In Ruby 2.4, BigDecimal(), as used by the Decimal cast, was changed so
that it will raise ArgumentError when passed an invalid string, in order
to be more consistent with Integer(), Float(), etc. The other numeric
types use ex. to_i and to_f.
Unfortunately, we can't simply change BigDecimal() to to_d. String#to_d
raises errors like BigDecimal(), unlike all the other to_* methods (this
should probably be filed as a ruby bug).
Instead, this simulates the existing behaviour and the behaviour of the
other to_* methods by finding a numeric string at the start of the
passed in value, and parsing that using BigDecimal().
See also
https://bugs.ruby-lang.org/issues/102863081a627ce
This issue is only appear when you try to call `define_attribute_method`
and passing a symbol in Active Record. It does not appear in isolation
in Active Model itself.
Before this patch, when you run `User.define_attribute_method :foo`, you
will get:
NoMethodError: undefined method `unpack' for :foo:Symbol
from activerecord/lib/active_record/attribute_methods/read.rb:28:in `define_method_attribute'
from activerecord/lib/active_record/attribute_methods/primary_key.rb:61:in `define_method_attribute'
from activemodel/lib/active_model/attribute_methods.rb:292:in `block in define_attribute_method'
from activemodel/lib/active_model/attribute_methods.rb:285:in `each'
from activemodel/lib/active_model/attribute_methods.rb:285:in `define_attribute_method'
This patch contains both a fix in Active Model and a test in Active
Record for this error.
`ActiveModel::TestCase` is used only for the test of Active Model.
Also, it is a private API and can not be used in applications.
Therefore, it is not necessary to include it in lib.
Right now it behaves differently on JRuby:
```
--- expected
+++ actual
@@ -1 +1 @@
-#<BigDecimal:5f3c866c,'0.333333333333333333',18(20)>
+#<BigDecimal:16e0afab,'0.3333333333333333',16(20)>
```
My initial PR (https://github.com/rails/rails/pull/27324)
offered to let the precision to be decided by the platform and
change the test expection, but other contributors suggested
that we should change the default precision in Rails
to be consistent of all platforms.
The value (18) comes from the max default precision that comes
from casting Rational(1/3) to BigDecimal.
As pointed out by @matthewd this change makes ImmutableString aware
of MysqlString's existence whereas previously MysqlString was only
overriding public API.
cc @kamipo
This reverts commit e632c2fa4cb60072a778ce95c952a0fa95e5b074, reversing
changes made to 334a7dcf107cd3ff1697163d331d289d6d65dcd7.
The only difference between `Type::ImmutableString` and its subclasses
is the representation of the casted booleans. Prefer extracting
`casted_true`/`casted_false` and override these by subclasses.
If a Error object was serialized in the database as YAML in the Rails
4.2 version, if we load in the Rails 5.0 version it will miss the
@details instance variable so methods like #clear and #add will start to
fail.
In Rails 4.1, Action View was extracted from Action Pack, but this
change was not reflected in the API documentation of ActiveModel::Model.
This commits makes it explicit in the documentation that Active Model
also interacts with Action View as well as Action Pack.
[ci skip]
In ActiveModel::Model API documentation, referrences to Rails
components were tagged with fixed-width font and named as if
they were modules.This fixes the inconsistency to match API
documentation conventions.
[ci skip]
Regexp#match? should be considered to be part of the Ruby core library. We are
emulating it for < 2.4, but not having to require the extension is part of the
illusion of the emulation.
This change now creates a method `#authenticate_XXX` where XXX is
the configured attribute name on `#has_secure_password`. `#authenticate`
is now an alias to this method when the attribute name is the default
'password'
existing `#has_secure_password`. This can be useful when one would
like to store some secure field as a digest, just like a password.
The method still defaults to `password`. It now also allows using the
same `#authenticate` method which now accepts a second argument for
specifying the attribute to be authenticated, or will default to 'password`.
A new method is also added for generating a new token for an attribute by
calling `#regenerate_XXXX` where `XXXX` is the attribute name.
All indentation was normalized by rubocop auto-correct at 80e66cc4d90bf8c15d1a5f6e3152e90147f00772.
But comments was still kept absolute position. This commit aligns
comments with method definitions for consistency.
Recently, the Rails team made an effort to keep the source code consistent, using Ruboco
(bb1ecdcc677bf6e68e0252505509c089619b5b90 and below). Some of the case
statements were missed.
This changes the case statements' formatting and is consistent with changes
in 810dff7c9fa9b2a38eb1560ce0378d760529ee6b and db63406cb007ab3756d2a96d2e0b5d4e777f8231.
NumericalityValidator#validate_each is never called when allow_nil is true and
the value is nil because it is already skipped in EachValidator#validate.
A few have been left for aesthetic reasons, but have made a pass
and removed most of them.
Note that if the method `foo` returns an array, `foo << 1`
is a regular push, nothing to do with assignments, so
no self required.
Currently `Type::Date#serialize` does not cast a value to a date object.
It should be cast to a date object for finding by date column correctly
working.
Fixes#25354.
We now use default procs inside of the errors object, which gets
included by default when marshaling anything that includes
`ActiveModel::Validations`. This means that Active Record objects cannot
be marshalled. We strip and apply the default proc ourselves. This will
ensure the objects are YAML serializable as well, since YAML falls back
to marshal implementations now. This is less important, however, as the
errors aren't included when dumping Active Record objects.
This commit does not include a changelog entry, as 5.0 is still in RC
status at the time of writing, and 5.0.0 will not release with the bug
this fixes.
Fixes#25165
Ruby 2.4 unifies Fixnum and Bignum into Integer: https://bugs.ruby-lang.org/issues/12005
* Forward compat with new unified Integer class in Ruby 2.4+.
* Backward compat with separate Fixnum/Bignum in Ruby 2.2 & 2.3.
* Drops needless Fixnum distinction in docs, preferring Integer.
- we are ending sentences properly
- fixing of space issues
- fixed continuity issues in some sentences.
Reverts 8fc97d198e .
This change reverts making sure we add '.' at end of deprecation sentences.
This is to keep sentences within Rails itself consistent and with a '.' at the end.
- Pass object to I18n helper so that when calling message proc, it will
pass that object as argument to the proc and we can generate custom
error messages based on current record being validated.
- Based on https://github.com/rails/rails/issues/856.
[Łukasz Bandzarewicz, Prathamesh Sonpatki]
Since precision is always larger than scale, it can actually change
rounding behavior. Given a precision of 5 and a scale of 3, when you
apply the precision of 5 to `1.25047`, the result is `1.2505`, which
when the scale is applied would be `1.251` instead of the expected
`1.250`.
This issue appears to only occur with floats, as scale doesn't apply to
other numeric types, and the bigdecimal constructor actually ignores
precision entirely when working with strings. There's no way we could
handle this for the "unknown object which responds to `to_d`" case, as
we can't assume an interface for applying the scale.
Fixes#24235
From: https://github.com/rails/rails/issues/24279
Problem:
By doing `record.errors.include? :foo`, it adds a new key to the
@messages hash that defaults to an empty array.
This happens because of a combination of these 2 commits:
b97035df64
(Added in Rails 4.1)
and
6ec8ba16d8 (diff-fdcf8b65b5fb954372c6fe1ddf284c78R76)
(Rails 5.0)
By adding the default proc that returns an array for non-existing keys,
ruby adds that key to the hash.
Solution:
Change `#include?` to check with `has_key?` and then check if that value is
`present?`.
Add test case for ActiveModels::Errors#include?
In order to fix issue #17621 we added a check to validations that
determined if a record should be validated. Based on the existing tests
and behavior we wanted we determined the best way to do that was by
checking if `!record.peristed? || record.changed? || record.marked_for_destruction?`
This change didn't make it into a release until now. When #23790 was
opened we realized that `valid?` and `invalid?` were broken and did not
work on persisted records because of the `!record.persisted?`.
While there is still a bug that #17621 brought up, this change was too
drastic and should not be a RC blocker. I will work on fixing this so
that we don't break `valid?` but also aren't validating parent records
through child records if that parent record is validate false. This
change removes the code changes to validate and the corresponding tests.
It adds tests for two of the bugs found since Rails 5 beta2 release.
Fixes#17621
Fixes#23645
When you're using an `attr_accessor` for a record instead of an
attribute in the database there's no way for the record to know if it
has `changed?` unless you tell it `attribute_will_change!("attribute")`.
The change made in 27aa4dd updated validations to check if a record was
`changed?` or `marked_for_destruction?` or not `persisted?`. It did not
take into account virtual attributes that do not affect the model's
dirty status.
The only way to fix this is to always validate the record if the
attribute does not belong to the set of attributes the record expects
(in `record.attributes`) because virtual attributes will not be in that
hash.
I think we should consider deprecating this particular behavior in the
future and requiring that the user mark the record dirty by noting that
the virtual attribute will change. Unfortunately this isn't easy because
we have no way of knowing that you did the "right thing" in your
application by marking it dirty and will get the deprecation warning
even if you are doing the correct thing.
For now this restores expected behavior when using a virtual attribute
by always validating the record, as well as adds tests for this case.
I was going to add the `!record.attributes.include?(attribute)` to the
`should_validate?` method but `uniqueness` cannot validate a virtual
attribute with nothing to hold on to the attribute. Because of this
`should_validate?` was about to become a very messy method so I decided
to split them up so we can handle it specifically for each case.
The `attribute` parameter is not used inside the `normalize_detail`
method. This does not need to go through a deprecation cycle, since the
method is private.
* 5-0-beta-sec:
bumping version
fix version update task to deal with .beta1.1
Eliminate instance level writers for class accessors
allow :file to be outside rails root, but anything else must be inside the rails view directory
Don't short-circuit reject_if proc
stop caching mime types globally
use secure string comparisons for basic auth username / password
Instance level writers can have an impact on how the Active Model /
Record objects are saved. Specifically, they can be used to bypass
validations. This is a problem if mass assignment protection is
disabled and specific attributes are passed to the constructor.
CVE-2016-0753
This is an alternate implementation to #22875, that generalizes a lot of
the logic that type decorators are going to need, in order to have them
work with arrays, ranges, etc. The types have the ability to map over a
value, with the default implementation being to just yield that given
value. Array and Range give more appropriate definitions.
This does not automatically make ranges time zone aware, as they need to
be added to the `time_zone_aware` types config, but we could certainly
make that change if we feel it is appropriate. I do think this would be
a breaking change however, and should at least have a deprecation cycle.
Closes#22875.
/cc @matthewd
A regression (#22744) introduced in 7500dae caused certain numericality
validations to raise an error when run against an attribute with a
string value. Previously, these validations would successfully run
against string values because the value was cast to a numeric class.
This commit resolves the regression by converting string values to
floats before performing numericality comparison validations.
[fixes#22744]
We currently generate an unbounded number of prepared statements when
`limit` or `offset` are called with a dynamic argument. This changes
`LIMIT` and `OFFSET` to use bind params, eliminating the problem.
`Type::Value#hash` needed to be implemented, as it turns out we busted
the query cache if the type object used wasn't exactly the same object.
This drops support for passing an `Arel::Nodes::SqlLiteral` to `limit`.
Doing this relied on AR internals, and was never officially supported
usage.
Fixes#22250.
I seriously don't even know why we handle booleans, but those strings
should technically be frozen. Additionally, we don't need to actually
check the class in the mutable string type, since the `cast_value`
function will always return a string.
This type adds an escape hatch to apps for which string duping causes
unacceptable memory growth. The reason we are duping them is in order to
detect mutation, which was a feature added to 4.2 in #15674. The string
type was modified to support this behavior in #15788.
Memory growth is really only a concern for string types, as it's the
only mutable type where the act of coersion does not create a new object
regardless (as we're usually returning an object of a different class).
I do feel strongly that if we are going to support detecting mutation,
we should do it universally for any type which is mutable. While it is
less common and ideomatic to mutate strings than arrays or hashes, there
shouldn't be rules or gotchas to understanding our behavior.
However, I also appreciate that for apps which are using a lot of string
columns, this would increase the number of allocations by a large
factor. To ensure that we keep our contract, if you'd like to opt out of
mutation detection on strings, you'll also be option out of mutation of
those strings.
I'm not completely married to the thought that strings coming out of
this actually need to be frozen -- and I think the name is correct
either way, as the purpose of this is to provide a string type which
does not detect mutation.
In the new implementation, I'm only overriding `cast_value`. I did not
port over the duping in `serialize`. I cannot think of a reason we'd
need to dup the string there, and the tests pass without it.
Unfortunately that line was introduced at a time where I was not nearly
as good about writing my commit messages, so I have no context as to
why I added it. Thanks past Sean. You are a jerk.
This explanation was change by https://github.com/rails/rails/commit/7a27de2b.
This change reversed the including module (`ActiveModel::Serializers::JSON`)
and the included module (`ActiveModel::Serialization`) by mistake.
The implementation of `attribute_method?` on Active Record requires
establishing a database connection and querying the schema. As a general
rule, we don't want to require database connections for any class macro,
as the class should be able to be loaded without a database (e.g. for
things like compiling assets).
Instead of eagerly defining these methods, we do it lazily the first
time they are accessed via `method_missing`. This should not cause any
performance hits, as it will only hit `method_missing` once for the
entire class.
The biggest source of the performance regression in these methods
occurred because dirty tracking required eagerly materializing and type
casting the assigned values. In the previous commits, I've changed dirty
tracking to perform the comparisons lazily. However, all of this is moot
when calling `save`, since `changes_applied` will be called, which just
ends up eagerly materializing everything, anyway. With the new mutation
tracker, it's easy to just compare the previous two hashes in the same
lazy fashion.
We will not have aliasing issues with this setup, which is proven by the
fact that we're able to detect nested mutation.
Before:
User.create! 2.007k (± 7.1%) i/s - 10.098k
After:
User.create! 2.557k (± 3.5%) i/s - 12.789k
Fixes#19859
This moves a bit more of the logic required for dirty checking into the
attribute objects. I had hoped to remove the `with_value_from_database`
stuff, but unfortunately just calling `dup` on the attribute objects
isn't enough, since the values might contain deeply nested data
structures. I think this can be cleaned up further.
This makes most dirty checking become lazy, and reduces the number of
object allocations and amount of CPU time when assigning a value. This
opens the door (but doesn't quite finish) to improving the performance
of writes to a place comparable to 4.1
Timestamp column can have less precision than ruby timestamp
In result in how big a fraction of a second can be stored in the
database.
m = Model.create!
m.created_at.usec == m.reload.created_at.usec
# => false
# due to different seconds precision in Time.now and database column
If the precision is low enough, (mysql default is 0, so it is always low
enough by default) the value changes when model is reloaded from the
database. This patch fixes that issue ensuring that any timestamp
assigned as an attribute is converted to column precision under the
attribute.
In Active Record, it appears these were either autoloaded, which
actually was likely due to test ordering since the method `Float#to_d`
wouldn't trigger it. This makes it explicit, and unlikely to fail in the
future.
The `override` option is only a thing for Active Record registrations.
We should figure out how to make this properly error out without doing
anything too weird to the code.
Any tests for a type which is not overridden by Active Record, and does
not test the specifics of the attributes API interacting in more complex
ways have no reason to be in the Active Record suite. Doing this
revealed that the implementation of the date and time types in AM was
actually completely broken, and incapable of returning any value other
than `nil`.
Things like decorations, overrides, and priorities only matter for
Active Record, so the Active Model registry can be implemented much more
simply. At this point, I wonder if having Active Record's registry
inherit from Active Model's is even worth the trouble?
The Active Model class was also missing test cases, which have been
backfilled.
This removes the error when two types are registered with the same name,
but given that Active Model is meant to be significantly more generic, I
do not think this is an issue for now. If we want, we can raise an error
at the point that someone tries to register it.
We do not need to require each file from AM individually, the type
module does that for us. Even if the classes are extremely small right
now, I'd rather keep any custom classes needed by AR in their own files,
as they can easily have more complex changes in the future.
These are used by the connection adapters to convert SQL type
information into the appropriate type object, and makes no sense outside
of the context of Active Record
The thread_safe gem is being deprecated and all its code has been merged
into the concurrent-ruby gem. The new class, Concurrent::Map, is exactly
the same as its predecessor except for fixes to two bugs discovered
during the merge.
Example:
```ruby
class Person
include ActiveModel::Validations
attr_reader :name, :title
validates_presence_of :name, on: :create
validates_presence_of :title, on: :update
end
person = Person.new
person.valid?([:create, :update]) # => true
person.errors.messages # => {:name=>["can't be blank"], :title=>["can't be blank"]}
```
This reverts commit 51dd2588433457960cca592d5b5dac6e0537feac, reversing
changes made to ecb4e4b21b3222b823fa24d4a0598b1f2f63ecfb.
This broke Active Record tests
It was not expecting the new `case_insensitive` option to be passed to
`generate_message`, instead of fixing the test we can just not pass this
option down since it is specific to the confirmation validator and not
necessary for the error message.
Case :- 1. In case of email confirmation one needs case insensitive comparison
2. In case of password confirmation one needs case sensitive comparison
[ci skip] Update Guides for case_sensitive option in confirmation validation
Example:
```ruby
class Person
include ActiveModel::Validations
attr_reader :name, :title
validates_presence_of :name, on: :create
validates_presence_of :title, on: :update
end
person = Person.new
person.valid?([:create, :update]) # => true
person.errors.messages # => {:name=>["can't be blank"], :title=>["can't be blank"]}
```
I wrote a utility that helps find areas where you could optimize your program using a frozen string instead of a string literal, it's called [let_it_go](https://github.com/schneems/let_it_go). After going through the output and adding `.freeze` I was able to eliminate the creation of 1,114 string objects on EVERY request to [codetriage](codetriage.com). How does this impact execution?
To look at memory:
```ruby
require 'get_process_mem'
mem = GetProcessMem.new
GC.start
GC.disable
1_114.times { " " }
before = mem.mb
after = mem.mb
GC.enable
puts "Diff: #{after - before} mb"
```
Creating 1,114 string objects results in `Diff: 0.03125 mb` of RAM allocated on every request. Or 1mb every 32 requests.
To look at raw speed:
```ruby
require 'benchmark/ips'
number_of_objects_reduced = 1_114
Benchmark.ips do |x|
x.report("freeze") { number_of_objects_reduced.times { " ".freeze } }
x.report("no-freeze") { number_of_objects_reduced.times { " " } }
end
```
We get the results
```
Calculating -------------------------------------
freeze 1.428k i/100ms
no-freeze 609.000 i/100ms
-------------------------------------------------
freeze 14.363k (± 8.5%) i/s - 71.400k
no-freeze 6.084k (± 8.1%) i/s - 30.450k
```
Now we can do some maths:
```ruby
ips = 6_226k # iterations / 1 second
call_time_before = 1.0 / ips # seconds per iteration
ips = 15_254 # iterations / 1 second
call_time_after = 1.0 / ips # seconds per iteration
diff = call_time_before - call_time_after
number_of_objects_reduced * diff * 100
# => 0.4530373333993266 miliseconds saved per request
```
So we're shaving off 1 second of execution time for every 220 requests.
Is this going to be an insane speed boost to any Rails app: nope. Should we merge it: yep.
p.s. If you know of a method call that doesn't modify a string input such as [String#gsub](b0e2da69f0/lib/let_it_go/core_ext/string.rb (L37)) please [give me a pull request to the appropriate file](b0e2da69f0/lib/let_it_go/core_ext/string.rb (L37)), or open an issue in LetItGo so we can track and freeze more strings.
Keep those strings Frozen
This fixes the issue where you may be comparing (using a numeric
validator such as `greater_than`) numbers of a specific Numeric type
such as `BigDecimal`.
Previous behavior took the numeric value to be validated and
unconditionally converted to Float. For example, due to floating point
precision, this can cause issues when comparing a Float to a BigDecimal.
Consider the following:
```
validates :sub_total, numericality: {
greater_than: BigDecimal('97.18')
}
```
If the `:sub_total` value BigDecimal.new('97.18') was validated against
the above, the following would be valid since `:sub_total` is converted
to a Float regardless of its original type. The result therefore becomes
Kernel.Float(97.18) > BigDecimal.new('97.18')
The above illustrated behavior is corrected with this patch by
conditionally converting the value to validate to float.
Use the post-type-cast version of the attribute to validate numericality
[Roque Pinel & Trevor Wistaff]
[ci skip]
Closes#20792.
Custom validation methods are implemented in terms of
callbacks. The `validate` callback chain can't be halted using return
values of individual callbacks.
[ci skip]
While this :nodoc: did hide the constant it also removed the following
methods from the API docs:
- #attribute_method?
- #clear_validators!
- #validate
- #validators
- #validators_on
Those are public API and should be visible.
Issue was caused by dee4fbc
/cc @zzak
Only one constraint option can be used at a time (except for the minimum
and maximum ones that can eventually be combined). However, other
options can be used with them (e.g. the validation failure message).
So let's make the distinction between these two different options
categories.
[Yves Senn, Matthew Draper & Robin Dupret]
`ActiveModel::Dirty#[attr_name]_previous_change` to improve access
to recorded changes after the model has been saved.
It makes the dirty-attributes query methods consistent before and after
saving.
As demonstrated by #19570, this option is severely limited, and
satisfies an extremely specific use case. Realistically, there's not
much reason for this option to exist. Its functionality can be trivially
replicated with a normal Ruby method. Let's deprecate this option, in
favor of the simpler solution.
Rails 5.0 changes to ActiveModel::Errors include addition of `details`
that also accidentally changed the return value of `delete`. Since
there was no test for that behavior it went unnoticed. This commit
adds a test and fixes the regression.
Small improvements to comments have also been made. Since `get` is
getting deprecated it is better to use `[]` in other methods' code
examples. Also, in the module usage example, `def Person.method`
was replaced with a more commonly used `def self.method` code style.
Deprecation messages in ActiveModel::Errors are using String#squish
from ActiveSupport but were not explicitly requiring it, causing failures
when used outside rails.
The name `ActiveModel::AttributeAssignment::UnknownAttributeError` is
too implementation specific so let's move the constant directly under
the ActiveModel namespace.
Also since this constant used to be under the ActiveRecord namespace, to
make the upgrade path easier, let's avoid raising the former constant
when we deal with this error on the Active Record side.
Previously, calling `User.model_name.to_json` would result in an infinite
recursion as `.model_name` inherited its `.as_json` behavior from Object. This
patch fixes that unexpected behavior by delegating `.as_json` to :name.
The existing example seems somewhat forced: is it realistic
to have a model that accepts state in its initializer but
considers it has not been changed? By allowing state changes
to happen only via accessors it seems more natural that new
instances are considered to be unchanged (as they are in AR).
[ci skip]
Fixes#17621. This 5 year old (or older) issue causes validations to fire
when a parent record has `validate: false` option and a child record is
saved. It's not the responsibility of the model to validate an
associated object unless the object was created or modified by the
parent.
Clean up tests related to validations
`assert_nothing_raised` is not benefiting us in these tests
Corrected spelling of "respects"
It's better to use `assert_not_operator` over `assert !r.valid`
To be able to return type of validator, one can now call `details`
on Errors instance:
```ruby
class User < ActiveRecord::Base
validates :name, presence: true
end
```
```ruby
user = User.new; user.valid?; user.errors.details
=> {name: [{error: :blank}]}
```
When an attribute is assigned, we determine if it was already marked as
changed so we can determine if we need to clear the changes, or mark it
as changed. Since this only affects the `attributes_changed_by_setter`
hash, in-place changes are irrelevant to this process. Since calculating
in-place changes can be expensive, we can just skip it here.
I also added a test for the only edge case I could think of that would
be affected by this change.
This commit changes the original documentation of ActiveModel::Lint::Tests
introduced in dbf20c2d to focus less on *why* the tests exist and more on
*what* the tests do.
For instance, `test_to_key` was documented as:
> Returns an Enumerable of all (primary) key attributes...
whereas `test_to_key` is simply a test meant to *fail* or *pass*, and the
documentation above refers to `to_key`.
[ci skip]
Introduce explicit way of halting callback chains by throwing :abort. Deprecate current implicit behavior of halting callback chains by returning `false` in apps ported to Rails 5.0. Completely remove that behavior in brand new Rails 5.0 apps.
Conflicts:
railties/CHANGELOG.md
Before this commit, returning `false` in an ActiveModel `before_` callback
such as `before_create` would halt the callback chain.
After this commit, the behavior is deprecated: will still work until
the next release of Rails but will also display a deprecation warning.
The preferred way to halt a callback chain is to explicitly `throw(:abort)`.
Before this commit, returning `false` in an ActiveModel validation
callback such as `before_validation` would halt the callback chain.
After this commit, the behavior is deprecated: will still work until
the next release of Rails but will also display a deprecation warning.
The preferred way to halt a callback chain is to explicitly `throw(:abort)`.
This commit changes arguments and default value of CallbackChain's :terminator
option.
After this commit, Chains of callbacks defined **without** an explicit
`:terminator` option will be halted as soon as a `before_` callback throws
`:abort`.
Chains of callbacks defined **with** a `:terminator` option will maintain their
existing behavior of halting as soon as a `before_` callback matches the
terminator's expectation. For instance, ActiveModel's callbacks will still
halt the chain when a `before_` callback returns `false`.
Calling `changed_attributes` will ultimately check if every mutable
attribute has changed in place. Since this gets called whenever an
attribute is assigned, it's extremely slow. Instead, we can avoid this
calculation until we actually need it.
Fixes#18029
The default value for the argument `message` in
`ActiveModel::Errors#add` has a new behavior
since ca99ab2481d44d67bc392d0ec1125ff1439e9f94.
Before
person.errors.add(:name, nil)
# => ["is invalid"]
After
person.errors.add(:name, nil)
# => [nil]
I'm not sure what's the use case for this, but apparently it broke some apps.
Since it was not the intended result from #16210 I fixed it to not raise an
exception anymore. However, I didn't add documentation for it because I don't
know if this should be officially supported without knowing how it's meant to
be used.
In general, validations should be side-effect-free (other than adding to the
error message to `@errors`). Order-dependent validations seems like a bad idea.
Fixes#18002
Active Record defines `attribute_method_suffix :?`. That suffix will
match any predicate method when the lookup occurs in Active Model. This
will make it incorrectly decide that `id_changed?` should not exist,
because it attempts to determine if the attribute `id_changed` is
present, rather than `id` with the `_changed?` suffix. Instead, we will
look for any correct match.
The detection of in-place changes caused a weird unexpected issue with
numericality validations. That validator (out of necessity) works on the
`_before_type_cast` version of the attribute, since on an `:integer`
type column, a non-numeric string would type cast to 0.
However, strings are mutable, and we changed strings to ensure that the
post type cast version of the attribute was a different instance than
the before type cast version (so the mutation detection can work
properly).
Even though strings are the only mutable type for which a numericality
validation makes sense, special casing strings would feel like a strange
change to make here. Instead, we can make the assumption that for all
mutable types, we should work on the post-type-cast version of the
attribute, since all cases which would return 0 for non-numeric strings
are immutable.
Fixes#17852
