[boost] [shifted_ptr] Neural networks

From: Phil Bouchard (philippe_at_[hidden])
Date: 2008-05-28 06:51:00

Greetings:

I have made some changes into my sandbox and I am happy to say some tests 
send by Steven a while ago are now working correctly.  The interface was 
previously supporting only allocations made by "new_sh".  I was aware of 
this but now I am adding a solution to make shifted_ptr generic enough.

The solution is still a non-portable one but will be corrected shortly; it 
consists of detecting the stack frame and I am using gcc's 
__builtin_frame_address().  I also replaced Nedmalloc for Boost.Pool because 
I temporarily wish to reduce problems I might face.  The problem with 
Boost.Pool is that it doesn't handle arbitrary sized chunks and the 
complexity goes down to O(n) for array allocations.  All in all shifted_ptr 
behaves differently when a pointer resides either on the:
1. stack & data segment
2. local heap
3. system heap

Moreover the test file shifted_ptr_test3.cpp which uses BOOST_TEST macros is 
crashing on gcc for Windows.  The same tests appears in 
shifted_ptr_test2.cpp and work correctly.

Finally yes I definitely need to write documentation about this but I will 
basically need to write something similar to a book.  For now I just want to 
present a working example and here it is.  The ultimate objective of 
shifted_ptr is to handle neural networks for artificial intelligence 
testings so feel free using the pointer in your tests to help me finding 
flaws that might still exist.

Regards,
-Phil