http://www.perlmonks.org?node_id=85188


in reply to Re: Find duplicate files.
in thread Find duplicate files.

Yes, but there is a fundamental difference...

The first script will only do MD5 hashes on files if there is more than one file with the same file size, then compares the MD5s for the files of that size. Yours MD5's *everything*, then compares *all* the MD5s. If a file has a unique filesize, it *can't* have a duplicate.

Depending on the make up of the files, this can have a dramatic effect:

Files: 15272 Duplicates: 999 Bytes: 15073525
Results:
First script: real 0m11.855s user 0m2.590s sys 0m1.640s Second script: real 0m49.589s user 0m17.110s sys 0m6.500s
The second script is four times slower than the first...

Admittedly, if all your files were the same size there would be no difference, but in most cases, the first script will win. But hey...

Replies are listed 'Best First'.
Re: Re: Re: Find duplicate files.
by sarabob (Novice) on Jun 02, 2001 at 20:16 UTC
    Me again (posted as Anonymous Monk last time) As above, I drew a blank on google, and was amazed that even stuff written in C and posted on Freshmeat was stupidly slow at finding duplicate files. I knew I should have checked Perlmonks sooner :-) Just to point out the differences in real-time speed depending on how much of how many files you read in in the comparison step, I coded up another version in a little (fully functional though) utility.

    The code for script three is below

    Benchmarks:

    scenario 1: a bundle of MP3 files
    number of files: 5969
    Number of duplicates: 11
    Total size of all files: 16,560,706,048 (~16 gigs)

    (script one - original, MD5 hash files with same file size)

    ... 37,411,688 bytes in duplicated files real 5m1.753s user 1m48.760s sys 0m31.010s
    (script two - MD5 hash calculated on *all* files)
    Duplicates: 11 Bytes: 37411688 real 17m1.262s user 6m56.060s sys 1m59.830s
    fdupes (C program from freshmeat.net - uses same algorithm as script two)
    real 6m25.131s user 2m42.310s sys 0m32.450s
    (script three - see below, read up to first \n char for initial check, then read whole file in for full check. No MD5s calculated at all)
    37,411,688 bytes in duplicated files real 0m48.545s user 0m2.150s sys 0m1.460s
    Yes, that *is* 48 seconds rather than 5 or 17 minutes. This is because script 3 reads the first line in as a comparison first - creating an MD5 hash requires that the whole file is read in.

    Scenario 2: home directory
    number of files: 15280
    Number of duplicates: 998
    Total size of all files: 677,698,560 (677 megs)

    Script one results

    15,073,517 bytes in duplicated files real 0m9.745s user 0m2.610s sys 0m1.220s
    Script two results
    Duplicates: 998 Bytes: 15073517 real 0m51.197s user 0m17.520s sys 0m6.700s
    fdupes (C program from freshmeat.net - uses same algorithm as script two)
    real 0m18.332s user 0m8.080s sys 0m5.270s
    Script three results
    15,069,331 bytes in duplicated files real 0m12.924s user 0m3.110s sys 0m2.510s
    (Note less duplicates found by script three as it skips all the small files < 100 bytes)

    The third script is slower than the first in this situation as it must do multiple compares (ie a with b, a with c, a with d) rather than using the MD5 hashing technique It would be even slower if we counted small files (timed at around 23 seconds). Both 1 and 3 are still *much* faster than 2 though. The fdupes benchmarks are just in there for comparison to show how a bad algorithm can slow down a fast language.

    Also note that not using MD5 hashes means I suffer if there are three or more identical, large, files, but I wanted to be *absolutely* sure not to get any false positives and MD5 hashing doesn't (quite) do that. So I do a byte-for-byte comparison between possible pairs.

    There is almost certainally another way - we could do two passes using the MD5 technique, creating MD5 hashes for the first (say) 200 bytes of each file in the first pass, then MD5-ing on the whole file if the first ones match. This should give us good performance on both large numbers of duplicated small files *and* small numbers of duplicates of large files. But that's something for another day, and I somehow *prefer* to do byte-by-byte checks. Paranoia, I guess.

    Anyway - here's the code...

    fdupes.pl (usage: fdupes.pl <start dir>):

    #!/usr/bin/perl -w # Usage: ./fdupes.pl <start directory> use strict; use Term::ReadKey; use File::Find; # testing - 0 for interactive mode, 1 to skip all deletion etc my $testing = 0; # skip files smaller than 100 bytes. Set to zero if you like... my $minsize = 100; my $filecount = my $bytecount = my $fileschecked = my $wasted = 0; my %files = (); &usage unless (@ARGV); sub wanted { return unless -f; my $filesize = (stat($_))[7]; $bytecount += $filesize; return unless $filesize > $minsize; # skip small files $filecount++; push @{$files{$filesize}}, $File::Find::name; } find(\&wanted, $ARGV[0] || "."); # update progress display 1000 times maximum my $update_period = int($filecount/1000)+1; if ($fileschecked % $update_period == 0) { print "Progress: $fileschecked/$filecount\r"; # note \r does carriage return, but NO LINE FEED # for progress display } my @dupesets; # list of lists - @{$dupesets[0]} = (file1, file2) # where file1 and file2 are dupes foreach my $size (keys %files) { my @entries = @{$files{$size}}; my $samesizecount = scalar @entries; if (@{$files{$size}} == 1) { # unique size $fileschecked++; next; } # duplicates by file size.. Check if files are the same while (my $base = shift @entries) { # get first entry in list under filesize my @dupes = (); my $count = 0; while ($count <= $#entries) { # go through all @entries my $compare = $entries[$count]; if (&same($base, $compare)) { # remove "compare" from list so it can't be used # on next run splice(@entries, $count,1); # removed "compare" from list - update progress if (++$fileschecked % $update_period == 0) { print "Progress: $fileschecked/$filecount\r"; } if (@dupes) { # already have some dupes - just add duplicate # #n to list push @dupes, $compare; $wasted += $size; } else { # no dupes yet - include base file and duplicate # #1 in list push @dupes, ($base, $compare); $wasted += $size; } } else { $count++; # only increase counter if not a dupe - note splice # will break $array[$position] loop otherwise } } if (@dupes) { push @dupesets, \@dupes; } # "base" file removed from list of files to check - update # progress meter if (++$fileschecked % $update_period == 0) { print "Progress: $fileschecked/$filecount\r"; } } } if (@dupesets) { my @deletelist = (); # at least one set of duplicates exists # number of sets of duplicates my $dupesetcount = scalar(@dupesets); my $dupesetcounter = 0; foreach my $setref (@dupesets) { if ($testing) { print @$setref, "\n"; next; } $dupesetcounter++; my @dupes = @$setref; print "Duplicates found ($dupesetcounter / $dupesetcount)", "... Should I keep...\n"; my $count = 0; # print up list of options of which file to keep while ($count <= $#dupes) { # go through all @entries my $entry = $dupes[$count]; print $count + 1, " : $entry\n"; $count++; } # alternative options - keep all files, skip to end print "0: All\n"; print "A: Skip all remaining duplicates\n"; # use ReadKey to get user input ReadMode 4; # Turn off controls keys my $key = ''; while (not defined ($key = ReadKey(-1))) { # No key yet } ReadMode 0; # Reset tty mode before exiting if ($key eq 'A') { # skip any remaining dupes and get to deletion bit last; } # not a number or 'A' - default to zero (ie keep all files) $key = '0' unless ($key =~ /^\d+$/); if ($key == 0) { # ALL - don't delete anything #print "you chose: ALL\n"; } elsif (defined $dupes[$key-1]) { print "you chose: ", $dupes[$key-1], "\n"; my @list_to_delete = @dupes; # remove file to keep from list splice(@list_to_delete, $key-1, 1); # add rest to deletelist push @deletelist, @list_to_delete; } else { #print "you chose: invalid number... (nothing will", # " be deleted)\n"; } print "\n"; } # confirm deletion if any files are needing deleting if (@deletelist) { print "\n------------------------\n"; print "list of files to delete:\n"; foreach (@deletelist) { print "$_\n"; } print "\nAre you *sure* you want to delete all these files?", " (Y/N)\n"; ReadMode 4; # Turn off controls keys my $key = ''; while (not defined ($key = ReadKey(-1))) { # No key yet } ReadMode 0; # Reset tty mode before exiting if (lc($key) eq 'y') { print "deleting\n"; unlink @deletelist; } else { print "wussing out\n"; } } 1 while $wasted =~ s/^([-+]?\d+)(\d{3})/$1,$2/; print "$wasted bytes in duplicated files\n"; } # routine to check equivalence in files. pass 1 checks first # "line" of file (up to \n char), rest of file checked if 1st # line matches sub same { local($a, $b) = @_; open(A, $a) || die; open(B, $b) || die; if (<A> ne <B>) { # FIRST LINE is not the same return 0; # not duplicates } else { # try WHOLE FILE local $/ = undef; return <A> eq <B>; } } sub usage { print "Usage: $0 <start directory>\n"; exit; }
      When there are many, many duplicates and you don't care to comb through the list, this version keeps the first one and happily delete all others on confirmation. It also skips symlinks, because the original will happily delete the original and keep the link!