Thankyou Aristotle.

In it's veryCompact form the output is exactly what I was looking for:

P:\test>perl -MDumpvalue -e"
$d=new Dumpvalue; 
$h{ $_ } = [ map{ $_ & 1 ? { 'a'..'z' } : [ 1..100 ] } 1 .. 10 ] 
for 'a' .. 'j'; 
$d->veryCompact(1); 
$d->dumpValue( \%h )"

'a' => ARRAY(0x1871700)
   0  'a' => 'b', 'c' => 'd', 'e' => 'f', 'g' => 'h', 'i' =>
   1  0..99  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
   2  'a' => 'b', 'c' => 'd', 'e' => 'f', 'g' => 'h', 'i' =>
   3  0..99  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
   4  'a' => 'b', 'c' => 'd', 'e' => 'f', 'g' => 'h', 'i' =>
   5  0..99  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
   6  'a' => 'b', 'c' => 'd', 'e' => 'f', 'g' => 'h', 'i' =>
   7  0..99  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
   8  'a' => 'b', 'c' => 'd', 'e' => 'f', 'g' => 'h', 'i' =>
   9  0..99  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
'b' => ARRAY(0x18757e4)
   0  'a' => 'b', 'c' => 'd', 'e' => 'f', 'g' => 'h', 'i' =>
   1  0..99  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
   2  'a' => 'b', 'c' => 'd', 'e' => 'f', 'g' => 'h', 'i' =>
   3  0..99  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
   4  'a' => 'b', 'c' => 'd', 'e' => 'f', 'g' => 'h', 'i' =>
   5  0..99  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
   6  'a' => 'b', 'c' => 'd', 'e' => 'f', 'g' => 'h', 'i' =>
   7  0..99  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
   8  'a' => 'b', 'c' => 'd', 'e' => 'f', 'g' => 'h', 'i' =>
   9  0..99  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
'c' => ARRAY(0x18794fc)
   0  'a' => 'b', 'c' => 'd', 'e' => 'f', 'g' => 'h', 'i' =>
   1  0..99  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
[download]

And it even has an option to see through overloading and handle what comes out -- at least part way.

P:\test>p1
perl> use Math::Pari qw[ :int factorint sqrtint divisors PARI ];;
perl> $f = factorint 1000000;;
perl> use Dumpvalue;;
perl> $d = new Dumpvalue;;
perl> $d->veryCompact( 1 );;
perl> $d->set( bareStringify => 1 );;
perl> { local $\; $d->dumpValue( $f ) };;
0  Math::Pari=ARRAY(0x1a3702c)
   0  Math::Pari=SCALAR(0x1a3c9ec)
      -> 33884400
   1  Math::Pari=SCALAR(0x1a3ec7c)
      -> 33884376
1  Math::Pari=ARRAY(0x1a36f54)
   0  Math::Pari=SCALAR(0x1a3ec70)
      -> 33884388
   1  Math::Pari=SCALAR(0x1a3ec94)
      -> 33884364
perl> Terminating on signal SIGINT(2)
[download]

All it needs is to not de-overload the final SCALAR values so that Math::Pari will return the numbers, which as it's in Perl, I can fix.

It also does circularity testing:

perl> $r = \$r;;
perl> $d->dumpValue( $r );;
->
REF(0x22ac9c)

   -> REUSED_ADDRESS
[download]

But dumping my testcase above, it uses less than 50% extra memory--a considerable saving over the 250% of Data::Dumper. Though I now realise that a large proportion of the extra memory is DD consumes is used by building the output in memory rather than dumping straight the select'd output handle.

I think I can see how to reduce that further still--though it may slow it down a little.

And it's been sat there on my machine the whole time! It's a bit embarrassing that I've never noticed it, but I don't ever recall it being mentioned.

Not only did someone else see the need for what I was asking for, they wrote it, covered all the bases and dropped it on my machine without telling me:)

Once again, thanks Aristotle.