From patchwork Wed Jun  9 12:11:59 2010
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Subject: [c++] use XALLOCAVEC in the C++ front-end
From: =?utf-8?b?TWFudWVsIEzDs3Blei1JYsOhw7Fleg==?= <lopezibanez@gmail.com>
X-Patchwork-Id: 55086
Message-Id: <AANLkTinAxE2rJ4yNt-YWfafb1vld3CA-jqEFq-1C-NBR@mail.gmail.com>
To: Gabriel Dos Reis <gdr@integrable-solutions.net>
Cc: Jason Merrill <jason@redhat.com>, Nathan Froyd <froydnj@codesourcery.com>, 
 gcc-patches@gcc.gnu.org, "Joseph S. Myers" <joseph@codesourcery.com>
Date: Wed, 9 Jun 2010 14:11:59 +0200

On 9 June 2010 14:08, Gabriel Dos Reis <gdr@integrable-solutions.net> wrote:
> On Wed, Jun 9, 2010 at 6:53 AM, Manuel López-Ibáñez
> <lopezibanez@gmail.com> wrote:
>> On 9 June 2010 13:33, Gabriel Dos Reis <gdr@integrable-solutions.net> wrote:
>>> On Wed, Jun 9, 2010 at 4:53 AM, Manuel López-Ibáñez
>>> <lopezibanez@gmail.com> wrote:
>>>> On 8 June 2010 18:14, Jason Merrill <jason@redhat.com> wrote:
>>>>> On 06/08/2010 10:43 AM, Manuel López-Ibáñez wrote:
>>>>>>
>>>>>> On 8 June 2010 16:06, Jason Merrill<jason@redhat.com>  wrote:
>>>>>
>>>>>>> The error message is unclear, but the issue is that converting from
>>>>>>> char**
>>>>>>> to char const** is unsafe, while converting to char const*const* is safe.
>>>>>>
>>>>>> What does it mean unsafe in this context? I could produce a patch that
>>>>>> changes the message but I would like to be sure what to say.
>>>>>>
>>>>>> semantics.c:1223:7: error: new ‘const’ qualifier in cast from 'char
>>>>>> **' to 'const char **' is unsafe [-Werror=cast-qual]
>>>>> Explanation from the C++ standard (4.4 [conv.qual]):
>>>>>
>>>>> [ Note: if a program could assign a pointer of type T** to a pointer of type
>>>>> const T** (that is, if line #1 below were allowed), a program could
>>>>> inadvertently modify a const object (as it is done on line #2). For example,
>>>>>
>>>>>  int main() {
>>>>>    const char c = ’c’;
>>>>>    char* pc;
>>>>>    const char** pcc = &pc;// #1: not allowed
>>>>>    *pcc = &c;
>>>>>    *pc = ’C’; // #2: modifies a const object
>>>>>  }
>>>>> --end note ]
>>>>
>>>> So what do you think about the following message:
>>>>
>>>> error: new ‘const’ qualifier in cast from 'char **' to 'const char **'
>>>> is unsafe [-Werror=cast-qual]
>>>> note: to be safe all intermediate pointers must be equally qualified
>>>
>>> that is OK; but I find the leading 'new' hard to parse.   Please, just
>>> remove it.
>>>
>>
>> Anyway, this is the C FE.
>
> Yes, I know.
>
>> The C++ FE gives the following:
>>
>> test.C:4:23: warning: invalid conversion from ‘char**’ to ‘const
>> char**’ [-fpermissive]
>>
>> but the code is too intricate to give a clearer error in the case of
>> C++ or to factor out the common check.
>
> Removing the leading 'new' does not require any of that.

Of course, it was a just a comment since consistency between C and C++
is a goal, so we should always check for low hanging-fruit. Not this
case unfortunately.

I am bootstrapping this:


There will be some testcases that need adjusting.

Cheers,

Manuel.


Index: gcc/c-typeck.c
===================================================================
--- gcc/c-typeck.c      (revision 160384)
+++ gcc/c-typeck.c      (working copy)
@@ -4493,9 +4493,12 @@
          && !is_const)
        {
          int added = TYPE_QUALS (in_type) &~ TYPE_QUALS (in_otype);
-         warning (OPT_Wcast_qual,
-                  ("new %qv qualifier in middle of multi-level non-const cast "
-                   "is unsafe"), added);
+         if (warning_at (input_location, OPT_Wcast_qual,
+                          "%qv qualifier in cast from %qT to %qT is unsafe",
+                          added, in_type, in_otype))
+              inform (input_location,
+                      G_("to be safe all intermediate pointers"
+                         " must be equally qualified"));
          break;
        }
       if (is_const)