Comparing Parameterized types to Structured types

Parameterized constraints are built into core Moose and you are probably already familiar with the type constraints "HashRef" and "ArrayRef". Structured types have similar functionality, so their syntax is likewise similar. For example, you could define a parameterized constraint like:

    subtype ArrayOfInts,
     as ArrayRef[Int];

which would constrain a value to something like [1,2,3,...] and so on. On the other hand, a structured type constraint explicitly names all it's allowed 'internal' type parameter constraints. For the example:

    subtype StringFollowedByInt,
     as Tuple[Str,Int];

would constrain its value to things like "['hello', 111]" but "['hello', 'world']" would fail, as well as "['hello', 111, 'world']" and so on. Here's another example:

        package MyApp::Types;
    use MooseX::Types -declare [qw(StringIntOptionalHashRef)];
    use MooseX::Types::Moose qw(Str Int);
    use MooseX::Types::Structured qw(Tuple Optional);
    subtype StringIntOptionalHashRef,
     as Tuple[
        Str, Int,
        Optional[HashRef]
     ];

This defines a type constraint that validates values like:

    ['Hello', 100, {key1 => 'value1', key2 => 'value2'}];
    ['World', 200];

Notice that the last type constraint in the structure is optional. This is enabled via the helper "Optional" type constraint, which is a variation of the core Moose type constraint "Maybe". The main difference is that "Optional" type constraints are required to validate if they exist, while "Maybe" permits undefined values. So the following example would not validate:

    StringIntOptionalHashRef->validate(['Hello Undefined', 1000, undef]);

Please note the subtle difference between undefined and null. If you wish to allow both null and undefined, you should use the core Moose "Maybe" type constraint instead:

    package MyApp::Types;
    use MooseX::Types -declare [qw(StringIntMaybeHashRef)];
    use MooseX::Types::Moose qw(Str Int Maybe);
    use MooseX::Types::Structured qw(Tuple);
    subtype StringIntMaybeHashRef,
     as Tuple[
        Str, Int, Maybe[HashRef]
     ];

This would validate the following:

    ['Hello', 100, {key1 => 'value1', key2 => 'value2'}];
    ['World', 200, undef];
    ['World', 200];

Structured constraints are not limited to arrays. You can define a structure against a "HashRef" with the "Dict" type constraint as in this example:

    subtype FirstNameLastName,
     as Dict[
        firstname => Str,
        lastname => Str,
     ];

This would constrain a "HashRef" that validates something like:

    {firstname => 'Christopher', lastname => 'Parsons'};

but all the following would fail validation:

    ## Incorrect keys
    {first => 'Christopher', last => 'Parsons'};
    ## Too many keys
    {firstname => 'Christopher', lastname => 'Parsons', middlename => 'Allen'};
    ## Not a HashRef
    ['Christopher', 'Parsons'];

These structures can be as simple or elaborate as you wish. You can even combine various structured, parameterized and simple constraints all together:

    subtype Crazy,
     as Tuple[
        Int,
        Dict[name=>Str, age=>Int],
        ArrayRef[Int]
     ];

Which would match:

    [1, {name=>'John', age=>25},[10,11,12]];

Please notice how the type parameters can be visually arranged to your liking and to improve the clarity of your meaning. You don't need to run then altogether onto a single line. Additionally, since the "Dict" type constraint defines a hash constraint, the key order is not meaningful. For example:

    subtype AnyKeyOrder,
      as Dict[
        key1=>Int,
        key2=>Str,
        key3=>Int,
     ];

Would validate both:

    {key1 => 1, key2 => "Hi!", key3 => 2};
    {key2 => "Hi!", key1 => 100, key3 => 300};

As you would expect, since underneath it's just a plain old Perl hash at work.

Alternatives

You should exercise some care as to whether or not your complex structured constraints would be better off contained by a real object as in the following example:

    package MyApp::MyStruct;
    use Moose;
    ## lazy way to make a bunch of attributes
    has $_ for qw(full_name age_in_years);
    package MyApp::MyClass;
    use Moose;
    has person => (isa => 'MyApp::MyStruct');
    my $instance = MyApp::MyClass->new(
        person=>MyApp::MyStruct->new(
            full_name => 'John',
            age_in_years => 39,
        ),
    );

This method may take some additional time to set up but will give you more flexibility. However, structured constraints are highly compatible with this method, granting some interesting possibilities for coercion. Try:

    package MyApp::MyClass;
    use Moose;
    use MyApp::MyStruct;
    ## It's recommended your type declarations live in a separate class in order
    ## to promote reusability and clarity.  Inlined here for brevity.
    use MooseX::Types::DateTime qw(DateTime);
    use MooseX::Types -declare [qw(MyStruct)];
    use MooseX::Types::Moose qw(Str Int);
    use MooseX::Types::Structured qw(Dict);
    ## Use class_type to create an ISA type constraint if your object doesn't
    ## inherit from Moose::Object.
    class_type 'MyApp::MyStruct';
    ## Just a shorter version really.
    subtype MyStruct,
     as 'MyApp::MyStruct';
    ## Add the coercions.
    coerce MyStruct,
     from Dict[
        full_name=>Str,
        age_in_years=>Int
     ], via {
        MyApp::MyStruct->new(%$_);
     },
     from Dict[
        lastname=>Str,
        firstname=>Str,
        dob=>DateTime
     ], via {
        my $name = $_->{firstname} .' '. $_->{lastname};
        my $age = DateTime->now - $_->{dob};
        MyApp::MyStruct->new(
            full_name=>$name,
            age_in_years=>$age->years,
        );
     };
    has person => (isa=>MyStruct);

This would allow you to instantiate with something like:

    my $obj = MyApp::MyClass->new( person => {
        full_name=>'John Napiorkowski',
        age_in_years=>39,
    });

Or even:

    my $obj = MyApp::MyClass->new( person => {
        lastname=>'John',
        firstname=>'Napiorkowski',
        dob=>DateTime->new(year=>1969),
    });

If you are not familiar with how coercions work, check out the Moose cookbook entry Moose::Cookbook::Recipe5 for an explanation. The section "Coercions" has additional examples and discussion.

Subtyping a Structured type constraint

You need to exercise some care when you try to subtype a structured type as in this example:

    subtype Person,
     as Dict[name => Str];
    subtype FriendlyPerson,
     as Person[
        name => Str,
        total_friends => Int,
     ];

This will actually work BUT you have to take care that the subtype has a structure that does not contradict the structure of it's parent. For now the above works, but I will clarify the syntax for this at a future point, so it's recommended to avoid (should not really be needed so much anyway). For now this is supported in an EXPERIMENTAL way. Your thoughts, test cases and patches are welcomed for discussion. If you find a good use for this, please let me know.

Coercions

Coercions currently work for 'one level' deep. That is you can do:

    subtype Person,
     as Dict[
        name => Str,
        age => Int
    ];
    subtype Fullname,
     as Dict[
        first => Str,
        last => Str
     ];
    coerce Person,
     ## Coerce an object of a particular class
     from BlessedPersonObject, via {
        +{
            name=>$_->name,
            age=>$_->age,
        };
     },
     ## Coerce from [$name, $age]
     from ArrayRef, via {
        +{
            name=>$_->[0],
            age=>$_->[1],
        },
     },
     ## Coerce from {fullname=>{first=>...,last=>...}, dob=>$DateTimeObject}
     from Dict[fullname=>Fullname, dob=>DateTime], via {
        my $age = $_->dob - DateTime->now;
        my $firstn = $_->{fullname}->{first};
        my $lastn = $_->{fullname}->{last}
        +{
            name => $_->{fullname}->{first} .' '. ,
            age =>$age->years
        }
     };

And that should just work as expected. However, if there are any 'inner' coercions, such as a coercion on "Fullname" or on "DateTime", that coercion won't currently get activated.

Please see the test 07-coerce.t for a more detailed example. Discussion on extending coercions to support this welcome on the Moose development channel or mailing list.

Recursion

Newer versions of MooseX::Types support recursive type constraints. That is you can include a type constraint as a contained type constraint of itself. For example:

    subtype Person,
     as Dict[
         name=>Str,
         friends=>Optional[
             ArrayRef[Person]
         ],
     ];

This would declare a "Person" subtype that contains a name and an optional "ArrayRef" of "Person"s who are friends as in:

    {
        name => 'Mike',
        friends => [
            { name => 'John' },
            { name => 'Vincent' },
            {
                name => 'Tracey',
                friends => [
                    { name => 'Stephenie' },
                    { name => 'Ilya' },
                ],
            },
        ],
    };

Please take care to make sure the recursion node is either "Optional", or declare a union with an non-recursive option such as:

    subtype Value
     as Tuple[
         Str,
         Str|Tuple,
     ];

Which validates:

    [
        'Hello', [
            'World', [
                'Is', [
                    'Getting',
                    'Old',
                ],
            ],
        ],
    ];

Otherwise you will define a subtype that is impossible to validate since it is infinitely recursive. For more information about defining recursive types, please see the documentation in MooseX::Types and the test cases.

Tuple[@constraints]

This defines an ArrayRef based constraint which allows you to validate a specific list of contained constraints. For example:

    Tuple[Int,Str]; ## Validates [1,'hello']
    Tuple[Str|Object, Int]; ## Validates ['hello', 1] or [$object, 2]

The Values of @constraints should ideally be MooseX::Types declared type constraints. We do support 'old style' Moose string based constraints to a limited degree but these string type constraints are considered deprecated. There will be limited support for bugs resulting from mixing string and MooseX::Types in your structures. If you encounter such a bug and really need it fixed, we will required a detailed test case at the minimum.

Dict[%constraints]

This defines a HashRef based constraint which allowed you to validate a specific hashref. For example:

    Dict[name=>Str, age=>Int]; ## Validates {name=>'John', age=>39}

The keys in %constraints follow the same rules as @constraints in the above section.

Map[ $key_constraint, $value_constraint ]

This defines a "HashRef"-based constraint in which both the keys and values are required to meet certain constraints. For example, to map hostnames to IP addresses, you might say:

  Map[ HostName, IPAddress ]

The type constraint would only be met if every key was a valid "HostName" and every value was a valid "IPAddress".

Optional[$constraint]

This is primarily a helper constraint for "Dict" and "Tuple" type constraints. What this allows is for you to assert that a given type constraint is allowed to be null (but NOT undefined). If the value is null, then the type constraint passes but if the value is defined it must validate against the type constraint. This makes it easy to make a Dict where one or more of the keys doesn't have to exist or a tuple where some of the values are not required. For example:

    subtype Name() => as Dict[
        first=>Str,
        last=>Str,
        middle=>Optional[Str],
    ];

...creates a constraint that validates against a hashref with the keys 'first' and 'last' being strings and required while an optional key 'middle' is must be a string if it appears but doesn't have to appear. So in this case both the following are valid:

    {first=>'John', middle=>'James', last=>'Napiorkowski'}
    {first=>'Vanessa', last=>'Li'}

If you use the "Maybe" type constraint instead, your values will also validate against "undef", which may be incorrect for you.

slurpy

Structured type constraints by their nature are closed; that is validation will depend on an exact match between your structure definition and the arguments to be checked. Sometimes you might wish for a slightly looser amount of validation. For example, you may wish to validate the first 3 elements of an array reference and allow for an arbitrary number of additional elements. At first thought you might think you could do it this way:

    #  I want to validate stuff like: [1,"hello", $obj, 2,3,4,5,6,...]
    subtype AllowTailingArgs,
     as Tuple[
       Int,
       Str,
       Object,
       ArrayRef[Int],
     ];

However what this will actually validate are structures like this:

    [10,"Hello", $obj, [11,12,13,...] ]; # Notice element 4 is an ArrayRef

In order to allow structured validation of, "and then some", arguments, you can use the "slurpy" method against a type constraint. For example:

    use MooseX::Types::Structured qw(Tuple slurpy);
    subtype AllowTailingArgs,
     as Tuple[
       Int,
       Str,
       Object,
       slurpy ArrayRef[Int],
     ];

This will now work as expected, validating ArrayRef structures such as:

    [1,"hello", $obj, 2,3,4,5,6,...]

A few caveats apply. First, the slurpy type constraint must be the last one in the list of type constraint parameters. Second, the parent type of the slurpy type constraint must match that of the containing type constraint. That means that a "Tuple" can allow a slurpy "ArrayRef" (or children of "ArrayRef"s, including another "Tuple") and a "Dict" can allow a slurpy "HashRef" (or children/subtypes of HashRef, also including other "Dict" constraints).

Please note the technical way this works 'under the hood' is that the slurpy keyword transforms the target type constraint into a coderef. Please do not try to create your own custom coderefs; always use the slurpy method. The underlying technology may change in the future but the slurpy keyword will be supported.

Normalize a HashRef

You need a hashref to conform to a canonical structure but are required accept a bunch of different incoming structures. You can normalize using the "Dict" type constraint and coercions. This example also shows structured types mixed which other MooseX::Types libraries.

    package Test::MooseX::Meta::TypeConstraint::Structured::Examples::Normalize;
    use Moose;
    use DateTime;
    use MooseX::Types::Structured qw(Dict Tuple);
    use MooseX::Types::DateTime qw(DateTime);
    use MooseX::Types::Moose qw(Int Str Object);
    use MooseX::Types -declare => [qw(Name Age Person)];
    subtype Person,
     as Dict[
         name=>Str,
         age=>Int,
     ];
    coerce Person,
     from Dict[
         first=>Str,
         last=>Str,
         years=>Int,
     ], via { +{
        name => "$_->{first} $_->{last}",
        age => $_->{years},
     }},
     from Dict[
         fullname=>Dict[
             last=>Str,
             first=>Str,
         ],
         dob=>DateTime,
     ],
     ## DateTime needs to be inside of single quotes here to disambiguate the
     ## class package from the DataTime type constraint imported via the
     ## line "use MooseX::Types::DateTime qw(DateTime);"
     via { +{
        name => "$_->{fullname}{first} $_->{fullname}{last}",
        age => ($_->{dob} - 'DateTime'->now)->years,
     }};
    has person => (is=>'rw', isa=>Person, coerce=>1);

And now you can instantiate with all the following:

    __PACKAGE__->new(
        person=>{
            name=>'John Napiorkowski',
            age=>39,
        },
    );
    __PACKAGE__->new(
        person=>{
            first=>'John',
            last=>'Napiorkowski',
            years=>39,
        },
    );
    __PACKAGE__->new(
        person=>{
            fullname => {
                first=>'John',
                last=>'Napiorkowski'
            },
            dob => 'DateTime'->new(
                year=>1969,
                month=>2,
                day=>13
            ),
        },
    );

This technique is a way to support various ways to instantiate your class in a clean and declarative way.