Comment 20 for bug 1438136

Chris,

Your second query plan is still very strange, and I'm not 100% sure why. The second query should be using the tsvector index, like Jason's. Based on a little empirical testing, I think you may need to look into your indexing, as all title searches are taking at least 4 seconds, which is consistent with your output linked above. I don't think your indexing is quite right...

But, it's also obviously better than before.

I don't have time right this moment to test the inverse of this, where a very rare attribute is selected and a (relatively speaking) common search term is used. We don't want to lose the gains made in those cases, and that's where the complicated code comes into play. If anyone has time to apply the same structural change so such a core query, I would be very appreciative.

TIA.