-/* Copyright (C) 1995,96,97,2002, 2004, 2007 Free Software Foundation, Inc.
+/* Copyright (C) 1995-2019 Free Software Foundation, Inc.
This file is part of the GNU C Library.
Written by Ulrich Drepper <drepper@gnu.org>, 1995.
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
- License along with the GNU C Library; if not, write to the Free
- Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
- 02111-1307 USA. */
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
#include <assert.h>
#include <locale.h>
#include <stddef.h>
#include <stdint.h>
-#include <stdlib.h>
#include <string.h>
+#include <sys/param.h>
+#include <libc-diag.h>
#ifndef STRING_TYPE
# define STRING_TYPE char
# define USTRING_TYPE unsigned char
# define STRCOLL __strcoll_l
# define STRCMP strcmp
-# define STRLEN strlen
# define WEIGHT_H "../locale/weight.h"
# define SUFFIX MB
# define L(arg) arg
#define CONCAT1(a,b) a##b
#include "../locale/localeinfo.h"
+#include WEIGHT_H
+
+/* Track status while looking for sequences in a string. */
+typedef struct
+{
+ int len; /* Length of the current sequence. */
+ size_t val; /* Position of the sequence relative to the
+ previous non-ignored sequence. */
+ size_t idxmax; /* Maximum index in sequences. */
+ size_t idxcnt; /* Current count of indices. */
+ size_t backw; /* Current Backward sequence index. */
+ size_t backw_stop; /* Index where the backward sequences stop. */
+ const USTRING_TYPE *us; /* The string. */
+ unsigned char rule; /* Saved rule for the first sequence. */
+ int32_t idx; /* Index to weight of the current sequence. */
+ int32_t save_idx; /* Save looked up index of a forward
+ sequence after the last backward
+ sequence. */
+ const USTRING_TYPE *back_us; /* Beginning of the backward sequence. */
+} coll_seq;
+
+/* Get next sequence. Traverse the string as required. */
+static __always_inline void
+get_next_seq (coll_seq *seq, int nrules, const unsigned char *rulesets,
+ const USTRING_TYPE *weights, const int32_t *table,
+ const USTRING_TYPE *extra, const int32_t *indirect,
+ int pass)
+{
+ size_t val = seq->val = 0;
+ int len = seq->len;
+ size_t backw_stop = seq->backw_stop;
+ size_t backw = seq->backw;
+ size_t idxcnt = seq->idxcnt;
+ size_t idxmax = seq->idxmax;
+ int32_t idx = seq->idx;
+ const USTRING_TYPE *us = seq->us;
+
+ while (len == 0)
+ {
+ ++val;
+ if (backw_stop != ~0ul)
+ {
+ /* There is something pushed. */
+ if (backw == backw_stop)
+ {
+ /* The last pushed character was handled. Continue
+ with forward characters. */
+ if (idxcnt < idxmax)
+ {
+ idx = seq->save_idx;
+ backw_stop = ~0ul;
+ }
+ else
+ {
+ /* Nothing anymore. The backward sequence ended with
+ the last sequence in the string. Note that len is
+ still zero. */
+ idx = 0;
+ break;
+ }
+ }
+ else
+ {
+ /* XXX Traverse BACKW sequences from the beginning of
+ BACKW_STOP to get the next sequence. Is ther a quicker way
+ to do this? */
+ size_t i = backw_stop;
+ us = seq->back_us;
+ while (i < backw)
+ {
+ int32_t tmp = findidx (table, indirect, extra, &us, -1);
+ idx = tmp & 0xffffff;
+ i++;
+ }
+ --backw;
+ us = seq->us;
+ }
+ }
+ else
+ {
+ backw_stop = idxmax;
+ int32_t prev_idx = idx;
+
+ while (*us != L('\0'))
+ {
+ int32_t tmp = findidx (table, indirect, extra, &us, -1);
+ unsigned char rule = tmp >> 24;
+ prev_idx = idx;
+ idx = tmp & 0xffffff;
+ idxcnt = idxmax++;
+
+ /* Save the rule for the first sequence. */
+ if (__glibc_unlikely (idxcnt == 0))
+ seq->rule = rule;
+
+ if ((rulesets[rule * nrules + pass]
+ & sort_backward) == 0)
+ /* No more backward characters to push. */
+ break;
+ ++idxcnt;
+ }
+
+ if (backw_stop >= idxcnt)
+ {
+ /* No sequence at all or just one. */
+ if (idxcnt == idxmax || backw_stop > idxcnt)
+ /* Note that len is still zero. */
+ break;
+
+ backw_stop = ~0ul;
+ }
+ else
+ {
+ /* We pushed backward sequences. If the stream ended with the
+ backward sequence, then we process the last sequence we
+ found. Otherwise we process the sequence before the last
+ one since the last one was a forward sequence. */
+ seq->back_us = seq->us;
+ seq->us = us;
+ backw = idxcnt;
+ if (idxmax > idxcnt)
+ {
+ backw--;
+ seq->save_idx = idx;
+ idx = prev_idx;
+ }
+ if (backw > backw_stop)
+ backw--;
+ }
+ }
+
+ /* With GCC 5.3 when compiling with -Os the compiler complains
+ that idx, taken from seq->idx (seq1 or seq2 from STRCOLL) may
+ be used uninitialized. In general this can't possibly be true
+ since seq1.idx and seq2.idx are initialized to zero in the
+ outer function. Only one case where seq->idx is restored from
+ seq->save_idx might result in an uninitialized idx value, but
+ it is guarded by a sequence of checks against backw_stop which
+ ensures that seq->save_idx was saved to first and contains a
+ valid value. */
+ DIAG_PUSH_NEEDS_COMMENT;
+ DIAG_IGNORE_Os_NEEDS_COMMENT (5, "-Wmaybe-uninitialized");
+ len = weights[idx++];
+ DIAG_POP_NEEDS_COMMENT;
+ /* Skip over indices of previous levels. */
+ for (int i = 0; i < pass; i++)
+ {
+ idx += len;
+ len = weights[idx];
+ idx++;
+ }
+ }
+
+ /* Update the structure. */
+ seq->val = val;
+ seq->len = len;
+ seq->backw_stop = backw_stop;
+ seq->backw = backw;
+ seq->idxcnt = idxcnt;
+ seq->idxmax = idxmax;
+ seq->us = us;
+ seq->idx = idx;
+}
+
+/* Compare two sequences. */
+static __always_inline int
+do_compare (coll_seq *seq1, coll_seq *seq2, int position,
+ const USTRING_TYPE *weights)
+{
+ int seq1len = seq1->len;
+ int seq2len = seq2->len;
+ size_t val1 = seq1->val;
+ size_t val2 = seq2->val;
+ int idx1 = seq1->idx;
+ int idx2 = seq2->idx;
+ int result = 0;
+
+ /* Test for position if necessary. */
+ if (position && val1 != val2)
+ {
+ result = val1 > val2 ? 1 : -1;
+ goto out;
+ }
+
+ /* Compare the two sequences. */
+ do
+ {
+ if (weights[idx1] != weights[idx2])
+ {
+ /* The sequences differ. */
+ result = weights[idx1] - weights[idx2];
+ goto out;
+ }
+
+ /* Increment the offsets. */
+ ++idx1;
+ ++idx2;
+
+ --seq1len;
+ --seq2len;
+ }
+ while (seq1len > 0 && seq2len > 0);
+
+ if (position && seq1len != seq2len)
+ result = seq1len - seq2len;
+
+out:
+ seq1->len = seq1len;
+ seq2->len = seq2len;
+ seq1->idx = idx1;
+ seq2->idx = idx2;
+ return result;
+}
int
-STRCOLL (s1, s2, l)
- const STRING_TYPE *s1;
- const STRING_TYPE *s2;
- __locale_t l;
+STRCOLL (const STRING_TYPE *s1, const STRING_TYPE *s2, locale_t l)
{
- struct locale_data *current = l->__locales[LC_COLLATE];
+ struct __locale_data *current = l->__locales[LC_COLLATE];
uint_fast32_t nrules = current->values[_NL_ITEM_INDEX (_NL_COLLATE_NRULES)].word;
/* We don't assign the following values right away since it might be
unnecessary in case there are no rules. */
const USTRING_TYPE *weights;
const USTRING_TYPE *extra;
const int32_t *indirect;
- uint_fast32_t pass;
- int result = 0;
- const USTRING_TYPE *us1;
- const USTRING_TYPE *us2;
- size_t s1len;
- size_t s2len;
- int32_t *idx1arr;
- int32_t *idx2arr;
- unsigned char *rule1arr;
- unsigned char *rule2arr;
- size_t idx1max;
- size_t idx2max;
- size_t idx1cnt;
- size_t idx2cnt;
- size_t idx1now;
- size_t idx2now;
- size_t backw1_stop;
- size_t backw2_stop;
- size_t backw1;
- size_t backw2;
- int val1;
- int val2;
- int position;
- int seq1len;
- int seq2len;
- int use_malloc;
-
-#include WEIGHT_H
if (nrules == 0)
return STRCMP (s1, s2);
+ /* Catch empty strings. */
+ if (__glibc_unlikely (*s1 == '\0') || __glibc_unlikely (*s2 == '\0'))
+ return (*s1 != '\0') - (*s2 != '\0');
+
rulesets = (const unsigned char *)
current->values[_NL_ITEM_INDEX (_NL_COLLATE_RULESETS)].string;
table = (const int32_t *)
current->values[_NL_ITEM_INDEX (CONCAT(_NL_COLLATE_EXTRA,SUFFIX))].string;
indirect = (const int32_t *)
current->values[_NL_ITEM_INDEX (CONCAT(_NL_COLLATE_INDIRECT,SUFFIX))].string;
- use_malloc = 0;
assert (((uintptr_t) table) % __alignof__ (table[0]) == 0);
assert (((uintptr_t) weights) % __alignof__ (weights[0]) == 0);
assert (((uintptr_t) extra) % __alignof__ (extra[0]) == 0);
assert (((uintptr_t) indirect) % __alignof__ (indirect[0]) == 0);
- /* We need this a few times. */
- s1len = STRLEN (s1);
- s2len = STRLEN (s2);
-
- /* Catch empty strings. */
- if (__builtin_expect (s1len == 0, 0) || __builtin_expect (s2len == 0, 0))
- return (s1len != 0) - (s2len != 0);
-
- /* We need the elements of the strings as unsigned values since they
- are used as indeces. */
- us1 = (const USTRING_TYPE *) s1;
- us2 = (const USTRING_TYPE *) s2;
-
- /* Perform the first pass over the string and while doing this find
- and store the weights for each character. Since we want this to
- be as fast as possible we are using `alloca' to store the temporary
- values. But since there is no limit on the length of the string
- we have to use `malloc' if the string is too long. We should be
- very conservative here.
-
- Please note that the localedef programs makes sure that `position'
- is not used at the first level. */
- if (! __libc_use_alloca ((s1len + s2len) * (sizeof (int32_t) + 1)))
+ int result = 0, rule = 0;
+
+ /* With GCC 7 when compiling with -Os the compiler warns that
+ seq1.back_us and seq2.back_us might be used uninitialized.
+ Sometimes this warning appears at locations in locale/weightwc.h
+ where the actual use is, but on architectures other than x86_64,
+ x86 and s390x, a warning appears at the definitions of seq1 and
+ seq2. This uninitialized use is impossible for the same reason
+ as described in comments in locale/weightwc.h. */
+ DIAG_PUSH_NEEDS_COMMENT;
+ DIAG_IGNORE_Os_NEEDS_COMMENT (7, "-Wmaybe-uninitialized");
+ coll_seq seq1, seq2;
+ DIAG_POP_NEEDS_COMMENT;
+ seq1.len = 0;
+ seq1.idxmax = 0;
+ seq1.rule = 0;
+ seq2.len = 0;
+ seq2.idxmax = 0;
+
+ for (int pass = 0; pass < nrules; ++pass)
{
- idx1arr = (int32_t *) malloc ((s1len + s2len) * (sizeof (int32_t) + 1));
- idx2arr = &idx1arr[s1len];
- rule1arr = (unsigned char *) &idx2arr[s2len];
- rule2arr = &rule1arr[s1len];
-
- if (idx1arr == NULL)
- /* No memory. Well, go with the stack then.
-
- XXX Once this implementation is stable we will handle this
- differently. Instead of precomputing the indeces we will
- do this in time. This means, though, that this happens for
- every pass again. */
- goto try_stack;
- use_malloc = 1;
- }
- else
- {
- try_stack:
- idx1arr = (int32_t *) alloca (s1len * sizeof (int32_t));
- idx2arr = (int32_t *) alloca (s2len * sizeof (int32_t));
- rule1arr = (unsigned char *) alloca (s1len);
- rule2arr = (unsigned char *) alloca (s2len);
- }
-
- idx1cnt = 0;
- idx2cnt = 0;
- idx1max = 0;
- idx2max = 0;
- idx1now = 0;
- idx2now = 0;
- backw1_stop = ~0ul;
- backw2_stop = ~0ul;
- backw1 = ~0ul;
- backw2 = ~0ul;
- seq1len = 0;
- seq2len = 0;
- position = rulesets[0] & sort_position;
- while (1)
- {
- val1 = 0;
- val2 = 0;
-
- /* Get the next non-IGNOREd element for string `s1'. */
- if (seq1len == 0)
- do
- {
- ++val1;
-
- if (backw1_stop != ~0ul)
- {
- /* The is something pushed. */
- if (backw1 == backw1_stop)
- {
- /* The last pushed character was handled. Continue
- with forward characters. */
- if (idx1cnt < idx1max)
- {
- idx1now = idx1cnt;
- backw1_stop = ~0ul;
- }
- else
- /* Nothing anymore. The backward sequence ended with
- the last sequence in the string. Note that seq1len
- is still zero. */
- break;
- }
- else
- idx1now = --backw1;
- }
- else
- {
- backw1_stop = idx1max;
-
- while (*us1 != L('\0'))
- {
- int32_t tmp = findidx (&us1);
- rule1arr[idx1max] = tmp >> 24;
- idx1arr[idx1max] = tmp & 0xffffff;
- idx1cnt = idx1max++;
-
- if ((rulesets[rule1arr[idx1cnt] * nrules]
- & sort_backward) == 0)
- /* No more backward characters to push. */
- break;
- ++idx1cnt;
- }
-
- if (backw1_stop >= idx1cnt)
- {
- /* No sequence at all or just one. */
- if (idx1cnt == idx1max || backw1_stop > idx1cnt)
- /* Note that seq1len is still zero. */
- break;
-
- backw1_stop = ~0ul;
- idx1now = idx1cnt;
- }
- else
- /* We pushed backward sequences. */
- idx1now = backw1 = idx1cnt - 1;
- }
- }
- while ((seq1len = weights[idx1arr[idx1now]++]) == 0);
-
- /* And the same for string `s2'. */
- if (seq2len == 0)
- do
- {
- ++val2;
-
- if (backw2_stop != ~0ul)
- {
- /* The is something pushed. */
- if (backw2 == backw2_stop)
- {
- /* The last pushed character was handled. Continue
- with forward characters. */
- if (idx2cnt < idx2max)
- {
- idx2now = idx2cnt;
- backw2_stop = ~0ul;
- }
- else
- /* Nothing anymore. The backward sequence ended with
- the last sequence in the string. Note that seq2len
- is still zero. */
- break;
- }
- else
- idx2now = --backw2;
- }
- else
- {
- backw2_stop = idx2max;
-
- while (*us2 != L('\0'))
- {
- int32_t tmp = findidx (&us2);
- rule2arr[idx2max] = tmp >> 24;
- idx2arr[idx2max] = tmp & 0xffffff;
- idx2cnt = idx2max++;
-
- if ((rulesets[rule2arr[idx2cnt] * nrules]
- & sort_backward) == 0)
- /* No more backward characters to push. */
- break;
- ++idx2cnt;
- }
-
- if (backw2_stop >= idx2cnt)
- {
- /* No sequence at all or just one. */
- if (idx2cnt == idx2max || backw2_stop > idx2cnt)
- /* Note that seq1len is still zero. */
- break;
-
- backw2_stop = ~0ul;
- idx2now = idx2cnt;
- }
- else
- /* We pushed backward sequences. */
- idx2now = backw2 = idx2cnt - 1;
- }
- }
- while ((seq2len = weights[idx2arr[idx2now]++]) == 0);
-
- /* See whether any or both strings are empty. */
- if (seq1len == 0 || seq2len == 0)
- {
- if (seq1len == seq2len)
- /* Both ended. So far so good, both strings are equal at the
- first level. */
- break;
-
- /* This means one string is shorter than the other. Find out
- which one and return an appropriate value. */
- result = seq1len == 0 ? -1 : 1;
- goto free_and_return;
- }
-
- /* Test for position if necessary. */
- if (position && val1 != val2)
- {
- result = val1 - val2;
- goto free_and_return;
- }
-
- /* Compare the two sequences. */
- do
- {
- if (weights[idx1arr[idx1now]] != weights[idx2arr[idx2now]])
- {
- /* The sequences differ. */
- result = weights[idx1arr[idx1now]] - weights[idx2arr[idx2now]];
- goto free_and_return;
- }
-
- /* Increment the offsets. */
- ++idx1arr[idx1now];
- ++idx2arr[idx2now];
+ seq1.idxcnt = 0;
+ seq1.idx = 0;
+ seq2.idx = 0;
+ seq1.backw_stop = ~0ul;
+ seq1.backw = ~0ul;
+ seq2.idxcnt = 0;
+ seq2.backw_stop = ~0ul;
+ seq2.backw = ~0ul;
+
+ /* We need the elements of the strings as unsigned values since they
+ are used as indices. */
+ seq1.us = (const USTRING_TYPE *) s1;
+ seq2.us = (const USTRING_TYPE *) s2;
- --seq1len;
- --seq2len;
- }
- while (seq1len > 0 && seq2len > 0);
-
- if (position && seq1len != seq2len)
- {
- result = seq1len - seq2len;
- goto free_and_return;
- }
- }
-
- /* Now the remaining passes over the weights. We now use the
- indeces we found before. */
- for (pass = 1; pass < nrules; ++pass)
- {
/* We assume that if a rule has defined `position' in one section
- this is true for all of them. */
- idx1cnt = 0;
- idx2cnt = 0;
- backw1_stop = ~0ul;
- backw2_stop = ~0ul;
- backw1 = ~0ul;
- backw2 = ~0ul;
- position = rulesets[rule1arr[0] * nrules + pass] & sort_position;
+ this is true for all of them. Please note that the localedef programs
+ makes sure that `position' is not used at the first level. */
+
+ int position = rulesets[rule * nrules + pass] & sort_position;
while (1)
{
- val1 = 0;
- val2 = 0;
-
- /* Get the next non-IGNOREd element for string `s1'. */
- if (seq1len == 0)
- do
- {
- ++val1;
-
- if (backw1_stop != ~0ul)
- {
- /* The is something pushed. */
- if (backw1 == backw1_stop)
- {
- /* The last pushed character was handled. Continue
- with forward characters. */
- if (idx1cnt < idx1max)
- {
- idx1now = idx1cnt;
- backw1_stop = ~0ul;
- }
- else
- {
- /* Nothing anymore. The backward sequence
- ended with the last sequence in the string. */
- idx1now = ~0ul;
- break;
- }
- }
- else
- idx1now = --backw1;
- }
- else
- {
- backw1_stop = idx1cnt;
-
- while (idx1cnt < idx1max)
- {
- if ((rulesets[rule1arr[idx1cnt] * nrules + pass]
- & sort_backward) == 0)
- /* No more backward characters to push. */
- break;
- ++idx1cnt;
- }
-
- if (backw1_stop == idx1cnt)
- {
- /* No sequence at all or just one. */
- if (idx1cnt == idx1max)
- /* Note that seq1len is still zero. */
- break;
-
- backw1_stop = ~0ul;
- idx1now = idx1cnt++;
- }
- else
- /* We pushed backward sequences. */
- idx1now = backw1 = idx1cnt - 1;
- }
- }
- while ((seq1len = weights[idx1arr[idx1now]++]) == 0);
-
- /* And the same for string `s2'. */
- if (seq2len == 0)
- do
- {
- ++val2;
-
- if (backw2_stop != ~0ul)
- {
- /* The is something pushed. */
- if (backw2 == backw2_stop)
- {
- /* The last pushed character was handled. Continue
- with forward characters. */
- if (idx2cnt < idx2max)
- {
- idx2now = idx2cnt;
- backw2_stop = ~0ul;
- }
- else
- {
- /* Nothing anymore. The backward sequence
- ended with the last sequence in the string. */
- idx2now = ~0ul;
- break;
- }
- }
- else
- idx2now = --backw2;
- }
- else
- {
- backw2_stop = idx2cnt;
-
- while (idx2cnt < idx2max)
- {
- if ((rulesets[rule2arr[idx2cnt] * nrules + pass]
- & sort_backward) == 0)
- /* No more backward characters to push. */
- break;
- ++idx2cnt;
- }
-
- if (backw2_stop == idx2cnt)
- {
- /* No sequence at all or just one. */
- if (idx2cnt == idx2max)
- /* Note that seq2len is still zero. */
- break;
-
- backw2_stop = ~0ul;
- idx2now = idx2cnt++;
- }
- else
- /* We pushed backward sequences. */
- idx2now = backw2 = idx2cnt - 1;
- }
- }
- while ((seq2len = weights[idx2arr[idx2now]++]) == 0);
-
+ get_next_seq (&seq1, nrules, rulesets, weights, table,
+ extra, indirect, pass);
+ get_next_seq (&seq2, nrules, rulesets, weights, table,
+ extra, indirect, pass);
/* See whether any or both strings are empty. */
- if (seq1len == 0 || seq2len == 0)
+ if (seq1.len == 0 || seq2.len == 0)
{
- if (seq1len == seq2len)
- /* Both ended. So far so good, both strings are equal
- at this level. */
- break;
+ if (seq1.len == seq2.len)
+ {
+ /* Both strings ended and are equal at this level. Do a
+ byte-level comparison to ensure that we don't waste time
+ going through multiple passes for totally equal strings
+ before proceeding to subsequent passes. */
+ if (pass == 0 && STRCMP (s1, s2) == 0)
+ return result;
+ else
+ break;
+ }
/* This means one string is shorter than the other. Find out
which one and return an appropriate value. */
- result = seq1len == 0 ? -1 : 1;
- goto free_and_return;
+ return seq1.len == 0 ? -1 : 1;
}
- /* Test for position if necessary. */
- if (position && val1 != val2)
- {
- result = val1 - val2;
- goto free_and_return;
- }
-
- /* Compare the two sequences. */
- do
- {
- if (weights[idx1arr[idx1now]] != weights[idx2arr[idx2now]])
- {
- /* The sequences differ. */
- result = (weights[idx1arr[idx1now]]
- - weights[idx2arr[idx2now]]);
- goto free_and_return;
- }
-
- /* Increment the offsets. */
- ++idx1arr[idx1now];
- ++idx2arr[idx2now];
-
- --seq1len;
- --seq2len;
- }
- while (seq1len > 0 && seq2len > 0);
-
- if (position && seq1len != seq2len)
- {
- result = seq1len - seq2len;
- goto free_and_return;
- }
+ result = do_compare (&seq1, &seq2, position, weights);
+ if (result != 0)
+ return result;
}
- }
- /* Free the memory if needed. */
- free_and_return:
- if (use_malloc)
- free (idx1arr);
+ rule = seq1.rule;
+ }
return result;
}
projects / thirdparty / glibc.git / blobdiff