This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[C++ PATCH] Allow some function pointer conversions


This patch implements the smallest parts of DR195 that I think are
appropriate right now.  DR195 is in drafting, but has been kicking around
for several years. The patch fixes the regressions caused by Mark's patch
for 14035 in the C++ testsuite.

The new testcase g++.dg/expr/cast2.C, still fails.  It think it
should probably be removed as the testcase with this patch is doing
more complete checking.  Also I changed g++.old-deja/g++.mike/p10148.C.
I couldn't really see the point of the c-cast there -- it was converting
the function pointer to a base (of the current class) pointer.  I suspect
some parens might have been missed.

I've not documented the casts this patch permits, but will do so once
we've agreed on the desired semantics. Specifically, this patch permits

*) pointer-to-function to be explicitly converted to pointer-to-void,
provided there is no loss of precision.
*) pointer-to-void can be explicitly converted to pointer-to-function,
regardless of precision loss.  The reason for a lack of precision check
here, is so one can convert a function-pointer to a void-pointer and back
again, so long as the void-pointer has at least as many bits.

DR195 does not single out pointer-to-void, but I consider void pointers
to be a reasonable restriction right now.  We can always relax it later.

Let me know if you don't think this is the correct approach.

booted & tested on i686-pc-linux-gnu.

nathan
--
Nathan Sidwell    ::   http://www.codesourcery.com   ::     CodeSourcery LLC
nathan@codesourcery.com    ::     http://www.planetfall.pwp.blueyonder.co.uk

2004-10-20  Nathan Sidwell  <nathan@codesourcery.com>

	* typeck.c (composite_pointer_type): Add comment about DR 195
	(build_reinterpret_cast_1): Add for_reinterpret_cast_p parameter.
	Allow function pointer conversions that DR195 suggests.
	(build_reinterpret_cast, build_c_cast): Update
	build_reinterpret_cast_1 calls. 

2004-10-20  Nathan Sidwell  <nathan@codesourcery.com>

	* g++.dg/conversion/dr195.C: New.
	* g++.old-deja/g++.mike/p10148.C: Remove ill-formed cast.

Index: cp/typeck.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/cp/typeck.c,v
retrieving revision 1.584
diff -c -3 -p -r1.584 typeck.c
*** cp/typeck.c	19 Oct 2004 23:24:18 -0000	1.584
--- cp/typeck.c	20 Oct 2004 12:04:41 -0000
*************** composite_pointer_type (tree t1, tree t2
*** 507,512 ****
--- 507,514 ----
        tree result_type;
  
        if (pedantic && TYPE_PTRFN_P (t2))
+ 	/* Although DR195 suggests allowing this when no precision is
+ 	   lost, that is only allowed in a reinterpret_cast.  */
  	pedwarn ("ISO C++ forbids %s between pointer of type %<void *%> "
                   "and pointer-to-function", location);
        result_type 
*************** convert_member_func_to_ptr (tree type, t
*** 4803,4809 ****
  
  static tree
  build_reinterpret_cast_1 (tree type, tree expr, bool c_cast_p,
! 			  bool *valid_p)
  {
    tree intype;
  
--- 4805,4811 ----
  
  static tree
  build_reinterpret_cast_1 (tree type, tree expr, bool c_cast_p,
! 			  bool for_reinterpret_ref_p, bool *valid_p)
  {
    tree intype;
  
*************** build_reinterpret_cast_1 (tree type, tre
*** 4843,4849 ****
        expr = build_unary_op (ADDR_EXPR, expr, 0);
        if (expr != error_mark_node)
  	expr = build_reinterpret_cast_1
! 	  (build_pointer_type (TREE_TYPE (type)), expr, c_cast_p,
  	   valid_p);
        if (expr != error_mark_node)
  	expr = build_indirect_ref (expr, 0);
--- 4845,4851 ----
        expr = build_unary_op (ADDR_EXPR, expr, 0);
        if (expr != error_mark_node)
  	expr = build_reinterpret_cast_1
! 	  (build_pointer_type (TREE_TYPE (type)), expr, c_cast_p, true,
  	   valid_p);
        if (expr != error_mark_node)
  	expr = build_indirect_ref (expr, 0);
*************** build_reinterpret_cast_1 (tree type, tre
*** 4922,4928 ****
  	   || (TYPE_PTRFN_P (intype) && TYPE_PTROBV_P (type)))
      {
        if (pedantic || !c_cast_p)
! 	pedwarn ("ISO C++ forbids casting between pointer-to-function and pointer-to-object");
        expr = decl_constant_value (expr);
        return fold_if_not_in_template (build_nop (type, expr));
      }
--- 4924,4947 ----
  	   || (TYPE_PTRFN_P (intype) && TYPE_PTROBV_P (type)))
      {
        if (pedantic || !c_cast_p)
! 	{
! 	  /* DR 195 suggests allowing such casts if they do not lose
! 	     precision.  We allow conversion to pointer-to-void, if it
! 	     does not lose precision, and we allow conversion from
! 	     pointer-to-void regardless, so that one may convert
! 	     back again without warning.  Such conversions are not
! 	     permitted when we are recursively called to deal with
! 	     reinterpretting reference casts. */
! 	  if (!for_reinterpret_ref_p && VOID_TYPE_P (TREE_TYPE (type)))
! 	    {
! 	      if (TYPE_PRECISION (type) < TYPE_PRECISION (intype))
! 		warning ("conversion from %qT to %qT loses precision",
! 			 intype, type);
! 	    }
! 	  else if (for_reinterpret_ref_p || !VOID_TYPE_P (TREE_TYPE (intype)))
! 	    pedwarn ("ISO C++ forbids casting between pointer-to-function and pointer-to-object");
! 	}
!       
        expr = decl_constant_value (expr);
        return fold_if_not_in_template (build_nop (type, expr));
      }
*************** build_reinterpret_cast (tree type, tree 
*** 4955,4960 ****
--- 4974,4980 ----
      }
  
    return build_reinterpret_cast_1 (type, expr, /*c_cast_p=*/false,
+ 				   /*for_reinterpret_ref=*/false,
  				   /*valid_p=*/NULL);
  }
  
*************** build_c_cast (tree type, tree expr)
*** 5157,5162 ****
--- 5177,5183 ----
    /* Or a reinterpret_cast.  */
    if (!valid_p)
      result = build_reinterpret_cast_1 (type, value, /*c_cast_p=*/true,
+ 				       /*for_reinterpret_ref_p=*/false,
  				       &valid_p);
    /* The static_cast or reinterpret_cast may be followed by a
       const_cast.  */
Index: testsuite/g++.dg/conversion/dr195.C
===================================================================
RCS file: testsuite/g++.dg/conversion/dr195.C
diff -N testsuite/g++.dg/conversion/dr195.C
*** /dev/null	1 Jan 1970 00:00:00 -0000
--- testsuite/g++.dg/conversion/dr195.C	20 Oct 2004 13:29:45 -0000
***************
*** 0 ****
--- 1,25 ----
+ // Copyright (C) 2004 Free Software Foundation, Inc.
+ // Contributed by Nathan Sidwell 20 Oct 2004 <nathan@codesourcery.com>
+ 
+ // DR 195 allows conversions between function and object pointers
+ // under some circumstances.
+ 
+ typedef void (*PF)(void);
+ typedef void *PV;
+ typedef int *PO;
+ 
+ 
+ void foo ()
+ {
+   PF pf;
+   PV pv;
+   PO po;
+ 
+   pf = reinterpret_cast <PF>(pv);
+   pv = reinterpret_cast <PV>(pf);
+   pf = reinterpret_cast <PF>(po); // { dg-error "casting between" "" }
+   po = reinterpret_cast <PO>(pf); // { dg-error "casting between" "" }
+ 
+   pv = pf; // { dg-error "invalid conversion" "" }
+   pf = pv; // { dg-error "invalid conversion" "" }
+ }
Index: testsuite/g++.old-deja/g++.mike/p10148.C
===================================================================
RCS file: /cvs/gcc/gcc/gcc/testsuite/g++.old-deja/g++.mike/p10148.C,v
retrieving revision 1.4
diff -c -3 -p -r1.4 p10148.C
*** testsuite/g++.old-deja/g++.mike/p10148.C	1 May 2003 02:02:45 -0000	1.4
--- testsuite/g++.old-deja/g++.mike/p10148.C	20 Oct 2004 13:29:54 -0000
*************** public:
*** 23,29 ****
  };
  
  void TCRCB::eat () {
!  void *vp = (TIRD*)this->itc;
   this->itc();
  }
  
--- 23,29 ----
  };
  
  void TCRCB::eat () {
!  void *vp = (void *)((TIRD*)this)->itc;
   this->itc();
  }
  

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]