Recipe 5.17 Program: dutree

The dutree program, shown in Example 5-3, turns the output of du:

% du pcb
19      pcb/fix
20      pcb/rev/maybe/yes
10      pcb/rev/maybe/not
705     pcb/rev/maybe
54      pcb/rev/web
1371    pcb/rev
3       pcb/pending/mine
1016    pcb/pending
2412    pcb

into sorted, indented output:

2412 pcb
   |    1371 rev
   |       |    705 maybe
   |       |      |      675 .
   |       |      |       20 yes
   |       |      |       10 not
   |       |    612 .
   |       |     54 web
   |    1016 pending
   |       |        1013 .
   |       |           3 mine
   |      19 fix
   |       6 .

The arguments you give dutree are passed through to du. That way you could call dutree in any of these ways, or maybe more if your du supports other options:

% dutree
% dutree /usr
% dutree -a
% dutree -a /bin

The %Dirsize hash maintains the mapping of names to sizes. For example, $Dirsize{"pcb"} contains 2412 in this sample run. We'll use that hash both for output and for sorting each directory's subdirectories by size.

%Kids is more interesting. For any given path PATH, $Kids{PATH} contains a (reference to an) array of names of subdirectories of this one. The "pcb" entry contains a reference to an anonymous array containing "fix", "rev", and "pending". The "rev" entry contains "maybe" and "web". The "maybe" entry contains "yes" and "not", which do not have their own entries because they are end nodes in the tree.

The output function is passed the start of the treethe last line read in from the output of du. First it prints that directory and its size. Then the function sorts the directory's children (if any) so that those with the most disk usage float to the top. Finally, output calls itself, recursing on each child in order. The extra arguments are used in formatting.

This program is inherently recursive because the filesystem is recursive. However, its data structure is not recursive; at least, not the way a circular linked list is. Each value is an array of further keys to process. The recursion resides in the processing, not in the storage.

Example 5-3. dutree
  #!/usr/bin/perl -w
  # dutree - print sorted indented rendition of du output
  use strict;
  my %Dirsize;
  my %Kids;
  getdots(my $topdir = input( ));
  # run du, read in input, save sizes and kids
  # return last directory (file?) read
  sub input {
      my($size, $name, $parent);
      @ARGV = ("du @ARGV |");         # prep the arguments
      while (<>) {              # magic open is our friend 
          ($size, $name) = split;
          $Dirsize{$name} = $size;
          ($parent = $name) =~ s#/[^/]+$##;   # dirname
          push @{ $Kids{$parent} }, $name unless eof;
      return $name;
  # figure out how much is taken up in each directory
  # that isn't stored in subdirectories.  add a new
  # fake kid called "." containing that much.
  sub getdots {
      my $root = $_[0];
      my($size, $cursize);
      $size = $cursize = $Dirsize{$root};
      if ($Kids{$root}) {
          for my $kid (@{ $Kids{$root} }) {
              $cursize -= $Dirsize{$kid};
      if ($size != $cursize) {
          my $dot = "$root/.";
          $Dirsize{$dot} = $cursize;
          push @{ $Kids{$root} }, $dot;
  # recursively output everything,
  # passing padding and number width in as well
  # on recursive calls
  sub output {
      my($root, $prefix, $width) = (shift, shift || '', shift || 0);
      my $path;
      ($path = $root) =~ s#.*/##;     # basename
      my $size = $Dirsize{$root};
      my $line = sprintf("%${width}d %s", $size, $path);
      print $prefix, $line, "\n";
      for ($prefix .= $line) {        # build up more output
          s/\d /| /;
          s/[^|]/ /g;
      if ($Kids{$root}) {             # not a bachelor node
          my @Kids = @{ $Kids{$root} };
          @Kids = sort { $Dirsize{$b} <=> $Dirsize{$a} } @Kids;
          $Dirsize{$Kids[0]} =~ /(\d+)/;
          my $width = length $1;
          for my $kid (@Kids) { output($kid, $prefix, $width) }

Before Perl supported hashes of arrays directly, Herculean efforts were required to emulate these higher order constructs. Some folks used repeated calls to split and join, but these were exceedingly slow.

Example 5-4 is a version of dutree from those days of Perl antiquity. Because we didn't have proper array references, we had to usurp the Perl symbol table itself. This program created variables on the fly with bizarre names. Can you find which hash this program is using?

The @{"pcb"} array contains "pcb/fix", "pcb/rev", and "pcb/pending". The @{"pcb/rev"} array contains "pcb/rev/maybe" and "pcb/rev/web". The @{"pcb/rev/maybe"} array contains "pcb/rev/yes" and "pcb/rev/not".

When you assign something like "pcb/fix" to *kid, it promotes the string on the righthand side to a typeglob. This makes @kid an alias for @{"pcb/fix"}among other things. It would also alias &kid to &{"pcb/fix"}, and so on.

If that isn't interesting enough, consider how the local is using dynamic scoping of global variables to avoid passing in extra arguments. Check out what's happening with the $width variable in the output routine.

Example 5-4. dutree-orig
  # dutree_orig: the old version pre-perl5 (early 90s)
  @lines = `du @ARGV`;
  &input($top = pop @lines);
  sub input {
      local($root, *kid, $him) = @_[0,0];
      while (@lines && &childof($root, $lines[$#lines])) {
          &input($him = pop(@lines));
          push(@kid, $him);
      if (@kid) {
          local($mysize) = ($root =~ /^(\d+)/);
          for (@kid) { $mysize -= (/^(\d+)/)[0]; }
          push(@kid, "$mysize .") if $size != $mysize;
      @kid = &sizesort(*kid);
  sub output {
      local($root, *kid, $prefix) = @_[0,0,1];
      local($size, $path) = split(' ', $root);
      $path =~ s!.*/!!;
      $line = sprintf("%${width}d %s", $size, $path);
      print $prefix, $line, "\n";
      $prefix .= $line;
      $prefix =~ s/\d /| /;
      $prefix =~ s/[^|]/ /g;
      local($width) = $kid[0] =~ /(\d+)/ && length("$1");
      for (@kid) { &output($_, $prefix); };
  sub sizesort {
      local(*list, @index) = shift;
      sub bynum { $index[$b] <=> $index[$a]; }
      for (@list) { push(@index, /(\d+)/); }
      @list[sort bynum 0..$#list];
  sub childof {
      local(@pair) = @_;
      for (@pair) { s/^\d+\s+//g; s/$/\//; }
      index($pair[1], $pair[0]) >= 0;

The answer to the question posed earlier "Which hash is the old dutree using?" is %main::, that is, the Perl symbol table itself. Needless to say, this program will never run under use strict. We're happy to report that the updated version runs three times as fast as the old one. That's because the old one keeps looking up variables in the symbol table, and the new one doesn't have to. It's also because we avoid all that slow splitting of the space used and the directory name. But we thought we'd show you the old version because it is instructive, too.