jimm@amc-gw.amc.com (Jim McElroy) (05/04/90)
In the most recent issue of _The C++ Report_, Tom Cargill has an article titled: "Multiple Inheritance: Still a Research Project?" He writes that in the various published examples that he has seen, the code can be rewritten to use single inheritance without the code becomming larger or more complex. The rewriting generally involves making one of the classes a member of the other (composition, instead of a second line of inheritance). I would like to know the opinions of the readers of this group on the following points: Multiple inheritance is -- -- not worth it's weight in complexity -- convenient in a few instances, but non-essential -- very convenient -- absolutely essential Also, can anyone show examples of program organization in which multiple inheritance is a big win? Jim McElroy Meta-Analytics Bellevue, Washington jimm@amc.com
hardin@hpindda.HP.COM (John Hardin) (05/05/90)
jimm@amc-gw.amc.com (Jim McElroy) writes: > I would like to know the opinions of the readers of this group on > the following points: > > Multiple inheritance is -- > > -- not worth it's weight in complexity > > -- convenient in a few instances, but non-essential > > -- very convenient > > -- absolutely essential ---------- Last year I asked one of the developers of Borland's Turbo Pascal 5.5 (the one with object oriented extensions) how serious they were about OOP in Pascal, like were they planning on offering multiple inheritance in the future. I was told that they were serious, but that I only thought I needed multiple inheritance and they would not be offering it in Pascal. I admit to a lot of amazement on my part at this attitude. If I always developed all of my projects from the ground up, I could probably do fairly well without multiple inheritance. It would often simplify my job to have it, but it would come under "very convenient" on the above scale. However, IMHO, the main commercial draw to OOP (i.e., why for-profit companies are jumping on the OOP bandwagon) is largely for its benefits in the area of promoting code re-use. Faced with an extensive library of classes (which will hopefully be the case a few years from now), if I can inherit from only one of them I will have to choose which one gives me the most leverage and then re-implement (or probably copy code) from the others I need. I suspect that this "do we need multiple inheritance" discussion is due to the early stage of OOP in most commercial enterprises. Ten years from now we will all wonder how we could have been so naive. John Hardin hardin@hpindgh.hp.com ---------------
aaron@rruxh.UUCP (Akman) (06/08/90)
A few C++ Questions:
In Stroustrup pp. 163-166:
1. Is it still true that setting 'this' and zeroing it tell the
compiler not to use the free store? I understand it this way (am I
correct?): The compiler checks the constructor for 'this =
something' and if it finds that it won't bother getting storage for
'this' on entry to the constructor; otherwise the constructor is
called with a 'this' pointer that has already been given some storage.
2. Can someone explain the infinite loop in 5.5.6? I'm a little confused.
3. In 5.5.8, we're familiar with this variable-sized object trick
from the good 'ol days of C, but it's a little confusing to see:
class char_stack {
// whatever...
char s[1];
public:
char_stack(int sz);
// and so on...
};
because we (I) think of structure members as being contiguous data
in memory. I realize that char_stack is not data, but a function
member. My question is: What is really happening internally here?
I'm guessing that all the data parts are collected in contiguous
memory, no per-object space is allocated for any of the function
members, but some housekeeping type pointer helps an object find its
class functions in a shared place. So, really, there is nothing
that comes after 'char s[1]' in that example?
4. Do I have this right?
char_stack s1(100);
char_stack s2 = s1; // from p. 157
s2 = s1;
The 1st one says: allocate a char_stack and call the constructor
that expects an int (or whatever). The second says, allocate a
char_stack and call the X(X&) constructor. The second case is NOT
an assignment. The 3rd case says call 'operator=' for s2 with s1 as
its argument. Is the default 'operator=' bitwise copy, or is it
undefined? I understand why these assignments are dangerous in the
example shown.
--
-----------
Aaron Akman, 201-699-8019, bellcore!rruxh!aaron, RRC 4D-728hudgens@SCRI.FSU.EDU (Jim Hudgens) (10/29/90)
Does anyone have any C++ wrappers for curses? Or a set of classes built around curses? Thanks in advance. -- Disclaimer: I didn't do it. Jim Hudgens Supercomputer Computations Research Institute hudgens@sun13.scri.fsu.edu
rms@AI.MIT.EDU (Richard Stallman) (11/12/90)
I recently sent a message to a GNU mailing list, help-g++@prep.ai.mit.edu, talking about the use of that very list (and its associated newsgroup, gnu.g++.help (?)) to promote porting GNU software to Apple systems. It appears that my message was copied onto comp.lang.c++. Probably the mailing list is gatewayed there. Some people seem to think it is wrong of me to send such a message there. In fact, I did not do so. If you saw it on that newsgroup, please keep in mind my message was about the use of the GNU mailing list (and its equivalent newsgroup). I do urge the readers of comp.lang.c++ not to support the use of Apple equipment, but I would not assert that such support is misuse of comp.lang.c++. I'm sorry if anyone who got the wrong impression felt offended. There will be such messages on help-g++ from time to time. The only way to avoid having them appear on comp.lang.c++ would be to turn off the gateway. The readers of comp.lang.c++ ought to be the ones to decide whether to do this. Note, however, that the only choices available are to gateway nothing, to gateway everything, or to find a person who would like to do the work of gatewaying selectively. It may be that the gateway fails to indicate that the messages come from help-g++. If so, maybe an improvement in the gateway could avoid confusion. I don't know what the message looks like after it goes through the gateway. I am sending this message to help-g++ as a way of getting it onto comp.lang.c++. I don't post net news directly.
tower@AI.MIT.EDU (Leonard H. Tower Jr.) (11/15/90)
From: rms@ai.mit.edu (Richard Stallman) Date: Sun, 11 Nov 90 19:17:09 -0500 I recently sent a message to a GNU mailing list, help-g++@prep.ai.mit.edu, talking about the use of that very list (and its associated newsgroup, gnu.g++.help (?)) to promote porting GNU software to Apple systems. It appears that my message was copied onto comp.lang.c++. Probably the mailing list is gatewayed there. It is. Some people seem to think it is wrong of me to send such a message there. In fact, I did not do so. If you saw it on that newsgroup, please keep in mind my message was about the use of the GNU mailing list (and its equivalent newsgroup). I do urge the readers of comp.lang.c++ not to support the use of Apple equipment, but I would not assert that such support is misuse of comp.lang.c++. I'm sorry if anyone who got the wrong impression felt offended. There will be such messages on help-g++ from time to time. The only way to avoid having them appear on comp.lang.c++ would be to turn off the gateway. The readers of comp.lang.c++ ought to be the ones to decide whether to do this. Note, however, that the only choices available are to gateway nothing, to gateway everything, or to find a person who would like to do the work of gatewaying selectively. The readers of comp.lang.c++ did approve the one way gating in the Spring of 1988 by ~84 yes votes to ~6 no votes. It may be that the gateway fails to indicate that the messages come from help-g++. If so, maybe an improvement in the gateway could avoid confusion. I don't know what the message looks like after it goes through the gateway. I've appended the USENET news header from a gatewayed message. The Newsgroups:, Organization:, and Followup-To: header lines make it clear what's going on. Even so, we plan to change the Organization: field to make things even clearer. I am sending this message to help-g++ as a way of getting it onto comp.lang.c++. I don't post net news directly. thanx -len ---------------------------------------------------------------------- Xref: mit-eddie gnu.g++.help:106 comp.lang.c++:10837 Path: mit-eddie!bu.edu!att!tut.cis.ohio-state.edu!AI.MIT.EDU!rms From: rms@ai.mit.edu (Richard Stallman) Newsgroups: gnu.g++.help,comp.lang.c++ Subject: comp.lang.c++ Date: 12 Nov 90 00:17:09 GMT Sender: daemon@tut.cis.ohio-state.edu Followup-To: gnu.g++.help Organization: GNUs Not Usenet ----------------------------------------------------------------------