Changeset: e7382ec341ec for MonetDB
URL: https://dev.monetdb.org/hg/MonetDB/rev/e7382ec341ec
Modified Files:
monetdb5/modules/mal/pcre.c
Branch: ascii-flag
Log Message:
Implemented LIKE _ matching without PCRE.
diffs (truncated from 750 to 300 lines):
diff --git a/monetdb5/modules/mal/pcre.c b/monetdb5/modules/mal/pcre.c
--- a/monetdb5/modules/mal/pcre.c
+++ b/monetdb5/modules/mal/pcre.c
@@ -53,40 +53,15 @@ typedef regex_t pcre;
struct RE {
char *k;
bool search:1, atend:1, case_ignore:1;
- size_t len; /* number of bytes in
string */
- size_t ulen; /* number of codepoints in
string */
+ size_t skip; /* number of codepoints to skip before
matching */
+ size_t len; /* number of bytes in string */
+ size_t ulen; /* number of codepoints in string */
struct RE *n;
};
/* We cannot use strcasecmp and strncasecmp since they work byte for
* byte and don't deal with multibyte encodings (such as UTF-8). */
-/* returns true if the pattern does not contain unescaped `_' (single
- * character match) and ends with unescaped `%' (any sequence
- * match) */
-static inline bool
-re_simple(const char *pat, unsigned char esc)
-{
- bool escaped = false;
-
- if (pat == 0)
- return false;
- if (*pat == '%') {
- pat++;
- }
- while (*pat) {
- if (escaped) {
- escaped = false;
- } else if ((unsigned char) *pat == esc) {
- escaped = true;
- } else if (*pat == '_') {
- return false;
- }
- pat++;
- }
- return true;
-}
-
static inline bool
re_is_pattern_properly_escaped(const char *pat, unsigned char esc)
{
@@ -124,14 +99,77 @@ re_match(const char *restrict s, const s
const struct RE *r;
for (r = pattern; r; r = r->n) {
- if (*r->k == 0 && (r->search || *s == 0))
+ for (size_t i = 0; i < r->skip; s++) {
+ if (*s == 0)
+ return false;
+ i += (*s & 0xC0) != 0x80;
+ }
+ if (r->search) {
+ if (r->atend) {
+ /* we're searching for a string at the end, so
just skip
+ * over everything and just compare with the
tail of the
+ * haystack */
+ size_t slen = strlen(s);
+ if (slen < r->ulen) {
+ /* remaining string too short: each
codepoint
+ * requires at least one byte */
+ return false;
+ }
+ const char *e = s + slen;
+ if (!r->case_ignore) {
+ if (slen < r->len) {
+ /* remaining string is too
short to match */
+ return false;
+ }
+ e -= r->len;
+ if ((*e & 0xC0) == 0x80) {
+ /* not at start of a Unicode
character, so
+ * cannot match (this test not
strictly
+ * required: the strcmp should
also return
+ * unequal) */
+ return false;
+ }
+ return strcmp(e, r->k) == 0;
+ }
+ size_t ulen = r->ulen;
+ while (e > s && ulen != 0) {
+ ulen -= (*--e & 0xC0) != 0x80;
+ }
+ /* ulen != 0 means remaining string is too
short */
+ return ulen == 0 && GDKstrcasecmp(e, r->k) == 0;
+ }
+ /* in case we have a pattern consisting of % followed
by _,
+ * we need to backtrack, so use recursion; here we know
we
+ * have the %, look for an _ in the rest of the pattern
+ * (note %_ and _% are equivalent and is taken care of
by
+ * the pattern construction in re_create) */
+ for (const struct RE *p = r->n; p; p = p->n) {
+ if (p->skip != 0) {
+ struct RE pat = *r;
+ pat.search = false;
+ pat.skip = 0;
+ do {
+ if (re_match(s, &pat))
+ return true;
+ do
+ s++;
+ while (*s && (*s & 0xC0) ==
0x80);
+ } while (*s != 0);
+ return false;
+ }
+ }
+ }
+ if (r->k[0] == 0 && (r->search || *s == 0))
return true;
if (r->case_ignore) {
for (;;) {
+ if (r->search && (s = GDKstrcasestr(s, r->k))
== NULL)
+ return false;
if (*s == '\0')
return false;
/* in "atend" comparison, include NUL byte in
the compare */
- if (GDKstrncasecmp(s, r->k, SIZE_MAX, r->len +
r->atend) != 0) {
+ if ((!r->search || r->atend) &&
+ GDKstrncasecmp(s, r->k, SIZE_MAX,
r->len + r->atend) != 0) {
/* no match */
if (!r->search)
return false;
@@ -147,14 +185,14 @@ re_match(const char *restrict s, const s
break;
}
} else {
- /* search for first byte in pattern */
- if (r->search && (s = strchr(s, r->k[0])) == NULL)
- return false;
for (;;) {
+ if (r->search && (s = strstr(s, r->k)) == NULL)
+ return false;
if (*s == '\0')
return false;
/* in "atend" comparison, include NUL byte in
the compare */
- if (strncmp(s, r->k, r->len + r->atend) != 0) {
+ if ((!r->search || r->atend) &&
+ strncmp(s, r->k, r->len + r->atend) !=
0) {
/* no match */
if (!r->search)
return false;
@@ -204,18 +242,26 @@ re_create(const char *pat, bool caseigno
if (r == NULL)
return NULL;
- *r = (struct RE) {.atend = true };
+ *r = (struct RE) {
+ .atend = true,
+ .case_ignore = caseignore,
+ };
- while (esc != '%' && *pat == '%') {
- pat++; /* skip % */
- r->search = true;
+ for (;;) {
+ if (esc != '%' && *pat == '%') {
+ pat++; /* skip % */
+ r->search = true;
+ } else if (esc != '_' && *pat == '_') {
+ pat++;
+ r->skip++;
+ } else {
+ break;
+ }
}
if ((p = GDKstrdup(pat)) == NULL) {
GDKfree(r);
return NULL;
}
- if (caseignore)
- n->case_ignore = true;
r->k = p;
q = p;
@@ -227,25 +273,34 @@ re_create(const char *pat, bool caseigno
escaped = false;
} else if ((unsigned char) *p == esc) {
escaped = true;
- } else if (*p == '%') {
+ } else if (*p == '%' || *p == '_') {
n->atend = false;
- while (p[1] == '%')
+ bool search = false;
+ size_t skip = 0;
+ for (;;) {
+ if (*p == '_')
+ skip++;
+ else if (*p == '%')
+ search = true;
+ else
+ break;
p++;
- if (p[1]) {
+ }
+ if (*p || skip != 0) {
n = n->n = GDKmalloc(sizeof(struct RE));
if (n == NULL)
goto bailout;
*n = (struct RE) {
- .search = true,
+ .search = search,
.atend = true,
- .k = p + 1
+ .skip = skip,
+ .k = p,
+ .case_ignore = caseignore,
};
- if (caseignore) {
- n->case_ignore = true;
- }
}
*q = 0;
- q = p + 1;
+ q = p;
+ continue; /* skip increment, we
already did it */
} else {
*q++ = *p;
n->len++;
@@ -765,11 +820,11 @@ pcre_match_with_flags(bit *ret, const ch
#ifdef HAVE_LIBPCRE
/* special characters in PCRE that need to be escaped */
-static const char *pcre_specials = ".+?*()[]{}|^$\\";
+static const char pcre_specials[] = "$()*+.?[\\]^{|}";
#else
/* special characters in POSIX basic regular expressions that need to
* be escaped */
-static const char *pcre_specials = "^.[$()|*+?{\\";
+static const char pcre_specials[] = "$()*+.?[\\^{|";
#endif
/* change SQL LIKE pattern into PCRE pattern */
@@ -1055,7 +1110,7 @@ PCREsql2pcre(str *ret, const str *pat, c
}
static inline str
-choose_like_path(char **ppat, bool *use_re, bool *use_strcmp, bool *empty,
+choose_like_path(bool *use_re, bool *use_strcmp, bool *empty,
const char *pat, const char *esc)
{
str res = MAL_SUCCEED;
@@ -1074,17 +1129,8 @@ choose_like_path(char **ppat, bool *use_
if (is_strcmpable(pat, esc)) {
*use_re = true;
*use_strcmp = true;
- } else if (re_simple(pat, (unsigned char) *esc)) {
+ } else {
*use_re = true;
- } else {
- if ((res = sql2pcre(ppat, pat, esc)) != MAL_SUCCEED)
- return res;
- if (strNil(*ppat)) {
- GDKfree(*ppat);
- *ppat = NULL;
- *use_re = true;
- *use_strcmp = true;
- }
}
}
return res;
@@ -1095,11 +1141,10 @@ PCRElike_imp(bit *ret, const str *s, con
const bit *isens)
{
str res = MAL_SUCCEED;
- char *ppat = NULL;
bool use_re = false, use_strcmp = false, empty = false;
struct RE *re = NULL;
- if ((res = choose_like_path(&ppat, &use_re, &use_strcmp, &empty,
+ if ((res = choose_like_path(&use_re, &use_strcmp, &empty,
*pat, *esc)) !=
MAL_SUCCEED)
return res;
@@ -1110,7 +1155,7 @@ PCRElike_imp(bit *ret, const str *s, con
if (strNil(*s) || empty) {
*ret = bit_nil;
- } else if (use_re) {
+ } else {
if (use_strcmp) {
*ret = *isens ? GDKstrcasecmp(*s, *pat) == 0
: strcmp(*s, *pat) == 0;
@@ -1121,13 +1166,10 @@ PCRElike_imp(bit *ret, const str *s, con
else
*ret = re_match(*s, re);
}
- } else {
- res = *isens ? PCREimatch(ret, s, &ppat) : PCREmatch(ret, s,
&ppat);
}
if (re)
re_destroy(re);
- GDKfree(ppat);
return res;
}
_______________________________________________
checkin-list mailing list -- checkin-list@monetdb.org
To unsubscribe send an email to checkin-list-le...@monetdb.org
