TITLE: the empty base optimization

(Source: citations from "Traits on Steriods" a C++ Report column)

AUTHOR: Andrei Alexandrescu

[snip] in C++ even empty objects have a nonzero size. This rule enables much
of the language to remain sound (otherwise, for instance, how would you
build arrays of zero-sized objects?).

[snip]

Fortunately, the standard provides another rule that can be of help
regarding object sizes. It's known as the empty base optimization. If a base
class B of a class D is "empty" (i.e., has no nonstatic data members), the B
subobject in a D object can have an effective size zero. This doesn't break
the language rules because the B subobject is "melted" within the D object;
of course, as soon as you extract a standalone B object it has, again,
nonzero size. Whether you get the empty base optimization or not depends
heavily on your compiler—it's an option, not a requirement. Compilers such
as Metrowerks' Code-Warrior 5.x and Microsoft Visual C++ 6.0 do perform
empty base optimization—an added incentive to try taking advantage of this
optimization.*

[snip]

* Conveniently enough, the latest implementations of the standard C++
library for the mentioned compilers take advantage of the empty base
optimization, which is applicable to containers. Every standard container
aggregates an allocator object, and the default allocator is usually an
empty class.







_______________________________________________
cpptips mailing list
http://cpptips.hyperformix.com