orbitcpp-list

[orbitcpp-list] question about ORBit-tools

[Ronald G. <rga...@pi...>]

hey guys,
     got a question about some code I found in orbitcpp_tools.hh.

There is some code which goes like so:

#ifdef ORBITCPP_HAVE_BAD_CAST
	#define ORBITCPP_NARROW_OR_NULL(type,item) \
		type *result; \
		try { \
			result = dynamic_cast<type *>(item); \
		} \
		catch (bad_cast) {\
		    result = NULL; \
		}
#else
	#warning ORBit-C++ will not detect narrowing errors on pseudo objects.
	#define ORBITCPP_NARROW_OR_NULL(type,item) \
		type *result = static_cast<type *>(item);
#endif


a few questions/comments:

     1)  In C++, dynamic casting to a pointer type doesn't throw
exceptions, it just returns a 1.  bad_cast is thrown on dynamic casts
to references...ie, I don't think that code will work.

     2)  I looked at the code to set ORBITCPP_HAVE_BAD_CAST in
acinclude.m4... as far as I recall, in g++, bad_cast can be found in
<typeinfo> instead of <exception>.  I think this is the wrong place
according to c++ standards, but if we're going to use it, we might as
well support the most common compiler for this stuff.

     3)  I think that the above could be implemented as a template,
n'est-ce pas?  That would be the c++ way

just wondering what's thought about that...

ron

Re: [orbitcpp-list] question about ORBit-tools

[Phil D. <ph...@us...>]

Ronald Garcia wrote:
> 
> hey guys,
>      got a question about some code I found in orbitcpp_tools.hh.
> 
> There is some code which goes like so:
> 
> #ifdef ORBITCPP_HAVE_BAD_CAST
>         #define ORBITCPP_NARROW_OR_NULL(type,item) \
>                 type *result; \
>                 try { \
>                         result = dynamic_cast<type *>(item); \
>                 } \
>                 catch (bad_cast) {\
>                     result = NULL; \
>                 }
> #else
>         #warning ORBit-C++ will not detect narrowing errors on pseudo objects.
>         #define ORBITCPP_NARROW_OR_NULL(type,item) \
>                 type *result = static_cast<type *>(item);
> #endif
> 
> a few questions/comments:
> 
>      1)  In C++, dynamic casting to a pointer type doesn't throw
> exceptions, it just returns a 1.  bad_cast is thrown on dynamic casts
> to references...ie, I don't think that code will work.
> 

I don't think the dynamic cast thing will work at all, since it requires
rtti information AFAIK, and we're using reinterpret_casted C objects
which won't have this information (since no vtable etc..). Andreas?

Cheers,

Phil.

Re: [orbitcpp-list] question about ORBit-tools

[Andreas K. <ak...@ix...>]

Phil Dawes wrote:
> I don't think the dynamic cast thing will work at all, since it requires
> rtti information AFAIK, and we're using reinterpret_casted C objects
> which won't have this information (since no vtable etc..). Andreas?

The only point where this _is_ indeed needed is narrowing of exceptions,
which is in turn barely needed anywhere. Additionally, as Ron correctly
stated in his original message, the code is broken. If you feel like
eliminating this, go ahead.

bye
andy

Re: [orbitcpp-list] question about ORBit-tools

[Andreas K. <ak...@ix...>]

Ronald Garcia wrote:
> 
> hey guys,
>      got a question about some code I found in orbitcpp_tools.hh.
> 
> There is some code which goes like so:
> 
> #ifdef ORBITCPP_HAVE_BAD_CAST
>         #define ORBITCPP_NARROW_OR_NULL(type,item) \
>                 type *result; \
>                 try { \
>                         result = dynamic_cast<type *>(item); \
>                 } \
>                 catch (bad_cast) {\
>                     result = NULL; \
>                 }
> #else
>         #warning ORBit-C++ will not detect narrowing errors on pseudo objects.
>         #define ORBITCPP_NARROW_OR_NULL(type,item) \
>                 type *result = static_cast<type *>(item);
> #endif
> 
> a few questions/comments:
> 
>      1)  In C++, dynamic casting to a pointer type doesn't throw
> exceptions, it just returns a 1.  bad_cast is thrown on dynamic casts
> to references...ie, I don't think that code will work.

whoops. you're right.

> 
>      2)  I looked at the code to set ORBITCPP_HAVE_BAD_CAST in
> acinclude.m4... as far as I recall, in g++, bad_cast can be found in
> <typeinfo> instead of <exception>.  I think this is the wrong place
> according to c++ standards, but if we're going to use it, we might as
> well support the most common compiler for this stuff.

if we decide to keep the code, we should probably add g++ compatibility
and fix it.

>      3)  I think that the above could be implemented as a template,
> n'est-ce pas?  That would be the c++ way

true. it's just those damn old habits, you know :))

bye
andy