Date: Wed, 14 Sep 94 22:21 PDT
From: jdp AT polstra DOT com (John Polstra)
To: djgpp AT polstra DOT com
Subject: Solved: GCC Segmentation Violation after 1st Undeclared Variable

As has been reported here earlier by somebody else, GCC in DJGPP112
quits with a segmentation violation shortly after reporting the first
undeclared variable in a C++ program.  I tracked down the cause of the
problem this evening, with the help of the new GDB in the DJGPP112
release.  (Thanks, DJ!)

The problem turns out to be a bug in GCC and is not specific to DJGPP.
It just happens to cause a segmentation violation under DJGPP.  Here is
a patch:

[Beginning of patch]
*** src/gcc-260/cp/typeck.c-	Sat Jul  9 00:12:25 1994
--- src/gcc-260/cp/typeck.c	Wed Sep 14 22:04:28 1994
***************
*** 77,83 ****
  {
    tree type = TREE_TYPE (value);
  
!   /* First, detect a valid value with a complete type.  */
    if (TYPE_SIZE (type) != 0
        && type != void_type_node
        && ! (TYPE_LANG_SPECIFIC (type)
--- 77,87 ----
  {
    tree type = TREE_TYPE (value);
  
!   /* If the value is the error mark node, leave it alone.  */
!   if (value == error_mark_node)
!     return value;
! 
!   /* Detect a valid value with a complete type.  */
    if (TYPE_SIZE (type) != 0
        && type != void_type_node
        && ! (TYPE_LANG_SPECIFIC (type)
[End of patch]

The macro TYPE_LANG_SPECIFIC was referencing a tree node field that does
not exist (is not allocated) in the special error_mark_node.  The
error_mark_node gets put into the tree when the undeclared variable is
detected.

I will report the bug and the patch to the GNU folks.

   John Polstra                                       jdp AT polstra DOT com
   John D. Polstra & Co., Inc.                   Phone (206) 932-6482
   Seattle, Washington USA                         Fax (206) 935-1262
   "Self-knowledge is always bad news."                 -- John Barth