Version 2.2.1-beta, 13 nov 2013

added ALTER TABLE that can add attributes to disk and RT indexes on the fly
added ATTACH support for non-empty RT target indexes
added Chinese segmentation with RLP (Rosette Linguistics platform) support
added English, German lemmatization support
added HAVING support to SELECT statement, filtering on aggregate values is now possible
added N-best GROUP BY extension to return more than 1 row per group
added RT index support for index_field_lengths=1, bitfield attributes, and multiforms
added CSV, TSV data sources
added full JSON attributes support, arbitrary JSON documents (with subobjects etc) can now be stored
added in-place JSON updates for scalar values
added index type=template directive (allows CALL KEYWORDS, CALL SNIPPETS)
added ondisk_attrs, ondisk_attrs_default directives that keep attributes on disk
added table functions mechanism, and REMOVE_REPEATS() table function
added support for arbitrary expressions in WHERE for DELETE queries

added OPTION local_df=1, an option to aggregate IDFs over local indexes (shards)
added UDF XXX_reinit() method to reload UDFs with workers=prefork
added comma-separated syntax to OPTION idf, tfidf_unnormalized and tfidf_normalized flags
added lccs, wlccs, exact_order, min_gaps, and atc ranking factors
added sphinx_get_XXX_factors(), a faster interface to access PACKEDFACTORS() in UDFs
added support for exact_hit, exact_order field factors when using more than 32 fields (exact_hit, exact_order)

added DESCRIBE and --dumpheader support for tokencount attributes (generated by index_field_lengths=1 directive)
added RT index query profile, percentages, totals to SHOW PROFILE
added predicted_time, dist_predicted_time, fetched_docs, fetched_hits counters to SHOW META
added total_tokens and disk_bytes counters to SHOW INDEX STATUS

added ALL(), ANY() and INDEXOF() functions for JSON subarrays
added MIN_TOP_WEIGHT(), MIN_TOP_SORTVAL() functions
added TOP() aggregate function to expression ranker
added a check for duplicated tail hit positions in indextool --check
added compact_in option to query_log_format=sphinxql
added distance units and calculation method options to GEODIST() function, optimized it a lot
added embedded stopwords/exceptions/wordforms to --dumpheader
added indexer --nohup and indextool --rotate switches to check index files before rotating them
added scientific notation support for JSON attributes (as per RFC 4627)
added several SphinxQL statements to fix MySQL Workbench connection issues (LIKE for session variables, etc.)
added shutdown_timeout directive to searchd config section
added signed values support for INTEGER() and UINT() function
added snippet generation options to SNIPPET() function
added string filter support in distributed queries, SphinxAPI, SphinxQL query log
added support for mixed distributed and local index queries (SELECT * FROM dist1,dist2,local3), and index_weights option for that case

optimized JSON attributes access (1.12x to 2.0x+ total query speedup depending on the JSON data)
optimized SELECT (1.02x to 3.5x speedup, depending on index schema size)
optimized UPDATE (up to 3x faster on big updates)
optimized away internal threads table mutex contention with workers=threads and 1000s of threads
changed [emptyword -foo] query behavior in cases when emptyword is a stopword or an overshort word, made such queries computable rather than erroneous
changed post-morphology wordforms behavior, now it works as 'if ( stem(token)==stem(abc) ) emit(def)'
changed the config defaults to id64, dict=keywords, charset_type=utf-8, enable_star=1, workers=threads, mem_limit=128M, rt_mem_limit=128M
changed the default SphinxAPI matching mode to SPH_MATCH_EXTENDED2
disallowed dashes in index names in API requests (just like in SphinxQL)
removed legacy xmlpipe data source v1, compat_sphinxql_magics directive, SetWeights() SphinxAPI call, and SPH_SORT_CUSTOM SphinxAPI mode

fixed #1734, unquoted literal in json subscript could cause a crash, returns 'unknown column' now.
fixed #1683, under certain conditions stopwords were not taken into account in RT indexes
fixed #1648, #1644, when using AOT lemmas with snippet generation, not all the forms got highlighted
fixed #1549, OPTIONidf=tfidf_normalized was ignored for distributed queries
fixed that ORDER BY RAND() was not affected by index_weights
fixed that float updates with integer values in SphinxQL mistakenly set the float to 0
fixed that predicted_time was not accumulated with dist_threads
fixed GROUP_CONCAT result length limit (was implicitly limited by 1024 bytes)
fixed agent query distribution in HA mirroring
fixed duplicates check for quorum operator, it works ok now for expanded keywords
fixed off-by-1 query positions of words in indexes with wordforms and blended characters
fixed wrong lcs and min_best_span_pos ranking factor values when any expansion (expand_keywords or lemmatize) occurred
fixed a crash while creating indexes with sql_joined_field