Function fullTextMatch

Build a dialect-correct full-text-search WHERE-predicate SQL fragment.

Returns { $literal: sql } (see module header for the supported/verified usage patterns for this shape).

Per-dialect behavior:

• mysql/mariadb: MATCH(col1, col2) AGAINST('term' IN NATURAL LANGUAGE MODE | IN BOOLEAN MODE). mode: 'phrase' uses boolean mode with the term wrapped in "..." for an exact-phrase match (MySQL boolean-mode phrase syntax). NOTE: requires a FULLTEXT index on the given column(s) — see createFulltextIndex() in src/dialects/mysql/index.ts / mariadb/index.ts.
• postgres/cockroachdb: to_tsvector('english', col1 || ' ' || col2) @@ plainto_tsquery('english', 'term'). mode: 'phrase' uses websearch_to_tsquery instead of plainto_tsquery for more natural, phrase-aware parsing (supports quoted phrases, -, OR). NOTE: works on any text column; a GIN index (createFulltextIndex()) is only needed for performance, not correctness.
• mssql: mode: 'natural' (default) uses FREETEXT(col, 'term'); mode: 'boolean' / mode: 'phrase' uses CONTAINS(col, 'term') (phrase mode wraps the term in "...", which is exact-phrase syntax for CONTAINS). NOTE: requires a full-text index/catalog — see createFulltextIndex() in src/dialects/mssql/index.ts.
• oracle: CONTAINS(col, 'term', 1) > 0 (Oracle Text). Multiple columns are OR-ed together, since Oracle Text indexes and scores one column at a time. NOTE: requires an Oracle Text index — see createFulltextIndex() in src/dialects/oracle/index.ts.
• sqlite: col MATCH 'term'. STRUCTURAL REQUIREMENT (unlike every other dialect above): this only works against an FTS5 virtual table/column (see createFulltextIndex() / createFTS5Table() in src/dialects/sqlite/index.ts) — SQLite has no secondary full-text index you can bolt onto an ordinary table the way MySQL/Postgres/MSSQL do; the table itself must be an fts5 virtual table.
• clickhouse: uses hasToken(col, 'term') for a single-token exact match, or multiSearchAny(col, ['term1', 'term2', ...]) when searchTerm contains multiple whitespace-separated words. Multiple columns are OR-ed together. See createFulltextIndex() (ClickHouse ngrambf_v1/tokenbf_v1 index) in src/dialects/clickhouse/index.ts.
• db2: THROWS. Db2's optional Text Search feature (Net Search Extender) has its own CONTAINS-like predicate syntax, but it could not be verified with confidence here, and createFulltextIndex() in src/dialects/db2/index.ts itself throws "Fulltext indexes are not supported in Db2 LUW (requires Net Search Extender)" — so there is no verified index-creation path to pair a query against either. Throwing here rather than guessing avoids shipping silently-wrong SQL.
• snowflake: FALLBACK, not true full-text search. Snowflake has no Postgres-style tsvector/tsquery full-text search primitive (its closest offerings are Cortex Search / Search Optimization Service, which are not plain-SQL predicates); createFulltextIndex() in src/dialects/snowflake/index.ts is a documented no-op for the same reason. Falls back to ILIKE '%term%' (case-insensitive substring match — no tokenization, stemming, or relevance ranking).
• redshift: FALLBACK, not true full-text search. Redshift (unlike Postgres, which it forked from) does not implement tsvector/tsquery; createFulltextIndex() in src/dialects/redshift/index.ts throws accordingly. Falls back to ILIKE '%term%', same caveat as Snowflake.