Migrating to Solaris 9 - C++ Multithreaded X.25 application turning undead

Question

We have an older multithreaded C++ X.25 application that's been running without a problem for years (24x7). We upgraded the sun server from Solaris 8 to 9 and now, occasionally, the application stops working, but doesn't stop running. X.25 continues at a lower level - interrupts and RRs go back and forth, but the application level doesn't respond. The application's trace log simply stops and no errors are generated. The application has to be killed and restarted. Then after a while, it does it again about once or twice a week, I believe.
The people who do support have looked at it, and can't find anything, so I've been asked to look into it.
But it's on a production server on a customer site, so I don't actually have access to the server. All I can do is make changes to the application and have it installed on their server to run again.
I looked at the changes from Solaris 8 to 9 and the only one that seemed like it could have an impact is the changes to the multithreading implementation. I did come across a hint that there was a change somehow to interrupts and the last trace log messages was printed during a sigalrm interrupt. But that's not as likely because the change seemed to refer more to whether interrupts were considered idle time or not.
This application doesn't actually need to be multithreaded - the guy who wrote it 9 years ago was new to C++ and mutlithreading (had just taken a course) and got a bit carried away. It's a bit of a convoluted monster and everyone hates to touch it - which is why I (Ms. contractor) get it :-) One option is to remove the threads, but this application is in a lot of sites and the company is reluctant right now to let me make such a drastic change. It mike be required though, because they are planning to upgrade other customers to Solaris 9 and they don't want to face this problem with the busier customers.
Has anyone come across any problems with threading moving to Solaris 9? Or does anyone have any suggestion for any debugging that I can add to the applicaton itself to help clear up the mystery?
Any help or advice is appreciated,
Stefani

Answer 1

You replacement for operator new violates the requirements of the C++ standard, and a program using it has undefined behavior. The program could fail to compile, fail to link, fail in unknown ways at run time, or (by accident) do what you want it to do.
You can write your own "placement new" operator that has any exception behavior that you want. But if you replace the library version of operator new, you must follow exactly the requirements listed in the C++ standard, sections 3.7.3 "Dynamic storage duration", and 18.4 "Dynamic memory management".
I can't reproduce all the text here, but in particular a replacement operator new must have the signature
void* operator new(std::size_t) throw(std::bad_alloc); // single-object form
void* operator new(std::size_t) throw(std::bad_alloc); // array form