summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorlshprung <lshprung@yahoo.com>2021-10-24 11:02:25 -0700
committerlshprung <lshprung@yahoo.com>2021-10-24 11:02:25 -0700
commit736f1c3c9c87af0255373398b4cc706a2d4de58c (patch)
tree4f27d5356e24c102eb312d9fa45f1cc9e35d2a8d
parent329eba4771e625c0ce3d6cead0cac1d47c3bf8e5 (diff)
Separators are working correctlyfeature/working_sep
-rwxr-xr-xbuild_playlists.plx87
1 files changed, 66 insertions, 21 deletions
diff --git a/build_playlists.plx b/build_playlists.plx
index a774652..df204f2 100755
--- a/build_playlists.plx
+++ b/build_playlists.plx
@@ -50,6 +50,57 @@ sub build_m3u {
}
}
+# Subroutine to determine the output files for a given file
+# @_[0] -> reference to tag hashes
+sub get_output_files {
+ my @output;
+ my %tag_hash = @_;
+ my %count_hash; #keep track of index from tag_hash
+ my %value_hash; #keep track of value from tag_hash
+ my @keys_arr; #array of keys from tag_hash
+
+ # Remove PATH from tag_hash
+ delete($tag_hash{"PATH"});
+
+ # Initialize count_hash and value_hash
+ for my $i (sort(keys %tag_hash)){
+ push(@keys_arr, $i);
+ #print("@keys_arr\n");
+ $count_hash{$i} = 0;
+ #print("$i -> $count_hash{$i}\n");
+ $value_hash{$i} = @{$tag_hash{$i}}[0];
+ #print("$i -> $value_hash{$i}\n");
+ }
+
+ # Loop through all possible combinations of tags and append to @output
+ OUTER: while(1){
+ push(@output, $output_pattern);
+ $output[-1] =~ s/[{]([^}]*)[}]/$value_hash{$1}/g;
+
+ # Update
+ INNER: for my $key_index (@keys_arr){
+ $count_hash{$key_index} = (($count_hash{$key_index}+1) % scalar(@{$tag_hash{$key_index}}));
+ $value_hash{$key_index} = @{$tag_hash{$key_index}}[$count_hash{$key_index}];
+ if($count_hash{$key_index} > 0){
+ next OUTER;
+ }
+ }
+ last OUTER;
+ }
+
+ return @output;
+}
+
+
+ #for my $i (keys %tag_hash){
+ # if (!($i eq "PATH")){
+ # print("$i -> @{$tag_hash{$i}}\n");
+ # }
+ # else{
+ # print("$i -> $tag_hash{$i}\n");
+ # }
+ #}
+
# Print a help message
sub print_help {
print
@@ -165,6 +216,9 @@ else {
# Go through each row by ID
for my $i (1..$row_count){
+ # Reset @output_files
+ @output_files = ();
+
# Get output for the PATH, plus each tag of interest; store it in tag_hash
$statement = join(',', @tags_of_interest);
@db_output = flatten_array(db_cmd($dbh, "SELECT $statement FROM $table_name WHERE ID=$i;"));
@@ -183,29 +237,20 @@ else {
}
- # TODO determine array of output files (consider making this a separate subroutine)
- #$output_file = $output_pattern;
- #$output_file =~ s/[{]([^}]*)[}]/$tag_hash{$1}/g;
- # TODO Remove duplicates from @output_file
- # TODO break up by semicolon (signifying array of tag values)
-
- #for my $j (keys %tag_hash){
- # for my $k (split(';', $tag_hash{$j})){
- # print("$j -> $k\n");
- # }
- #}
- next;
+ # determine array of output files (consider making this a separate subroutine)
+ @output_files = get_output_files(%tag_hash);
- # Determine output_file
-
- # Open the file for writing
- #open FH, ">> $output_file" or die $!;
- # DEBUG
- if (!$options{quiet}){
- #print "Opened $output_file\n";
+ # Loop to add to proper files:
+ for my $output_file (@output_files){
+ # Open the file for writing
+ open FH, ">> $output_file" or die $!;
+ # DEBUG
+ if (!$options{quiet}){
+ print "Opened $output_file\n";
+ }
+ build_m3u("$output_file", *FH, $tag_hash{PATH});
+ close FH;
}
- #build_m3u("$output_file", *FH, $tag_hash{PATH});
- close FH;
}
}