NAME
    Class::Struct::FIELDS - Combine Class::Struct, base and fields

SYNOPSIS
    (This page documents 'Class::Struct::FIELDS' v.1.0.)

        use Class::Struct::FIELDS;
                # declare struct, based on fields, explicit class name:
        struct (CLASS_NAME => { ELEMENT_NAME => ELEMENT_TYPE, ... });

        use Class::Struct::FIELDS;
                # declare struct, based on fields, explicit class name
                # with inheritance:
        struct (CLASS_NAME => [qw(BASE_CLASSES ...)],
                { ELEMENT_NAME => ELEMENT_TYPE, ... });

        package CLASS_NAME;
        use Class::Struct::FIELDS;
                # declare struct, based on fields, implicit class name:
        struct (ELEMENT_NAME => ELEMENT_TYPE, ...);

        package CLASS_NAME;
        use Class::Struct::FIELDS;
                # declare struct, based on fields, implicit class name
                # with inheritance:
        struct ([qw(BASE_CLASSES ...)], ELEMENT_NAME => ELEMENT_TYPE, ...);

        package MyObj;
        use Class::Struct::FIELDS;
                # declare struct with four types of elements:
        struct (s => '$', a => '@', h => '%', x => '&', c => 'My_Other_Class');

        $obj = new MyObj;               # constructor

                                        # scalar type accessor:
        $element_value = $obj->s;           # element value
        $obj->s ('new value');              # assign to element

                                        # array type accessor:
        $ary_ref = $obj->a;                 # reference to whole array
        $ary_element_value = $obj->a->[2];  # array element value
        $ary_element_value = $obj->a (2);   # same thing
        $obj->a->[2] = 'new value';         # assign to array element
        $obj->a (2, 'newer value');         # same thing

                                        # hash type accessor:
        $hash_ref = $obj->h;                # reference to whole hash
        $hash_element_value = $obj->h->{x}; # hash element value
        $hash_element_value = $obj->h (x);  # same thing
        $obj->h->{x} = 'new value';         # assign to hash element
        $obj->h (x, 'newer value');         # same thing

                                        # code type accessor:
        $code_ref = $obj->x;                # reference to code
        $obj->x->(...);                     # call code
        $obj->x (sub {...});                # assign to element

                                        # regexp type accessor:
        $regexp = $obj->r;                  # reference to code
        $string =~ m/$obj->r/;              # match regexp
        $obj->r (qr/ ... /);                # assign to element

                                        # class type accessor:
        $element_value = $obj->c;            # object reference
        $obj->c->method (...);               # call method of object
        $obj->c (My_Other_Class::->new);     # assign a new object

DESCRIPTION
    'Class::Struct::FIELDS' exports a single function, 'struct'. Given a
    list of element names and types, and optionally a class name and/or an
    array reference of base classes, 'struct' creates a Perl 5 class that
    implements a "struct-like" data structure with inheritance.

    The new class is given a constructor method, 'new', for creating struct
    objects.

    Each element in the struct data has an accessor method, which is used to
    assign to the element and to fetch its value. The default accessor can
    be overridden by declaring a 'sub' of the same name in the package. (See
    Example 2.)

    Each element's type can be scalar, array, hash, code or class.

  Differences from 'Class::Struct' 'base' and 'fields'

    'Class::Struct::FIELDS' is a combination of 'Class::Struct', 'base' and
    'fields'.

    Unlike 'Class::Struct', inheritance is explicitly supported, and there
    is better support for user overrides of constructed accessor methods.
    One result is that you may no longer use the array ('[]') notation for
    indicating internal representation. Also, 'Class::Struct::FIELDS' relies
    on 'fields' for internal representation.

    Also, 'Class::Struct::FIELDS' supports code and regular expression
    elements. ('Class::Struct' handles code and regular expressions as
    scalars.)

    Lastly, 'Class::Struct::FIELDS' passes it's import list, if any, from
    the call to 'use Class::Struct::FIELDS ...' to 'struct' so that you may
    create new packages at compile-time.

    Unlike 'fields', each element has a data type, and is automatically
    created at first access.

  Calling 'use Class::Struct::FIELDS'

    You may call 'use Class::Struct::FIELDS' just as with any module
    library:

        use Class::Struct::FIELDS;
        struct Bob => [];

    However, if you try 'my Dog $spot' syntax with this example:

        use Class::Struct::FIELDS;
        struct Bob => [];
        my Bob $bob = Bob::->new;

    you will get a compile-time error:

        No such class Bob at <filename> line <number>, near "my Bob"
        Bareword "Bob::" refers to nonexistent package at <filename> line
        <number>.

    since the compiler has not seen your class declarations yet until after
    the call to 'struct', by which time it has already seen your 'my'
    declarations. Oops, too late. Instead, create the package for 'Bob'
    during compilation:

        use Class::Struct::FIELDS qw(Bob);
        my Bob $bob = Bob::->new;

    This compiles without error as 'import' for 'Class::Struct::FIELDS'
    calls 'struct' for you if you have any arguments in the 'use' statement.
    A more interesting example is:

        use Class::Struct::FIELDS Bob => { a => '$' };
        use Class::Struct::FIELDS Fred => [qw(Bob)];
        my Bob $bob = Bob::->new;
        my Fred $fred = Fred::->new;

  The 'struct' subroutine

    The 'struct' subroutine has three forms of parameter-list:

        struct (CLASS_NAME => { ELEMENT_LIST });
        struct (CLASS_NAME, ELEMENT_LIST);
        struct (ELEMENT_LIST);

    The first form explicitly identifies the name of the class being
    created. The second form is equivalent. The second form assumes the
    current package name as the class name. The second and third forms are
    distinguished by the parity of the argument list: an odd number of
    arguments is taken to be of the second form.

    Optionally, you may specify base classes with an array reference as the
    first non-class-name argument:

        struct (CLASS_NAME => [qw(BASE_CLASSES ...)], { ELEMENT_LIST });
        struct (CLASS_NAME => [qw(BASE_CLASSES ...)], ELEMENT_LIST);
        struct ([qw(BASE_CLASSES ...)], { ELEMENT_LIST });
        struct ([qw(BASE_CLASSES ...)], ELEMENT_LIST);

    (Since there is no ambiguity between CLASS_NAME and ELEMENT_LIST with
    the interposing array reference, you may always make ELEMENT_LIST a list
    or a hash reference with this form.)

    The class created by 'struct' may be either a subclass or superclass of
    other classes. See the base manpage and the fields manpage for details.

    The *ELEMENT_LIST* has the form

        NAME => TYPE, ...

    Each name-type pair declares one element of the struct. Each element
    name will be usually be defined as an accessor method of the same name
    as the field, unless a method by that name is explicitly defined (called
    a "user override") by the caller prior to the 'use' statement for
    'Class::Struct::FIELDS'. (See the Replacing member access methods with
    user overrides entry elsewhere in this document.)

    'struct' returns the name of the newly-constructed package.

  Element Types and Accessor Methods

    The five element types -- scalar, array, hash, code and class -- are
    represented by strings -- '$', '@', '%', '&', '/' and a class name.

    The accessor method provided by 'struct' for an element depends on the
    declared type of the element.

    Scalar ('$', '\$' or '*$')
        The element is a scalar, and by default is initialized to 'undef'
        (but see the Initializing with new entry elsewhere in this
        document).

        The accessor's argument, if any, is assigned to the element.

        If the element type is '$', the value of the element (after
        assignment) is returned. If the element type is '\$' or '*$', a
        reference to the element is returned.

    Array ('@', '\@' or '*@')
        The element is an array, initialized by default to '()'.

        With no argument, the accessor returns a reference to the element's
        whole array (whether or not the element was specified as '@', '\@'
        or '*@').

        With one or two arguments, the first argument is an index specifying
        one element of the array; the second argument, if present, is
        assigned to the array element. If the element type is '@', the
        accessor returns the array element value. If the element type is
        '\@' or '*@', a reference to the array element is returned.

    Hash ('%', '\%' or '*%')
        The element is a hash, initialized by default to '()'.

        With no argument, the accessor returns a reference to the element's
        whole hash (whether or not the element was specified as '%', '\%' or
        '*%').

        With one or two arguments, the first argument is a key specifying
        one element of the hash; the second argument, if present, is
        assigned to the hash element. If the element type is '%', the
        accessor returns the hash element value. If the element type is '\%'
        or '*%', a reference to the hash element is returned.

    Code ('&', '\&' or '*&')
        The element is code, and by default is initialized to 'undef' (but
        see the Initializing with new entry elsewhere in this document).

        The accessor's argument, if any, is assigned to the element.

        If the element type is '&', the value of the element (after
        assignment) is returned. If the element type is '\&' or '*&', a
        reference to the element is returned. (It is unclear of what value
        this facility is. XXX)

    Regexp ('/', '\/' or '*/')
        If the element type is '/', the value of the element (after
        assignment) is returned. If the element type is '\/' or '*/', a
        reference to the element is returned. (It is unclear of what value
        this facility is. XXX)

        Regular expressions really are special in that you create them with
        special syntax, not with a call to a constructor:

          $obj->r (qr/^$/); # fine
          $obj->r (Regexp->new); # WRONG

    Class ('Class_Name', '\Class_Name' or '*Class_Name')
        The element's value must be a reference blessed to the named class
        or to one of its subclasses. The element is initialized to the
        result of calling the 'new' constructor of the named class.

        The accessor's argument, if any, is assigned to the element. The
        accessor will 'croak' if this is not an appropriate object
        reference.

        If the element type does not start with a '\' or '*', the accessor
        returns the element value (after assignment). If the element type
        starts with a '\' or '*', a reference to the element itself is
        returned.

        The class is automatically required for you so that, for example,
        you can safely write:

            struct MyObj {io => 'IO::Scalar'};

        and access 'io' immediately. The same applies for nested structs:

            BEGIN {
              struct Alice { when => '$' };
              struct Bob { who => 'Alice' };
            }

            my Bob $b = Bob::->new;
            $b->who->when ('what');

        Note, however, the 'BEGIN' block so that this example can use the
        'my Dog $spot' syntax for 'my Bob $b'. Also, no actual import
        happens for the caller -- the automatic use is only for convenience
        in auto-constructing members, not magic. Another way to do this is:

            { package Bob; use Class::Struct::FIELDS; struct }
            my Bob $b = Bob::->new;

        And of course the best way to do this is simply:

            use Class::Struct::FIELDS qw(Bob);
            my Bob $b = Bob::->new;

    What about globs ('*') and other funny types?
        At present, 'Class::Struct::FIELDS' does not support special
        notation for other intrinsic types. Use a scalar to hold a reference
        to globs and other unusual specimens, or wrap them in a class such
        as 'IO::Handle' (globs). XXX

  Initializing with 'new'

    'struct' always creates a constructor called 'new'. That constructor may
    take a list of initializers for the various elements of the new struct.

    Each initializer is a pair of values: *element name*' => '*value*. The
    initializer value for a scalar element is just a scalar value. The
    initializer for an array element is an array reference. The initializer
    for a hash is a hash reference. The initializer for code is a code
    reference.

    The initializer for a class element is also a hash reference, and the
    contents of that hash are passed to the element's own constructor.

    'new' tries to be as clever as possible in deducing what type of object
    to construct. All of these are valid:

        use Class::Struct::FIELDS qw(Bob);
        my Bob $b = Bob::->new; # good style
        my Bob $b2 = $b->new; # works fine
        my Bob $b3 = &Bob::new; # if you insist
        my Bob $b4 = Bob::new (apple => 3, banana => 'four'); # WRONG!

    The last case doesn't behave as hoped for: 'new' tries to construct an
    object of package 'apple' (and hopefully fails, unless you actually have
    a package named 'apple'), not an object of package 'Bob'.

    See Example 3 below for an example of initialization.

  Initializing with 'init'

    You may also use 'init' as a constructor to assign initial values to new
    objects. (In fact, this is the preferred method.) 'struct' will see to
    it that you have a ready object to work with, and pass you any arguments
    used in the call to 'new':

      sub init {
        my MyObj $self = shift;

        @self->a->[0..3] = (a..d);

        return $self;
      }

    It is essential that you return an object from 'init', as this is
    returned to the caller of 'new'. You may return a different object if
    you wish, but this would be rather uncommon.

    First, 'new' arranges for any constructor argument list to be processed
    first before calling 'init'.

    Second, 'new' arranges to call 'init' for base classes, calling them in
    bottom-up order, before calling 'init'. This is so that ancestors may
    construct an object before descendents.

    There is no corresponding facility for DESTROY. XXX

  Replacing member access methods with user overrides

    You might want to create custom access methods, or user overrides. The
    most straight forward way to do this and still retain 'string' and
    'warnings' is:

        use strict;
        use warnings;

        sub Bob::ff ($;$$); # declare sub so Class::Struct::FIELDS can see

        use Class::Struct::FIELDS Bob => { ff => '$' };

        sub Bob::ff ($;$$) {
          my Bob $self = shift;

          &some_other_sub (@_);
        }

    If you do not declare the user override prior to the 'use' statement, a
    warning is issued if the warning flag (-w) is set.

    Notice that we changed the default sub signature for *ff* from '($;$)'
    to '($;$$)'. Normally, this might generate a warning if we redefine the
    sub, but declaring the sub ahead of time keeps 'strict' and 'warnings'
    happy. You might prefer this construction:

        { package Bob; }

        sub Bob::ff ($;$$) {
          my Bob $self = shift;

          &some_other_sub (@_);
        }

        use Class::Struct::FIELDS Bob => { ff => '$' };

    You might still want the advantages of the the constructed accessor
    methods, even with user overrides (for example, checking that an
    assigned value is the right type or package). 'Class::Struct::FIELDS'
    constructs the accessor with a special name, so that you may use it
    yourself in the user override. That special name is the regular field
    name prepended by a double underscore, '__'. You can access these so:

        use strict;
        use warnings;

        sub Bob::ff ($;$); # declare sub so Class::Struct::FIELDS can see
        sub Bob::gg ($;$); # declare sub so Class::Struct::FIELDS can see

        use Class::Struct::FIELDS Bob => { ff => '$', gg => '$' };

        # This example is identical to having no user override.
        sub Bob::ff ($;$) {
          my Bob $self = shift;
          $self->__ff (@_);
        }

        # This example illustrates a workaround for v5.6.0.
        sub Bob::gg ($;$) {
          # This silliness is due to a bug in 5.6.0: it thinks you can't
          # fiddle with @_ if you've given it a prototype.  XXX
          my @args = @_;
          $args[1] *= 2 if @args == 2 and defined $args[1];
          @_ = @args;
          goto &Bob::__gg;
        }

  Private fields

    Fields starting with a leading underscore, '_', are private: they are
    still valid fields, but 'Class::Struct::FIELDS' does not create
    subroutines to access them. Instead, you should access them the usual
    way for hash members:

        $self->{_private_key}; # ok
        $self->_private_key; # Compilation error

    See the fields manpage for more details.

EXAMPLES
    Example 1
        Giving a struct element a class type that is also a struct is how
        structs are nested. Here, 'timeval' represents a time (seconds and
        microseconds), and 'rusage' has two elements, each of which is of
        type 'timeval'.

            use Class::Struct::FIELDS;

            struct (rusage => {
              ru_utime => timeval,  # seconds
              ru_stime => timeval,  # microseconds
            });

            struct (timeval => {
              tv_secs  => '$',
              tv_usecs => '$',
            });

                # create an object:
            my $t = new rusage;

                # $t->ru_utime and $t->ru_stime are objects of type timeval.
                # set $t->ru_utime to 100.0 sec and $t->ru_stime to 5.0 sec.
            $t->ru_utime->tv_secs (100);
            $t->ru_utime->tv_usecs (0);
            $t->ru_stime->tv_secs (5);
            $t->ru_stime->tv_usecs (0);

    Example 2
        An accessor function can be redefined in order to provide additional
        checking of values, etc. Here, we want the 'count' element always to
        be nonnegative, so we redefine the 'count' accessor accordingly.

            package MyObj;
            use Class::Struct::FIELDS;

            # declare the struct
            struct (MyObj => {count => '$', stuff => '%'});

            # override the default accessor method for 'count'
            sub count {
              my MyObj $self = shift;

              if (@_) {
                die 'count must be nonnegative' if $_[0] < 0;
                $self->{count} = shift;
                warn "Too many args to count" if @_;
              }

              return $self->{count};
            }

            package main;
            $x = new MyObj;
            print "\$x->count (5) = ", $x->count (5), "\n";
                                    # prints '$x->count (5) = 5'

            print "\$x->count = ", $x->count, "\n";
                                    # prints '$x->count = 5'

            print "\$x->count (-5) = ", $x->count (-5), "\n";
                                    # dies due to negative argument!

    Example 3
        The constructor of a generated class can be passed a list of
        *element*=>*value* pairs, with which to initialize the struct. If no
        initializer is specified for a particular element, its default
        initialization is performed instead. Initializers for non-existent
        elements are silently ignored.

        Note that the initializer for a nested struct is specified as an
        anonymous hash of initializers, which is passed on to the nested
        struct's constructor.

            use Class::Struct::FIELDS;

            struct Breed =>
            {
              name  => '$',
              cross => '$',
            };

            struct Cat =>
            {
              name     => '$',
              kittens  => '@',
              markings => '%',
              breed    => 'Breed',
            };

            my $cat = Cat->new
              (name     => 'Socks',
               kittens  => ['Monica', 'Kenneth'],
               markings => { socks => 1, blaze => "white" },
               breed    => { name => 'short-hair', cross => 1 });

            print "Once a cat called ", $cat->name, "\n";
            print "(which was a ", $cat->breed->name, ")\n";
            print "had two kittens: ", join(' and ', @{$cat->kittens}), "\n";

    Example 4
        'Class::Struct::FIELDS' has a very elegant idiom for creating
        inheritance trees:

            use Class::Struct::FIELDS Fred => [];
            use Class::Struct::FIELDS Barney => [qw(Fred)];
            use Class::Struct::FIELDS Wilma => [qw(Barney)],
              { aa => '@',
                bb => 'IO::Scalar' };

        That's all the code it takes!

EXPORTS
    'Class::Struct::FIELDS' export 'struct' for backwards-compatibility with
    'Class::Struct'.

DIAGNOSTICS
    The following are diagnostics generated by Class::Struct::Fields. Items
    marked "(W)" are non-fatal (invoke 'Carp::carp'); those marked "(F)" are
    fatal (invoke 'Carp::croak').

    'struct' usage error
        (F) The caller failed to read the documentation for
        'Class::Struct::FIELDS' and follow the advice therein.

    Accessor '%s' exists in package '%s' hides method in base class
        (W) There is already a subroutine, with the name of one of the
        accessors, located in a base class of the given package. You should
        consider renaming the field with the given name.

    Method '%s' exists in package '%s' overrides accessor
        (W) There is already a subroutine, with the name of one of the
        accessors, located in the given package. You may have intended this,
        however, if defining your own custom accessors.

    Method 'new' already exists in package '%s'
        (W) There is already a 'new' subroutine located in the given
        package. As long as the caveats for defining your own 'new' are
        followed, this warning is harmless; otherwise your objects may not
        be properly initialized.

    Initializer for '%s' must be %s reference
        (F) At runtime, the caller tried to assign the wrong type of
        argument to the element. An example which triggers this message:

            use Class::Struct::FIELDS Bob => { ary => '@' };
            my $b = Bob::->new;
            $b->ary ({hash => 'reference'}); # croaks

        The last statement will croak with the message, "Initializer for
        'ary' must be ARRAY reference".

    Initializer for '%s' must be %s object
        (F) At runtime, the caller tried to assign the wrong class of
        argument to the element. An example which triggers this message:

            use Class::Struct::FIELDS Bob => { mary => 'Mary' };
            use Class::Struct::FIELDS qw(Fred); # NOT inherit from Mary
            my $b = Bob::->new;
            $b->ary (Fred::->new); # croaks

        The last statement will croak with the message, "Initializer for
        'aa' must be Mary object".

BUGS AND CAVEATS
    Please see the TODO list.

CREDITS
    This documentation is amazingly like that of 'Class::Struct'. I wonder
    why. Credit to Dr. Damian Conway <damian@conway.org>.

AUTHOR
    B. K. Oxley (binkley) <binkley@bigfoot.com>

    Copyright (c) 2000 B. K. Oxley (binkley). All rights reserved. This
    program is free software; you can redistribute it and/or modify it under
    the same terms as Perl itself.

SEE ALSO
    the Class::Contract manpage
        'Class::Contract' is an extension module by Damian Conway for
        writing in a design-by-contract object-oriented style. It has many
        of the features of 'Class::Struct::FIELDS', and many more besides.

    the Class::Struct manpage
        'Class::Struct' is a standard module for creating simple,
        uninherited data structures.

    the base manpage
        'base' is a standard module for establishing IS-A relationships with
        base classes at compile time.

    the fields manpage
        'fields' is a standard module for imbuing your class with efficient
        pseudo-hashes for data members.