C++ PATCH to handling of exception specs in system headers

C++ PATCH to handling of exception specs in system headers

[Jason Merrill]

[-- Attachment #1: Type: text/plain, Size: 330 bytes --]

-pedantic shouldn't change something from OK into an error, but it was 
doing so for redeclaration of a declaration from a system header with a 
mismatched exception specification.  And whether we are strict about 
things in system headers should be controlled by -Wsystem-headers.

Tested x86_64-pc-linux-gnu, applying to trunk.

[-- Attachment #2: ped-sys.patch --]
[-- Type: text/x-patch, Size: 2085 bytes --]

commit 8c029a40a7d4348c432f09eb2512a485cfe9bda8
Author: Jason Merrill <jason@redhat.com>
Date:   Tue May 26 11:47:27 2015 -0400

    	* decl.c (check_redeclaration_exception_specification): Depend on
    	-Wsystem-headers rather than -pedantic.

diff --git a/gcc/cp/decl.c b/gcc/cp/decl.c
index 420c7f40..a8cb358 100644
--- a/gcc/cp/decl.c
+++ b/gcc/cp/decl.c
@@ -1213,14 +1213,19 @@ check_redeclaration_exception_specification (tree new_decl,
      all declarations, including the definition and an explicit
      specialization, of that function shall have an
      exception-specification with the same set of type-ids.  */
-  if ((pedantic || ! DECL_IN_SYSTEM_HEADER (old_decl))
-      && ! DECL_IS_BUILTIN (old_decl)
+  if (! DECL_IS_BUILTIN (old_decl)
       && flag_exceptions
       && !comp_except_specs (new_exceptions, old_exceptions, ce_normal))
     {
-      error ("declaration of %q+F has a different exception specifier",
-	     new_decl);
-      inform (0, "from previous declaration %q+F", old_decl);
+      const char *msg
+	= "declaration of %q+F has a different exception specifier";
+      bool complained = true;
+      if (! DECL_IN_SYSTEM_HEADER (old_decl))
+	error (msg, new_decl);
+      else
+	complained = pedwarn (0, OPT_Wsystem_headers, msg, new_decl);
+      if (complained)
+	inform (0, "from previous declaration %q+F", old_decl);
     }
 }
 
diff --git a/gcc/testsuite/g++.dg/warn/Wsystem-headers1.C b/gcc/testsuite/g++.dg/warn/Wsystem-headers1.C
new file mode 100644
index 0000000..ac5c30d
--- /dev/null
+++ b/gcc/testsuite/g++.dg/warn/Wsystem-headers1.C
@@ -0,0 +1,3 @@
+#include <stdlib.h>
+
+extern double atof (const char *);
diff --git a/gcc/testsuite/g++.dg/warn/Wsystem-headers1a.C b/gcc/testsuite/g++.dg/warn/Wsystem-headers1a.C
new file mode 100644
index 0000000..68f6ea1
--- /dev/null
+++ b/gcc/testsuite/g++.dg/warn/Wsystem-headers1a.C
@@ -0,0 +1,6 @@
+// { dg-do compile { target *-*-*gnu } }
+// { dg-options "-Wsystem-headers" }
+
+#include <stdlib.h>
+
+extern double atof (const char *); // { dg-warning "different exception spec" }

Re: C++ PATCH to handling of exception specs in system headers

[Rainer Orth]

Jason Merrill <jason@redhat.com> writes:

> -pedantic shouldn't change something from OK into an error, but it was
> doing so for redeclaration of a declaration from a system header with a
> mismatched exception specification.  And whether we are strict about things
> in system headers should be controlled by -Wsystem-headers.
>
> Tested x86_64-pc-linux-gnu, applying to trunk.

The new g++.dg/warn/Wsystem-headers1.C test FAILs on Solaris 10 and 11:

FAIL: g++.dg/warn/Wsystem-headers1.C  -std=c++98 (test for excess errors)

/vol/gcc/src/hg/trunk/local/gcc/testsuite/g++.dg/warn/Wsystem-headers1.C:3:33: error: 'double atof(const char*)' conflicts with a previous declaration
In file included from /usr/include/stdlib.h:17:0,
                 from /vol/gcc/src/hg/trunk/local/gcc/testsuite/g++.dg/warn/Wsystem-headers1.C:1:
/var/gcc/regression/trunk/10-gcc/build/gcc/include-fixed/iso/stdlib_iso.h:119:15: note: previous declaration 'double std::atof(const char*)'

	Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University

Re: C++ PATCH to handling of exception specs in system headers

[Jason Merrill]

On 06/01/2015 08:13 AM, Rainer Orth wrote:
> Jason Merrill <jason@redhat.com> writes:
>
>> -pedantic shouldn't change something from OK into an error, but it was
>> doing so for redeclaration of a declaration from a system header with a
>> mismatched exception specification.  And whether we are strict about things
>> in system headers should be controlled by -Wsystem-headers.
>>
>> Tested x86_64-pc-linux-gnu, applying to trunk.
>
> The new g++.dg/warn/Wsystem-headers1.C test FAILs on Solaris 10 and 11:
>
> FAIL: g++.dg/warn/Wsystem-headers1.C  -std=c++98 (test for excess errors)
>
> /vol/gcc/src/hg/trunk/local/gcc/testsuite/g++.dg/warn/Wsystem-headers1.C:3:33: error: 'double atof(const char*)' conflicts with a previous declaration
> In file included from /usr/include/stdlib.h:17:0,
>                   from /vol/gcc/src/hg/trunk/local/gcc/testsuite/g++.dg/warn/Wsystem-headers1.C:1:
> /var/gcc/regression/trunk/10-gcc/build/gcc/include-fixed/iso/stdlib_iso.h:119:15: note: previous declaration 'double std::atof(const char*)'

It's not clear what the conflict is.  Can you quote the declaration from 
stdlib_iso.h?

Jason

Re: C++ PATCH to handling of exception specs in system headers

[Rainer Orth]

Jason Merrill <jason@redhat.com> writes:

> On 06/01/2015 08:13 AM, Rainer Orth wrote:
>> Jason Merrill <jason@redhat.com> writes:
>>
>>> -pedantic shouldn't change something from OK into an error, but it was
>>> doing so for redeclaration of a declaration from a system header with a
>>> mismatched exception specification.  And whether we are strict about things
>>> in system headers should be controlled by -Wsystem-headers.
>>>
>>> Tested x86_64-pc-linux-gnu, applying to trunk.
>>
>> The new g++.dg/warn/Wsystem-headers1.C test FAILs on Solaris 10 and 11:
>>
>> FAIL: g++.dg/warn/Wsystem-headers1.C  -std=c++98 (test for excess errors)
>>
>> /vol/gcc/src/hg/trunk/local/gcc/testsuite/g++.dg/warn/Wsystem-headers1.C:3:33:
>> error: 'double atof(const char*)' conflicts with a previous declaration
>> In file included from /usr/include/stdlib.h:17:0,
>>                   from
>> /vol/gcc/src/hg/trunk/local/gcc/testsuite/g++.dg/warn/Wsystem-headers1.C:1:
>> /var/gcc/regression/trunk/10-gcc/build/gcc/include-fixed/iso/stdlib_iso.h:119:15:
>> note: previous declaration 'double std::atof(const char*)'
>
> It's not clear what the conflict is.  Can you quote the declaration from
> stdlib_iso.h?

sure: it's

extern double atof(const char *);

which is inside

#if __cplusplus >= 199711L
namespace std {
#endif

#ifdef  __cplusplus
extern "C" {
#endif

Isn't this about global vs. std namespace?

	Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University

Re: C++ PATCH to handling of exception specs in system headers

[Jason Merrill]

On 06/01/2015 11:34 AM, Rainer Orth wrote:
> Isn't this about global vs. std namespace?

Probably.  Does making the declaration in the testcase extern "C" help?

Jason

Re: C++ PATCH to handling of exception specs in system headers

[Rainer Orth]

Jason Merrill <jason@redhat.com> writes:

> On 06/01/2015 11:34 AM, Rainer Orth wrote:
>> Isn't this about global vs. std namespace?
>
> Probably.  Does making the declaration in the testcase extern "C" help?

It does indeed, and the test still passes on Linux.

	Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University