I32 sawback; /* Did we see \1, ...? */
U32 seen;
I32 size; /* Code size. */
- I32 npar; /* () count. */
+ I32 npar; /* Capture buffer count, (OPEN). */
+ I32 cpar; /* Capture buffer count, (CLOSE). */
+ I32 nestroot; /* root parens we are in - used by accept */
I32 extralen;
I32 seen_zerolen;
I32 seen_evals;
- regnode **parens; /* offsets of each paren */
+ regnode **open_parens; /* pointers to open parens */
+ regnode **close_parens; /* pointers to close parens */
+ regnode *opend; /* END node in program */
I32 utf8;
HV *charnames; /* cache of named sequences */
HV *paren_names; /* Paren names */
+ regnode **recurse; /* Recurse regops */
+ I32 recurse_count; /* Number of recurse regops */
#if ADD_TO_REGEXEC
char *starttry; /* -Dr: where regtry was called. */
#define RExC_starttry (pRExC_state->starttry)
#define RExC_seen (pRExC_state->seen)
#define RExC_size (pRExC_state->size)
#define RExC_npar (pRExC_state->npar)
+#define RExC_cpar (pRExC_state->cpar)
+#define RExC_nestroot (pRExC_state->nestroot)
#define RExC_extralen (pRExC_state->extralen)
#define RExC_seen_zerolen (pRExC_state->seen_zerolen)
#define RExC_seen_evals (pRExC_state->seen_evals)
#define RExC_utf8 (pRExC_state->utf8)
#define RExC_charnames (pRExC_state->charnames)
-#define RExC_parens (pRExC_state->parens)
+#define RExC_open_parens (pRExC_state->open_parens)
+#define RExC_close_parens (pRExC_state->close_parens)
+#define RExC_opend (pRExC_state->opend)
#define RExC_paren_names (pRExC_state->paren_names)
+#define RExC_recurse (pRExC_state->recurse)
+#define RExC_recurse_count (pRExC_state->recurse_count)
#define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
#define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
#endif
+
+#define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
+#define PBITVAL(paren) (1 << ((paren) & 7))
+#define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
+#define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
+#define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
+
+
/* About scan_data_t.
During optimisation we recurse through the regexp program performing
#define SCF_WHILEM_VISITED_POS 0x2000
#define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
-
+#define SCF_SEEN_ACCEPT 0x8000
#define UTF (RExC_utf8 != 0)
#define LOC ((RExC_flags & PMf_LOCALE) != 0)
#define Set_Node_Offset_To_R(node,byte) STMT_START { \
if (! SIZE_ONLY) { \
MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
- __LINE__, (node), (int)(byte))); \
+ __LINE__, (int)(node), (int)(byte))); \
if((node) < 0) { \
Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
} else { \
#define DEBUG_STUDYDATA(data,depth) \
DEBUG_OPTIMISE_MORE_r(if(data){ \
PerlIO_printf(Perl_debug_log, \
- "%*s"/* Len:%"IVdf"/%"IVdf" */" Pos:%"IVdf"/%"IVdf \
+ "%*s"/* Len:%"IVdf"/%"IVdf" */"Pos:%"IVdf"/%"IVdf \
" Flags: %"IVdf" Whilem_c: %"IVdf" Lcp: %"IVdf" ", \
(int)(depth)*2, "", \
(IV)((data)->pos_min), \
S_cl_and(struct regnode_charclass_class *cl,
const struct regnode_charclass_class *and_with)
{
+
+ assert(and_with->type == ANYOF);
if (!(and_with->flags & ANYOF_CLASS)
&& !(cl->flags & ANYOF_CLASS)
&& (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
PerlIO_printf( Perl_debug_log, "\n");
- for( state = 1 ; state < trie->laststate ; state++ ) {
+ for( state = 1 ; state < trie->statecount ; state++ ) {
const U32 base = trie->states[ state ].trans.base;
PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
(SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
PERL_PV_ESCAPE_FIRSTCHAR
) ,
- TRIE_LIST_ITEM(state,charid).forid,
- (UV)TRIE_LIST_ITEM(state,charid).newstate
- );
- }
+ TRIE_LIST_ITEM(state,charid).forid,
+ (UV)TRIE_LIST_ITEM(state,charid).newstate
+ );
+ if (!(charid % 10))
+ PerlIO_printf(Perl_debug_log, "\n%*s| ",
+ (int)((depth * 2) + 14), "");
+ }
}
PerlIO_printf( Perl_debug_log, "\n");
}
*(d++) = uv;
*/
-#define TRIE_STORE_REVCHAR \
+#define TRIE_STORE_REVCHAR \
STMT_START { \
- SV *tmp = Perl_newSVpvf_nocontext( "%c", (int)uvc ); \
+ SV *tmp = newSVpvs(""); \
if (UTF) SvUTF8_on(tmp); \
+ Perl_sv_catpvf( aTHX_ tmp, "%c", (int)uvc ); \
av_push( TRIE_REVCHARMAP(trie), tmp ); \
} STMT_END
#define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
- TRIE_LIST_LEN( state ) *= 2; \
- Renew( trie->states[ state ].trans.list, \
- TRIE_LIST_LEN( state ), reg_trie_trans_le ); \
+ U32 ging = TRIE_LIST_LEN( state ) *= 2; \
+ Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
} \
TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
if ( noper_next < tail ) { \
if (!trie->jump) \
Newxz( trie->jump, word_count + 1, U16); \
- trie->jump[curword] = (U16)(tail - noper_next); \
+ trie->jump[curword] = (U16)(noper_next - convert); \
if (!jumper) \
jumper = noper_next; \
if (!nextbranch) \
U32 next_alloc = 0;
regnode *jumper = NULL;
regnode *nextbranch = NULL;
+ regnode *convert = NULL;
/* we just use folder as a flag in utf8 */
const U8 * const folder = ( flags == EXACTF
? PL_fold
REG_NODE_NUM(last), REG_NODE_NUM(tail),
(int)depth);
});
+
+ /* Find the node we are going to overwrite */
+ if ( first == startbranch && OP( last ) != BRANCH ) {
+ /* whole branch chain */
+ convert = first;
+ } else {
+ /* branch sub-chain */
+ convert = NEXTOPER( first );
+ }
+
/* -- First loop and Setup --
We first traverse the branches and scan each word to determine if it
STRLEN transcount = 1;
+ DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
+ "%*sCompiling trie using list compiler\n",
+ (int)depth * 2 + 2, ""));
+
Newxz( trie->states, TRIE_CHARCOUNT(trie) + 2, reg_trie_state );
TRIE_LIST_NEW(1);
next_alloc = 2;
} /* end second pass */
- trie->laststate = next_alloc;
+ /* next alloc is the NEXT state to be allocated */
+ trie->statecount = next_alloc;
Renew( trie->states, next_alloc, reg_trie_state );
/* and now dump it out before we compress it */
DEBUG_TRIE_COMPILE_MORE_r(
dump_trie_interim_list(trie,next_alloc,depth+1)
- );
+ );
Newxz( trie->trans, transcount ,reg_trie_trans );
{
use TRIE_NODENUM() to convert.
*/
-
+ DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
+ "%*sCompiling trie using table compiler\n",
+ (int)depth * 2 + 2, ""));
Newxz( trie->trans, ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1,
reg_trie_trans );
const U32 laststate = TRIE_NODENUM( next_alloc );
U32 state, charid;
U32 pos = 0, zp=0;
- trie->laststate = laststate;
+ trie->statecount = laststate;
for ( state = 1 ; state < laststate ; state++ ) {
U8 flag = 0;
}
}
trie->lasttrans = pos + 1;
- Renew( trie->states, laststate + 1, reg_trie_state);
+ Renew( trie->states, laststate, reg_trie_state);
DEBUG_TRIE_COMPILE_MORE_r(
PerlIO_printf( Perl_debug_log,
"%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
} /* end table compress */
}
+ DEBUG_TRIE_COMPILE_MORE_r(
+ PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
+ (int)depth * 2 + 2, "",
+ (UV)trie->statecount,
+ (UV)trie->lasttrans)
+ );
/* resize the trans array to remove unused space */
Renew( trie->trans, trie->lasttrans, reg_trie_trans);
);
{ /* Modify the program and insert the new TRIE node*/
- regnode *convert;
U8 nodetype =(U8)(flags & 0xFF);
char *str=NULL;
the whole branch sequence, including the first.
*/
/* Find the node we are going to overwrite */
- if ( first == startbranch && OP( last ) != BRANCH ) {
- /* whole branch chain */
- convert = first;
- DEBUG_r({
- const regnode *nop = NEXTOPER( convert );
- mjd_offset= Node_Offset((nop));
- mjd_nodelen= Node_Length((nop));
- });
- } else {
+ if ( first != startbranch || OP( last ) == BRANCH ) {
/* branch sub-chain */
- convert = NEXTOPER( first );
NEXT_OFF( first ) = (U16)(last - first);
DEBUG_r({
mjd_offset= Node_Offset((convert));
mjd_nodelen= Node_Length((convert));
});
+ /* whole branch chain */
+ } else {
+ DEBUG_r({
+ const regnode *nop = NEXTOPER( convert );
+ mjd_offset= Node_Offset((nop));
+ mjd_nodelen= Node_Length((nop));
+ });
}
+
DEBUG_OPTIMISE_r(
PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
(int)depth * 2 + 2, "",
trie->startstate= 1;
if ( trie->bitmap && !trie->widecharmap && !trie->jump ) {
U32 state;
- DEBUG_OPTIMISE_r(
- PerlIO_printf(Perl_debug_log, "%*sLaststate:%"UVuf"\n",
- (int)depth * 2 + 2, "",
- (UV)trie->laststate)
- );
- for ( state = 1 ; state < trie->laststate-1 ; state++ ) {
+ for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
U32 ofs = 0;
I32 idx = -1;
U32 count = 0;
jump[0], which is otherwise unused by the jump logic.
We use this when dumping a trie and during optimisation. */
if (trie->jump)
- trie->jump[0] = (U16)(tail - nextbranch);
+ trie->jump[0] = (U16)(nextbranch - convert);
/* XXXX */
if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
reg_trie_data *trie=(reg_trie_data *)RExC_rx->data->data[ARG(source)];
U32 *q;
const U32 ucharcount = trie->uniquecharcount;
- const U32 numstates = trie->laststate;
+ const U32 numstates = trie->statecount;
const U32 ubound = trie->lasttrans + ucharcount;
U32 q_read = 0;
U32 q_write = 0;
RExC_rx->data->data[ data_slot ] = (void*)aho;
aho->trie=trie;
aho->states=(reg_trie_state *)savepvn((const char*)trie->states,
- (trie->laststate+1)*sizeof(reg_trie_state));
+ numstates * sizeof(reg_trie_state));
Newxz( q, numstates, U32);
Newxz( aho->fail, numstates, U32 );
aho->refcount = 1;
*/
fail[ 0 ] = fail[ 1 ] = 0;
DEBUG_TRIE_COMPILE_r({
- PerlIO_printf(Perl_debug_log, "%*sStclass Failtable: 0", (int)(depth * 2), "");
+ PerlIO_printf(Perl_debug_log,
+ "%*sStclass Failtable (%"UVuf" states): 0",
+ (int)(depth * 2), "", (UV)numstates
+ );
for( q_read=1; q_read<numstates; q_read++ ) {
PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
}
#endif
#define DEBUG_PEEP(str,scan,depth) \
- DEBUG_OPTIMISE_r({ \
+ DEBUG_OPTIMISE_r({if (scan){ \
SV * const mysv=sv_newmortal(); \
regnode *Next = regnext(scan); \
regprop(RExC_rx, mysv, scan); \
- PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s [%d]\n", \
+ PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
(int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
Next ? (REG_NODE_NUM(Next)) : 0 ); \
- });
+ }});
/* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
to the position after last scanned or to NULL. */
-
+#define INIT_AND_WITHP \
+ assert(!and_withp); \
+ Newx(and_withp,1,struct regnode_charclass_class); \
+ SAVEFREEPV(and_withp)
+
+/* this is a chain of data about sub patterns we are processing that
+ need to be handled seperately/specially in study_chunk. Its so
+ we can simulate recursion without losing state. */
+struct scan_frame;
+typedef struct scan_frame {
+ regnode *last; /* last node to process in this frame */
+ regnode *next; /* next node to process when last is reached */
+ struct scan_frame *prev; /*previous frame*/
+ I32 stop; /* what stopparen do we use */
+} scan_frame;
STATIC I32
-S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
+S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
I32 *minlenp, I32 *deltap,
- regnode *last, scan_data_t *data, U32 flags, U32 depth)
+ regnode *last,
+ scan_data_t *data,
+ I32 stopparen,
+ U8* recursed,
+ struct regnode_charclass_class *and_withp,
+ U32 flags, U32 depth)
/* scanp: Start here (read-write). */
/* deltap: Write maxlen-minlen here. */
/* last: Stop before this one. */
+ /* data: string data about the pattern */
+ /* stopparen: treat close N as END */
+ /* recursed: which subroutines have we recursed into */
+ /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
{
dVAR;
I32 min = 0, pars = 0, code;
int is_inf_internal = 0; /* The studied chunk is infinite */
I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
scan_data_t data_fake;
- struct regnode_charclass_class and_with; /* Valid if flags & SCF_DO_STCLASS_OR */
SV *re_trie_maxbuff = NULL;
regnode *first_non_open = scan;
-
-
+ I32 stopmin = I32_MAX;
+ scan_frame last_frame= { last, NULL, NULL, stopparen };
+ scan_frame *frame=&last_frame;
+
GET_RE_DEBUG_FLAGS_DECL;
+
#ifdef DEBUGGING
- StructCopy(&zero_scan_data, &data_fake, scan_data_t);
+ StructCopy(&zero_scan_data, &data_fake, scan_data_t);
#endif
+
if ( depth == 0 ) {
- while (first_non_open && OP(first_non_open) == OPEN)
+ while (first_non_open && OP(first_non_open) == OPEN)
first_non_open=regnext(first_non_open);
}
+ while (frame) {
+
+ DEBUG_PEEP("FBEG",scan,depth);
+ while ( scan && OP(scan) != END && scan < frame->last ) {
+ /* Peephole optimizer: */
+ DEBUG_STUDYDATA(data,depth);
+ DEBUG_PEEP("Peep",scan,depth);
+ JOIN_EXACT(scan,&min,0);
+
+ /* Follow the next-chain of the current node and optimize
+ away all the NOTHINGs from it. */
+ if (OP(scan) != CURLYX) {
+ const int max = (reg_off_by_arg[OP(scan)]
+ ? I32_MAX
+ /* I32 may be smaller than U16 on CRAYs! */
+ : (I32_MAX < U16_MAX ? 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))
+ && ((PL_regkind[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;
+ }
- while (scan && OP(scan) != END && scan < last) {
- /* Peephole optimizer: */
- DEBUG_STUDYDATA(data,depth);
- DEBUG_PEEP("Peep",scan,depth);
- JOIN_EXACT(scan,&min,0);
-
- /* Follow the next-chain of the current node and optimize
- away all the NOTHINGs from it. */
- if (OP(scan) != CURLYX) {
- const int max = (reg_off_by_arg[OP(scan)]
- ? I32_MAX
- /* I32 may be smaller than U16 on CRAYs! */
- : (I32_MAX < U16_MAX ? 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))
- && ((PL_regkind[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;
- }
-
-
-
- /* The principal pseudo-switch. Cannot be a switch, since we
- look into several different things. */
- if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
- || OP(scan) == IFTHEN || OP(scan) == SUSPEND) {
- next = regnext(scan);
- code = OP(scan);
- /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
-
- if (OP(next) == code || code == IFTHEN || code == SUSPEND) {
- /* NOTE - There is similar code to this block below for handling
- TRIE nodes on a re-study. If you change stuff here check there
- too. */
- I32 max1 = 0, min1 = I32_MAX, num = 0;
- struct regnode_charclass_class accum;
- regnode * const startbranch=scan;
-
- if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
- scan_commit(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
- if (flags & SCF_DO_STCLASS)
- cl_init_zero(pRExC_state, &accum);
+ /* The principal pseudo-switch. Cannot be a switch, since we
+ look into several different things. */
+ if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
+ || OP(scan) == IFTHEN) {
+ next = regnext(scan);
+ code = OP(scan);
+ /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
+
+ if (OP(next) == code || code == IFTHEN) {
+ /* NOTE - There is similar code to this block below for handling
+ TRIE nodes on a re-study. If you change stuff here check there
+ too. */
+ I32 max1 = 0, min1 = I32_MAX, num = 0;
+ struct regnode_charclass_class accum;
+ regnode * const startbranch=scan;
+
+ if (flags & SCF_DO_SUBSTR)
+ scan_commit(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
+ if (flags & SCF_DO_STCLASS)
+ cl_init_zero(pRExC_state, &accum);
- while (OP(scan) == code) {
- I32 deltanext, minnext, f = 0, fake;
- struct regnode_charclass_class this_class;
+ while (OP(scan) == code) {
+ I32 deltanext, minnext, f = 0, fake;
+ struct regnode_charclass_class this_class;
- num++;
- data_fake.flags = 0;
- if (data) {
- data_fake.whilem_c = data->whilem_c;
- data_fake.last_closep = data->last_closep;
- }
- else
- data_fake.last_closep = &fake;
- next = regnext(scan);
- scan = NEXTOPER(scan);
- if (code != BRANCH)
+ num++;
+ data_fake.flags = 0;
+ if (data) {
+ data_fake.whilem_c = data->whilem_c;
+ data_fake.last_closep = data->last_closep;
+ }
+ else
+ data_fake.last_closep = &fake;
+ next = regnext(scan);
scan = NEXTOPER(scan);
- if (flags & SCF_DO_STCLASS) {
- cl_init(pRExC_state, &this_class);
- data_fake.start_class = &this_class;
- f = SCF_DO_STCLASS_AND;
- }
- if (flags & SCF_WHILEM_VISITED_POS)
- f |= SCF_WHILEM_VISITED_POS;
-
- /* we suppose the run is continuous, last=next...*/
- minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
- next, &data_fake, f,depth+1);
- if (min1 > minnext)
- min1 = minnext;
- if (max1 < minnext + deltanext)
- max1 = minnext + deltanext;
- if (deltanext == I32_MAX)
- is_inf = is_inf_internal = 1;
- scan = next;
- if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
- pars++;
- if (data) {
- if (data_fake.flags & SF_HAS_EVAL)
- data->flags |= SF_HAS_EVAL;
- data->whilem_c = data_fake.whilem_c;
+ if (code != BRANCH)
+ scan = NEXTOPER(scan);
+ if (flags & SCF_DO_STCLASS) {
+ cl_init(pRExC_state, &this_class);
+ data_fake.start_class = &this_class;
+ f = SCF_DO_STCLASS_AND;
+ }
+ if (flags & SCF_WHILEM_VISITED_POS)
+ f |= SCF_WHILEM_VISITED_POS;
+
+ /* we suppose the run is continuous, last=next...*/
+ minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
+ next, &data_fake,
+ stopparen, recursed, NULL, f,depth+1);
+ if (min1 > minnext)
+ min1 = minnext;
+ if (max1 < minnext + deltanext)
+ max1 = minnext + deltanext;
+ if (deltanext == I32_MAX)
+ is_inf = is_inf_internal = 1;
+ scan = next;
+ if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
+ pars++;
+ if (data_fake.flags & SCF_SEEN_ACCEPT) {
+ if ( stopmin > minnext)
+ stopmin = min + min1;
+ flags &= ~SCF_DO_SUBSTR;
+ if (data)
+ data->flags |= SCF_SEEN_ACCEPT;
+ }
+ if (data) {
+ if (data_fake.flags & SF_HAS_EVAL)
+ data->flags |= SF_HAS_EVAL;
+ data->whilem_c = data_fake.whilem_c;
+ }
+ if (flags & SCF_DO_STCLASS)
+ cl_or(pRExC_state, &accum, &this_class);
}
- if (flags & SCF_DO_STCLASS)
- cl_or(pRExC_state, &accum, &this_class);
- 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;
- if (flags & SCF_DO_STCLASS_OR) {
- cl_or(pRExC_state, data->start_class, &accum);
- if (min1) {
- cl_and(data->start_class, &and_with);
- flags &= ~SCF_DO_STCLASS;
+ 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);
}
- }
- else if (flags & SCF_DO_STCLASS_AND) {
- if (min1) {
- cl_and(data->start_class, &accum);
- flags &= ~SCF_DO_STCLASS;
+ min += min1;
+ delta += max1 - min1;
+ if (flags & SCF_DO_STCLASS_OR) {
+ cl_or(pRExC_state, data->start_class, &accum);
+ if (min1) {
+ cl_and(data->start_class, and_withp);
+ flags &= ~SCF_DO_STCLASS;
+ }
}
- else {
- /* Switch to OR mode: cache the old value of
- * data->start_class */
- StructCopy(data->start_class, &and_with,
- struct regnode_charclass_class);
- flags &= ~SCF_DO_STCLASS_AND;
- StructCopy(&accum, data->start_class,
- struct regnode_charclass_class);
- flags |= SCF_DO_STCLASS_OR;
- data->start_class->flags |= ANYOF_EOS;
+ else if (flags & SCF_DO_STCLASS_AND) {
+ if (min1) {
+ cl_and(data->start_class, &accum);
+ flags &= ~SCF_DO_STCLASS;
+ }
+ else {
+ /* Switch to OR mode: cache the old value of
+ * data->start_class */
+ INIT_AND_WITHP;
+ StructCopy(data->start_class, and_withp,
+ struct regnode_charclass_class);
+ flags &= ~SCF_DO_STCLASS_AND;
+ StructCopy(&accum, data->start_class,
+ struct regnode_charclass_class);
+ flags |= SCF_DO_STCLASS_OR;
+ data->start_class->flags |= ANYOF_EOS;
+ }
}
- }
- if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
- /* demq.
+ if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
+ /* demq.
- Assuming this was/is a branch we are dealing with: 'scan' now
- points at the item that follows the branch sequence, whatever
- it is. We now start at the beginning of the sequence and look
- for subsequences of
+ Assuming this was/is a branch we are dealing with: 'scan' now
+ points at the item that follows the branch sequence, whatever
+ it is. We now start at the beginning of the sequence and look
+ for subsequences of
- BRANCH->EXACT=>x1
- BRANCH->EXACT=>x2
- tail
+ BRANCH->EXACT=>x1
+ BRANCH->EXACT=>x2
+ tail
- which would be constructed from a pattern like /A|LIST|OF|WORDS/
+ which would be constructed from a pattern like /A|LIST|OF|WORDS/
- If we can find such a subseqence we need to turn the first
- element into a trie and then add the subsequent branch exact
- strings to the trie.
+ If we can find such a subseqence we need to turn the first
+ element into a trie and then add the subsequent branch exact
+ strings to the trie.
- We have two cases
+ We have two cases
- 1. patterns where the whole set of branch can be converted.
+ 1. patterns where the whole set of branch can be converted.
- 2. patterns where only a subset can be converted.
+ 2. patterns where only a subset can be converted.
- In case 1 we can replace the whole set with a single regop
- for the trie. In case 2 we need to keep the start and end
- branchs so
+ In case 1 we can replace the whole set with a single regop
+ for the trie. In case 2 we need to keep the start and end
+ branchs so
- 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
- becomes BRANCH TRIE; BRANCH X;
+ 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
+ becomes BRANCH TRIE; BRANCH X;
- There is an additional case, that being where there is a
- common prefix, which gets split out into an EXACT like node
- preceding the TRIE node.
+ There is an additional case, that being where there is a
+ common prefix, which gets split out into an EXACT like node
+ preceding the TRIE node.
- If x(1..n)==tail then we can do a simple trie, if not we make
- a "jump" trie, such that when we match the appropriate word
- we "jump" to the appopriate tail node. Essentailly we turn
- a nested if into a case structure of sorts.
+ If x(1..n)==tail then we can do a simple trie, if not we make
+ a "jump" trie, such that when we match the appropriate word
+ we "jump" to the appopriate tail node. Essentailly we turn
+ a nested if into a case structure of sorts.
- */
-
- int made=0;
- if (!re_trie_maxbuff) {
- re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
- if (!SvIOK(re_trie_maxbuff))
- sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
- }
- if ( SvIV(re_trie_maxbuff)>=0 ) {
- regnode *cur;
- regnode *first = (regnode *)NULL;
- regnode *last = (regnode *)NULL;
- regnode *tail = scan;
- U8 optype = 0;
- U32 count=0;
+ */
+
+ int made=0;
+ if (!re_trie_maxbuff) {
+ re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
+ if (!SvIOK(re_trie_maxbuff))
+ sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
+ }
+ if ( SvIV(re_trie_maxbuff)>=0 ) {
+ regnode *cur;
+ regnode *first = (regnode *)NULL;
+ regnode *last = (regnode *)NULL;
+ regnode *tail = scan;
+ U8 optype = 0;
+ U32 count=0;
#ifdef DEBUGGING
- SV * const mysv = sv_newmortal(); /* for dumping */
+ SV * const mysv = sv_newmortal(); /* for dumping */
#endif
- /* var tail is used because there may be a TAIL
- regop in the way. Ie, the exacts will point to the
- thing following the TAIL, but the last branch will
- point at the TAIL. So we advance tail. If we
- have nested (?:) we may have to move through several
- tails.
- */
-
- while ( OP( tail ) == TAIL ) {
- /* this is the TAIL generated by (?:) */
- tail = regnext( tail );
- }
+ /* var tail is used because there may be a TAIL
+ regop in the way. Ie, the exacts will point to the
+ thing following the TAIL, but the last branch will
+ point at the TAIL. So we advance tail. If we
+ have nested (?:) we may have to move through several
+ tails.
+ */
+
+ while ( OP( tail ) == TAIL ) {
+ /* this is the TAIL generated by (?:) */
+ tail = regnext( tail );
+ }
-
- DEBUG_OPTIMISE_r({
- regprop(RExC_rx, mysv, tail );
- PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
- (int)depth * 2 + 2, "",
- "Looking for TRIE'able sequences. Tail node is: ",
- SvPV_nolen_const( mysv )
- );
- });
-
- /*
-
- step through the branches, cur represents each
- branch, noper is the first thing to be matched
- as part of that branch and noper_next is the
- regnext() of that node. if noper is an EXACT
- and noper_next is the same as scan (our current
- position in the regex) then the EXACT branch is
- a possible optimization target. Once we have
- two or more consequetive such branches we can
- create a trie of the EXACT's contents and stich
- it in place. If the sequence represents all of
- the branches we eliminate the whole thing and
- replace it with a single TRIE. If it is a
- subsequence then we need to stitch it in. This
- means the first branch has to remain, and needs
- to be repointed at the item on the branch chain
- following the last branch optimized. This could
- be either a BRANCH, in which case the
- subsequence is internal, or it could be the
- item following the branch sequence in which
- case the subsequence is at the end.
-
- */
-
- /* dont use tail as the end marker for this traverse */
- for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
- regnode * const noper = NEXTOPER( cur );
-#if defined(DEBUGGING) || defined(NOJUMPTRIE)
- regnode * const noper_next = regnext( noper );
-#endif
- DEBUG_OPTIMISE_r({
- regprop(RExC_rx, mysv, cur);
- PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
- (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
+ DEBUG_OPTIMISE_r({
+ regprop(RExC_rx, mysv, tail );
+ PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
+ (int)depth * 2 + 2, "",
+ "Looking for TRIE'able sequences. Tail node is: ",
+ SvPV_nolen_const( mysv )
+ );
+ });
+
+ /*
+
+ step through the branches, cur represents each
+ branch, noper is the first thing to be matched
+ as part of that branch and noper_next is the
+ regnext() of that node. if noper is an EXACT
+ and noper_next is the same as scan (our current
+ position in the regex) then the EXACT branch is
+ a possible optimization target. Once we have
+ two or more consequetive such branches we can
+ create a trie of the EXACT's contents and stich
+ it in place. If the sequence represents all of
+ the branches we eliminate the whole thing and
+ replace it with a single TRIE. If it is a
+ subsequence then we need to stitch it in. This
+ means the first branch has to remain, and needs
+ to be repointed at the item on the branch chain
+ following the last branch optimized. This could
+ be either a BRANCH, in which case the
+ subsequence is internal, or it could be the
+ item following the branch sequence in which
+ case the subsequence is at the end.
- regprop(RExC_rx, mysv, noper);
- PerlIO_printf( Perl_debug_log, " -> %s",
- SvPV_nolen_const(mysv));
+*/
- if ( noper_next ) {
- regprop(RExC_rx, mysv, noper_next );
- PerlIO_printf( Perl_debug_log,"\t=> %s\t",
- SvPV_nolen_const(mysv));
- }
- PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
- REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
- });
- if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
- : PL_regkind[ OP( noper ) ] == EXACT )
- || OP(noper) == NOTHING )
+ /* dont use tail as the end marker for this traverse */
+ for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
+ regnode * const noper = NEXTOPER( cur );
+#if defined(DEBUGGING) || defined(NOJUMPTRIE)
+ regnode * const noper_next = regnext( noper );
+#endif
+
+ DEBUG_OPTIMISE_r({
+ regprop(RExC_rx, mysv, cur);
+ PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
+ (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
+
+ regprop(RExC_rx, mysv, noper);
+ PerlIO_printf( Perl_debug_log, " -> %s",
+ SvPV_nolen_const(mysv));
+
+ if ( noper_next ) {
+ regprop(RExC_rx, mysv, noper_next );
+ PerlIO_printf( Perl_debug_log,"\t=> %s\t",
+ SvPV_nolen_const(mysv));
+ }
+ PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
+ REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
+ });
+ if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
+ : PL_regkind[ OP( noper ) ] == EXACT )
+ || OP(noper) == NOTHING )
#ifdef NOJUMPTRIE
- && noper_next == tail
+ && noper_next == tail
#endif
- && count < U16_MAX)
- {
- count++;
- if ( !first || optype == NOTHING ) {
- if (!first) first = cur;
- optype = OP( noper );
- } else {
- last = cur;
- }
- } else {
- if ( last ) {
- make_trie( pRExC_state,
- startbranch, first, cur, tail, count,
- optype, depth+1 );
- }
- if ( PL_regkind[ OP( noper ) ] == EXACT
+ && count < U16_MAX)
+ {
+ count++;
+ if ( !first || optype == NOTHING ) {
+ if (!first) first = cur;
+ optype = OP( noper );
+ } else {
+ last = cur;
+ }
+ } else {
+ if ( last ) {
+ make_trie( pRExC_state,
+ startbranch, first, cur, tail, count,
+ optype, depth+1 );
+ }
+ if ( PL_regkind[ OP( noper ) ] == EXACT
#ifdef NOJUMPTRIE
- && noper_next == tail
+ && noper_next == tail
#endif
- ){
- count = 1;
- first = cur;
- optype = OP( noper );
- } else {
- count = 0;
- first = NULL;
- optype = 0;
- }
- last = NULL;
- }
- }
- DEBUG_OPTIMISE_r({
- regprop(RExC_rx, mysv, cur);
- PerlIO_printf( Perl_debug_log,
- "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
- "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
-
- });
- if ( last ) {
- made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
+ ) {
+ count = 1;
+ first = cur;
+ optype = OP( noper );
+ } else {
+ count = 0;
+ first = NULL;
+ optype = 0;
+ }
+ last = NULL;
+ }
+ }
+ DEBUG_OPTIMISE_r({
+ regprop(RExC_rx, mysv, cur);
+ PerlIO_printf( Perl_debug_log,
+ "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
+ "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
+
+ });
+ if ( last ) {
+ made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
#ifdef TRIE_STUDY_OPT
- if ( ((made == MADE_EXACT_TRIE &&
- startbranch == first)
- || ( first_non_open == first )) &&
- depth==0 )
- flags |= SCF_TRIE_RESTUDY;
+ if ( ((made == MADE_EXACT_TRIE &&
+ startbranch == first)
+ || ( first_non_open == first )) &&
+ depth==0 ) {
+ flags |= SCF_TRIE_RESTUDY;
+ if ( startbranch == first
+ && scan == tail )
+ {
+ RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
+ }
+ }
#endif
- }
- }
-
- } /* do trie */
-
- }
- 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) {
- I32 l = STR_LEN(scan);
- UV uc;
- if (UTF) {
- const U8 * const s = (U8*)STRING(scan);
- l = utf8_length(s, s + l);
- uc = utf8_to_uvchr(s, NULL);
- } else {
- uc = *((U8*)STRING(scan));
- }
- min += l;
- if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
- /* 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;
+ }
+ }
+
+ } /* do trie */
+
}
- sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
- if (UTF)
- SvUTF8_on(data->last_found);
- {
- SV * const sv = data->last_found;
- MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
- mg_find(sv, PERL_MAGIC_utf8) : NULL;
- if (mg && mg->mg_len >= 0)
- mg->mg_len += utf8_length((U8*)STRING(scan),
- (U8*)STRING(scan)+STR_LEN(scan));
+ 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) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
+ scan_frame *newframe = NULL;
+ I32 paren;
+ regnode *start;
+ regnode *end;
+
+ if (OP(scan) != SUSPEND) {
+ /* set the pointer */
+ if (OP(scan) == GOSUB) {
+ paren = ARG(scan);
+ RExC_recurse[ARG2L(scan)] = scan;
+ start = RExC_open_parens[paren-1];
+ end = RExC_close_parens[paren-1];
+ } else {
+ paren = 0;
+ start = RExC_rx->program + 1;
+ end = RExC_opend;
+ }
+ if (!recursed) {
+ Newxz(recursed, (((RExC_npar)>>3) +1), U8);
+ SAVEFREEPV(recursed);
+ }
+ if (!PAREN_TEST(recursed,paren+1)) {
+ PAREN_SET(recursed,paren+1);
+ Newx(newframe,1,scan_frame);
+ } else {
+ if (flags & SCF_DO_SUBSTR) {
+ scan_commit(pRExC_state,data,minlenp);
+ data->longest = &(data->longest_float);
+ }
+ is_inf = is_inf_internal = 1;
+ if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
+ cl_anything(pRExC_state, data->start_class);
+ flags &= ~SCF_DO_STCLASS;
+ }
+ } else {
+ Newx(newframe,1,scan_frame);
+ paren = stopparen;
+ start = scan+2;
+ end = regnext(scan);
}
- data->last_end = data->pos_min + l;
- data->pos_min += l; /* As in the first entry. */
- data->flags &= ~SF_BEFORE_EOL;
- }
- if (flags & SCF_DO_STCLASS_AND) {
- /* Check whether it is compatible with what we know already! */
- int compat = 1;
-
- if (uc >= 0x100 ||
- (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
- && !ANYOF_BITMAP_TEST(data->start_class, uc)
- && (!(data->start_class->flags & ANYOF_FOLD)
- || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
- )
- compat = 0;
- ANYOF_CLASS_ZERO(data->start_class);
- ANYOF_BITMAP_ZERO(data->start_class);
- if (compat)
- ANYOF_BITMAP_SET(data->start_class, uc);
- data->start_class->flags &= ~ANYOF_EOS;
- if (uc < 0x100)
- data->start_class->flags &= ~ANYOF_UNICODE_ALL;
- }
- else if (flags & SCF_DO_STCLASS_OR) {
- /* false positive possible if the class is case-folded */
- if (uc < 0x100)
- ANYOF_BITMAP_SET(data->start_class, uc);
- else
- data->start_class->flags |= ANYOF_UNICODE_ALL;
- data->start_class->flags &= ~ANYOF_EOS;
- cl_and(data->start_class, &and_with);
- }
- flags &= ~SCF_DO_STCLASS;
- }
- else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
- I32 l = STR_LEN(scan);
- UV uc = *((U8*)STRING(scan));
-
- /* Search for fixed substrings supports EXACT only. */
- if (flags & SCF_DO_SUBSTR) {
- assert(data);
- scan_commit(pRExC_state, data, minlenp);
+ if (newframe) {
+ assert(start);
+ assert(end);
+ SAVEFREEPV(newframe);
+ newframe->next = regnext(scan);
+ newframe->last = end;
+ newframe->stop = stopparen;
+ newframe->prev = frame;
+ frame = newframe;
+ scan = start;
+ stopparen = paren;
+ continue;
+ }
}
- if (UTF) {
- const U8 * const s = (U8 *)STRING(scan);
- l = utf8_length(s, s + l);
- uc = utf8_to_uvchr(s, NULL);
- }
- min += l;
- if (flags & SCF_DO_SUBSTR)
- data->pos_min += l;
- if (flags & SCF_DO_STCLASS_AND) {
- /* Check whether it is compatible with what we know already! */
- int compat = 1;
-
- if (uc >= 0x100 ||
- (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
- && !ANYOF_BITMAP_TEST(data->start_class, uc)
- && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
- compat = 0;
- ANYOF_CLASS_ZERO(data->start_class);
- ANYOF_BITMAP_ZERO(data->start_class);
- if (compat) {
- ANYOF_BITMAP_SET(data->start_class, uc);
+ else if (OP(scan) == EXACT) {
+ I32 l = STR_LEN(scan);
+ UV uc;
+ if (UTF) {
+ const U8 * const s = (U8*)STRING(scan);
+ l = utf8_length(s, s + l);
+ uc = utf8_to_uvchr(s, NULL);
+ } else {
+ uc = *((U8*)STRING(scan));
+ }
+ min += l;
+ if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
+ /* 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, STRING(scan), STR_LEN(scan));
+ if (UTF)
+ SvUTF8_on(data->last_found);
+ {
+ SV * const sv = data->last_found;
+ MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
+ mg_find(sv, PERL_MAGIC_utf8) : NULL;
+ if (mg && mg->mg_len >= 0)
+ mg->mg_len += utf8_length((U8*)STRING(scan),
+ (U8*)STRING(scan)+STR_LEN(scan));
+ }
+ data->last_end = data->pos_min + l;
+ data->pos_min += l; /* As in the first entry. */
+ data->flags &= ~SF_BEFORE_EOL;
+ }
+ if (flags & SCF_DO_STCLASS_AND) {
+ /* Check whether it is compatible with what we know already! */
+ int compat = 1;
+
+ if (uc >= 0x100 ||
+ (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
+ && !ANYOF_BITMAP_TEST(data->start_class, uc)
+ && (!(data->start_class->flags & ANYOF_FOLD)
+ || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
+ )
+ compat = 0;
+ ANYOF_CLASS_ZERO(data->start_class);
+ ANYOF_BITMAP_ZERO(data->start_class);
+ if (compat)
+ ANYOF_BITMAP_SET(data->start_class, uc);
data->start_class->flags &= ~ANYOF_EOS;
- data->start_class->flags |= ANYOF_FOLD;
- if (OP(scan) == EXACTFL)
- data->start_class->flags |= ANYOF_LOCALE;
+ if (uc < 0x100)
+ data->start_class->flags &= ~ANYOF_UNICODE_ALL;
}
- }
- else if (flags & SCF_DO_STCLASS_OR) {
- if (data->start_class->flags & ANYOF_FOLD) {
- /* false positive possible if the class is case-folded.
- Assume that the locale settings are the same... */
+ else if (flags & SCF_DO_STCLASS_OR) {
+ /* false positive possible if the class is case-folded */
if (uc < 0x100)
ANYOF_BITMAP_SET(data->start_class, uc);
+ else
+ data->start_class->flags |= ANYOF_UNICODE_ALL;
data->start_class->flags &= ~ANYOF_EOS;
+ cl_and(data->start_class, and_withp);
}
- cl_and(data->start_class, &and_with);
+ flags &= ~SCF_DO_STCLASS;
}
- flags &= ~SCF_DO_STCLASS;
- }
- else if (strchr((const char*)PL_varies,OP(scan))) {
- I32 mincount, maxcount, minnext, deltanext, fl = 0;
- I32 f = flags, pos_before = 0;
- regnode * const oscan = scan;
- struct regnode_charclass_class this_class;
- struct regnode_charclass_class *oclass = NULL;
- I32 next_is_eval = 0;
-
- switch (PL_regkind[OP(scan)]) {
- case WHILEM: /* End of (?:...)* . */
- scan = NEXTOPER(scan);
- goto finish;
- case PLUS:
- if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
- next = NEXTOPER(scan);
- if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
- 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:
- if (flags & SCF_DO_STCLASS) {
- mincount = 0;
- maxcount = REG_INFTY;
- next = regnext(scan);
- scan = NEXTOPER(scan);
- goto do_curly;
- }
- is_inf = is_inf_internal = 1;
- scan = regnext(scan);
- if (flags & SCF_DO_SUBSTR) {
- scan_commit(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
- data->longest = &(data->longest_float);
- }
- goto optimize_curly_tail;
- case CURLY:
- mincount = ARG1(scan);
- maxcount = ARG2(scan);
- next = regnext(scan);
- if (OP(scan) == CURLYX) {
- I32 lp = (data ? *(data->last_closep) : 0);
- scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
- }
- scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
- next_is_eval = (OP(scan) == EVAL);
- do_curly:
+ else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
+ I32 l = STR_LEN(scan);
+ UV uc = *((U8*)STRING(scan));
+
+ /* Search for fixed substrings supports EXACT only. */
if (flags & SCF_DO_SUBSTR) {
- if (mincount == 0) scan_commit(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
- 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;
+ assert(data);
+ scan_commit(pRExC_state, data, minlenp);
}
- if (flags & SCF_DO_STCLASS) {
- cl_init(pRExC_state, &this_class);
- oclass = data->start_class;
- data->start_class = &this_class;
- f |= SCF_DO_STCLASS_AND;
- f &= ~SCF_DO_STCLASS_OR;
+ if (UTF) {
+ const U8 * const s = (U8 *)STRING(scan);
+ l = utf8_length(s, s + l);
+ uc = utf8_to_uvchr(s, NULL);
}
- /* These are the cases when once a subexpression
- fails at a particular position, it cannot succeed
- even after backtracking at the enclosing scope.
-
- XXXX what if minimal match and we are at the
- initial run of {n,m}? */
- if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
- f &= ~SCF_WHILEM_VISITED_POS;
-
- /* This will finish on WHILEM, setting scan, or on NULL: */
- minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext, last, data,
- (mincount == 0
- ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
-
- if (flags & SCF_DO_STCLASS)
- data->start_class = oclass;
- if (mincount == 0 || minnext == 0) {
- if (flags & SCF_DO_STCLASS_OR) {
- cl_or(pRExC_state, data->start_class, &this_class);
- }
- else if (flags & SCF_DO_STCLASS_AND) {
- /* Switch to OR mode: cache the old value of
- * data->start_class */
- StructCopy(data->start_class, &and_with,
- struct regnode_charclass_class);
- flags &= ~SCF_DO_STCLASS_AND;
- StructCopy(&this_class, data->start_class,
- struct regnode_charclass_class);
- flags |= SCF_DO_STCLASS_OR;
- data->start_class->flags |= ANYOF_EOS;
- }
- } else { /* Non-zero len */
- if (flags & SCF_DO_STCLASS_OR) {
- cl_or(pRExC_state, data->start_class, &this_class);
- cl_and(data->start_class, &and_with);
+ min += l;
+ if (flags & SCF_DO_SUBSTR)
+ data->pos_min += l;
+ if (flags & SCF_DO_STCLASS_AND) {
+ /* Check whether it is compatible with what we know already! */
+ int compat = 1;
+
+ if (uc >= 0x100 ||
+ (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
+ && !ANYOF_BITMAP_TEST(data->start_class, uc)
+ && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
+ compat = 0;
+ ANYOF_CLASS_ZERO(data->start_class);
+ ANYOF_BITMAP_ZERO(data->start_class);
+ if (compat) {
+ ANYOF_BITMAP_SET(data->start_class, uc);
+ data->start_class->flags &= ~ANYOF_EOS;
+ data->start_class->flags |= ANYOF_FOLD;
+ if (OP(scan) == EXACTFL)
+ data->start_class->flags |= ANYOF_LOCALE;
}
- else if (flags & SCF_DO_STCLASS_AND)
- cl_and(data->start_class, &this_class);
- flags &= ~SCF_DO_STCLASS;
}
- if (!scan) /* It was not CURLYX, but CURLY. */
- scan = next;
- if ( /* ? quantifier ok, except for (?{ ... }) */
- (next_is_eval || !(mincount == 0 && maxcount == 1))
- && (minnext == 0) && (deltanext == 0)
- && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
- && maxcount <= REG_INFTY/3 /* Complement check for big count */
- && ckWARN(WARN_REGEXP))
- {
- vWARN(RExC_parse,
- "Quantifier unexpected on zero-length expression");
+ else if (flags & SCF_DO_STCLASS_OR) {
+ if (data->start_class->flags & ANYOF_FOLD) {
+ /* false positive possible if the class is case-folded.
+ Assume that the locale settings are the same... */
+ if (uc < 0x100)
+ ANYOF_BITMAP_SET(data->start_class, uc);
+ data->start_class->flags &= ~ANYOF_EOS;
+ }
+ cl_and(data->start_class, and_withp);
}
+ flags &= ~SCF_DO_STCLASS;
+ }
+ else if (strchr((const char*)PL_varies,OP(scan))) {
+ I32 mincount, maxcount, minnext, deltanext, fl = 0;
+ I32 f = flags, pos_before = 0;
+ regnode * const oscan = scan;
+ struct regnode_charclass_class this_class;
+ struct regnode_charclass_class *oclass = NULL;
+ I32 next_is_eval = 0;
+
+ switch (PL_regkind[OP(scan)]) {
+ case WHILEM: /* End of (?:...)* . */
+ scan = NEXTOPER(scan);
+ goto finish;
+ case PLUS:
+ if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
+ next = NEXTOPER(scan);
+ if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
+ 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:
+ if (flags & SCF_DO_STCLASS) {
+ mincount = 0;
+ maxcount = REG_INFTY;
+ next = regnext(scan);
+ scan = NEXTOPER(scan);
+ goto do_curly;
+ }
+ is_inf = is_inf_internal = 1;
+ scan = regnext(scan);
+ if (flags & SCF_DO_SUBSTR) {
+ scan_commit(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
+ data->longest = &(data->longest_float);
+ }
+ goto optimize_curly_tail;
+ case CURLY:
+ if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
+ && (scan->flags == stopparen))
+ {
+ mincount = 1;
+ maxcount = 1;
+ } else {
+ mincount = ARG1(scan);
+ maxcount = ARG2(scan);
+ }
+ next = regnext(scan);
+ if (OP(scan) == CURLYX) {
+ I32 lp = (data ? *(data->last_closep) : 0);
+ scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
+ }
+ scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
+ next_is_eval = (OP(scan) == EVAL);
+do_curly:
+ if (flags & SCF_DO_SUBSTR) {
+ if (mincount == 0) scan_commit(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
+ 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;
+ }
+ if (flags & SCF_DO_STCLASS) {
+ cl_init(pRExC_state, &this_class);
+ oclass = data->start_class;
+ data->start_class = &this_class;
+ f |= SCF_DO_STCLASS_AND;
+ f &= ~SCF_DO_STCLASS_OR;
+ }
+ /* These are the cases when once a subexpression
+ fails at a particular position, it cannot succeed
+ even after backtracking at the enclosing scope.
+
+ XXXX what if minimal match and we are at the
+ initial run of {n,m}? */
+ if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
+ f &= ~SCF_WHILEM_VISITED_POS;
+
+ /* This will finish on WHILEM, setting scan, or on NULL: */
+ minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
+ last, data, stopparen, recursed, NULL,
+ (mincount == 0
+ ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
+
+ if (flags & SCF_DO_STCLASS)
+ data->start_class = oclass;
+ if (mincount == 0 || minnext == 0) {
+ if (flags & SCF_DO_STCLASS_OR) {
+ cl_or(pRExC_state, data->start_class, &this_class);
+ }
+ else if (flags & SCF_DO_STCLASS_AND) {
+ /* Switch to OR mode: cache the old value of
+ * data->start_class */
+ INIT_AND_WITHP;
+ StructCopy(data->start_class, and_withp,
+ struct regnode_charclass_class);
+ flags &= ~SCF_DO_STCLASS_AND;
+ StructCopy(&this_class, data->start_class,
+ struct regnode_charclass_class);
+ flags |= SCF_DO_STCLASS_OR;
+ data->start_class->flags |= ANYOF_EOS;
+ }
+ } else { /* Non-zero len */
+ if (flags & SCF_DO_STCLASS_OR) {
+ cl_or(pRExC_state, data->start_class, &this_class);
+ cl_and(data->start_class, and_withp);
+ }
+ else if (flags & SCF_DO_STCLASS_AND)
+ cl_and(data->start_class, &this_class);
+ flags &= ~SCF_DO_STCLASS;
+ }
+ if (!scan) /* It was not CURLYX, but CURLY. */
+ scan = next;
+ if ( /* ? quantifier ok, except for (?{ ... }) */
+ (next_is_eval || !(mincount == 0 && maxcount == 1))
+ && (minnext == 0) && (deltanext == 0)
+ && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
+ && maxcount <= REG_INFTY/3 /* Complement check for big count */
+ && ckWARN(WARN_REGEXP))
+ {
+ vWARN(RExC_parse,
+ "Quantifier unexpected on zero-length expression");
+ }
- min += minnext * mincount;
- is_inf_internal |= ((maxcount == REG_INFTY
- && (minnext + deltanext) > 0)
- || deltanext == I32_MAX);
- is_inf |= is_inf_internal;
- 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 * const nxt1 = nxt;
+ min += minnext * mincount;
+ is_inf_internal |= ((maxcount == REG_INFTY
+ && (minnext + deltanext) > 0)
+ || deltanext == I32_MAX);
+ is_inf |= is_inf_internal;
+ 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 * const nxt1 = nxt;
#ifdef DEBUGGING
- regnode *nxt2;
+ regnode *nxt2;
#endif
- /* Skip open. */
- nxt = regnext(nxt);
- if (!strchr((const char*)PL_simple,OP(nxt))
- && !(PL_regkind[OP(nxt)] == EXACT
- && STR_LEN(nxt) == 1))
- goto nogo;
+ /* Skip open. */
+ nxt = regnext(nxt);
+ if (!strchr((const char*)PL_simple,OP(nxt))
+ && !(PL_regkind[OP(nxt)] == EXACT
+ && STR_LEN(nxt) == 1))
+ goto nogo;
#ifdef DEBUGGING
- nxt2 = nxt;
+ nxt2 = nxt;
#endif
- nxt = regnext(nxt);
- if (OP(nxt) != CLOSE)
- goto nogo;
- /* Now we know that nxt2 is the only contents: */
- oscan->flags = (U8)ARG(nxt);
- OP(oscan) = CURLYN;
- OP(nxt1) = NOTHING; /* was OPEN. */
+ nxt = regnext(nxt);
+ if (OP(nxt) != CLOSE)
+ goto nogo;
+ if (RExC_open_parens) {
+ RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
+ RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
+ }
+ /* Now we know that nxt2 is the only contents: */
+ oscan->flags = (U8)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. */
+ 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 /* atom is fixed width */
- && minnext != 0 /* CURLYM can't handle zero width */
- ) {
- /* 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 = (U8)ARG(nxt);
- OP(nxt1) = OPTIMIZED; /* was OPEN. */
- OP(nxt) = OPTIMIZED; /* was CLOSE. */
+ }
+nogo:
+
+ /* Try optimization CURLYX => CURLYM. */
+ if ( OP(oscan) == CURLYX && data
+ && !(data->flags & SF_HAS_PAR)
+ && !(data->flags & SF_HAS_EVAL)
+ && !deltanext /* atom is fixed width */
+ && minnext != 0 /* CURLYM can't handle zero width */
+ ) {
+ /* 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 = (U8)ARG(nxt);
+ if (RExC_open_parens) {
+ RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
+ RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
+ }
+ 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. */
+ 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 */
+ 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(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
+ NULL, stopparen, recursed, NULL, 0,depth+1);
}
- nxt1 = nnxt;
+ else
+ oscan->flags = 0;
}
-#endif
- /* Optimize again: */
- study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
- NULL, 0,depth+1);
- }
- else
- oscan->flags = 0;
- }
- else if ((OP(oscan) == CURLYX)
- && (flags & SCF_WHILEM_VISITED_POS)
- /* See the comment on a similar expression above.
- However, this time it not a subexpression
- we care about, but the expression itself. */
- && (maxcount == REG_INFTY)
- && data && ++data->whilem_c < 16) {
- /* This stays as CURLYX, we can put the count/of pair. */
- /* Find WHILEM (as in regexec.c) */
- regnode *nxt = oscan + NEXT_OFF(oscan);
-
- if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
- nxt += ARG(nxt);
- PREVOPER(nxt)->flags = (U8)(data->whilem_c
- | (RExC_whilem_seen << 4)); /* On WHILEM */
- }
- if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
- pars++;
- if (flags & SCF_DO_SUBSTR) {
- SV *last_str = NULL;
- int counted = mincount != 0;
+ else if ((OP(oscan) == CURLYX)
+ && (flags & SCF_WHILEM_VISITED_POS)
+ /* See the comment on a similar expression above.
+ However, this time it not a subexpression
+ we care about, but the expression itself. */
+ && (maxcount == REG_INFTY)
+ && data && ++data->whilem_c < 16) {
+ /* This stays as CURLYX, we can put the count/of pair. */
+ /* Find WHILEM (as in regexec.c) */
+ regnode *nxt = oscan + NEXT_OFF(oscan);
+
+ if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
+ nxt += ARG(nxt);
+ PREVOPER(nxt)->flags = (U8)(data->whilem_c
+ | (RExC_whilem_seen << 4)); /* On WHILEM */
+ }
+ if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
+ pars++;
+ if (flags & SCF_DO_SUBSTR) {
+ SV *last_str = NULL;
+ int counted = mincount != 0;
- if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
+ if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
#if defined(SPARC64_GCC_WORKAROUND)
- I32 b = 0;
- STRLEN l = 0;
- const char *s = NULL;
- I32 old = 0;
+ I32 b = 0;
+ STRLEN l = 0;
+ const char *s = NULL;
+ I32 old = 0;
- if (pos_before >= data->last_start_min)
- b = pos_before;
- else
- b = data->last_start_min;
+ if (pos_before >= data->last_start_min)
+ b = pos_before;
+ else
+ b = data->last_start_min;
- l = 0;
- s = SvPV_const(data->last_found, l);
- old = b - data->last_start_min;
+ l = 0;
+ s = SvPV_const(data->last_found, l);
+ old = b - data->last_start_min;
#else
- I32 b = pos_before >= data->last_start_min
- ? pos_before : data->last_start_min;
- STRLEN l;
- const char * const s = SvPV_const(data->last_found, l);
- I32 old = b - data->last_start_min;
+ I32 b = pos_before >= data->last_start_min
+ ? pos_before : data->last_start_min;
+ STRLEN l;
+ const char * const s = SvPV_const(data->last_found, l);
+ I32 old = b - data->last_start_min;
#endif
- if (UTF)
- old = utf8_hop((U8*)s, old) - (U8*)s;
-
- l -= old;
- /* Get the added string: */
- last_str = newSVpvn(s + old, l);
- if (UTF)
- SvUTF8_on(last_str);
- 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_const(last_str), l, mincount - 1);
- SvCUR_set(last_str, SvCUR(last_str) * mincount);
- /* Add additional parts. */
- SvCUR_set(data->last_found,
- SvCUR(data->last_found) - l);
- sv_catsv(data->last_found, last_str);
- {
- SV * sv = data->last_found;
- MAGIC *mg =
- SvUTF8(sv) && SvMAGICAL(sv) ?
+ if (UTF)
+ old = utf8_hop((U8*)s, old) - (U8*)s;
+
+ l -= old;
+ /* Get the added string: */
+ last_str = newSVpvn(s + old, l);
+ if (UTF)
+ SvUTF8_on(last_str);
+ 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_const(last_str), l, mincount - 1);
+ SvCUR_set(last_str, SvCUR(last_str) * mincount);
+ /* Add additional parts. */
+ SvCUR_set(data->last_found,
+ SvCUR(data->last_found) - l);
+ sv_catsv(data->last_found, last_str);
+ {
+ SV * sv = data->last_found;
+ MAGIC *mg =
+ SvUTF8(sv) && SvMAGICAL(sv) ?
+ mg_find(sv, PERL_MAGIC_utf8) : NULL;
+ if (mg && mg->mg_len >= 0)
+ mg->mg_len += CHR_SVLEN(last_str);
+ }
+ data->last_end += l * (mincount - 1);
+ }
+ } else {
+ /* start offset must point into the last copy */
+ data->last_start_min += minnext * (mincount - 1);
+ data->last_start_max += is_inf ? I32_MAX
+ : (maxcount - 1) * (minnext + data->pos_delta);
+ }
+ }
+ /* It is counted once already... */
+ data->pos_min += minnext * (mincount - counted);
+ data->pos_delta += - counted * deltanext +
+ (minnext + deltanext) * maxcount - minnext * mincount;
+ if (mincount != maxcount) {
+ /* Cannot extend fixed substrings found inside
+ the group. */
+ scan_commit(pRExC_state,data,minlenp);
+ if (mincount && last_str) {
+ SV * const sv = data->last_found;
+ MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
mg_find(sv, PERL_MAGIC_utf8) : NULL;
- if (mg && mg->mg_len >= 0)
- mg->mg_len += CHR_SVLEN(last_str);
+
+ if (mg)
+ mg->mg_len = -1;
+ sv_setsv(sv, last_str);
+ data->last_end = data->pos_min;
+ data->last_start_min =
+ data->pos_min - CHR_SVLEN(last_str);
+ data->last_start_max = is_inf
+ ? I32_MAX
+ : data->pos_min + data->pos_delta
+ - CHR_SVLEN(last_str);
}
- data->last_end += l * (mincount - 1);
+ data->longest = &(data->longest_float);
}
- } else {
- /* start offset must point into the last copy */
- data->last_start_min += minnext * (mincount - 1);
- data->last_start_max += is_inf ? I32_MAX
- : (maxcount - 1) * (minnext + data->pos_delta);
+ SvREFCNT_dec(last_str);
}
- }
- /* It is counted once already... */
- data->pos_min += minnext * (mincount - counted);
- data->pos_delta += - counted * deltanext +
- (minnext + deltanext) * maxcount - minnext * mincount;
- if (mincount != maxcount) {
- /* Cannot extend fixed substrings found inside
- the group. */
- scan_commit(pRExC_state,data,minlenp);
- if (mincount && last_str) {
- SV * const sv = data->last_found;
- MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
- mg_find(sv, PERL_MAGIC_utf8) : NULL;
-
- if (mg)
- mg->mg_len = -1;
- sv_setsv(sv, last_str);
- data->last_end = data->pos_min;
- data->last_start_min =
- data->pos_min - CHR_SVLEN(last_str);
- data->last_start_max = is_inf
- ? I32_MAX
- : data->pos_min + data->pos_delta
- - CHR_SVLEN(last_str);
+ if (data && (fl & SF_HAS_EVAL))
+ data->flags |= SF_HAS_EVAL;
+optimize_curly_tail:
+ if (OP(oscan) != CURLYX) {
+ while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
+ && NEXT_OFF(next))
+ NEXT_OFF(oscan) += NEXT_OFF(next);
}
- data->longest = &(data->longest_float);
- }
- SvREFCNT_dec(last_str);
- }
- if (data && (fl & SF_HAS_EVAL))
- data->flags |= SF_HAS_EVAL;
- optimize_curly_tail:
- if (OP(oscan) != CURLYX) {
- while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
- && NEXT_OFF(next))
- NEXT_OFF(oscan) += NEXT_OFF(next);
+ continue;
+ default: /* REF and CLUMP only? */
+ if (flags & SCF_DO_SUBSTR) {
+ scan_commit(pRExC_state,data,minlenp); /* Cannot expect anything... */
+ data->longest = &(data->longest_float);
+ }
+ is_inf = is_inf_internal = 1;
+ if (flags & SCF_DO_STCLASS_OR)
+ cl_anything(pRExC_state, data->start_class);
+ flags &= ~SCF_DO_STCLASS;
+ break;
}
- continue;
- default: /* REF and CLUMP only? */
+ }
+ else if (strchr((const char*)PL_simple,OP(scan))) {
+ int value = 0;
+
if (flags & SCF_DO_SUBSTR) {
- scan_commit(pRExC_state,data,minlenp); /* Cannot expect anything... */
- data->longest = &(data->longest_float);
+ scan_commit(pRExC_state,data,minlenp);
+ data->pos_min++;
+ }
+ min++;
+ if (flags & SCF_DO_STCLASS) {
+ data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
+
+ /* Some of the logic below assumes that switching
+ locale on will only add false positives. */
+ switch (PL_regkind[OP(scan)]) {
+ case SANY:
+ default:
+do_default:
+ /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
+ if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
+ cl_anything(pRExC_state, data->start_class);
+ break;
+ case REG_ANY:
+ if (OP(scan) == SANY)
+ goto do_default;
+ if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
+ value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
+ || (data->start_class->flags & ANYOF_CLASS));
+ cl_anything(pRExC_state, data->start_class);
+ }
+ if (flags & SCF_DO_STCLASS_AND || !value)
+ ANYOF_BITMAP_CLEAR(data->start_class,'\n');
+ break;
+ case ANYOF:
+ if (flags & SCF_DO_STCLASS_AND)
+ cl_and(data->start_class,
+ (struct regnode_charclass_class*)scan);
+ else
+ cl_or(pRExC_state, data->start_class,
+ (struct regnode_charclass_class*)scan);
+ break;
+ case ALNUM:
+ if (flags & SCF_DO_STCLASS_AND) {
+ if (!(data->start_class->flags & ANYOF_LOCALE)) {
+ ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
+ for (value = 0; value < 256; value++)
+ if (!isALNUM(value))
+ ANYOF_BITMAP_CLEAR(data->start_class, value);
+ }
+ }
+ else {
+ if (data->start_class->flags & ANYOF_LOCALE)
+ ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
+ else {
+ for (value = 0; value < 256; value++)
+ if (isALNUM(value))
+ ANYOF_BITMAP_SET(data->start_class, value);
+ }
+ }
+ break;
+ case ALNUML:
+ if (flags & SCF_DO_STCLASS_AND) {
+ if (data->start_class->flags & ANYOF_LOCALE)
+ ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
+ }
+ else {
+ ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
+ data->start_class->flags |= ANYOF_LOCALE;
+ }
+ break;
+ case NALNUM:
+ if (flags & SCF_DO_STCLASS_AND) {
+ if (!(data->start_class->flags & ANYOF_LOCALE)) {
+ ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
+ for (value = 0; value < 256; value++)
+ if (isALNUM(value))
+ ANYOF_BITMAP_CLEAR(data->start_class, value);
+ }
+ }
+ else {
+ if (data->start_class->flags & ANYOF_LOCALE)
+ ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
+ else {
+ for (value = 0; value < 256; value++)
+ if (!isALNUM(value))
+ ANYOF_BITMAP_SET(data->start_class, value);
+ }
+ }
+ break;
+ case NALNUML:
+ if (flags & SCF_DO_STCLASS_AND) {
+ if (data->start_class->flags & ANYOF_LOCALE)
+ ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
+ }
+ else {
+ data->start_class->flags |= ANYOF_LOCALE;
+ ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
+ }
+ break;
+ case SPACE:
+ if (flags & SCF_DO_STCLASS_AND) {
+ if (!(data->start_class->flags & ANYOF_LOCALE)) {
+ ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
+ for (value = 0; value < 256; value++)
+ if (!isSPACE(value))
+ ANYOF_BITMAP_CLEAR(data->start_class, value);
+ }
+ }
+ else {
+ if (data->start_class->flags & ANYOF_LOCALE)
+ ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
+ else {
+ for (value = 0; value < 256; value++)
+ if (isSPACE(value))
+ ANYOF_BITMAP_SET(data->start_class, value);
+ }
+ }
+ break;
+ case SPACEL:
+ if (flags & SCF_DO_STCLASS_AND) {
+ if (data->start_class->flags & ANYOF_LOCALE)
+ ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
+ }
+ else {
+ data->start_class->flags |= ANYOF_LOCALE;
+ ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
+ }
+ break;
+ case NSPACE:
+ if (flags & SCF_DO_STCLASS_AND) {
+ if (!(data->start_class->flags & ANYOF_LOCALE)) {
+ ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
+ for (value = 0; value < 256; value++)
+ if (isSPACE(value))
+ ANYOF_BITMAP_CLEAR(data->start_class, value);
+ }
+ }
+ else {
+ if (data->start_class->flags & ANYOF_LOCALE)
+ ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
+ else {
+ for (value = 0; value < 256; value++)
+ if (!isSPACE(value))
+ ANYOF_BITMAP_SET(data->start_class, value);
+ }
+ }
+ break;
+ case NSPACEL:
+ if (flags & SCF_DO_STCLASS_AND) {
+ if (data->start_class->flags & ANYOF_LOCALE) {
+ ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
+ for (value = 0; value < 256; value++)
+ if (!isSPACE(value))
+ ANYOF_BITMAP_CLEAR(data->start_class, value);
+ }
+ }
+ else {
+ data->start_class->flags |= ANYOF_LOCALE;
+ ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
+ }
+ break;
+ case DIGIT:
+ if (flags & SCF_DO_STCLASS_AND) {
+ ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
+ for (value = 0; value < 256; value++)
+ if (!isDIGIT(value))
+ ANYOF_BITMAP_CLEAR(data->start_class, value);
+ }
+ else {
+ if (data->start_class->flags & ANYOF_LOCALE)
+ ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
+ else {
+ for (value = 0; value < 256; value++)
+ if (isDIGIT(value))
+ ANYOF_BITMAP_SET(data->start_class, value);
+ }
+ }
+ break;
+ case NDIGIT:
+ if (flags & SCF_DO_STCLASS_AND) {
+ ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
+ for (value = 0; value < 256; value++)
+ if (isDIGIT(value))
+ ANYOF_BITMAP_CLEAR(data->start_class, value);
+ }
+ else {
+ if (data->start_class->flags & ANYOF_LOCALE)
+ ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
+ else {
+ for (value = 0; value < 256; value++)
+ if (!isDIGIT(value))
+ ANYOF_BITMAP_SET(data->start_class, value);
+ }
+ }
+ break;
+ }
+ if (flags & SCF_DO_STCLASS_OR)
+ cl_and(data->start_class, and_withp);
+ flags &= ~SCF_DO_STCLASS;
}
- is_inf = is_inf_internal = 1;
- if (flags & SCF_DO_STCLASS_OR)
- cl_anything(pRExC_state, data->start_class);
- flags &= ~SCF_DO_STCLASS;
- break;
}
- }
- else if (strchr((const char*)PL_simple,OP(scan))) {
- int value = 0;
-
- if (flags & SCF_DO_SUBSTR) {
- scan_commit(pRExC_state,data,minlenp);
- data->pos_min++;
+ else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
+ data->flags |= (OP(scan) == MEOL
+ ? SF_BEFORE_MEOL
+ : SF_BEFORE_SEOL);
}
- min++;
- if (flags & SCF_DO_STCLASS) {
- data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
+ else if ( PL_regkind[OP(scan)] == BRANCHJ
+ /* Lookbehind, or need to calculate parens/evals/stclass: */
+ && (scan->flags || data || (flags & SCF_DO_STCLASS))
+ && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
+ if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
+ || OP(scan) == UNLESSM )
+ {
+ /* Negative Lookahead/lookbehind
+ In this case we can't do fixed string optimisation.
+ */
- /* Some of the logic below assumes that switching
- locale on will only add false positives. */
- switch (PL_regkind[OP(scan)]) {
- case SANY:
- default:
- do_default:
- /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
- if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
- cl_anything(pRExC_state, data->start_class);
- break;
- case REG_ANY:
- if (OP(scan) == SANY)
- goto do_default;
- if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
- value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
- || (data->start_class->flags & ANYOF_CLASS));
- cl_anything(pRExC_state, data->start_class);
+ I32 deltanext, minnext, fake = 0;
+ regnode *nscan;
+ struct regnode_charclass_class intrnl;
+ int f = 0;
+
+ data_fake.flags = 0;
+ if (data) {
+ data_fake.whilem_c = data->whilem_c;
+ data_fake.last_closep = data->last_closep;
}
- if (flags & SCF_DO_STCLASS_AND || !value)
- ANYOF_BITMAP_CLEAR(data->start_class,'\n');
- break;
- case ANYOF:
- if (flags & SCF_DO_STCLASS_AND)
- cl_and(data->start_class,
- (struct regnode_charclass_class*)scan);
else
- cl_or(pRExC_state, data->start_class,
- (struct regnode_charclass_class*)scan);
- break;
- case ALNUM:
- if (flags & SCF_DO_STCLASS_AND) {
- if (!(data->start_class->flags & ANYOF_LOCALE)) {
- ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
- for (value = 0; value < 256; value++)
- if (!isALNUM(value))
- ANYOF_BITMAP_CLEAR(data->start_class, value);
- }
- }
- else {
- if (data->start_class->flags & ANYOF_LOCALE)
- ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
- else {
- for (value = 0; value < 256; value++)
- if (isALNUM(value))
- ANYOF_BITMAP_SET(data->start_class, value);
- }
- }
- break;
- case ALNUML:
- if (flags & SCF_DO_STCLASS_AND) {
- if (data->start_class->flags & ANYOF_LOCALE)
- ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
- }
- else {
- ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
- data->start_class->flags |= ANYOF_LOCALE;
- }
- break;
- case NALNUM:
- if (flags & SCF_DO_STCLASS_AND) {
- if (!(data->start_class->flags & ANYOF_LOCALE)) {
- ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
- for (value = 0; value < 256; value++)
- if (isALNUM(value))
- ANYOF_BITMAP_CLEAR(data->start_class, value);
- }
- }
- else {
- if (data->start_class->flags & ANYOF_LOCALE)
- ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
- else {
- for (value = 0; value < 256; value++)
- if (!isALNUM(value))
- ANYOF_BITMAP_SET(data->start_class, value);
- }
- }
- break;
- case NALNUML:
- if (flags & SCF_DO_STCLASS_AND) {
- if (data->start_class->flags & ANYOF_LOCALE)
- ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
- }
- else {
- data->start_class->flags |= ANYOF_LOCALE;
- ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
+ data_fake.last_closep = &fake;
+ if ( flags & SCF_DO_STCLASS && !scan->flags
+ && OP(scan) == IFMATCH ) { /* Lookahead */
+ cl_init(pRExC_state, &intrnl);
+ data_fake.start_class = &intrnl;
+ f |= SCF_DO_STCLASS_AND;
}
- break;
- case SPACE:
- if (flags & SCF_DO_STCLASS_AND) {
- if (!(data->start_class->flags & ANYOF_LOCALE)) {
- ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
- for (value = 0; value < 256; value++)
- if (!isSPACE(value))
- ANYOF_BITMAP_CLEAR(data->start_class, value);
+ if (flags & SCF_WHILEM_VISITED_POS)
+ f |= SCF_WHILEM_VISITED_POS;
+ next = regnext(scan);
+ nscan = NEXTOPER(NEXTOPER(scan));
+ minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
+ last, &data_fake, stopparen, recursed, NULL, f, depth+1);
+ if (scan->flags) {
+ if (deltanext) {
+ vFAIL("Variable length lookbehind not implemented");
}
- }
- else {
- if (data->start_class->flags & ANYOF_LOCALE)
- ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
- else {
- for (value = 0; value < 256; value++)
- if (isSPACE(value))
- ANYOF_BITMAP_SET(data->start_class, value);
+ else if (minnext > (I32)U8_MAX) {
+ vFAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
}
+ scan->flags = (U8)minnext;
}
- break;
- case SPACEL:
- if (flags & SCF_DO_STCLASS_AND) {
- if (data->start_class->flags & ANYOF_LOCALE)
- ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
+ if (data) {
+ if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
+ pars++;
+ if (data_fake.flags & SF_HAS_EVAL)
+ data->flags |= SF_HAS_EVAL;
+ data->whilem_c = data_fake.whilem_c;
}
- else {
- data->start_class->flags |= ANYOF_LOCALE;
- ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
+ if (f & SCF_DO_STCLASS_AND) {
+ const int was = (data->start_class->flags & ANYOF_EOS);
+
+ cl_and(data->start_class, &intrnl);
+ if (was)
+ data->start_class->flags |= ANYOF_EOS;
}
- break;
- case NSPACE:
- if (flags & SCF_DO_STCLASS_AND) {
- if (!(data->start_class->flags & ANYOF_LOCALE)) {
- ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
- for (value = 0; value < 256; value++)
- if (isSPACE(value))
- ANYOF_BITMAP_CLEAR(data->start_class, value);
+ }
+#if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
+ else {
+ /* Positive Lookahead/lookbehind
+ In this case we can do fixed string optimisation,
+ but we must be careful about it. Note in the case of
+ lookbehind the positions will be offset by the minimum
+ length of the pattern, something we won't know about
+ until after the recurse.
+ */
+ I32 deltanext, fake = 0;
+ regnode *nscan;
+ struct regnode_charclass_class intrnl;
+ int f = 0;
+ /* We use SAVEFREEPV so that when the full compile
+ is finished perl will clean up the allocated
+ minlens when its all done. This was we don't
+ have to worry about freeing them when we know
+ they wont be used, which would be a pain.
+ */
+ I32 *minnextp;
+ Newx( minnextp, 1, I32 );
+ SAVEFREEPV(minnextp);
+
+ if (data) {
+ StructCopy(data, &data_fake, scan_data_t);
+ if ((flags & SCF_DO_SUBSTR) && data->last_found) {
+ f |= SCF_DO_SUBSTR;
+ if (scan->flags)
+ scan_commit(pRExC_state, &data_fake,minlenp);
+ data_fake.last_found=newSVsv(data->last_found);
}
}
- else {
- if (data->start_class->flags & ANYOF_LOCALE)
- ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
- else {
- for (value = 0; value < 256; value++)
- if (!isSPACE(value))
- ANYOF_BITMAP_SET(data->start_class, value);
- }
+ else
+ data_fake.last_closep = &fake;
+ data_fake.flags = 0;
+ if (is_inf)
+ data_fake.flags |= SF_IS_INF;
+ if ( flags & SCF_DO_STCLASS && !scan->flags
+ && OP(scan) == IFMATCH ) { /* Lookahead */
+ cl_init(pRExC_state, &intrnl);
+ data_fake.start_class = &intrnl;
+ f |= SCF_DO_STCLASS_AND;
}
- break;
- case NSPACEL:
- if (flags & SCF_DO_STCLASS_AND) {
- if (data->start_class->flags & ANYOF_LOCALE) {
- ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
- for (value = 0; value < 256; value++)
- if (!isSPACE(value))
- ANYOF_BITMAP_CLEAR(data->start_class, value);
+ if (flags & SCF_WHILEM_VISITED_POS)
+ f |= SCF_WHILEM_VISITED_POS;
+ next = regnext(scan);
+ nscan = NEXTOPER(NEXTOPER(scan));
+
+ *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
+ last, &data_fake, stopparen, recursed, NULL, f,depth+1);
+ if (scan->flags) {
+ if (deltanext) {
+ vFAIL("Variable length lookbehind not implemented");
}
- }
- else {
- data->start_class->flags |= ANYOF_LOCALE;
- ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
- }
- break;
- case DIGIT:
- if (flags & SCF_DO_STCLASS_AND) {
- ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
- for (value = 0; value < 256; value++)
- if (!isDIGIT(value))
- ANYOF_BITMAP_CLEAR(data->start_class, value);
- }
- else {
- if (data->start_class->flags & ANYOF_LOCALE)
- ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
- else {
- for (value = 0; value < 256; value++)
- if (isDIGIT(value))
- ANYOF_BITMAP_SET(data->start_class, value);
+ else if (*minnextp > (I32)U8_MAX) {
+ vFAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
}
+ scan->flags = (U8)*minnextp;
}
- break;
- case NDIGIT:
- if (flags & SCF_DO_STCLASS_AND) {
- ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
- for (value = 0; value < 256; value++)
- if (isDIGIT(value))
- ANYOF_BITMAP_CLEAR(data->start_class, value);
+
+ *minnextp += min;
+
+ if (f & SCF_DO_STCLASS_AND) {
+ const int was = (data->start_class->flags & ANYOF_EOS);
+
+ cl_and(data->start_class, &intrnl);
+ if (was)
+ data->start_class->flags |= ANYOF_EOS;
}
- else {
- if (data->start_class->flags & ANYOF_LOCALE)
- ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
- else {
- for (value = 0; value < 256; value++)
- if (!isDIGIT(value))
- ANYOF_BITMAP_SET(data->start_class, value);
+ if (data) {
+ if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
+ pars++;
+ if (data_fake.flags & SF_HAS_EVAL)
+ data->flags |= SF_HAS_EVAL;
+ data->whilem_c = data_fake.whilem_c;
+ if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
+ if (RExC_rx->minlen<*minnextp)
+ RExC_rx->minlen=*minnextp;
+ scan_commit(pRExC_state, &data_fake, minnextp);
+ SvREFCNT_dec(data_fake.last_found);
+
+ if ( data_fake.minlen_fixed != minlenp )
+ {
+ data->offset_fixed= data_fake.offset_fixed;
+ data->minlen_fixed= data_fake.minlen_fixed;
+ data->lookbehind_fixed+= scan->flags;
+ }
+ if ( data_fake.minlen_float != minlenp )
+ {
+ data->minlen_float= data_fake.minlen_float;
+ data->offset_float_min=data_fake.offset_float_min;
+ data->offset_float_max=data_fake.offset_float_max;
+ data->lookbehind_float+= scan->flags;
+ }
}
}
- break;
- }
- if (flags & SCF_DO_STCLASS_OR)
- cl_and(data->start_class, &and_with);
- flags &= ~SCF_DO_STCLASS;
- }
- }
- else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
- data->flags |= (OP(scan) == MEOL
- ? SF_BEFORE_MEOL
- : SF_BEFORE_SEOL);
- }
- else if ( PL_regkind[OP(scan)] == BRANCHJ
- /* Lookbehind, or need to calculate parens/evals/stclass: */
- && (scan->flags || data || (flags & SCF_DO_STCLASS))
- && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
- if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
- || OP(scan) == UNLESSM )
- {
- /* Negative Lookahead/lookbehind
- In this case we can't do fixed string optimisation.
- */
- I32 deltanext, minnext, fake = 0;
- regnode *nscan;
- struct regnode_charclass_class intrnl;
- int f = 0;
- data_fake.flags = 0;
- if (data) {
- data_fake.whilem_c = data->whilem_c;
- data_fake.last_closep = data->last_closep;
- }
- else
- data_fake.last_closep = &fake;
- if ( flags & SCF_DO_STCLASS && !scan->flags
- && OP(scan) == IFMATCH ) { /* Lookahead */
- cl_init(pRExC_state, &intrnl);
- data_fake.start_class = &intrnl;
- f |= SCF_DO_STCLASS_AND;
}
- if (flags & SCF_WHILEM_VISITED_POS)
- f |= SCF_WHILEM_VISITED_POS;
- next = regnext(scan);
- nscan = NEXTOPER(NEXTOPER(scan));
- minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext, last, &data_fake, f,depth+1);
- if (scan->flags) {
- if (deltanext) {
- vFAIL("Variable length lookbehind not implemented");
- }
- else if (minnext > (I32)U8_MAX) {
- vFAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
- }
- scan->flags = (U8)minnext;
- }
- if (data) {
- if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
- pars++;
- if (data_fake.flags & SF_HAS_EVAL)
- data->flags |= SF_HAS_EVAL;
- data->whilem_c = data_fake.whilem_c;
- }
- if (f & SCF_DO_STCLASS_AND) {
- const int was = (data->start_class->flags & ANYOF_EOS);
-
- cl_and(data->start_class, &intrnl);
- if (was)
- data->start_class->flags |= ANYOF_EOS;
- }
+#endif
}
-#if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
- else {
- /* Positive Lookahead/lookbehind
- In this case we can do fixed string optimisation,
- but we must be careful about it. Note in the case of
- lookbehind the positions will be offset by the minimum
- length of the pattern, something we won't know about
- until after the recurse.
- */
- I32 deltanext, fake = 0;
- regnode *nscan;
- struct regnode_charclass_class intrnl;
- int f = 0;
- /* We use SAVEFREEPV so that when the full compile
- is finished perl will clean up the allocated
- minlens when its all done. This was we don't
- have to worry about freeing them when we know
- they wont be used, which would be a pain.
- */
- I32 *minnextp;
- Newx( minnextp, 1, I32 );
- SAVEFREEPV(minnextp);
-
- if (data) {
- StructCopy(data, &data_fake, scan_data_t);
- if ((flags & SCF_DO_SUBSTR) && data->last_found) {
- f |= SCF_DO_SUBSTR;
- if (scan->flags)
- scan_commit(pRExC_state, &data_fake,minlenp);
- data_fake.last_found=newSVsv(data->last_found);
- }
- }
- else
- data_fake.last_closep = &fake;
- data_fake.flags = 0;
- if (is_inf)
- data_fake.flags |= SF_IS_INF;
- if ( flags & SCF_DO_STCLASS && !scan->flags
- && OP(scan) == IFMATCH ) { /* Lookahead */
- cl_init(pRExC_state, &intrnl);
- data_fake.start_class = &intrnl;
- f |= SCF_DO_STCLASS_AND;
- }
- if (flags & SCF_WHILEM_VISITED_POS)
- f |= SCF_WHILEM_VISITED_POS;
- next = regnext(scan);
- nscan = NEXTOPER(NEXTOPER(scan));
-
- *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext, last, &data_fake, f,depth+1);
- if (scan->flags) {
- if (deltanext) {
- vFAIL("Variable length lookbehind not implemented");
- }
- else if (*minnextp > (I32)U8_MAX) {
- vFAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
- }
- scan->flags = (U8)*minnextp;
- }
-
- *minnextp += min;
-
-
- if (f & SCF_DO_STCLASS_AND) {
- const int was = (data->start_class->flags & ANYOF_EOS);
-
- cl_and(data->start_class, &intrnl);
- if (was)
- data->start_class->flags |= ANYOF_EOS;
- }
- if (data) {
- if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
- pars++;
- if (data_fake.flags & SF_HAS_EVAL)
- data->flags |= SF_HAS_EVAL;
- data->whilem_c = data_fake.whilem_c;
- if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
- if (RExC_rx->minlen<*minnextp)
- RExC_rx->minlen=*minnextp;
- scan_commit(pRExC_state, &data_fake, minnextp);
- SvREFCNT_dec(data_fake.last_found);
-
- if ( data_fake.minlen_fixed != minlenp )
- {
- data->offset_fixed= data_fake.offset_fixed;
- data->minlen_fixed= data_fake.minlen_fixed;
- data->lookbehind_fixed+= scan->flags;
- }
- if ( data_fake.minlen_float != minlenp )
- {
- data->minlen_float= data_fake.minlen_float;
- data->offset_float_min=data_fake.offset_float_min;
- data->offset_float_max=data_fake.offset_float_max;
- data->lookbehind_float+= scan->flags;
- }
- }
- }
-
-
+ else if (OP(scan) == OPEN) {
+ if (stopparen != (I32)ARG(scan))
+ pars++;
}
-#endif
- }
- else if (OP(scan) == OPEN) {
- pars++;
- }
- else if (OP(scan) == CLOSE) {
- if ((I32)ARG(scan) == is_par) {
- next = regnext(scan);
+ else if (OP(scan) == CLOSE) {
+ if (stopparen == (I32)ARG(scan)) {
+ break;
+ }
+ if ((I32)ARG(scan) == is_par) {
+ next = regnext(scan);
- if ( next && (OP(next) != WHILEM) && next < last)
- is_par = 0; /* Disable optimization */
+ if ( next && (OP(next) != WHILEM) && next < last)
+ is_par = 0; /* Disable optimization */
+ }
+ if (data)
+ *(data->last_closep) = ARG(scan);
}
- if (data)
- *(data->last_closep) = ARG(scan);
- }
- else if (OP(scan) == EVAL) {
+ else if (OP(scan) == EVAL) {
if (data)
data->flags |= SF_HAS_EVAL;
- }
- else if ( (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
- || OP(scan)==RECURSE) /* recursion */
- {
- if (OP(scan)==RECURSE) {
- ARG2L_SET( scan, RExC_parens[ARG(scan)-1] - scan );
- }
+ }
+ else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
+ if (flags & SCF_DO_SUBSTR) {
+ scan_commit(pRExC_state,data,minlenp);
+ flags &= ~SCF_DO_SUBSTR;
+ }
+ if (data && OP(scan)==ACCEPT) {
+ data->flags |= SCF_SEEN_ACCEPT;
+ if (stopmin > min)
+ stopmin = min;
+ }
+ }
+ else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
+ {
if (flags & SCF_DO_SUBSTR) {
scan_commit(pRExC_state,data,minlenp);
data->longest = &(data->longest_float);
if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
cl_anything(pRExC_state, data->start_class);
flags &= ~SCF_DO_STCLASS;
- }
+ }
#ifdef TRIE_STUDY_OPT
-#ifdef FULL_TRIE_STUDY
- else if (PL_regkind[OP(scan)] == TRIE) {
- /* NOTE - There is similar code to this block above for handling
- BRANCH nodes on the initial study. If you change stuff here
- check there too. */
- regnode *tail= regnext(scan);
- reg_trie_data *trie = (reg_trie_data*)RExC_rx->data->data[ ARG(scan) ];
- I32 max1 = 0, min1 = I32_MAX;
- struct regnode_charclass_class accum;
-
- if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
- scan_commit(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
- if (flags & SCF_DO_STCLASS)
- cl_init_zero(pRExC_state, &accum);
-
- if (!trie->jump) {
- min1= trie->minlen;
- max1= trie->maxlen;
- } else {
- const regnode *nextbranch= NULL;
- U32 word;
-
- for ( word=1 ; word <= trie->wordcount ; word++)
- {
- I32 deltanext=0, minnext=0, f = 0, fake;
- struct regnode_charclass_class this_class;
-
- data_fake.flags = 0;
- if (data) {
- data_fake.whilem_c = data->whilem_c;
- data_fake.last_closep = data->last_closep;
- }
- else
- data_fake.last_closep = &fake;
-
- if (flags & SCF_DO_STCLASS) {
- cl_init(pRExC_state, &this_class);
- data_fake.start_class = &this_class;
- f = SCF_DO_STCLASS_AND;
- }
- if (flags & SCF_WHILEM_VISITED_POS)
- f |= SCF_WHILEM_VISITED_POS;
-
- if (trie->jump[word]) {
- if (!nextbranch)
- nextbranch = tail - trie->jump[0];
- scan= tail - trie->jump[word];
- /* We go from the jump point to the branch that follows
- it. Note this means we need the vestigal unused branches
- even though they arent otherwise used.
- */
- minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
- (regnode *)nextbranch, &data_fake, f,depth+1);
- }
- if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
- nextbranch= regnext((regnode*)nextbranch);
-
- if (min1 > (I32)(minnext + trie->minlen))
- min1 = minnext + trie->minlen;
- if (max1 < (I32)(minnext + deltanext + trie->maxlen))
- max1 = minnext + deltanext + trie->maxlen;
- if (deltanext == I32_MAX)
- is_inf = is_inf_internal = 1;
-
- if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
- pars++;
-
- if (data) {
- if (data_fake.flags & SF_HAS_EVAL)
- data->flags |= SF_HAS_EVAL;
- data->whilem_c = data_fake.whilem_c;
- }
- if (flags & SCF_DO_STCLASS)
- cl_or(pRExC_state, &accum, &this_class);
- }
- }
- 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;
- if (flags & SCF_DO_STCLASS_OR) {
- cl_or(pRExC_state, data->start_class, &accum);
- if (min1) {
- cl_and(data->start_class, &and_with);
- flags &= ~SCF_DO_STCLASS;
- }
- }
- else if (flags & SCF_DO_STCLASS_AND) {
- if (min1) {
- cl_and(data->start_class, &accum);
- flags &= ~SCF_DO_STCLASS;
- }
- else {
- /* Switch to OR mode: cache the old value of
- * data->start_class */
- StructCopy(data->start_class, &and_with,
- struct regnode_charclass_class);
- flags &= ~SCF_DO_STCLASS_AND;
- StructCopy(&accum, data->start_class,
- struct regnode_charclass_class);
- flags |= SCF_DO_STCLASS_OR;
- data->start_class->flags |= ANYOF_EOS;
- }
- }
- scan= tail;
- continue;
- }
+#ifdef FULL_TRIE_STUDY
+ else if (PL_regkind[OP(scan)] == TRIE) {
+ /* NOTE - There is similar code to this block above for handling
+ BRANCH nodes on the initial study. If you change stuff here
+ check there too. */
+ regnode *trie_node= scan;
+ regnode *tail= regnext(scan);
+ reg_trie_data *trie = (reg_trie_data*)RExC_rx->data->data[ ARG(scan) ];
+ I32 max1 = 0, min1 = I32_MAX;
+ struct regnode_charclass_class accum;
+
+ if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
+ scan_commit(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
+ if (flags & SCF_DO_STCLASS)
+ cl_init_zero(pRExC_state, &accum);
+
+ if (!trie->jump) {
+ min1= trie->minlen;
+ max1= trie->maxlen;
+ } else {
+ const regnode *nextbranch= NULL;
+ U32 word;
+
+ for ( word=1 ; word <= trie->wordcount ; word++)
+ {
+ I32 deltanext=0, minnext=0, f = 0, fake;
+ struct regnode_charclass_class this_class;
+
+ data_fake.flags = 0;
+ if (data) {
+ data_fake.whilem_c = data->whilem_c;
+ data_fake.last_closep = data->last_closep;
+ }
+ else
+ data_fake.last_closep = &fake;
+
+ if (flags & SCF_DO_STCLASS) {
+ cl_init(pRExC_state, &this_class);
+ data_fake.start_class = &this_class;
+ f = SCF_DO_STCLASS_AND;
+ }
+ if (flags & SCF_WHILEM_VISITED_POS)
+ f |= SCF_WHILEM_VISITED_POS;
+
+ if (trie->jump[word]) {
+ if (!nextbranch)
+ nextbranch = trie_node + trie->jump[0];
+ scan= trie_node + trie->jump[word];
+ /* We go from the jump point to the branch that follows
+ it. Note this means we need the vestigal unused branches
+ even though they arent otherwise used.
+ */
+ minnext = study_chunk(pRExC_state, &scan, minlenp,
+ &deltanext, (regnode *)nextbranch, &data_fake,
+ stopparen, recursed, NULL, f,depth+1);
+ }
+ if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
+ nextbranch= regnext((regnode*)nextbranch);
+
+ if (min1 > (I32)(minnext + trie->minlen))
+ min1 = minnext + trie->minlen;
+ if (max1 < (I32)(minnext + deltanext + trie->maxlen))
+ max1 = minnext + deltanext + trie->maxlen;
+ if (deltanext == I32_MAX)
+ is_inf = is_inf_internal = 1;
+
+ if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
+ pars++;
+ if (data_fake.flags & SCF_SEEN_ACCEPT) {
+ if ( stopmin > min + min1)
+ stopmin = min + min1;
+ flags &= ~SCF_DO_SUBSTR;
+ if (data)
+ data->flags |= SCF_SEEN_ACCEPT;
+ }
+ if (data) {
+ if (data_fake.flags & SF_HAS_EVAL)
+ data->flags |= SF_HAS_EVAL;
+ data->whilem_c = data_fake.whilem_c;
+ }
+ if (flags & SCF_DO_STCLASS)
+ cl_or(pRExC_state, &accum, &this_class);
+ }
+ }
+ 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;
+ if (flags & SCF_DO_STCLASS_OR) {
+ cl_or(pRExC_state, data->start_class, &accum);
+ if (min1) {
+ cl_and(data->start_class, and_withp);
+ flags &= ~SCF_DO_STCLASS;
+ }
+ }
+ else if (flags & SCF_DO_STCLASS_AND) {
+ if (min1) {
+ cl_and(data->start_class, &accum);
+ flags &= ~SCF_DO_STCLASS;
+ }
+ else {
+ /* Switch to OR mode: cache the old value of
+ * data->start_class */
+ INIT_AND_WITHP;
+ StructCopy(data->start_class, and_withp,
+ struct regnode_charclass_class);
+ flags &= ~SCF_DO_STCLASS_AND;
+ StructCopy(&accum, data->start_class,
+ struct regnode_charclass_class);
+ flags |= SCF_DO_STCLASS_OR;
+ data->start_class->flags |= ANYOF_EOS;
+ }
+ }
+ scan= tail;
+ continue;
+ }
#else
- else if (PL_regkind[OP(scan)] == TRIE) {
- reg_trie_data *trie = (reg_trie_data*)RExC_rx->data->data[ ARG(scan) ];
- U8*bang=NULL;
-
- min += trie->minlen;
- delta += (trie->maxlen - trie->minlen);
- flags &= ~SCF_DO_STCLASS; /* xxx */
- if (flags & SCF_DO_SUBSTR) {
- scan_commit(pRExC_state,data,minlenp); /* Cannot expect anything... */
- data->pos_min += trie->minlen;
- data->pos_delta += (trie->maxlen - trie->minlen);
- if (trie->maxlen != trie->minlen)
- data->longest = &(data->longest_float);
- }
- if (trie->jump) /* no more substrings -- for now /grr*/
- flags &= ~SCF_DO_SUBSTR;
- }
+ else if (PL_regkind[OP(scan)] == TRIE) {
+ reg_trie_data *trie = (reg_trie_data*)RExC_rx->data->data[ ARG(scan) ];
+ U8*bang=NULL;
+
+ min += trie->minlen;
+ delta += (trie->maxlen - trie->minlen);
+ flags &= ~SCF_DO_STCLASS; /* xxx */
+ if (flags & SCF_DO_SUBSTR) {
+ scan_commit(pRExC_state,data,minlenp); /* Cannot expect anything... */
+ data->pos_min += trie->minlen;
+ data->pos_delta += (trie->maxlen - trie->minlen);
+ if (trie->maxlen != trie->minlen)
+ data->longest = &(data->longest_float);
+ }
+ if (trie->jump) /* no more substrings -- for now /grr*/
+ flags &= ~SCF_DO_SUBSTR;
+ }
#endif /* old or new */
-#endif /* TRIE_STUDY_OPT */
- /* Else: zero-length, ignore. */
- scan = regnext(scan);
+#endif /* TRIE_STUDY_OPT */
+ /* Else: zero-length, ignore. */
+ scan = regnext(scan);
+ }
+ DEBUG_PEEP("FEND",scan,depth);
+ scan = frame->next;
+ stopparen = frame->stop;
+ frame = frame->prev;
+
}
finish:
data->flags &= ~SF_IN_PAR;
}
if (flags & SCF_DO_STCLASS_OR)
- cl_and(data->start_class, &and_with);
+ cl_and(data->start_class, and_withp);
if (flags & SCF_TRIE_RESTUDY)
data->flags |= SCF_TRIE_RESTUDY;
DEBUG_STUDYDATA(data,depth);
- return min;
+ return min < stopmin ? min : stopmin;
}
STATIC I32
S_add_data(RExC_state_t *pRExC_state, I32 n, const char *s)
{
if (RExC_rx->data) {
+ const U32 count = RExC_rx->data->count;
Renewc(RExC_rx->data,
- sizeof(*RExC_rx->data) + sizeof(void*) * (RExC_rx->data->count + n - 1),
+ sizeof(*RExC_rx->data) + sizeof(void*) * (count + n - 1),
char, struct reg_data);
- Renew(RExC_rx->data->what, RExC_rx->data->count + n, U8);
+ Renew(RExC_rx->data->what, count + n, U8);
RExC_rx->data->count += n;
}
else {
* Beware that the optimization-preparation code in here knows about some
* of the structure of the compiled regexp. [I'll say.]
*/
+
+
+
#ifndef PERL_IN_XSUB_RE
-#define CORE_ONLY_BLOCK(c) {c}{
#define RE_ENGINE_PTR &PL_core_reg_engine
#else
-#define CORE_ONLY_BLOCK(c) {
extern const struct regexp_engine my_reg_engine;
#define RE_ENGINE_PTR &my_reg_engine
#endif
+/* these make a few things look better, to avoid indentation */
+#define BEGIN_BLOCK {
#define END_BLOCK }
regexp *
dVAR;
GET_RE_DEBUG_FLAGS_DECL;
DEBUG_r(if (!PL_colorset) reginitcolors());
- CORE_ONLY_BLOCK(
+#ifndef PERL_IN_XSUB_RE
+ BEGIN_BLOCK
/* Dispatch a request to compile a regexp to correct
regexp engine. */
HV * const table = GvHV(PL_hintgv);
if (table) {
SV **ptr= hv_fetchs(table, "regcomp", FALSE);
- if (ptr && SvIOK(*ptr)) {
+ if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
DEBUG_COMPILE_r({
PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
});
return CALLREGCOMP_ENG(eng, exp, xend, pm);
}
- })
+ }
+ END_BLOCK
+#endif
+ BEGIN_BLOCK
register regexp *r;
regnode *scan;
regnode *first;
RExC_end = xend;
RExC_naughty = 0;
RExC_npar = 1;
+ RExC_cpar = 1;
+ RExC_nestroot = 0;
RExC_size = 0L;
RExC_emit = &PL_regdummy;
RExC_whilem_seen = 0;
RExC_charnames = NULL;
- RExC_parens = NULL;
+ RExC_open_parens = NULL;
+ RExC_close_parens = NULL;
+ RExC_opend = NULL;
RExC_paren_names = NULL;
+ RExC_recurse = NULL;
+ RExC_recurse_count = 0;
#if 0 /* REGC() is (currently) a NOP at the first pass.
* Clever compilers notice this and complain. --jhi */
if (RExC_whilem_seen > 15)
RExC_whilem_seen = 15;
+#ifdef DEBUGGING
+ /* Make room for a sentinel value at the end of the program */
+ RExC_size++;
+#endif
+
/* Allocate space and zero-initialize. Note, the two step process
of zeroing when in debug mode, thus anything assigned has to
happen after that */
r->substrs = 0; /* Useful during FAIL. */
r->startp = 0; /* Useful during FAIL. */
r->endp = 0;
+ r->swap = NULL;
r->paren_names = 0;
if (RExC_seen & REG_SEEN_RECURSE) {
- Newx(RExC_parens, RExC_npar,regnode *);
- SAVEFREEPV(RExC_parens);
+ Newxz(RExC_open_parens, RExC_npar,regnode *);
+ SAVEFREEPV(RExC_open_parens);
+ Newxz(RExC_close_parens,RExC_npar,regnode *);
+ SAVEFREEPV(RExC_close_parens);
}
/* Useful during FAIL. */
RExC_end = xend;
RExC_naughty = 0;
RExC_npar = 1;
+ RExC_cpar = 1;
RExC_emit_start = r->program;
RExC_emit = r->program;
+#ifdef DEBUGGING
+ /* put a sentinal on the end of the program so we can check for
+ overwrites */
+ r->program[RExC_size].type = 255;
+#endif
/* Store the count of eval-groups for security checks: */
RExC_emit->next_off = (RExC_seen_evals > (I32)U16_MAX) ? U16_MAX : (U16)RExC_seen_evals;
REGC((U8)REG_MAGIC, (char*) RExC_emit++);
/* XXXX To minimize changes to RE engine we always allocate
3-units-long substrs field. */
Newx(r->substrs, 1, struct reg_substr_data);
+ if (RExC_recurse_count) {
+ Newxz(RExC_recurse,RExC_recurse_count,regnode *);
+ SAVEFREEPV(RExC_recurse);
+ }
reStudy:
r->minlen = minlen = sawplus = sawopen = 0;
Zero(r->substrs, 1, struct reg_substr_data);
- StructCopy(&zero_scan_data, &data, scan_data_t);
#ifdef TRIE_STUDY_OPT
if ( restudied ) {
+ U32 seen=RExC_seen;
DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
- RExC_state=copyRExC_state;
+
+ RExC_state = copyRExC_state;
+ if (seen & REG_TOP_LEVEL_BRANCHES)
+ RExC_seen |= REG_TOP_LEVEL_BRANCHES;
+ else
+ RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
if (data.last_found) {
SvREFCNT_dec(data.longest_fixed);
SvREFCNT_dec(data.longest_float);
SvREFCNT_dec(data.last_found);
}
+ StructCopy(&zero_scan_data, &data, scan_data_t);
} else {
- copyRExC_state=RExC_state;
+ StructCopy(&zero_scan_data, &data, scan_data_t);
+ copyRExC_state = RExC_state;
}
+#else
+ StructCopy(&zero_scan_data, &data, scan_data_t);
#endif
/* Dig out information for optimizations. */
first = NEXTOPER(first);
goto again;
}
- else if (!sawopen && (OP(first) == STAR &&
+ else if ((!sawopen || !RExC_sawback) &&
+ (OP(first) == STAR &&
PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
!(r->reganch & ROPT_ANCH) )
{
* it happens that c_offset_min has been invalidated, since the
* earlier string may buy us something the later one won't.]
*/
- minlen = 0;
-
+
data.longest_fixed = newSVpvs("");
data.longest_float = newSVpvs("");
data.last_found = newSVpvs("");
} else /* XXXX Check for BOUND? */
stclass_flag = 0;
data.last_closep = &last_close;
-
+
minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
- &data, SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
+ &data, -1, NULL, NULL,
+ SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
CHECK_RESTUDY_GOTO;
struct regnode_charclass_class ch_class;
I32 last_close = 0;
- DEBUG_COMPILE_r(PerlIO_printf(Perl_debug_log, "\n"));
+ DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
scan = r->program + 1;
cl_init(pRExC_state, &ch_class);
data.start_class = &ch_class;
data.last_closep = &last_close;
+
minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
- &data, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
-
+ &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
+
CHECK_RESTUDY_GOTO;
r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
/* Guard against an embedded (?=) or (?<=) with a longer minlen than
the "real" pattern. */
+ DEBUG_OPTIMISE_r({
+ PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
+ minlen, r->minlen);
+ });
+ r->minlenret = minlen;
if (r->minlen < minlen)
r->minlen = minlen;
r->reganch |= ROPT_EVAL_SEEN;
if (RExC_seen & REG_SEEN_CANY)
r->reganch |= ROPT_CANY_SEEN;
+ if (RExC_seen & REG_SEEN_VERBARG)
+ r->reganch |= ROPT_VERBARG_SEEN;
+ if (RExC_seen & REG_SEEN_CUTGROUP)
+ r->reganch |= ROPT_CUTGROUP_SEEN;
if (RExC_paren_names)
r->paren_names = (HV*)SvREFCNT_inc(RExC_paren_names);
else
r->paren_names = NULL;
+ if (RExC_recurse_count) {
+ for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
+ const regnode *scan = RExC_recurse[RExC_recurse_count-1];
+ ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
+ }
+ }
Newxz(r->startp, RExC_npar, I32);
Newxz(r->endp, RExC_npar, I32);
-
- DEBUG_r( RX_DEBUG_on(r) );
+ /* assume we don't need to swap parens around before we match */
+
DEBUG_DUMP_r({
PerlIO_printf(Perl_debug_log,"Final program:\n");
regdump(r);
#undef END_BLOCK
#undef RE_ENGINE_PTR
+#ifndef PERL_IN_XSUB_RE
SV*
Perl_reg_named_buff_sv(pTHX_ SV* namesv)
{
return GvSVn(gv_paren);
}
}
-
+#endif
/* Scans the name of a named buffer from the pattern.
* If flags is REG_RSN_RETURN_NULL returns null.
#define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
#endif
+/* this idea is borrowed from STR_WITH_LEN in handy.h */
+#define CHECK_WORD(s,v,l) \
+ (((sizeof(s)-1)==(l)) && (strnEQ(start_verb, (s ""), (sizeof(s)-1))))
+
STATIC regnode *
S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
/* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
/* Make an OPEN node, if parenthesized. */
if (paren) {
+ if ( *RExC_parse == '*') { /* (*VERB:ARG) */
+ char *start_verb = RExC_parse;
+ STRLEN verb_len = 0;
+ char *start_arg = NULL;
+ unsigned char op = 0;
+ int argok = 1;
+ int internal_argval = 0; /* internal_argval is only useful if !argok */
+ while ( *RExC_parse && *RExC_parse != ')' ) {
+ if ( *RExC_parse == ':' ) {
+ start_arg = RExC_parse + 1;
+ break;
+ }
+ RExC_parse++;
+ }
+ ++start_verb;
+ verb_len = RExC_parse - start_verb;
+ if ( start_arg ) {
+ RExC_parse++;
+ while ( *RExC_parse && *RExC_parse != ')' )
+ RExC_parse++;
+ if ( *RExC_parse != ')' )
+ vFAIL("Unterminated verb pattern argument");
+ if ( RExC_parse == start_arg )
+ start_arg = NULL;
+ } else {
+ if ( *RExC_parse != ')' )
+ vFAIL("Unterminated verb pattern");
+ }
+
+ switch ( *start_verb ) {
+ case 'A': /* (*ACCEPT) */
+ if ( CHECK_WORD("ACCEPT",start_verb,verb_len) ) {
+ op = ACCEPT;
+ internal_argval = RExC_nestroot;
+ }
+ break;
+ case 'C': /* (*COMMIT) */
+ if ( CHECK_WORD("COMMIT",start_verb,verb_len) )
+ op = COMMIT;
+ break;
+ case 'F': /* (*FAIL) */
+ if ( verb_len==1 || CHECK_WORD("FAIL",start_verb,verb_len) ) {
+ op = OPFAIL;
+ argok = 0;
+ }
+ break;
+ case ':': /* (*:NAME) */
+ case 'M': /* (*MARK:NAME) */
+ if ( verb_len==0 || CHECK_WORD("MARK",start_verb,verb_len) ) {
+ op = MARKPOINT;
+ argok = -1;
+ }
+ break;
+ case 'P': /* (*PRUNE) */
+ if ( CHECK_WORD("PRUNE",start_verb,verb_len) )
+ op = PRUNE;
+ break;
+ case 'S': /* (*SKIP) */
+ if ( CHECK_WORD("SKIP",start_verb,verb_len) )
+ op = SKIP;
+ break;
+ case 'T': /* (*THEN) */
+ /* [19:06] <TimToady> :: is then */
+ if ( CHECK_WORD("THEN",start_verb,verb_len) ) {
+ op = CUTGROUP;
+ RExC_seen |= REG_SEEN_CUTGROUP;
+ }
+ break;
+ }
+ if ( ! op ) {
+ RExC_parse++;
+ vFAIL3("Unknown verb pattern '%.*s'",
+ verb_len, start_verb);
+ }
+ if ( argok ) {
+ if ( start_arg && internal_argval ) {
+ vFAIL3("Verb pattern '%.*s' may not have an argument",
+ verb_len, start_verb);
+ } else if ( argok < 0 && !start_arg ) {
+ vFAIL3("Verb pattern '%.*s' has a mandatory argument",
+ verb_len, start_verb);
+ } else {
+ ret = reganode(pRExC_state, op, internal_argval);
+ if ( ! internal_argval && ! SIZE_ONLY ) {
+ if (start_arg) {
+ SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
+ ARG(ret) = add_data( pRExC_state, 1, "S" );
+ RExC_rx->data->data[ARG(ret)]=(void*)sv;
+ ret->flags = 0;
+ } else {
+ ret->flags = 1;
+ }
+ }
+ }
+ if (!internal_argval)
+ RExC_seen |= REG_SEEN_VERBARG;
+ } else if ( start_arg ) {
+ vFAIL3("Verb pattern '%.*s' may not have an argument",
+ verb_len, start_verb);
+ } else {
+ ret = reg_node(pRExC_state, op);
+ }
+ nextchar(pRExC_state);
+ return ret;
+ } else
if (*RExC_parse == '?') { /* (?...) */
U32 posflags = 0, negflags = 0;
U32 *flagsp = &posflags;
case '=': /* (?=...) */
case '!': /* (?!...) */
RExC_seen_zerolen++;
+ if (*RExC_parse == ')') {
+ ret=reg_node(pRExC_state, OPFAIL);
+ nextchar(pRExC_state);
+ return ret;
+ }
case ':': /* (?:...) */
case '>': /* (?>...) */
break;
case 'R' : /* (?R) */
if (*RExC_parse != ')')
FAIL("Sequence (?R) not terminated");
- reg_node(pRExC_state, SRECURSE);
- break; /* (?PARNO) */
+ ret = reg_node(pRExC_state, GOSTART);
+ nextchar(pRExC_state);
+ return ret;
+ /*notreached*/
{ /* named and numeric backreferences */
I32 num;
char * parse_start;
}
goto gen_recurse_regop;
/* NOT REACHED */
+ case '+':
+ if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
+ RExC_parse++;
+ vFAIL("Illegal pattern");
+ }
+ goto parse_recursion;
+ /* NOT REACHED*/
+ case '-': /* (?-1) */
+ if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
+ RExC_parse--; /* rewind to let it be handled later */
+ goto parse_flags;
+ }
+ /*FALLTHROUGH */
case '1': case '2': case '3': case '4': /* (?1) */
case '5': case '6': case '7': case '8': case '9':
RExC_parse--;
+ parse_recursion:
num = atoi(RExC_parse);
parse_start = RExC_parse - 1; /* MJD */
+ if (*RExC_parse == '-')
+ RExC_parse++;
while (isDIGIT(*RExC_parse))
RExC_parse++;
if (*RExC_parse!=')')
vFAIL("Expecting close bracket");
gen_recurse_regop:
- ret = reganode(pRExC_state, RECURSE, num);
+ if ( paren == '-' ) {
+ /*
+ Diagram of capture buffer numbering.
+ Top line is the normal capture buffer numbers
+ Botton line is the negative indexing as from
+ the X (the (?-2))
+
+ + 1 2 3 4 5 X 6 7
+ /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
+ - 5 4 3 2 1 X x x
+
+ */
+ num = RExC_npar + num;
+ if (num < 1) {
+ RExC_parse++;
+ vFAIL("Reference to nonexistent group");
+ }
+ } else if ( paren == '+' ) {
+ num = RExC_npar + num - 1;
+ }
+
+ ret = reganode(pRExC_state, GOSUB, num);
if (!SIZE_ONLY) {
if (num > (I32)RExC_rx->nparens) {
RExC_parse++;
vFAIL("Reference to nonexistent group");
}
- ARG2L_SET( ret, 0);
+ ARG2L_SET( ret, RExC_recurse_count++);
RExC_emit++;
DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
- "Recurse #%"UVuf" to %"IVdf"\n", ARG(ret), ARG2L(ret)));
+ "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
} else {
RExC_size++;
}
SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
}
- ret = reganode(pRExC_state,RECURSEP,parno);
+ ret = reganode(pRExC_state,INSUBP,parno);
goto insert_if_check_paren;
}
else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
capturing_parens:
parno = RExC_npar;
RExC_npar++;
+
ret = reganode(pRExC_state, OPEN, parno);
- if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
- DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
- "Setting paren #%"IVdf" to %d\n", parno, REG_NODE_NUM(ret)));
- RExC_parens[parno-1]= ret;
-
+ if (!SIZE_ONLY ){
+ if (!RExC_nestroot)
+ RExC_nestroot = parno;
+ if (RExC_seen & REG_SEEN_RECURSE) {
+ DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
+ "Setting open paren #%"IVdf" to %d\n",
+ (IV)parno, REG_NODE_NUM(ret)));
+ RExC_open_parens[parno-1]= ret;
+ }
}
Set_Node_Length(ret, 1); /* MJD */
Set_Node_Offset(ret, RExC_parse); /* MJD */
ender = reg_node(pRExC_state, TAIL);
break;
case 1:
+ RExC_cpar++;
ender = reganode(pRExC_state, CLOSE, parno);
+ if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
+ DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
+ "Setting close paren #%"IVdf" to %d\n",
+ (IV)parno, REG_NODE_NUM(ender)));
+ RExC_close_parens[parno-1]= ender;
+ if (RExC_nestroot == parno)
+ RExC_nestroot = 0;
+ }
Set_Node_Offset(ender,RExC_parse+1); /* MJD */
Set_Node_Length(ender,1); /* MJD */
break;
break;
case 0:
ender = reg_node(pRExC_state, END);
+ if (!SIZE_ONLY) {
+ assert(!RExC_opend); /* there can only be one! */
+ RExC_opend = ender;
+ }
break;
}
REGTAIL(pRExC_state, lastbr, ender);
*flagp = flags;
return(ret);
}
- /* else if (OP(ret)==RECURSE) {
- RExC_parse++;
- vFAIL("Illegal quantifier on recursion group");
- } */
#if 0 /* Now runtime fix should be reliable. */
origparse);
}
- if (*RExC_parse == '?') {
+ if (RExC_parse < RExC_end && *RExC_parse == '?') {
nextchar(pRExC_state);
reginsert(pRExC_state, MINMOD, ret, depth+1);
REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
}
- if (ISMULT2(RExC_parse)) {
+#ifndef REG_ALLOW_MINMOD_SUSPEND
+ else
+#endif
+ if (RExC_parse < RExC_end && *RExC_parse == '+') {
+ regnode *ender;
+ nextchar(pRExC_state);
+ ender = reg_node(pRExC_state, SUCCEED);
+ REGTAIL(pRExC_state, ret, ender);
+ reginsert(pRExC_state, SUSPEND, ret, depth+1);
+ ret->flags = 0;
+ ender = reg_node(pRExC_state, TAIL);
+ REGTAIL(pRExC_state, ret, ender);
+ /*ret= ender;*/
+ }
+
+ if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
RExC_parse++;
vFAIL("Nested quantifiers");
}
}
+/*
+ * reg_recode
+ *
+ * It returns the code point in utf8 for the value in *encp.
+ * value: a code value in the source encoding
+ * encp: a pointer to an Encode object
+ *
+ * If the result from Encode is not a single character,
+ * it returns U+FFFD (Replacement character) and sets *encp to NULL.
+ */
+STATIC UV
+S_reg_recode(pTHX_ const char value, SV **encp)
+{
+ STRLEN numlen = 1;
+ SV * const sv = sv_2mortal(newSVpvn(&value, numlen));
+ const char * const s = encp && *encp ? sv_recode_to_utf8(sv, *encp)
+ : SvPVX(sv);
+ const STRLEN newlen = SvCUR(sv);
+ UV uv = UNICODE_REPLACEMENT;
+
+ if (newlen)
+ uv = SvUTF8(sv)
+ ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
+ : *(U8*)s;
+
+ if (!newlen || numlen != newlen) {
+ uv = UNICODE_REPLACEMENT;
+ if (encp)
+ *encp = NULL;
+ }
+ return uv;
+}
+
/*
- regatom - the lowest level
case 'c':
case '0':
goto defchar;
+ case 'R':
case '1': case '2': case '3': case '4':
case '5': case '6': case '7': case '8': case '9':
{
- const I32 num = atoi(RExC_parse);
-
+ I32 num;
+ bool isrel=(*RExC_parse=='R');
+ if (isrel)
+ RExC_parse++;
+ num = atoi(RExC_parse);
+ if (isrel) {
+ num = RExC_cpar - num;
+ if (num < 1)
+ vFAIL("Reference to nonexistent or unclosed group");
+ }
if (num > 9 && num >= RExC_npar)
goto defchar;
else {
while (isDIGIT(*RExC_parse))
RExC_parse++;
- if (!SIZE_ONLY && num > (I32)RExC_rx->nparens)
- vFAIL("Reference to nonexistent group");
+ if (!SIZE_ONLY) {
+ if (num > (I32)RExC_rx->nparens)
+ vFAIL("Reference to nonexistent group");
+ /* People make this error all the time apparently.
+ So we cant fail on it, even though we should
+
+ else if (num >= RExC_cpar)
+ vFAIL("Reference to unclosed group will always match");
+ */
+ }
RExC_sawback = 1;
ret = reganode(pRExC_state,
(U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
case 'p':
case 'P':
case 'N':
+ case 'R':
--p;
goto loopdone;
case 'n':
ender = grok_hex(p, &numlen, &flags, NULL);
p += numlen;
}
+ if (PL_encoding && ender < 0x100)
+ goto recode_encoding;
break;
case 'c':
p++;
--p;
goto loopdone;
}
+ if (PL_encoding && ender < 0x100)
+ goto recode_encoding;
+ break;
+ recode_encoding:
+ {
+ SV* enc = PL_encoding;
+ ender = reg_recode((const char)(U8)ender, &enc);
+ if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
+ vWARN(p, "Invalid escape in the specified encoding");
+ RExC_utf8 = 1;
+ }
break;
case '\0':
if (p >= RExC_end)
break;
}
- /* If the encoding pragma is in effect recode the text of
- * any EXACT-kind nodes. */
- if (ret && PL_encoding && PL_regkind[OP(ret)] == EXACT) {
- const STRLEN oldlen = STR_LEN(ret);
- SV * const sv = sv_2mortal(newSVpvn(STRING(ret), oldlen));
-
- if (RExC_utf8)
- SvUTF8_on(sv);
- if (sv_utf8_downgrade(sv, TRUE)) {
- const char * const s = sv_recode_to_utf8(sv, PL_encoding);
- const STRLEN newlen = SvCUR(sv);
-
- if (SvUTF8(sv))
- RExC_utf8 = 1;
- if (!SIZE_ONLY) {
- GET_RE_DEBUG_FLAGS_DECL;
- DEBUG_COMPILE_r(PerlIO_printf(Perl_debug_log, "recode %*s to %*s\n",
- (int)oldlen, STRING(ret),
- (int)newlen, s));
- Copy(s, STRING(ret), newlen, char);
- STR_LEN(ret) += newlen - oldlen;
- RExC_emit += STR_SZ(newlen) - STR_SZ(oldlen);
- } else
- RExC_size += STR_SZ(newlen) - STR_SZ(oldlen);
- }
- }
-
return(ret);
}
value = grok_hex(RExC_parse, &numlen, &flags, NULL);
RExC_parse += numlen;
}
+ if (PL_encoding && value < 0x100)
+ goto recode_encoding;
break;
case 'c':
value = UCHARAT(RExC_parse++);
break;
case '0': case '1': case '2': case '3': case '4':
case '5': case '6': case '7': case '8': case '9':
- {
- I32 flags = 0;
- numlen = 3;
- value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
- RExC_parse += numlen;
- break;
- }
+ {
+ I32 flags = 0;
+ numlen = 3;
+ value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
+ RExC_parse += numlen;
+ if (PL_encoding && value < 0x100)
+ goto recode_encoding;
+ break;
+ }
+ recode_encoding:
+ {
+ SV* enc = PL_encoding;
+ value = reg_recode((const char)(U8)value, &enc);
+ if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
+ vWARN(RExC_parse,
+ "Invalid escape in the specified encoding");
+ break;
+ }
default:
if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
vWARN2(RExC_parse,
RExC_size += 1;
return(ret);
}
+#ifdef DEBUGGING
+ if (OP(RExC_emit) == 255)
+ Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %s: %d ",
+ reg_name[op], OP(RExC_emit));
+#endif
NODE_ALIGN_FILL(ret);
ptr = ret;
FILL_ADVANCE_NODE(ptr, op);
}
RExC_emit = ptr;
-
return(ret);
}
*/
return(ret);
}
-
+#ifdef DEBUGGING
+ if (OP(RExC_emit) == 255)
+ Perl_croak(aTHX_ "panic: reganode overwriting end of allocated program space");
+#endif
NODE_ALIGN_FILL(ret);
ptr = ret;
FILL_ADVANCE_NODE_ARG(ptr, op, arg);
}
RExC_emit = ptr;
-
return(ret);
}
src = RExC_emit;
RExC_emit += size;
dst = RExC_emit;
- if (RExC_parens) {
+ if (RExC_open_parens) {
int paren;
+ DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);
for ( paren=0 ; paren < RExC_npar ; paren++ ) {
- if ( RExC_parens[paren] >= src )
- RExC_parens[paren] += size;
- }
+ if ( RExC_open_parens[paren] >= opnd ) {
+ DEBUG_PARSE_FMT("open"," - %d",size);
+ RExC_open_parens[paren] += size;
+ } else {
+ DEBUG_PARSE_FMT("open"," - %s","ok");
+ }
+ if ( RExC_close_parens[paren] >= opnd ) {
+ DEBUG_PARSE_FMT("close"," - %d",size);
+ RExC_close_parens[paren] += size;
+ } else {
+ DEBUG_PARSE_FMT("close"," - %s","ok");
+ }
+ }
}
-
+
while (src > opnd) {
StructCopy(--src, --dst, regnode);
if (RExC_offsets) { /* MJD 20010112 */
GET_RE_DEBUG_FLAGS_DECL;
sv_setpvn(sv, "", 0);
+
if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
/* It would be nice to FAIL() here, but this may be called from
regexec.c, and it would be hard to supply pRExC_state. */
Perl_sv_catpvf(aTHX_ sv,
"<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
(UV)trie->startstate,
- (IV)trie->laststate-1,
+ (IV)trie->statecount-1, /* -1 because of the unused 0 element */
(UV)trie->wordcount,
(UV)trie->minlen,
(UV)trie->maxlen,
}
else if (k == WHILEM && o->flags) /* Ordinal/of */
Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
- else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP)
+ else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT)
Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
- else if (k == RECURSE)
+ else if (k == GOSUB)
Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
- else if (k == LOGICAL)
+ else if (k == VERB) {
+ if (!o->flags)
+ Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
+ (SV*)prog->data->data[ ARG( o ) ]);
+ } else if (k == LOGICAL)
Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
else if (k == ANYOF) {
int i, rangestart = -1;
DEBUG_COMPILE_r({
if (!PL_colorset)
reginitcolors();
- if (RX_DEBUG(r)){
- SV *dsv= sv_newmortal();
+ {
+ SV *dsv= sv_newmortal();
RE_PV_QUOTED_DECL(s, (r->reganch & ROPT_UTF8),
dsv, r->precomp, r->prelen, 60);
PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
if (trie->nextword)
Safefree(trie->nextword);
#ifdef DEBUGGING
- if (RX_DEBUG(r)) {
- if (trie->words)
- SvREFCNT_dec((SV*)trie->words);
- if (trie->revcharmap)
- SvREFCNT_dec((SV*)trie->revcharmap);
- }
+ if (trie->words)
+ SvREFCNT_dec((SV*)trie->words);
+ if (trie->revcharmap)
+ SvREFCNT_dec((SV*)trie->revcharmap);
#endif
Safefree(r->data->data[n]); /* do this last!!!! */
}
}
Safefree(r->startp);
Safefree(r->endp);
+ if (r->swap) {
+ Safefree(r->swap->startp);
+ Safefree(r->swap->endp);
+ Safefree(r->swap);
+ }
Safefree(r);
}
Copy(r->startp, ret->startp, npar, I32);
Newx(ret->endp, npar, I32);
Copy(r->startp, ret->startp, npar, I32);
+ if(r->swap) {
+ Newx(ret->swap, 1, regexp_paren_ofs);
+ /* no need to copy these */
+ Newx(ret->swap->startp, npar, I32);
+ Newx(ret->swap->endp, npar, I32);
+ } else {
+ ret->swap = NULL;
+ }
Newx(ret->substrs, 1, struct reg_substr_data);
for (s = ret->substrs->data, i = 0; i < 3; i++, s++) {
for (i = 0; i < count; i++) {
d->what[i] = r->data->what[i];
switch (d->what[i]) {
- /* legal options are one of: sfpont
+ /* legal options are one of: sSfpont
see also regcomp.h and pregfree() */
case 's':
case 'S':
ret->precomp = SAVEPVN(r->precomp, r->prelen);
ret->refcnt = r->refcnt;
ret->minlen = r->minlen;
+ ret->minlenret = r->minlenret;
ret->prelen = r->prelen;
ret->nparens = r->nparens;
ret->lastparen = r->lastparen;
}
#endif
+/*
+ reg_stringify()
+
+ converts a regexp embedded in a MAGIC struct to its stringified form,
+ caching the converted form in the struct and returns the cached
+ string.
+
+ If lp is nonnull then it is used to return the length of the
+ resulting string
+
+ If flags is nonnull and the returned string contains UTF8 then
+ (flags & 1) will be true.
+
+ If haseval is nonnull then it is used to return whether the pattern
+ contains evals.
+
+ Normally called via macro:
+
+ CALLREG_STRINGIFY(mg,0,0);
+
+ And internally with
+
+ CALLREG_AS_STR(mg,lp,flags,haseval)
+
+ See sv_2pv_flags() in sv.c for an example of internal usage.
+
+ */
+
+char *
+Perl_reg_stringify(pTHX_ MAGIC *mg, STRLEN *lp, U32 *flags, I32 *haseval ) {
+ dVAR;
+ const regexp * const re = (regexp *)mg->mg_obj;
+
+ if (!mg->mg_ptr) {
+ const char *fptr = "msix";
+ char reflags[6];
+ char ch;
+ int left = 0;
+ int right = 4;
+ bool need_newline = 0;
+ U16 reganch = (U16)((re->reganch & PMf_COMPILETIME) >> 12);
+
+ while((ch = *fptr++)) {
+ if(reganch & 1) {
+ reflags[left++] = ch;
+ }
+ else {
+ reflags[right--] = ch;
+ }
+ reganch >>= 1;
+ }
+ if(left != 4) {
+ reflags[left] = '-';
+ left = 5;
+ }
+
+ mg->mg_len = re->prelen + 4 + left;
+ /*
+ * If /x was used, we have to worry about a regex ending with a
+ * comment later being embedded within another regex. If so, we don't
+ * want this regex's "commentization" to leak out to the right part of
+ * the enclosing regex, we must cap it with a newline.
+ *
+ * So, if /x was used, we scan backwards from the end of the regex. If
+ * we find a '#' before we find a newline, we need to add a newline
+ * ourself. If we find a '\n' first (or if we don't find '#' or '\n'),
+ * we don't need to add anything. -jfriedl
+ */
+ if (PMf_EXTENDED & re->reganch) {
+ const char *endptr = re->precomp + re->prelen;
+ while (endptr >= re->precomp) {
+ const char c = *(endptr--);
+ if (c == '\n')
+ break; /* don't need another */
+ if (c == '#') {
+ /* we end while in a comment, so we need a newline */
+ mg->mg_len++; /* save space for it */
+ need_newline = 1; /* note to add it */
+ break;
+ }
+ }
+ }
+
+ Newx(mg->mg_ptr, mg->mg_len + 1 + left, char);
+ mg->mg_ptr[0] = '(';
+ mg->mg_ptr[1] = '?';
+ Copy(reflags, mg->mg_ptr+2, left, char);
+ *(mg->mg_ptr+left+2) = ':';
+ Copy(re->precomp, mg->mg_ptr+3+left, re->prelen, char);
+ if (need_newline)
+ mg->mg_ptr[mg->mg_len - 2] = '\n';
+ mg->mg_ptr[mg->mg_len - 1] = ')';
+ mg->mg_ptr[mg->mg_len] = 0;
+ }
+ if (haseval)
+ *haseval = re->program[0].next_off;
+ if (flags)
+ *flags = ((re->reganch & ROPT_UTF8) ? 1 : 0);
+
+ if (lp)
+ *lp = mg->mg_len;
+ return mg->mg_ptr;
+}
+
+
#ifndef PERL_IN_XSUB_RE
/*
- regnext - dig the "next" pointer out of a node
DUMPUNTIL(NEXTOPER(node), next);
}
else if ( PL_regkind[(U8)op] == TRIE ) {
+ const regnode *this_trie = node;
const char op = OP(node);
const I32 n = ARG(node);
const reg_ac_data * const ac = op>=AHOCORASICK ?
PL_colors[0], PL_colors[1],
(SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
PERL_PV_PRETTY_ELIPSES |
- PERL_PV_PRETTY_LTGT
+ PERL_PV_PRETTY_LTGT
)
: "???"
);
if (trie->jump) {
U16 dist= trie->jump[word_idx+1];
- PerlIO_printf(Perl_debug_log, "(%u)\n",(next - dist) - start);
+ PerlIO_printf(Perl_debug_log, "(%u)\n",
+ (dist ? this_trie + dist : next) - start);
if (dist) {
if (!nextbranch)
- nextbranch= next - trie->jump[0];
- DUMPUNTIL(next - dist, nextbranch);
- }
+ nextbranch= this_trie + trie->jump[0];
+ DUMPUNTIL(this_trie + dist, nextbranch);
+ }
if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
nextbranch= regnext((regnode *)nextbranch);
} else {
https://perl5.git.perl.org / perl5.git / blobdiff