After talking with Richard Smith, I will be working on some changes to the SelfReferenceVisitor. The HandleExpr was a bit of a hack when first introduced and should be removed. The call to the self reference checker should be moved later into the function so that the initial Expr will be processed first (this processing will get rid of the ParenListExpr). This should clean up the visitor a bit. Stay tuned for a patch.
On Fri, Sep 28, 2012 at 1:34 PM, Pan, Wei <[email protected]> wrote: > I discovered this starting from the following example**** > > ** ** > > extern int &foo();**** > > int &m([&]() { m = foo(); return m; }());**** > > ** ** > > This should be also warned too.**** > > ** ** > > *From:* [email protected] [mailto: > [email protected]] *On Behalf Of *Richard Smith > *Sent:* Friday, September 28, 2012 4:15 PM > *To:* Jordan Rose; Richard Trieu > *Cc:* llvm cfe > *Subject:* Re: [cfe-commits] [Patch] Warn about self-init of refs using > parenthesized initializers**** > > ** ** > > I don't think the patch is right. My recollection of this class is (though > perhaps Richard Trieu can correct me):**** > > ** ** > > HandleExpr is intended to be called for the expression which is the > top-level initializer of the declaration.**** > > HandleValue is intended to be called for an expression which is "used" > (either directly in the initialization of the result, or as the operand of > an lvalue-to-rvalue conversion).**** > > Visit* are called for all evaluated subexpressions (including ones we > don't consider to be "used").**** > > ** ** > > So... I think that we shouldn't be calling HandleExpr (nor HandleValue) > from VisitParenListExpr, since it doesn't imply a use, in the required > sense. I think the right way to deal with this is to add special cases for > ParenListExpr and InitListExpr to HandleExpr. We also need to decide > whether to warn in these cases:**** > > ** ** > > int n(n); // Warn here, or is this treated like int n = n;?**** > > int n{n}; // Warn here, or is this treated like int n = n;?**** > > ** ** > > ** ** > > FWIW, it looks like we're missing an IgnoreParenCasts call from > HandleExpr, and the IgnoreParenImpCasts call in HandleValue should be > IgnoreParenCasts. We don't currently warn on cases like these, for instance: > **** > > ** ** > > int &r = static_cast<int&>(r);**** > > ** ** > > ... nor ...**** > > ** ** > > struct S {} s = static_cast<S&>(s);**** > > ** ** > > On Fri, Sep 28, 2012 at 9:18 AM, Jordan Rose <[email protected]> > wrote:**** > > This still doesn't handle C++11 initialization, does it? > > void g() { > int &a{a}; > } > > Jordan**** > > > > On Sep 28, 2012, at 4:05 , Hans Wennborg <[email protected]> wrote: > > > Hi, > > > > As pointed out in Wei's email [1], Clang currently fails to warn about > > self-initialized references when using parenthesized initializers: > > > > void f() { > > int &a(a); > > } > > > > The attached patch fixes this. Please take a look. > > > > Thanks, > > Hans > > > > [1]. > http://lists.cs.uiuc.edu/pipermail/cfe-dev/2012-September/024582.html**** > > > > <ref_self_init_parenthesized_initializers.patch>_______________________________________________ > > cfe-commits mailing list > > [email protected] > > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits > > _______________________________________________ > cfe-commits mailing list > [email protected] > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits**** > > ** ** >
_______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
