so why is perl wasting so many compute cycles getting that information?

If the docs don't explain, Use The Source Luke

http://perl5.git.perl.org/perl.git/tree/HEAD:/ext/File-Glob

http://perl5.git.perl.org/perl.git/blob/HEAD:/ext/File-Glob/bsd_glob.c#l650

/*
 651  * The functions glob2 and glob3 are mutually recursive; there is
+ one level
 652  * of recursion for each segment in the pattern that contains one
+ or more
 653  * meta characters.
 654  */
[download]

http://perl5.git.perl.org/perl.git/blob/HEAD:/ext/File-Glob/bsd_glob.c#l670

So in  "$path/d[0-9]*/d[0-9]*" it needs to determine if d[0-9]* is a directory, so it needs to call stat

Could parts of bsd_glob.c be optimized? Maybe

These days I usually use File::Find::Rule, but this glob converted is not as elegant

use File::Find::Rule;
my @dirs = find( 'dir', 'name', qr/^d\d/i, 'maxdepth', 1, 'in', $path 
+);
my @files = find( 'file', 'name', qr/^d\d/i, 'maxdepth', 1, 'in', \@di
+rs );
[download]