How can I adjust the rendering of objects in a longmess?

debugging, perl, stack-trace

Solution

The problem is in this part of `Carp.pm`:

sub format_arg {
    my $arg = shift;
    if ( ref($arg) ) {
        $arg = defined($overload::VERSION) ? overload::StrVal($arg) : "$arg";
    }
    ...
}

That is, when an argument could be an overloaded object, then any stringification overloading is circumvented with the `StrVal` helper, which forces default stringification.

Unfortunately, there is no straightforward way around that. All we can do is monkey-patch the `Carp::format_arg` sub, e.g.

BEGIN {
  use overload ();
  use Carp ();
  no warnings 'redefine';
  my $orig = \&Carp::format_arg;

  *Carp::format_arg = sub {
    my ($arg) = @_;
    if (ref $arg and my $stringify = overload::Method($arg, '""')) {
      $_[0] = $stringify->($arg);
    }
    goto &$orig;
  };
}

As it is, this is inelegant, and should be put into a pragma:

File `Carp/string_overloading.pm`:

package Carp::string_overloading;

use strict; use warnings;

use overload ();
use Carp ();

# remember the original format_arg method
my $orig = \&Carp::format_arg;
# This package is internal to Perl's warning system.
$Carp::CarpInternal{ __PACKAGE__() }++;

{
    no warnings 'redefine';
    *Carp::format_arg = sub {
        my ($arg) = @_;
        if (    ref($arg)
            and in_effect(1 + Carp::long_error_loc)
            and my $stringify = overload::Method($arg, '""')
        ) {
            $_[0] = $stringify->($arg);
        }
        goto &$orig;
    };
}

sub import   { $^H{__PACKAGE__ . "/in_effect"} = 1 }

sub unimport { $^H{__PACKAGE__ . "/in_effect"} = 0 }

sub in_effect {
    my $level = shift // 1;
    return (caller $level)[10]{__PACKAGE__ . "/in_effect"};
}

1;

Then the code

use strict; use warnings;

package Foo {
    use Carp ();

    use overload '""' => sub {
        my $self = shift;
        return sprintf '%s[%s]', ref $self, join ", ", @$self;
    };

    use Carp::string_overloading;
    sub foo { Carp::confess "as requested" }

    no Carp::string_overloading;
    sub bar { Carp::confess "as requested" }
}

my $foo = bless [1..3] => 'Foo';

eval { $foo->foo("foo") };
print $@;
eval { $foo->bar("bar") };
print $@;

outputs:

as requested at test.pl line 12.
        Foo::foo('Foo[1, 2, 3]', 'foo') called at test.pl line 20
        eval {...} called at test.pl line 20
as requested at test.pl line 15.
        Foo::bar('Foo=ARRAY(0x85468ec)', 'bar') called at test.pl line 22
        eval {...} called at test.pl line 22

Problem

We're slowly refactoring our large Perl application towards object-oriented interfaces, especially for data models. The annoying part is that stack traces get less useful. To give a fabricated example: Before. ``` sub send_message { my ($user_id, $message) = @_; ... Carp::confess('test'); } # output: test at example.pm line 23 foo('42', 'Hello World') called at example.pl line 5 ``` After. ``` sub send_message { my ($user, $message) = @_; ... Carp::confess('test'); } # output: test at example.pm line 23 foo('MyApp::Model::User=HASH(0x2c94f68)', 'Hello World') called at example.pl line 5 ``` So now I cannot see which user was passed to `foo()`, I only see the class name (which is already documented) and some memory address of an object. I tried to install a stringification operator on the model class using overload.pm: ``` use overload ( '""' => \&stringify ); sub stringify { my ($self) = @_; return sprintf '%s[id=%d]', ref($self), $self->id; } ``` But this does not affect the longmess. What I would like is something like this: ``` test at example.pm line 23 foo('MyApp::Model::User[id=42]', 'Hello World') called at example.pl line 5 ``` That is, the first parameter to `foo()` should be displayed using the object's `stringify()` method. How can I achieve that?

Original source