Re: Patches for --enable-checking

[Franz Sirl <Franz dot Sirl-kernel at lauterbach dot com>]

Am Fri, 02 May 2036 schrieb Richard Kenner:
>When I ran with --enable-checking, it showed quite a number of problems,
>the patches for which are below.  There are still about a half dozen
>g++ testsuite crashes on checking errors, but they are all in error
>recovery cases where an error message has already been given.  They are
>beyond my ability to fix, so I'll leave them for somebody else.
>
>Sun Mar 26 20:15:26 2000  Richard Kenner  <kenner@vlsi1.ultra.nyu.edu>
>
>	* c-convert.c (convert): Return if output or input type is ERROR_MARK.
>	* c-decl.c (duplicate_decls): Only look at DECL_BUILT_IN_NONANSI
>	and DECL_INLINE if FUNCTION_DECL.
>	(pushdecl, redeclaration_error_message): Likewise, for DECL_INLINE.
>	(store_parm_decls): Check for type of PARM_DECL being ERROR_MARK.
>	Use DECL_WEAK, not DECL_RESULT, to flag for already seen.
>	(combine_parm_decls): Likewise.
>	* ggc-common.c (gcc_mark_tree_children, case 'd'): Use DECL_RESULT_FLD.
>	* print-tree.c (print_node): Likewise.
>	Only test DECL_PACKED, DECL_INLINE, DECL_BIT_FIELD, and
>	DECL_TRANSPARENT_UNION on proper decl types.
>	Properly handly DECL_INCOMING_RTL and DECL_SAVED_INSNS.
>	* stor-layout.c (layout_decl): Only check DECL_PACKED and
>	DECL_BIT_FIELD of FIELD_DECL.
>	* tree.h (DECL_RESULT_FLD): New macro.
>	* cp/class.c (check_field_decl): Fix typo.
>	(build_vtbl_or_vbase_field): Don't clear DECL_SAVED_INSNS.
>	(check_methods): Likewise.
>	(check_field_decls): Likewise.
>	Use DECL_CONTEXT, not DECL_FIELD_CONTEXT.
>	* cp-tree.h (DECL_SHADOWED_FOR_VAR, DECL_TEMPLATE_RESULT):
>	Use DECL_RESULT_FLD, not DECL_RESULT.
>	* cp/decl.c (xref_tag): Use DECL_TEMPLATE_RESULT.
>	* cp/lex.c (identifier_type): Likewise.
>	* cp/pt.c (determine_specialization, lookup_template_class): Likewise.
>	(tsubst_friend_function, tsubst_decl, instantiate_template): Likewise.
>	(resolve_overloaded_unification, more_specialized): Likewise.
>	* cp/semantics.c (finish_member_declaration): Likewise.
>	* cp/typeck.c (build_x_function_call): Likewise.

Great! The testsuite is nearly back to normal now. I think you just missed the
Fortran stuff. The appended patch fixes that.

OK to commit?

Franz.


Index: ChangeLog
===================================================================
RCS file: /cvs/gcc/egcs/gcc/f/ChangeLog,v
retrieving revision 1.240
diff -u -p -r1.240 ChangeLog
--- ChangeLog	2000/03/25 18:34:13	1.240
+++ ChangeLog	2000/03/28 19:36:15
@@ -1,3 +1,9 @@
+2000-03-28  Franz Sirl  <Franz.Sirl-kernel@lauterbach.com>
+
+	* com.c (ffecom_decl_field): Use DECL_ALIGN and DECL_OFFSET_ALIGN
+	for a FIELD_DECL.
+	(ffecom_init_0): Likewise.
+
 Sat Mar 25 09:12:10 2000  Richard Kenner  <kenner@vlsi1.ultra.nyu.edu>
 
 	* com.c (ffecom_tree_canonize_ptr_): Use bitsize_zero_node.
Index: com.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/f/com.c,v
retrieving revision 1.82
diff -u -p -r1.82 com.c
--- com.c	2000/03/25 18:34:13	1.82
+++ com.c	2000/03/28 19:36:31
@@ -11000,7 +11000,8 @@ ffecom_decl_field (tree context, tree pr
 
   field = build_decl (FIELD_DECL, get_identifier (name), type);
   DECL_CONTEXT (field) = context;
-  DECL_FRAME_SIZE (field) = 0;
+  DECL_ALIGN (field) = 0;
+  DECL_OFFSET_ALIGN (field) = 0;
   if (prevfield != NULL_TREE)
     TREE_CHAIN (prevfield) = field;
 
@@ -12056,7 +12057,8 @@ ffecom_init_0 ()
 						 ffecom_tree_type[i][j]);
 	DECL_CONTEXT (ffecom_multi_fields_[i][j])
 	  = ffecom_multi_type_node_;
-	DECL_FRAME_SIZE (ffecom_multi_fields_[i][j]) = 0;
+	DECL_ALIGN (ffecom_multi_fields_[i][j]) = 0;
+	DECL_OFFSET_ALIGN (ffecom_multi_fields_[i][j]) = 0;
 	TREE_CHAIN (ffecom_multi_fields_[i][j]) = field;
 	field = ffecom_multi_fields_[i][j];
       }