In short if gcc has to return a struct, it first puts it in the stack of the callee and then copies it to the stack of the caller. I have described the problem at the following blog http://blog.360.yahoo.com/blog-Soi.8NUyaKJGQJEysjIbGQ--?p=2 I thought it was a bug .. but that may not ne true .. someone one told me that this was done to support overloaded copy constructor's in C++, I am curious to know why this was done. In particular why following transformation was not done ie given main () { class X x; x = foo(); } X foo () { return X(2,3); } now the gcc may make x = foo(); => foo( &x); so foo gets addr of x in main's stack .. and foo becomes void foo( X * x) { return x->copy_constructor (2,3 ) } in this case no copy from foo's stack to main's stack will be needed .. Thanks and Regards, Ravi Krishna
