2006-01-24 Frank Ch. Eigler <fche@elastic.org>

	PR 2060 etc.
	* tapsets.cxx (visit_target_symbol): Tolerate failed resolution by
	letting target_symbol instance pass through to optimizer and
	type checker.
	* elaborate.cxx (semantic_pass_optimize): New family of functions and
	associated visitor classes.
	(visit_for_loop): Tolerate absent init/incr clauses.
	(semantic_pass): Invoke unless unoptimized (-u) option given.
	* main.cxx, session.h: Add support for flag.
	* staptree.cxx (visit_for_loop): Tolerate absent init/incr clauses.
	(traversing_visitor::visit_arrayindex): Visit the index expressions.
	(functioncall_traversing_visitor): New class.
	(varuse_tracking_visitor): New class.
	* staptree.h: Corresponding changes.
	* parse.cxx (parse_for_loop): Represent absent init/incr expressions
	with null statement pointer instead of optimized-out dummy numbers.
	* stap.1.in: Document optimization.
	* testsuite/{semko,transko}/*.stp: Added "-u" or other code to many
	tests to check bad code without optimizer elision.
	* testsuite/semok/optimize.stp: New test.

	* elaborate.cxx (unresolved, invalid, mismatch): Standardize error
	message wording.
	* stapfuncs.5.in: Tweak print/printf docs.
	* tapset/logging.stp: Remove redundant "print" auxiliary function,
	since it's a translator built-in.
	* testsuite/transok/five.stp: Extend test.
	* translate.cxx (emit_symbol_data): Put symbol table into a separate
	temporary header file, to make "-p3" output easier on the eyes.
	* buildrun.cxx (compile_pass): Eliminate test-mode support throughout.
	* main.cxx, session.h, translate.cxx: Ditto.
	* main.cxx (main): For last-pass=2 runs, print post-optimization ASTs.

1 files changed, 12 insertions, 23 deletions

diff --git a/parse.cxx b/parse.cxx
index 56495a56..1d8e2947 100644
--- a/parse.cxx
+++ b/parse.cxx
@@ -1232,11 +1232,8 @@ parser::parse_for_loop ()
   t = peek ();
   if (t && t->type == tok_operator && t->content == ";")
     {
-      literal_number* l = new literal_number(0);
-      expr_statement* es = new expr_statement;
-      es->value = l;
-      s->init = es;
-      es->value->tok = es->tok = next ();
+      s->init = 0;
+      next ();
     }
   else
     {
@@ -1266,11 +1263,8 @@ parser::parse_for_loop ()
   t = peek ();
   if (t && t->type == tok_operator && t->content == ")")
     {
-      literal_number* l = new literal_number(2);
-      expr_statement* es = new expr_statement;
-      es->value = l;
-      s->incr = es;
-      es->value->tok = es->tok = next ();
+      s->incr = 0;
+      next ();
     }
   else
     {
@@ -1301,23 +1295,12 @@ parser::parse_while_loop ()
     throw parse_error ("expected '('");
 
   // dummy init and incr fields
-  literal_number* l = new literal_number(0);
-  expr_statement* es = new expr_statement;
-  es->value = l;
-  s->init = es;
-  es->value->tok = es->tok = t;
-
-  l = new literal_number(2);
-  es = new expr_statement;
-  es->value = l;
-  s->incr = es;
-  es->value->tok = es->tok = t;
-
+  s->init = 0;
+  s->incr = 0;
 
   // condition
   s->cond = parse_expression ();
 
-
   t = next ();
   if (! (t->type == tok_operator && t->content == ")"))
     throw parse_error ("expected ')'");
@@ -1808,6 +1791,12 @@ parser::parse_unary ()
 expression*
 parser::parse_crement () // as in "increment" / "decrement"
 {
+  // NB: Ideally, we'd parse only a symbol as an operand to the
+  // *crement operators, instead of a general expression value.  We'd
+  // need more complex lookahead code to tell apart the postfix cases.
+  // So we just punt, and leave it to pass-3 to signal errors on
+  // cases like "4++".
+
   const token* t = peek ();
   if (t && t->type == tok_operator 
       && (t->content == "++" || t->content == "--"))