[C++] Weffc++/Wnon-virtual-dtor confusion

[Nathan Sidwell <nathan at acm dot org>]

This patch fixes some unexpected behaviour of the Weffc++ and Wnon-virtual-dtor 
flags.  The documentation for the latter says it's also enabled by Weffc++, but 
that's untrue.

The current behaviour of Weffc++ is to warn about any non-virtual dtor in a 
non-polymorphic base class (relying on the below described -Wnon-virtual-dtor 
case to catch polymorphic classes).  Since Edition 3 of Scott Meyer's book, the 
rule has been that it only applies to polymorphic classes.  I.e.  a polymorphic 
class should not have an accessible non-virtual destructor and all of its bases 
should be likewise. (it's possible earlier editions stated this rule too, but 
3rd ed was what I could find).

The current behaviour of Wnon-virtual-dtor is to complain about accessible 
non-virtual dtors in polymorphic classes.  This is what one expects.

One surprising outcome of the current implementation is that -Weffc++ 
-Wno-non-virtual-dtor still complains about the lack of virtual dtors in base 
classes.

This patch:
1) Clarifies the documentation for -Wnon-virtual-dtor to be for polymorphic 
classes and bases of such classes. (i.e. move the 3rd ed Weffc++ behaviour into 
this flag).

2) As different editions of Scott's book renumber the rules, I removed the 
numbering in Weffc++ and merged the two separate lists.

3) Made -Wnon-virtual-dtor get set from -Weffc++.

4) Moved the check for nonvirtual dtors in base classes to after we determine 
the class is polymorphic.  Warn for all bases, not just the non-polymorphic ones.

5) Made that check trigger on -Wnon-virtual-dtor, not Weffc++.

I removed the 3 scattered tests for non-virtual dtor checking and updated and 
cloned the warn/Wnvdtor.C test to check for the interactions between the two flags.

built & tested on i686-pc-linux-gnu.  I'll commit in a few days, unless there 
are comments.

nathan

Attachment: all.diff
Description: Text document