Description:
------------
See bug #22072.

Since the above is now closed, I am creating this entry. Please accept my apologies if it would have been better added to 22072 instead.

I wasn't aware of this potential problem until I saw it in the change log for 4.3.3, so I thought I'd test it. Something is still not right. I'm using PHP 4.3.3 with Apache 2.0.47.

My first test was to see whether ignore_user_abort() works as expected. I used the very simple:

  ignore_user_abort ( false );
  while ( 1 );

This makes it easy to spot in the process list using 'top'. I tested both using PHP as a dynamically loaded Apache module and the CGI binary. The latter had safe mode enabled, the former didn't; I don't know whether this can/should make any difference. In both cases the script continued to execute after I clicked on the browser's stop button and until its PHP enforced time limit was reached.

I then went on to test ( using only the Apache module ) using the code suggested in bug report 22072:

  http://www.mpfreescene.com/test/?option=source

As rather expected now, this continued to execute after I'd used the stop button.

I went on to test by modifying the code as follows:

  - I set set_time_limit(15); so it would eventually be
      aborted by this limit.
  - I made $m very large.
  - Have the result of connection_status() written to a
      file each time round the loop.
  - I removed the sleep() call.

I ran the script and used the stop button while it was running. It continued to execute until it had used 15 seconds of CPU time. Log file shows: httpd: PHP Fatal error:  Maximum execution time of 15 seconds exceeded in...
The final ( large ) output file contained all zeros and a final 2.

So, it seems that PHP correctly detects and reports the case of a timeout, but not an abort.

I noticed one other possibly odd thing while doing this. If PHP has output buffering enabled, the call to flush() results in the script being immediately shut down ( execution doesn't get to the line following flush() ). The result of connection_status() when called in exitfp() is then 1. If the flush() call is removed, the results as previously described are the same whether or not output buffering is enabled.

I hope someone can make sense of all this! Do let me know if you would like any further information.