[tomasc]

The search now respects SCI.

From the included spec:

class MyDoc
  include Mongoid::Document
  include Mongoid::FullTextSearch

  field :title
  fulltext_search_in :title
end

class MyInheritedDoc < MyDoc
end

MyDoc.fulltext_search(…) # => will return both MyDoc as well as MyInheritedDoc documents
MyInheritedDoc.fulltext_search(…) # => will return only MyInheritedDoc documents

It is also possible to pre-empt the search with Monogid criteria:

MyDoc.where(value: 10).fulltext_search(…)

Please note that this will obviously not work in case an index is shared by multiple classes (that are not connected through inheritance), since a criteria applies only to one class.

[tomasc]

builds upon #35 , will rebased once that one is merged

[tomasc]

@dblock here we keep failing miserably on Mongoid 3 (but not 3.1). No idea why, again …

[dblock]

I can reproduce this locally and it looks like a clear bug introduced in this PR. We could drop support for 3.0, but I'd like to understand what's going on.

With Mongoid 3.

Both ExternalArtist and ExternalArtwork are using the same index, mongoid_fulltext.artworks_and_artists.

I am supposed to get back 2 objects, an external artist and artwork via ExternalArtwork.fulltext_search. The correct result is:

[#<ExternalArtist _id: 5af21a8f9b0b58e214000001, _type: nil, full_name: "Pablo Picasso">,
 #<ExternalArtwork _id: 5af21a8f9b0b58e214000002, _type: nil, title: "Portrait of Picasso">]

In Mongoid 3 the query is {"ngram"=>"picasso", "class"=>{"$in"=>["ExternalArtwork"]}} which is incorrect.

The code in document_type_filters does return {} unless fields['_type'].present?, in Mongoid 3 fields['_type'] returns #<Mongoid::Fields::Standard:...>", and in Mongoid > 3.1 that's nil. I am not sure what that means ;)

[dblock]

I tried fields['_type'].default_val.call instead, and that passed all tests, which tells me you're missing a test here for a descendant type, that is never not nil in the entire test suite.

[dblock]

This is simpler and works for all versions:

return {} if descendants.none?

[dblock]

@tomasc Got a moment to wrap this up?

[tomasc]

@dblock will try to get to that this week. In the end I ended up not using the gem, but yes, let's wrap this up.

[dblock]

AI posting on behalf of @dblock: adapted this proposal onto the modernized code in #57, with @tomasc credited in the changelog, PR description, and commit coauthor. The replacement adds regression coverage for result limits after filtering, dangling references with strict Mongoid settings, scores, descendants, and unscoped/shared-index compatibility, including Mongoid 3.0. Its complete fork matrix is green: https://github.com/dblock/mongoid_fulltext/actions/runs/38996842235. Closing this older PR in favor of #57; thank you for the original implementation.