Re: 252.eon / current GCC

Re: 252.eon / current GCC

[Michael Matz]

Hi,

On Thu, 21 Feb 2002, Andreas Jaeger wrote:

> [... a compile error with 252.eon from SPEC2000 ...]

A reduced testcase for this problem is:

-----------------
template <class T>
class Wrapper {
  public:
    Wrapper (T& a);
    Wrapper (const Wrapper<char>& ref);
};

template <class T>
class Element {
public:
  T * operator[](int x);
};

void test()
{
  char bla = 42;
  Element< Wrapper <unsigned char> > elem;
  elem[1][1] = Wrapper<char> (bla);
}
------------------

The error is cause by this patch:

2002-02-20  Jakub Jelinek  <jakub@redhat.com>
        * typeck.c (cp_pointer_int_sum): Renamed from
        pointer_int_sum, call pointer_int_sum.

In that process a certain thing got lost, namely the completion of the
inner TREE_TYPE on which size_in_bytes() is applied.  The below patch
fixes it for me.  I did not test the patch further.  Neither bootstrapped
nor regtested.  No time because of moving, sorry.


Ciao,
Michael.
-- 
2002-02-21  Michael Matz  <matz@suse.de>

	* typeck.c (cp_pointer_int_sum): Complete inner type which
	is used later by size_in_bytes().

Index: typeck.c
===================================================================
RCS file: /cvsroot/gcc/gcc/gcc/cp/typeck.c,v
retrieving revision 1.388
diff -u -p -r1.388 typeck.c
--- typeck.c	20 Feb 2002 23:05:42 -0000	1.388
+++ typeck.c	22 Feb 2002 02:54:39 -0000
@@ -4076,8 +4076,17 @@ cp_pointer_int_sum (resultcode, ptrop, i
      enum tree_code resultcode;
      register tree ptrop, intop;
 {
-  if (!complete_type_or_else (TREE_TYPE (ptrop), ptrop))
+  tree res_type = TREE_TYPE (ptrop);
+
+  if (!complete_type_or_else (res_type, ptrop))
     return error_mark_node;
+
+  /* pointer_int_sum() uses size_in_bytes() on the TREE_TYPE(res_type)
+     in certain circumstance (when it's valid to do so).  So we need
+     to make sure it's complete.  We don't need to check here, if we
+     can actually complete it at all, as those checks will be done in
+     pointer_int_sum() anyway.  */
+  (void) complete_type (TREE_TYPE (res_type));

   return pointer_int_sum (resultcode, ptrop, fold (intop));
 }

Re: 252.eon / current GCC

[Andreas Jaeger]

Michael Matz <matzmich@cs.tu-berlin.de> writes:

> Hi,
>
> On Thu, 21 Feb 2002, Andreas Jaeger wrote:
>
>> [... a compile error with 252.eon from SPEC2000 ...]
>
> A reduced testcase for this problem is:
>
> -----------------
> template <class T>
> class Wrapper {
>   public:
>     Wrapper (T& a);
>     Wrapper (const Wrapper<char>& ref);
> };
>
> template <class T>
> class Element {
> public:
>   T * operator[](int x);
> };
>
> void test()
> {
>   char bla = 42;
>   Element< Wrapper <unsigned char> > elem;
>   elem[1][1] = Wrapper<char> (bla);
> }
> ------------------
>
> The error is cause by this patch:
>
> 2002-02-20  Jakub Jelinek  <jakub@redhat.com>
>         * typeck.c (cp_pointer_int_sum): Renamed from
>         pointer_int_sum, call pointer_int_sum.
>
> In that process a certain thing got lost, namely the completion of the
> inner TREE_TYPE on which size_in_bytes() is applied.  The below patch
> fixes it for me.  I did not test the patch further.  Neither bootstrapped
> nor regtested.  No time because of moving, sorry.
>
>
> Ciao,
> Michael.


I've bootstrapped and regtested the patch on i686-linux-gnu.  Ok to
commit?

Thanks Michael for investigating this.

Andreas

P.S. Here's the patch again since I'm including gcc-patches this time.


2002-02-21  Michael Matz  <matz@suse.de>

	* typeck.c (cp_pointer_int_sum): Complete inner type which
	is used later by size_in_bytes().

Index: typeck.c
===================================================================
RCS file: /cvsroot/gcc/gcc/gcc/cp/typeck.c,v
retrieving revision 1.388
diff -u -p -r1.388 typeck.c
--- typeck.c	20 Feb 2002 23:05:42 -0000	1.388
+++ typeck.c	22 Feb 2002 02:54:39 -0000
@@ -4076,8 +4076,17 @@ cp_pointer_int_sum (resultcode, ptrop, i
      enum tree_code resultcode;
      register tree ptrop, intop;
 {
-  if (!complete_type_or_else (TREE_TYPE (ptrop), ptrop))
+  tree res_type = TREE_TYPE (ptrop);
+
+  if (!complete_type_or_else (res_type, ptrop))
     return error_mark_node;
+
+  /* pointer_int_sum() uses size_in_bytes() on the TREE_TYPE(res_type)
+     in certain circumstance (when it's valid to do so).  So we need
+     to make sure it's complete.  We don't need to check here, if we
+     can actually complete it at all, as those checks will be done in
+     pointer_int_sum() anyway.  */
+  (void) complete_type (TREE_TYPE (res_type));

   return pointer_int_sum (resultcode, ptrop, fold (intop));
 }


-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj