[boost] Direct vs. indirect shared_ptr

From: Nathan Myers (ncm_at_[hidden])
Date: 1998-12-06 18:04:14

Beman Dawes writes:
 > Greg has two implementations (with identical interfaces) of
 > shared_ptr:
 > 
 >    (1) An "direct" version which privately holds the T* and a pointer
 > to a reference counter on the heap.
 > 
 >    (2) An "indirect" version which privately holds a single pointer
 > to a heap allocated struc which contains the T* and reference
 > counter.
 >
 > What should we do:
 > 
 >    * Pick one implementation?  If so, which one? (Greg's preference.)
 >    * Supply both with different names?  What are good names, then?

There is a third option: parameterize.  However, I wouldn't
advocate that.  Ultimately we should let Greg choose, but 
if I were to advise him I would choose the "direct" version.
Simple benchmarks are unlikely to distinguish the two, but
on a heavily-loaded system the indirect version is more
vulnerable to locality effects and to memory manager 
fragmentation.

An intrusive version for use where you can put the reference-counter
in the referent would be useful too, but should have a different name.

Incidentally, these will be interesting to make thread-safe, and
more interesting to make efficiently thread-safe.  Such thread-safety
is inherently non-portable, but where better to encapsulate it than
in the boost library?  

Some work on encapsulation of concurrency primitives for a variety 
of architectures is found in GNU glibc-2.1.  

Nathan Myers
ncm_at_[hidden]
------------------------------------------------------------------------
Free Web-based e-mail groups -- http://www.eGroups.com