egroupware_official/etemplate/inc/class.bo_merge.inc.php

623 lines
24 KiB
PHP

<?php
/**
* EGroupware - Document merge print
*
* @link http://www.egroupware.org
* @author Ralf Becker <RalfBecker-AT-outdoor-training.de>
* @package addressbook
* @copyright (c) 2007-9 by Ralf Becker <RalfBecker-AT-outdoor-training.de>
* @license http://opensource.org/licenses/gpl-license.php GPL - GNU General Public License
* @version $Id$
*/
/**
* Document merge print
*/
abstract class bo_merge
{
/**
* Instance of the addressbook_bo class
*
* @var addressbook_bo
*/
var $contacts;
/**
* Datetime format according to user preferences
*
* @var string
*/
var $datetime_format = 'Y-m-d H:i';
/**
* Mimetype of document processed by merge
*
* @var string
*/
var $mimetype;
/**
* Plugins registered by extending class to create a table with multiple rows
*
* $$table/$plugin$$ ... $$endtable$$
*
* Callback returns replacements for row $n (stringing with 0) or null if no further rows
*
* @var array $plugin => array callback($plugin,$id,$n)
*/
var $table_plugins = array();
/**
* Constructor
*
* @return bo_merge
*/
function __construct()
{
$this->contacts = new addressbook_bo();
$this->datetime_format = $GLOBALS['egw_info']['user']['preferences']['common']['dateformat'].' '.
($GLOBALS['egw_info']['user']['preferences']['common']['timeformat']==12 ? 'h:i a' : 'H:i');
}
/**
* Get all replacements, must be implemented in extending class
*
* Can use eg. the following high level methods:
* - contact_replacements($contact_id,$prefix='')
* - format_datetime($time,$format=null)
*
* @param int $id id of entry
* @param string &$content=null content to create some replacements only if they are use
* @return array|boolean array with replacements or false if entry not found
*/
abstract protected function get_replacements($id,&$content=null);
/**
* Return if merge-print is implemented for given mime-type (and/or extension)
*
* @param string $mimetype eg. text/plain
* @param string $extension only checked for applications/msword and .rtf
*/
static public function is_implemented($mimetype,$extension=null)
{
static $zip_available;
if (is_null($zip_available))
{
$zip_available = check_load_extension('zip') &&
class_exists('ZipArchive'); // some PHP has zip extension, but no ZipArchive (eg. RHEL5!)
}
switch ($mimetype)
{
case 'application/msword':
if (strtolower($extension) != '.rtf') break;
case 'application/rtf':
case 'text/rtf':
return true; // rtf files
case 'application/vnd.oasis.opendocument.text': // oo text
case 'application/vnd.oasis.opendocument.spreadsheet': // oo spreadsheet
if (!$zip_available) break;
return true; // open office write xml files
case 'application/vnd.openxmlformats-officedocument.wordprocessingml.document': // ms word 2007 xml format
case 'application/vnd.openxmlformats-officedocument.wordprocessingml.d': // mimetypes in vfs are limited to 64 chars
case 'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet': // ms excel 2007 xml format
case 'application/vnd.openxmlformats-officedocument.spreadsheetml.shee':
if (!$zip_available) break;
return true; // ms word xml format
case 'application/xml':
return true; // alias for text/xml, eg. ms office 2003 word format
default:
if (substr($mimetype,0,5) == 'text/')
{
return true; // text files
}
break;
}
return false;
// As browsers not always return correct mime types, one could use a negative list instead
//return !($mimetype == egw_vfs::DIR_MIME_TYPE || substr($mimetype,0,6) == 'image/');
}
/**
* Return replacements for a contact
*
* @param int|string|array $contact contact-array or id
* @param string $prefix='' prefix like eg. 'user'
* @return array
*/
public function contact_replacements($contact,$prefix='')
{
if (!is_array($contact))
{
$contact = $this->contacts->read($contact);
}
if (!is_array($contact)) return array();
$replacements = array();
foreach(array_keys($this->contacts->contact_fields) as $name)
{
$value = $contact[$name];
switch($name)
{
case 'created': case 'modified':
$value = $this->format_datetime($value);
break;
case 'bday':
if ($value)
{
list($y,$m,$d) = explode('-',$value);
$value = common::dateformatorder($y,$m,$d,true);
}
break;
case 'owner': case 'creator': case 'modifier':
$value = common::grab_owner_name($value);
break;
case 'cat_id':
if ($value)
{
// if cat-tree is displayed, we return a full category path not just the name of the cat
$use = $GLOBALS['egw_info']['server']['cat_tab'] == 'Tree' ? 'path' : 'name';
$cats = array();
foreach(is_array($value) ? $value : explode(',',$value) as $cat_id)
{
$cats[] = $GLOBALS['egw']->categories->id2name($cat_id,$use);
}
$value = implode(', ',$cats);
}
break;
case 'jpegphoto': // returning a link might make more sense then the binary photo
if ($contact['photo'])
{
$value = ($GLOBALS['egw_info']['server']['webserver_url'][0] == '/' ?
($_SERVER['HTTPS'] ? 'https://' : 'http://').$_SERVER['HTTP_HOST'] : '').
$GLOBALS['egw']->link('/index.php',$contact['photo']);
}
break;
case 'tel_prefer':
if ($value && $contact[$value])
{
$value = $contact[$value];
}
break;
case 'account_id':
if ($value)
{
$replacements['$$'.($prefix ? $prefix.'/':'').'account_lid$$'] = $GLOBALS['egw']->accounts->id2name($value);
}
break;
}
if ($name != 'photo') $replacements['$$'.($prefix ? $prefix.'/':'').$name.'$$'] = $value;
}
// set custom fields
foreach($this->contacts->customfields as $name => $field)
{
$name = '#'.$name;
$value = (string)$contact[$name];
switch($field['type'])
{
case 'select-account':
if ($value) $value = common::grab_owner_name($value);
break;
case 'select':
if (count($field['values']) == 1 && isset($field['values']['@']))
{
$field['values'] = customfields_widget::_get_options_from_file($field['values']['@']);
}
$values = array();
foreach($field['rows'] > 1 ? explode(',',$value) : (array) $value as $value)
{
$values[] = $field['values'][$value];
}
$value = implode(', ',$values);
break;
case 'date':
case 'date-time':
if ($value)
{
$format = $field['len'] ? $field['len'] : ($field['type'] == 'date' ? 'Y-m-d' : 'Y-m-d H:i:s');
$date = array_combine(preg_split('/[\\/. :-]/',$format),preg_split('/[\\/. :-]/',$value));
$value = common::dateformatorder($date['Y'],$date['m'],$date['d'],true);
if (isset($date['H'])) $value .= ' '.common::formattime($date['H'],$date['i']);
}
break;
}
$replacements['$$'.($prefix ? $prefix.'/':'').$name.'$$'] = $value;
}
return $replacements;
}
/**
* Format a datetime
*
* @param int|string|DateTime $time unix timestamp or Y-m-d H:i:s string (in user time!)
* @param string $format=null format string, default $this->datetime_format
* @deprecated use egw_time::to($time='now',$format='')
* @return string
*/
protected function format_datetime($time,$format=null)
{
if (is_null($format)) $format = $this->datetime_format;
return egw_time::to($time,$format);
}
/**
* Merges a given document with contact data
*
* @param string $document path/url of document
* @param array $ids array with contact id(s)
* @param string &$err error-message on error
* @param string $mimetype mimetype of complete document, eg. text/*, application/vnd.oasis.opendocument.text, application/rtf
* @param array $fix=null regular expression => replacement pairs eg. to fix garbled placeholders
* @return string|boolean merged document or false on error
*/
public function &merge($document,$ids,&$err,$mimetype,array $fix=null)
{
if (!($content = file_get_contents($document)))
{
$err = lang("Document '%1' does not exist or is not readable for you!",$document);
return false;
}
// alternative syntax using double curly brackets (eg. {{cat_id}} instead $$cat_id$$),
// agressivly removing all xml-tags eg. Word adds within placeholders
$content = preg_replace_callback('/{{[^}]+}}/i',create_function('$p','return \'$$\'.strip_tags(substr($p[0],2,-2)).\'$$\';'),$content);
// make currently processed mimetype available to class methods;
$this->mimetype = $mimetype;
// fix garbled placeholders
if ($fix && is_array($fix))
{
$content = preg_replace(array_keys($fix),array_values($fix),$content);
//die("<pre>".htmlspecialchars($content)."</pre>\n");
}
list($contentstart,$contentrepeat,$contentend) = preg_split('/\$\$pagerepeat\$\$/',$content,-1, PREG_SPLIT_NO_EMPTY); //get differt parts of document, seperatet by Pagerepeat
if ($mimetype == 'application/vnd.oasis.opendocument.text' && count($ids) > 1)
{
//for odt files we have to split the content and add a style for page break to the style area
list($contentstart,$contentrepeat,$contentend) = preg_split('/office:body>/',$content,-1, PREG_SPLIT_NO_EMPTY); //get differt parts of document, seperatet by Pagerepeat
$contentstart = substr($contentstart,0,strlen($contentstart)-1); //remove "<"
$contentrepeat = substr($contentrepeat,0,strlen($contentrepeat)-2); //remove "</";
// need to add page-break style to the style list
list($stylestart,$stylerepeat,$styleend) = preg_split('/<\/office:automatic-styles>/',$content,-1, PREG_SPLIT_NO_EMPTY); //get differt parts of document style sheets
$contentstart = $stylestart.'<style:style style:name="P200" style:family="paragraph" style:parent-style-name="Standard"><style:paragraph-properties fo:break-before="page"/></style:style></office:automatic-styles>';
$contentstart .= '<office:body>';
$contentend = '</office:body></office:document-content>';
}
if ($mimetype == 'application/vnd.openxmlformats-officedocument.wordprocessingml.document' && count($ids) > 1)
{
//for Word 2007 XML files we have to split the content and add a style for page break to the style area
list($contentstart,$contentrepeat,$contentend) = preg_split('/w:body>/',$content,-1, PREG_SPLIT_NO_EMPTY); //get differt parts of document, seperatet by Pagerepeat
$contentstart = substr($contentstart,0,strlen($contentstart)-1); //remove "</"
$contentrepeat = substr($contentrepeat,0,strlen($contentrepeat)-2); //remove "</";
$contentstart .= '<w:body>';
$contentend = '</w:body></w:document>';
}
list($Labelstart,$Labelrepeat,$Labeltend) = preg_split('/\$\$label\$\$/',$contentrepeat,-1, PREG_SPLIT_NO_EMPTY); //get the Lable content
preg_match_all('/\$\$labelplacement\$\$/',$contentrepeat,$countlables, PREG_SPLIT_NO_EMPTY);
$countlables = count($countlables[0])+1;
preg_replace('/\$\$labelplacement\$\$/','',$Labelrepeat,1);
if ($countlables > 1) $lableprint = true;
if (count($ids) > 1 && !$contentrepeat)
{
$err = lang('for more then one contact in a document use the tag pagerepeat!');
return false;
}
foreach ((array)$ids as $id)
{
if ($contentrepeat) $content = $contentrepeat; //content to repeat
if ($lableprint) $content = $Labelrepeat;
// generate replacements
if (!($replacements = $this->get_replacements($id,$content)))
{
$err = lang('Entry not found!');
return false;
}
// some general replacements: current user, date and time
if (strpos($content,'$$user/') !== null && ($user = $GLOBALS['egw']->accounts->id2name($GLOBALS['egw_info']['user']['account_id'],'person_id')))
{
$replacements += $this->contact_replacements($user,'user');
}
$replacements['$$date$$'] = egw_time::to('now',true);
$replacements['$$datetime$$'] = egw_time::to('now');
$replacements['$$time$$'] = egw_time::to('now',false);
// does our extending class registered table-plugins AND document contains table tags
if ($this->table_plugins && preg_match_all('/\\$\\$table\\/([A-Za-z0-9_]+)\\$\\$(.*?)\\$\\$endtable\\$\\$/s',$content,$matches,PREG_SET_ORDER))
{
// process each table
foreach($matches as $match)
{
$plugin = $match[1]; // plugin name
$callback = $this->table_plugins[$plugin];
$repeat = $match[2]; // line to repeat
$repeats = '';
if (isset($callback))
{
for($n = 0; ($row_replacements = ExecMethod2($callback,$plugin,$id,$n)); ++$n)
{
$repeats .= $this->replace($repeat,$row_replacements,$mimetype);
}
}
$content = str_replace($match[0],$repeats,$content);
}
}
$content = $this->replace($content,$replacements,$mimetype);
if (strpos($content,'$$IF'))
{ //Example use to use: $$IF n_prefix~Herr~Sehr geehrter~Sehr geehrte$$
$this->replacements =& $replacements;
$content = preg_replace_callback('/\$\$IF ([0-9a-z_-]+)~(.*)~(.*)~(.*)\$\$/imU',Array($this,'replace_callback'),$content);
unset($this->replacements);
}
if (strpos($content,'$$NELF'))
{ //Example: $$NEPBR org_unit$$ sets a LF and value of org_unit, only if there is a value
$this->replacements =& $replacements;
$content = preg_replace_callback('/\$\$NELF ([0-9a-z_-]+)\$\$/imU',Array($this,'replace_callback'),$content);
unset($this->replacements);
}
if (strpos($content,'$$NENVLF'))
{ //Example: $$NEPBRNV org_unit$$ sets only a LF if there is a value for org_units, but did not add any value
$this->replacements =& $replacements;
$content = preg_replace_callback('/\$\$NENVLF ([0-9a-z_-]+)\$\$/imU',Array($this,'replace_callback'),$content);
unset($this->replacements);
}
if (strpos($content,'$$LETTERPREFIX$$'))
{ //Example use to use: $$LETTERPREFIX$$
$LETTERPREFIXCUSTOM = '$$LETTERPREFIXCUSTOM n_prefix title n_family$$';
$content = str_replace('$$LETTERPREFIX$$',$LETTERPREFIXCUSTOM,$content);
}
if (strpos($content,'$$LETTERPREFIXCUSTOM'))
{ //Example use to use for a custom Letter Prefix: $$LETTERPREFIX n_prefix title n_family$$
$this->replacements =& $replacements;
$content = preg_replace_callback('/\$\$LETTERPREFIXCUSTOM ([0-9a-z_-]+)(.*)\$\$/imU',Array($this,'replace_callback'),$content);
unset($this->replacements);
}
// remove not existing replacements (eg. from calendar array)
if (strpos($content,'$$') !== null)
{
$content = preg_replace('/\$\$[a-z0-9_\/]+\$\$/i','',$content);
}
if ($contentrepeat) $contentrep[$id] = $content;
}
if ($Labelrepeat)
{
$countpage=0;
$count=0;
$contentrepeatpages[$countpage] = $Labelstart.$Labeltend;
foreach ($contentrep as $Label)
{
$count=$count+1;
if ($count % $countlables == 0)
{
$countpage=$countpage+1;
$contentrepeatpages[$countpage] = $Labelstart.$Labeltend;
}
$contentrepeatpages[$countpage] = preg_replace('/\$\$labelplacement\$\$/',$Label,$contentrepeatpages[$countpage],1);
}
$contentrepeatpages[$countpage] = preg_replace('/\$\$labelplacement\$\$/','',$contentrepeatpages[$countpage],-1); //clean empty fields
switch($mimetype)
{
case 'application/msword':
if (strtolower(substr($document,-4)) != '.rtf') break; // no binary word documents
case 'application/rtf':
case 'text/rtf':
return $contentstart.implode('\\par \\page\\pard\\plain',$contentrepeatpages).$contentend;
case 'application/vnd.oasis.opendocument.text':
case 'application/vnd.oasis.opendocument.spreadsheet':
// todo OO writer files
break;
case 'application/vnd.openxmlformats-officedocument.wordprocessingml.document':
case 'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet':
return $contentstart.implode('<w:br w:type="page" />',$contentrep).$contentend;
}
$err = lang('%1 not implemented for %2!','$$labelplacement$$',$mimetype);
return false;
}
if ($contentrepeat)
{
switch($mimetype)
{
case 'application/msword':
if (strtolower(substr($document,-4)) != '.rtf') break; // no binary word documents
case 'application/rtf':
case 'text/rtf':
return $contentstart.implode('\\par \\page\\pard\\plain',$contentrep).$contentend;
case 'application/vnd.oasis.opendocument.text':
case 'application/vnd.oasis.opendocument.spreadsheet':
case 'application/xml':
return $contentstart.implode('',$contentrep).$contentend;
case 'application/vnd.openxmlformats-officedocument.wordprocessingml.document':
case 'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet':
return $contentstart.implode('<w:br w:type="page" />',$contentrep).$contentend;
}
$err = lang('%1 not implemented for %2!','$$pagerepeat$$',$mimetype);
return false;
}
return $content;
}
/**
* Replace placeholders in $content of $mimetype with $replacements
*
* @param string $content
* @param array $replacements name => replacement pairs
* @param string $mimetype mimetype of content
* @return string
*/
protected function replace($content,array $replacements,$mimetype)
{
switch($mimetype)
{
case 'application/vnd.oasis.opendocument.text': // open office
case 'application/vnd.oasis.opendocument.spreadsheet':
case 'application/vnd.openxmlformats-officedocument.wordprocessingml.document': // ms office 2007
case 'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet':
case 'application/xml':
case 'text/xml':
$is_xml = true;
$charset = 'utf-8'; // xml files --> always use utf-8
break;
case 'text/html':
$is_xml = true;
// fall through
default: // div. text files --> use our export-charset, defined in addressbook prefs
$charset = $this->contacts->prefs['csv_charset'];
break;
}
//error_log(__METHOD__."('$document', ... ,$mimetype) --> $charset (egw=".$GLOBALS['egw']->translation->charset().', export='.$this->contacts->prefs['csv_charset'].')');
// do we need to convert charset
if ($charset && $charset != $GLOBALS['egw']->translation->charset())
{
$replacements = $GLOBALS['egw']->translation->convert($replacements,$GLOBALS['egw']->translation->charset(),$charset);
}
if ($is_xml) // zip'ed xml document (eg. OO) --> need to encode &,<,> to not mess up xml
{
$replacements = str_replace(array('&amp;','&','<','>'),array('&amp;','&amp;','&lt;','&gt;'),$replacements);
}
return str_replace(array_keys($replacements),array_values($replacements),$content);
}
/**
* Callback for preg_replace to process $$IF
*
* @param array $param
* @return string
*/
private function replace_callback($param)
{
if (array_key_exists('$$'.$param[4].'$$',$this->replacements)) $param[4] = $this->replacements['$$'.$param[4].'$$'];
if (array_key_exists('$$'.$param[3].'$$',$this->replacements)) $param[3] = $this->replacements['$$'.$param[3].'$$'];
$replace = preg_match('/'.$param[2].'/',$this->replacements['$$'.$param[1].'$$']) ? $param[3] : $param[4];
switch($this->mimetype)
{
case 'application/rtf':
case 'text/rtf':
$LF = '}\par \pard\plain{';
break;
case 'application/vnd.oasis.opendocument.text':
case 'application/vnd.oasis.opendocument.spreadsheet':
$LF ='</text:p><text:p>';
break;
case 'application/vnd.openxmlformats-officedocument.wordprocessingml.document':
case 'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet':
$LF ='</w:r></w:p><w:r><w:t>';
break;
}
if (strpos($param[0],'$$NELF') === 0)
{ //sets a Pagebreak and value, only if the field has a value
if ($this->replacements['$$'.$param[1].'$$'] !='') $replace = $LF.$this->replacements['$$'.$param[1].'$$'];
}
if (strpos($param[0],'$$NENVLF') === 0)
{ //sets a Pagebreak without any value, only if the field has a value
if ($this->replacements['$$'.$param[1].'$$'] !='') $replace = $LF;
}
if (strpos($param[0],'$$LETTERPREFIXCUSTOM') === 0)
{ //sets a Letterprefix
$replaceprefix = array();
$replaceprefix = explode(' ',substr($param[0],21,strlen($contentstart)-2));
foreach ($replaceprefix as $key => $nameprefix)
{
if ($this->replacements['$$'.$nameprefix.'$$'] !='') $replaceprefixsort[] = $this->replacements['$$'.$nameprefix.'$$'];
}
$replace = implode($replaceprefixsort,' ');
}
return $replace;
}
/**
* Download document merged with contact(s)
*
* @param string $document vfs-path of document
* @param array $ids array with contact id(s)
* @return string with error-message on error, otherwise it does NOT return
*/
public function download($document,$ids)
{
$content_url = egw_vfs::PREFIX.$document;
switch (($mimetype = egw_vfs::mime_content_type($document)))
{
case 'application/vnd.oasis.opendocument.text':
case 'application/vnd.oasis.opendocument.spreadsheet':
$ext = $mimetype == 'application/vnd.oasis.opendocument.text' ? '.odt' : '.ods';
$archive = tempnam($GLOBALS['egw_info']['server']['temp_dir'], basename($document,$ext).'-').$ext;
copy($content_url,$archive);
$content_url = 'zip://'.$archive.'#'.($content_file = 'content.xml');
break;
case 'application/vnd.openxmlformats-officedocument.wordprocessingml.d': // mimetypes in vfs are limited to 64 chars
$mimetype = 'application/vnd.openxmlformats-officedocument.wordprocessingml.document';
case 'application/vnd.openxmlformats-officedocument.wordprocessingml.document':
$archive = tempnam($GLOBALS['egw_info']['server']['temp_dir'], basename($document,'.docx').'-').'.docx';
copy($content_url,$archive);
$content_url = 'zip://'.$archive.'#'.($content_file = 'word/document.xml');
$fix = array( // regular expression to fix garbled placeholders
'/'.preg_quote('$$</w:t></w:r><w:proofErr w:type="spellStart"/><w:r><w:t>','/').'([a-z0-9_]+)'.
preg_quote('</w:t></w:r><w:proofErr w:type="spellEnd"/><w:r><w:t>','/').'/i' => '$$\\1$$',
'/'.preg_quote('$$</w:t></w:r><w:proofErr w:type="spellStart"/><w:r><w:rPr><w:lang w:val="','/').
'([a-z]{2}-[A-Z]{2})'.preg_quote('"/></w:rPr><w:t>','/').'([a-z0-9_]+)'.
preg_quote('</w:t></w:r><w:proofErr w:type="spellEnd"/><w:r><w:rPr><w:lang w:val="','/').
'([a-z]{2}-[A-Z]{2})'.preg_quote('"/></w:rPr><w:t>$$','/').'/i' => '$$\\2$$',
'/'.preg_quote('$</w:t></w:r><w:proofErr w:type="spellStart"/><w:r><w:t>','/').'([a-z0-9_]+)'.
preg_quote('</w:t></w:r><w:proofErr w:type="spellEnd"/><w:r><w:t>','/').'/i' => '$\\1$',
'/'.preg_quote('$ $</w:t></w:r><w:proofErr w:type="spellStart"/><w:r><w:t>','/').'([a-z0-9_]+)'.
preg_quote('</w:t></w:r><w:proofErr w:type="spellEnd"/><w:r><w:t>','/').'/i' => '$ $\\1$ $',
);
break;
case 'application/vnd.openxmlformats-officedocument.spreadsheetml.shee':
$mimetype = 'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet';
case 'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet':
$archive = tempnam($GLOBALS['egw_info']['server']['temp_dir'], basename($document,'.xlsx').'-').'.xlsx';
copy($content_url,$archive);
$content_url = 'zip://'.$archive.'#'.($content_file = 'xl/sharedStrings.xml');
break;
}
if (!($merged =& $this->merge($content_url,$ids,$err,$mimetype,$fix)))
{
return $err;
}
if (isset($archive))
{
$zip = new ZipArchive;
if ($zip->open($archive,ZIPARCHIVE::CHECKCONS) !== true) throw new Exception("!ZipArchive::open('$archive',ZIPARCHIVE::OVERWRITE)");
if ($zip->addFromString($content_file,$merged) !== true) throw new Exception("!ZipArchive::addFromString('$content_file',\$merged)");
if ($zip->close() !== true) throw new Exception("!ZipArchive::close()");
unset($zip);
unset($merged);
if (substr($mimetype,0,35) == 'application/vnd.oasis.opendocument.' && // only open office archives need that, ms word files brake
file_exists('/usr/bin/zip') && version_compare(PHP_VERSION,'5.3.1','<')) // fix broken zip archives generated by current php
{
exec('/usr/bin/zip -F '.escapeshellarg($archive));
}
html::content_header(basename($document),$mimetype,filesize($archive));
readfile($archive,'r');
}
else
{
if ($mimetype == 'application/xml')
{
if (strpos($merged,'<?mso-application progid="Word.Document"?>') !== false)
{
$mimetype = 'application/msword'; // to open it automatically in word or oowriter
}
elseif (strpos($merged,'<?mso-application progid="Excel.Sheet"?>') !== false)
{
$mimetype = 'application/vnd.ms-excel'; // to open it automatically in excel or oocalc
}
}
ExecMethod2('phpgwapi.browser.content_header',basename($document),$mimetype);
echo $merged;
}
common::egw_exit();
}
}