Re: [PATCH v2] c++: Mitigate -Wuseless-cast with classes [PR85043]

[Marek Polacek <polacek@redhat.com>]

On Wed, Oct 19, 2022 at 02:37:23PM -0400, Jason Merrill wrote:
> On 10/19/22 12:38, Marek Polacek wrote:
> > On Wed, Oct 19, 2022 at 09:27:27AM -0400, Jason Merrill wrote:
> > > On 10/18/22 13:38, Marek Polacek wrote:
> > > > -Wuseless-cast (not part of -Wall/-Wextra) warns here:
> > > > 
> > > >     struct S { };
> > > >     void g (S&&);
> > > >     void f (S&& arg)
> > > >     {
> > > >       g (S(arg)); // warning: useless cast to type 'struct S'
> > > >     }
> > > > 
> > > > which is wrong: the code will not compile without the cast because
> > > > "arg" is an lvalue which cannot bind to S&&.
> > > > 
> > > > I'd like to disable the warning when a class object is cast to
> > > > a non-reference type, which seems like a minimal patch that fixes the
> > > > problems reported in our Bugzilla.  Of course, the cast in "(int)i"
> > > > may not be useless, either, but I'm not changing that here.
> > > 
> > > As I commented on PR 14710, "I wouldn't warn about a cast that changes the
> > > type or value category of an expression at all...."
> > 
> > Ah, I didn't see that comment.
> > 
> > > The code currently checks the value category when casting to a reference,
> > > but not when casting to a non-reference; that seems like the thing to fix.
> > 
> > OK, how about this?  The nice thing is that it handles non-classes as well,
> > while we still warn about truly useless casts as in "X(X{})".  Thanks,
> > 
> > Bootstrapped/regtested on x86_64-pc-linux-gnu, ok for trunk?
> > 
> > -- >8 --
> > -Wuseless-cast (not part of -Wall/-Wextra) warns here:
> > 
> >    struct S { };
> >    void g (S&&);
> >    void f (S&& arg)
> >    {
> >      g (S(arg)); // warning: useless cast to type 'struct S'
> >    }
> > 
> > which is wrong: the code will not compile without the cast because
> > "arg" is an lvalue which cannot bind to S&&.
> > 
> > This patch disables the warning when an object that isn't a prvalue
> > is cast to a non-reference type.  Therefore we still warn about the
> > useless cast in "X(X{})".
> > 
> > 	PR c++/85043
> > 
> > gcc/cp/ChangeLog:
> > 
> > 	* typeck.cc (maybe_warn_about_useless_cast): Don't warn when
> > 	a glvalue is cast to a non-reference type.
> > 
> > gcc/ChangeLog:
> > 
> > 	* doc/invoke.texi: Update documentation of -Wuseless-cast.
> > 
> > gcc/testsuite/ChangeLog:
> > 
> > 	* g++.dg/warn/Wuseless-cast.C: Remove dg-warning.
> > 	* g++.dg/warn/Wuseless-cast3.C: New test.
> > ---
> >   gcc/cp/typeck.cc                           |  4 ++-
> >   gcc/doc/invoke.texi                        | 13 +++++++++-
> >   gcc/testsuite/g++.dg/warn/Wuseless-cast.C  | 12 ++++-----
> >   gcc/testsuite/g++.dg/warn/Wuseless-cast3.C | 29 ++++++++++++++++++++++
> >   4 files changed, 50 insertions(+), 8 deletions(-)
> >   create mode 100644 gcc/testsuite/g++.dg/warn/Wuseless-cast3.C
> > 
> > diff --git a/gcc/cp/typeck.cc b/gcc/cp/typeck.cc
> > index da0e1427b97..8e1d14ee6b0 100644
> > --- a/gcc/cp/typeck.cc
> > +++ b/gcc/cp/typeck.cc
> > @@ -8108,7 +8108,9 @@ maybe_warn_about_useless_cast (location_t loc, tree type, tree expr,
> >   	   && (TYPE_REF_IS_RVALUE (type)
> >   	       ? xvalue_p (expr) : lvalue_p (expr))
> >   	   && same_type_p (TREE_TYPE (expr), TREE_TYPE (type)))
> > -	  || same_type_p (TREE_TYPE (expr), type))
> > +	  /* Don't warn when converting a class object to a non-reference type,
> > +	     because that's a common way to create a temporary.  */
> > +	  || (!glvalue_p (expr) && same_type_p (TREE_TYPE (expr), type)))
> 
> Might use ?: instead of || so we check this only when !TYPE_REF_P.  OK with
> that change.

Ah nice, pushed with this instead of the ||:

      if (TYPE_REF_P (type)
          ? ((TYPE_REF_IS_RVALUE (type)
              ? xvalue_p (expr) : lvalue_p (expr))
             && same_type_p (TREE_TYPE (expr), TREE_TYPE (type)))
          /* Don't warn when converting a class object to a non-reference type,
             because that's a common way to create a temporary.  */
          : (!glvalue_p (expr) && same_type_p (TREE_TYPE (expr), type)))


Thanks,

Marek