DB-independent fulltext indexing

Hi everyone, My problem in a nutshell: How do I use, or at least roughly emulate, fulltext indexing in a db-portable way? My project requires near perfect portability, so I'm already using PEAR::DB and standard ansi sql extensively, but I need to provide the ability to search large text fields. I see two possible routes, but each has its drawbacks. I could: 1) Ignore the variety of proprietary implementations and roll my own solution, something with a keyword table and a keyword_to_content mapping table that would track occurrences of keywords in text records. Benefits: -entirely db-independent -easy to add soundex, synonym lookups, controlled vocabularies or anything else -quick reads/selects Problems: -inserts/updates: every unique word in a new text record must be located in the keyword table or else inserted there (one or two queries); then that keyword_id and associated content_id must be inserted in the mapping table (another query), so for a 500-word string of text, that could mean 1001 (best case) or 1501 (worst case) queries! 2) Or, since I'm already using PEAR DB, I could extend it, or each of the drivers, to somehow modify WHERE clauses to have a particular rdbms's fulltext syntax (e.g. "WHERE MATCH(cols) AGAINST('searchString')" for mysql, "WHERE CONTAINS(cols, 'searchString')" for mssql, "WHERE CONTAINS(cols, 'searchString') > 0" for oracle, etc.) This seems really kludgy, though, and it also means that, for oracle and mssql at least, there's an additional wrinkle of having to schedule regular index refreshes/updates. Any ideas? I realize this questions isn't totally PEAR-specific, but I ask here in case anyone has tried something like this or extended DB like this. Thanks, Andy