path: root/hacks/xscreensaver-getimage-file

#!/usr/bin/perl -w
# Copyright © 2001-2020 Jamie Zawinski <jwz@jwz.org>.
#
# Permission to use, copy, modify, distribute, and sell this software and its
# documentation for any purpose is hereby granted without fee, provided that
# the above copyright notice appear in all copies and that both that
# copyright notice and this permission notice appear in supporting
# documentation.  No representations are made about the suitability of this
# software for any purpose.  It is provided "as is" without express or 
# implied warranty.
#
# This program chooses a random file from under the given directory, and
# prints its name.  The file will be an image file whose dimensions are
# larger than a certain minimum size.
#
# If the directory is a URL, it is assumed to be an RSS or Atom feed.
# The images from that feed will be downloaded, cached, and selected from
# at random.  The feed will be re-polled periodically, as needed.
#
# The various xscreensaver hacks that manipulate images ("jigsaw", etc.) get
# the image to manipulate by running the "xscreensaver-getimage" program.
#
# Under X11, the "xscreensaver-getimage" program invokes this script,
# depending on the value of the "chooseRandomImages" and "imageDirectory"
# settings in the ~/.xscreensaver file (or .../app-defaults/XScreenSaver).
# The screen savers invoke "xscreensaver-getimage" via utils/grabclient.c,
# which then invokes this script.
#
# Under Cocoa, this script lives inside the .saver bundle, and is invoked
# directly from utils/grabclient.c.
#
# Created: 12-Apr-01.

require 5;
#use diagnostics;	# Fails on some MacOS 10.5 systems
use strict;

use POSIX;
use Fcntl;

use Fcntl ':flock'; # import LOCK_* constants

use POSIX ':fcntl_h';				# S_ISDIR was here in Perl 5.6
import Fcntl ':mode' unless defined &S_ISUID;	# but it is here in Perl 5.8
	# but in Perl 5.10, both of these load, and cause errors!
	# So we have to check for S_ISUID instead of S_ISDIR?  WTF?

use Digest::MD5 qw(md5_base64);

# Some Linux systems don't install LWP by default!
# Only error out if we're actually loading a URL instead of local data.
BEGIN { eval 'use LWP::Simple;' }


my $progname = $0; $progname =~ s@.*/@@g;
my ($version) = ('$Revision: 1.57 $' =~ m/\s(\d[.\d]+)\s/s);

my $verbose = 0;

# Whether to use MacOS X's Spotlight to generate the list of files.
# When set to -1, uses Spotlight if "mdfind" exists.
#
# (In my experience, this isn't actually any faster, and might not find
# everything if your Spotlight index is out of date, which happens often.)
#
my $use_spotlight_p = 0;

# Whether to cache the results of the last run.
#
my $cache_p = 1;

# Regenerate the cache if it is older than this many seconds.
#
my $cache_max_age = 60 * 60 * 3;   # 3 hours

# Re-poll RSS/Atom feeds when local copy is older than this many seconds.
#
my $feed_max_age = $cache_max_age;


# This matches files that we are allowed to use as images (case-insensitive.)
# Anything not matching this is ignored.  This is so you can point your
# imageDirectory at directory trees that have things other than images in
# them, but it assumes that you gave your images sensible file extensions.
#
my @good_extensions = ('jpg', 'jpeg', 'pjpeg', 'pjpg', 'png', 'gif',
                       'tif', 'tiff', 'xbm', 'xpm', 'svg');
my $good_file_re = '\.(' . join("|", @good_extensions) . ')$';

# This matches file extensions that might occur in an image directory,
# and that are never used in the name of a subdirectory.  This is an
# optimization that prevents us from having to stat() those files to
# tell whether they are directories or not.  (It speeds things up a
# lot.  Don't give your directories stupid names.)
#
my @nondir_extensions = ('ai', 'bmp', 'bz2', 'cr2', 'crw', 'db',
   'dmg', 'eps', 'gz', 'hqx', 'htm', 'html', 'icns', 'ilbm', 'mov',
   'nef', 'pbm', 'pdf', 'php', 'pl', 'ppm', 'ps', 'psd', 'sea', 'sh',
   'shtml', 'tar', 'tgz', 'thb', 'txt', 'xcf', 'xmp', 'Z', 'zip' );
my $nondir_re = '\.(' . join("|", @nondir_extensions) . ')$';


# JPEG, GIF, and PNG files that are are smaller than this are rejected:
# this is so that you can use an image directory that contains both big
# images and thumbnails, and have it only select the big versions.
# But, if all of your images are smaller than this, all will be rejected.
#
my $min_image_width  = 500;
my $min_image_height = 500;

my @all_files = ();         # list of "good" files we've collected
my %seen_inodes;            # for breaking recursive symlink loops

# For diagnostic messages:
#
my $dir_count = 1;          # number of directories seen
my $stat_count = 0;	    # number of files/dirs stat'ed
my $skip_count_unstat = 0;  # number of files skipped without stat'ing
my $skip_count_stat = 0;    # number of files skipped after stat

my $config_file = $ENV{HOME} . "/.xscreensaver";
my $image_directory = undef;


sub find_all_files($);
sub find_all_files($) {
  my ($dir) = @_;

  print STDERR "$progname:  + reading dir $dir/...\n" if ($verbose > 1);

  my $dd;
  if (! opendir ($dd, $dir)) {
    print STDERR "$progname: couldn't open $dir: $!\n" if ($verbose);
    return;
  }
  my @files = readdir ($dd);
  closedir ($dd);

  my @dirs = ();

  foreach my $file (@files) {
    next if ($file =~ m/^\./);      # silently ignore dot files/dirs

    if ($file =~ m/[~%\#]$/) {      # ignore backup files (and dirs...)
      $skip_count_unstat++;
      print STDERR "$progname:  - skip file  $file\n" if ($verbose > 1);
    }

    $file = "$dir/$file";

    if ($file =~ m/$good_file_re/io) {
      #
      # Assume that files ending in .jpg exist and are not directories.
      #
      push @all_files, $file;
      print STDERR "$progname:  - found file $file\n" if ($verbose > 1);

    } elsif ($file =~ m/$nondir_re/io) {
      #
      # Assume that files ending in .html are not directories.
      #
      $skip_count_unstat++;
      print STDERR "$progname: -- skip file  $file\n" if ($verbose > 1);

    } else {
      #
      # Now we need to stat the file to see if it's a subdirectory.
      #
      # Note: we could use the trick of checking "nlinks" on the parent
      # directory to see if this directory contains any subdirectories,
      # but that would exclude any symlinks to directories.
      #
      my @st = stat($file);
      my ($dev,$ino,$mode,$nlink,$uid,$gid,$rdev,$size,
          $atime,$mtime,$ctime,$blksize,$blocks) = @st;

      $stat_count++;

      if ($#st == -1) {
        if ($verbose) {
          my $ll = readlink $file;
          if (defined ($ll)) {
            print STDERR "$progname: + dangling symlink: $file -> $ll\n";
          } else {
            print STDERR "$progname: + unreadable: $file\n";
          }
        }
        next;
      }

      next if ($seen_inodes{"$dev:$ino"}); # break symlink loops
      $seen_inodes{"$dev:$ino"} = 1;

      if (S_ISDIR($mode)) {
        push @dirs, $file;
        $dir_count++;
        print STDERR "$progname:  + found dir  $file\n" if ($verbose > 1);

      } else {
        $skip_count_stat++;
        print STDERR "$progname:  + skip file  $file\n" if ($verbose > 1);
      }
    }
  }

  foreach (@dirs) {
    find_all_files ($_);
  }
}


sub spotlight_all_files($) {
  my ($dir) = @_;

  my @terms = ();
  # "public.image" matches all (indexed) images, including Photoshop, etc.
#  push @terms, "kMDItemContentTypeTree == 'public.image'";
  foreach (@good_extensions) {

    # kMDItemFSName hits the file system every time: much worse than "find".
#    push @terms, "kMDItemFSName == '*.$_'";

    # kMDItemDisplayName matches against the name in the Spotlight index,
    # but won't find files that (for whatever reason) didn't get indexed.
    push @terms, "kMDItemDisplayName == '*.$_'";
  }

  $dir =~ s@([^-_/a-z\d.,])@\\$1@gsi;  # quote for sh
  my $cmd = "mdfind -onlyin $dir \"" . join (' || ', @terms) . "\"";

  print STDERR "$progname: executing: $cmd\n" if ($verbose > 1);
  @all_files = split (/[\r\n]+/, `$cmd`);
}


# If we're using cacheing, read the cache file and return its contents,
# if any.  This also holds an exclusive lock on the cache file, which 
# has the additional benefit that if two copies of this program are
# running at once, one will wait for the other, instead of both of
# them spanking the same file system at the same time.
#
my $cache_fd = undef;
my $cache_file_name = undef;
my $read_cache_p = 0;

sub read_cache($) {
  my ($dir) = @_;

  return () unless ($cache_p);

  my $dd = "$ENV{HOME}/Library/Caches";    # MacOS location
  if (-d $dd) {
    $cache_file_name = "$dd/org.jwz.xscreensaver.getimage.cache";
  } elsif (-d "$ENV{HOME}/.cache") {	   # Gnome "FreeDesktop XDG" location
    $dd = "$ENV{HOME}/.cache/xscreensaver";
    if (! -d $dd) { mkdir ($dd) || error ("mkdir $dd: $!"); }
    $cache_file_name = "$dd/xscreensaver-getimage.cache"
  } elsif (-d "$ENV{HOME}/tmp") {	   # If ~/tmp/ exists, use it.
    $cache_file_name = "$ENV{HOME}/tmp/.xscreensaver-getimage.cache";
  } else {
    $cache_file_name = "$ENV{HOME}/.xscreensaver-getimage.cache";
  }

  print STDERR "$progname: awaiting lock: $cache_file_name\n"
    if ($verbose > 1);

  my $file = $cache_file_name;
  open ($cache_fd, '+>>', $file) || error ("unable to write $file: $!");
  flock ($cache_fd, LOCK_EX)     || error ("unable to lock $file: $!");
  seek ($cache_fd, 0, 0)         || error ("unable to rewind $file: $!");

  my $mtime = (stat($cache_fd))[9];

  if ($mtime + $cache_max_age < time) {
    print STDERR "$progname: cache is too old\n" if ($verbose);
    return ();
  }

  my $odir = <$cache_fd>;
  $odir =~ s/[\r\n]+$//s if defined ($odir);
  if (!defined ($odir) || ($dir ne $odir)) {
    print STDERR "$progname: cache is for $odir, not $dir\n"
      if ($verbose && $odir);
    return ();
  }

  my @files = ();
  while (<$cache_fd>) { 
    s/[\r\n]+$//s;
    push @files, "$odir/$_";
  }

  print STDERR "$progname: " . ($#files+1) . " files in cache\n"
    if ($verbose);

  $read_cache_p = 1;
  return @files;
}


sub write_cache($) {
  my ($dir) = @_;

  return unless ($cache_p);

  # If we read the cache, just close it without rewriting it.
  # If we didn't read it, then write it now.

  if (! $read_cache_p) {

    truncate ($cache_fd, 0) ||
      error ("unable to truncate $cache_file_name: $!");
    seek ($cache_fd, 0, 0) ||
      error ("unable to rewind $cache_file_name: $!");

    if ($#all_files >= 0) {
      print $cache_fd "$dir\n";
      foreach (@all_files) {
        my $f = $_; # stupid Perl. do this to avoid modifying @all_files!
        $f =~ s@^\Q$dir/@@so || die;  # remove $dir from front
        print $cache_fd "$f\n";
      }
    }

    print STDERR "$progname: cached " . ($#all_files+1) . " files\n"
      if ($verbose);
  }

  flock ($cache_fd, LOCK_UN) ||
    error ("unable to unlock $cache_file_name: $!");
  close ($cache_fd);
  $cache_fd = undef;
}


sub html_unquote($) {
  my ($h) = @_;

  # This only needs to handle entities that occur in RSS, not full HTML.
  my %ent = ( 'amp' => '&', 'lt' => '<', 'gt' => '>', 
              'quot' => '"', 'apos' => "'" );
  $h =~ s/(&(\#)?([[:alpha:]\d]+);?)/
    {
     my ($o, $c) = ($1, $3);
     if (! defined($2)) {
       $c = $ent{$c};			# for &lt;
     } else {
       if ($c =~ m@^x([\dA-F]+)$@si) {	# for &#x41;
         $c = chr(hex($1));
       } elsif ($c =~ m@^\d+$@si) {	# for &#65;
         $c = chr($c);
       } else {
         $c = undef;
       }
     }
     ($c || $o);
    }
   /gexi;
  return $h;
}



# Figure out what the proxy server should be, either from environment
# variables or by parsing the output of the (MacOS) program "scutil",
# which tells us what the system-wide proxy settings are.
#
sub set_proxy($) {
  my ($ua) = @_;

  my $proxy_data = `scutil --proxy 2>/dev/null`;
  foreach my $proto ('http', 'https') {
    my ($server) = ($proxy_data =~ m/\b${proto}Proxy\s*:\s*([^\s]+)/si);
    my ($port)   = ($proxy_data =~ m/\b${proto}Port\s*:\s*([^\s]+)/si);
    my ($enable) = ($proxy_data =~ m/\b${proto}Enable\s*:\s*([^\s]+)/si);

    if ($server && $enable) {
      # Note: this ignores the "ExceptionsList".
      my $proto2 = 'http';
      $ENV{"${proto}_proxy"} = ("${proto2}://" . $server .
                                ($port ? ":$port" : "") . "/");
      print STDERR "$progname: MacOS $proto proxy: " .
                   $ENV{"${proto}_proxy"} . "\n"
        if ($verbose > 2);
    }
  }

  $ua->env_proxy();
}


sub init_lwp() {
  if (! defined ($LWP::Simple::ua)) {
    error ("\n\n\tPerl is broken. Do this to repair it:\n" .
           "\n\tsudo cpan LWP::Simple LWP::Protocol::https Mozilla::CA\n");
  }
  set_proxy ($LWP::Simple::ua);
}


sub sanity_check_lwp() {
  my $url1 = 'https://www.mozilla.org/';
  my $url2 =  'http://www.mozilla.org/';
  my $body = (LWP::Simple::get($url1) || '');
  if (length($body) < 10240) {
    my $err = "";
    $body = (LWP::Simple::get($url2) || '');
    if (length($body) < 10240) {
      $err = "Perl is broken: neither HTTP nor HTTPS URLs work.";
    } else {
      $err = "Perl is broken: HTTP URLs work but HTTPS URLs don't.";
    }
    $err .= "\nMaybe try: sudo cpan -f Mozilla::CA LWP::Protocol::https";
    $err =~ s/^/\t/gm;
    error ("\n\n$err\n");
  }
}


# If the URL does not already end with an extension appropriate for the
# content-type, add it after a "#" search.
#
# This is for when we know the content type of the URL, but the URL is
# some crazy thing without an extension. The files on disk need to have
# proper extensions.
#
sub force_extension($$) {
  my ($url, $ct) = @_;
  return $url unless (defined($url) && defined($ct));
  my ($ext) = ($ct =~ m@^image/([-a-z\d]+)@si);
  return $url unless $ext;
  $ext = lc($ext);
  $ext = 'jpg' if ($ext eq 'jpeg');
  return $url if ($url =~ m/\.$ext$/si);
  return "$url#.$ext";
}


# Returns a list of the image enclosures in the RSS or Atom feed.
# Elements of the list are references, [ "url", "guid" ].
#
sub parse_feed($);
sub parse_feed($) {
  my ($url) = @_;

  init_lwp();
  $LWP::Simple::ua->agent ("$progname/$version");
  $LWP::Simple::ua->timeout (10);  # bail sooner than the default of 3 minutes


  # Half the time, random Linux systems don't have Mozilla::CA installed,
  # which results in "Can't verify SSL peers without knowning which
  # Certificate Authorities to trust".
  #
  # In xscreensaver-text we just disabled certificate checks. However,
  # malicious images really do exist, so for xscreensaver-getimage-file,
  # let's actually require that SSL be installed properly.

  print STDERR "$progname: loading $url\n" if ($verbose);
  my $body = (LWP::Simple::get($url) || '');

  if ($body !~ m@^\s*<(\?xml|rss)\b@si) {
    # Not an RSS/Atom feed.  Try RSS autodiscovery.

    # (Great news, everybody: Flickr no longer provides RSS for "Sets",
    # only for "Photostreams", and only the first 20 images of those.
    # Thanks, assholes.)

    if ($body =~ m/^\s*$/s) {
      sanity_check_lwp();
      error ("null response: $url");
    }

    error ("not an RSS or Atom feed, or HTML: $url")
      unless ($body =~ m@<(HEAD|BODY|A|IMG)\b@si);

    # Find the first <link> with RSS or Atom in it, and use that instead.

    $body =~ s@<LINK\s+([^<>]*)>@{
      my $p = $1;
      if ($p =~ m! \b REL  \s* = \s* ['"]? alternate \b!six &&
          $p =~ m! \b TYPE \s* = \s* ['"]? application/(atom|rss) !six &&
          $p =~ m! \b HREF \s* = \s* ['"]  ( [^<>'"]+ ) !six
         ) {
        my $u2 = html_unquote ($1);
        if ($u2 =~ m!^/!s) {
          my ($h) = ($url =~ m!^([a-z]+://[^/]+)!si);
          $u2 = "$h$u2";
        }
        print STDERR "$progname: found feed: $u2\n"
          if ($verbose);
        return parse_feed ($u2);
      }
      '';
    }@gsexi;

    error ("no RSS or Atom feed for HTML page: $url");
  }


  $body =~ s@(<ENTRY|<ITEM)@\001$1@gsi;
  my @items = split(/\001/, $body);
  shift @items;

  my @imgs = ();
  my %ids;

  foreach my $item (@items) {
    my $iurl = undef;
    my $id = undef;

    # First look for <link rel="enclosure" href="...">
    #
    if (! $iurl) {
      foreach my $link ($item =~ m@<LINK[^<>]*>@gsi) {
        last if $iurl;
        my ($href) = ($link =~ m/\bHREF\s*=\s*[\"\']([^<>\'\"]+)/si);
        my ($type) = ($link =~ m/\bTYPE\s*=\s*[\"\']?([^<>\'\"]+)/si);
        my ($rel)  = ($link =~ m/\bREL\s*=\s*[\"\']?([^<>\'\"]+)/si);
        $href = undef unless (lc($rel || '') eq 'enclosure');
        $href = undef if ($type && $type !~ m@^image/@si);  # omit videos
        $iurl = html_unquote($href) if $href;
        $iurl = force_extension ($iurl, $type);
      }
    }

    # Then look for <media:content url="...">
    #
    if (! $iurl) {
      foreach my $link ($item =~ m@<MEDIA:CONTENT[^<>]*>@gsi) {
        last if $iurl;
        my ($href) = ($link =~ m/\bURL\s*=\s*[\"\']([^<>\'\"]+)/si);
        my ($type) = ($link =~ m/\bTYPE\s*=\s*[\"\']?([^<>\'\"]+)/si);
        my ($med)  = ($link =~ m/\bMEDIUM\s*=\s*[\"\']?([^<>\'\"]+)/si);
        $type = 'image/jpeg' if (!$type && lc($med || '') eq 'image');
        $href = undef if ($type && $type !~ m@^image/@si);  # omit videos
        $iurl = html_unquote($href) if $href;
        $iurl = force_extension ($iurl, $type);
      }
    }

    # Then look for <enclosure url="..."/> 
    #
    if (! $iurl) {
      foreach my $link ($item =~ m@<ENCLOSURE[^<>]*>@gsi) {
        last if $iurl;
        my ($href) = ($link =~ m/\bURL\s*=\s*[\"\']([^<>\'\"]+)/si);
        my ($type) = ($link =~ m/\bTYPE\s*=\s*[\"\']?([^<>\'\"]+)/si);
        $href = undef if ($type && $type !~ m@^image/@si);  # omit videos
        $iurl = html_unquote($href) if ($href);
        $iurl = force_extension ($iurl, $type);
      }
    }

    # Ok, maybe there's an image in the <url> field?
    #
    if (! $iurl) {
      foreach my $link ($item =~ m@<URL\b[^<>]*>([^<>]*)@gsi) {
        last if $iurl;
        my $u2 = $1;
        $iurl = html_unquote($u2) if ($u2 =~ m/$good_file_re/io);
        if (! $iurl) {
          my $u3 = $u2;
          $u3 =~ s/#.*$//gs;
          $u3 =~ s/[?&].*$//gs;
          $iurl = html_unquote($u2) if ($u3 =~ m/$good_file_re/io);
        }
      }
    }

    # Then look for <content:encoded> or <description>... with an
    # <img src="..."> inside. If more than one image, take the first.
    #
    foreach my $t ('content:encoded', 'description') {
      last if $iurl;
      foreach my $link ($item =~ m@<$t[^<>]*>(.*?)</$t>@gsi) {
        last if $iurl;
        my $desc = $1;
        if ($desc =~ m@<!\[CDATA\[\s*(.*?)\s*\]\]>@gs) {
          $desc = $1;
        } else {
          $desc = html_unquote($desc);
        }
        my ($href) = ($desc =~ m@<IMG[^<>]*\bSRC=[\"\']?([^\"\'<>]+)@si);
        $iurl = html_unquote($href) if ($href);
        # If IMG SRC has a bogus extension, pretend it's a JPEG.
        $iurl = force_extension ($iurl, 'image/jpeg')
          if ($iurl && $iurl !~ m/$good_file_re/io);
      }
    }

    # Find a unique ID for this image, to defeat image farms.
    # First look for <id>...</id>
    ($id) = ($item =~ m!<ID\b[^<>]*>\s*([^<>]+?)\s*</ID>!si) unless $id;

    # Then look for <guid isPermaLink=...> ... </guid>
    ($id) = ($item =~ m!<GUID\b[^<>]*>\s*([^<>]+?)\s*</GUID>!si) unless $id;

    # Then look for <link> ... </link>
    ($id) = ($item =~ m!<LINK\b[^<>]*>\s*([^<>]+?)\s*</LINK>!si) unless $id;

    # If we only have a GUID or LINK, but it's an image, use that.
    $iurl = $id if (!$iurl && $id && $id =~ m/$good_file_re/io);

    if ($iurl) {
      $id = $iurl unless $id;
      my $o = $ids{$id};
      if (! $o) {
        $ids{$id} = $iurl;
        my @P = ($iurl, $id);
        push @imgs, \@P;
      } elsif ($iurl ne $o) {
        print STDERR "$progname: WARNING: dup ID \"$id\"" .
                     " for \"$o\" and \"$iurl\"\n";
      }
    }
  }

  return @imgs;
}


# Like md5_base64 but uses filename-safe characters.
#
sub md5_file($) {
  my ($s) = @_;
  $s = md5_base64($s);
  $s =~ s@[/]@_@gs;
  $s =~ s@[+]@-@gs;
  return $s;
}


# expands the first URL relative to the second.
#
sub expand_url($$) {
  my ($url, $base) = @_;

  $url =~ s/^\s+//gs;  # lose whitespace at front and back
  $url =~ s/\s+$//gs;

  if (! ($url =~ m/^[a-z]+:/)) {

    $base =~ s@(\#.*)$@@;       # strip anchors
    $base =~ s@(\?.*)$@@;       # strip arguments
    $base =~ s@/[^/]*$@/@;      # take off trailing file component

    my $tail = '';
    if ($url =~ s@(\#.*)$@@) { $tail = $1; }         # save anchors
    if ($url =~ s@(\?.*)$@@) { $tail = "$1$tail"; }  # save arguments

    my $base2 = $base;

    $base2 =~ s@^([a-z]+:/+[^/]+)/.*@$1@        # if url is an absolute path
      if ($url =~ m@^/@);

    my $ourl = $url;

    $url = $base2 . $url;
    $url =~ s@/\./@/@g;                         # expand "."
    1 while ($url =~ s@/[^/]+/\.\./@/@s);       # expand ".."

    $url .= $tail;                              # put anchors/args back

    print STDERR "$progname: relative URL: $ourl --> $url\n"
      if ($verbose > 1);

  } else {
    print STDERR "$progname: absolute URL: $url\n"
      if ($verbose > 2);
  }

  return $url;
}


# Given the URL of an image, download it into the given directory
# and return the file name.
#
sub download_image($$$) {
  my ($url, $uid, $dir) = @_;

  my $url2 = $url;
  $url2 =~ s/\#.*$//s;		# Omit search terms after file extension
  $url2 =~ s/\?.*$//s;
  my ($ext) = ($url  =~ m@\.([a-z\d]+)$@si);
     ($ext) = ($url2 =~ m@\.([a-z\d]+)$@si) unless $ext;

  # If the feed hasn't put a sane extension on their URLs, nothing's going
  # to work. This code assumes that file names have extensions, even the
  # ones in the cache directory.
  #
  if (! $ext) {
    print STDERR "$progname: skipping extensionless URL: $url\n"
      if ($verbose > 1);
    return undef;
  }

  # Don't bother downloading files that we will reject anyway.
  #
  if (! ($url  =~ m/$good_file_re/io ||
         $url2 =~ m/$good_file_re/io)) {
    print STDERR "$progname: skipping non-image URL: $url\n"
      if ($verbose > 1);
    return undef;
  }

  my $file = md5_file ($uid);
  $file .= '.' . lc($ext) if $ext;

  # Don't bother doing If-Modified-Since to see if the URL has changed.
  # If we have already downloaded it, assume it's good.
  if (-f "$dir/$file") {
    print STDERR "$progname: exists: $dir/$file for $uid / $url\n" 
      if ($verbose > 1);
    return $file;
  }

  # Special-case kludge for Flickr:
  # Their RSS feeds sometimes include only the small versions of the images.
  # So if the URL ends in one of the "small-size" letters, change it to "b".
  #
  #     _o  orig,  1600 +
  #     _k  large, 2048 max
  #     _h  large, 1600 max
  #     _b  large, 1024 max
  #     _c  medium, 800 max
  #     _z  medium, 640 max
  #     ""  medium, 500 max
  #     _n  small,  320 max
  #     _m  small,  240 max
  #     _t  thumb,  100 max
  #     _q  square, 150x150
  #     _s  square,  75x75
  #
  # Note: if we wanted to get the _k or _o version instead of the _b or _h
  # version, we'd need to crack the DRM -- which is easy: see crack_secret
  # in "https://www.jwz.org/hacks/galdown".
  #
  $url =~ s@_[sqtmnzc](\.[a-z]+)$@_b$1@si
    if ($url =~ m@^https?://[^/?#&]*?flickr\.com/@si);

  print STDERR "$progname: downloading: $dir/$file for $uid / $url\n" 
    if ($verbose > 1);
  init_lwp();
  $LWP::Simple::ua->agent ("$progname/$version");

  $url =~ s/\#.*$//s;		# Omit search terms
  my $status = LWP::Simple::mirror ($url, "$dir/$file");
  if (!LWP::Simple::is_success ($status)) {
    print STDERR "$progname: error $status: $url\n";   # keep going
  }

  return $file;
}


sub mirror_feed($) {
  my ($url) = @_;

  if ($url !~ m/^https?:/si) {   # not a URL: local directory.
    return (undef, $url);
  }

  my $dir = "$ENV{HOME}/Library/Caches";    # MacOS location
  if (-d $dir) {
    $dir = "$dir/org.jwz.xscreensaver.feeds";
  } elsif (-d "$ENV{HOME}/.cache") {	   # Gnome "FreeDesktop XDG" location
    $dir = "$ENV{HOME}/.cache/xscreensaver";
    if (! -d $dir) { mkdir ($dir) || error ("mkdir $dir: $!"); }
    $dir .= "/feeds";
    if (! -d $dir) { mkdir ($dir) || error ("mkdir $dir: $!"); }
  } elsif (-d "$ENV{HOME}/tmp") {	   # If ~/tmp/ exists, use it.
    $dir = "$ENV{HOME}/tmp/.xscreensaver-feeds";
  } else {
    $dir = "$ENV{HOME}/.xscreensaver-feeds";
  }

  if (! -d $dir) {
    mkdir ($dir) || error ("mkdir $dir: $!");
    print STDERR "$progname: mkdir $dir/\n" if ($verbose);
  }

  # MD5 for directory name to use for cache of a feed URL.
  $dir .= '/' . md5_file ($url);

  if (! -d $dir) {
    mkdir ($dir) || error ("mkdir $dir: $!");
    print STDERR "$progname: mkdir $dir/ for $url\n" if ($verbose);
  }

  # At this point, we have the directory corresponding to this URL.
  # Now check to see if the files in it are up to date, and download
  # them if not.

  my $stamp = '.timestamp';
  my $lock = "$dir/$stamp";

  print STDERR "$progname: awaiting lock: $lock\n"
    if ($verbose > 1);

  my $mtime = ((stat($lock))[9]) || 0;

  my $lock_fd;
  open ($lock_fd, '+>>', $lock) || error ("unable to write $lock: $!");
  flock ($lock_fd, LOCK_EX)     || error ("unable to lock $lock: $!");
  seek ($lock_fd, 0, 0)         || error ("unable to rewind $lock: $!");

  my $poll_p = ($mtime + $feed_max_age < time);

  # --no-cache cmd line arg means poll again right now.
  $poll_p = 1 unless ($cache_p);

  # Even if the cache is young, make sure there is at least one file,
  # and re-check if not.
  #
  if (! $poll_p) {
    my $count = 0;
    opendir (my $dirh, $dir) || error ("$dir: $!");
    foreach my $f (readdir ($dirh)) {
      next if ($f =~ m/^\./s);
      $count++;
      last;
    }
    closedir $dirh;

    if ($count <= 0) {
      print STDERR "$progname: no image files in cache of $url\n"
        if ($verbose);
      $poll_p = 1;
    }
  }

  if ($poll_p) {

    print STDERR "$progname: loading $url\n" if ($verbose);

    my %files;
    opendir (my $dirh, $dir) || error ("$dir: $!");
    foreach my $f (readdir ($dirh)) {
      next if ($f eq '.' || $f eq '..');
      $files{$f} = 0;  # 0 means "file exists, should be deleted"
    }
    closedir $dirh;

    $files{$stamp} = 1;

    # Download each image currently in the feed.
    #
    my $count = 0;
    my @urls = parse_feed ($url);
    print STDERR "$progname: " . ($#urls + 1) . " images\n"
      if ($verbose > 1);
    my %seen_src_urls;
    foreach my $p (@urls) {
      my ($furl, $id) = @$p;
      $furl = expand_url ($furl, $url);

      # No need to download the same image twice, even if it was in the feed
      # multiple times under different GUIDs.
      next if ($seen_src_urls{$furl});
      $seen_src_urls{$furl} = 1;

      my $f = download_image ($furl, $id, $dir);
      next unless $f;
      $files{$f} = 1;    # Got it, don't delete
      $count++;
    }

    my $empty_p = ($count <= 0);

    # Now delete any files that are no longer in the feed.
    # But if there was nothing in the feed (network failure?)
    # then don't blow away the old files.
    #
    my $kept = 0;
    foreach my $f (keys(%files)) {
      if ($count <= 0) {
        $kept++;
      } elsif ($files{$f}) {
        $kept++;
      } else {
        if (unlink ("$dir/$f")) {
          print STDERR "$progname: rm $dir/$f\n" if ($verbose > 1);
        } else {
          print STDERR "$progname: rm $dir/$f: $!\n";   # don't bail
        }
      }
    }

    # Both feed and cache are empty. No files at all. Bail.
    error ("empty feed: $url") if ($kept <= 1);

    # Feed is empty, but we have some files from last time. Warn.
    print STDERR "$progname: empty feed: using cache: $url\n"
      if ($empty_p);

    $mtime = time();	# update the timestamp

  } else {

    # Not yet time to re-check the URL.
    print STDERR "$progname: using cache: $url\n" if ($verbose);

  }

  # Unlock and update the write date on the .timestamp file.
  #
  truncate ($lock_fd, 0) || error ("unable to truncate $lock: $!");
  seek ($lock_fd, 0, 0)  || error ("unable to rewind $lock: $!");
  utime ($mtime, $mtime, $lock_fd) || error ("unable to touch $lock: $!");
  flock ($lock_fd, LOCK_UN) || error ("unable to unlock $lock: $!");
  close ($lock_fd);
  $lock_fd = undef;
  print STDERR "$progname: unlocked $lock\n" if ($verbose > 1);

  # Don't bother using the imageDirectory cache.  We know that this directory
  # is flat, and we can assume that an RSS feed doesn't contain 100,000 images
  # like ~/Pictures/ might.
  #
  $cache_p = 0;

  # Return the URL and directory name of the files of that URL's local cache.
  #
  return ($url, $dir);
}


sub find_random_file($) {
  my ($dir) = @_;

  if ($use_spotlight_p == -1) {
    $use_spotlight_p = 0;
    if (-x '/usr/bin/mdfind') {
      $use_spotlight_p = 1;
    }
  }

  my $url;
  ($url, $dir) = mirror_feed ($dir);

  if ($url) {
    $use_spotlight_p = 0;
    print STDERR "$progname: $dir is cache for $url\n" if ($verbose > 1);
  }

  @all_files = read_cache ($dir);

  if ($#all_files >= 0) {
    # got it from the cache...

  } elsif ($use_spotlight_p) {
    print STDERR "$progname: spotlighting $dir...\n" if ($verbose);
    spotlight_all_files ($dir);
    print STDERR "$progname: found " . ($#all_files+1) .
                 " file" . ($#all_files == 0 ? "" : "s") .
                 " via Spotlight\n"
      if ($verbose);
  } else {
    print STDERR "$progname: recursively reading $dir...\n" if ($verbose);
    find_all_files ($dir);
    print STDERR "$progname: " .
                 "f=" . ($#all_files+1) . "; " .
                 "d=$dir_count; " .
                 "s=$stat_count; " .
                 "skip=${skip_count_unstat}+$skip_count_stat=" .
                  ($skip_count_unstat + $skip_count_stat) .
                 ".\n"
      if ($verbose);
  }

  write_cache ($dir);

  if ($#all_files < 0) {
    print STDERR "$progname: no image files in $dir\n";
    exit 1;
  }

  my $max_tries = 50;
  my $total_files = @all_files;
  my $sparse_p = ($total_files < 20);

  # If the directory has a lot of files in it:
  #   Make a pass through looking for hirez files (assume some are thumbs);
  #   If we found none, then, select any other file at random.
  # Otherwise if there are a small number of files:
  #   Just select one at random (in case there's like, just one hirez).

  for (my $check_size_p = $sparse_p ? 0 : 1;
       $check_size_p >= 0; $check_size_p--) {

    for (my $i = 0; $i < $max_tries; $i++) {
      my $n = int (rand ($total_files));
      my $file = $all_files[$n];
      if (!$check_size_p || large_enough_p ($file)) {
        if (! $url) {
          $file =~ s@^\Q$dir/@@so || die;  # remove $dir from front
        }
        return $file;
      }
    }
  }

  print STDERR "$progname: no suitable images in " . ($url || $dir) . " -- " .
               ($total_files <= $max_tries
                ? "all $total_files images"
                : "$max_tries of $total_files images") .
               " are smaller than ${min_image_width}x${min_image_height}.\n";

  # If we got here, blow away the cache.  Maybe it's stale.
  unlink $cache_file_name if $cache_file_name;

  exit 1;
}


sub large_enough_p($) {
  my ($file) = @_;

  my ($w, $h) = image_file_size ($file);

  if (!defined ($h)) {

    # Nonexistent files are obviously too small!
    # Already printed $verbose message about the file not existing.
    return 0 unless -f $file;

    print STDERR "$progname: $file: unable to determine image size\n"
      if ($verbose);
    # Assume that unknown files are of good sizes: this will happen if
    # they matched $good_file_re, but we don't have code to parse them.
    # (This will also happen if the file is junk...)
    return 1;
  }

  if ($w < $min_image_width || $h < $min_image_height) {
    print STDERR "$progname: $file: too small ($w x $h)\n" if ($verbose);
    return 0;
  }

  print STDERR "$progname: $file: $w x $h\n" if ($verbose);
  return 1;
}



# Given the raw body of a GIF document, returns the dimensions of the image.
#
sub gif_size($) {
  my ($body) = @_;
  my $type = substr($body, 0, 6);
  my $s;
  return () unless ($type =~ /GIF8[7,9]a/);
  $s = substr ($body, 6, 10);
  my ($a,$b,$c,$d) = unpack ("C"x4, $s);
  return (($b<<8|$a), ($d<<8|$c));
}

# Given the raw body of a JPEG document, returns the dimensions of the image.
#
sub jpeg_size($) {
  my ($body) = @_;
  my $i = 0;
  my $L = length($body);

  my $c1 = substr($body, $i, 1); $i++;
  my $c2 = substr($body, $i, 1); $i++;
  return () unless (ord($c1) == 0xFF && ord($c2) == 0xD8);

  my $ch = "0";
  while (ord($ch) != 0xDA && $i < $L) {
    # Find next marker, beginning with 0xFF.
    while (ord($ch) != 0xFF) {
      return () if (length($body) <= $i);
      $ch = substr($body, $i, 1); $i++;
    }
    # markers can be padded with any number of 0xFF.
    while (ord($ch) == 0xFF) {
      return () if (length($body) <= $i);
      $ch = substr($body, $i, 1); $i++;
    }

    # $ch contains the value of the marker.
    my $marker = ord($ch);

    if (($marker >= 0xC0) &&
        ($marker <= 0xCF) &&
        ($marker != 0xC4) &&
        ($marker != 0xCC)) {  # it's a SOFn marker
      $i += 3;
      return () if (length($body) <= $i);
      my $s = substr($body, $i, 4); $i += 4;
      my ($a,$b,$c,$d) = unpack("C"x4, $s);
      return (($c<<8|$d), ($a<<8|$b));

    } else {
      # We must skip variables, since FFs in variable names aren't
      # valid JPEG markers.
      return () if (length($body) <= $i);
      my $s = substr($body, $i, 2); $i += 2;
      my ($c1, $c2) = unpack ("C"x2, $s);
      my $length = ($c1 << 8) | $c2;
      return () if ($length < 2);
      $i += $length-2;
    }
  }
  return ();
}

# Given the raw body of a PNG document, returns the dimensions of the image.
#
sub png_size($) {
  my ($body) = @_;
  return () unless ($body =~ m/^\211PNG\r/s);
  my ($bits) = ($body =~ m/^.{12}(.{12})/s);
  return () unless defined ($bits);
  return () unless ($bits =~ /^IHDR/);
  my ($ign, $w, $h) = unpack("a4N2", $bits);
  return ($w, $h);
}


# Given the raw body of an SVG document, returns the dimensions of the image.
#
sub svg_size($) {
  my ($body) = @_;
  return () unless ($body =~ m/^<\?xml\s/s);
  return () unless ($body =~ m/<svg\s/si);
  my ($w) = ($body =~ m@\swidth=[\"\'](\d+)[\"\']@si);
  my ($h) = ($body =~ m@\sheight=[\"\'](\d+)[\"\']@si);
  return () unless (defined ($w) && defined ($h));
  return ($w, $h);
}


# Given the raw body of a GIF, JPEG, PNG or SVG document, returns the
# dimensions of the image.
#
sub image_size($) {
  my ($body) = @_;
  return () if (length($body) < 10);
  my ($w, $h) = gif_size ($body);
  if ($w && $h) { return ($w, $h); }
  ($w, $h) = jpeg_size ($body);
  if ($w && $h) { return ($w, $h); }
  ($w, $h) = svg_size ($body);
  if ($w && $h) { return ($w, $h); }
  # #### TODO: need image parsers for TIFF, XPM, XBM.
  return png_size ($body);
}

# Returns the dimensions of the image file.
#
sub image_file_size($) {
  my ($file) = @_;
  my $in;
  if (! open ($in, '<:raw', $file)) {
    print STDERR "$progname: $file: $!\n" if ($verbose);
    return ();
  }
  my $body = '';
  sysread ($in, $body, 1024 * 50);  # The first 50k should be enough.
  close $in;			    # (It's not for certain huge jpegs...
  return image_size ($body);	    # but we know they're huge!)
}


# Reads the prefs we use from ~/.xscreensaver
#
sub get_x11_prefs() {
  my $got_any_p = 0;

  if (open (my $in, '<', $config_file)) {
    print STDERR "$progname: reading $config_file\n" if ($verbose > 1);
    local $/ = undef;  # read entire file
    my $body = <$in>;
    close $in;
    $got_any_p = get_x11_prefs_1 ($body);

  } elsif ($verbose > 1) {
    print STDERR "$progname: $config_file: $!\n";
  }

  if (! $got_any_p && defined ($ENV{DISPLAY})) {
    # We weren't able to read settings from the .xscreensaver file.
    # Fall back to any settings in the X resource database
    # (/usr/X11R6/lib/X11/app-defaults/XScreenSaver)
    #
    print STDERR "$progname: reading X resources\n" if ($verbose > 1);
    my $body = `appres XScreenSaver xscreensaver -1`;
    $got_any_p = get_x11_prefs_1 ($body);
  }
}


sub get_x11_prefs_1($) {
  my ($body) = @_;

  my $got_any_p = 0;
  my $choosep = 1;
  $body =~ s@\\\n@@gs;
  $body =~ s@^[ \t]*#[^\n]*$@@gm;

  foreach my $line (split (/\n/, $body)) {
    $line =~ s/#.*//s;
    if ($line =~ m/^[.*]*imageDirectory:[ \t]*([^\s]+)\s*$/si) {
      $image_directory = $1;
      $got_any_p = 1;
    } elsif ($line =~ m/^[.*]*chooseRandomImages:[ \t]*([^\s]+)\s*$/si) {
      $choosep = ($1 =~ m/^true$/si ? 1 : 0);
      $got_any_p = 1;
    }
  }

  # Don't allow image files to be loaded if chooseRandomImages is false.
  $image_directory = undef unless $choosep;

  return $got_any_p;
}


sub get_cocoa_prefs($) {
  my ($id) = @_;
  print STDERR "$progname: reading Cocoa prefs: \"$id\"\n" if ($verbose > 1);
  my $v = get_cocoa_pref_1 ($id, "imageDirectory");
  $v = '~/Pictures' unless defined ($v);  # Match default in XScreenSaverView
  $image_directory = $v if defined ($v);
}


sub get_cocoa_pref_1($$) {
  my ($id, $key) = @_;
  # make sure there's nothing stupid/malicious in either string.
  $id  =~ s/[^-a-z\d. ]/_/gsi;
  $key =~ s/[^-a-z\d. ]/_/gsi;
  my $cmd = "defaults -currentHost read \"$id\" \"$key\"";

  print STDERR "$progname: executing $cmd\n"
    if ($verbose > 3);

  my $val = `$cmd 2>/dev/null`;
  $val =~ s/^\s+//s;
  $val =~ s/\s+$//s;

  print STDERR "$progname: Cocoa: $id $key = \"$val\"\n"
    if ($verbose > 2);

  $val = undef if ($val =~ m/^$/s);

  return $val;
}


sub error($) {
  my ($err) = @_;
  print STDERR "$progname: $err\n";
  exit 1;
}

sub usage() {
  print STDERR "usage: $progname [--verbose] [ directory-or-feed-url ]\n\n" .
  "       Prints the name of a randomly-selected image file.  The directory\n" .
  "       is searched recursively.  Images smaller than " .
         "${min_image_width}x${min_image_height} are excluded.\n" .
  "\n" .
  "       The directory may also be the URL of an RSS/Atom feed.  Enclosed\n" .
  "       images will be downloaded and cached locally.\n" .
  "\n";
  exit 1;
}

sub main() {
  my $cocoa_id = undef;
  my $abs_p = 0;

  while ($_ = $ARGV[0]) {
    shift @ARGV;
    if    (m/^--?verbose$/s)      { $verbose++; }
    elsif (m/^-v+$/s)             { $verbose += length($_)-1; }
    elsif (m/^--?name$/s)         { }   # ignored, for compatibility
    elsif (m/^--?spotlight$/s)    { $use_spotlight_p = 1; }
    elsif (m/^--?no-spotlight$/s) { $use_spotlight_p = 0; }
    elsif (m/^--?cache$/s)        { $cache_p = 1; }
    elsif (m/^--?no-?cache$/s)    { $cache_p = 0; }
    elsif (m/^--?flush-?cache$/s) { $feed_max_age = $cache_max_age = 0; }
    elsif (m/^--?cocoa$/)         { $cocoa_id = shift @ARGV; }
    elsif (m/^--?abs(olute)?$/)   { $abs_p = 1; }
    elsif (m/^-./)                { usage; }
    elsif (!defined($image_directory)) { $image_directory = $_; }
    else                          { usage; }
  }

  # Most hacks (X11 and Cocoa) pass a --directory value on the command line,
  # but if they don't, look it up from the resources.  Currently this only
  # happens with "glitchpeg" which invokes xscreensaver-getimage-file
  # directly instead of going through the traditional path.
  #
  if (! $image_directory) {
    if (!defined ($cocoa_id)) {
      # see OSX/XScreenSaverView.m
      $cocoa_id = $ENV{XSCREENSAVER_CLASSPATH};
    }

    if (defined ($cocoa_id)) {
      get_cocoa_prefs ($cocoa_id);
    } else {
      get_x11_prefs();
    }
    error ("image file loading is not configured") unless $image_directory;
  }

  usage unless (defined($image_directory));

  $image_directory =~ s@^feed:@http:@si;

  if ($image_directory =~ m/^https?:/si) {
    # ok
  } else {
    $image_directory =~ s@^~/@$ENV{HOME}/@s;     # allow literal "~/"
    $image_directory =~ s@/+$@@s;		 # omit trailing /

    if (! -d $image_directory) {
      print STDERR "$progname: $image_directory not a directory or URL\n";
      usage;
    }
  }

  my $file = find_random_file ($image_directory);

  # With --absolute return fully qualified paths instead of relative to --dir.
  if ($abs_p &&
      $file !~ m@^/@ &&
      $image_directory =~ m@^/@s) {
    $file = "$image_directory/$file";
    $file =~ s@//+@/@gs;
  }

  print STDOUT "$file\n";
}

main;
exit 0;