Re: [Libfoundation-users] Debugging Retain Cycles

[Sven C. K. <sc...@sn...>]

On Mon, Aug 20, 2001 at 11:56:54AM +0200, Helge Hess wrote:
> As long as ObjC objects are concerned Boehm GC *does* use typed memory
> and is exact ! It only needs to guess on stack and global variables.
> While it may not find all leaks it will find 99.99% of the leaks.

Cool!

> > One could read the return address from the stack, and translate it
> > afterwards via binutils's addr2line program, but that's very evil!
> 
> More important: it's very slow which makes it unusable. You can get the
> callee information in the RETAIN macros, but even then tracking all RC
> owners will be a huge memory problem !
> 
> #define RETAIN(x) \
>   [x retainFromFunction:__PRETTY_FUNCTION__ \
>      file:__FILE__ line:__LINE__];

I am afraid, my source code doesn't use the RETAIN et al. macros like lF
does.  So, bascially I would wrap retain et al., and write the callees
address and the address of the object to retain to a logfile.
Afterwards, I would read the logfile with a script and search for retain 
cycles.  Then I would  translate all the callee addresses that are part
of cycles to line numbers via addr2line or using a generated addresses
map to speed up processing.

But I think this scenario is pretty fictional, since I hope GCObject will
do this already for me. 

Regards,

Sven