Directly attach an after_action function to the current change.

See Ash.Changeset.after_action/3 for more information.

Provide the option prepend?: true to place the hook before all other hooks instead of after.

Example

change after_action(fn changeset, record, _context ->
  Logger.debug("Successfully executed action #{changeset.action.name} on #{inspect(changeset.resource)}")
  {:ok, record}
end)

Directly attach an after_transaction function to the current change.

See Ash.Changeset.after_transaction/3 for more information.

Provide the option prepend?: true to place the hook before all other hooks instead of after.

Example

change after_transaction(fn
  changeset, {:ok, record}, _context ->
    Logger.debug("Successfully executed transaction for action #{changeset.action.name} on #{inspect(changeset.resource)}")
    {:ok, record}
  changeset, {:error, reason}, _context ->
    Logger.debug("Failed to execute transaction for action #{changeset.action.name} on #{inspect(changeset.resource)}, reason: #{inspect(reason)}")
    {:error, reason}
end)

@spec atomic_update(attribute :: atom(), expr :: Ash.Expr.t(), opts :: Keyword.t()) ::
  Ash.Resource.Change.ref()

Updates an attribute using an expression. See Ash.Changeset.atomic_update/3 for more.

Options:

• :cast_atomic? - set to false to ignore atomic type casting logic. Defaults to true.

Directly attach a before_action function to the current change.

See Ash.Changeset.before_action/3 for more information.

Provide the option append?: true to place the hook after all other hooks instead of before.

Example

change before_action(fn changeset, _context ->
  Logger.debug("About to execute #{changeset.action.name} on #{inspect(changeset.resource)}")

  changeset
end)

Directly attach a before_transaction function to the current change.

See Ash.Changeset.before_transaction/3 for more information.

Provide the option append?: true to place the hook after all other hooks instead of before.

Example

change before_transaction(fn changeset, _context ->
  Logger.debug("About to execute transaction for #{changeset.action.name} on #{inspect(changeset.resource)}")

  changeset
end)

Cascade this resource's destroy action to a related resource's destroy action.

Adds an after-action hook that explicitly calls destroy on any records related via the named relationship. It will optimise for bulk destroys where possible.

Beware database constraints

Think carefully before using this change with data layers which enforce referential integrity (ie PostgreSQL and SQLite) and you may need to defer constraints for the relationship in question.

See also:

1. postgres.references.reference.deferrable DSL
2. sqlite.references.reference.deferrable DSL
3. PostgreSQL's SET CONSTRAINTS documentation
4. SQLite's PRAGMA defer_foreign_keys documentation

Cascading notifications

By default notifications are disabled for the related destroy. This is to avoid potentially sending a lot of notifications for high-cardinality relationships.

Options

• :relationship (atom/0) - Required. The name of the relationship to work on

• :action (atom/0) - The name of the destroy action to call on the related resource. Uses the primary destroy by default.

• :read_action (atom/0) - The name of the read action to call on the related resource to find results to be destroyed

• :return_notifications? (boolean/0) - Return notifications for all destroyed records? The default value is false.

• :domain (Ash.Domain) -

Example

change cascade_destroy(:relationship)

Cascade a resource's update action to a related resource's update action.

Adds an after-action hook that explicitly calls update on any records related via the named relationship. It will optimise for bulk updates where possible.

Allows you to copy fields from the arguments or changes to the destination, this way you can cascade a bunch of changes downstream.

Beware database constraints

Think carefully before using this change with data layers which enforce referential integrity (ie PostgreSQL and SQLite) and you may need to defer constraints for the relationship in question.

See also:

1. postgres.references.reference.deferrable DSL
2. sqlite.references.reference.deferrable DSL
3. PostgreSQL's SET CONSTRAINTS documentation
4. SQLite's PRAGMA defer_foreign_keys documentation

Cascading notifications

By default notifications are disabled for the related updates. This is to avoid potentially sending a lot of notifications for high-cardinality relationships.

Options

• :relationship (atom/0) - Required. The name of the relationship to work on

• :action (atom/0) - The name of the update action to call on the related resource. Uses the primary update by default.

• :copy_inputs (list of atom/0) - A list of fields to copy & pass on to the downstream update. The source action cannot be atomic. The default value is [].

• :read_action (atom/0) - The name of the read action to call on the related resource to find results to be updated

• :return_notifications? (boolean/0) - Return notifications for all updated records? The default value is false.

• :domain (Ash.Domain) -

Example

change cascade_update(:relationship1)
change cascade_update(:relationship2, copy_inputs: [:field1, :field2])

@spec ensure_selected(select :: atom() | [atom()]) :: Ash.Resource.Change.ref()

Passes the provided value into Ash.Changeset.ensure_selected/2

If the value is not already selected, this makes sure it is. Does not deselect anything else.

Example

change ensure_selected([:necessary_field])

@spec filter(expr :: Ash.Expr.t()) :: Ash.Resource.Change.ref()

Applies a filter to the changeset. Has no effect for create actions.

This ensures that only things matching the provided filter are updated or destroyed.

@spec get_and_lock(lock :: Ash.DataLayer.lock_type()) :: Ash.Resource.Change.ref()

Re-fetches the record being updated and locks it with the given type.

This happens in a before_action hook (so that it is done as part of the transaction).

If your resource has global validations (in the top level validations block), you may want to add delay_global_validations? true to your action to ensure they happen on the locked record.

@spec get_and_lock_for_update() :: Ash.Resource.Change.ref()

Re-fetches the record being updated and locks it for update.

Only usable with data layers that support locking :for_update.

This happens in a before_action hook (so that it is done as part of the transaction).

If your resource has global validations (in the top level validations block), you may want to add delay_global_validations? true to your action to ensure they happen on the locked record.

@spec increment(attribute :: atom(), opts :: Keyword.t()) :: Ash.Resource.Change.ref()

Increments an attribute's value by the amount specified, which defaults to 1.

Options:

• :amount - Defaults to 1
• :overflow_limit - Defaults to nil. If the value is over the overflow limit it will roll-over to the amount being incremented by (for common database limit support)

@spec load(load :: term()) :: Ash.Resource.Change.ref()

Passes the provided value into Ash.load after the action has completed.

Example

change load(:comments)
change load([:friend_count, :friends])

@spec manage_relationship(
  argument :: atom(),
  relationship_name :: atom() | nil,
  opts :: Keyword.t()
) :: Ash.Resource.Change.ref()

Calls Ash.Changeset.manage_relationship/4 with the changeset and relationship provided, using the value provided for the named argument.

If relationship_name is not specified, it is assumed to be the same as the argument.

For information on the available options, see Ash.Changeset.manage_relationship/4.

Examples

change manage_relationship(:comments, type: :append)
change manage_relationship(:remove_comments, :comments, type: :remove)

Apply an "optimistic lock" on a record being updated or destroyed.

See Ash.Resource.Change.OptimisticLock for more.

@spec prevent_change(attribute :: atom()) :: Ash.Resource.Change.ref()

Clears a change off of the changeset before the action runs.

Does not fail if it is being changed, but ensures it is cleared just before the action.

Can be useful if a change is only used in validations but shouldn't ultimately be written to the data layer.

Examples

change prevent_change(:email)

@spec relate_actor(relationship :: atom(), opts :: Keyword.t()) ::
  Ash.Resource.Change.ref()

Relates the actor to the data being changed, as the provided relationship.

Options

• :allow_nil? (boolean/0) - Whether or not to allow the actor to be nil, in which case nothing will happen. The default value is false.

• :field (atom/0) - The field of the actor to set the relationship to

Examples

change relate_actor(:owner, allow_nil?: true)

@spec select(select :: atom() | [atom()]) :: Ash.Resource.Change.ref()

Passes the provided value into Ash.Changeset.select/3

Keep in mind, this will limit the fields that are selected. You may want ensure_selected/1 if you want to make sure that something is selected, without deselecting anything else.

Selecting in changesets does not actually do a select in the data layer. It nils out any fields that were not selected after completing the action. This can be useful if you are writing policies that have to do with specific fields being selected.

Example

change select([:name])

@spec set_attribute(
  attribute :: atom(),
  (-> term()) | {:_arg, :status} | term(),
  opts :: Keyword.t()
) :: Ash.Resource.Change.ref()

Sets the attribute to the value provided.

If a zero argument function is provided, it is called to determine the value.

Use arg(:argument_name) to use the value of the given argument. If the argument is not supplied then nothing happens.

Options

• :set_when_nil? (boolean/0) - When false, decline setting the attribute if it is nil. The default value is true.

• :new? (boolean/0) - When true, sets the attribute to the value provided if the attribute is not already being changed. The default value is false.

Examples

change set_attribute(:active, false)
change set_attribute(:opened_at, &DateTime.utc_now/0)
change set_attribute(:status, arg(:status))
change set_attribute(:encrypted_data, arg(:data), set_when_nil?: false)

@spec set_context(context :: map() | mfa()) :: Ash.Resource.Change.ref()

Merges the given query context.

If an MFA is provided, it will be called with the changeset. The MFA should return {:ok, context_to_be_merged} or {:error, term}

Examples

change set_context(%{something_used_internally: true})
change set_context({MyApp.Context, :set_context, []})

@spec set_new_attribute(
  relationship :: atom(),
  (-> term()) | {:_arg, :status} | term()
) ::
  Ash.Resource.Change.ref()

Sets the attribute to the value provided if the attribute is not already being changed.

If a zero argument function is provided, it is called to determine the value.

Use arg(:argument_name) to use the value of the given argument. If the argument is not supplied then nothing happens.

Examples

change set_new_attribute(:active, false)
change set_new_attribute(:opened_at, &DateTime.utc_now/0)
change set_new_attribute(:status, arg(:status))

Updates an existing attribute change by applying a function to it.

The update function gets called with the value already cast to the correct type, and only gets called on valid changesets, so the value is guaranteed to have passed validations and constraints.