References#

A reference is a scalar that points to another value. References are how Perl builds nested data structures — arrays of arrays, hash of hashes, hash of arrays of hashes — because the basic container types (arrays, hashes) flatten when they appear in lists, and a reference is a single scalar that survives the flattening.

my @arr = (1, 2, 3);
my $ref = \@arr;             # take a reference to @arr
$$ref[0]                     # full deref form     — 1
$ref->[0]                    # arrow shorthand     — 1
@$ref                        # whole-array deref   — (1, 2, 3)

A reference is itself a scalar. It can be passed through any list-flattening calling convention without losing its identity, stored in arrays and hashes, returned from subs, weakened, serialised — anywhere a scalar can go.

Things you can reference#

Seven things take a reference:

\$scalar             # SCALAR ref
\@array              # ARRAY ref
\%hash               # HASH ref
\&subroutine         # CODE ref
\*glob               # GLOB ref
\\$scalar            # REF (a reference to a reference)
\$file_handle        # IO ref (reference to a filehandle)

Plus the anonymous constructors, which build the value and return a reference in one step:

my $aref = [1, 2, 3];                  # anonymous array
my $href = { name => 'Alice', age => 30 };  # anonymous hash
my $cref = sub { $_[0] * 2 };          # anonymous sub

The bracket-vs-paren distinction is essential: (1, 2, 3) is a list, [1, 2, 3] is an anonymous array reference. The first flattens; the second does not.

Dereferencing#

Four spellings, in increasing order of how often they appear in modern code:

${$sref}             # full sigil-deref form
$$sref               # short sigil-deref form (only if name is simple)
$sref->{key}         # arrow form (postfix; for arrays and hashes)
$sref->[0]           # same, for arrays

The -> arrow form is the idiomatic spelling for nested access:

$config->{database}{host}            # arrow elision between subscripts
$config->{database}->{host}          # explicit; same meaning
${$config->{database}}{host}         # full form; the same again

Between two consecutive subscripts the -> is optional (the «arrow elision» rule); at the head of a chain it is required:

$ref->[0]                     # required arrow at the head
$ref->[0][1]                  # second arrow elided
$ref->[0]->[1]                # equivalent
[1,2,3]->[0]                  # required arrow on an anonymous ref

Whole-container deref:

@$aref                        # the whole array
%$href                        # the whole hash, as flat key/value list
&$cref(@args)                 # call the sub
$cref->(@args)                # idiomatic call form

Use the curly-braced form @{ EXPR } / %{ EXPR } when the deref target is an expression more complex than a bare scalar variable:

my @found = @{ $config->{users} };   # deref a chain — needs braces

See arrow for the operator and subscript for the bracket family.

What `ref` returns#

ref reports the kind of thing a reference points to:

ref \$scalar         # 'SCALAR'
ref \@array          # 'ARRAY'
ref \%hash           # 'HASH'
ref \&sub            # 'CODE'
ref \*FH             # 'GLOB'
ref \\$scalar        # 'REF'    — a ref-to-ref
ref \$io_handle      # 'IO'     — wrapper around a file handle

ref [1, 2, 3]        # 'ARRAY'
ref { a => 1 }       # 'HASH'
ref sub {}           # 'CODE'

ref bless {}, 'Foo'  # 'Foo'    — blessed; class name not 'HASH'
ref 42               # ''       — not a reference at all
ref undef            # ''

For blessed objects, ref returns the class. To distinguish a blessed hash-ref from a plain hash-ref, use Scalar::Util::reftype (which always returns the underlying container shape regardless of bless) or isa (the inheritance-aware check).

Postfix dereference (`@`, `%`, `$*`)#

Perl 5.20+ added a postfix syntax that mirrors the arrow chain:

$ref->@*               # equivalent to @$ref
$ref->%*               # equivalent to %$ref
$ref->$*               # equivalent to $$ref

$ref->@[1, 3]          # array slice through ref — equivalent to @{$ref}[1,3]
$ref->%[1, 3]          # key/value slice
$ref->@{qw(a b)}       # hash slice

The postfix forms read left-to-right with the rest of an arrow chain, which is often clearer when the deref target is itself a chain:

$config->{users}->@*             # all users — chains naturally
@{$config->{users}}              # same; reads right-to-left

Pick the form that reads better in context. Postfix is the clearer choice in chains; the leading-sigil form is shorter for plain scalar refs.

Autovivification#

Reading or writing through a deref into a slot that doesn’t exist yet auto-creates the intermediate structure:

my %h;
$h{a}{b}{c} = 1;
# %h is now (a => { b => { c => 1 } })  — three nested hashes built

The mechanism: when the assignment target is $h{a}{b}{c}, Perl sees $h{a} is undef, the next subscript is {b} (a hash subscript), so it makes $h{a} a fresh hash ref. Then $h{a}{b} is undef, next subscript is {c} (also hash), so make it a fresh hash ref. Then assign 1 to the leaf.

The same happens for array subscripts:

my %by_dept;
push @{ $by_dept{eng} }, 'Alice';
# $by_dept{eng} did not exist; the push autoviv'd it as an arrayref

Autoviv on the left of an assignment (or a push/pop/… mutator) is what makes this work. Autoviv on the right of an expression is more dangerous — reading $h{a}{b}{c} when $h{a} is undef also autovivifies, populating the hash as a side effect of inspection:

my %h;
exists $h{a}{b};         # accidentally creates $h{a} = {} as a hash!
                          # %h is now (a => {})

The fix when you genuinely just want to test: chain exists:

exists $h{a} && exists $h{a}{b};        # short-circuits if $h{a} absent

This is the most-bitten autoviv trap. Tests of nested existence must short-circuit, or they leave hash slots behind.

Hash of arrays — the canonical example#

my @people = (
    { name => 'Alice', dept => 'eng'   },
    { name => 'Bob',   dept => 'sales' },
    { name => 'Carol', dept => 'eng'   },
    { name => 'Dan',   dept => 'sales' },
);

# Group by department:
my %by_dept;
for my $p (@people) {
    push @{ $by_dept{ $p->{dept} } }, $p->{name};
}
# %by_dept = (
#     eng   => ['Alice', 'Carol'],
#     sales => ['Bob', 'Dan'],
# )

# Iterate:
for my $dept (sort keys %by_dept) {
    print "$dept: @{ $by_dept{$dept} }\n";
}
# eng: Alice Carol
# sales: Bob Dan

The @{ $by_dept{...} } is the deref-with-braces form. It’s necessary because the deref target is an expression ($by_dept{$dept}), not a bare variable.

The same shape with the postfix form:

print "$dept: ", join(' ', $by_dept{$dept}->@*), "\n";

Pick whichever you find clearer; both are idiomatic in modern Perl.

Weak references and circular structures#

A reference holds the referent alive (refcount-bumped) until the reference itself goes out of scope. This becomes a problem when two structures point at each other:

my $parent = { name => 'p' };
my $child  = { name => 'c', parent => $parent };
push @{ $parent->{children} }, $child;

# parent refers to child via the children arrayref
# child refers back to parent via the parent slot
# When both go out of scope, neither's refcount can reach 0 — leak.

The fix is a weak reference — a reference that doesn’t bump the refcount. The $child->{parent} slot is the one that should be weakened (the back-pointer):

use Scalar::Util qw(weaken);

my $parent = { name => 'p' };
my $child  = { name => 'c', parent => $parent };
weaken $child->{parent};                 # break the cycle
push @{ $parent->{children} }, $child;

After weakening, $child->{parent} reads as the parent while it exists, and silently becomes undef when the parent is freed. The canonical place for weaken is on every back-pointer in a tree structure where children point back at their parent.

The Scalar::Util::isweak predicate tests whether a given reference is weakened.

Reference equality#

Two references are equal (numerically ==, stringwise eq) if and only if they refer to the same underlying value:

my @a = (1, 2, 3);
my $r1 = \@a;
my $r2 = \@a;
my $r3 = [1, 2, 3];      # different anonymous array

$r1 == $r2               # TRUE — same address
$r1 == $r3               # FALSE — different array, even though contents match

Numeric comparison of refs effectively compares their machine address. Two refs with the same content but different identities compare unequal. Deep-equal needs a helper from Test::More or a hand-rolled walker.

A reference stringifies to something like ARRAY(0x55ab12cd) — class name plus hex address. Don’t rely on the format; use it only as a debug print.