: the custom Sort object seems a bit more direct. : : I'm not very familiar with the solr source. Can you give me some idea : of how to get started -- maybe this is now a better discussion for : solr-dev . . .
solr-user is fine ... writing plugins are a user level discussion topic (although maybe we should consider adding solr-plugin-dev list so people on solr-user who don't know (and don't want to know) anything about java don't have to read them) Since the custom sorting you want to do is all based on a single field, the smallest granularity of a solr plugin you could write would be a new FieldType with a getSortField method that returns a new custom instance of SortField encapsulating the logic you want. You can subclass the existing DateField class and just override that one method, to use your new code, you would just specify YourDateField in your schema.xml using it's full name (with java package). that's all the special Solr work you need to do ... everything else is just extending some Classes the underlying Lucene Java API provides -- you can even test that most of your stuff is working using Lucene directly without ever worrying about Solr... Your new SortField class can be very simple -- all you need to define is a new SortComparatorSource. this is where the interesting stuff happens, SortComparatorSource instances are expected to decide, given an IndexReader and a fieldName how compare two documents. Most of them use the Lucene FieldCache, which requires a single value per field -- you can make your SortComparatorSource do anything you want, as long as it produces something called a "ScoreDocComparator" which knows how to compare two documents and decide which one comes first. All of these classes i mentioned (besides DateField) are Lucene classes that are documented in the Lucene javadocs... http://lucene.apache.org/java/2_3_1/api/index.html ...the best examples of how to use/extend them are probably the Unit tests. (just search for the class anmes) Now: all that said, i'm honestly not sure what the best way to achieve your goal would be. One appraoch would be to build an array of values like the FieldCache where you take into cosideration the current time while building it to decide which single value to give each doc. the downside is that you can't cache that data stucture for very long, because time marches forward. Another approach would be a more complicated data structure that still provides fast lookups per docid (array of arrays where the items in the subarrays are sorted so you can do a binary search) -Hoss
