Fix extern "C" naming issues on OSX #3020
Conversation
Signed-off-by: Nazim Uddin Bhuiyan <nazimudd@ualberta.ca>
Clang adds another underscore to these names, and results in linker issues. While global/extern symbols written in assembly need to have leading underscores on OSX, they can not be referred to with leading underscores in c/cpp. Signed-off-by: Nazim Uddin Bhuiyan <nazimudd@ualberta.ca>
|
Jenkins test sanity xlinux,win |
| @@ -4707,7 +4707,11 @@ bool TR::CompilationInfo::isQueuedForCompilation(J9Method * method, void *oldSta | |||
| return linkageInfo->isBeingCompiled(); | |||
| } | |||
|
|
|||
| #if defined(OSX) | |||
| JIT_HELPER(initialInvokeExactThunkGlue); | |||
There was a problem hiding this comment.
Can the JIT_HELPER macro be updated to include _ on OSX? That would minimize the other required changes.
There was a problem hiding this comment.
The leading _ is needed for platforms other than OSX. So what we would need is to have JIT_HELPER(x) defined to extern "C" void x() on mac and extern "C" void _x() on other platforms. We would then also need to enclose any calls to these functions in C/C++ in another macro, say, CALL_JIT_HELPER.
However, I did not go with that solution because the problem is that there is a lack of consistency here. The method calls/declarations with leading underscore in C also are named with leading underscores in the MASM code. Most of the methods are missing a leading underscore (and also in the assembly code), and so using our single JIT_HELPER macro that expands differently on OSX would not work. We would then need to divide them into 2 different categories, requiring 4 different macros: JIT_HELPER(x), JIT_HELPER_UNDERSCORE(x), CALL_JIT_HELPER(x), CALL_JIT_HELPER_UNDERSCORE(x).
Did you have a different idea in mind?
There was a problem hiding this comment.
The VM handles this somewhat differently by hiding the platform differences in the macros. Looking at the JIT code, the approach there is to handle it with ifdefs throughout.
I don't want this to block this PR from moving forward. Though I think it would be worth looking at cleaning this up in the future.
There was a problem hiding this comment.
This will be cleaned up once the NASM assembly files are set up to be used by Windows and xLinux. We will no longer need to have the code in the #else blocks of #if defined(OSX) in this PR.
|
Jenkins test sanity |
|
I agree with @DanHeidinga that hiding the macOS specifics behind the |
This is the other side of #3004, where references from c/cpp to assembly symbols should not have a leading underscore, since Clang adds an underscore to them anyway. Currently, this fix is implemented just for macOS, but in the future, there will be a cleaner, more unified solution once NASM is used across all x86 operating systems.
Signed-off-by: Nazim Uddin Bhuiyan nazim.uddin.bhuiyan@ibm.com
Issue: #36