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,
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_internal *ri;
STRLEN plen;
char *exp;
- char* xend;
regnode *scan;
I32 flags;
I32 minlen = 0;
U32 rx_flags;
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 sawplus = 0;
I32 sawopen = 0;
regex_charset initial_charset = get_regex_charset(orig_rx_flags);
- bool code_is_utf8 = 0;
bool recompile = 0;
bool runtime_code = 0;
scan_data_t data;
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) {
- /* skip if we have SVs as well as OPs. In this case,
- * a) we decide utf8 based on SVs not OPs;
- * b) the current pad may not match that which the ops
- * were compiled in, so, so on threaded builds,
- * cSVOPo_sv would look in the wrong pad */
- if (!pat_count && 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);
- xend = exp + plen;
if (!eng->op_comp) {
if ((SvUTF8(pat) && IN_BYTES)
PL_colors[4],PL_colors[5],s);
});
- if (0) {
- redo_first_pass:
- {
- U8 *const src = (U8*)exp;
- U8 *dst;
- int n=0;
- STRLEN s = 0, d = 0;
- bool do_end = 0;
+ 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() */
-
- 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) {
- 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);
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 */
+ S_pat_upgrade_to_utf8(aTHX_ pRExC_state, &exp, &plen,
+ pRExC_state->num_code_blocks);
goto redo_first_pass;
}
}
/* 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;
SvLEN_set(code_blocksv, 1); /*sufficient to make sv_clear free it*/
}
if (reg(pRExC_state, 0, &flags,1) == 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=%#X", flags);
+ 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 */
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);
- Perl_croak(aTHX_ "panic: reg returned NULL to re_op_compile for generation pass, flags=%#X", flags);
+ 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;
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;
}
nextchar(pRExC_state);
return ret;
- } else
- if (*RExC_parse == '?') { /* (?...) */
+ }
+ else 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 '#': /* (?#...) */
+ /* 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");
+ nextchar(pRExC_state);
+ *flagp = TRYAGAIN;
+ return NULL;
case '0' : /* (?0) */
case 'R' : /* (?R) */
if (*RExC_parse != ')')
*flagp = RESTART_UTF8;
return NULL;
}
- FAIL2("panic: regbranch returned NULL, flags=%#X",
- flags);
+ FAIL2("panic: regbranch returned NULL, flags=%#"UVxf"",
+ (UV) flags);
} else
REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
c = *nextchar(pRExC_state);
*flagp = RESTART_UTF8;
return NULL;
}
- FAIL2("panic: regbranch returned NULL, flags=%#X",
- flags);
+ FAIL2("panic: regbranch returned NULL, flags=%#"UVxf"",
+ (UV) flags);
}
REGTAIL(pRExC_state, ret, lastbr);
if (flags&HASWIDTH)
*flagp = RESTART_UTF8;
return NULL;
}
- FAIL2("panic: regbranch returned NULL, flags=%#X", flags);
+ FAIL2("panic: regbranch returned NULL, flags=%#"UVxf"", (UV) flags);
}
if (*RExC_parse == '|') {
if (!SIZE_ONLY && RExC_extralen) {
*flagp = RESTART_UTF8;
return NULL;
}
- FAIL2("panic: regbranch returned NULL, flags=%#X", flags);
+ FAIL2("panic: regbranch returned NULL, flags=%#"UVxf"", (UV) flags);
}
REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
lastbr = br;
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 (");
*flagp = RESTART_UTF8;
return NULL;
}
- FAIL2("panic: regpiece returned NULL, flags=%#X", flags);
+ FAIL2("panic: regpiece returned NULL, flags=%#"UVxf"", (UV) flags);
}
else if (ret == NULL)
ret = latest;
if (flags & (TRYAGAIN|RESTART_UTF8))
*flagp |= flags & (TRYAGAIN|RESTART_UTF8);
else
- FAIL2("panic: regatom returned NULL, flags=%#X", flags);
+ FAIL2("panic: regatom returned NULL, flags=%#"UVxf"", (UV) flags);
return(NULL);
}
*flagp = RESTART_UTF8;
return FALSE;
}
- FAIL2("panic: reg returned NULL to grok_bslash_N, flags=%#X",
- flags);
+ FAIL2("panic: reg returned NULL to grok_bslash_N, flags=%#"UVxf"",
+ (UV) flags);
}
*flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
* 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);
if (ret == NULL) {
if (*flagp & RESTART_UTF8)
return NULL;
- FAIL2("panic: regclass returned NULL to regatom, flags=%#X",
- *flagp);
+ FAIL2("panic: regclass returned NULL to regatom, flags=%#"UVxf"",
+ (UV) *flagp);
}
nextchar(pRExC_state);
Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
}
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) {
*flagp = RESTART_UTF8;
return NULL;
}
- FAIL2("panic: reg returned NULL to regatom, flags=%#X", flags);
+ FAIL2("panic: reg returned NULL to regatom, flags=%#"UVxf"", (UV) flags);
}
*flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
break;
/* regclass() can only return RESTART_UTF8 if multi-char folds
are allowed. */
if (!ret)
- FAIL2("panic: regclass returned NULL to regatom, flags=%#X",
- *flagp);
+ FAIL2("panic: regclass returned NULL to regatom, flags=%#"UVxf"",
+ (UV) *flagp);
RExC_parse--;
goto loopdone;
}
- if (FOLD) {
- if (UTF
- /* See comments for join_exact() as to why we fold
- * this non-UTF at compile time */
- || (node_type == EXACTFU
- && ender == LATIN_SMALL_LETTER_SHARP_S))
- {
+ if (! FOLD) {
+ if (UTF) {
+ const STRLEN unilen = reguni(pRExC_state, ender, s);
+ if (unilen > 0) {
+ s += unilen;
+ len += unilen;
+ }
+ /* The loop increments <len> each time, as all but this
+ * path (and one other) through it add a single byte to
+ * the EXACTish node. But this one has changed len to
+ * be the correct final value, so subtract one to
+ * cancel out the increment that follows */
+ len--;
+ }
+ else {
+ REGC((char)ender, s++);
+ }
+ }
+ else /* FOLD */
+ if (! ( UTF
+ /* See comments for join_exact() as to why we fold this
+ * non-UTF at compile time */
+ || (node_type == EXACTFU
+ && ender == LATIN_SMALL_LETTER_SHARP_S)))
+ {
+ *(s++) = (char) ender;
+ maybe_exact &= ! IS_IN_SOME_FOLD_L1(ender);
+ }
+ else {
- /* Prime the casefolded buffer. Locale rules, which
- * apply only to code points < 256, aren't known until
- * execution, so for them, just output the original
- * character using utf8. If we start to fold non-UTF
- * patterns, be sure to update join_exact() */
- if (LOC && ender < 256) {
- if (UNI_IS_INVARIANT(ender)) {
- *s = (U8) ender;
- foldlen = 1;
- } else {
- *s = UTF8_TWO_BYTE_HI(ender);
- *(s + 1) = UTF8_TWO_BYTE_LO(ender);
- foldlen = 2;
- }
+ /* Prime the casefolded buffer. Locale rules, which apply
+ * only to code points < 256, aren't known until execution,
+ * so for them, just output the original character using
+ * utf8. If we start to fold non-UTF patterns, be sure to
+ * update join_exact() */
+ if (LOC && ender < 256) {
+ if (UNI_IS_INVARIANT(ender)) {
+ *s = (U8) ender;
+ foldlen = 1;
+ } else {
+ *s = UTF8_TWO_BYTE_HI(ender);
+ *(s + 1) = UTF8_TWO_BYTE_LO(ender);
+ foldlen = 2;
}
- else {
- UV folded = _to_uni_fold_flags(
- ender,
- (U8 *) s,
- &foldlen,
- FOLD_FLAGS_FULL
- | ((LOC) ? FOLD_FLAGS_LOCALE
- : (ASCII_FOLD_RESTRICTED)
- ? FOLD_FLAGS_NOMIX_ASCII
- : 0)
- );
-
- /* If this node only contains non-folding code
- * points so far, see if this new one is also
- * non-folding */
- if (maybe_exact) {
- if (folded != ender) {
- maybe_exact = FALSE;
+ }
+ else {
+ UV folded = _to_uni_fold_flags(
+ ender,
+ (U8 *) s,
+ &foldlen,
+ FOLD_FLAGS_FULL
+ | ((LOC) ? FOLD_FLAGS_LOCALE
+ : (ASCII_FOLD_RESTRICTED)
+ ? FOLD_FLAGS_NOMIX_ASCII
+ : 0)
+ );
+
+ /* If this node only contains non-folding code points
+ * so far, see if this new one is also non-folding */
+ if (maybe_exact) {
+ if (folded != ender) {
+ maybe_exact = FALSE;
+ }
+ else {
+ /* Here the fold is the original; we have
+ * to check further to see if anything
+ * folds to it */
+ if (! PL_utf8_foldable) {
+ SV* swash = swash_init("utf8",
+ "_Perl_Any_Folds",
+ &PL_sv_undef, 1, 0);
+ PL_utf8_foldable =
+ _get_swash_invlist(swash);
+ SvREFCNT_dec_NN(swash);
}
- else {
- /* Here the fold is the original; we have
- * to check further to see if anything
- * folds to it */
- if (! PL_utf8_foldable) {
- SV* swash = swash_init("utf8",
- "_Perl_Any_Folds",
- &PL_sv_undef, 1, 0);
- PL_utf8_foldable =
- _get_swash_invlist(swash);
- SvREFCNT_dec_NN(swash);
- }
- if (_invlist_contains_cp(PL_utf8_foldable,
- ender))
- {
- maybe_exact = FALSE;
- }
+ if (_invlist_contains_cp(PL_utf8_foldable,
+ ender))
+ {
+ maybe_exact = FALSE;
}
}
- ender = folded;
}
- s += foldlen;
-
- /* The loop increments <len> each time, as all but this
- * path (and the one just below for UTF) through it add
- * a single byte to the EXACTish node. But this one
- * has changed len to be the correct final value, so
- * subtract one to cancel out the increment that
- * follows */
- len += foldlen - 1;
- }
- else {
- *(s++) = (char) ender;
- maybe_exact &= ! IS_IN_SOME_FOLD_L1(ender);
+ ender = folded;
}
+ s += foldlen;
+
+ /* The loop increments <len> each time, as all but this
+ * path (and one other) through it add a single byte to the
+ * EXACTish node. But this one has changed len to be the
+ * correct final value, so subtract one to cancel out the
+ * increment that follows */
+ len += foldlen - 1;
}
- else if (UTF) {
- const STRLEN unilen = reguni(pRExC_state, ender, s);
- if (unilen > 0) {
- s += unilen;
- len += unilen;
- }
-
- /* See comment just above for - 1 */
- len--;
- }
- else {
- REGC((char)ender, s++);
- }
if (next_is_quantifier) {
FALSE, /* don't allow multi-char folds */
TRUE, /* silence non-portable warnings. */
¤t))
- FAIL2("panic: regclass returned NULL to handle_sets, flags=%#X",
- *flagp);
+ FAIL2("panic: regclass returned NULL to handle_sets, flags=%#"UVxf"",
+ (UV) *flagp);
/* function call leaves parse pointing to the ']', except
* if we faked it */
FALSE, /* don't allow multi-char folds */
FALSE, /* don't silence non-portable warnings. */
¤t))
- FAIL2("panic: regclass returned NULL to handle_sets, flags=%#X",
- *flagp);
+ 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--;
FALSE, /* don't allow multi-char folds */
FALSE, /* don't silence non-portable warnings. */
¤t))
- FAIL2("panic: regclass returned NULL to handle_sets, flags=%#X",
- *flagp);
+ 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) {
they're valid on this machine */
NULL);
if (!node)
- FAIL2("panic: regclass returned NULL to handle_sets, flags=%#X", flagp);
+ FAIL2("panic: regclass returned NULL to handle_sets, flags=%#"UVxf,
+ PTR2UV(flagp));
if (save_fold) {
RExC_flags |= RXf_PMf_FOLD;
}
/* <multi_char_matches> is actually an array of arrays.
* There will be one or two top-level elements: [2],
* and/or [3]. The [2] element is an array, each
- * element thereof is a character which folds to two
- * characters; likewise for [3]. (Unicode guarantees a
- * maximum of 3 characters in any fold.) When we
- * rewrite the character class below, we will do so
- * such that the longest folds are written first, so
- * that it prefers the longest matching strings first.
- * This is done even if it turns out that any
- * quantifier is non-greedy, out of programmer
- * laziness. Tom Christiansen has agreed that this is
- * ok. This makes the test for the ligature 'ffi' come
- * before the test for 'ff' */
+ * element thereof is a character which folds to TWO
+ * characters; [3] is for folds to THREE characters.
+ * (Unicode guarantees a maximum of 3 characters in any
+ * fold.) When we rewrite the character class below,
+ * we will do so such that the longest folds are
+ * written first, so that it prefers the longest
+ * matching strings first. This is done even if it
+ * turns out that any quantifier is non-greedy, out of
+ * programmer laziness. Tom Christiansen has agreed
+ * that this is ok. This makes the test for the
+ * ligature 'ffi' come before the test for 'ff' */
if (av_exists(multi_char_matches, cp_count)) {
this_array_ptr = (AV**) av_fetch(multi_char_matches,
cp_count, FALSE);
* doesn't allow them between above and below 256 */
if ((ASCII_FOLD_RESTRICTED
&& (isASCII(c) != isASCII(j)))
- || (LOC && ((c < 256) != (j < 256))))
- {
+ || (LOC && c < 256)) {
continue;
}
https://perl5.git.perl.org / perl5.git / blobdiff