I'm wondering what people's experiences have been converting projects from MRC (manual reference counting) to ARC (automatic reference counting) with regard to performance and memory usage.
I'm interested primarily in medium/large projects and quantified/profiled results in multiple domains (time, memory, etc), rather than small programs and 'sensed' changes.
I've played around with it in small test cases, and I saw some interesting differences, but I've certainly not migrated full/complex programs at this time.
I'm also curious to hear how much effort you made to minimize memory usage/reference count operations when the converted program was initially developed using MRC: e.g., did you create autorelease pools in the right places? did you favor autorelease on creation, or releasing in the following statement? did you measure and improve memory usage often? I ask because one developer's approach from another's can make a significant difference in the results.
Cheers