The London Perl and Raku Workshop takes place on 26th Oct 2024. If your company depends on Perl, please consider sponsoring and/or attending.


MooseX::Attribute::TypeConstraint::CustomizeFatal - Control how failed type constraint checks are handled


    package Class;
    use Moose;
    use MooseX::Types::Moose ':all';
    use MooseX::Attribute::TypeConstraint::CustomizeFatal;

    my %attributes = (
        a => "warning",
        b => "default",
        c => "default_no_warning",
        d => "error",

    while (my ($attribute, $on_typeconstraint_failure) = each %attributes) {
        has $attribute => (
            is                        => 'ro',
            isa                       => Int,
            default                   => 12345,

            traits                    => ['TypeConstraint::CustomizeFatal'],
            on_typeconstraint_failure => $on_typeconstraint_failure,

    package main;

        a => "foo", # will be "foo" but will warn
        b => "foo", # will be 12345 but will warn
        c => "foo", # will be 12345 but won't warn
        # d => "foo", # will die, just like Moose does by default

    my $class = Class->new;
    # Does *NOT* work (even if "is" was "rw"), all of this dies
    # $class->(a => "foo");
    # $class->(b => "foo");
    # $class->(c => "foo");
    # $class->(d => "foo");

    1; # *NOT* __PACKAGE__->meta->make_immutable;


By default Moose will just die if you give an attribute a typeconstraint that fails. This trait allows you to customize that behavior to make failures either issue an error like Moose does by default (this is the default), a warning and keep the invalid value, or falling back to the default value either silently or with a warning.


This only works when you construct your class (i.e. with new) and not when setting the values later with an accessor. Therefore you can use this to construct a validating object that has only ro attributes, but not to incrementally add data to the rw attributes of a constructed object (we'll just die on invalid data as Moose does by default).

This also doesn't work if you make your classes immutable. I.e. with __PACKAGE__-meta->make_immutable;> at the end of the class in question instead of 1;.

I might address these caveats in future versions (I'd like to, suggestions/patches welcome), but don't count on it.

But even with these caveats this is still really useful for constructing a validated object all at once from say a HTTP request or to validate data by piggy-backing on the Moose type system without inventing your own validation logic.


A lot of our existing validation code just warns on or ignores invalid values, whereas Moose will die on invalid values.

This makes converting a lot of code inconvenient, particularly code that has values that are supplied directly by the user, usually we don't want the whole process to die if we have some invalid parameter.

We're dealing with these in some cases by wrapping a Moose constructor in try/catch, but that only allows us to either construct an object or not, this trait gives us a much more finely grained control over this not an attribute level.

Now you can define on an attribute level what should be done with invalid values, we can either throw an error (the default), warn and keep the invalid value, fall back on the default either with a warning or without one.

This makes it easy to write user facing code that dispatches directly to a Moose class. If the user provides us with correct values: great! If he doesn't we can just ignore the provided value, replace it with a default, and move on.

In other cases you might want to spew warnings when you start getting unexpected values, but you don't want to die and return an error to the user just because some minor subsystem rendering the page is getting a value it didn't expect.


This module was originally developed at and for With approval from, this module was generalized and put on CPAN, for which the authors would like to express their gratitude.


Ævar Arnfjörð Bjarmason <>