Ranking results with complex conditions using Rails and Squeel

Question

I'm probably doing something silly here - and I'm open to other ways of doing - but I'm trying to order my results set based on a computed field:

Client.select{['clients.*',
               (cast((surname == matching_surname).as int) * 10 +
                cast((given_names == matching_given_names).as int) + 
                cast((date_of_birth == matching_date_of_birth).as int).as(ranking)]}.
       where{(surname =~ matching_surname) | 
             (given_names =~ matching_given_names) | 
             (date_of_birth == matching_date_of_birth)}.
       order{`ranking`.desc}

My problem is that date_of_birth could be nil. This causes the cast((...).as int) call to return three different values - 1 if the expression evaluated to true; 0 if the expression evaluated to false; and nil if the underlying column value was nil.

The nil values from the expressions cause the whole ranking to evaluate to NIL - which means that even if I have a record that matches exactly on surname and given_names, if the date_of_birth column is nil, the ranking for the record is nil.

I have tried to use a complex expression in the cast that checks if not nil or the matching_value, but it fails with a Squeel exception using | and ruby evaluates it when using || and or.

I've also tried to use a predicates in the order for aliased columns:

order{[`ranking` != nil, `ranking`.desc]}

but that throws an ActiveRecordexception complaining that the column ranking does not exist.

I'm at the end of my rope... any ideas?

Answer 1

After a bit of a dance, I was able to calculate the ranking using a series of outer joins to other scopes as follows:

def self.weighted_by_any (client)
  scope = 
    select{[`clients.*`,
            [
             ((cast((`not rank_A.id is null`).as int) * 100) if client[:social_insurance_number].present?),
             ((cast((`not rank_B.id is null`).as int) * 10) if client[:surname].present?),
             ((cast((`not rank_C.id is null`).as int) * 1) if client[:given_names].present?), 
             ((cast((`not rank_D.id is null`).as int) * 1) if client[:date_of_birth].present?)
            ].compact.reduce(:+).as(`ranking`)
          ]}.by_any(client)

  scope = scope.joins{"left join (" + Client.weigh_social_insurance_number(client).to_sql + ") AS rank_A ON rank_A.id = clients.id"} if client[:social_insurance_number].present?
  scope = scope.joins{"left join (" + Client.weigh_surname(client).to_sql + ") AS rank_B on rank_B.id = clients.id"} if client[:surname].present?
  scope = scope.joins{"left join (" + Client.weigh_given_names(client).to_sql + ") AS rank_C on rank_C.id = clients.id"} if client[:given_names].present?
  scope = scope.joins{"left join (" + Client.weigh_date_of_birth(client).to_sql + ") AS rank_D on rank_D.id = clients.id"} if client[:date_of_birth].present?
  scope.order{`ranking`.desc}
end

where Client.weigh_<attribute>(client) is another scope that looks like the following:

def self.weigh_social_insurance_number (client)
  select{[:id]}.where{social_insurance_number == client[:social_insurance_number]}
end

This allowed me to break out the comparison of the value from the check for nil and so removed the third value in my boolean calculation (TRUE => 1, FALSE => 0).

Clean? Efficient? Elegant? Maybe not... but working. :)

EDIT base on new information

I've refactored this into something much more beautiful, thanks to Bigxiang's answer. Here's what I've come up with:

First, i replaced the weigh_<attribute>(client) scopes with sifters. I'd previously discovered that you can use sifters in the select{} portion of the scope - which we will be using in a minute.

sifter :weigh_social_insurance_number do |token|
  # check if the token is present - we don't want to match on nil, but we want the column in the results
  # cast the comparison of the token to the column to an integer -> nil = nil, true = 1, false = 0
  # use coalesce to replace the nil value with `0` (for no match)
  (token.present? ? coalesce(cast((social_insurance_number == token).as int), `0`) : `0`).as(weight_social_insurance_number)
end

sifter :weigh_surname do |token|
  (token.present? ? coalesce(cast((surname == token).as int), `0`) :`0`).as(weight_surname)
end

sifter :weigh_given_names do |token|
  (token.present? ? coalesce(cast((given_names == token).as int), `0`) : `0`).as(weight_given_names)
end

sifter :weigh_date_of_birth do |token|
  (token.present? ? coalesce(cast((date_of_birth == token).as int), `0`) : `0`).as(weight_date_of_birth)
end

So, let's create a scope using the sifters to weigh all our criteria:

def self.weigh_criteria (client)
  select{[`*`, 
          sift(weigh_social_insurance_number, client[:social_insurance_number]),
          sift(weigh_surname, client[:surname]),
          sift(weigh_given_names, client[:given_names]),
          sift(weigh_date_of_birth, client[:date_of_birth])
        ]}
end

Now that we can determine if the criteria provided match the column value, we compute our ranking using another sifter:

sifter :ranking do
  (weight_social_insurance_number * 100 + weight_surname * 10 + weight_date_of_birth * 5 + weight_given_names).as(ranking)
end

And adding it all together to make our scope that includes all the model attributes and our computed attributes:

def self.weighted_by_any (client)
  # check if the date is valid 
  begin 
    client[:date_of_birth] = Date.parse(client[:date_of_birth])
  rescue => e
    client.delete(:date_of_birth)
  end

  select{[`*`, sift(ranking)]}.from("(#{weigh_criteria(client).by_any(client).to_sql}) clients").order{`ranking`.desc}
end

So, I can now search for a client and have the results ranked by how closely they match the provided criteria with:

irb(main): Client.weighted_by_any(client)
  Client Load (8.9ms)  SELECT *, 
                              "clients"."weight_social_insurance_number" * 100 + 
                              "clients"."weight_surname" * 10 + 
                              "clients"."weight_date_of_birth" * 5 + 
                              "clients"."weight_given_names" AS ranking 
                       FROM (
                             SELECT *, 
                                    coalesce(cast("clients"."social_insurance_number" = '<sin>' AS int), 0) AS weight_social_insurance_number, 
                                    coalesce(cast("clients"."surname" = '<surname>' AS int), 0) AS weight_surname, 
                                    coalesce(cast("clients"."given_names" = '<given_names>' AS int), 0) AS weight_given_names,         0 AS weight_date_of_birth 
                             FROM "clients" 
                             WHERE ((("clients"."social_insurance_number" = '<sin>' 
                                   OR "clients"."surname" ILIKE '<surname>%') 
                                   OR "clients"."given_names" ILIKE '<given_names>%'))
                            ) clients 
                       ORDER BY ranking DESC

Cleaner, more elegant, and working better!