} \
} STMT_END
+
static void restore_pos(pTHX_ void *arg);
STATIC CHECKPOINT
if (paren_elems_to_push < 0)
Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
-#define REGCP_OTHER_ELEMS 6
+#define REGCP_OTHER_ELEMS 8
SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS);
+
for (p = PL_regsize; p > parenfloor; p--) {
/* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
SSPUSHINT(PL_regendp[p]);
));
}
/* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
+ SSPUSHPTR(PL_regstartp);
+ SSPUSHPTR(PL_regendp);
SSPUSHINT(PL_regsize);
SSPUSHINT(*PL_reglastparen);
SSPUSHINT(*PL_reglastcloseparen);
*PL_reglastcloseparen = SSPOPINT;
*PL_reglastparen = SSPOPINT;
PL_regsize = SSPOPINT;
+ PL_regendp=(I32 *) SSPOPPTR;
+ PL_regstartp=(I32 *) SSPOPPTR;
+
/* Now restore the parentheses context. */
for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
i > 0; i -= REGCP_PAREN_ELEMS) {
srch_end_shift -= ((strbeg - s) - srch_start_shift);
srch_start_shift = strbeg - s;
}
- DEBUG_OPTIMISE_r({
+ DEBUG_OPTIMISE_MORE_r({
PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
(IV)prog->check_offset_min,
(IV)srch_start_shift,
start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
end_point= HOP3(strend, -srch_end_shift, strbeg);
}
- DEBUG_OPTIMISE_r({
+ DEBUG_OPTIMISE_MORE_r({
PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
(int)(end_point - start_point),
(int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
- DEBUG_OPTIMISE_r(
+ DEBUG_OPTIMISE_MORE_r(
PerlIO_printf(Perl_debug_log,
"Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
(IV)prog->check_offset_min,
}
}
if (last == NULL) {
- DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
- "%sCan't trim the tail, match fails (should not happen)%s\n",
- PL_colors[4], PL_colors[5]));
+ DEBUG_EXECUTE_r(
+ PerlIO_printf(Perl_debug_log,
+ "%sCan't trim the tail, match fails (should not happen)%s\n",
+ PL_colors[4], PL_colors[5]));
goto phooey; /* Should not happen! */
}
dontbother = strend - last + prog->float_min_offset;
the same. */
restore_pos(aTHX_ prog);
}
+ if (prog->paren_names)
+ (void)hv_iterinit(prog->paren_names);
/* make sure $`, $&, $', and $digit will work later */
if ( !(flags & REXEC_NOT_FIRST) ) {
return 0;
}
+
/*
- regtry - try match at specific point
*/
PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
(IV)(PL_stack_sp - PL_stack_base));
));
- SAVEI32(cxstack[cxstack_ix].blk_oldsp);
+ SAVESTACK_CXPOS();
cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
/* Otherwise OP_NEXTSTATE will free whatever on stack now. */
SAVETMPS;
prog->subbeg = PL_bostr;
prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
}
+ DEBUG_EXECUTE_r(PL_reg_starttry = startpos);
prog->startp[0] = startpos - PL_bostr;
PL_reginput = startpos;
- PL_regstartp = prog->startp;
- PL_regendp = prog->endp;
PL_reglastparen = &prog->lastparen;
PL_reglastcloseparen = &prog->lastcloseparen;
prog->lastparen = 0;
prog->lastcloseparen = 0;
PL_regsize = 0;
- DEBUG_EXECUTE_r(PL_reg_starttry = startpos);
+ PL_regstartp = prog->startp;
+ PL_regendp = prog->endp;
if (PL_reg_start_tmpl <= prog->nparens) {
PL_reg_start_tmpl = prog->nparens*3/2 + 3;
if(PL_reg_start_tmp)
#define sayYES goto yes
#define sayNO goto no
#define sayNO_SILENT goto no_silent
-#define saySAME(x) if (x) goto yes; else goto no
/* we dont use STMT_START/END here because it leads to
"unreachable code" warnings, which are bogus, but distracting. */
#define CACHEsayNO \
- if (st->u.whilem.cache_offset | st->u.whilem.cache_bit) \
- PL_reg_poscache[st->u.whilem.cache_offset] |= \
- (1<<st->u.whilem.cache_bit); \
+ if (ST.cache_mask) \
+ PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
sayNO
/* this is used to determine how far from the left messages like
return SLAB_FIRST(s);
}
-/* simulate a recursive call to regmatch */
-
-#define REGMATCH(ns, where) \
- st->scan = scan; \
- scan = (ns); \
- st->resume_state = resume_##where; \
- goto start_recurse; \
- resume_point_##where:
/* push a new state then goto it */
/*
- - regmatch - main matching routine
- *
- * Conceptually the strategy is simple: check to see whether the current
- * node matches, call self recursively to see whether the rest matches,
- * and then act accordingly. In practice we make some effort to avoid
- * recursion, in particular by going through "ordinary" nodes (that don't
- * need to know whether the rest of the match failed) by a loop instead of
- * by recursion.
- */
-/* [lwall] I've hoisted the register declarations to the outer block in order to
- * maybe save a little bit of pushing and popping on the stack. It also takes
- * advantage of machines that use a register save mask on subroutine entry.
- *
- * This function used to be heavily recursive, but since this had the
- * effect of blowing the CPU stack on complex regexes, it has been
- * restructured to be iterative, and to save state onto the heap rather
- * than the stack. Essentially whereever regmatch() used to be called, it
- * pushes the current state, notes where to return, then jumps back into
- * the main loop.
- *
- * Originally the structure of this function used to look something like
- S_regmatch() {
- int a = 1, b = 2;
+regmatch() - main matching routine
+
+This is basically one big switch statement in a loop. We execute an op,
+set 'next' to point the next op, and continue. If we come to a point which
+we may need to backtrack to on failure such as (A|B|C), we push a
+backtrack state onto the backtrack stack. On failure, we pop the top
+state, and re-enter the loop at the state indicated. If there are no more
+states to pop, we return failure.
+
+Sometimes we also need to backtrack on success; for example /A+/, where
+after successfully matching one A, we need to go back and try to
+match another one; similarly for lookahead assertions: if the assertion
+completes successfully, we backtrack to the state just before the assertion
+and then carry on. In these cases, the pushed state is marked as
+'backtrack on success too'. This marking is in fact done by a chain of
+pointers, each pointing to the previous 'yes' state. On success, we pop to
+the nearest yes state, discarding any intermediate failure-only states.
+Sometimes a yes state is pushed just to force some cleanup code to be
+called at the end of a successful match or submatch; e.g. (??{$re}) uses
+it to free the inner regex.
+
+Note that failure backtracking rewinds the cursor position, while
+success backtracking leaves it alone.
+
+A pattern is complete when the END op is executed, while a subpattern
+such as (?=foo) is complete when the SUCCESS op is executed. Both of these
+ops trigger the "pop to last yes state if any, otherwise return true"
+behaviour.
+
+A common convention in this function is to use A and B to refer to the two
+subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
+the subpattern to be matched possibly multiple times, while B is the entire
+rest of the pattern. Variable and state names reflect this convention.
+
+The states in the main switch are the union of ops and failure/success of
+substates associated with with that op. For example, IFMATCH is the op
+that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
+'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
+successfully matched A and IFMATCH_A_fail is a state saying that we have
+just failed to match A. Resume states always come in pairs. The backtrack
+state we push is marked as 'IFMATCH_A', but when that is popped, we resume
+at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
+on success or failure.
+
+The struct that holds a backtracking state is actually a big union, with
+one variant for each major type of op. The variable st points to the
+top-most backtrack struct. To make the code clearer, within each
+block of code we #define ST to alias the relevant union.
+
+Here's a concrete example of a (vastly oversimplified) IFMATCH
+implementation:
+
+ switch (state) {
+ ....
+
+#define ST st->u.ifmatch
+
+ case IFMATCH: // we are executing the IFMATCH op, (?=A)B
+ ST.foo = ...; // some state we wish to save
...
- while (scan != NULL) {
- a++; // do stuff with a and b
- ...
- switch (OP(scan)) {
- case FOO: {
- int local = 3;
- ...
- if (regmatch(...)) // recurse
- goto yes;
- }
- ...
- }
- }
- yes:
- return 1;
- }
+ // push a yes backtrack state with a resume value of
+ // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
+ // first node of A:
+ PUSH_YES_STATE_GOTO(IFMATCH_A, A);
+ // NOTREACHED
+
+ case IFMATCH_A: // we have successfully executed A; now continue with B
+ next = B;
+ bar = ST.foo; // do something with the preserved value
+ break;
+
+ case IFMATCH_A_fail: // A failed, so the assertion failed
+ ...; // do some housekeeping, then ...
+ sayNO; // propagate the failure
+
+#undef ST
- * Now it looks something like this:
+ ...
+ }
- typedef struct {
- int a, b, local;
- int resume_state;
- } regmatch_state;
+For any old-timers reading this who are familiar with the old recursive
+approach, the code above is equivalent to:
- S_regmatch() {
- regmatch_state *st = new();
- int depth=0;
- st->a++; // do stuff with a and b
+ case IFMATCH: // we are executing the IFMATCH op, (?=A)B
+ {
+ int foo = ...
...
- while (scan != NULL) {
- ...
- switch (OP(scan)) {
- case FOO: {
- st->local = 3;
- ...
- st->scan = scan;
- scan = ...;
- st->resume_state = resume_FOO;
- goto start_recurse; // recurse
-
- resume_point_FOO:
- if (result)
- goto yes;
- }
- ...
- }
- start_recurse:
- st = new(); push a new state
- st->a = 1; st->b = 2;
- depth++;
- }
- yes:
- result = 1;
- if (depth--) {
- st = pop();
- switch (resume_state) {
- case resume_FOO:
- goto resume_point_FOO;
- ...
- }
+ if (regmatch(A)) {
+ next = B;
+ bar = foo;
+ break;
}
- return result
+ ...; // do some housekeeping, then ...
+ sayNO; // propagate the failure
}
-
- * WARNING: this means that any line in this function that contains a
- * REGMATCH() or TRYPAREN() is actually simulating a recursive call to
- * regmatch() using gotos instead. Thus the values of any local variables
- * not saved in the regmatch_state structure will have been lost when
- * execution resumes on the next line .
- *
- * States (ie the st pointer) are allocated in slabs of about 4K in size.
- * PL_regmatch_state always points to the currently active state, and
- * PL_regmatch_slab points to the slab currently containing PL_regmatch_state.
- * The first time regmatch is called, the first slab is allocated, and is
- * never freed until interpreter desctruction. When the slab is full,
- * a new one is allocated chained to the end. At exit from regmatch, slabs
- * allocated since entry are freed.
- */
+
+The topmost backtrack state, pointed to by st, is usually free. If you
+want to claim it, populate any ST.foo fields in it with values you wish to
+save, then do one of
+
+ PUSH_STATE_GOTO(resume_state, node);
+ PUSH_YES_STATE_GOTO(resume_state, node);
+
+which sets that backtrack state's resume value to 'resume_state', pushes a
+new free entry to the top of the backtrack stack, then goes to 'node'.
+On backtracking, the free slot is popped, and the saved state becomes the
+new free state. An ST.foo field in this new top state can be temporarily
+accessed to retrieve values, but once the main loop is re-entered, it
+becomes available for reuse.
+
+Note that the depth of the backtrack stack constantly increases during the
+left-to-right execution of the pattern, rather than going up and down with
+the pattern nesting. For example the stack is at its maximum at Z at the
+end of the pattern, rather than at X in the following:
+
+ /(((X)+)+)+....(Y)+....Z/
+
+The only exceptions to this are lookahead/behind assertions and the cut,
+(?>A), which pop all the backtrack states associated with A before
+continuing.
+
+Bascktrack state structs are allocated in slabs of about 4K in size.
+PL_regmatch_state and st always point to the currently active state,
+and PL_regmatch_slab points to the slab currently containing
+PL_regmatch_state. The first time regmatch() is called, the first slab is
+allocated, and is never freed until interpreter destruction. When the slab
+is full, a new one is allocated and chained to the end. At exit from
+regmatch(), slabs allocated since entry are freed.
+
+*/
#define DEBUG_STATE_pp(pp) \
#endif
+/* reg_check_named_buff_matched()
+ * Checks to see if a named buffer has matched. The data array of
+ * buffer numbers corresponding to the buffer is expected to reside
+ * in the regexp->data->data array in the slot stored in the ARG() of
+ * node involved. Note that this routine doesn't actually care about the
+ * name, that information is not preserved from compilation to execution.
+ * Returns the index of the leftmost defined buffer with the given name
+ * or 0 if non of the buffers matched.
+ */
+STATIC I32
+S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan) {
+ I32 n;
+ SV *sv_dat=(SV*)rex->data->data[ ARG( scan ) ];
+ I32 *nums=(I32*)SvPVX(sv_dat);
+ for ( n=0; n<SvIVX(sv_dat); n++ ) {
+ if ((I32)*PL_reglastparen >= nums[n] &&
+ PL_regendp[nums[n]] != -1)
+ {
+ return nums[n];
+ }
+ }
+ return 0;
+}
+
STATIC I32 /* 0 failure, 1 success */
S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog)
{
/* cache heavy used fields of st in registers */
register regnode *scan;
register regnode *next;
- register I32 n = 0; /* initialize to shut up compiler warning */
+ register I32 n = 0; /* general value; init to avoid compiler warning */
+ register I32 ln = 0; /* len or last; init to avoid compiler warning */
register char *locinput = PL_reginput;
-
- /* these variables are NOT saved during a recusive RFEGMATCH: */
register I32 nextchr; /* is always set to UCHARAT(locinput) */
+
bool result = 0; /* return value of S_regmatch */
- int depth = 0; /* depth of recursion */
+ int depth = 0; /* depth of backtrack stack */
+ int nochange_depth = 0; /* depth of RECURSE recursion with nochange*/
regmatch_state *yes_state = NULL; /* state to pop to on success of
subpattern */
regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
U32 state_num;
-
- I32 parenfloor = 0;
+
+ /* these three flags are set by various ops to signal information to
+ * the very next op. They have a useful lifetime of exactly one loop
+ * iteration, and are not preserved or restored by state pushes/pops
+ */
+ bool sw = 0; /* the condition value in (?(cond)a|b) */
+ bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
+ int logical = 0; /* the following EVAL is:
+ 0: (?{...})
+ 1: (?(?{...})X|Y)
+ 2: (??{...})
+ or the following IFMATCH/UNLESSM is:
+ false: plain (?=foo)
+ true: used as a condition: (?(?=foo))
+ */
#ifdef DEBUGGING
GET_RE_DEBUG_FLAGS_DECL;
if (st > SLAB_LAST(PL_regmatch_slab))
st = PL_regmatch_state = S_push_slab(aTHX);
- st->minmod = 0;
- st->sw = 0;
- st->logical = 0;
- cur_curlyx = NULL;
-
/* Note that nextchr is a byte even in UTF */
nextchr = UCHARAT(locinput);
scan = prog;
ST.accepted = 0; /* how many accepting states we have seen */
ST.B = next;
ST.jump = trie->jump;
-
-#ifdef DEBUGGING
ST.me = scan;
-#endif
-
-
/*
traverse the TRIE keeping track of all accepting states
we transition through until we get to a failing node.
locinput = PL_reginput;
nextchr = UCHARAT(locinput);
- if ( !ST.jump )
+ if ( !ST.jump || !ST.jump[ST.accept_buff[0].wordnum])
scan = ST.B;
else
- scan = ST.B - ST.jump[ST.accept_buff[0].wordnum];
+ scan = ST.me + ST.jump[ST.accept_buff[0].wordnum];
continue; /* execute rest of RE */
}
SV ** const tmp = RX_DEBUG(reginfo->prog)
? av_fetch( trie->words, ST.accept_buff[ best ].wordnum - 1, 0 )
: NULL;
- regnode *nextop=!ST.jump ?
+ regnode *nextop=(!ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) ?
ST.B :
- ST.B - ST.jump[ST.accept_buff[best].wordnum];
+ ST.me + ST.jump[ST.accept_buff[best].wordnum];
PerlIO_printf( Perl_debug_log,
"%*s %strying alternation #%d <%s> at node #%d %s\n",
REPORT_CODE_OFF+depth*2, "", PL_colors[4],
best = ST.accepted;
}
PL_reginput = (char *)ST.accept_buff[ best ].endpos;
- if ( !ST.jump ) {
+ if ( !ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) {
PUSH_STATE_GOTO(TRIE_next, ST.B);
/* NOTREACHED */
} else {
- PUSH_STATE_GOTO(TRIE_next, ST.B - ST.jump[ST.accept_buff[best].wordnum]);
+ PUSH_STATE_GOTO(TRIE_next, ST.me + ST.jump[ST.accept_buff[best].wordnum]);
/* NOTREACHED */
}
/* NOTREACHED */
case EXACT: {
char *s = STRING(scan);
- st->ln = STR_LEN(scan);
+ ln = STR_LEN(scan);
if (do_utf8 != UTF) {
/* The target and the pattern have differing utf8ness. */
char *l = locinput;
- const char * const e = s + st->ln;
+ const char * const e = s + ln;
if (do_utf8) {
/* The target is utf8, the pattern is not utf8. */
/* Inline the first character, for speed. */
if (UCHARAT(s) != nextchr)
sayNO;
- if (PL_regeol - locinput < st->ln)
+ if (PL_regeol - locinput < ln)
sayNO;
- if (st->ln > 1 && memNE(s, locinput, st->ln))
+ if (ln > 1 && memNE(s, locinput, ln))
sayNO;
- locinput += st->ln;
+ locinput += ln;
nextchr = UCHARAT(locinput);
break;
}
/* FALL THROUGH */
case EXACTF: {
char * const s = STRING(scan);
- st->ln = STR_LEN(scan);
+ ln = STR_LEN(scan);
if (do_utf8 || UTF) {
/* Either target or the pattern are utf8. */
const char * const l = locinput;
char *e = PL_regeol;
- if (ibcmp_utf8(s, 0, st->ln, (bool)UTF,
+ if (ibcmp_utf8(s, 0, ln, (bool)UTF,
l, &e, 0, do_utf8)) {
/* One more case for the sharp s:
* pack("U0U*", 0xDF) =~ /ss/i,
* byte sequence for the U+00DF. */
if (!(do_utf8 &&
toLOWER(s[0]) == 's' &&
- st->ln >= 2 &&
+ ln >= 2 &&
toLOWER(s[1]) == 's' &&
(U8)l[0] == 0xC3 &&
e - l >= 2 &&
UCHARAT(s) != ((OP(scan) == EXACTF)
? PL_fold : PL_fold_locale)[nextchr])
sayNO;
- if (PL_regeol - locinput < st->ln)
+ if (PL_regeol - locinput < ln)
sayNO;
- if (st->ln > 1 && (OP(scan) == EXACTF
- ? ibcmp(s, locinput, st->ln)
- : ibcmp_locale(s, locinput, st->ln)))
+ if (ln > 1 && (OP(scan) == EXACTF
+ ? ibcmp(s, locinput, ln)
+ : ibcmp_locale(s, locinput, ln)))
sayNO;
- locinput += st->ln;
+ locinput += ln;
nextchr = UCHARAT(locinput);
break;
}
/* was last char in word? */
if (do_utf8) {
if (locinput == PL_bostr)
- st->ln = '\n';
+ ln = '\n';
else {
const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
- st->ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
+ ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
}
if (OP(scan) == BOUND || OP(scan) == NBOUND) {
- st->ln = isALNUM_uni(st->ln);
+ ln = isALNUM_uni(ln);
LOAD_UTF8_CHARCLASS_ALNUM();
n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
}
else {
- st->ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(st->ln));
+ ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
n = isALNUM_LC_utf8((U8*)locinput);
}
}
else {
- st->ln = (locinput != PL_bostr) ?
+ ln = (locinput != PL_bostr) ?
UCHARAT(locinput - 1) : '\n';
if (OP(scan) == BOUND || OP(scan) == NBOUND) {
- st->ln = isALNUM(st->ln);
+ ln = isALNUM(ln);
n = isALNUM(nextchr);
}
else {
- st->ln = isALNUM_LC(st->ln);
+ ln = isALNUM_LC(ln);
n = isALNUM_LC(nextchr);
}
}
- if (((!st->ln) == (!n)) == (OP(scan) == BOUND ||
+ if (((!ln) == (!n)) == (OP(scan) == BOUND ||
OP(scan) == BOUNDL))
sayNO;
break;
locinput++;
nextchr = UCHARAT(locinput);
break;
+
+ case NREFFL:
+ {
+ char *s;
+ char type;
+ PL_reg_flags |= RF_tainted;
+ /* FALL THROUGH */
+ case NREF:
+ case NREFF:
+ type = OP(scan);
+ n = reg_check_named_buff_matched(rex,scan);
+
+ if ( n ) {
+ type = REF + ( type - NREF );
+ goto do_ref;
+ } else {
+ sayNO;
+ }
+ /* unreached */
case REFFL:
PL_reg_flags |= RF_tainted;
/* FALL THROUGH */
case REF:
- case REFF: {
- char *s;
+ case REFF:
n = ARG(scan); /* which paren pair */
- st->ln = PL_regstartp[n];
+ type = OP(scan);
+ do_ref:
+ ln = PL_regstartp[n];
PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
- if ((I32)*PL_reglastparen < n || st->ln == -1)
+ if ((I32)*PL_reglastparen < n || ln == -1)
sayNO; /* Do not match unless seen CLOSEn. */
- if (st->ln == PL_regendp[n])
+ if (ln == PL_regendp[n])
break;
- s = PL_bostr + st->ln;
- if (do_utf8 && OP(scan) != REF) { /* REF can do byte comparison */
+ s = PL_bostr + ln;
+ if (do_utf8 && type != REF) { /* REF can do byte comparison */
char *l = locinput;
const char *e = PL_bostr + PL_regendp[n];
/*
* in the 8-bit case (no pun intended) because in Unicode we
* have to map both upper and title case to lower case.
*/
- if (OP(scan) == REFF) {
+ if (type == REFF) {
while (s < e) {
STRLEN ulen1, ulen2;
U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
/* Inline the first character, for speed. */
if (UCHARAT(s) != nextchr &&
- (OP(scan) == REF ||
- (UCHARAT(s) != ((OP(scan) == REFF
- ? PL_fold : PL_fold_locale)[nextchr]))))
+ (type == REF ||
+ (UCHARAT(s) != (type == REFF
+ ? PL_fold : PL_fold_locale)[nextchr])))
sayNO;
- st->ln = PL_regendp[n] - st->ln;
- if (locinput + st->ln > PL_regeol)
+ ln = PL_regendp[n] - ln;
+ if (locinput + ln > PL_regeol)
sayNO;
- if (st->ln > 1 && (OP(scan) == REF
- ? memNE(s, locinput, st->ln)
- : (OP(scan) == REFF
- ? ibcmp(s, locinput, st->ln)
- : ibcmp_locale(s, locinput, st->ln))))
+ if (ln > 1 && (type == REF
+ ? memNE(s, locinput, ln)
+ : (type == REFF
+ ? ibcmp(s, locinput, ln)
+ : ibcmp_locale(s, locinput, ln))))
sayNO;
- locinput += st->ln;
+ locinput += ln;
nextchr = UCHARAT(locinput);
break;
- }
-
+ }
case NOTHING:
case TAIL:
break;
#undef ST
#define ST st->u.eval
-
- case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
{
SV *ret;
+ regexp *re;
+ regnode *startpoint;
+
+ case SRECURSE:
+ case RECURSE: /* /(...(?1))/ */
+ if (cur_eval && cur_eval->locinput==locinput) {
+ if (cur_eval->u.eval.close_paren == ARG(scan))
+ Perl_croak(aTHX_ "Infinite recursion in RECURSE in regexp");
+ if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH )
+ Perl_croak(aTHX_ "RECURSE without pos change exceeded limit in regexp");
+ } else {
+ nochange_depth = 0;
+ }
+ re = rex;
+ (void)ReREFCNT_inc(rex);
+ if (OP(scan)==RECURSE) {
+ startpoint = scan + ARG2L(scan);
+ ST.close_paren = ARG(scan);
+ } else {
+ startpoint = re->program+1;
+ ST.close_paren = 0;
+ }
+ goto eval_recurse_doit;
+ /* NOTREACHED */
+ case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
+ if (cur_eval && cur_eval->locinput==locinput) {
+ if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH )
+ Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regexp");
+ } else {
+ nochange_depth = 0;
+ }
{
/* execute the code in the {...} */
dSP;
PL_op = oop;
PAD_RESTORE_LOCAL(old_comppad);
PL_curcop = ocurcop;
- if (!st->logical) {
+ if (!logical) {
/* /(?{...})/ */
sv_setsv(save_scalar(PL_replgv), ret);
break;
}
}
- if (st->logical == 2) { /* Postponed subexpression: /(??{...})/ */
- regexp *re;
+ if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
+ logical = 0;
{
/* extract RE object from returned value; compiling if
* necessary */
PL_regsize = osize;
}
}
+ DEBUG_EXECUTE_r(
+ debug_start_match(re, do_utf8, locinput, PL_regeol,
+ "Matching embedded");
+ );
+ startpoint = re->program + 1;
+ ST.close_paren = 0; /* only used for RECURSE */
+ /* borrowed from regtry */
+ if (PL_reg_start_tmpl <= re->nparens) {
+ PL_reg_start_tmpl = re->nparens*3/2 + 3;
+ if(PL_reg_start_tmp)
+ Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
+ else
+ Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
+ }
+ eval_recurse_doit: /* Share code with RECURSE below this line */
/* run the pattern returned from (??{...}) */
ST.cp = regcppush(0); /* Save *all* the positions. */
REGCP_SET(ST.lastcp);
+
+ PL_regstartp = re->startp; /* essentially NOOP on RECURSE */
+ PL_regendp = re->endp; /* essentially NOOP on RECURSE */
+
*PL_reglastparen = 0;
*PL_reglastcloseparen = 0;
PL_reginput = locinput;
+ PL_regsize = 0;
/* XXXX This is too dramatic a measure... */
PL_reg_maxiter = 0;
- st->logical = 0;
ST.toggle_reg_flags = PL_reg_flags;
if (re->reganch & ROPT_UTF8)
PL_reg_flags |= RF_utf8;
ST.B = next;
ST.prev_eval = cur_eval;
cur_eval = st;
-
- DEBUG_EXECUTE_r(
- debug_start_match(re, do_utf8, locinput, PL_regeol,
- "Matching embedded");
- );
/* now continue from first node in postoned RE */
- PUSH_YES_STATE_GOTO(EVAL_AB, re->program + 1);
+ PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
/* NOTREACHED */
}
- /* /(?(?{...})X|Y)/ */
- st->sw = (bool)SvTRUE(ret);
- st->logical = 0;
+ /* logical is 1, /(?(?{...})X|Y)/ */
+ sw = (bool)SvTRUE(ret);
+ logical = 0;
break;
}
/* XXXX This is too dramatic a measure... */
PL_reg_maxiter = 0;
sayNO_SILENT;
-
#undef ST
case OPEN:
n = ARG(scan); /* which paren pair */
PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
PL_regendp[n] = locinput - PL_bostr;
+ /*if (n > PL_regsize)
+ PL_regsize = n;*/
if (n > (I32)*PL_reglastparen)
*PL_reglastparen = n;
*PL_reglastcloseparen = n;
+ if (cur_eval && cur_eval->u.eval.close_paren == (U32)n) {
+ goto fake_end;
+ }
break;
case GROUPP:
n = ARG(scan); /* which paren pair */
- st->sw = (bool)((I32)*PL_reglastparen >= n && PL_regendp[n] != -1);
+ sw = (bool)((I32)*PL_reglastparen >= n && PL_regendp[n] != -1);
break;
+ case NGROUPP:
+ /* reg_check_named_buff_matched returns 0 for no match */
+ sw = (bool)(0 < reg_check_named_buff_matched(rex,scan));
+ break;
+ case RECURSEP:
+ n = ARG(scan);
+ sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
+ break;
+ case DEFINEP:
+ sw = 0;
+ break;
case IFTHEN:
PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
- if (st->sw)
+ if (sw)
next = NEXTOPER(NEXTOPER(scan));
else {
next = scan + ARG(scan);
}
break;
case LOGICAL:
- st->logical = scan->flags;
+ logical = scan->flags;
break;
+
/*******************************************************************
- cur_curlyx points to the regmatch_state associated with the most recent CURLYX.
- This struct contains info about the innermost (...)* loop (an
- "infoblock"), and a pointer to the next outer cur_curlyx.
- Here is how Y(A)*Z is processed (if it is compiled into CURLYX/WHILEM):
+The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
+pattern, where A and B are subpatterns. (For simple A, CURLYM or
+STAR/PLUS/CURLY/CURLYN are used instead.)
+
+A*B is compiled as <CURLYX><A><WHILEM><B>
- 1) After matching Y, regnode for CURLYX is processed;
+On entry to the subpattern, CURLYX is called. This pushes a CURLYX
+state, which contains the current count, initialised to -1. It also sets
+cur_curlyx to point to this state, with any previous value saved in the
+state block.
- 2) This regnode populates cur_curlyx, and calls regmatch() recursively
- with the starting point at WHILEM node;
+CURLYX then jumps straight to the WHILEM op, rather than executing A,
+since the pattern may possibly match zero times (i.e. it's a while {} loop
+rather than a do {} while loop).
- 3) Each hit of WHILEM node tries to match A and Z (in the order
- depending on the current iteration, min/max of {min,max} and
- greediness). The information about where are nodes for "A"
- and "Z" is read from cur_curlyx, as is info on how many times "A"
- was already matched, and greediness.
+Each entry to WHILEM represents a successful match of A. The count in the
+CURLYX block is incremented, another WHILEM state is pushed, and execution
+passes to A or B depending on greediness and the current count.
- 4) After A matches, the same WHILEM node is hit again.
+For example, if matching against the string a1a2a3b (where the aN are
+substrings that match /A/), then the match progresses as follows: (the
+pushed states are interspersed with the bits of strings matched so far):
- 5) Each time WHILEM is hit, cur_curlyx is the infoblock created by CURLYX
- of the same pair. Thus when WHILEM tries to match Z, it temporarily
- resets cur_curlyx, since this Y(A)*Z can be a part of some other loop:
- as in (Y(A)*Z)*. If Z matches, the automaton will hit the WHILEM node
- of the external loop.
+ <CURLYX cnt=-1>
+ <CURLYX cnt=0><WHILEM>
+ <CURLYX cnt=1><WHILEM> a1 <WHILEM>
+ <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
+ <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
+ <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
- Currently present infoblocks form a tree with a stem formed by cur_curlyx
- and whatever it mentions via ->next, and additional attached trees
- corresponding to temporarily unset infoblocks as in "5" above.
+(Contrast this with something like CURLYM, which maintains only a single
+backtrack state:
- In the following picture, infoblocks for outer loop of
- (Y(A)*?Z)*?T are denoted O, for inner I. NULL starting block
- is denoted by x. The matched string is YAAZYAZT. Temporarily postponed
- infoblocks are drawn below the "reset" infoblock.
+ <CURLYM cnt=0> a1
+ a1 <CURLYM cnt=1> a2
+ a1 a2 <CURLYM cnt=2> a3
+ a1 a2 a3 <CURLYM cnt=3> b
+)
- In fact in the picture below we do not show failed matches for Z and T
- by WHILEM blocks. [We illustrate minimal matches, since for them it is
- more obvious *why* one needs to *temporary* unset infoblocks.]
+Each WHILEM state block marks a point to backtrack to upon partial failure
+of A or B, and also contains some minor state data related to that
+iteration. The CURLYX block, pointed to by cur_curlyx, contains the
+overall state, such as the count, and pointers to the A and B ops.
- Matched REx position InfoBlocks Comment
- (Y(A)*?Z)*?T x
- Y(A)*?Z)*?T x <- O
- Y (A)*?Z)*?T x <- O
- Y A)*?Z)*?T x <- O <- I
- YA )*?Z)*?T x <- O <- I
- YA A)*?Z)*?T x <- O <- I
- YAA )*?Z)*?T x <- O <- I
- YAA Z)*?T x <- O # Temporary unset I
- I
+This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
+must always point to the *current* CURLYX block, the rules are:
- YAAZ Y(A)*?Z)*?T x <- O
- I
+When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
+and set cur_curlyx to point the new block.
- YAAZY (A)*?Z)*?T x <- O
- I
+When popping the CURLYX block after a successful or unsuccessful match,
+restore the previous cur_curlyx.
- YAAZY A)*?Z)*?T x <- O <- I
- I
+When WHILEM is about to execute B, save the current cur_curlyx, and set it
+to the outer one saved in the CURLYX block.
- YAAZYA )*?Z)*?T x <- O <- I
- I
+When popping the WHILEM block after a successful or unsuccessful B match,
+restore the previous cur_curlyx.
- YAAZYA Z)*?T x <- O # Temporary unset I
- I,I
+Here's an example for the pattern (AI* BI)*BO
+I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
- YAAZYAZ )*?T x <- O
- I,I
+cur_
+curlyx backtrack stack
+------ ---------------
+NULL
+CO <CO prev=NULL> <WO>
+CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
+CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
+NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
- YAAZYAZ T x # Temporary unset O
- O
- I,I
+At this point the pattern succeeds, and we work back down the stack to
+clean up, restoring as we go:
- YAAZYAZT x
- O
- I,I
- *******************************************************************/
+CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
+CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
+CO <CO prev=NULL> <WO>
+NULL
- case CURLYX: {
- /* No need to save/restore up to this paren */
- parenfloor = scan->flags;
-
- /* Dave says:
-
- CURLYX and WHILEM are always paired: they're the moral
- equivalent of pp_enteriter anbd pp_iter.
-
- The only time next could be null is if the node tree is
- corrupt. This was mentioned on p5p a few days ago.
-
- See http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2006-04/msg00556.html
- So we'll assert that this is true:
- */
- assert(next);
- if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
- next += ARG(next);
- /* XXXX Probably it is better to teach regpush to support
- parenfloor > PL_regsize... */
- if (parenfloor > (I32)*PL_reglastparen)
- parenfloor = *PL_reglastparen; /* Pessimization... */
-
- st->u.curlyx.cp = PL_savestack_ix;
- st->u.curlyx.outercc = cur_curlyx;
- cur_curlyx = st;
- /* these fields contain the state of the current curly.
- * they are accessed by subsequent WHILEMs;
- * cur and lastloc are also updated by WHILEM */
- st->u.curlyx.parenfloor = parenfloor;
- st->u.curlyx.cur = -1; /* this will be updated by WHILEM */
- st->u.curlyx.min = ARG1(scan);
- st->u.curlyx.max = ARG2(scan);
- st->u.curlyx.scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
- st->u.curlyx.lastloc = 0;
- /* st->next and st->minmod are also read by WHILEM */
+*******************************************************************/
- PL_reginput = locinput;
- REGMATCH(PREVOPER(next), CURLYX); /* start on the WHILEM */
- /*** all unsaved local vars undefined at this point */
- regcpblow(st->u.curlyx.cp);
- cur_curlyx = st->u.curlyx.outercc;
- saySAME(result);
- }
+#define ST st->u.curlyx
+
+ case CURLYX: /* start of /A*B/ (for complex A) */
+ {
+ /* No need to save/restore up to this paren */
+ I32 parenfloor = scan->flags;
+
+ assert(next); /* keep Coverity happy */
+ if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
+ next += ARG(next);
+
+ /* XXXX Probably it is better to teach regpush to support
+ parenfloor > PL_regsize... */
+ if (parenfloor > (I32)*PL_reglastparen)
+ parenfloor = *PL_reglastparen; /* Pessimization... */
+
+ ST.prev_curlyx= cur_curlyx;
+ cur_curlyx = st;
+ ST.cp = PL_savestack_ix;
+
+ /* these fields contain the state of the current curly.
+ * they are accessed by subsequent WHILEMs */
+ ST.parenfloor = parenfloor;
+ ST.min = ARG1(scan);
+ ST.max = ARG2(scan);
+ ST.A = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
+ ST.B = next;
+ ST.minmod = minmod;
+ minmod = 0;
+ ST.count = -1; /* this will be updated by WHILEM */
+ ST.lastloc = NULL; /* this will be updated by WHILEM */
+
+ PL_reginput = locinput;
+ PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
/* NOTREACHED */
- case WHILEM: {
- /*
- * This is really hard to understand, because after we match
- * what we're trying to match, we must make sure the rest of
- * the REx is going to match for sure, and to do that we have
- * to go back UP the parse tree by recursing ever deeper. And
- * if it fails, we have to reset our parent's current state
- * that we can try again after backing off.
- */
+ }
- /* Dave says:
-
- cur_curlyx gets initialised by CURLYX ready for use by WHILEM.
- So again, unless somethings been corrupted, cur_curlyx cannot
- be null at that point in WHILEM.
-
- See http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2006-04/msg00556.html
- So we'll assert that this is true:
- */
- assert(cur_curlyx);
- st->u.whilem.lastloc = cur_curlyx->u.curlyx.lastloc; /* Detection of 0-len. */
- st->u.whilem.cache_offset = 0;
- st->u.whilem.cache_bit = 0;
-
- n = cur_curlyx->u.curlyx.cur + 1; /* how many we know we matched */
- PL_reginput = locinput;
+ case CURLYX_end: /* just finished matching all of A*B */
+ regcpblow(ST.cp);
+ cur_curlyx = ST.prev_curlyx;
+ sayYES;
+ /* NOTREACHED */
- DEBUG_EXECUTE_r(
- PerlIO_printf(Perl_debug_log,
- "%*s %ld out of %ld..%ld cc=%"UVxf"\n",
- REPORT_CODE_OFF+depth*2, "",
- (long)n, (long)cur_curlyx->u.curlyx.min,
- (long)cur_curlyx->u.curlyx.max,
- PTR2UV(cur_curlyx))
- );
+ case CURLYX_end_fail: /* just failed to match all of A*B */
+ regcpblow(ST.cp);
+ cur_curlyx = ST.prev_curlyx;
+ sayNO;
+ /* NOTREACHED */
- /* If degenerate scan matches "", assume scan done. */
- if (locinput == cur_curlyx->u.curlyx.lastloc && n >=
- cur_curlyx->u.curlyx.min)
- {
- st->u.whilem.savecc = cur_curlyx;
- cur_curlyx = cur_curlyx->u.curlyx.outercc;
- DEBUG_EXECUTE_r(
- PerlIO_printf(Perl_debug_log,
- "%*s empty match detected, try continuation...\n",
- REPORT_CODE_OFF+depth*2, "")
- );
- REGMATCH(st->u.whilem.savecc->next, WHILEM1);
- /*** all unsaved local vars undefined at this point */
- cur_curlyx = st->u.whilem.savecc;
- if (result)
- sayYES;
- sayNO;
- }
+#undef ST
+#define ST st->u.whilem
- /* First just match a string of min scans. */
-
- if (n < cur_curlyx->u.curlyx.min) {
- cur_curlyx->u.curlyx.cur = n;
- cur_curlyx->u.curlyx.lastloc = locinput;
- REGMATCH(cur_curlyx->u.curlyx.scan, WHILEM2);
- /*** all unsaved local vars undefined at this point */
- if (result)
- sayYES;
- cur_curlyx->u.curlyx.cur = n - 1;
- cur_curlyx->u.curlyx.lastloc = st->u.whilem.lastloc;
- sayNO;
- }
+ case WHILEM: /* just matched an A in /A*B/ (for complex A) */
+ {
+ /* see the discussion above about CURLYX/WHILEM */
+ I32 n;
+ assert(cur_curlyx); /* keep Coverity happy */
+ n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
+ ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
+ ST.cache_offset = 0;
+ ST.cache_mask = 0;
+
+ PL_reginput = locinput;
+
+ DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
+ "%*s whilem: matched %ld out of %ld..%ld\n",
+ REPORT_CODE_OFF+depth*2, "", (long)n,
+ (long)cur_curlyx->u.curlyx.min,
+ (long)cur_curlyx->u.curlyx.max)
+ );
+
+ /* First just match a string of min A's. */
+
+ if (n < cur_curlyx->u.curlyx.min) {
+ cur_curlyx->u.curlyx.lastloc = locinput;
+ PUSH_STATE_GOTO(WHILEM_A_pre, cur_curlyx->u.curlyx.A);
+ /* NOTREACHED */
+ }
+
+ /* If degenerate A matches "", assume A done. */
+
+ if (locinput == cur_curlyx->u.curlyx.lastloc) {
+ DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
+ "%*s whilem: empty match detected, trying continuation...\n",
+ REPORT_CODE_OFF+depth*2, "")
+ );
+ goto do_whilem_B_max;
+ }
+
+ /* super-linear cache processing */
+
+ if (scan->flags) {
- if (scan->flags) {
- /* Check whether we already were at this position.
- Postpone detection until we know the match is not
- *that* much linear. */
if (!PL_reg_maxiter) {
+ /* start the countdown: Postpone detection until we
+ * know the match is not *that* much linear. */
PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
/* possible overflow for long strings and many CURLYX's */
if (PL_reg_maxiter < 0)
PL_reg_maxiter = I32_MAX;
PL_reg_leftiter = PL_reg_maxiter;
}
+
if (PL_reg_leftiter-- == 0) {
+ /* initialise cache */
const I32 size = (PL_reg_maxiter + 7)/8;
if (PL_reg_poscache) {
if ((I32)PL_reg_poscache_size < size) {
PL_reg_poscache_size = size;
Newxz(PL_reg_poscache, size, char);
}
- DEBUG_EXECUTE_r(
- PerlIO_printf(Perl_debug_log,
- "%sDetected a super-linear match, switching on caching%s...\n",
- PL_colors[4], PL_colors[5])
- );
+ DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
+ "%swhilem: Detected a super-linear match, switching on caching%s...\n",
+ PL_colors[4], PL_colors[5])
+ );
}
+
if (PL_reg_leftiter < 0) {
- st->u.whilem.cache_offset = locinput - PL_bostr;
-
- st->u.whilem.cache_offset = (scan->flags & 0xf) - 1
- + st->u.whilem.cache_offset * (scan->flags>>4);
- st->u.whilem.cache_bit = st->u.whilem.cache_offset % 8;
- st->u.whilem.cache_offset /= 8;
- if (PL_reg_poscache[st->u.whilem.cache_offset] & (1<<st->u.whilem.cache_bit)) {
- DEBUG_EXECUTE_r(
- PerlIO_printf(Perl_debug_log,
- "%*s already tried at this position...\n",
- REPORT_CODE_OFF+depth*2, "")
+ /* have we already failed at this position? */
+ I32 offset, mask;
+ offset = (scan->flags & 0xf) - 1
+ + (locinput - PL_bostr) * (scan->flags>>4);
+ mask = 1 << (offset % 8);
+ offset /= 8;
+ if (PL_reg_poscache[offset] & mask) {
+ DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
+ "%*s whilem: (cache) already tried at this position...\n",
+ REPORT_CODE_OFF+depth*2, "")
);
sayNO; /* cache records failure */
}
+ ST.cache_offset = offset;
+ ST.cache_mask = mask;
}
- }
+ }
- /* Prefer next over scan for minimal matching. */
-
- if (cur_curlyx->minmod) {
- st->u.whilem.savecc = cur_curlyx;
- cur_curlyx = cur_curlyx->u.curlyx.outercc;
- st->u.whilem.cp = regcppush(st->u.whilem.savecc->u.curlyx.parenfloor);
- REGCP_SET(st->u.whilem.lastcp);
- REGMATCH(st->u.whilem.savecc->next, WHILEM3);
- /*** all unsaved local vars undefined at this point */
- cur_curlyx = st->u.whilem.savecc;
- if (result) {
- regcpblow(st->u.whilem.cp);
- sayYES; /* All done. */
- }
- REGCP_UNWIND(st->u.whilem.lastcp);
- regcppop(rex);
-
- if (n >= cur_curlyx->u.curlyx.max) { /* Maximum greed exceeded? */
- if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
- && !(PL_reg_flags & RF_warned)) {
- PL_reg_flags |= RF_warned;
- Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
- "Complex regular subexpression recursion",
- REG_INFTY - 1);
- }
- CACHEsayNO;
- }
+ /* Prefer B over A for minimal matching. */
- DEBUG_EXECUTE_r(
- PerlIO_printf(Perl_debug_log,
- "%*s trying longer...\n",
- REPORT_CODE_OFF+depth*2, "")
- );
- /* Try scanning more and see if it helps. */
- PL_reginput = locinput;
- cur_curlyx->u.curlyx.cur = n;
- cur_curlyx->u.curlyx.lastloc = locinput;
- st->u.whilem.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
- REGCP_SET(st->u.whilem.lastcp);
- REGMATCH(cur_curlyx->u.curlyx.scan, WHILEM4);
- /*** all unsaved local vars undefined at this point */
- if (result) {
- regcpblow(st->u.whilem.cp);
- sayYES;
- }
- REGCP_UNWIND(st->u.whilem.lastcp);
- regcppop(rex);
- cur_curlyx->u.curlyx.cur = n - 1;
- cur_curlyx->u.curlyx.lastloc = st->u.whilem.lastloc;
- CACHEsayNO;
- }
+ if (cur_curlyx->u.curlyx.minmod) {
+ ST.save_curlyx = cur_curlyx;
+ cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
+ ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
+ REGCP_SET(ST.lastcp);
+ PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
+ /* NOTREACHED */
+ }
- /* Prefer scan over next for maximal matching. */
-
- if (n < cur_curlyx->u.curlyx.max) { /* More greed allowed? */
- st->u.whilem.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
- cur_curlyx->u.curlyx.cur = n;
- cur_curlyx->u.curlyx.lastloc = locinput;
- REGCP_SET(st->u.whilem.lastcp);
- REGMATCH(cur_curlyx->u.curlyx.scan, WHILEM5);
- /*** all unsaved local vars undefined at this point */
- if (result) {
- regcpblow(st->u.whilem.cp);
- sayYES;
- }
- REGCP_UNWIND(st->u.whilem.lastcp);
- regcppop(rex); /* Restore some previous $<digit>s? */
- PL_reginput = locinput;
- DEBUG_EXECUTE_r(
- PerlIO_printf(Perl_debug_log,
- "%*s failed, try continuation...\n",
- REPORT_CODE_OFF+depth*2, "")
- );
- }
- if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
- && !(PL_reg_flags & RF_warned)) {
+ /* Prefer A over B for maximal matching. */
+
+ if (n < cur_curlyx->u.curlyx.max) { /* More greed allowed? */
+ ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
+ cur_curlyx->u.curlyx.lastloc = locinput;
+ REGCP_SET(ST.lastcp);
+ PUSH_STATE_GOTO(WHILEM_A_max, cur_curlyx->u.curlyx.A);
+ /* NOTREACHED */
+ }
+ goto do_whilem_B_max;
+ }
+ /* NOTREACHED */
+
+ case WHILEM_B_min: /* just matched B in a minimal match */
+ case WHILEM_B_max: /* just matched B in a maximal match */
+ cur_curlyx = ST.save_curlyx;
+ sayYES;
+ /* NOTREACHED */
+
+ case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
+ cur_curlyx = ST.save_curlyx;
+ cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
+ cur_curlyx->u.curlyx.count--;
+ CACHEsayNO;
+ /* NOTREACHED */
+
+ case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
+ REGCP_UNWIND(ST.lastcp);
+ regcppop(rex);
+ /* FALL THROUGH */
+ case WHILEM_A_pre_fail: /* just failed to match even minimal A */
+ cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
+ cur_curlyx->u.curlyx.count--;
+ CACHEsayNO;
+ /* NOTREACHED */
+
+ case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
+ REGCP_UNWIND(ST.lastcp);
+ regcppop(rex); /* Restore some previous $<digit>s? */
+ PL_reginput = locinput;
+ DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
+ "%*s whilem: failed, trying continuation...\n",
+ REPORT_CODE_OFF+depth*2, "")
+ );
+ do_whilem_B_max:
+ if (cur_curlyx->u.curlyx.count >= REG_INFTY
+ && ckWARN(WARN_REGEXP)
+ && !(PL_reg_flags & RF_warned))
+ {
+ PL_reg_flags |= RF_warned;
+ Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
+ "Complex regular subexpression recursion",
+ REG_INFTY - 1);
+ }
+
+ /* now try B */
+ ST.save_curlyx = cur_curlyx;
+ cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
+ PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
+ /* NOTREACHED */
+
+ case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
+ cur_curlyx = ST.save_curlyx;
+ REGCP_UNWIND(ST.lastcp);
+ regcppop(rex);
+
+ if (cur_curlyx->u.curlyx.count >= cur_curlyx->u.curlyx.max) {
+ /* Maximum greed exceeded */
+ if (cur_curlyx->u.curlyx.count >= REG_INFTY
+ && ckWARN(WARN_REGEXP)
+ && !(PL_reg_flags & RF_warned))
+ {
PL_reg_flags |= RF_warned;
- Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
- "Complex regular subexpression recursion",
- REG_INFTY - 1);
+ Perl_warner(aTHX_ packWARN(WARN_REGEXP),
+ "%s limit (%d) exceeded",
+ "Complex regular subexpression recursion",
+ REG_INFTY - 1);
}
-
- /* Failed deeper matches of scan, so see if this one works. */
- st->u.whilem.savecc = cur_curlyx;
- cur_curlyx = cur_curlyx->u.curlyx.outercc;
- REGMATCH(st->u.whilem.savecc->next, WHILEM6);
- /*** all unsaved local vars undefined at this point */
- cur_curlyx = st->u.whilem.savecc;
- if (result)
- sayYES;
- cur_curlyx->u.curlyx.cur = n - 1;
- cur_curlyx->u.curlyx.lastloc = st->u.whilem.lastloc;
+ cur_curlyx->u.curlyx.count--;
CACHEsayNO;
}
+
+ DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
+ "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
+ );
+ /* Try grabbing another A and see if it helps. */
+ PL_reginput = locinput;
+ cur_curlyx->u.curlyx.lastloc = locinput;
+ ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
+ REGCP_SET(ST.lastcp);
+ PUSH_STATE_GOTO(WHILEM_A_min, ST.save_curlyx->u.curlyx.A);
/* NOTREACHED */
#undef ST
for (n = *PL_reglastparen; n > ST.lastparen; n--)
PL_regendp[n] = -1;
*PL_reglastparen = n;
+ /*dmq: *PL_reglastcloseparen = n; */
scan = ST.next_branch;
/* no more branches? */
if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ))
/* NOTREACHED */
case MINMOD:
- st->minmod = 1;
+ minmod = 1;
break;
#undef ST
ST.B = next;
ST.alen = 0;
ST.count = 0;
- ST.minmod = st->minmod;
- st->minmod = 0;
+ ST.minmod = minmod;
+ minmod = 0;
ST.c1 = CHRTEST_UNINIT;
REGCP_SET(ST.cp);
);
locinput = PL_reginput;
- if (ST.count < (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)))
+
+ if (cur_eval && cur_eval->u.eval.close_paren &&
+ cur_eval->u.eval.close_paren == ST.me->flags)
+ goto fake_end;
+
+ if ( ST.count < (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)) )
goto curlym_do_A; /* try to match another A */
goto curlym_do_B; /* try to match B */
case CURLYM_A_fail: /* just failed to match an A */
REGCP_UNWIND(ST.cp);
- if (ST.minmod || ST.count < ARG1(ST.me) /* min*/ )
+
+ if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
+ || (cur_eval && cur_eval->u.eval.close_paren &&
+ cur_eval->u.eval.close_paren == ST.me->flags))
sayNO;
curlym_do_B: /* execute the B in /A{m,n}B/ */
PL_regstartp[paren]
= HOPc(PL_reginput, -ST.alen) - PL_bostr;
PL_regendp[paren] = PL_reginput - PL_bostr;
+ /*dmq: *PL_reglastcloseparen = paren; */
}
else
PL_regendp[paren] = -1;
+ if (cur_eval && cur_eval->u.eval.close_paren &&
+ cur_eval->u.eval.close_paren == ST.me->flags)
+ {
+ if (ST.count)
+ goto fake_end;
+ else
+ sayNO;
+ }
}
+
PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
/* NOTREACHED */
if (success) { \
PL_regstartp[paren] = HOPc(locinput, -1) - PL_bostr; \
PL_regendp[paren] = locinput - PL_bostr; \
+ *PL_reglastcloseparen = paren; \
} \
else \
PL_regendp[paren] = -1; \
*PL_reglastparen = ST.paren;
ST.min = ARG1(scan); /* min to match */
ST.max = ARG2(scan); /* max to match */
+ if (cur_eval && cur_eval->u.eval.close_paren &&
+ cur_eval->u.eval.close_paren == ST.paren) {
+ ST.min=1;
+ ST.max=1;
+ }
scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
goto repeat;
case CURLY: /* /A{m,n}B/ where A is width 1 */
ST.A = scan;
ST.B = next;
PL_reginput = locinput;
- if (st->minmod) {
- st->minmod = 0;
+ if (minmod) {
+ minmod = 0;
if (ST.min && regrepeat(rex, ST.A, ST.min) < ST.min)
sayNO;
ST.count = ST.min;
}
PL_reginput = locinput;
CURLY_SETPAREN(ST.paren, ST.count);
+ if (cur_eval && cur_eval->u.eval.close_paren &&
+ cur_eval->u.eval.close_paren == ST.paren) {
+ goto fake_end;
+ }
PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
}
/* NOTREACHED */
{
curly_try_B_min:
CURLY_SETPAREN(ST.paren, ST.count);
+ if (cur_eval && cur_eval->u.eval.close_paren &&
+ cur_eval->u.eval.close_paren == ST.paren) {
+ goto fake_end;
+ }
PUSH_STATE_GOTO(CURLY_B_min, ST.B);
}
}
/* If it could work, try it. */
if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
CURLY_SETPAREN(ST.paren, ST.count);
+ if (cur_eval && cur_eval->u.eval.close_paren &&
+ cur_eval->u.eval.close_paren == ST.paren) {
+ goto fake_end;
+ }
PUSH_STATE_GOTO(CURLY_B_max, ST.B);
/* NOTREACHED */
}
#undef ST
-
case END:
+ fake_end:
if (cur_eval) {
/* we've just finished A in /(??{A})B/; now continue with B */
I32 tmpix;
st->u.eval.prev_eval = cur_eval;
cur_eval = cur_eval->u.eval.prev_eval;
DEBUG_EXECUTE_r(
- PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ...\n",
- REPORT_CODE_OFF+depth*2, ""););
+ PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %x\n",
+ REPORT_CODE_OFF+depth*2, "",(int)cur_eval););
PUSH_YES_STATE_GOTO(EVAL_AB,
st->u.eval.prev_eval->u.eval.B); /* match B */
}
char * const s = HOPBACKc(locinput, scan->flags);
if (!s) {
/* trivial fail */
- if (st->logical) {
- st->logical = 0;
- st->sw = 1 - (bool)ST.wanted;
+ if (logical) {
+ logical = 0;
+ sw = 1 - (bool)ST.wanted;
}
else if (ST.wanted)
sayNO;
do_ifmatch:
ST.me = scan;
+ ST.logical = logical;
/* execute body of (?...A) */
PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
/* NOTREACHED */
/* FALL THROUGH */
case IFMATCH_A: /* body of (?...A) succeeded */
- if (st->logical) {
- st->logical = 0;
- st->sw = (bool)ST.wanted;
+ if (ST.logical) {
+ sw = (bool)ST.wanted;
}
else if (!ST.wanted)
sayNO;
if (next == scan)
next = NULL;
break;
+ case OPFAIL:
+ sayNO;
default:
PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
PTR2UV(scan), OP(scan));
if (newst > SLAB_LAST(PL_regmatch_slab))
newst = S_push_slab(aTHX);
PL_regmatch_state = newst;
- /* XXX probably don't need to initialise these */
- newst->minmod = 0;
- newst->sw = 0;
- newst->logical = 0;
-
locinput = PL_reginput;
nextchr = UCHARAT(locinput);
continue;
/* NOTREACHED */
}
-
- /* simulate recursively calling regmatch(), but without actually
- * recursing - ie save the current state on the heap rather than on
- * the stack, then re-enter the loop. This avoids complex regexes
- * blowing the processor stack */
-
- start_recurse:
- {
- /* push new state */
- regmatch_state *oldst = st;
-
- DEBUG_STATE_pp("push");
- depth++;
- st->u.yes.prev_yes_state = yes_state;
- yes_state = st;
-
- /* grab the next free state slot */
- st++;
- if (st > SLAB_LAST(PL_regmatch_slab))
- st = S_push_slab(aTHX);
- PL_regmatch_state = st;
-
- oldst->next = next;
- oldst->n = n;
- oldst->locinput = locinput;
-
- locinput = PL_reginput;
- nextchr = UCHARAT(locinput);
- st->minmod = 0;
- st->sw = 0;
- st->logical = 0;
-
- }
}
/*
yes_state = st->u.yes.prev_yes_state;
PL_regmatch_state = st;
- switch (st->resume_state) {
- case resume_CURLYX:
- case resume_WHILEM1:
- case resume_WHILEM2:
- case resume_WHILEM3:
- case resume_WHILEM4:
- case resume_WHILEM5:
- case resume_WHILEM6:
- result = 1;
- /* restore previous state and re-enter */
- scan = st->scan;
- next = st->next;
- n = st->n;
- locinput= st->locinput;
- nextchr = UCHARAT(locinput);
- switch (st->resume_state) {
- case resume_CURLYX:
- goto resume_point_CURLYX;
- case resume_WHILEM1:
- goto resume_point_WHILEM1;
- case resume_WHILEM2:
- goto resume_point_WHILEM2;
- case resume_WHILEM3:
- goto resume_point_WHILEM3;
- case resume_WHILEM4:
- goto resume_point_WHILEM4;
- case resume_WHILEM5:
- goto resume_point_WHILEM5;
- case resume_WHILEM6:
- goto resume_point_WHILEM6;
- }
- Perl_croak(aTHX_ "unexpected whilem resume state");
-
- case EVAL_AB:
- case IFMATCH_A:
- case CURLYM_A:
- state_num = st->resume_state;
- goto reenter_switch;
-
- case CURLYM_B:
- case BRANCH_next:
- case TRIE_next:
- case CURLY_B_max:
- default:
- Perl_croak(aTHX_ "unexpected yes resume state");
- }
+ state_num = st->resume_state;
+ goto reenter_switch;
}
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
);
no_silent:
- result = 0;
-
if (depth) {
/* there's a previous state to backtrack to */
st--;
st = SLAB_LAST(PL_regmatch_slab);
}
PL_regmatch_state = st;
- scan = st->scan;
- next = st->next;
- n = st->n;
locinput= st->locinput;
nextchr = UCHARAT(locinput);
if (yes_state == st)
yes_state = st->u.yes.prev_yes_state;
- switch (st->resume_state) {
- case resume_CURLYX:
- goto resume_point_CURLYX;
- case resume_WHILEM1:
- goto resume_point_WHILEM1;
- case resume_WHILEM2:
- goto resume_point_WHILEM2;
- case resume_WHILEM3:
- goto resume_point_WHILEM3;
- case resume_WHILEM4:
- goto resume_point_WHILEM4;
- case resume_WHILEM5:
- goto resume_point_WHILEM5;
- case resume_WHILEM6:
- goto resume_point_WHILEM6;
-
- case TRIE_next:
- case EVAL_AB:
- case BRANCH_next:
- case CURLYM_A:
- case CURLYM_B:
- case IFMATCH_A:
- case CURLY_B_max:
- case CURLY_B_min:
- case CURLY_B_min_known:
- state_num = st->resume_state + 1; /* failure = success + 1 */
- goto reenter_switch;
-
- default:
- Perl_croak(aTHX_ "regexp resume memory corruption");
- }
+ state_num = st->resume_state + 1; /* failure = success + 1 */
+ goto reenter_switch;
}
+ result = 0;
final_exit: