9 GCC HEAD regressions, 7 new, with your patch on 2005-02-18T07:48:16Z.

[Jakub Jelinek]

On Fri, Feb 18, 2005 at 08:56:56PM +0000, GCC regression checker wrote:
> With your recent patch, GCC HEAD has some regression test failures, 
> which used to pass.  There are 7 new failures, and 2
> failures that existed before and after that patch; 1 failures
> have been fixed.
> 
> The new failures are:
> native g++.sum g++.dg/lookup/koenig1.C
> native g++.sum g++.dg/template/error1.C
> native g++.sum g++.old-deja/g++.other/crash24.C
> native g++.sum g++.old-deja/g++.other/vaarg3.C
> native g++.sum g++.old-deja/g++.pt/instantiate8.C
> native g++.sum g++.old-deja/g++.pt/partial2.C
> native g++.sum g++.old-deja/g++.robertl/eb4.C

Some of these are caused by my patch:
> +2005-02-18  Jakub Jelinek  <jakub@redhat.com>
> +
> +	PR c++/19813
> +	* decl.c (start_decl_1): Clear TREE_READONLY flag if
> +	its type has TYPE_NEEDS_CONSTRUCTING.
> +	(complete_vars): Likewise.

Apparently my bootstraps of this patch were all with --enable-checking=release
and so the problems did not pop up.  Although start_decl_1 exits early
if type is error_mark_node, it can set type to error_mark_node afterwards.

Committed as obvious.  Sorry.

2005-02-19  Jakub Jelinek  <jakub@redhat.com>

	* decl.c (start_decl_1): Only check TYPE_NEEDS_CONSTRUCTING
	if type is not error_mark_node.

--- gcc/cp/decl.c.jj	2005-02-18 07:51:10.000000000 +0100
+++ gcc/cp/decl.c	2005-02-19 00:16:01.988566483 +0100
@@ -3829,7 +3829,7 @@ start_decl_1 (tree decl)
      so we need to be conservative and unset TREE_READONLY for types
      with constructors.  Otherwise aliasing code will ignore stores in
      an inline constructor.  */
-   if (TYPE_NEEDS_CONSTRUCTING (type))
+   if (type != error_mark_node && TYPE_NEEDS_CONSTRUCTING (type))
      TREE_READONLY (decl) = 0;
 }
 
	Jakub