gcj/83: Failure to compile try/finally?

[Alexandre Petit-Bianco]

The following reply was made to PR gcj/83; it has been noted by GNATS.

From: Alexandre Petit-Bianco <apbianco@cygnus.com>
To: gback@cs.utah.edu
Cc: java-gnats@sourceware.cygnus.com
Subject: Re: gcj/83: Failure to compile try/finally?
Date: Fri, 18 Feb 2000 16:23:37 -0800 (PST)

 gback@cs.utah.edu writes:
 
 > The following program produces incorrect results when
 > compiled from the java source.
 > ...
 > gcj cannot compile the .class file it generates with -C.
 
 This PR actually exposes two problems. The first one has to do with
 gcj not enforcing a correct order of evaluation for something like:
 
 	foo (a+b);
 
 With `a' and `b' being of type String or returning something of type
 String.
 
 `b' is currently evaluated before `a', if `a' modifies `b', its new
 value won't be used to construct the string. I believe this is fixed
 with the patch below.
 
 The second problem is a gcj verifier problem. I think we should file a
 second PR with it.
 
 On thing. I tried to rebuild libgcj with the patch, and I think it
 exposes an other problem in check_init (I might have a patch for
 that.) So, before I get a definite answer, I'm not checking anything
 in.
 
 Here's the patch to fix the order of evaluation problem. I'm filing an
 other PR for the gcj's generated class to native.
 
 ./A
 
 Fri Feb 18 16:18:04 2000  Alexandre Petit-Bianco  <apbianco@cygnus.com>
 
 	* parse.y (patch_string): Call force_evaluation_order on the
  	completed string concatenation tree.
 	* expr.c (force_evaluation_order): Call force_evaluation_order on
 	function's arguments too.
 
 Index: parse.y
 ===================================================================
 RCS file: /cvs/gcc/egcs/gcc/java/parse.y,v
 retrieving revision 1.135
 diff -u -p -r1.135 parse.y
 --- parse.y	2000/02/15 22:54:21	1.135
 +++ parse.y	2000/02/19 00:19:06
 @@ -10272,6 +10272,8 @@ patch_string (node)
        /* Temporary disable forbid the use of `this'. */
        ctxp->explicit_constructor_p = 0;
        ret = java_complete_tree (make_qualified_primary (node, invoke, 0));
 +      /* String concatenation arguments must be evaluated in order too. */
 +      ret = force_evaluation_order (ret);
        /* Restore it at its previous value */
        ctxp->explicit_constructor_p = saved;
        return ret;
 Index: expr.c
 ===================================================================
 RCS file: /cvs/gcc/egcs/gcc/java/expr.c,v
 retrieving revision 1.59
 diff -u -p -r1.59 expr.c
 --- expr.c	2000/02/18 18:17:37	1.59
 +++ expr.c	2000/02/19 00:19:06
 @@ -2694,7 +2694,7 @@ force_evaluation_order (node)
        for (cmp = NULL_TREE, arg = TREE_OPERAND (node, 1); 
  	   arg; arg = TREE_CHAIN (arg))
  	{
 -	  tree saved = save_expr (TREE_VALUE (arg));
 +	  tree saved = save_expr (force_evaluation_order (TREE_VALUE (arg)));
  	  cmp = (cmp == NULL_TREE ? saved :
  		 build (COMPOUND_EXPR, void_type_node, cmp, saved));
  	  TREE_VALUE (arg) = saved;