optimize double anchors such as ^$

2008-05-15  Paolo Bonzini  <bonzini@gnu.org>

	* lib/regcomp.c (optimize_utf8): Add a note on why we test
	opr.ctx_type.
	(calc_first): Initialize constraint field.
	(duplicate_node_closure): Use it instead of special casing ANCHORS.
	Fix grammar.
	(duplicate_node): Merge constraint field for all node types.
	(calc_eclosure_iter): Look at constraint field for all node types.
	* lib/regex_internal.c (create_cd_newstate): Don't look at
	opr.ctx_type.

diff --git a/ChangeLog b/ChangeLog
index d53d6dd4df2144dd70dc98fa34c19a4fb248c2b2..6a2635b088b953ae922e51677156cc4cb68fcf4d 100644 (file)
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,15 @@
+2008-05-15  Paolo Bonzini  <bonzini@gnu.org>
+
+       * lib/regcomp.c (optimize_utf8): Add a note on why we test
+       opr.ctx_type.
+       (calc_first): Initialize constraint field.
+       (duplicate_node_closure): Use it instead of special casing ANCHORS.
+       Fix grammar.
+       (duplicate_node): Merge constraint field for all node types.
+       (calc_eclosure_iter): Look at constraint field for all node types.
+       * lib/regex_internal.c (create_cd_newstate): Don't look at
+       opr.ctx_type.
+
 2008-05-14  Bruno Haible  <bruno@clisp.org>
 
        Help GCC to do better code generation.

diff --git a/lib/regcomp.c b/lib/regcomp.c
index a02418df7baf9799b289e3b69fc791d9686871a1..dc15666503622c57f1821033805f472dec13899e 100644 (file)
--- a/lib/regcomp.c
+++ b/lib/regcomp.c
@@ -1057,7 +1057,9 @@ optimize_utf8 (re_dfa_t *dfa)
          case BUF_LAST:
            break;
          default:
-           /* Word anchors etc. cannot be handled.  */
+           /* Word anchors etc. cannot be handled.  It's okay to test
+              opr.ctx_type since constraints (for all DFA nodes) are
+              created by ORing one or more opr.ctx_type values.  */
            return;
          }
        break;
@@ -1344,6 +1346,8 @@ calc_first (void *extra, bin_tree_t *node)
       node->node_idx = re_dfa_add_node (dfa, node->token);
       if (BE (node->node_idx == REG_MISSING, 0))
         return REG_ESPACE;
+      if (node->token.type == ANCHOR)
+        dfa->nodes[node->node_idx].constraint = node->token.opr.ctx_type;
     }
   return REG_NOERROR;
 }
@@ -1473,21 +1477,18 @@ duplicate_node_closure (re_dfa_t *dfa, Idx top_org_node, Idx top_clone_node,
             destination.  */
          org_dest = dfa->edests[org_node].elems[0];
          re_node_set_empty (dfa->edests + clone_node);
-         if (dfa->nodes[org_node].type == ANCHOR)
+         clone_dest = search_duplicated_node (dfa, org_dest, constraint);
+         /* If the node is root_node itself, it means the epsilon closure
+            has a loop.  Then tie it to the destination of the root_node.  */
+         if (org_node == root_node && clone_node != org_node)
            {
-             /* In case of the node has another constraint, append it.  */
-             if (org_node == root_node && clone_node != org_node)
-               {
-                 /* ...but if the node is root_node itself, it means the
-                    epsilon closure have a loop, then tie it to the
-                    destination of the root_node.  */
-                 ok = re_node_set_insert (dfa->edests + clone_node, org_dest);
-                 if (BE (! ok, 0))
-                   return REG_ESPACE;
-                 break;
-               }
-             constraint |= dfa->nodes[org_node].opr.ctx_type;
+             ok = re_node_set_insert (dfa->edests + clone_node, org_dest);
+             if (BE (! ok, 0))
+               return REG_ESPACE;
+             break;
            }
+         /* In case the node has another constraint, append it.  */
+         constraint |= dfa->nodes[org_node].constraint;
          clone_dest = duplicate_node (dfa, org_dest, constraint);
          if (BE (clone_dest == REG_MISSING, 0))
            return REG_ESPACE;
@@ -1505,7 +1506,7 @@ duplicate_node_closure (re_dfa_t *dfa, Idx top_org_node, Idx top_clone_node,
          clone_dest = search_duplicated_node (dfa, org_dest, constraint);
          if (clone_dest == REG_MISSING)
            {
-             /* There are no such a duplicated node, create a new one.  */
+             /* There is no such duplicated node, create a new one.  */
              reg_errcode_t err;
              clone_dest = duplicate_node (dfa, org_dest, constraint);
              if (BE (clone_dest == REG_MISSING, 0))
@@ -1520,7 +1521,7 @@ duplicate_node_closure (re_dfa_t *dfa, Idx top_org_node, Idx top_clone_node,
            }
          else
            {
-             /* There are a duplicated node which satisfy the constraint,
+             /* There is a duplicated node which satisfy the constraint,
                 use it to avoid infinite loop.  */
              ok = re_node_set_insert (dfa->edests + clone_node, clone_dest);
              if (BE (! ok, 0))
@@ -1569,8 +1570,7 @@ duplicate_node (re_dfa_t *dfa, Idx org_idx, unsigned int constraint)
   if (BE (dup_idx != REG_MISSING, 1))
     {
       dfa->nodes[dup_idx].constraint = constraint;
-      if (dfa->nodes[org_idx].type == ANCHOR)
-       dfa->nodes[dup_idx].constraint |= dfa->nodes[org_idx].opr.ctx_type;
+      dfa->nodes[dup_idx].constraint |= dfa->nodes[org_idx].constraint;
       dfa->nodes[dup_idx].duplicated = 1;
 
       /* Store the index of the original node.  */
@@ -1652,7 +1652,6 @@ static reg_errcode_t
 calc_eclosure_iter (re_node_set *new_set, re_dfa_t *dfa, Idx node, bool root)
 {
   reg_errcode_t err;
-  unsigned int constraint;
   Idx i;
   bool incomplete;
   bool ok;
@@ -1666,15 +1665,14 @@ calc_eclosure_iter (re_node_set *new_set, re_dfa_t *dfa, Idx node, bool root)
      We reference this value to avoid infinite loop.  */
   dfa->eclosures[node].nelem = REG_MISSING;
 
-  constraint = ((dfa->nodes[node].type == ANCHOR)
-               ? dfa->nodes[node].opr.ctx_type : 0);
-  /* If the current node has constraints, duplicate all nodes.
-     Since they must inherit the constraints.  */
-  if (constraint
+  /* If the current node has constraints, duplicate all nodes
+     since they must inherit the constraints.  */
+  if (dfa->nodes[node].constraint
       && dfa->edests[node].nelem
       && !dfa->nodes[dfa->edests[node].elems[0]].duplicated)
     {
-      err = duplicate_node_closure (dfa, node, node, node, constraint);
+      err = duplicate_node_closure (dfa, node, node, node,
+                                   dfa->nodes[node].constraint);
       if (BE (err != REG_NOERROR, 0))
        return err;
     }

diff --git a/lib/regex_internal.c b/lib/regex_internal.c
index 21298889a3161b644bf9bc653d08149de6fbb53f..7f9a3ae0e5662b3e80da8dd5b98c74e7f9a4d837 100644 (file)
--- a/lib/regex_internal.c
+++ b/lib/regex_internal.c
@@ -1689,11 +1689,9 @@ create_cd_newstate (const re_dfa_t *dfa, const re_node_set *nodes,
 
   for (i = 0 ; i < nodes->nelem ; i++)
     {
-      unsigned int constraint = 0;
       re_token_t *node = dfa->nodes + nodes->elems[i];
       re_token_type_t type = node->type;
-      if (node->constraint)
-       constraint = node->constraint;
+      unsigned int constraint = node->constraint;
 
       if (type == CHARACTER && !constraint)
        continue;
@@ -1706,8 +1704,6 @@ create_cd_newstate (const re_dfa_t *dfa, const re_node_set *nodes,
        newstate->halt = 1;
       else if (type == OP_BACK_REF)
        newstate->has_backref = 1;
-      else if (type == ANCHOR)
-       constraint = node->opr.ctx_type;
 
       if (constraint)
        {