I've read similar bug report from several of our users (we have free hosting server with aprox. 60,000 users, currently running latest RC of php-4.3.0 with apache-2.0.43), but I've not been able to reproduce it. Can you please paste here part of your code ?

I'm running Apache 1.3.27.  I haven't tested it under *nix.  As far as a code snippet, I will assume that it is in this code, somewhere:

if ($admin_from_web==true) {
	$row = 0;
	$insert_msg = '';
	$fp = fopen("$path_to_gb",'r') or die("$unable_to_access_file_msg $path_to_gb");
	set_file_buffer($fp,$buffersize);
	flock($fp,1);
	$num = 16;
	for ($i=0;$i<$num;$i++) {
		$j = $i+1;
		$head_array[$i] = ${"col$j"};
	}
	while ($data = fgetcsv($fp,$csv_buffer_size,",")) {
		$row++;
		if ($row<2) continue;
		$new_msg = "";
		if ($data[0]==$msgid) {
			if (!empty($update)) {
				reset ($HTTP_POST_VARS);
				while (list ($key, $val) = each ($HTTP_POST_VARS)) {			
					$found=false;
					for ($i=0;$i<count($head_array);$i++) {
						if ($key==$head_array[$i]) {
							$rec[$i] = $val;
							$found=true;
						}
						if ($found) break;
					}
				}
				$msg = $label_name."$data[4] (ip = ".getenv(REMOTE_ADDR).") - $head_array[0]:$msgid ".$on." ".date($date_format)." $at ".date($time_format)."\n";
				for ($c=0;$c<$num;$c++) {
					if ($c==0) $comma=""; 
					else $comma=",";
					if ($c==13&&$row>1) {
						$rec[$c] = htmlspecialchars($rec[$c],ENT_QUOTES);
						$rec[$c] = str_replace("&amp;#","&#",$rec[$c]);						
						$new_msg .= $comma."\"$rec[$c]\"";						
					}
					elseif ($c<4) $new_msg .= $comma."$rec[$c]";
					else $new_msg .= $comma."\"$rec[$c]\"";
					if (chop($data[$c])!=chop($rec[$c])) {
							$msg .= "$head_array[$c]: [$data[$c]]\n\n$head_array[$c]: [$rec[$c]]\n\n\n";
					}
				}
				$new_msg .= "\n";
				if ($admin_email_advice_on_gb_edit && !empty($admin_email_address)) {
					$to = "$admin_email_address,$admin_email_address_addl";
					$subject = "$admin_email_subject_on_gb_edit";
					@mail($to, $subject, "$msg","From: $admin_email_address") ;
				}
			}
		}
		else {
			for ($c=0;$c<$num;$c++) {
				$val = $data[$c];
				if ($c==0) $comma=""; 
				else $comma=",";
				if ($c==13&&$row>1) $new_msg .= $comma."\"$val\"";
				elseif ($c<4&&$row>1) $new_msg .= $comma."$val";
				else $new_msg .= $comma."\"$val\"";
			}
			$new_msg .= "\n";						
		}
		if (!empty($new_msg)) $newgb[] = $new_msg;
	}
	flock($fp,3);
	fclose($fp);
	unset($delete);
	unset($update);
	$buffersize = round(filesize($path_to_gb)*1.5);	
	$fp = fopen("$path_to_gb",'w') or die("$unable_to_access_file_msg $path_to_gb");
	set_file_buffer($fp,$buffersize);
	flock($fp,1);	
	for ($i=0;$i<count($newgb);$i++) {
		if ($i==0) $prefx = "<? Header(\"HTTP/1.0 403 Forbidden\");exit; ?>\n";
		else $prefx = "";
		fwrite($fp,$prefx.$newgb[$i]);
	}
	flock($fp,3);
	fclose($fp);
}

Please try and cut down the script to a complete self-contained example that I can run here.
I'm not going to spend hours debugging a long, incomplete code fragment.
What I'm looking for it something like this:

$fp = fopen($filename, "w");
$len = fwrite($fp, $data);
fclose($fp);
echo "Wrote $len bytes";

If that works, try adding in the flock calls.
If that works, try adding in those set_file_buffer calls
(NB: there does not seem to be a need to play with file
buffering in this case).

Please help me to help you.

Sorry for the misunderstanding.  First of all I did not know that you were a developer working the problem. Nothing in your comment told me that.  Secondly, I have already stated that I will try to produce a smaller script (see my original post).  I just haven't had time to work the problem.  As to the buffering, without it, the full text is not written out.  It gets truncated.

Keep this at feedback.
(and note that I am not mira-junk@cekit.cz and that my previous comment on this was my first).

It appears to be the flock() call.  This fails.  However, if you comment out the flock() calls it works.

$filename = "c:/phptest.txt";
$data = "This is test text";
$fp = fopen("$filename",'w');
flock($fp,1);
$len = fwrite($fp,$data);
flock($fp,3);
fclose($fp);
echo "Wrote $len bytes";

Small note, the 2nd flock (flock($fp,3);) is entirely optional. The problem is occures if prior to write the LOCK_SH is set on the file, the problem does not happen if LOCK_EX is used.

Just a further note.  It also/still errors on RC4.  Also, if you use flock($fp,2) instead of flock($fp,1) it works.

Update summary.
LOCK_EX works fine.
Additionally, writing to the file prior to obtaining the lock also works fine, but defeats the purpose of the lock.

Your script is broken :)
It's taken a few hours for us to realize this and it's pretty easy to overlook.

You are locking the file for shared access and then writing to it.  This doesn't make sense, because you should really be using an exclusive lock (LOCK_EX) to do this.

Under unix systems, flock is an advisory lock, meaning that processes can ignore the lock status if they wish.

Under windows, the locking is enforced by the OS which forbids the write access, hence the zero-size file.

Additionally, the win32 docs state that closing a file that was locked (but not unlocked) leads to undefined behaviour, so that flock($fp, LOCK_UN) call *is* required for win32.

I am so sorry, but why does it [appear to] work under previous releases and only breaks under 4.3.0RC's?  have you tightened up the code?

I can't explain the sudden change in behaviour either.
The entire filesystem IO layer was rewritten since 4.2,
but AFAICT, the same functions are called in 4.2 as are
being called in 4.3, in the same order.
The only difference is that 4.3 prefers to use fd's rather
than FILE*'s, but in my tests here, using the FILE* gave
the same error.

*shrug*