Re: missing warning

[bonnardv at pratique dot fr (Valentin Bonnard)]

A (At) 1:51 26/11/98, Alexander V. Lukyanov ecrivait (wrote):
>On Wed, Nov 25, 1998 at 10:26:00PM +0100, Valentin Bonnard wrote:
>> >So a warning could be wrong
>> >in any of the cases, thus better to have no warning.
>> 
>> Sorry I don't follow this resonning.
>
>I mean, someone who uses something like this for purpose will not like the
>warning. 

That's why it isn't an error: it can make sens in particular cases.

>E.g. it's possible to have base class without virtual destructor
>and never delete pointer to base class, or delete it when all derived
>classes don't have dtors too, etc. and the like.
>
>Well, maybe I'm wrong on this. Probably there are a lot of warnings for
>possibly legal code :)  At least such warnings require a commonly accepted
>idea of good style, IMHO.

That's the point of -Wall not being the default.

Using public non virtual destructors in base classes is 
discouraged by all style guides, by Effective C++, and is 
generally a terrible idea.

When it's ok:

- for traits classes which inherit from other trait classes

  There is a well-know example: BidirectionnalIteratorTag 
  inherits from ForwardIteratorTag, InputIteratorTag ect

- for meta programming, where a class is just a value, 
  and one never create instances of it (much delete them)

Most style rules aren't universally accepted; they apply 
to one coding style but may be utter non-sens in another. 
That's why they shouldn't be mandatory.

Making the dtor virtual when it is public is a generally 
accepted rule. The rare exceptions don't IMO imply that we 
should leave it outside -Wall, much less not implement
it at all (again there might be other reasons why it won't 
be implemented).


Valentin Bonnard                mailto:bonnardv@pratique.fr
info about C++/a propos du C++: http://pages.pratique.fr/~bonnardv/