leo@atcmp.nl (Leo Willems) (07/16/90)
The next question was in comp.lang.c++ before, but is repeated since I think it never reached the `world': The next program calls a default constructor at an unexpected place, I think it is a bug. ================= struct A{ A(){ puts("I am default constructor A()"); } }; struct B{ B() { puts("I am default constructor B()"); } B(B&){ puts("I am copy constructor B(B&)"); } }; struct C{ C(){ puts("I am default constructor C()");} A a; B b; }; main() { puts(" first a 'default' constructor:"); C defc; puts("\n now a copy constructor: "); C copc(defc); } ======outputs: (after compilation with Glsp. 2.0 for sparc station): first a 'default' constructor: I am default constructor A() I am default constructor B() I am default constructor C() now a copy constructor: I am default constructor A() <===== offending constructor call I am copy constructor B(B&) =============================== This is the first reason why I think this is a bug: The A() constructor is used on the A part of variable copc,( not on some temporary) so the A part of copc is first initialised using A::A(), THEN overwitten using the default memberwise initialisation. (In practice, this will not often lead to an error) This is the second reason why I think this is a bug: When in class C the declarations of the members A and B are switched the `error' is gone, the output becomes: ================ first a 'default' constructor: I am default constructor B() I am default constructor A() I am default constructor C() now a copy constructor: I am copy constructor B(B&) =============== Is this fixed in 2.1? (If it's a bug......) Thanks Leo Leo Willems Internet: leo@atcmp.nl AT Computing UUCP: mcsun!hp4nl!kunivv1!atcmpe!leo P. O. Box 1428 6501 BK Nijmegen Phone: +31-80-566880 The Netherlands Fax: +31-80-555887
brianl@wenti.Berkeley.EDU (Brian Lee) (07/31/90)
In Article 8713, leo@atcmp.nl (Leo Willems) describes how the default constructor for a member class is sometimes called instead of the copy constructor. I am also using AT&T C++ Release 2.0, and a similar thing happens for base classes. I'm relatively new to this newsgroup, so please forgive me if this is old news. Consider the following program #include <iostream.h> class X { public: X() { cout << "default X::X()\n"; } X(const X&) { cout << "copy X::X(const X&)\n"; }}; class Y : public X { public: Y() { cout << "default Y::Y()\n"; } // Y(const Y&) { cout << "copy Y::Y(const Y&)\n"; } }; class Z : public Y { public: Z() { cout << "default Z::Z()\n"; } Z(const Z&) { cout << "copy Z::Z(const Z&)\n"; } }; main() { cout << "\ndefault:\n"; Z k; cout << "\ncopy:\n"; Z l = k; } It produces the following output default: default X::X() default Y::Y() default Z::Z() copy: default X::X() default Y::Y() copy Z::Z(const Z&) Note that the default constructor for Y is called instead of a copy constructor being generated and that calling the default constructor for Y causes the default constructor for X to be called instead of the copy constructor for X. Assuming that the semantics of member-wise initialization extend to base class initialization, this is wrong. Try increasing the length of the inheritance chain and/or playing with different combinations of defined/undefined copy constructors. It appears that to be safe, one should define a default and copy constructor for all classes. Brian Lee U.C. Berkeley Cadgroup brianl@ic.Berkely.EDU