[LLVMbugs] [Bug 2505] New: Support for autorelease pools and 'autorelease' messages in memory leak analysis

[bugzilla-daemon at cs.uiuc.edu]

http://llvm.org/bugs/show_bug.cgi?id=2505

           Summary: Support for autorelease pools and 'autorelease' messages
                    in memory leak analysis
           Product: clang
           Version: unspecified
          Platform: Macintosh
        OS/Version: MacOS X
            Status: NEW
          Keywords: new-feature
          Severity: enhancement
          Priority: P2
         Component: Semantic Analyzer
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: nikita at zhuk.fi
                CC: llvmbugs at cs.uiuc.edu


In a non-GC enrivonment, clang static analyzer stops tracking reference counts
as soon as it sees an "autorelease" message sent to an object (tracking is
stopped in CFRefCount.cpp:879). 

This causes clang to miss possible memory leaks and uses after release. Some
concrete examples:

- (void)useAfterRelease
{
  NSAutoreleasePool *pool = [[NSAutoreleasePool alloc] init];
  NSString *str = [[NSString alloc] initWithString:@"some string data"]; //
rc(str)=1

  [str autorelease]; // rc(str)=1, added to local autorelease pool 'pool'
  [pool release]; // rc(str)=0, since local autorelease pool 'pool' is released

  NSLog(@"%@", str); // 'str' is used after release. Not reported by clang
checker.
}

- (void)memoryLeak
{
  NSString *str = [[NSString alloc] initWithString:@"some string data"]; //
rc(str)=1

  [str retain]; // rc(str)=2
  [str autorelease]; // rc(str)=2, and will be 1 after the current event.
   // 'str' is leaked, since it goes out of scope. Not reported by clang
checker.
}

Clang should probably track local autorelease pools (which may be nested) and
handle 'autorelease' messages just as it handles 'release' messages, but taking
into account the knowledge about autorelease pools. 

Clang revision used: 52881


-- 
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.