenumber in csv results (#91)
  • CSV: renamed "result_address" to "result_label" (#92)
  • no BOM by default in UTF-8