* REGNODE_SIMPLE */
#define SIMPLE 0x02
#define SPSTART 0x04 /* Starts with * or + */
-#define TRYAGAIN 0x08 /* Weeded out a declaration. */
-#define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
+#define POSTPONED 0x08 /* (?1),(?&name), (??{...}) or similar */
+#define TRYAGAIN 0x10 /* Weeded out a declaration. */
+#define RESTART_UTF8 0x20 /* Restart, need to calcuate sizes as UTF-8 */
#define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
#define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
#define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
-/* If not already in utf8, do a longjmp back to the beginning */
-#define UTF8_LONGJMP 42 /* Choose a value not likely to ever conflict */
#define REQUIRE_UTF8 STMT_START { \
- if (! UTF) JMPENV_JUMP(UTF8_LONGJMP); \
+ if (!UTF) { \
+ *flagp = RESTART_UTF8; \
+ return NULL; \
+ } \
} STMT_END
/* This converts the named class defined in regcomp.h to its equivalent class
data->offset_float_min = l ? data->last_start_min : data->pos_min;
data->offset_float_max = (l
? data->last_start_max
- : data->pos_min + data->pos_delta);
+ : (data->pos_delta == I32_MAX ? I32_MAX : data->pos_min + data->pos_delta));
if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
data->offset_float_max = I32_MAX;
if (data->flags & SF_BEFORE_EOL)
/* OR char bitmap and class bitmap separately */
for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
cl->bitmap[i] |= or_with->bitmap[i];
- ANYOF_CLASS_OR(or_with, cl);
+ if (or_with->flags & ANYOF_CLASS) {
+ ANYOF_CLASS_OR(or_with, cl);
+ }
}
else { /* XXXX: logic is complicated, leave it along for a moment. */
cl_anything(pRExC_state, cl);
OP(scan) = EXACTFU_SS;
s += 2;
}
- else if (len == 6 /* len is the same in both ASCII and EBCDIC for these */
+ else if (len == 6 /* len is the same in both ASCII and EBCDIC
+ for these */
&& (memEQ(s, GREEK_SMALL_LETTER_IOTA_UTF8
COMBINING_DIAERESIS_UTF8
COMBINING_ACUTE_ACCENT_UTF8,
stopparen, recursed, NULL, f,depth+1);
if (min1 > minnext)
min1 = minnext;
- if (max1 < minnext + deltanext)
- max1 = minnext + deltanext;
- if (deltanext == I32_MAX)
+ if (deltanext == I32_MAX) {
is_inf = is_inf_internal = 1;
+ max1 = I32_MAX;
+ } else if (max1 < minnext + deltanext)
+ max1 = minnext + deltanext;
scan = next;
if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
pars++;
min1 = 0;
if (flags & SCF_DO_SUBSTR) {
data->pos_min += min1;
- data->pos_delta += max1 - min1;
+ if (data->pos_delta >= I32_MAX - (max1 - min1))
+ data->pos_delta = I32_MAX;
+ else
+ data->pos_delta += max1 - min1;
if (max1 != min1 || is_inf)
data->longest = &(data->longest_float);
}
min += min1;
- delta += max1 - min1;
+ if (delta == I32_MAX || I32_MAX - delta - (max1 - min1) < 0)
+ delta = I32_MAX;
+ else
+ delta += max1 - min1;
if (flags & SCF_DO_STCLASS_OR) {
cl_or(pRExC_state, data->start_class, &accum);
if (min1) {
}
min += minnext * mincount;
- is_inf_internal |= ((maxcount == REG_INFTY
- && (minnext + deltanext) > 0)
- || deltanext == I32_MAX);
+ is_inf_internal |= deltanext == I32_MAX
+ || (maxcount == REG_INFTY && minnext + deltanext > 0);
is_inf |= is_inf_internal;
- delta += (minnext + deltanext) * maxcount - minnext * mincount;
+ if (is_inf)
+ delta = I32_MAX;
+ else
+ delta += (minnext + deltanext) * maxcount - minnext * mincount;
/* Try powerful optimization CURLYX => CURLYN. */
if ( OP(oscan) == CURLYX && data
}
/* It is counted once already... */
data->pos_min += minnext * (mincount - counted);
- data->pos_delta += - counted * deltanext +
+#if 0
+PerlIO_printf(Perl_debug_log, "counted=%d deltanext=%d I32_MAX=%d minnext=%d maxcount=%d mincount=%d\n",
+ counted, deltanext, I32_MAX, minnext, maxcount, mincount);
+if (deltanext != I32_MAX)
+PerlIO_printf(Perl_debug_log, "LHS=%d RHS=%d\n", -counted * deltanext + (minnext + deltanext) * maxcount - minnext * mincount, I32_MAX - data->pos_delta);
+#endif
+ if (deltanext == I32_MAX || -counted * deltanext + (minnext + deltanext) * maxcount - minnext * mincount >= I32_MAX - data->pos_delta)
+ data->pos_delta = I32_MAX;
+ else
+ data->pos_delta += - counted * deltanext +
(minnext + deltanext) * maxcount - minnext * mincount;
if (mincount != maxcount) {
/* Cannot extend fixed substrings found inside
NEXT_OFF(oscan) += NEXT_OFF(next);
}
continue;
- default: /* REF, ANYOFV, and CLUMP only? */
+ default: /* REF, and CLUMP only? */
if (flags & SCF_DO_SUBSTR) {
SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
data->longest = &(data->longest_float);
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)
+ if (deltanext == I32_MAX) {
is_inf = is_inf_internal = 1;
+ max1 = I32_MAX;
+ } else if (max1 < (I32)(minnext + deltanext + trie->maxlen))
+ max1 = minnext + deltanext + trie->maxlen;
if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
pars++;
NULL, NULL, rx_flags, 0);
}
+
+/* upgrade pattern pat_p of length plen_p to UTF8, and if there are code
+ * blocks, recalculate the indices. Update pat_p and plen_p in-place to
+ * point to the realloced string and length.
+ *
+ * This is essentially a copy of Perl_bytes_to_utf8() with the code index
+ * stuff added */
+
+static void
+S_pat_upgrade_to_utf8(pTHX_ RExC_state_t * const pRExC_state,
+ char **pat_p, STRLEN *plen_p, int num_code_blocks)
+{
+ U8 *const src = (U8*)*pat_p;
+ U8 *dst;
+ int n=0;
+ STRLEN s = 0, d = 0;
+ bool do_end = 0;
+ GET_RE_DEBUG_FLAGS_DECL;
+
+ DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
+ "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
+
+ Newx(dst, *plen_p * 2 + 1, U8);
+
+ while (s < *plen_p) {
+ const UV uv = NATIVE_TO_ASCII(src[s]);
+ if (UNI_IS_INVARIANT(uv))
+ dst[d] = (U8)UTF_TO_NATIVE(uv);
+ else {
+ dst[d++] = (U8)UTF8_EIGHT_BIT_HI(uv);
+ dst[d] = (U8)UTF8_EIGHT_BIT_LO(uv);
+ }
+ if (n < num_code_blocks) {
+ if (!do_end && pRExC_state->code_blocks[n].start == s) {
+ pRExC_state->code_blocks[n].start = d;
+ assert(dst[d] == '(');
+ do_end = 1;
+ }
+ else if (do_end && pRExC_state->code_blocks[n].end == s) {
+ pRExC_state->code_blocks[n].end = d;
+ assert(dst[d] == ')');
+ do_end = 0;
+ n++;
+ }
+ }
+ s++;
+ d++;
+ }
+ dst[d] = '\0';
+ *plen_p = d;
+ *pat_p = (char*) dst;
+ SAVEFREEPV(*pat_p);
+ RExC_orig_utf8 = RExC_utf8 = 1;
+}
+
+
+
+/* S_concat_pat(): concatenate a list of args to the pattern string pat,
+ * while recording any code block indices, and handling overloading,
+ * nested qr// objects etc. If pat is null, it will allocate a new
+ * string, or just return the first arg, if there's only one.
+ *
+ * Returns the malloced/updated pat.
+ * patternp and pat_count is the array of SVs to be concatted;
+ * oplist is the optional list of ops that generated the SVs;
+ * recompile_p is a pointer to a boolean that will be set if
+ * the regex will need to be recompiled.
+ * delim, if non-null is an SV that will be inserted between each element
+ */
+
+static SV*
+S_concat_pat(pTHX_ RExC_state_t * const pRExC_state,
+ SV *pat, SV ** const patternp, int pat_count,
+ OP *oplist, bool *recompile_p, SV *delim)
+{
+ SV **svp;
+ int n = 0;
+ bool use_delim = FALSE;
+ bool alloced = FALSE;
+
+ /* if we know we have at least two args, create an empty string,
+ * then concatenate args to that. For no args, return an empty string */
+ if (!pat && pat_count != 1) {
+ pat = newSVpvn("", 0);
+ SAVEFREESV(pat);
+ alloced = TRUE;
+ }
+
+ for (svp = patternp; svp < patternp + pat_count; svp++) {
+ SV *sv;
+ SV *rx = NULL;
+ STRLEN orig_patlen = 0;
+ bool code = 0;
+ SV *msv = use_delim ? delim : *svp;
+
+ /* if we've got a delimiter, we go round the loop twice for each
+ * svp slot (except the last), using the delimiter the second
+ * time round */
+ if (use_delim) {
+ svp--;
+ use_delim = FALSE;
+ }
+ else if (delim)
+ use_delim = TRUE;
+
+ if (SvTYPE(msv) == SVt_PVAV) {
+ /* we've encountered an interpolated array within
+ * the pattern, e.g. /...@a..../. Expand the list of elements,
+ * then recursively append elements.
+ * The code in this block is based on S_pushav() */
+
+ AV *const av = (AV*)msv;
+ const I32 maxarg = AvFILL(av) + 1;
+ SV **array;
+
+ if (oplist) {
+ assert(oplist->op_type == OP_PADAV
+ || oplist->op_type == OP_RV2AV);
+ oplist = oplist->op_sibling;;
+ }
+
+ if (SvRMAGICAL(av)) {
+ U32 i;
+
+ Newx(array, maxarg, SV*);
+ SAVEFREEPV(array);
+ for (i=0; i < (U32)maxarg; i++) {
+ SV ** const svp = av_fetch(av, i, FALSE);
+ array[i] = svp ? *svp : &PL_sv_undef;
+ }
+ }
+ else
+ array = AvARRAY(av);
+
+ pat = S_concat_pat(aTHX_ pRExC_state, pat,
+ array, maxarg, NULL, recompile_p,
+ /* $" */
+ GvSV((gv_fetchpvs("\"", GV_ADDMULTI, SVt_PV))));
+
+ continue;
+ }
+
+
+ /* we make the assumption here that each op in the list of
+ * op_siblings maps to one SV pushed onto the stack,
+ * except for code blocks, with have both an OP_NULL and
+ * and OP_CONST.
+ * This allows us to match up the list of SVs against the
+ * list of OPs to find the next code block.
+ *
+ * Note that PUSHMARK PADSV PADSV ..
+ * is optimised to
+ * PADRANGE PADSV PADSV ..
+ * so the alignment still works. */
+
+ if (oplist) {
+ if (oplist->op_type == OP_NULL
+ && (oplist->op_flags & OPf_SPECIAL))
+ {
+ assert(n < pRExC_state->num_code_blocks);
+ pRExC_state->code_blocks[n].start = pat ? SvCUR(pat) : 0;
+ pRExC_state->code_blocks[n].block = oplist;
+ pRExC_state->code_blocks[n].src_regex = NULL;
+ n++;
+ code = 1;
+ oplist = oplist->op_sibling; /* skip CONST */
+ assert(oplist);
+ }
+ oplist = oplist->op_sibling;;
+ }
+
+ /* apply magic and QR overloading to arg */
+
+ SvGETMAGIC(msv);
+ if (SvROK(msv) && SvAMAGIC(msv)) {
+ SV *sv = AMG_CALLunary(msv, regexp_amg);
+ if (sv) {
+ if (SvROK(sv))
+ sv = SvRV(sv);
+ if (SvTYPE(sv) != SVt_REGEXP)
+ Perl_croak(aTHX_ "Overloaded qr did not return a REGEXP");
+ msv = sv;
+ }
+ }
+
+ /* try concatenation overload ... */
+ if (pat && (SvAMAGIC(pat) || SvAMAGIC(msv)) &&
+ (sv = amagic_call(pat, msv, concat_amg, AMGf_assign)))
+ {
+ sv_setsv(pat, sv);
+ /* overloading involved: all bets are off over literal
+ * code. Pretend we haven't seen it */
+ pRExC_state->num_code_blocks -= n;
+ n = 0;
+ }
+ else {
+ /* ... or failing that, try "" overload */
+ while (SvAMAGIC(msv)
+ && (sv = AMG_CALLunary(msv, string_amg))
+ && sv != msv
+ && !( SvROK(msv)
+ && SvROK(sv)
+ && SvRV(msv) == SvRV(sv))
+ ) {
+ msv = sv;
+ SvGETMAGIC(msv);
+ }
+ if (SvROK(msv) && SvTYPE(SvRV(msv)) == SVt_REGEXP)
+ msv = SvRV(msv);
+
+ if (pat) {
+ /* this is a partially unrolled
+ * sv_catsv_nomg(pat, msv);
+ * that allows us to adjust code block indices if
+ * needed */
+ STRLEN slen, dlen;
+ char *dst = SvPV_force_nomg(pat, dlen);
+ const char *src = SvPV_flags_const(msv, slen, 0);
+ orig_patlen = dlen;
+ if (SvUTF8(msv) && !SvUTF8(pat)) {
+ S_pat_upgrade_to_utf8(aTHX_ pRExC_state, &dst, &dlen, n);
+ sv_setpvn(pat, dst, dlen);
+ SvUTF8_on(pat);
+ }
+ sv_catpvn_nomg(pat, src, slen);
+ rx = msv;
+ }
+ else
+ pat = msv;
+
+ if (code)
+ pRExC_state->code_blocks[n-1].end = SvCUR(pat)-1;
+ }
+
+ /* extract any code blocks within any embedded qr//'s */
+ if (rx && SvTYPE(rx) == SVt_REGEXP
+ && RX_ENGINE((REGEXP*)rx)->op_comp)
+ {
+
+ RXi_GET_DECL(ReANY((REGEXP *)rx), ri);
+ if (ri->num_code_blocks) {
+ int i;
+ /* the presence of an embedded qr// with code means
+ * we should always recompile: the text of the
+ * qr// may not have changed, but it may be a
+ * different closure than last time */
+ *recompile_p = 1;
+ Renew(pRExC_state->code_blocks,
+ pRExC_state->num_code_blocks + ri->num_code_blocks,
+ struct reg_code_block);
+ pRExC_state->num_code_blocks += ri->num_code_blocks;
+
+ for (i=0; i < ri->num_code_blocks; i++) {
+ struct reg_code_block *src, *dst;
+ STRLEN offset = orig_patlen
+ + ReANY((REGEXP *)rx)->pre_prefix;
+ assert(n < pRExC_state->num_code_blocks);
+ src = &ri->code_blocks[i];
+ dst = &pRExC_state->code_blocks[n];
+ dst->start = src->start + offset;
+ dst->end = src->end + offset;
+ dst->block = src->block;
+ dst->src_regex = (REGEXP*) SvREFCNT_inc( (SV*)
+ src->src_regex
+ ? src->src_regex
+ : (REGEXP*)rx);
+ n++;
+ }
+ }
+ }
+ }
+ /* avoid calling magic multiple times on a single element e.g. =~ $qr */
+ if (alloced)
+ SvSETMAGIC(pat);
+
+ return pat;
+}
+
+
+
/* see if there are any run-time code blocks in the pattern.
* False positives are allowed */
static bool
-S_has_runtime_code(pTHX_ RExC_state_t * const pRExC_state, OP *expr,
- U32 pm_flags, char *pat, STRLEN plen)
+S_has_runtime_code(pTHX_ RExC_state_t * const pRExC_state,
+ char *pat, STRLEN plen)
{
int n = 0;
STRLEN s;
- /* avoid infinitely recursing when we recompile the pattern parcelled up
- * as qr'...'. A single constant qr// string can't have have any
- * run-time component in it, and thus, no runtime code. (A non-qr
- * string, however, can, e.g. $x =~ '(?{})') */
- if ((pm_flags & PMf_IS_QR) && expr && expr->op_type == OP_CONST)
- return 0;
-
for (s = 0; s < plen; s++) {
if (n < pRExC_state->num_code_blocks
&& s == pRExC_state->code_blocks[n].start)
SAVETMPS;
save_re_context();
PUSHSTACKi(PERLSI_REQUIRE);
- /* this causes the toker to collapse \\ into \ when parsing
- * qr''; normally only q'' does this. It also alters hints
- * handling */
- PL_reg_state.re_reparsing = TRUE;
- eval_sv(sv, G_SCALAR);
+ /* G_RE_REPARSING causes the toker to collapse \\ into \ when
+ * parsing qr''; normally only q'' does this. It also alters
+ * hints handling */
+ eval_sv(sv, G_SCALAR|G_RE_REPARSING);
SvREFCNT_dec_NN(sv);
SPAGAIN;
qr_ref = POPs;
S_setup_longest(pTHX_ RExC_state_t *pRExC_state, SV* sv_longest, SV** rx_utf8, SV** rx_substr, I32* rx_end_shift, I32 lookbehind, I32 offset, I32 *minlen, STRLEN longest_length, bool eol, bool meol)
{
/* This is the common code for setting up the floating and fixed length
- * string data extracted from Perlre_op_compile() below. Returns a boolean
+ * string data extracted from Perl_re_op_compile() below. Returns a boolean
* as to whether succeeded or not */
I32 t,ml;
REGEXP *
Perl_re_op_compile(pTHX_ SV ** const patternp, int pat_count,
- OP *expr, const regexp_engine* eng, REGEXP *VOL old_re,
+ OP *expr, const regexp_engine* eng, REGEXP *old_re,
bool *is_bare_re, U32 orig_rx_flags, U32 pm_flags)
{
dVAR;
struct regexp *r;
regexp_internal *ri;
STRLEN plen;
- char * VOL exp;
- char* xend;
+ char *exp;
regnode *scan;
I32 flags;
I32 minlen = 0;
U32 rx_flags;
- SV * VOL pat;
- SV * VOL code_blocksv = NULL;
+ SV *pat;
+ SV *code_blocksv = NULL;
+ SV** new_patternp = patternp;
/* these are all flags - maybe they should be turned
* into a single int with different bit masks */
I32 sawlookahead = 0;
I32 sawplus = 0;
I32 sawopen = 0;
- bool used_setjump = FALSE;
regex_charset initial_charset = get_regex_charset(orig_rx_flags);
- bool code_is_utf8 = 0;
- bool VOL recompile = 0;
+ bool recompile = 0;
bool runtime_code = 0;
- U8 jump_ret = 0;
- dJMPENV;
scan_data_t data;
RExC_state_t RExC_state;
RExC_state_t * const pRExC_state = &RExC_state;
#ifdef TRIE_STUDY_OPT
- int restudied;
+ int restudied = 0;
RExC_state_t copyRExC_state;
#endif
GET_RE_DEBUG_FLAGS_DECL;
if (expr && (expr->op_type == OP_LIST ||
(expr->op_type == OP_NULL && expr->op_targ == OP_LIST))) {
-
- /* is the source UTF8, and how many code blocks are there? */
+ /* allocate code_blocks if needed */
OP *o;
int ncode = 0;
- for (o = cLISTOPx(expr)->op_first; o; o = o->op_sibling) {
- if (o->op_type == OP_CONST && SvUTF8(cSVOPo_sv))
- code_is_utf8 = 1;
- else if (o->op_type == OP_NULL && (o->op_flags & OPf_SPECIAL))
- /* count of DO blocks */
- ncode++;
- }
+ for (o = cLISTOPx(expr)->op_first; o; o = o->op_sibling)
+ if (o->op_type == OP_NULL && (o->op_flags & OPf_SPECIAL))
+ ncode++; /* count of DO blocks */
if (ncode) {
pRExC_state->num_code_blocks = ncode;
Newx(pRExC_state->code_blocks, ncode, struct reg_code_block);
}
}
- if (pat_count) {
- /* handle a list of SVs */
+ if (!pat_count) {
+ /* compile-time pattern with just OP_CONSTs and DO blocks */
- SV **svp;
+ int n;
+ OP *o;
- /* apply magic and RE overloading to each arg */
- for (svp = patternp; svp < patternp + pat_count; svp++) {
- SV *rx = *svp;
- SvGETMAGIC(rx);
- if (SvROK(rx) && SvAMAGIC(rx)) {
- SV *sv = AMG_CALLunary(rx, regexp_amg);
- if (sv) {
- if (SvROK(sv))
- sv = SvRV(sv);
- if (SvTYPE(sv) != SVt_REGEXP)
- Perl_croak(aTHX_ "Overloaded qr did not return a REGEXP");
- *svp = sv;
- }
- }
- }
+ /* find how many CONSTs there are */
+ assert(expr);
+ n = 0;
+ if (expr->op_type == OP_CONST)
+ n = 1;
+ else
+ for (o = cLISTOPx(expr)->op_first; o; o = o->op_sibling) {
+ if (o->op_type == OP_CONST)
+ n++;
+ }
- if (pat_count > 1) {
- /* concat multiple args and find any code block indexes */
+ /* fake up an SV array */
- OP *o = NULL;
- int n = 0;
- bool utf8 = 0;
- STRLEN orig_patlen = 0;
+ assert(!new_patternp);
+ Newx(new_patternp, n, SV*);
+ SAVEFREEPV(new_patternp);
+ pat_count = n;
- if (pRExC_state->num_code_blocks) {
- o = cLISTOPx(expr)->op_first;
- assert( o->op_type == OP_PUSHMARK
- || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
- || o->op_type == OP_PADRANGE);
- o = o->op_sibling;
- }
+ n = 0;
+ if (expr->op_type == OP_CONST)
+ new_patternp[n] = cSVOPx_sv(expr);
+ else
+ for (o = cLISTOPx(expr)->op_first; o; o = o->op_sibling) {
+ if (o->op_type == OP_CONST)
+ new_patternp[n++] = cSVOPo_sv;
+ }
- pat = newSVpvn("", 0);
- SAVEFREESV(pat);
-
- /* determine if the pattern is going to be utf8 (needed
- * in advance to align code block indices correctly).
- * XXX This could fail to be detected for an arg with
- * overloading but not concat overloading; but the main effect
- * in this obscure case is to need a 'use re eval' for a
- * literal code block */
- for (svp = patternp; svp < patternp + pat_count; svp++) {
- if (SvUTF8(*svp))
- utf8 = 1;
- }
- if (utf8)
- SvUTF8_on(pat);
-
- for (svp = patternp; svp < patternp + pat_count; svp++) {
- SV *sv, *msv = *svp;
- SV *rx;
- bool code = 0;
- /* we make the assumption here that each op in the list of
- * op_siblings maps to one SV pushed onto the stack,
- * except for code blocks, with have both an OP_NULL and
- * and OP_CONST.
- * This allows us to match up the list of SVs against the
- * list of OPs to find the next code block.
- *
- * Note that PUSHMARK PADSV PADSV ..
- * is optimised to
- * PADRANGE NULL NULL ..
- * so the alignment still works. */
- if (o) {
- if (o->op_type == OP_NULL && (o->op_flags & OPf_SPECIAL)) {
- assert(n < pRExC_state->num_code_blocks);
- pRExC_state->code_blocks[n].start = SvCUR(pat);
- pRExC_state->code_blocks[n].block = o;
- pRExC_state->code_blocks[n].src_regex = NULL;
- n++;
- code = 1;
- o = o->op_sibling; /* skip CONST */
- assert(o);
- }
- o = o->op_sibling;;
- }
+ }
- if ((SvAMAGIC(pat) || SvAMAGIC(msv)) &&
- (sv = amagic_call(pat, msv, concat_amg, AMGf_assign)))
- {
- sv_setsv(pat, sv);
- /* overloading involved: all bets are off over literal
- * code. Pretend we haven't seen it */
- pRExC_state->num_code_blocks -= n;
- n = 0;
- rx = NULL;
+ DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
+ "Assembling pattern from %d elements%s\n", pat_count,
+ orig_rx_flags & RXf_SPLIT ? " for split" : ""));
- }
- else {
- while (SvAMAGIC(msv)
- && (sv = AMG_CALLunary(msv, string_amg))
- && sv != msv
- && !( SvROK(msv)
- && SvROK(sv)
- && SvRV(msv) == SvRV(sv))
- ) {
- msv = sv;
- SvGETMAGIC(msv);
- }
- if (SvROK(msv) && SvTYPE(SvRV(msv)) == SVt_REGEXP)
- msv = SvRV(msv);
- orig_patlen = SvCUR(pat);
- sv_catsv_nomg(pat, msv);
- rx = msv;
- if (code)
- pRExC_state->code_blocks[n-1].end = SvCUR(pat)-1;
- }
+ /* set expr to the first arg op */
- /* extract any code blocks within any embedded qr//'s */
- if (rx && SvTYPE(rx) == SVt_REGEXP
- && RX_ENGINE((REGEXP*)rx)->op_comp)
- {
+ if (pRExC_state->num_code_blocks
+ && expr->op_type != OP_CONST)
+ {
+ expr = cLISTOPx(expr)->op_first;
+ assert( expr->op_type == OP_PUSHMARK
+ || (expr->op_type == OP_NULL && expr->op_targ == OP_PUSHMARK)
+ || expr->op_type == OP_PADRANGE);
+ expr = expr->op_sibling;
+ }
- RXi_GET_DECL(ReANY((REGEXP *)rx), ri);
- if (ri->num_code_blocks) {
- int i;
- /* the presence of an embedded qr// with code means
- * we should always recompile: the text of the
- * qr// may not have changed, but it may be a
- * different closure than last time */
- recompile = 1;
- Renew(pRExC_state->code_blocks,
- pRExC_state->num_code_blocks + ri->num_code_blocks,
- struct reg_code_block);
- pRExC_state->num_code_blocks += ri->num_code_blocks;
- for (i=0; i < ri->num_code_blocks; i++) {
- struct reg_code_block *src, *dst;
- STRLEN offset = orig_patlen
- + ReANY((REGEXP *)rx)->pre_prefix;
- assert(n < pRExC_state->num_code_blocks);
- src = &ri->code_blocks[i];
- dst = &pRExC_state->code_blocks[n];
- dst->start = src->start + offset;
- dst->end = src->end + offset;
- dst->block = src->block;
- dst->src_regex = (REGEXP*) SvREFCNT_inc( (SV*)
- src->src_regex
- ? src->src_regex
- : (REGEXP*)rx);
- n++;
- }
- }
- }
- }
- SvSETMAGIC(pat);
- }
- else {
- SV *sv;
- pat = *patternp;
- while (SvAMAGIC(pat)
- && (sv = AMG_CALLunary(pat, string_amg))
- && sv != pat)
- {
- pat = sv;
- SvGETMAGIC(pat);
- }
- }
+ pat = S_concat_pat(aTHX_ pRExC_state, NULL, new_patternp, pat_count,
+ expr, &recompile, NULL);
- /* handle bare regex: foo =~ $re */
- {
- SV *re = pat;
- if (SvROK(re))
- re = SvRV(re);
- if (SvTYPE(re) == SVt_REGEXP) {
- if (is_bare_re)
- *is_bare_re = TRUE;
- SvREFCNT_inc(re);
- Safefree(pRExC_state->code_blocks);
- return (REGEXP*)re;
- }
- }
- }
- else {
- /* not a list of SVs, so must be a list of OPs */
- assert(expr);
- if (expr->op_type == OP_LIST) {
- int i = -1;
- bool is_code = 0;
- OP *o;
-
- pat = newSVpvn("", 0);
- SAVEFREESV(pat);
- if (code_is_utf8)
- SvUTF8_on(pat);
-
- /* given a list of CONSTs and DO blocks in expr, append all
- * the CONSTs to pat, and record the start and end of each
- * code block in code_blocks[] (each DO{} op is followed by an
- * OP_CONST containing the corresponding literal '(?{...})
- * text)
- */
- for (o = cLISTOPx(expr)->op_first; o; o = o->op_sibling) {
- if (o->op_type == OP_CONST) {
- sv_catsv(pat, cSVOPo_sv);
- if (is_code) {
- pRExC_state->code_blocks[i].end = SvCUR(pat)-1;
- is_code = 0;
- }
- }
- else if (o->op_type == OP_NULL && (o->op_flags & OPf_SPECIAL)) {
- assert(i+1 < pRExC_state->num_code_blocks);
- pRExC_state->code_blocks[++i].start = SvCUR(pat);
- pRExC_state->code_blocks[i].block = o;
- pRExC_state->code_blocks[i].src_regex = NULL;
- is_code = 1;
- }
- }
- }
- else {
- assert(expr->op_type == OP_CONST);
- pat = cSVOPx_sv(expr);
- }
+ /* handle bare (possibly after overloading) regex: foo =~ $re */
+ {
+ SV *re = pat;
+ if (SvROK(re))
+ re = SvRV(re);
+ if (SvTYPE(re) == SVt_REGEXP) {
+ if (is_bare_re)
+ *is_bare_re = TRUE;
+ SvREFCNT_inc(re);
+ Safefree(pRExC_state->code_blocks);
+ DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
+ "Precompiled pattern%s\n",
+ orig_rx_flags & RXf_SPLIT ? " for split" : ""));
+
+ return (REGEXP*)re;
+ }
}
exp = SvPV_nomg(pat, plen);
RExC_contains_locale = 0;
pRExC_state->runtime_code_qr = NULL;
- /****************** LONG JUMP TARGET HERE***********************/
- /* Longjmp back to here if have to switch in midstream to utf8 */
- if (! RExC_orig_utf8) {
- JMPENV_PUSH(jump_ret);
- used_setjump = TRUE;
- }
-
- if (jump_ret == 0) { /* First time through */
- xend = exp + plen;
-
- DEBUG_COMPILE_r({
+ DEBUG_COMPILE_r({
SV *dsv= sv_newmortal();
- RE_PV_QUOTED_DECL(s, RExC_utf8,
- dsv, exp, plen, 60);
+ RE_PV_QUOTED_DECL(s, RExC_utf8, dsv, exp, plen, 60);
PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
- PL_colors[4],PL_colors[5],s);
+ PL_colors[4],PL_colors[5],s);
});
- }
- else { /* longjumped back */
- U8 *src, *dst;
- int n=0;
- STRLEN s = 0, d = 0;
- bool do_end = 0;
-
- /* If the cause for the longjmp was other than changing to utf8, pop
- * our own setjmp, and longjmp to the correct handler */
- if (jump_ret != UTF8_LONGJMP) {
- JMPENV_POP;
- JMPENV_JUMP(jump_ret);
- }
- GET_RE_DEBUG_FLAGS;
+ redo_first_pass:
+ /* we jump here if we upgrade the pattern to utf8 and have to
+ * recompile */
- /* It's possible to write a regexp in ascii that represents Unicode
- codepoints outside of the byte range, such as via \x{100}. If we
- detect such a sequence we have to convert the entire pattern to utf8
- and then recompile, as our sizing calculation will have been based
- on 1 byte == 1 character, but we will need to use utf8 to encode
- at least some part of the pattern, and therefore must convert the whole
- thing.
- -- dmq */
- DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
- "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
-
- /* upgrade pattern to UTF8, and if there are code blocks,
- * recalculate the indices.
- * This is essentially an unrolled Perl_bytes_to_utf8() */
-
- src = (U8*)SvPV_nomg(pat, plen);
- Newx(dst, plen * 2 + 1, U8);
-
- while (s < plen) {
- const UV uv = NATIVE_TO_ASCII(src[s]);
- if (UNI_IS_INVARIANT(uv))
- dst[d] = (U8)UTF_TO_NATIVE(uv);
- else {
- dst[d++] = (U8)UTF8_EIGHT_BIT_HI(uv);
- dst[d] = (U8)UTF8_EIGHT_BIT_LO(uv);
- }
- if (n < pRExC_state->num_code_blocks) {
- if (!do_end && pRExC_state->code_blocks[n].start == s) {
- pRExC_state->code_blocks[n].start = d;
- assert(dst[d] == '(');
- do_end = 1;
- }
- else if (do_end && pRExC_state->code_blocks[n].end == s) {
- pRExC_state->code_blocks[n].end = d;
- assert(dst[d] == ')');
- do_end = 0;
- n++;
- }
- }
- s++;
- d++;
- }
- dst[d] = '\0';
- plen = d;
- exp = (char*) dst;
- xend = exp + plen;
- SAVEFREEPV(exp);
- RExC_orig_utf8 = RExC_utf8 = 1;
- }
+ if ((pm_flags & PMf_USE_RE_EVAL)
+ /* this second condition covers the non-regex literal case,
+ * i.e. $foo =~ '(?{})'. */
+ || (IN_PERL_COMPILETIME && (PL_hints & HINT_RE_EVAL))
+ )
+ runtime_code = S_has_runtime_code(aTHX_ pRExC_state, exp, plen);
/* return old regex if pattern hasn't changed */
+ /* XXX: note in the below we have to check the flags as well as the pattern.
+ *
+ * Things get a touch tricky as we have to compare the utf8 flag independently
+ * from the compile flags.
+ */
if ( old_re
&& !recompile
- && !!RX_UTF8(old_re) == !!RExC_utf8
+ && !!RX_UTF8(old_re) == !!RExC_utf8
+ && ( RX_COMPFLAGS(old_re) == ( orig_rx_flags & RXf_PMf_FLAGCOPYMASK ) )
&& RX_PRECOMP(old_re)
&& RX_PRELEN(old_re) == plen
- && memEQ(RX_PRECOMP(old_re), exp, plen))
+ && memEQ(RX_PRECOMP(old_re), exp, plen)
+ && !runtime_code /* with runtime code, always recompile */ )
{
- /* with runtime code, always recompile */
- runtime_code = S_has_runtime_code(aTHX_ pRExC_state, expr, pm_flags,
- exp, plen);
- if (!runtime_code) {
- if (used_setjump) {
- JMPENV_POP;
- }
- Safefree(pRExC_state->code_blocks);
- return old_re;
- }
+ Safefree(pRExC_state->code_blocks);
+ return old_re;
}
- else if ((pm_flags & PMf_USE_RE_EVAL)
- /* this second condition covers the non-regex literal case,
- * i.e. $foo =~ '(?{})'. */
- || ( !PL_reg_state.re_reparsing && IN_PERL_COMPILETIME
- && (PL_hints & HINT_RE_EVAL))
- )
- runtime_code = S_has_runtime_code(aTHX_ pRExC_state, expr, pm_flags,
- exp, plen);
-
-#ifdef TRIE_STUDY_OPT
- restudied = 0;
-#endif
rx_flags = orig_rx_flags;
if (!S_compile_runtime_code(aTHX_ pRExC_state, exp, plen)) {
/* whoops, we have a non-utf8 pattern, whilst run-time code
* got compiled as utf8. Try again with a utf8 pattern */
- JMPENV_JUMP(UTF8_LONGJMP);
+ S_pat_upgrade_to_utf8(aTHX_ pRExC_state, &exp, &plen,
+ pRExC_state->num_code_blocks);
+ goto redo_first_pass;
}
}
assert(!pRExC_state->runtime_code_qr);
/* First pass: determine size, legality. */
RExC_parse = exp;
RExC_start = exp;
- RExC_end = xend;
+ RExC_end = exp + plen;
RExC_naughty = 0;
RExC_npar = 1;
RExC_nestroot = 0;
pRExC_state->code_blocks. We cannot SAVEFREEPV it now, as we may
need it to survive as long as the regexp (qr/(?{})/).
We must check that code_blocksv is not already set, because we may
- have longjmped back. */
+ have jumped back to restart the sizing pass. */
if (pRExC_state->code_blocks && !code_blocksv) {
code_blocksv = newSV_type(SVt_PV);
SAVEFREESV(code_blocksv);
SvLEN_set(code_blocksv, 1); /*sufficient to make sv_clear free it*/
}
if (reg(pRExC_state, 0, &flags,1) == NULL) {
- RExC_precomp = NULL;
- return(NULL);
+ /* It's possible to write a regexp in ascii that represents Unicode
+ codepoints outside of the byte range, such as via \x{100}. If we
+ detect such a sequence we have to convert the entire pattern to utf8
+ and then recompile, as our sizing calculation will have been based
+ on 1 byte == 1 character, but we will need to use utf8 to encode
+ at least some part of the pattern, and therefore must convert the whole
+ thing.
+ -- dmq */
+ if (flags & RESTART_UTF8) {
+ S_pat_upgrade_to_utf8(aTHX_ pRExC_state, &exp, &plen,
+ pRExC_state->num_code_blocks);
+ goto redo_first_pass;
+ }
+ Perl_croak(aTHX_ "panic: reg returned NULL to re_op_compile for sizing pass, flags=%#"UVxf"", (UV) flags);
}
if (code_blocksv)
SvLEN_set(code_blocksv,0); /* no you can't have it, sv_clear */
- /* Here, finished first pass. Get rid of any added setjmp */
- if (used_setjump) {
- JMPENV_POP;
- }
-
DEBUG_PARSE_r({
PerlIO_printf(Perl_debug_log,
"Required size %"IVdf" nodes\n"
RXi_SET( r, ri );
r->engine= eng;
r->extflags = rx_flags;
+ RXp_COMPFLAGS(r) = orig_rx_flags & RXf_PMf_FLAGCOPYMASK;
+
if (pm_flags & PMf_IS_QR) {
ri->code_blocks = pRExC_state->code_blocks;
ri->num_code_blocks = pRExC_state->num_code_blocks;
RExC_flags = rx_flags; /* don't let top level (?i) bleed */
RExC_pm_flags = pm_flags;
RExC_parse = exp;
- RExC_end = xend;
+ RExC_end = exp + plen;
RExC_naughty = 0;
RExC_npar = 1;
RExC_emit_start = ri->program;
REGC((U8)REG_MAGIC, (char*) RExC_emit++);
if (reg(pRExC_state, 0, &flags,1) == NULL) {
ReREFCNT_dec(rx);
- return(NULL);
+ Perl_croak(aTHX_ "panic: reg returned NULL to re_op_compile for generation pass, flags=%#"UVxf"", (UV) flags);
}
/* XXXX To minimize changes to RE engine we always allocate
3-units-long substrs field. */
if (RExC_seen & REG_SEEN_GPOS)
r->extflags |= RXf_GPOS_SEEN;
if (RExC_seen & REG_SEEN_LOOKBEHIND)
- r->extflags |= RXf_LOOKBEHIND_SEEN;
+ r->extflags |= RXf_NO_INPLACE_SUBST; /* inplace might break the lookbehind */
if (pRExC_state->num_code_blocks)
r->extflags |= RXf_EVAL_SEEN;
if (RExC_seen & REG_SEEN_CANY)
if (RExC_seen & REG_SEEN_VERBARG)
{
r->intflags |= PREGf_VERBARG_SEEN;
- r->extflags |= RXf_MODIFIES_VARS;
+ r->extflags |= RXf_NO_INPLACE_SUBST; /* don't understand this! Yves */
}
if (RExC_seen & REG_SEEN_CUTGROUP)
r->intflags |= PREGf_CUTGROUP_SEEN;
else
RXp_PAREN_NAMES(r) = NULL;
-#ifdef STUPID_PATTERN_CHECKS
- if (RX_PRELEN(rx) == 0)
- r->extflags |= RXf_NULL;
- if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3))
- r->extflags |= RXf_WHITE;
- else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^')
- r->extflags |= RXf_START_ONLY;
-#else
{
regnode *first = ri->program + 1;
U8 fop = OP(first);
+ regnode *next = NEXTOPER(first);
+ U8 nop = OP(next);
- if (PL_regkind[fop] == NOTHING && OP(NEXTOPER(first)) == END)
+ if (PL_regkind[fop] == NOTHING && nop == END)
r->extflags |= RXf_NULL;
- else if (PL_regkind[fop] == BOL && OP(NEXTOPER(first)) == END)
+ else if (PL_regkind[fop] == BOL && nop == END)
r->extflags |= RXf_START_ONLY;
- else if (fop == PLUS && PL_regkind[OP(NEXTOPER(first))] == POSIXD && FLAGS(NEXTOPER(first)) == _CC_SPACE
- && OP(regnext(first)) == END)
- r->extflags |= RXf_WHITE;
+ else if (fop == PLUS && PL_regkind[nop] == POSIXD && FLAGS(next) == _CC_SPACE && OP(regnext(first)) == END)
+ r->extflags |= RXf_WHITE;
+ else if ( r->extflags & RXf_SPLIT && fop == EXACT && STR_LEN(first) == 1 && *(STRING(first)) == ' ' && OP(regnext(first)) == END )
+ r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
+
}
-#endif
#ifdef DEBUGGING
if (RExC_paren_names) {
ri->name_list_idx = add_data( pRExC_state, 1, "a" );
#define WASTED_O 0x01
#define WASTED_G 0x02
#define WASTED_C 0x04
-#define WASTED_GC (0x02|0x04)
+#define WASTED_GC (WASTED_G|WASTED_C)
I32 wastedflags = 0x00;
U32 posflags = 0, negflags = 0;
U32 *flagsp = &posflags;
#define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
#endif
+/* Returns NULL, setting *flagp to TRYAGAIN at the end of (?) that only sets
+ flags. Returns NULL, setting *flagp to RESTART_UTF8 if the sizing scan
+ needs to be restarted.
+ Otherwise would only return NULL if regbranch() returns NULL, which
+ cannot happen. */
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. */
+ /* paren: Parenthesized? 0=top; 1,2=inside '(': changed to letter.
+ * 2 is like 1, but indicates that nextchar() has been called to advance
+ * RExC_parse beyond the '('. Things like '(?' are indivisible tokens, and
+ * this flag alerts us to the need to check for that */
{
dVAR;
regnode *ret; /* Will be the head of the group. */
/* Make an OPEN node, if parenthesized. */
if (paren) {
+
+ /* Under /x, space and comments can be gobbled up between the '(' and
+ * here (if paren ==2). The forms '(*VERB' and '(?...' disallow such
+ * intervening space, as the sequence is a token, and a token should be
+ * indivisible */
+ bool has_intervening_patws = paren == 2 && *(RExC_parse - 1) != '(';
+
if ( *RExC_parse == '*') { /* (*VERB:ARG) */
char *start_verb = RExC_parse;
STRLEN verb_len = 0;
unsigned char op = 0;
int argok = 1;
int internal_argval = 0; /* internal_argval is only useful if !argok */
+
+ if (has_intervening_patws && SIZE_ONLY) {
+ ckWARNregdep(RExC_parse + 1, "In '(*VERB...)', splitting the initial '(*' is deprecated");
+ }
while ( *RExC_parse && *RExC_parse != ')' ) {
if ( *RExC_parse == ':' ) {
start_arg = RExC_parse + 1;
if (*RExC_parse == '?') { /* (?...) */
bool is_logical = 0;
const char * const seqstart = RExC_parse;
+ if (has_intervening_patws && SIZE_ONLY) {
+ ckWARNregdep(RExC_parse + 1, "In '(?...)', splitting the initial '(?' is deprecated");
+ }
RExC_parse++;
paren = *RExC_parse++;
case '@': /* (?@...) */
vFAIL2("Sequence (?%c...) not implemented", (int)paren);
break;
- case '#': /* (?#...) */
- while (*RExC_parse && *RExC_parse != ')')
+ case '#': /* (?#...) */
+ /* XXX As soon as we disallow separating the '?' and '*' (by
+ * spaces or (?#...) comment), it is believed that this case
+ * will be unreachable and can be removed. See
+ * [perl #117327] */
+ while (*RExC_parse && *RExC_parse != ')')
RExC_parse++;
if (*RExC_parse != ')')
FAIL("Sequence (?#... not terminated");
|| RExC_parse[1] == '<'
|| RExC_parse[1] == '{') { /* Lookahead or eval. */
I32 flag;
+ regnode *tail;
ret = reg_node(pRExC_state, LOGICAL);
if (!SIZE_ONLY)
ret->flags = 1;
- REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
+
+ tail = reg(pRExC_state, 1, &flag, depth+1);
+ if (flag & RESTART_UTF8) {
+ *flagp = RESTART_UTF8;
+ return NULL;
+ }
+ REGTAIL(pRExC_state, ret, tail);
goto insert_if;
}
}
insert_if:
REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
br = regbranch(pRExC_state, &flags, 1,depth+1);
- if (br == NULL)
- br = reganode(pRExC_state, LONGJMP, 0);
- else
+ if (br == NULL) {
+ if (flags & RESTART_UTF8) {
+ *flagp = RESTART_UTF8;
+ return NULL;
+ }
+ FAIL2("panic: regbranch returned NULL, flags=%#"UVxf"",
+ (UV) flags);
+ } else
REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
c = *nextchar(pRExC_state);
if (flags&HASWIDTH)
if (is_define)
vFAIL("(?(DEFINE)....) does not allow branches");
lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
- regbranch(pRExC_state, &flags, 1,depth+1);
+ if (!regbranch(pRExC_state, &flags, 1,depth+1)) {
+ if (flags & RESTART_UTF8) {
+ *flagp = RESTART_UTF8;
+ return NULL;
+ }
+ FAIL2("panic: regbranch returned NULL, flags=%#"UVxf"",
+ (UV) flags);
+ }
REGTAIL(pRExC_state, ret, lastbr);
if (flags&HASWIDTH)
*flagp |= HASWIDTH;
/* branch_len = (paren != 0); */
- if (br == NULL)
- return(NULL);
+ if (br == NULL) {
+ if (flags & RESTART_UTF8) {
+ *flagp = RESTART_UTF8;
+ return NULL;
+ }
+ FAIL2("panic: regbranch returned NULL, flags=%#"UVxf"", (UV) flags);
+ }
if (*RExC_parse == '|') {
if (!SIZE_ONLY && RExC_extralen) {
reginsert(pRExC_state, BRANCHJ, br, depth+1);
}
br = regbranch(pRExC_state, &flags, 0, depth+1);
- if (br == NULL)
- return(NULL);
+ if (br == NULL) {
+ if (flags & RESTART_UTF8) {
+ *flagp = RESTART_UTF8;
+ return NULL;
+ }
+ FAIL2("panic: regbranch returned NULL, flags=%#"UVxf"", (UV) flags);
+ }
REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
lastbr = br;
*flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
case ':':
ender = reg_node(pRExC_state, TAIL);
break;
- case 1:
+ case 1: case 2:
ender = reganode(pRExC_state, CLOSE, parno);
if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
/* Check for proper termination. */
if (paren) {
- RExC_flags = oregflags;
+ /* restore original flags, but keep (?p) */
+ RExC_flags = oregflags | (RExC_flags & RXf_PMf_KEEPCOPY);
if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
RExC_parse = oregcomp_parse;
vFAIL("Unmatched (");
- regbranch - one alternative of an | operator
*
* Implements the concatenation operator.
+ *
+ * Returns NULL, setting *flagp to RESTART_UTF8 if the sizing scan needs to be
+ * restarted.
*/
STATIC regnode *
S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
if (latest == NULL) {
if (flags & TRYAGAIN)
continue;
- return(NULL);
+ if (flags & RESTART_UTF8) {
+ *flagp = RESTART_UTF8;
+ return NULL;
+ }
+ FAIL2("panic: regpiece returned NULL, flags=%#"UVxf"", (UV) flags);
}
else if (ret == NULL)
ret = latest;
* both the endmarker for their branch list and the body of the last branch.
* It might seem that this node could be dispensed with entirely, but the
* endmarker role is not redundant.
+ *
+ * Returns NULL, setting *flagp to TRYAGAIN if regatom() returns NULL with
+ * TRYAGAIN.
+ * Returns NULL, setting *flagp to RESTART_UTF8 if the sizing scan needs to be
+ * restarted.
*/
STATIC regnode *
S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
ret = regatom(pRExC_state, &flags,depth+1);
if (ret == NULL) {
- if (flags & TRYAGAIN)
- *flagp |= TRYAGAIN;
+ if (flags & (TRYAGAIN|RESTART_UTF8))
+ *flagp |= flags & (TRYAGAIN|RESTART_UTF8);
+ else
+ FAIL2("panic: regatom returned NULL, flags=%#"UVxf"", (UV) flags);
return(NULL);
}
The function raises an error (via vFAIL), and doesn't return for various
syntax errors. Otherwise it returns TRUE and sets <node_p> or <valuep> on
- success; it returns FALSE otherwise.
+ success; it returns FALSE otherwise. Returns FALSE, setting *flagp to
+ RESTART_UTF8 if the sizing scan needs to be restarted. Such a restart is
+ only possible if node_p is non-NULL.
+
If <valuep> is non-null, it means the caller can accept an input sequence
consisting of a just a single code point; <*valuep> is set to that value
/* The values are Unicode, and therefore not subject to recoding */
RExC_override_recoding = 1;
- *node_p = reg(pRExC_state, 1, &flags, depth+1);
+ if (!(*node_p = reg(pRExC_state, 1, &flags, depth+1))) {
+ if (flags & RESTART_UTF8) {
+ *flagp = RESTART_UTF8;
+ return FALSE;
+ }
+ FAIL2("panic: reg returned NULL to grok_bslash_N, flags=%#"UVxf"",
+ (UV) flags);
+ }
*flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
RExC_parse = endbrace;
* additionally will populate the node's STRING with <code_point>, if <len>
* is 0. In both cases <*flagp> is appropriately set
*
- * It knows that under FOLD, UTF characters and the Latin Sharp S must be
- * folded (the latter only when the rules indicate it can match 'ss') */
+ * It knows that under FOLD, the Latin Sharp S and UTF characters above
+ * 255, must be folded (the former only when the rules indicate it can
+ * match 'ss') */
bool len_passed_in = cBOOL(len != 0);
U8 character[UTF8_MAXBYTES_CASE+1];
if (! len_passed_in) {
if (UTF) {
- if (FOLD) {
- to_uni_fold(NATIVE_TO_UNI(code_point), character, &len);
+ if (FOLD && (! LOC || code_point > 255)) {
+ _to_uni_fold_flags(NATIVE_TO_UNI(code_point),
+ character,
+ &len,
+ FOLD_FLAGS_FULL | ((LOC)
+ ? FOLD_FLAGS_LOCALE
+ : (ASCII_FOLD_RESTRICTED)
+ ? FOLD_FLAGS_NOMIX_ASCII
+ : 0));
}
else {
uvchr_to_utf8( character, code_point);
escape sequences, with the one for handling literal escapes requiring
a dummy entry for all of the special escapes that are actually handled
by the other.
+
+ Returns NULL, setting *flagp to TRYAGAIN if reg() returns NULL with
+ TRYAGAIN.
+ Returns NULL, setting *flagp to RESTART_UTF8 if the sizing scan needs to be
+ restarted.
+ Otherwise does not return NULL.
*/
STATIC regnode *
RExC_parse = oregcomp_parse;
vFAIL("Unmatched [");
}
+ if (ret == NULL) {
+ if (*flagp & RESTART_UTF8)
+ return NULL;
+ FAIL2("panic: regclass returned NULL to regatom, flags=%#"UVxf"",
+ (UV) *flagp);
+ }
nextchar(pRExC_state);
Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
break;
}
case '(':
nextchar(pRExC_state);
- ret = reg(pRExC_state, 1, &flags,depth+1);
+ ret = reg(pRExC_state, 2, &flags,depth+1);
if (ret == NULL) {
if (flags & TRYAGAIN) {
if (RExC_parse == RExC_end) {
}
goto tryagain;
}
- return(NULL);
+ if (flags & RESTART_UTF8) {
+ *flagp = RESTART_UTF8;
+ return NULL;
+ }
+ FAIL2("panic: reg returned NULL to regatom, flags=%#"UVxf"", (UV) flags);
}
*flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
break;
It would be a bug if these returned
non-portables */
NULL);
+ /* regclass() can only return RESTART_UTF8 if multi-char folds
+ are allowed. */
+ if (!ret)
+ FAIL2("panic: regclass returned NULL to regatom, flags=%#"UVxf"",
+ (UV) *flagp);
RExC_parse--;
++RExC_parse;
if (! grok_bslash_N(pRExC_state, &ret, NULL, flagp, depth, FALSE,
FALSE /* not strict */ )) {
+ if (*flagp & RESTART_UTF8)
+ return NULL;
RExC_parse--;
goto defchar;
}
flagp, depth, FALSE,
FALSE /* not strict */ ))
{
+ if (*flagp & RESTART_UTF8)
+ FAIL("panic: grok_bslash_N set RESTART_UTF8");
RExC_parse = p = oldp;
goto loopdone;
}
#define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
PERL_STATIC_INLINE I32
-S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value, SV *free_me,
- const bool strict)
+S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value, const bool strict)
{
dVAR;
I32 namedclass = OOB_NAMEDCLASS;
the class closes */
while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
RExC_parse++;
- SvREFCNT_dec(free_me);
vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
}
} else {
RExC_parse++;
}
- (void) regclass(pRExC_state, flagp,depth+1,
- is_posix_class, /* parse the whole char
- class only if not a
- posix class */
- FALSE, /* don't allow multi-char folds */
- TRUE, /* silence non-portable warnings. */
- ¤t);
+ /* regclass() can only return RESTART_UTF8 if multi-char
+ folds are allowed. */
+ if (!regclass(pRExC_state, flagp,depth+1,
+ is_posix_class, /* parse the whole char
+ class only if not a
+ posix class */
+ FALSE, /* don't allow multi-char folds */
+ TRUE, /* silence non-portable warnings. */
+ ¤t))
+ FAIL2("panic: regclass returned NULL to handle_sets, flags=%#"UVxf"",
+ (UV) *flagp);
+
/* function call leaves parse pointing to the ']', except
* if we faked it */
if (is_posix_class) {
vFAIL("Unexpected character");
case '\\':
- (void) regclass(pRExC_state, flagp,depth+1,
- TRUE, /* means parse just the next thing */
- FALSE, /* don't allow multi-char folds */
- FALSE, /* don't silence non-portable warnings.
- */
- ¤t);
+ /* regclass() can only return RESTART_UTF8 if multi-char
+ folds are allowed. */
+ if (!regclass(pRExC_state, flagp,depth+1,
+ TRUE, /* means parse just the next thing */
+ FALSE, /* don't allow multi-char folds */
+ FALSE, /* don't silence non-portable warnings. */
+ ¤t))
+ FAIL2("panic: regclass returned NULL to handle_sets, flags=%#"UVxf"",
+ (UV) *flagp);
/* regclass() will return with parsing just the \ sequence,
* leaving the parse pointer at the next thing to parse */
RExC_parse--;
RExC_parse++;
}
- (void) regclass(pRExC_state, flagp,depth+1,
- is_posix_class, /* parse the whole char class
- only if not a posix class */
- FALSE, /* don't allow multi-char folds */
- FALSE, /* don't silence non-portable warnings.
- */
- ¤t);
+ /* regclass() can only return RESTART_UTF8 if multi-char
+ folds are allowed. */
+ if(!regclass(pRExC_state, flagp,depth+1,
+ is_posix_class, /* parse the whole char class
+ only if not a posix class */
+ FALSE, /* don't allow multi-char folds */
+ FALSE, /* don't silence non-portable warnings. */
+ ¤t))
+ FAIL2("panic: regclass returned NULL to handle_sets, flags=%#"UVxf"",
+ (UV) *flagp);
/* function call leaves parse pointing to the ']', except if we
* faked it */
if (is_posix_class) {
* already has all folding taken into consideration, and we don't want
* regclass() to add to that */
RExC_flags &= ~RXf_PMf_FOLD;
+ /* regclass() can only return RESTART_UTF8 if multi-char folds are allowed.
+ */
node = regclass(pRExC_state, flagp,depth+1,
FALSE, /* means parse the whole char class */
FALSE, /* don't allow multi-char folds */
well have generated non-portable code points, but
they're valid on this machine */
NULL);
+ if (!node)
+ FAIL2("panic: regclass returned NULL to handle_sets, flags=%#"UVxf,
+ PTR2UV(flagp));
if (save_fold) {
RExC_flags |= RXf_PMf_FOLD;
}
* corresponding bit set if that character is in the list. For characters
* above 255, a range list or swash is used. There are extra bits for \w,
* etc. in locale ANYOFs, as what these match is not determinable at
- * compile time */
+ * compile time
+ *
+ * Returns NULL, setting *flagp to RESTART_UTF8 if the sizing scan needs
+ * to be restarted. This can only happen if ret_invlist is non-NULL.
+ */
dVAR;
UV prevvalue = OOB_UNICODE, save_prevvalue = OOB_UNICODE;
if (LOC) {
ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
}
- listsv = newSVpvs("# comment\n");
+ listsv = newSVpvs_flags("# comment\n", SVs_TEMP);
initial_listsv_len = SvCUR(listsv);
+ SvTEMP_off(listsv); /* Grr, TEMPs and mortals are conflated. */
}
if (skip_white) {
s++;
if (*s && c == *s && s[1] == ']') {
SAVEFREESV(RExC_rx_sv);
- SAVEFREESV(listsv);
ckWARN3reg(s+2,
"POSIX syntax [%c %c] belongs inside character classes",
c, c);
(void)ReREFCNT_inc(RExC_rx_sv);
- SvREFCNT_inc_simple_void_NN(listsv);
}
}
&& RExC_parse < RExC_end
&& POSIXCC(UCHARAT(RExC_parse)))
{
- namedclass = regpposixcc(pRExC_state, value, listsv, strict);
+ namedclass = regpposixcc(pRExC_state, value, strict);
}
else if (value == '\\') {
if (UTF) {
TRUE, /* => charclass */
strict))
{
+ if (*flagp & RESTART_UTF8)
+ FAIL("panic: grok_bslash_N set RESTART_UTF8");
goto parseit;
}
}
value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
RExC_parse += numlen;
if (numlen != 3) {
- SAVEFREESV(listsv); /* In case warnings are fatalized */
if (strict) {
RExC_parse += (UTF) ? UTF8SKIP(RExC_parse) : 1;
vFAIL("Need exactly 3 octal digits");
form_short_octal_warning(RExC_parse, numlen));
(void)ReREFCNT_inc(RExC_rx_sv);
}
- SvREFCNT_inc_simple_void_NN(listsv);
}
if (PL_encoding && value < 0x100)
goto recode_encoding;
default:
/* Allow \_ to not give an error */
if (!SIZE_ONLY && isWORDCHAR(value) && value != '_') {
- SAVEFREESV(listsv);
if (strict) {
vFAIL2("Unrecognized escape \\%c in character class",
(int)value);
(int)value);
(void)ReREFCNT_inc(RExC_rx_sv);
}
- SvREFCNT_inc_simple_void_NN(listsv);
}
break;
} /* End of switch on char following backslash */
const int w = (RExC_parse >= rangebegin)
? RExC_parse - rangebegin
: 0;
- SAVEFREESV(listsv); /* in case of fatal warnings */
if (strict) {
vFAIL4("False [] range \"%*.*s\"", w, w, rangebegin);
}
cp_list = add_cp_to_invlist(cp_list, '-');
cp_list = add_cp_to_invlist(cp_list, prevvalue);
}
- SvREFCNT_inc_simple_void_NN(listsv);
}
range = 0; /* this was not a true range */
#ifndef EBCDIC
cp_list = _add_range_to_invlist(cp_list, prevvalue, value);
#else
- UV* this_range = _new_invlist(1);
+ SV* this_range = _new_invlist(1);
_append_range_to_invlist(this_range, prevvalue, value);
/* In EBCDIC, the ranges 'A-Z' and 'a-z' are each not contiguous.
&& (prevvalue >= 'a' && value <= 'z')
|| (prevvalue >= 'A' && value <= 'Z'))
{
- _invlist_intersection(this_range, PL_ASCII, &this_range, );
- _invlist_intersection(this_range, PL_Alpha, &this_range, );
+ _invlist_intersection(this_range, PL_Posix_ptrs[_CC_ALPHA],
+ &this_range);
}
_invlist_union(cp_list, this_range, &cp_list);
literal_endpoint = 0;
ret = reg(pRExC_state, 1, ®_flags, depth+1);
- *flagp |= reg_flags&(HASWIDTH|SIMPLE|SPSTART|POSTPONED);
+ *flagp |= reg_flags&(HASWIDTH|SIMPLE|SPSTART|POSTPONED|RESTART_UTF8);
RExC_parse = save_parse;
RExC_end = save_end;
RExC_in_multi_char_class = 0;
SvREFCNT_dec_NN(multi_char_matches);
- SvREFCNT_dec_NN(listsv);
return ret;
}
RExC_parse = (char *) cur_parse;
SvREFCNT_dec(posixes);
- SvREFCNT_dec_NN(listsv);
SvREFCNT_dec(cp_list);
return ret;
}
else {
RExC_emit = orig_emit;
}
- return END;
+ return orig_emit;
}
/* If we didn't do folding, it's because some information isn't available
}
SvREFCNT_dec_NN(cp_list);
- SvREFCNT_dec_NN(listsv);
return ret;
}
}
&& ! HAS_NONLOCALE_RUNTIME_PROPERTY_DEFINITION)
{
ARG_SET(ret, ANYOF_NONBITMAP_EMPTY);
- SvREFCNT_dec_NN(listsv);
}
else {
/* av[0] stores the character class description in its textual form:
SV *rv;
av_store(av, 0, (HAS_NONLOCALE_RUNTIME_PROPERTY_DEFINITION)
- ? listsv
- : (SvREFCNT_dec_NN(listsv), &PL_sv_undef));
+ ? SvREFCNT_inc(listsv) : &PL_sv_undef);
if (swash) {
av_store(av, 1, swash);
SvREFCNT_dec_NN(cp_list);
npar = r->nparens+1;
Newx(ret->offs, npar, regexp_paren_pair);
Copy(r->offs, ret->offs, npar, regexp_paren_pair);
- if(ret->swap) {
- /* no need to copy these */
- Newx(ret->swap, npar, regexp_paren_pair);
- }
if (ret->substrs) {
/* Do it this way to avoid reading from *r after the StructCopy().
EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all
ones (binary 1111 1111, hexadecimal FF). It is similar, but not
- identical, to the ASCII delete (DEL) or rubout control character.
- ) So the old condition can be simplified to !isPRINT(c) */
+ identical, to the ASCII delete (DEL) or rubout control character. ...
+ it is typically mapped to hexadecimal code 9F, in order to provide a
+ unique character mapping in both directions)
+
+ So the old condition can be simplified to !isPRINT(c) */
if (!isPRINT(c)) {
if (c < 256) {
Perl_sv_catpvf(aTHX_ sv, "\\x%02x", c);