Specs

add_error(t(), atom(), String.t(), Keyword.t()) :: t()

Adds an error to the changeset.

An additional keyword list keys can be passed to provide additional contextual information for the error. This is useful when using traverse_errors/2

Examples

iex> changeset = change(%Post{}, %{title: ""})
iex> changeset = add_error(changeset, :title, "empty")
iex> changeset.errors
[title: {"empty", []}]
iex> changeset.valid?
false

iex> changeset = change(%Post{}, %{title: ""})
iex> changeset = add_error(changeset, :title, "empty", additional: "info")
iex> changeset.errors
[title: {"empty", [additional: "info"]}]
iex> changeset.valid?
false

Specs

apply_action(t(), atom()) :: {:ok, Ecto.Schema.t() | data()} | {:error, t()}

Applies the changeset action only if the changes are valid.

If the changes are valid, all changes are applied to the changeset data. If the changes are invalid, no changes are applied, and an error tuple is returned with the changeset containing the action that was attempted to be applied.

The action may be any atom.

Examples

iex> {:ok, data} = apply_action(changeset, :update)

iex> {:error, changeset} = apply_action(changeset, :update)
%Ecto.Changeset{action: :update}

Specs

apply_action!(t(), atom()) :: Ecto.Schema.t() | data()

Applies the changeset action if the changes are valid or raises an error.

Examples

iex> changeset = change(%Post{author: "bar"}, %{title: "foo"})
iex> apply_action!(changeset, :update)
%Post{author: "bar", title: "foo"}

iex> changeset = change(%Post{author: "bar"}, %{title: :bad})
iex> apply_action!(changeset, :update)
** (Ecto.InvalidChangesetError) could not perform update because changeset is invalid.

See apply_action/2 for more information.

Specs

apply_changes(t()) :: Ecto.Schema.t() | data()

Applies the changeset changes to the changeset data.

This operation will return the underlying data with changes regardless if the changeset is valid or not.

Examples

iex> changeset = change(%Post{author: "bar"}, %{title: "foo"})
iex> apply_changes(changeset)
%Post{author: "bar", title: "foo"}

Specs

assoc_constraint(t(), atom(), Keyword.t()) :: t()

Checks the associated field exists.

This is similar to foreign_key_constraint/3 except that the field is inferred from the association definition. This is useful to guarantee that a child will only be created if the parent exists in the database too. Therefore, it only applies to belongs_to associations.

As the name says, a constraint is required in the database for this function to work. Such constraint is often added as a reference to the child table:

create table(:comments) do
  add :post_id, references(:posts)
end

Now, when inserting a comment, it is possible to forbid any comment to be added if the associated post does not exist:

comment
|> Ecto.Changeset.cast(params, [:post_id])
|> Ecto.Changeset.assoc_constraint(:post)
|> Repo.insert

Options

• :message - the message in case the constraint check fails, defaults to "does not exist"
• :name - the constraint name. By default, the constraint name is inferred from the table + association field. May be required explicitly for complex cases

Specs

cast(
  Ecto.Schema.t() | t() | {data(), types()},
  %{required(binary()) => term()} | %{required(atom()) => term()} | :invalid,
  [atom()],
  Keyword.t()
) :: t()

Applies the given params as changes for the given data according to the given set of permitted keys. Returns a changeset.

The given data may be either a changeset, a schema struct or a {data, types} tuple. The second argument is a map of params that are cast according to the type information from data. params is a map with string keys or a map with atom keys containing potentially invalid data.

During casting, all permitted parameters whose values match the specified type information will have their key name converted to an atom and stored together with the value as a change in the :changes field of the changeset. All parameters that are not explicitly permitted are ignored.

If casting of all fields is successful, the changeset is returned as valid.

Note that cast/4 validates the types in the params, but not in the given data.

Options

• :empty_values - a list of values to be considered as empty when casting. All empty values are discarded on cast. Defaults to [""]

Examples

iex> changeset = cast(post, params, [:title])
iex> if changeset.valid? do
...>   Repo.update!(changeset)
...> end

Passing a changeset as the first argument:

iex> changeset = cast(post, %{title: "Hello"}, [:title])
iex> new_changeset = cast(changeset, %{title: "Foo", body: "World"}, [:body])
iex> new_changeset.params
%{"title" => "Hello", "body" => "World"}

Or creating a changeset from a simple map with types:

iex> data = %{title: "hello"}
iex> types = %{title: :string}
iex> changeset = cast({data, types}, %{title: "world"}, [:title])
iex> apply_changes(changeset)
%{title: "world"}

Composing casts

cast/4 also accepts a changeset as its first argument. In such cases, all the effects caused by the call to cast/4 (additional errors and changes) are simply added to the ones already present in the argument changeset. Parameters are merged (not deep-merged) and the ones passed to cast/4 take precedence over the ones already in the changeset.

Casts the given association with the changeset parameters.

This function should be used when working with the entire association at once (and not a single element of a many-style association) and receiving data external to the application.

cast_assoc/3 works matching the records extracted from the database and compares it with the parameters received from an external source. Therefore, it is expected that the data in the changeset has explicitly preloaded the association being cast and that all of the IDs exist and are unique.

For example, imagine a user has many addresses relationship where post data is sent as follows

%{"name" => "john doe", "addresses" => [
  %{"street" => "somewhere", "country" => "brazil", "id" => 1},
  %{"street" => "elsewhere", "country" => "poland"},
]}

and then

User
|> Repo.get!(id)
|> Repo.preload(:addresses) # Only required when updating data
|> Ecto.Changeset.cast(params, [])
|> Ecto.Changeset.cast_assoc(:addresses, with: &MyApp.Address.changeset/2)

The parameters for the given association will be retrieved from changeset.params. Those parameters are expected to be a map with attributes, similar to the ones passed to cast/4. Once parameters are retrieved, cast_assoc/3 will match those parameters with the associations already in the changeset record.

Once cast_assoc/3 is called, Ecto will compare each parameter with the user's already preloaded addresses and act as follows:

• If the parameter does not contain an ID, the parameter data will be passed to MyApp.Address.changeset/2 with a new struct and become an insert operation
• If the parameter contains an ID and there is no associated child with such ID, the parameter data will be passed to MyApp.Address.changeset/2 with a new struct and become an insert operation
• If the parameter contains an ID and there is an associated child with such ID, the parameter data will be passed to MyApp.Address.changeset/2 with the existing struct and become an update operation
• If there is an associated child with an ID and its ID is not given as parameter, the :on_replace callback for that association will be invoked (see the "On replace" section on the module documentation)

Every time the MyApp.Address.changeset/2 function is invoked, it must return a changeset. Once the parent changeset is given to an Ecto.Repo function, all entries will be inserted/updated/deleted within the same transaction.

Note developers are allowed to explicitly set the :action field of a changeset to instruct Ecto how to act in certain situations. Let's suppose that, if one of the associations has only empty fields, you want to ignore the entry altogether instead of showing an error. The changeset function could be written like this:

def changeset(struct, params) do
  struct
  |> cast(params, [:title, :body])
  |> validate_required([:title, :body])
  |> case do
    %{valid?: false, changes: changes} = changeset when changes == %{} ->
      # If the changeset is invalid and has no changes, it is
      # because all required fields are missing, so we ignore it.
      %{changeset | action: :ignore}
    changeset ->
      changeset
  end
end

Partial changes for many-style associations

By preloading an association using a custom query you can confine the behavior of cast_assoc/3. This opens up the possibility to work on a subset of the data, instead of all associations in the database.

Taking the initial example of users having addresses imagine those addresses are set up to belong to a country. If you want to allow users to bulk edit all addresses that belong to a single country, you can do so by changing the preload query:

query = from MyApp.Address, where: [country: ^edit_country]

User
|> Repo.get!(id)
|> Repo.preload(addresses: query)
|> Ecto.Changeset.cast(params, [])
|> Ecto.Changeset.cast_assoc(:addresses)

This will allow you to cast and update only the association for the given country. The important point for partial changes is that any addresses, which were not preloaded won't be changed.

Options

• :required - if the association is a required field
• :required_message - the message on failure, defaults to "can't be blank"
• :invalid_message - the message on failure, defaults to "is invalid"
• :force_update_on_change - force the parent record to be updated in the repository if there is a change, defaults to true
• :with - the function to build the changeset from params. Defaults to the changeset/2 function of the associated module. It can be changed by passing an anonymous function or an MFA tuple. If using an MFA, the default changeset and parameters arguments will be prepended to the given args. For example, using with: {Author, :special_changeset, ["hello"]} will be invoked as Author.special_changeset(changeset, params, "hello")

Casts the given embed with the changeset parameters.

The parameters for the given embed will be retrieved from changeset.params. Those parameters are expected to be a map with attributes, similar to the ones passed to cast/4. Once parameters are retrieved, cast_embed/3 will match those parameters with the embeds already in the changeset record. See cast_assoc/3 for an example of working with casts and associations which would also apply for embeds.

The changeset must have been previously cast using cast/4 before this function is invoked.

Options

• :required - if the embed is a required field
• :required_message - the message on failure, defaults to "can't be blank"
• :invalid_message - the message on failure, defaults to "is invalid"
• :force_update_on_change - force the parent record to be updated in the repository if there is a change, defaults to true
• :with - the function to build the changeset from params. Defaults to the changeset/2 function of the embedded module. It can be changed by passing an anonymous function or an MFA tuple. If using an MFA, the default changeset and parameters arguments will be prepended to the given args. For example, using with: {Author, :special_changeset, ["hello"]} will be invoked as Author.special_changeset(changeset, params, "hello")

Specs

change(
  Ecto.Schema.t() | t() | {data(), types()},
  %{required(atom()) => term()} | Keyword.t()
) :: t()

Wraps the given data in a changeset or adds changes to a changeset.

changes is a map or keyword where the key is an atom representing a field, association or embed and the value is a term. Note the value is directly stored in the changeset with no validation whatsoever. For this reason, this function is meant for working with data internal to the application.

When changing embeds and associations, see put_assoc/4 for a complete reference on the accepted values.

This function is useful for:

• wrapping a struct inside a changeset
• directly changing a struct without performing castings nor validations
• directly bulk-adding changes to a changeset

Changed attributes will only be added if the change does not have the same value as the field in the data.

When a changeset is passed as the first argument, the changes passed as the second argument are merged over the changes already in the changeset if they differ from the values in the struct.

When a {data, types} is passed as the first argument, a changeset is created with the given data and types and marked as valid.

See cast/4 if you'd prefer to cast and validate external parameters.

Examples

iex> changeset = change(%Post{})
%Ecto.Changeset{...}
iex> changeset.valid?
true
iex> changeset.changes
%{}

iex> changeset = change(%Post{author: "bar"}, title: "title")
iex> changeset.changes
%{title: "title"}

iex> changeset = change(%Post{title: "title"}, title: "title")
iex> changeset.changes
%{}

iex> changeset = change(changeset, %{title: "new title", body: "body"})
iex> changeset.changes.title
"new title"
iex> changeset.changes.body
"body"

Checks for a check constraint in the given field.

The check constraint works by relying on the database to check if the check constraint has been violated or not and, if so, Ecto converts it into a changeset error.

In order to use the check constraint, the first step is to define the check constraint in a migration:

create constraint("users", :price_must_be_positive, check: "price > 0")

Now that a constraint exists, when modifying users, we could annotate the changeset with a check constraint so Ecto knows how to convert it into an error message:

cast(user, params, [:price])
|> check_constraint(:price, name: :price_must_be_positive)

Now, when invoking Repo.insert/2 or Repo.update/2, if the price is not positive, it will be converted into an error and {:error, changeset} returned by the repository. Note that the error will occur only after hitting the database so it will not be visible until all other validations pass.

Options

• :message - the message in case the constraint check fails. Defaults to "is invalid"
• :name - the name of the constraint. Required.
• :match - how the changeset constraint name is matched against the repo constraint, may be :exact, :suffix or :prefix. Defaults to :exact. :suffix matches any repo constraint which ends_with? :name to this changeset constraint. :prefix matches any repo constraint which starts_with? :name to this changeset constraint.

Specs

constraints(t()) :: [constraint()]

Returns all constraints in a changeset.

A constraint is a map with the following fields:

• :type - the type of the constraint that will be checked in the database, such as :check, :unique, etc
• :constraint - the database constraint name as a string
• :match - the type of match Ecto will perform on a violated constraint against the :constraint value. It is :exact, :suffix or :prefix
• :field - the field a violated constraint will apply the error to
• :error_message - the error message in case of violated constraints
• :error_type - the type of error that identifies the error message

Specs

delete_change(t(), atom()) :: t()

Deletes a change with the given key.

Examples

iex> changeset = change(%Post{}, %{title: "foo"})
iex> changeset = delete_change(changeset, :title)
iex> get_change(changeset, :title)
nil

Checks for an exclusion constraint in the given field.

The exclusion constraint works by relying on the database to check if the exclusion constraint has been violated or not and, if so, Ecto converts it into a changeset error.

Options

• :message - the message in case the constraint check fails, defaults to "violates an exclusion constraint"
• :name - the constraint name. By default, the constraint name is inferred from the table + field. May be required explicitly for complex cases
• :match - how the changeset constraint name is matched against the repo constraint, may be :exact or :suffix. Defaults to :exact. :suffix matches any repo constraint which ends_with? :name to this changeset constraint.

Specs

fetch_change(t(), atom()) :: {:ok, term()} | :error

Fetches a change from the given changeset.

This function only looks at the :changes field of the given changeset and returns {:ok, value} if the change is present or :error if it's not.

Examples

iex> changeset = change(%Post{body: "foo"}, %{title: "bar"})
iex> fetch_change(changeset, :title)
{:ok, "bar"}
iex> fetch_change(changeset, :body)
:error

Specs

fetch_change!(t(), atom()) :: term()

Same as fetch_change/2 but returns the value or raises if the given key was not found.

Examples

iex> changeset = change(%Post{body: "foo"}, %{title: "bar"})
iex> fetch_change!(changeset, :title)
"bar"
iex> fetch_change!(changeset, :body)
** (KeyError) key :body not found in: %{title: "bar"}

Specs

fetch_field(t(), atom()) :: {:changes, term()} | {:data, term()} | :error

Fetches the given field from changes or from the data.

While fetch_change/2 only looks at the current changes to retrieve a value, this function looks at the changes and then falls back on the data, finally returning :error if no value is available.

For relations, these functions will return the changeset original data with changes applied. To retrieve raw changesets, please use fetch_change/2.

Examples

iex> post = %Post{title: "Foo", body: "Bar baz bong"}
iex> changeset = change(post, %{title: "New title"})
iex> fetch_field(changeset, :title)
{:changes, "New title"}
iex> fetch_field(changeset, :body)
{:data, "Bar baz bong"}
iex> fetch_field(changeset, :not_a_field)
:error

Specs

fetch_field!(t(), atom()) :: term()

Same as fetch_field/2 but returns the value or raises if the given key was not found.

Examples

iex> post = %Post{title: "Foo", body: "Bar baz bong"}
iex> changeset = change(post, %{title: "New title"})
iex> fetch_field!(changeset, :title)
"New title"
iex> fetch_field!(changeset, :other)
** (KeyError) key :other not found in: %Post{...}

Specs

force_change(t(), atom(), term()) :: t()

Forces a change on the given key with value.

If the change is already present, it is overridden with the new value.

Examples

iex> changeset = change(%Post{author: "bar"}, %{title: "foo"})
iex> changeset = force_change(changeset, :title, "bar")
iex> changeset.changes
%{title: "bar"}

iex> changeset = force_change(changeset, :author, "bar")
iex> changeset.changes
%{title: "bar", author: "bar"}

Specs

foreign_key_constraint(t(), atom(), Keyword.t()) :: t()

Checks for foreign key constraint in the given field.

The foreign key constraint works by relying on the database to check if the associated data exists or not. This is useful to guarantee that a child will only be created if the parent exists in the database too.

In order to use the foreign key constraint the first step is to define the foreign key in a migration. This is often done with references. For example, imagine you are creating a comments table that belongs to posts. One would have:

create table(:comments) do
  add :post_id, references(:posts)
end

By default, Ecto will generate a foreign key constraint with name "comments_post_id_fkey" (the name is configurable).

Now that a constraint exists, when creating comments, we could annotate the changeset with foreign key constraint so Ecto knows how to convert it into an error message:

cast(comment, params, [:post_id])
|> foreign_key_constraint(:post_id)

Now, when invoking Repo.insert/2 or Repo.update/2, if the associated post does not exist, it will be converted into an error and {:error, changeset} returned by the repository.

Options

• :message - the message in case the constraint check fails, defaults to "does not exist"
• :name - the constraint name. By default, the constraint name is inferred from the table + field. May be required explicitly for complex cases

Specs

get_change(t(), atom(), term()) :: term()

Gets a change or returns a default value.

Examples

iex> changeset = change(%Post{body: "foo"}, %{title: "bar"})
iex> get_change(changeset, :title)
"bar"
iex> get_change(changeset, :body)
nil

Specs

get_field(t(), atom(), term()) :: term()

Gets a field from changes or from the data.

While get_change/3 only looks at the current changes to retrieve a value, this function looks at the changes and then falls back on the data, finally returning default if no value is available.

For relations, these functions will return the changeset data with changes applied. To retrieve raw changesets, please use get_change/3.

iex> post = %Post{title: "A title", body: "My body is a cage"}
iex> changeset = change(post, %{title: "A new title"})
iex> get_field(changeset, :title)
"A new title"
iex> get_field(changeset, :not_a_field, "Told you, not a field!")
"Told you, not a field!"

Specs

merge(t(), t()) :: t()

Merges two changesets.

This function merges two changesets provided they have been applied to the same data (their :data field is equal); if the data differs, an ArgumentError exception is raised. If one of the changesets has a :repo field which is not nil, then the value of that field is used as the :repo field of the resulting changeset; if both changesets have a non-nil and different :repo field, an ArgumentError exception is raised.

The other fields are merged with the following criteria:

• params - params are merged (not deep-merged) giving precedence to the params of changeset2 in case of a conflict. If both changesets have their :params fields set to nil, the resulting changeset will have its params set to nil too.
• changes - changes are merged giving precedence to the changeset2 changes.
• errors and validations - they are simply concatenated.
• required - required fields are merged; all the fields that appear in the required list of both changesets are moved to the required list of the resulting changeset.

Examples

iex> changeset1 = cast(%Post{}, %{title: "Title"}, [:title])
iex> changeset2 = cast(%Post{}, %{title: "New title", body: "Body"}, [:title, :body])
iex> changeset = merge(changeset1, changeset2)
iex> changeset.changes
%{body: "Body", title: "New title"}

iex> changeset1 = cast(%Post{body: "Body"}, %{title: "Title"}, [:title])
iex> changeset2 = cast(%Post{}, %{title: "New title"}, [:title])
iex> merge(changeset1, changeset2)
** (ArgumentError) different :data when merging changesets

Specs

no_assoc_constraint(t(), atom(), Keyword.t()) :: t()

Checks the associated field does not exist.

This is similar to foreign_key_constraint/3 except that the field is inferred from the association definition. This is useful to guarantee that parent can only be deleted (or have its primary key changed) if no child exists in the database. Therefore, it only applies to has_* associations.

As the name says, a constraint is required in the database for this function to work. Such constraint is often added as a reference to the child table:

create table(:comments) do
  add :post_id, references(:posts)
end

Now, when deleting the post, it is possible to forbid any post to be deleted if they still have comments attached to it:

post
|> Ecto.Changeset.change
|> Ecto.Changeset.no_assoc_constraint(:comments)
|> Repo.delete

Options

• :message - the message in case the constraint check fails, defaults to "is still associated with this entry" (for has_one) and "are still associated with this entry" (for has_many)
• :name - the constraint name. By default, the constraint name is inferred from the association table + association field. May be required explicitly for complex cases

Specs

optimistic_lock(Ecto.Schema.t() | t(), atom(), (term() -> term())) :: t()

Applies optimistic locking to the changeset.

Optimistic locking (or optimistic concurrency control) is a technique that allows concurrent edits on a single record. While pessimistic locking works by locking a resource for an entire transaction, optimistic locking only checks if the resource changed before updating it.

This is done by regularly fetching the record from the database, then checking whether another user has made changes to the record only when updating the record. This behaviour is ideal in situations where the chances of concurrent updates to the same record are low; if they're not, pessimistic locking or other concurrency patterns may be more suited.

Usage

Optimistic locking works by keeping a "version" counter for each record; this counter gets incremented each time a modification is made to a record. Hence, in order to use optimistic locking, a field must exist in your schema for versioning purpose. Such field is usually an integer but other types are supported.

Examples

Assuming we have a Post schema (stored in the posts table), the first step is to add a version column to the posts table:

alter table(:posts) do
  add :lock_version, :integer, default: 1
end

The column name is arbitrary and doesn't need to be :lock_version. Now add a field to the schema too:

defmodule Post do
  use Ecto.Schema

  schema "posts" do
    field :title, :string
    field :lock_version, :integer, default: 1
  end

  def changeset(:update, struct, params \\ %{}) do
    struct
    |> Ecto.Changeset.cast(params, [:title])
    |> Ecto.Changeset.optimistic_lock(:lock_version)
  end
end

Now let's take optimistic locking for a spin:

iex> post = Repo.insert!(%Post{title: "foo"})
%Post{id: 1, title: "foo", lock_version: 1}
iex> valid_change = Post.changeset(:update, post, %{title: "bar"})
iex> stale_change = Post.changeset(:update, post, %{title: "baz"})
iex> Repo.update!(valid_change)
%Post{id: 1, title: "bar", lock_version: 2}
iex> Repo.update!(stale_change)
** (Ecto.StaleEntryError) attempted to update a stale entry:

%Post{id: 1, title: "baz", lock_version: 1}

When a conflict happens (a record which has been previously fetched is being updated, but that same record has been modified since it was fetched), an Ecto.StaleEntryError exception is raised.

Optimistic locking also works with delete operations. Just call the optimistic_lock/3 function with the data before delete:

iex> changeset = Ecto.Changeset.optimistic_lock(post, :lock_version)
iex> Repo.delete(changeset)

optimistic_lock/3 by default assumes the field being used as a lock is an integer. If you want to use another type, you need to pass the third argument customizing how the next value is generated:

iex> Ecto.Changeset.optimistic_lock(post, :lock_uuid, fn _ -> Ecto.UUID.generate end)

Specs

prepare_changes(t(), (t() -> t())) :: t()

Provides a function executed by the repository on insert/update/delete.

If the changeset given to the repository is valid, the function given to prepare_changes/2 will be called with the changeset and must return a changeset, allowing developers to do final adjustments to the changeset or to issue data consistency commands. The repository itself can be accessed inside the function under the repo field in the changeset. If the changeset given to the repository is invalid, the function will not be invoked.

The given function is guaranteed to run inside the same transaction as the changeset operation for databases that do support transactions.

Example

A common use case is updating a counter cache, in this case updating a post's comment count when a comment is created:

def create_comment(comment, params) do
  comment
  |> cast(params, [:body, :post_id])
  |> prepare_changes(fn changeset ->
       if post_id = get_change(changeset, :post_id) do
         query = from Post, where: [id: ^post_id]
         changeset.repo.update_all(query, inc: [comment_count: 1])
       end
       changeset
     end)
end

We retrieve the repo from the comment changeset itself and use update_all to update the counter cache in one query. Finally, the original changeset must be returned.

Puts the given association entry or entries as a change in the changeset.

This function is used to work with associations as a whole. For example, if a Post has many Comments, it allows you to add, remove or change all comments at once. If your goal is to simply add a new comment to a post, then it is preferred to do so manually, as we will describe later in the "Example: Adding a comment to a post" section.

This function requires the associated data to have been preloaded, except when the parent changeset has been newly built and not yet persisted. Missing data will invoke the :on_replace behaviour defined on the association.

For associations with cardinality one, nil can be used to remove the existing entry. For associations with many entries, an empty list may be given instead.

If the association has no changes, it will be skipped. If the association is invalid, the changeset will be marked as invalid. If the given value is not any of values below, it will raise.

The associated data may be given in different formats:

• a map or a keyword list representing changes to be applied to the associated data. A map or keyword list can be given to update the associated data as long as they have matching primary keys. For example, put_assoc(changeset, :comments, [%{id: 1, title: "changed"}]) will locate the comment with :id of 1 and update its title. If no comment with such id exists, one is created on the fly. Since only a single comment was given, any other associated comment will be replaced. On all cases, it is expected the keys to be atoms. This API is mostly used in scripts and tests, to make it straight- forward to create schemas with associations at once, such as:

  Ecto.Changeset.change(
    %Post{},
    title: "foo",
    comments: [
      %{body: "first"},
      %{body: "second"}
    ]
  )

• changesets or structs - when a changeset or struct is given, they are treated as the canonical data and the associated data currently stored in the association is ignored. For instance, the operation put_assoc(changeset, :comments, [%Comment{id: 1, title: "changed"}]) will send the Comment as is to the database, ignoring any comment currently associated, even if a matching ID is found. If the comment is already persisted to the database, then put_assoc/4 only takes care of guaranteeing that the comments and the parent data are associated. This extremely useful when associating existing data, as we will see in the "Example: Adding tags to a post" section.

Once the parent changeset is given to an Ecto.Repo function, all entries will be inserted/updated/deleted within the same transaction.

Example: Adding a comment to a post

Imagine a relationship where Post has many comments and you want to add a new comment to an existing post. While it is possible to use put_assoc/4 for this, it would be unnecessarily complex. Let's see an example.

First, let's fetch the post with all existing comments:

post = Post |> Repo.get!(1) |> Repo.preload(:comments)

The following approach is wrong:

post
|> Ecto.Changeset.change()
|> Ecto.Changeset.put_assoc(:comments, [%Comment{body: "bad example!"}])
|> Repo.update!()

The reason why the example above is wrong is because put_assoc/4 always works with the full data. So the example above will effectively erase all previous comments and only keep the comment you are currently adding. Instead, you could try:

post
|> Ecto.Changeset.change()
|> Ecto.Changeset.put_assoc(:comments, [%Comment{body: "so-so example!"} | post.comments])
|> Repo.update!()

In this example, we prepend the new comment to the list of existing comments. Ecto will diff the list of comments currently in post with the list of comments given, and correctly insert the new comment to the database. Note, however, Ecto is doing a lot of work just to figure out something we knew since the beginning, which is that there is only one new comment.

In cases like above, when you want to work only on a single entry, it is much easier to simply work on the associated directly. For example, we could instead set the post association in the comment:

%Comment{body: "better example"}
|> Ecto.Changeset.change()
|> Ecto.Changeset.put_assoc(:post, post)
|> Repo.insert!()

Alternatively, we can make sure that when we create a comment, it is already associated to the post:

Ecto.build_assoc(post, :comments)
|> Ecto.Changeset.change(body: "great example!")
|> Repo.insert!()

Or we can simply set the post_id in the comment itself:

%Comment{body: "better example", post_id: post.id}
|> Repo.insert!()

In other words, when you find yourself wanting to work only with a subset of the data, then using put_assoc/4 is most likely unnecessary. Instead, you want to work on the other side of the association.

Let's see an example where using put_assoc/4 is a good fit.

Example: Adding tags to a post

Imagine you are receiving a set of tags you want to associate to a post. Let's imagine that those tags exist upfront and are all persisted to the database. Imagine we get the data in this format:

params = %{"title" => "new post", "tags" => ["learner"]}

Now, since the tags already exist, we will bring all of them from the database and put them directly in the post:

tags = Repo.all(from t in Tag, where: t.name in ^params["tags"])

post
|> Repo.preload(:tags)
|> Ecto.Changeset.cast(params, [:title]) # No need to allow :tags as we put them directly
|> Ecto.Changeset.put_assoc(:tags, tags) # Explicitly set the tags

Since in this case we always require the user to pass all tags directly, using put_assoc/4 is a great fit. It will automatically remove any tag not given and properly associate all of the given tags with the post.

Furthermore, since the tag information is given as structs read directly from the database, Ecto will treat the data as correct and only do the minimum necessary to guarantee that posts and tags are associated, without trying to update or diff any of the fields in the tag struct.

Although it accepts an opts argument, there are no options currently supported by put_assoc/4.

Specs

put_change(t(), atom(), term()) :: t()

Puts a change on the given key with value.

key is an atom that represents any field, embed or association in the changeset. Note the value is directly stored in the changeset with no validation whatsoever. For this reason, this function is meant for working with data internal to the application.

If the change is already present, it is overridden with the new value. If the change has the same value as in the changeset data, it is not added to the list of changes.

When changing embeds and associations, see put_assoc/4 for a complete reference on the accepted values.

Examples

iex> changeset = change(%Post{author: "bar"}, %{title: "foo"})
iex> changeset = put_change(changeset, :title, "bar")
iex> changeset.changes
%{title: "bar"}

iex> changeset = put_change(changeset, :author, "bar")
iex> changeset.changes
%{title: "bar"}

Puts the given embed entry or entries as a change in the changeset.

This function is used to work with embeds as a whole. For embeds with cardinality one, nil can be used to remove the existing entry. For embeds with many entries, an empty list may be given instead.

If the embed has no changes, it will be skipped. If the embed is invalid, the changeset will be marked as invalid.

The list of supported values and their behaviour is described in put_assoc/4. If the given value is not any of values listed there, it will raise.

Although this function accepts an opts argument, there are no options currently supported by put_embed/4.

Specs

traverse_errors(
  t(),
  (error() -> String.t()) | (t(), atom(), error() -> String.t())
) :: %{required(atom()) => [String.t() | map()]}

Traverses changeset errors and applies the given function to error messages.

This function is particularly useful when associations and embeds are cast in the changeset as it will traverse all associations and embeds and place all errors in a series of nested maps.

A changeset is supplied along with a function to apply to each error message as the changeset is traversed. The error message function receives an error tuple {msg, opts}, for example:

{"should be at least %{count} characters", [count: 3, validation: :length, min: 3]}

Examples

iex> traverse_errors(changeset, fn {msg, opts} ->
...>   Enum.reduce(opts, msg, fn {key, value}, acc ->
...>     String.replace(acc, "%{#{key}}", to_string(value))
...>   end)
...> end)
%{title: ["should be at least 3 characters"]}

Optionally function can accept three arguments: changeset, field and error tuple {msg, opts}. It is useful whenever you want to extract validations rules from changeset.validations to build detailed error description.

Specs

unique_constraint(t(), atom() | [atom(), ...], Keyword.t()) :: t()

Checks for a unique constraint in the given field or list of fields.

The unique constraint works by relying on the database to check if the unique constraint has been violated or not and, if so, Ecto converts it into a changeset error.

In order to use the uniqueness constraint, the first step is to define the unique index in a migration:

create unique_index(:users, [:email])

Now that a constraint exists, when modifying users, we could annotate the changeset with a unique constraint so Ecto knows how to convert it into an error message:

cast(user, params, [:email])
|> unique_constraint(:email)

Now, when invoking Repo.insert/2 or Repo.update/2, if the email already exists, it will be converted into an error and {:error, changeset} returned by the repository. Note that the error will occur only after hitting the database so it will not be visible until all other validations pass.

Options

• :message - the message in case the constraint check fails, defaults to "has already been taken"

• :name - the constraint name. By default, the constraint name is inferred from the table + field(s). May be required explicitly for complex cases

• :match - how the changeset constraint name is matched against the repo constraint, may be :exact or :suffix. Defaults to :exact. :suffix matches any repo constraint which ends_with? :name to this changeset constraint.

Complex constraints

Because the constraint logic is in the database, we can leverage all the database functionality when defining them. For example, let's suppose the e-mails are scoped by company id:

# In migration
create unique_index(:users, [:email, :company_id])

# In the changeset function
cast(user, params, [:email])
|> unique_constraint([:email, :company_id])

The first field name, :email in this case, will be used as the error key to the changeset errors keyword list. For example, the above unique_constraint/3 would generate something like:

Repo.insert!(%User{email: "john@elixir.org", company_id: 1})
changeset = User.changeset(%User{}, %{email: "john@elixir.org", company_id: 1})
{:error, changeset} = Repo.insert(changeset)
changeset.errors #=> [email: {"has already been taken", []}]

In complex cases, instead of relying on name inference, it may be best to set the constraint name explicitly:

# In the migration
create unique_index(:users, [:email, :company_id], name: :users_email_company_id_index)

# In the changeset function
cast(user, params, [:email])
|> unique_constraint(:email, name: :users_email_company_id_index)

Partitioning

If your table is partitioned, then your unique index might look different per partition, e.g. Postgres adds p<number> to the middle of your key, like:

users_p0_email_key
users_p1_email_key
...
users_p99_email_key

In this case you can use the name and suffix options together to match on these dynamic indexes, like:

cast(user, params, [:email])
|> unique_constraint(:email, name: :email_key, match: :suffix)

Case sensitivity

Unfortunately, different databases provide different guarantees when it comes to case-sensitiveness. For example, in MySQL, comparisons are case-insensitive by default. In Postgres, users can define case insensitive column by using the :citext type/extension. In your migration:

execute "CREATE EXTENSION IF NOT EXISTS citext"
create table(:users) do
  ...
  add :email, :citext
  ...
end

If for some reason your database does not support case insensitive columns, you can explicitly downcase values before inserting/updating them:

cast(data, params, [:email])
|> update_change(:email, &String.downcase/1)
|> unique_constraint(:email)

Specs

unsafe_validate_unique(t(), atom() | [atom(), ...], Ecto.Repo.t(), Keyword.t()) ::
  t()

Validates that no existing record with a different primary key has the same values for these fields.

This function exists to provide quick feedback to users of your application. It should not be relied on for any data guarantee as it has race conditions and is inherently unsafe. For example, if this check happens twice in the same time interval (because the user submitted a form twice), both checks may pass and you may end-up with duplicate entries in the database. Therefore, a unique_constraint/3 should also be used to ensure your data won't get corrupted.

However, because constraints are only checked if all validations succeed, this function can be used as an early check to provide early feedback to users, since most conflicting data will have been inserted prior to the current validation phase.

Options

• :message - the message in case the constraint check fails, defaults to "has already been taken".

• :match - how the changeset constraint name is matched against the repo constraint, may be :exact or :suffix. Defaults to :exact. :suffix matches any repo constraint which ends_with? :name to this changeset constraint.

• :error_key - the key to which changeset error will be added when check fails, defaults to the first field name of the given list of fields.

• :prefix - The prefix to run the query on (such as the schema path in Postgres or the database in MySQL). See Ecto.Repo documentation for more information.

Examples

unsafe_validate_unique(changeset, :city_name, repo)
unsafe_validate_unique(changeset, [:city_name, :state_name], repo)
unsafe_validate_unique(changeset, [:city_name, :state_name], repo, message: "city must be unique within state")
unsafe_validate_unique(changeset, [:city_name, :state_name], repo, prefix: "public")

Specs

update_change(t(), atom(), (term() -> term())) :: t()

Updates a change.

The given function is invoked with the change value only if there is a change for the given key. Note that the value of the change can still be nil (unless the field was marked as required on validate_required/3).

Examples

iex> changeset = change(%Post{}, %{impressions: 1})
iex> changeset = update_change(changeset, :impressions, &(&1 + 1))
iex> changeset.changes.impressions
2

Specs

validate_acceptance(t(), atom(), Keyword.t()) :: t()

Validates the given parameter is true.

Note this validation only checks the parameter itself is true, never the field in the schema. That's because acceptance parameters do not need to be persisted, as by definition they would always be stored as true.

Options

• :message - the message on failure, defaults to "must be accepted"

Examples

validate_acceptance(changeset, :terms_of_service)
validate_acceptance(changeset, :rules, message: "please accept rules")

Specs

validate_change(
  t(),
  atom(),
  (atom(), term() ->
     [{atom(), String.t()} | {atom(), {String.t(), Keyword.t()}}])
) :: t()

Validates the given field change.

It invokes the validator function to perform the validation only if a change for the given field exists and the change value is not nil. The function must return a list of errors (with an empty list meaning no errors).

In case there's at least one error, the list of errors will be appended to the :errors field of the changeset and the :valid? flag will be set to false.

Examples

iex> changeset = change(%Post{}, %{title: "foo"})
iex> changeset = validate_change changeset, :title, fn :title, title  ->
...>   # Value must not be "foo"!
...>   if title == "foo" do
...>     [title: "cannot be foo"]
...>   else
...>     []
...>   end
...> end
iex> changeset.errors
[title: {"cannot be foo", []}]

Specs

validate_change(
  t(),
  atom(),
  term(),
  (atom(), term() ->
     [{atom(), String.t()} | {atom(), {String.t(), Keyword.t()}}])
) :: t()

Stores the validation metadata and validates the given field change.

Similar to validate_change/3 but stores the validation metadata into the changeset validators. The validator metadata is often used as a reflection mechanism, to automatically generate code based on the available validations.

Examples

iex> changeset = change(%Post{}, %{title: "foo"})
iex> changeset = validate_change changeset, :title, :useless_validator, fn
...>   _, _ -> []
...> end
iex> changeset.validations
[title: :useless_validator]

Specs

validate_confirmation(t(), atom(), Keyword.t()) :: t()

Validates that the given parameter matches its confirmation.

By calling validate_confirmation(changeset, :email), this validation will check if both "email" and "email_confirmation" in the parameter map matches. Note this validation only looks at the parameters themselves, never the fields in the schema. As such as, the "email_confirmation" field does not need to be added as a virtual field in your schema.

Note that if the confirmation field is nil or missing, this does not add a validation error. You can specify that the confirmation parameter is required in the options (see below).

Options

• :message - the message on failure, defaults to "does not match confirmation"
• :required - boolean, sets whether existence of confirmation parameter is required for addition of error. Defaults to false

Examples

validate_confirmation(changeset, :email)
validate_confirmation(changeset, :password, message: "does not match password")

cast(data, params, [:password])
|> validate_confirmation(:password, message: "does not match password")

Specs

validate_exclusion(t(), atom(), Enum.t(), Keyword.t()) :: t()

Validates a change is not included in the given enumerable.

Options

• :message - the message on failure, defaults to "is reserved"

Examples

validate_exclusion(changeset, :name, ~w(admin superadmin))

Specs

validate_format(t(), atom(), Regex.t(), Keyword.t()) :: t()

Validates a change has the given format.

The format has to be expressed as a regular expression.

Options

• :message - the message on failure, defaults to "has invalid format"

Examples

validate_format(changeset, :email, ~r/@/)

Specs

validate_inclusion(t(), atom(), Enum.t(), Keyword.t()) :: t()

Validates a change is included in the given enumerable.

Options

• :message - the message on failure, defaults to "is invalid"

Examples

validate_inclusion(changeset, :gender, ["man", "woman", "other", "prefer not to say"])
validate_inclusion(changeset, :age, 0..99)

Specs

validate_length(t(), atom(), Keyword.t()) :: t()

Validates a change is a string or list of the given length.

Note that the length of a string is counted in graphemes by default. If using this validation to match a character limit of a database backend, it's likely that the limit ignores graphemes and limits the number of unicode characters. Then consider using the :count option to limit the number of codepoints (:codepoints), or limit the number of bytes (:bytes).

Options

• :is - the length must be exactly this value
• :min - the length must be greater than or equal to this value
• :max - the length must be less than or equal to this value
• :count - what length to count for string, :graphemes (default), :codepoints or :bytes
• :message - the message on failure, depending on the validation, is one of:
  • for strings:
    • "should be %{count} character(s)"
    • "should be at least %{count} character(s)"
    • "should be at most %{count} character(s)"
  • for binary:
    • "should be %{count} byte(s)"
    • "should be at least %{count} byte(s)"
    • "should be at most %{count} byte(s)"
  • for lists:
    • "should have %{count} item(s)"
    • "should have at least %{count} item(s)"
    • "should have at most %{count} item(s)"

Examples

validate_length(changeset, :title, min: 3)
validate_length(changeset, :title, max: 100)
validate_length(changeset, :title, min: 3, max: 100)
validate_length(changeset, :code, is: 9)
validate_length(changeset, :topics, is: 2)
validate_length(changeset, :icon, count: :bytes, max: 1024 * 16)

Specs

validate_number(t(), atom(), Keyword.t()) :: t()

Validates the properties of a number.

Options

• :less_than
• :greater_than
• :less_than_or_equal_to
• :greater_than_or_equal_to
• :equal_to
• :not_equal_to
• :message - the message on failure, defaults to one of:
  • "must be less than %{number}"
  • "must be greater than %{number}"
  • "must be less than or equal to %{number}"
  • "must be greater than or equal to %{number}"
  • "must be equal to %{number}"
  • "must be not equal to %{number}"

Examples

validate_number(changeset, :count, less_than: 3)
validate_number(changeset, :pi, greater_than: 3, less_than: 4)
validate_number(changeset, :the_answer_to_life_the_universe_and_everything, equal_to: 42)

Specs

validate_required(t(), list() | atom(), Keyword.t()) :: t()

Validates that one or more fields are present in the changeset.

You can pass a single field name or a list of field names that are required.

If the value of a field is nil or a string made only of whitespace, the changeset is marked as invalid, the field is removed from the changeset's changes, and an error is added. An error won't be added if the field already has an error.

If a field is given to validate_required/3 but it has not been passed as parameter during cast/3 (i.e. it has not been changed), then validate_required/3 will check for its current value in the data. If the data contains an non-empty value for the field, then no error is added. This allows developers to use validate_required/3 to perform partial updates. For example, on insert all fields would be required, because their default values on the data are all nil, but on update, if you don't want to change a field that has been previously set, you are not required to pass it as a paramater, since validate_required/3 won't add an error for missing changes as long as the value in the data given to the changeset is not empty.

Do not use this function to validate associations are required, instead pass the :required option to cast_assoc/3.

Opposite to other validations, calling this function does not store the validation under the changeset.validations key. Instead, it stores all required fields under changeset.required.

Options

• :message - the message on failure, defaults to "can't be blank"
• :trim - a boolean that sets whether whitespaces are removed before running the validation on binaries/strings, defaults to true

Examples

validate_required(changeset, :title)
validate_required(changeset, [:title, :body])

Specs

validate_subset(t(), atom(), Enum.t(), Keyword.t()) :: t()

Validates a change, of type enum, is a subset of the given enumerable. Like validate_inclusion/4 for lists.

Options

• :message - the message on failure, defaults to "has an invalid entry"

Examples

validate_subset(changeset, :pets, ["cat", "dog", "parrot"])
validate_subset(changeset, :lottery_numbers, 0..99)

Specs

validations(t()) :: [{atom(), term()}]

Returns a keyword list of the validations for this changeset.

The keys in the list are the names of fields, and the values are a validation associated with the field. A field may occur multiple times in the list.

Example

%Post{}
|> change()
|> validate_format(:title, ~r/^\w+:\s/, message: "must start with a topic")
|> validate_length(:title, max: 100)
|> validations()
#=> [
  title: {:length, [ max: 100 ]},
  title: {:format, ~r/^\w+:\s/}
]

The following validations may be included in the result. The list is not necessarily exhaustive. For example, custom validations written by the developer will also appear in our return value.

This first group contains validations that take a keyword list of validators, where the validators are shown immediately following the validation type. This list may also include a message: key.

• {:length, [option]}

  • min: n
  • max: n
  • is: n
  • count: :graphemes | :codepoints

• {:number, [option]}

  • equal_to: n
  • greater_than: n
  • greater_than_or_equal_to: n
  • less_than: n
  • less_than_or_equal_to: n

The other validators simply take a value:

• {:exclusion, Enum.t}
• {:format, ~r/pattern/}
• {:inclusion, Enum.t}
• {:subset, Enum.t}