/*
* altdissect - determine alternative subexpression matches (uncomplicated)
*/
static int /* regexec return code */
altdissect(struct vars * v,
struct subre * t,
chr *begin, /* beginning of relevant substring */
chr *end) /* end of same */
{
struct dfa *d;
int i;
assert(t != NULL);
assert(t->op == '|');
for (i = 0; t != NULL; t = t->right, i++)
{
MDEBUG(("trying %dth\n", i));
assert(t->left != NULL && t->left->cnfa.nstates > 0);
d = newdfa(v, &t->left->cnfa, &v->g->cmap, &v->dfa1);
if (ISERR())
return v->err;
if (longest(v, d, begin, end, (int *) NULL) == end)
{
MDEBUG(("success\n"));
freedfa(d);
return dissect(v, t->left, begin, end);
}
freedfa(d);
}
return REG_ASSERT; /* none of them matched?!? */
}
/*
- lacon - lookahead-constraint checker for miss()
^ static int lacon(struct vars *, struct cnfa *, chr *, pcolor);
*/
static int /* predicate: constraint satisfied? */
lacon(
struct vars *v,
struct cnfa *pcnfa, /* parent cnfa */
chr *cp,
pcolor co) /* "color" of the lookahead constraint */
{
int n;
struct subre *sub;
struct dfa *d;
struct smalldfa sd;
chr *end;
n = co - pcnfa->ncolors;
assert(n < v->g->nlacons && v->g->lacons != NULL);
FDEBUG(("=== testing lacon %d\n", n));
sub = &v->g->lacons[n];
d = newdfa(v, &sub->cnfa, &v->g->cmap, &sd);
if (d == NULL) {
ERR(REG_ESPACE);
return 0;
}
end = longest(v, d, cp, v->stop, (int *)NULL);
freedfa(d);
FDEBUG(("=== lacon %d match %d\n", n, (end != NULL)));
return (sub->subno) ? (end != NULL) : (end == NULL);
}
/*
* getsubdfa - create or re-fetch the DFA for a subre node
*
* We only need to create the DFA once per overall regex execution.
* The DFA will be freed by the cleanup step in pg_regexec().
*/
static struct dfa *
getsubdfa(struct vars * v,
struct subre * t)
{
if (v->subdfas[t->id] == NULL)
{
v->subdfas[t->id] = newdfa(v, &t->cnfa, &v->g->cmap, DOMALLOC);
if (ISERR())
return NULL;
}
return v->subdfas[t->id];
}
/*
* cfind - find a match for the main NFA (with complications)
*/
static int
cfind(struct vars * v,
struct cnfa * cnfa,
struct colormap * cm)
{
struct dfa *s;
struct dfa *d;
chr *cold;
int ret;
s = newdfa(v, &v->g->search, cm, &v->dfa1);
NOERR();
d = newdfa(v, cnfa, cm, &v->dfa2);
if (ISERR())
{
assert(d == NULL);
freedfa(s);
return v->err;
}
ret = cfindloop(v, cnfa, cm, d, s, &cold);
freedfa(d);
freedfa(s);
NOERR();
if (v->g->cflags & REG_EXPECT)
{
assert(v->details != NULL);
if (cold != NULL)
v->details->rm_extend.rm_so = OFF(cold);
else
v->details->rm_extend.rm_so = OFF(v->stop);
v->details->rm_extend.rm_eo = OFF(v->stop); /* unknown */
}
return ret;
}
/*
- caltdissect - determine alternative subexpression matches (w. complications)
^ static int caltdissect(struct vars *, struct subre *, chr *, chr *);
*/
static int /* regexec return code */
caltdissect(
struct vars *v,
struct subre *t,
chr *begin, /* beginning of relevant substring */
chr *end) /* end of same */
{
struct dfa *d;
int er;
#define UNTRIED 0 /* not yet tried at all */
#define TRYING 1 /* top matched, trying submatches */
#define TRIED 2 /* top didn't match or submatches exhausted */
if (t == NULL) {
return REG_NOMATCH;
}
assert(t->op == '|');
if (v->mem[t->retry] == TRIED) {
return caltdissect(v, t->right, begin, end);
}
MDEBUG(("calt n%d\n", t->retry));
assert(t->left != NULL);
if (v->mem[t->retry] == UNTRIED) {
d = newdfa(v, &t->left->cnfa, &v->g->cmap, DOMALLOC);
if (ISERR()) {
return v->err;
}
if (longest(v, d, begin, end, NULL) != end) {
freedfa(d);
v->mem[t->retry] = TRIED;
return caltdissect(v, t->right, begin, end);
}
freedfa(d);
MDEBUG(("calt matched\n"));
v->mem[t->retry] = TRYING;
}
er = cdissect(v, t->left, begin, end);
if (er != REG_NOMATCH) {
return er;
}
v->mem[t->retry] = TRIED;
return caltdissect(v, t->right, begin, end);
}
/*
* crevdissect - determine backref shortest-first subexpression matches
* The retry memory stores the offset of the trial midpoint from begin,
* plus 1 so that 0 uniquely means "clean slate".
*/
static int /* regexec return code */
crevdissect(struct vars * v,
struct subre * t,
chr *begin, /* beginning of relevant substring */
chr *end) /* end of same */
{
struct dfa *d;
struct dfa *d2;
chr *mid;
int er;
assert(t->op == '.');
assert(t->left != NULL && t->left->cnfa.nstates > 0);
assert(t->right != NULL && t->right->cnfa.nstates > 0);
assert(t->left->flags & SHORTER);
/* concatenation -- need to split the substring between parts */
d = newdfa(v, &t->left->cnfa, &v->g->cmap, DOMALLOC);
if (ISERR())
return v->err;
d2 = newdfa(v, &t->right->cnfa, &v->g->cmap, DOMALLOC);
if (ISERR())
{
freedfa(d);
return v->err;
}
MDEBUG(("crev %d\n", t->retry));
/* pick a tentative midpoint */
if (v->mem[t->retry] == 0)
{
mid = shortest(v, d, begin, begin, end, (chr **) NULL, (int *) NULL);
if (mid == NULL)
{
freedfa(d);
freedfa(d2);
return REG_NOMATCH;
}
MDEBUG(("tentative midpoint %ld\n", LOFF(mid)));
v->mem[t->retry] = (mid - begin) + 1;
}
else
{
mid = begin + (v->mem[t->retry] - 1);
MDEBUG(("working midpoint %ld\n", LOFF(mid)));
}
/* iterate until satisfaction or failure */
for (;;)
{
/* try this midpoint on for size */
er = cdissect(v, t->left, begin, mid);
if (er == REG_OKAY &&
longest(v, d2, mid, end, (int *) NULL) == end &&
(er = cdissect(v, t->right, mid, end)) ==
REG_OKAY)
break; /* NOTE BREAK OUT */
if (er != REG_OKAY && er != REG_NOMATCH)
{
freedfa(d);
freedfa(d2);
return er;
}
/* that midpoint didn't work, find a new one */
if (mid == end)
{
/* all possibilities exhausted */
MDEBUG(("%d no midpoint\n", t->retry));
freedfa(d);
freedfa(d2);
return REG_NOMATCH;
}
mid = shortest(v, d, begin, mid + 1, end, (chr **) NULL, (int *) NULL);
if (mid == NULL)
{
/* failed to find a new one */
MDEBUG(("%d failed midpoint\n", t->retry));
freedfa(d);
freedfa(d2);
return REG_NOMATCH;
}
MDEBUG(("%d: new midpoint %ld\n", t->retry, LOFF(mid)));
v->mem[t->retry] = (mid - begin) + 1;
zapmem(v, t->left);
zapmem(v, t->right);
}
/* satisfaction */
MDEBUG(("successful\n"));
freedfa(d);
freedfa(d2);
return REG_OKAY;
}
/*
* condissect - determine concatenation subexpression matches (uncomplicated)
*/
static int /* regexec return code */
condissect(struct vars * v,
struct subre * t,
chr *begin, /* beginning of relevant substring */
chr *end) /* end of same */
{
struct dfa *d;
struct dfa *d2;
chr *mid;
int i;
int shorter = (t->left->flags & SHORTER) ? 1 : 0;
chr *stop = (shorter) ? end : begin;
assert(t->op == '.');
assert(t->left != NULL && t->left->cnfa.nstates > 0);
assert(t->right != NULL && t->right->cnfa.nstates > 0);
d = newdfa(v, &t->left->cnfa, &v->g->cmap, &v->dfa1);
NOERR();
d2 = newdfa(v, &t->right->cnfa, &v->g->cmap, &v->dfa2);
if (ISERR())
{
assert(d2 == NULL);
freedfa(d);
return v->err;
}
/* pick a tentative midpoint */
if (shorter)
mid = shortest(v, d, begin, begin, end, (chr **) NULL,
(int *) NULL);
else
mid = longest(v, d, begin, end, (int *) NULL);
if (mid == NULL)
{
freedfa(d);
freedfa(d2);
return REG_ASSERT;
}
MDEBUG(("tentative midpoint %ld\n", LOFF(mid)));
/* iterate until satisfaction or failure */
while (longest(v, d2, mid, end, (int *) NULL) != end)
{
/* that midpoint didn't work, find a new one */
if (mid == stop)
{
/* all possibilities exhausted! */
MDEBUG(("no midpoint!\n"));
freedfa(d);
freedfa(d2);
return REG_ASSERT;
}
if (shorter)
mid = shortest(v, d, begin, mid + 1, end, (chr **) NULL,
(int *) NULL);
else
mid = longest(v, d, begin, mid - 1, (int *) NULL);
if (mid == NULL)
{
/* failed to find a new one! */
MDEBUG(("failed midpoint!\n"));
freedfa(d);
freedfa(d2);
return REG_ASSERT;
}
MDEBUG(("new midpoint %ld\n", LOFF(mid)));
}
/* satisfaction */
MDEBUG(("successful\n"));
freedfa(d);
freedfa(d2);
i = dissect(v, t->left, begin, mid);
if (i != REG_OKAY)
return i;
return dissect(v, t->right, mid, end);
}
/*
* find - find a match for the main NFA (no-complications case)
*/
static int
find(struct vars * v,
struct cnfa * cnfa,
struct colormap * cm)
{
struct dfa *s;
struct dfa *d;
chr *begin;
chr *end = NULL;
chr *cold;
chr *open; /* open and close of range of possible starts */
chr *close;
int hitend;
int shorter = (v->g->tree->flags & SHORTER) ? 1 : 0;
/* first, a shot with the search RE */
s = newdfa(v, &v->g->search, cm, &v->dfa1);
assert(!(ISERR() && s != NULL));
NOERR();
MDEBUG(("\nsearch at %ld\n", LOFF(v->start)));
cold = NULL;
close = shortest(v, s, v->search_start, v->search_start, v->stop,
&cold, (int *) NULL);
freedfa(s);
NOERR();
if (v->g->cflags & REG_EXPECT)
{
assert(v->details != NULL);
if (cold != NULL)
v->details->rm_extend.rm_so = OFF(cold);
else
v->details->rm_extend.rm_so = OFF(v->stop);
v->details->rm_extend.rm_eo = OFF(v->stop); /* unknown */
}
if (close == NULL) /* not found */
return REG_NOMATCH;
if (v->nmatch == 0) /* found, don't need exact location */
return REG_OKAY;
/* find starting point and match */
assert(cold != NULL);
open = cold;
cold = NULL;
MDEBUG(("between %ld and %ld\n", LOFF(open), LOFF(close)));
d = newdfa(v, cnfa, cm, &v->dfa1);
assert(!(ISERR() && d != NULL));
NOERR();
for (begin = open; begin <= close; begin++)
{
MDEBUG(("\nfind trying at %ld\n", LOFF(begin)));
if (shorter)
end = shortest(v, d, begin, begin, v->stop,
(chr **) NULL, &hitend);
else
end = longest(v, d, begin, v->stop, &hitend);
NOERR();
if (hitend && cold == NULL)
cold = begin;
if (end != NULL)
break; /* NOTE BREAK OUT */
}
assert(end != NULL); /* search RE succeeded so loop should */
freedfa(d);
/* and pin down details */
assert(v->nmatch > 0);
v->pmatch[0].rm_so = OFF(begin);
v->pmatch[0].rm_eo = OFF(end);
if (v->g->cflags & REG_EXPECT)
{
if (cold != NULL)
v->details->rm_extend.rm_so = OFF(cold);
else
v->details->rm_extend.rm_so = OFF(v->stop);
v->details->rm_extend.rm_eo = OFF(v->stop); /* unknown */
}
if (v->nmatch == 1) /* no need for submatches */
return REG_OKAY;
/* submatches */
zapsubs(v->pmatch, v->nmatch);
return dissect(v, v->g->tree, begin, end);
}
/*
- ccondissect - concatenation subexpression matches (with complications)
* The retry memory stores the offset of the trial midpoint from begin, plus 1
* so that 0 uniquely means "clean slate".
^ static int ccondissect(struct vars *, struct subre *, chr *, chr *);
*/
static int /* regexec return code */
ccondissect(
struct vars *v,
struct subre *t,
chr *begin, /* beginning of relevant substring */
chr *end) /* end of same */
{
struct dfa *d;
struct dfa *d2;
chr *mid;
int er;
assert(t->op == '.');
assert(t->left != NULL && t->left->cnfa.nstates > 0);
assert(t->right != NULL && t->right->cnfa.nstates > 0);
if (t->left->flags&SHORTER) { /* reverse scan */
return crevdissect(v, t, begin, end);
}
d = newdfa(v, &t->left->cnfa, &v->g->cmap, DOMALLOC);
if (ISERR()) {
return v->err;
}
d2 = newdfa(v, &t->right->cnfa, &v->g->cmap, DOMALLOC);
if (ISERR()) {
freedfa(d);
return v->err;
}
MDEBUG(("cconcat %d\n", t->retry));
/*
* Pick a tentative midpoint.
*/
if (v->mem[t->retry] == 0) {
mid = longest(v, d, begin, end, NULL);
if (mid == NULL) {
freedfa(d);
freedfa(d2);
return REG_NOMATCH;
}
MDEBUG(("tentative midpoint %ld\n", LOFF(mid)));
v->mem[t->retry] = (mid - begin) + 1;
} else {
mid = begin + (v->mem[t->retry] - 1);
MDEBUG(("working midpoint %ld\n", LOFF(mid)));
}
/*
* Iterate until satisfaction or failure.
*/
for (;;) {
/*
* Try this midpoint on for size.
*/
er = cdissect(v, t->left, begin, mid);
if ((er == REG_OKAY) && (longest(v, d2, mid, end, NULL) == end)
&& (er = cdissect(v, t->right, mid, end)) == REG_OKAY) {
break; /* NOTE BREAK OUT */
}
if ((er != REG_OKAY) && (er != REG_NOMATCH)) {
freedfa(d);
freedfa(d2);
return er;
}
/*
* That midpoint didn't work, find a new one.
*/
if (mid == begin) {
/*
* All possibilities exhausted.
*/
MDEBUG(("%d no midpoint\n", t->retry));
freedfa(d);
freedfa(d2);
return REG_NOMATCH;
}
mid = longest(v, d, begin, mid-1, NULL);
if (mid == NULL) {
/*
* Failed to find a new one.
*/
MDEBUG(("%d failed midpoint\n", t->retry));
freedfa(d);
freedfa(d2);
return REG_NOMATCH;
}
MDEBUG(("%d: new midpoint %ld\n", t->retry, LOFF(mid)));
v->mem[t->retry] = (mid - begin) + 1;
zapmem(v, t->left);
zapmem(v, t->right);
}
/*
* Satisfaction.
*/
MDEBUG(("successful\n"));
freedfa(d);
freedfa(d2);
return REG_OKAY;
}
