validate_uniqueness_of validation fails when having more than one validates to the same field #830
Labels
Milestone
Comments
|
Yes, starting in version 3, the matcher will look at the existing uniqueness validation that is on the attribute in question and will compare the scopes defined there with the scopes you are passing to the So you've definitely found a bug 🪲 |
|
+1 to this issue |
|
@luismaia did you find a work around by chance? |
|
@durhamka I decide to stay (for now) with version 2.8.0. Anyway I guess that if you create a custom validator (containing the logic of your multiple validators) it will work. |
|
Thanks! @luismaia |
|
the same, but for my cases 3.0.1 works pretty well |
|
@lolgear Can you elaborate how 3.0.1 was an improvement? |
|
|
I've got a partial fix for the original issue mentioned at the very top here: 28bd9a1. The reason why it's a partial fix is because the matcher now knows about multiple uniqueness validations that are on the same attribute, so you won't get the exact failure you got again. But there's still an issue lurking here, and this happens to be one of those issues that is actually a little complicated to fix. (I'm surprised it even worked in 2.8.0.) Remember that the uniqueness matcher works like this: the matcher will take (or create) an existing record, then make a duplicate of it (this becomes the "new" record). This record is invalid out of the gate, so the matcher will then set certain attributes on the new record that are different than corresponding attributes in the existing record, then assert that the record is valid with the changes (because it is now unique). The attributes it chooses to set are based on the list you provide to So when you say So the new record that it is working with won't be completely valid, because it won't be completely unique. I know that wasn't the best explanation, but the point is, I might need to add another qualifier to properly handle this case, and I'm not sure what that looks like yet. So I'll have to think about it. But the failure message you saw is at least fixed. |
|
FYI I'm getting a slightly different error: |
|
@s2t2 Can you provide a backtrace? |
|
Sure. See below. Hope this is helpful. |
|
@s2t2 I think you're getting that error because the matcher is trying to set I'm curious -- how would you write a manual test for this? |
|
@mcmire something like this (model spec for describe "uniqueness of composite key which includes polymorphic association" do
context "when there is an existing other attribute for a given polymorphic resource" do
let(:poly){ create(:other_model) }
let!(:thing){ create(:thing, poly: poly, other_attribute: "MyVal")}
let(:duplicate){ create(:thing, poly: poly, other_attribute: "MyVal")}
let(:different){ create(:thing, poly: poly, other_attribute: "OtherVal")}
it "should not allow duplicate combinations of the other attribute and the polymorphic resource" do
expect{ duplicate }.to raise_error(ActiveRecord::RecordInvalid, "Validation failed: Other attribute has already been taken")
end
it "should allow different other attributes to be associated with the same polymorphic resource" do
expect{ different }.to change{ Thing.count }.by(1)
end
end
end |
|
I'm going to go ahead and close this issue since I think the original issue has been fixed and I can't quite understand the edge case I posted above. If there any more issues with the uniqueness matcher in the future I think it's better to start a new issue so I can keep track of it better. |
netbsd-srcmastr
pushed a commit
to NetBSD/pkgsrc
that referenced
this issue
Sep 23, 2018
# 3.1.2 ### Deprecations * This is the **last version** that supports Rails 4.0 and 4.1 and Ruby 2.0 and 2.1. ### Bug fixes * When the `permit` matcher was used without `#on`, the controller did not use `params#require`, the params object was duplicated, and the matcher did not recognize the `#permit` call inside the controller. This behavior happened because the matcher overwrote double registries with the same parameter hash whenever ActionController::Parameters was instantiated. * *Commit: [44c019]* * *Issue: [#899]* * *Pull request: [#902]* # 3.1.1 ### Bug fixes * Some matchers make use of ActiveSupport's `in?` method, but do not include the file where this is defined in ActiveSupport. This causes problems with projects using shoulda-matchers that do not include all of ActiveSupport by default. To fix this, replace `in?` with Ruby's builtin `include?`. * *Pull request: [#879]* * `validate_uniqueness_of` works by creating a record if it doesn't exist, and then testing against a new record with various attributes set that are equal to (or different than) corresponding attributes in the existing record. In 3.1.0 a change was made whereby when the uniqueness matcher is given a new record and creates an existing record out of it, it ensures that the record is valid before continuing on. This created a problem because if the subject, before it was saved, was empty and therefore in an invalid state, it could not effectively be saved. While ideally this should be enforced, doing so would be a backward-incompatible change, so this behavior has been rolled back. ([#880], [#884], [#885]) * *Commit: [45de869]* * *Issues: [#880], [#884], [#885]* * Fix an issue with `validate_uniqueness_of` + `scoped_to` when used against a model where the attribute has multiple uniqueness validations and each validation has a different set of scopes. In this case, a test written for the first validation (and its scopes) would pass, but tests for the other validations (and their scopes) would not, as the matcher only considered the first set of scopes as the *actual* set of scopes. * *Commit: [28bd9a1]* * *Issues: [#830]* ### Improvements * Update `validate_uniqueness_of` so that if an existing record fails to be created because a column is non-nullable and was not filled in, raise an ExistingRecordInvalid exception with details on how to fix the test. * *Commit: [78ccfc5]* [#879]: thoughtbot/shoulda-matchers#879 [45de869]: thoughtbot/shoulda-matchers@45de869 [#880]: thoughtbot/shoulda-matchers#880 [#884]: thoughtbot/shoulda-matchers#884 [#885]: thoughtbot/shoulda-matchers#885 [78ccfc5]: thoughtbot/shoulda-matchers@78ccfc5 [28bd9a1]: thoughtbot/shoulda-matchers@28bd9a1 [#830]: thoughtbot/shoulda-matchers#830 # 3.1.0 ### Bug fixes * Update `validate_numericality_of` so that submatchers are applied lazily instead of immediately. Previously, qualifiers were order-dependent, meaning that if you used `strict` before you used, say, `odd`, then `strict` wouldn't actually apply to `odd`. Now the order that you specify qualifiers doesn't matter. * *Source: [6c67a5e]* * Fix `allow_value` so that it does not raise an AttributeChangedValueError (formerly CouldNotSetAttributeError) when used against an attribute that is an enum in an ActiveRecord model. * *Source: [9e8603e]* * Add a `ignoring_interference_by_writer` qualifier to all matchers, not just `allow_value`. *This is enabled by default, which means that you should never get a CouldNotSetAttributeError again.* (You may get some more information if a test fails, however.) * *Source: [1189934], [5532f43]* * *Fixes: [#786], [#799], [#801], [#804], [#817], [#841], [#849], [#872], [#873], and [#874]* * Fix `validate_numericality_of` so that it does not blow up when used against a virtual attribute defined in an ActiveRecord model (that is, an attribute that is not present in the database but is defined using `attr_accessor`). * *Source: [#822]* * Update `validate_numericality_of` so that it no longer raises an IneffectiveTestError if used against a numeric column. * *Source: [5ed0362]* * *Fixes: [#832]* [6c67a5e]: thoughtbot/shoulda-matchers@6c67a5e [9e8603e]: thoughtbot/shoulda-matchers@9e8603e [1189934]: thoughtbot/shoulda-matchers@1189934 [5532f43]: thoughtbot/shoulda-matchers@5532f43 [#786]: thoughtbot/shoulda-matchers#786 [#799]: thoughtbot/shoulda-matchers#799 [#801]: thoughtbot/shoulda-matchers#801 [#804]: thoughtbot/shoulda-matchers#804 [#817]: thoughtbot/shoulda-matchers#817 [#841]: thoughtbot/shoulda-matchers#841 [#849]: thoughtbot/shoulda-matchers#849 [#872]: thoughtbot/shoulda-matchers#872 [#873]: thoughtbot/shoulda-matchers#873 [#874]: thoughtbot/shoulda-matchers#874 [#822]: thoughtbot/shoulda-matchers#822 [5ed0362]: thoughtbot/shoulda-matchers@5ed0362 [#832]: thoughtbot/shoulda-matchers#832 ### Features * Add a new qualifier, `ignoring_case_sensitivity`, to `validate_uniqueness_of`. This provides a way to test uniqueness of an attribute whose case is normalized, either in a custom writer method for that attribute, or in a custom `before_validation` callback. * *Source: [#840]* * *Fixes: [#836]* [#840]: thoughtbot/shoulda-matchers#840 [#836]: thoughtbot/shoulda-matchers#836 ### Improvements * Improve failure messages and descriptions of all matchers across the board so that it is easier to understand what the matcher was doing when it failed. (You'll see a huge difference in the output of the numericality and uniqueness matchers in particular.) * Matchers now raise an error if any attributes that the matcher is attempting to set do not exist on the model. * *Source: [2962112]* * Update `validate_numericality_of` so that it doesn't always run all of the submatchers, but stops on the first one that fails. Since failure messages now contain information as to what value the matcher set on the attribute when it failed, this change guarantees that the correct value will be shown. * *Source: [8e24a6e]* * Continue to detect if attributes change incoming values, but now instead of immediately seeing a CouldNotSetAttributeError, you will only be informed about it if the test you've written fails. * *Source: [1189934]* * Add an additional check to `define_enum_for` to ensure that the column that underlies the enum attribute you're testing is an integer column. * *Source: [68dd70a]* * Add a test for `validate_numericality_of` so that it officially supports money columns. * *Source: [a559713]* * *Refs: [#841]* [2962112]: thoughtbot/shoulda-matchers@2962112 [8e24a6e]: thoughtbot/shoulda-matchers@8e24a6e [68dd70a]: thoughtbot/shoulda-matchers@68dd70a [a559713]: thoughtbot/shoulda-matchers@a559713 # 3.0.1 ### Bug fixes * Fix `validate_inclusion_of` + `in_array` when used against a date or datetime column/attribute so that it does not raise a CouldNotSetAttributeError. ([#783], [8fa97b4]) * Fix `validate_numericality_of` when used against a numeric column so that it no longer raises a CouldNotSetAttributeError if the matcher has been qualified in any way (`only_integer`, `greater_than`, `odd`, etc.). ([#784], [#812]) ### Improvements * `validate_uniqueness_of` now raises a NonCaseSwappableValueError if the value the matcher is using to test uniqueness cannot be case-swapped -- in other words, if it doesn't contain any alpha characters. When this is the case, the matcher cannot work effectively. ([#789], [ada9bd3]) [#783]: thoughtbot/shoulda-matchers#783 [8fa97b4]: thoughtbot/shoulda-matchers@8fa97b4 [#784]: thoughtbot/shoulda-matchers#784 [#789]: thoughtbot/shoulda-matchers#789 [ada9bd3]: thoughtbot/shoulda-matchers@ada9bd3 [#812]: thoughtbot/shoulda-matchers#812 # 3.0.0 ### Backward-incompatible changes * We've dropped support for Rails 3.x, Ruby 1.9.2, and Ruby 1.9.3, and RSpec 2. All of these have been end-of-lifed. ([a4045a1], [b7fe87a], [32c0e62]) * The gem no longer detects the test framework you're using or mixes itself into that framework automatically. [History][no-auto-integration-1] has [shown][no-auto-integration-2] that performing any kind of detection is prone to bugs and more complicated than it should be. Here are the updated instructions: * You no longer need to say `require: false` in your Gemfile; you can include the gem as normal. * You'll need to add the following somewhere in your `rails_helper` (for RSpec) or `test_helper` (for Minitest / Test::Unit): ``` ruby Shoulda::Matchers.configure do |config| config.integrate do |with| # Choose a test framework: with.test_framework :rspec with.test_framework :minitest with.test_framework :minitest_4 with.test_framework :test_unit # Choose one or more libraries: with.library :active_record with.library :active_model with.library :action_controller # Or, choose the following (which implies all of the above): with.library :rails end end ``` ([1900071]) * Previously, under RSpec, all of the matchers were mixed into all of the example groups. This created a problem because some gems, such as [active_model_serializers-matchers], provide matchers that share the same name as some of our own matchers. Now, matchers are only mixed into whichever example group they belong to: * ActiveModel and ActiveRecord matchers are available only in model example groups. * ActionController matchers are available only in controller example groups. * The `route` matcher is available only in routing example groups. ([af98a23], [8cf449b]) * There are two changes to `allow_value`: * The negative form of `allow_value` has been changed so that instead of asserting that any of the given values is an invalid value (allowing good values to pass through), assert that *all* values are invalid values (allowing good values not to pass through). This means that this test which formerly passed will now fail: ``` ruby expect(record).not_to allow_value('good value', *bad_values) ``` ([19ce8a6]) * `allow_value` now raises a CouldNotSetAttributeError if in setting the attribute, the value of the attribute from reading the attribute back is different from the one used to set it. This would happen if the writer method for that attribute has custom logic to ignore certain incoming values or change them in any way. Here are three examples we've seen: * You're attempting to assert that an attribute should not allow nil, yet the attribute's writer method contains a conditional to do nothing if the attribute is set to nil: ``` ruby class Foo include ActiveModel::Model attr_reader :bar def bar=(value) return if value.nil? @bar = value end end describe Foo do it do foo = Foo.new foo.bar = "baz" # This will raise a CouldNotSetAttributeError since `foo.bar` is now "123" expect(foo).not_to allow_value(nil).for(:bar) end end ``` * You're attempting to assert that an numeric attribute should not allow a string that contains non-numeric characters, yet the writer method for that attribute strips out non-numeric characters: ``` ruby class Foo include ActiveModel::Model attr_reader :bar def bar=(value) @bar = value.gsub(/\D+/, '') end end describe Foo do it do foo = Foo.new # This will raise a CouldNotSetAttributeError since `foo.bar` is now "123" expect(foo).not_to allow_value("abc123").for(:bar) end end ``` * You're passing a value to `allow_value` that the model typecasts into another value: ``` ruby describe Foo do # Assume that `attr` is a string # This will raise a CouldNotSetAttributeError since `attr` typecasts `[]` to `"[]"` it { should_not allow_value([]).for(:attr) } end ``` With all of these failing examples, why are we making this change? We want to guard you (as the developer) from writing a test that you think acts one way but actually acts a different way, as this could lead to a confusing false positive or negative. If you understand the problem and wish to override this behavior so that you do not get a CouldNotSetAttributeError, you can add the `ignoring_interference_by_writer` qualifier like so. Note that this will not always cause the test to pass. ``` ruby it { should_not allow_value([]).for(:attr).ignoring_interference_by_writer } ``` ([9d9dc4e]) * `validate_uniqueness_of` is now properly case-sensitive by default, to match the default behavior of the validation itself. This is a backward-incompatible change because this test which incorrectly passed before will now fail: ``` ruby class Product < ActiveRecord::Base validates_uniqueness_of :name, case_sensitive: false end describe Product do it { is_expected.to validate_uniqueness_of(:name) } end ``` ([57a1922]) * `ensure_inclusion_of`, `ensure_exclusion_of`, and `ensure_length_of` have been removed in favor of their `validate_*` counterparts. ([55c8d09]) * `set_the_flash` and `set_session` have been changed to more closely align with each other: * `set_the_flash` has been removed in favor of `set_flash`. ([801f2c7]) * `set_session('foo')` is no longer valid syntax, please use `set_session['foo']` instead. ([535fe05]) * `set_session['key'].to(nil)` will no longer pass when the key in question has not been set yet. ([535fe05]) * Change `set_flash` so that `set_flash[:foo].now` is no longer valid syntax. You'll want to use `set_flash.now[:foo]` instead. This was changed in order to more closely align with how `flash.now` works when used in a controller. ([#755], [#752]) * Change behavior of `validate_uniqueness_of` when the matcher is not qualified with any scopes, but your validation is. Previously the following test would pass when it now fails: ``` ruby class Post < ActiveRecord::Base validate :slug, uniqueness: { scope: :user_id } end describe Post do it { should validate_uniqueness_of(:slug) } end ``` ([6ac7b81]) [active_model_serializers-matchers]: https://github.com/adambarber/active_model_serializers-matchers [no-auto-integration-1]: freerange/mocha@049080c [no-auto-integration-2]: rr/rr#29 [1900071]: thoughtbot/shoulda-matchers@1900071 [b7fe87a]: thoughtbot/shoulda-matchers@b7fe87a [a4045a1]: thoughtbot/shoulda-matchers@a4045a1 [57a1922]: thoughtbot/shoulda-matchers@57a1922 [19ce8a6]: thoughtbot/shoulda-matchers@19c38a6 [eaaa2d8]: thoughtbot/shoulda-matchers@eaaa2d8 [55c8d09]: thoughtbot/shoulda-matchers@55c8d09 [801f2c7]: thoughtbot/shoulda-matchers@801f2c7 [535fe05]: thoughtbot/shoulda-matchers@535fe05 [6ac7b81]: thoughtbot/shoulda-matchers@6ac7b81 [#755]: thoughtbot/shoulda-matchers#755 [#752]: thoughtbot/shoulda-matchers#752 [9d9dc4e]: thoughtbot/shoulda-matchers@9d9dc4e [32c0e62]: thoughtbot/shoulda-matchers@32c0e62 [af98a23]: thoughtbot/shoulda-matchers@af98a23 [8cf449b]: thoughtbot/shoulda-matchers@8cf449b ### Bug fixes * So far the tests for the gem have been running against only SQLite. Now they run against PostgreSQL, too. As a result we were able to fix some Postgres-related bugs, specifically around `validate_uniqueness_of`: * When scoped to a UUID column that ends in an "f", the matcher is able to generate a proper "next" value without erroring. ([#402], [#587], [#662]) * Support scopes that are PostgreSQL array columns. Please note that this is only supported for Rails 4.2 and greater, as versions before this cannot handle array columns correctly, particularly in conjunction with the uniqueness validator. ([#554]) * Fix so that when scoped to a text column and the scope is set to nil before running it through the matcher, the matcher does not fail. ([#521], [#607]) * Fix `define_enum_for` so that it actually tests that the attribute is present in the list of defined enums, as you could fool it by merely defining a class method that was the pluralized version of the attribute name. In the same vein, passing a pluralized version of the attribute name to `define_enum_for` would erroneously pass, and now it fails. ([#641]) * Fix `permit` so that it does not break the functionality of ActionController::Parameters#require. ([#648], [#675]) * Fix `validate_uniqueness_of` + `scoped_to` so that it does not raise an error if a record exists where the scoped attribute is nil. ([#677]) * Fix `route` matcher so if your route includes a default `format`, you can specify this as a symbol or string. ([#693]) * Fix `validate_uniqueness_of` so that it allows you to test against scoped attributes that are boolean columns. ([#457], [#694]) * Fix failure message for `validate_numericality_of` as it sometimes didn't provide the reason for failure. ([#699]) * Fix `shoulda/matchers/independent` so that it can be required independently, without having to require all of the gem. ([#746], [e0a0200]) ### Features * Add `on` qualifier to `permit`. This allows you to make an assertion that a restriction was placed on a slice of the `params` hash and not the entire `params` hash. Although we don't require you to use this qualifier, we do recommend it, as it's a more precise check. ([#675]) * Add `strict` qualifier to `validate_numericality_of`. ([#620]) * Add `on` qualifier to `validate_numericality_of`. ([9748869]; h/t [#356], [#358]) * Add `join_table` qualifier to `have_and_belong_to_many`. ([#556]) * `allow_values` is now an alias for `allow_value`. This makes more sense when checking against multiple values: ``` ruby it { should allow_values('this', 'and', 'that') } ``` ([#692]) [9748869]: thoughtbot/shoulda-matchers@9748869 [#402]: thoughtbot/shoulda-matchers#402 [#587]: thoughtbot/shoulda-matchers#587 [#662]: thoughtbot/shoulda-matchers#662 [#554]: thoughtbot/shoulda-matchers#554 [#641]: thoughtbot/shoulda-matchers#641 [#521]: thoughtbot/shoulda-matchers#521 [#607]: thoughtbot/shoulda-matchers#607 [#648]: thoughtbot/shoulda-matchers#648 [#675]: thoughtbot/shoulda-matchers#675 [#677]: thoughtbot/shoulda-matchers#677 [#620]: thoughtbot/shoulda-matchers#620 [#693]: thoughtbot/shoulda-matchers#693 [#356]: thoughtbot/shoulda-matchers#356 [#358]: thoughtbot/shoulda-matchers#358 [#556]: thoughtbot/shoulda-matchers#556 [#457]: thoughtbot/shoulda-matchers#457 [#694]: thoughtbot/shoulda-matchers#694 [#692]: thoughtbot/shoulda-matchers#692 [#699]: thoughtbot/shoulda-matchers#699 [#746]: thoughtbot/shoulda-matchers#746
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I have the following validations in one of my models:
And my RSpec is:
When upgrading from version 2.8.0 to 3.0.1 I start having this error:
If I exchange the validates order, in the model I got this error:
My conclusion is that shoulda-matchers (in version 3) is only executing the first validates definition per field. Is this conclusion correct?
I tested against master and I get the same error.
Is this an issue or am I doing something wrong?
The text was updated successfully, but these errors were encountered: