file_redirection duplicates are removed too in maintenance script

This commit is contained in:
Mikael Nordfeldth 2015-04-15 23:56:23 +02:00
parent 5f1e9e80c2
commit 0337a7b866

View File

@ -25,7 +25,7 @@ $longoptions = array('yes');
$helptext = <<<END_OF_HELP $helptext = <<<END_OF_HELP
remove_duplicate_file_urls.php [options] remove_duplicate_file_urls.php [options]
Remove duplicate URL entries in the file table because they for some reason were not unique. Remove duplicate URL entries in the file and file_redirection tables because they for some reason were not unique.
-y --yes do not wait for confirmation -y --yes do not wait for confirmation
@ -34,7 +34,7 @@ END_OF_HELP;
require_once INSTALLDIR.'/scripts/commandline.inc'; require_once INSTALLDIR.'/scripts/commandline.inc';
if (!have_option('y', 'yes')) { if (!have_option('y', 'yes')) {
print "About to remove duplicate file URL entries. Are you sure? [y/N] "; print "About to remove duplicate URL entries in file and file_redirection tables. Are you sure? [y/N] ";
$response = fgets(STDIN); $response = fgets(STDIN);
if (strtolower(trim($response)) != 'y') { if (strtolower(trim($response)) != 'y') {
print "Aborting.\n"; print "Aborting.\n";
@ -44,16 +44,43 @@ if (!have_option('y', 'yes')) {
$file = new File(); $file = new File();
$file->query('SELECT id, url, COUNT(*) AS c FROM file GROUP BY url HAVING c > 1'); $file->query('SELECT id, url, COUNT(*) AS c FROM file GROUP BY url HAVING c > 1');
print "\nFound {$file->N} URLs with duplicate entries in file table";
while ($file->fetch()) { while ($file->fetch()) {
// We've got a URL that is duplicated in the file table // We've got a URL that is duplicated in the file table
$dupfile = new File(); $dupfile = new File();
$dupfile->url = $file->url; $dupfile->url = $file->url;
if ($dupfile->find(true)) { if ($dupfile->find(true)) {
print "\nDeleting duplicate entries in file table for URL: {$file->url} [";
// Leave one of the URLs in the database by using ->find(true) // Leave one of the URLs in the database by using ->find(true)
// and only deleting starting with this fetch. // and only deleting starting with this fetch.
while($dupfile->fetch()) { while($dupfile->fetch()) {
print ".";
$dupfile->delete(); $dupfile->delete();
} }
print "]\n";
} else {
print "\nWarning! URL suddenly disappeared from database: {$file->url}\n";
}
}
$file = new File_redirection();
$file->query('SELECT file_id, url, COUNT(*) AS c FROM file_redirection GROUP BY url HAVING c > 1');
print "\nFound {$file->N} URLs with duplicate entries in file_redirection table";
while ($file->fetch()) {
// We've got a URL that is duplicated in the file_redirection table
$dupfile = new File_redirection();
$dupfile->url = $file->url;
if ($dupfile->find(true)) {
print "\nDeleting duplicate entries in file table for URL: {$file->url} [";
// Leave one of the URLs in the database by using ->find(true)
// and only deleting starting with this fetch.
while($dupfile->fetch()) {
print ".";
$dupfile->delete();
}
print "]\n";
} else {
print "\nWarning! URL suddenly disappeared from database: {$file->url}\n";
} }
} }
print "\nDONE.\n"; print "\nDONE.\n";