* string/Makefile (distribute): Add str-two-way.h.

2008-03-29  Eric Blake	<ebb9@byu.net>

	Rewrite string searches to O(n) rather than O(n^2).
	* string/str-two-way.h: New file.  For linear fixed-allocation
	string searching.
	* string/memmem.c: New implementation.
	* string/strstr.c: New implementation.
	* string/strcasestr.c: New implementation.

	* sysdeps/posix/getaddrinfo.c (getaddrinfo): Call _res_hconf_init

ChangeLog

@@ -1,3 +1,16 @@
+2008-05-14  Ulrich Drepper  <drepper@redhat.com>
+
+	* string/Makefile (distribute): Add str-two-way.h.
+
+2008-03-29  Eric Blake	<ebb9@byu.net>
+
+	Rewrite string searches to O(n) rather than O(n^2).
+	* string/str-two-way.h: New file.  For linear fixed-allocation
+	string searching.
+	* string/memmem.c: New implementation.
+	* string/strstr.c: New implementation.
+	* string/strcasestr.c: New implementation.
+
 2008-04-11  Paolo Bonzini  <bonzini@gnu.org>
 
 	* posix/regcomp.c (optimize_utf8): Add a note on why we test
@@ -47,7 +60,7 @@
 	(match_prefix): Don't treat IPv4 loopback address special when
 	converting to v4 mapped addressed.
 
-	* sysdeps/posix/getaddrinfo.c (getaddrinfo): Add _res_hconf_init
+	* sysdeps/posix/getaddrinfo.c (getaddrinfo): Call _res_hconf_init
 	if necessary.
 	* posix/tst-rfc3484.c: Add dummy definition of _res_hconf_init.
 	* posix/tst-rfc3484-2.c: Likewise.

NEWS

@@ -15,6 +15,13 @@ Version 2.9
 
 * getaddrinfo now handles DCCP and UDPlite.
   Implemented by Ulrich Drepper.
+
+* New fixed-size conversion macros: htobe16, htole16, be16toh, le16toh,
+  htobe32, htole32, be32toh, le32toh, htobe64, htole64, be64toh, le64toh.
+  Implemented by Ulrich Drepper.
+
+* New implementation of memmem, strstr, and strcasestr which is O(n).
+  Implemented by Eric Blake.
 
 Version 2.8
 

posix/regcomp.c

@@ -1038,7 +1038,9 @@ 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;
@@ -1318,6 +1320,8 @@ calc_first (void *extra, bin_tree_t *node)
       node->node_idx = re_dfa_add_node (dfa, node->token);
       if (BE (node->node_idx == -1, 0))
         return REG_ESPACE;
+      if (node->token.type == ANCHOR)
+        dfa->nodes[node->node_idx].constraint = node->token.opr.ctx_type;
     }
   return REG_NOERROR;
 }
@@ -1446,22 +1450,17 @@ duplicate_node_closure (re_dfa_t *dfa, int top_org_node, int 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)
+	  /* If the node is root_node itself, it means the epsilon clsoure
+	     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.  */
-		  ret = re_node_set_insert (dfa->edests + clone_node,
-					    org_dest);
-		  if (BE (ret < 0, 0))
-		    return REG_ESPACE;
-		  break;
-		}
-	      constraint |= dfa->nodes[org_node].opr.ctx_type;
+	      ret = re_node_set_insert (dfa->edests + clone_node, org_dest);
+	      if (BE (ret < 0, 0))
+		return REG_ESPACE;
+	      break;
 	    }
+	  /* In case of the node has another constraint, add it.  */
+	  constraint |= dfa->nodes[org_node].constraint;
 	  clone_dest = duplicate_node (dfa, org_dest, constraint);
 	  if (BE (clone_dest == -1, 0))
 	    return REG_ESPACE;
@@ -1479,7 +1478,7 @@ duplicate_node_closure (re_dfa_t *dfa, int top_org_node, int top_clone_node,
 	  clone_dest = search_duplicated_node (dfa, org_dest, constraint);
 	  if (clone_dest == -1)
 	    {
-	      /* 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 == -1, 0))
@@ -1494,7 +1493,7 @@ duplicate_node_closure (re_dfa_t *dfa, int top_org_node, int top_clone_node,
 	    }
 	  else
 	    {
-	      /* There are a duplicated node which satisfy the constraint,
+	      /* There is a duplicated node which satisfies the constraint,
 		 use it to avoid infinite loop.  */
 	      ret = re_node_set_insert (dfa->edests + clone_node, clone_dest);
 	      if (BE (ret < 0, 0))
@@ -1543,8 +1542,7 @@ duplicate_node (re_dfa_t *dfa, int org_idx, unsigned int constraint)
   if (BE (dup_idx != -1, 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.  */
@@ -1624,7 +1622,6 @@ static reg_errcode_t
 calc_eclosure_iter (re_node_set *new_set, re_dfa_t *dfa, int node, int root)
 {
   reg_errcode_t err;
-  unsigned int constraint;
   int i, incomplete;
   re_node_set eclosure;
   incomplete = 0;
@@ -1636,15 +1633,14 @@ calc_eclosure_iter (re_node_set *new_set, re_dfa_t *dfa, int node, int root)
      We reference this value to avoid infinite loop.  */
   dfa->eclosures[node].nelem = -1;
 
-  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;
     }

posix/regex_internal.c

@@ -1665,11 +1665,9 @@ 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;
@@ -1682,8 +1680,6 @@ 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)
 	{

string/Makefile

@@ -55,7 +55,8 @@ tests		:= tester inl-tester noinl-tester testcopy test-ffs	\
 		   tst-strtok tst-strxfrm bug-strcoll1 tst-strfry	\
 		   bug-strtok1 $(addprefix test-,$(strop-tests))	\
 		   bug-envz1 tst-strxfrm2 tst-endian
-distribute	:= memcopy.h pagecopy.h tst-svc.expect test-string.h
+distribute	:= memcopy.h pagecopy.h tst-svc.expect test-string.h	\
+		   str-two-way.h
 
 
 include ../Rules

string/memmem.c

@@ -1,4 +1,4 @@
-/* Copyright (C) 1991,92,93,94,96,97,98,2000,2004 Free Software Foundation, Inc.
+/* Copyright (C) 1991,92,93,94,96,97,98,2000,2004,2008 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -16,26 +16,36 @@
    Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
    02111-1307 USA.  */
 
-#include <stddef.h>
+/* This particular implementation was written by Eric Blake, 2008.  */
+
+#ifndef _LIBC
+# include <config.h>
+#endif
+
+/* Specification of memmem.  */
 #include <string.h>
 
 #ifndef _LIBC
 # define __builtin_expect(expr, val)   (expr)
 #endif
 
+#define RETURN_TYPE void *
+#define AVAILABLE(h, h_l, j, n_l) ((j) <= (h_l) - (n_l))
+#include "str-two-way.h"
+
 #undef memmem
 
-/* Return the first occurrence of NEEDLE in HAYSTACK.  */
+/* Return the first occurrence of NEEDLE in HAYSTACK.  Return HAYSTACK
+   if NEEDLE_LEN is 0, otherwise NULL if NEEDLE is not found in
+   HAYSTACK.  */
 void *
-memmem (haystack, haystack_len, needle, needle_len)
-     const void *haystack;
-     size_t haystack_len;
-     const void *needle;
-     size_t needle_len;
+memmem (const void *haystack_start, size_t haystack_len,
+	const void *needle_start, size_t needle_len)
 {
-  const char *begin;
-  const char *const last_possible
-    = (const char *) haystack + haystack_len - needle_len;
+  /* Abstract memory is considered to be an array of 'unsigned char' values,
+     not an array of 'char' values.  See ISO C 99 section 6.2.6.1.  */
+  const unsigned char *haystack = (const unsigned char *) haystack_start;
+  const unsigned char *needle = (const unsigned char *) needle_start;
 
   if (needle_len == 0)
     /* The first occurrence of the empty string is deemed to occur at
@@ -47,12 +57,22 @@ memmem (haystack, haystack_len, needle, needle_len)
   if (__builtin_expect (haystack_len < needle_len, 0))
     return NULL;
 
-  for (begin = (const char *) haystack; begin <= last_possible; ++begin)
-    if (begin[0] == ((const char *) needle)[0] &&
-	!memcmp ((const void *) &begin[1],
-		 (const void *) ((const char *) needle + 1),
-		 needle_len - 1))
-      return (void *) begin;
-
-  return NULL;
+  /* Use optimizations in memchr when possible, to reduce the search
+     size of haystack using a linear algorithm with a smaller
+     coefficient.  However, avoid memchr for long needles, since we
+     can often achieve sublinear performance.  */
+  if (needle_len < LONG_NEEDLE_THRESHOLD)
+    {
+      haystack = memchr (haystack, *needle, haystack_len);
+      if (!haystack || __builtin_expect (needle_len == 1, 0))
+	return (void *) haystack;
+      haystack_len -= haystack - (const unsigned char *) haystack_start;
+      if (haystack_len < needle_len)
+	return NULL;
+      return two_way_short_needle (haystack, haystack_len, needle, needle_len);
+    }
+  else
+    return two_way_long_needle (haystack, haystack_len, needle, needle_len);
 }
+
+#undef LONG_NEEDLE_THRESHOLD