Sat Mar 23 17:52:49 1996 Ulrich Drepper <drepper@gnu.ai.mit.edu>
authorroland <roland>
Thu, 28 Mar 1996 07:44:19 +0000 (07:44 +0000)
committerroland <roland>
Thu, 28 Mar 1996 07:44:19 +0000 (07:44 +0000)
* locale/weight.h: Functions to access collation tables.

locale/weight.h [new file with mode: 0644]

diff --git a/locale/weight.h b/locale/weight.h
new file mode 100644 (file)
index 0000000..5a9bcd7
--- /dev/null
@@ -0,0 +1,171 @@
+/* Copyright (C) 1996 Free Software Foundation, Inc.
+This file is part of the GNU C Library.
+Written by Ulrich Drepper, <drepper@gnu.ai.mit.edu>.
+
+The GNU C Library is free software; you can redistribute it and/or
+modify it under the terms of the GNU Library General Public License as
+published by the Free Software Foundation; either version 2 of the
+License, or (at your option) any later version.
+
+The GNU C Library is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+Library General Public License for more details.
+
+You should have received a copy of the GNU Library General Public
+License along with the GNU C Library; see the file COPYING.LIB.  If
+not, write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+Boston, MA 02111-1307, USA.  */
+
+#include <alloca.h>
+
+
+#ifndef STRING_TYPE
+# error STRING_TYPE not defined
+#endif
+
+#ifndef USTRING_TYPE
+# error USTRING_TYPE not defined
+#endif
+
+typedef struct weight_t
+{
+  struct weight_t *prev;
+  struct weight_t *next;
+  struct data_pair {
+    size_t number;
+    u32_t *value;
+  } data[0];
+} weight_t;
+
+
+/* The following five macros grant access to the non-byte order
+   dependend values in the collate locale file.  */
+#define collate_nrules \
+  (_NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES))
+#define collate_hash_size \
+  (_NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_HASH_SIZE))
+#define collate_hash_layers \
+  (_NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_HASH_LAYERS))
+#define collate_undefined \
+  (_NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_UNDEFINED))
+#define collate_rules \
+  (_NL_CURRENT (LC_COLLATE, _NL_COLLATE_RULES))
+
+
+static __inline int get_weight (const STRING_TYPE **str, weight_t *result);
+static __inline int
+get_weight (const STRING_TYPE **str, weight_t *result)
+{
+  unsigned int ch = *((USTRING_TYPE *) (*str))++;
+  size_t slot;
+
+  if (sizeof (STRING_TYPE) == 1)
+    slot = ch * (collate_nrules + 1);
+  else
+    {
+      const size_t level_size = collate_hash_size * (collate_nrules + 1);
+      size_t level;
+
+      slot = (ch * (collate_nrules + 1)) % collate_hash_size;
+
+      level = 0;
+      while (__collate_table[slot] != (u32_t) ch)
+       {
+         if (__collate_table[slot + 1] == 0
+             || ++level >= collate_hash_layers)
+           {
+             size_t idx = collate_undefined;
+             size_t cnt;
+
+             for (cnt = 0; cnt < collate_nrules; ++cnt)
+               {
+                 result->data[cnt].number = __collate_extra[idx++];
+                 result->data[cnt].value = &__collate_extra[idx];
+                 idx += result->data[cnt].number;
+               }
+             return 0;
+           }
+         slot += level_size;
+       }
+    }
+
+  if (__collate_table[slot + 1] != FORWARD_CHAR)
+    {
+      /* We have a simple form.  One one value for each weight.  */
+      size_t cnt;
+
+      for (cnt = 0; cnt < collate_nrules; ++cnt)
+       {
+         result->data[cnt].number = 1;
+         result->data[cnt].value = &__collate_table[slot + 1 + cnt];
+       }
+      return ch == 0;
+    }
+
+  /* We now look for any collation element which starts with CH.
+     There might none, but the last list member is a catch-all case
+     because it is simple the character CH.  The value of this entry
+     might be the same as UNDEFINED.  */
+  slot = __collate_table[slot + 2];
+
+  while (1)
+    {
+      size_t idx;
+
+      /* This is a comparison between a u32_t array (aka wchar_t) and
+        an 8-bit string.  */
+      for (idx = 0; __collate_extra[slot + 2 + idx] != 0; ++idx)
+       if (__collate_extra[slot + 2 + idx] != (u32_t) str[idx])
+         break;
+
+      /* When the loop finished with all character of the collation
+        element used, we found the longest prefix.  */
+      if (__collate_extra[slot + 2 + idx] == 0)
+       {
+         size_t cnt;
+
+         idx += slot + 3;
+         for (cnt = 0; cnt < collate_nrules; ++cnt)
+           {
+             result->data[cnt].number = __collate_extra[idx++];
+             result->data[cnt].value = &__collate_extra[idx];
+             idx += result->data[cnt].number;
+           }
+         return 0;
+       }
+
+      /* To next entry in list.  */
+      slot += __collate_extra[slot];
+    }
+  /* NOTREACHED */
+  return 0;    /* To calm down gcc.  */
+}
+
+
+/* To process a string efficiently we retrieve all information about
+   the string at once.  The following macro constructs a double linked
+   list of this information.  It is a macro because we use `alloca'
+   and we use a double linked list because of the backward collation
+   order.  */
+#define get_string(str, forw, backw)                                         \
+  do                                                                         \
+    {                                                                        \
+      weight_t *newp;                                                        \
+      do                                                                     \
+       {                                                                     \
+         newp = (weight_t *) alloca (sizeof (weight_t)                       \
+                                     + (collate_nrules                       \
+                                        * sizeof (struct data_pair)));       \
+                                                                             \
+         newp->prev = backw;                                                 \
+         if (backw == NULL)                                                  \
+           forw = newp;                                                      \
+         else                                                                \
+           backw->next = newp;                                               \
+         newp->next = NULL;                                                  \
+         backw = newp;                                                       \
+       }                                                                     \
+      while (get_weight (&str, newp) == 0);                                  \
+    }                                                                        \
+  while (0)