CFG: @​finally code is executed for any 'exit' from a @​try block

[llvmbot]

Bugzilla Link	3952
Resolution	FIXED
Resolved on	Oct 27, 2016 02:02
Version	unspecified
OS	All
Attachments	Project demonstrating the problem.
Reporter	LLVM Bugzilla Contributor
CC	@tkremenek

Extended Description

This may or may not be related to http://llvm.org/bugs/show_bug.cgi?id=3888; I'm not clang-savvy enough to know for sure.

Code like the following will falsely warn that 'obj' is leaked:

NSObject *obj = [[NSObject alloc] init];
@​try {
    return nil;
} @​catch (NSException *exc) {
    [obj release];
    obj = nil;
    [exc raise];
} @​finally {
    [obj release];
}

return nil;

But both the @​catch and @​finally block clean it up. Really only the @​finally is needed, but this example was trimmed from code that needed different cleanup on both paths. The assignment obj=nil avoids @​finally hitting a zombie.

Attached a full Xcode project that can be used with scan-build to show the problem.

[llvmbot]

assigned to @tkremenek

[tkremenek]

This is exposing our limited support in the analyzer's control-flow graph for @​try...@catch. We're only approximating the semantics of these blocks right now, and plan on implementing better support for @​try...@catch in a couple months. Thanks for reporting this; this is a good issue to track.

[llvmbot]

Many moons later, I'm no longer able to reproduce this problem, with Xcode 8.1 b3.