We have done some more research, and narrowed this down to an apparent bug in the Sun Hotspot compiler in JDK 1.4.1 (and we believe 1.4.2). We found that it can happen outside of Xalan calls, so it is not Xalan related.
If we turn of Hotspot the problem goes away. In addition, we found that there are two modes you can run Hotspot in (-client and -server). The -client mode is the default, and we found if we switched to the -server mode, the problem went away. The -server mode is only available with the JDK (not the JRE). We did find this listed as an old bug in JDK 1.3, which was fixed (and apparently reintroduced) at some point. Hope that helps. Cory -----Original Message----- From: Chris McCabe [mailto:[EMAIL PROTECTED] Sent: Friday, June 27, 2003 3:52 PM To: [EMAIL PROTECTED] Subject: Re: Xalan Memory Problem using JDK 1.4 When we ran into this problem, I tried to create a testcase so I could submit it as a bug, but I was unsuccessful, possibly because our production environment runs on Solaris while we develop under Linux. The problem only happened in production, and I could not debug our production environment. Chris Joseph Kesselman wrote: >Reminder, folks; It's very hard to say, or do, much about this kind of >report without testcases to analyse. > >(If they're big, consider making them available on request or attaching 'em >to a Bugzilla report rather than posting them into the mailing list.) > > > > -- Chris P. McCabe - Principal Engineer Choice Hotels International - Information Systems [EMAIL PROTECTED] - 602-953-4416
