[cfe-commits] r59232 - /cfe/trunk/lib/Analysis/GRExprEngine.cpp

[Ted Kremenek]

On Nov 16, 2008, at 9:22 PM, Zhongxing Xu wrote:

>
>
> On Mon, Nov 17, 2008 at 1:14 PM, Zhongxing Xu  
> <xuzhongxing at gmail.com> wrote:
>
>
> On Sun, Nov 16, 2008 at 12:27 PM, Zhongxing Xu  
> <xuzhongxing at gmail.com> wrote:
>
>
> On Sun, Nov 16, 2008 at 11:28 AM, Ted Kremenek <kremenek at apple.com>  
> wrote:
>
> On Nov 15, 2008, at 5:55 AM, Zhongxing Xu wrote:
>
>
> One side problem is that I cannot make the MemRegion* in the  
> ImmutableList const. If I did that, I would get compile error,  
> related to FoldingSet Profile() overloading.
>
> It seems that ImmutableList cannot take a const type as template  
> parameter, otherwise the  two Profile functions in the partial  
> specialization of FoldingSetTrait would have the same signature.  
> (FoldingSet.h:444, 447).
>
> Could we provide a partial specialization of FoldingSetTrait just  
> for ImmutableList?  Or, does (something like) this work (i.e., add  
> another partial specialization for const types):
>
> template<typename const T> struct FoldingSetTrait<const T*> {
>  static inline void Profile(const T* X, FoldingSetNodeID& ID) {
>    ID.AddPointer(X);
>  }
> };
>
> Yeah, this works! Thanks. Incidentally, do you think the old problem  
> is caused by a g++ bug, because, at FoldingSet.h:234, the parameter  
> for Add() already contains 'const', T should not contain 'const'  
> when being instantiated.
>
> I was wrong. That was not g++ bug.
>
> The template argument const T& need a const type. But const  
> MemRegion* is not a const type,
> but a non-const pointer to const MemRegion. So T must match the  
> entire 'const MemRegion*'.
>
> If T only matches 'MemRegion*', then the substituted type would be  
> 'MemRegion* const', that is a const pointer to non-const MemRegion.
>
> So if we write
>
> template <typename T>
> inline void Add(T const & x) { FoldingSetTrait<T>::Profile(x,  
> *this); }
>
> that would be less confusing.

The problem is that we want a 'const MemRegion*' so that Profile()  
operators on constant objects.  If we had:

   const MemRegion* R = ..
   FoldingSetNodeID& ID;
   ID.Add(R);

then (I don't believe) that this wouldn't work if we change Add() to  
take a "T const& x" because it would require that the const qualifier  
in 'const MemRegion*' to be cast way.

Is it possible to add a partial specialization just for pointer  
types?  e.g.:

<template const T*>
inline void Add(const T* X) { .... }

This seems simpler and also means that the pointer is passed-by-value  
instead of passed-by-reference (which is less error prone).
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20081117/ff0afd9b/attachment.html>