[ http://nagoya.apache.org/jira/browse/XERCESC-477?page=history ]
Alberto Massari updated XERCESC-477:
------------------------------------
Priority: Major
> Memory Leak in SAX Parser
> -------------------------
>
> Key: XERCESC-477
> URL: http://nagoya.apache.org/jira/browse/XERCESC-477
> Project: Xerces-C++
> Type: Bug
> Components: SAX/SAX2
> Versions: 1.6.0
> Environment: Operating System: Solaris
> Platform: Sun
> Reporter: Thomas
> Assignee: Xerces-C Developers Mailing List
>
> I am experiencing a large memory leak using the Xerces 1.6 SAX Parser on a
> Solaris 8.0 machine. The O.S. is installed using 32 bit mode. The code is
> single threaded, built with -DAPP_NO_THREADS directive.
>
> Schema and Namespaces are set to false. Validation is set to true. The
> MemBufInputSource class is the source for the xml document.
> The application is initialized using the XMLPlatform::initialize() method and a
> new Parser is created for each XML buffer that gets parsed. After each use,
> the XMLPlatform::terminate() method is called to force cleanup of any remaining
> static data structures based upon recommendations found in the FAQ's. The
> following code snippet illustrates how I am using the parser.
> try {
> XMLPlatformUtils::Initialize();
> }catch (const XMLException& toCatch) {
> ....
> }
> // create new parser for each request
> SAXParser parser;
> parser.setValidationScheme(valScheme);
> parser.setDoNamespaces(doNamespaces);
> parser.setDoSchema(doSchema);
> parser.setDocumentHandler(&handler);
> parser.setErrorHandler(&handler);
> const char* appName = "AppName";
> MemBufInputSource* memBufIS = new MemBufInputSource
> ((const XMLByte*)ar->getMsg(), strlen(ar->getMsg())
> , appName, false);
> parser.parse(*memBufIS);
> // use the terminate method based upon advice in Xerces FAQ
> XMLInterface::terminate();
> The process size grows rapidly as seen using the ps -efl command. The Forte
> workshop environment, dbx, etc. , does not detect the leak, however.
> In an attempt to isolate the leak, the code was instrumented using the ioctl()
> method to get the process byte size from the system process table. Using this
> primitive method, it is possible to see where the code is growing.
> It appears to be occuring somewhere inside the XMLScanner class. The best I can
> do to isolate it is somewhere before the constructor of the XMLReader class.
> The output shows the process size increases somewhere between where the
> ReaderMgr creates the new XMLReader and the constructor of the XMLReader. The
> following is taken from ReaderMgr.cpp.
> 439 try {
> 440 if (src.getEncoding())
> 441 {
> 442 cerr << ": creatReader: " << getProcSizeInBytes() << endl;
> 443 retVal = new XMLReader
> 444 (
> 445 src.getPublicId()
> 446 , src.getSystemId()
> 447 , newStream
> 448 , src.getEncoding()
> 449 , refFrom
> 450 , type
> 451 , source
> 452 );
> 453 cerr << ": creatReader: " << getProcSizeInBytes() << endl;
> 454 }
> 455 else
> 456 {
> 457 cerr <<": creatReader: " << getProcSizeInBytes() << endl;
> 458 retVal = new XMLReader
> 459 (
> 460 src.getPublicId()
> 461 , src.getSystemId()
> 462 , newStream
> 463 , refFrom
> 464 , type
> 465 , source
> 466 );
> cerr <<": creatReader: " << getProcSizeInBytes() << endl;
> The size of the leak is not consistent either. The following data shows the
> delta's in process size in between each request. Each request sends the same xml
> message and gets the same response back. The column on the left details the
> size of the process and the column on the right details the change in size from
> the previous request.
> 16,850,944 0
> 17,817,600 966,656
> 18,006,016 188,416
> 18,006,016 0
> 18,006,016 0
> 18,006,016 0
> 18,104,320 98,304
> 18,169,856 65,536
> 18,194,432 24,576
> 18,194,432 0
> 18,194,432 0
> 18,243,584 49,152
> 18,259,968 16,384
> 18,276,352 16,384
> 18,276,352 0
> 18,276,352 0
> 18,276,352 0
> 18,341,888 65,536
> 18,341,888 0
> 18,448,384 106,496
> 18,538,496 90,112
> 18,538,496 0
> 18,538,496 0
> 18,538,496 0
> 18,538,496 0
> 18,620,416 81,920
> 18,644,992 24,576
> 18,661,376 16,384
> 18,677,760 16,384
> 18,677,760 0
> 18,677,760 0
> 18,677,760 0
> 18,677,760 0
> 18,833,408 155,648
> 18,833,408 0
> 18,833,408 0
> 18,833,408 0
> 18,833,408 0
> 18,833,408 0
> 18,833,408 0
> 18,874,368 40,960
> 18,874,368 0
> 18,874,368 0
> 18,972,672 98,304
> 18,972,672 0
> 19,062,784 90,112
> 19,079,168 16,384
> 19,161,088 81,920
> 19,161,088 0
> 19,161,088 0
> 19,218,432 57,344
> 19,234,816 16,384
> 19,251,200 16,384
> 19,267,584 16,384
> 19,267,584 0
> 19,267,584 0
> 19,267,584 0
> 19,267,584 0
> 19,267,584 0
> 19,267,584 0
> 19,267,584 0
> 19,267,584 0
> 19,333,120 65,536
> 19,415,040 81,920
> 19,415,040 0
> 19,415,040 0
> 19,480,576 65,536
> 19,480,576 0
> 19,570,588 90,012
> 19,570,588 0
> 19,668,992 98,404
> 19,668,992 0
> 19,668,992 0
> 19,668,992 0
> 19,775,488 106,496
> 19,775,488 0
> 19,841,024 65,536
> 19,841,024 0
> 19,841,024 0
> 19,922,944 81,920
> 19,947,520 24,576
> 19,963,904 16,384
> 19,980,288 16,384
> 19,980,288 0
> 19,980,288 0
> 20,037,632 57,344
> 20,054,016 16,384
> 20,054,016 0
> 20,054,016 0
> 20,054,016 0
> 20,054,016 0
> 20,054,016 0
> 20,119,552 65,536
> 20,119,552 0
> 20,160,512 40,960
> 20,176,896 16,384
> 20,193,280 16,384
> 20,217,856 24,576
> 20,234,240 16,384
> 20,250,264 16,024
> 20,250,264 0
> 20,250,264 0
> 20,307,968 57,704
>
>
> What is interesting is that the process size does not always increase, and when
> it does, it does not increase at the same rate. It almost looks like some
> resizing of internal cache or hashtables is taking place at certain intervals.
> As stated earlier, the debugger in workshop was not able to detect the leak.
> Sun was contacted to see why this was the case. After a reviewing the problem,
> they thought it might be related to some known problems with their compiler and
> debugger. After installing some upgrades and patches, the memory leak was still
> not showing up using their tools. At this point Sun believes it is a programming
> logic problem with the xerces library.
>
> This bug appears to be related to bug 8984 and 7278 which have not had any
> change in status since they were entered back in march and early may,
> respectively.
> Thank you for your assistance.
> Thomas
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://nagoya.apache.org/jira/secure/Administrators.jspa
-
If you want more information on JIRA, or have a bug to report see:
http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
