(no subject)

[haberg]

. <address@hidden> <address@hidden>
Mime-Version: 1.0 (Apple Message framework v746.2)
Content-Type: text/plain; charset=US-ASCII; delsp=yes; format=flowed
Message-Id: <address@hidden>
Cc: address@hidden
Content-Transfer-Encoding: 7bit
From: Hans Aberg <address@hidden>
Subject: Re: too many warnings from Bison CVS for Pike
Date: Thu, 9 Feb 2006 19:45:44 +0100
To: Paul Eggert <address@hidden>
X-Mailer: Apple Mail (2.746.2)

On 9 Feb 2006, at 01:48, Paul Eggert wrote:

> If a compiler defines __cplusplus, then yacc.c by default never calls
> memcpy, so there should never be a correctness issue about bypassing
> user-defined C++ assignment operators.  There is a way to override
> this, but it's not documented and nobody should use it unless they
> really know what they're doing, and (again) it's on them to get it to
> work.  For details, please see data/yacc.c; for an example, see
> src/parse-gram.c.

I can give three C++ possibilities:

Static C-parserstack. This seems what Bison is implementing for C++,  
i.e., yyoverflow is not defined, leaving up to the user to implement  
parser stack extensions. This will work with all C++ types, as long  
as the parser stack does not overflows.

Dynamic C-parserstack. This is what is Bison implements if  
__cplusplus is not defined. However, this should work under C++ as  
well for POD types. By the segment:

#if (! defined (yyoverflow) \
      && (! defined (__cplusplus) \
         || (defined (YYLTYPE_IS_TRIVIAL) && YYLTYPE_IS_TRIVIAL \
              && defined (YYSTYPE_IS_TRIVIAL) && YYSTYPE_IS_TRIVIAL)))

/* A type that is properly aligned for any stack member.  */
union yyalloc
{
   short int yyss;
   YYSTYPE yyvs;
     YYLTYPE yyls;
};
...

you already have a test against non-POD types, as these cannot be  
used in a union. So if one would want to use this dynamic stack with  
a POD type, just remove the (! defined (__cplusplus) above.

Dynamic C-parserstack with C++ extensions. This would be the same as  
the dynamic C-parserstack, but with some extra C++ extensions (using  
__cplusplus) so that also non-POD semantic types can be used, and  
possibly even with C++ exceptions. This was (I think) discussed  
before, somebody even posting examples how to call assignment  
operators and destructors explicitly. One might at this point also  
even provide proper cleanup using the %destructor implementation. But  
it seems quite complicated.

But one might supply a macro, say CPP_SEMANTIC_POD, which signals  
that the C++ semantic type is a POD, and that the dynamic C- 
parserstack should be used. The code above would then look like (I  
think, if I cmoputed it right):
#if (! defined (yyoverflow) \
      && (! ((defined (__cplusplus) && !(defined CPP_SEMANTIC_POD))) \
         || (defined (YYLTYPE_IS_TRIVIAL) && YYLTYPE_IS_TRIVIAL \
              && defined (YYSTYPE_IS_TRIVIAL) && YYSTYPE_IS_TRIVIAL)))
I.e., if __cplusplus is defined, and also CPP_SEMANTIC_POD, then  
yyoverflow should be defined as well.

- Also, there is an error in the comment of the block right above:
#  if defined (__STDC__) || defined (__cplusplus)
#   include <stdlib.h> /* INFRINGES ON USER NAME SPACE */
#   define YYSIZE_T size_t
I think this comment is due to the compiler I formerly used, which  
added a "using namespace std" in its C compatibility header. However,  
this is wrong according to the C++ standard: One just adds a new name  
for each name defined in the C-header. For example, the C++ of malloc  
is std::malloc, but when the C-compatibility header is added, one  
just gets the name ::malloc added. So this comment can simply be  
taken away.

   Hans Aberg