NAME Zydeco - Jazz up your Perl SYNOPSIS MyApp.pm use v5.14; use strict; use warnings; package MyApp { use Zydeco; class Person { has name ( type => Str, required => true ); has gender ( type => Str ); factory new_man (Str $name) { return $class->new(name => $name, gender => 'male'); } factory new_woman (Str $name) { return $class->new(name => $name, gender => 'female'); } method greet (Person *friend, Str *greeting = "Hello") { printf("%s, %s!\n", $arg->greeting, $arg->friend->name); } coerce from Str via from_string { return $class->new(name => $_); } } } my_script.pl use v5.14; use strict; use warnings; use MyApp; use MyApp::Types qw( is_Person ); # Create a new MyApp::Person object. # my $alice = MyApp->new_woman("Alice"); is_Person($alice) or die; # The string "Bob" will be coerced to a MyApp::Person. # $alice->greet(friend => "Bob", greeting => 'Hi'); DESCRIPTION Zydeco is a Perl module to jazz up your object-oriented programming. It fuses together: * Classes, roles, and interfaces, including parameterizable classes and roles (a.k.a. class generators and role generators). * Factories to help your objects make other objects. * Powerful and concise attribute definitions * Methods with signatures, type constraints, and coercion. * Method modifiers to easily wrap or override inherited methods. * Multimethods. * Powerful delegation features. * True private methods and attributes. * Syntactic sugar as sweet as pecan pie. Important Concepts The Factory Package and Prefix Zydeco assumes that all the classes and roles you are building with it will be defined under the same namespace prefix. For example "MyApp::Person" and "MyApp::Company" are both defined under the common prefix of "MyApp". It also assumes there will be a factory package that can be used to build new instances of your class. Rather than creating a new person object with `MyApp::Person->new()`, you should create a new person object with `MyApp->new_person()`. Calling `MyApp::Person->new()` directly is only encouraged from within the "MyApp::Person" class itself, and from within the factory. Everywhere else, you should call `MyApp->new_person()` instead. By default, the factory package and the prefix are the same: they are the caller that you imported Zydeco into. But they can be set to whatever: use Zydeco ( prefix => 'MyApp::Objects', factory_package => 'MyApp::Makers', ); (If Zydeco is imported from `main`, then prefix defaults to undef, and factory_package defaults to "Local". But you will get a warning about that unless it's a `perl -e` one-liner.) Zydeco assumes that you are defining all the classes and roles within this namespace prefix in a single Perl module file. This Perl module file would normally be named based on the prefix, so in the example above, it would be "MyApp/Objects.pm" and in the example from the SYNOPSIS, it would be "MyApp.pm". But see also the documentation for `include`. Of course, there is nothing to stop you from having multiple prefixes for different logical parts of a larger codebase, but Zydeco assumes that if it's been set up for a prefix, it owns that prefix and everything under it, and it's all defined in the same Perl module. Each object defined by Zydeco will have a `FACTORY` method, so you can do: $person_object->FACTORY And it will return the string "MyApp". This allows for stuff like: class Person { method give_birth { return $self->FACTORY->new_person(); } } The Type Library While building your classes and objects, Zydeco will also build type constraints for each of them. So for the "MyApp::Person" class above, it also builds a Person type constraint. This can be used in Moo/Moose attribute definitions like: use MyApp; use MyApp::Types qw( Person ); use Moose; has boss => (is => 'rw', isa => Person); And just anywhere a type constraint may be used generally. You should know this stuff by now. Note that we had to `use MyApp` before we could `use MyApp::Types`. This is because there isn't a physical "MyApp/Types.pm" file on disk; it is defined entirely by "MyApp.pm". Your type library will be the same as your namespace prefix, with "::Types" added at the end. But you can change that: use Zydeco ( prefix => 'MyApp::Objects', factory_package => 'MyApp::Makers', type_library => 'MyApp::TypeLibrary', ); It can sometimes be helpful to pre-warn Zydeco about the types you're going to define before you define them, just so it is able to allow them as barewords in some places... use Zydeco ( prefix => 'MyApp::Objects', factory_package => 'MyApp::Makers', type_library => 'MyApp::TypeLibrary', declare => [qw( Person Company )], ); See also Type::Tiny::Manual. KEYWORDS `class` Define a very basic class: class Person; Define a more complicated class: class Person { ...; } Note that for the `class` keyword without a block, it does *not* act like the `package` keyword by changing the "ambient" package. It just defines a totally empty class with no methods or attributes. The prefix will automatically be added to the class name, so if the prefix is MyApp, the above will create a class called MyApp::Person. It will also create a factory method `MyApp->new_person`. (The name is generated by stripping the prefix from the class name, replacing any "::" with an underscore, lowercasing, and prefixing it with "new_".) And it will create a type called Person in the type library. (Same rules to generate the name apart from lowercasing and adding "new_".) Classes can be given more complex names: class Person::Neanderthal { ...; } Will create "MyApp::Person::Neanderthal" class, a factory method called `MyApp->new_person_neanderthal`, and a Person_Neanderthal type. It is possible to create a class without the prefix: class ::Person { ...; } The class name will now be "Person" instead of "MyApp::Person"! Nested classes `class` blocks can be nested. This establishes an inheritance heirarchy. class Animal { has name; class Mammal { class Primate { class Monkey; class Gorilla; class Human { class Superhuman; } } } class Bird; class Fish { class Shark; } } my $superman = MyApp->new_superhuman( name => 'Kal El' ); See also `extends` as an alternative way of declaring inheritance. It is possible to prefix a class name with a plus sign: package MyApp { use Zydeco; class Person { has name; class +Employee { has job_title; } } } Now the employee class will be named `MyApp::Person::Employee` instead of the usual `MyApp::Employee`. Abstract classes Classes can be declared as abstract: package MyApp { use Zydeco; abstract class Animal { class Cat; class Dog; } } For abstract classes, there is no constructor or factory, so you cannot create an Animal instance directly; but you can create instances of the subclasses. It is usually better to use roles than abstract classes, but sometimes the abstract class makes more intuitive sense. `role` Define a very basic role: role Person; Define a more complicated role: role Person { ...; } This is just the same as `class` but defines a role instead of a class. Roles cannot be nested within each other, nor can roles be nested in classes, nor classes in roles. `interface` An interface is a lightweight role. It cannot define attributes, methods, multimethods, or method modifiers, but otherwise functions as a role. (It may have `requires` statements and define constants.) package MyApp; use Zydeco; interface Serializer { requires serialize; } interface Deserializer { requires deserialize; } class MyJSON { with Serializer, Deserialize; method serialize ($value) { ... } method deserialize ($value) { ... } } my $obj = MyApp->new_myjson; $obj->does('MyApp::Serializer'); # true `toolkit` Use a different toolkit instead of Moo. # use Mouse class Foo { toolkit Mouse; } # use Moose # use MooseX::Aliases # use MooseX::StrictConstructor class Bar { toolkit Moose ( Aliases, StrictConstructor ); } You can of course specify you want to use Moo: class Baz { toolkit Moo; } Not all MooseX/MouseX/MooX packages will work, but *X::StrictConstructor will. Although it is not possible to use the `toolkit` keyword outside of `class`, `abstract class`, `role`, and `interface` blocks, it is possible to specify a default toolkit when you import Zydeco. use Zydeco ( ..., toolkit => 'Moose', ); use Zydeco ( ..., toolkit => 'Mouse', ); `extends` Defines a parent class. Only for use with `class` and `abstract class` blocks. class Person { extends Animal; } This works: class Person { extends ::Animal; # no prefix } It is possible to "lift" `extends` outside the class definition block: class Person extends Animal { ...; } If there are no methods, etc to define, you don't need the block at all: class Person extends Animal; `with` Composes roles and interfaces. class Person { with Employable, Consumer; } role Consumer; role Worker; role Payable; role Employable { with Worker, Payable; } It is possible to compose a role which does not have its own definition by adding a question mark to the end of the name: class Person { with Employable, Consumer?; } role Employable { with Worker?, Payable?; } This is equivalent to declaring an empty role. The `with` keyword can only be used with `class`, `abstract class`, `role`, and `interface` blocks. Similarly to `extends`, `with` can be "lifted" outside its block: class Company with Taxable { ...; } interface Storage with Serializer, Deserializer; If lifting both `extends` and `with` outside a block, `extends` must come first. class Person extends Animal with Thinking::Rational { ...; } `begin` This code gets run early on in the definition of a class or role. class Person { begin { say "Defining $package"; } } At the time the code gets run, none of the class's attributes or methods will be defined yet. The lexical variables $package and $kind are defined within the block. $kind will be either 'class' or 'role'. The `begin` keyword cannot be used outside of `class`, `abstract class`, `role`, and `interface` blocks, though it is possible to define a global default for it: use Zydeco ( ..., begin => sub { my ($package, $kind) = @_; ...; }, ); Per-package `begin` overrides the global `begin`. Unlike Perl's `BEGIN` keyword, a package can only have one `begin`. If `class` definitions are nested, `begin` blocks will be inherited by child classes. If a parent class is specified via `extends`, `begin` blocks will not be inherited. `end` This code gets run late in the definition of a class or role. class Person { end { say "Finished defining $package"; } } The lexical variables $package and $kind are defined within the block. $kind will be either 'class' or 'role'. The `end` keyword cannot be used outside of `class`, `abstract class`, `role`, and `interface` blocks, though it is possible to define a global default for it: use Zydeco ( ..., end => sub { my ($package, $kind) = @_; ...; }, ); Per-package `end` overrides the global `end`. Unlike Perl's `END` keyword, a package can only have one `end`. If `class` definitions are nested, `end` blocks will be inherited by child classes. If a parent class is specified via `extends`, `end` blocks will not be inherited. `has` Defines an attribute. class Person { has name; has age; } my $bob = MyApp->new_person(name => "Bob", age => 21); Cannot be used outside of `class`, `abstract class`, and `role` blocks. Moo-style attribute specifications may be given: class Person { has name ( is => rw, type => Str, required => true ); has age ( is => rw, type => Int ); } Note there is no fat comma after the attribute name! It is a bareword. Use a plus sign before an attribute name to modify an attribute defined in a parent class. class Animal { has name ( type => Str, required => false ); class Person { has +name ( required => true ); } } `rw`, `rwp`, `ro`, `lazy`, `bare`, `private`, `true`, and `false` are allowed as barewords for readability, but `is` is optional, and defaults to `rw`. The names of attributes can start with an asterisk: has *foo; This adds no extra meaning, but is supported for consistency with the syntax of named parameters in method signatures. (Depending on your text editor, it may also improve syntax highlighting.) If you need to decide an attribute name on-the-fly, you can replace the name with a block that returns the name as a string. class Employee { extends Person; has { $ENV{LOCALE} eq 'GB' ? 'national_insurance_no' : 'social_security_no' } (type => Str) } my $bob = Employee->new( name => 'Bob', social_security_no => 1234, ); You can think of the syntax as being kind of like `print`. print BAREWORD_FILEHANDLE @strings; print { block_returning_filehandle(); } @strings; The block is called in scalar context, so you'll need a loop to define a list like this: class Person { my @attrs = qw( name age ); # this does not work has {@attrs} ( required => true ); # this works for my $attr (@attrs) { has {$attr} ( required => true ); } } Type constraints for attributes Note `type` instead of `isa`. Any type constraints from Types::Standard, Types::Common::Numeric, and Types::Common::String will be avaiable as barewords. Also, any pre-declared types can be used as barewords. It's possible to quote types as strings, in which case you don't need to have pre-declared them. class Person { has name ( type => Str, required => true ); has age ( type => Int ); has spouse ( type => 'Person' ); has kids ( is => lazy, type => 'ArrayRef[Person]', builder => sub { [] }, ); } Note that when `type` is a string, Zydeco will consult your type library to figure out what it means. It is also possible to use `isa => 'SomeClass'` or `does => 'SomeRole'` to force strings to be treated as class names or role names instead of type names. class Person { has name ( type => Str, required => true ); has age ( type => Int ); has spouse ( isa => 'Person' ); has pet ( isa => '::Animal' ); # no prefix } For enumerations, you can define them like this: class Person { ...; has status ( enum => ['alive', 'dead', 'undead'] ); } Delegation Zydeco integrates support for MooX::Enumeration (and MooseX::Enumeration, but MouseX::Enumeration doesn't exist). class Person { ...; has status ( enum => ['alive', 'dead', 'undead'], default => 'alive', handles => 1, ); } my $bob = MyApp->new_person; if ( $bob->is_alive ) { ...; } `handles => 1` creates methods named `is_alive`, `is_dead`, and `is_undead`, and `handles => 2` creates methods named `status_is_alive`, `status_is_dead`, and `status_is_undead`. Checking `$bob->status eq 'alvie'` is prone to typos, but `$bob->status_is_alvie` will cause a runtime error because the method is not defined. Zydeco also integrates support for Sub::HandlesVia allowing you to delegate certain methods to unblessed references and non-reference values. For example: class Person { has age ( type => 'Int', default => 0, handles_via => 'Counter', handles => { birthday => 'inc', # increment age }, ); after birthday { if ($self->age < 30) { say "yay!"; } else { say "urgh!"; } } } `handles` otherwise works as you'd expect from Moo and Moose. Required versus optional attributes and defaults A trailing `!` indicates a required attribute. class Person { has name!; } It is possible to give a default using an equals sign. class WidgetCollection { has name = "Widgets"; has count (type => Num) = 0; } Note that the default comes after the spec, so in cases where the spec is long, it may be clearer to express the default inside the spec: class WidgetCollection { has name = "Widgets"; has count ( type => Num, lazy => true, required => false, default => 0, ); } Defaults given this way will be eager (non-lazy), but can be made lazy using the spec: class WidgetCollection { has name = "Widgets"; has count (is => lazy) = 0; } Defaults *can* use the $self object: class WidgetCollection { has name = "Widgets"; has display_name = $self->name; } Any default that includes $self will automatically be lazy, but can be made eager using the spec. (It is almost certainly a bad idea to do so though.) class WidgetCollection { has name = "Widgets"; has display_name ( lazy => false ) = $self->name; } Specifying multiple attributes at once Commas may be used to separate multiple attributes: class WidgetCollection { has name, display_name ( type => Str ); } The specification and defaults are applied to every attribute in the list. Private attributes If an attribute name starts with a dollar sign, it is a private (lexical) attribute. Private attributes cannot be set in the constructor, and cannot be directly accessed outside the class's lexical scope. class Foo { has $ua = HTTP::Tiny->new; method fetch_data ( Str $url ) { my $response = $self->$ua->get($url); $response->{is_success} or confess('request failed'); return $response->{content}; } } Note how `$self->$ua` is still called as a method. You don't just do `$ua->get()`. The invocant is still required, just like it would be with a normal public attribute: class Foo { has ua = HTTP::Tiny->new; method fetch_data ( Str $url ) { my $response = $self->ua->get($url); $response->{is_success} or confess('request failed'); return $response->{content}; } } Private attributes can have delegated methods (`handles`): class Foo { has $ua ( default => sub { HTTP::Tiny->new }, handles => [ http_get => 'get', http_post => 'post', ], ); method fetch_data ( Str $url ) { my $response = $self->http_get($url); $response->{is_success} or confess('request failed'); return $response->{content}; } } These can even be made lexical too: class Foo { my ($http_get, $http_post); # predeclare has $ua ( default => sub { HTTP::Tiny->new }, handles => [ \$http_get => 'get', \$http_post => 'post', ], ); method fetch_data ( Str $url ) { my $response = $self->$http_get($url); $response->{is_success} or confess('request failed'); return $response->{content}; } } Note how an arrayref is used for `handles` instead of a hashref. This is because scalarrefs don't work as hashref keys. Although constructors ignore private attributes, you may set them in a factory method. class Foo { has $ua; factory new_foo (%args) { my $instance = $class->new(%args); $instance->$ua( HTTP::Tiny->new ); return $instance; } } `has $foo` is just a shortcut for: my $foo; has foo => (is => "private", accessor => \$foo); You can use `is => "private"` to create even *more* private attributes without even having that lexical accessor: has foo => (is => "private"); If it seems like an attribute that can't be set in the constructor and doesn't have accessors would be useless, you're wrong. Because it can still have delegations and a default value. Private attributes use lexical variables, so are visible to subclasses only if the subclass definition is nested in the base class. From version 0.506 onwards, not only can private attributes have public delegations and accessors, but public attributes can have private delegations and attributes: class Foo { my ($clear_ua, $http_get, $http_post); # predeclare has ua ( is => ro, # public attribute! clearer => \$clear_ua, # private clearer default => sub { HTTP::Tiny->new }, handles => [ # private delegations \$http_get => 'get', # \$http_post => 'post', # ], ); ...; } `constant` Defines a constant. class Person { extends Animal; constant latin_name = 'Homo sapiens'; } `MyApp::Person->latin_name`, `MyApp::Person::latin_name`, and `$person_object->latin_name` will return 'Homo sapiens'. Outside of `class`, `abstract class`, `role`, and `interface` blocks, will define a constant in the caller package. (That is, usually the factory.) `method` Defines a method. class Person { has spouse; method marry { my ($self, $partner) = @_; $self->spouse($partner); $partner->spouse($self); return $self; } } `sub { ... }` will not work as a way to define methods within the class. Use `method { ... }` instead. Outside of `class`, `abstract class`, `role`, and `interface` blocks, `method` will define a method in the caller package. (Usually the factory.) The variables $self and $class will be automatically defined within all methods. $self is set to $_[0] (though the invocant is not shifted off @_). $class is set to `ref($self)||$self`. If the method is called as a class method, both $self and $class will be the same thing: the full class name as a string. If the method is called as an object method, $self is the object and $class is its class. Like with `has`, you may use a block that returns a string instead of a bareword name for the method. method {"ma"."rry"} { ...; } Zydeco supports method signatures for named arguments and positional arguments. A mixture of named and positional arguments is allowed, with some limitations. For anything more complicates, you should define the method with no signature at all, and unpack @_ within the body of the method. Signatures for Named Arguments class Person { has spouse; method marry ( Person *partner, Object *date = DateTime->now ) { $self->spouse( $arg->partner ); $arg->partner->spouse( $self ); return $self; } } The syntax for each named argument is: Type *name = default The type is a type name, which will be parsed using Type::Parser. (So it can include the `~`, `|`, and `&`, operators, and can include parameters in `[ ]` brackets. Type::Parser can handle whitespace in the type, but not comments. Alternatively, you can provide a block which returns a type name as a string or returns a blessed Type::Tiny object. For very complex types, where you're expressing additional coercions or value constraints, this is probably what you want. The asterisk indicates that the argument is named, not positional. The name may be followed by a question mark to indicate an optional argument. method marry ( Person *partner, Object *date? ) { ...; } Or it may be followed by an equals sign to set a default value. Comments may be included in the signature, but not in the middle of a type constraint. method marry ( # comment here is okay Person # comment here is fine too *partner # and here ) { ... } method marry ( Person # comment here is not okay! | Horse *partner ) { ... } As with signature-free methods, $self and $class wll be defined for you in the body of the method. However, when a signature has been used $self *is* shifted off @_. Also within the body of the method, a variable called $arg is provided. This is a hashref of the named arguments. So you can access the partner argument in the above example like this: $arg->{partner} But because $arg is blessed, you can also do: $arg->partner The latter style is encouraged as it looks neater, plus it helps catch typos. (`$ars->{pratner}` for example!) However, accessing it as a plain hashref is supported and shouldn't be considered to be breaking encapsulation. For optional arguments you can check: exists($arg->{date}) Or: $arg->has_date For types which have a coercion defined, the value will be automatically coerced. Methods with named arguments can be called with a hash or hashref. $alice->marry( partner => $bob ); # okay $alice->marry({ partner => $bob }); # also okay Signatures for Positional Arguments method marry ( Person $partner, Object $date? ) { $self->spouse( $partner ); $partner->spouse( $self ); return $self; } The dollar sign is used instead of an asterisk to indicate a positional argument. As with named arguments, $self is automatically shifted off @_ and $class exists. Unlike named arguments, there is no $arg variable, and instead a scalar variable is defined for each named argument. Optional arguments and defaults are supported in the same way as named arguments. It is possible to include a slurpy hash or array at the end of the list of positional arguments. method marry ( $partner, $date, @vows ) { ...; } If you need to perform a type check on the slurpy parameter, you should pretend it is a hashref or arrayref. method marry ( $partner, $date, ArrayRef[Str] @vows ) { ...; } Signatures with Mixed Arguments You may mix named and positional arguments with the following limitations: * Positional arguments must appear at the beginning and/or end of the list. They cannot be surrounded by named arguments. * Positional arguments cannot be optional and cannot have a default. They must be required. (Named arguments can be optional and have defaults.) * No slurpies! method print_html ($tag, Str $text, *htmlver?, *xml?, $fh) { confess "update your HTML" if $arg->htmlver < 5; if (length $text) { print $fh "$text"; } elsif ($arg->xml) { print $fh ""; } else { print $fh ""; } } $obj->print_html('h1', 'Hello World', { xml => true }, \*STDOUT); $obj->print_html('h1', 'Hello World', xml => true , \*STDOUT); $obj->print_html('h1', 'Hello World', \*STDOUT); Mixed signatures are basically implemented like named signatures, but prior to interpreting @_ as a hash, some parameters are spliced off the head and tail. We need to know how many elements to splice off each end, so that is why there are restrictions on slurpies and optional parameters. Empty Signatures There is a difference between the following two methods: method foo { ...; } method foo () { ...; } In the first, you have not provided a signature and are expected to deal with @_ in the body of the method. In the second, there is a signature, but it is a signature showing that the method expects no arguments (other than the invocant of course). Optimizing Methods For a slight compile-time penalty, you can improve the speed which methods run at using the `:optimize` attribute: method foo :optimize (...) { ...; } Optimized methods must not close over any lexical (`my` or `our`) variables; they can only access the variables declared in their, signature, $self, $class, @_, and globals. They cannot access private attributes unless those private attributes have public accessors. Anonymous Methods It *is* possible to use `method` without a name to return an anonymous method (coderef): use Zydeco prefix => 'MyApp'; class MyClass { method get_method ($foo) { method ($bar) { return $foo . $bar; } } } my $obj = MyApp->new_myclass; my $anon = $obj->get_method("FOO"); say ref($anon); # CODE say $obj->$anon("BAR"); # FOOBAR Note that while $anon is a coderef, it is still a method, and still expects to be passed an object as $self. Due to limitations with Keyword::Simple, keywords are always complete statements, so `method ...` has an implicit semicolon before and after it. This means that this won't work: my $x = method { ... }; Because it gets treated as: my $x = ; method { ... }; A workaround is to wrap it in a `do { ... }` block. my $x = do { method { ... } }; Private methods A shortcut for the pattern of: my $x = do { method { ... } }; Is this: method $x { ... } Zydeco will declare the variable `my $x` for you, assign the coderef to the variable, and you don't need to worry about a `do` block to wrap it. Multimethods Multi methods should *Just Work [tm]* if you prefix them with the keyword `multi` use Zydeco prefix => 'MyApp'; class Widget { multi method foo :alias(quux) (Any $x) { say "Buzz"; } multi method foo (HashRef $h) { say "Fizz"; } } my $thing = MyApp->new_widget; $thing->foo( {} ); # Fizz $thing->foo( 42 ); # Buzz $thing->quux( {} ); # Buzz Outside of `class`, `abstract class`, `role`, and `interface` blocks, `multi method` will define a multi method in the caller package. (That is, usually the factory.) Multimethods cannot be anonymous or private. `requires` Indicates that a role requires classes to fulfil certain methods. role Payable { requires account; requires deposit (Num $amount); } class Employee { extends Person; with Payable; has account!; method deposit (Num $amount) { ...; } } Required methods have an optional signature; this is usually ignored, but if Devel::StrictMode determines that strict behaviour is being used, the signature will be applied to the method via an `around` modifier. Or to put it another way, this: role Payable { requires account; requires deposit (Num $amount); } Is a shorthand for this: role Payable { requires account; requires deposit; use Devel::StrictMode 'STRICT'; if (STRICT) { around deposit (Num $amount) { $self->$next(@_); } } } Can only be used in `role` and `interface` blocks. `before` before marry { say "Speak now or forever hold your peace!"; } As with `method`, $self and $class are defined. As with `method`, you can provide a signature: before marry ( Person $partner, Object $date? ) { say "Speak now or forever hold your peace!"; } Note that this will result in the argument types being checked/coerced twice; once by the before method modifier and once by the method itself. Sometimes this may be desirable, but at other times your before method modifier might not care about the types of the arguments, so can omit checking them. before marry ( $partner, $date? ) { say "Speak now or forever hold your peace!"; } Commas may be used to modify multiple methods: before marry, sky_dive (@args) { say "wish me luck!"; } The `:optimize` attribute is supported for `before`. Method modifiers do work outside of `class`, `abstract class`, `role`, and `interface` blocks, modifying methods in the caller package, which is usually the factory package. `after` There's not much to say about `after`. It's just like `before`. after marry { say "You may kiss the bride!"; } after marry ( Person $partner, Object $date? ) { say "You may kiss the bride!"; } after marry ( $partner, $date? ) { say "You may kiss the bride!"; } Commas may be used to modify multiple methods: after marry, finished_school_year (@args) { $self->go_on_holiday(); } The `:optimize` attribute is supported for `after`. Method modifiers do work outside of `class`, `abstract class`, `role`, and `interface` blocks, modifying methods in the caller package, which is usually the factory package. `around` The `around` method modifier is somewhat more interesting. around marry ( Person $partner, Object $date? ) { say "Speak now or forever hold your peace!"; my $return = $self->$next(@_); say "You may kiss the bride!"; return $return; } The $next variable holds a coderef pointing to the "original" method that is being modified. This gives your method modifier the ability to munge the arguments seen by the "original" method, and munge any return values. (I say "original" in quote marks because it may not really be the original method but another wrapper!) $next and $self are both shifted off @_. If you use the signature-free version then $next and $self are not shifted off @_ for you, but the variables are still defined. around marry { say "Speak now or forever hold your peace!"; my $return = $self->$next($_[2], $_[3]); say "You may kiss the bride!"; return $return; } Commas may be used to modify multiple methods: around insert, update ($dbh, @args) { $dbh->begin_transaction; my $return = $self->$next(@_); $dbh->commit_transaction; return $return; } The `:optimize` attribute is supported for `around`. Note that `SUPER::` won't work as expected in Zydeco, so `around` should be used instead. Method modifiers do work outside of `class`, `abstract class`, `role`, and `interface` blocks, modifying methods in the caller package, which is usually the factory package. `factory` The `factory` keyword is used to define alternative constructors for your class. class Person { has name ( type => Str, required => true ); has gender ( type => Str ); factory new_man (Str $name) { return $class->new(name => $name, gender => 'male'); } factory new_woman (Str $name) { return $class->new(name => $name, gender => 'female'); } } But here's the twist. These methods are defined within the factory package, not within the class. So you can call: MyApp->new_man("Bob") # yes But not: MyApp::Person->new_man("Bob") # no Note that if your class defines *any* factory methods like this, then the default factory method (in this case `MyApp->new_person` will no longer be automatically created. But you can create the default one easily: class Person { has name ( type => Str, required => true ); has gender ( type => Str ); factory new_man (Str $name) { ... } factory new_woman (Str $name) { ... } factory new_person; # no method signature or body! } Within a factory method body, the variable $class is defined, just like normal methods, but $self is not defined. There is also a variable $factory which is a string containing the factory package name. This is because you sometimes need to create more than just one object in a factory method. class Wheel; class Car { has wheels = []; factory new_three_wheeler () { return $class->new( wheels => [ $factory->new_wheel, $factory->new_wheel, $factory->new_wheel, ] ); } factory new_four_wheeler () { return $class->new( wheels => [ $factory->new_wheel, $factory->new_wheel, $factory->new_wheel, $factory->new_wheel, ] ); } } As with `method` and the method modifiers, if you provide a signature, $factory and $class will be shifted off @_. If you don't provide a signature, the variables will be defined, but not shifted off @_. An alternative way to provide additional constructors is with `method` and then use `factory` to proxy them. class Person { has name ( type => Str, required => true ); has gender ( type => Str ); method new_guy (Str $name) { ... } method new_gal (Str $name) { ... } factory new_person; factory new_man via new_guy; factory new_woman via new_gal; } Now `MyApp->new_man` will call `MyApp::Person->new_guy`. `factory new_person` with no `via` or method body is basically like saying `via new`. The `:optimize` attribute is supported for `factory`. The `factory` keyword can only be used inside `class` blocks. Implementing a singleton Factories make it pretty easy to implement the singleton pattern. class AppConfig { ...; factory get_appconfig () { state $config = $class->new(); } } Now `MyApp->get_appconfig` will always return the same AppConfig object. Because any explicit use of the `factory` keyword in a class definition suppresses the automatic creation of a factory method for the class, there will be no `MyApp->new_appconfig` method for creating new objects of that class. (People can still manually call `MyApp::AppConfig->new` to get a new AppConfig object, but remember Zydeco discourages calling constructors directly, and encourages you to use the factory package for instantiating objects!) `type_name` class Homo::Sapiens { type_name Human; } The class will still be called MyApp::Homo::Sapiens but the type in the type library will be called Human instead of Homo_Sapiens. Can only be used in `class`, `abstract class`, `role`, and `interface` blocks. `coerce` class Person { has name ( type => Str, required => true ); has gender ( type => Str ); coerce from Str via from_string { $class->new(name => $_); } } class Company { has owner ( type => 'Person', required => true ); } my $acme = MyApp->new_company( owner => "Bob" ); Note that the company owner is supposed to be a person object, not a string, but the Person class knows how create a person object from a string. Coercions are automatically enabled in a lot of places for types that have a coercion. For example, types in signatures, and types in attribute definitions. Note that the coercion body doesn't allow signatures, and the value being coerced will be found in $_. If you want to have signatures, you can define a coercion as a normal method first: class Person { has name ( type => Str, required => true ); has gender ( type => Str ); method from_string ( Str $name ) { $class->new(name => $name); } coerce from Str via from_string; } In both cases, a `MyApp::Person->from_string` method is generated which can be called to manually coerce a string into a person object. They keyword `from` is technically optional, but does make the statement more readable. coerce Str via from_string { # this works $class->new(name => $_); } The `:optimize` attribute is not currently supported for `coerce`. Can only be used in `class` blocks. `overload` class Collection { has items = []; overload '@{}' => sub { shift->list }; } The list passed to `overload` is passed to overload with no other processing. Can only be used in `class` blocks. `version` class Person { version '1.0'; } This just sets $MyApp::Person::VERSION. Can only be used with `class`, `abstract class`, `role`, and `interface` blocks. You can set a default version for all packages like this: use Zydeco ( ..., version => 1.0, ); If `class` definitions are nested, `version` will be inherited by child classes. If a parent class is specified via `extends`, `version` will not be inherited. Versions may also be expressed like: class Animal 1.0; class Person 1.1 extends Animal { ...; } `authority` class Person { authority 'cpan:TOBYINK'; } This just sets $MyApp::Person::AUTHORITY. It is used to indicate who is the maintainer of the package. Can only be used in `class`, `abstract class`, `role`, and `interface` blocks. use Zydeco ( ..., version => 1.0, authority => 'cpan:TOBYINK', ); If `class` definitions are nested, `authority` will be inherited by child classes. If a parent class is specified via `extends`, `authority` will not be inherited. `include` `include` is the Zydeco equivalent of Perl's `require`. package MyApp { use Zydeco; include Database; include Classes; include Roles; } It works somewhat more crudely than `require` and `use`, evaluating the included file pretty much as if it had been copy and pasted into the file that included it. The names of the files to load are processsed using the same rules for prefixes as classes and roles (so MyApp::Database, etc in the example), and @INC is searched just like `require` and `use` do, but instead of looking for a file called "MyApp/Database.pm", Zydeco will look for "MyApp/Database.pl" (yes, ".pl"). This naming convention ensures people won't accidentally load MyApp::Database using `use` or `require` because it isn't intended to be loaded outside the context of the MyApp package. The file "MyApp/Database.pl" might look something like this: class Database { has dbh = DBI->connect(...); factory get_db { state $instance = $class->new; } } Note that it doesn't start with a `package` statement, nor `use Zydeco`. It's just straight on to the definitions. There's no `1;` needed at the end. `use strict` and `use warnings` are safe to put in the file if you need them to satisfy linters, but they're not necessary because the contents of the file are evaluated as if they had been copied and pasted into the main MyApp module. There are *no* checks to prevent a file from being included more than once, and there are *no* checks to deal with cyclical inclusions. Inclusions are currently only supported at the top level, and not within class and role definitions. `Zydeco::PACKAGE_SPEC()` This function can be used while a class or role is being compiled to tweak the specification for the class/role. class Foo { has foo; Zydeco::PACKAGE_SPEC->{has}{foo}{type} = Int; } It returns a hashref of attributes, methods, etc. MooX::Press should give you an idea about how the hashref is structured, but Zydeco only supports a subset of what MooX::Press supports. For example, MooX::Press allows `has` to be an arrayref or a hashref, but Zydeco only supports a hashref. The exact subset that Zydeco supports is subject to change without notice. This can be used to access MooX::Press features that Zydeco doesn't expose. IMPORTS Zydeco also exports constants `true` and `false` into your namespace. These show clearer boolean intent in code than using 1 and 0. Zydeco exports `rw`, `ro`, `rwp`, `lazy`, `bare`, and `private` constants which make your attribute specs a little cleaner looking. Zydeco exports `blessed` from Scalar::Util because that can be handy to have, and `confess` from Carp. Zydeco's copy of `confess` is super-powered and runs its arguments through `sprintf`. before vote { if ($self->age < 18) { confess("Can't vote, only %d", $self->age); } } Zydeco turns on strict, warnings, and the following modern Perl features: # Perl 5.14 and Perl 5.16 say state unicode_strings # Perl 5.18 or above say state unicode_strings unicode_eval evalbytes current_sub fc If you're wondering why not other features, it's because I didn't want to enable any that are currently classed as experimental, nor any that require a version of Perl above 5.18. The `switch` feature is mostly seen as a failed experiment these days, and `lexical_subs` cannot be called as methods so are less useful in object-oriented code. You can, of course, turn on extra features yourself. package MyApp { use Zydeco; use feature qw( lexical_subs postderef ); ...; } (The `current_sub` feature is unlikely to work fully unless you have `:optimize` switched on for that method, or the method does not include a signature. For non-optimized methods with a signature, a wrapper is installed that handles checks, coercions, and defaults. `__SUB__` will point to the "inner" sub, minus the wrapper.) Zydeco exports Syntax::Keyword::Try for you. Useful to have. And last but not least, it exports all the types, `is_*` functions, and `assert_*` functions from Types::Standard, Types::Common::String, and Types::Common::Numeric. You can choose which parts of Zydeco you import: package MyApp { use Zydeco keywords => [qw/ -booleans -privacy -utils -types -is -assert -features try class abstract role interface include toolkit begin end extends with requires has constant method multi factory before after around type_name coerce version authority overload /]; It should mostly be obvious what they all do, but `-privacy` is `ro`, `rw`, `rwp`, etc; `-types` is bareword type constraints (though even without this export, they should work in method signatures), `-is` are the functions like `is_NonEmptyStr` and `is_Object`, `-assert` are functions like `assert_Int`, `-utils` gives you `blessed` and `confess`. `no Zydeco` will clear up: class abstract role interface include toolkit begin end extends with requires has constant method multi factory before after around type_name coerce version authority overload But won't clear up things Zydeco imported for you from other packages. Use `no MooX::Press::Keywords`, `no Types::Standard`, etc to do that, or just use namespace::autoclean. FEATURES Helper Subs Earlier it was stated that `sub` cannot be used to define methods in classes and roles. This is true, but that doesn't mean that it has no use. package MyApp { use Zydeco; sub helper_function { ... } class Foo { method foo { ...; helper_function(...); ...; } } class Bar { sub other_helper { ... } method bar { ...; helper_function(...); other_helper(...); ...; } } } The subs defined by `sub` end up in the "MyApp" package, not "MyApp::Foo" or "MyApp::Bar". They can be called by any of the classes and roles defined in MyApp. This makes them suitable for helper subs like logging, List::Util/Scalar::Util sorts of functions, and so on. package MyApp { use Zydeco; use List::Util qw( any all first reduce ); # the above functions are now available within # all of MyApp's classes and roles, but won't # pollute any of their namespaces. use namespace::clean; # And now they won't even pollute MyApp's namespace. # Though I'm pretty sure this will also stop them # from working in any methods that used ":optimize". class Foo { ... } role Bar { ... } role Baz { ... } } `sub` is also usually your best option for those tiny little coderefs that need to be defined here and there: has foo ( is => lazy, type => ArrayRef[Str], builder => sub { [] }, ); Though consider using Sub::Quote if you're using Moo. Anonymous Classes and Roles Anonymous classes It is possible to make anonymous classes: my $class = do { class; }; my $object = $class->new; The `do { ... }` block is necessary because of a limitation in Keyword::Simple, where any keywords it defines must be complete statements. Anonymous classes can have methods and attributes and so on: my $class = do { class { has foo (type => Int); has bar (type => Int); }}; my $object = $class->new(foo => 1, bar => 2); Anonymous classes *do not* implicitly inherit from their parent like named nested classes do. Named classes nested inside anonymous classes *do not* implicitly inherit from the anonymous class. Having one anonymous class inherit from another can be done though: my $base = do { class; } my $derived = do { class { extends {"::$k1"}; }}; This works because `extends` accepts a block which returns a string for the package name, and the string needs to begin with "::" to avoid the auto prefix mechanism. Anonymous roles Anonymous roles work in much the same way. Parameterizable Classes and Roles Parameterizable classes package MyApp { use Zydeco; class Animal { has name; } class Species ( Str $common_name, Str $binomial ) { extends Animal; constant common_name = $common_name; constant binomial = $binomial; } class Dog { extends Species('dog', 'Canis familiaris'); method bark () { say "woof!"; } } } Here, "MyApp::Species" isn't a class in the usual sense; you cannot create instances of it. It's like a template for generating classes. Then "MyApp::Dog" generates a class from the template and inherits from that. my $Cat = MyApp->generate_species('cat', 'Felis catus'); my $mog = $Cat->new(name => 'Mog'); $mog->isa('MyApp::Animal'); # true $mog->isa('MyApp::Species'); # false!!! $mog->isa($Cat); # true Because there are never any instances of "MyApp::Species", it doesn't make sense to have a Species type constraint. Instead there are SpeciesClass and SpeciesInstance type constraints. use MyApp::Types -is; my $lassie = MyApp->new_dog; is_Animal( $lassie ); # true is_Dog( $lassie ); # true is_SpeciesInstance( $lassie ); # true is_SpeciesClass( ref($lassie) ); # true Subclasses cannot be nested inside parameterizable classes, but parameterizable classes can be nested inside regular classes, in which case the classes they generate will inherit from the outer class. package MyApp { use Zydeco; class Animal { has name; class Species ( Str $common_name, Str $binomial ) { constant common_name = $common_name; constant binomial = $binomial; } } class Dog { extends Species('dog', 'Canis familiaris'); method bark () { say "woof!"; } } } Anonymous parameterizable classes are possible: my $generator = do { class ($footype, $bartype) { has foo (type => $footype); has bar (type => $bartype); } }; my $class = $generator->generate_package(Int, Num); my $object = $class->new(foo => 42, bar => 4.2); Parameterizable roles Often it makes more sense to parameterize roles than classes. package MyApp { use Zydeco; class Animal { has name; } role Species ( Str $common_name, Str $binomial ) { constant common_name = $common_name; constant binomial = $binomial; } class Dog { extends Animal; with Species('dog', 'Canis familiaris'), GoodBoi?; method bark () { say "woof!"; } } } Anonymous parameterizable roles are possible. BUGS Please report any bugs to . TODO Plugin system Zydeco can often load MooX/MouseX/MooseX plugins and work fine with them, but some things won't work, like plugins that rely on being able to wrap `has`. So it would be nice to have a plugin system that extensions can hook into. If you're interested in extending Zydeco, file a bug report about it and let's have a conversation about the best way for that to happen. I probably won't start a plugin API until someone actually wants to write a plugin, because that will give me a better idea about what kind of API is required. SEE ALSO Zydeco website: . Less magic version: MooX::Press. Important underlying technologies: Moo, Type::Tiny::Manual, Sub::HandlesVia, Sub::MultiMethod, Lexical::Accessor, Syntax::Keyword::Try. Similar modules: Moops, Kavorka, Dios, MooseX::Declare. Zydeco vs MooX::Press Zydeco is mostly just a syntax wrapper over MooX::Press. You can see the class and role definition that Zydeco passes to MooX::Press using: use Zydeco debug => 1; Rewriting your classes and roles to use MooX::Press directly would give you a small speed boost, and fewer dependencies, though MooX::Press is a little more frustrating to work with, needing a *lot* of punctuation. Zydeco vs Moops Because I also wrote Moops, people are likely to wonder what the difference is, and why re-invent the wheel? Zydeco has fewer dependencies than Moops, and crucially doesn't rely on Package::Keyword and Devel::CallParser which have... issues. Zydeco uses Damian Conway's excellent PPR to handle most parsing needs, so parsing should be more predictable. Moops is faster in most circumstances though. Here are a few key syntax and feature differences. Declaring a class Moops and Zydeco use different logic for determining whether a class name is "absolute" or "relative". In Moops, classes containing a "::" are seen as absolute class names; in Zydeco, only classes *starting with* "::" are taken to be absolute; all others are given the prefix. Moops: package MyApp { use Moops; class Foo { class Bar { class Baz { # Nesting class blocks establishes a naming # heirarchy so this is MyApp::Foo::Bar::Baz! } } } } Zydeco: package MyApp { use Zydeco; class Foo { class Bar { class Baz { # This is only MyApp::Baz, but nesting # establishes an @ISA chain instead. } } } } How namespacing works Moops: use feature 'say'; package MyApp { use Moops; use List::Util qw(uniq); class Foo { say __PACKAGE__; # MyApp::Foo say for uniq(1,2,1,3); # ERROR! sub foo { ... } # MyApp::Foo::foo() } } Zydeco: use feature 'say'; package MyApp { use Zydeco; use List::Util qw(uniq); class Foo { say __PACKAGE__; # MyApp say for uniq(1,2,1,3); # this works fine sub foo { ... } # MyApp::foo() } } This is why you can't use `sub` to define methods in Zydeco. You need to use the `method` keyword. In Zydeco, all the code in the class definition block is still executing in the parent package's namespace! Multimethods Moops/Kavorka multimethods are faster, but Zydeco is smarter at picking the best candidate to dispatch to, and intelligently selecting candidates across inheritance hierarchies and role compositions. Other crazy Kavorka features Kavorka allows you to mark certain parameters as read-only or aliases, allows you to specify multiple names for named parameters, allows you to rename the invocant, allows you to give methods and parameters attributes, allows you to specify a method's return type, etc, etc. Zydeco's `method` keyword is unlikely to ever offer as many features as that. It is unlikely to offer many more features than it currently offers. If you need fine-grained control over how @_ is handled, just don't use a signature and unpack @_ inside your method body however you need to. Lexical accessors Zydeco has tighter integration with Lexical::Accessor, allowing you to use the same keyword `has` to declare private and public attributes. Factories Zydeco puts an emphasis on having a factory package for instantiating objects. Moops didn't have anything similar. `augment` and `override` These are Moose method modifiers that are not implemented by Moo. Moops allows you to use these in Moose and Mouse classes, but not Moo classes. Zydeco simply doesn't support them. Type Libraries Moops allowed you to declare multiple type libraries, define type constraints in each, and specify for each class and role which type libraries you want it to use. Zydeco automatically creates a single type library for all your classes and roles within a module to use, and automatically populates it with the types it thinks you might want. If you need to use other type constraints: package MyApp { use Zydeco; # Just import types into the factory package! use Types::Path::Tiny qw( Path ); class DataSource { has file => ( type => Path ); method set_file ( Path $file ) { $self->file( $file ); } } } my $ds = MyApp->new_datasource; $ds->set_file('blah.txt'); # coerce Str to Path print $ds->file->slurp_utf8; Constants Moops: class Foo { define PI = 3.2; } Zydeco: class Foo { constant PI = 3.2; } Parameterizable classes and roles These were always on my todo list for Moops; I doubt they'll ever be done. They work nicely in Zydeco though. AUTHOR Toby Inkster . COPYRIGHT AND LICENCE This software is copyright (c) 2020 by Toby Inkster. This is free software; you can redistribute it and/or modify it under the same terms as the Perl 5 programming language system itself. DISCLAIMER OF WARRANTIES THIS PACKAGE IS PROVIDED "AS IS" AND WITHOUT ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE.