Next
| Up
What is reference counting
Reference counting is not full garbage collection, since it cannot detect detached cycles
of unreference objects. However it is of use in limited domains where cycles are not generated.
An example is in OS's where the mapping between resources and resource-users is maintained via
reference counting.
Each object that is the destination of a pointer has a count field, and this is
maintained to record exactly how many pointers exist that point to the object. The operations of
allocating/pointer-creation, duplicating a pointer, and destroying a pointer all have to maintain this
count.
When the act of destroying a pointer reduces count to zero, then the object is garbage
and is reclaimed (if it itself contains pointers, then these are recursively subject to the
destroy-pointer operation).
Problems with reference counting
Despite being a very straightforward technique, most languages and compilers supporting pointers
do not support the hooks required on pointer creation, duplication and deletion, and so it has to
be explicitly coded, and the cases of pointer duplication have to be accurately identified by
the programmer.
It is quite inefficient - no other technique involves any overhead in copying pointers, for
instance, nor of recursively visiting an objects contents when the last pointer to it vanishes.
It doesn't detect or reclaim cycles.
It requires a count field in each object.
Advantages of reference counting
It recycles dead objects immediately upon their becoming unreachable - promptness - which is very useful for
finalization purposes.
It is easy to implement correctly, and objects to do not have to be moved or copied.
Last updated by markt@chaos.org.uk
Tue 16 January 1996