Re: Use of valgrinder on GNUstep

[Adam Fedor]

Fred Kiefer wrote:
> I did run a few GNUstep applications under valgrind control and was 
> surprised how few memeory leaks showed up. I solved some of them and 
> some looked like valgrind has a bit of a problem with our object 
> handling, but there are a few other cases where I cannot make any 
> progress. Perhaps someone with a bit more understanding of libobjc could 
> have a look?.
>
> ==5172== 212 bytes in 1 blocks are definitely lost in loss record 50 of 90
> ==5172==    at 0x40163B5B: calloc (vg_clientfuncs.c:245)
> ==5172==    by 0x406A4D6B: objc_calloc (in /usr/lib/libobjc.so.1.0.0)
> ==5172==    by 0x406A6C00: __objc_register_instance_methods_to_class (in 
> /usr/lib/libobjc.so.1.0.0)

Looks like in the runtime, a list is allocated to hold class methods but 
isn't dealocated if there aren't any.


> ==5172== 3381 bytes in 221 blocks are definitely lost in loss record 83 
> of 90
> ==5172==    at 0x401636E3: malloc (vg_clientfuncs.c:103)
> ==5172==    by 0x406A4BE4: objc_malloc (in /usr/lib/libobjc.so.1.0.0)
> ==5172==    by 0x406A73E9: __sel_register_typed_name (in 
> /usr/lib/libobjc.so.1.0.0)
> ==5172==    by 0x406A75EB: sel_register_typed_name (in 
> /usr/lib/libobjc.so.1.0.0)

Harder to figure out without looking deeper. The code looks ok. How does 
it know the bytes are lost? Does that mean they aren't referenced anywhere?