* with the POSIX routines of the same names.
*/
+#ifdef IN_XSUB_RE
+/* We *really* need to overwrite these symbols: */
+# define Perl_pregcomp my_regcomp
+# define Perl_regdump my_regdump
+# define Perl_regprop my_regprop
+/* *These* symbols are masked to allow static link. */
+# define Perl_pregfree my_regfree
+# define Perl_regnext my_regnext
+#endif
+
/*SUPPRESS 112*/
/*
* pregcomp and pregexec -- regsub and regerror are not used in perl
*/
#include "EXTERN.h"
#include "perl.h"
-#include "INTERN.h"
+
+#ifndef IN_XSUB_RE
+# include "INTERN.h"
+#endif
+
+#define REG_COMP_C
#include "regcomp.h"
+#ifdef op
+#undef op
+#endif /* op */
+
#ifdef MSDOS
# if defined(BUGGY_MSC6)
/* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
* Flags to be passed up and down.
*/
#define WORST 0 /* Worst case. */
-#define HASWIDTH 0x1 /* Known never to match null string. */
+#define HASWIDTH 0x1 /* Known to match non-null strings. */
#define SIMPLE 0x2 /* Simple enough to be STAR/PLUS operand. */
#define SPSTART 0x4 /* Starts with * or +. */
#define TRYAGAIN 0x8 /* Weeded out a declaration. */
* Forward declarations for pregcomp()'s friends.
*/
-static char *reg _((I32, I32 *));
-static char *reganode _((char, unsigned short));
-static char *regatom _((I32 *));
-static char *regbranch _((I32 *));
-static void regc _((char));
-static char *regclass _((void));
+#ifndef PERL_OBJECT
+static regnode *reg _((I32, I32 *));
+static regnode *reganode _((U8, U32));
+static regnode *regatom _((I32 *));
+static regnode *regbranch _((I32 *, I32));
+static void regc _((U8, char *));
+static regnode *regclass _((void));
STATIC I32 regcurly _((char *));
-static char *regnode _((char));
-static char *regpiece _((I32 *));
-static void reginsert _((char, char *));
-static void regoptail _((char *, char *));
-static void regset _((char *, I32));
-static void regtail _((char *, char *));
+static regnode *reg_node _((U8));
+static regnode *regpiece _((I32 *));
+static void reginsert _((U8, regnode *));
+static void regoptail _((regnode *, regnode *));
+static void regtail _((regnode *, regnode *));
static char* regwhite _((char *, char *));
static char* nextchar _((void));
+static void re_croak2 _((const char* pat1,const char* pat2,...)) __attribute__((noreturn));
+#endif
+
+/* Length of a variant. */
+
+#ifndef PERL_OBJECT
+typedef struct {
+ I32 len_min;
+ I32 len_delta;
+ I32 pos_min;
+ I32 pos_delta;
+ SV *last_found;
+ I32 last_end; /* min value, <0 unless valid. */
+ I32 last_start_min;
+ I32 last_start_max;
+ SV **longest; /* Either &l_fixed, or &l_float. */
+ SV *longest_fixed;
+ I32 offset_fixed;
+ SV *longest_float;
+ I32 offset_float_min;
+ I32 offset_float_max;
+ I32 flags;
+} scan_data_t;
+#endif
+
+static scan_data_t zero_scan_data = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+
+#define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
+#define SF_BEFORE_SEOL 0x1
+#define SF_BEFORE_MEOL 0x2
+#define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
+#define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
+
+#ifdef NO_UNARY_PLUS
+# define SF_FIX_SHIFT_EOL (0+2)
+# define SF_FL_SHIFT_EOL (0+4)
+#else
+# define SF_FIX_SHIFT_EOL (+2)
+# define SF_FL_SHIFT_EOL (+4)
+#endif
+
+#define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
+#define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
+
+#define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
+#define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
+#define SF_IS_INF 0x40
+#define SF_HAS_PAR 0x80
+#define SF_IN_PAR 0x100
+#define SF_HAS_EVAL 0x200
+#define SCF_DO_SUBSTR 0x400
+
+STATIC void
+scan_commit(scan_data_t *data)
+{
+ STRLEN l = SvCUR(data->last_found);
+ STRLEN old_l = SvCUR(*data->longest);
+
+ if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
+ sv_setsv(*data->longest, data->last_found);
+ if (*data->longest == data->longest_fixed) {
+ data->offset_fixed = l ? data->last_start_min : data->pos_min;
+ if (data->flags & SF_BEFORE_EOL)
+ data->flags
+ |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
+ else
+ data->flags &= ~SF_FIX_BEFORE_EOL;
+ } else {
+ data->offset_float_min = l ? data->last_start_min : data->pos_min;
+ data->offset_float_max = (l
+ ? data->last_start_max
+ : data->pos_min + data->pos_delta);
+ if (data->flags & SF_BEFORE_EOL)
+ data->flags
+ |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
+ else
+ data->flags &= ~SF_FL_BEFORE_EOL;
+ }
+ }
+ SvCUR_set(data->last_found, 0);
+ data->last_end = -1;
+ data->flags &= ~SF_BEFORE_EOL;
+}
+
+/* Stops at toplevel WHILEM as well as at `last'. At end *scanp is set
+ to the position after last scanned or to NULL. */
+
+STATIC I32
+study_chunk(regnode **scanp, I32 *deltap, regnode *last, scan_data_t *data, U32 flags)
+ /* scanp: Start here (read-write). */
+ /* deltap: Write maxlen-minlen here. */
+ /* last: Stop before this one. */
+{
+ dTHR;
+ I32 min = 0, pars = 0, code;
+ regnode *scan = *scanp, *next;
+ I32 delta = 0;
+ int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
+ I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
+ scan_data_t data_fake;
+
+ while (scan && OP(scan) != END && scan < last) {
+ /* Peephole optimizer: */
+
+ if (regkind[(U8)OP(scan)] == EXACT) {
+ regnode *n = regnext(scan);
+ U32 stringok = 1;
+#ifdef DEBUGGING
+ regnode *stop = scan;
+#endif
+
+ next = scan + (*OPERAND(scan) + 2 - 1)/sizeof(regnode) + 2;
+ /* Skip NOTHING, merge EXACT*. */
+ while (n &&
+ ( regkind[(U8)OP(n)] == NOTHING ||
+ (stringok && (OP(n) == OP(scan))))
+ && NEXT_OFF(n)
+ && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
+ if (OP(n) == TAIL || n > next)
+ stringok = 0;
+ if (regkind[(U8)OP(n)] == NOTHING) {
+ NEXT_OFF(scan) += NEXT_OFF(n);
+ next = n + NODE_STEP_REGNODE;
+#ifdef DEBUGGING
+ if (stringok)
+ stop = n;
+#endif
+ n = regnext(n);
+ } else {
+ int oldl = *OPERAND(scan);
+ regnode *nnext = regnext(n);
+
+ if (oldl + *OPERAND(n) > U8_MAX)
+ break;
+ NEXT_OFF(scan) += NEXT_OFF(n);
+ *OPERAND(scan) += *OPERAND(n);
+ next = n + (*OPERAND(n) + 2 - 1)/sizeof(regnode) + 2;
+ /* Now we can overwrite *n : */
+ Move(OPERAND(n) + 1, OPERAND(scan) + oldl + 1,
+ *OPERAND(n) + 1, char);
+#ifdef DEBUGGING
+ if (stringok)
+ stop = next - 1;
+#endif
+ n = nnext;
+ }
+ }
+#ifdef DEBUGGING
+ /* Allow dumping */
+ n = scan + (*OPERAND(scan) + 2 - 1)/sizeof(regnode) + 2;
+ while (n <= stop) {
+ /* Purify reports a benign UMR here sometimes, because we
+ * don't initialize the OP() slot of a node when that node
+ * is occupied by just the trailing null of the string in
+ * an EXACT node */
+ if (regkind[(U8)OP(n)] != NOTHING || OP(n) == NOTHING) {
+ OP(n) = OPTIMIZED;
+ NEXT_OFF(n) = 0;
+ }
+ n++;
+ }
+#endif
+
+ }
+ if (OP(scan) != CURLYX) {
+ int max = (reg_off_by_arg[OP(scan)] ? I32_MAX : U16_MAX);
+ int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
+ int noff;
+ regnode *n = scan;
+
+ /* Skip NOTHING and LONGJMP. */
+ while ((n = regnext(n))
+ && ((regkind[(U8)OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
+ || ((OP(n) == LONGJMP) && (noff = ARG(n))))
+ && off + noff < max)
+ off += noff;
+ if (reg_off_by_arg[OP(scan)])
+ ARG(scan) = off;
+ else
+ NEXT_OFF(scan) = off;
+ }
+ if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
+ || OP(scan) == IFTHEN || OP(scan) == SUSPEND) {
+ next = regnext(scan);
+ code = OP(scan);
+
+ if (OP(next) == code || code == IFTHEN || code == SUSPEND) {
+ I32 max1 = 0, min1 = I32_MAX, num = 0;
+
+ if (flags & SCF_DO_SUBSTR)
+ scan_commit(data);
+ while (OP(scan) == code) {
+ I32 deltanext, minnext;
+
+ num++;
+ data_fake.flags = 0;
+ next = regnext(scan);
+ scan = NEXTOPER(scan);
+ if (code != BRANCH)
+ scan = NEXTOPER(scan);
+ /* We suppose the run is continuous, last=next...*/
+ minnext = study_chunk(&scan, &deltanext, next,
+ &data_fake, 0);
+ if (min1 > minnext)
+ min1 = minnext;
+ if (max1 < minnext + deltanext)
+ max1 = minnext + deltanext;
+ if (deltanext == I32_MAX)
+ is_inf = 1;
+ scan = next;
+ if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
+ pars++;
+ if (data && (data_fake.flags & SF_HAS_EVAL))
+ data->flags |= SF_HAS_EVAL;
+ if (code == SUSPEND)
+ break;
+ }
+ if (code == IFTHEN && num < 2) /* Empty ELSE branch */
+ min1 = 0;
+ if (flags & SCF_DO_SUBSTR) {
+ data->pos_min += min1;
+ data->pos_delta += max1 - min1;
+ if (max1 != min1 || is_inf)
+ data->longest = &(data->longest_float);
+ }
+ min += min1;
+ delta += max1 - min1;
+ } else if (code == BRANCHJ) /* single branch is optimized. */
+ scan = NEXTOPER(NEXTOPER(scan));
+ else /* single branch is optimized. */
+ scan = NEXTOPER(scan);
+ continue;
+ } else if (OP(scan) == EXACT) {
+ min += *OPERAND(scan);
+ if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
+ I32 l = *OPERAND(scan);
+
+ /* The code below prefers earlier match for fixed
+ offset, later match for variable offset. */
+ if (data->last_end == -1) { /* Update the start info. */
+ data->last_start_min = data->pos_min;
+ data->last_start_max = is_inf
+ ? I32_MAX : data->pos_min + data->pos_delta;
+ }
+ sv_catpvn(data->last_found, (char *)(OPERAND(scan)+1), l);
+ data->last_end = data->pos_min + l;
+ data->pos_min += l; /* As in the first entry. */
+ data->flags &= ~SF_BEFORE_EOL;
+ }
+ } else if (regkind[(U8)OP(scan)] == EXACT) {
+ if (flags & SCF_DO_SUBSTR)
+ scan_commit(data);
+ min += *OPERAND(scan);
+ if (data && (flags & SCF_DO_SUBSTR))
+ data->pos_min += *OPERAND(scan);
+ } else if (strchr(varies,OP(scan))) {
+ I32 mincount, maxcount, minnext, deltanext, pos_before, fl;
+ regnode *oscan = scan;
+
+ switch (regkind[(U8)OP(scan)]) {
+ case WHILEM:
+ scan = NEXTOPER(scan);
+ goto finish;
+ case PLUS:
+ if (flags & SCF_DO_SUBSTR) {
+ next = NEXTOPER(scan);
+ if (OP(next) == EXACT) {
+ mincount = 1;
+ maxcount = REG_INFTY;
+ next = regnext(scan);
+ scan = NEXTOPER(scan);
+ goto do_curly;
+ }
+ }
+ if (flags & SCF_DO_SUBSTR)
+ data->pos_min++;
+ min++;
+ /* Fall through. */
+ case STAR:
+ is_inf = 1;
+ scan = regnext(scan);
+ if (flags & SCF_DO_SUBSTR) {
+ scan_commit(data);
+ data->longest = &(data->longest_float);
+ }
+ goto optimize_curly_tail;
+ case CURLY:
+ mincount = ARG1(scan);
+ maxcount = ARG2(scan);
+ next = regnext(scan);
+ scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
+ do_curly:
+ if (flags & SCF_DO_SUBSTR) {
+ if (mincount == 0) scan_commit(data);
+ pos_before = data->pos_min;
+ }
+ if (data) {
+ fl = data->flags;
+ data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
+ if (is_inf)
+ data->flags |= SF_IS_INF;
+ }
+ /* This will finish on WHILEM, setting scan, or on NULL: */
+ minnext = study_chunk(&scan, &deltanext, last, data,
+ mincount == 0
+ ? (flags & ~SCF_DO_SUBSTR) : flags);
+ if (!scan) /* It was not CURLYX, but CURLY. */
+ scan = next;
+ if (dowarn && (minnext + deltanext == 0)
+ && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
+ && maxcount <= 10000) /* Complement check for big count */
+ warn("Strange *+?{} on zero-length expression");
+ min += minnext * mincount;
+ is_inf |= (maxcount == REG_INFTY && (minnext + deltanext) > 0
+ || deltanext == I32_MAX);
+ delta += (minnext + deltanext) * maxcount - minnext * mincount;
+
+ /* Try powerful optimization CURLYX => CURLYN. */
+ if ( OP(oscan) == CURLYX && data
+ && data->flags & SF_IN_PAR
+ && !(data->flags & SF_HAS_EVAL)
+ && !deltanext && minnext == 1 ) {
+ /* Try to optimize to CURLYN. */
+ regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
+ regnode *nxt1 = nxt, *nxt2;
+
+ /* Skip open. */
+ nxt = regnext(nxt);
+ if (!strchr(simple,OP(nxt))
+ && !(regkind[(U8)OP(nxt)] == EXACT
+ && *OPERAND(nxt) == 1))
+ goto nogo;
+ nxt2 = nxt;
+ nxt = regnext(nxt);
+ if (OP(nxt) != CLOSE)
+ goto nogo;
+ /* Now we know that nxt2 is the only contents: */
+ oscan->flags = ARG(nxt);
+ OP(oscan) = CURLYN;
+ OP(nxt1) = NOTHING; /* was OPEN. */
+#ifdef DEBUGGING
+ OP(nxt1 + 1) = OPTIMIZED; /* was count. */
+ NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
+ NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
+ OP(nxt) = OPTIMIZED; /* was CLOSE. */
+ OP(nxt + 1) = OPTIMIZED; /* was count. */
+ NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
+#endif
+ }
+ nogo:
+
+ /* Try optimization CURLYX => CURLYM. */
+ if ( OP(oscan) == CURLYX && data
+ && !(data->flags & SF_HAS_PAR)
+ && !(data->flags & SF_HAS_EVAL)
+ && !deltanext ) {
+ /* XXXX How to optimize if data == 0? */
+ /* Optimize to a simpler form. */
+ regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
+ regnode *nxt2;
+
+ OP(oscan) = CURLYM;
+ while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
+ && (OP(nxt2) != WHILEM))
+ nxt = nxt2;
+ OP(nxt2) = SUCCEED; /* Whas WHILEM */
+ /* Need to optimize away parenths. */
+ if (data->flags & SF_IN_PAR) {
+ /* Set the parenth number. */
+ regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
+
+ if (OP(nxt) != CLOSE)
+ FAIL("panic opt close");
+ oscan->flags = ARG(nxt);
+ OP(nxt1) = OPTIMIZED; /* was OPEN. */
+ OP(nxt) = OPTIMIZED; /* was CLOSE. */
+#ifdef DEBUGGING
+ OP(nxt1 + 1) = OPTIMIZED; /* was count. */
+ OP(nxt + 1) = OPTIMIZED; /* was count. */
+ NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
+ NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
+#endif
+#if 0
+ while ( nxt1 && (OP(nxt1) != WHILEM)) {
+ regnode *nnxt = regnext(nxt1);
+
+ if (nnxt == nxt) {
+ if (reg_off_by_arg[OP(nxt1)])
+ ARG_SET(nxt1, nxt2 - nxt1);
+ else if (nxt2 - nxt1 < U16_MAX)
+ NEXT_OFF(nxt1) = nxt2 - nxt1;
+ else
+ OP(nxt) = NOTHING; /* Cannot beautify */
+ }
+ nxt1 = nnxt;
+ }
+#endif
+ /* Optimize again: */
+ study_chunk(&nxt1, &deltanext, nxt, NULL, 0);
+ } else
+ oscan->flags = 0;
+ }
+ if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
+ pars++;
+ if (flags & SCF_DO_SUBSTR) {
+ SV *last_str = Nullsv;
+ int counted = mincount != 0;
+
+ if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
+ I32 b = pos_before >= data->last_start_min
+ ? pos_before : data->last_start_min;
+ STRLEN l;
+ char *s = SvPV(data->last_found, l);
+
+ l -= b - data->last_start_min;
+ /* Get the added string: */
+ last_str = newSVpv(s + b - data->last_start_min, l);
+ if (deltanext == 0 && pos_before == b) {
+ /* What was added is a constant string */
+ if (mincount > 1) {
+ SvGROW(last_str, (mincount * l) + 1);
+ repeatcpy(SvPVX(last_str) + l,
+ SvPVX(last_str), l, mincount - 1);
+ SvCUR(last_str) *= mincount;
+ /* Add additional parts. */
+ SvCUR_set(data->last_found,
+ SvCUR(data->last_found) - l);
+ sv_catsv(data->last_found, last_str);
+ data->last_end += l * (mincount - 1);
+ }
+ }
+ }
+ /* It is counted once already... */
+ data->pos_min += minnext * (mincount - counted);
+ data->pos_delta += - counted * deltanext +
+ (minnext + deltanext) * maxcount - minnext * mincount;
+ if (mincount != maxcount) {
+ scan_commit(data);
+ if (mincount && last_str) {
+ sv_setsv(data->last_found, last_str);
+ data->last_end = data->pos_min;
+ data->last_start_min =
+ data->pos_min - SvCUR(last_str);
+ data->last_start_max = is_inf
+ ? I32_MAX
+ : data->pos_min + data->pos_delta
+ - SvCUR(last_str);
+ }
+ data->longest = &(data->longest_float);
+ }
+ }
+ if (data && (fl & SF_HAS_EVAL))
+ data->flags |= SF_HAS_EVAL;
+ optimize_curly_tail:
+ if (OP(oscan) != CURLYX) {
+ while (regkind[(U8)OP(next = regnext(oscan))] == NOTHING
+ && NEXT_OFF(next))
+ NEXT_OFF(oscan) += NEXT_OFF(next);
+ }
+ continue;
+ default: /* REF only? */
+ if (flags & SCF_DO_SUBSTR) {
+ scan_commit(data);
+ data->longest = &(data->longest_float);
+ }
+ is_inf = 1;
+ break;
+ }
+ } else if (strchr(simple,OP(scan))) {
+ if (flags & SCF_DO_SUBSTR) {
+ scan_commit(data);
+ data->pos_min++;
+ }
+ min++;
+ } else if (regkind[(U8)OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
+ data->flags |= (OP(scan) == MEOL
+ ? SF_BEFORE_MEOL
+ : SF_BEFORE_SEOL);
+ } else if (regkind[(U8)OP(scan)] == BRANCHJ
+ && (scan->flags || data)
+ && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
+ I32 deltanext, minnext;
+ regnode *nscan;
+
+ data_fake.flags = 0;
+ next = regnext(scan);
+ nscan = NEXTOPER(NEXTOPER(scan));
+ minnext = study_chunk(&nscan, &deltanext, last, &data_fake, 0);
+ if (scan->flags) {
+ if (deltanext) {
+ FAIL("variable length lookbehind not implemented");
+ } else if (minnext > U8_MAX) {
+ FAIL2("lookbehind longer than %d not implemented", U8_MAX);
+ }
+ scan->flags = minnext;
+ }
+ if (data && data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
+ pars++;
+ if (data && (data_fake.flags & SF_HAS_EVAL))
+ data->flags |= SF_HAS_EVAL;
+ } else if (OP(scan) == OPEN) {
+ pars++;
+ } else if (OP(scan) == CLOSE && ARG(scan) == is_par) {
+ next = regnext(scan);
+
+ if ( next && (OP(next) != WHILEM) && next < last)
+ is_par = 0; /* Disable optimization */
+ } else if (OP(scan) == EVAL) {
+ if (data)
+ data->flags |= SF_HAS_EVAL;
+ }
+ /* Else: zero-length, ignore. */
+ scan = regnext(scan);
+ }
+
+ finish:
+ *scanp = scan;
+ *deltap = is_inf ? I32_MAX : delta;
+ if (flags & SCF_DO_SUBSTR && is_inf)
+ data->pos_delta = I32_MAX - data->pos_min;
+ if (is_par > U8_MAX)
+ is_par = 0;
+ if (is_par && pars==1 && data) {
+ data->flags |= SF_IN_PAR;
+ data->flags &= ~SF_HAS_PAR;
+ } else if (pars && data) {
+ data->flags |= SF_HAS_PAR;
+ data->flags &= ~SF_IN_PAR;
+ }
+ return min;
+}
+
+STATIC I32
+add_data(I32 n, char *s)
+{
+ dTHR;
+ if (regcomp_rx->data) {
+ Renewc(regcomp_rx->data,
+ sizeof(*regcomp_rx->data) + sizeof(void*) * (regcomp_rx->data->count + n - 1),
+ char, struct reg_data);
+ Renew(regcomp_rx->data->what, regcomp_rx->data->count + n, U8);
+ regcomp_rx->data->count += n;
+ } else {
+ Newc(1207, regcomp_rx->data, sizeof(*regcomp_rx->data) + sizeof(void*) * (n - 1),
+ char, struct reg_data);
+ New(1208, regcomp_rx->data->what, n, U8);
+ regcomp_rx->data->count = n;
+ }
+ Copy(s, regcomp_rx->data->what + regcomp_rx->data->count - n, n, U8);
+ return regcomp_rx->data->count - n;
+}
/*
- pregcomp - compile a regular expression into internal code
* of the structure of the compiled regexp. [I'll say.]
*/
regexp *
-pregcomp(exp,xend,pm)
-char* exp;
-char* xend;
-PMOP* pm;
+pregcomp(char *exp, char *xend, PMOP *pm)
{
+ dTHR;
register regexp *r;
- register char *scan;
- register SV *longish;
- SV *longest;
- register I32 len;
- register char *first;
+ regnode *scan;
+ SV **longest;
+ SV *longest_fixed;
+ SV *longest_float;
+ regnode *first;
I32 flags;
- I32 backish;
- I32 backest;
- I32 curback;
I32 minlen = 0;
I32 sawplus = 0;
I32 sawopen = 0;
-#define MAX_REPEAT_DEPTH 12
- struct {
- char *opcode;
- I32 count;
- } repeat_stack[MAX_REPEAT_DEPTH];
- I32 repeat_depth = 0;
- I32 repeat_count = 1; /* We start unmultiplied. */
if (exp == NULL)
- croak("NULL regexp argument");
+ FAIL("NULL regexp argument");
regprecomp = savepvn(exp, xend - exp);
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "compiling RE `%*s'\n",
+ xend - exp, regprecomp));
regflags = pm->op_pmflags;
regsawback = 0;
+ regseen = 0;
+ seen_zerolen = *exp == '^' ? -1 : 0;
+ seen_evals = 0;
+ extralen = 0;
+
/* First pass: determine size, legality. */
- regparse = exp;
+ regcomp_parse = exp;
regxend = xend;
regnaughty = 0;
regnpar = 1;
regsize = 0L;
regcode = ®dummy;
- regc((char)MAGIC);
+ regc((U8)MAGIC, (char*)regcode);
if (reg(0, &flags) == NULL) {
Safefree(regprecomp);
regprecomp = Nullch;
return(NULL);
}
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "size %d ", regsize));
+
+ DEBUG_r(
+ if (!colorset) {
+ int i = 0;
+ char *s = PerlEnv_getenv("TERMCAP_COLORS");
+
+ colorset = 1;
+ if (s) {
+ colors[0] = s = savepv(s);
+ while (++i < 4) {
+ s = strchr(s, '\t');
+ if (!s)
+ FAIL("Not enough TABs in TERMCAP_COLORS");
+ *s = '\0';
+ colors[i] = ++s;
+ }
+ } else {
+ while (i < 4)
+ colors[i++] = "";
+ }
+ /* Reset colors: */
+ PerlIO_printf(Perl_debug_log, "%s%s%s%s",
+ colors[0],colors[1],colors[2],colors[3]);
+ }
+ );
- /* Small enough for pointer-storage convention? */
- if (regsize >= 32767L) /* Probably could be 65535L. */
- FAIL("regexp too big");
+ /* Small enough for pointer-storage convention?
+ If extralen==0, this means that we will not need long jumps. */
+ if (regsize >= 0x10000L && extralen)
+ regsize += extralen;
+ else
+ extralen = 0;
/* Allocate space and initialize. */
- Newc(1001, r, sizeof(regexp) + (unsigned)regsize, char, regexp);
+ Newc(1001, r, sizeof(regexp) + (unsigned)regsize * sizeof(regnode),
+ char, regexp);
if (r == NULL)
FAIL("regexp out of space");
+ r->refcnt = 1;
r->prelen = xend - exp;
r->precomp = regprecomp;
r->subbeg = r->subbase = NULL;
+ r->nparens = regnpar - 1; /* set early to validate backrefs */
+ regcomp_rx = r;
/* Second pass: emit code. */
- regparse = exp;
+ regcomp_parse = exp;
regxend = xend;
regnaughty = 0;
regnpar = 1;
regcode = r->program;
- regc((char)MAGIC);
+ /* Store the count of eval-groups for security checks: */
+ regcode->next_off = ((seen_evals > U16_MAX) ? U16_MAX : seen_evals);
+ regc((U8)MAGIC, (char*) regcode++);
+ r->data = 0;
if (reg(0, &flags) == NULL)
return(NULL);
/* Dig out information for optimizations. */
+ r->reganch = pm->op_pmflags & PMf_COMPILETIME;
pm->op_pmflags = regflags;
- r->regstart = Nullsv; /* Worst-case defaults. */
- r->reganch = 0;
- r->regmust = Nullsv;
- r->regback = -1;
- r->regstclass = Nullch;
+ r->regstclass = NULL;
r->naughty = regnaughty >= 10; /* Probably an expensive pattern. */
- scan = r->program+1; /* First BRANCH. */
- if (OP(regnext(scan)) == END) {/* Only one top-level choice. */
- scan = NEXTOPER(scan);
+ scan = r->program + 1; /* First BRANCH. */
+
+ /* XXXX To minimize changes to RE engine we always allocate
+ 3-units-long substrs field. */
+ Newz(1004, r->substrs, 1, struct reg_substr_data);
+ if (OP(scan) != BRANCH) { /* Only one top-level choice. */
+ scan_data_t data;
+ I32 fake;
+ STRLEN longest_float_length, longest_fixed_length;
+
+ StructCopy(&zero_scan_data, &data, scan_data_t);
first = scan;
+ /* Skip introductions and multiplicators >= 1. */
while ((OP(first) == OPEN && (sawopen = 1)) ||
(OP(first) == BRANCH && OP(regnext(first)) != BRANCH) ||
(OP(first) == PLUS) ||
/* Starting-point info. */
again:
- if (OP(first) == EXACT) {
- r->regstart = newSVpv(OPERAND(first)+1,*OPERAND(first));
- if (SvCUR(r->regstart) > !sawstudy)
- fbm_compile(r->regstart);
- (void)SvUPGRADE(r->regstart, SVt_PVBM);
- }
+ if (OP(first) == EXACT); /* Empty, get anchored substr later. */
else if (strchr(simple+2,OP(first)))
r->regstclass = first;
else if (regkind[(U8)OP(first)] == BOUND ||
regkind[(U8)OP(first)] == NBOUND)
r->regstclass = first;
else if (regkind[(U8)OP(first)] == BOL) {
- r->reganch |= ROPT_ANCH_BOL;
+ r->reganch |= (OP(first) == MBOL ? ROPT_ANCH_MBOL: ROPT_ANCH_BOL);
first = NEXTOPER(first);
goto again;
}
if (sawplus && (!sawopen || !regsawback))
r->reganch |= ROPT_SKIP; /* x+ must match 1st of run */
- DEBUG_r(PerlIO_printf(Perl_debug_log, "first %d next %d offset %d\n",
- OP(first), OP(NEXTOPER(first)), first - scan));
+ /* Scan is after the zeroth branch, first is atomic matcher. */
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "first at %d\n",
+ first - scan + 1));
/*
* If there's something expensive in the r.e., find the
* longest literal string that must appear and make it the
* and avoiding duplication strengthens checking. Not a
* strong reason, but sufficient in the absence of others.
* [Now we resolve ties in favor of the earlier string if
- * it happens that curback has been invalidated, since the
+ * it happens that c_offset_min has been invalidated, since the
* earlier string may buy us something the later one won't.]
*/
- longish = newSVpv("",0);
- longest = newSVpv("",0);
- len = 0;
minlen = 0;
- curback = 0;
- backish = 0;
- backest = 0;
- while (OP(scan) != END) {
- if (OP(scan) == BRANCH) {
- if (OP(regnext(scan)) == BRANCH) {
- curback = -30000;
- while (OP(scan) == BRANCH)
- scan = regnext(scan);
- }
- else /* single branch is ok */
- scan = NEXTOPER(scan);
- continue;
- }
- if (OP(scan) == UNLESSM) {
- curback = -30000;
- scan = regnext(scan);
- continue;
- }
- if (OP(scan) == EXACT) {
- char *t;
-
- first = scan;
- while ((t = regnext(scan)) && OP(t) == CLOSE)
- scan = t;
- minlen += *OPERAND(first) * repeat_count;
- if (curback - backish == len) {
- sv_catpvn(longish, OPERAND(first)+1,
- *OPERAND(first));
- len += *OPERAND(first);
- curback += *OPERAND(first);
- first = regnext(scan);
- }
- else if (*OPERAND(first) >= len + (curback >= 0)) {
- len = *OPERAND(first);
- sv_setpvn(longish, OPERAND(first)+1,len);
- backish = curback;
- curback += len;
- first = regnext(scan);
- }
- else
- curback += *OPERAND(first);
- }
- else if (strchr(varies,OP(scan))) {
- int tcount;
- char *next;
-
- if (repeat_depth < MAX_REPEAT_DEPTH
- && ((OP(scan) == PLUS
- && (tcount = 1)
- && (next = NEXTOPER(scan)))
- || (regkind[(U8)OP(scan)] == CURLY
- && (tcount = ARG1(scan))
- && (next = NEXTOPER(scan)+4))))
- {
- /* We treat (abc)+ as (abc)(abc)*. */
-
- /* Mark the place to return back. */
- repeat_stack[repeat_depth].opcode = regnext(scan);
- repeat_stack[repeat_depth].count = repeat_count;
- repeat_depth++;
- repeat_count *= tcount;
-
- /* Go deeper: */
- scan = next;
- continue;
- }
- else {
- curback = -30000;
- len = 0;
- if (SvCUR(longish) > SvCUR(longest)) {
- sv_setsv(longest,longish);
- backest = backish;
- }
- sv_setpvn(longish,"",0);
- }
- }
- else if (strchr(simple,OP(scan))) {
- curback++;
- minlen += repeat_count;
- len = 0;
- if (SvCUR(longish) > SvCUR(longest)) {
- sv_setsv(longest,longish);
- backest = backish;
- }
- sv_setpvn(longish,"",0);
- }
- scan = regnext(scan);
- if (!scan) { /* Go up PLUS or CURLY. */
- if (!repeat_depth--)
- croak("panic: re scan");
- scan = repeat_stack[repeat_depth].opcode;
- repeat_count = repeat_stack[repeat_depth].count;
- /* Need to submit the longest string found: */
- curback = -30000;
- len = 0;
- if (SvCUR(longish) > SvCUR(longest)) {
- sv_setsv(longest,longish);
- backest = backish;
- }
- sv_setpvn(longish,"",0);
- }
- }
-
- /* Prefer earlier on tie, unless we can tail match latter */
- if (SvCUR(longish) + (first && regkind[(U8)OP(first)] == EOL)
- > SvCUR(longest))
- {
- sv_setsv(longest,longish);
- backest = backish;
+ data.longest_fixed = newSVpv("",0);
+ data.longest_float = newSVpv("",0);
+ data.last_found = newSVpv("",0);
+ data.longest = &(data.longest_fixed);
+ first = scan;
+
+ minlen = study_chunk(&first, &fake, scan + regsize, /* Up to end */
+ &data, SCF_DO_SUBSTR);
+ if ( regnpar == 1 && data.longest == &(data.longest_fixed)
+ && data.last_start_min == 0 && data.last_end > 0
+ && !seen_zerolen
+ && (!(regseen & REG_SEEN_GPOS) || (r->reganch & ROPT_ANCH_GPOS)))
+ r->reganch |= ROPT_CHECK_ALL;
+ scan_commit(&data);
+ SvREFCNT_dec(data.last_found);
+
+ longest_float_length = SvCUR(data.longest_float);
+ if (longest_float_length
+ || (data.flags & SF_FL_BEFORE_EOL
+ && (!(data.flags & SF_FL_BEFORE_MEOL)
+ || (regflags & PMf_MULTILINE)))) {
+ if (SvCUR(data.longest_fixed)
+ && data.offset_fixed == data.offset_float_min)
+ goto remove; /* Like in (a)+. */
+
+ r->float_substr = data.longest_float;
+ r->float_min_offset = data.offset_float_min;
+ r->float_max_offset = data.offset_float_max;
+ fbm_compile(r->float_substr, 0);
+ BmUSEFUL(r->float_substr) = 100;
+ if (data.flags & SF_FL_BEFORE_EOL /* Cannot have SEOL and MULTI */
+ && (!(data.flags & SF_FL_BEFORE_MEOL)
+ || (regflags & PMf_MULTILINE)))
+ SvTAIL_on(r->float_substr);
+ } else {
+ remove:
+ r->float_substr = Nullsv;
+ SvREFCNT_dec(data.longest_float);
+ longest_float_length = 0;
}
- else
- sv_setpvn(longish,"",0);
- if (SvCUR(longest)
- && (!r->regstart
- || !fbm_instr((unsigned char*) SvPVX(r->regstart),
- (unsigned char *) (SvPVX(r->regstart)
- + SvCUR(r->regstart)),
- longest)))
- {
- r->regmust = longest;
- if (backest < 0)
- backest = -1;
- r->regback = backest;
- if (SvCUR(longest) > !(sawstudy ||
- (first && regkind[(U8)OP(first)] == EOL)))
- fbm_compile(r->regmust);
- (void)SvUPGRADE(r->regmust, SVt_PVBM);
- BmUSEFUL(r->regmust) = 100;
- if (first && regkind[(U8)OP(first)] == EOL && SvCUR(longish))
- SvTAIL_on(r->regmust);
+
+ longest_fixed_length = SvCUR(data.longest_fixed);
+ if (longest_fixed_length
+ || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
+ && (!(data.flags & SF_FIX_BEFORE_MEOL)
+ || (regflags & PMf_MULTILINE)))) {
+ r->anchored_substr = data.longest_fixed;
+ r->anchored_offset = data.offset_fixed;
+ fbm_compile(r->anchored_substr, 0);
+ BmUSEFUL(r->anchored_substr) = 100;
+ if (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
+ && (!(data.flags & SF_FIX_BEFORE_MEOL)
+ || (regflags & PMf_MULTILINE)))
+ SvTAIL_on(r->anchored_substr);
+ } else {
+ r->anchored_substr = Nullsv;
+ SvREFCNT_dec(data.longest_fixed);
+ longest_fixed_length = 0;
}
- else {
- SvREFCNT_dec(longest);
- longest = Nullsv;
+
+ /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
+ if (longest_fixed_length > longest_float_length) {
+ r->check_substr = r->anchored_substr;
+ r->check_offset_min = r->check_offset_max = r->anchored_offset;
+ if (r->reganch & ROPT_ANCH_SINGLE)
+ r->reganch |= ROPT_NOSCAN;
+ } else {
+ r->check_substr = r->float_substr;
+ r->check_offset_min = data.offset_float_min;
+ r->check_offset_max = data.offset_float_max;
}
- SvREFCNT_dec(longish);
+ } else {
+ /* Several toplevels. Best we can is to set minlen. */
+ I32 fake;
+
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "\n"));
+ scan = r->program + 1;
+ minlen = study_chunk(&scan, &fake, scan + regsize, NULL, 0);
+ r->check_substr = r->anchored_substr = r->float_substr = Nullsv;
}
- r->nparens = regnpar - 1;
r->minlen = minlen;
+ if (regseen & REG_SEEN_GPOS)
+ r->reganch |= ROPT_GPOS_SEEN;
+ if (regseen & REG_SEEN_LOOKBEHIND)
+ r->reganch |= ROPT_LOOKBEHIND_SEEN;
+ if (regseen & REG_SEEN_EVAL)
+ r->reganch |= ROPT_EVAL_SEEN;
Newz(1002, r->startp, regnpar, char*);
Newz(1002, r->endp, regnpar, char*);
DEBUG_r(regdump(r));
* is a trifle forced, but the need to tie the tails of the branches to what
* follows makes it hard to avoid.
*/
-static char *
-reg(paren, flagp)
-I32 paren; /* Parenthesized? */
-I32 *flagp;
+STATIC regnode *
+reg(I32 paren, I32 *flagp)
+ /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
{
- register char *ret;
- register char *br;
- register char *ender = 0;
+ dTHR;
+ register regnode *ret; /* Will be the head of the group. */
+ register regnode *br;
+ register regnode *lastbr;
+ register regnode *ender = 0;
register I32 parno = 0;
- I32 flags;
+ I32 flags, oregflags = regflags, have_branch = 0, open = 0;
+ char c;
- *flagp = HASWIDTH; /* Tentatively. */
+ *flagp = 0; /* Tentatively. */
/* Make an OPEN node, if parenthesized. */
if (paren) {
- if (*regparse == '?') {
- regparse++;
- paren = *regparse++;
- ret = NULL;
+ if (*regcomp_parse == '?') {
+ U16 posflags = 0, negflags = 0;
+ U16 *flagsp = &posflags;
+
+ regcomp_parse++;
+ paren = *regcomp_parse++;
+ ret = NULL; /* For look-ahead/behind. */
switch (paren) {
- case ':':
+ case '<':
+ regseen |= REG_SEEN_LOOKBEHIND;
+ if (*regcomp_parse == '!')
+ paren = ',';
+ if (*regcomp_parse != '=' && *regcomp_parse != '!')
+ goto unknown;
+ regcomp_parse++;
case '=':
case '!':
+ seen_zerolen++;
+ case ':':
+ case '>':
break;
case '$':
case '@':
- croak("Sequence (?%c...) not implemented", (int)paren);
+ FAIL2("Sequence (?%c...) not implemented", (int)paren);
break;
case '#':
- while (*regparse && *regparse != ')')
- regparse++;
- if (*regparse != ')')
- croak("Sequence (?#... not terminated");
+ while (*regcomp_parse && *regcomp_parse != ')')
+ regcomp_parse++;
+ if (*regcomp_parse != ')')
+ FAIL("Sequence (?#... not terminated");
nextchar();
*flagp = TRYAGAIN;
return NULL;
+ case '{':
+ {
+ dTHR;
+ I32 count = 1, n = 0;
+ char c;
+ char *s = regcomp_parse;
+ SV *sv;
+ OP_4tree *sop, *rop;
+
+ seen_zerolen++;
+ regseen |= REG_SEEN_EVAL;
+ while (count && (c = *regcomp_parse)) {
+ if (c == '\\' && regcomp_parse[1])
+ regcomp_parse++;
+ else if (c == '{')
+ count++;
+ else if (c == '}')
+ count--;
+ regcomp_parse++;
+ }
+ if (*regcomp_parse != ')')
+ FAIL("Sequence (?{...}) not terminated or not {}-balanced");
+ if (!SIZE_ONLY) {
+ AV *av;
+
+ if (regcomp_parse - 1 - s)
+ sv = newSVpv(s, regcomp_parse - 1 - s);
+ else
+ sv = newSVpv("", 0);
+
+ rop = sv_compile_2op(sv, &sop, "re", &av);
+
+ n = add_data(3, "nso");
+ regcomp_rx->data->data[n] = (void*)rop;
+ regcomp_rx->data->data[n+1] = (void*)av;
+ regcomp_rx->data->data[n+2] = (void*)sop;
+ SvREFCNT_dec(sv);
+ } else { /* First pass */
+ if (reginterp_cnt < ++seen_evals && curcop != &compiling)
+ /* No compiled RE interpolated, has runtime
+ components ===> unsafe. */
+ FAIL("Eval-group not allowed at runtime, use re 'eval'");
+ if (tainted)
+ FAIL("Eval-group in insecure regular expression");
+ }
+
+ nextchar();
+ return reganode(EVAL, n);
+ }
+ case '(':
+ {
+ if (regcomp_parse[0] == '?') {
+ if (regcomp_parse[1] == '=' || regcomp_parse[1] == '!'
+ || regcomp_parse[1] == '<'
+ || regcomp_parse[1] == '{') { /* Lookahead or eval. */
+ I32 flag;
+
+ ret = reg_node(LOGICAL);
+ regtail(ret, reg(1, &flag));
+ goto insert_if;
+ }
+ } else if (regcomp_parse[0] >= '1' && regcomp_parse[0] <= '9' ) {
+ parno = atoi(regcomp_parse++);
+
+ while (isDIGIT(*regcomp_parse))
+ regcomp_parse++;
+ ret = reganode(GROUPP, parno);
+ if ((c = *nextchar()) != ')')
+ FAIL2("Switch (?(number%c not recognized", c);
+ insert_if:
+ regtail(ret, reganode(IFTHEN, 0));
+ br = regbranch(&flags, 1);
+ if (br == NULL)
+ br = reganode(LONGJMP, 0);
+ else
+ regtail(br, reganode(LONGJMP, 0));
+ c = *nextchar();
+ if (c == '|') {
+ lastbr = reganode(IFTHEN, 0); /* Fake one for optimizer. */
+ regbranch(&flags, 1);
+ regtail(ret, lastbr);
+ c = *nextchar();
+ } else
+ lastbr = NULL;
+ if (c != ')')
+ FAIL("Switch (?(condition)... contains too many branches");
+ ender = reg_node(TAIL);
+ regtail(br, ender);
+ if (lastbr) {
+ regtail(lastbr, ender);
+ regtail(NEXTOPER(NEXTOPER(lastbr)), ender);
+ } else
+ regtail(ret, ender);
+ return ret;
+ } else {
+ FAIL2("Unknown condition for (?(%.2s", regcomp_parse);
+ }
+ }
+ case 0:
+ FAIL("Sequence (? incomplete");
+ break;
default:
- --regparse;
- while (*regparse && strchr("iogmsx", *regparse))
- pmflag(®flags, *regparse++);
- if (*regparse != ')')
- croak("Sequence (?%c...) not recognized", *regparse);
+ --regcomp_parse;
+ parse_flags:
+ while (*regcomp_parse && strchr("iogcmsx", *regcomp_parse)) {
+ if (*regcomp_parse != 'o')
+ pmflag(flagsp, *regcomp_parse);
+ ++regcomp_parse;
+ }
+ if (*regcomp_parse == '-') {
+ flagsp = &negflags;
+ ++regcomp_parse;
+ goto parse_flags;
+ }
+ regflags |= posflags;
+ regflags &= ~negflags;
+ if (*regcomp_parse == ':') {
+ regcomp_parse++;
+ paren = ':';
+ break;
+ }
+ unknown:
+ if (*regcomp_parse != ')')
+ FAIL2("Sequence (?%c...) not recognized", *regcomp_parse);
nextchar();
*flagp = TRYAGAIN;
return NULL;
parno = regnpar;
regnpar++;
ret = reganode(OPEN, parno);
+ open = 1;
}
} else
ret = NULL;
/* Pick up the branches, linking them together. */
- br = regbranch(&flags);
+ br = regbranch(&flags, 1);
if (br == NULL)
return(NULL);
- if (ret != NULL)
- regtail(ret, br); /* OPEN -> first. */
- else
+ if (*regcomp_parse == '|') {
+ if (!SIZE_ONLY && extralen) {
+ reginsert(BRANCHJ, br);
+ } else
+ reginsert(BRANCH, br);
+ have_branch = 1;
+ if (SIZE_ONLY)
+ extralen += 1; /* For BRANCHJ-BRANCH. */
+ } else if (paren == ':') {
+ *flagp |= flags&SIMPLE;
+ }
+ if (open) { /* Starts with OPEN. */
+ regtail(ret, br); /* OPEN -> first. */
+ } else if (paren != '?') /* Not Conditional */
ret = br;
- if (!(flags&HASWIDTH))
- *flagp &= ~HASWIDTH;
+ if (flags&HASWIDTH)
+ *flagp |= HASWIDTH;
*flagp |= flags&SPSTART;
- while (*regparse == '|') {
+ lastbr = br;
+ while (*regcomp_parse == '|') {
+ if (!SIZE_ONLY && extralen) {
+ ender = reganode(LONGJMP,0);
+ regtail(NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
+ }
+ if (SIZE_ONLY)
+ extralen += 2; /* Account for LONGJMP. */
nextchar();
- br = regbranch(&flags);
+ br = regbranch(&flags, 0);
if (br == NULL)
return(NULL);
- regtail(ret, br); /* BRANCH -> BRANCH. */
- if (!(flags&HASWIDTH))
- *flagp &= ~HASWIDTH;
+ regtail(lastbr, br); /* BRANCH -> BRANCH. */
+ lastbr = br;
+ if (flags&HASWIDTH)
+ *flagp |= HASWIDTH;
*flagp |= flags&SPSTART;
}
- /* Make a closing node, and hook it on the end. */
- switch (paren) {
- case ':':
- ender = regnode(NOTHING);
- break;
- case 1:
- ender = reganode(CLOSE, parno);
- break;
- case '=':
- case '!':
- ender = regnode(SUCCEED);
- *flagp &= ~HASWIDTH;
- break;
- case 0:
- ender = regnode(END);
- break;
+ if (have_branch || paren != ':') {
+ /* Make a closing node, and hook it on the end. */
+ switch (paren) {
+ case ':':
+ ender = reg_node(TAIL);
+ break;
+ case 1:
+ ender = reganode(CLOSE, parno);
+ break;
+ case '<':
+ case ',':
+ case '=':
+ case '!':
+ *flagp &= ~HASWIDTH;
+ /* FALL THROUGH */
+ case '>':
+ ender = reg_node(SUCCEED);
+ break;
+ case 0:
+ ender = reg_node(END);
+ break;
+ }
+ regtail(lastbr, ender);
+
+ if (have_branch) {
+ /* Hook the tails of the branches to the closing node. */
+ for (br = ret; br != NULL; br = regnext(br)) {
+ regoptail(br, ender);
+ }
+ }
}
- regtail(ret, ender);
- /* Hook the tails of the branches to the closing node. */
- for (br = ret; br != NULL; br = regnext(br))
- regoptail(br, ender);
+ {
+ char *p;
+ static char parens[] = "=!<,>";
- if (paren == '=') {
- reginsert(IFMATCH,ret);
- regtail(ret, regnode(NOTHING));
- }
- else if (paren == '!') {
- reginsert(UNLESSM,ret);
- regtail(ret, regnode(NOTHING));
+ if (paren && (p = strchr(parens, paren))) {
+ int node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
+ int flag = (p - parens) > 1;
+
+ if (paren == '>')
+ node = SUSPEND, flag = 0;
+ reginsert(node,ret);
+ ret->flags = flag;
+ regtail(ret, reg_node(TAIL));
+ }
}
/* Check for proper termination. */
- if (paren && (regparse >= regxend || *nextchar() != ')')) {
+ if (paren && (regcomp_parse >= regxend || *nextchar() != ')')) {
FAIL("unmatched () in regexp");
- } else if (!paren && regparse < regxend) {
- if (*regparse == ')') {
+ } else if (!paren && regcomp_parse < regxend) {
+ if (*regcomp_parse == ')') {
FAIL("unmatched () in regexp");
} else
FAIL("junk on end of regexp"); /* "Can't happen". */
/* NOTREACHED */
}
+ if (paren != 0) {
+ regflags = oregflags;
+ }
return(ret);
}
*
* Implements the concatenation operator.
*/
-static char *
-regbranch(flagp)
-I32 *flagp;
+STATIC regnode *
+regbranch(I32 *flagp, I32 first)
{
- register char *ret;
- register char *chain;
- register char *latest;
- I32 flags = 0;
+ dTHR;
+ register regnode *ret;
+ register regnode *chain = NULL;
+ register regnode *latest;
+ I32 flags = 0, c = 0;
- *flagp = WORST; /* Tentatively. */
+ if (first)
+ ret = NULL;
+ else {
+ if (!SIZE_ONLY && extralen)
+ ret = reganode(BRANCHJ,0);
+ else
+ ret = reg_node(BRANCH);
+ }
+
+ if (!first && SIZE_ONLY)
+ extralen += 1; /* BRANCHJ */
+
+ *flagp = WORST; /* Tentatively. */
- ret = regnode(BRANCH);
- chain = NULL;
- regparse--;
+ regcomp_parse--;
nextchar();
- while (regparse < regxend && *regparse != '|' && *regparse != ')') {
+ while (regcomp_parse < regxend && *regcomp_parse != '|' && *regcomp_parse != ')') {
flags &= ~TRYAGAIN;
latest = regpiece(&flags);
if (latest == NULL) {
if (flags & TRYAGAIN)
continue;
return(NULL);
- }
+ } else if (ret == NULL)
+ ret = latest;
*flagp |= flags&HASWIDTH;
- if (chain == NULL) /* First piece. */
+ if (chain == NULL) /* First piece. */
*flagp |= flags&SPSTART;
else {
regnaughty++;
regtail(chain, latest);
}
chain = latest;
+ c++;
+ }
+ if (chain == NULL) { /* Loop ran zero times. */
+ chain = reg_node(NOTHING);
+ if (ret == NULL)
+ ret = chain;
+ }
+ if (c == 1) {
+ *flagp |= flags&SIMPLE;
}
- if (chain == NULL) /* Loop ran zero times. */
- (void) regnode(NOTHING);
return(ret);
}
* It might seem that this node could be dispensed with entirely, but the
* endmarker role is not redundant.
*/
-static char *
-regpiece(flagp)
-I32 *flagp;
+STATIC regnode *
+regpiece(I32 *flagp)
{
- register char *ret;
+ dTHR;
+ register regnode *ret;
register char op;
register char *next;
I32 flags;
- char *origparse = regparse;
+ char *origparse = regcomp_parse;
char *maxpos;
I32 min;
- I32 max = 32767;
+ I32 max = REG_INFTY;
ret = regatom(&flags);
if (ret == NULL) {
return(NULL);
}
- op = *regparse;
- if (op == '(' && regparse[1] == '?' && regparse[2] == '#') {
- while (op && op != ')')
- op = *++regparse;
- if (op) {
- nextchar();
- op = *regparse;
- }
- }
+ op = *regcomp_parse;
- if (op == '{' && regcurly(regparse)) {
- next = regparse + 1;
+ if (op == '{' && regcurly(regcomp_parse)) {
+ next = regcomp_parse + 1;
maxpos = Nullch;
while (isDIGIT(*next) || *next == ',') {
if (*next == ',') {
if (*next == '}') { /* got one */
if (!maxpos)
maxpos = next;
- regparse++;
- min = atoi(regparse);
+ regcomp_parse++;
+ min = atoi(regcomp_parse);
if (*maxpos == ',')
maxpos++;
else
- maxpos = regparse;
+ maxpos = regcomp_parse;
max = atoi(maxpos);
if (!max && *maxpos != '0')
- max = 32767; /* meaning "infinity" */
- regparse = next;
+ max = REG_INFTY; /* meaning "infinity" */
+ else if (max >= REG_INFTY)
+ FAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
+ regcomp_parse = next;
nextchar();
do_curly:
}
else {
regnaughty += 4 + regnaughty; /* compound interest */
- regtail(ret, regnode(WHILEM));
+ regtail(ret, reg_node(WHILEM));
+ if (!SIZE_ONLY && extralen) {
+ reginsert(LONGJMP,ret);
+ reginsert(NOTHING,ret);
+ NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
+ }
reginsert(CURLYX,ret);
- regtail(ret, regnode(NOTHING));
+ if (!SIZE_ONLY && extralen)
+ NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
+ regtail(ret, reg_node(NOTHING));
+ if (SIZE_ONLY)
+ extralen += 3;
}
+ ret->flags = 0;
if (min > 0)
- *flagp = (WORST|HASWIDTH);
+ *flagp = WORST;
+ if (max > 0)
+ *flagp |= HASWIDTH;
if (max && max < min)
- croak("Can't do {n,m} with n > m");
- if (regcode != ®dummy) {
-#ifdef REGALIGN
- *(unsigned short *)(ret+3) = min;
- *(unsigned short *)(ret+5) = max;
-#else
- ret[3] = min >> 8; ret[4] = min & 0377;
- ret[5] = max >> 8; ret[6] = max & 0377;
-#endif
+ FAIL("Can't do {n,m} with n > m");
+ if (!SIZE_ONLY) {
+ ARG1_SET(ret, min);
+ ARG2_SET(ret, max);
}
goto nest_check;
*flagp = flags;
return(ret);
}
+
+#if 0 /* Now runtime fix should be reliable. */
+ if (!(flags&HASWIDTH) && op != '?')
+ FAIL("regexp *+ operand could be empty");
+#endif
+
nextchar();
- *flagp = (op != '+') ? (WORST|SPSTART) : (WORST|HASWIDTH);
+ *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
if (op == '*' && (flags&SIMPLE)) {
reginsert(STAR, ret);
+ ret->flags = 0;
regnaughty += 4;
}
else if (op == '*') {
goto do_curly;
} else if (op == '+' && (flags&SIMPLE)) {
reginsert(PLUS, ret);
+ ret->flags = 0;
regnaughty += 3;
}
else if (op == '+') {
goto do_curly;
}
nest_check:
- if (dowarn && regcode != ®dummy && !(flags&HASWIDTH) && max > 10000) {
+ if (dowarn && !SIZE_ONLY && !(flags&HASWIDTH) && max > 10000) {
warn("%.*s matches null string many times",
- regparse - origparse, origparse);
+ regcomp_parse - origparse, origparse);
}
- if (*regparse == '?') {
+ if (*regcomp_parse == '?') {
nextchar();
reginsert(MINMOD, ret);
-#ifdef REGALIGN
- regtail(ret, ret + 4);
-#else
- regtail(ret, ret + 3);
-#endif
+ regtail(ret, ret + NODE_STEP_REGNODE);
}
- if (ISMULT2(regparse))
+ if (ISMULT2(regcomp_parse))
FAIL("nested *?+ in regexp");
return(ret);
*
* [Yes, it is worth fixing, some scripts can run twice the speed.]
*/
-static char *
-regatom(flagp)
-I32 *flagp;
+STATIC regnode *
+regatom(I32 *flagp)
{
- register char *ret = 0;
+ dTHR;
+ register regnode *ret = 0;
I32 flags;
*flagp = WORST; /* Tentatively. */
tryagain:
- switch (*regparse) {
+ switch (*regcomp_parse) {
case '^':
+ seen_zerolen++;
nextchar();
if (regflags & PMf_MULTILINE)
- ret = regnode(MBOL);
+ ret = reg_node(MBOL);
else if (regflags & PMf_SINGLELINE)
- ret = regnode(SBOL);
+ ret = reg_node(SBOL);
else
- ret = regnode(BOL);
+ ret = reg_node(BOL);
break;
case '$':
+ if (regcomp_parse[1])
+ seen_zerolen++;
nextchar();
if (regflags & PMf_MULTILINE)
- ret = regnode(MEOL);
+ ret = reg_node(MEOL);
else if (regflags & PMf_SINGLELINE)
- ret = regnode(SEOL);
+ ret = reg_node(SEOL);
else
- ret = regnode(EOL);
+ ret = reg_node(EOL);
break;
case '.':
nextchar();
if (regflags & PMf_SINGLELINE)
- ret = regnode(SANY);
+ ret = reg_node(SANY);
else
- ret = regnode(ANY);
+ ret = reg_node(ANY);
regnaughty++;
*flagp |= HASWIDTH|SIMPLE;
break;
case '[':
- regparse++;
+ regcomp_parse++;
ret = regclass();
*flagp |= HASWIDTH|SIMPLE;
break;
goto tryagain;
return(NULL);
}
- *flagp |= flags&(HASWIDTH|SPSTART);
+ *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE);
break;
case '|':
case ')':
*flagp |= TRYAGAIN;
return NULL;
}
- croak("internal urp in regexp at /%s/", regparse);
+ FAIL2("internal urp in regexp at /%s/", regcomp_parse);
/* Supposed to be caught earlier. */
break;
case '{':
- if (!regcurly(regparse)) {
- regparse++;
+ if (!regcurly(regcomp_parse)) {
+ regcomp_parse++;
goto defchar;
}
/* FALL THROUGH */
FAIL("?+*{} follows nothing in regexp");
break;
case '\\':
- switch (*++regparse) {
+ switch (*++regcomp_parse) {
case 'A':
- ret = regnode(SBOL);
+ seen_zerolen++;
+ ret = reg_node(SBOL);
*flagp |= SIMPLE;
nextchar();
break;
case 'G':
- ret = regnode(GPOS);
+ ret = reg_node(GPOS);
+ regseen |= REG_SEEN_GPOS;
*flagp |= SIMPLE;
nextchar();
break;
case 'Z':
- ret = regnode(SEOL);
+ ret = reg_node(SEOL);
+ *flagp |= SIMPLE;
+ nextchar();
+ break;
+ case 'z':
+ ret = reg_node(EOS);
*flagp |= SIMPLE;
+ seen_zerolen++; /* Do not optimize RE away */
nextchar();
break;
case 'w':
- ret = regnode((regflags & PMf_LOCALE) ? ALNUML : ALNUM);
+ ret = reg_node((regflags & PMf_LOCALE) ? ALNUML : ALNUM);
*flagp |= HASWIDTH|SIMPLE;
nextchar();
break;
case 'W':
- ret = regnode((regflags & PMf_LOCALE) ? NALNUML : NALNUM);
+ ret = reg_node((regflags & PMf_LOCALE) ? NALNUML : NALNUM);
*flagp |= HASWIDTH|SIMPLE;
nextchar();
break;
case 'b':
- ret = regnode((regflags & PMf_LOCALE) ? BOUNDL : BOUND);
+ seen_zerolen++;
+ ret = reg_node((regflags & PMf_LOCALE) ? BOUNDL : BOUND);
*flagp |= SIMPLE;
nextchar();
break;
case 'B':
- ret = regnode((regflags & PMf_LOCALE) ? NBOUNDL : NBOUND);
+ seen_zerolen++;
+ ret = reg_node((regflags & PMf_LOCALE) ? NBOUNDL : NBOUND);
*flagp |= SIMPLE;
nextchar();
break;
case 's':
- ret = regnode((regflags & PMf_LOCALE) ? SPACEL : SPACE);
+ ret = reg_node((regflags & PMf_LOCALE) ? SPACEL : SPACE);
*flagp |= HASWIDTH|SIMPLE;
nextchar();
break;
case 'S':
- ret = regnode((regflags & PMf_LOCALE) ? NSPACEL : NSPACE);
+ ret = reg_node((regflags & PMf_LOCALE) ? NSPACEL : NSPACE);
*flagp |= HASWIDTH|SIMPLE;
nextchar();
break;
case 'd':
- ret = regnode(DIGIT);
+ ret = reg_node(DIGIT);
*flagp |= HASWIDTH|SIMPLE;
nextchar();
break;
case 'D':
- ret = regnode(NDIGIT);
+ ret = reg_node(NDIGIT);
*flagp |= HASWIDTH|SIMPLE;
nextchar();
break;
case '1': case '2': case '3': case '4':
case '5': case '6': case '7': case '8': case '9':
{
- I32 num = atoi(regparse);
+ I32 num = atoi(regcomp_parse);
if (num > 9 && num >= regnpar)
goto defchar;
else {
+ if (!SIZE_ONLY && num > regcomp_rx->nparens)
+ FAIL("reference to nonexistent group");
regsawback = 1;
- ret = reganode(REF, num);
+ ret = reganode((regflags & PMf_FOLD)
+ ? ((regflags & PMf_LOCALE) ? REFFL : REFF)
+ : REF, num);
*flagp |= HASWIDTH;
- while (isDIGIT(*regparse))
- regparse++;
- regparse--;
+ while (isDIGIT(*regcomp_parse))
+ regcomp_parse++;
+ regcomp_parse--;
nextchar();
}
}
break;
case '\0':
- if (regparse >= regxend)
+ if (regcomp_parse >= regxend)
FAIL("trailing \\ in regexp");
/* FALL THROUGH */
default:
case '#':
if (regflags & PMf_EXTENDED) {
- while (regparse < regxend && *regparse != '\n') regparse++;
- if (regparse < regxend)
+ while (regcomp_parse < regxend && *regcomp_parse != '\n') regcomp_parse++;
+ if (regcomp_parse < regxend)
goto tryagain;
}
/* FALL THROUGH */
default: {
register I32 len;
- register char ender;
+ register U8 ender;
register char *p;
- char *oldp;
+ char *oldp, *s;
I32 numlen;
- regparse++;
+ regcomp_parse++;
defchar:
- ret = regnode((regflags & PMf_FOLD)
+ ret = reg_node((regflags & PMf_FOLD)
? ((regflags & PMf_LOCALE) ? EXACTFL : EXACTF)
: EXACT);
- regc(0); /* save spot for len */
- for (len = 0, p = regparse - 1;
+ s = (char *) OPERAND(ret);
+ regc(0, s++); /* save spot for len */
+ for (len = 0, p = regcomp_parse - 1;
len < 127 && p < regxend;
len++)
{
case 'A':
case 'G':
case 'Z':
+ case 'z':
case 'w':
case 'W':
case 'b':
p = oldp;
else {
len++;
- regc(ender);
+ regc(ender, s++);
}
break;
}
- regc(ender);
+ regc(ender, s++);
}
loopdone:
- regparse = p - 1;
+ regcomp_parse = p - 1;
nextchar();
if (len < 0)
FAIL("internal disaster in regexp");
*flagp |= HASWIDTH;
if (len == 1)
*flagp |= SIMPLE;
- if (regcode != ®dummy)
+ if (!SIZE_ONLY)
*OPERAND(ret) = len;
- regc('\0');
+ regc('\0', s++);
+ if (SIZE_ONLY) {
+ regsize += (len + 2 + sizeof(regnode) - 1) / sizeof(regnode);
+ } else {
+ regcode += (len + 2 + sizeof(regnode) - 1) / sizeof(regnode);
+ }
}
break;
}
return(ret);
}
-static char *
-regwhite(p, e)
-char *p;
-char *e;
+STATIC char *
+regwhite(char *p, char *e)
{
while (p < e) {
if (isSPACE(*p))
return p;
}
-static void
-regset(opnd, c)
-char *opnd;
-register I32 c;
-{
- if (opnd == ®dummy)
- return;
- c &= 0xFF;
- opnd[1 + (c >> 3)] |= (1 << (c & 7));
-}
-
-static char *
-regclass()
+STATIC regnode *
+regclass(void)
{
- register char *opnd;
- register I32 class;
+ dTHR;
+ register char *opnd, *s;
+ register I32 Class;
register I32 lastclass = 1234;
register I32 range = 0;
- register char *ret;
+ register regnode *ret;
register I32 def;
I32 numlen;
- ret = regnode(ANYOF);
- opnd = regcode;
- for (class = 0; class < 33; class++)
- regc(0);
- if (*regparse == '^') { /* Complement of range. */
+ s = opnd = (char *) OPERAND(regcode);
+ ret = reg_node(ANYOF);
+ for (Class = 0; Class < 33; Class++)
+ regc(0, s++);
+ if (*regcomp_parse == '^') { /* Complement of range. */
regnaughty++;
- regparse++;
- if (opnd != ®dummy)
+ regcomp_parse++;
+ if (!SIZE_ONLY)
*opnd |= ANYOF_INVERT;
}
- if (opnd != ®dummy) {
+ if (!SIZE_ONLY) {
+ regcode += ANY_SKIP;
if (regflags & PMf_FOLD)
*opnd |= ANYOF_FOLD;
if (regflags & PMf_LOCALE)
*opnd |= ANYOF_LOCALE;
+ } else {
+ regsize += ANY_SKIP;
}
- if (*regparse == ']' || *regparse == '-')
+ if (*regcomp_parse == ']' || *regcomp_parse == '-')
goto skipcond; /* allow 1st char to be ] or - */
- while (regparse < regxend && *regparse != ']') {
+ while (regcomp_parse < regxend && *regcomp_parse != ']') {
skipcond:
- class = UCHARAT(regparse++);
- if (class == '\\') {
- class = UCHARAT(regparse++);
- switch (class) {
+ Class = UCHARAT(regcomp_parse++);
+ if (Class == '[' && regcomp_parse + 1 < regxend &&
+ /* I smell either [: or [= or [. -- POSIX has been here, right? */
+ (*regcomp_parse == ':' || *regcomp_parse == '=' || *regcomp_parse == '.')) {
+ char posixccc = *regcomp_parse;
+ char* posixccs = regcomp_parse++;
+
+ while (regcomp_parse < regxend && *regcomp_parse != posixccc)
+ regcomp_parse++;
+ if (regcomp_parse == regxend)
+ /* Grandfather lone [:, [=, [. */
+ regcomp_parse = posixccs;
+ else {
+ regcomp_parse++; /* skip over the posixccc */
+ if (*regcomp_parse == ']') {
+ /* Not Implemented Yet.
+ * (POSIX Extended Character Classes, that is)
+ * The text between e.g. [: and :] would start
+ * at posixccs + 1 and stop at regcomp_parse - 2. */
+ if (dowarn && !SIZE_ONLY)
+ warn("Character class syntax [%c %c] is reserved for future extensions", posixccc, posixccc);
+ regcomp_parse++; /* skip over the ending ] */
+ }
+ }
+ }
+ if (Class == '\\') {
+ Class = UCHARAT(regcomp_parse++);
+ switch (Class) {
case 'w':
- if (regflags & PMf_LOCALE) {
- if (opnd != ®dummy)
+ if (!SIZE_ONLY) {
+ if (regflags & PMf_LOCALE)
*opnd |= ANYOF_ALNUML;
- }
- else {
- for (class = 0; class < 256; class++)
- if (isALNUM(class))
- regset(opnd, class);
+ else {
+ for (Class = 0; Class < 256; Class++)
+ if (isALNUM(Class))
+ ANYOF_SET(opnd, Class);
+ }
}
lastclass = 1234;
continue;
case 'W':
- if (regflags & PMf_LOCALE) {
- if (opnd != ®dummy)
+ if (!SIZE_ONLY) {
+ if (regflags & PMf_LOCALE)
*opnd |= ANYOF_NALNUML;
- }
- else {
- for (class = 0; class < 256; class++)
- if (!isALNUM(class))
- regset(opnd, class);
+ else {
+ for (Class = 0; Class < 256; Class++)
+ if (!isALNUM(Class))
+ ANYOF_SET(opnd, Class);
+ }
}
lastclass = 1234;
continue;
case 's':
- if (regflags & PMf_LOCALE) {
- if (opnd != ®dummy)
+ if (!SIZE_ONLY) {
+ if (regflags & PMf_LOCALE)
*opnd |= ANYOF_SPACEL;
- }
- else {
- for (class = 0; class < 256; class++)
- if (isSPACE(class))
- regset(opnd, class);
+ else {
+ for (Class = 0; Class < 256; Class++)
+ if (isSPACE(Class))
+ ANYOF_SET(opnd, Class);
+ }
}
lastclass = 1234;
continue;
case 'S':
- if (regflags & PMf_LOCALE) {
- if (opnd != ®dummy)
+ if (!SIZE_ONLY) {
+ if (regflags & PMf_LOCALE)
*opnd |= ANYOF_NSPACEL;
- }
- else {
- for (class = 0; class < 256; class++)
- if (!isSPACE(class))
- regset(opnd, class);
+ else {
+ for (Class = 0; Class < 256; Class++)
+ if (!isSPACE(Class))
+ ANYOF_SET(opnd, Class);
+ }
}
lastclass = 1234;
continue;
case 'd':
- for (class = '0'; class <= '9'; class++)
- regset(opnd, class);
+ if (!SIZE_ONLY) {
+ for (Class = '0'; Class <= '9'; Class++)
+ ANYOF_SET(opnd, Class);
+ }
lastclass = 1234;
continue;
case 'D':
- for (class = 0; class < '0'; class++)
- regset(opnd, class);
- for (class = '9' + 1; class < 256; class++)
- regset(opnd, class);
+ if (!SIZE_ONLY) {
+ for (Class = 0; Class < '0'; Class++)
+ ANYOF_SET(opnd, Class);
+ for (Class = '9' + 1; Class < 256; Class++)
+ ANYOF_SET(opnd, Class);
+ }
lastclass = 1234;
continue;
case 'n':
- class = '\n';
+ Class = '\n';
break;
case 'r':
- class = '\r';
+ Class = '\r';
break;
case 't':
- class = '\t';
+ Class = '\t';
break;
case 'f':
- class = '\f';
+ Class = '\f';
break;
case 'b':
- class = '\b';
+ Class = '\b';
break;
case 'e':
- class = '\033';
+ Class = '\033';
break;
case 'a':
- class = '\007';
+ Class = '\007';
break;
case 'x':
- class = scan_hex(regparse, 2, &numlen);
- regparse += numlen;
+ Class = scan_hex(regcomp_parse, 2, &numlen);
+ regcomp_parse += numlen;
break;
case 'c':
- class = UCHARAT(regparse++);
- class = toCTRL(class);
+ Class = UCHARAT(regcomp_parse++);
+ Class = toCTRL(Class);
break;
case '0': case '1': case '2': case '3': case '4':
case '5': case '6': case '7': case '8': case '9':
- class = scan_oct(--regparse, 3, &numlen);
- regparse += numlen;
+ Class = scan_oct(--regcomp_parse, 3, &numlen);
+ regcomp_parse += numlen;
break;
}
}
if (range) {
- if (lastclass > class)
+ if (lastclass > Class)
FAIL("invalid [] range in regexp");
range = 0;
}
else {
- lastclass = class;
- if (*regparse == '-' && regparse+1 < regxend &&
- regparse[1] != ']') {
- regparse++;
+ lastclass = Class;
+ if (*regcomp_parse == '-' && regcomp_parse+1 < regxend &&
+ regcomp_parse[1] != ']') {
+ regcomp_parse++;
range = 1;
continue; /* do it next time */
}
}
- for ( ; lastclass <= class; lastclass++)
- regset(opnd, lastclass);
- lastclass = class;
+ if (!SIZE_ONLY) {
+ for ( ; lastclass <= Class; lastclass++)
+ ANYOF_SET(opnd, lastclass);
+ }
+ lastclass = Class;
}
- if (*regparse != ']')
+ if (*regcomp_parse != ']')
FAIL("unmatched [] in regexp");
nextchar();
+ /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
+ if (!SIZE_ONLY && (*opnd & (0xFF ^ ANYOF_INVERT)) == ANYOF_FOLD) {
+ for (Class = 0; Class < 256; ++Class) {
+ if (ANYOF_TEST(opnd, Class)) {
+ I32 cf = fold[Class];
+ ANYOF_SET(opnd, cf);
+ }
+ }
+ *opnd &= ~ANYOF_FOLD;
+ }
+ /* optimize inverted simple patterns (e.g. [^a-z]) */
+ if (!SIZE_ONLY && (*opnd & 0xFF) == ANYOF_INVERT) {
+ for (Class = 0; Class < 32; ++Class)
+ opnd[1 + Class] ^= 0xFF;
+ *opnd = 0;
+ }
return ret;
}
-static char*
-nextchar()
+STATIC char*
+nextchar(void)
{
- char* retval = regparse++;
+ dTHR;
+ char* retval = regcomp_parse++;
for (;;) {
- if (*regparse == '(' && regparse[1] == '?' &&
- regparse[2] == '#') {
- while (*regparse && *regparse != ')')
- regparse++;
- regparse++;
+ if (*regcomp_parse == '(' && regcomp_parse[1] == '?' &&
+ regcomp_parse[2] == '#') {
+ while (*regcomp_parse && *regcomp_parse != ')')
+ regcomp_parse++;
+ regcomp_parse++;
continue;
}
if (regflags & PMf_EXTENDED) {
- if (isSPACE(*regparse)) {
- regparse++;
+ if (isSPACE(*regcomp_parse)) {
+ regcomp_parse++;
continue;
}
- else if (*regparse == '#') {
- while (*regparse && *regparse != '\n')
- regparse++;
- regparse++;
+ else if (*regcomp_parse == '#') {
+ while (*regcomp_parse && *regcomp_parse != '\n')
+ regcomp_parse++;
+ regcomp_parse++;
continue;
}
}
}
/*
-- regnode - emit a node
+- reg_node - emit a node
*/
-#ifdef CAN_PROTOTYPE
-static char * /* Location. */
-regnode(char op)
-#else
-static char * /* Location. */
-regnode(op)
-char op;
-#endif
+STATIC regnode * /* Location. */
+reg_node(U8 op)
{
- register char *ret;
- register char *ptr;
+ dTHR;
+ register regnode *ret;
+ register regnode *ptr;
ret = regcode;
- if (ret == ®dummy) {
-#ifdef REGALIGN
- if (!(regsize & 1))
- regsize++;
-#endif
- regsize += 3;
+ if (SIZE_ONLY) {
+ SIZE_ALIGN(regsize);
+ regsize += 1;
return(ret);
}
-#ifdef REGALIGN
-#ifndef lint
- if (!((long)ret & 1))
- *ret++ = 127;
-#endif
-#endif
+ NODE_ALIGN_FILL(ret);
ptr = ret;
- *ptr++ = op;
- *ptr++ = '\0'; /* Null "next" pointer. */
- *ptr++ = '\0';
+ FILL_ADVANCE_NODE(ptr, op);
regcode = ptr;
return(ret);
/*
- reganode - emit a node with an argument
*/
-#ifdef CAN_PROTOTYPE
-static char * /* Location. */
-reganode(char op, unsigned short arg)
-#else
-static char * /* Location. */
-reganode(op, arg)
-char op;
-unsigned short arg;
-#endif
+STATIC regnode * /* Location. */
+reganode(U8 op, U32 arg)
{
- register char *ret;
- register char *ptr;
+ dTHR;
+ register regnode *ret;
+ register regnode *ptr;
ret = regcode;
- if (ret == ®dummy) {
-#ifdef REGALIGN
- if (!(regsize & 1))
- regsize++;
-#endif
- regsize += 5;
+ if (SIZE_ONLY) {
+ SIZE_ALIGN(regsize);
+ regsize += 2;
return(ret);
}
-#ifdef REGALIGN
-#ifndef lint
- if (!((long)ret & 1))
- *ret++ = 127;
-#endif
-#endif
+ NODE_ALIGN_FILL(ret);
ptr = ret;
- *ptr++ = op;
- *ptr++ = '\0'; /* Null "next" pointer. */
- *ptr++ = '\0';
-#ifdef REGALIGN
- *(unsigned short *)(ret+3) = arg;
-#else
- ret[3] = arg >> 8; ret[4] = arg & 0377;
-#endif
- ptr += 2;
+ FILL_ADVANCE_NODE_ARG(ptr, op, arg);
regcode = ptr;
return(ret);
/*
- regc - emit (if appropriate) a byte of code
*/
-#ifdef CAN_PROTOTYPE
-static void
-regc(char b)
-#else
-static void
-regc(b)
-char b;
-#endif
+STATIC void
+regc(U8 b, char* s)
{
- if (regcode != ®dummy)
- *regcode++ = b;
- else
- regsize++;
+ dTHR;
+ if (!SIZE_ONLY)
+ *s = b;
}
/*
*
* Means relocating the operand.
*/
-#ifdef CAN_PROTOTYPE
-static void
-reginsert(char op, char *opnd)
-#else
-static void
-reginsert(op, opnd)
-char op;
-char *opnd;
-#endif
+STATIC void
+reginsert(U8 op, regnode *opnd)
{
- register char *src;
- register char *dst;
- register char *place;
- register int offset = (regkind[(U8)op] == CURLY ? 4 : 0);
-
- if (regcode == ®dummy) {
-#ifdef REGALIGN
- regsize += 4 + offset;
-#else
- regsize += 3 + offset;
-#endif
+ dTHR;
+ register regnode *src;
+ register regnode *dst;
+ register regnode *place;
+ register int offset = regarglen[(U8)op];
+
+/* (regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
+
+ if (SIZE_ONLY) {
+ regsize += NODE_STEP_REGNODE + offset;
return;
}
src = regcode;
-#ifdef REGALIGN
- regcode += 4 + offset;
-#else
- regcode += 3 + offset;
-#endif
+ regcode += NODE_STEP_REGNODE + offset;
dst = regcode;
while (src > opnd)
- *--dst = *--src;
+ StructCopy(--src, --dst, regnode);
place = opnd; /* Op node, where operand used to be. */
- *place++ = op;
- *place++ = '\0';
- *place++ = '\0';
- while (offset-- > 0)
- *place++ = '\0';
-#ifdef REGALIGN
- *place++ = '\177';
-#endif
+ src = NEXTOPER(place);
+ FILL_ADVANCE_NODE(place, op);
+ Zero(src, offset, regnode);
}
/*
-- regtail - set the next-pointer at the end of a node chain
+- regtail - set the next-pointer at the end of a node chain of p to val.
*/
-static void
-regtail(p, val)
-char *p;
-char *val;
+STATIC void
+regtail(regnode *p, regnode *val)
{
- register char *scan;
- register char *temp;
+ dTHR;
+ register regnode *scan;
+ register regnode *temp;
register I32 offset;
- if (p == ®dummy)
+ if (SIZE_ONLY)
return;
/* Find last node. */
scan = temp;
}
-#ifdef REGALIGN
- offset = val - scan;
-#ifndef lint
- *(short*)(scan+1) = offset;
-#else
- offset = offset;
-#endif
-#else
- if (OP(scan) == BACK)
- offset = scan - val;
- else
- offset = val - scan;
- *(scan+1) = (offset>>8)&0377;
- *(scan+2) = offset&0377;
-#endif
+ if (reg_off_by_arg[OP(scan)]) {
+ ARG_SET(scan, val - scan);
+ } else {
+ NEXT_OFF(scan) = val - scan;
+ }
}
/*
- regoptail - regtail on operand of first argument; nop if operandless
*/
-static void
-regoptail(p, val)
-char *p;
-char *val;
+STATIC void
+regoptail(regnode *p, regnode *val)
{
+ dTHR;
/* "Operandless" and "op != BRANCH" are synonymous in practice. */
- if (p == NULL || p == ®dummy || regkind[(U8)OP(p)] != BRANCH)
+ if (p == NULL || SIZE_ONLY)
+ return;
+ if (regkind[(U8)OP(p)] == BRANCH) {
+ regtail(NEXTOPER(p), val);
+ } else if ( regkind[(U8)OP(p)] == BRANCHJ) {
+ regtail(NEXTOPER(NEXTOPER(p)), val);
+ } else
return;
- regtail(NEXTOPER(p), val);
}
/*
- regcurly - a little FSA that accepts {\d+,?\d*}
*/
STATIC I32
-regcurly(s)
-register char *s;
+regcurly(register char *s)
{
if (*s++ != '{')
return FALSE;
return TRUE;
}
-#ifdef DEBUGGING
-/*
- - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
- */
-void
-regdump(r)
-regexp *r;
+STATIC regnode *
+dumpuntil(regnode *start, regnode *node, regnode *last, SV* sv, I32 l)
{
- register char *s;
+#ifdef DEBUGGING
register char op = EXACT; /* Arbitrary non-END op. */
- register char *next;
-
-
- s = r->program + 1;
- while (op != END) { /* While that wasn't END last time... */
-#ifdef REGALIGN
- if (!((long)s & 1))
- s++;
-#endif
- op = OP(s);
- PerlIO_printf(Perl_debug_log, "%2d%s", s-r->program, regprop(s)); /* Where, what. */
- next = regnext(s);
- s += regarglen[(U8)op];
+ register regnode *next, *onode;
+
+ while (op != END && (!last || node < last)) {
+ /* While that wasn't END last time... */
+
+ NODE_ALIGN(node);
+ op = OP(node);
+ if (op == CLOSE)
+ l--;
+ next = regnext(node);
+ /* Where, what. */
+ if (OP(node) == OPTIMIZED)
+ goto after_print;
+ regprop(sv, node);
+ PerlIO_printf(Perl_debug_log, "%4d:%*s%s", node - start,
+ 2*l + 1, "", SvPVX(sv));
if (next == NULL) /* Next ptr. */
PerlIO_printf(Perl_debug_log, "(0)");
else
- PerlIO_printf(Perl_debug_log, "(%d)", (s-r->program)+(next-s));
- s += 3;
- if (op == ANYOF) {
- s += 33;
- }
- if (regkind[(U8)op] == EXACT) {
- /* Literal string, where present. */
- s++;
- (void)PerlIO_putc(Perl_debug_log, ' ');
- (void)PerlIO_putc(Perl_debug_log, '<');
- while (*s != '\0') {
- (void)PerlIO_putc(Perl_debug_log,*s);
- s++;
- }
- (void)PerlIO_putc(Perl_debug_log, '>');
- s++;
- }
+ PerlIO_printf(Perl_debug_log, "(%d)", next - start);
(void)PerlIO_putc(Perl_debug_log, '\n');
+ after_print:
+ if (regkind[(U8)op] == BRANCHJ) {
+ register regnode *nnode = (OP(next) == LONGJMP
+ ? regnext(next)
+ : next);
+ if (last && nnode > last)
+ nnode = last;
+ node = dumpuntil(start, NEXTOPER(NEXTOPER(node)), nnode, sv, l + 1);
+ } else if (regkind[(U8)op] == BRANCH) {
+ node = dumpuntil(start, NEXTOPER(node), next, sv, l + 1);
+ } else if ( op == CURLY) { /* `next' might be very big: optimizer */
+ node = dumpuntil(start, NEXTOPER(node) + EXTRA_STEP_2ARGS,
+ NEXTOPER(node) + EXTRA_STEP_2ARGS + 1, sv, l + 1);
+ } else if (regkind[(U8)op] == CURLY && op != CURLYX) {
+ node = dumpuntil(start, NEXTOPER(node) + EXTRA_STEP_2ARGS,
+ next, sv, l + 1);
+ } else if ( op == PLUS || op == STAR) {
+ node = dumpuntil(start, NEXTOPER(node), NEXTOPER(node) + 1, sv, l + 1);
+ } else if (op == ANYOF) {
+ node = NEXTOPER(node);
+ node += ANY_SKIP;
+ } else if (regkind[(U8)op] == EXACT) {
+ /* Literal string, where present. */
+ node += ((*OPERAND(node)) + 2 + sizeof(regnode) - 1) / sizeof(regnode);
+ node = NEXTOPER(node);
+ } else {
+ node = NEXTOPER(node);
+ node += regarglen[(U8)op];
+ }
+ if (op == CURLYX || op == OPEN)
+ l++;
+ else if (op == WHILEM)
+ l--;
}
+#endif /* DEBUGGING */
+ return node;
+}
+
+/*
+ - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
+ */
+void
+regdump(regexp *r)
+{
+#ifdef DEBUGGING
+ dTHR;
+ SV *sv = sv_newmortal();
+
+ (void)dumpuntil(r->program, r->program + 1, NULL, sv, 0);
/* Header fields of interest. */
- if (r->regstart)
- PerlIO_printf(Perl_debug_log, "start `%s' ", SvPVX(r->regstart));
- if (r->regstclass)
- PerlIO_printf(Perl_debug_log, "stclass `%s' ", regprop(r->regstclass));
+ if (r->anchored_substr)
+ PerlIO_printf(Perl_debug_log, "anchored `%s%s%s'%s at %d ",
+ colors[0],
+ SvPVX(r->anchored_substr),
+ colors[1],
+ SvTAIL(r->anchored_substr) ? "$" : "",
+ r->anchored_offset);
+ if (r->float_substr)
+ PerlIO_printf(Perl_debug_log, "floating `%s%s%s'%s at %d..%u ",
+ colors[0],
+ SvPVX(r->float_substr),
+ colors[1],
+ SvTAIL(r->float_substr) ? "$" : "",
+ r->float_min_offset, r->float_max_offset);
+ if (r->check_substr)
+ PerlIO_printf(Perl_debug_log,
+ r->check_substr == r->float_substr
+ ? "(checking floating" : "(checking anchored");
+ if (r->reganch & ROPT_NOSCAN)
+ PerlIO_printf(Perl_debug_log, " noscan");
+ if (r->reganch & ROPT_CHECK_ALL)
+ PerlIO_printf(Perl_debug_log, " isall");
+ if (r->check_substr)
+ PerlIO_printf(Perl_debug_log, ") ");
+
+ if (r->regstclass) {
+ regprop(sv, r->regstclass);
+ PerlIO_printf(Perl_debug_log, "stclass `%s' ", SvPVX(sv));
+ }
if (r->reganch & ROPT_ANCH) {
PerlIO_printf(Perl_debug_log, "anchored");
if (r->reganch & ROPT_ANCH_BOL)
PerlIO_printf(Perl_debug_log, "(BOL)");
+ if (r->reganch & ROPT_ANCH_MBOL)
+ PerlIO_printf(Perl_debug_log, "(MBOL)");
if (r->reganch & ROPT_ANCH_GPOS)
PerlIO_printf(Perl_debug_log, "(GPOS)");
PerlIO_putc(Perl_debug_log, ' ');
}
+ if (r->reganch & ROPT_GPOS_SEEN)
+ PerlIO_printf(Perl_debug_log, "GPOS ");
if (r->reganch & ROPT_SKIP)
PerlIO_printf(Perl_debug_log, "plus ");
if (r->reganch & ROPT_IMPLICIT)
PerlIO_printf(Perl_debug_log, "implicit ");
- if (r->regmust != NULL)
- PerlIO_printf(Perl_debug_log, "must have \"%s\" back %ld ", SvPVX(r->regmust),
- (long) r->regback);
PerlIO_printf(Perl_debug_log, "minlen %ld ", (long) r->minlen);
+ if (r->reganch & ROPT_EVAL_SEEN)
+ PerlIO_printf(Perl_debug_log, "with eval ");
PerlIO_printf(Perl_debug_log, "\n");
+#endif /* DEBUGGING */
}
/*
- regprop - printable representation of opcode
*/
-char *
-regprop(op)
-char *op;
+void
+regprop(SV *sv, regnode *o)
{
+#ifdef DEBUGGING
+ dTHR;
register char *p = 0;
- (void) strcpy(buf, ":");
-
- switch (OP(op)) {
+ sv_setpvn(sv, "", 0);
+ switch (OP(o)) {
case BOL:
p = "BOL";
break;
case EOL:
p = "EOL";
break;
+ case EOS:
+ p = "EOS";
+ break;
case MEOL:
p = "MEOL";
break;
p = "BRANCH";
break;
case EXACT:
- p = "EXACT";
+ sv_catpvf(sv, "EXACT <%s%s%s>", colors[0], OPERAND(o) + 1, colors[1]);
break;
case EXACTF:
- p = "EXACTF";
+ sv_catpvf(sv, "EXACTF <%s%s%s>", colors[0], OPERAND(o) + 1, colors[1]);
break;
case EXACTFL:
- p = "EXACTFL";
+ sv_catpvf(sv, "EXACTFL <%s%s%s>", colors[0], OPERAND(o) + 1, colors[1]);
break;
case NOTHING:
p = "NOTHING";
break;
+ case TAIL:
+ p = "TAIL";
+ break;
case BACK:
p = "BACK";
break;
p = "NBOUNDL";
break;
case CURLY:
- (void)sprintf(buf+strlen(buf), "CURLY {%d,%d}", ARG1(op),ARG2(op));
- p = NULL;
+ sv_catpvf(sv, "CURLY {%d,%d}", ARG1(o), ARG2(o));
+ break;
+ case CURLYM:
+ sv_catpvf(sv, "CURLYM[%d] {%d,%d}", o->flags, ARG1(o), ARG2(o));
+ break;
+ case CURLYN:
+ sv_catpvf(sv, "CURLYN[%d] {%d,%d}", o->flags, ARG1(o), ARG2(o));
break;
case CURLYX:
- (void)sprintf(buf+strlen(buf), "CURLYX {%d,%d}", ARG1(op),ARG2(op));
- p = NULL;
+ sv_catpvf(sv, "CURLYX {%d,%d}", ARG1(o), ARG2(o));
break;
case REF:
- (void)sprintf(buf+strlen(buf), "REF%d", ARG1(op));
- p = NULL;
+ sv_catpvf(sv, "REF%d", ARG(o));
+ break;
+ case REFF:
+ sv_catpvf(sv, "REFF%d", ARG(o));
+ break;
+ case REFFL:
+ sv_catpvf(sv, "REFFL%d", ARG(o));
break;
case OPEN:
- (void)sprintf(buf+strlen(buf), "OPEN%d", ARG1(op));
- p = NULL;
+ sv_catpvf(sv, "OPEN%d", ARG(o));
break;
case CLOSE:
- (void)sprintf(buf+strlen(buf), "CLOSE%d", ARG1(op));
+ sv_catpvf(sv, "CLOSE%d", ARG(o));
p = NULL;
break;
case STAR:
p = "GPOS";
break;
case UNLESSM:
- p = "UNLESSM";
+ sv_catpvf(sv, "UNLESSM[-%d]", o->flags);
break;
case IFMATCH:
- p = "IFMATCH";
+ sv_catpvf(sv, "IFMATCH[-%d]", o->flags);
break;
case SUCCEED:
p = "SUCCEED";
case NSPACEL:
p = "NSPACEL";
break;
+ case EVAL:
+ p = "EVAL";
+ break;
+ case LONGJMP:
+ p = "LONGJMP";
+ break;
+ case BRANCHJ:
+ p = "BRANCHJ";
+ break;
+ case IFTHEN:
+ p = "IFTHEN";
+ break;
+ case GROUPP:
+ sv_catpvf(sv, "GROUPP%d", ARG(o));
+ break;
+ case LOGICAL:
+ p = "LOGICAL";
+ break;
+ case SUSPEND:
+ p = "SUSPEND";
+ break;
+ case RENUM:
+ p = "RENUM";
+ break;
+ case OPTIMIZED:
+ p = "OPTIMIZED";
+ break;
default:
FAIL("corrupted regexp opcode");
}
- if (p != NULL)
- (void) strcat(buf, p);
- return(buf);
+ if (p)
+ sv_catpv(sv, p);
+#endif /* DEBUGGING */
}
-#endif /* DEBUGGING */
void
-pregfree(r)
-struct regexp *r;
+pregfree(struct regexp *r)
{
- if (!r)
+ dTHR;
+ if (!r || (--r->refcnt > 0))
return;
- if (r->precomp) {
+ if (r->precomp)
Safefree(r->precomp);
- r->precomp = Nullch;
- }
- if (r->subbase) {
+ if (r->subbase)
Safefree(r->subbase);
- r->subbase = Nullch;
- }
- if (r->regmust) {
- SvREFCNT_dec(r->regmust);
- r->regmust = Nullsv;
+ if (r->substrs) {
+ if (r->anchored_substr)
+ SvREFCNT_dec(r->anchored_substr);
+ if (r->float_substr)
+ SvREFCNT_dec(r->float_substr);
+ Safefree(r->substrs);
}
- if (r->regstart) {
- SvREFCNT_dec(r->regstart);
- r->regstart = Nullsv;
+ if (r->data) {
+ int n = r->data->count;
+ while (--n >= 0) {
+ switch (r->data->what[n]) {
+ case 's':
+ SvREFCNT_dec((SV*)r->data->data[n]);
+ break;
+ case 'o':
+ op_free((OP_4tree*)r->data->data[n]);
+ break;
+ case 'n':
+ break;
+ default:
+ FAIL2("panic: regfree data code '%c'", r->data->what[n]);
+ }
+ }
+ Safefree(r->data->what);
+ Safefree(r->data);
}
Safefree(r->startp);
Safefree(r->endp);
Safefree(r);
}
+
+/*
+ - regnext - dig the "next" pointer out of a node
+ *
+ * [Note, when REGALIGN is defined there are two places in regmatch()
+ * that bypass this code for speed.]
+ */
+regnode *
+regnext(register regnode *p)
+{
+ dTHR;
+ register I32 offset;
+
+ if (p == ®dummy)
+ return(NULL);
+
+ offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
+ if (offset == 0)
+ return(NULL);
+
+ return(p+offset);
+}
+
+STATIC void
+re_croak2(const char* pat1,const char* pat2,...)
+{
+ va_list args;
+ STRLEN l1 = strlen(pat1);
+ STRLEN l2 = strlen(pat2);
+ char buf[512];
+ char *message;
+
+ if (l1 > 510)
+ l1 = 510;
+ if (l1 + l2 > 510)
+ l2 = 510 - l1;
+ Copy(pat1, buf, l1 , char);
+ Copy(pat2, buf + l1, l2 , char);
+ buf[l1 + l2] = '\n';
+ buf[l1 + l2 + 1] = '\0';
+ va_start(args, pat2);
+ message = mess(buf, &args);
+ va_end(args);
+ l1 = strlen(message);
+ if (l1 > 512)
+ l1 = 512;
+ Copy(message, buf, l1 , char);
+ buf[l1] = '\0'; /* Overwrite \n */
+ croak("%s", buf);
+}
https://perl5.git.perl.org / perl5.git / blobdiff