java/4715

[Tom Tromey <tromey@redhat.com>]

The following reply was made to PR java/4715; it has been noted by GNATS.

From: Tom Tromey <tromey@redhat.com>
To: Bryce McKinlay <bryce@waitaki.otago.ac.nz>
Cc: gcc-gnats@gcc.gnu.org
Subject: java/4715
Date: 20 Dec 2001 23:43:17 -0700

 I looked at this PR a bit tonight.  I didn't really get very far.
 This scenario seems too complex to figure out in one night (for me
 anyway).
 
 It wasn't clear to me that simply returning NULL_TREE when we
 encountered an incomplete type was correct.  I came up with the
 appended patch, which seems to work.  However, I don't vouch for its
 correctness.
 
 Tom
 
 Index: parse.y
 ===================================================================
 RCS file: /cvs/gcc/gcc/gcc/java/parse.y,v
 retrieving revision 1.338
 diff -u -r1.338 parse.y
 --- parse.y 2001/12/21 05:28:27 1.338
 +++ parse.y 2001/12/21 06:29:48
 @@ -5232,14 +5232,23 @@
  
    if (!basetype_vec)
      return NULL_TREE;
 -  
 +
    for (i = 0; i < TREE_VEC_LENGTH (basetype_vec); i++)
      {
 -      tree su = BINFO_TYPE (TREE_VEC_ELT (basetype_vec, i));
 +      tree su;
 +
 +      /* We can end up with a NULL_TREE or an incomplete type here if
 +	 we are compiling multiple files at once.  It's safe to simply
 +	 ignore these cases.  */
 +      if (TREE_VEC_ELT (basetype_vec, i) == NULL_TREE)
 +	continue;
 +      su = BINFO_TYPE (TREE_VEC_ELT (basetype_vec, i));
 +      if (INCOMPLETE_TYPE_P (su))
 +	continue;
  
        if (inherits_from_p (su, target))
  	return lookup_cl (TYPE_NAME (su));
 -      
 +
        for (ctx = DECL_CONTEXT (TYPE_NAME (su)); ctx; ctx = DECL_CONTEXT (ctx))
  	{
  	  /* An enclosing context shouldn't be TARGET */