# include <config.h>
#endif
+#include <errno.h>
#include <error.h>
#include <stdlib.h>
+#include <wchar.h>
+#include <sys/param.h>
#include "charmap.h"
#include "localeinfo.h"
#include "linereader.h"
#include "locfile.h"
#include "localedef.h"
+#include "elem-hash.h"
/* Uncomment the following line in the production version. */
/* #define NDEBUG 1 */
struct element_t *last;
/* These are the rules for this section. */
enum coll_sort_rule *rules;
+ /* Index of the rule set in the appropriate section of the output file. */
+ int ruleidx;
};
struct element_t;
/* Data type for collating element. */
struct element_t
{
+ const char *name;
+
const char *mbs;
+ size_t nmbs;
const uint32_t *wcs;
- int order;
+ size_t nwcs;
+ int *mborder;
+ int wcorder;
+
+ /* The following is a bit mask which bits are set if this element is
+ used in the appropriate level. Interesting for the singlebyte
+ weight computation.
+
+ XXX The type here restricts the number of levels to 32. It could
+ we changed if necessary but I doubt this is necessary. */
+ unsigned int used_in_level;
struct element_list_t *weights;
+ /* Nonzero if this is a real character definition. */
+ int is_character;
+
/* Where does the definition come from. */
const char *file;
size_t line;
/* Predecessor and successor in the order list. */
struct element_t *last;
struct element_t *next;
+
+ /* Next element in multibyte output list. */
+ struct element_t *mbnext;
+
+ /* Next element in wide character output list. */
+ struct element_t *wcnext;
};
+/* Special element value. */
+#define ELEMENT_ELLIPSIS2 ((struct element_t *) 1)
+#define ELEMENT_ELLIPSIS3 ((struct element_t *) 2)
+#define ELEMENT_ELLIPSIS4 ((struct element_t *) 3)
+
/* Data type for collating symbol. */
struct symbol_t
{
/* To make handling of errors easier we have another section. */
struct section_list error_section;
- /* Number of sorting rules given in order_start line. */
- uint32_t nrules;
-
/* Start of the order list. */
struct element_t *start;
/* This is the cursor for `reorder_after' insertions. */
struct element_t *cursor;
- /* Remember whether last weight was an ellipsis. */
- int was_ellipsis;
+ /* This value is used when handling ellipsis. */
+ struct element_t ellipsis_weight;
/* Known collating elements. */
hash_table elem_table;
that the definitions from more than one input file contains information.
Therefore we keep all relevant input in a list. */
struct locale_collate_t *next;
+
+ /* Arrays with heads of the list for each of the leading bytes in
+ the multibyte sequences. */
+ struct element_t *mbheads[256];
+
+ /* Table size of wide character hash table. */
+ size_t plane_size;
+ size_t plane_cnt;
+
+ /* Arrays with heads of the list for each of the leading bytes in
+ the multibyte sequences. */
+ struct element_t **wcheads;
};
/* We have a few global variables which are used for reading all
LC_COLLATE category descriptions in all files. */
-static int nrules;
+static uint32_t nrules;
+
+
+/* These are definitions used by some of the functions for handling
+ UTF-8 encoding below. */
+static const uint32_t encoding_mask[] =
+{
+ ~0x7ff, ~0xffff, ~0x1fffff, ~0x3ffffff
+};
+
+static const unsigned char encoding_byte[] =
+{
+ 0xc0, 0xe0, 0xf0, 0xf8, 0xfc
+};
+
+
+/* We need UTF-8 encoding of numbers. */
+static inline int
+utf8_encode (char *buf, int val)
+{
+ char *startp = buf;
+ int retval;
+
+ if (val < 0x80)
+ {
+ *buf++ = (char) val;
+ retval = 1;
+ }
+ else
+ {
+ int step;
+
+ for (step = 2; step < 6; ++step)
+ if ((val & encoding_mask[step - 2]) == 0)
+ break;
+ retval = step;
+
+ *buf = encoding_byte[step - 2];
+ --step;
+ do
+ {
+ buf[step] = 0x80 | (val & 0x3f);
+ val >>= 6;
+ }
+ while (--step > 0);
+ *buf |= val;
+ }
+
+ return buf - startp;
+}
static struct section_list *
static struct element_t *
-new_element (struct locale_collate_t *collate, const char *mbs,
- size_t len, const uint32_t *wcs)
+new_element (struct locale_collate_t *collate, const char *mbs, size_t mbslen,
+ const uint32_t *wcs, const char *name, size_t namelen,
+ int is_character)
{
struct element_t *newp;
newp = (struct element_t *) obstack_alloc (&collate->mempool,
sizeof (*newp));
- newp->mbs = obstack_copy0 (&collate->mempool, mbs, len);
- newp->wcs = wcs;
- newp->order = 0;
+ newp->name = name == NULL ? NULL : obstack_copy0 (&collate->mempool,
+ name, namelen);
+ if (mbs != NULL)
+ {
+ newp->mbs = obstack_copy0 (&collate->mempool, mbs, mbslen);
+ newp->nmbs = mbslen;
+ }
+ else
+ {
+ newp->mbs = NULL;
+ newp->nmbs = 0;
+ }
+ if (wcs != NULL)
+ {
+ size_t nwcs = wcslen ((wchar_t *) wcs);
+ uint32_t zero = 0;
+ obstack_grow (&collate->mempool, wcs, nwcs * sizeof (uint32_t));
+ obstack_grow (&collate->mempool, &zero, sizeof (uint32_t));
+ newp->wcs = (uint32_t *) obstack_finish (&collate->mempool);
+ newp->nwcs = nwcs;
+ }
+ else
+ {
+ newp->wcs = NULL;
+ newp->nwcs = 0;
+ }
+ newp->mborder = NULL;
+ newp->wcorder = 0;
+ newp->used_in_level = 0;
+ newp->is_character = is_character;
/* Will be allocated later. */
newp->weights = NULL;
newp->file = NULL;
newp->line = 0;
- newp->section = NULL;
+ newp->section = collate->current_section;
newp->last = NULL;
newp->next = NULL;
+ newp->mbnext = NULL;
+
return newp;
}
if (! warned)
{
lr_error (ldfile, _("\
-%s: `%s' mentioned twice in definition of weight %d"),
+%s: `%s' mentioned more than once in definition of weight %d"),
"LC_COLLATE", "forward", cnt + 1);
}
}
if (! warned)
{
lr_error (ldfile, _("\
-%s: `%s' mentioned twice in definition of weight %d"),
+%s: `%s' mentioned more than once in definition of weight %d"),
"LC_COLLATE", "backward", cnt + 1);
}
}
if (! warned)
{
lr_error (ldfile, _("\
-%s: `%s' mentioned twice in definition of weight %d in category `%s'"),
+%s: `%s' mentioned more than once in definition of weight %d"),
"LC_COLLATE", "position", cnt + 1);
}
}
/* See whether we have to increment the counter. */
if (arg->tok != tok_comma && rules[cnt] != 0)
- ++cnt;
+ {
+ /* Add the default `forward' if we have seen only `position'. */
+ if (rules[cnt] == sort_position)
+ rules[cnt] = sort_position | sort_forward;
+
+ ++cnt;
+ }
if (arg->tok == tok_eof || arg->tok == tok_eol)
/* End of line or file, so we exit the loop. */
result = sym->order;
if (result == NULL)
- result = sym->order = new_element (collate, str, len, NULL);
+ result = sym->order = new_element (collate, NULL, 0, NULL,
+ NULL, 0, 0);
}
else if (find_entry (&collate->elem_table, str, len,
(void **) &result) != 0)
{
- /* It's also no collation element. So it is an element defined
- later. */
- result = new_element (collate, str, len, wcstr);
+ /* It's also no collation element. So it is a character
+ element defined later. */
+ result = new_element (collate, NULL, 0, NULL, str, len, 1);
if (result != NULL)
/* Insert it into the sequence table. */
insert_entry (&collate->seq_table, str, len, result);
}
+static void
+unlink_element (struct locale_collate_t *collate)
+{
+ if (collate->cursor == collate->start)
+ {
+ assert (collate->cursor->next == NULL);
+ assert (collate->cursor->last == NULL);
+ collate->cursor = NULL;
+ }
+ else
+ {
+ if (collate->cursor->next != NULL)
+ collate->cursor->next->last = collate->cursor->last;
+ if (collate->cursor->last != NULL)
+ collate->cursor->last->next = collate->cursor->next;
+ collate->cursor = collate->cursor->last;
+ }
+}
+
+
static void
insert_weights (struct linereader *ldfile, struct element_t *elem,
struct charmap_t *charmap, struct repertoire_t *repertoire,
- struct locale_collate_t *collate)
+ struct locale_collate_t *collate, enum token_t ellipsis)
{
int weight_cnt;
struct token *arg;
elem->line = ldfile->lineno;
elem->last = collate->cursor;
elem->next = collate->cursor ? collate->cursor->next : NULL;
+ elem->section = collate->current_section;
+ if (collate->cursor != NULL)
+ collate->cursor->next = elem;
+ if (collate->start == NULL)
+ {
+ assert (collate->cursor == NULL);
+ collate->start = elem;
+ }
elem->weights = (struct element_list_t *)
obstack_alloc (&collate->mempool, nrules * sizeof (struct element_list_t));
memset (elem->weights, '\0', nrules * sizeof (struct element_list_t));
elem->weights[weight_cnt].w = (struct element_t **)
obstack_alloc (&collate->mempool, sizeof (struct element_t *));
elem->weights[weight_cnt].w[0] = NULL;
- elem->weights[weight_cnt].cnt = 0;
+ elem->weights[weight_cnt].cnt = 1;
}
else if (arg->tok == tok_bsymbol)
{
const char *cp = arg->val.str.startmb;
int cnt = 0;
struct element_t *charelem;
- void *base = obstack_base (&collate->mempool);
+ struct element_t **weights = NULL;
+ int max = 0;
if (*cp == '\0')
{
if (*cp == '<')
{
/* Ahh, it's a bsymbol. That's what we want. */
- const char *startp = cp;
+ const char *startp = ++cp;
- while (*++cp != '>')
+ while (*cp != '>')
{
if (*cp == ldfile->escape_char)
++cp;
if (*cp == '\0')
- {
- /* It's a syntax error. */
- obstack_free (&collate->mempool, base);
- goto syntax;
- }
+ /* It's a syntax error. */
+ goto syntax;
+
+ ++cp;
}
charelem = find_element (ldfile, collate, startp,
what this means. We interpret all characters in the
string as if that would be bsymbols. Otherwise we
would have to match back to bsymbols somehow and this
- is also not what people normally expect. */
+ is normally not what people normally expect. */
charelem = find_element (ldfile, collate, cp++, 1, NULL);
}
}
/* Add the pointer. */
- obstack_ptr_grow (&collate->mempool, charelem);
- ++cnt;
+ if (cnt >= max)
+ {
+ struct element_t **newp;
+ max += 10;
+ newp = (struct element_t **)
+ alloca (max * sizeof (struct element_t *));
+ memcpy (newp, weights, cnt * sizeof (struct element_t *));
+ weights = newp;
+ }
+ weights[cnt++] = charelem;
}
while (*cp != '\0');
/* Now store the information. */
elem->weights[weight_cnt].w = (struct element_t **)
- obstack_finish (&collate->mempool);
+ obstack_alloc (&collate->mempool,
+ cnt * sizeof (struct element_t *));
+ memcpy (elem->weights[weight_cnt].w, weights,
+ cnt * sizeof (struct element_t *));
elem->weights[weight_cnt].cnt = cnt;
/* We don't need the string anymore. */
free (arg->val.str.startmb);
}
+ else if (ellipsis != tok_none
+ && (arg->tok == tok_ellipsis2
+ || arg->tok == tok_ellipsis3
+ || arg->tok == tok_ellipsis4))
+ {
+ /* It must be the same ellipsis as used in the initial column. */
+ if (arg->tok != ellipsis)
+ lr_error (ldfile, _("\
+%s: weights must use the same ellipsis symbol as the name"),
+ "LC_COLLATE");
+
+ /* The weight for this level has to be ignored. We use the
+ null pointer to indicate this. */
+ elem->weights[weight_cnt].w = (struct element_t **)
+ obstack_alloc (&collate->mempool, sizeof (struct element_t *));
+ elem->weights[weight_cnt].w[0] = ELEMENT_ELLIPSIS2;
+ elem->weights[weight_cnt].cnt = 1;
+ }
else
{
syntax:
}
-static void
+static int
insert_value (struct linereader *ldfile, struct token *arg,
struct charmap_t *charmap, struct repertoire_t *repertoire,
struct locale_collate_t *collate)
uint32_t wc;
struct element_t *elem = NULL;
- /* First determine the wide character. There must be such a value,
- otherwise we ignore it (if it is no collatio symbol or element). */
- wc = repertoire_find_value (repertoire, arg->val.str.startmb,
- arg->val.str.lenmb);
-
/* Try to find the character in the charmap. */
seq = charmap_find_value (charmap, arg->val.str.startmb, arg->val.str.lenmb);
+ /* Determine the wide character. */
+ if (seq == NULL || seq->ucs4 == UNINITIALIZED_CHAR_VALUE)
+ {
+ wc = repertoire_find_value (repertoire, arg->val.str.startmb,
+ arg->val.str.lenmb);
+ if (seq != NULL)
+ seq->ucs4 = wc;
+ }
+ else
+ wc = seq->ucs4;
+
if (wc == ILLEGAL_CHAR_VALUE && seq == NULL)
{
/* It's no character, so look through the collation elements and
elem = sym->order;
if (elem == NULL)
- elem = sym->order = new_element (collate, arg->val.str.startmb,
- arg->val.str.lenmb,
- arg->val.str.startwc);
+ elem = sym->order = new_element (collate, NULL, 0, NULL, NULL, 0,
+ 0);
}
else if (find_entry (&collate->elem_table, arg->val.str.startmb,
arg->val.str.lenmb, (void **) &elem) != 0)
{
/* It's also no collation element. Therefore ignore it. */
lr_ignore_rest (ldfile, 0);
- return;
+ return 1;
}
}
else
if (find_entry (&collate->seq_table, arg->val.str.startmb,
arg->val.str.lenmb, (void **) &elem) != 0)
{
+ uint32_t wcs[2] = { wc, 0 };
+
/* We have to allocate an entry. */
- elem = new_element (collate, arg->val.str.startmb,
- arg->val.str.lenmb,
- arg->val.str.startwc);
+ elem = new_element (collate, seq != NULL ? seq->bytes : NULL,
+ seq != NULL ? seq->nbytes : 0,
+ wc == ILLEGAL_CHAR_VALUE ? NULL : wcs,
+ arg->val.str.startmb, arg->val.str.lenmb, 1);
/* And add it to the table. */
if (insert_entry (&collate->seq_table, arg->val.str.startmb,
arg->val.str.lenmb, elem) != 0)
/* This cannot happen. */
- abort ();
+ assert (! "Internal error");
+ }
+ else
+ {
+ /* Maybe the character was used before the definition. In this case
+ we have to insert the byte sequences now. */
+ if (elem->mbs == NULL && seq != NULL)
+ {
+ elem->mbs = obstack_copy0 (&collate->mempool,
+ seq->bytes, seq->nbytes);
+ elem->nmbs = seq->nbytes;
+ }
+
+ if (elem->wcs == NULL && wc != ILLEGAL_CHAR_VALUE)
+ {
+ uint32_t wcs[2] = { wc, 0 };
+
+ elem->wcs = obstack_copy (&collate->mempool, wcs, sizeof (wcs));
+ elem->nwcs = 1;
+ }
}
}
if (elem->next != NULL || (collate->cursor != NULL
&& elem->next == collate->cursor))
{
- lr_error (ldfile, _("order for `%.*s' already defined at %s:%Z"),
- arg->val.str.lenmb, arg->val.str.startmb,
+ lr_error (ldfile, _("order for `%.*s' already defined at %s:%zu"),
+ (int) arg->val.str.lenmb, arg->val.str.startmb,
elem->file, elem->line);
lr_ignore_rest (ldfile, 0);
- return;
+ return 1;
}
- insert_weights (ldfile, elem, charmap, repertoire, collate);
+ insert_weights (ldfile, elem, charmap, repertoire, collate, tok_none);
+
+ return 0;
}
static void
-collate_startup (struct linereader *ldfile, struct localedef_t *locale,
- struct localedef_t *copy_locale, int ignore_content)
+handle_ellipsis (struct linereader *ldfile, struct token *arg,
+ enum token_t ellipsis, struct charmap_t *charmap,
+ struct repertoire_t *repertoire,
+ struct locale_collate_t *collate)
{
- if (!ignore_content)
+ struct element_t *startp;
+ struct element_t *endp;
+
+ /* Unlink the entry added for the ellipsis. */
+ unlink_element (collate);
+ startp = collate->cursor;
+
+ /* Process and add the end-entry. */
+ if (arg != NULL
+ && insert_value (ldfile, arg, charmap, repertoire, collate))
+ /* Something went wrong with inserting the to-value. This means
+ we cannot process the ellipsis. */
+ return;
+
+ /* Reset the cursor. */
+ collate->cursor = startp;
+
+ /* Now we have to handle many different situations:
+ - we have to distinguish between the three different ellipsis forms
+ - the is the ellipsis at the beginning, in the middle, or at the end.
+ */
+ endp = collate->cursor->next;
+ assert (arg == NULL || endp != NULL);
+
+ /* Both, the start and the end symbol, must stand for characters. */
+ if ((startp != NULL && (startp->name == NULL || ! startp->is_character))
+ || (endp != NULL && (endp->name == NULL|| ! endp->is_character)))
{
- struct locale_collate_t *collate;
-
- if (copy_locale == NULL)
- collate = locale->categories[LC_COLLATE].collate =
- (struct locale_collate_t *) xcalloc (1,
- sizeof (struct locale_collate_t));
- else
- collate = locale->categories[LC_COLLATE].collate =
- copy_locale->categories[LC_COLLATE].collate;
-
- /* Init the various data structures. */
- init_hash (&collate->elem_table, 100);
- init_hash (&collate->sym_table, 100);
- init_hash (&collate->seq_table, 500);
- obstack_init (&collate->mempool);
-
- collate->col_weight_max = -1;
+ lr_error (ldfile, _("\
+%s: the start end the end symbol of a range must stand for characters"),
+ "LC_COLLATE");
+ return;
}
- ldfile->translate_strings = 0;
- ldfile->return_widestr = 0;
-}
+ if (ellipsis == tok_ellipsis3)
+ {
+ /* One requirement we make here: the length of the byte
+ sequences for the first and end character must be the same.
+ This is mainly to prevent unwanted effects and this is often
+ not what is wanted. */
+ size_t len = (startp->mbs != NULL ? startp->nmbs
+ : (endp->mbs != NULL ? endp->nmbs : 0));
+ char mbcnt[len + 1];
+ char mbend[len + 1];
+
+ /* Well, this should be caught somewhere else already. Just to
+ make sure. */
+ assert (startp == NULL || startp->wcs == NULL || startp->wcs[1] == 0);
+ assert (endp == NULL || endp->wcs == NULL || endp->wcs[1] == 0);
+
+ if (startp != NULL && endp != NULL
+ && startp->mbs != NULL && endp->mbs != NULL
+ && startp->nmbs != endp->nmbs)
+ {
+ lr_error (ldfile, _("\
+%s: byte sequences of first and last character must have the same length"),
+ "LC_COLLATE");
+ return;
+ }
+
+ /* Determine whether we have to generate multibyte sequences. */
+ if ((startp == NULL || startp->mbs != NULL)
+ && (endp == NULL || endp->mbs != NULL))
+ {
+ int cnt;
+ int ret;
+
+ /* Prepare the beginning byte sequence. This is either from the
+ beginning byte sequence or it is all nulls if it was an
+ initial ellipsis. */
+ if (startp == NULL || startp->mbs == NULL)
+ memset (mbcnt, '\0', len);
+ else
+ {
+ memcpy (mbcnt, startp->mbs, len);
+ /* And increment it so that the value is the first one we will
+ try to insert. */
+ for (cnt = len - 1; cnt >= 0; --cnt)
+ if (++mbcnt[cnt] != '\0')
+ break;
+ }
+ mbcnt[len] = '\0';
-void
-collate_finish (struct localedef_t *locale, struct charmap_t *charmap)
-{
-}
+ /* And the end sequence. */
+ if (endp == NULL || endp->mbs == NULL)
+ memset (mbend, '\0', len);
+ else
+ memcpy (mbend, endp->mbs, len);
+ mbend[len] = '\0';
+ /* Test whether we have a correct range. */
+ ret = memcmp (mbcnt, mbend, len);
+ if (ret >= 0)
+ {
+ if (ret > 0)
+ lr_error (ldfile, _("%s: byte sequence of first character of \
+sequence is not lower than that of the last character"), "LC_COLLATE");
+ return;
+ }
-void
-collate_output (struct localedef_t *locale, struct charmap_t *charmap,
- const char *output_path)
-{
-}
+ /* Generate the byte sequences data. */
+ while (1)
+ {
+ struct charseq *seq;
+ /* Quite a bit of work ahead. We have to find the character
+ definition for the byte sequence and then determine the
+ wide character belonging to it. */
+ seq = charmap_find_symbol (charmap, mbcnt, len);
+ if (seq != NULL)
+ {
+ struct element_t *elem;
+ size_t namelen;
+
+ if (seq->ucs4 == UNINITIALIZED_CHAR_VALUE)
+ seq->ucs4 = repertoire_find_value (repertoire, seq->name,
+ strlen (seq->name));
+
+ /* I don't this this can ever happen. */
+ assert (seq->name != NULL);
+ namelen = strlen (seq->name);
+
+ /* Now we are ready to insert the new value in the
+ sequence. Find out whether the element is
+ already known. */
+ if (find_entry (&collate->seq_table, seq->name, namelen,
+ (void **) &elem) != 0)
+ {
+ uint32_t wcs[2] = { seq->ucs4, 0 };
+
+ /* We have to allocate an entry. */
+ elem = new_element (collate, mbcnt, len,
+ seq->ucs4 == ILLEGAL_CHAR_VALUE
+ ? NULL : wcs, seq->name,
+ namelen, 1);
+
+ /* And add it to the table. */
+ if (insert_entry (&collate->seq_table, seq->name,
+ namelen, elem) != 0)
+ /* This cannot happen. */
+ assert (! "Internal error");
+ }
-void
-collate_read (struct linereader *ldfile, struct localedef_t *result,
- struct charmap_t *charmap, const char *repertoire_name,
- int ignore_content)
-{
- struct repertoire_t *repertoire = NULL;
- struct locale_collate_t *collate;
- struct token *now;
- struct token *arg = NULL;
- enum token_t nowtok;
- int state = 0;
- int was_ellipsis = 0;
- struct localedef_t *copy_locale = NULL;
+ /* Test whether this element is not already in the list. */
+ if (elem->next != NULL || (collate->cursor != NULL
+ && elem->next == collate->cursor))
+ {
+ lr_error (ldfile, _("\
+order for `%.*s' already defined at %s:%zu"),
+ (int) namelen, seq->name,
+ elem->file, elem->line);
+ goto increment;
+ }
- /* Get the repertoire we have to use. */
- if (repertoire_name != NULL)
- repertoire = repertoire_read (repertoire_name);
+ /* Enqueue the new element. */
+ elem->last = collate->cursor;
+ if (collate->cursor != NULL)
+ elem->next = NULL;
+ else
+ {
+ elem->next = collate->cursor->next;
+ elem->last->next = elem;
+ if (elem->next != NULL)
+ elem->next->last = elem;
+ }
+ if (collate->start == NULL)
+ {
+ assert (collate->cursor == NULL);
+ collate->start = elem;
+ }
+ collate->cursor = elem;
+
+ /* Add the weight value. We take them from the
+ `ellipsis_weights' member of `collate'. */
+ elem->weights = (struct element_list_t *)
+ obstack_alloc (&collate->mempool,
+ nrules * sizeof (struct element_list_t));
+ for (cnt = 0; cnt < nrules; ++cnt)
+ if (collate->ellipsis_weight.weights[cnt].cnt == 1
+ && (collate->ellipsis_weight.weights[cnt].w[0]
+ == ELEMENT_ELLIPSIS2))
+ {
+ elem->weights[cnt].w = (struct element_t **)
+ obstack_alloc (&collate->mempool,
+ sizeof (struct element_t *));
+ elem->weights[cnt].w[0] = elem;
+ elem->weights[cnt].cnt = 1;
+ }
+ else
+ {
+ /* Simly use the weight from `ellipsis_weight'. */
+ elem->weights[cnt].w =
+ collate->ellipsis_weight.weights[cnt].w;
+ elem->weights[cnt].cnt =
+ collate->ellipsis_weight.weights[cnt].cnt;
+ }
+ }
- /* The rest of the line containing `LC_COLLATE' must be free. */
- lr_ignore_rest (ldfile, 1);
+ /* Increment for the next round. */
+ increment:
+ for (cnt = len - 1; cnt >= 0; --cnt)
+ if (++mbcnt[cnt] != '\0')
+ break;
- do
- {
- now = lr_token (ldfile, charmap, NULL);
- nowtok = now->tok;
+ /* Find out whether this was all. */
+ if (cnt < 0 || memcmp (mbcnt, mbend, len) >= 0)
+ /* Yep, that's all. */
+ break;
+ }
+ }
}
- while (nowtok == tok_eol);
-
- if (nowtok == tok_copy)
+ else
{
- state = 2;
- now = lr_token (ldfile, charmap, NULL);
- if (now->tok != tok_string)
+ /* For symbolic range we naturally must have a beginning and an
+ end specified by the user. */
+ if (startp == NULL)
+ lr_error (ldfile, _("\
+%s: symbolic range ellipsis must not directly follow `order_start'"),
+ "LC_COLLATE");
+ else if (endp == NULL)
+ lr_error (ldfile, _("\
+%s: symbolic range ellipsis must not be direct followed by `order_end'"),
+ "LC_COLLATE");
+ else
{
- SYNTAX_ERROR (_("%s: syntax error"), "LC_COLLATE");
-
- skip_category:
- do
- now = lr_token (ldfile, charmap, NULL);
- while (now->tok != tok_eof && now->tok != tok_end);
-
- if (now->tok != tok_eof
- || (now = lr_token (ldfile, charmap, NULL), now->tok == tok_eof))
- lr_error (ldfile, _("%s: premature end of file"), "LC_COLLATE");
- else if (now->tok != tok_lc_collate)
+ /* Determine the range. To do so we have to determine the
+ common prefix of the both names and then the numeric
+ values of both ends. */
+ size_t lenfrom = strlen (startp->name);
+ size_t lento = strlen (endp->name);
+ char buf[lento + 1];
+ int preflen = 0;
+ long int from;
+ long int to;
+ char *cp;
+ int base = ellipsis == tok_ellipsis2 ? 16 : 10;
+
+ if (lenfrom != lento)
{
+ invalid_range:
lr_error (ldfile, _("\
-%1$s: definition does not end with `END %1$s'"), "LC_COLLATE");
- lr_ignore_rest (ldfile, 0);
+`%s' and `%.*s' are no valid names for symbolic range"),
+ startp->name, (int) lento, endp->name);
+ return;
}
- else
- lr_ignore_rest (ldfile, 1);
-
- return;
- }
-
- /* Get the locale definition. */
- copy_locale = find_locale (LC_COLLATE, now->val.str.startmb,
- repertoire_name, charmap);
- if ((copy_locale->avail & COLLATE_LOCALE) == 0)
- {
- /* Not yet loaded. So do it now. */
- if (locfile_read (copy_locale, charmap) != 0)
- goto skip_category;
- }
-
- lr_ignore_rest (ldfile, 1);
- now = lr_token (ldfile, charmap, NULL);
- nowtok = now->tok;
- }
+ while (startp->name[preflen] == endp->name[preflen])
+ if (startp->name[preflen] == '\0')
+ /* Nothing to be done. The start and end point are identical
+ and while inserting the end point we have already given
+ the user an error message. */
+ return;
+ else
+ ++preflen;
- /* Prepare the data structures. */
- collate_startup (ldfile, result, copy_locale, ignore_content);
- collate = result->categories[LC_COLLATE].collate;
+ errno = 0;
+ from = strtol (startp->name + preflen, &cp, base);
+ if ((from == UINT_MAX && errno == ERANGE) || *cp != '\0')
+ goto invalid_range;
- while (1)
- {
- /* Of course we don't proceed beyond the end of file. */
- if (nowtok == tok_eof)
- break;
+ errno = 0;
+ to = strtol (endp->name + preflen, &cp, base);
+ if ((to == UINT_MAX && errno == ERANGE) || *cp != '\0')
+ goto invalid_range;
- /* Ingore empty lines. */
- if (nowtok == tok_eol)
- {
- now = lr_token (ldfile, charmap, NULL);
- nowtok = now->tok;
- continue;
- }
+ /* Copy the prefix. */
+ memcpy (buf, startp->name, preflen);
- switch (nowtok)
- {
- case tok_coll_weight_max:
- /* Ignore the rest of the line if we don't need the input of
- this line. */
- if (ignore_content)
+ /* Loop over all values. */
+ for (++from; from < to; ++from)
{
- lr_ignore_rest (ldfile, 0);
- break;
- }
+ struct element_t *elem = NULL;
+ struct charseq *seq;
+ uint32_t wc;
+ int cnt;
- if (state != 0)
- goto err_label;
+ /* Generate the the name. */
+ sprintf (buf + preflen, base == 10 ? "%d" : "%x", from);
- arg = lr_token (ldfile, charmap, NULL);
- if (arg->tok != tok_number)
- goto err_label;
- if (collate->col_weight_max != -1)
- lr_error (ldfile, _("%s: duplicate definition of `%s'"),
- "LC_COLLATE", "col_weight_max");
- else
- collate->col_weight_max = arg->val.num;
- lr_ignore_rest (ldfile, 1);
- break;
+ /* Look whether this name is already defined. */
+ if (find_entry (&collate->seq_table, arg->val.str.startmb,
+ arg->val.str.lenmb, (void **) &elem) == 0)
+ {
+ if (elem->next != NULL || (collate->cursor != NULL
+ && elem->next == collate->cursor))
+ {
+ lr_error (ldfile, _("\
+%s: order for `%.*s' already defined at %s:%zu"),
+ "LC_COLLATE", (int) lenfrom, buf,
+ elem->file, elem->line);
+ continue;
+ }
- case tok_section_symbol:
- /* Ignore the rest of the line if we don't need the input of
- this line. */
- if (ignore_content)
- {
- lr_ignore_rest (ldfile, 0);
- break;
- }
+ if (elem->name == NULL)
+ {
+ lr_error (ldfile, _("%s: `%s' must be a charater"),
+ "LC_COLLATE", buf);
+ continue;
+ }
+ }
- if (state != 0)
- goto err_label;
+ if (elem == NULL || (elem->mbs == NULL && elem->wcs == NULL))
+ {
+ /* Search for a character of this name. */
+ seq = charmap_find_value (charmap, buf, lenfrom);
+ if (seq == NULL || seq->ucs4 == UNINITIALIZED_CHAR_VALUE)
+ {
+ wc = repertoire_find_value (repertoire, buf, lenfrom);
- arg = lr_token (ldfile, charmap, repertoire);
- if (arg->tok != tok_bsymbol)
- goto err_label;
- else if (!ignore_content)
- {
- /* Check whether this section is already known. */
- struct section_list *known = collate->sections;
- while (known != NULL)
- if (strcmp (known->name, arg->val.str.startmb) == 0)
- break;
+ if (seq != NULL)
+ seq->ucs4 = wc;
+ }
+ else
+ wc = seq->ucs4;
- if (known != NULL)
- {
- lr_error (ldfile,
- _("%s: duplicate declaration of section `%s'"),
- "LC_COLLATE", arg->val.str.startmb);
- free (arg->val.str.startmb);
+ if (wc == ILLEGAL_CHAR_VALUE && seq == NULL)
+ /* We don't know anything about a character with this
+ name. XXX Should we warn? */
+ continue;
+
+ if (elem == NULL)
+ {
+ uint32_t wcs[2] = { wc, 0 };
+
+ /* We have to allocate an entry. */
+ elem = new_element (collate,
+ seq != NULL ? seq->bytes : NULL,
+ seq != NULL ? seq->nbytes : 0,
+ wc == ILLEGAL_CHAR_VALUE
+ ? NULL : wcs, buf, lenfrom, 1);
+ }
+ else
+ {
+ /* Update the element. */
+ if (seq != NULL)
+ {
+ elem->mbs = obstack_copy0 (&collate->mempool,
+ seq->bytes, seq->nbytes);
+ elem->nmbs = seq->nbytes;
+ }
+
+ if (wc != ILLEGAL_CHAR_VALUE)
+ {
+ uint32_t zero = 0;
+
+ obstack_grow (&collate->mempool,
+ &wc, sizeof (uint32_t));
+ obstack_grow (&collate->mempool,
+ &zero, sizeof (uint32_t));
+ elem->wcs = obstack_finish (&collate->mempool);
+ elem->nwcs = 1;
+ }
+ }
+
+ elem->file = ldfile->fname;
+ elem->line = ldfile->lineno;
+ elem->section = collate->current_section;
}
- else
- collate->sections = make_seclist_elem (collate,
- arg->val.str.startmb,
- collate->sections);
- lr_ignore_rest (ldfile, known == NULL);
- }
- else
- {
- free (arg->val.str.startmb);
- lr_ignore_rest (ldfile, 0);
+ /* Enqueue the new element. */
+ elem->last = collate->cursor;
+ elem->next = collate->cursor->next;
+ elem->last->next = elem;
+ if (elem->next != NULL)
+ elem->next->last = elem;
+ collate->cursor = elem;
+
+ /* Now add the weights. They come from the `ellipsis_weights'
+ member of `collate'. */
+ elem->weights = (struct element_list_t *)
+ obstack_alloc (&collate->mempool,
+ nrules * sizeof (struct element_list_t));
+ for (cnt = 0; cnt < nrules; ++cnt)
+ if (collate->ellipsis_weight.weights[cnt].cnt == 1
+ && (collate->ellipsis_weight.weights[cnt].w[0]
+ == ELEMENT_ELLIPSIS2))
+ {
+ elem->weights[cnt].w = (struct element_t **)
+ obstack_alloc (&collate->mempool,
+ sizeof (struct element_t *));
+ elem->weights[cnt].w[0] = elem;
+ elem->weights[cnt].cnt = 1;
+ }
+ else
+ {
+ /* Simly use the weight from `ellipsis_weight'. */
+ elem->weights[cnt].w =
+ collate->ellipsis_weight.weights[cnt].w;
+ elem->weights[cnt].cnt =
+ collate->ellipsis_weight.weights[cnt].cnt;
+ }
}
+ }
+ }
+}
+
+
+static void
+collate_startup (struct linereader *ldfile, struct localedef_t *locale,
+ struct localedef_t *copy_locale, int ignore_content)
+{
+ if (!ignore_content)
+ {
+ struct locale_collate_t *collate;
+
+ if (copy_locale == NULL)
+ {
+ collate = locale->categories[LC_COLLATE].collate =
+ (struct locale_collate_t *)
+ xcalloc (1, sizeof (struct locale_collate_t));
+
+ /* Init the various data structures. */
+ init_hash (&collate->elem_table, 100);
+ init_hash (&collate->sym_table, 100);
+ init_hash (&collate->seq_table, 500);
+ obstack_init (&collate->mempool);
+
+ collate->col_weight_max = -1;
+ }
+ else
+ collate = locale->categories[LC_COLLATE].collate =
+ copy_locale->categories[LC_COLLATE].collate;
+ }
+
+ ldfile->translate_strings = 0;
+ ldfile->return_widestr = 0;
+}
+
+
+void
+collate_finish (struct localedef_t *locale, struct charmap_t *charmap)
+{
+ /* Now is the time when we can assign the individual collation
+ values for all the symbols. We have possibly different values
+ for the wide- and the multibyte-character symbols. This is done
+ since it might make a difference in the encoding if there is in
+ some cases no multibyte-character but there are wide-characters.
+ (The other way around it is not important since theencoded
+ collation value in the wide-character case is 32 bits wide and
+ therefore requires no encoding).
+
+ The lowest collation value assigned is 2. Zero is reserved for
+ the NUL byte terminating the strings in the `strxfrm'/`wcsxfrm'
+ functions and 1 is used to separate the individual passes for the
+ different rules.
+
+ We also have to construct is list with all the bytes/words which
+ can come first in a sequence, followed by all the elements which
+ also start with this byte/word. The order is reverse which has
+ among others the important effect that longer strings are located
+ first in the list. This is required for the output data since
+ the algorithm used in `strcoll' etc depends on this.
+
+ The multibyte case is easy. We simply sort into an array with
+ 256 elements. */
+ struct locale_collate_t *collate = locale->categories[LC_COLLATE].collate;
+ int mbact[nrules];
+ int wcact;
+ struct element_t *runp;
+ int i;
+ int need_undefined = 0;
+ struct section_list *sect;
+ int ruleidx;
+ int nr_wide_elems = 0;
+ size_t min_total;
+ size_t act_size;
+
+ if (collate == NULL)
+ {
+ /* No data, no check. */
+ if (! be_quiet)
+ error (0, 0, _("No definition for %s category found"), "LC_COLLATE");
+ return;
+ }
+
+ /* If this assertion is hit change the type in `element_t'. */
+ assert (nrules <= sizeof (runp->used_in_level) * 8);
+
+ /* Make sure that the `position' rule is used either in all sections
+ or in none. */
+ for (i = 0; i < nrules; ++i)
+ for (sect = collate->sections; sect != NULL; sect = sect->next)
+ if ((sect->rules[i] & sort_position)
+ != (collate->sections->rules[i] & sort_position))
+ {
+ error (0, 0, _("\
+%s: `position' must be used for a specific level in all sections or none"),
+ "LC_COLLATE");
break;
+ }
- case tok_collating_element:
- /* Ignore the rest of the line if we don't need the input of
- this line. */
- if (ignore_content)
+ /* Find out which elements are used at which level. At the same
+ time we find out whether we have any undefined symbols. */
+ runp = collate->start;
+ while (runp != NULL)
+ {
+ if (runp->mbs != NULL)
+ {
+ for (i = 0; i < nrules; ++i)
{
- lr_ignore_rest (ldfile, 0);
- break;
+ int j;
+
+ for (j = 0; j < runp->weights[i].cnt; ++j)
+ /* A NULL pointer as the weight means IGNORE. */
+ if (runp->weights[i].w[j] != NULL)
+ {
+ if (runp->weights[i].w[j]->weights == NULL)
+ {
+ error_at_line (0, 0, runp->file, runp->line,
+ _("symbol `%s' not defined"),
+ runp->weights[i].w[j]->name);
+
+ need_undefined = 1;
+ runp->weights[i].w[j] = &collate->undefined;
+ }
+ else
+ /* Set the bit for the level. */
+ runp->weights[i].w[j]->used_in_level |= 1 << i;
+ }
}
+ }
- if (state != 0)
- goto err_label;
+ /* Up to the next entry. */
+ runp = runp->next;
+ }
- arg = lr_token (ldfile, charmap, repertoire);
- if (arg->tok != tok_bsymbol)
- goto err_label;
- else
- {
- const char *symbol = arg->val.str.startmb;
- size_t symbol_len = arg->val.str.lenmb;
+ /* Walk through the list of defined sequences and assign weights. Also
+ create the data structure which will allow generating the single byte
+ character based tables.
+
+ Since at each time only the weights for each of the rules are
+ only compared to other weights for this rule it is possible to
+ assign more compact weight values than simply counting all
+ weights in sequence. We can assign weights from 3, one for each
+ rule individually and only for those elements, which are actually
+ used for this rule.
+
+ Why is this important? It is not for the wide char table. But
+ it is for the singlebyte output since here larger numbers have to
+ be encoded to make it possible to emit the value as a byte
+ string. */
+ for (i = 0; i < nrules; ++i)
+ mbact[i] = 2;
+ wcact = 2;
+ runp = collate->start;
+ while (runp != NULL)
+ {
+ /* Determine the order. */
+ if (runp->used_in_level != 0)
+ {
+ runp->mborder = (int *) obstack_alloc (&collate->mempool,
+ nrules * sizeof (int));
- /* Next the `from' keyword. */
- arg = lr_token (ldfile, charmap, repertoire);
- if (arg->tok != tok_from)
- {
- free ((char *) symbol);
- goto err_label;
- }
+ for (i = 0; i < nrules; ++i)
+ if ((runp->used_in_level & (1 << i)) != 0)
+ runp->mborder[i] = mbact[i]++;
+ else
+ runp->mborder[i] = 0;
+ }
- ldfile->return_widestr = 1;
+ if (runp->mbs != NULL)
+ {
+ struct element_t **eptr;
- /* Finally the string with the replacement. */
- arg = lr_token (ldfile, charmap, repertoire);
- ldfile->return_widestr = 0;
- if (arg->tok != tok_string)
- goto err_label;
+ /* Find the point where to insert in the list. */
+ eptr = &collate->mbheads[((unsigned char *) runp->mbs)[0]];
+ while (*eptr != NULL)
+ {
+ if ((*eptr)->nmbs < runp->nmbs)
+ break;
- if (!ignore_content)
+ if ((*eptr)->nmbs == runp->nmbs)
{
- if (symbol == NULL)
- lr_error (ldfile, _("\
-%s: unknown character in collating element name"),
- "LC_COLLATE");
- if (arg->val.str.startmb == NULL)
- lr_error (ldfile, _("\
-%s: unknown character in collating element definition"),
- "LC_COLLATE");
- if (arg->val.str.startwc == NULL)
- lr_error (ldfile, _("\
-%s: unknown wide character in collating element definition"),
- "LC_COLLATE");
- else if (arg->val.str.lenwc < 2)
- lr_error (ldfile, _("\
-%s: substitution string in collating element definition must have at least two characters"),
- "LC_COLLATE");
+ int c = memcmp ((*eptr)->mbs, runp->mbs, runp->nmbs);
- if (symbol != NULL)
+ if (c == 0)
{
- /* The name is already defined. */
- if (check_duplicate (ldfile, collate, charmap,
- repertoire, symbol, symbol_len))
- goto col_elem_free;
-
- if (insert_entry (&collate->elem_table,
- symbol, symbol_len,
- new_element (collate,
- arg->val.str.startmb,
- arg->val.str.lenmb,
- arg->val.str.startwc))
- < 0)
- lr_error (ldfile, _("\
-error while adding collating element"));
+ /* This should not happen. It means that we have
+ to symbols with the same byte sequence. It is
+ of course an error. */
+ error_at_line (0, 0, (*eptr)->file, (*eptr)->line,
+ _("symbol `%s' has same encoding as"),
+ (*eptr)->name);
+ error_at_line (0, 0, runp->file, runp->line,
+ _("symbol `%s'"), runp->name);
+ goto dont_insert;
}
- else
- goto col_elem_free;
- }
- else
- {
- col_elem_free:
- if (symbol != NULL)
- free ((char *) symbol);
- if (arg->val.str.startmb != NULL)
- free (arg->val.str.startmb);
- if (arg->val.str.startwc != NULL)
- free (arg->val.str.startwc);
+ else if (c < 0)
+ /* Insert it here. */
+ break;
}
- lr_ignore_rest (ldfile, 1);
- }
- break;
- case tok_collating_symbol:
- /* Ignore the rest of the line if we don't need the input of
- this line. */
- if (ignore_content)
- {
- lr_ignore_rest (ldfile, 0);
- break;
+ /* To the next entry. */
+ eptr = &(*eptr)->mbnext;
}
- if (state != 0)
- goto err_label;
+ /* Set the pointers. */
+ runp->mbnext = *eptr;
+ *eptr = runp;
+ dont_insert:
+ }
- arg = lr_token (ldfile, charmap, repertoire);
- if (arg->tok != tok_bsymbol)
- goto err_label;
- else
+ if (runp->wcs != NULL)
+ {
+ runp->wcorder = wcact++;
+
+ /* We take the opportunity to count the elements which have
+ wide characters. */
+ ++nr_wide_elems;
+ }
+
+ /* Up to the next entry. */
+ runp = runp->next;
+ }
+
+ /* Find out whether any of the `mbheads' entries is unset. In this
+ case we use the UNDEFINED entry. */
+ for (i = 1; i < 256; ++i)
+ if (collate->mbheads[i] == NULL)
+ {
+ need_undefined = 1;
+ collate->mbheads[i] = &collate->undefined;
+ }
+
+ /* Now to the wide character case. Here we have to find first a good
+ mapping function to get the wide range of wide character values
+ (0x00000000 to 0x7fffffff) to a managable table. This might take
+ some time so we issue a warning.
+
+ We use a very trivial hashing function to store the sparse
+ table. CH % TABSIZE is used as an index. To solve multiple hits
+ we have N planes. This guarantees a fixed search time for a
+ character [N / 2]. In the following code we determine the minimum
+ value for TABSIZE * N, where TABSIZE >= 256.
+
+ Some people complained that this algorithm takes too long. Well,
+ go on, improve it. But changing the step size is *not* an
+ option. Some people changed this to use only sizes of prime
+ numbers. Think again, do some math. We are looking for the
+ optimal solution, not something which works in general. Unless
+ somebody can provide a dynamic programming solution I think this
+ implementation is as good as it can get. */
+ if (nr_wide_elems > 512 && !be_quiet)
+ fputs (_("\
+Computing table size for collation table might take a while..."),
+ stderr);
+
+ min_total = UINT_MAX;
+ act_size = 256;
+
+ /* While we want to have a small total size we are willing to use a
+ little bit larger table if this reduces the number of layers.
+ Therefore we add a little penalty to the number of planes.
+ Maybe this constant has to be adjusted a bit. */
+#define PENALTY 128
+ do
+ {
+ size_t cnt[act_size];
+ struct element_t *elem[act_size];
+ size_t act_planes = 1;
+
+ memset (cnt, '\0', sizeof cnt);
+ memset (elem, '\0', sizeof elem);
+
+ runp = collate->start;
+ while (runp != NULL)
+ {
+ if (runp->wcs != NULL)
{
- const char *symbol = arg->val.str.startmb;
- size_t symbol_len = arg->val.str.lenmb;
+ size_t nr = runp->wcs[0] % act_size;
+ struct element_t *elemp = elem[nr];
- if (!ignore_content)
+ while (elemp != NULL)
{
- if (symbol == NULL)
- lr_error (ldfile, _("\
-%s: unknown character in collating symbol name"),
- "LC_COLLATE");
- else
- {
- /* The name is already defined. */
- if (check_duplicate (ldfile, collate, charmap,
- repertoire, symbol, symbol_len))
- goto col_sym_free;
-
- if (insert_entry (&collate->sym_table,
- symbol, symbol_len,
- new_symbol (collate)) < 0)
- lr_error (ldfile, _("\
-error while adding collating symbol"));
- }
+ if (elemp->wcs[0] == runp->wcs[0])
+ break;
+ elemp = elemp->wcnext;
}
- else
+
+ if (elemp == NULL && ++cnt[nr] > act_planes)
{
- col_sym_free:
- if (symbol != NULL)
- free ((char *) symbol);
+ act_planes = cnt[nr];
+
+ runp->wcnext = elem[nr];
+ elem[nr] = runp;
+
+ if ((act_size + PENALTY) * act_planes >= min_total)
+ break;
}
- lr_ignore_rest (ldfile, 1);
}
- break;
- case tok_symbol_equivalence:
- /* Ignore the rest of the line if we don't need the input of
- this line. */
- if (ignore_content)
- {
- lr_ignore_rest (ldfile, 0);
- break;
- }
+ /* Up to the next entry. */
+ runp = runp->next;
+ }
- if (state != 0)
- goto err_label;
+ if ((act_size + PENALTY) * act_planes < min_total)
+ {
+ min_total = (act_size + PENALTY) * act_planes;
+ collate->plane_size = act_size;
+ collate->plane_cnt = act_planes;
+ }
- arg = lr_token (ldfile, charmap, repertoire);
- if (arg->tok != tok_bsymbol)
- goto err_label;
- else
+ ++act_size;
+ }
+ while (act_size < min_total);
+
+ if (nr_wide_elems > 512 && !be_quiet)
+ fputs (_(" done\n"), stderr);
+
+ /* Now that we know how large the table has to be we are able to
+ allocate the array and start adding the characters to the lists
+ in the same way we did it for the multibyte characters. */
+ collate->wcheads = (struct element_t **)
+ obstack_alloc (&collate->mempool, (collate->plane_size
+ * collate->plane_cnt
+ * sizeof (struct element_t *)));
+ memset (collate->wcheads, '\0', (collate->plane_size
+ * collate->plane_cnt
+ * sizeof (struct element_t *)));
+
+ /* Start adding. */
+ runp = collate->start;
+ while (runp != NULL)
+ {
+ if (runp->wcs != NULL)
+ {
+ struct element_t **eptr;
+ size_t idx;
+
+ /* Find a free index. */
+ idx = runp->wcs[0] % collate->plane_size;
+ while (collate->wcheads[idx] != NULL)
{
- const char *newname = arg->val.str.startmb;
- size_t newname_len = arg->val.str.lenmb;
- const char *symname;
- size_t symname_len;
- struct symbol_t *symval;
+ /* Stop if this is an entry with the same starting character. */
+ if (collate->wcheads[idx]->wcs[0] == runp->wcs[0])
+ break;
- arg = lr_token (ldfile, charmap, repertoire);
- if (arg->tok != tok_bsymbol)
- {
- if (newname != NULL)
- free ((char *) newname);
- goto err_label;
- }
+ idx += collate->plane_size;
+ }
- symname = arg->val.str.startmb;
- symname_len = arg->val.str.lenmb;
+ /* Find the point where to insert in the list. */
+ eptr = &collate->wcheads[idx];
+ while (*eptr != NULL)
+ {
+ if ((*eptr)->nwcs < runp->nwcs)
+ break;
- if (!ignore_content)
+ if ((*eptr)->nwcs == runp->nwcs)
{
- if (newname == NULL)
- {
- lr_error (ldfile, _("\
-%s: unknown character in equivalent definition name"),
- "LC_COLLATE");
- goto sym_equiv_free;
- }
- if (symname == NULL)
- {
- lr_error (ldfile, _("\
-%s: unknown character in equivalent definition value"),
- "LC_COLLATE");
- goto sym_equiv_free;
- }
- /* The name is already defined. */
- if (check_duplicate (ldfile, collate, charmap,
- repertoire, symname, symname_len))
- goto col_sym_free;
-
- /* See whether the symbol name is already defined. */
- if (find_entry (&collate->sym_table, symname, symname_len,
- (void **) &symval) != 0)
- {
- lr_error (ldfile, _("\
-%s: unknown symbol `%s' in equivalent definition"),
- "LC_COLLATE", symname);
- goto col_sym_free;
- }
+ int c = wmemcmp ((wchar_t *) (*eptr)->wcs,
+ (wchar_t *) runp->wcs, runp->nwcs);
- if (insert_entry (&collate->sym_table,
- newname, newname_len, symval) < 0)
+ if (c == 0)
{
- lr_error (ldfile, _("\
-error while adding equivalent collating symbol"));
- goto sym_equiv_free;
+ /* This should not happen. It means that we have
+ to symbols with the same byte sequence. It is
+ of course an error. */
+ error_at_line (0, 0, (*eptr)->file, (*eptr)->line,
+ _("symbol `%s' has same encoding as"),
+ (*eptr)->name);
+ error_at_line (0, 0, runp->file, runp->line,
+ _("symbol `%s'"), runp->name);
+ goto dont_insertwc;
}
-
- free ((char *) symname);
- }
- else
- {
- sym_equiv_free:
- if (newname != NULL)
- free ((char *) newname);
- if (symname != NULL)
- free ((char *) symname);
+ else if (c < 0)
+ /* Insert it here. */
+ break;
}
- lr_ignore_rest (ldfile, 1);
- }
- break;
- case tok_order_start:
- /* Ignore the rest of the line if we don't need the input of
- this line. */
- if (ignore_content)
- {
- lr_ignore_rest (ldfile, 0);
- break;
+ /* To the next entry. */
+ eptr = &(*eptr)->wcnext;
}
- if (state != 0 && state != 1)
- goto err_label;
- state = 1;
+ /* Set the pointers. */
+ runp->wcnext = *eptr;
+ *eptr = runp;
+ dont_insertwc:
+ }
- /* The 14652 draft does not specify whether all `order_start' lines
- must contain the same number of sort-rules, but 14651 does. So
- we require this here as well. */
- arg = lr_token (ldfile, charmap, repertoire);
- if (arg->tok == tok_bsymbol)
- {
- /* This better should be a section name. */
- struct section_list *sp = collate->sections;
- while (sp != NULL
- && strcmp (sp->name, arg->val.str.startmb) != 0)
- sp = sp->next;
+ /* Up to the next entry. */
+ runp = runp->next;
+ }
- if (sp == NULL)
- {
- lr_error (ldfile, _("\
-%s: unknown section name `%s'"),
- "LC_COLLATE", arg->val.str.startmb);
- /* We use the error section. */
- collate->current_section = &collate->error_section;
- }
- else
- {
- /* Remember this section. */
- collate->current_section = sp;
+ /* Now determine whether the UNDEFINED entry is needed and if yes,
+ whether it was defined. */
+ collate->undefined.used_in_level = need_undefined ? ~0ul : 0;
+ if (collate->undefined.file == NULL)
+ {
+ if (need_undefined)
+ {
+ error (0, 0, _("no definition of `UNDEFINED'"));
- /* One should not be allowed to open the same
- section twice. */
- if (sp->first != NULL)
- lr_error (ldfile, _("\
-%s: multiple order definitions for section `%s'"),
- "LC_COLLATE", sp->name);
+ /* Add UNDEFINED at the end. */
+ collate->undefined.mborder =
+ (int *) obstack_alloc (&collate->mempool, nrules * sizeof (int));
- /* Next should come the end of the line or a semicolon. */
- arg = lr_token (ldfile, charmap, repertoire);
- if (arg->tok == tok_eol)
- {
- uint32_t cnt;
+ for (i = 0; i < nrules; ++i)
+ collate->undefined.mborder[i] = mbact[i]++;
+ }
- /* This means we have exactly one rule: `forward'. */
- if (collate->nrules > 1)
- lr_error (ldfile, _("\
-%s: invalid number of sorting rules"),
- "LC_COLLATE");
- else
- collate->nrules = 1;
- sp->rules = obstack_alloc (&collate->mempool,
- (sizeof (enum coll_sort_rule)
- * collate->nrules));
- for (cnt = 0; cnt < collate->nrules; ++cnt)
- sp->rules[cnt] = sort_forward;
+ /* In any case we will need the definition for the wide character
+ case. But we will not complain that it is missing since the
+ specification strangely enough does not seem to account for
+ this. */
+ collate->undefined.wcorder = wcact++;
+ }
- /* Next line. */
- break;
- }
+ /* Finally, try to unify the rules for the sections. Whenever the rules
+ for a section are the same as those for another section give the
+ ruleset the same index. Since there are never many section we can
+ use an O(n^2) algorithm here. */
+ sect = collate->sections;
+ assert (sect != NULL);
+ ruleidx = 0;
+ do
+ {
+ struct section_list *osect = collate->sections;
- /* Get the next token. */
- arg = lr_token (ldfile, charmap, repertoire);
- }
- }
- else
- {
- /* There is no section symbol. Therefore we use the unnamed
- section. */
- collate->current_section = &collate->unnamed_section;
+ while (osect != sect)
+ if (memcmp (osect->rules, sect->rules, nrules) == 0)
+ break;
+ else
+ osect = osect->next;
- if (collate->unnamed_section.first != NULL)
- lr_error (ldfile, _("\
-%s: multiple order definitions for unnamed section"),
- "LC_COLLATE");
- }
+ if (osect == sect)
+ sect->ruleidx = ruleidx++;
+ else
+ sect->ruleidx = osect->ruleidx;
- /* Now read the direction names. */
- read_directions (ldfile, arg, charmap, repertoire, collate);
+ /* Next section. */
+ sect = sect->next;
+ }
+ while (sect != NULL);
+ /* We are currently not prepared for more than 256 rulesets. But this
+ should never really be a problem. */
+ assert (ruleidx <= 256);
+}
- /* From now be need the strings untranslated. */
- ldfile->translate_strings = 0;
- break;
- case tok_order_end:
- /* Ignore the rest of the line if we don't need the input of
- this line. */
- if (ignore_content)
- {
- lr_ignore_rest (ldfile, 0);
- break;
- }
+static int32_t
+output_weight (struct obstack *pool, struct locale_collate_t *collate,
+ struct element_t *elem)
+{
+ size_t cnt;
+ int32_t retval;
- if (state != 1)
- goto err_label;
- state = 2;
- lr_ignore_rest (ldfile, 1);
- break;
+ /* Optimize the use of UNDEFINED. */
+ if (elem == &collate->undefined)
+ /* The weights are already inserted. */
+ return 0;
- case tok_reorder_after:
- /* Ignore the rest of the line if we don't need the input of
- this line. */
- if (ignore_content)
- {
- lr_ignore_rest (ldfile, 0);
- break;
- }
+ /* This byte can start exactly one collation element and this is
+ a single byte. We can directly give the index to the weights. */
+ retval = obstack_object_size (pool);
- if (state != 2 && state != 3)
- goto err_label;
- state = 3;
+ /* Construct the weight. */
+ for (cnt = 0; cnt < nrules; ++cnt)
+ {
+ char buf[elem->weights[cnt].cnt * 7];
+ int len = 0;
+ int i;
+
+ for (i = 0; i < elem->weights[cnt].cnt; ++i)
+ /* Encode the weight value. We do nothing for IGNORE entries. */
+ if (elem->weights[cnt].w[i] != NULL)
+ len += utf8_encode (&buf[len],
+ elem->weights[cnt].w[i]->mborder[cnt]);
+
+ /* And add the buffer content. */
+ obstack_1grow (pool, len);
+ obstack_grow (pool, buf, len);
+ }
- arg = lr_token (ldfile, charmap, repertoire);
- if (arg->tok == tok_bsymbol)
- {
- /* Find this symbol in the sequence table. */
- struct element_t *insp;
- int no_error = 1;
+ return retval | ((elem->section->ruleidx & 0x7f) << 24);
+}
- if (find_entry (&collate->seq_table, arg->val.str.startmb,
- arg->val.str.lenmb, (void **) &insp) == 0)
- /* Yes, the symbol exists. Simply point the cursor
- to it. */
- collate->cursor = insp;
- else
- {
- /* This is bad. The symbol after which we have to
- insert does not exist. */
- lr_error (ldfile, _("\
-%s: cannot reorder after %.*s: symbol not known"),
- "LC_COLLATE", arg->val.str.lenmb,
- arg->val.str.startmb);
- collate->cursor = NULL;
- no_error = 0;
- }
- lr_ignore_rest (ldfile, no_error);
- }
- else
- /* This must not happen. */
- goto err_label;
- break;
+static int32_t
+output_weightwc (struct obstack *pool, struct locale_collate_t *collate,
+ struct element_t *elem)
+{
+ size_t cnt;
+ int32_t retval;
- case tok_reorder_end:
- /* Ignore the rest of the line if we don't need the input of
- this line. */
- if (ignore_content)
- break;
+ /* Optimize the use of UNDEFINED. */
+ if (elem == &collate->undefined)
+ /* The weights are already inserted. */
+ return 0;
- if (state != 3)
- goto err_label;
- state = 4;
- lr_ignore_rest (ldfile, 1);
- break;
+ /* This byte can start exactly one collation element and this is
+ a single byte. We can directly give the index to the weights. */
+ retval = obstack_object_size (pool);
- case tok_bsymbol:
- /* Ignore the rest of the line if we don't need the input of
- this line. */
- if (ignore_content)
- {
- lr_ignore_rest (ldfile, 0);
- break;
- }
+ /* Construct the weight. */
+ for (cnt = 0; cnt < nrules; ++cnt)
+ {
+ int32_t buf[elem->weights[cnt].cnt];
+ int32_t i;
- if (state != 1 && state != 3)
- goto err_label;
+ for (i = 0; i < elem->weights[cnt].cnt; ++i)
+ if (elem->weights[cnt].w[i] != NULL)
+ buf[i] = elem->weights[cnt].w[i]->wcorder;
- if (state == 3)
- {
- /* It is possible that we already have this collation sequence.
- In this case we move the entry. */
- struct element_t *seqp;
-
- /* If the symbol after which we have to insert was not found
- ignore all entries. */
- if (collate->cursor == NULL)
- {
- lr_ignore_rest (ldfile, 0);
- break;
- }
+ /* And add the buffer content. */
+ if (sizeof (int) == sizeof (int32_t))
+ obstack_int_grow (pool, i);
+ else
+ obstack_grow (pool, &i, sizeof (int32_t));
- if (find_entry (&collate->seq_table, arg->val.str.startmb,
- arg->val.str.lenmb, (void **) &seqp) == 0)
- {
- /* Remove the entry from the old position. */
- if (seqp->last == NULL)
- collate->start = seqp->next;
- else
- seqp->last->next = seqp->next;
- if (seqp->next != NULL)
- seqp->next->last = seqp->last;
+ obstack_grow (pool, buf, i * sizeof (int32_t));
+ }
- /* We also have to check whether this entry is the
- first or last of a section. */
- if (seqp->section->first == seqp)
- {
- if (seqp->section->first == seqp->section->last)
- /* This setion has no content anymore. */
- seqp->section->first = seqp->section->last = NULL;
- else
- seqp->section->first = seqp->next;
- }
- else if (seqp->section->last == seqp)
- seqp->section->last = seqp->last;
+ return retval | ((elem->section->ruleidx & 0x7f) << 24);
+}
- /* Now insert it in the new place. */
- seqp->next = collate->cursor->next;
- seqp->last = collate->cursor;
- collate->cursor->next = seqp;
- if (seqp->next != NULL)
- seqp->next->last = seqp;
- seqp->section = collate->cursor->section;
- if (seqp->section->last == collate->cursor)
- seqp->section->last = seqp;
+void
+collate_output (struct localedef_t *locale, struct charmap_t *charmap,
+ const char *output_path)
+{
+ struct locale_collate_t *collate = locale->categories[LC_COLLATE].collate;
+ const size_t nelems = _NL_ITEM_INDEX (_NL_NUM_LC_COLLATE);
+ struct iovec iov[2 + nelems];
+ struct locale_file data;
+ uint32_t idx[nelems];
+ size_t cnt;
+ size_t ch;
+ int32_t tablemb[256];
+ struct obstack weightpool;
+ struct obstack extrapool;
+ struct obstack indirectpool;
+ struct section_list *sect;
+ uint32_t *names;
+ uint32_t *tablewc;
+ size_t table_size;
+ uint32_t elem_size;
+ uint32_t *elem_table;
+ int i;
+ struct element_t *runp;
- break;
- }
+ data.magic = LIMAGIC (LC_COLLATE);
+ data.n = nelems;
+ iov[0].iov_base = (void *) &data;
+ iov[0].iov_len = sizeof (data);
- /* Otherwise we just add a new entry. */
- }
+ iov[1].iov_base = (void *) idx;
+ iov[1].iov_len = sizeof (idx);
- /* Now insert in the new place. */
- insert_value (ldfile, arg, charmap, repertoire, collate);
- break;
+ idx[0] = iov[0].iov_len + iov[1].iov_len;
+ cnt = 0;
- case tok_undefined:
- /* Ignore the rest of the line if we don't need the input of
- this line. */
- if (ignore_content)
- {
- lr_ignore_rest (ldfile, 0);
- break;
- }
+ assert (cnt == _NL_ITEM_INDEX (_NL_COLLATE_NRULES));
+ iov[2 + cnt].iov_base = &nrules;
+ iov[2 + cnt].iov_len = sizeof (uint32_t);
+ idx[1 + cnt] = idx[cnt] + iov[2 + cnt].iov_len;
+ ++cnt;
- if (state != 1)
- goto err_label;
+ /* If we have no LC_COLLATE data emit only the number of rules as zero. */
+ if (collate == NULL)
+ {
+ int32_t dummy = 0;
- /* See whether UNDEFINED already appeared somewhere. */
- if (collate->undefined.next != NULL
- || (collate->cursor != NULL
- && collate->undefined.next == collate->cursor))
+ while (cnt < _NL_ITEM_INDEX (_NL_NUM_LC_COLLATE))
+ {
+ /* The words have to be handled specially. */
+ if (cnt == _NL_ITEM_INDEX (_NL_COLLATE_HASH_SIZE)
+ || cnt == _NL_ITEM_INDEX (_NL_COLLATE_HASH_LAYERS)
+ || cnt == _NL_ITEM_INDEX (_NL_COLLATE_SYMB_HASH_SIZEMB))
{
- lr_error (ldfile, _("order for `%.*s' already defined at %s:%Z"),
- 9, "UNDEFINED", collate->undefined.file,
- collate->undefined.line);
- lr_ignore_rest (ldfile, 0);
+ iov[2 + cnt].iov_base = &dummy;
+ iov[2 + cnt].iov_len = sizeof (int32_t);
}
else
- /* Parse the weights. */
- insert_weights (ldfile, &collate->undefined, charmap,
- repertoire, collate);
- break;
-
- case tok_ellipsis3:
- /* Ignore the rest of the line if we don't need the input of
- this line. */
- if (ignore_content)
{
- lr_ignore_rest (ldfile, 0);
- break;
+ iov[2 + cnt].iov_base = (char *) "";
+ iov[2 + cnt].iov_len = 0;
}
- if (state != 1 && state != 3)
- goto err_label;
+ if (cnt + 1 < _NL_ITEM_INDEX (_NL_NUM_LC_COLLATE))
+ idx[1 + cnt] = idx[cnt] + iov[2 + cnt].iov_len;
+ ++cnt;
+ }
- was_ellipsis = 1;
- /* XXX Read the remainder of the line and remember what are
- the weights. */
- break;
+ assert (cnt == _NL_ITEM_INDEX (_NL_NUM_LC_COLLATE));
- case tok_end:
- /* Next we assume `LC_COLLATE'. */
- if (!ignore_content)
- {
- if (state == 0)
- /* We must either see a copy statement or have
- ordering values. */
- lr_error (ldfile,
- _("%s: empty category description not allowed"),
- "LC_COLLATE");
- else if (state == 1)
- lr_error (ldfile, _("%s: missing `order_end' keyword"),
- "LC_COLLATE");
- else if (state == 3)
- error (0, 0, _("%s: missing `reorder-end' keyword"),
- "LC_COLLATE");
- }
- arg = lr_token (ldfile, charmap, NULL);
- if (arg->tok == tok_eof)
- break;
- if (arg->tok == tok_eol)
- lr_error (ldfile, _("%s: incomplete `END' line"), "LC_COLLATE");
- else if (arg->tok != tok_lc_collate)
- lr_error (ldfile, _("\
-%1$s: definition does not end with `END %1$s'"), "LC_COLLATE");
- lr_ignore_rest (ldfile, arg->tok == tok_lc_collate);
- return;
+ write_locale_data (output_path, "LC_COLLATE", 2 + cnt, iov);
- default:
- err_label:
- SYNTAX_ERROR (_("%s: syntax error"), "LC_COLLATE");
- }
+ return;
+ }
- /* Prepare for the next round. */
- now = lr_token (ldfile, charmap, NULL);
- nowtok = now->tok;
+ obstack_init (&weightpool);
+ obstack_init (&extrapool);
+ obstack_init (&indirectpool);
+
+ /* Since we are using the sign of an integer to mark indirection the
+ offsets in the arrays we are indirectly referring to must not be
+ zero since -0 == 0. Therefore we add a bit of dummy content. */
+ if (sizeof (int) == sizeof (int32_t))
+ {
+ obstack_int_grow (&extrapool, 0);
+ obstack_int_grow (&indirectpool, 0);
+ }
+ else
+ {
+ int32_t zero = 0;
+ obstack_grow (&extrapool, &zero, sizeof (zero));
+ obstack_grow (&indirectpool, &zero, sizeof (zero));
}
- /* When we come here we reached the end of the file. */
- lr_error (ldfile, _("%s: premature end of file"), "LC_COLLATE");
-}
+ /* Prepare the ruleset table. */
+ for (sect = collate->sections, i = 0; sect != NULL; sect = sect->next)
+ if (sect->ruleidx == i)
+ {
+ int j;
+ obstack_make_room (&weightpool, nrules);
-#if 0
+ for (j = 0; j < nrules; ++j)
+ obstack_1grow_fast (&weightpool, sect->rules[j]);
+ ++i;
+ }
+ /* And align the output. */
+ i = (nrules * i) % __alignof__ (int32_t);
+ if (i > 0)
+ do
+ obstack_1grow (&weightpool, '\0');
+ while (++i < __alignof__ (int32_t));
+
+ assert (cnt == _NL_ITEM_INDEX (_NL_COLLATE_RULESETS));
+ iov[2 + cnt].iov_len = obstack_object_size (&weightpool);
+ iov[2 + cnt].iov_base = obstack_finish (&weightpool);
+ idx[1 + cnt] = idx[cnt] + iov[2 + cnt].iov_len;
+ ++cnt;
+
+ /* Generate the 8-bit table. Walk through the lists of sequences
+ starting with the same byte and add them one after the other to
+ the table. In case we have more than one sequence starting with
+ the same byte we have to use extra indirection.
+
+ First add a record for the NUL byte. This entry will never be used
+ so it does not matter. */
+ tablemb[0] = 0;
+
+ /* Now insert the `UNDEFINED' value if it is used. Since this value
+ will probably be used more than once it is good to store the
+ weights only once. */
+ if (collate->undefined.used_in_level != 0)
+ output_weight (&weightpool, collate, &collate->undefined);
+
+ for (ch = 1; ch < 256; ++ch)
+ if (collate->mbheads[ch]->mbnext == NULL
+ && collate->mbheads[ch]->nmbs == 1)
+ {
+ tablemb[ch] = output_weight (&weightpool, collate,
+ collate->mbheads[ch]);
+ }
+ else
+ {
+ /* The entries in the list are sorted by length and then
+ alphabetically. This is the order in which we will add the
+ elements to the collation table. This allows to simply
+ walk the table in sequence and stop at the first matching
+ entry. Since the longer sequences are coming first in the
+ list they have the possibility to match first, just as it
+ has to be. In the worst case we are walking to the end of
+ the list where we put, if no singlebyte sequence is defined
+ in the locale definition, the weights for UNDEFINED.
+
+ To reduce the length of the search list we compress them a bit.
+ This happens by collecting sequences of consecutive byte
+ sequences in one entry (having and begin and end byte sequence)
+ and add only one index into the weight table. We can find the
+ consecutive entries since they are also consecutive in the list. */
+ struct element_t *runp = collate->mbheads[ch];
+ struct element_t *lastp;
+
+ tablemb[ch] = -obstack_object_size (&extrapool);
+
+ do
+ {
+ /* Store the current index in the weight table. We know that
+ the current position in the `extrapool' is aligned on a
+ 32-bit address. */
+ int32_t weightidx;
+ int added;
+
+ /* Output the weight info. */
+ weightidx = output_weight (&weightpool, collate, runp);
+
+ /* Find out wether this is a single entry or we have more than
+ one consecutive entry. */
+ if (runp->mbnext != NULL
+ && runp->nmbs == runp->mbnext->nmbs
+ && memcmp (runp->mbs, runp->mbnext->mbs, runp->nmbs - 1) == 0
+ && (runp->mbs[runp->nmbs - 1] + 1
+ == runp->mbnext->mbs[runp->nmbs - 1]))
+ {
+ int i;
+
+ /* Now add first the initial byte sequence. */
+ added = ((sizeof (int32_t) + 1 + 2 * (runp->nmbs - 1)
+ + __alignof__ (int32_t) - 1)
+ & ~(__alignof__ (int32_t) - 1));
+ obstack_make_room (&extrapool, added);
+
+ /* More than one consecutive entry. We mark this by having
+ a negative index into the indirect table. */
+ if (sizeof (int32_t) == sizeof (int))
+ obstack_int_grow_fast (&extrapool,
+ obstack_object_size (&indirectpool)
+ / sizeof (int32_t));
+ else
+ {
+ int32_t i = (obstack_object_size (&indirectpool)
+ / sizeof (int32_t));
+ obstack_grow (&extrapool, &i, sizeof (int32_t));
+ }
+ obstack_1grow_fast (&extrapool, runp->nmbs - 1);
+ for (i = 1; i < runp->nmbs; ++i)
+ obstack_1grow_fast (&extrapool, runp->mbs[i]);
-/* What kind of symbols get defined? */
-enum coll_symbol
-{
- undefined,
- ellipsis,
- character,
- element,
- symbol
-};
+ /* Now find the end of the consecutive sequence and
+ add all the indeces in the indirect pool. */
+ while (1)
+ {
+ if (sizeof (int32_t) == sizeof (int))
+ obstack_int_grow (&extrapool, weightidx);
+ else
+ obstack_grow (&extrapool, &weightidx, sizeof (int32_t));
+
+ runp = runp->next;
+ if (runp->mbnext == NULL
+ || runp->nmbs != runp->mbnext->nmbs
+ || memcmp (runp->mbs, runp->mbnext->mbs,
+ runp->nmbs - 1) != 0
+ || (runp->mbs[runp->nmbs - 1] + 1
+ != runp->mbnext->mbs[runp->nmbs - 1]))
+ break;
+ /* Insert the weight. */
+ weightidx = output_weight (&weightpool, collate, runp);
+ }
-typedef struct patch_t
-{
- const char *fname;
- size_t lineno;
- const char *token;
- union
- {
- unsigned int *pos;
- size_t idx;
- } where;
- struct patch_t *next;
-} patch_t;
-
-
-typedef struct element_t
-{
- const char *namemb;
- const uint32_t *namewc;
- unsigned int this_weight;
+ /* And add the end byte sequence. Without length this
+ time. */
+ for (i = 1; i < runp->nmbs; ++i)
+ obstack_1grow_fast (&extrapool, runp->mbs[i]);
- struct element_t *next;
+ weightidx = output_weight (&weightpool, collate, runp);
+ if (sizeof (int32_t) == sizeof (int))
+ obstack_int_grow (&extrapool, weightidx);
+ else
+ obstack_grow (&extrapool, &weightidx, sizeof (int32_t));
+ }
+ else
+ {
+ /* A single entry. Simply add the index and the length and
+ string (except for the first character which is already
+ tested for). */
+ int i;
+
+ added = ((sizeof (int32_t) + 1 + runp->nmbs - 1
+ + __alignof__ (int32_t) - 1)
+ & ~(__alignof__ (int32_t) - 1));
+ obstack_make_room (&extrapool, added);
+
+ if (sizeof (int32_t) == sizeof (int))
+ obstack_int_grow_fast (&extrapool, weightidx);
+ else
+ obstack_grow (&extrapool, &weightidx, sizeof (int32_t));
+ obstack_1grow_fast (&extrapool, runp->nmbs - 1);
+ for (i = 1; i < runp->nmbs; ++i)
+ obstack_1grow_fast (&extrapool, runp->mbs[i]);
+ }
+
+ /* Add alignment bytes if necessary. */
+ i = added % __alignof__ (int32_t);
+ if (i > 0)
+ do
+ obstack_1grow_fast (&extrapool, '\0');
+ while (++i != __alignof__ (int32_t));
- unsigned int *ordering;
- size_t ordering_len;
-} element_t;
+ /* Next entry. */
+ lastp = runp;
+ runp = runp->mbnext;
+ }
+ while (runp != NULL);
+ /* If the final entry in the list is not a single character we
+ add an UNDEFINED entry here. */
+ if (lastp->nmbs != 1)
+ {
+ int added = ((sizeof (int32_t) + 1 + 1 + __alignof__ (int32_t))
+ & ~(__alignof__ (int32_t) - 1));
+ obstack_make_room (&extrapool, added);
-/* The real definition of the struct for the LC_COLLATE locale. */
-struct locale_collate_t
-{
- /* Collate symbol table. Simple mapping to number. */
- hash_table symbols;
+ if (sizeof (int32_t) == sizeof (int))
+ obstack_int_grow_fast (&extrapool, 0);
+ else
+ {
+ int32_t zero = 0;
+ obstack_grow (&extrapool, &zero, sizeof (int32_t));
+ }
+ /* XXX What rule? We just pick the first. */
+ obstack_1grow_fast (&extrapool, 0);
+ /* Length is zero. */
+ obstack_1grow_fast (&extrapool, 0);
+
+ /* Add alignment bytes if necessary. */
+ i = added % __alignof__ (int32_t);
+ if (i > 0)
+ do
+ obstack_1grow_fast (&extrapool, '\0');
+ while (++i != __alignof__ (int32_t));
+ }
+ }
- /* The collation elements. */
- hash_table elements;
- struct obstack element_mem;
+ /* Now add the four tables. */
+ assert (cnt == _NL_ITEM_INDEX (_NL_COLLATE_TABLEMB));
+ iov[2 + cnt].iov_base = tablemb;
+ iov[2 + cnt].iov_len = sizeof (tablemb);
+ idx[1 + cnt] = idx[cnt] + iov[2 + cnt].iov_len;
+ ++cnt;
+
+ assert (cnt == _NL_ITEM_INDEX (_NL_COLLATE_WEIGHTMB));
+ iov[2 + cnt].iov_len = obstack_object_size (&weightpool);
+ iov[2 + cnt].iov_base = obstack_finish (&weightpool);
+ idx[1 + cnt] = idx[cnt] + iov[2 + cnt].iov_len;
+ ++cnt;
+
+ assert (cnt == _NL_ITEM_INDEX (_NL_COLLATE_EXTRAMB));
+ iov[2 + cnt].iov_len = obstack_object_size (&extrapool);
+ iov[2 + cnt].iov_base = obstack_finish (&extrapool);
+ idx[1 + cnt] = idx[cnt] + iov[2 + cnt].iov_len;
+ ++cnt;
+
+ assert (cnt == _NL_ITEM_INDEX (_NL_COLLATE_INDIRECTMB));
+ iov[2 + cnt].iov_len = obstack_object_size (&indirectpool);
+ iov[2 + cnt].iov_base = obstack_finish (&indirectpool);
+ idx[1 + cnt] = idx[cnt] + iov[2 + cnt].iov_len;
+ ++cnt;
+
+
+ /* Now the same for the wide character table. We need to store some
+ more information here. */
+ assert (cnt == _NL_ITEM_INDEX (_NL_COLLATE_HASH_SIZE));
+ iov[2 + cnt].iov_base = &collate->plane_size;
+ iov[2 + cnt].iov_len = sizeof (collate->plane_size);
+ idx[1 + cnt] = idx[cnt] + iov[2 + cnt].iov_len;
+ ++cnt;
+
+ assert (cnt == _NL_ITEM_INDEX (_NL_COLLATE_HASH_LAYERS));
+ iov[2 + cnt].iov_base = &collate->plane_cnt;
+ iov[2 + cnt].iov_len = sizeof (collate->plane_cnt);
+ idx[1 + cnt] = idx[cnt] + iov[2 + cnt].iov_len;
+ ++cnt;
+
+ /* Construct a table with the names. The size of the table is the same
+ as the table with the pointers. */
+ table_size = collate->plane_size * collate->plane_cnt;
+ names = (uint32_t *) alloca (table_size * sizeof (uint32_t));
+ for (ch = 0; ch < table_size; ++ch)
+ if (collate->wcheads[ch] == NULL)
+ names[ch] = 0;
+ else
+ names[ch] = collate->wcheads[ch]->wcs[0];
+
+ assert (cnt == _NL_ITEM_INDEX (_NL_COLLATE_NAMES));
+ iov[2 + cnt].iov_base = names;
+ iov[2 + cnt].iov_len = table_size * sizeof (uint32_t);
+ idx[1 + cnt] = idx[cnt] + iov[2 + cnt].iov_len;
+ ++cnt;
+
+ /* Since we are using the sign of an integer to mark indirection the
+ offsets in the arrays we are indirectly referring to must not be
+ zero since -0 == 0. Therefore we add a bit of dummy content. */
+ if (sizeof (int) == sizeof (int32_t))
+ {
+ obstack_int_grow (&extrapool, 0);
+ obstack_int_grow (&indirectpool, 0);
+ }
+ else
+ {
+ int32_t zero = 0;
+ obstack_grow (&extrapool, &zero, sizeof (zero));
+ obstack_grow (&indirectpool, &zero, sizeof (zero));
+ }
- /* The result tables. */
- hash_table resultmb;
- hash_table resultwc;
+ /* Now insert the `UNDEFINED' value if it is used. Since this value
+ will probably be used more than once it is good to store the
+ weights only once. */
+ output_weightwc (&weightpool, collate, &collate->undefined);
+
+ /* Generate the table. Walk through the lists of sequences
+ starting with the same byte and add them one after the other to
+ the table. In case we have more than one sequence starting with
+ the same byte we have to use extra indirection. */
+ tablewc = (uint32_t *) alloca (table_size * sizeof (uint32_t));
+ for (ch = 0; ch < table_size; ++ch)
+ if (collate->wcheads[ch] == NULL)
+ {
+ /* Set the entry to zero. */
+ tablewc[ch] = 0;
+ }
+ else if (collate->wcheads[ch]->wcnext == NULL
+ && collate->wcheads[ch]->nwcs == 1)
+ {
+ tablewc[ch] = output_weightwc (&weightpool, collate,
+ collate->wcheads[ch]);
+ }
+ else
+ {
+ /* As for the singlebyte table, we recognize sequences and
+ compress them. */
+ struct element_t *runp = collate->wcheads[ch];
+ struct element_t *lastp;
- /* Sorting rules given in order_start line. */
- uint32_t nrules;
- enum coll_sort_rule *rules;
+ tablewc[ch] = -obstack_object_size (&extrapool);
- /* Used while recognizing symbol composed of multiple tokens
- (collating-element). */
- const char *combine_token;
- size_t combine_token_len;
-
- /* How many sorting order specifications so far. */
- unsigned int order_cnt;
-
- /* Was lastline ellipsis? */
- int was_ellipsis;
- /* Value of last entry if was character. */
- uint32_t last_char;
- /* Current element. */
- element_t *current_element;
- /* What kind of symbol is current element. */
- enum coll_symbol kind;
-
- /* Patch lists. */
- patch_t *current_patch;
- patch_t *all_patches;
-
- /* Room for the UNDEFINED information. */
- element_t undefined;
- unsigned int undefined_len;
-
- /* Script information. */
- const char **scripts;
- unsigned int nscripts;
-};
+ do
+ {
+ /* Store the current index in the weight table. We know that
+ the current position in the `extrapool' is aligned on a
+ 32-bit address. */
+ int32_t weightidx;
+ int added;
+
+ /* Output the weight info. */
+ weightidx = output_weightwc (&weightpool, collate, runp);
+
+ /* Find out wether this is a single entry or we have more than
+ one consecutive entry. */
+ if (runp->wcnext != NULL
+ && runp->nwcs == runp->wcnext->nwcs
+ && wmemcmp ((wchar_t *) runp->wcs,
+ (wchar_t *)runp->wcnext->wcs, runp->nwcs - 1) == 0
+ && (runp->wcs[runp->nwcs - 1] + 1
+ == runp->wcnext->wcs[runp->nwcs - 1]))
+ {
+ int i;
+ /* Now add first the initial byte sequence. */
+ added = (1 + 1 + 2 * (runp->nwcs - 1)) * sizeof (int32_t);
+ if (sizeof (int32_t) == sizeof (int))
+ obstack_make_room (&extrapool, added);
-/* Be verbose? Defined in localedef.c. */
-extern int verbose;
+ /* More than one consecutive entry. We mark this by having
+ a negative index into the indirect table. */
+ if (sizeof (int32_t) == sizeof (int))
+ {
+ obstack_int_grow_fast (&extrapool,
+ obstack_object_size (&indirectpool)
+ / sizeof (int32_t));
+ obstack_int_grow_fast (&extrapool, runp->nwcs - 1);
+ }
+ else
+ {
+ int32_t i = (obstack_object_size (&indirectpool)
+ / sizeof (int32_t));
+ obstack_grow (&extrapool, &i, sizeof (int32_t));
+ i = runp->nwcs - 1;
+ obstack_grow (&extrapool, &i, sizeof (int32_t));
+ }
+ for (i = 1; i < runp->nwcs; ++i)
+ if (sizeof (int32_t) == sizeof (int))
+ obstack_int_grow_fast (&extrapool, runp->wcs[i]);
+ else
+ obstack_grow (&extrapool, &runp->wcs[i], sizeof (int32_t));
+ /* Now find the end of the consecutive sequence and
+ add all the indeces in the indirect pool. */
+ while (1)
+ {
+ if (sizeof (int32_t) == sizeof (int))
+ obstack_int_grow (&extrapool, weightidx);
+ else
+ obstack_grow (&extrapool, &weightidx, sizeof (int32_t));
+
+ runp = runp->next;
+ if (runp->wcnext == NULL
+ || runp->nwcs != runp->wcnext->nwcs
+ || wmemcmp ((wchar_t *) runp->wcs,
+ (wchar_t *) runp->wcnext->wcs,
+ runp->nwcs - 1) != 0
+ || (runp->wcs[runp->nwcs - 1] + 1
+ != runp->wcnext->wcs[runp->nwcs - 1]))
+ break;
+ /* Insert the weight. */
+ weightidx = output_weightwc (&weightpool, collate, runp);
+ }
-#define obstack_chunk_alloc malloc
-#define obstack_chunk_free free
+ /* And add the end byte sequence. Without length this
+ time. */
+ for (i = 1; i < runp->nwcs; ++i)
+ if (sizeof (int32_t) == sizeof (int))
+ obstack_int_grow (&extrapool, runp->wcs[i]);
+ else
+ obstack_grow (&extrapool, &runp->wcs[i], sizeof (int32_t));
+ weightidx = output_weightwc (&weightpool, collate, runp);
+ if (sizeof (int32_t) == sizeof (int))
+ obstack_int_grow (&extrapool, weightidx);
+ else
+ obstack_grow (&extrapool, &weightidx, sizeof (int32_t));
+ }
+ else
+ {
+ /* A single entry. Simply add the index and the length and
+ string (except for the first character which is already
+ tested for). */
+ int i;
-/* Prototypes for local functions. */
-static void collate_startup (struct linereader *ldfile,
- struct localedef_t *locale,
- struct charmap_t *charmap, int ignore_content);
+ added = (1 + 1 + runp->nwcs - 1) * sizeof (int32_t);
+ if (sizeof (int) == sizeof (int32_t))
+ obstack_make_room (&extrapool, added);
+ if (sizeof (int32_t) == sizeof (int))
+ {
+ obstack_int_grow_fast (&extrapool, weightidx);
+ obstack_int_grow_fast (&extrapool, runp->nwcs - 1);
+ }
+ else
+ {
+ int32_t l = runp->nwcs - 1;
+ obstack_grow (&extrapool, &weightidx, sizeof (int32_t));
+ obstack_grow (&extrapool, &l, sizeof (int32_t));
+ }
+ for (i = 1; i < runp->nwcs; ++i)
+ if (sizeof (int32_t) == sizeof (int))
+ obstack_int_grow_fast (&extrapool, runp->wcs[i]);
+ else
+ obstack_grow (&extrapool, &runp->wcs[i], sizeof (int32_t));
+ }
-static void
-collate_startup (struct linereader *ldfile, struct localedef_t *locale,
- struct charmap_t *charset, int ignore_content)
-{
- struct locale_collate_t *collate;
+ /* Next entry. */
+ lastp = runp;
+ runp = runp->wcnext;
+ }
+ while (runp != NULL);
+ }
- /* Allocate the needed room. */
- locale->categories[LC_COLLATE].collate = collate =
- (struct locale_collate_t *) xmalloc (sizeof (struct locale_collate_t));
+ /* Now add the four tables. */
+ assert (cnt == _NL_ITEM_INDEX (_NL_COLLATE_TABLEWC));
+ iov[2 + cnt].iov_base = tablewc;
+ iov[2 + cnt].iov_len = table_size * sizeof (int32_t);
+ idx[1 + cnt] = idx[cnt] + iov[2 + cnt].iov_len;
+ ++cnt;
+
+ assert (cnt == _NL_ITEM_INDEX (_NL_COLLATE_WEIGHTWC));
+ iov[2 + cnt].iov_len = obstack_object_size (&weightpool);
+ iov[2 + cnt].iov_base = obstack_finish (&weightpool);
+ idx[1 + cnt] = idx[cnt] + iov[2 + cnt].iov_len;
+ ++cnt;
+
+ assert (cnt == _NL_ITEM_INDEX (_NL_COLLATE_EXTRAWC));
+ iov[2 + cnt].iov_len = obstack_object_size (&extrapool);
+ iov[2 + cnt].iov_base = obstack_finish (&extrapool);
+ idx[1 + cnt] = idx[cnt] + iov[2 + cnt].iov_len;
+ ++cnt;
+
+ assert (cnt == _NL_ITEM_INDEX (_NL_COLLATE_INDIRECTWC));
+ iov[2 + cnt].iov_len = obstack_object_size (&indirectpool);
+ iov[2 + cnt].iov_base = obstack_finish (&indirectpool);
+ idx[1 + cnt] = idx[cnt] + iov[2 + cnt].iov_len;
+ ++cnt;
+
+
+ /* Finally write the table with collation element names out. It is
+ a hash table with a simple function which gets the name of the
+ character as the input. One character might have many names. The
+ value associated with the name is an index into the weight table
+ where we are then interested in the first-level weight value.
+
+ To determine how large the table should be we are counting the
+ elements have to put in. Since we are using internal chaining
+ using a secondary hash function we have to make the table a bit
+ larger to avoid extremely long search times. We can achieve
+ good results with a 40% larger table than there are entries. */
+ elem_size = 0;
+ runp = collate->start;
+ while (runp != NULL)
+ {
+ if (runp->mbs != NULL && runp->weights != NULL)
+ /* Yep, the element really counts. */
+ ++elem_size;
+
+ runp = runp->next;
+ }
+ /* Add 40% and find the next prime number. */
+ elem_size = MIN (next_prime (elem_size * 1.4), 257);
+
+ /* Allocate the table. Each entry consists of two words: the hash
+ value and an index in a secondary table which provides the index
+ into the weight table and the string itself (so that a match can
+ be determined). */
+ elem_table = (uint32_t *) obstack_alloc (&extrapool,
+ elem_size * 2 * sizeof (uint32_t));
+ memset (elem_table, '\0', elem_size * 2 * sizeof (uint32_t));
+
+ /* Now add the elements. */
+ runp = collate->start;
+ while (runp != NULL)
+ {
+ if (runp->mbs != NULL && runp->weights != NULL)
+ {
+ /* Compute the hash value of the name. */
+ uint32_t namelen = strlen (runp->name);
+ uint32_t hash = elem_hash (runp->name, namelen);
+ size_t idx = hash % elem_size;
+
+ if (elem_table[idx * 2] != 0)
+ {
+ /* The spot is already take. Try iterating using the value
+ from the secondary hashing function. */
+ size_t iter = hash % (elem_size - 2);
- /* Allocate hash table for collating elements. */
- if (init_hash (&collate->elements, 512))
- error (4, 0, _("memory exhausted"));
- collate->combine_token = NULL;
- obstack_init (&collate->element_mem);
+ do
+ {
+ idx += iter;
+ if (idx >= elem_size)
+ idx -= elem_size;
+ }
+ while (elem_table[idx * 2] != 0);
+
+ /* This is the spot where we will insert the value. */
+ elem_table[idx * 2] = hash;
+ elem_table[idx * 2 + 1] = obstack_object_size (&extrapool);
+
+ /* The the string itself including length. */
+ obstack_1grow (&extrapool, namelen);
+ obstack_grow (&extrapool, runp->name, namelen);
+
+ /* And the multibyte representation. */
+ obstack_1grow (&extrapool, runp->nmbs);
+ obstack_grow (&extrapool, runp->mbs, runp->nmbs);
+
+ /* And align again to 32 bits. */
+ if ((1 + namelen + 1 + runp->nmbs) % sizeof (int32_t) != 0)
+ obstack_grow (&extrapool, "\0\0",
+ (sizeof (int32_t)
+ - ((1 + namelen + 1 + runp->nmbs)
+ % sizeof (int32_t))));
+ }
+ }
- /* Allocate hash table for collating elements. */
- if (init_hash (&collate->symbols, 64))
- error (4, 0, _("memory exhausted"));
+ runp = runp->next;
+ }
- /* Allocate hash table for result. */
- if (init_hash (&collate->result, 512))
- error (4, 0, _("memory exhausted"));
+ /* Prepare to write out this data. */
+ assert (cnt == _NL_ITEM_INDEX (_NL_COLLATE_SYMB_HASH_SIZEMB));
+ iov[2 + cnt].iov_base = &elem_size;
+ iov[2 + cnt].iov_len = sizeof (int32_t);
+ idx[1 + cnt] = idx[cnt] + iov[2 + cnt].iov_len;
+ ++cnt;
- collate->nrules = 0;
- collate->nrules_max = 10;
- collate->rules
- = (enum coll_sort_rule *) xmalloc (collate->nrules_max
- * sizeof (enum coll_sort_rule));
+ assert (cnt == _NL_ITEM_INDEX (_NL_COLLATE_SYMB_TABLEMB));
+ iov[2 + cnt].iov_base = elem_table;
+ iov[2 + cnt].iov_len = elem_size * 2 * sizeof (int32_t);
+ idx[1 + cnt] = idx[cnt] + iov[2 + cnt].iov_len;
+ ++cnt;
- collate->order_cnt = 1; /* The smallest weight is 2. */
+ assert (cnt == _NL_ITEM_INDEX (_NL_COLLATE_SYMB_EXTRAMB));
+ iov[2 + cnt].iov_len = obstack_object_size (&extrapool);
+ iov[2 + cnt].iov_base = obstack_finish (&extrapool);
+ ++cnt;
- collate->was_ellipsis = 0;
- collate->last_char = L'\0'; /* 0 because leading ellipsis is allowed. */
- collate->all_patches = NULL;
+ assert (cnt == _NL_ITEM_INDEX (_NL_NUM_LC_COLLATE));
- /* This tells us no UNDEFINED entry was found until now. */
- memset (&collate->undefined, '\0', sizeof (collate->undefined));
+ write_locale_data (output_path, "LC_COLLATE", 2 + cnt, iov);
- ldfile->translate_strings = 0;
- ldfile->return_widestr = 0;
+ obstack_free (&weightpool, NULL);
+ obstack_free (&extrapool, NULL);
+ obstack_free (&indirectpool, NULL);
}
void
-collate_finish (struct localedef_t *locale, struct charset_t *charset,
- struct repertoire_t *repertoire)
+collate_read (struct linereader *ldfile, struct localedef_t *result,
+ struct charmap_t *charmap, const char *repertoire_name,
+ int ignore_content)
{
- struct locale_collate_t *collate = locale->categories[LC_COLLATE].collate;
- patch_t *patch;
- size_t cnt;
+ struct repertoire_t *repertoire = NULL;
+ struct locale_collate_t *collate;
+ struct token *now;
+ struct token *arg = NULL;
+ enum token_t nowtok;
+ int state = 0;
+ enum token_t was_ellipsis = tok_none;
+ struct localedef_t *copy_locale = NULL;
- /* Patch the constructed table so that forward references are
- correctly filled. */
- for (patch = collate->all_patches; patch != NULL; patch = patch->next)
- {
- uint32_t wch;
- size_t toklen = strlen (patch->token);
- void *ptmp;
- unsigned int value = 0;
+ /* Get the repertoire we have to use. */
+ if (repertoire_name != NULL)
+ repertoire = repertoire_read (repertoire_name);
- wch = charset_find_value (&charset->char_table, patch->token, toklen);
- if (wch != ILLEGAL_CHAR_VALUE)
- {
- element_t *runp;
-
- if (find_entry (&collate->result, &wch, sizeof (uint32_t),
- (void *) &runp) < 0)
- runp = NULL;
- for (; runp != NULL; runp = runp->next)
- if (runp->name[0] == wch && runp->name[1] == L'\0')
- break;
-
- value = runp == NULL ? 0 : runp->this_weight;
- }
- else if (find_entry (&collate->elements, patch->token, toklen, &ptmp)
- >= 0)
- {
- value = ((element_t *) ptmp)->this_weight;
- }
- else if (find_entry (&collate->symbols, patch->token, toklen, &ptmp)
- >= 0)
- {
- value = (unsigned long int) ptmp;
- }
- else
- value = 0;
+ /* The rest of the line containing `LC_COLLATE' must be free. */
+ lr_ignore_rest (ldfile, 1);
- if (value == 0)
- {
- if (!be_quiet)
- error_at_line (0, 0, patch->fname, patch->lineno,
- _("no weight defined for symbol `%s'"),
- patch->token);
- }
- else
- *patch->where.pos = value;
+ do
+ {
+ now = lr_token (ldfile, charmap, NULL);
+ nowtok = now->tok;
}
+ while (nowtok == tok_eol);
- /* If no definition for UNDEFINED is given, all characters in the
- given charset must be specified. */
- if (collate->undefined.ordering == NULL)
+ if (nowtok == tok_copy)
{
- /**************************************************************\
- |* XXX We should test whether really an unspecified character *|
- |* exists before giving the message. *|
- \**************************************************************/
- uint32_t weight;
+ state = 2;
+ now = lr_token (ldfile, charmap, NULL);
+ if (now->tok != tok_string)
+ {
+ SYNTAX_ERROR (_("%s: syntax error"), "LC_COLLATE");
+
+ skip_category:
+ do
+ now = lr_token (ldfile, charmap, NULL);
+ while (now->tok != tok_eof && now->tok != tok_end);
- if (!be_quiet)
- error (0, 0, _("no definition of `UNDEFINED'"));
+ if (now->tok != tok_eof
+ || (now = lr_token (ldfile, charmap, NULL), now->tok == tok_eof))
+ lr_error (ldfile, _("%s: premature end of file"), "LC_COLLATE");
+ else if (now->tok != tok_lc_collate)
+ {
+ lr_error (ldfile, _("\
+%1$s: definition does not end with `END %1$s'"), "LC_COLLATE");
+ lr_ignore_rest (ldfile, 0);
+ }
+ else
+ lr_ignore_rest (ldfile, 1);
- collate->undefined.ordering_len = collate->nrules;
- weight = ++collate->order_cnt;
+ return;
+ }
- for (cnt = 0; cnt < collate->nrules; ++cnt)
+ /* Get the locale definition. */
+ copy_locale = load_locale (LC_COLLATE, now->val.str.startmb,
+ repertoire_name, charmap);
+ if ((copy_locale->avail & COLLATE_LOCALE) == 0)
{
- uint32_t one = 1;
- obstack_grow (&collate->element_mem, &one, sizeof (one));
+ /* Not yet loaded. So do it now. */
+ if (locfile_read (copy_locale, charmap) != 0)
+ goto skip_category;
}
- for (cnt = 0; cnt < collate->nrules; ++cnt)
- obstack_grow (&collate->element_mem, &weight, sizeof (weight));
+ lr_ignore_rest (ldfile, 1);
- collate->undefined.ordering = obstack_finish (&collate->element_mem);
+ now = lr_token (ldfile, charmap, NULL);
+ nowtok = now->tok;
}
- collate->undefined_len = 2; /* For the name: 1 x uint32_t + L'\0'. */
- for (cnt = 0; cnt < collate->nrules; ++cnt)
- collate->undefined_len += 1 + collate->undefined.ordering[cnt];
-}
-
-
+ /* Prepare the data structures. */
+ collate_startup (ldfile, result, copy_locale, ignore_content);
+ collate = result->categories[LC_COLLATE].collate;
-void
-collate_output (struct localedef_t *locale, struct charset_t *charset,
- struct repertoire_t *repertoire, const char *output_path)
-{
- struct locale_collate_t *collate = locale->categories[LC_COLLATE].collate;
- uint32_t table_size, table_best, level_best, sum_best;
- void *last;
- element_t *pelem;
- uint32_t *name;
- size_t len;
- const size_t nelems = _NL_ITEM_INDEX (_NL_NUM_LC_COLLATE);
- struct iovec iov[2 + nelems];
- struct locale_file data;
- uint32_t idx[nelems];
- struct obstack non_simple;
- struct obstack string_pool;
- size_t cnt, entry_size;
- uint32_t undefined_offset = UINT_MAX;
- uint32_t *table, *extra, *table2, *extra2;
- size_t extra_len;
- uint32_t element_hash_tab_size;
- uint32_t *element_hash_tab;
- uint32_t *element_hash_tab_ob;
- uint32_t element_string_pool_size;
- char *element_string_pool;
- uint32_t element_value_size;
- uint32_t *element_value;
- uint32_t *element_value_ob;
- uint32_t symbols_hash_tab_size;
- uint32_t *symbols_hash_tab;
- uint32_t *symbols_hash_tab_ob;
- uint32_t symbols_string_pool_size;
- char *symbols_string_pool;
- uint32_t symbols_class_size;
- uint32_t *symbols_class;
- uint32_t *symbols_class_ob;
- hash_table *hash_tab;
- unsigned int dummy_weights[collate->nrules + 1];
-
- sum_best = UINT_MAX;
- table_best = 0xffff;
- level_best = 0xffff;
-
- /* Compute table size. */
- if (!be_quiet)
- fputs (_("\
-Computing table size for collation information might take a while..."),
- stderr);
- for (table_size = 256; table_size < sum_best; ++table_size)
+ while (1)
{
- size_t hits[table_size];
- unsigned int worst = 1;
- size_t cnt;
-
- last = NULL;
-
- for (cnt = 0; cnt < 256; ++cnt)
- hits[cnt] = 1;
- memset (&hits[256], '\0', sizeof (hits) - 256 * sizeof (size_t));
-
- while (iterate_table (&collate->result, &last, (const void **) &name,
- &len, (void **) &pelem) >= 0)
- if (pelem->ordering != NULL && pelem->name[0] > 0xff)
- if (++hits[(unsigned int) pelem->name[0] % table_size] > worst)
- {
- worst = hits[(unsigned int) pelem->name[0] % table_size];
- if (table_size * worst > sum_best)
- break;
- }
+ /* Of course we don't proceed beyond the end of file. */
+ if (nowtok == tok_eof)
+ break;
- if (table_size * worst < sum_best)
+ /* Ingore empty lines. */
+ if (nowtok == tok_eol)
{
- sum_best = table_size * worst;
- table_best = table_size;
- level_best = worst;
+ now = lr_token (ldfile, charmap, NULL);
+ nowtok = now->tok;
+ continue;
}
- }
- assert (table_best != 0xffff || level_best != 0xffff);
- if (!be_quiet)
- fputs (_(" done\n"), stderr);
-
- obstack_init (&non_simple);
- obstack_init (&string_pool);
-
- data.magic = LIMAGIC (LC_COLLATE);
- data.n = nelems;
- iov[0].iov_base = (void *) &data;
- iov[0].iov_len = sizeof (data);
-
- iov[1].iov_base = (void *) idx;
- iov[1].iov_len = sizeof (idx);
-
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_NRULES)].iov_base = &collate->nrules;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_NRULES)].iov_len = sizeof (uint32_t);
-
- table = (uint32_t *) alloca (collate->nrules * sizeof (uint32_t));
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_RULES)].iov_base = table;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_RULES)].iov_len
- = collate->nrules * sizeof (uint32_t);
- /* Another trick here. Describing the collation method needs only a
- few bits (3, to be exact). But the binary file should be
- accessible by machines with both endianesses and so we store both
- forms in the same word. */
- for (cnt = 0; cnt < collate->nrules; ++cnt)
- table[cnt] = collate->rules[cnt] | bswap_32 (collate->rules[cnt]);
-
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_HASH_SIZE)].iov_base = &table_best;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_HASH_SIZE)].iov_len = sizeof (uint32_t);
-
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_HASH_LAYERS)].iov_base = &level_best;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_HASH_LAYERS)].iov_len
- = sizeof (uint32_t);
-
- entry_size = 1 + MAX (collate->nrules, 2);
-
- table = (uint32_t *) alloca (table_best * level_best * entry_size
- * sizeof (table[0]));
- memset (table, '\0', table_best * level_best * entry_size
- * sizeof (table[0]));
-
-
- /* Macros for inserting in output table. */
-#define ADD_VALUE(expr) \
- do { \
- uint32_t to_write = (uint32_t) expr; \
- obstack_grow (&non_simple, &to_write, sizeof (to_write)); \
- } while (0)
-
-#define ADD_ELEMENT(pelem, len) \
- do { \
- size_t cnt, idx; \
- \
- ADD_VALUE (len); \
- \
- wlen = wcslen (pelem->name); \
- obstack_grow (&non_simple, pelem->name, (wlen + 1) * sizeof (uint32_t)); \
- \
- idx = collate->nrules; \
- for (cnt = 0; cnt < collate->nrules; ++cnt) \
- { \
- size_t disp; \
- \
- ADD_VALUE (pelem->ordering[cnt]); \
- for (disp = 0; disp < pelem->ordering[cnt]; ++disp) \
- ADD_VALUE (pelem->ordering[idx++]); \
- } \
- } while (0)
-
-#define ADD_FORWARD(pelem) \
- do { \
- /* We leave a reference in the main table and put all \
- information in the table for the extended entries. */ \
- element_t *runp; \
- element_t *has_simple = NULL; \
- size_t wlen; \
- \
- table[(level * table_best + slot) * entry_size + 1] \
- = FORWARD_CHAR; \
- table[(level * table_best + slot) * entry_size + 2] \
- = obstack_object_size (&non_simple) / sizeof (uint32_t); \
- \
- /* Here we have to construct the non-simple table entry. First \
- compute the total length of this entry. */ \
- for (runp = (pelem); runp != NULL; runp = runp->next) \
- if (runp->ordering != NULL) \
- { \
- uint32_t value; \
- size_t cnt; \
- \
- value = 1 + wcslen (runp->name) + 1; \
- \
- for (cnt = 0; cnt < collate->nrules; ++cnt) \
- /* We have to take care for entries without ordering \
- information. While reading them they get inserted in the \
- table and later not removed when something goes wrong with \
- reading its weights. */ \
- value += 1 + runp->ordering[cnt]; \
- \
- if (runp->name[1] == L'\0') \
- has_simple = runp; \
- \
- ADD_ELEMENT (runp, value); \
- } \
- \
- if (has_simple == NULL) \
- { \
- size_t idx, cnt; \
- \
- ADD_VALUE (collate->undefined_len + 1); \
- \
- /* Add the name. */ \
- ADD_VALUE ((pelem)->name[0]); \
- ADD_VALUE (0); \
- \
- idx = collate->nrules; \
- for (cnt = 0; cnt < collate->nrules; ++cnt) \
- { \
- size_t disp; \
- \
- ADD_VALUE (collate->undefined.ordering[cnt]); \
- for (disp = 0; disp < collate->undefined.ordering[cnt]; ++disp) \
- { \
- if ((uint32_t) collate->undefined.ordering[idx] \
- == ELLIPSIS_CHAR) \
- ADD_VALUE ((pelem)->name[0]); \
- else \
- ADD_VALUE (collate->undefined.ordering[idx++]); \
- ++idx; \
- } \
- } \
- } \
- } while (0)
-
-
-
- /* Fill the table now. First we look for all the characters which
- fit into one single byte. This speeds up the 8-bit string
- functions. */
- last = NULL;
- while (iterate_table (&collate->result, &last, (const void **) &name,
- &len, (void **) &pelem) >= 0)
- if (pelem->name[0] <= 0xff)
- {
- /* We have a single byte name. Now we must distinguish
- between entries in simple form (i.e., only one value per
- weight and no collation element starting with the same
- character) and those which are not. */
- size_t slot = ((size_t) pelem->name[0]);
- const size_t level = 0;
-
- table[slot * entry_size] = pelem->name[0];
-
- if (pelem->name[1] == L'\0' && pelem->next == NULL
- && pelem->ordering_len == collate->nrules)
- {
- /* Yes, we have a simple one. Lucky us. */
- size_t cnt;
-
- for (cnt = 0; cnt < collate->nrules; ++cnt)
- table[slot * entry_size + 1 + cnt]
- = pelem->ordering[collate->nrules + cnt];
- }
- else
- ADD_FORWARD (pelem);
- }
-
- /* Now check for missing single byte entries. If one exist we fill
- with the UNDEFINED entry. */
- for (cnt = 0; cnt < 256; ++cnt)
- /* The first weight is never 0 for existing entries. */
- if (table[cnt * entry_size + 1] == 0)
- {
- /* We have to fill in the information from the UNDEFINED
- entry. */
- table[cnt * entry_size] = (uint32_t) cnt;
-
- if (collate->undefined.ordering_len == collate->nrules)
- {
- size_t inner;
-
- for (inner = 0; inner < collate->nrules; ++inner)
- if ((uint32_t)collate->undefined.ordering[collate->nrules
- + inner]
- == ELLIPSIS_CHAR)
- table[cnt * entry_size + 1 + inner] = cnt;
- else
- table[cnt * entry_size + 1 + inner]
- = collate->undefined.ordering[collate->nrules + inner];
- }
- else
- {
- if (undefined_offset != UINT_MAX)
- {
- table[cnt * entry_size + 1] = FORWARD_CHAR;
- table[cnt * entry_size + 2] = undefined_offset;
- }
- else
- {
- const size_t slot = cnt;
- const size_t level = 0;
-
- ADD_FORWARD (&collate->undefined);
- undefined_offset = table[cnt * entry_size + 2];
- }
- }
- }
-
- /* Now we are ready for inserting the whole rest. */
- last = NULL;
- while (iterate_table (&collate->result, &last, (const void **) &name,
- &len, (void **) &pelem) >= 0)
- if (pelem->name[0] > 0xff)
- {
- /* Find the position. */
- size_t slot = ((size_t) pelem->name[0]) % table_best;
- size_t level = 0;
-
- while (table[(level * table_best + slot) * entry_size + 1] != 0)
- ++level;
- assert (level < level_best);
-
- if (pelem->name[1] == L'\0' && pelem->next == NULL
- && pelem->ordering_len == collate->nrules)
- {
- /* Again a simple entry. */
- size_t inner;
-
- for (inner = 0; inner < collate->nrules; ++inner)
- table[(level * table_best + slot) * entry_size + 1 + inner]
- = pelem->ordering[collate->nrules + inner];
- }
- else
- ADD_FORWARD (pelem);
- }
- /* Add the UNDEFINED entry. */
- {
- /* Here we have to construct the non-simple table entry. */
- size_t idx, cnt;
-
- undefined_offset = obstack_object_size (&non_simple);
-
- idx = collate->nrules;
- for (cnt = 0; cnt < collate->nrules; ++cnt)
- {
- size_t disp;
-
- ADD_VALUE (collate->undefined.ordering[cnt]);
- for (disp = 0; disp < collate->undefined.ordering[cnt]; ++disp)
- ADD_VALUE (collate->undefined.ordering[idx++]);
- }
- }
-
- /* Finish the extra block. */
- extra_len = obstack_object_size (&non_simple);
- extra = (uint32_t *) obstack_finish (&non_simple);
- assert ((extra_len % sizeof (uint32_t)) == 0);
-
- /* Now we have to build the two array for the other byte ordering. */
- table2 = (uint32_t *) alloca (table_best * level_best * entry_size
- * sizeof (table[0]));
- extra2 = (uint32_t *) alloca (extra_len);
-
- for (cnt = 0; cnt < table_best * level_best * entry_size; ++cnt)
- table2[cnt] = bswap_32 (table[cnt]);
-
- for (cnt = 0; cnt < extra_len / sizeof (uint32_t); ++cnt)
- extra2[cnt] = bswap_32 (extra2[cnt]);
-
- /* We need a simple hashing table to get a collation-element->chars
- mapping. We again use internal hashing using a secondary hashing
- function.
-
- Each string has an associate hashing value V, computed by a
- fixed function. To locate the string we use open addressing with
- double hashing. The first index will be V % M, where M is the
- size of the hashing table. If no entry is found, iterating with
- a second, independent hashing function takes place. This second
- value will be 1 + V % (M - 2). The approximate number of probes
- will be
-
- for unsuccessful search: (1 - N / M) ^ -1
- for successful search: - (N / M) ^ -1 * ln (1 - N / M)
-
- where N is the number of keys.
-
- If we now choose M to be the next prime bigger than 4 / 3 * N,
- we get the values 4 and 1.85 resp. Because unsuccessful searches
- are unlikely this is a good value. Formulas: [Knuth, The Art of
- Computer Programming, Volume 3, Sorting and Searching, 1973,
- Addison Wesley] */
- if (collate->elements.filled == 0)
- {
- /* We don't need any element table since there are no collating
- elements. */
- element_hash_tab_size = 0;
- element_hash_tab = NULL;
- element_hash_tab_ob = NULL;
- element_string_pool_size = 0;
- element_string_pool = NULL;
- element_value_size = 0;
- element_value = NULL;
- element_value_ob = NULL;
- }
- else
- {
- void *ptr; /* Running pointer. */
- const char *key; /* Key for current bucket. */
- size_t keylen; /* Length of key data. */
- const element_t *data; /* Data, i.e., the character sequence. */
-
- element_hash_tab_size = next_prime ((collate->elements.filled * 4) / 3);
- if (element_hash_tab_size < 7)
- /* We need a minimum to make the following code work. */
- element_hash_tab_size = 7;
-
- element_hash_tab = obstack_alloc (&non_simple, (2 * element_hash_tab_size
- * sizeof (uint32_t)));
- memset (element_hash_tab, '\377', (2 * element_hash_tab_size
- * sizeof (uint32_t)));
-
- ptr = NULL;
- while (iterate_table (&collate->elements, &ptr, (const void **) &key,
- &keylen, (void **) &data) == 0)
+ switch (nowtok)
{
- size_t hash_val = hash_string (key, keylen);
- size_t idx = hash_val % element_hash_tab_size;
-
- if (element_hash_tab[2 * idx] != (~((uint32_t) 0)))
+ case tok_coll_weight_max:
+ /* Ignore the rest of the line if we don't need the input of
+ this line. */
+ if (ignore_content)
{
- /* We need the second hashing function. */
- size_t c = 1 + (hash_val % (element_hash_tab_size - 2));
-
- do
- if (idx >= element_hash_tab_size - c)
- idx -= element_hash_tab_size - c;
- else
- idx += c;
- while (element_hash_tab[2 * idx] != (~((uint32_t) 0)));
+ lr_ignore_rest (ldfile, 0);
+ break;
}
- element_hash_tab[2 * idx] = obstack_object_size (&non_simple);
- element_hash_tab[2 * idx + 1] = (obstack_object_size (&string_pool)
- / sizeof (uint32_t));
+ if (state != 0)
+ goto err_label;
- obstack_grow0 (&non_simple, key, keylen);
- obstack_grow (&string_pool, data->name,
- (wcslen (data->name) + 1) * sizeof (uint32_t));
- }
+ arg = lr_token (ldfile, charmap, NULL);
+ if (arg->tok != tok_number)
+ goto err_label;
+ if (collate->col_weight_max != -1)
+ lr_error (ldfile, _("%s: duplicate definition of `%s'"),
+ "LC_COLLATE", "col_weight_max");
+ else
+ collate->col_weight_max = arg->val.num;
+ lr_ignore_rest (ldfile, 1);
+ break;
- if (obstack_object_size (&non_simple) % 4 != 0)
- obstack_blank (&non_simple,
- 4 - (obstack_object_size (&non_simple) % 4));
- element_string_pool_size = obstack_object_size (&non_simple);
- element_string_pool = obstack_finish (&non_simple);
-
- element_value_size = obstack_object_size (&string_pool);
- element_value = obstack_finish (&string_pool);
-
- /* Create the tables for the other byte order. */
- element_hash_tab_ob = obstack_alloc (&non_simple,
- (2 * element_hash_tab_size
- * sizeof (uint32_t)));
- for (cnt = 0; cnt < 2 * element_hash_tab_size; ++cnt)
- element_hash_tab_ob[cnt] = bswap_U32 (element_hash_tab[cnt]);
-
- element_value_ob = obstack_alloc (&string_pool, element_value_size);
- for (cnt = 0; cnt < element_value_size / 4; ++cnt)
- element_value_ob[cnt] = bswap_32 (element_value[cnt]);
- }
+ case tok_section_symbol:
+ /* Ignore the rest of the line if we don't need the input of
+ this line. */
+ if (ignore_content)
+ {
+ lr_ignore_rest (ldfile, 0);
+ break;
+ }
- /* Store collation elements as map to collation class. There are
- three kinds of symbols:
- - simple characters
- - collation elements
- - collation symbols
- We need to make a table which lets the user to access the primary
- weight based on the symbol string. */
- symbols_hash_tab_size = next_prime ((4 * (charset->char_table.filled
- + collate->elements.filled
- + collate->symbols.filled)) / 3);
- symbols_hash_tab = obstack_alloc (&non_simple, (2 * symbols_hash_tab_size
- * sizeof (uint32_t)));
- memset (symbols_hash_tab, '\377', (2 * symbols_hash_tab_size
- * sizeof (uint32_t)));
-
- /* Now fill the array. First the symbols from the character set,
- then the collation elements and last the collation symbols. */
- hash_tab = &charset->char_table;
- while (1)
- {
- void *ptr; /* Running pointer. */
- const char *key; /* Key for current bucket. */
- size_t keylen; /* Length of key data. */
- void *data; /* Data. */
-
- ptr = NULL;
- while (iterate_table (hash_tab, &ptr, (const void **) &key,
- &keylen, (void **) &data) == 0)
- {
- size_t hash_val;
- size_t idx;
- uint32_t word;
- unsigned int *weights;
+ if (state != 0)
+ goto err_label;
- if (hash_tab == &charset->char_table
- || hash_tab == &collate->elements)
+ arg = lr_token (ldfile, charmap, repertoire);
+ if (arg->tok != tok_bsymbol)
+ goto err_label;
+ else if (!ignore_content)
{
- element_t *lastp, *firstp;
- uint32_t dummy_name[2];
- const uint32_t *name;
- size_t name_len;
+ /* Check whether this section is already known. */
+ struct section_list *known = collate->sections;
+ while (known != NULL)
+ if (strcmp (known->name, arg->val.str.startmb) == 0)
+ break;
- if (hash_tab == &charset->char_table)
- {
- dummy_name[0] = (uint32_t) ((unsigned long int) data);
- dummy_name[1] = L'\0';
- name = dummy_name;
- name_len = sizeof (uint32_t);
- }
- else
+ if (known != NULL)
{
- element_t *elemp = (element_t *) data;
- name = elemp->name;
- name_len = wcslen (name) * sizeof (uint32_t);
+ lr_error (ldfile,
+ _("%s: duplicate declaration of section `%s'"),
+ "LC_COLLATE", arg->val.str.startmb);
+ free (arg->val.str.startmb);
}
-
- /* First check whether this character is used at all. */
- if (find_entry (&collate->result, name, name_len,
- (void *) &firstp) < 0)
- /* The symbol is not directly mentioned in the collation.
- I.e., we use the value for UNDEFINED. */
- lastp = &collate->undefined;
else
- {
- /* The entry for the simple character is always found at
- the end. */
- lastp = firstp;
- while (lastp->next != NULL && wcscmp (name, lastp->name))
- lastp = lastp->next;
- }
+ collate->sections = make_seclist_elem (collate,
+ arg->val.str.startmb,
+ collate->sections);
- weights = lastp->ordering;
+ lr_ignore_rest (ldfile, known == NULL);
}
else
{
- dummy_weights[0] = 1;
- dummy_weights[collate->nrules]
- = (unsigned int) ((unsigned long int) data);
-
- weights = dummy_weights;
- }
-
- /* In LASTP->ordering we now have the collation class.
- Determine the place in the hashing table next. */
- hash_val = hash_string (key, keylen);
- idx = hash_val % symbols_hash_tab_size;
-
- if (symbols_hash_tab[2 * idx] != (~((uint32_t) 0)))
- {
- /* We need the second hashing function. */
- size_t c = 1 + (hash_val % (symbols_hash_tab_size - 2));
-
- do
- if (idx >= symbols_hash_tab_size - c)
- idx -= symbols_hash_tab_size - c;
- else
- idx += c;
- while (symbols_hash_tab[2 * idx] != (~((uint32_t) 0)));
+ free (arg->val.str.startmb);
+ lr_ignore_rest (ldfile, 0);
}
+ break;
- symbols_hash_tab[2 * idx] = obstack_object_size (&string_pool);
- symbols_hash_tab[2 * idx + 1] = (obstack_object_size (&non_simple)
- / sizeof (uint32_t));
-
- obstack_grow0 (&string_pool, key, keylen);
- /* Adding the first weight looks complicated. We have to deal
- with the kind it is stored and with the fact that original
- form uses `unsigned int's while we need `uint32_t' here. */
- word = weights[0];
- obstack_grow (&non_simple, &word, sizeof (uint32_t));
- for (cnt = 0; cnt < weights[0]; ++cnt)
+ case tok_collating_element:
+ /* Ignore the rest of the line if we don't need the input of
+ this line. */
+ if (ignore_content)
{
- word = weights[collate->nrules + cnt];
- obstack_grow (&non_simple, &word, sizeof (uint32_t));
+ lr_ignore_rest (ldfile, 0);
+ break;
}
- }
-
- if (hash_tab == &charset->char_table)
- hash_tab = &collate->elements;
- else if (hash_tab == &collate->elements)
- hash_tab = &collate->symbols;
- else
- break;
- }
-
- /* Now we have the complete tables. */
- if (obstack_object_size (&string_pool) % 4 != 0)
- obstack_blank (&non_simple, 4 - (obstack_object_size (&string_pool) % 4));
- symbols_string_pool_size = obstack_object_size (&string_pool);
- symbols_string_pool = obstack_finish (&string_pool);
-
- symbols_class_size = obstack_object_size (&non_simple);
- symbols_class = obstack_finish (&non_simple);
- /* Generate tables with other byte order. */
- symbols_hash_tab_ob = obstack_alloc (&non_simple, (2 * symbols_hash_tab_size
- * sizeof (uint32_t)));
- for (cnt = 0; cnt < 2 * symbols_hash_tab_size; ++cnt)
- symbols_hash_tab_ob[cnt] = bswap_32 (symbols_hash_tab[cnt]);
-
- symbols_class_ob = obstack_alloc (&non_simple, symbols_class_size);
- for (cnt = 0; cnt < symbols_class_size / 4; ++cnt)
- symbols_class_ob[cnt] = bswap_32 (symbols_class[cnt]);
-
-
- /* Store table addresses and lengths. */
-#if __BYTE_ORDER == __BIG_ENDIAN
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_TABLE_EB)].iov_base = table;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_TABLE_EB)].iov_len
- = table_best * level_best * entry_size * sizeof (table[0]);
+ if (state != 0)
+ goto err_label;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_TABLE_EL)].iov_base = table2;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_TABLE_EL)].iov_len
- = table_best * level_best * entry_size * sizeof (table[0]);
+ arg = lr_token (ldfile, charmap, repertoire);
+ if (arg->tok != tok_bsymbol)
+ goto err_label;
+ else
+ {
+ const char *symbol = arg->val.str.startmb;
+ size_t symbol_len = arg->val.str.lenmb;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_EXTRA_EB)].iov_base = extra;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_EXTRA_EB)].iov_len = extra_len;
+ /* Next the `from' keyword. */
+ arg = lr_token (ldfile, charmap, repertoire);
+ if (arg->tok != tok_from)
+ {
+ free ((char *) symbol);
+ goto err_label;
+ }
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_EXTRA_EL)].iov_base = extra2;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_EXTRA_EL)].iov_len = extra_len;
-#else
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_TABLE_EB)].iov_base = table2;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_TABLE_EB)].iov_len
- = table_best * level_best * entry_size * sizeof (table[0]);
+ ldfile->return_widestr = 1;
+ ldfile->translate_strings = 1;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_TABLE_EL)].iov_base = table;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_TABLE_EL)].iov_len
- = table_best * level_best * entry_size * sizeof (table[0]);
+ /* Finally the string with the replacement. */
+ arg = lr_token (ldfile, charmap, repertoire);
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_EXTRA_EB)].iov_base = extra2;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_EXTRA_EB)].iov_len = extra_len;
+ ldfile->return_widestr = 0;
+ ldfile->translate_strings = 0;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_EXTRA_EL)].iov_base = extra;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_EXTRA_EL)].iov_len = extra_len;
-#endif
+ if (arg->tok != tok_string)
+ goto err_label;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_UNDEFINED)].iov_base = &undefined_offset;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_UNDEFINED)].iov_len = sizeof (uint32_t);
-
-
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_HASH_SIZE)].iov_base
- = &element_hash_tab_size;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_HASH_SIZE)].iov_len
- = sizeof (uint32_t);
-
-#if __BYTE_ORDER == __BIG_ENDIAN
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_HASH_EB)].iov_base
- = element_hash_tab;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_HASH_EB)].iov_len
- = 2 * element_hash_tab_size * sizeof (uint32_t);
-
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_HASH_EL)].iov_base
- = element_hash_tab_ob;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_HASH_EL)].iov_len
- = 2 * element_hash_tab_size * sizeof (uint32_t);
-#else
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_HASH_EL)].iov_base
- = element_hash_tab;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_HASH_EL)].iov_len
- = 2 * element_hash_tab_size * sizeof (uint32_t);
-
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_HASH_EB)].iov_base
- = element_hash_tab_ob;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_HASH_EB)].iov_len
- = 2 * element_hash_tab_size * sizeof (uint32_t);
-#endif
+ if (!ignore_content && symbol != NULL)
+ {
+ /* The name is already defined. */
+ if (check_duplicate (ldfile, collate, charmap,
+ repertoire, symbol, symbol_len))
+ goto col_elem_free;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_STR_POOL)].iov_base
- = element_string_pool;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_STR_POOL)].iov_len
- = element_string_pool_size;
-
-#if __BYTE_ORDER == __BIG_ENDIAN
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_VAL_EB)].iov_base
- = element_value;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_VAL_EB)].iov_len
- = element_value_size;
-
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_VAL_EL)].iov_base
- = element_value_ob;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_VAL_EL)].iov_len
- = element_value_size;
-#else
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_VAL_EL)].iov_base
- = element_value;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_VAL_EL)].iov_len
- = element_value_size;
-
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_VAL_EB)].iov_base
- = element_value_ob;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_ELEM_VAL_EB)].iov_len
- = element_value_size;
-#endif
-
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_HASH_SIZE)].iov_base
- = &symbols_hash_tab_size;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_HASH_SIZE)].iov_len
- = sizeof (uint32_t);
-
-#if __BYTE_ORDER == __BIG_ENDIAN
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_HASH_EB)].iov_base
- = symbols_hash_tab;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_HASH_EB)].iov_len
- = 2 * symbols_hash_tab_size * sizeof (uint32_t);
-
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_HASH_EL)].iov_base
- = symbols_hash_tab_ob;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_HASH_EL)].iov_len
- = 2 * symbols_hash_tab_size * sizeof (uint32_t);
-#else
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_HASH_EL)].iov_base
- = symbols_hash_tab;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_HASH_EL)].iov_len
- = 2 * symbols_hash_tab_size * sizeof (uint32_t);
-
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_HASH_EB)].iov_base
- = symbols_hash_tab_ob;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_HASH_EB)].iov_len
- = 2 * symbols_hash_tab_size * sizeof (uint32_t);
-#endif
-
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_STR_POOL)].iov_base
- = symbols_string_pool;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_STR_POOL)].iov_len
- = symbols_string_pool_size;
-
-#if __BYTE_ORDER == __BIG_ENDIAN
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_CLASS_EB)].iov_base
- = symbols_class;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_CLASS_EB)].iov_len
- = symbols_class_size;
-
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_CLASS_EL)].iov_base
- = symbols_class_ob;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_CLASS_EL)].iov_len
- = symbols_class_size;
-#else
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_CLASS_EL)].iov_base
- = symbols_class;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_CLASS_EL)].iov_len
- = symbols_class_size;
-
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_CLASS_EB)].iov_base
- = symbols_class_ob;
- iov[2 + _NL_ITEM_INDEX (_NL_COLLATE_SYMB_CLASS_EB)].iov_len
- = symbols_class_size;
-#endif
-
- /* Update idx array. */
- idx[0] = iov[0].iov_len + iov[1].iov_len;
- for (cnt = 1; cnt < nelems; ++cnt)
- idx[cnt] = idx[cnt - 1] + iov[1 + cnt].iov_len;
-
- write_locale_data (output_path, "LC_COLLATE", 2 + nelems, iov);
-
- obstack_free (&non_simple, NULL);
- obstack_free (&string_pool, NULL);
-}
-
-
-static int
-collate_element_to (struct linereader *ldfile,
- struct locale_collate_t *collate,
- struct token *code, struct charmap_t *charmap,
- struct repertoire_t *repertoire)
-{
- struct charseq *seq;
- uint32_t value;
- void *not_used;
-
- seq = charmap_find_value (charmap, code->val.str.start, code->val.str.len);
- if (seq != NULL)
- {
- lr_error (ldfile, _("symbol for multicharacter collating element "
- "`%.*s' duplicates symbolic name in charmap"),
- (int) code->val.str.len, code->val.str.start);
- return 1;
- }
-
- value = repertoire_find_value (repertoire, code->val.str.start,
- code->val.str.len);
- if (value != ILLEGAL_CHAR_VALUE)
- {
- lr_error (ldfile, _("symbol for multicharacter collating element "
- "`%.*s' duplicates symbolic name in repertoire"),
- (int) code->val.str.len, code->val.str.start);
- return 1;
- }
-
- if (find_entry (&collate->elements, code->val.str.start, code->val.str.len,
- ¬_used) >= 0)
- {
- lr_error (ldfile, _("symbol for multicharacter collating element "
- "`%.*s' duplicates other element definition"),
- (int) code->val.str.len, code->val.str.start);
- return 1;
- }
-
- if (find_entry (&collate->elements, code->val.str.start, code->val.str.len,
- ¬_used) >= 0)
- {
- lr_error (ldfile, _("symbol for multicharacter collating element "
- "`%.*s' duplicates symbol definition"),
- (int) code->val.str.len, code->val.str.start);
- return 1;
- }
-
- return 0;
-}
-
-
-static void
-collate_element_from (struct linereader *ldfile,
- struct locale_collate_t *collate,
- const char *to_str, struct token *code,
- struct charmap_t *charmap,
- struct repertoire_t *repertoire)
-{
- element_t *elemp, *runp;
-
- /* CODE is a string. */
- elemp = (element_t *) obstack_alloc (&collate->element_mem,
- sizeof (element_t));
-
- /* We have to translate the string. It may contain <...> character
- names. */
- elemp->namemb = code->val.str.startmb;
- elemp->namewc = code->val.str.startwc;
- elemp->this_weight = 0;
- elemp->ordering = NULL;
- elemp->ordering_len = 0;
-
- if (elemp->namemb == NULL && elemp->namewc == NULL)
- {
- /* The string contains characters which are not in the charmap nor
- in the repertoire. Ignore the string. */
- if (verbose)
- lr_error (ldfile, _("\
-`from' string in collation element declaration contains unknown character"));
- return;
- }
-
- /* The entries in the linked lists of RESULT are sorting in
- descending order. The order is important for the `strcoll' and
- `wcscoll' functions. */
- if (find_entry (&collate->resultwc, elemp->namewc, sizeof (uint32_t),
- (void *) &runp) >= 0)
- {
- /* We already have an entry with this key. Check whether it is
- identical. */
- element_t *prevp = NULL;
- int cmpres;
-
- do
- {
- cmpres = wcscmp (elemp->namewc, runp->namewc);
- if (cmpres <= 0)
- break;
- prevp = runp;
- }
- while ((runp = runp->next) != NULL);
+ if (insert_entry (&collate->elem_table,
+ symbol, symbol_len,
+ new_element (collate,
+ arg->val.str.startmb,
+ arg->val.str.lenmb - 1,
+ arg->val.str.startwc,
+ symbol, symbol_len, 0)) < 0)
+ lr_error (ldfile, _("\
+error while adding collating element"));
+ }
+ else
+ {
+ col_elem_free:
+ if (symbol != NULL)
+ free ((char *) symbol);
+ if (arg->val.str.startmb != NULL)
+ free (arg->val.str.startmb);
+ if (arg->val.str.startwc != NULL)
+ free (arg->val.str.startwc);
+ }
+ lr_ignore_rest (ldfile, 1);
+ }
+ break;
- if (cmpres == 0)
- lr_error (ldfile, _("\
-duplicate collating element definition (repertoire)"));
- else
- {
- elemp->next = runp;
- if (prevp == NULL)
+ case tok_collating_symbol:
+ /* Ignore the rest of the line if we don't need the input of
+ this line. */
+ if (ignore_content)
{
- if (set_entry (&collate->resultwc, elemp->namewc,
- sizeof (uint32_t), elemp) < 0)
- error (EXIT_FAILURE, 0, _("\
-error while inserting collation element into hash table"));
+ lr_ignore_rest (ldfile, 0);
+ break;
}
- else
- prevp->next = elemp;
- }
- }
- else
- {
- elemp->next = NULL;
- if (insert_entry (&collate->resultwc, elemp->namewc, sizeof (uint32_t),
- elemp) < 0)
- error (EXIT_FAILURE, errno, _("error while inserting to hash table"));
- }
- /* Now also insert the element definition in the multibyte table. */
- if (find_entry (&collate->resultmb, elemp->namemb, 1, (void *) &runp) >= 0)
- {
- /* We already have an entry with this key. Check whether it is
- identical. */
- element_t *prevp = NULL;
- int cmpres;
-
- do
- {
- cmpres = strcmp (elemp->namemb, runp->namemb);
- if (cmpres <= 0)
- break;
- prevp = runp;
- }
- while ((runp = runp->next) != NULL);
+ if (state != 0)
+ goto err_label;
- if (cmpres == 0)
- lr_error (ldfile, _("\
-duplicate collating element definition (charmap)"));
- else
- {
- elemp->next = runp;
- if (prevp == NULL)
- {
- if (set_entry (&collate->resultmb, elemp->namemb, 1, elemp) < 0)
- error (EXIT_FAILURE, 0, _("\
-error while inserting collation element into hash table"));
- }
+ arg = lr_token (ldfile, charmap, repertoire);
+ if (arg->tok != tok_bsymbol)
+ goto err_label;
else
- prevp->next = elemp;
- }
- }
- else
- {
- elemp->next = NULL;
- if (insert_entry (&collate->resultmb, elemp->namemb, 1, elemp) < 0)
- error (EXIT_FAILURE, errno, _("error while inserting to hash table"));
- }
-
- /* Finally install the mapping from the `to'-name to the `from'-name. */
- if (insert_entry (&collate->elements, to_str, strlen (to_str),
- (void *) elemp) < 0)
- lr_error (ldfile, _("cannot insert new collating symbol definition: %s"),
- strerror (errno));
-}
+ {
+ const char *symbol = arg->val.str.startmb;
+ size_t symbol_len = arg->val.str.lenmb;
+ if (!ignore_content)
+ {
+ if (symbol == NULL)
+ lr_error (ldfile, _("\
+%s: unknown character in collating symbol name"),
+ "LC_COLLATE");
+ else
+ {
+ /* The name is already defined. */
+ if (check_duplicate (ldfile, collate, charmap,
+ repertoire, symbol, symbol_len))
+ goto col_sym_free;
-static void
-collate_symbol (struct linereader *ldfile, struct locale_collate_t *collate,
- struct token *code, struct charmap_t *charmap,
- struct repertoire_t *repertoire)
-{
- uint32_t value;
- struct charseq *seq;
- void *not_used;
+ if (insert_entry (&collate->sym_table,
+ symbol, symbol_len,
+ new_symbol (collate)) < 0)
+ lr_error (ldfile, _("\
+error while adding collating symbol"));
+ }
+ }
+ else
+ {
+ col_sym_free:
+ if (symbol != NULL)
+ free ((char *) symbol);
+ }
+ lr_ignore_rest (ldfile, 1);
+ }
+ break;
- seq = charset_find_value (charmap, code->val.str.start, code->val.str.len);
- if (seq != NULL)
- {
- lr_error (ldfile, _("symbol for multicharacter collating element "
- "`%.*s' duplicates symbolic name in charmap"),
- (int) code->val.str.len, code->val.str.start);
- return;
- }
+ case tok_symbol_equivalence:
+ /* Ignore the rest of the line if we don't need the input of
+ this line. */
+ if (ignore_content)
+ {
+ lr_ignore_rest (ldfile, 0);
+ break;
+ }
- value = repertoire (repertoire, code->val.str.start, code->val.str.len);
- if (value != ILLEGAL_CHAR_VALUE)
- {
- lr_error (ldfile, _("symbol for multicharacter collating element "
- "`%.*s' duplicates symbolic name in repertoire"),
- (int) code->val.str.len, code->val.str.start);
- return;
- }
+ if (state != 0)
+ goto err_label;
- if (find_entry (&collate->elements, code->val.str.start, code->val.str.len,
- ¬_used) >= 0)
- {
- lr_error (ldfile, _("symbol for multicharacter collating element "
- "`%.*s' duplicates element definition"),
- (int) code->val.str.len, code->val.str.start);
- return;
- }
+ arg = lr_token (ldfile, charmap, repertoire);
+ if (arg->tok != tok_bsymbol)
+ goto err_label;
+ else
+ {
+ const char *newname = arg->val.str.startmb;
+ size_t newname_len = arg->val.str.lenmb;
+ const char *symname;
+ size_t symname_len;
+ struct symbol_t *symval;
- if (find_entry (&collate->symbols, code->val.str.start, code->val.str.len,
- ¬_used) >= 0)
- {
- lr_error (ldfile, _("symbol for multicharacter collating element "
- "`%.*s' duplicates other symbol definition"),
- (int) code->val.str.len, code->val.str.start);
- return;
- }
+ arg = lr_token (ldfile, charmap, repertoire);
+ if (arg->tok != tok_bsymbol)
+ {
+ if (newname != NULL)
+ free ((char *) newname);
+ goto err_label;
+ }
- if (insert_entry (&collate->symbols, code->val.str.start, code->val.str.len,
- (void *) 0) < 0)
- lr_error (ldfile, _("cannot insert new collating symbol definition: %s"),
- strerror (errno));
-}
+ symname = arg->val.str.startmb;
+ symname_len = arg->val.str.lenmb;
+ if (!ignore_content)
+ {
+ if (newname == NULL)
+ {
+ lr_error (ldfile, _("\
+%s: unknown character in equivalent definition name"),
+ "LC_COLLATE");
+ goto sym_equiv_free;
+ }
+ if (symname == NULL)
+ {
+ lr_error (ldfile, _("\
+%s: unknown character in equivalent definition value"),
+ "LC_COLLATE");
+ goto sym_equiv_free;
+ }
+ /* The name is already defined. */
+ if (check_duplicate (ldfile, collate, charmap,
+ repertoire, symname, symname_len))
+ goto col_sym_free;
-void
-collate_new_order (struct linereader *ldfile, struct localedef_t *locale,
- enum coll_sort_rule sort_rule)
-{
- struct locale_collate_t *collate = locale->categories[LC_COLLATE].collate;
+ /* See whether the symbol name is already defined. */
+ if (find_entry (&collate->sym_table, symname, symname_len,
+ (void **) &symval) != 0)
+ {
+ lr_error (ldfile, _("\
+%s: unknown symbol `%s' in equivalent definition"),
+ "LC_COLLATE", symname);
+ goto col_sym_free;
+ }
- if (collate->nrules >= collate->nrules_max)
- {
- collate->nrules_max *= 2;
- collate->rules
- = (enum coll_sort_rule *) xrealloc (collate->rules,
- collate->nrules_max
- * sizeof (enum coll_sort_rule));
- }
+ if (insert_entry (&collate->sym_table,
+ newname, newname_len, symval) < 0)
+ {
+ lr_error (ldfile, _("\
+error while adding equivalent collating symbol"));
+ goto sym_equiv_free;
+ }
- collate->rules[collate->nrules++] = sort_rule;
-}
+ free ((char *) symname);
+ }
+ else
+ {
+ sym_equiv_free:
+ if (newname != NULL)
+ free ((char *) newname);
+ if (symname != NULL)
+ free ((char *) symname);
+ }
+ lr_ignore_rest (ldfile, 1);
+ }
+ break;
+ case tok_order_start:
+ /* Ignore the rest of the line if we don't need the input of
+ this line. */
+ if (ignore_content)
+ {
+ lr_ignore_rest (ldfile, 0);
+ break;
+ }
-void
-collate_build_arrays (struct linereader *ldfile, struct localedef_t *locale)
-{
- struct locale_collate_t *collate = locale->categories[LC_COLLATE].collate;
+ if (state != 0 && state != 1)
+ goto err_label;
+ state = 1;
- collate->rules
- = (enum coll_sort_rule *) xrealloc (collate->rules,
- collate->nrules
- * sizeof (enum coll_sort_rule));
+ /* The 14652 draft does not specify whether all `order_start' lines
+ must contain the same number of sort-rules, but 14651 does. So
+ we require this here as well. */
+ arg = lr_token (ldfile, charmap, repertoire);
+ if (arg->tok == tok_bsymbol)
+ {
+ /* This better should be a section name. */
+ struct section_list *sp = collate->sections;
+ while (sp != NULL
+ && strcmp (sp->name, arg->val.str.startmb) != 0)
+ sp = sp->next;
- /* Allocate arrays for temporary weights. */
- collate->weight_cnt = (int *) xmalloc (collate->nrules * sizeof (int));
+ if (sp == NULL)
+ {
+ lr_error (ldfile, _("\
+%s: unknown section name `%s'"),
+ "LC_COLLATE", arg->val.str.startmb);
+ /* We use the error section. */
+ collate->current_section = &collate->error_section;
- /* Choose arbitrary start value for table size. */
- collate->nweight_max = 5 * collate->nrules;
- collate->weight = (int *) xmalloc (collate->nweight_max * sizeof (int));
-}
+ if (collate->error_section.first == NULL)
+ {
+ collate->error_section.next = collate->sections;
+ collate->sections = &collate->error_section;
+ }
+ }
+ else
+ {
+ /* Remember this section. */
+ collate->current_section = sp;
+ /* One should not be allowed to open the same
+ section twice. */
+ if (sp->first != NULL)
+ lr_error (ldfile, _("\
+%s: multiple order definitions for section `%s'"),
+ "LC_COLLATE", sp->name);
+ else
+ {
+ sp->next = collate->sections;
+ collate->sections = sp;
+ }
-int
-collate_order_elem (struct linereader *ldfile, struct localedef_t *locale,
- struct token *code, struct charset_t *charset)
-{
- const uint32_t zero = L'\0';
- struct locale_collate_t *collate = locale->categories[LC_COLLATE].collate;
- int result = 0;
- uint32_t value;
- void *tmp;
- unsigned int i;
+ /* Next should come the end of the line or a semicolon. */
+ arg = lr_token (ldfile, charmap, repertoire);
+ if (arg->tok == tok_eol)
+ {
+ uint32_t cnt;
- switch (code->tok)
- {
- case tok_bsymbol:
- /* We have a string to find in one of the three hashing tables. */
- value = charset_find_value (&charset->char_table, code->val.str.start,
- code->val.str.len);
- if (value != ILLEGAL_CHAR_VALUE)
- {
- element_t *lastp, *firstp;
+ /* This means we have exactly one rule: `forward'. */
+ if (nrules > 1)
+ lr_error (ldfile, _("\
+%s: invalid number of sorting rules"),
+ "LC_COLLATE");
+ else
+ nrules = 1;
+ sp->rules = obstack_alloc (&collate->mempool,
+ (sizeof (enum coll_sort_rule)
+ * nrules));
+ for (cnt = 0; cnt < nrules; ++cnt)
+ sp->rules[cnt] = sort_forward;
- collate->kind = character;
+ /* Next line. */
+ break;
+ }
- if (find_entry (&collate->result, &value, sizeof (uint32_t),
- (void *) &firstp) < 0)
- firstp = lastp = NULL;
+ /* Get the next token. */
+ arg = lr_token (ldfile, charmap, repertoire);
+ }
+ }
else
{
- /* The entry for the simple character is always found at
- the end. */
- lastp = firstp;
- while (lastp->next != NULL)
- lastp = lastp->next;
+ /* There is no section symbol. Therefore we use the unnamed
+ section. */
+ collate->current_section = &collate->unnamed_section;
- if (lastp->name[0] == value && lastp->name[1] == L'\0')
+ if (collate->unnamed_section.first != NULL)
+ lr_error (ldfile, _("\
+%s: multiple order definitions for unnamed section"),
+ "LC_COLLATE");
+ else
{
- lr_error (ldfile,
- _("duplicate definition for character `%.*s'"),
- (int) code->val.str.len, code->val.str.start);
- lr_ignore_rest (ldfile, 0);
- result = -1;
- break;
+ collate->unnamed_section.next = collate->sections;
+ collate->sections = &collate->unnamed_section;
}
}
- collate->current_element
- = (element_t *) obstack_alloc (&collate->element_mem,
- sizeof (element_t));
-
- obstack_grow (&collate->element_mem, &value, sizeof (value));
- obstack_grow (&collate->element_mem, &zero, sizeof (zero));
-
- collate->current_element->name =
- (const uint32_t *) obstack_finish (&collate->element_mem);
-
- collate->current_element->this_weight = ++collate->order_cnt;
-
- collate->current_element->next = NULL;
+ /* Now read the direction names. */
+ read_directions (ldfile, arg, charmap, repertoire, collate);
- if (firstp == NULL)
- {
- if (insert_entry (&collate->result, &value, sizeof (uint32_t),
- (void *) collate->current_element) < 0)
- {
- lr_error (ldfile, _("cannot insert collation element `%.*s'"),
- (int) code->val.str.len, code->val.str.start);
- exit (4);
- }
- }
- else
- lastp->next = collate->current_element;
- }
- else if (find_entry (&collate->elements, code->val.str.start,
- code->val.str.len, &tmp) >= 0)
- {
- collate->current_element = (element_t *) tmp;
+ /* From now be need the strings untranslated. */
+ ldfile->translate_strings = 0;
+ break;
- if (collate->current_element->this_weight != 0)
+ case tok_order_end:
+ /* Ignore the rest of the line if we don't need the input of
+ this line. */
+ if (ignore_content)
{
- lr_error (ldfile, _("\
-collation element `%.*s' appears more than once: ignore line"),
- (int) code->val.str.len, code->val.str.start);
lr_ignore_rest (ldfile, 0);
- result = -1;
break;
}
- collate->kind = element;
- collate->current_element->this_weight = ++collate->order_cnt;
- }
- else if (find_entry (&collate->symbols, code->val.str.start,
- code->val.str.len, &tmp) >= 0)
- {
- unsigned int order = ++collate->order_cnt;
+ if (state != 1)
+ goto err_label;
- if ((unsigned long int) tmp != 0ul)
+ /* Handle ellipsis at end of list. */
+ if (was_ellipsis != tok_none)
{
- lr_error (ldfile, _("\
-collation symbol `%.*s' appears more than once: ignore line"),
- (int) code->val.str.len, code->val.str.start);
- lr_ignore_rest (ldfile, 0);
- result = -1;
- break;
+ handle_ellipsis (ldfile, NULL, was_ellipsis, charmap, repertoire,
+ collate);
+ was_ellipsis = tok_none;
}
- collate->kind = symbol;
+ state = 2;
+ lr_ignore_rest (ldfile, 1);
+ break;
- if (set_entry (&collate->symbols, code->val.str.start,
- code->val.str.len, (void *) order) < 0)
+ case tok_reorder_after:
+ /* Ignore the rest of the line if we don't need the input of
+ this line. */
+ if (ignore_content)
{
- lr_error (ldfile, _("cannot process order specification"));
- exit (4);
+ lr_ignore_rest (ldfile, 0);
+ break;
}
- }
- else
- {
- if (verbose)
- lr_error (ldfile, _("unknown symbol `%.*s': line ignored"),
- (int) code->val.str.len, code->val.str.start);
- lr_ignore_rest (ldfile, 0);
-
- result = -1;
- }
- break;
-
- case tok_undefined:
- collate->kind = undefined;
- collate->current_element = &collate->undefined;
- break;
-
- case tok_ellipsis:
- if (collate->was_ellipsis)
- {
- lr_error (ldfile, _("\
-two lines in a row containing `...' are not allowed"));
- result = -1;
- }
- else if (collate->kind != character)
- {
- /* An ellipsis requires the previous line to be an
- character definition. */
- lr_error (ldfile, _("\
-line before ellipsis does not contain definition for character constant"));
- lr_ignore_rest (ldfile, 0);
- result = -1;
- }
- else
- collate->kind = ellipsis;
- break;
- default:
- assert (! "illegal token in `collate_order_elem'");
- }
-
- /* Now it's time to handle the ellipsis in the previous line. We do
- this only when the last line contained an definition for a
- character, the current line also defines an character, the
- character code for the later is bigger than the former. */
- if (collate->was_ellipsis)
- {
- if (collate->kind != character)
- {
- lr_error (ldfile, _("\
-line after ellipsis must contain character definition"));
- lr_ignore_rest (ldfile, 0);
- result = -1;
- }
- else if (collate->last_char > value)
- {
- lr_error (ldfile, _("end point of ellipsis range is bigger then start"));
- lr_ignore_rest (ldfile, 0);
- result = -1;
- }
- else
- {
- /* We can fill the arrays with the information we need. */
- uint32_t name[2];
- unsigned int *data;
- size_t *ptr;
- size_t cnt;
-
- name[0] = collate->last_char + 1;
- name[1] = L'\0';
-
- data = (unsigned int *) alloca ((collate->nrules + collate->nweight)
- * sizeof (unsigned int));
- ptr = (size_t *) alloca (collate->nrules * sizeof (size_t));
-
- /* Prepare data. Because the characters covered by an
- ellipsis all have equal values we prepare the data once
- and only change the variable number (if there are any).
- PTR[...] will point to the entries which will have to be
- fixed during the output loop. */
- for (cnt = 0; cnt < collate->nrules; ++cnt)
+ if (state == 1)
{
- data[cnt] = collate->weight_cnt[cnt];
- ptr[cnt] = (cnt == 0
- ? collate->nweight
- : ptr[cnt - 1] + collate->weight_cnt[cnt - 1]);
- }
-
- for (cnt = 0; cnt < collate->nweight; ++cnt)
- data[collate->nrules + cnt] = collate->weight[cnt];
-
- for (cnt = 0; cnt < collate->nrules; ++cnt)
- if ((uint32_t) data[ptr[cnt]] != ELLIPSIS_CHAR)
- ptr[cnt] = 0;
+ lr_error (ldfile, _("%s: missing `order_end' keyword"),
+ "LC_COLLATE");
+ state = 2;
- while (name[0] <= value)
- {
- element_t *pelem;
-
- pelem = (element_t *) obstack_alloc (&collate->element_mem,
- sizeof (element_t));
- pelem->name
- = (const uint32_t *) obstack_copy (&collate->element_mem,
- name, 2 * sizeof (uint32_t));
- pelem->this_weight = ++collate->order_cnt;
-
- pelem->ordering_len = collate->nweight;
- pelem->ordering
- = (unsigned int *) obstack_copy (&collate->element_mem, data,
- (collate->nrules
- + pelem->ordering_len)
- * sizeof (unsigned int));
-
- /* `...' weights need to be adjusted. */
- for (cnt = 0; cnt < collate->nrules; ++cnt)
- if (ptr[cnt] != 0)
- pelem->ordering[ptr[cnt]] = pelem->this_weight;
-
- /* Insert new entry into result table. */
- if (find_entry (&collate->result, name, sizeof (uint32_t),
- (void *) &pelem->next) >= 0)
+ /* Handle ellipsis at end of list. */
+ if (was_ellipsis != tok_none)
{
- if (set_entry (&collate->result, name, sizeof (uint32_t),
- (void *) pelem) < 0)
- error (4, 0, _("cannot insert into result table"));
+ handle_ellipsis (ldfile, arg, was_ellipsis, charmap,
+ repertoire, collate);
+ was_ellipsis = tok_none;
}
+ }
+ else if (state != 2 && state != 3)
+ goto err_label;
+ state = 3;
+
+ arg = lr_token (ldfile, charmap, repertoire);
+ if (arg->tok == tok_bsymbol)
+ {
+ /* Find this symbol in the sequence table. */
+ struct element_t *insp;
+ int no_error = 1;
+
+ if (find_entry (&collate->seq_table, arg->val.str.startmb,
+ arg->val.str.lenmb, (void **) &insp) == 0)
+ /* Yes, the symbol exists. Simply point the cursor
+ to it. */
+ collate->cursor = insp;
else
{
- pelem->next = NULL;
- if (insert_entry (&collate->result, name, sizeof (uint32_t),
- (void *) pelem) < 0)
- error (4, 0, _("cannot insert into result table"));
+ /* This is bad. The symbol after which we have to
+ insert does not exist. */
+ lr_error (ldfile, _("\
+%s: cannot reorder after %.*s: symbol not known"),
+ "LC_COLLATE", (int) arg->val.str.lenmb,
+ arg->val.str.startmb);
+ collate->cursor = NULL;
+ no_error = 0;
}
- /* Increment counter. */
- ++name[0];
+ lr_ignore_rest (ldfile, no_error);
}
- }
- }
-
- /* Reset counters for weights. */
- collate->weight_idx = 0;
- collate->nweight = 0;
- for (i = 0; i < collate->nrules; ++i)
- collate->weight_cnt[i] = 0;
- collate->current_patch = NULL;
-
- return result;
-}
-
-
-int
-collate_weight_bsymbol (struct linereader *ldfile, struct localedef_t *locale,
- struct token *code, struct charset_t *charset)
-{
- struct locale_collate_t *collate = locale->categories[LC_COLLATE].collate;
- unsigned int here_weight;
- uint32_t value;
- void *tmp;
-
- assert (code->tok == tok_bsymbol);
-
- value = charset_find_value (&charset->char_table, code->val.str.start,
- code->val.str.len);
- if (value != ILLEGAL_CHAR_VALUE)
- {
- element_t *runp;
-
- if (find_entry (&collate->result, &value, sizeof (uint32_t),
- (void *)&runp) < 0)
- runp = NULL;
-
- while (runp != NULL
- && (runp->name[0] != value || runp->name[1] != L'\0'))
- runp = runp->next;
-
- here_weight = runp == NULL ? 0 : runp->this_weight;
- }
- else if (find_entry (&collate->elements, code->val.str.start,
- code->val.str.len, &tmp) >= 0)
- {
- element_t *runp = (element_t *) tmp;
-
- here_weight = runp->this_weight;
- }
- else if (find_entry (&collate->symbols, code->val.str.start,
- code->val.str.len, &tmp) >= 0)
- {
- here_weight = (unsigned int) tmp;
- }
- else
- {
- if (verbose)
- lr_error (ldfile, _("unknown symbol `%.*s': line ignored"),
- (int) code->val.str.len, code->val.str.start);
- lr_ignore_rest (ldfile, 0);
- return -1;
- }
-
- /* When we currently work on a collation symbol we do not expect any
- weight. */
- if (collate->kind == symbol)
- {
- lr_error (ldfile, _("\
-specification of sorting weight for collation symbol does not make sense"));
- lr_ignore_rest (ldfile, 0);
- return -1;
- }
-
- /* Add to the current collection of weights. */
- if (collate->nweight >= collate->nweight_max)
- {
- collate->nweight_max *= 2;
- collate->weight = (unsigned int *) xrealloc (collate->weight,
- collate->nweight_max);
- }
-
- /* If the weight is currently not known, we remember to patch the
- resulting tables. */
- if (here_weight == 0)
- {
- patch_t *newp;
-
- newp = (patch_t *) obstack_alloc (&collate->element_mem,
- sizeof (patch_t));
- newp->fname = ldfile->fname;
- newp->lineno = ldfile->lineno;
- newp->token = (const char *) obstack_copy0 (&collate->element_mem,
- code->val.str.start,
- code->val.str.len);
- newp->where.idx = collate->nweight++;
- newp->next = collate->current_patch;
- collate->current_patch = newp;
- }
- else
- collate->weight[collate->nweight++] = here_weight;
- ++collate->weight_cnt[collate->weight_idx];
-
- return 0;
-}
-
-
-int
-collate_next_weight (struct linereader *ldfile, struct localedef_t *locale)
-{
- struct locale_collate_t *collate = locale->categories[LC_COLLATE].collate;
-
- if (collate->kind == symbol)
- {
- lr_error (ldfile, _("\
-specification of sorting weight for collation symbol does not make sense"));
- lr_ignore_rest (ldfile, 0);
- return -1;
- }
-
- ++collate->weight_idx;
- if (collate->weight_idx >= collate->nrules)
- {
- lr_error (ldfile, _("too many weights"));
- lr_ignore_rest (ldfile, 0);
- return -1;
- }
-
- return 0;
-}
-
-
-int
-collate_simple_weight (struct linereader *ldfile, struct localedef_t *locale,
- struct token *code, struct charset_t *charset)
-{
- struct locale_collate_t *collate = locale->categories[LC_COLLATE].collate;
- unsigned int value = 0;
-
- /* There current tokens can be `IGNORE', `...', or a string. */
- switch (code->tok)
- {
- case tok_ignore:
- /* This token is allowed in all situations. */
- value = IGNORE_CHAR;
- break;
-
- case tok_ellipsis:
- /* The ellipsis is only allowed for the `...' or `UNDEFINED'
- entry. */
- if (collate->kind != ellipsis && collate->kind != undefined)
- {
- lr_error (ldfile, _("\
-`...' must only be used in `...' and `UNDEFINED' entries"));
- lr_ignore_rest (ldfile, 0);
- return -1;
- }
- value = ELLIPSIS_CHAR;
- break;
-
- case tok_string:
- /* This can become difficult. We have to get the weights which
- correspond to the single wide chars in the string. But some
- of the `chars' might not be real characters, but collation
- elements or symbols. And so the string decoder might have
- signaled errors. The string at this point is not translated.
- I.e., all <...> sequences are still there. */
- {
- char *runp = code->val.str.start;
- void *tmp;
-
- while (*runp != '\0')
- {
- char *startp = (char *) runp;
- char *putp = (char *) runp;
- uint32_t wch;
-
- /* Lookup weight for char and store it. */
- if (*runp == '<')
- {
- while (*++runp != '\0' && *runp != '>')
- {
- if (*runp == ldfile->escape_char)
- if (*++runp == '\0')
- {
- lr_error (ldfile, _("unterminated weight name"));
- lr_ignore_rest (ldfile, 0);
- return -1;
- }
- *putp++ = *runp;
- }
- if (*runp == '>')
- ++runp;
-
- if (putp == startp)
- {
- lr_error (ldfile, _("empty weight name: line ignored"));
- lr_ignore_rest (ldfile, 0);
- return -1;
- }
-
- wch = charset_find_value (&charset->char_table, startp,
- putp - startp);
- if (wch != ILLEGAL_CHAR_VALUE)
- {
- element_t *pelem;
-
- if (find_entry (&collate->result, &wch, sizeof (uint32_t),
- (void *)&pelem) < 0)
- pelem = NULL;
-
- while (pelem != NULL
- && (pelem->name[0] != wch
- || pelem->name[1] != L'\0'))
- pelem = pelem->next;
-
- value = pelem == NULL ? 0 : pelem->this_weight;
- }
- else if (find_entry (&collate->elements, startp, putp - startp,
- &tmp) >= 0)
- {
- element_t *pelem = (element_t *) tmp;
-
- value = pelem->this_weight;
- }
- else if (find_entry (&collate->symbols, startp, putp - startp,
- &tmp) >= 0)
- {
- value = (unsigned int) tmp;
- }
- else
- {
- if (verbose)
- lr_error (ldfile, _("unknown symbol `%.*s': line ignored"),
- (int) (putp - startp), startp);
- lr_ignore_rest (ldfile, 0);
- return -1;
- }
- }
- else
- {
- element_t *wp;
- uint32_t wch;
-
- if (*runp == ldfile->escape_char)
- {
- static const char digits[] = "0123456789abcdef";
- const char *dp;
- int base;
-
- ++runp;
- if (tolower (*runp) == 'x')
- {
- ++runp;
- base = 16;
- }
- else if (tolower (*runp) == 'd')
- {
- ++runp;
- base = 10;
- }
- else
- base = 8;
-
- dp = strchr (digits, tolower (*runp));
- if (dp == NULL || (dp - digits) >= base)
- {
- illegal_char:
- lr_error (ldfile, _("\
-illegal character constant in string"));
- lr_ignore_rest (ldfile, 0);
- return -1;
- }
- wch = dp - digits;
- ++runp;
-
- dp = strchr (digits, tolower (*runp));
- if (dp == NULL || (dp - digits) >= base)
- goto illegal_char;
- wch *= base;
- wch += dp - digits;
- ++runp;
-
- if (base != 16)
- {
- dp = strchr (digits, tolower (*runp));
- if (dp != NULL && (dp - digits < base))
- {
- wch *= base;
- wch += dp - digits;
- ++runp;
- }
- }
- }
- else
- wch = (uint32_t) *runp++;
-
- /* Lookup the weight for WCH. */
- if (find_entry (&collate->result, &wch, sizeof (wch),
- (void *)&wp) < 0)
- wp = NULL;
-
- while (wp != NULL
- && (wp->name[0] != wch || wp->name[1] != L'\0'))
- wp = wp->next;
-
- value = wp == NULL ? 0 : wp->this_weight;
-
- /* To get the correct name for the error message. */
- putp = runp;
+ else
+ /* This must not happen. */
+ goto err_label;
+ break;
- /**************************************************\
- |* I know here is something wrong. Characters in *|
- |* the string which are not in the <...> form *|
- |* cannot be declared forward for now!!! *|
- \**************************************************/
- }
+ case tok_reorder_end:
+ /* Ignore the rest of the line if we don't need the input of
+ this line. */
+ if (ignore_content)
+ break;
- /* Store in weight array. */
- if (collate->nweight >= collate->nweight_max)
- {
- collate->nweight_max *= 2;
- collate->weight
- = (unsigned int *) xrealloc (collate->weight,
- collate->nweight_max);
- }
+ if (state != 3)
+ goto err_label;
+ state = 4;
+ lr_ignore_rest (ldfile, 1);
+ break;
- if (value == 0)
- {
- patch_t *newp;
-
- newp = (patch_t *) obstack_alloc (&collate->element_mem,
- sizeof (patch_t));
- newp->fname = ldfile->fname;
- newp->lineno = ldfile->lineno;
- newp->token
- = (const char *) obstack_copy0 (&collate->element_mem,
- startp, putp - startp);
- newp->where.idx = collate->nweight++;
- newp->next = collate->current_patch;
- collate->current_patch = newp;
- }
- else
- collate->weight[collate->nweight++] = value;
- ++collate->weight_cnt[collate->weight_idx];
- }
- }
- return 0;
+ case tok_reorder_sections_after:
+ /* Ignore the rest of the line if we don't need the input of
+ this line. */
+ if (ignore_content)
+ {
+ lr_ignore_rest (ldfile, 0);
+ break;
+ }
- default:
- assert (! "should not happen");
- }
+ if (state == 1)
+ {
+ lr_error (ldfile, _("%s: missing `order_end' keyword"),
+ "LC_COLLATE");
+ state = 2;
+ /* Handle ellipsis at end of list. */
+ if (was_ellipsis != tok_none)
+ {
+ handle_ellipsis (ldfile, NULL, was_ellipsis, charmap,
+ repertoire, collate);
+ was_ellipsis = tok_none;
+ }
+ }
+ else if (state == 3)
+ {
+ error (0, 0, _("%s: missing `reorder-end' keyword"),
+ "LC_COLLATE");
+ state = 4;
+ }
+ else if (state != 2 && state != 4)
+ goto err_label;
+ state = 5;
- if (collate->nweight >= collate->nweight_max)
- {
- collate->nweight_max *= 2;
- collate->weight = (unsigned int *) xrealloc (collate->weight,
- collate->nweight_max);
- }
+ /* Get the name of the sections we are adding after. */
+ arg = lr_token (ldfile, charmap, repertoire);
+ if (arg->tok == tok_bsymbol)
+ {
+ /* Now find a section with this name. */
+ struct section_list *runp = collate->sections;
- collate->weight[collate->nweight++] = value;
- ++collate->weight_cnt[collate->weight_idx];
+ while (runp != NULL)
+ {
+ if (runp->name != NULL
+ && strlen (runp->name) == arg->val.str.lenmb
+ && memcmp (runp->name, arg->val.str.startmb,
+ arg->val.str.lenmb) == 0)
+ break;
- return 0;
-}
+ runp = runp->next;
+ }
+ if (runp != NULL)
+ collate->current_section = runp;
+ else
+ {
+ /* This is bad. The section after which we have to
+ reorder does not exist. Therefore we cannot
+ process the whole rest of this reorder
+ specification. */
+ lr_error (ldfile, _("%s: section `%.*s' not known"),
+ "LC_COLLATE", (int) arg->val.str.lenmb,
+ arg->val.str.startmb);
-void
-collate_end_weight (struct linereader *ldfile, struct localedef_t *locale)
-{
- struct locale_collate_t *collate = locale->categories[LC_COLLATE].collate;
- element_t *pelem = collate->current_element;
+ do
+ {
+ lr_ignore_rest (ldfile, 0);
- if (collate->kind == symbol)
- {
- /* We don't have to do anything. */
- collate->was_ellipsis = 0;
- return;
- }
+ now = lr_token (ldfile, charmap, NULL);
+ }
+ while (now->tok == tok_reorder_sections_after
+ || now->tok == tok_reorder_sections_end
+ || now->tok == tok_end);
- if (collate->kind == ellipsis)
- {
- /* Before the next line is processed the ellipsis is handled. */
- collate->was_ellipsis = 1;
- return;
- }
+ /* Process the token we just saw. */
+ nowtok = now->tok;
+ continue;
+ }
+ }
+ else
+ /* This must not happen. */
+ goto err_label;
+ break;
- assert (collate->kind == character || collate->kind == element
- || collate->kind == undefined);
+ case tok_reorder_sections_end:
+ /* Ignore the rest of the line if we don't need the input of
+ this line. */
+ if (ignore_content)
+ break;
- /* Fill in the missing weights. */
- while (++collate->weight_idx < collate->nrules)
- {
- collate->weight[collate->nweight++] = pelem->this_weight;
- ++collate->weight_cnt[collate->weight_idx];
- }
+ if (state != 5)
+ goto err_label;
+ state = 6;
+ lr_ignore_rest (ldfile, 1);
+ break;
- /* Now we know how many ordering weights the current
- character/element has. Allocate room in the element structure
- and copy information. */
- pelem->ordering_len = collate->nweight;
+ case tok_bsymbol:
+ /* Ignore the rest of the line if we don't need the input of
+ this line. */
+ if (ignore_content)
+ {
+ lr_ignore_rest (ldfile, 0);
+ break;
+ }
- /* First we write an array with the number of values for each
- weight. */
- obstack_grow (&collate->element_mem, collate->weight_cnt,
- collate->nrules * sizeof (unsigned int));
+ if (state != 1 && state != 3)
+ goto err_label;
- /* Now the weights itselves. */
- obstack_grow (&collate->element_mem, collate->weight,
- collate->nweight * sizeof (unsigned int));
+ if (state == 3)
+ {
+ /* It is possible that we already have this collation sequence.
+ In this case we move the entry. */
+ struct element_t *seqp;
- /* Get result. */
- pelem->ordering = obstack_finish (&collate->element_mem);
+ /* If the symbol after which we have to insert was not found
+ ignore all entries. */
+ if (collate->cursor == NULL)
+ {
+ lr_ignore_rest (ldfile, 0);
+ break;
+ }
- /* Now we handle the "patches". */
- while (collate->current_patch != NULL)
- {
- patch_t *this_patch;
+ if (find_entry (&collate->seq_table, arg->val.str.startmb,
+ arg->val.str.lenmb, (void **) &seqp) == 0)
+ {
+ /* Remove the entry from the old position. */
+ if (seqp->last == NULL)
+ collate->start = seqp->next;
+ else
+ seqp->last->next = seqp->next;
+ if (seqp->next != NULL)
+ seqp->next->last = seqp->last;
- this_patch = collate->current_patch;
+ /* We also have to check whether this entry is the
+ first or last of a section. */
+ if (seqp->section->first == seqp)
+ {
+ if (seqp->section->first == seqp->section->last)
+ /* This setion has no content anymore. */
+ seqp->section->first = seqp->section->last = NULL;
+ else
+ seqp->section->first = seqp->next;
+ }
+ else if (seqp->section->last == seqp)
+ seqp->section->last = seqp->last;
- this_patch->where.pos = &pelem->ordering[collate->nrules
- + this_patch->where.idx];
+ /* Now insert it in the new place. */
+ seqp->next = collate->cursor->next;
+ seqp->last = collate->cursor;
+ collate->cursor->next = seqp;
+ if (seqp->next != NULL)
+ seqp->next->last = seqp;
- collate->current_patch = this_patch->next;
- this_patch->next = collate->all_patches;
- collate->all_patches = this_patch;
- }
+ seqp->section = collate->cursor->section;
+ if (seqp->section->last == collate->cursor)
+ seqp->section->last = seqp;
- /* Set information for next round. */
- collate->was_ellipsis = 0;
- if (collate->kind != undefined)
- collate->last_char = pelem->name[0];
-}
+ break;
+ }
+ /* Otherwise we just add a new entry. */
+ }
+ else if (state == 5)
+ {
+ /* We are reordering sections. Find the named section. */
+ struct section_list *runp = collate->sections;
+ struct section_list *prevp = NULL;
-/* The parser for the LC_CTYPE section of the locale definition. */
-void
-read_lc_collate (struct linereader *ldfile, struct localedef_t *result,
- struct charmap_t *charmap, struct repertoire_t *repertoire,
- int ignore_content)
-{
- struct locale_collate_t *collate;
- int did_copy = 0;
- const char *save_str;
+ while (runp != NULL)
+ {
+ if (runp->name != NULL
+ && strlen (runp->name) == arg->val.str.lenmb
+ && memcmp (runp->name, arg->val.str.startmb,
+ arg->val.str.lenmb) == 0)
+ break;
+
+ prevp = runp;
+ runp = runp->next;
+ }
- /* The rest of the line containing `LC_COLLATE' must be free. */
- lr_ignore_rest (ldfile, 1);
+ if (runp == NULL)
+ {
+ lr_error (ldfile, _("%s: section `%.*s' not known"),
+ "LC_COLLATE", (int) arg->val.str.lenmb,
+ arg->val.str.startmb);
+ lr_ignore_rest (ldfile, 0);
+ }
+ else
+ {
+ if (runp != collate->current_section)
+ {
+ /* Remove the named section from the old place and
+ insert it in the new one. */
+ prevp->next = runp->next;
- now = lr_token (ldfile, charmap, NULL);
- nowtok = now->tok;
+ runp->next = collate->current_section->next;
+ collate->current_section->next = runp;
+ collate->current_section = runp;
+ }
- /* If we see `copy' now we are almost done. */
- if (nowtok == tok_copy)
- {
- handle_copy (ldfile, charmap, repertoire, result, tok_lc_collate,
- LC_COLLATE, "LC_COLLATE", ignore_content);
- did_copy = 1;
- }
+ /* Process the rest of the line which might change
+ the collation rules. */
+ arg = lr_token (ldfile, charmap, repertoire);
+ if (arg->tok != tok_eof && arg->tok != tok_eol)
+ read_directions (ldfile, arg, charmap, repertoire,
+ collate);
+ }
+ break;
+ }
+ else if (was_ellipsis != tok_none)
+ {
+ /* Using the information in the `ellipsis_weight'
+ element and this and the last value we have to handle
+ the ellipsis now. */
+ assert (state == 1);
- /* Prepare the data structures. */
- collate_startup (ldfile, result, charmap, ignore_content);
- collate = result->categories[LC_COLLATE].collate;
+ handle_ellipsis (ldfile, arg, was_ellipsis, charmap, repertoire,
+ collate);
- while (1)
- {
- /* Of course we don't proceed beyond the end of file. */
- if (nowtok == tok_eof)
- break;
+ /* Remember that we processed the ellipsis. */
+ was_ellipsis = tok_none;
- /* Ignore empty lines. */
- if (nowtok == tok_eol)
- {
- now = lr_token (ldfile, charmap, NULL);
- nowtok = now->tok;
- continue;
- }
+ /* And don't add the value a second time. */
+ break;
+ }
- switch (nowtok)
- {
- case tok_coll_weight_max:
- if (did_copy)
- goto err_label;
- /* The rest of the line must be a single integer value. */
- now = lr_token (ldfile, charmap, NULL);
- if (now->tok != tok_number)
- goto err_label;
- /* We simply forget about the value we just read, the implementation
- has no fixed limits. */
- lr_ignore_rest (ldfile, 1);
+ /* Now insert in the new place. */
+ insert_value (ldfile, arg, charmap, repertoire, collate);
break;
- case tok_script:
- if (did_copy)
- goto err_label;
- /* We expect the name of the script in brackets. */
- now = lr_token (ldfile, charmap, NULL);
- if (now->tok != tok_bsymbol && now->tok != tok_ucs4)
- goto err_label;
- if (now->tok != tok_bsymbol)
+ case tok_undefined:
+ /* Ignore the rest of the line if we don't need the input of
+ this line. */
+ if (ignore_content)
{
- lr_error (ldfile, _("\
-script name `%s' must not duplicate any known name"),
- tok->val.str.startmb);
lr_ignore_rest (ldfile, 0);
break;
}
- collate->scripts = xmalloc (collate->scripts,
- (collate->nscripts
- * sizeof (const char *)));
- collate->scripts[collate->nscripts++] = tok->val.str.startmb;
- lr_ignore_rest (ldfile, 1);
- break;
- case tok_collating_element:
- if (did_copy)
- goto err_label;
- /* Get the first argument, a symbol in brackets. */
- now = lr_token (ldfile, charmap, NULL);
- if (now->tok != tok_bsymbol)
+ if (state != 1)
goto err_label;
- /* Test it. */
- if (collate_element_to (ldfile, collate, now, charmap, repertoire))
+
+ if (was_ellipsis != tok_none)
+ {
+ lr_error (ldfile,
+ _("%s: cannot have `%s' as end of ellipsis range"),
+ "LC_COLLATE", "UNDEFINED");
+
+ unlink_element (collate);
+ was_ellipsis = tok_none;
+ }
+
+ /* See whether UNDEFINED already appeared somewhere. */
+ if (collate->undefined.next != NULL
+ || (collate->cursor != NULL
+ && collate->undefined.next == collate->cursor))
{
- /* An error occurred. */
+ lr_error (ldfile,
+ _("%s: order for `%.*s' already defined at %s:%zu"),
+ "LC_COLLATE", 9, "UNDEFINED", collate->undefined.file,
+ collate->undefined.line);
lr_ignore_rest (ldfile, 0);
- break;
}
- save_str = tok->val.str.startmb;
- /* Next comes `from'. */
- now = lr_token (ldfile, charmap, NULL);
- if (now->tok != tok_from)
- goto err_label;
- /* Now comes a string. */
- now = lr_token (ldfile, charmap, repertoire);
- if (now->tok != tok_string)
- goto err_label;
- collate_element_from (ldfile, collate, save_str, now, charmap,
- repertoire);
- /* The rest of the line should be empty. */
- lr_ignore_rest (ldfile, 1);
+ else
+ /* Parse the weights. */
+ insert_weights (ldfile, &collate->undefined, charmap,
+ repertoire, collate, tok_none);
break;
- case tok_collating_symbol:
- if (did_copy)
- goto err_label;
- /* Get the argument, a single symbol in brackets. */
- now = lr_token (ldfile, charmap, NULL);
- if (now->tok != tok_bsymbol)
+ case tok_ellipsis2:
+ case tok_ellipsis3:
+ case tok_ellipsis4:
+ /* This is the symbolic (decimal or hexadecimal) or absolute
+ ellipsis. */
+ if (was_ellipsis != tok_none)
goto err_label;
- collate_symbol (ldfile, collate, now, charmap, repertoire);
- break;
- case tok_order_start:
- if (did_copy)
+ if (state != 1 && state != 3)
goto err_label;
- /* We expect now a scripting symbol or start right away
- with the order keywords. Or we have no argument at all
- in which means `forward'. */
- now = lr_token (ldfile, charmap, NULL);
- if (now->tok == tok_eol)
- {
- static enum coll_sort_rule default_rule = sort_forward;
- /* Use a single `forward' rule. */
- collate->nrules = 1;
- collate->rules = &default_rule;
- }
- else
- {
- /* XXX We don't recognize the ISO 14651 extensions yet. */
- uint32_t nrules = 0;
- uint32_t nrules_max = 32;
- enum coll_sort_rule *rules = alloca (nrules_max
- * sizeof (*rules));
- int saw_semicolon = 0;
-
- memset (rules, '\0', nrules_max * sizeof (*rules));
- do
- {
- if (now->tok != tok_forward && now->tok != tok_backward
- && now->tok != tok_position)
- goto err_label;
+ was_ellipsis = nowtok;
- if (saw_semicolon)
- {
- if (nrules == nrules_max)
- {
- newp = alloca (nrules_max * 2 * sizeof (*rules));
- rules = memcpy (newp, rules,
- nrules_max * sizeof (*rules));
- memset (&rules[nrules_max], '\0',
- nrules_max * sizeof (*rules));
- nrules_max *= 2;
- }
- ++nrules;
- }
+ insert_weights (ldfile, &collate->ellipsis_weight, charmap,
+ repertoire, collate, nowtok);
+ break;
- switch (now->tok)
- {
- case tok_forward:
- if ((rules[nrules] & sort_backward) != 0)
- {
- lr_error (ldfile, _("\
-`forward' and `backward' order exclude each other"));
- lr_ignore_rest (ldfile, 0);
- goto error_sort;
- }
- rules[nrules] |= sort_forward;
- break;
- case tok_backward:
- if ((rules[nrules] & sort_forward) != 0)
- {
- lr_error (ldfile, _("\
-`forward' and `backward' order exclude each other"));
- lr_ignore_rest (ldfile, 0);
- goto error_sort;
- }
- rules[nrules] |= sort_backward;
- break;
- case tok_position:
- rules[nrules] |= tok_position;
- break;
- }
+ case tok_end:
+ /* Next we assume `LC_COLLATE'. */
+ if (!ignore_content)
+ {
+ if (state == 0)
+ /* We must either see a copy statement or have
+ ordering values. */
+ lr_error (ldfile,
+ _("%s: empty category description not allowed"),
+ "LC_COLLATE");
+ else if (state == 1)
+ {
+ lr_error (ldfile, _("%s: missing `order_end' keyword"),
+ "LC_COLLATE");
- /* Get the next token. This is either the end of the line,
- a comma or a semicolon. */
- now = lr_token (ldfile, charmap, NULL);
- if (now->tok == tok_comma || now->tok == tok_semicolon)
+ /* Handle ellipsis at end of list. */
+ if (was_ellipsis != tok_none)
{
- saw_semicolon = now->tok == tok_semicolon;
- now = lr_token (ldfile, charmap, NULL);
+ handle_ellipsis (ldfile, NULL, was_ellipsis, charmap,
+ repertoire, collate);
+ was_ellipsis = tok_none;
}
}
- while (now->tok != tok_eol || now->tok != tok_eof);
-
- error_sort:
- collate->nrules = nrules;
- collate->rules = memcpy (xmalloc (nrules * sizeof (*rules)),
- rules, nrules * sizeof (*rules));
+ else if (state == 3)
+ error (0, 0, _("%s: missing `reorder-end' keyword"),
+ "LC_COLLATE");
+ else if (state == 5)
+ error (0, 0, _("%s: missing `reorder-sections-end' keyword"),
+ "LC_COLLATE");
}
+ arg = lr_token (ldfile, charmap, NULL);
+ if (arg->tok == tok_eof)
+ break;
+ if (arg->tok == tok_eol)
+ lr_error (ldfile, _("%s: incomplete `END' line"), "LC_COLLATE");
+ else if (arg->tok != tok_lc_collate)
+ lr_error (ldfile, _("\
+%1$s: definition does not end with `END %1$s'"), "LC_COLLATE");
+ lr_ignore_rest (ldfile, arg->tok == tok_lc_collate);
+ return;
- /* Now read the rules. */
- read_rules (ldfile, collate, charmap, repertoire);
- break;
-
- case tok_reorder_after:
- break;
-
- case tok_reorder_script_after:
- break;
-
- default:
- err_label:
- if (now->tok != tok_eof)
- SYNTAX_ERROR (_("syntax error in %s locale definition"),
- "LC_COLLATE");
+ default:
+ err_label:
+ SYNTAX_ERROR (_("%s: syntax error"), "LC_COLLATE");
}
/* Prepare for the next round. */
}
/* When we come here we reached the end of the file. */
- lr_error (ldfile, _("premature end of file while reading category `%s'"),
- "LC_COLLATE");
+ lr_error (ldfile, _("%s: premature end of file"), "LC_COLLATE");
}
-
-#endif
git://sourceware.org / glibc.git / blobdiff