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_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
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), "", numstates
+ );
for( q_read=1; q_read<numstates; q_read++ ) {
PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
}
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)
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;
-
-
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);
}
num++;
data_fake.flags = 0;
- if (data) {
+ if (data) {
data_fake.whilem_c = data->whilem_c;
data_fake.last_closep = data->last_closep;
}
/* we suppose the run is continuous, last=next...*/
minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
- next, &data_fake, f,depth+1);
+ next, &data_fake,
+ stopparen, recursed, NULL, f,depth+1);
if (min1 > minnext)
min1 = minnext;
if (max1 < minnext + deltanext)
if (flags & SCF_DO_STCLASS_OR) {
cl_or(pRExC_state, data->start_class, &accum);
if (min1) {
- cl_and(data->start_class, &and_with);
+ 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,
+ INIT_AND_WITHP;
+ StructCopy(data->start_class, and_withp,
struct regnode_charclass_class);
flags &= ~SCF_DO_STCLASS_AND;
StructCopy(&accum, data->start_class,
else
data->start_class->flags |= ANYOF_UNICODE_ALL;
data->start_class->flags &= ~ANYOF_EOS;
- cl_and(data->start_class, &and_with);
+ cl_and(data->start_class, and_withp);
}
flags &= ~SCF_DO_STCLASS;
}
ANYOF_BITMAP_SET(data->start_class, uc);
data->start_class->flags &= ~ANYOF_EOS;
}
- cl_and(data->start_class, &and_with);
+ cl_and(data->start_class, and_withp);
}
flags &= ~SCF_DO_STCLASS;
}
}
goto optimize_curly_tail;
case CURLY:
- mincount = ARG1(scan);
- maxcount = ARG2(scan);
+ 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);
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,
+ minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
+ last, data, stopparen, recursed, NULL,
(mincount == 0
? (f & ~SCF_DO_SUBSTR) : f),depth+1);
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,
+ INIT_AND_WITHP;
+ StructCopy(data->start_class, and_withp,
struct regnode_charclass_class);
flags &= ~SCF_DO_STCLASS_AND;
StructCopy(&this_class, data->start_class,
} 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);
+ cl_and(data->start_class, and_withp);
}
else if (flags & SCF_DO_STCLASS_AND)
cl_and(data->start_class, &this_class);
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. */
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. */
#endif
/* Optimize again: */
study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
- NULL, 0,depth+1);
+ NULL, stopparen, recursed, NULL, 0,depth+1);
}
else
oscan->flags = 0;
break;
}
if (flags & SCF_DO_STCLASS_OR)
- cl_and(data->start_class, &and_with);
+ cl_and(data->start_class, and_withp);
flags &= ~SCF_DO_STCLASS;
}
}
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);
+ 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");
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);
+
+ *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");
}
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++;
#endif
}
else if (OP(scan) == OPEN) {
- pars++;
+ if (stopparen != (I32)ARG(scan))
+ pars++;
}
else if (OP(scan) == CLOSE) {
+ if (stopparen == (I32)ARG(scan)) {
+ break;
+ }
if ((I32)ARG(scan) == is_par) {
next = regnext(scan);
if (data)
*(data->last_closep) = ARG(scan);
}
+ else if (OP(scan) == GOSUB || OP(scan) == GOSTART) {
+ /* set the pointer */
+ I32 paren;
+ regnode *start;
+ regnode *end;
+ 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;
+ }
+ assert(start);
+ assert(end);
+ if (!recursed) {
+ Newxz(recursed, (((RExC_npar)>>3) +1), U8);
+ SAVEFREEPV(recursed);
+ }
+ if (!PAREN_TEST(recursed,paren+1)) {
+ I32 deltanext = 0;
+ PAREN_SET(recursed,paren+1);
+
+ DEBUG_PEEP("goto",start,depth);
+ min += study_chunk(
+ pRExC_state,
+ &start,
+ minlenp,
+ &deltanext,
+ end+1,
+ data,
+ paren,
+ recursed,
+ and_withp,
+ flags,depth+1);
+ delta+=deltanext;
+ if (deltanext == I32_MAX) {
+ is_inf = is_inf_internal = 1;
+ delta=deltanext;
+ }
+ DEBUG_PEEP("rtrn",end,depth);
+ PAREN_UNSET(recursed,paren+1);
+ } 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 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 ( OP(scan)==OPFAIL ) {
+ if (flags & SCF_DO_SUBSTR) {
+ scan_commit(pRExC_state,data,minlenp);
+ flags &= ~SCF_DO_SUBSTR;
+ }
+ }
+ 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);
flags &= ~SCF_DO_STCLASS;
}
#ifdef TRIE_STUDY_OPT
-#ifdef FULL_TRIE_STUDY
+#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
+ 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;
if (trie->jump[word]) {
if (!nextbranch)
- nextbranch = tail - trie->jump[0];
- scan= tail - trie->jump[word];
+ 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, f,depth+1);
+ 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 (flags & SCF_DO_STCLASS_OR) {
cl_or(pRExC_state, data->start_class, &accum);
if (min1) {
- cl_and(data->start_class, &and_with);
+ 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,
+ INIT_AND_WITHP;
+ StructCopy(data->start_class, and_withp,
struct regnode_charclass_class);
flags &= ~SCF_DO_STCLASS_AND;
StructCopy(&accum, data->start_class,
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;
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 {
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",
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 */
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. */
/* 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 ) {
SvREFCNT_dec(data.longest_float);
SvREFCNT_dec(data.last_found);
}
+ StructCopy(&zero_scan_data, &data, scan_data_t);
} else {
+ 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. */
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;
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;
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);
RExC_parse++;
case '=': /* (?=...) */
case '!': /* (?!...) */
+ if (*RExC_parse == ')')
+ goto do_op_fail;
RExC_seen_zerolen++;
case ':': /* (?:...) */
case '>': /* (?>...) */
break;
+ case 'C': /* (?CUT) and (?COMMIT) */
+ if (RExC_parse[0] == 'O' &&
+ RExC_parse[1] == 'M' &&
+ RExC_parse[2] == 'M' &&
+ RExC_parse[3] == 'I' &&
+ RExC_parse[4] == 'T' &&
+ RExC_parse[5] == ')')
+ {
+ RExC_parse+=5;
+ ret = reg_node(pRExC_state, COMMIT);
+ } else if (
+ RExC_parse[0] == 'U' &&
+ RExC_parse[1] == 'T' &&
+ RExC_parse[2] == ')')
+ {
+ RExC_parse+=2;
+ ret = reg_node(pRExC_state, CUT);
+ } else {
+ vFAIL("Sequence (?C... not terminated");
+ }
+ nextchar(pRExC_state);
+ return ret;
+ break;
+ case 'E': /* (?ERROR) */
+ if (RExC_parse[0] == 'R' &&
+ RExC_parse[1] == 'R' &&
+ RExC_parse[2] == 'O' &&
+ RExC_parse[3] == 'R' &&
+ RExC_parse[4] == ')')
+ {
+ RExC_parse+=4;
+ ret = reg_node(pRExC_state, OPERROR);
+ } else {
+ vFAIL("Sequence (?E... not terminated");
+ }
+ nextchar(pRExC_state);
+ return ret;
+ break;
+ case 'F':
+ if (RExC_parse[0] == 'A' &&
+ RExC_parse[1] == 'I' &&
+ RExC_parse[2] == 'L')
+ RExC_parse+=3;
+ if (*RExC_parse != ')')
+ vFAIL("Sequence (?FAIL) or (?F) not terminated");
+ do_op_fail:
+ ret = reg_node(pRExC_state, OPFAIL);
+ nextchar(pRExC_state);
+ return ret;
+ break;
case '$': /* (?$...) */
case '@': /* (?@...) */
vFAIL2("Sequence (?%c...) not implemented", (int)paren);
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;
vFAIL("Expecting close bracket");
gen_recurse_regop:
- ret = reganode(pRExC_state, RECURSE, num);
+ 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", (UV)ARG(ret), (IV)ARG2L(ret)));
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' ) {
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", (IV)parno, REG_NODE_NUM(ret)));
- RExC_parens[parno-1]= ret;
-
+ "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 */
break;
case 1:
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;
+ }
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);
src = RExC_emit;
RExC_emit += size;
dst = RExC_emit;
- if (RExC_parens) {
+ if (RExC_open_parens) {
int paren;
+ DEBUG_PARSE_FMT("inst"," - %d",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 */
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,
Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP)
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)
Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
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 {
projects / p5sagit/p5-mst-13.2.git / blobdiff