Apache Doris dialect: add parsing support for inverted-index MATCH operators

[nico-gsantos]

Apache Doris supports full-text search through inverted indexes, exposed via the SQL functions and operators MATCH, MATCH_ALL, and MATCH_ANY. These constructs are essential for enabling inverted-index acceleration and are first-class features for text search workloads in Doris.

However, sqlglot currently fails to parse queries that use MATCH, MATCH_ALL, or MATCH_ANY, raising parsing errors such as “Unexpected token: MATCH”.

This limitation becomes a practical issue when sqlglot is used as a SQL parser or validator (for example, in Apache Superset). As a consequence, valid and highly performant Apache Doris SQL queries cannot be used, even though they execute correctly in Doris itself.

Using LIKE is not a viable workaround, because Apache Doris does not apply inverted-index acceleration or term normalization to LIKE predicates. As documented by Apache Doris, full-text search using MATCH_ANY with inverted indexes can yield significant performance improvements (e.g., up to a 9× speedup in documented examples) and may return different result sets due to index-level normalization (such as automatic lowercasing). Therefore, LIKE is neither functionally nor performance-equivalent to MATCH_* queries

[georgesittas]

Hey,

Doris is a community-contributed dialect, so improving it is out of scope for the core SQLGlot team. Well-tested & documented PRs are welcome.