Lars Aronsson wrote:
Would it be reasonable to update the search index each time a new version of a page is saved? In that case, the search would still be indexed (and fast), but it would always be up-to-date.
This is true now, since the pages are in a true database with Magnus's new software. In the old version, all the data was just stored in text files on disk. I wrote a program to go through and analyze the keywords from all the pages and titles, and construct a search index from that.
I always wanted to put it on a cron job to update nightly, but it was so inefficient that I didn't feel comfortable letting it run without supervision, and I didn't feel comfortable running it all that often.
Now that everything is in a real database, it should be true that with a little playing around and tweaking, we can get decent results that are fast and also always instantly updated.
The current version is a very simple SQL query. It doesn't work so well in terms of being intelligent about returning what you probably want.