Re: Crash with Clang 3.2 and @syncrhonize

[Doug Warren]

Hi David,

That makes sense to me, but I have to ask undefined based on what?  The most recent "Objective-C Programming Language" I found didn't say anything on the subject.  (Chapter 11 Threading.)  I'll change the code here but was hoping to point out something to the other engineers as to why that's a bad idea and not to do it in the future.  

Now I just need to come up with a failing test that shows our other major issue, class_getProperty not returning a property that's clearly in the class.

On Thu, Jul 18, 2013 at 2:06 AM, David Chisnall <address@hidden> wrote:

This looks like expected behaviour. address@hidden(nil) (which is what your code is doing via a somewhat convoluted path) is undefined behaviour.  It appears to work with recent versions of Apple libobjc, but it crashes with older versions.  Code that relies on it is fundamentally broken (what should it mean?  Don't acquire a lock at all, and potentially introduce races?  Acquire a global lock, and potentially introduce deadlock from unrelated bits of code holding the same lock?) and should be fixed.

David

On 18 Jul 2013, at 00:23, Doug Warren <address@hidden> wrote:

> I'm looking at an odd bug where very simple synchronize tests crash IE:
> @interface TestSync : NSObject
> @property(nonatomic, retain) NSString* sync;
>
> - (BOOL) doSync;
> @end
>
> @implementation TestSync
>
> @synthesize sync;
>
> - (BOOL) doSync
> {
>       @synchronized(sync)
>       {
>               BOOL bTest = YES;
>               return bTest;
>       }
> }
>
> @end
>
> TEST(objc_synchronized, sync)
> {
>     id localSync = [[TestSync alloc] init];
>     printf("Hi: %d\n", localSync->isa);
>     EXPECT_TRUE([localSync doSync]);
> }
>
> Results in:
> [----------] 1 test from objc_synchronized
> [ RUN      ] objc_synchronized.sync
> Hi: 705172864
> Segmentation fault
>
> With a stack trace of:
> ********** Crash dump: **********
> Build fingerprint: 'generic/sdk/generic:4.2.2/JB_MR1.1/576024:eng/test-keys'
> pid: 31728, tid: 31728, name: UNKNOWN  >>> /data/local/tmp/gtestnations <<<
> signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr 00000000
> Stack frame #00  pc 0000c85c  /data/local/tmp/libobjc.so
> Stack frame #01  pc 0000d148  /data/local/tmp/libobjc.so (objc_sync_enter+64)
> Stack frame #02  pc 000234ac  /data/local/tmp/gtest
> Stack frame #03  pc 000236fc  /data/local/tmp/gtest
>
> When looking at it under gdb, when referenceListForObject in associate.m is called, object is NULL, and we're dereferencing NULL with the variable access of isa.  Is there anyway to further debug what is going on from clang converting @synthesize to the call to objc_sync_enter?

> _______________________________________________
> Gnustep-dev mailing list
> address@hidden
> https://lists.gnu.org/mailman/listinfo/gnustep-dev




-- Sent from my STANTEC-ZEBRA