10 replies to this topic

[jeroenb]

I recently encountered a piece of code like the following (not saying anything about whether it is good or not):

[source lang="cpp"]class Ptr{public: Ptr(void* ptr = NULL, bool owned= true); ...}Ptr ptr(false);[/source]

This seems to compile pretty well, and the false is implicitly converted to a null pointer! In case you give true as argument to this constructor, you get a compile error.

Now my question is: how can you prefend that the bool is converted automatically to a null pointer. I tried an explicit constructor with only one bool argument. But that made it impossible to send NULL as pointer to the first.

[mhagain]

Yup, that's the beauty of C++ - false is 0 and so is NULL.

[samoth]

Does not compile just fine (though it does compile) with gcc:
warning: converting 'false' to pointer type for argument 1 of 'Ptr::Ptr(void*, bool)'

Strictly speaking, your compiler is correct about doing this conversion. The standard says: (4.5-6) "A prvalue of type bool can be converted to a prvalue of type int, with false becoming zero and true becoming one." and (4.10) "A null pointer constant is an integral constant expression (5.19) prvalue of integer type that evaluates to zero or a prvalue of type std::nullptr_t. A null pointer constant can be converted to a pointer type".

This explains why it is legal for false, but not for true (the former is a null pointer constant, the latter is not). It's converting a bool constant to an integer constant, which happens to be a null pointer constant, which it converts to a (null) pointer.

Edited by samoth, 21 June 2012 - 03:48 AM.

[jeroenb]

That it is valid, I agree on. But now I would like to prevend that this constructor is called by accident with only a bool argument. Adding a constructor explicit Ptr(bool owned) again makes it impossible to call the constructor with the NULL value, as that will result in an ambiguous call to either constructor. Any idea's?

ps. I am compiling with the microsoft compiler, which does not produce by default a warning for this.

[ApEk]

http://ideone.com/qiHP9
Default arguments in constructors should almost always be implemented as different constructors along with associating initialization lists.

[Ameise]

http://ideone.com/qiHP9
Default arguments in constructors should almost always be implemented as different constructors along with associating initialization lists.

That makes sense, if your compiler supports delegating constructors. VC10/11 do not. Otherwise, you are duplicating a lot of code, or using initializer private methods.

[frob]

Can you remove the default parameter? That would be easiest.

As for delegating constructors, it isn't hard for your constructors to run along the lines:

MyObject::MyObject()
{
   ConstructInternal( NULL, false, true);
}
MyObject::MyObject(Something* ptr )
{
  ConstructInternal( ptr, false, true);
}
MyObject::MyObject(Something* ptr, bool owned )
{
  ConstructInternal( ptr, false, owned);
}
MyObject::MyObject(bool aFlag)
{
  ConstructInternal( NULL, aFlag, true);
}
...

We do this all the time in our game code for exactly the reason you describe in your post.

True it is not as pretty as default parameters, but it does help resolve the problem of implicit conversions.


As the project grows and a new parameter becomes needed we just add another signature so we don't break existing code. It solves so many headaches when sharing code between multiple consoles.

Edit: as a side note, try to avoid plain bool as a flag. Use an enum and specifically state what the thing is or is not. Today there are two choices, but tomorrow you may need a third.

Edited by frob, 21 June 2012 - 05:23 PM.

[Narf the Mouse]

Edit: as a side note, try to avoid plain bool as a flag. Use an enum and specifically state what the thing is or is not. Today there are two choices, but tomorrow you may need a third.

Plus, new Object( MakeBobYourUncle.True ) ; is much less ambiguous than new Object( true ) ;

Edited by Narf the Mouse, 22 June 2012 - 01:43 PM.

[jeroenb]

Frob, thats a good idea, and indeed is much clearer. Thanks for the suggestions all, this helps a lot!

[tbergelt]

I recently encountered a piece of code like the following (not saying anything about whether it is good or not):

[source lang="cpp"]class Ptr{public: Ptr(void* ptr = NULL, bool owned= true); ...}Ptr ptr(false);[/source]

if you define false as !true this will no longer compile.

#define false !true

[frob]

Please don't revive old threads. This conversation was done three months ago.