|
|
29e444 |
commit e91a48408662e591e1320b2cab6d3fcda7793b18
|
|
|
29e444 |
Author: Siddhesh Poyarekar <siddhesh@redhat.com>
|
|
|
29e444 |
Date: Fri Jun 28 16:24:16 2013 +0530
|
|
|
29e444 |
|
|
|
29e444 |
Simplify strcoll implementation
|
|
|
29e444 |
|
|
|
12745e |
diff --git glibc-2.17-c758a686/string/strcoll_l.c glibc-2.17-c758a686/string/strcoll_l.c
|
|
|
29e444 |
index ecda08f..1bb9e23 100644
|
|
|
12745e |
--- glibc-2.17-c758a686/string/strcoll_l.c
|
|
|
12745e |
+++ glibc-2.17-c758a686/string/strcoll_l.c
|
|
|
29e444 |
@@ -41,11 +41,244 @@
|
|
|
29e444 |
|
|
|
29e444 |
#include "../locale/localeinfo.h"
|
|
|
29e444 |
|
|
|
29e444 |
+/* Track status while looking for sequences in a string. */
|
|
|
29e444 |
+typedef struct
|
|
|
29e444 |
+{
|
|
|
29e444 |
+ int len; /* Length of the current sequence. */
|
|
|
29e444 |
+ int val; /* Position of the sequence relative to the
|
|
|
29e444 |
+ previous non-ignored sequence. */
|
|
|
29e444 |
+ size_t idxnow; /* Current index in sequences. */
|
|
|
29e444 |
+ size_t idxmax; /* Maximum index in sequences. */
|
|
|
29e444 |
+ size_t idxcnt; /* Current count of indeces. */
|
|
|
29e444 |
+ size_t backw; /* Current Backward sequence index. */
|
|
|
29e444 |
+ size_t backw_stop; /* Index where the backward sequences stop. */
|
|
|
29e444 |
+ const USTRING_TYPE *us; /* The string. */
|
|
|
29e444 |
+ int32_t *idxarr; /* Array to cache weight indeces. */
|
|
|
29e444 |
+ unsigned char *rulearr; /* Array to cache rules. */
|
|
|
29e444 |
+} coll_seq;
|
|
|
29e444 |
+
|
|
|
29e444 |
+/* Get next sequence. The weight indeces are cached, so we don't need to
|
|
|
29e444 |
+ traverse the string. */
|
|
|
29e444 |
+static void
|
|
|
29e444 |
+get_next_seq_cached (coll_seq *seq, int nrules, int pass,
|
|
|
29e444 |
+ const unsigned char *rulesets,
|
|
|
29e444 |
+ const USTRING_TYPE *weights)
|
|
|
29e444 |
+{
|
|
|
29e444 |
+ int val = seq->val = 0;
|
|
|
29e444 |
+ int len = seq->len;
|
|
|
29e444 |
+ size_t backw_stop = seq->backw_stop;
|
|
|
29e444 |
+ size_t backw = seq->backw;
|
|
|
29e444 |
+ size_t idxcnt = seq->idxcnt;
|
|
|
29e444 |
+ size_t idxmax = seq->idxmax;
|
|
|
29e444 |
+ size_t idxnow = seq->idxnow;
|
|
|
29e444 |
+ unsigned char *rulearr = seq->rulearr;
|
|
|
29e444 |
+ int32_t *idxarr = seq->idxarr;
|
|
|
29e444 |
+
|
|
|
29e444 |
+ while (len == 0)
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ ++val;
|
|
|
29e444 |
+ if (backw_stop != ~0ul)
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ /* The is something pushed. */
|
|
|
29e444 |
+ if (backw == backw_stop)
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ /* The last pushed character was handled. Continue
|
|
|
29e444 |
+ with forward characters. */
|
|
|
29e444 |
+ if (idxcnt < idxmax)
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ idxnow = idxcnt;
|
|
|
29e444 |
+ backw_stop = ~0ul;
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+ else
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ /* Nothing anymore. The backward sequence
|
|
|
29e444 |
+ ended with the last sequence in the string. */
|
|
|
29e444 |
+ idxnow = ~0ul;
|
|
|
29e444 |
+ break;
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+ else
|
|
|
29e444 |
+ idxnow = --backw;
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+ else
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ backw_stop = idxcnt;
|
|
|
29e444 |
+
|
|
|
29e444 |
+ while (idxcnt < idxmax)
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ if ((rulesets[rulearr[idxcnt] * nrules + pass]
|
|
|
29e444 |
+ & sort_backward) == 0)
|
|
|
29e444 |
+ /* No more backward characters to push. */
|
|
|
29e444 |
+ break;
|
|
|
29e444 |
+ ++idxcnt;
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+
|
|
|
29e444 |
+ if (backw_stop == idxcnt)
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ /* No sequence at all or just one. */
|
|
|
29e444 |
+ if (idxcnt == idxmax)
|
|
|
29e444 |
+ /* Note that seq1len is still zero. */
|
|
|
29e444 |
+ break;
|
|
|
29e444 |
+
|
|
|
29e444 |
+ backw_stop = ~0ul;
|
|
|
29e444 |
+ idxnow = idxcnt++;
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+ else
|
|
|
29e444 |
+ /* We pushed backward sequences. */
|
|
|
29e444 |
+ idxnow = backw = idxcnt - 1;
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+ len = weights[idxarr[idxnow]++];
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+
|
|
|
29e444 |
+ /* Update the structure. */
|
|
|
29e444 |
+ seq->val = val;
|
|
|
29e444 |
+ seq->len = len;
|
|
|
29e444 |
+ seq->backw_stop = backw_stop;
|
|
|
29e444 |
+ seq->backw = backw;
|
|
|
29e444 |
+ seq->idxcnt = idxcnt;
|
|
|
29e444 |
+ seq->idxnow = idxnow;
|
|
|
29e444 |
+}
|
|
|
29e444 |
+
|
|
|
29e444 |
+/* Get next sequence. Traverse the string as required. */
|
|
|
29e444 |
+static void
|
|
|
29e444 |
+get_next_seq (coll_seq *seq, int nrules, const unsigned char *rulesets,
|
|
|
29e444 |
+ const USTRING_TYPE *weights, const int32_t *table,
|
|
|
29e444 |
+ const USTRING_TYPE *extra, const int32_t *indirect)
|
|
|
29e444 |
+{
|
|
|
29e444 |
+#include WEIGHT_H
|
|
|
29e444 |
+ int val = seq->val = 0;
|
|
|
29e444 |
+ int len = seq->len;
|
|
|
29e444 |
+ size_t backw_stop = seq->backw_stop;
|
|
|
29e444 |
+ size_t backw = seq->backw;
|
|
|
29e444 |
+ size_t idxcnt = seq->idxcnt;
|
|
|
29e444 |
+ size_t idxmax = seq->idxmax;
|
|
|
29e444 |
+ size_t idxnow = seq->idxnow;
|
|
|
29e444 |
+ unsigned char *rulearr = seq->rulearr;
|
|
|
29e444 |
+ int32_t *idxarr = seq->idxarr;
|
|
|
29e444 |
+ const USTRING_TYPE *us = seq->us;
|
|
|
29e444 |
+
|
|
|
29e444 |
+ while (len == 0)
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ ++val;
|
|
|
29e444 |
+ if (backw_stop != ~0ul)
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ /* The is something pushed. */
|
|
|
29e444 |
+ if (backw == backw_stop)
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ /* The last pushed character was handled. Continue
|
|
|
29e444 |
+ with forward characters. */
|
|
|
29e444 |
+ if (idxcnt < idxmax)
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ idxnow = idxcnt;
|
|
|
29e444 |
+ backw_stop = ~0ul;
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+ else
|
|
|
29e444 |
+ /* Nothing anymore. The backward sequence ended with
|
|
|
29e444 |
+ the last sequence in the string. Note that seq2len
|
|
|
29e444 |
+ is still zero. */
|
|
|
29e444 |
+ break;
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+ else
|
|
|
29e444 |
+ idxnow = --backw;
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+ else
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ backw_stop = idxmax;
|
|
|
29e444 |
+
|
|
|
29e444 |
+ while (*us != L('\0'))
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ int32_t tmp = findidx (&us, -1);
|
|
|
29e444 |
+ rulearr[idxmax] = tmp >> 24;
|
|
|
29e444 |
+ idxarr[idxmax] = tmp & 0xffffff;
|
|
|
29e444 |
+ idxcnt = idxmax++;
|
|
|
29e444 |
+
|
|
|
29e444 |
+ if ((rulesets[rulearr[idxcnt] * nrules]
|
|
|
29e444 |
+ & sort_backward) == 0)
|
|
|
29e444 |
+ /* No more backward characters to push. */
|
|
|
29e444 |
+ break;
|
|
|
29e444 |
+ ++idxcnt;
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+
|
|
|
29e444 |
+ if (backw_stop >= idxcnt)
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ /* No sequence at all or just one. */
|
|
|
29e444 |
+ if (idxcnt == idxmax || backw_stop > idxcnt)
|
|
|
29e444 |
+ /* Note that seq1len is still zero. */
|
|
|
29e444 |
+ break;
|
|
|
29e444 |
+
|
|
|
29e444 |
+ backw_stop = ~0ul;
|
|
|
29e444 |
+ idxnow = idxcnt;
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+ else
|
|
|
29e444 |
+ /* We pushed backward sequences. */
|
|
|
29e444 |
+ idxnow = backw = idxcnt - 1;
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+ len = weights[idxarr[idxnow]++];
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+
|
|
|
29e444 |
+ /* Update the structure. */
|
|
|
29e444 |
+ seq->val = val;
|
|
|
29e444 |
+ seq->len = len;
|
|
|
29e444 |
+ seq->backw_stop = backw_stop;
|
|
|
29e444 |
+ seq->backw = backw;
|
|
|
29e444 |
+ seq->idxcnt = idxcnt;
|
|
|
29e444 |
+ seq->idxmax = idxmax;
|
|
|
29e444 |
+ seq->idxnow = idxnow;
|
|
|
29e444 |
+ seq->us = us;
|
|
|
29e444 |
+}
|
|
|
29e444 |
+
|
|
|
29e444 |
+/* Compare two sequences. */
|
|
|
29e444 |
+static int
|
|
|
29e444 |
+do_compare (coll_seq *seq1, coll_seq *seq2, int position,
|
|
|
29e444 |
+ const USTRING_TYPE *weights)
|
|
|
29e444 |
+{
|
|
|
29e444 |
+ int seq1len = seq1->len;
|
|
|
29e444 |
+ int seq2len = seq2->len;
|
|
|
29e444 |
+ int val1 = seq1->val;
|
|
|
29e444 |
+ int val2 = seq2->val;
|
|
|
29e444 |
+ int32_t *idx1arr = seq1->idxarr;
|
|
|
29e444 |
+ int32_t *idx2arr = seq2->idxarr;
|
|
|
29e444 |
+ int idx1now = seq1->idxnow;
|
|
|
29e444 |
+ int idx2now = seq2->idxnow;
|
|
|
29e444 |
+ int result = 0;
|
|
|
29e444 |
+
|
|
|
29e444 |
+ /* Test for position if necessary. */
|
|
|
29e444 |
+ if (position && val1 != val2)
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ result = val1 - val2;
|
|
|
29e444 |
+ goto out;
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+
|
|
|
29e444 |
+ /* Compare the two sequences. */
|
|
|
29e444 |
+ do
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ if (weights[idx1arr[idx1now]] != weights[idx2arr[idx2now]])
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ /* The sequences differ. */
|
|
|
29e444 |
+ result = weights[idx1arr[idx1now]] - weights[idx2arr[idx2now]];
|
|
|
29e444 |
+ goto out;
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+
|
|
|
29e444 |
+ /* Increment the offsets. */
|
|
|
29e444 |
+ ++idx1arr[idx1now];
|
|
|
29e444 |
+ ++idx2arr[idx2now];
|
|
|
29e444 |
+
|
|
|
29e444 |
+ --seq1len;
|
|
|
29e444 |
+ --seq2len;
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+ while (seq1len > 0 && seq2len > 0);
|
|
|
29e444 |
+
|
|
|
29e444 |
+ if (position && seq1len != seq2len)
|
|
|
29e444 |
+ result = seq1len - seq2len;
|
|
|
29e444 |
+
|
|
|
29e444 |
+out:
|
|
|
29e444 |
+ seq1->len = seq1len;
|
|
|
29e444 |
+ seq2->len = seq2len;
|
|
|
29e444 |
+ return result;
|
|
|
29e444 |
+}
|
|
|
29e444 |
+
|
|
|
29e444 |
int
|
|
|
29e444 |
-STRCOLL (s1, s2, l)
|
|
|
29e444 |
- const STRING_TYPE *s1;
|
|
|
29e444 |
- const STRING_TYPE *s2;
|
|
|
29e444 |
- __locale_t l;
|
|
|
29e444 |
+STRCOLL (const STRING_TYPE *s1, const STRING_TYPE *s2, __locale_t l)
|
|
|
29e444 |
{
|
|
|
29e444 |
struct __locale_data *current = l->__locales[LC_COLLATE];
|
|
|
29e444 |
uint_fast32_t nrules = current->values[_NL_ITEM_INDEX (_NL_COLLATE_NRULES)].word;
|
|
|
29e444 |
@@ -56,34 +289,6 @@ STRCOLL (s1, s2, l)
|
|
|
29e444 |
const USTRING_TYPE *weights;
|
|
|
29e444 |
const USTRING_TYPE *extra;
|
|
|
29e444 |
const int32_t *indirect;
|
|
|
29e444 |
- uint_fast32_t pass;
|
|
|
29e444 |
- int result = 0;
|
|
|
29e444 |
- const USTRING_TYPE *us1;
|
|
|
29e444 |
- const USTRING_TYPE *us2;
|
|
|
29e444 |
- size_t s1len;
|
|
|
29e444 |
- size_t s2len;
|
|
|
29e444 |
- int32_t *idx1arr;
|
|
|
29e444 |
- int32_t *idx2arr;
|
|
|
29e444 |
- unsigned char *rule1arr;
|
|
|
29e444 |
- unsigned char *rule2arr;
|
|
|
29e444 |
- size_t idx1max;
|
|
|
29e444 |
- size_t idx2max;
|
|
|
29e444 |
- size_t idx1cnt;
|
|
|
29e444 |
- size_t idx2cnt;
|
|
|
29e444 |
- size_t idx1now;
|
|
|
29e444 |
- size_t idx2now;
|
|
|
29e444 |
- size_t backw1_stop;
|
|
|
29e444 |
- size_t backw2_stop;
|
|
|
29e444 |
- size_t backw1;
|
|
|
29e444 |
- size_t backw2;
|
|
|
29e444 |
- int val1;
|
|
|
29e444 |
- int val2;
|
|
|
29e444 |
- int position;
|
|
|
29e444 |
- int seq1len;
|
|
|
29e444 |
- int seq2len;
|
|
|
29e444 |
- int use_malloc;
|
|
|
29e444 |
-
|
|
|
29e444 |
-#include WEIGHT_H
|
|
|
29e444 |
|
|
|
29e444 |
if (nrules == 0)
|
|
|
29e444 |
return STRCMP (s1, s2);
|
|
|
29e444 |
@@ -98,7 +303,6 @@ STRCOLL (s1, s2, l)
|
|
|
29e444 |
current->values[_NL_ITEM_INDEX (CONCAT(_NL_COLLATE_EXTRA,SUFFIX))].string;
|
|
|
29e444 |
indirect = (const int32_t *)
|
|
|
29e444 |
current->values[_NL_ITEM_INDEX (CONCAT(_NL_COLLATE_INDIRECT,SUFFIX))].string;
|
|
|
29e444 |
- use_malloc = 0;
|
|
|
29e444 |
|
|
|
29e444 |
assert (((uintptr_t) table) % __alignof__ (table[0]) == 0);
|
|
|
29e444 |
assert (((uintptr_t) weights) % __alignof__ (weights[0]) == 0);
|
|
|
29e444 |
@@ -106,18 +310,13 @@ STRCOLL (s1, s2, l)
|
|
|
29e444 |
assert (((uintptr_t) indirect) % __alignof__ (indirect[0]) == 0);
|
|
|
29e444 |
|
|
|
29e444 |
/* We need this a few times. */
|
|
|
29e444 |
- s1len = STRLEN (s1);
|
|
|
29e444 |
- s2len = STRLEN (s2);
|
|
|
29e444 |
+ size_t s1len = STRLEN (s1);
|
|
|
29e444 |
+ size_t s2len = STRLEN (s2);
|
|
|
29e444 |
|
|
|
29e444 |
/* Catch empty strings. */
|
|
|
29e444 |
- if (__builtin_expect (s1len == 0, 0) || __builtin_expect (s2len == 0, 0))
|
|
|
29e444 |
+ if (__glibc_unlikely (s1len == 0) || __glibc_unlikely (s2len == 0))
|
|
|
29e444 |
return (s1len != 0) - (s2len != 0);
|
|
|
29e444 |
|
|
|
29e444 |
- /* We need the elements of the strings as unsigned values since they
|
|
|
29e444 |
- are used as indeces. */
|
|
|
29e444 |
- us1 = (const USTRING_TYPE *) s1;
|
|
|
29e444 |
- us2 = (const USTRING_TYPE *) s2;
|
|
|
29e444 |
-
|
|
|
29e444 |
/* Perform the first pass over the string and while doing this find
|
|
|
29e444 |
and store the weights for each character. Since we want this to
|
|
|
29e444 |
be as fast as possible we are using `alloca' to store the temporary
|
|
|
29e444 |
@@ -127,14 +326,27 @@ STRCOLL (s1, s2, l)
|
|
|
29e444 |
|
|
|
29e444 |
Please note that the localedef programs makes sure that `position'
|
|
|
29e444 |
is not used at the first level. */
|
|
|
29e444 |
+
|
|
|
29e444 |
+ coll_seq seq1, seq2;
|
|
|
29e444 |
+ bool use_malloc = false;
|
|
|
29e444 |
+ int result = 0;
|
|
|
29e444 |
+
|
|
|
29e444 |
+ memset (&seq1, 0, sizeof (seq1));
|
|
|
29e444 |
+ seq2 = seq1;
|
|
|
29e444 |
+
|
|
|
29e444 |
+ /* We need the elements of the strings as unsigned values since they
|
|
|
29e444 |
+ are used as indeces. */
|
|
|
29e444 |
+ seq1.us = (const USTRING_TYPE *) s1;
|
|
|
29e444 |
+ seq2.us = (const USTRING_TYPE *) s2;
|
|
|
29e444 |
+
|
|
|
29e444 |
if (! __libc_use_alloca ((s1len + s2len) * (sizeof (int32_t) + 1)))
|
|
|
29e444 |
{
|
|
|
29e444 |
- idx1arr = (int32_t *) malloc ((s1len + s2len) * (sizeof (int32_t) + 1));
|
|
|
29e444 |
- idx2arr = &idx1arr[s1len];
|
|
|
29e444 |
- rule1arr = (unsigned char *) &idx2arr[s2len];
|
|
|
29e444 |
- rule2arr = &rule1arr[s1len];
|
|
|
29e444 |
+ seq1.idxarr = (int32_t *) malloc ((s1len + s2len) * (sizeof (int32_t) + 1));
|
|
|
29e444 |
+ seq2.idxarr = &seq1.idxarr[s1len];
|
|
|
29e444 |
+ seq1.rulearr = (unsigned char *) &seq2.idxarr[s2len];
|
|
|
29e444 |
+ seq2.rulearr = &seq1.rulearr[s1len];
|
|
|
29e444 |
|
|
|
29e444 |
- if (idx1arr == NULL)
|
|
|
29e444 |
+ if (seq1.idxarr == NULL)
|
|
|
29e444 |
/* No memory. Well, go with the stack then.
|
|
|
29e444 |
|
|
|
29e444 |
XXX Once this implementation is stable we will handle this
|
|
|
29e444 |
@@ -142,396 +354,73 @@ STRCOLL (s1, s2, l)
|
|
|
29e444 |
do this in time. This means, though, that this happens for
|
|
|
29e444 |
every pass again. */
|
|
|
29e444 |
goto try_stack;
|
|
|
29e444 |
- use_malloc = 1;
|
|
|
29e444 |
+ use_malloc = true;
|
|
|
29e444 |
}
|
|
|
29e444 |
else
|
|
|
29e444 |
{
|
|
|
29e444 |
try_stack:
|
|
|
29e444 |
- idx1arr = (int32_t *) alloca (s1len * sizeof (int32_t));
|
|
|
29e444 |
- idx2arr = (int32_t *) alloca (s2len * sizeof (int32_t));
|
|
|
29e444 |
- rule1arr = (unsigned char *) alloca (s1len);
|
|
|
29e444 |
- rule2arr = (unsigned char *) alloca (s2len);
|
|
|
29e444 |
+ seq1.idxarr = (int32_t *) alloca (s1len * sizeof (int32_t));
|
|
|
29e444 |
+ seq2.idxarr = (int32_t *) alloca (s2len * sizeof (int32_t));
|
|
|
29e444 |
+ seq1.rulearr = (unsigned char *) alloca (s1len);
|
|
|
29e444 |
+ seq2.rulearr = (unsigned char *) alloca (s2len);
|
|
|
29e444 |
}
|
|
|
29e444 |
|
|
|
29e444 |
- idx1cnt = 0;
|
|
|
29e444 |
- idx2cnt = 0;
|
|
|
29e444 |
- idx1max = 0;
|
|
|
29e444 |
- idx2max = 0;
|
|
|
29e444 |
- idx1now = 0;
|
|
|
29e444 |
- idx2now = 0;
|
|
|
29e444 |
- backw1_stop = ~0ul;
|
|
|
29e444 |
- backw2_stop = ~0ul;
|
|
|
29e444 |
- backw1 = ~0ul;
|
|
|
29e444 |
- backw2 = ~0ul;
|
|
|
29e444 |
- seq1len = 0;
|
|
|
29e444 |
- seq2len = 0;
|
|
|
29e444 |
- position = rulesets[0] & sort_position;
|
|
|
29e444 |
- while (1)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- val1 = 0;
|
|
|
29e444 |
- val2 = 0;
|
|
|
29e444 |
-
|
|
|
29e444 |
- /* Get the next non-IGNOREd element for string `s1'. */
|
|
|
29e444 |
- if (seq1len == 0)
|
|
|
29e444 |
- do
|
|
|
29e444 |
- {
|
|
|
29e444 |
- ++val1;
|
|
|
29e444 |
-
|
|
|
29e444 |
- if (backw1_stop != ~0ul)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- /* The is something pushed. */
|
|
|
29e444 |
- if (backw1 == backw1_stop)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- /* The last pushed character was handled. Continue
|
|
|
29e444 |
- with forward characters. */
|
|
|
29e444 |
- if (idx1cnt < idx1max)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- idx1now = idx1cnt;
|
|
|
29e444 |
- backw1_stop = ~0ul;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- else
|
|
|
29e444 |
- /* Nothing anymore. The backward sequence ended with
|
|
|
29e444 |
- the last sequence in the string. Note that seq1len
|
|
|
29e444 |
- is still zero. */
|
|
|
29e444 |
- break;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- else
|
|
|
29e444 |
- idx1now = --backw1;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- else
|
|
|
29e444 |
- {
|
|
|
29e444 |
- backw1_stop = idx1max;
|
|
|
29e444 |
-
|
|
|
29e444 |
- while (*us1 != L('\0'))
|
|
|
29e444 |
- {
|
|
|
29e444 |
- int32_t tmp = findidx (&us1, -1);
|
|
|
29e444 |
- rule1arr[idx1max] = tmp >> 24;
|
|
|
29e444 |
- idx1arr[idx1max] = tmp & 0xffffff;
|
|
|
29e444 |
- idx1cnt = idx1max++;
|
|
|
29e444 |
-
|
|
|
29e444 |
- if ((rulesets[rule1arr[idx1cnt] * nrules]
|
|
|
29e444 |
- & sort_backward) == 0)
|
|
|
29e444 |
- /* No more backward characters to push. */
|
|
|
29e444 |
- break;
|
|
|
29e444 |
- ++idx1cnt;
|
|
|
29e444 |
- }
|
|
|
29e444 |
-
|
|
|
29e444 |
- if (backw1_stop >= idx1cnt)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- /* No sequence at all or just one. */
|
|
|
29e444 |
- if (idx1cnt == idx1max || backw1_stop > idx1cnt)
|
|
|
29e444 |
- /* Note that seq1len is still zero. */
|
|
|
29e444 |
- break;
|
|
|
29e444 |
-
|
|
|
29e444 |
- backw1_stop = ~0ul;
|
|
|
29e444 |
- idx1now = idx1cnt;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- else
|
|
|
29e444 |
- /* We pushed backward sequences. */
|
|
|
29e444 |
- idx1now = backw1 = idx1cnt - 1;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- }
|
|
|
29e444 |
- while ((seq1len = weights[idx1arr[idx1now]++]) == 0);
|
|
|
29e444 |
-
|
|
|
29e444 |
- /* And the same for string `s2'. */
|
|
|
29e444 |
- if (seq2len == 0)
|
|
|
29e444 |
- do
|
|
|
29e444 |
- {
|
|
|
29e444 |
- ++val2;
|
|
|
29e444 |
-
|
|
|
29e444 |
- if (backw2_stop != ~0ul)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- /* The is something pushed. */
|
|
|
29e444 |
- if (backw2 == backw2_stop)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- /* The last pushed character was handled. Continue
|
|
|
29e444 |
- with forward characters. */
|
|
|
29e444 |
- if (idx2cnt < idx2max)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- idx2now = idx2cnt;
|
|
|
29e444 |
- backw2_stop = ~0ul;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- else
|
|
|
29e444 |
- /* Nothing anymore. The backward sequence ended with
|
|
|
29e444 |
- the last sequence in the string. Note that seq2len
|
|
|
29e444 |
- is still zero. */
|
|
|
29e444 |
- break;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- else
|
|
|
29e444 |
- idx2now = --backw2;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- else
|
|
|
29e444 |
- {
|
|
|
29e444 |
- backw2_stop = idx2max;
|
|
|
29e444 |
-
|
|
|
29e444 |
- while (*us2 != L('\0'))
|
|
|
29e444 |
- {
|
|
|
29e444 |
- int32_t tmp = findidx (&us2, -1);
|
|
|
29e444 |
- rule2arr[idx2max] = tmp >> 24;
|
|
|
29e444 |
- idx2arr[idx2max] = tmp & 0xffffff;
|
|
|
29e444 |
- idx2cnt = idx2max++;
|
|
|
29e444 |
-
|
|
|
29e444 |
- if ((rulesets[rule2arr[idx2cnt] * nrules]
|
|
|
29e444 |
- & sort_backward) == 0)
|
|
|
29e444 |
- /* No more backward characters to push. */
|
|
|
29e444 |
- break;
|
|
|
29e444 |
- ++idx2cnt;
|
|
|
29e444 |
- }
|
|
|
29e444 |
-
|
|
|
29e444 |
- if (backw2_stop >= idx2cnt)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- /* No sequence at all or just one. */
|
|
|
29e444 |
- if (idx2cnt == idx2max || backw2_stop > idx2cnt)
|
|
|
29e444 |
- /* Note that seq1len is still zero. */
|
|
|
29e444 |
- break;
|
|
|
29e444 |
-
|
|
|
29e444 |
- backw2_stop = ~0ul;
|
|
|
29e444 |
- idx2now = idx2cnt;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- else
|
|
|
29e444 |
- /* We pushed backward sequences. */
|
|
|
29e444 |
- idx2now = backw2 = idx2cnt - 1;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- }
|
|
|
29e444 |
- while ((seq2len = weights[idx2arr[idx2now]++]) == 0);
|
|
|
29e444 |
-
|
|
|
29e444 |
- /* See whether any or both strings are empty. */
|
|
|
29e444 |
- if (seq1len == 0 || seq2len == 0)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- if (seq1len == seq2len)
|
|
|
29e444 |
- /* Both ended. So far so good, both strings are equal at the
|
|
|
29e444 |
- first level. */
|
|
|
29e444 |
- break;
|
|
|
29e444 |
-
|
|
|
29e444 |
- /* This means one string is shorter than the other. Find out
|
|
|
29e444 |
- which one and return an appropriate value. */
|
|
|
29e444 |
- result = seq1len == 0 ? -1 : 1;
|
|
|
29e444 |
- goto free_and_return;
|
|
|
29e444 |
- }
|
|
|
29e444 |
+ seq1.rulearr[0] = 0;
|
|
|
29e444 |
|
|
|
29e444 |
- /* Test for position if necessary. */
|
|
|
29e444 |
- if (position && val1 != val2)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- result = val1 - val2;
|
|
|
29e444 |
- goto free_and_return;
|
|
|
29e444 |
- }
|
|
|
29e444 |
-
|
|
|
29e444 |
- /* Compare the two sequences. */
|
|
|
29e444 |
- do
|
|
|
29e444 |
- {
|
|
|
29e444 |
- if (weights[idx1arr[idx1now]] != weights[idx2arr[idx2now]])
|
|
|
29e444 |
- {
|
|
|
29e444 |
- /* The sequences differ. */
|
|
|
29e444 |
- result = weights[idx1arr[idx1now]] - weights[idx2arr[idx2now]];
|
|
|
29e444 |
- goto free_and_return;
|
|
|
29e444 |
- }
|
|
|
29e444 |
-
|
|
|
29e444 |
- /* Increment the offsets. */
|
|
|
29e444 |
- ++idx1arr[idx1now];
|
|
|
29e444 |
- ++idx2arr[idx2now];
|
|
|
29e444 |
-
|
|
|
29e444 |
- --seq1len;
|
|
|
29e444 |
- --seq2len;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- while (seq1len > 0 && seq2len > 0);
|
|
|
29e444 |
-
|
|
|
29e444 |
- if (position && seq1len != seq2len)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- result = seq1len - seq2len;
|
|
|
29e444 |
- goto free_and_return;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- }
|
|
|
29e444 |
-
|
|
|
29e444 |
- /* Now the remaining passes over the weights. We now use the
|
|
|
29e444 |
- indeces we found before. */
|
|
|
29e444 |
- for (pass = 1; pass < nrules; ++pass)
|
|
|
29e444 |
+ /* Cache values in the first pass and if needed, use them in subsequent
|
|
|
29e444 |
+ passes. */
|
|
|
29e444 |
+ for (int pass = 0; pass < nrules; ++pass)
|
|
|
29e444 |
{
|
|
|
29e444 |
+ seq1.idxcnt = 0;
|
|
|
29e444 |
+ seq1.backw_stop = ~0ul;
|
|
|
29e444 |
+ seq1.backw = ~0ul;
|
|
|
29e444 |
+ seq2.idxcnt = 0;
|
|
|
29e444 |
+ seq2.backw_stop = ~0ul;
|
|
|
29e444 |
+ seq2.backw = ~0ul;
|
|
|
29e444 |
+
|
|
|
29e444 |
/* We assume that if a rule has defined `position' in one section
|
|
|
29e444 |
this is true for all of them. */
|
|
|
29e444 |
- idx1cnt = 0;
|
|
|
29e444 |
- idx2cnt = 0;
|
|
|
29e444 |
- backw1_stop = ~0ul;
|
|
|
29e444 |
- backw2_stop = ~0ul;
|
|
|
29e444 |
- backw1 = ~0ul;
|
|
|
29e444 |
- backw2 = ~0ul;
|
|
|
29e444 |
- position = rulesets[rule1arr[0] * nrules + pass] & sort_position;
|
|
|
29e444 |
+ int position = rulesets[seq1.rulearr[0] * nrules + pass] & sort_position;
|
|
|
29e444 |
|
|
|
29e444 |
while (1)
|
|
|
29e444 |
{
|
|
|
29e444 |
- val1 = 0;
|
|
|
29e444 |
- val2 = 0;
|
|
|
29e444 |
-
|
|
|
29e444 |
- /* Get the next non-IGNOREd element for string `s1'. */
|
|
|
29e444 |
- if (seq1len == 0)
|
|
|
29e444 |
- do
|
|
|
29e444 |
- {
|
|
|
29e444 |
- ++val1;
|
|
|
29e444 |
-
|
|
|
29e444 |
- if (backw1_stop != ~0ul)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- /* The is something pushed. */
|
|
|
29e444 |
- if (backw1 == backw1_stop)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- /* The last pushed character was handled. Continue
|
|
|
29e444 |
- with forward characters. */
|
|
|
29e444 |
- if (idx1cnt < idx1max)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- idx1now = idx1cnt;
|
|
|
29e444 |
- backw1_stop = ~0ul;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- else
|
|
|
29e444 |
- {
|
|
|
29e444 |
- /* Nothing anymore. The backward sequence
|
|
|
29e444 |
- ended with the last sequence in the string. */
|
|
|
29e444 |
- idx1now = ~0ul;
|
|
|
29e444 |
- break;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- }
|
|
|
29e444 |
- else
|
|
|
29e444 |
- idx1now = --backw1;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- else
|
|
|
29e444 |
- {
|
|
|
29e444 |
- backw1_stop = idx1cnt;
|
|
|
29e444 |
-
|
|
|
29e444 |
- while (idx1cnt < idx1max)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- if ((rulesets[rule1arr[idx1cnt] * nrules + pass]
|
|
|
29e444 |
- & sort_backward) == 0)
|
|
|
29e444 |
- /* No more backward characters to push. */
|
|
|
29e444 |
- break;
|
|
|
29e444 |
- ++idx1cnt;
|
|
|
29e444 |
- }
|
|
|
29e444 |
-
|
|
|
29e444 |
- if (backw1_stop == idx1cnt)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- /* No sequence at all or just one. */
|
|
|
29e444 |
- if (idx1cnt == idx1max)
|
|
|
29e444 |
- /* Note that seq1len is still zero. */
|
|
|
29e444 |
- break;
|
|
|
29e444 |
-
|
|
|
29e444 |
- backw1_stop = ~0ul;
|
|
|
29e444 |
- idx1now = idx1cnt++;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- else
|
|
|
29e444 |
- /* We pushed backward sequences. */
|
|
|
29e444 |
- idx1now = backw1 = idx1cnt - 1;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- }
|
|
|
29e444 |
- while ((seq1len = weights[idx1arr[idx1now]++]) == 0);
|
|
|
29e444 |
-
|
|
|
29e444 |
- /* And the same for string `s2'. */
|
|
|
29e444 |
- if (seq2len == 0)
|
|
|
29e444 |
- do
|
|
|
29e444 |
- {
|
|
|
29e444 |
- ++val2;
|
|
|
29e444 |
-
|
|
|
29e444 |
- if (backw2_stop != ~0ul)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- /* The is something pushed. */
|
|
|
29e444 |
- if (backw2 == backw2_stop)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- /* The last pushed character was handled. Continue
|
|
|
29e444 |
- with forward characters. */
|
|
|
29e444 |
- if (idx2cnt < idx2max)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- idx2now = idx2cnt;
|
|
|
29e444 |
- backw2_stop = ~0ul;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- else
|
|
|
29e444 |
- {
|
|
|
29e444 |
- /* Nothing anymore. The backward sequence
|
|
|
29e444 |
- ended with the last sequence in the string. */
|
|
|
29e444 |
- idx2now = ~0ul;
|
|
|
29e444 |
- break;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- }
|
|
|
29e444 |
- else
|
|
|
29e444 |
- idx2now = --backw2;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- else
|
|
|
29e444 |
- {
|
|
|
29e444 |
- backw2_stop = idx2cnt;
|
|
|
29e444 |
-
|
|
|
29e444 |
- while (idx2cnt < idx2max)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- if ((rulesets[rule2arr[idx2cnt] * nrules + pass]
|
|
|
29e444 |
- & sort_backward) == 0)
|
|
|
29e444 |
- /* No more backward characters to push. */
|
|
|
29e444 |
- break;
|
|
|
29e444 |
- ++idx2cnt;
|
|
|
29e444 |
- }
|
|
|
29e444 |
-
|
|
|
29e444 |
- if (backw2_stop == idx2cnt)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- /* No sequence at all or just one. */
|
|
|
29e444 |
- if (idx2cnt == idx2max)
|
|
|
29e444 |
- /* Note that seq2len is still zero. */
|
|
|
29e444 |
- break;
|
|
|
29e444 |
-
|
|
|
29e444 |
- backw2_stop = ~0ul;
|
|
|
29e444 |
- idx2now = idx2cnt++;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- else
|
|
|
29e444 |
- /* We pushed backward sequences. */
|
|
|
29e444 |
- idx2now = backw2 = idx2cnt - 1;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- }
|
|
|
29e444 |
- while ((seq2len = weights[idx2arr[idx2now]++]) == 0);
|
|
|
29e444 |
+ if (pass == 0)
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ get_next_seq (&seq1, nrules, rulesets, weights, table, extra,
|
|
|
29e444 |
+ indirect);
|
|
|
29e444 |
+ get_next_seq (&seq2, nrules, rulesets, weights, table, extra,
|
|
|
29e444 |
+ indirect);
|
|
|
29e444 |
+ }
|
|
|
29e444 |
+ else
|
|
|
29e444 |
+ {
|
|
|
29e444 |
+ get_next_seq_cached (&seq1, nrules, pass, rulesets, weights);
|
|
|
29e444 |
+ get_next_seq_cached (&seq2, nrules, pass, rulesets, weights);
|
|
|
29e444 |
+ }
|
|
|
29e444 |
|
|
|
29e444 |
/* See whether any or both strings are empty. */
|
|
|
29e444 |
- if (seq1len == 0 || seq2len == 0)
|
|
|
29e444 |
+ if (seq1.len == 0 || seq2.len == 0)
|
|
|
29e444 |
{
|
|
|
29e444 |
- if (seq1len == seq2len)
|
|
|
29e444 |
+ if (seq1.len == seq2.len)
|
|
|
29e444 |
/* Both ended. So far so good, both strings are equal
|
|
|
29e444 |
at this level. */
|
|
|
29e444 |
break;
|
|
|
29e444 |
|
|
|
29e444 |
/* This means one string is shorter than the other. Find out
|
|
|
29e444 |
which one and return an appropriate value. */
|
|
|
29e444 |
- result = seq1len == 0 ? -1 : 1;
|
|
|
29e444 |
+ result = seq1.len == 0 ? -1 : 1;
|
|
|
29e444 |
goto free_and_return;
|
|
|
29e444 |
}
|
|
|
29e444 |
|
|
|
29e444 |
- /* Test for position if necessary. */
|
|
|
29e444 |
- if (position && val1 != val2)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- result = val1 - val2;
|
|
|
29e444 |
- goto free_and_return;
|
|
|
29e444 |
- }
|
|
|
29e444 |
-
|
|
|
29e444 |
- /* Compare the two sequences. */
|
|
|
29e444 |
- do
|
|
|
29e444 |
- {
|
|
|
29e444 |
- if (weights[idx1arr[idx1now]] != weights[idx2arr[idx2now]])
|
|
|
29e444 |
- {
|
|
|
29e444 |
- /* The sequences differ. */
|
|
|
29e444 |
- result = (weights[idx1arr[idx1now]]
|
|
|
29e444 |
- - weights[idx2arr[idx2now]]);
|
|
|
29e444 |
- goto free_and_return;
|
|
|
29e444 |
- }
|
|
|
29e444 |
-
|
|
|
29e444 |
- /* Increment the offsets. */
|
|
|
29e444 |
- ++idx1arr[idx1now];
|
|
|
29e444 |
- ++idx2arr[idx2now];
|
|
|
29e444 |
-
|
|
|
29e444 |
- --seq1len;
|
|
|
29e444 |
- --seq2len;
|
|
|
29e444 |
- }
|
|
|
29e444 |
- while (seq1len > 0 && seq2len > 0);
|
|
|
29e444 |
-
|
|
|
29e444 |
- if (position && seq1len != seq2len)
|
|
|
29e444 |
- {
|
|
|
29e444 |
- result = seq1len - seq2len;
|
|
|
29e444 |
- goto free_and_return;
|
|
|
29e444 |
- }
|
|
|
29e444 |
+ result = do_compare (&seq1, &seq2, position, weights);
|
|
|
29e444 |
+ if (result != 0)
|
|
|
29e444 |
+ goto free_and_return;
|
|
|
29e444 |
}
|
|
|
29e444 |
}
|
|
|
29e444 |
|
|
|
29e444 |
/* Free the memory if needed. */
|
|
|
29e444 |
free_and_return:
|
|
|
29e444 |
if (use_malloc)
|
|
|
29e444 |
- free (idx1arr);
|
|
|
29e444 |
+ free (seq1.idxarr);
|
|
|
29e444 |
|
|
|
29e444 |
return result;
|
|
|
29e444 |
}
|