Posted by Al on 02/09/05 20:28

Many thanks to Richard and Skippy.

Skippy: Your suggestion to set non-blocking mode (stream_set_blocking) was the
magic key to success.

Here is the code I ended up with, I left some test stuff included so anyone who
may need it has a good start.

> $fp= fopen($file, 'rb' );
> if (!$fp) {die ("Failed to open $file, errors: $errstr ($errno)<br>\n");
> }//end if
>
> stream_set_blocking($fp, FALSE );
> $html= ''; $chunk= '';
> $start= time();
> while (!feof($fp)){
> sleep($interval);
> $chunk= fread($fp, 10000); //chunks, packets are limited anyhow
> $length= strlen($chunk);
> if($length)== 0){echo "<div style=\"color:blue\">Read zero bytes</div>";}
> $html .= $chunk;
> echo "<div>At " . date("h:s") . ">> $length bytes read</div>";
> ob_flush();
> if (time() - $start > $timelimit){echo("<div>style= \"color:red\">Timeout!</div>\n"); break;}
> }//end while

Putting the sleep() first thing helps everything get started better. Seems as if
fread() needs a little time after fopen().

ob_flush() is neat. It forces the client browser to render progress. I've
always been told you can't make a progress bar or list with php. Well here it
is, at least it works with IE6 and Mozilla.


Skippy wrote:
> Quoting Al <news@ridersite.org>:
>
>>Darn, I left out an important function, the fread(). Code snip should be:
>> $fp= fopen("http://www.anything.com/foo.html, 'rb');
>> if(!fp) {do something different}
>> stream_set_timeout($fp, 2);
>> $contents= fread($fp, 200000);
>> $status= stream_get_meta_data($fp);
>> if($status[timed_out] {do something};
>>
>>It appears to me there is a basic logic problem. The script must get
>>past the fread() function before it gets to the stream_get_meta_data($fp).
>>But, it hangs up on fread() and the script times out.
>
>
> Set the stream to non-blocking mode (stream_set_blocking). Now all
> read functions will return instantly, regardless if any data came
> through. You can test whether you got back anything by checking the
> returned data. If you now read from the non-blocked stream in a cycle,
> you will have no "dead" times and you will always retain control.
> Only now can you rely on comparing time() with a control value set
> before entering the cycle and decide to abort if a number of seconds
> has passed.
>
> stream_set_timeout() will be pretty much superfluous at this point
> and may even hinder your reading from the stream.
>
> Important tip: in a non-blocking cycle you MUST have an usleep()
> somewhere. Otherwise that cycle will consume 100% CPU while it
> runs. An usleep(1000) will suffice, adjust as needed.
>

[Back to original message]