MooseX::AttributeShortcuts - Shorthand for common attribute options
This document describes version 0.037 of MooseX::AttributeShortcuts - released November 20, 2017 as part of MooseX-AttributeShortcuts.
package Some::Class;
use Moose;
use MooseX::AttributeShortcuts;
# same as:
# is => 'ro', lazy => 1, builder => '_build_foo'
has foo => (is => 'lazy');
# same as: is => 'ro', writer => '_set_foo'
has foo => (is => 'rwp');
# same as: is => 'ro', builder => '_build_bar'
has bar => (is => 'ro', builder => 1);
# same as: is => 'ro', clearer => 'clear_bar'
has bar => (is => 'ro', clearer => 1);
# same as: is => 'ro', predicate => 'has_bar'
has bar => (is => 'ro', predicate => 1);
# works as you'd expect for "private": predicate => '_has_bar'
has _bar => (is => 'ro', predicate => 1);
# extending? Use the "Shortcuts" trait alias
extends 'Some::OtherClass';
has '+bar' => (traits => [Shortcuts], builder => 1, ...);
Ever find yourself repeatedly specifying writers and builders, because there's
no good shortcut to specifying them? Sometimes you want an attribute to have
a read-only public interface, but a private writer. And wouldn't it be easier
to just say builder => 1 and have the attribute construct the canonical
_build_$name builder name for you?
This package causes an attribute trait to be applied to all attributes defined to the using class. This trait extends the attribute option processing to handle the above variations. All attribute options as described in Moose or Class::MOP::Attribute remain usable, just as when this trait is not applied.
Moose has long had a lazy_build attribute option. It was
once considered a best practice, but that has, ah, changed. This trait began
as a desire to still leverage bits of lazy_build (and a tacit
acknowledgment that fat-finger bugs rank among the most embarrassing, right up
there with "the TV was unplugged the entire time").
This author does not recommend you use lazy_build, unless you know exactly
what you're doing (probably) and that it's a good idea (probably not).
Nonetheless, this lazy_build option is why we set certain options the way
we do below; while lazy_build in its entirety is not optimal, it had the
right idea: regular, predictable accessor names for regular, predictable
attribute options.
As an example, just looking at the below it doesn't seem logical that:
has _foo => (is => 'ro', clearer => 1);
...becomes:
has _foo => (is => 'ro', clearer => '_clear_foo');
After reading the lazy_build attribute option, however, we see that the choice had already been made for us.
This package automatically applies an attribute metaclass trait. Simply using this package causes the trait to be applied by default to your attribute's metaclasses.
If you're extending a class and trying to extend its attributes as well,
you'll find out that the trait is only applied to attributes defined locally
in the class. This package exports a trait shortcut function Shortcuts
that will help you apply this to the extended attribute:
has '+something' => (traits => [Shortcuts], ...);
Unless specified here, all options defined by Moose::Meta::Attribute and Class::MOP::Attribute remain unchanged.
Want to see additional options? Ask, or better yet, fork on GitHub and send a pull request. If the shortcuts you're asking for already exist in Moo or Mouse or elsewhere, please note that as it will carry significant weight.
For the following, $name should be read as the attribute name; and the
various prefixes should be read using the defaults.
Specifying is => 'rwp' will cause the following options to be set:
is => 'ro'
writer => "_set_$name"
rwp can be read as "read + write private".
Specifying is => 'lazy' will cause the following options to be set:
is => 'ro'
builder => "_build_$name"
lazy => 1
NOTE: Since 0.009 we no longer set init_arg => undef if no init_arg
is explicitly provided. This is a change made in parallel with Moo, based
on a large number of people surprised that lazy also made one's init_def
undefined.
Specifying is => 'lazy' and a default will cause the following options to be
set:
is => 'ro'
lazy => 1
default => ... # as provided
That is, if you specify is => 'lazy' and also provide a default, then
we won't try to set a builder, as well.
Specifying builder => 1 will cause the following options to be set:
builder => "_build_$name"
Passing a coderef to builder will cause that coderef to be installed in the
class this attribute is associated with the name you'd expect, and
builder => 1 to be set.
e.g., in your class (or role),
has foo => (is => 'ro', builder => sub { 'bar!' });
...is effectively the same as...
has foo => (is => 'ro', builder => '_build_foo');
sub _build_foo { 'bar!' }
The behaviour of this option in roles changed in 0.030, and the builder methods will be installed in the role itself. This means you can alias/exclude/etc builder methods in roles, just as you can with any other method.
Specifying clearer => 1 will cause the following options to be set:
clearer => "clear_$name"
or, if your attribute name begins with an underscore:
clearer => "_clear$name"
(that is, an attribute named _foo would get _clear_foo)
Specifying predicate => 1 will cause the following options to be set:
predicate => "has_$name"
or, if your attribute name begins with an underscore:
predicate => "_has$name"
(that is, an attribute named _foo would get _has_foo)
This is a somewhat esoteric shortcut; you probably don't want to use this (or even read this section).
Specifying init_arg => 1 will cause the following options to be set:
# attribute: "name"
init_arg => 'name'
# or, attribute: "_name"
init_arg => '_name'
...while init_arg => -1 will cause the following options to be set:
# attribute: "name"
init_arg => '_name'
# or, attribute: "_name"
init_arg => 'name'
Specifying trigger => 1 will cause the attribute to be created with a trigger
that calls a named method in the class with the options passed to the trigger.
By default, the method name the trigger calls is the name of the attribute
prefixed with _trigger_.
e.g., for an attribute named foo this would be equivalent to:
trigger => sub { shift->_trigger_foo(@_) }
For an attribute named _foo:
trigger => sub { shift->_trigger__foo(@_) }
This naming scheme, in which the trigger is always private, is the same as the builder naming scheme (just with a different prefix).
Creating a delegation with a coderef will now create a new, "custom accessor"
for the attribute. These coderefs will be installed and called as methods on
the associated class (just as readers, writers, and other accessors are), and
will have the attribute metaclass available in $_. Anything the accessor
is called with it will have access to in @_, just as you'd expect of a
method.
e.g., the following example creates an attribute named bar with a standard
reader accessor named bar and two custom accessors named foo and
foo_too.
has bar => (
is => 'ro',
isa => 'Int',
handles => {
foo => sub {
my $self = shift @_;
return $_->get_value($self) + 1;
},
foo_too => sub {
my $self = shift @_;
return $self->bar + 1;
},
# ...as you'd expect.
bar => 'bar',
},
);
...and later,
Note that in this example both foo() and foo_too() do effectively the same thing: return the attribute's current value plus 1. However, foo() accesses the attribute value directly through the metaclass, the pros and cons of which this author leaves as an exercise for the reader to determine.
You may choose to use the installed accessors to get at the attribute's value, or use the direct metaclass access, your choice.
"Abusus non tollit usum."
Note that we create new, anonymous subtypes whenever the constraint or coercion options are specified in such a way that the Shortcuts trait (this one) is invoked. It's fully supported to use both constraint and coerce options at the same time.
This facility is intended to assist with the creation of one-off type constraints and coercions. It is not possible to deliberately reuse the subtypes we create, and if you find yourself using a particular isa / constraint / coerce option triplet in more than one place you should really think about creating a type that you can reuse. MooseX::Types provides the facilities to easily do this, or even a simple constant definition at the package level with an anonymous type stashed away for local use.
has foo => (
is => 'rw',
# $_ == $_[0] == the value to be validated
isa => sub { die unless $_[0] == 1 },
);
# passes constraint
$thing->foo(1);
# fails constraint
$thing->foo(5);
Given a coderef, create a type constraint for the attribute. This constraint will fail if the coderef dies, and pass otherwise.
Astute users will note that this is the same way Moo constraints work; we use MooseX::Meta::TypeConstraint::Mooish to implement the constraint.
Given a package name, this option will create an isa type constraint that
requires the value of the attribute be an instance of the class (or a
descendant class) given. That is,
has foo => (is => 'ro', isa_instance_of => 'SomeThing');
...is effectively the same as:
use Moose::TypeConstraints 'class_type';
has foo => (
is => 'ro',
isa => class_type('SomeThing'),
);
...but a touch less awkward.
Specifying the constraint option with a coderef will cause a new subtype
constraint to be created, with the parent type being the type specified in the
isa option and the constraint being the coderef supplied here.
For example, only integers greater than 10 will pass this attribute's type constraint:
# value must be an integer greater than 10 to pass the constraint
has thinger => (
isa => 'Int',
constraint => sub { $_ > 10 },
# ...
);
Note that if you supply a constraint, you must also provide an isa.
Supplying a constraint and asking for coercion will "Just Work", that is, any
coercions that the isa type has will still work.
For example, let's say that you're using the File type constraint from
MooseX::Types::Path::Class, and you want an additional constraint that the
file must exist:
has thinger => (
is => 'ro',
isa => File,
constraint => sub { !! $_->stat },
coerce => 1,
);
thinger will correctly coerce the string "/etc/passwd" to a
Path::Class:File, and will only accept the coerced result as a value if
the file exists.
Specifying the coerce option with a hashref will cause a new subtype to be created and used (just as with the constraint option, above), with the specified coercions added to the list. In the passed hashref, the keys are Moose types (well, strings resolvable to Moose types), and the values are coderefs that will coerce a given type to our type.
has bar => (
is => 'ro',
isa => 'Str',
coerce => [
Int => sub { "$_" },
Object => sub { 'An instance of ' . ref $_ },
],
);
Sometimes attribute traits interact in surprising ways. This trait is well behaved; if you have discovered any interactions with other traits (good, bad, indifferent, etc), please report this so that it can be worked around, fixed, or documented, as appropriate.
MooseX::SemiAffordanceAccessor changes how the is => 'rw' and
accessor => ... attribute options work. If our trait detects that an
attribute has had the
MooseX::SemiAffordanceAccessor attribute trait
applied, then we change our behaviour to conform to its expectations:
-
is => 'rwp'This:
has foo => (is => 'rwp'); has _bar => (is => 'rwp');...is now effectively equivalent to:
has foo => (is => 'ro', writer => '_set_foo'); has _bar => (is => 'ro', writer => '_set_bar') -
-writer_prefixis ignored...as MooseX::SemiAffordanceAccessor has its own specific ideas as to how writers should look.
Please see those modules/websites for more information related to this module.
Please report any bugs or feature requests on the bugtracker website https://github.com/RsrchBoy/moosex-attributeshortcuts/issues
When submitting a bug or request, please include a test-file or a patch to an existing test-file that illustrates the bug or desired feature.
Chris Weyl cweyl@alumni.drew.edu
- David Steinbrunner dsteinbrunner@pobox.com
- Graham Knop haarg@haarg.org
- Karen Etheridge ether@cpan.org
- Olaf Alders olaf@wundersolutions.com
This software is Copyright (c) 2017, 2015, 2014, 2013, 2012, 2011 by Chris Weyl.
This is free software, licensed under:
The GNU Lesser General Public License, Version 2.1, February 1999