[LLVMbugs] [Bug 11838] New: Clang fails to call temporary destructors in temporaries

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=11838

             Bug #: 11838
           Summary: Clang fails to call temporary destructors in
                    temporaries
           Product: clang
           Version: unspecified
          Platform: Macintosh
        OS/Version: MacOS X
            Status: NEW
          Severity: release blocker
          Priority: P
         Component: C++
        AssignedTo: unassignedclangbugs at nondot.org
        ReportedBy: asomerfield at serif.com
                CC: dgregor at apple.com, llvmbugs at cs.uiuc.edu
    Classification: Unclassified


Hi,

We've just hit a blocker with Clang+LLVM ("Apple Compiler 3.0") in the latest
XCode.

Basically, Clang does not call destructors of temporaries inside a templated
function. GCC+LLVM does not exhibit this behaviour.

Example:



#include <iostream>

class SomeObject
{
public:
    SomeObject()
    {
        printf("Constructed\r\n");
    }

    SomeObject(const SomeObject& rhs)
    {
        printf("Copied\r\n");
    }

    ~SomeObject()
    {
        printf("Destructed\r\n");
    }
};

class Test
{
public:

    template<typename T> // Comment me out to see working case
    void DoTest()
    {
        GetObject();
    }

    SomeObject GetObject()
    {
        return m_object;
    }

    SomeObject m_object;
};

int main (int argc, const char * argv[])
{
    // Works
    //Test test;
    //test.DoTest();

    // Broken
    Test test;
    test.DoTest<int>();
    return 0;
}



Running this code yields:

Constructed
Copied
Destructed

If you a.) Run this with GCC+LLVM OR b.) remove the template<typename T> and
comment in the "Works" code in main, you will get the correct output of:

Constructed
Copied
Destructed
Destructed

We noticed this while tracking down an issue in our code which manifested as
"ref-counted pointers failing to decrement sometimes :/" (imagine if
GetObject() returned a ref-counted object by value).

I don't think this is already fixed - I tried to find similar things in
BugZilla but got nothing - we are using the latest Xcode on latest Mac OS X
Lion using Sandybridge X86 hardware. The bug manifests in both -O0 (debug) and
-O3 (release) - which leads me to guess it's not a dead-code optimisation
problem.

Cheers,

A

-- 
Configure bugmail: http://llvm.org/bugs/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.