Here is a brief pseudo-grammar for XDumper's output. Note that whitespace is irrelevant and only used for clarity.

item    -> var | special | label
var     -> ( label ':' )? classname? '<ro>'? (scalar |
           array | hash | glob | code | io | lvalue )
scalar  -> number | string | ref | 'undef'
ref     -> '<weak>'? ( '\' item | '[' list? ']' | '{' hlist? '}' )
array   -> '@(' list? ')'
hash    -> '%(' hlist? ')'
list    -> item (',' item)*
hlist   -> key '=>' item (',' key '=>' item)*
glob    -> '<anon>'? '*' package? name
code    -> '<format>'? '&(' filename ':' linenum ')'
io      -> '<io>'
lvalue  -> substr | pos | vec
substr  -> 'substr(' item ',' number ',' number ')'
pos     -> 'pos(' item ')'
vec     -> 'vec(' item ',' number ',' number ')'
special -> '<undef>' | '<yes>' | '<no>'
[download]

In particular note that [...] is an abbreviation for \@(...) and {...} is an abbreviation for \%(...).

The abbeviated forms are used whenever possible, but sometimes it has to use the expanded form, for example when the array or hash has a prefix (blessing, read-only) or when it has a label.

For example $L001: <ro> \Foo @(1, 2) means that label $L001 names a read-only reference to a Foo-blessed array containing the numbers 1 and 2.

For people who know the perl guts: 'var', 'ref', 'array', 'hash' etc in the above grammar correspond one-to-one with SV, RV, AV, HV etc.

•Update: added format and io
•Update: added lvalues (completely forgot them earlier)