Description: ------------ An E_NOTICE error, eg for an uninitialised variable, in a SOAP server causes the Apache process to either run at 100% until max_execution_time is reached, or to die with a segmentation fault. The server function completes and the Apache access log records a 200 result, but the client sees either no response when the process has run to the full time, or a bad HTTP response after a segfault. The problem is seen regardless of the error_reporting() setting but a local error handler set and restored around the offending code will allow the expected result. Reproduce code: --------------- Minimal client (client1.php) and server (crashy.php) in http://falmouth.roundpoint.co.uk/~ajf/example.tar.gz reproduces the running till max_execution_time reached case. Expected result: ---------------- Either the string "Hello stuff" or a soap_fault object regarding the uninitialised variable depending on the level set by error_reporting(). Actual result: -------------- Either a soap_fault object for a timed out read from the server or for a bad HTTP response.

the error handler should ignore '0' errors, eg. hidden by @ - see the set_error_handler manual page.. $soap_server_fault = null; function SOAP_ServerErrorHandler($errno, $errmsg, $filename, $linenum, $vars) { if (!$errno) { return; } global $soap_server_fault; $detail = "Errno: $errno\nFilename: $filename\nLineno: $linenum\n"; // XXX very strange behaviour with error handling if we =& here. $soap_server_fault = new SOAP_Fault($errmsg, 'Server', 'PHP', $detail); }

To build on Alan's fix, this would also skip E_NOTICE messages. Not sure the PEAR way (or likely even the best way) to actually do this, but the following at least works. function SOAP_ServerErrorHandler($errno, $errmsg, $filename, $linenum, $vars) { if (! $errno || $errno == E_NOTICE) { return; } global $soap_server_fault; $detail = "Errno: $errno\nFilename: $filename\nLineno: $linenum\n"; // XXX very strange behaviour with error handling if we =& here. $soap_server_fault = new SOAP_Fault($errmsg, 'Server', 'PHP', $detail); }

The segfault or max. exec. time-out occurs due to Pear::SOAP trying to serialize a recursive array or object, which causes the crash (I will file this as a separate bug). This recursive array is created due to a backtrace array, which is included in the SOAP_Fault object that gets created due to an E_NOTICE error. (see line 69 in Fault.php in the 0.8RC3 release (function SOAP_Fault->message()) Appearantly the $this->backtrace variable is assumed to be a string, but isn't really. An easy (dirty) fix would be to use print_r to change the backtrace array into a string (print_r catches recursion in arrays or objects), like this: (line 69 would change from:) $params[] =& new SOAP_Value('detail', 'string', $this->backtrace); (into:) $params[] =& new SOAP_Value('detail', 'string', print_r($this->backtrace, true)); A clean (and much tougher) fix would be to make sure recursive arrays or objects are caught in the serialize routines and properly dealt with (probably by throwing a SOAP_Fault - I don't know if the SOAP spec provides in this).

Can you have a go with CVS version, I committed both patches

Thank you for your bug report. This issue has been fixed in the latest released version of the package, which you can download at http://pear.php.net/get/SOAP