PATCH RFA: Remove extern "C" from java header file

[Ian Lance Taylor-3]

The java frontend header file java/verify.h has an extern "C" wrapper.
This breaks when compiling java with C++.  I can't think of any reason
for having it. This patch removes it.  Bootstrapped and tested on
x86_64-unknown-linux-gnu.  OK for mainline?

Ian


2009-06-09  Ian Lance Taylor  <iant@...>

        * verify.h: Remove extern "C".



Index: java/verify.h
===================================================================
--- java/verify.h (revision 148327)
+++ java/verify.h (working copy)
@@ -1,5 +1,6 @@
 /* Declarations to interface gcj with bytecode verifier.
-   Copyright (C) 2003, 2004, 2005, 2006, 2007 Free Software Foundation, Inc.
+   Copyright (C) 2003, 2004, 2005, 2006, 2007, 2009
+   Free Software Foundation, Inc.
 
 This file is part of GCC.
 
@@ -26,11 +27,6 @@
 #ifndef GCC_VERIFY_H
 #define GCC_VERIFY_H
 
-#ifdef __cplusplus
-extern "C"
-{
-#endif
-
 #include "system.h"
 #include "coretypes.h"
 #include "jcf.h"
@@ -155,8 +151,4 @@
 
 int verify_method (vfy_method *meth);
 
-#ifdef __cplusplus
-}
-#endif
-
 #endif /* ! GCC_VERIFY_H */

[Andrew Pinski-2]

On Tue, Jun 9, 2009 at 9:58 PM, Ian Lance Taylor<iant@...> wrote:
> The java frontend header file java/verify.h has an extern "C" wrapper.
> This breaks when compiling java with C++.  I can't think of any reason
> for having it. This patch removes it.  Bootstrapped and tested on
> x86_64-unknown-linux-gnu.  OK for mainline?

How does it break?  The original code in verify.h came from the
runtime which was originally written in C++ so they wanted to provide
a C interface of the C++ code.

Thanks,
Andrew Pinski

[Ian Lance Taylor-3]

Andrew Pinski <pinskia@...> writes:

> On Tue, Jun 9, 2009 at 9:58 PM, Ian Lance Taylor<iant@...> wrote:
>> The java frontend header file java/verify.h has an extern "C" wrapper.
>> This breaks when compiling java with C++.  I can't think of any reason
>> for having it. This patch removes it.  Bootstrapped and tested on
>> x86_64-unknown-linux-gnu.  OK for mainline?
>
> How does it break?  The original code in verify.h came from the
> runtime which was originally written in C++ so they wanted to provide
> a C interface of the C++ code.

It breaks because the extern "C" is used around the gcc header files
such as system.h, coretypes.h, etc.  When gcc is compiled with C++,
those header files declare C++ functions, not C functions.  Thus there
is confusion as to what the functions should be named.

If it is important that this file be C++ ready when the compiler is
compiled with C, then I'll need to use some sort of #if to move the
extern "C" after the #include's.  Let me know.

Ian

[Bryce McKinlay-3]

On Wed, Jun 10, 2009 at 3:29 AM, Ian Lance Taylor<iant@...> wrote:

> Andrew Pinski <pinskia@...> writes:
>
>> On Tue, Jun 9, 2009 at 9:58 PM, Ian Lance Taylor<iant@...> wrote:
>>> The java frontend header file java/verify.h has an extern "C" wrapper.
>>> This breaks when compiling java with C++.  I can't think of any reason
>>> for having it. This patch removes it.  Bootstrapped and tested on
>>> x86_64-unknown-linux-gnu.  OK for mainline?
>>
>> How does it break?  The original code in verify.h came from the
>> runtime which was originally written in C++ so they wanted to provide
>> a C interface of the C++ code.
>
> It breaks because the extern "C" is used around the gcc header files
> such as system.h, coretypes.h, etc.  When gcc is compiled with C++,
> those header files declare C++ functions, not C functions.  Thus there
> is confusion as to what the functions should be named.

I think this is just a hangover from when the verifier was originally
ported from the C++ version in the runtime. I can't think of a reason
to have it - the verifier implementation is not actually shared
between the front end and the runtime, so it can be removed.

Bryce

[Andrew Haley]

Bryce McKinlay wrote:
>
> I think this is just a hangover from when the verifier was originally
> ported from the C++ version in the runtime. I can't think of a reason
> to have it - the verifier implementation is not actually shared
> between the front end and the runtime, so it can be removed.

Thanks for the reminder.  I agree, the patch is fine.

Andrew.