Array of strings not converted to arrays of integers - ruby-on-rails

Question

I'm trying to understand the following behaviour:
irb(main):016:0* pg = ProductGroup.find(1)
irb(main):017:0> pg.good_type_ids
=> [1]
irb(main):018:0> pg.update({"good_type_ids"=>["", "1"]})
irb(main):019:0> pg.changed?
=> false
irb(main):020:0> pg.good_type_ids
=> ["1"]
The code above mimics a form update. The good_type_ids are selected ids's of a multi select. Although the value of the array changes from [1] to ["1"], the changed? method returns false. The update method seems also be smart enough to strip the empty string.
irb(main):021:0> pg = ProductGroup.find(1)
irb(main):022:0> pg.good_type_ids
=> [1]
irb(main):023:0> pg.attributes = {"good_type_ids"=>["", "1"]}
=> {"good_type_ids"=>["", "1"]}
irb(main):024:0> pg.changed?
=> true
irb(main):025:0> pg.good_type_ids
=> ["", "1"]
In this second example I'm trying to apply the changed params from the form to the object. I don't want to save it to the dbs!
Somehow the attributes method behaves different than the update method and good_type_ids will be ["","1"] after the method was invoked.
It seems that the type of the column in the dbs is taken into account by the "update" and "attributes" method, only for arrays (postgress) this doesn't seem to work.
irb(main):030:0* pg.attributes = {"recourse_days"=>5}
=> {"recourse_days"=>5}
irb(main):031:0> pg.changed?
=> true
irb(main):039:0* pg.recourse_days
=> 5
recourse_days are defined as integer, during post the controller receives a string and converts it correctly to an integer.
Schema.rb: t.integer "good_type_ids", default: [], array: true
As temporary work arround I have put this in the update method of the controller
params["good_type_ids"].reject!(&:blank?)
params["good_type_ids"].map!(&:to_i)

Answer 1

UPD
Strings conversion issues and nil elements saving to pgsql seems to be fixed in rails 4.2.0, i've heard whole type casting system is pretty much revised there:
> doc.test_digit = ["1"] #=> ["1"]
> doc.test_digit #=> [1]
> doc.test_string #=> ["2"]
> doc.test_string = [2] #=> [2]
> doc.test_string #=> ["2"]
> doc.attributes = {'test_digit'=>["", "2"]}
=> {"test_digit"=>["", "2"]}
> doc.test_digit #=> [nil, 2]
> doc.save
(0.9ms) BEGIN
SQL (1.8ms) UPDATE "documents" SET "test_digit" = $1, "updated_at" = $2 WHERE "documents"."id" = $3 [["test_digit", "{NULL,2}"], ["updated_at", "2015-02-26 00:32:25.326735"], ["id", 1]]
(3.7ms) COMMIT
=> true
> doc.reload
Document Load (1.7ms) SELECT "documents".* FROM "documents" WHERE "documents"."id" = $1 LIMIT 1 [["id", 1]]
=> #<Document id: 1, name: "test1", test_digit: [nil, 2], test_string: ["2"], created_at: "2015-02-25 23:51:32", updated_at: "2015-02-26 00:32:25">
> doc.test_digit #=> [nil, 2]
I used following migrations:
t.integer :test_digit, array: true, default: []
t.string :test_string, array: true, default: []
Update method updates the db, thus your model attributes are no longer dirty.
Good guide on this
The update method seems also be smart enough to strip the empty string.
This is not normal behavior, according to the source of pgsql adapter/array_parser, empty array elements should be saved and loaded as any other values. (this functionality was added on Aug 20, 2012 to the adapter just in case you are using an older version)