@Araq, you are ahead of me again, and again I'm glad as it means more 
simplicity, with the thread users somewhat responsible for choosing to use 
atomic accesses or locks as necessary and tuning the allocation/deallocation 
process to ensure thread safety.

> There is no reason to take the cost of a lock for every allocation, there are 
> many ways to avoid that.

Interesting, I judged what would be done from your implementation of the shared 
allocations/deallocations in the current source and it seems you are already 
thinking of something better. I am already thinking of ways that could be done.

> Also, you will be able to setup a thread-local allocator for sections of your 
> code. The allocator proposal is not part of the "destructor's spec".

I didn't know about that. Is the proposal documented somewhere?

In trying to implement functional programming concepts in C/C++, I find it to 
be comparatively slow due to the allocation/deallocation of many small areas of 
memory as required aren't tuned for that. Perhaps the facilities that will be 
made available with this new proposal will help one work around that limitation 
as in allocating larger chunks of memory from the system and then serving out 
these small allocations from that smaller pooled area; that is how Haskell 
handles it so much more quickly.

Thanks for your patience in walking me through these questions...

Reply via email to