I have a model, Item, that comes in three types, implemented using Single Table Inheritance. Item has a tree hierarchy that is represented using a has_many :through relationship for both the parents (called groups) and the children (called sub_items). One of the subclasses (we'll call it ItemA) must have only one parent at all times (but the other subclasses can have 0 or more parents). I couldn't figure out how to implement validations that would enforce the rules of the tree hierarchy, so I left them out.
ItemA has a helper method, parent, to get its parent. Sometimes this method will raise undefined method 'first' for nil:NilClass. It doesn't always happen, but it does in certain situations.
Situation 1
I am using jqGrid to list all the ItemAs. I use sorting functionality for some of the columns, including the parent. The grid would initially load the ItemAs successfully, showing that they all had parents as they should. However, when I tried to sort the parent column, I would get the error as if they suddenly disappeared. It went away when I removed includes(:groups) from the sorting method. I don't understand why that would help, but I figured the problem was solved. Until...
Situation 2
I am testing my app with Rspec, Factory Girl, and Selenium. In one of my tests, I call the parent method on an instance of ItemA created using Factory Girl. It raises the error. My tests that use the parent method indirectly don't fail. For example, the parent method is called on the index page for ItemA, and many tests visit that page without problem. This situation has not been resolved. My app no longer has any calls to includes, so that isn't part of it this time.
Relevant Code
ItemA
class ItemA < Item
# This method is called in Situation 1
def self.sort_by_parent sort_order
all.sort_by(&:parent_name).tap do |i|
i.reverse! if sort_order == :desc
end
end
def parent
groups.take
end
def parent_name
parent.component_name
end
end
Item
class Item < ActiveRecord::Base
has_many :item_groups, foreign_key: 'sub_item_id', dependent: :destroy
has_many :groups, through: :item_groups
has_many :group_items, class_name: 'ItemGroup', foreign_key: 'group_id', dependent: :destroy
has_many :sub_items, through: :group_items
end
update_spec
feature 'ItemA editing', js: true do
given!(:item_a) { create(:item_a) }
given!(:parent) { create(:item_b) }
scenario 'when parent', focus: true do
itemas_page = ItemAsPage.visit # This is a custom page object
# Situation 2 occurs here
itemas_page.edit_parent item_a.parent_name, parent.component_name
expect(itemas_page).to have_item_a_with parent.component_name
end
end
Why is the parent method sometimes reading as nil, and how do I get it to always produce a value?
EDIT: As I'm changing my code, I'm coming across more situations that cause this error. I checked the source. Here's a snapshot of ActiveRecord::FinderMethods:
module ActiveRecord::FinderMethods
def take(limit = nil)
limit ? limit(limit).to_a : find_take
end
private
def find_take
if loaded?
#records.first
else
#take ||= limit(1).to_a.first
end
end
end
For debugging purposes, I modified the parent method to look like this:
def parent
groups.tap {|g| puts '#records: ' + g.instance_variable_get(:#records).inspect }.take
end
#records is nil. I tried changing it to groups.reload.take to get #records to load, but it's not working. I'm using groups.limit(1).to_a.first right now, which is working, but I'm curious to find out what sort of bug in my app is causing this problem.
It turns out that despite calling reload, Rails still uses the cached version of the query. The same problem is described here. This fixes it:
class ItemA < Item
def parent
ActiveRecord::Base.uncached { groups.take }
end
end
Related
I am trying to create a rspec test for custom validation in a spree extension(like a gem)
I need to validate uniqueness of a variants
option values for a product (all Spree models)
Here is the basic structure of models(Although they are part of spree, a rails based e-commerce building):
class Product
has_many :variants
has_many :option_values, through: :variants #defined in the spree extension, not in actual spree core
has_many :product_option_types
has_many :option_types, through: :product_option_types
end
class Variant
belongs_to :product, touch: true
has_many :option_values_variants
has_many :option_values, through: option_values
end
class OptionType
has_many :option_values
has_many :product_option_types
has_many :products, through: :product_option_types
end
class OptionValue
belongs_to :option_type
has_many :option_value_variants
has_many :variants, through: :option_value_variants
end
So I have created a custom validation to check the uniqueness of a variants option values for a certain product. That is a product(lets say product1) can have many variants. And a variant with option values lets say (Red(Option_type: Color) and Circle(Option_type: Shape)) have to unique for this product
Anyway this is the custom validator
validate :uniqueness_of_option_values
def uniqueness_of_option_values
#The problem is in product.variants, When I use it the product.variants collection is returning be empty. And I don't get why.
product.variants.each do |v|
#This part inside the each block doesn't matter though for here.
variant_option_values = v.option_values.ids
this_option_values = option_values.collect(&:id)
matches_with_another_variant = (variant_option_values.length == this_option_values.length) && (variant_option_values - this_option_values).empty?
if !option_values.empty? && !(persisted? && v.id == id) && matches_with_another_variant
errors.add(:base, :already_created)
end
end
end
And finally here are the specs
require 'spec_helper'
describe Spree::Variant do
let(:product) { FactoryBot.create(:product) }
let(:variant1) { FactoryBot.create(:variant, product: product) }
describe "#option_values" do
context "on create" do
before do
#variant2 = FactoryBot.create(:variant, product: product, option_values: variant1.option_values)
end
it "should validate that option values are unique for every variant" do
#This is the main test. This should return false according to my uniqueness validation. But its not since in the custom uniqueness validation method product.variants returns empty and hence its not going inside the each block.
puts #variant2.valid?
expect(true).to be true #just so that the test will pass. Not actually what I want to put here
end
end
end
end
Anybody know whats wrong here. Thanks in advance
I have a guess at what's happening. I think a fix would be to change your validation with the following line:
product.variants.reload.each do |v|
What I think is happing is that when you call variant1 in your test, it is running the validation for variant1, which calls variants on the product object. This queries the database for related variants, and gets an empty result. However, since variant2 has the same actual product object, that product object will not re-query the database, and remembers (incorrectly) that its variants is an empty result.
Another change which might make your test run is to change your test as follows:
before do
#variant2 = FactoryBot.create(:variant, product_id: product.id, option_values: variant1.option_values)
end
It is subtle and I'd like to know if it works. This sets the product_id field on variant2, but does not set the product object for the association to be the actual same product object that variant1 has. (In practice this is more likely to happen in your actual code, that the product object is not shared between variant objects.)
Another thing for your correct solution (if all this is right) is to do the reload but put all your save code (and your update code) in a transaction. That way there won't be a race condition of two variants which would conflict, because in a transaction the first must complete the validation and save before the second one does its validation, so it will be sure to detect the other one which just saved.
Some suggested debugging techniques:
If possible, watch the log to see when queries are made. You might have caught that the second validation did not query for variants.
Check the object_id. You might have caught that the product objects were in fact the same object.
Also check new_record? to make sure that variant1 saved before you tested variant2. I think it does save, but it would have be nice to know you checked that.
Let's say I have the following relationship in my Rails app:
class Parent < ActiveRecord::Base
has_many :kids
end
class Kid < ActiveRecord::Base
belongs_to :parent
end
I want parents to be able to see a list of their chatty kids, and use the count in paginating through that list. Here's a way to do that (I know it's a little odd, bear with me):
class Parent < ActiveRecord::Base
has_many :kids do
def for_chatting
proxy_association.owner.kids.where(:chatty => true)
end
end
end
But! Some parents have millions of kids, and p.kids.for_chatting.count takes too long to run, even with good database indexes. I'm pretty sure this cannot be directly fixed. But! I can set up a Parent#chatty_kids_count attribute and keep it correctly updated with database triggers. Then, I can:
class Parent < ActiveRecord::Base
has_many :kids do
def for_chatting
parent = proxy_association.owner
kid_assoc = parent.kids.where(:chatty => true)
def kid_assoc.count
parent.chatty_kids_count
end
end
end
end
And then parent.kids.for_chatting.count uses the cached count and my pagination is fast.
But! Overriding count() on singleton association objects makes the uh-oh, I am being way too clever part of my brain light up big-time.
I feel like there's a clearer way to approach this. Is there? Or is this a "yeah it's weird, leave a comment about why you're doing this and it'll be fine" kind of situation?
Edit:
I checked the code of will_paginate, seems like it is not using count method of AR relation, but i found that you can provide option total_entries for paginate
#kids = #parent.kids.for_chatting.paginate(
page: params[:page],
total_entries: parent.chatty_kids_count
)
This is not working
You can use wrapper for collection like here
https://github.com/kaminari/kaminari/pull/818#issuecomment-252788488,
just override count method.
class RelationWrapper < SimpleDelegator
def initialize(relation, total_count)
super(relation)
#total_count = total_count
end
def count
#total_count
end
end
# in a controller:
relation = RelationWrapper.new(#parent.kids.for_chatting, parent.chatty_kids_count)
I'm struggling with the following problem: how to force object to destroy when it has no childs (or more precisly last child got destroyed). I tried following soltions:
Adding counter cache to the parent and destroing him if it is equal to 0, but then i couldn't create it using Factory Girl it was immediately destroyed.
Last solution from Delete Orphaned Parent, unfortunatelly it won't work with destroy_all - I'm getting stack overflow.
I cannot find any other up-to-date solution. Any one have any idea how can it be done?
One way to handle this is in an after_destroy callback on the the child object. Here's an example with two models named Parent and Child:
class Parent < ActiveRecord::Base
has_many :children
end
class Child < ActiveRecord::Base
belongs_to :parent
after_destroy: :destroy_orphaned_parent
def destroy_orphaned_parent
parent.destroy if parent.children.empty?
end
end
This solution works fine with destroy_all and will work for creating Parent records with no children in FactoryGirl.
I would suggest the following:
(1) Child model should not be able to call parent_model.destory, which is against model design pattern.
(2) In parent model after_update callback, check if children.count > 0. destroy if no associations found.
The Setup
I have an STI setup like so:
class Transaction < ActiveRecord::Base
belongs_to :account
scope :deposits, -> { where type: Deposit }
end
class Deposit < Transaction
scope :pending, -> { where state: :pending }
end
class Account < ActiveRecord::Base
has_many :transactions
end
If I call:
> a = Account.first
> a.transactions.deposits
...then I get what I expect, a collection of Deposit instances, however if I look at the class of what's returned:
> a.transactions.deposits.class
...then it's actually not a Deposit collection, it's still a Transaction collection, ie. it's a Transaction::ActiveRecord_AssociationRelation
The Problem
So, to the problem, if I then want to call one of the Deposit scopes on that collection it fails:
> a.transactions.deposits.pending
NoMethodError: undefined method `pending' for #<Transaction::ActiveRecord_Associations_CollectionProxy:0x007f8ac1252d00>
Things I've Checked
I've tried changing the scope to Deposit.where... which had no effect, and also to Deposit.unscoped.where... which actually returns the right collection object, but it strips all the scope, so I lose the account_id=123 part of the query so it fails on that side.
I've checked this and the problem exists for both Rails 4.1 and 4.2. Thanks for any pointers on how to make this work.
I know there's a workaround, but...
I know I could work around the issue by adding a has_many :deposits into Account, but I'm trying to avoid that (in reality I have many associated tables and many different transaction subclasses, and I'm trying to avoid adding the dozens of extra associations that would require).
Question
How can I get what's returned by the deposits scope to actually be a Deposit::ActiveRecord_Association... so that I can chain my scopes from Deposit class?
I created an isolated test for your issue here:https://gist.github.com/aalvarado/4ce836699d0ffb8b3782#file-sti_scope-rb and it has the error you mentioned.
I came across this post from pivotal http://pivotallabs.com/merging-scopes-with-sti-models/ about using were_values in a scope to get all the conditions. I then used them on unscope to force the expected class, basically this:
def self.deposits
conditions = where(nil).where_values.reduce(&:and)
Deposit.unscoped.where(conditions)
end
This test asserts that it returns a Deposit::ActiveRecord_Relation https://gist.github.com/aalvarado/4ce836699d0ffb8b3782#file-sti_scope2-rb
Update
You can also write this as a scope if you prefer:
scope :deposits, -> { Deposit.unscoped.where where(nil).where_values.reduce &:and }
As a quick workaround you can do > a.transactions.deposits.merge(Deposit.pending), but can't think of a different way of solving it. I'll think and try more options later and come back if I find anything.
You might want to say that an Account has_many :deposits
class Account < ActiveRecord::Base
has_many :transactions
has_many :deposits
end
Then you should be able to query
a.deposits.pending
Updated
Appears to be a precedence error and nothing to do with the question I originally asked. See discussion below.
Original question
Is it possible to use active record associations in callbacks? I've tested this code in the console and it works fine as long as it isn't in a callback. I'm trying to create callbacks that pull attributes from other associated models and I keep getting errors of nil.attribute.
If callbacks are not the correct approach to take, how would one do a similar action in rails? If the associations are simple, you could use create_association(attributes => ), but as associations get more complex this starts to get messy.
For example...
class User < ActiveRecord::Base
belongs_to :b
before_validation_on_create {|user| user.create_b} #note, other logic prevents creating multiple b
end
class B < ActiveRecord::Base
has_many :users, :dependent => destroy
after_create{ |b| b.create_c }
has_one :c
end
class C < ActiveRecord::Base
belongs_to :b
after_create :create_alert_email
private
def create_alert_email
self.alert_email = User.find_by_b_id(self.b_id).email #error, looks for nil.email
end
end
Off course associations are available in your callbacks. After all, the create_after_email is simply a method. You can even call it alone, without using a callback. ActiveRecord doesn't apply any special flag to callback methods to prevent them from working as any other method.
Also notice you are running a User#find query directly without taking advantage of any association method. An other reason why ActiveRecord association feature should not be the guilty in this case.
The reason why you are getting the error should probably searched somewhere else.
Be sure self.b_id is set and references a valid record. Perhaps it is nil or actually there's no User record with that value. In fact, you don't test whether the query returns a record or nil: you are assuming a record with that value always exists. Are you sure this assumption is always statisfied?