The root of this issue: in the JRex makefile there is a build
variable SHARED_LIBRARY_LIBS which links mozilla object files
directly into the JRex dylib.. but the JRex code we have also loads
mozilla libraries dynamically, duplicating some symbols.
Removing the SHARED_LIBRARY_LIBS variable solves the issues.
Somehow we were getting away lightly when using GCC 3.3..
Zoltan
On 28 Jun 2007, at 13:59, Zoltan Foley-Fisher wrote:
Hello there,
We are having an issue with compilation differences between GCC 3.3
and 4.0, OS 10.4.10, on PowerPC using 10.3.9 SDK, on Intel using
10.4u SDK.
We are working with the JRex library to embed Gecko framework in a
Universal Java application. When we compile using GCC 3.3, the
appropriate dynamic libraries are created without problem and
packaged into a Java application bundle, which launches correctly.
But when we compile the same code with GCC 4.0, the dynamic
libraries are created, but fail to load into our Java application
correctly. We see some strange behaviours when using GDB, that
occur every time the program is run, and only in very specific
places in the large codebase:
On PPC, a C++ function is called
nsIXPConnect *xpc = nsContentUtils::XPConnect();
but returns null. nsContentUtils::XPConnect is defined as:
static nsIXPConnect *XPConnect()
{
return sXPConnect;
}
When we follow the call with GDB we notice that there are two
sXPConnect symbols...
On Intel, another C++ function is called:
nsresult rv = CallGetService(NS_SCRIPTSECURITYMANAGER_CONTRACTID,
&sSecurityManager);
but the call results in sSecurityManager null. We follow the call
with GDB and see correct values and code flow until this function
return, where it mysteriously "drops" the value to be assigned to
sSecurityManager.
Even more mysteriously, if we move the program counter back with
GDB and repeat the function a second time, it returns the value for
sSecurityManager correctly!
Have any of you guys encountered such behaviour before?
Thanks for any help,
Zoltan Foley-Fisher
Atimi Software Inc.
