diff options
Diffstat (limited to 'framework/DataAccess/adodb/tohtml.inc.php')
-rw-r--r-- | framework/DataAccess/adodb/tohtml.inc.php | 195 |
1 files changed, 0 insertions, 195 deletions
diff --git a/framework/DataAccess/adodb/tohtml.inc.php b/framework/DataAccess/adodb/tohtml.inc.php deleted file mode 100644 index b7dca463..00000000 --- a/framework/DataAccess/adodb/tohtml.inc.php +++ /dev/null @@ -1,195 +0,0 @@ -<?php -/* - V4.72 21 Feb 2006 (c) 2000-2006 John Lim (jlim@natsoft.com.my). All rights reserved. - Released under both BSD license and Lesser GPL library license. - Whenever there is any discrepancy between the two licenses, - the BSD license will take precedence. - - Some pretty-printing by Chris Oxenreider <oxenreid@state.net> -*/ - -// specific code for tohtml -GLOBAL $gSQLMaxRows,$gSQLBlockRows,$ADODB_ROUND; - -$ADODB_ROUND=4; // rounding -$gSQLMaxRows = 1000; // max no of rows to download -$gSQLBlockRows=20; // max no of rows per table block - -// RecordSet to HTML Table -//------------------------------------------------------------ -// Convert a recordset to a html table. Multiple tables are generated -// if the number of rows is > $gSQLBlockRows. This is because -// web browsers normally require the whole table to be downloaded -// before it can be rendered, so we break the output into several -// smaller faster rendering tables. -// -// $rs: the recordset -// $ztabhtml: the table tag attributes (optional) -// $zheaderarray: contains the replacement strings for the headers (optional) -// -// USAGE: -// include('adodb.inc.php'); -// $db = ADONewConnection('mysql'); -// $db->Connect('mysql','userid','password','database'); -// $rs = $db->Execute('select col1,col2,col3 from table'); -// rs2html($rs, 'BORDER=2', array('Title1', 'Title2', 'Title3')); -// $rs->Close(); -// -// RETURNS: number of rows displayed - - -function rs2html(&$rs,$ztabhtml=false,$zheaderarray=false,$htmlspecialchars=true,$echo = true) -{ -$s ='';$rows=0;$docnt = false; -GLOBAL $gSQLMaxRows,$gSQLBlockRows,$ADODB_ROUND; - - if (!$rs) { - printf(ADODB_BAD_RS,'rs2html'); - return false; - } - - if (! $ztabhtml) $ztabhtml = "BORDER='1' WIDTH='98%'"; - //else $docnt = true; - $typearr = array(); - $ncols = $rs->FieldCount(); - $hdr = "<TABLE COLS=$ncols $ztabhtml><tr>\n\n"; - for ($i=0; $i < $ncols; $i++) { - $field = $rs->FetchField($i); - if ($field) { - if ($zheaderarray) $fname = $zheaderarray[$i]; - else $fname = htmlspecialchars($field->name); - $typearr[$i] = $rs->MetaType($field->type,$field->max_length); - //print " $field->name $field->type $typearr[$i] "; - } else { - $fname = 'Field '.($i+1); - $typearr[$i] = 'C'; - } - if (strlen($fname)==0) $fname = ' '; - $hdr .= "<TH>$fname</TH>"; - } - $hdr .= "\n</tr>"; - if ($echo) print $hdr."\n\n"; - else $html = $hdr; - - // smart algorithm - handles ADODB_FETCH_MODE's correctly by probing... - $numoffset = isset($rs->fields[0]) ||isset($rs->fields[1]) || isset($rs->fields[2]); - while (!$rs->EOF) { - - $s .= "<TR valign=top>\n"; - - for ($i=0; $i < $ncols; $i++) { - if ($i===0) $v=($numoffset) ? $rs->fields[0] : reset($rs->fields); - else $v = ($numoffset) ? $rs->fields[$i] : next($rs->fields); - - $type = $typearr[$i]; - switch($type) { - case 'D': - if (empty($v)) $s .= "<TD> </TD>\n"; - else if (!strpos($v,':')) { - $s .= " <TD>".$rs->UserDate($v,"D d, M Y") ." </TD>\n"; - } - break; - case 'T': - if (empty($v)) $s .= "<TD> </TD>\n"; - else $s .= " <TD>".$rs->UserTimeStamp($v,"D d, M Y, h:i:s") ." </TD>\n"; - break; - - case 'N': - if (abs($v) - round($v,0) < 0.00000001) - $v = round($v); - else - $v = round($v,$ADODB_ROUND); - case 'I': - $s .= " <TD align=right>".stripslashes((trim($v))) ." </TD>\n"; - - break; - /* - case 'B': - if (substr($v,8,2)=="BM" ) $v = substr($v,8); - $mtime = substr(str_replace(' ','_',microtime()),2); - $tmpname = "tmp/".uniqid($mtime).getmypid(); - $fd = @fopen($tmpname,'a'); - @ftruncate($fd,0); - @fwrite($fd,$v); - @fclose($fd); - if (!function_exists ("mime_content_type")) { - function mime_content_type ($file) { - return exec("file -bi ".escapeshellarg($file)); - } - } - $t = mime_content_type($tmpname); - $s .= (substr($t,0,5)=="image") ? " <td><img src='$tmpname' alt='$t'></td>\\n" : " <td><a - href='$tmpname'>$t</a></td>\\n"; - break; - */ - - default: - if ($htmlspecialchars) $v = htmlspecialchars(trim($v)); - $v = trim($v); - if (strlen($v) == 0) $v = ' '; - $s .= " <TD>". str_replace("\n",'<br>',stripslashes($v)) ."</TD>\n"; - - } - } // for - $s .= "</TR>\n\n"; - - $rows += 1; - if ($rows >= $gSQLMaxRows) { - $rows = "<p>Truncated at $gSQLMaxRows</p>"; - break; - } // switch - - $rs->MoveNext(); - - // additional EOF check to prevent a widow header - if (!$rs->EOF && $rows % $gSQLBlockRows == 0) { - - //if (connection_aborted()) break;// not needed as PHP aborts script, unlike ASP - if ($echo) print $s . "</TABLE>\n\n"; - else $html .= $s ."</TABLE>\n\n"; - $s = $hdr; - } - } // while - - if ($echo) print $s."</TABLE>\n\n"; - else $html .= $s."</TABLE>\n\n"; - - if ($docnt) if ($echo) print "<H2>".$rows." Rows</H2>"; - - return ($echo) ? $rows : $html; - } - -// pass in 2 dimensional array -function arr2html(&$arr,$ztabhtml='',$zheaderarray='') -{ - if (!$ztabhtml) $ztabhtml = 'BORDER=1'; - - $s = "<TABLE $ztabhtml>";//';print_r($arr); - - if ($zheaderarray) { - $s .= '<TR>'; - for ($i=0; $i<sizeof($zheaderarray); $i++) { - $s .= " <TH>{$zheaderarray[$i]}</TH>\n"; - } - $s .= "\n</TR>"; - } - - for ($i=0; $i<sizeof($arr); $i++) { - $s .= '<TR>'; - $a = &$arr[$i]; - if (is_array($a)) - for ($j=0; $j<sizeof($a); $j++) { - $val = $a[$j]; - if (empty($val)) $val = ' '; - $s .= " <TD>$val</TD>\n"; - } - else if ($a) { - $s .= ' <TD>'.$a."</TD>\n"; - } else $s .= " <TD> </TD>\n"; - $s .= "\n</TR>\n"; - } - $s .= '</TABLE>'; - print $s; -} - -?>
\ No newline at end of file |