Multiple inheritance without virtual functions in c++

Question

I came across the diamond problem and found different solutions for different cases with a single diamond. However I couldn't find a solution for 'chained' diamonds.

According to the structure: yes, I want to have multiple baseclasses everytime, so virtual inheritance isn't a solution (is it even called diamond then?). I also wanted to avoid get/set-functions for every middle-layer of a diamond.

p   p
|   |
k   k
 \ /
  s

class parent { int val; };
class kid1 : public parent {};
class kid2 : public parent {};
class school : public kid1, public kid2 {};

Accessing val in the parent class works now like follows:

school* s = new school;
s->kid1::val=1; // works

But what about the next 'chained' diamond:

p   p   p   p
|   |   |   |
k   k   k   k
 \ /     \ /
  s       s
  |       |
  c       c
    \   /
      w

class country1 : public school {};
class country2 : public school {};
class world : public country1, public country2 {};

Accessing val via:

world* w = new world;
w->country1::kid1::val=1; // error

results in:

error: ‘kid1’ is an ambiguous base of ‘world’

Why? Isn't the route to the value well defined?

@Nikolay: I think @nem knows that. But for each country we have an unambiguous base kid1. Hence the question, why this doesn't work — Armen Tsirunyan
– Armen Tsirunyan, Commented Mar 18, 2011 at 14:26
I'm pretty sure the school : kid relationship violates the LSP. At least I have never been to a school that was two kids :) — fredoverflow
– fredoverflow, Commented Mar 18, 2011 at 14:40
I fear for the sanity of whoever ends up maintaining this code base - it even drove poor Visual Studio insane. The IDE registers this as an error, but the compiler doesn't. — Jon
– Jon, Commented Mar 18, 2011 at 14:46
this was just hypothetical :) i have a different implementation. — nem
– nem, Commented Mar 18, 2011 at 14:53
@Fred, @Jon: agreed, it would be much better to use composition here than inheritance — Ben Voigt
– Ben Voigt, Commented Mar 18, 2011 at 14:53

Ben Voigt · Accepted Answer · 2011-03-18 14:31:44Z

3

s->kid1::val does not mean "val from the kid1 subobject". It's just a name qualified by the type (not the subobject) that contains it.

I don't know why country1::kid1 is even accepted at all, but apparently it's a typedef for ::kid1. Two data members in world both have the qualified name ::kid1::val.

What you want is:

world* w = new world;
country1* const c1 = world;
c1->kid1::val = 1;

answered Mar 18, 2011 at 14:31

Ben Voigt

286k45 gold badges444 silver badges764 bronze badges

Sign up to request clarification or add additional context in comments.

Comments

James Kanze · Accepted Answer · 2011-03-18 17:43:48Z

1

It is. The error is due to a bug in your compiler.

answered Mar 18, 2011 at 17:43

James Kanze

155k20 gold badges191 silver badges338 bronze badges

12 Comments

Ben Voigt Over a year ago

Considering I've already explained why the code is wrong, any claim the code is right should come with a citation of the particular part of the C++ standard that you think allows it :)

James Kanze Over a year ago

§3.4.3 Qualified name lookup is quite clear. Given w->country1::kid1::val, the compiler uses §3.4.5 Class member access to look up country1; according to §3.4.5/4, if the id-expression is a qualified-id (it is), this lookup takes place in both the class scope, and the context of the complete expression, so the compiler finds ::country1. Following that, the usual rules in §3.4.3 apply.

Ben Voigt Over a year ago

@James: And then the rule is applied again (there are two scope resolution operators), and the compiler finds ::kid1 (as I said in my answer). But ::kid1::val is ambiguous, the compiler is correct. There is no country1::kid1 apart from ::kid1, scope resolution operators work on type names, not subobject names (and base subobjects aren't named, so you can't say w->country1.kid1.val).

James Kanze Over a year ago

@Ben Voigt Which rule? After the compiler has found country1, it's in a qualified name lookup (§3.4.3), and doesn't (or shouldn't) look anywhere but in country1.

Ben Voigt Over a year ago

@James: What is w->country1::kid1? A type? Then w->country1::kid1::val is ambiguous, because that type is twice a base class of world. A subobject? Then it's illegal to follow it with ::.

|

Collectives™ on Stack Overflow

Multiple inheritance without virtual functions in c++

2 Answers 2

Comments

12 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

Comments

12 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related