Output Control 函数
在线手册:中文 英文
PHP手册

ob_start

(PHP 4, PHP 5)

ob_start打开输出控制缓冲

说明

bool ob_start ([ callback $output_callback [, int $chunk_size [, bool $erase ]]] )

此函数将打开输出缓冲。当输出缓冲是活跃的时候,没有输出能从脚本送出(除http标头外),相反输 出的内容被存储在内部缓冲区中。

内部缓冲区的内容可以用 ob_get_contents() 函数复制到一个字符串变量中。 想要输出存储在内部缓冲区中的内容,可以使用 ob_end_flush() 函数。另外, 使用ob_end_clean() 函数会静默丢弃掉缓冲区的内容。

Warning

当有正在调用的回调函数时,一些网络服务器(例如Apache)会改变一个脚本的工作目录。 你可以通过例如chdir(dirname($_SERVER['SCRIPT_FILENAME'])) 在回调函数中再把它改回来。

输出缓冲区是可堆叠的,这即意谓着,当有一个ob_start() 是活跃的时, 你可以调用另一个 ob_start() 。 只要确保又正确调用了 ob_end_flush() 恰当的次数即可。 如果有多重输出回调函数是活跃的,输出内容会一直按嵌套的顺序依次通过它们而被过滤。

参数

output_callback

可选参数 output_callback 函数可以被指定。 此函数把一个字符串当作参数并返回一个字符串。 此函数在下情况会被调用到:输出缓冲区被( ob_flush(), ob_clean() 或者相似的函数)冲刷(送出)或者被清洗的时候;或者在请求结束之际输出缓冲区内容被冲刷到浏览器时。 当 output_callback 被调用时,它将收到输出缓冲区的内容作为参数 并预期返回一个新的输出缓冲区作为结果,这个新返回的输出缓冲区内容将被送到浏览器。 如果这个 output_callback 不是一个可以调用的函数,此函数 会返回 FALSE

如果回调函数有两个参数,第二个参数会由一个位域补充,该位域由 PHP_OUTPUT_HANDLER_START, PHP_OUTPUT_HANDLER_CONTPHP_OUTPUT_HANDLER_END 组成。

如果 output_callback 返回 FALSE ,其原来的输入 内容被直接送到浏览器。

这个参数 output_callback 可以通过直接给一个 NULL 值而避开。

ob_end_clean(), ob_end_flush(), ob_clean(), ob_flush() and ob_start() 不能从一个回调函数中调用。 如果从回调函数中调用了它们,产生的行为是不明确的。 如果想要删除缓冲区的内容,从回调函数中返回一个"" (空字符串)。 更不能从一个回调函数中使用像print_r($expression, true)highlight_file($filename, true) 一样的输出缓冲函数。

Note:

在PHP 4.0.4中, ob_gzhandler() 被引入是为了简化把gz编码过 数据发送到支持压缩网页的浏览器。 ob_gzhandler() 会判定浏览器可以接受哪种类型的编码内容,并返回相应 的输出。

chunk_size

如果可选参数 chunk_size 被赋值了,在任何一个能引起缓冲区的长度等于 或超过 chunk_size 的输出操作后,缓冲区都会被刷送。 默认值 0 意味着函数仅在最后被调用,其余的特殊值可以将 chunk_size 从 1 设定到 4096。

erase

如果可选参数 erase 被赋成 FALSE,直到脚本执行完成缓冲区才被删除。 这使得,如果调用了冲刷和清洗(清除)函数,会抛出一个“notice”,并返回 FALSE 值。

返回值

成功时返回 TRUE, 或者在失败时返回 FALSE.

更新日志

版本 说明
4.3.2 在传递的 output_callback 不能被执行时,此函数 被改成返回 FALSE
4.2.0 添加了 erase 参数。

范例

Example #1 用户自定义回调函数的例子

<?php

function callback($buffer)
{
  
// replace all the apples with oranges
  
return (str_replace("apples""oranges"$buffer));
}

ob_start("callback");

?>
<html>
<body>
<p>It's like comparing apples to oranges.</p>
</body>
</html>
<?php

ob_end_flush
();

?>

以上例程会输出:

<html>
<body>
<p>It's like comparing oranges to oranges.</p>
</body>
</html>

参见


Output Control 函数
在线手册:中文 英文
PHP手册
PHP手册 - N: 打开输出控制缓冲

用户评论:

Filip Dalge (22-Mar-2012 08:24)

When a fatal error is thrown, PHP will output the current buffer of Output-Control without postprocessing before printing the error message. If you are working with several output control levels, this might not result in the desired behavior.

You can use an output callback handler to handle this and discard the output.

Therefore, use ob_start("ob_error_handler") in connection with the following:

function ob_error_handler($str) {
    $error = error_get_last();
    if ($error && $error["type"] == E_USER_ERROR || $error["type"] == E_ERROR) {
        return ini_get("error_prepend_string").
          "\nFatal error: $error[message] in $error[file] on line $error[line]\n".
          ini_get("error_append_string");
    }
    return $str;
}

loksly at gmail dot com (24-Oct-2011 01:30)

It's easy to add a function that caches the whole page you're going to generate so resources are saved. I wrote it in a file called "cacheable.php", and I included in each file that I want to be cached. It cannot manage files with ob_start inside.

Here is the code:
file "cacheable.php":
----------------------------------
<?php
   
global $cachefile;

   
$cachefilename = urlencode($_SERVER['REQUEST_URI'])."-".md5($_SERVER['REQUEST_URI']);

   
$cachefile = sys_get_temp_dir ().DIRECTORY_SEPARATOR.$cachefilename.'.cache';

   
// How long to keep cache file?
   
$cachetime = 60*5;    //5min

    // Is cache file still fresh? If so, serve it.   
   
if (file_exists($cachefile) && time() - $cachetime < filemtime($cachefile)) {
     
readfile($cachefile);
      exit;
    }

    function
cacheOutputtoFile()
    {
        global
$cachefile;
       
$webpage = ob_get_contents();
        if (
$fp = fopen($cachefile, 'w+'))
        {
            if (
flock($fp, LOCK_EX))
            {
// do an exclusive lock
               
ftruncate($fp, 0); // truncate file
                //compress unnecesary spaces
               
$busca = array('/\>[^\S ]+/s','/[^\S ]+\</s','/(\s)+/s');
               
$reemplaza = array('>','<','\\1');
               
$webpage = preg_replace($busca, $reemplaza, $webpage);
               
               
fwrite($fp, $webpage);
               
fclose($fp);
            }
        }
        echo
$webpage;
    }

   
ob_start();
   
register_shutdown_function("cacheOutputtoFile");

//note you must use register_shutdown_function instead of using "cacheOutputtoFile" as parameter of ob_start as socket is already closed when callback function is called, then the client not get the echo $webpage output.
?>

----------------------------------
usage:

<?php
require_once(realpath(dirname(__FILE__).DIRECTORY_SEPARATOR).DIRECTORY_SEPARATOR."cacheable.php");
?>
----------------------------------

Anonymous (08-Feb-2011 03:32)

In case you're in for some readable representation for flags, this is a variant as a private class member:

<?php
class foo {
    private function
getFlagsReadable($flags) {
       
$flagNames = array('PHP_OUTPUT_HANDLER_START', 'PHP_OUTPUT_HANDLER_CONT', 'PHP_OUTPUT_HANDLER_END');
       
$readable = '';
        foreach(
$flagNames as $flagName)
            if (
$flags & constant($flagName) )
               
$readable .= (strlen($readable) ? ' | ' : '') . $flagName
           
;       
        return
$readable;
    }
}
?>

Chris (25-Nov-2010 12:35)

Careful with while using functions that change headers of a page; that change will not be undone when ending output buffering.

If you for instance have a class that generates an image and sets the appropriate headers, they will still be in place after the end of ob.

For instance:
<?php
  ob_start
();
 
myClass::renderPng(); //header("Content-Type: image/png"); in here
 
$pngString = ob_get_contents();
 
ob_end_clean();
?>

will put the image bytes into $pngString, and set the content type to image/png. Though the image will not be sent to the client, the png header is still in place; if you do html output here, the browser will most likely display "image error, cannot be viewed", at least firefox does.

You need to set the correct image type (text/html) manually in this case.

sonicx (21-Nov-2009 02:58)

Hey folks,
as dan already noted, using the ob_gzhandler will make it impossible to get the content-length,
and thusly will make the download of files a little ugly.
the client cant show progress as it wont know how large it's targeted download actually is.
even if you write a custom handler, it seems impossible to set the buffer's size to a global.
the ob_handler seems to not share a global scope.
i wrote a small php to solve that problem, by not using the ob_* functions, but rather gzencode manually:

<?php
    $file
= $_REQUEST["file"] or die("Please specify the 'file' parameter.");
   
   
$allowedExtensions = array(
       
"jpg" => "image",
       
"png" => "image",
       
"gif" => "image"
   
);
   
   
$path = pathinfo($file);
    if(!
in_array(strtolower($path["extension"]), array_keys($allowedExtensions))) {
       
header("HTTP/1.0 403 Access denied");
        die(
"403<br/>You dont have access to that file, wether it exists or not.");
    }
   
   
$pack = true;
    if(!empty(
$_SERVER["HTTP_ACCEPT_ENCODING"]) && strpos("gzip",$_SERVER["HTTP_ACCEPT_ENCODING"]) === NULL)
       
$pack = false;
       
    if(!
is_file($file) || !is_readable($file)) {
       
header("HTTP/1.0 404 Not Found");
        die(
"404<br/>The file you are looking for is not available.");
    }
   
   
$buffer = file_get_contents($file);
    if(
$pack) {
       
header("Content-Encoding: gzip");
       
$buffer = gzencode($buffer,9,true);
    }
   
$length = strlen($buffer);
   
header("Content-Type: {$allowedExtensions[$path["extension"]]}/{$path["extension"]}");   
    
header("Content-Length: ".$length);
    die(
$buffer);
?>

this will proxy images to a client, gzipping them at the highest compression-level.
to save the compression's overhead one could cache compressed outputs.

Bitwise (21-Oct-2009 08:41)

There is no start flag problem. One just has to notice that the second parameter is not a mode but consists of bitwise-OR'ed flags.

<?php
function ob_handler($string, $flags) {
        static
$input = array();
        if (
$flags & PHP_OUTPUT_HANDLER_START )
               
$flags_sent[] = "PHP_OUTPUT_HANDLER_START";
        if (
$flags & PHP_OUTPUT_HANDLER_CONT )
               
$flags_sent[] = "PHP_OUTPUT_HANDLER_CONT";
        if (
$flags & PHP_OUTPUT_HANDLER_END )
               
$flags_sent[] = "PHP_OUTPUT_HANDLER_END";
       
$input[] = implode(' | ', $flags_sent) . " ($flags): $string<br />";
       
$output  = "$string<br />";
        if (
$flags & PHP_OUTPUT_HANDLER_END ) {
               
$output .= '<br />';
                foreach(
$input as $k => $v) $output .= "$k: $v";
        }
        return
$output;
}

ob_start('ob_handler');

echo
'flush';
ob_flush();

echo
'flush 2';
ob_flush();

echo
'clean';
ob_clean();

echo
'flush 3';
ob_flush();

echo
'end flush';
ob_end_flush();
?>

flush
flush 2
flush 3
end flush

0: PHP_OUTPUT_HANDLER_START | PHP_OUTPUT_HANDLER_CONT (3): flush
1: PHP_OUTPUT_HANDLER_CONT (2): flush 2
2: PHP_OUTPUT_HANDLER_CONT (2): clean
3: PHP_OUTPUT_HANDLER_CONT (2): flush 3
4: PHP_OUTPUT_HANDLER_END (4): end flush

clancy hood at gmail dot com (20-Sep-2009 01:31)

With ob callback: note that the second parameter sent to your method won't help you differentiate between flush calls and calls to ob_clean, but the buffer contents is sent in both cases, so you end up parsing data that isn't going to be used. Also, note that the constant PHP_OUTPUT_HANDLER_START is never actually sent, rather the integer "3" turns up on first flush:
<?php

   
function ob_handler($string, $flag){
        static
$input = array();
       
$done = false;
        switch(
$flag){
            case
PHP_OUTPUT_HANDLER_START:   
               
$flag_sent = "PHP_OUTPUT_HANDLER_START ($flag)";
                break; 
            case
PHP_OUTPUT_HANDLER_CONT:
               
$flag_sent = "PHP_OUTPUT_HANDLER_CONT ($flag)";
                break;
            case
PHP_OUTPUT_HANDLER_END:
               
$done = true;
               
$flag_sent = "PHP_OUTPUT_HANDLER_END ($flag)";
                break;
            default:
               
$flag_sent = "Flag is not a constant ($flag)";
        }
       
$input[] = "$flag_sent: $string<br />";
       
$output = "$string<br />";
        if(!
$done) return $output;
       
// print_r($input, 1) causes an error and var_export just doesn't work
       
$output .= '<br />';
        foreach(
$input as $k=>$v) $output .= "$k: $v";
        return
$output;
    }

   
ob_start('ob_handler');

    echo
'flush';
   
ob_flush();

    echo
'flush 2';
   
ob_flush();

    echo
'clean';
   
ob_clean();

    echo
'flush 3';
   
ob_flush();

    echo
'end flush';
   
ob_end_flush();
?>
?
flush
flush 2
flush 3
end flush

0: Flag is not a constant (3): flush
1: PHP_OUTPUT_HANDLER_CONT (2): flush 2
2: PHP_OUTPUT_HANDLER_CONT (2): clean
3: PHP_OUTPUT_HANDLER_CONT (2): flush 3
4: PHP_OUTPUT_HANDLER_END (4): end flush

I suppose the START flag problem *may* be a bug but I'm not able to upgrade before reporting since I must have the same version as my server (I'm on PHP 5.2.6). If anyone has 5.2.11 or other stable version feel free to test/report as you see fit.

mariusads at helpedia dot com (02-Jul-2009 11:02)

Make sure the editor you use does not add the UTF8/UTF16 BOM at the start of the scripts if you want to use ob_start("ob_gzhandler");

If those three characters are present, browsers like Firefox won't be able to decode the pages and will report:

Content Encoding Error

The page you are trying to view cannot be shown because it uses an invalid or unsupported form of compression.

The page you are trying to view cannot be shown because it uses an invalid or unsupported form of compression.

Google Chrome will simply report "Error 2 (net::ERR_FAILED): Unknown error."

With the ob_start command commented out, the page is successfully load and the browser will usually detect the BOM and not show it on the page, so everything's hard to debug.

Hudson (14-May-2009 10:16)

Hi,
Just wanted to share the problem I had with a fresh installation of Windows Server 2008 using IIS7 with PHP5.2.9-2 (windows MSI installer). Could also be applicable to earlier PHP builds on that platform. It was installed as "IIS fastCGI". I've not really used windows server until a recent project came up and came across a really annoying issue with output buffering. No alteration of php.ini was required.

As you know, OB can be useful for user feedback e.g. timer or % complete effect.

The problem:
Everything installed perfectly (using binary mods as well). Scripts all ready in test rig i.e. "default website" folder on IIS. They worked as normal except for output buffering - nothing worked. The browser wouldn't flush anything until script execution had ceased. By browser I refer to all common (IE, Firefox, Opera, Safari). Even installing latest build Apache, bizarrely, (in case it was an IIS issue) yielded the same problem.

Example:

<?php
ob_start
();  // Start output buffer (if not enabled in php.ini)
for($i=0; $i < 10; $i++)
{
    echo
$i, ' ';
   
ob_flush();
   
flush();
   
sleep(1);
}
?>

Expected: 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 appearing incrementally every second in the browser.

Result: 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 at once AFTER ten seconds.

I just couldn't get output buffering to work.

The Solution (requires admin rights):
Go to your server system drive e.g. c:\ and do the following...

C: > Windows > System32 > inetsrv > Config > applicationHost.Config

Go to the bottom of the page or look for your website e.g. <location path="Default Web Site"> Default could be whatever your server is.

ORIGINAL LINE: <add name="php" path="*.php" verb="*" modules="IsapiModule" scriptProcessor="C:\Program Files\PHP\php5isapi.dll" resourceType="Unspecified"  />

CHANGE TO: <add name="php" path="*.php" verb="*" modules="IsapiModule" scriptProcessor="C:\Program Files\PHP\php5isapi.dll" resourceType="Unspecified" responseBufferLimit="0" />

You MUST add responseBufferLimit="0"

By default, IIS buffers everything up to a massive 4MB (think it's to do with ASP) only flushing after the script has finished execution. You need to override that and it's not really documented, which is rather annoying.

Hooray... You will instantly find that output buffering should now functions as normal.

robin (17-Apr-2009 02:53)

When using brame buffer from time to time, you can see this warning
Warning: session_start() [function.session-start]: Cannot send session cache limiter - headers already sent
One of the many reasons may be in bad encoding of document. My experience, the document had header of utf-8 encoding but in fact it was win1252. After reconding to the utf-8 warning has disapeard.

ozana at omdesign dot cz (15-Jan-2009 04:54)

Simple cache system for HTML:

<?php
$cacheFile
= 'cache.html';

if ( (
file_exists($cacheFile)) && ((fileatime($cacheFile) + 600) > time()) )
{
   
$content = file_get_contents($cacheFile);
    echo
$content;
} else
{
   
ob_start();
   
// write content
   
echo '<h1>Hello world to cache</h1>';
   
$content = ob_get_contents();
   
ob_end_clean();
   
file_put_contents($cacheFile,$content);
    echo
$content;
}
?>

cj at ceejayoz dot com (29-Dec-2008 04:45)

Note that since PHP 5.1.x, all objects have their destructors called before the output buffer callback function executes. Thus, globalised objects will not be available as expected in the function.

This is stated to be intended behaviour, per http://bugs.php.net/bug.php?id=40104

codextasy at gmail dot com (19-Sep-2008 09:26)

There is a difference between the documentation and real callback functions invocation.
Manual says: "The function will be called when ob_end_flush() is called, or when the output buffer is flushed to the browser at the end of the request."

Actually, the callback function, once set by ob_start(), will be called regardless.
Here are the functions that invoke callback function immediately:
ob_clean
ob_end_clean
ob_end_flush
ob_flush
ob_get_clean

BUT only two of them return the result returned by the callback (ob_end_flush, ob_flush), other functions discard it.

At the end of the request, even if none of the functions listed above is called, the callback will be called anyway, and its result will be returned to the browser (well, at least this is corresponding to the manual).

There is one more trick:
If you set callback function with chunk_size > 1, callback function will be called each time output buffer is equal or exceeds chunk_size and its result will be output to the browser, even if you call any of ob_clean(), ob_end_clean(), ob_get_clean() later, so be aware of this fact.

fordiman at gmail dot com (27-Jun-2008 05:35)

Here's a nifty function I use daily.  Essentially: include a PHP file - but render its output to a variable, rather than to the buffer.  It's also set up to load the script with a variable set, and automagically loads globals into the script's namespace, making it an effective templating scheme.  It also has error handling, so that you're not flying blind when using output buffering.

<?php
$GLOBALS
['BufferedErrors']=Array();
function
errorParse($errno, $errstr, $errfile, $errline, $errcontext) {
   
$errorTypes = Array(
       
E_ERROR => 'Fatal Error',
       
E_WARNING => 'Warning',
       
E_PARSE => 'Parse Error',
       
E_NOTICE => 'Notice',
       
E_CORE_ERROR => 'Fatal Core Error',
       
E_CORE_WARNING => 'Core Warning',
       
E_COMPILE_ERROR => 'Compilation Error',
       
E_COMPILE_WARNING => 'Compilation Warning',
       
E_USER_ERROR => 'Triggered Error',
       
E_USER_WARNING => 'Triggered Warning',
       
E_USER_NOTICE => 'Triggered Notice',
       
E_STRICT => 'Deprecation Notice',
       
E_RECOVERABLE_ERROR => 'Catchable Fatal Error'
   
);
   
$ret=(object)Array(
       
'number'=>$errno,
       
'message'=>$errstr,
       
'file'=>$errfile,
       
'line'=>$errline,
       
'context'=>$errcontext,
       
'type'=>$errorTypes[$errno]
    );
   
$GLOBALS['BufferedErrors'][]=$ret;
    return
false;
}
function
parse($fileToInclude, $argumentsToFile=false) {
   
$bufferedErrorStack = $GLOBALS['BufferedErrors'];
   
set_error_handler('errorParse', error_reporting());
   
$GLOBALS['BufferedErrors']=Array();
   
    if (!
file_exists($fileToInclude))
        return
'';
    if (
$argumentsToFile === false)
       
$argumentsToFile = Array();
   
$argumentsToFile = array_merge($GLOBALS, $argumentsToFile);
    foreach (
$argumentsToFile as $variableName => $variableValue)
        $
$variableName = $variableValue;
   
ob_start();
    include(
$fileToInclude);
   
$ret = ob_get_contents();
   
ob_end_clean();
   
   
restore_error_handler();
   
$errors = $GLOBALS['BufferedErrors'];
   
$GLOBALS['BufferedErrors'] = $bufferedErrorStack;
    if (
count($errors)>0) {
       
$ret.='<ul class="error">';
        foreach (
$errors as $error)
           
$ret.=
               
'<li>'.
                   
'<b>'.$error->type.'</b>: '.
                   
$error->message.
                   
'<blockquote>'.
                       
'<i>file</i>: '.$error->file.'<br />'.
                       
'<i>line</i>: '.$error->line.
                   
'</blockquote>'.
               
'</li>';
       
$ret.='</ul>';
    }
    return
$ret;
}

petethebloke at gmail dot com (25-Apr-2008 01:45)

I was seriously stuck trying to get marketing emails into a site so that the designer could edit them in Dreamweaver while my app could dynamically write and send them to customers. Then I discovered that this works - I barely believed how clever PHP could be! Hope this helps someone else.
<?php
class marketing_email{
    function
marketing_email($customer_id){
       
//---stuff to get customer details from DB-------
        //----stuff---
        //I'm going to put 'newpart-' at the beginning and 'a' at the end of the separator string
        //I find that the 'a' is more reliable for a line end when it arrives at the client computer
       
$random_hash = md5(date('r', time()));
       
ob_start(); //Turn on output buffering
       
?>
--newpart-<?php echo $random_hash; ?>a
Content-Type: text/plain; charset="iso-8859-1"
Content-Transfer-Encoding: 7bit

Dear <?php echo $customer_name ?>
Your New special offer Blah Blah

--newpart-<?php echo $random_hash; ?>a
Content-Type: text/html; charset="iso-8859-1"
Content-Transfer-Encoding: 7bit

<head>
<style type="text/css">
<!--
p {
font-family: Verdana, Arial, Helvetica, sans-serif;
etc etc
}
-->
</style>
<title>Offer</title>
</head>
<body>
Blah Blah Blah
</body>

--newpart-<?php echo $random_hash; ?>a--
        <?php
        $pagecontent
=ob_get_contents();
       
ob_end_clean();
       
//define the headers we want passed. Note that they are separated with \r\n
       
$mail_headers = "MIME-Version: 1.0\r\nFrom: admin@domain.co.uk\r\nReply-To: admin@domain.co.uk";
       
//add boundary string and mime type specification
       
$mail_headers .= "\r\nContent-Type: multipart/alternative;\r\n\tboundary=\"newpart-".$random_hash."a\"";
       
$to=$customer_email;

        if(
mail($to,"Offer",$pagecontent,$mail_headers)){
            return
TRUE;
        }else{
            return
FALSE;
        }
    }
}
?>

Scott (16-Apr-2008 01:15)

Hello,

I wrote a pretty simple, yet very effective page caching script using output buffering.

Just one of the uses for such a collection of functions.

http://www.coderprofile.com/source-code/170/simple-page-caching

Kind regards,
Scott

coldshine at gmail dot com (07-Apr-2008 07:52)

Referring to dan at roteloftet dot com's comment:

RFC 2616 (HTTP) specifies a "transparent" Content-Encoding, "identity" (§ 3.5), that nicely suits what you tried to do with the (invalid) "None". So this equally working, and it's also RFC-compliant:

<?php
header
('Content-Encoding: identity', true);
?>

dan at roteloftet dot com (09-Mar-2008 10:19)

Some web hosting servers (mine do, at least) have in their php.ini the following setting:
output_handler = ob_gzhandler

This proved problematic for php-scripts which returns an image or a binary file in general, since there is no way to determine the content length of the compressed file.

Since I spent a lot of time scouring the net searching for a work-around (.htaccess-modifications were out of the picture for various reasons), I found this to work nicely to cancel out the ob_gzhandler specified in the php.ini:

<?php
while (ob_get_level())
       
ob_end_clean();
header("Content-Encoding: None", true);
?>

Put this at the top of the script before anything else is written to the page, and the script result will not be compressed.

dale3h (10-Jan-2008 03:06)

Here's a simple way, using register_shutdown_function, to easily maintain your global output filter:

File: outputfilter.inc.php
<?php
  initOutputFilter
();

  function
initOutputFilter() {
   
ob_start('cbOutputFilter');
   
register_shutdown_function('ob_end_flush');
  }

  function
cbOutputFilter($output) {
   
$search  = array(
     
'user@domain.com',
     
'user2@domain.com',
    );
   
$replace = array(
     
'<img src="/images/nospam001.jpg" border="0" alt="" />',
     
'<img src="/images/nospam002.jpg" border="0" alt="" />',
    );

    return
str_replace($search, $replace, $output);
  }
?>

Just include it in your global file, or in each file you want to filter.

This one in particular replaces any occurrence of "user@domain.com" or "user2@domain.com" with an image (that contains each email respectively), preventing spam bots from picking it up.

Eamon Straughn eamon at gizzle dot co dot uk (27-Dec-2007 07:06)

In regards to below. The best thing to do is create an error handler that catches all non-fatal errors prior and during your ob_start.

works for me all the time. and if there is an error you can format the errors in that script it executed in. From my experience i'm quite sure that would be the most logical choice. wouldn't it? Plus your using objects...Please do think before you can code.

Always plan else you'll be working blind. Anyways ob_start wouldn't exit;

Just plan ahead and anticipate. It's just like driving a car.. Happy new year.

Anonymous (22-Oct-2007 02:40)

As a follow up to my previous post :
(output seems to go to standard ouptput in command  line mode even with the use of output buffering (ob_start) )

Setting implicit_flush to false seems to do the trick :

   ini_set('implicit_flush',false); // (avoids output even with    ob_start, in command line mode)
   ob_start();
   include (realpath(dirname(__FILE__))."/".$template);
   $ret_str.=ob_get_contents();
   ob_end_clean();
   ini_set('implicit_flush',true);

(See :
Chapter 43. Using PHP from the command line
)

Francois Hill (22-Oct-2007 01:42)

Following clement dot ayme at st dot com 's remark :

In my experience it seems that the output IS buffered, but ALSO sent to the standard output !

Charlie Farrow (18-Oct-2007 05:49)

Under certain freak conditions, when an error ocours perfoming an action on an object that cannot be done (either because the object does not exist or the method does not exist) inside of an ob_start() the script will exit and print everything the current function generates before the error, but nothing else, including no error message.

I am at a loss to why no error message appears and am trying to get a working example for the developers that is simpler than my whole program!

So if you are using ob_start() and you get no output, check your objects.... you have made a mistake on them somewhere. The only trouble is you will not know where as there is no error!!

Asher Haig (ahaig at ridiculouspower dot com) (20-Aug-2007 09:17)

When a script ends, all buffered output is flushed (this is not a bug: http://bugs.php.net/bug.php?id=42334&thanks=4). What happens when the script throws an error (and thus ends) in the middle of an output buffer? The script spits out everything in the buffer before printing the error!

Here is the simplest solution I have been able to find. Put it at the beginning of the error handling function to clear all buffered data and print only the error:

$handlers = ob_list_handlers();
while ( ! empty($handlers) )    {
    ob_end_clean();
    $handlers = ob_list_handlers();
}

clement dot ayme at st dot com (27-Jul-2007 05:51)

ob_start() seems not compliant with command-line PHP.

just calling the function ahead of your unix-php script and nothing
happens. The STDOUT stream is not buffered.

tracey AT archive DOT org (29-Mar-2007 09:01)

Way to make all stdout and stderr write to a log
from *inside* a php script.
You simply need to make sure to call elog() every
once in awhile to get output.
It's a nice way to "daemonize" a script w.r.t. its logging.

// This allows us to capture all stdout and stderr (and error_log() calls)
// to this logfile...
// The "collected output" will be flushed anytime "elog()" is used...
ini_set("error_log", "/var/log/script.log");
ob_start();

function elog($str)
{
  // get anything written to stdout or stderr that did *NOT* use elog()
  // and write it now...
  $writeme = ob_get_contents();
  if ($writeme)
  {
    error_log($writeme);
    ob_end_clean();
    ob_start();
  }
  // now write message this method was called with
  error_log($str);
}

joebezucha at tlen dot pl (16-Mar-2007 10:05)

Hi, I use those functions for stripping unnecessary chars in my output code...because I have JavaScript placed in outpout code so I don't remove \n\r\t but just replace them with single space (it could cause errors in scripts)
Function stripBufferSkipTextareaTags skips tags Textarea. It's needed to don't loose \n\r when user edit some content...

sorry for my english ;)

<?php

function stripBufferSkipTextareaTags($buffer){
   
$poz_current = 0;
   
$poz_end = strlen($buffer)-1;
   
$result = "";
   
    while (
$poz_current < $poz_end){
       
$t_poz_start = stripos($buffer, "<textarea", $poz_current);
        if (
$t_poz_start === false){
           
$buffer_part_2strip = substr($buffer, $poz_current);
           
$temp = stripBuffer($buffer_part_2strip);
           
$result .= $temp;
           
$poz_current = $poz_end;
        }
        else{
           
$buffer_part_2strip = substr($buffer, $poz_current, $t_poz_start-$poz_current);
           
$temp = stripBuffer($buffer_part_2strip);
           
$result .= $temp;
           
$t_poz_end = stripos($buffer, "</textarea>", $t_poz_start);
           
$temp = substr($buffer, $t_poz_start, $t_poz_end-$t_poz_start);
           
$result .= $temp;
           
$poz_current = $t_poz_end;
        }
    }
    return
$result;
}

function
stripBuffer($buffer){
   
// change new lines and tabs to single spaces
   
$buffer = str_replace(array("\r\n", "\r", "\n", "\t"), ' ', $buffer);
   
// multispaces to single...
   
$buffer = ereg_replace(" {2,}", ' ',$buffer);
   
// remove single spaces between tags
   
$buffer = str_replace("> <", "><", $buffer);
   
// remove single spaces around &nbsp;
   
$buffer = str_replace(" &nbsp;", "&nbsp;", $buffer);
   
$buffer = str_replace("&nbsp; ", "&nbsp;", $buffer);
    return
$buffer;
}

ob_start("stripBufferSkipTextareaTags");

?>

denis at SPAM_WELCOME dot i39 dot ru (13-Mar-2007 02:55)

This function's behaviour  has been changed in php 5.2.0:

<?
global $AP;
$AP = new ap;
ob_start("ob_end");

function ob_end() {
        global $AP;
        $r = $AP->test();

        return $r;
}

class ap {
        function test() {
                return "debug";
        }
}
?>

In older versions it shows: "debug".
But latest php version causes error: PHP Fatal error: Call to a member function test() on a non-object.
And this is not a bug: http://bugs.php.net/bug.php?id=40104

Tobias Goldkamp (25-Dec-2006 11:02)

I use this to strip unnecessary characters from HTML output:

<?php

function sanitize_output($buffer)
{
   
$search = array(
       
'/\>[^\S ]+/s', //strip whitespaces after tags, except space
       
'/[^\S ]+\</s', //strip whitespaces before tags, except space
       
'/(\s)+/s'  // shorten multiple whitespace sequences
       
);
   
$replace = array(
       
'>',
       
'<',
       
'\\1'
       
);
 
$buffer = preg_replace($search, $replace, $buffer);
    return
$buffer;
}

ob_start("sanitize_output");

?>

lucky760 at yahoo dot com (21-Nov-2006 05:20)

Just for simplicity's sake (and because I had to rewrite it for use at http://www.VideoSift.com anyway), here's a very simplified, pre-PHP5 version. Just add one call to dump_css_cache() for each of your CSS files.

<?php

ob_start
('ob_gzhandler');

header('Content-Type: text/css; charset: UTF-8');
header('Cache-Control: must-revalidate');

$expire_offset = 0; // set to a reaonable interval, say 3600 (1 hr)
header('Expires: ' . gmdate('D, d M Y H:i:s', time() + $expire_offset) . ' GMT');

function
css_compress($buffer) {
 
$buffer = preg_replace('!/\*[^*]*\*+([^/][^*]*\*+)*/!', '', $buffer);// remove comments
 
$buffer = str_replace(array("\r\n", "\r", "\n", "\t", '  '), '', $buffer);// remove tabs, spaces, newlines, etc.
 
$buffer = str_replace('{ ', '{', $buffer);// remove unnecessary spaces.
 
$buffer = str_replace(' }', '}', $buffer);
 
$buffer = str_replace('; ', ';', $buffer);
 
$buffer = str_replace(', ', ',', $buffer);
 
$buffer = str_replace(' {', '{', $buffer);
 
$buffer = str_replace('} ', '}', $buffer);
 
$buffer = str_replace(': ', ':', $buffer);
 
$buffer = str_replace(' ,', ',', $buffer);
 
$buffer = str_replace(' ;', ';', $buffer);
  return
$buffer;
}

function
dump_css_cache($filename) {
 
$cwd = getcwd() . DIRECTORY_SEPARATOR;

 
$stat = stat($filename);
 
$current_cache = $cwd . '.' . $filename . '.' . $stat['size'] . '-' . $stat['mtime'] . '.cache';

 
// the cache exists - just dump it
 
if (is_file($current_cache)) {
    include(
$current_cache);
    return;
  }

 
// remove any old, lingering caches for this file
 
if ($dead_files = glob($cwd . '.' . $filename . '.*.cache', GLOB_NOESCAPE))
    foreach (
$dead_files as $dead_file)
     
unlink($dead_file);
 
  if (!
function_exists('file_put_contents')) {
    function
file_put_contents($filename, $contents) {
     
$handle = fopen($filename, 'w');
     
fwrite($handle, $contents);
     
fclose($handle);
    }
  }
 
 
$cache_contents = css_compress(file_get_contents($filename));
 
file_put_contents($current_cache, $cache_contents);
 
  echo
$cache_contents;
}

dump_css_cache('_general.css');

?>

lucky760 at yahoo dot com (21-Nov-2006 05:10)

In extension to the compress() function posted below, here's a nifty little class that improves the idea a bit. Basically, running that compress() function for all your CSS for every single page load is clearly far less than optimal, especially since the styles will change only infrequently at the very worst.

With this class you can simply specify an array of your CSS file names and call dump_style(). The contents of each file are saved in compress()'d form in a cache file that is only recreated when the corresponding source CSS changes.

It's intended for PHP5, but will work identically if you just un-OOP everything and possibly define file_put_contents.

Enjoy!

<?php

$CSS_FILES
= array(
 
'_general.css'
);

$css_cache = new CSSCache($CSS_FILES);
$css_cache->dump_style();

//
// class CSSCache
//

class CSSCache {
  private
$filenames = array();
  private
$cwd;
   
  public function
__construct($i_filename_arr) {
    if (!
is_array($i_filename_arr))
     
$i_filename_arr = array($i_filename_arr);
   
   
$this->filenames = $i_filename_arr;
   
$this->cwd = getcwd() . DIRECTORY_SEPARATOR;
   
    if (
$this->style_changed())
     
$expire = -72000;
    else
     
$expire = 3200;
   
   
header('Content-Type: text/css; charset: UTF-8');
   
header('Cache-Control: must-revalidate');
   
header('Expires: ' . gmdate('D, d M Y H:i:s', time() + $expire) . ' GMT');
  }
 
  public function
dump_style() {
   
ob_start('ob_gzhandler');
   
    foreach (
$this->filenames as $filename)
     
$this->dump_cache_contents($filename);
   
   
ob_end_flush();
  }
 
    private function
get_cache_name($filename, $wildcard = FALSE) {
   
$stat = stat($filename);
    return
$this->cwd . '.' . $filename . '.' .
      (
$wildcard ? '*' : ($stat['size'] . '-' . $stat['mtime'])) . '.cache';
  }
 
  private function
style_changed() {
    foreach (
$this->filenames as $filename)
      if (!
is_file($this->get_cache_name($filename)))
        return
TRUE;
    return
FALSE;
  }

  private function
compress($buffer) {
   
$buffer = preg_replace('!/\*[^*]*\*+([^/][^*]*\*+)*/!', '', $buffer);
   
$buffer = str_replace(array("\r\n", "\r", "\n", "\t", '  '), '', $buffer);
   
$buffer = str_replace('{ ', '{', $buffer);
   
$buffer = str_replace(' }', '}', $buffer);
   
$buffer = str_replace('; ', ';', $buffer);
   
$buffer = str_replace(', ', ',', $buffer);
   
$buffer = str_replace(' {', '{', $buffer);
   
$buffer = str_replace('} ', '}', $buffer);
   
$buffer = str_replace(': ', ':', $buffer);
   
$buffer = str_replace(' ,', ',', $buffer);
   
$buffer = str_replace(' ;', ';', $buffer);
    return
$buffer;
  }
 
  private function
dump_cache_contents($filename) {
   
$current_cache = $this->get_cache_name($filename);
   
   
// the cache exists - just dump it
   
if (is_file($current_cache)) {
      include(
$current_cache);
      return;
    }
   
   
// remove any old, lingering caches for this file
   
if ($dead_files = glob($this->get_cache_name($filename, TRUE), GLOB_NOESCAPE))
      foreach (
$dead_files as $dead_file)
       
unlink($dead_file);
   
   
$compressed = $this->compress(file_get_contents($filename));
   
file_put_contents($current_cache, $compressed);
   
    echo
$compressed;
  }
}

?>

(10-Nov-2006 06:34)

I'm sure some of you more brilliant minds could pare this down some more, but using the method found at fiftyfoureleven.com  for compressing, I got my 10090-byte stylesheet down to 3536 bytes and then again down to 2713 bytes, by stripping unecessary characters from the stylesheet. 2 ob_start calls and the CSS file is now 73% smaller. YMMV.

<?php
ob_start
("ob_gzhandler");
ob_start("compress");
header("Content-type: text/css; charset: UTF-8");
header("Cache-Control: must-revalidate");
$off = 0; # Set to a reaonable value later, say 3600 (1 hr);
$exp = "Expires: " . gmdate("D, d M Y H:i:s", time() + $off) . " GMT";
header($exp);

function
compress($buffer) {
   
$buffer = preg_replace('!/\*[^*]*\*+([^/][^*]*\*+)*/!', '', $buffer); // remove comments
   
$buffer = str_replace(array("\r\n", "\r", "\n", "\t", '  ', '    ', '    '), '', $buffer); // remove tabs, spaces, newlines, etc.
   
$buffer = str_replace('{ ', '{', $buffer); // remove unnecessary spaces.
   
$buffer = str_replace(' }', '}', $buffer);
   
$buffer = str_replace('; ', ';', $buffer);
   
$buffer = str_replace(', ', ',', $buffer);
   
$buffer = str_replace(' {', '{', $buffer);
   
$buffer = str_replace('} ', '}', $buffer);
   
$buffer = str_replace(': ', ':', $buffer);
   
$buffer = str_replace(' ,', ',', $buffer);
   
$buffer = str_replace(' ;', ';', $buffer);
    return
$buffer;
}

require_once(
'screen.css');
require_once(
'layout.css');
require_once(
'custom.php');
require_once(
'titles.css');
require_once(
'bus.css');

?>

feedback at realitymedias dot com (07-Nov-2006 11:35)

PHP Suggests: Some web servers (e.g. Apache) change the working directory of a script when calling the callback function. You can change it back by doing, for example, the following in the callback function: chdir(dirname($_SERVER['SCRIPT_FILENAME']))

The solution provided by PHP, does not function as intended when running PHP as a CGI (on CGI mode (CGI-BUILD (--enable-cgi) and/or CLI)).  In such a case, PHP is executed as a CGI-BIN and the web server daemon (e.g. Apache) sees SCRIPT_FILENAME as being the PHPCGI processor, and won't look deeper to find what file the PHPCGI processor is actually running/parsing; therefore the path returned by SCRIPT_FILENAME is wrong (most of the time, containing/ending with "cgi-system/php.cgi").

As SCRIPT_FILENAME is the safest way to proceed, but turns to be wrong in this exact situation; PATH_TRANSLATED is the next safe solution one would turn towards since it is populated with a different mechanism.

It would be correct to develop in the direction if the script filename itself is contained in the SCRIPT_FILENAME path value, then the SCRIPT_FILENAME content is reported correctly. If it is not, using PATH_TRANSLATED is the next logical choice we can use. The best reference in this case would be PHP_SELF as it is populated by PHP itself. Using SCRIPT_NAME as a reference would be an error as it is affected by the same problem (reports cgi-system and/or php.cgi as well).

The following is the revised code and should work on both the non-CGI and the CGI PHP processor types.

<?php
chdir
(dirname((strstr($_SERVER["SCRIPT_FILENAME"], $_SERVER["PHP_SELF"])
?
$_SERVER["SCRIPT_FILENAME"] : $_SERVER["PATH_TRANSLATED"])));
?>

Or the decomposed code as follows:

<?php
if (strstr($_SERVER["SCRIPT_FILENAME"], $_SERVER["PHP_SELF"])) {
$reference = $_SERVER["SCRIPT_FILENAME"];
} else {
   
$reference = $_SERVER["PATH_TRANSLATED"];
}
chdir(dirname($reference);
?>

This has been tested on Apache 1 & 2, PHP 4 & 5 and IIS 5.1

butch at enterpol dot pl (02-Oct-2006 08:04)

simple code to make phpsession $_GET nice for Valid XHTML 1.0 Transitional :)

function callback($buffer)
{
  $buffer = str_replace("&PHPSESSID", "&amp;PHPSESSID", $buffer);
  return $buffer;
}

ob_start("callback");

session_start();

net_navard at yahoo dot com (30-May-2006 03:09)

Hello firends

ob_start() opens a buffer in which all output is stored. So every time you do an echo, the output of that is added to the buffer. When the script finishes running, or you call ob_flush(), that stored output is sent to the browser (and gzipped first if you use ob_gzhandler, which means it downloads faster).

The most common reason to use ob_start is as a way to collect data that would otherwise be sent to the browser.

These are two usages of ob_start():

1-Well, you have more control over the output. Trivial example: say you want to show the user an error message, but the script has already sent some HTML to the browser. It'll look ugly, with a half-rendered page and then an error message. Using the output buffering functions, you can simply delete the buffer and sebuffer and send only the error message, which means it looks all nice and neat buffer and send
2-The reason output buffering was invented was to create a seamless transfer, from: php engine -> apache -> operating system -> web user

If you make sure each of those use the same buffer size, the system will use less writes, use less system resources and be able to handle more traffic.

With Regards, Hossein

(capitals) THE maYoR ANd thOSe@gmail (31-Jan-2006 10:14)

One of the notes below mentions that ob_end_flush() is called automatically at the end of the script if you called ob_start without an ob_end.

Because I couldn't find any other way to do it, I tried to use this fact to have some stuff run at the end of every script. It was a maintenance nightmare, so I'm putting a link here to the good way to do it, since it's nigh impossible to find with google.

http://php.net/register_shutdown_function

simon (24-Jan-2006 04:51)

Found that variables in class instances we're not being set after the call to ob_start().
Call ob_start after the variables are set however and it works but that didn't seem to solve the goal of a self contained templating class.
The fix was to assign the class by reference with '&new'
Here is a simplified working example:
<?php
class Buffer {
var
$template = ' - template set in class constructor';
function
Buffer() {
   
$this->startBuffer();
}
function
startBuffer() {
   
ob_start(array(&$this, 'doFlush'));
}
function
doFlush($buffer) {
   
/* simple string concat to show use of a
    template string and the buffer output */
   
return $buffer . $this->template;
}
}
/* template does not get set:
$buffer1 = new Buffer();
$buffer1->template = ' - template set in instance';
echo 'some buffer content';
*/
/* this works as expected */
$buffer2 = &new Buffer();
$buffer2->template = ' - template set in instance';
echo
'some buffer content';

ernest at vogelsinger dot at (08-Jan-2006 05:57)

When you rely on URL rewriting to pass the PHP session ID you should be careful with ob_get_contents(), as this might disable URL rewriting completely.

Example:
ob_start();
session_start();
echo '<a href=".">self link</a>';
$data = ob_get_contents();
ob_end_clean();
echo $data;

In the example above, URL rewriting will never occur. In fact, rewriting would occur if you ended the buffering envelope using ob_end_flush(). It seems to me that rewriting occurs in the very same buffering envelope where the session gets started, not at the final output stage.

If you need a scenario like the one above, using an "inner envelope" will help:

ob_start();
ob_start();   // add the inner buffering envelope
session_start();
echo '<a href=".">self link</a>';
ob_end_flush(); // closing the inner envelope will activate URL rewriting
$data = ob_get_contents();
ob_end_clean();
echo $data;

In case you're interested or believe like me that this is rather a design flaw instead of a feature, please visit bug #35933 (http://bugs.php.net/bug.php?id=35933) and comment on it.

cyrille.berliat[no spam]free.fr (16-Oct-2005 11:07)

If you're trying to use ob_start() in some PHP5 classes (probably works on PHP4 classes), this is the good way :

<?

class HTMLPage
{
//----------------------------------------------------------------- PUBLIC

//----------------------------------------------------- Mthodes publiques
   
    public static function ConvertIntoSGML( $source )
    // Mode d'emploi :
    //convertit une string en une SGML valide
    //
    // Renvoie :
    //la chaine traite
    //
    // Algorithme :
    //analyse char par char de la chaine. Si un caractre est de nombre ASCII > 127,
    //conversion en son code SGML.
    {
        $newString = '';
       
        for( $i = 0; $i < strlen( $source ) ; $i++ ) {
            $o = ord( $source{ $i } );
           
            $newString .= ( ( $o > 127 ) ? '&#'.$o.';': $source{ $i } );
        }
       
        return $newString;
    }
   
    public function FlushSite( $source )
    {
        return $this->ConvertIntoSGML ( $source );
    }
   
//-------------------------------------------- Constructeurs - destructeur
    function __construct()
    // Mode d'emploi (constructeur) :
    //initialise la buffurisation
    //
    // Contrat :
    //
    {
        ob_start( array (  & $this , 'FlushSite' ) );
    } //---- Fin du constructeur
   
//------------------------------------------------------ Mthodes Magiques

//------------------------------------------------------------------ PRIVE

}

// Example :

$webdesign = new HTMLPage ( );

echo 'Hllo world'; // Will produce the source 'h&#233;llo world'

?>

Without the & before $this, you'll loose your content because ob_start() will call the flushsite() function from a clone of the object and not the caller object himself.

Note : call_back function must be public because ob_start() is in an extern scope from your class :)

I hope this will help you!

php at bucksvsbytes dot com (16-Sep-2005 06:29)

The following should be added: "If outbut buffering is still active when the script ends, PHP outputs it automatically. In effect, every script ends with ob_end_flush()."

admin at bobfrank dot org (28-Aug-2005 04:50)

If you want to run code in the middle of a string that you made, but you want to wait the printing...
(so if you want to allow php in bb-code style, and you want to execute it in order, and print everything in order...)

phpRun($code) {
    ob_start();
    exec($code);
    $output = ob_get_contents();
    ob_end_clean();
    return $output;
}

$str = str_replace("]\n", "]", $str);
$match = array('#\[php\](.*?)\[\/php\]#se');
$replace = array( phpRun( stripslashes('$1') ) );
$str= preg_replace($match, $replace, $str);

echo $str;

kasper at johansen dot tdcadsl dot dk (28-Aug-2005 03:29)

Another way to make your code harder to copy, is to remove all line-breaks and tabs you have in it.

You can use this function to remove those.

You can choose to place "ob_start();" at the start of your main PHP-file, and "ob_end_clean();" at the end of it.

This is not the best solution though. If you are running Apache you might want to do something like this to a ".htaccess"-file:

php_value auto_prepend_file "/home/www/load_before.php"
php_value auto_append_file "/home/www/load_after.php"

(you should replace those files with files that actually exists).

In the "load_before.php" you can place the "ob_start();", and in the "load_after.php" you can do something like this:

<?
    $html = strtr(ob_get_contents(), array("\t" => "", "\n" => "", "\r" => ""));
    ob_end_clean();
   
    echo $html;
?>

This will clean your HTML of all kind of linebreaks (both \n and \r) and tabulators (\t). This will save your users some bandwidth, but it will also make your HTML, JavaScripts and more very difficult to read or copy. That way, making it harder for people to steal your code if you do not want them to.

This isnt in the spirit of OpenSource, but anyway you should be aware that this is possible.

Be aware that if you use PHP-files, to simulate pictures, all linebreaks will also be remove, making them corrupt. A solution to this, could be to check the headers (Content-Type), and if isnt set, or it is "text/html", you can go ahead and remove the linebreaks and tabs.

At my site, with more than 50.000 pictures and about 60 people online, I couldnt see any difference in the loadtime.

But still be aware, that your output will not be sent, before the script is finished, which will make your page slower to load that way also, since it cannot send any output while loading, but will have to wait until the load is finished.

geoffrey at nevra dot net (09-Aug-2005 08:05)

When using a callback with ob_start(), functions like ob_get_contents() don't make use of it, use ob_end_flush() instead.

nb: not tested with every ob_* functions, just ob_get_contents() and ob_end_flush()

(05-Aug-2005 03:54)

I usually create my pages in four parts - variable initialisation, import header (using the variables just declared to configure), main body (mostly non-PHP), import footer.  I wondered about making the main body examinable by another PHP script if the main page was included into it.  I found I could control output of the main body by ending the header with an unclosed function which finishes at the start of the footer, thus enclosing the main body.  Output buffering can then be used to read this into a variable.  As a demonstration of how this can be used to control the order of output look at this example:

<?php
$output
= "";

// Callback to process buffered output
function capture($buffer)
    {
   
$GLOBALS['output'] .= $buffer;
    return
"C ";
    }

// Calls the printE() function with output capture
function captureE()
    {
   
ob_start("capture");
   
printE();
   
ob_end_flush();
    }
?>

A
<?php
// Output 'E' (the main body in the example scenario)
function printE()
    {
// (End header after this line) ?>
E
    <?php // (Start footer with this line)
   
}
?>
B
<?php captureE(); ?>
D
<?php print $output; ?>
F
<?php printE(); ?>
G

The output is A B C D E F E G.

For the application I mentioned above there are two points to note:
 - The page when executed alone must output its main body but the inspection script should suppress this, perhaps by means of a variable set before the page is included and then checked for in the footer output lines.
 - Because the main body is now inside a function it has a different namespace, thus changes may be required to prevent code breaking (e.g. use of globals, handling of functions defined within the main body).

oersoep at gmail dot com (08-Jul-2005 01:46)

These are handy. First one has been mentioned before.

ob_start( array( 'lib_class', 'parse_output' ) );
ob_start( array( $this, 'parse_output' ) );

Note: $this is NOT a reference. Anything the callback saves or logs disappears in the clone ob_start works with.
It does enable the callback to work with the attributes of $this, like $this->ar_tpl_value or whatever your style is.

The manual says:
"If the optional parameter chunk_size is passed, the callback function is called on every first newline after chunk_size bytes of output. The output_callback parameter may be bypassed by passing a NULL value."
This doesn't work with my 4.3.11. Might be the Zend optimizer though. Daren't turn it off to go see.

Aleksey (25-May-2005 08:08)

This function dynamically changes title of HTML page:

  function change_title($new_title) {
    $output = ob_get_contents();
    ob_end_clean();

    $output = preg_replace("/<title>(.*?)<\/title>/", "<title>$new_title</title>", $output);
    echo $output;
  }

Example:
  ob_start();
  // ... some output
  change_title('NEW TITLE!');

jds1509 at NOSPAMrit dot edu (24-May-2005 03:06)

This code demonstrates the affect of providing a value to the chunk_size parameter. A value of 1 or 0 will be ignored by php. Here's the Code:

<?

function callback($buffer)
{
    return "TRAPPED:".$buffer."<br/>";
}

ob_start("callback",2);

echo "long string, so callback";
echo "X";
echo " - no callback, less than 2 chars";

?>newlines,
but
no
callback
!<?

// PHP block

?>PHP block initiates callback.<?

echo "One more callback at EOF...";

?>

The code above outputs:

TRAPPED:long string, so callback
TRAPPED:X - no callback, less than 2 chars
TRAPPED:newlines, but no callback !
TRAPPED:PHP block initiates callback.
TRAPPED:One more callback at EOF...
TRAPPED:

rafa dot chacon at factorydea dot com (10-May-2005 10:10)

If you're trying to include a php file inside a loop by require_once (in example, a dinamic email template) and change the value of some variables (in example, url to unsuscribe, different for each user), you should use

<?php

// ... some code

$usermail = array("email1", "email2", ...);

for(
$i = 0; $i < $MAX; $i++)
{
       
$usermail_unsuscribe = $usermail[$i];
       
ob_start();
        include(
"email_template.php");
       
ob_clean();
}
?>

Otherwise $usermail_unsuscribe will get only "email1" value.

JM (08-May-2005 08:17)

I don't claim to understand this--I would have expected the exact opposite--but it seems that
  ob_start() ... ob_end_flush()
can massively improve perfomance, by at least a factor of 10 (admittedly a small number of samples).

I tried this after discovering that I could move a large (100ms) bottleneck in one of my scripts into
   echo "<!-- about 40 characters of junk -->";
which clearly shouldn't have taken long to run.

My unfounded theory is that without buffering, the interaction between PHP4.3.4 and Apache is not optimized, whereas with buffering, PHP delivers the entire page at once, which Apache handles better.

I should add that this is under https.

Ray Paseur (Paseur ... ImagineDB.com) (01-Mar-2005 09:50)

You can use PHP to generate a static HTML page.  Useful if you have a complex script that, for performance reasons, you do not want site visitors to run repeatedly on demand.  A "cron" job can execute the PHP script to create the HTML page.  For example:

<?php // CREATE index.html
  
ob_start();
/* PERFORM COMLEX QUERY, ECHO RESULTS, ETC. */
  
$page = ob_get_contents();
  
ob_end_clean();
  
$cwd = getcwd();
  
$file = "$cwd" .'/'. "index.html";
   @
chmod($file,0755);
  
$fw = fopen($file, "w");
  
fputs($fw,$page, strlen($page));
  
fclose($fw);
   die();
?>

eddie (13-Feb-2005 04:09)

I use this function for deleting not needed characters within the html code before sending the whole stuff to the browser.

function callback($buffer){
    $buffer = str_replace("\n", "", $buffer);
    $buffer = str_replace("\t", "", $buffer);
    $buffer = str_replace(chr(13), "", $buffer);
    $buffer = ereg_replace("<!\-\- [\/\ a-zA-Z]* \-\->", "", $buffer);
    return $buffer;
}

First str_replace will delete any newlines, second any tabs and the third any carriage return. Finally the regular expression will delete any html-comment which consists of /, space, a-z or A-Z.
Using this saves about 1kb on every pageload.

FB (02-Feb-2005 02:59)

I've noticed a bug with MSIE for non cached contents if your page is less than 4096 octets : you have to refresh the page each time to view its content !

Here is the solution to prevent this stupid behaviour of MSIE : just insert this code at the top of your scripts :

function ob_callback($buffer)
{
    return $buffer . str_repeat(' ', max(0, 4097 - strlen($buffer)));
}

ob_start('ob_callback');

aaron at offtone.com (14-Nov-2004 01:19)

My callback is stored in a function class, and using ob_start ('Class::callback') wasn't working. Not wanting to instantiate the class (no need, it's a function class) I tried this and it worked a charm:

ob_start (array (Class, 'callback'));

PHP 4.3.4

dev at kiwicore dot org (28-Oct-2004 09:49)

I wanted to do things a very particular way with output buffering and shutdown functions; using register_shutdown_function instead of the built in callback feature of this function. However, one should note that this won't work, because the contents of the buffer are no longer in scope when PHP is calling the shutdown functions. This would have been easy to see EXCEPT that PHP graciously flushes any unsent buffers at the end of the script, or when calling exit. So:

<?php
    ob_start
();
    echo
'hi';
    exit;
?>

Prints "hi". In a nutshell, if you want it to have a shutdown function that handles an output buffer, just specify it in ob_start() and let PHP automatically call it at the end of the script.

jkloss at hotmail dot com (17-Mar-2004 03:20)

If ob_start does not seem to be working for you, note that with Apache 2 the flush() function causes PHP to send headers regardless of whether ob_start had been called before flush.

ob_start();
echo 'test';
flush();

will cause Apache 2 to send whatever headers may be stacked up - which means you can't use a header(location:xxx) after the flush.  To fix, remove the flush().  Spent several hours discovering this.  Apache 1.x didn't work this way.

mjr (10-Mar-2004 05:10)

If you're using object-orientated code in PHP you may, like me, want to use a call-back function that is inside an object (i.e. a class function). In this case you send ob_start a two-element array as its single argument. The first element is the name of the object (without the $ at the start), and the second is the function to call. So to use a function 'indent' in an object called '$template' you would use <?php ob_start(array('template', 'indent')); ?>.

zeisss at web dot de (23-Nov-2003 05:03)

Note that the current working directory changes in the callback procedure (Changed from htdocs\ to windows\system32\ on my system).

You have to use absolut paths if you want to open files on your local system.

ed.oohay (a) suamhcs_rodnan (22-Nov-2003 01:18)

Output Buffering even works in nested scopes or might be applied in recursive structures... thought this might save someone a little time guessing and testing :)

<pre><?php
   
    ob_start
();              // start output buffer 1
   
echo "a";                // fill ob1
       
       
ob_start();              // start output buffer 2
       
echo "b";                // fill ob2
       
$s1 = ob_get_contents(); // read ob2 ("b")
       
ob_end_flush();          // flush ob2 to ob1
       
   
echo "c";                // continue filling ob1
   
$s2 = ob_get_contents(); // read ob1 ("a" . "b" . "c")
   
ob_end_flush();          // flush ob1 to browser
   
    // echoes "b" followed by "abc", as supposed to:
   
echo "<HR>$s1<HR>$s2<HR>";
   
?></pre>

... at least works on Apache 1.3.28

Nandor =)

venky_athome at yahoo dot com (17-Jan-2003 05:58)

IE 55. sp2 and IE6 as on the date of adding this note have problems with content type gzip and caching http headers. The pages are never cached. I think this combination of http headers can also crash the browser.

see http://support.microsoft.com/default.aspx?scid=kb;en-us;321722

(05-Dec-2002 08:02)

If you're using Apache (1.3x or 2.0), you might consider adding automatic compression capability to your delivered pages.

I assume you all know how to build compression classes and use them in your programs, but none has yet to offer the speed and robustness of a binary-compiled module. Furthermore, such modules also log the "compressable" hit in the web log file, thus allowing your favorite web anaysing program to show you reports of bandwidth saved.

Having said that, you might consider the following two modules for Apache:

1) Apache 1.3x: use mod_gzip, available from:
http://sourceforge.net/projects/mod-gzip/

2) Apache 2.x: use mod_gz, see here:
http://www.mail-archive.com/dev@httpd.apache.org/msg00734.html

3) Apache 1.3x: you may also want to use mod_defalte, from:
ftp://ftp.lexa.ru/pub/apache-rus/contrib/

Hope it helps.

cliff at NOSPAMtravelguides dot com (07-Aug-2002 11:36)

good article on output buffering on devshed:

http://www.devshed.com/c/a/PHP/Output-Buffering-With-PHP