This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
make PADOFFSET be SSizet_t
[perl5.git] / pp_hot.c
CommitLineData
a0d0e21e
LW
1/* pp_hot.c
2 *
1129b882
NC
3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 by Larry Wall and others
a0d0e21e
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
12 * Then he heard Merry change the note, and up went the Horn-cry of Buckland,
13 * shaking the air.
14 *
4ac71550
TC
15 * Awake! Awake! Fear, Fire, Foes! Awake!
16 * Fire, Foes! Awake!
17 *
18 * [p.1007 of _The Lord of the Rings_, VI/viii: "The Scouring of the Shire"]
a0d0e21e
LW
19 */
20
166f8a29
DM
21/* This file contains 'hot' pp ("push/pop") functions that
22 * execute the opcodes that make up a perl program. A typical pp function
23 * expects to find its arguments on the stack, and usually pushes its
24 * results onto the stack, hence the 'pp' terminology. Each OP structure
25 * contains a pointer to the relevant pp_foo() function.
26 *
27 * By 'hot', we mean common ops whose execution speed is critical.
28 * By gathering them together into a single file, we encourage
29 * CPU cache hits on hot code. Also it could be taken as a warning not to
30 * change any code in this file unless you're sure it won't affect
31 * performance.
32 */
33
a0d0e21e 34#include "EXTERN.h"
864dbfa3 35#define PERL_IN_PP_HOT_C
a0d0e21e
LW
36#include "perl.h"
37
38/* Hot code. */
39
40PP(pp_const)
41{
39644a26 42 dSP;
996c9baa 43 XPUSHs(cSVOP_sv);
a0d0e21e
LW
44 RETURN;
45}
46
47PP(pp_nextstate)
48{
533c011a 49 PL_curcop = (COP*)PL_op;
a0d0e21e 50 TAINT_NOT; /* Each statement is presumed innocent */
4ebe6e95 51 PL_stack_sp = PL_stack_base + CX_CUR()->blk_oldsp;
a0d0e21e 52 FREETMPS;
f410a211 53 PERL_ASYNC_CHECK();
a0d0e21e
LW
54 return NORMAL;
55}
56
57PP(pp_gvsv)
58{
39644a26 59 dSP;
924508f0 60 EXTEND(SP,1);
5d9574c1 61 if (UNLIKELY(PL_op->op_private & OPpLVAL_INTRO))
1d7c1841 62 PUSHs(save_scalar(cGVOP_gv));
a0d0e21e 63 else
c69033f2 64 PUSHs(GvSVn(cGVOP_gv));
a0d0e21e
LW
65 RETURN;
66}
67
b1c05ba5
DM
68
69/* also used for: pp_lineseq() pp_regcmaybe() pp_scalar() pp_scope() */
70
a0d0e21e
LW
71PP(pp_null)
72{
73 return NORMAL;
74}
75
3dd9d4e4
FC
76/* This is sometimes called directly by pp_coreargs, pp_grepstart and
77 amagic_call. */
a0d0e21e
LW
78PP(pp_pushmark)
79{
3280af22 80 PUSHMARK(PL_stack_sp);
a0d0e21e
LW
81 return NORMAL;
82}
83
84PP(pp_stringify)
85{
20b7effb 86 dSP; dTARGET;
4cc783ef
DD
87 SV * const sv = TOPs;
88 SETs(TARG);
89 sv_copypv(TARG, sv);
90 SvSETMAGIC(TARG);
91 /* no PUTBACK, SETs doesn't inc/dec SP */
92 return NORMAL;
a0d0e21e
LW
93}
94
95PP(pp_gv)
96{
20b7effb 97 dSP;
ad64d0ec 98 XPUSHs(MUTABLE_SV(cGVOP_gv));
a0d0e21e
LW
99 RETURN;
100}
101
b1c05ba5
DM
102
103/* also used for: pp_andassign() */
104
a0d0e21e
LW
105PP(pp_and)
106{
f410a211 107 PERL_ASYNC_CHECK();
4cc783ef
DD
108 {
109 /* SP is not used to remove a variable that is saved across the
110 sv_2bool_flags call in SvTRUE_NN, if a RISC/CISC or low/high machine
111 register or load/store vs direct mem ops macro is introduced, this
112 should be a define block between direct PL_stack_sp and dSP operations,
113 presently, using PL_stack_sp is bias towards CISC cpus */
114 SV * const sv = *PL_stack_sp;
115 if (!SvTRUE_NN(sv))
116 return NORMAL;
117 else {
118 if (PL_op->op_type == OP_AND)
119 --PL_stack_sp;
120 return cLOGOP->op_other;
121 }
a0d0e21e
LW
122 }
123}
124
125PP(pp_sassign)
126{
20b7effb 127 dSP;
3e75a3c4
RU
128 /* sassign keeps its args in the optree traditionally backwards.
129 So we pop them differently.
130 */
131 SV *left = POPs; SV *right = TOPs;
748a9306 132
533c011a 133 if (PL_op->op_private & OPpASSIGN_BACKWARDS) {
0bd48802
AL
134 SV * const temp = left;
135 left = right; right = temp;
a0d0e21e 136 }
d48c660d
DM
137 assert(TAINTING_get || !TAINT_get);
138 if (UNLIKELY(TAINT_get) && !SvTAINTED(right))
a0d0e21e 139 TAINT_NOT;
5d9574c1
DM
140 if (UNLIKELY(PL_op->op_private & OPpASSIGN_CV_TO_GV)) {
141 /* *foo =\&bar */
3e75a3c4 142 SV * const cv = SvRV(right);
e26df76a 143 const U32 cv_type = SvTYPE(cv);
3e75a3c4 144 const bool is_gv = isGV_with_GP(left);
6136c704 145 const bool got_coderef = cv_type == SVt_PVCV || cv_type == SVt_PVFM;
e26df76a
NC
146
147 if (!got_coderef) {
148 assert(SvROK(cv));
149 }
150
3e75a3c4
RU
151 /* Can do the optimisation if left (LVALUE) is not a typeglob,
152 right (RVALUE) is a reference to something, and we're in void
e26df76a 153 context. */
13be902c 154 if (!got_coderef && !is_gv && GIMME_V == G_VOID) {
e26df76a 155 /* Is the target symbol table currently empty? */
3e75a3c4 156 GV * const gv = gv_fetchsv_nomg(left, GV_NOINIT, SVt_PVGV);
bb112e5a 157 if (SvTYPE(gv) != SVt_PVGV && !SvOK(gv)) {
e26df76a
NC
158 /* Good. Create a new proxy constant subroutine in the target.
159 The gv becomes a(nother) reference to the constant. */
160 SV *const value = SvRV(cv);
161
ad64d0ec 162 SvUPGRADE(MUTABLE_SV(gv), SVt_IV);
1ccdb730 163 SvPCS_IMPORTED_on(gv);
e26df76a 164 SvRV_set(gv, value);
b37c2d43 165 SvREFCNT_inc_simple_void(value);
3e75a3c4 166 SETs(left);
e26df76a
NC
167 RETURN;
168 }
169 }
170
171 /* Need to fix things up. */
13be902c 172 if (!is_gv) {
e26df76a 173 /* Need to fix GV. */
3e75a3c4 174 left = MUTABLE_SV(gv_fetchsv_nomg(left,GV_ADD, SVt_PVGV));
e26df76a
NC
175 }
176
177 if (!got_coderef) {
178 /* We've been returned a constant rather than a full subroutine,
179 but they expect a subroutine reference to apply. */
53a42478 180 if (SvROK(cv)) {
d343c3ef 181 ENTER_with_name("sassign_coderef");
53a42478
NC
182 SvREFCNT_inc_void(SvRV(cv));
183 /* newCONSTSUB takes a reference count on the passed in SV
184 from us. We set the name to NULL, otherwise we get into
185 all sorts of fun as the reference to our new sub is
186 donated to the GV that we're about to assign to.
187 */
3e75a3c4 188 SvRV_set(right, MUTABLE_SV(newCONSTSUB(GvSTASH(left), NULL,
ad64d0ec 189 SvRV(cv))));
fc2b2dca 190 SvREFCNT_dec_NN(cv);
d343c3ef 191 LEAVE_with_name("sassign_coderef");
53a42478
NC
192 } else {
193 /* What can happen for the corner case *{"BONK"} = \&{"BONK"};
194 is that
195 First: ops for \&{"BONK"}; return us the constant in the
196 symbol table
197 Second: ops for *{"BONK"} cause that symbol table entry
198 (and our reference to it) to be upgraded from RV
199 to typeblob)
200 Thirdly: We get here. cv is actually PVGV now, and its
201 GvCV() is actually the subroutine we're looking for
202
203 So change the reference so that it points to the subroutine
204 of that typeglob, as that's what they were after all along.
205 */
159b6efe 206 GV *const upgraded = MUTABLE_GV(cv);
53a42478
NC
207 CV *const source = GvCV(upgraded);
208
209 assert(source);
210 assert(CvFLAGS(source) & CVf_CONST);
211
0ad694a7 212 SvREFCNT_inc_simple_void_NN(source);
fc2b2dca 213 SvREFCNT_dec_NN(upgraded);
3e75a3c4 214 SvRV_set(right, MUTABLE_SV(source));
53a42478 215 }
e26df76a 216 }
53a42478 217
e26df76a 218 }
8fe85e3f 219 if (
5d9574c1 220 UNLIKELY(SvTEMP(left)) && !SvSMAGICAL(left) && SvREFCNT(left) == 1 &&
3e75a3c4 221 (!isGV_with_GP(left) || SvFAKE(left)) && ckWARN(WARN_MISC)
8fe85e3f
FC
222 )
223 Perl_warner(aTHX_
224 packWARN(WARN_MISC), "Useless assignment to a temporary"
225 );
3e75a3c4
RU
226 SvSetMagicSV(left, right);
227 SETs(left);
a0d0e21e
LW
228 RETURN;
229}
230
231PP(pp_cond_expr)
232{
20b7effb 233 dSP;
f410a211 234 PERL_ASYNC_CHECK();
a0d0e21e 235 if (SvTRUEx(POPs))
1a67a97c 236 RETURNOP(cLOGOP->op_other);
a0d0e21e 237 else
1a67a97c 238 RETURNOP(cLOGOP->op_next);
a0d0e21e
LW
239}
240
241PP(pp_unstack)
242{
f5319de9 243 PERL_CONTEXT *cx;
8f3964af 244 PERL_ASYNC_CHECK();
a0d0e21e 245 TAINT_NOT; /* Each statement is presumed innocent */
4ebe6e95 246 cx = CX_CUR();
f5319de9 247 PL_stack_sp = PL_stack_base + cx->blk_oldsp;
a0d0e21e 248 FREETMPS;
eae48c89 249 if (!(PL_op->op_flags & OPf_SPECIAL)) {
93661e56 250 assert(CxTYPE(cx) == CXt_BLOCK || CxTYPE_is_LOOP(cx));
dfe0f39b 251 CX_LEAVE_SCOPE(cx);
eae48c89 252 }
a0d0e21e
LW
253 return NORMAL;
254}
255
a0d0e21e
LW
256PP(pp_concat)
257{
20b7effb 258 dSP; dATARGET; tryAMAGICbin_MG(concat_amg, AMGf_assign);
748a9306
LW
259 {
260 dPOPTOPssrl;
8d6d96c1
HS
261 bool lbyte;
262 STRLEN rlen;
d4c19fe8 263 const char *rpv = NULL;
a6b599c7 264 bool rbyte = FALSE;
a9c4fd4e 265 bool rcopied = FALSE;
8d6d96c1 266
6f1401dc
DM
267 if (TARG == right && right != left) { /* $r = $l.$r */
268 rpv = SvPV_nomg_const(right, rlen);
c75ab21a 269 rbyte = !DO_UTF8(right);
59cd0e26 270 right = newSVpvn_flags(rpv, rlen, SVs_TEMP);
349d4f2f 271 rpv = SvPV_const(right, rlen); /* no point setting UTF-8 here */
db79b45b 272 rcopied = TRUE;
8d6d96c1 273 }
7889fe52 274
89734059 275 if (TARG != left) { /* not $l .= $r */
a9c4fd4e 276 STRLEN llen;
6f1401dc 277 const char* const lpv = SvPV_nomg_const(left, llen);
90f5826e 278 lbyte = !DO_UTF8(left);
8d6d96c1
HS
279 sv_setpvn(TARG, lpv, llen);
280 if (!lbyte)
281 SvUTF8_on(TARG);
282 else
283 SvUTF8_off(TARG);
284 }
18ea7bf2
SM
285 else { /* $l .= $r and left == TARG */
286 if (!SvOK(left)) {
51f69a24
AC
287 if ((left == right /* $l .= $l */
288 || (PL_op->op_private & OPpTARGET_MY)) /* $l = $l . $r */
289 && ckWARN(WARN_UNINITIALIZED)
290 )
291 report_uninit(left);
76f68e9b 292 sv_setpvs(left, "");
c75ab21a 293 }
18ea7bf2
SM
294 else {
295 SvPV_force_nomg_nolen(left);
296 }
583a5589 297 lbyte = !DO_UTF8(left);
90f5826e 298 if (IN_BYTES)
18ea7bf2 299 SvUTF8_off(left);
8d6d96c1 300 }
a12c0f56 301
c75ab21a 302 if (!rcopied) {
6f1401dc 303 rpv = SvPV_nomg_const(right, rlen);
c75ab21a
RH
304 rbyte = !DO_UTF8(right);
305 }
8d6d96c1
HS
306 if (lbyte != rbyte) {
307 if (lbyte)
308 sv_utf8_upgrade_nomg(TARG);
309 else {
db79b45b 310 if (!rcopied)
59cd0e26 311 right = newSVpvn_flags(rpv, rlen, SVs_TEMP);
8d6d96c1 312 sv_utf8_upgrade_nomg(right);
6f1401dc 313 rpv = SvPV_nomg_const(right, rlen);
69b47968 314 }
a0d0e21e 315 }
8d6d96c1 316 sv_catpvn_nomg(TARG, rpv, rlen);
43ebc500 317
a0d0e21e
LW
318 SETTARG;
319 RETURN;
748a9306 320 }
a0d0e21e
LW
321}
322
d5524600
DM
323/* push the elements of av onto the stack.
324 * XXX Note that padav has similar code but without the mg_get().
325 * I suspect that the mg_get is no longer needed, but while padav
326 * differs, it can't share this function */
327
f9ae8fb6 328STATIC void
d5524600
DM
329S_pushav(pTHX_ AV* const av)
330{
331 dSP;
c70927a6 332 const SSize_t maxarg = AvFILL(av) + 1;
d5524600 333 EXTEND(SP, maxarg);
5d9574c1 334 if (UNLIKELY(SvRMAGICAL(av))) {
c70927a6
FC
335 PADOFFSET i;
336 for (i=0; i < (PADOFFSET)maxarg; i++) {
d5524600
DM
337 SV ** const svp = av_fetch(av, i, FALSE);
338 /* See note in pp_helem, and bug id #27839 */
339 SP[i+1] = svp
340 ? SvGMAGICAL(*svp) ? (mg_get(*svp), *svp) : *svp
341 : &PL_sv_undef;
342 }
343 }
344 else {
c70927a6
FC
345 PADOFFSET i;
346 for (i=0; i < (PADOFFSET)maxarg; i++) {
ce0d59fd 347 SV * const sv = AvARRAY(av)[i];
5d9574c1 348 SP[i+1] = LIKELY(sv) ? sv : &PL_sv_undef;
ce0d59fd 349 }
d5524600
DM
350 }
351 SP += maxarg;
352 PUTBACK;
353}
354
355
a7fd8ef6
DM
356/* ($lex1,@lex2,...) or my ($lex1,@lex2,...) */
357
358PP(pp_padrange)
359{
20b7effb 360 dSP;
a7fd8ef6
DM
361 PADOFFSET base = PL_op->op_targ;
362 int count = (int)(PL_op->op_private) & OPpPADRANGE_COUNTMASK;
363 int i;
d5524600
DM
364 if (PL_op->op_flags & OPf_SPECIAL) {
365 /* fake the RHS of my ($x,$y,..) = @_ */
366 PUSHMARK(SP);
367 S_pushav(aTHX_ GvAVn(PL_defgv));
368 SPAGAIN;
369 }
370
a7fd8ef6
DM
371 /* note, this is only skipped for compile-time-known void cxt */
372 if ((PL_op->op_flags & OPf_WANT) != OPf_WANT_VOID) {
373 EXTEND(SP, count);
374 PUSHMARK(SP);
375 for (i = 0; i <count; i++)
376 *++SP = PAD_SV(base+i);
377 }
378 if (PL_op->op_private & OPpLVAL_INTRO) {
4e09461c
DM
379 SV **svp = &(PAD_SVl(base));
380 const UV payload = (UV)(
381 (base << (OPpPADRANGE_COUNTSHIFT + SAVE_TIGHT_SHIFT))
382 | (count << SAVE_TIGHT_SHIFT)
383 | SAVEt_CLEARPADRANGE);
6d59e610 384 STATIC_ASSERT_STMT(OPpPADRANGE_COUNTMASK + 1 == (1 << OPpPADRANGE_COUNTSHIFT));
d081a355
DM
385 assert((payload >> (OPpPADRANGE_COUNTSHIFT+SAVE_TIGHT_SHIFT))
386 == (Size_t)base);
a3444cc5
DM
387 {
388 dSS_ADD;
389 SS_ADD_UV(payload);
390 SS_ADD_END(1);
391 }
4e09461c 392
a7fd8ef6 393 for (i = 0; i <count; i++)
4e09461c 394 SvPADSTALE_off(*svp++); /* mark lexical as active */
a7fd8ef6
DM
395 }
396 RETURN;
397}
398
399
a0d0e21e
LW
400PP(pp_padsv)
401{
20b7effb 402 dSP;
6c28b496
DD
403 EXTEND(SP, 1);
404 {
405 OP * const op = PL_op;
406 /* access PL_curpad once */
407 SV ** const padentry = &(PAD_SVl(op->op_targ));
408 {
409 dTARG;
410 TARG = *padentry;
411 PUSHs(TARG);
412 PUTBACK; /* no pop/push after this, TOPs ok */
8ec5e241 413 }
6c28b496
DD
414 if (op->op_flags & OPf_MOD) {
415 if (op->op_private & OPpLVAL_INTRO)
416 if (!(op->op_private & OPpPAD_STATE))
417 save_clearsv(padentry);
418 if (op->op_private & OPpDEREF) {
8f90a16d
FC
419 /* TOPs is equivalent to TARG here. Using TOPs (SP) rather
420 than TARG reduces the scope of TARG, so it does not
421 span the call to save_clearsv, resulting in smaller
422 machine code. */
6c28b496
DD
423 TOPs = vivify_ref(TOPs, op->op_private & OPpDEREF);
424 }
425 }
426 return op->op_next;
4633a7c4 427 }
a0d0e21e
LW
428}
429
430PP(pp_readline)
431{
30901a8a
FC
432 dSP;
433 if (TOPs) {
434 SvGETMAGIC(TOPs);
fc99edcf 435 tryAMAGICunTARGETlist(iter_amg, 0);
30901a8a
FC
436 PL_last_in_gv = MUTABLE_GV(*PL_stack_sp--);
437 }
438 else PL_last_in_gv = PL_argvgv, PL_stack_sp--;
6e592b3a
BM
439 if (!isGV_with_GP(PL_last_in_gv)) {
440 if (SvROK(PL_last_in_gv) && isGV_with_GP(SvRV(PL_last_in_gv)))
159b6efe 441 PL_last_in_gv = MUTABLE_GV(SvRV(PL_last_in_gv));
8efb3254 442 else {
f5284f61 443 dSP;
ad64d0ec 444 XPUSHs(MUTABLE_SV(PL_last_in_gv));
f5284f61 445 PUTBACK;
897d3989 446 Perl_pp_rv2gv(aTHX);
159b6efe 447 PL_last_in_gv = MUTABLE_GV(*PL_stack_sp--);
84ee769f
FC
448 if (PL_last_in_gv == (GV *)&PL_sv_undef)
449 PL_last_in_gv = NULL;
450 else
451 assert(isGV_with_GP(PL_last_in_gv));
f5284f61
IZ
452 }
453 }
a0d0e21e
LW
454 return do_readline();
455}
456
457PP(pp_eq)
458{
20b7effb 459 dSP;
33efebe6
DM
460 SV *left, *right;
461
a42d0242 462 tryAMAGICbin_MG(eq_amg, AMGf_set|AMGf_numeric);
33efebe6
DM
463 right = POPs;
464 left = TOPs;
465 SETs(boolSV(
466 (SvIOK_notUV(left) && SvIOK_notUV(right))
467 ? (SvIVX(left) == SvIVX(right))
468 : ( do_ncmp(left, right) == 0)
469 ));
470 RETURN;
a0d0e21e
LW
471}
472
b1c05ba5 473
4c2c3128 474/* also used for: pp_i_preinc() */
b1c05ba5 475
a0d0e21e
LW
476PP(pp_preinc)
477{
4c2c3128
DM
478 SV *sv = *PL_stack_sp;
479
480 if (LIKELY(((sv->sv_flags &
481 (SVf_THINKFIRST|SVs_GMG|SVf_IVisUV|
482 SVf_IOK|SVf_NOK|SVf_POK|SVp_NOK|SVp_POK|SVf_ROK))
483 == SVf_IOK))
484 && SvIVX(sv) != IV_MAX)
485 {
486 SvIV_set(sv, SvIVX(sv) + 1);
487 }
488 else /* Do all the PERL_PRESERVE_IVUV and hard cases in sv_inc */
489 sv_inc(sv);
490 SvSETMAGIC(sv);
491 return NORMAL;
492}
493
494
495/* also used for: pp_i_predec() */
496
497PP(pp_predec)
498{
499 SV *sv = *PL_stack_sp;
500
501 if (LIKELY(((sv->sv_flags &
502 (SVf_THINKFIRST|SVs_GMG|SVf_IVisUV|
503 SVf_IOK|SVf_NOK|SVf_POK|SVp_NOK|SVp_POK|SVf_ROK))
504 == SVf_IOK))
505 && SvIVX(sv) != IV_MIN)
55497cff 506 {
4c2c3128 507 SvIV_set(sv, SvIVX(sv) - 1);
748a9306 508 }
4c2c3128
DM
509 else /* Do all the PERL_PRESERVE_IVUV and hard cases in sv_dec */
510 sv_dec(sv);
511 SvSETMAGIC(sv);
a0d0e21e
LW
512 return NORMAL;
513}
514
b1c05ba5
DM
515
516/* also used for: pp_orassign() */
517
a0d0e21e
LW
518PP(pp_or)
519{
20b7effb 520 dSP;
f410a211 521 PERL_ASYNC_CHECK();
a0d0e21e
LW
522 if (SvTRUE(TOPs))
523 RETURN;
524 else {
c960fc3b
SP
525 if (PL_op->op_type == OP_OR)
526 --SP;
a0d0e21e
LW
527 RETURNOP(cLOGOP->op_other);
528 }
529}
530
b1c05ba5
DM
531
532/* also used for: pp_dor() pp_dorassign() */
533
25a55bd7 534PP(pp_defined)
c963b151 535{
20b7effb 536 dSP;
eb578fdb 537 SV* sv;
6136c704 538 bool defined;
25a55bd7 539 const int op_type = PL_op->op_type;
ea5195b7 540 const bool is_dor = (op_type == OP_DOR || op_type == OP_DORASSIGN);
c963b151 541
6136c704 542 if (is_dor) {
f410a211 543 PERL_ASYNC_CHECK();
25a55bd7 544 sv = TOPs;
5d9574c1 545 if (UNLIKELY(!sv || !SvANY(sv))) {
2bd49cfc
NC
546 if (op_type == OP_DOR)
547 --SP;
25a55bd7
SP
548 RETURNOP(cLOGOP->op_other);
549 }
b7c44293
RGS
550 }
551 else {
552 /* OP_DEFINED */
25a55bd7 553 sv = POPs;
5d9574c1 554 if (UNLIKELY(!sv || !SvANY(sv)))
25a55bd7 555 RETPUSHNO;
b7c44293 556 }
25a55bd7 557
6136c704 558 defined = FALSE;
c963b151
BD
559 switch (SvTYPE(sv)) {
560 case SVt_PVAV:
561 if (AvMAX(sv) >= 0 || SvGMAGICAL(sv) || (SvRMAGICAL(sv) && mg_find(sv, PERL_MAGIC_tied)))
25a55bd7 562 defined = TRUE;
c963b151
BD
563 break;
564 case SVt_PVHV:
565 if (HvARRAY(sv) || SvGMAGICAL(sv) || (SvRMAGICAL(sv) && mg_find(sv, PERL_MAGIC_tied)))
25a55bd7 566 defined = TRUE;
c963b151
BD
567 break;
568 case SVt_PVCV:
569 if (CvROOT(sv) || CvXSUB(sv))
25a55bd7 570 defined = TRUE;
c963b151
BD
571 break;
572 default:
5b295bef 573 SvGETMAGIC(sv);
c963b151 574 if (SvOK(sv))
25a55bd7 575 defined = TRUE;
6136c704 576 break;
c963b151 577 }
6136c704
AL
578
579 if (is_dor) {
c960fc3b
SP
580 if(defined)
581 RETURN;
582 if(op_type == OP_DOR)
583 --SP;
25a55bd7 584 RETURNOP(cLOGOP->op_other);
25a55bd7 585 }
d9aa96a4
SP
586 /* assuming OP_DEFINED */
587 if(defined)
588 RETPUSHYES;
589 RETPUSHNO;
c963b151
BD
590}
591
230ee21f
DM
592
593
a0d0e21e
LW
594PP(pp_add)
595{
20b7effb 596 dSP; dATARGET; bool useleft; SV *svl, *svr;
230ee21f 597
6f1401dc
DM
598 tryAMAGICbin_MG(add_amg, AMGf_assign|AMGf_numeric);
599 svr = TOPs;
600 svl = TOPm1s;
601
28e5dec8 602#ifdef PERL_PRESERVE_IVUV
230ee21f
DM
603
604 /* special-case some simple common cases */
605 if (!((svl->sv_flags|svr->sv_flags) & (SVf_IVisUV|SVs_GMG))) {
606 IV il, ir;
607 U32 flags = (svl->sv_flags & svr->sv_flags);
608 if (flags & SVf_IOK) {
609 /* both args are simple IVs */
610 UV topl, topr;
611 il = SvIVX(svl);
612 ir = SvIVX(svr);
613 do_iv:
614 topl = ((UV)il) >> (UVSIZE * 8 - 2);
615 topr = ((UV)ir) >> (UVSIZE * 8 - 2);
616
617 /* if both are in a range that can't under/overflow, do a
618 * simple integer add: if the top of both numbers
619 * are 00 or 11, then it's safe */
620 if (!( ((topl+1) | (topr+1)) & 2)) {
621 SP--;
622 TARGi(il + ir, 0); /* args not GMG, so can't be tainted */
623 SETs(TARG);
624 RETURN;
625 }
626 goto generic;
627 }
628 else if (flags & SVf_NOK) {
629 /* both args are NVs */
630 NV nl = SvNVX(svl);
631 NV nr = SvNVX(svr);
632
3336af0b
DD
633 if (
634#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
635 !Perl_isnan(nl) && nl == (NV)(il = (IV)nl)
636 && !Perl_isnan(nr) && nr == (NV)(ir = (IV)nr)
637#else
638 nl == (NV)(il = (IV)nl) && nr == (NV)(ir = (IV)nr)
639#endif
640 )
230ee21f
DM
641 /* nothing was lost by converting to IVs */
642 goto do_iv;
643 SP--;
644 TARGn(nl + nr, 0); /* args not GMG, so can't be tainted */
645 SETs(TARG);
646 RETURN;
647 }
648 }
649
650 generic:
651
652 useleft = USE_LEFT(svl);
28e5dec8
JH
653 /* We must see if we can perform the addition with integers if possible,
654 as the integer code detects overflow while the NV code doesn't.
655 If either argument hasn't had a numeric conversion yet attempt to get
656 the IV. It's important to do this now, rather than just assuming that
657 it's not IOK as a PV of "9223372036854775806" may not take well to NV
658 addition, and an SV which is NOK, NV=6.0 ought to be coerced to
659 integer in case the second argument is IV=9223372036854775806
660 We can (now) rely on sv_2iv to do the right thing, only setting the
661 public IOK flag if the value in the NV (or PV) slot is truly integer.
662
663 A side effect is that this also aggressively prefers integer maths over
7dca457a
NC
664 fp maths for integer values.
665
a00b5bd3 666 How to detect overflow?
7dca457a
NC
667
668 C 99 section 6.2.6.1 says
669
670 The range of nonnegative values of a signed integer type is a subrange
671 of the corresponding unsigned integer type, and the representation of
672 the same value in each type is the same. A computation involving
673 unsigned operands can never overflow, because a result that cannot be
674 represented by the resulting unsigned integer type is reduced modulo
675 the number that is one greater than the largest value that can be
676 represented by the resulting type.
677
678 (the 9th paragraph)
679
680 which I read as "unsigned ints wrap."
681
682 signed integer overflow seems to be classed as "exception condition"
683
684 If an exceptional condition occurs during the evaluation of an
685 expression (that is, if the result is not mathematically defined or not
686 in the range of representable values for its type), the behavior is
687 undefined.
688
689 (6.5, the 5th paragraph)
690
691 I had assumed that on 2s complement machines signed arithmetic would
692 wrap, hence coded pp_add and pp_subtract on the assumption that
693 everything perl builds on would be happy. After much wailing and
694 gnashing of teeth it would seem that irix64 knows its ANSI spec well,
695 knows that it doesn't need to, and doesn't. Bah. Anyway, the all-
696 unsigned code below is actually shorter than the old code. :-)
697 */
698
01f91bf2 699 if (SvIV_please_nomg(svr)) {
28e5dec8
JH
700 /* Unless the left argument is integer in range we are going to have to
701 use NV maths. Hence only attempt to coerce the right argument if
702 we know the left is integer. */
eb578fdb 703 UV auv = 0;
9c5ffd7c 704 bool auvok = FALSE;
7dca457a
NC
705 bool a_valid = 0;
706
28e5dec8 707 if (!useleft) {
7dca457a
NC
708 auv = 0;
709 a_valid = auvok = 1;
710 /* left operand is undef, treat as zero. + 0 is identity,
711 Could SETi or SETu right now, but space optimise by not adding
712 lots of code to speed up what is probably a rarish case. */
713 } else {
714 /* Left operand is defined, so is it IV? */
01f91bf2 715 if (SvIV_please_nomg(svl)) {
800401ee
JH
716 if ((auvok = SvUOK(svl)))
717 auv = SvUVX(svl);
7dca457a 718 else {
eb578fdb 719 const IV aiv = SvIVX(svl);
7dca457a
NC
720 if (aiv >= 0) {
721 auv = aiv;
722 auvok = 1; /* Now acting as a sign flag. */
53e2bfb7
DM
723 } else {
724 auv = (aiv == IV_MIN) ? (UV)aiv : (UV)(-aiv);
7dca457a
NC
725 }
726 }
727 a_valid = 1;
28e5dec8
JH
728 }
729 }
7dca457a
NC
730 if (a_valid) {
731 bool result_good = 0;
732 UV result;
eb578fdb 733 UV buv;
800401ee 734 bool buvok = SvUOK(svr);
a00b5bd3 735
7dca457a 736 if (buvok)
800401ee 737 buv = SvUVX(svr);
7dca457a 738 else {
eb578fdb 739 const IV biv = SvIVX(svr);
7dca457a
NC
740 if (biv >= 0) {
741 buv = biv;
742 buvok = 1;
743 } else
53e2bfb7 744 buv = (biv == IV_MIN) ? (UV)biv : (UV)(-biv);
7dca457a
NC
745 }
746 /* ?uvok if value is >= 0. basically, flagged as UV if it's +ve,
602f51c4 747 else "IV" now, independent of how it came in.
7dca457a
NC
748 if a, b represents positive, A, B negative, a maps to -A etc
749 a + b => (a + b)
750 A + b => -(a - b)
751 a + B => (a - b)
752 A + B => -(a + b)
753 all UV maths. negate result if A negative.
754 add if signs same, subtract if signs differ. */
755
756 if (auvok ^ buvok) {
757 /* Signs differ. */
758 if (auv >= buv) {
759 result = auv - buv;
760 /* Must get smaller */
761 if (result <= auv)
762 result_good = 1;
763 } else {
764 result = buv - auv;
765 if (result <= buv) {
766 /* result really should be -(auv-buv). as its negation
767 of true value, need to swap our result flag */
768 auvok = !auvok;
769 result_good = 1;
28e5dec8
JH
770 }
771 }
7dca457a
NC
772 } else {
773 /* Signs same */
774 result = auv + buv;
775 if (result >= auv)
776 result_good = 1;
777 }
778 if (result_good) {
779 SP--;
780 if (auvok)
28e5dec8 781 SETu( result );
7dca457a
NC
782 else {
783 /* Negate result */
784 if (result <= (UV)IV_MIN)
53e2bfb7
DM
785 SETi(result == (UV)IV_MIN
786 ? IV_MIN : -(IV)result);
7dca457a
NC
787 else {
788 /* result valid, but out of range for IV. */
789 SETn( -(NV)result );
28e5dec8
JH
790 }
791 }
7dca457a
NC
792 RETURN;
793 } /* Overflow, drop through to NVs. */
28e5dec8
JH
794 }
795 }
230ee21f
DM
796
797#else
798 useleft = USE_LEFT(svl);
28e5dec8 799#endif
230ee21f 800
a0d0e21e 801 {
6f1401dc 802 NV value = SvNV_nomg(svr);
4efa5a16 803 (void)POPs;
28e5dec8
JH
804 if (!useleft) {
805 /* left operand is undef, treat as zero. + 0.0 is identity. */
806 SETn(value);
807 RETURN;
808 }
6f1401dc 809 SETn( value + SvNV_nomg(svl) );
28e5dec8 810 RETURN;
a0d0e21e
LW
811 }
812}
813
b1c05ba5
DM
814
815/* also used for: pp_aelemfast_lex() */
816
a0d0e21e
LW
817PP(pp_aelemfast)
818{
20b7effb 819 dSP;
93bad3fd 820 AV * const av = PL_op->op_type == OP_AELEMFAST_LEX
8f878375 821 ? MUTABLE_AV(PAD_SV(PL_op->op_targ)) : GvAVn(cGVOP_gv);
a3b680e6 822 const U32 lval = PL_op->op_flags & OPf_MOD;
7e169e84
DM
823 const I8 key = (I8)PL_op->op_private;
824 SV** svp;
825 SV *sv;
826
827 assert(SvTYPE(av) == SVt_PVAV);
828
f4484b87
DM
829 EXTEND(SP, 1);
830
7e169e84
DM
831 /* inlined av_fetch() for simple cases ... */
832 if (!SvRMAGICAL(av) && key >= 0 && key <= AvFILLp(av)) {
833 sv = AvARRAY(av)[key];
9fb994be 834 if (sv) {
7e169e84
DM
835 PUSHs(sv);
836 RETURN;
837 }
838 }
839
840 /* ... else do it the hard way */
841 svp = av_fetch(av, key, lval);
842 sv = (svp ? *svp : &PL_sv_undef);
b024352e
DM
843
844 if (UNLIKELY(!svp && lval))
7e169e84 845 DIE(aTHX_ PL_no_aelem, (int)key);
b024352e 846
39cf747a 847 if (!lval && SvRMAGICAL(av) && SvGMAGICAL(sv)) /* see note in pp_helem() */
fd69380d 848 mg_get(sv);
be6c24e0 849 PUSHs(sv);
a0d0e21e
LW
850 RETURN;
851}
852
853PP(pp_join)
854{
20b7effb 855 dSP; dMARK; dTARGET;
a0d0e21e
LW
856 MARK++;
857 do_join(TARG, *MARK, MARK, SP);
858 SP = MARK;
859 SETs(TARG);
860 RETURN;
861}
862
863PP(pp_pushre)
864{
20b7effb 865 dSP;
44a8e56a
PP
866#ifdef DEBUGGING
867 /*
868 * We ass_u_me that LvTARGOFF() comes first, and that two STRLENs
869 * will be enough to hold an OP*.
870 */
c4420975 871 SV* const sv = sv_newmortal();
44a8e56a
PP
872 sv_upgrade(sv, SVt_PVLV);
873 LvTYPE(sv) = '/';
533c011a 874 Copy(&PL_op, &LvTARGOFF(sv), 1, OP*);
44a8e56a
PP
875 XPUSHs(sv);
876#else
ad64d0ec 877 XPUSHs(MUTABLE_SV(PL_op));
44a8e56a 878#endif
a0d0e21e
LW
879 RETURN;
880}
881
882/* Oversized hot code. */
883
b1c05ba5
DM
884/* also used for: pp_say() */
885
a0d0e21e
LW
886PP(pp_print)
887{
20b7effb 888 dSP; dMARK; dORIGMARK;
eb578fdb 889 PerlIO *fp;
236988e4 890 MAGIC *mg;
159b6efe
NC
891 GV * const gv
892 = (PL_op->op_flags & OPf_STACKED) ? MUTABLE_GV(*++MARK) : PL_defoutgv;
9c9f25b8 893 IO *io = GvIO(gv);
5b468f54 894
9c9f25b8 895 if (io
ad64d0ec 896 && (mg = SvTIED_mg((const SV *)io, PERL_MAGIC_tiedscalar)))
5b468f54 897 {
01bb7c6d 898 had_magic:
68dc0745 899 if (MARK == ORIGMARK) {
1c846c1f 900 /* If using default handle then we need to make space to
a60c0954
NIS
901 * pass object as 1st arg, so move other args up ...
902 */
4352c267 903 MEXTEND(SP, 1);
68dc0745
PP
904 ++MARK;
905 Move(MARK, MARK + 1, (SP - MARK) + 1, SV*);
906 ++SP;
907 }
3e0cb5de 908 return Perl_tied_method(aTHX_ SV_CONST(PRINT), mark - 1, MUTABLE_SV(io),
94bc412f
NC
909 mg,
910 (G_SCALAR | TIED_METHOD_ARGUMENTS_ON_STACK
911 | (PL_op->op_type == OP_SAY
912 ? TIED_METHOD_SAY : 0)), sp - mark);
236988e4 913 }
9c9f25b8 914 if (!io) {
68b590d9 915 if ( gv && GvEGVx(gv) && (io = GvIO(GvEGV(gv)))
ad64d0ec 916 && (mg = SvTIED_mg((const SV *)io, PERL_MAGIC_tiedscalar)))
01bb7c6d 917 goto had_magic;
51087808 918 report_evil_fh(gv);
93189314 919 SETERRNO(EBADF,RMS_IFI);
a0d0e21e
LW
920 goto just_say_no;
921 }
922 else if (!(fp = IoOFP(io))) {
7716c5c5
NC
923 if (IoIFP(io))
924 report_wrongway_fh(gv, '<');
51087808 925 else
7716c5c5 926 report_evil_fh(gv);
93189314 927 SETERRNO(EBADF,IoIFP(io)?RMS_FAC:RMS_IFI);
a0d0e21e
LW
928 goto just_say_no;
929 }
930 else {
e23d9e2f 931 SV * const ofs = GvSV(PL_ofsgv); /* $, */
a0d0e21e 932 MARK++;
e23d9e2f 933 if (ofs && (SvGMAGICAL(ofs) || SvOK(ofs))) {
a0d0e21e
LW
934 while (MARK <= SP) {
935 if (!do_print(*MARK, fp))
936 break;
937 MARK++;
938 if (MARK <= SP) {
e23d9e2f
CS
939 /* don't use 'ofs' here - it may be invalidated by magic callbacks */
940 if (!do_print(GvSV(PL_ofsgv), fp)) {
a0d0e21e
LW
941 MARK--;
942 break;
943 }
944 }
945 }
946 }
947 else {
948 while (MARK <= SP) {
949 if (!do_print(*MARK, fp))
950 break;
951 MARK++;
952 }
953 }
954 if (MARK <= SP)
955 goto just_say_no;
956 else {
cfc4a7da
GA
957 if (PL_op->op_type == OP_SAY) {
958 if (PerlIO_write(fp, "\n", 1) == 0 || PerlIO_error(fp))
959 goto just_say_no;
960 }
961 else if (PL_ors_sv && SvOK(PL_ors_sv))
7889fe52 962 if (!do_print(PL_ors_sv, fp)) /* $\ */
a0d0e21e
LW
963 goto just_say_no;
964
965 if (IoFLAGS(io) & IOf_FLUSH)
760ac839 966 if (PerlIO_flush(fp) == EOF)
a0d0e21e
LW
967 goto just_say_no;
968 }
969 }
970 SP = ORIGMARK;
e52fd6f4 971 XPUSHs(&PL_sv_yes);
a0d0e21e
LW
972 RETURN;
973
974 just_say_no:
975 SP = ORIGMARK;
e52fd6f4 976 XPUSHs(&PL_sv_undef);
a0d0e21e
LW
977 RETURN;
978}
979
b1c05ba5
DM
980
981/* also used for: pp_rv2hv() */
bdaf10a5 982/* also called directly by pp_lvavref */
b1c05ba5 983
a0d0e21e
LW
984PP(pp_rv2av)
985{
20b7effb 986 dSP; dTOPss;
1c23e2bd 987 const U8 gimme = GIMME_V;
13c59d41
MH
988 static const char an_array[] = "an ARRAY";
989 static const char a_hash[] = "a HASH";
bdaf10a5
FC
990 const bool is_pp_rv2av = PL_op->op_type == OP_RV2AV
991 || PL_op->op_type == OP_LVAVREF;
d83b45b8 992 const svtype type = is_pp_rv2av ? SVt_PVAV : SVt_PVHV;
a0d0e21e 993
9026059d 994 SvGETMAGIC(sv);
a0d0e21e 995 if (SvROK(sv)) {
5d9574c1 996 if (UNLIKELY(SvAMAGIC(sv))) {
93d7320b 997 sv = amagic_deref_call(sv, is_pp_rv2av ? to_av_amg : to_hv_amg);
93d7320b 998 }
17ab7946 999 sv = SvRV(sv);
5d9574c1 1000 if (UNLIKELY(SvTYPE(sv) != type))
dcbac5bb 1001 /* diag_listed_as: Not an ARRAY reference */
13c59d41 1002 DIE(aTHX_ "Not %s reference", is_pp_rv2av ? an_array : a_hash);
5d9574c1
DM
1003 else if (UNLIKELY(PL_op->op_flags & OPf_MOD
1004 && PL_op->op_private & OPpLVAL_INTRO))
3da99855 1005 Perl_croak(aTHX_ "%s", PL_no_localize_ref);
a0d0e21e 1006 }
5d9574c1 1007 else if (UNLIKELY(SvTYPE(sv) != type)) {
67955e0c 1008 GV *gv;
1c846c1f 1009
6e592b3a 1010 if (!isGV_with_GP(sv)) {
13c59d41 1011 gv = Perl_softref2xv(aTHX_ sv, is_pp_rv2av ? an_array : a_hash,
dc3c76f8
NC
1012 type, &sp);
1013 if (!gv)
1014 RETURN;
35cd451c
GS
1015 }
1016 else {
159b6efe 1017 gv = MUTABLE_GV(sv);
a0d0e21e 1018 }
ad64d0ec 1019 sv = is_pp_rv2av ? MUTABLE_SV(GvAVn(gv)) : MUTABLE_SV(GvHVn(gv));
533c011a 1020 if (PL_op->op_private & OPpLVAL_INTRO)
ad64d0ec 1021 sv = is_pp_rv2av ? MUTABLE_SV(save_ary(gv)) : MUTABLE_SV(save_hash(gv));
9f527363
FC
1022 }
1023 if (PL_op->op_flags & OPf_REF) {
17ab7946 1024 SETs(sv);
a0d0e21e 1025 RETURN;
9f527363 1026 }
5d9574c1 1027 else if (UNLIKELY(PL_op->op_private & OPpMAYBE_LVSUB)) {
40c94d11
FC
1028 const I32 flags = is_lvalue_sub();
1029 if (flags && !(flags & OPpENTERSUB_INARGS)) {
cde874ca 1030 if (gimme != G_ARRAY)
042560a6 1031 goto croak_cant_return;
17ab7946 1032 SETs(sv);
78f9721b 1033 RETURN;
40c94d11 1034 }
a0d0e21e
LW
1035 }
1036
17ab7946 1037 if (is_pp_rv2av) {
502c6561 1038 AV *const av = MUTABLE_AV(sv);
636fe681 1039 /* The guts of pp_rv2av */
96913b52 1040 if (gimme == G_ARRAY) {
d5524600
DM
1041 SP--;
1042 PUTBACK;
1043 S_pushav(aTHX_ av);
1044 SPAGAIN;
1c846c1f 1045 }
96913b52
VP
1046 else if (gimme == G_SCALAR) {
1047 dTARGET;
c70927a6 1048 const SSize_t maxarg = AvFILL(av) + 1;
96913b52 1049 SETi(maxarg);
93965878 1050 }
17ab7946
NC
1051 } else {
1052 /* The guts of pp_rv2hv */
96913b52
VP
1053 if (gimme == G_ARRAY) { /* array wanted */
1054 *PL_stack_sp = sv;
981b7185 1055 return Perl_do_kv(aTHX);
96913b52 1056 }
c8fe3bdf 1057 else if ((PL_op->op_private & OPpTRUEBOOL
adc42c31 1058 || ( PL_op->op_private & OPpMAYBE_TRUEBOOL
c8fe3bdf
FC
1059 && block_gimme() == G_VOID ))
1060 && (!SvRMAGICAL(sv) || !mg_find(sv, PERL_MAGIC_tied)))
1061 SETs(HvUSEDKEYS(sv) ? &PL_sv_yes : sv_2mortal(newSViv(0)));
96913b52 1062 else if (gimme == G_SCALAR) {
1a8bdda9 1063 dTARG;
96913b52 1064 TARG = Perl_hv_scalar(aTHX_ MUTABLE_HV(sv));
96913b52
VP
1065 SETTARG;
1066 }
17ab7946 1067 }
be85d344 1068 RETURN;
042560a6
NC
1069
1070 croak_cant_return:
1071 Perl_croak(aTHX_ "Can't return %s to lvalue scalar context",
1072 is_pp_rv2av ? "array" : "hash");
77e217c6 1073 RETURN;
a0d0e21e
LW
1074}
1075
10c8fecd 1076STATIC void
fb8f4cf8 1077S_do_oddball(pTHX_ SV **oddkey, SV **firstkey)
10c8fecd 1078{
7918f24d
NC
1079 PERL_ARGS_ASSERT_DO_ODDBALL;
1080
fb8f4cf8 1081 if (*oddkey) {
6d822dc4 1082 if (ckWARN(WARN_MISC)) {
a3b680e6 1083 const char *err;
fb8f4cf8
RZ
1084 if (oddkey == firstkey &&
1085 SvROK(*oddkey) &&
1086 (SvTYPE(SvRV(*oddkey)) == SVt_PVAV ||
1087 SvTYPE(SvRV(*oddkey)) == SVt_PVHV))
10c8fecd 1088 {
a3b680e6 1089 err = "Reference found where even-sized list expected";
10c8fecd
GS
1090 }
1091 else
a3b680e6 1092 err = "Odd number of elements in hash assignment";
f1f66076 1093 Perl_warner(aTHX_ packWARN(WARN_MISC), "%s", err);
10c8fecd 1094 }
6d822dc4 1095
10c8fecd
GS
1096 }
1097}
1098
a5f48505
DM
1099
1100/* Do a mark and sweep with the SVf_BREAK flag to detect elements which
1101 * are common to both the LHS and RHS of an aassign, and replace them
1102 * with copies. All these copies are made before the actual list assign is
1103 * done.
1104 *
1105 * For example in ($a,$b) = ($b,$a), assigning the value of the first RHS
1106 * element ($b) to the first LH element ($a), modifies $a; when the
1107 * second assignment is done, the second RH element now has the wrong
1108 * value. So we initially replace the RHS with ($b, mortalcopy($a)).
1109 * Note that we don't need to make a mortal copy of $b.
1110 *
1111 * The algorithm below works by, for every RHS element, mark the
1112 * corresponding LHS target element with SVf_BREAK. Then if the RHS
1113 * element is found with SVf_BREAK set, it means it would have been
1114 * modified, so make a copy.
1115 * Note that by scanning both LHS and RHS in lockstep, we avoid
1116 * unnecessary copies (like $b above) compared with a naive
1117 * "mark all LHS; copy all marked RHS; unmark all LHS".
1118 *
1119 * If the LHS element is a 'my' declaration' and has a refcount of 1, then
1120 * it can't be common and can be skipped.
ebc643ce
DM
1121 *
1122 * On DEBUGGING builds it takes an extra boolean, fake. If true, it means
1123 * that we thought we didn't need to call S_aassign_copy_common(), but we
1124 * have anyway for sanity checking. If we find we need to copy, then panic.
a5f48505
DM
1125 */
1126
1127PERL_STATIC_INLINE void
1128S_aassign_copy_common(pTHX_ SV **firstlelem, SV **lastlelem,
ebc643ce
DM
1129 SV **firstrelem, SV **lastrelem
1130#ifdef DEBUGGING
1131 , bool fake
1132#endif
1133)
a5f48505
DM
1134{
1135 dVAR;
1136 SV **relem;
1137 SV **lelem;
1138 SSize_t lcount = lastlelem - firstlelem + 1;
1139 bool marked = FALSE; /* have we marked any LHS with SVf_BREAK ? */
1140 bool const do_rc1 = cBOOL(PL_op->op_private & OPpASSIGN_COMMON_RC1);
beb08a1e 1141 bool copy_all = FALSE;
a5f48505
DM
1142
1143 assert(!PL_in_clean_all); /* SVf_BREAK not already in use */
1144 assert(firstlelem < lastlelem); /* at least 2 LH elements */
1145 assert(firstrelem < lastrelem); /* at least 2 RH elements */
1146
ebc643ce
DM
1147
1148 lelem = firstlelem;
a5f48505
DM
1149 /* we never have to copy the first RH element; it can't be corrupted
1150 * by assigning something to the corresponding first LH element.
1151 * So this scan does in a loop: mark LHS[N]; test RHS[N+1]
1152 */
ebc643ce 1153 relem = firstrelem + 1;
a5f48505
DM
1154
1155 for (; relem <= lastrelem; relem++) {
1156 SV *svr;
1157
1158 /* mark next LH element */
1159
1160 if (--lcount >= 0) {
1161 SV *svl = *lelem++;
1162
1163 if (UNLIKELY(!svl)) {/* skip AV alias marker */
1164 assert (lelem <= lastlelem);
1165 svl = *lelem++;
1166 lcount--;
1167 }
1168
1169 assert(svl);
beb08a1e
TC
1170 if (SvSMAGICAL(svl)) {
1171 copy_all = TRUE;
1172 }
a5f48505
DM
1173 if (SvTYPE(svl) == SVt_PVAV || SvTYPE(svl) == SVt_PVHV) {
1174 if (!marked)
1175 return;
1176 /* this LH element will consume all further args;
1177 * no need to mark any further LH elements (if any).
1178 * But we still need to scan any remaining RHS elements;
1179 * set lcount negative to distinguish from lcount == 0,
1180 * so the loop condition continues being true
1181 */
1182 lcount = -1;
1183 lelem--; /* no need to unmark this element */
1184 }
1185 else if (!(do_rc1 && SvREFCNT(svl) == 1) && svl != &PL_sv_undef) {
1186 assert(!SvIMMORTAL(svl));
1187 SvFLAGS(svl) |= SVf_BREAK;
1188 marked = TRUE;
1189 }
1190 else if (!marked) {
1191 /* don't check RH element if no SVf_BREAK flags set yet */
1192 if (!lcount)
1193 break;
1194 continue;
1195 }
1196 }
1197
1198 /* see if corresponding RH element needs copying */
1199
1200 assert(marked);
1201 svr = *relem;
1202 assert(svr);
1203
5c1db569 1204 if (UNLIKELY(SvFLAGS(svr) & (SVf_BREAK|SVs_GMG) || copy_all)) {
a5f48505 1205
ebc643ce
DM
1206#ifdef DEBUGGING
1207 if (fake) {
9ae0115f 1208 /* op_dump(PL_op); */
ebc643ce
DM
1209 Perl_croak(aTHX_
1210 "panic: aassign skipped needed copy of common RH elem %"
1211 UVuf, (UV)(relem - firstrelem));
1212 }
1213#endif
1214
a5f48505
DM
1215 TAINT_NOT; /* Each item is independent */
1216
1217 /* Dear TODO test in t/op/sort.t, I love you.
1218 (It's relying on a panic, not a "semi-panic" from newSVsv()
1219 and then an assertion failure below.) */
1220 if (UNLIKELY(SvIS_FREED(svr))) {
1221 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p",
1222 (void*)svr);
1223 }
1224 /* avoid break flag while copying; otherwise COW etc
1225 * disabled... */
1226 SvFLAGS(svr) &= ~SVf_BREAK;
1227 /* Not newSVsv(), as it does not allow copy-on-write,
8c1e192f
DM
1228 resulting in wasteful copies.
1229 Also, we use SV_NOSTEAL in case the SV is used more than
1230 once, e.g. (...) = (f())[0,0]
1231 Where the same SV appears twice on the RHS without a ref
1232 count bump. (Although I suspect that the SV won't be
1233 stealable here anyway - DAPM).
1234 */
a5f48505
DM
1235 *relem = sv_mortalcopy_flags(svr,
1236 SV_GMAGIC|SV_DO_COW_SVSETSV|SV_NOSTEAL);
1237 /* ... but restore afterwards in case it's needed again,
1238 * e.g. ($a,$b,$c) = (1,$a,$a)
1239 */
1240 SvFLAGS(svr) |= SVf_BREAK;
1241 }
1242
1243 if (!lcount)
1244 break;
1245 }
1246
1247 if (!marked)
1248 return;
1249
1250 /*unmark LHS */
1251
1252 while (lelem > firstlelem) {
1253 SV * const svl = *(--lelem);
1254 if (svl)
1255 SvFLAGS(svl) &= ~SVf_BREAK;
1256 }
1257}
1258
1259
1260
a0d0e21e
LW
1261PP(pp_aassign)
1262{
27da23d5 1263 dVAR; dSP;
3280af22
NIS
1264 SV **lastlelem = PL_stack_sp;
1265 SV **lastrelem = PL_stack_base + POPMARK;
1266 SV **firstrelem = PL_stack_base + POPMARK + 1;
a0d0e21e
LW
1267 SV **firstlelem = lastrelem + 1;
1268
eb578fdb
KW
1269 SV **relem;
1270 SV **lelem;
a0d0e21e 1271
eb578fdb
KW
1272 SV *sv;
1273 AV *ary;
a0d0e21e 1274
1c23e2bd 1275 U8 gimme;
a0d0e21e 1276 HV *hash;
c70927a6 1277 SSize_t i;
a0d0e21e 1278 int magic;
a5f48505 1279 U32 lval;
a68090fe
DM
1280 /* PL_delaymagic is restored by JUMPENV_POP on dieing, so we
1281 * only need to save locally, not on the save stack */
1282 U16 old_delaymagic = PL_delaymagic;
ebc643ce
DM
1283#ifdef DEBUGGING
1284 bool fake = 0;
1285#endif
5637b936 1286
3280af22 1287 PL_delaymagic = DM_DELAY; /* catch simultaneous items */
a0d0e21e
LW
1288
1289 /* If there's a common identifier on both sides we have to take
1290 * special care that assigning the identifier on the left doesn't
1291 * clobber a value on the right that's used later in the list.
1292 */
acdea6f0 1293
beb08a1e
TC
1294 /* at least 2 LH and RH elements, or commonality isn't an issue */
1295 if (firstlelem < lastlelem && firstrelem < lastrelem) {
5c1db569
TC
1296 for (relem = firstrelem+1; relem <= lastrelem; relem++) {
1297 if (SvGMAGICAL(*relem))
1298 goto do_scan;
1299 }
beb08a1e
TC
1300 for (lelem = firstlelem; lelem <= lastlelem; lelem++) {
1301 if (*lelem && SvSMAGICAL(*lelem))
1302 goto do_scan;
a5f48505 1303 }
beb08a1e
TC
1304 if ( PL_op->op_private & (OPpASSIGN_COMMON_SCALAR|OPpASSIGN_COMMON_RC1) ) {
1305 if (PL_op->op_private & OPpASSIGN_COMMON_RC1) {
1306 /* skip the scan if all scalars have a ref count of 1 */
1307 for (lelem = firstlelem; lelem <= lastlelem; lelem++) {
1308 sv = *lelem;
1309 if (!sv || SvREFCNT(sv) == 1)
1310 continue;
1311 if (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVAV)
1312 goto do_scan;
1313 break;
1314 }
1315 }
1316 else {
1317 do_scan:
1318 S_aassign_copy_common(aTHX_
1319 firstlelem, lastlelem, firstrelem, lastrelem
ebc643ce 1320#ifdef DEBUGGING
beb08a1e 1321 , fake
ebc643ce 1322#endif
beb08a1e
TC
1323 );
1324 }
a5f48505 1325 }
a0d0e21e 1326 }
ebc643ce
DM
1327#ifdef DEBUGGING
1328 else {
1329 /* on debugging builds, do the scan even if we've concluded we
1330 * don't need to, then panic if we find commonality. Note that the
1331 * scanner assumes at least 2 elements */
1332 if (firstlelem < lastlelem && firstrelem < lastrelem) {
1333 fake = 1;
1334 goto do_scan;
1335 }
1336 }
1337#endif
a0d0e21e 1338
a5f48505
DM
1339 gimme = GIMME_V;
1340 lval = (gimme == G_ARRAY) ? (PL_op->op_flags & OPf_MOD || LVRET) : 0;
1341
a0d0e21e
LW
1342 relem = firstrelem;
1343 lelem = firstlelem;
4608196e
RGS
1344 ary = NULL;
1345 hash = NULL;
10c8fecd 1346
5d9574c1 1347 while (LIKELY(lelem <= lastlelem)) {
bdaf10a5 1348 bool alias = FALSE;
bbce6d69 1349 TAINT_NOT; /* Each item stands on its own, taintwise. */
a0d0e21e 1350 sv = *lelem++;
bdaf10a5
FC
1351 if (UNLIKELY(!sv)) {
1352 alias = TRUE;
1353 sv = *lelem++;
1354 ASSUME(SvTYPE(sv) == SVt_PVAV);
1355 }
a0d0e21e 1356 switch (SvTYPE(sv)) {
a5f48505
DM
1357 case SVt_PVAV: {
1358 bool already_copied = FALSE;
60edcf09 1359 ary = MUTABLE_AV(sv);
748a9306 1360 magic = SvMAGICAL(ary) != 0;
60edcf09
FC
1361 ENTER;
1362 SAVEFREESV(SvREFCNT_inc_simple_NN(sv));
a5f48505
DM
1363
1364 /* We need to clear ary. The is a danger that if we do this,
1365 * elements on the RHS may be prematurely freed, e.g.
1366 * @a = ($a[0]);
1367 * In the case of possible commonality, make a copy of each
1368 * RHS SV *before* clearing the array, and add a reference
1369 * from the tmps stack, so that it doesn't leak on death.
1370 * Otherwise, make a copy of each RHS SV only as we're storing
1371 * it into the array - that way we don't have to worry about
1372 * it being leaked if we die, but don't incur the cost of
1373 * mortalising everything.
1374 */
1375
1376 if ( (PL_op->op_private & OPpASSIGN_COMMON_AGG)
1377 && (relem <= lastrelem)
1378 && (magic || AvFILL(ary) != -1))
1379 {
1380 SV **svp;
1381 EXTEND_MORTAL(lastrelem - relem + 1);
1382 for (svp = relem; svp <= lastrelem; svp++) {
8c1e192f 1383 /* see comment in S_aassign_copy_common about SV_NOSTEAL */
a5f48505
DM
1384 *svp = sv_mortalcopy_flags(*svp,
1385 SV_GMAGIC|SV_DO_COW_SVSETSV|SV_NOSTEAL);
1386 TAINT_NOT;
1387 }
1388 already_copied = TRUE;
1389 }
1390
1391 av_clear(ary);
6768377c
DM
1392 if (relem <= lastrelem)
1393 av_extend(ary, lastrelem - relem);
1394
a0d0e21e
LW
1395 i = 0;
1396 while (relem <= lastrelem) { /* gobble up all the rest */
5117ca91 1397 SV **didstore;
bdaf10a5 1398 if (LIKELY(!alias)) {
a5f48505
DM
1399 if (already_copied)
1400 sv = *relem;
1401 else {
1402 if (LIKELY(*relem))
1403 /* before newSV, in case it dies */
1404 SvGETMAGIC(*relem);
1405 sv = newSV(0);
8c1e192f
DM
1406 /* see comment in S_aassign_copy_common about
1407 * SV_NOSTEAL */
1408 sv_setsv_flags(sv, *relem,
1409 (SV_DO_COW_SVSETSV|SV_NOSTEAL));
a5f48505
DM
1410 *relem = sv;
1411 }
bdaf10a5
FC
1412 }
1413 else {
a5f48505
DM
1414 if (!already_copied)
1415 SvGETMAGIC(*relem);
bdaf10a5
FC
1416 if (!SvROK(*relem))
1417 DIE(aTHX_ "Assigned value is not a reference");
1418 if (SvTYPE(SvRV(*relem)) > SVt_PVLV)
1419 /* diag_listed_as: Assigned value is not %s reference */
1420 DIE(aTHX_
1421 "Assigned value is not a SCALAR reference");
a5f48505 1422 if (lval && !already_copied)
bdaf10a5
FC
1423 *relem = sv_mortalcopy(*relem);
1424 /* XXX else check for weak refs? */
0ad694a7 1425 sv = SvREFCNT_inc_NN(SvRV(*relem));
bdaf10a5
FC
1426 }
1427 relem++;
a5f48505 1428 if (already_copied)
0ad694a7 1429 SvREFCNT_inc_simple_void_NN(sv); /* undo mortal free */
5117ca91
GS
1430 didstore = av_store(ary,i++,sv);
1431 if (magic) {
18024492
FC
1432 if (!didstore)
1433 sv_2mortal(sv);
8ef24240 1434 if (SvSMAGICAL(sv))
fb73857a 1435 mg_set(sv);
5117ca91 1436 }
bbce6d69 1437 TAINT_NOT;
a0d0e21e 1438 }
5d9574c1 1439 if (UNLIKELY(PL_delaymagic & DM_ARRAY_ISA))
ad64d0ec 1440 SvSETMAGIC(MUTABLE_SV(ary));
60edcf09 1441 LEAVE;
a0d0e21e 1442 break;
a5f48505
DM
1443 }
1444
10c8fecd 1445 case SVt_PVHV: { /* normal hash */
a0d0e21e 1446 SV *tmpstr;
1c4ea384
RZ
1447 int odd;
1448 int duplicates = 0;
45960564 1449 SV** topelem = relem;
1c4ea384 1450 SV **firsthashrelem = relem;
a5f48505 1451 bool already_copied = FALSE;
a0d0e21e 1452
60edcf09 1453 hash = MUTABLE_HV(sv);
748a9306 1454 magic = SvMAGICAL(hash) != 0;
1c4ea384
RZ
1455
1456 odd = ((lastrelem - firsthashrelem)&1)? 0 : 1;
5d9574c1 1457 if (UNLIKELY(odd)) {
fb8f4cf8 1458 do_oddball(lastrelem, firsthashrelem);
1d2b3927
HS
1459 /* we have firstlelem to reuse, it's not needed anymore
1460 */
1c4ea384
RZ
1461 *(lastrelem+1) = &PL_sv_undef;
1462 }
1463
60edcf09
FC
1464 ENTER;
1465 SAVEFREESV(SvREFCNT_inc_simple_NN(sv));
a5f48505
DM
1466
1467 /* We need to clear hash. The is a danger that if we do this,
1468 * elements on the RHS may be prematurely freed, e.g.
1469 * %h = (foo => $h{bar});
1470 * In the case of possible commonality, make a copy of each
1471 * RHS SV *before* clearing the hash, and add a reference
1472 * from the tmps stack, so that it doesn't leak on death.
1473 */
1474
1475 if ( (PL_op->op_private & OPpASSIGN_COMMON_AGG)
1476 && (relem <= lastrelem)
1477 && (magic || HvUSEDKEYS(hash)))
1478 {
1479 SV **svp;
1480 EXTEND_MORTAL(lastrelem - relem + 1);
1481 for (svp = relem; svp <= lastrelem; svp++) {
1482 *svp = sv_mortalcopy_flags(*svp,
1483 SV_GMAGIC|SV_DO_COW_SVSETSV|SV_NOSTEAL);
1484 TAINT_NOT;
1485 }
1486 already_copied = TRUE;
1487 }
1488
a0d0e21e 1489 hv_clear(hash);
a5f48505 1490
5d9574c1 1491 while (LIKELY(relem < lastrelem+odd)) { /* gobble up all the rest */
5117ca91 1492 HE *didstore;
1c4ea384 1493 assert(*relem);
632b9d6f
FC
1494 /* Copy the key if aassign is called in lvalue context,
1495 to avoid having the next op modify our rhs. Copy
1496 it also if it is gmagical, lest it make the
1497 hv_store_ent call below croak, leaking the value. */
a5f48505 1498 sv = (lval || SvGMAGICAL(*relem)) && !already_copied
632b9d6f
FC
1499 ? sv_mortalcopy(*relem)
1500 : *relem;
45960564 1501 relem++;
1c4ea384 1502 assert(*relem);
a5f48505
DM
1503 if (already_copied)
1504 tmpstr = *relem++;
1505 else {
1506 SvGETMAGIC(*relem);
1507 tmpstr = newSV(0);
1508 sv_setsv_nomg(tmpstr,*relem++); /* value */
1509 }
1510
a88bf2bc 1511 if (gimme == G_ARRAY) {
45960564
DM
1512 if (hv_exists_ent(hash, sv, 0))
1513 /* key overwrites an existing entry */
1514 duplicates += 2;
a88bf2bc 1515 else {
45960564 1516 /* copy element back: possibly to an earlier
1d2b3927
HS
1517 * stack location if we encountered dups earlier,
1518 * possibly to a later stack location if odd */
45960564
DM
1519 *topelem++ = sv;
1520 *topelem++ = tmpstr;
1521 }
1522 }
a5f48505 1523 if (already_copied)
0ad694a7 1524 SvREFCNT_inc_simple_void_NN(tmpstr); /* undo mortal free */
5117ca91 1525 didstore = hv_store_ent(hash,sv,tmpstr,0);
632b9d6f
FC
1526 if (magic) {
1527 if (!didstore) sv_2mortal(tmpstr);
1528 SvSETMAGIC(tmpstr);
1529 }
bbce6d69 1530 TAINT_NOT;
8e07c86e 1531 }
60edcf09 1532 LEAVE;
1c4ea384
RZ
1533 if (duplicates && gimme == G_ARRAY) {
1534 /* at this point we have removed the duplicate key/value
1535 * pairs from the stack, but the remaining values may be
1536 * wrong; i.e. with (a 1 a 2 b 3) on the stack we've removed
1537 * the (a 2), but the stack now probably contains
1538 * (a <freed> b 3), because { hv_save(a,1); hv_save(a,2) }
1539 * obliterates the earlier key. So refresh all values. */
1540 lastrelem -= duplicates;
1541 relem = firsthashrelem;
1542 while (relem < lastrelem+odd) {
1543 HE *he;
1544 he = hv_fetch_ent(hash, *relem++, 0, 0);
1545 *relem++ = (he ? HeVAL(he) : &PL_sv_undef);
1546 }
1547 }
1548 if (odd && gimme == G_ARRAY) lastrelem++;
a0d0e21e
LW
1549 }
1550 break;
1551 default:
6fc92669
GS
1552 if (SvIMMORTAL(sv)) {
1553 if (relem <= lastrelem)
1554 relem++;
1555 break;
a0d0e21e
LW
1556 }
1557 if (relem <= lastrelem) {
5d9574c1 1558 if (UNLIKELY(
1c70fb82
FC
1559 SvTEMP(sv) && !SvSMAGICAL(sv) && SvREFCNT(sv) == 1 &&
1560 (!isGV_with_GP(sv) || SvFAKE(sv)) && ckWARN(WARN_MISC)
5d9574c1 1561 ))
1c70fb82
FC
1562 Perl_warner(aTHX_
1563 packWARN(WARN_MISC),
1564 "Useless assignment to a temporary"
1565 );
a0d0e21e
LW
1566 sv_setsv(sv, *relem);
1567 *(relem++) = sv;
1568 }
1569 else
3280af22 1570 sv_setsv(sv, &PL_sv_undef);
8ef24240 1571 SvSETMAGIC(sv);
a0d0e21e
LW
1572 break;
1573 }
1574 }
5d9574c1 1575 if (UNLIKELY(PL_delaymagic & ~DM_DELAY)) {
985213f2 1576 /* Will be used to set PL_tainting below */
dfff4baf
BF
1577 Uid_t tmp_uid = PerlProc_getuid();
1578 Uid_t tmp_euid = PerlProc_geteuid();
1579 Gid_t tmp_gid = PerlProc_getgid();
1580 Gid_t tmp_egid = PerlProc_getegid();
985213f2 1581
b469f1e0 1582 /* XXX $> et al currently silently ignore failures */
3280af22 1583 if (PL_delaymagic & DM_UID) {
a0d0e21e 1584#ifdef HAS_SETRESUID
b469f1e0
JH
1585 PERL_UNUSED_RESULT(
1586 setresuid((PL_delaymagic & DM_RUID) ? PL_delaymagic_uid : (Uid_t)-1,
1587 (PL_delaymagic & DM_EUID) ? PL_delaymagic_euid : (Uid_t)-1,
1588 (Uid_t)-1));
56febc5e
AD
1589#else
1590# ifdef HAS_SETREUID
b469f1e0
JH
1591 PERL_UNUSED_RESULT(
1592 setreuid((PL_delaymagic & DM_RUID) ? PL_delaymagic_uid : (Uid_t)-1,
1593 (PL_delaymagic & DM_EUID) ? PL_delaymagic_euid : (Uid_t)-1));
56febc5e
AD
1594# else
1595# ifdef HAS_SETRUID
b28d0864 1596 if ((PL_delaymagic & DM_UID) == DM_RUID) {
b469f1e0 1597 PERL_UNUSED_RESULT(setruid(PL_delaymagic_uid));
b28d0864 1598 PL_delaymagic &= ~DM_RUID;
a0d0e21e 1599 }
56febc5e
AD
1600# endif /* HAS_SETRUID */
1601# ifdef HAS_SETEUID
b28d0864 1602 if ((PL_delaymagic & DM_UID) == DM_EUID) {
b469f1e0 1603 PERL_UNUSED_RESULT(seteuid(PL_delaymagic_euid));
b28d0864 1604 PL_delaymagic &= ~DM_EUID;
a0d0e21e 1605 }
56febc5e 1606# endif /* HAS_SETEUID */
b28d0864 1607 if (PL_delaymagic & DM_UID) {
985213f2 1608 if (PL_delaymagic_uid != PL_delaymagic_euid)
cea2e8a9 1609 DIE(aTHX_ "No setreuid available");
b469f1e0 1610 PERL_UNUSED_RESULT(PerlProc_setuid(PL_delaymagic_uid));
a0d0e21e 1611 }
56febc5e
AD
1612# endif /* HAS_SETREUID */
1613#endif /* HAS_SETRESUID */
04783dc7 1614
985213f2
AB
1615 tmp_uid = PerlProc_getuid();
1616 tmp_euid = PerlProc_geteuid();
a0d0e21e 1617 }
b469f1e0 1618 /* XXX $> et al currently silently ignore failures */
3280af22 1619 if (PL_delaymagic & DM_GID) {
a0d0e21e 1620#ifdef HAS_SETRESGID
b469f1e0
JH
1621 PERL_UNUSED_RESULT(
1622 setresgid((PL_delaymagic & DM_RGID) ? PL_delaymagic_gid : (Gid_t)-1,
1623 (PL_delaymagic & DM_EGID) ? PL_delaymagic_egid : (Gid_t)-1,
1624 (Gid_t)-1));
56febc5e
AD
1625#else
1626# ifdef HAS_SETREGID
b469f1e0
JH
1627 PERL_UNUSED_RESULT(
1628 setregid((PL_delaymagic & DM_RGID) ? PL_delaymagic_gid : (Gid_t)-1,
1629 (PL_delaymagic & DM_EGID) ? PL_delaymagic_egid : (Gid_t)-1));
56febc5e
AD
1630# else
1631# ifdef HAS_SETRGID
b28d0864 1632 if ((PL_delaymagic & DM_GID) == DM_RGID) {
b469f1e0 1633 PERL_UNUSED_RESULT(setrgid(PL_delaymagic_gid));
b28d0864 1634 PL_delaymagic &= ~DM_RGID;
a0d0e21e 1635 }
56febc5e
AD
1636# endif /* HAS_SETRGID */
1637# ifdef HAS_SETEGID
b28d0864 1638 if ((PL_delaymagic & DM_GID) == DM_EGID) {
b469f1e0 1639 PERL_UNUSED_RESULT(setegid(PL_delaymagic_egid));
b28d0864 1640 PL_delaymagic &= ~DM_EGID;
a0d0e21e 1641 }
56febc5e 1642# endif /* HAS_SETEGID */
b28d0864 1643 if (PL_delaymagic & DM_GID) {
985213f2 1644 if (PL_delaymagic_gid != PL_delaymagic_egid)
cea2e8a9 1645 DIE(aTHX_ "No setregid available");
b469f1e0 1646 PERL_UNUSED_RESULT(PerlProc_setgid(PL_delaymagic_gid));
a0d0e21e 1647 }
56febc5e
AD
1648# endif /* HAS_SETREGID */
1649#endif /* HAS_SETRESGID */
04783dc7 1650
985213f2
AB
1651 tmp_gid = PerlProc_getgid();
1652 tmp_egid = PerlProc_getegid();
a0d0e21e 1653 }
284167a5 1654 TAINTING_set( TAINTING_get | (tmp_uid && (tmp_euid != tmp_uid || tmp_egid != tmp_gid)) );
9a9b5ec9
DM
1655#ifdef NO_TAINT_SUPPORT
1656 PERL_UNUSED_VAR(tmp_uid);
1657 PERL_UNUSED_VAR(tmp_euid);
1658 PERL_UNUSED_VAR(tmp_gid);
1659 PERL_UNUSED_VAR(tmp_egid);
1660#endif
a0d0e21e 1661 }
a68090fe 1662 PL_delaymagic = old_delaymagic;
54310121 1663
54310121
PP
1664 if (gimme == G_VOID)
1665 SP = firstrelem - 1;
1666 else if (gimme == G_SCALAR) {
1667 dTARGET;
1668 SP = firstrelem;
231cbeb2 1669 SETi(lastrelem - firstrelem + 1);
54310121
PP
1670 }
1671 else {
1c4ea384 1672 if (ary || hash)
1d2b3927
HS
1673 /* note that in this case *firstlelem may have been overwritten
1674 by sv_undef in the odd hash case */
a0d0e21e 1675 SP = lastrelem;
1c4ea384 1676 else {
a0d0e21e 1677 SP = firstrelem + (lastlelem - firstlelem);
1c4ea384
RZ
1678 lelem = firstlelem + (relem - firstrelem);
1679 while (relem <= SP)
1680 *relem++ = (lelem <= lastlelem) ? *lelem++ : &PL_sv_undef;
1681 }
a0d0e21e 1682 }
08aeb9f7 1683
54310121 1684 RETURN;
a0d0e21e
LW
1685}
1686
8782bef2
GB
1687PP(pp_qr)
1688{
20b7effb 1689 dSP;
eb578fdb 1690 PMOP * const pm = cPMOP;
fe578d7f 1691 REGEXP * rx = PM_GETRE(pm);
10599a69 1692 SV * const pkg = rx ? CALLREG_PACKAGE(rx) : NULL;
c4420975 1693 SV * const rv = sv_newmortal();
d63c20f2
DM
1694 CV **cvp;
1695 CV *cv;
288b8c02
NC
1696
1697 SvUPGRADE(rv, SVt_IV);
c2123ae3
NC
1698 /* For a subroutine describing itself as "This is a hacky workaround" I'm
1699 loathe to use it here, but it seems to be the right fix. Or close.
1700 The key part appears to be that it's essential for pp_qr to return a new
1701 object (SV), which implies that there needs to be an effective way to
1702 generate a new SV from the existing SV that is pre-compiled in the
1703 optree. */
1704 SvRV_set(rv, MUTABLE_SV(reg_temp_copy(NULL, rx)));
288b8c02
NC
1705 SvROK_on(rv);
1706
8d919b0a 1707 cvp = &( ReANY((REGEXP *)SvRV(rv))->qr_anoncv);
5d9574c1 1708 if (UNLIKELY((cv = *cvp) && CvCLONE(*cvp))) {
d63c20f2 1709 *cvp = cv_clone(cv);
fc2b2dca 1710 SvREFCNT_dec_NN(cv);
d63c20f2
DM
1711 }
1712
288b8c02 1713 if (pkg) {
f815daf2 1714 HV *const stash = gv_stashsv(pkg, GV_ADD);
fc2b2dca 1715 SvREFCNT_dec_NN(pkg);
288b8c02
NC
1716 (void)sv_bless(rv, stash);
1717 }
1718
5d9574c1 1719 if (UNLIKELY(RX_ISTAINTED(rx))) {
e08e52cf 1720 SvTAINTED_on(rv);
9274aefd
DM
1721 SvTAINTED_on(SvRV(rv));
1722 }
c8c13c22 1723 XPUSHs(rv);
1724 RETURN;
8782bef2
GB
1725}
1726
a0d0e21e
LW
1727PP(pp_match)
1728{
20b7effb 1729 dSP; dTARG;
eb578fdb 1730 PMOP *pm = cPMOP;
d65afb4b 1731 PMOP *dynpm = pm;
eb578fdb 1732 const char *s;
5c144d81 1733 const char *strend;
99a90e59 1734 SSize_t curpos = 0; /* initial pos() or current $+[0] */
a0d0e21e 1735 I32 global;
7fadf4a7 1736 U8 r_flags = 0;
5c144d81 1737 const char *truebase; /* Start of string */
eb578fdb 1738 REGEXP *rx = PM_GETRE(pm);
b3eb6a9b 1739 bool rxtainted;
1c23e2bd 1740 const U8 gimme = GIMME_V;
a0d0e21e 1741 STRLEN len;
a3b680e6 1742 const I32 oldsave = PL_savestack_ix;
e60df1fa 1743 I32 had_zerolen = 0;
b1422d62 1744 MAGIC *mg = NULL;
a0d0e21e 1745
533c011a 1746 if (PL_op->op_flags & OPf_STACKED)
a0d0e21e 1747 TARG = POPs;
6ffceeb7 1748 else if (ARGTARG)
59f00321 1749 GETTARGET;
a0d0e21e 1750 else {
54b9620d 1751 TARG = DEFSV;
a0d0e21e
LW
1752 EXTEND(SP,1);
1753 }
d9f424b2 1754
c277df42 1755 PUTBACK; /* EVAL blocks need stack_sp. */
69dc4b30
FC
1756 /* Skip get-magic if this is a qr// clone, because regcomp has
1757 already done it. */
f1d31338 1758 truebase = ReANY(rx)->mother_re
69dc4b30
FC
1759 ? SvPV_nomg_const(TARG, len)
1760 : SvPV_const(TARG, len);
f1d31338 1761 if (!truebase)
2269b42e 1762 DIE(aTHX_ "panic: pp_match");
f1d31338 1763 strend = truebase + len;
284167a5
SM
1764 rxtainted = (RX_ISTAINTED(rx) ||
1765 (TAINT_get && (pm->op_pmflags & PMf_RETAINT)));
9212bbba 1766 TAINT_NOT;
a0d0e21e 1767
6c864ec2 1768 /* We need to know this in case we fail out early - pos() must be reset */
de0df3c0
MH
1769 global = dynpm->op_pmflags & PMf_GLOBAL;
1770
d65afb4b 1771 /* PMdf_USED is set after a ?? matches once */
c737faaf
YO
1772 if (
1773#ifdef USE_ITHREADS
1774 SvREADONLY(PL_regex_pad[pm->op_pmoffset])
1775#else
1776 pm->op_pmflags & PMf_USED
1777#endif
1778 ) {
e5dc5375 1779 DEBUG_r(PerlIO_printf(Perl_debug_log, "?? already matched once"));
de0df3c0 1780 goto nope;
a0d0e21e
LW
1781 }
1782
7e313637
FC
1783 /* empty pattern special-cased to use last successful pattern if
1784 possible, except for qr// */
8d919b0a 1785 if (!ReANY(rx)->mother_re && !RX_PRELEN(rx)
7e313637 1786 && PL_curpm) {
3280af22 1787 pm = PL_curpm;
aaa362c4 1788 rx = PM_GETRE(pm);
a0d0e21e 1789 }
d65afb4b 1790
389ecb56 1791 if (RX_MINLEN(rx) >= 0 && (STRLEN)RX_MINLEN(rx) > len) {
75d43e96
FC
1792 DEBUG_r(PerlIO_printf(Perl_debug_log, "String shorter than min possible regex match (%"
1793 UVuf" < %"IVdf")\n",
1794 (UV)len, (IV)RX_MINLEN(rx)));
de0df3c0 1795 goto nope;
e5dc5375 1796 }
c277df42 1797
8ef97b0e 1798 /* get pos() if //g */
de0df3c0 1799 if (global) {
b1422d62 1800 mg = mg_find_mglob(TARG);
8ef97b0e 1801 if (mg && mg->mg_len >= 0) {
25fdce4a 1802 curpos = MgBYTEPOS(mg, TARG, truebase, len);
8ef97b0e
DM
1803 /* last time pos() was set, it was zero-length match */
1804 if (mg->mg_flags & MGf_MINMATCH)
1805 had_zerolen = 1;
1806 }
a0d0e21e 1807 }
8ef97b0e 1808
6e240d0b 1809#ifdef PERL_SAWAMPERSAND
a41aa44d 1810 if ( RX_NPARENS(rx)
6502e081 1811 || PL_sawampersand
6502e081 1812 || (RX_EXTFLAGS(rx) & (RXf_EVAL_SEEN|RXf_PMf_KEEPCOPY))
5b0e71e9 1813 || (dynpm->op_pmflags & PMf_KEEPCOPY)
6e240d0b
FC
1814 )
1815#endif
1816 {
6502e081
DM
1817 r_flags |= (REXEC_COPY_STR|REXEC_COPY_SKIP_PRE);
1818 /* in @a =~ /(.)/g, we iterate multiple times, but copy the buffer
1819 * only on the first iteration. Therefore we need to copy $' as well
1820 * as $&, to make the rest of the string available for captures in
1821 * subsequent iterations */
1822 if (! (global && gimme == G_ARRAY))
1823 r_flags |= REXEC_COPY_SKIP_POST;
1824 };
5b0e71e9
DM
1825#ifdef PERL_SAWAMPERSAND
1826 if (dynpm->op_pmflags & PMf_KEEPCOPY)
1827 /* handle KEEPCOPY in pmop but not rx, eg $r=qr/a/; /$r/p */
1828 r_flags &= ~(REXEC_COPY_SKIP_PRE|REXEC_COPY_SKIP_POST);
1829#endif
22e551b9 1830
f1d31338
DM
1831 s = truebase;
1832
d7be1480 1833 play_it_again:
985afbc1 1834 if (global)
03c83e26 1835 s = truebase + curpos;
f722798b 1836
77da2310 1837 if (!CALLREGEXEC(rx, (char*)s, (char *)strend, (char*)truebase,
03c83e26 1838 had_zerolen, TARG, NULL, r_flags))
03b6c93d 1839 goto nope;
77da2310
NC
1840
1841 PL_curpm = pm;
985afbc1 1842 if (dynpm->op_pmflags & PMf_ONCE)
c737faaf 1843#ifdef USE_ITHREADS
77da2310 1844 SvREADONLY_on(PL_regex_pad[dynpm->op_pmoffset]);
c737faaf 1845#else
77da2310 1846 dynpm->op_pmflags |= PMf_USED;
c737faaf 1847#endif
a0d0e21e 1848
72311751
GS
1849 if (rxtainted)
1850 RX_MATCH_TAINTED_on(rx);
1851 TAINT_IF(RX_MATCH_TAINTED(rx));
35c2ccc3
DM
1852
1853 /* update pos */
1854
1855 if (global && (gimme != G_ARRAY || (dynpm->op_pmflags & PMf_CONTINUE))) {
b1422d62 1856 if (!mg)
35c2ccc3 1857 mg = sv_magicext_mglob(TARG);
25fdce4a 1858 MgBYTEPOS_set(mg, TARG, truebase, RX_OFFS(rx)[0].end);
adf51885
DM
1859 if (RX_ZERO_LEN(rx))
1860 mg->mg_flags |= MGf_MINMATCH;
1861 else
1862 mg->mg_flags &= ~MGf_MINMATCH;
35c2ccc3
DM
1863 }
1864
bf9dff51
DM
1865 if ((!RX_NPARENS(rx) && !global) || gimme != G_ARRAY) {
1866 LEAVE_SCOPE(oldsave);
1867 RETPUSHYES;
1868 }
1869
88ab22af
DM
1870 /* push captures on stack */
1871
bf9dff51 1872 {
07bc277f 1873 const I32 nparens = RX_NPARENS(rx);
a3b680e6 1874 I32 i = (global && !nparens) ? 1 : 0;
a0d0e21e 1875
c277df42 1876 SPAGAIN; /* EVAL blocks could move the stack. */
ffc61ed2
JH
1877 EXTEND(SP, nparens + i);
1878 EXTEND_MORTAL(nparens + i);
1879 for (i = !i; i <= nparens; i++) {
a0d0e21e 1880 PUSHs(sv_newmortal());
5d9574c1
DM
1881 if (LIKELY((RX_OFFS(rx)[i].start != -1)
1882 && RX_OFFS(rx)[i].end != -1 ))
1883 {
07bc277f 1884 const I32 len = RX_OFFS(rx)[i].end - RX_OFFS(rx)[i].start;
f1d31338 1885 const char * const s = RX_OFFS(rx)[i].start + truebase;
5d9574c1
DM
1886 if (UNLIKELY(RX_OFFS(rx)[i].end < 0 || RX_OFFS(rx)[i].start < 0
1887 || len < 0 || len > strend - s))
5637ef5b
NC
1888 DIE(aTHX_ "panic: pp_match start/end pointers, i=%ld, "
1889 "start=%ld, end=%ld, s=%p, strend=%p, len=%"UVuf,
1890 (long) i, (long) RX_OFFS(rx)[i].start,
1891 (long)RX_OFFS(rx)[i].end, s, strend, (UV) len);
a0d0e21e 1892 sv_setpvn(*SP, s, len);
cce850e4 1893 if (DO_UTF8(TARG) && is_utf8_string((U8*)s, len))
a197cbdd 1894 SvUTF8_on(*SP);
a0d0e21e
LW
1895 }
1896 }
1897 if (global) {
0e0b3e82 1898 curpos = (UV)RX_OFFS(rx)[0].end;
03c83e26 1899 had_zerolen = RX_ZERO_LEN(rx);
c277df42 1900 PUTBACK; /* EVAL blocks may use stack */
cf93c79d 1901 r_flags |= REXEC_IGNOREPOS | REXEC_NOT_FIRST;
a0d0e21e
LW
1902 goto play_it_again;
1903 }
4633a7c4 1904 LEAVE_SCOPE(oldsave);
a0d0e21e
LW
1905 RETURN;
1906 }
e5964223 1907 NOT_REACHED; /* NOTREACHED */
a0d0e21e 1908
7b52d656 1909 nope:
d65afb4b 1910 if (global && !(dynpm->op_pmflags & PMf_CONTINUE)) {
b1422d62
DM
1911 if (!mg)
1912 mg = mg_find_mglob(TARG);
1913 if (mg)
1914 mg->mg_len = -1;
a0d0e21e 1915 }
4633a7c4 1916 LEAVE_SCOPE(oldsave);
a0d0e21e
LW
1917 if (gimme == G_ARRAY)
1918 RETURN;
1919 RETPUSHNO;
1920}
1921
1922OP *
864dbfa3 1923Perl_do_readline(pTHX)
a0d0e21e 1924{
20b7effb 1925 dSP; dTARGETSTACKED;
eb578fdb 1926 SV *sv;
a0d0e21e
LW
1927 STRLEN tmplen = 0;
1928 STRLEN offset;
760ac839 1929 PerlIO *fp;
eb578fdb
KW
1930 IO * const io = GvIO(PL_last_in_gv);
1931 const I32 type = PL_op->op_type;
1c23e2bd 1932 const U8 gimme = GIMME_V;
a0d0e21e 1933
6136c704 1934 if (io) {
50db69d8 1935 const MAGIC *const mg = SvTIED_mg((const SV *)io, PERL_MAGIC_tiedscalar);
6136c704 1936 if (mg) {
3e0cb5de 1937 Perl_tied_method(aTHX_ SV_CONST(READLINE), SP, MUTABLE_SV(io), mg, gimme, 0);
6136c704 1938 if (gimme == G_SCALAR) {
50db69d8
NC
1939 SPAGAIN;
1940 SvSetSV_nosteal(TARG, TOPs);
1941 SETTARG;
6136c704 1942 }
50db69d8 1943 return NORMAL;
0b7c7b4f 1944 }
e79b0511 1945 }
4608196e 1946 fp = NULL;
a0d0e21e
LW
1947 if (io) {
1948 fp = IoIFP(io);
1949 if (!fp) {
1950 if (IoFLAGS(io) & IOf_ARGV) {
1951 if (IoFLAGS(io) & IOf_START) {
a0d0e21e 1952 IoLINES(io) = 0;
b9f2b683 1953 if (av_tindex(GvAVn(PL_last_in_gv)) < 0) {
1d7c1841 1954 IoFLAGS(io) &= ~IOf_START;
d5eb9a46 1955 do_open6(PL_last_in_gv, "-", 1, NULL, NULL, 0);
4bac9ae4 1956 SvTAINTED_off(GvSVn(PL_last_in_gv)); /* previous tainting irrelevant */
76f68e9b 1957 sv_setpvs(GvSVn(PL_last_in_gv), "-");
3280af22 1958 SvSETMAGIC(GvSV(PL_last_in_gv));
a2008d6d
GS
1959 fp = IoIFP(io);
1960 goto have_fp;
a0d0e21e
LW
1961 }
1962 }
157fb5a1 1963 fp = nextargv(PL_last_in_gv, PL_op->op_flags & OPf_SPECIAL);
a0d0e21e 1964 if (!fp) { /* Note: fp != IoIFP(io) */
3280af22 1965 (void)do_close(PL_last_in_gv, FALSE); /* now it does*/
a0d0e21e
LW
1966 }
1967 }
0d44d22b
NC
1968 else if (type == OP_GLOB)
1969 fp = Perl_start_glob(aTHX_ POPs, io);
a0d0e21e
LW
1970 }
1971 else if (type == OP_GLOB)
1972 SP--;
7716c5c5 1973 else if (IoTYPE(io) == IoTYPE_WRONLY) {
a5390457 1974 report_wrongway_fh(PL_last_in_gv, '>');
a00b5bd3 1975 }
a0d0e21e
LW
1976 }
1977 if (!fp) {
041457d9 1978 if ((!io || !(IoFLAGS(io) & IOf_START))
de7dabb6
TC
1979 && ckWARN(WARN_CLOSED)
1980 && type != OP_GLOB)
041457d9 1981 {
de7dabb6 1982 report_evil_fh(PL_last_in_gv);
3f4520fe 1983 }
54310121 1984 if (gimme == G_SCALAR) {
79628082 1985 /* undef TARG, and push that undefined value */
ba92458f 1986 if (type != OP_RCATLINE) {
aab1202a 1987 sv_setsv(TARG,NULL);
ba92458f 1988 }
a0d0e21e
LW
1989 PUSHTARG;
1990 }
1991 RETURN;
1992 }
a2008d6d 1993 have_fp:
54310121 1994 if (gimme == G_SCALAR) {
a0d0e21e 1995 sv = TARG;
0f722b55
RGS
1996 if (type == OP_RCATLINE && SvGMAGICAL(sv))
1997 mg_get(sv);
48de12d9
RGS
1998 if (SvROK(sv)) {
1999 if (type == OP_RCATLINE)
5668452f 2000 SvPV_force_nomg_nolen(sv);
48de12d9
RGS
2001 else
2002 sv_unref(sv);
2003 }
f7877b28 2004 else if (isGV_with_GP(sv)) {
5668452f 2005 SvPV_force_nomg_nolen(sv);
f7877b28 2006 }
862a34c6 2007 SvUPGRADE(sv, SVt_PV);
a0d0e21e 2008 tmplen = SvLEN(sv); /* remember if already alloced */
e3918bb7 2009 if (!tmplen && !SvREADONLY(sv) && !SvIsCOW(sv)) {
f72e8700
JJ
2010 /* try short-buffering it. Please update t/op/readline.t
2011 * if you change the growth length.
2012 */
2013 Sv_Grow(sv, 80);
2014 }
2b5e58c4
AMS
2015 offset = 0;
2016 if (type == OP_RCATLINE && SvOK(sv)) {
2017 if (!SvPOK(sv)) {
5668452f 2018 SvPV_force_nomg_nolen(sv);
2b5e58c4 2019 }
a0d0e21e 2020 offset = SvCUR(sv);
2b5e58c4 2021 }
a0d0e21e 2022 }
54310121 2023 else {
561b68a9 2024 sv = sv_2mortal(newSV(80));
54310121
PP
2025 offset = 0;
2026 }
fbad3eb5 2027
3887d568
AP
2028 /* This should not be marked tainted if the fp is marked clean */
2029#define MAYBE_TAINT_LINE(io, sv) \
2030 if (!(IoFLAGS(io) & IOf_UNTAINT)) { \
2031 TAINT; \
2032 SvTAINTED_on(sv); \
2033 }
2034
684bef36 2035/* delay EOF state for a snarfed empty file */
fbad3eb5 2036#define SNARF_EOF(gimme,rs,io,sv) \
684bef36 2037 (gimme != G_SCALAR || SvCUR(sv) \
b9fee9ba 2038 || (IoFLAGS(io) & IOf_NOLINE) || !RsSNARF(rs))
fbad3eb5 2039
a0d0e21e 2040 for (;;) {
09e8efcc 2041 PUTBACK;
fbad3eb5 2042 if (!sv_gets(sv, fp, offset)
2d726892
TF
2043 && (type == OP_GLOB
2044 || SNARF_EOF(gimme, PL_rs, io, sv)
2045 || PerlIO_error(fp)))
fbad3eb5 2046 {
760ac839 2047 PerlIO_clearerr(fp);
a0d0e21e 2048 if (IoFLAGS(io) & IOf_ARGV) {
157fb5a1 2049 fp = nextargv(PL_last_in_gv, PL_op->op_flags & OPf_SPECIAL);
a0d0e21e
LW
2050 if (fp)
2051 continue;
3280af22 2052 (void)do_close(PL_last_in_gv, FALSE);
a0d0e21e
LW
2053 }
2054 else if (type == OP_GLOB) {
a2a5de95
NC
2055 if (!do_close(PL_last_in_gv, FALSE)) {
2056 Perl_ck_warner(aTHX_ packWARN(WARN_GLOB),
2057 "glob failed (child exited with status %d%s)",
2058 (int)(STATUS_CURRENT >> 8),
2059 (STATUS_CURRENT & 0x80) ? ", core dumped" : "");
4eb79ab5 2060 }
a0d0e21e 2061 }
54310121 2062 if (gimme == G_SCALAR) {
ba92458f
AE
2063 if (type != OP_RCATLINE) {
2064 SV_CHECK_THINKFIRST_COW_DROP(TARG);
0c34ef67 2065 SvOK_off(TARG);
ba92458f 2066 }
09e8efcc 2067 SPAGAIN;
a0d0e21e
LW
2068 PUSHTARG;
2069 }
3887d568 2070 MAYBE_TAINT_LINE(io, sv);
a0d0e21e
LW
2071 RETURN;
2072 }
3887d568 2073 MAYBE_TAINT_LINE(io, sv);
a0d0e21e 2074 IoLINES(io)++;
b9fee9ba 2075 IoFLAGS(io) |= IOf_NOLINE;
71be2cbc 2076 SvSETMAGIC(sv);
09e8efcc 2077 SPAGAIN;
a0d0e21e 2078 XPUSHs(sv);
a0d0e21e 2079 if (type == OP_GLOB) {
349d4f2f 2080 const char *t1;
45a23732 2081 Stat_t statbuf;
a0d0e21e 2082
3280af22 2083 if (SvCUR(sv) > 0 && SvCUR(PL_rs) > 0) {
6136c704 2084 char * const tmps = SvEND(sv) - 1;
aa07b2f6 2085 if (*tmps == *SvPVX_const(PL_rs)) {
c07a80fd 2086 *tmps = '\0';
b162af07 2087 SvCUR_set(sv, SvCUR(sv) - 1);
c07a80fd
PP
2088 }
2089 }
349d4f2f 2090 for (t1 = SvPVX_const(sv); *t1; t1++)
b51c3e77
CB
2091#ifdef __VMS
2092 if (strchr("*%?", *t1))
2093#else
7ad1e72d 2094 if (strchr("$&*(){}[]'\";\\|?<>~`", *t1))
b51c3e77 2095#endif
a0d0e21e 2096 break;
45a23732 2097 if (*t1 && PerlLIO_lstat(SvPVX_const(sv), &statbuf) < 0) {
a0d0e21e
LW
2098 (void)POPs; /* Unmatched wildcard? Chuck it... */
2099 continue;
2100 }
2d79bf7f 2101 } else if (SvUTF8(sv)) { /* OP_READLINE, OP_RCATLINE */
d4c19fe8
AL
2102 if (ckWARN(WARN_UTF8)) {
2103 const U8 * const s = (const U8*)SvPVX_const(sv) + offset;
2104 const STRLEN len = SvCUR(sv) - offset;
2105 const U8 *f;
2106
2107 if (!is_utf8_string_loc(s, len, &f))
2108 /* Emulate :encoding(utf8) warning in the same case. */
2109 Perl_warner(aTHX_ packWARN(WARN_UTF8),
2110 "utf8 \"\\x%02X\" does not map to Unicode",
2111 f < (U8*)SvEND(sv) ? *f : 0);
2112 }
a0d0e21e 2113 }
54310121 2114 if (gimme == G_ARRAY) {
a0d0e21e 2115 if (SvLEN(sv) - SvCUR(sv) > 20) {
1da4ca5f 2116 SvPV_shrink_to_cur(sv);
a0d0e21e 2117 }
561b68a9 2118 sv = sv_2mortal(newSV(80));
a0d0e21e
LW
2119 continue;
2120 }
54310121 2121 else if (gimme == G_SCALAR && !tmplen && SvLEN(sv) - SvCUR(sv) > 80) {
a0d0e21e 2122 /* try to reclaim a bit of scalar space (only on 1st alloc) */
d5b5861b
NC
2123 const STRLEN new_len
2124 = SvCUR(sv) < 60 ? 80 : SvCUR(sv)+40; /* allow some slop */
1da4ca5f 2125 SvPV_renew(sv, new_len);
a0d0e21e
LW
2126 }
2127 RETURN;
2128 }
2129}
2130
a0d0e21e
LW
2131PP(pp_helem)
2132{
20b7effb 2133 dSP;
760ac839 2134 HE* he;
ae77835f 2135 SV **svp;
c445ea15 2136 SV * const keysv = POPs;
85fbaab2 2137 HV * const hv = MUTABLE_HV(POPs);
a3b680e6
AL
2138 const U32 lval = PL_op->op_flags & OPf_MOD || LVRET;
2139 const U32 defer = PL_op->op_private & OPpLVAL_DEFER;
be6c24e0 2140 SV *sv;
92970b93 2141 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
d30e492c 2142 bool preeminent = TRUE;
a0d0e21e 2143
6dfc73ea
SM
2144 if (SvTYPE(hv) != SVt_PVHV)
2145 RETPUSHUNDEF;
d4c19fe8 2146
92970b93 2147 if (localizing) {
d4c19fe8
AL
2148 MAGIC *mg;
2149 HV *stash;
d30e492c
VP
2150
2151 /* If we can determine whether the element exist,
2152 * Try to preserve the existenceness of a tied hash
2153 * element by using EXISTS and DELETE if possible.
2154 * Fallback to FETCH and STORE otherwise. */
2c5f48c2 2155 if (SvCANEXISTDELETE(hv))
d30e492c 2156 preeminent = hv_exists_ent(hv, keysv, 0);
d4c19fe8 2157 }
d30e492c 2158
5f9d7e2b 2159 he = hv_fetch_ent(hv, keysv, lval && !defer, 0);
d4c19fe8 2160 svp = he ? &HeVAL(he) : NULL;
a0d0e21e 2161 if (lval) {
746f6409 2162 if (!svp || !*svp || *svp == &PL_sv_undef) {
68dc0745
PP
2163 SV* lv;
2164 SV* key2;
2d8e6c8d 2165 if (!defer) {
be2597df 2166 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
2d8e6c8d 2167 }
68dc0745
PP
2168 lv = sv_newmortal();
2169 sv_upgrade(lv, SVt_PVLV);
2170 LvTYPE(lv) = 'y';
6136c704 2171 sv_magic(lv, key2 = newSVsv(keysv), PERL_MAGIC_defelem, NULL, 0);
fc2b2dca 2172 SvREFCNT_dec_NN(key2); /* sv_magic() increments refcount */
0ad694a7 2173 LvTARG(lv) = SvREFCNT_inc_simple_NN(hv);
68dc0745
PP
2174 LvTARGLEN(lv) = 1;
2175 PUSHs(lv);
2176 RETURN;
2177 }
92970b93 2178 if (localizing) {
bfcb3514 2179 if (HvNAME_get(hv) && isGV(*svp))
159b6efe 2180 save_gp(MUTABLE_GV(*svp), !(PL_op->op_flags & OPf_SPECIAL));
47cfc530
VP
2181 else if (preeminent)
2182 save_helem_flags(hv, keysv, svp,
2183 (PL_op->op_flags & OPf_SPECIAL) ? 0 : SAVEf_SETMAGIC);
2184 else
2185 SAVEHDELETE(hv, keysv);
5f05dabc 2186 }
9026059d
GG
2187 else if (PL_op->op_private & OPpDEREF) {
2188 PUSHs(vivify_ref(*svp, PL_op->op_private & OPpDEREF));
2189 RETURN;
2190 }
a0d0e21e 2191 }
746f6409 2192 sv = (svp && *svp ? *svp : &PL_sv_undef);
fd69380d
DM
2193 /* Originally this did a conditional C<sv = sv_mortalcopy(sv)>; this
2194 * was to make C<local $tied{foo} = $tied{foo}> possible.
2195 * However, it seems no longer to be needed for that purpose, and
2196 * introduced a new bug: stuff like C<while ($hash{taintedval} =~ /.../g>
2197 * would loop endlessly since the pos magic is getting set on the
2198 * mortal copy and lost. However, the copy has the effect of
2199 * triggering the get magic, and losing it altogether made things like
2200 * c<$tied{foo};> in void context no longer do get magic, which some
2201 * code relied on. Also, delayed triggering of magic on @+ and friends
2202 * meant the original regex may be out of scope by now. So as a
2203 * compromise, do the get magic here. (The MGf_GSKIP flag will stop it
2204 * being called too many times). */
39cf747a 2205 if (!lval && SvRMAGICAL(hv) && SvGMAGICAL(sv))
fd69380d 2206 mg_get(sv);
be6c24e0 2207 PUSHs(sv);
a0d0e21e
LW
2208 RETURN;
2209}
2210
fedf30e1
DM
2211
2212/* a stripped-down version of Perl_softref2xv() for use by
2213 * pp_multideref(), which doesn't use PL_op->op_flags */
2214
f9db5646 2215STATIC GV *
fedf30e1
DM
2216S_softref2xv_lite(pTHX_ SV *const sv, const char *const what,
2217 const svtype type)
2218{
2219 if (PL_op->op_private & HINT_STRICT_REFS) {
2220 if (SvOK(sv))
2221 Perl_die(aTHX_ PL_no_symref_sv, sv,
2222 (SvPOKp(sv) && SvCUR(sv)>32 ? "..." : ""), what);
2223 else
2224 Perl_die(aTHX_ PL_no_usym, what);
2225 }
2226 if (!SvOK(sv))
2227 Perl_die(aTHX_ PL_no_usym, what);
2228 return gv_fetchsv_nomg(sv, GV_ADD, type);
2229}
2230
2231
79815f56
DM
2232/* Handle one or more aggregate derefs and array/hash indexings, e.g.
2233 * $h->{foo} or $a[0]{$key}[$i] or f()->[1]
fedf30e1
DM
2234 *
2235 * op_aux points to an array of unions of UV / IV / SV* / PADOFFSET.
79815f56
DM
2236 * Each of these either contains a set of actions, or an argument, such as
2237 * an IV to use as an array index, or a lexical var to retrieve.
2238 * Several actions re stored per UV; we keep shifting new actions off the
2239 * one UV, and only reload when it becomes zero.
fedf30e1
DM
2240 */
2241
2242PP(pp_multideref)
2243{
2244 SV *sv = NULL; /* init to avoid spurious 'may be used uninitialized' */
2245 UNOP_AUX_item *items = cUNOP_AUXx(PL_op)->op_aux;
2246 UV actions = items->uv;
2247
2248 assert(actions);
2249 /* this tells find_uninit_var() where we're up to */
2250 PL_multideref_pc = items;
2251
2252 while (1) {
2253 /* there are three main classes of action; the first retrieve
2254 * the initial AV or HV from a variable or the stack; the second
2255 * does the equivalent of an unrolled (/DREFAV, rv2av, aelem),
2256 * the third an unrolled (/DREFHV, rv2hv, helem).
2257 */
2258 switch (actions & MDEREF_ACTION_MASK) {
2259
2260 case MDEREF_reload:
2261 actions = (++items)->uv;
2262 continue;
2263
2264 case MDEREF_AV_padav_aelem: /* $lex[...] */
2265 sv = PAD_SVl((++items)->pad_offset);
2266 goto do_AV_aelem;
2267
2268 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
2269 sv = UNOP_AUX_item_sv(++items);
2270 assert(isGV_with_GP(sv));
2271 sv = (SV*)GvAVn((GV*)sv);
2272 goto do_AV_aelem;
2273
2274 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
2275 {
2276 dSP;
2277 sv = POPs;
2278 PUTBACK;
2279 goto do_AV_rv2av_aelem;
2280 }
2281
2282 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
2283 sv = UNOP_AUX_item_sv(++items);
2284 assert(isGV_with_GP(sv));
2285 sv = GvSVn((GV*)sv);
2286 goto do_AV_vivify_rv2av_aelem;
2287
2288 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
2289 sv = PAD_SVl((++items)->pad_offset);
2290 /* FALLTHROUGH */
2291
2292 do_AV_vivify_rv2av_aelem:
2293 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
2294 /* this is the OPpDEREF action normally found at the end of
2295 * ops like aelem, helem, rv2sv */
2296 sv = vivify_ref(sv, OPpDEREF_AV);
2297 /* FALLTHROUGH */
2298
2299 do_AV_rv2av_aelem:
2300 /* this is basically a copy of pp_rv2av when it just has the
2301 * sKR/1 flags */
2302 SvGETMAGIC(sv);
2303 if (LIKELY(SvROK(sv))) {
2304 if (UNLIKELY(SvAMAGIC(sv))) {
2305 sv = amagic_deref_call(sv, to_av_amg);
2306 }
2307 sv = SvRV(sv);
2308 if (UNLIKELY(SvTYPE(sv) != SVt_PVAV))
2309 DIE(aTHX_ "Not an ARRAY reference");
2310 }
2311 else if (SvTYPE(sv) != SVt_PVAV) {
2312 if (!isGV_with_GP(sv))
2313 sv = (SV*)S_softref2xv_lite(aTHX_ sv, "an ARRAY", SVt_PVAV);
2314 sv = MUTABLE_SV(GvAVn((GV*)sv));
2315 }
2316 /* FALLTHROUGH */
2317
2318 do_AV_aelem:
2319 {
2320 /* retrieve the key; this may be either a lexical or package
2321 * var (whose index/ptr is stored as an item) or a signed
2322 * integer constant stored as an item.
2323 */
2324 SV *elemsv;
2325 IV elem = 0; /* to shut up stupid compiler warnings */
2326
2327
2328 assert(SvTYPE(sv) == SVt_PVAV);
2329
2330 switch (actions & MDEREF_INDEX_MASK) {
2331 case MDEREF_INDEX_none:
2332 goto finish;
2333 case MDEREF_INDEX_const:
2334 elem = (++items)->iv;
2335 break;
2336 case MDEREF_INDEX_padsv:
2337 elemsv = PAD_SVl((++items)->pad_offset);
2338 goto check_elem;
2339 case MDEREF_INDEX_gvsv:
2340 elemsv = UNOP_AUX_item_sv(++items);
2341 assert(isGV_with_GP(elemsv));
2342 elemsv = GvSVn((GV*)elemsv);
2343 check_elem:
2344 if (UNLIKELY(SvROK(elemsv) && !SvGAMAGIC(elemsv)
2345 && ckWARN(WARN_MISC)))
2346 Perl_warner(aTHX_ packWARN(WARN_MISC),
2347 "Use of reference \"%"SVf"\" as array index",
2348 SVfARG(elemsv));
2349 /* the only time that S_find_uninit_var() needs this
2350 * is to determine which index value triggered the
2351 * undef warning. So just update it here. Note that
2352 * since we don't save and restore this var (e.g. for
2353 * tie or overload execution), its value will be
2354 * meaningless apart from just here */
2355 PL_multideref_pc = items;
2356 elem = SvIV(elemsv);
2357 break;
2358 }
2359
2360
2361 /* this is basically a copy of pp_aelem with OPpDEREF skipped */
2362
2363 if (!(actions & MDEREF_FLAG_last)) {
2364 SV** svp = av_fetch((AV*)sv, elem, 1);
2365 if (!svp || ! (sv=*svp))
2366 DIE(aTHX_ PL_no_aelem, elem);
2367 break;
2368 }
2369
2370 if (PL_op->op_private &
2371 (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE))
2372 {
2373 if (PL_op->op_private & OPpMULTIDEREF_EXISTS) {
2374 sv = av_exists((AV*)sv, elem) ? &PL_sv_yes : &PL_sv_no;
2375 }
2376 else {
2377 I32 discard = (GIMME_V == G_VOID) ? G_DISCARD : 0;
2378 sv = av_delete((AV*)sv, elem, discard);
2379 if (discard)
2380 return NORMAL;
2381 if (!sv)
2382 sv = &PL_sv_undef;
2383 }
2384 }
2385 else {
2386 const U32 lval = PL_op->op_flags & OPf_MOD || LVRET;
2387 const U32 defer = PL_op->op_private & OPpLVAL_DEFER;
2388 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
2389 bool preeminent = TRUE;
2390 AV *const av = (AV*)sv;
2391 SV** svp;
2392
2393 if (UNLIKELY(localizing)) {
2394 MAGIC *mg;
2395 HV *stash;
2396
2397 /* If we can determine whether the element exist,
2398 * Try to preserve the existenceness of a tied array
2399 * element by using EXISTS and DELETE if possible.
2400 * Fallback to FETCH and STORE otherwise. */
2401 if (SvCANEXISTDELETE(av))
2402 preeminent = av_exists(av, elem);
2403 }
2404
2405 svp = av_fetch(av, elem, lval && !defer);
2406
2407 if (lval) {
2408 if (!svp || !(sv = *svp)) {
2409 IV len;
2410 if (!defer)
2411 DIE(aTHX_ PL_no_aelem, elem);
2412 len = av_tindex(av);
2413 sv = sv_2mortal(newSVavdefelem(av,
2414 /* Resolve a negative index now, unless it points
2415 * before the beginning of the array, in which
2416 * case record it for error reporting in
2417 * magic_setdefelem. */
2418 elem < 0 && len + elem >= 0
2419 ? len + elem : elem, 1));
2420 }
2421 else {
2422 if (UNLIKELY(localizing)) {
2423 if (preeminent) {
2424 save_aelem(av, elem, svp);
2425 sv = *svp; /* may have changed */
2426 }
2427 else
2428 SAVEADELETE(av, elem);
2429 }
2430 }
2431 }
2432 else {
2433 sv = (svp ? *svp : &PL_sv_undef);
2434 /* see note in pp_helem() */
2435 if (SvRMAGICAL(av) && SvGMAGICAL(sv))
2436 mg_get(sv);
2437 }
2438 }
2439
2440 }
2441 finish:
2442 {
2443 dSP;
2444 XPUSHs(sv);
2445 RETURN;
2446 }
2447 /* NOTREACHED */
2448
2449
2450
2451
2452 case MDEREF_HV_padhv_helem: /* $lex{...} */
2453 sv = PAD_SVl((++items)->pad_offset);
2454 goto do_HV_helem;
2455
2456 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
2457 sv = UNOP_AUX_item_sv(++items);
2458 assert(isGV_with_GP(sv));
2459 sv = (SV*)GvHVn((GV*)sv);
2460 goto do_HV_helem;
2461
2462 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
2463 {
2464 dSP;
2465 sv = POPs;
2466 PUTBACK;
2467 goto do_HV_rv2hv_helem;
2468 }
2469
2470 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
2471 sv = UNOP_AUX_item_sv(++items);
2472 assert(isGV_with_GP(sv));
2473 sv = GvSVn((GV*)sv);
2474 goto do_HV_vivify_rv2hv_helem;
2475
2476 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
2477 sv = PAD_SVl((++items)->pad_offset);
2478 /* FALLTHROUGH */
2479
2480 do_HV_vivify_rv2hv_helem:
2481 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
2482 /* this is the OPpDEREF action normally found at the end of
2483 * ops like aelem, helem, rv2sv */
2484 sv = vivify_ref(sv, OPpDEREF_HV);
2485 /* FALLTHROUGH */
2486
2487 do_HV_rv2hv_helem:
2488 /* this is basically a copy of pp_rv2hv when it just has the
2489 * sKR/1 flags (and pp_rv2hv is aliased to pp_rv2av) */
2490
2491 SvGETMAGIC(sv);
2492 if (LIKELY(SvROK(sv))) {
2493 if (UNLIKELY(SvAMAGIC(sv))) {
2494 sv = amagic_deref_call(sv, to_hv_amg);
2495 }
2496 sv = SvRV(sv);
2497 if (UNLIKELY(SvTYPE(sv) != SVt_PVHV))
2498 DIE(aTHX_ "Not a HASH reference");
2499 }
2500 else if (SvTYPE(sv) != SVt_PVHV) {
2501 if (!isGV_with_GP(sv))
2502 sv = (SV*)S_softref2xv_lite(aTHX_ sv, "a HASH", SVt_PVHV);
2503 sv = MUTABLE_SV(GvHVn((GV*)sv));
2504 }
2505 /* FALLTHROUGH */
2506
2507 do_HV_helem:
2508 {
2509 /* retrieve the key; this may be either a lexical / package
2510 * var or a string constant, whose index/ptr is stored as an
2511 * item
2512 */
2513 SV *keysv = NULL; /* to shut up stupid compiler warnings */
2514
2515 assert(SvTYPE(sv) == SVt_PVHV);
2516
2517 switch (actions & MDEREF_INDEX_MASK) {
2518 case MDEREF_INDEX_none:
2519 goto finish;
2520
2521 case MDEREF_INDEX_const:
2522 keysv = UNOP_AUX_item_sv(++items);
2523 break;
2524
2525 case MDEREF_INDEX_padsv:
2526 keysv = PAD_SVl((++items)->pad_offset);
2527 break;
2528
2529 case MDEREF_INDEX_gvsv:
2530 keysv = UNOP_AUX_item_sv(++items);
2531 keysv = GvSVn((GV*)keysv);
2532 break;
2533 }
2534
2535 /* see comment above about setting this var */
2536 PL_multideref_pc = items;
2537
2538
2539 /* ensure that candidate CONSTs have been HEKified */
2540 assert( ((actions & MDEREF_INDEX_MASK) != MDEREF_INDEX_const)
2541 || SvTYPE(keysv) >= SVt_PVMG
2542 || !SvOK(keysv)
2543 || SvROK(keysv)
2544 || SvIsCOW_shared_hash(keysv));
2545
2546 /* this is basically a copy of pp_helem with OPpDEREF skipped */
2547
2548 if (!(actions & MDEREF_FLAG_last)) {
2549 HE *he = hv_fetch_ent((HV*)sv, keysv, 1, 0);
2550 if (!he || !(sv=HeVAL(he)) || sv == &PL_sv_undef)
2551 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
2552 break;
2553 }
2554
2555 if (PL_op->op_private &
2556 (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE))
2557 {
2558 if (PL_op->op_private & OPpMULTIDEREF_EXISTS) {
2559 sv = hv_exists_ent((HV*)sv, keysv, 0)
2560 ? &PL_sv_yes : &PL_sv_no;
2561 }
2562 else {
2563 I32 discard = (GIMME_V == G_VOID) ? G_DISCARD : 0;
2564 sv = hv_delete_ent((HV*)sv, keysv, discard, 0);
2565 if (discard)
2566 return NORMAL;
2567 if (!sv)
2568 sv = &PL_sv_undef;
2569 }
2570 }
2571 else {
2572 const U32 lval = PL_op->op_flags & OPf_MOD || LVRET;
2573 const U32 defer = PL_op->op_private & OPpLVAL_DEFER;
2574 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
2575 bool preeminent = TRUE;
2576 SV **svp;
2577 HV * const hv = (HV*)sv;
2578 HE* he;
2579
2580 if (UNLIKELY(localizing)) {
2581 MAGIC *mg;
2582 HV *stash;
2583
2584 /* If we can determine whether the element exist,
2585 * Try to preserve the existenceness of a tied hash
2586 * element by using EXISTS and DELETE if possible.
2587 * Fallback to FETCH and STORE otherwise. */
2588 if (SvCANEXISTDELETE(hv))
2589 preeminent = hv_exists_ent(hv, keysv, 0);
2590 }
2591
2592 he = hv_fetch_ent(hv, keysv, lval && !defer, 0);
2593 svp = he ? &HeVAL(he) : NULL;
2594
2595
2596 if (lval) {
2597 if (!svp || !(sv = *svp) || sv == &PL_sv_undef) {
2598 SV* lv;
2599 SV* key2;
2600 if (!defer)
2601 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
2602 lv = sv_newmortal();
2603 sv_upgrade(lv, SVt_PVLV);
2604 LvTYPE(lv) = 'y';
2605 sv_magic(lv, key2 = newSVsv(keysv),
2606 PERL_MAGIC_defelem, NULL, 0);
2607 /* sv_magic() increments refcount */
2608 SvREFCNT_dec_NN(key2);
0ad694a7 2609 LvTARG(lv) = SvREFCNT_inc_simple_NN(hv);
fedf30e1
DM
2610 LvTARGLEN(lv) = 1;
2611 sv = lv;
2612 }
2613 else {
2614 if (localizing) {
2615 if (HvNAME_get(hv) && isGV(sv))
2616 save_gp(MUTABLE_GV(sv),
2617 !(PL_op->op_flags & OPf_SPECIAL));
2618 else if (preeminent) {
2619 save_helem_flags(hv, keysv, svp,
2620 (PL_op->op_flags & OPf_SPECIAL)
2621 ? 0 : SAVEf_SETMAGIC);
2622 sv = *svp; /* may have changed */
2623 }
2624 else
2625 SAVEHDELETE(hv, keysv);
2626 }
2627 }
2628 }
2629 else {
2630 sv = (svp && *svp ? *svp : &PL_sv_undef);
2631 /* see note in pp_helem() */
2632 if (SvRMAGICAL(hv) && SvGMAGICAL(sv))
2633 mg_get(sv);
2634 }
2635 }
2636 goto finish;
2637 }
2638
2639 } /* switch */
2640
2641 actions >>= MDEREF_SHIFT;
2642 } /* while */
2643 /* NOTREACHED */
2644}
2645
2646
a0d0e21e
LW
2647PP(pp_iter)
2648{
eb578fdb 2649 PERL_CONTEXT *cx;
7d6c2cef 2650 SV *oldsv;
1d7c1841 2651 SV **itersvp;
8a1f10dd 2652 SV *retsv;
a0d0e21e 2653
84f05d57
JH
2654 SV *sv;
2655 AV *av;
2656 IV ix;
2657 IV inc;
2658
4ebe6e95 2659 cx = CX_CUR();
1d7c1841 2660 itersvp = CxITERVAR(cx);
4b5c941e 2661 assert(itersvp);
a48ce6be
DM
2662
2663 switch (CxTYPE(cx)) {
17c91640 2664
b552b52c
DM
2665 case CXt_LOOP_LAZYSV: /* string increment */
2666 {
2667 SV* cur = cx->blk_loop.state_u.lazysv.cur;
2668 SV *end = cx->blk_loop.state_u.lazysv.end;
2669 /* If the maximum is !SvOK(), pp_enteriter substitutes PL_sv_no.
2670 It has SvPVX of "" and SvCUR of 0, which is what we want. */
2671 STRLEN maxlen = 0;
2672 const char *max = SvPV_const(end, maxlen);
5d9574c1 2673 if (UNLIKELY(SvNIOK(cur) || SvCUR(cur) > maxlen))
8a1f10dd 2674 goto retno;
b552b52c
DM
2675
2676 oldsv = *itersvp;
6d3ca00e
DM
2677 /* NB: on the first iteration, oldsv will have a ref count of at
2678 * least 2 (one extra from blk_loop.itersave), so the GV or pad
2679 * slot will get localised; on subsequent iterations the RC==1
2680 * optimisation may kick in and the SV will be reused. */
2681 if (oldsv && LIKELY(SvREFCNT(oldsv) == 1 && !SvMAGICAL(oldsv))) {
b552b52c
DM
2682 /* safe to reuse old SV */
2683 sv_setsv(oldsv, cur);
a48ce6be 2684 }
b552b52c
DM
2685 else
2686 {
2687 /* we need a fresh SV every time so that loop body sees a
2688 * completely new SV for closures/references to work as
2689 * they used to */
2690 *itersvp = newSVsv(cur);
6d3ca00e 2691 SvREFCNT_dec(oldsv);
b552b52c
DM
2692 }
2693 if (strEQ(SvPVX_const(cur), max))
2694 sv_setiv(cur, 0); /* terminate next time */
2695 else
2696 sv_inc(cur);
2697 break;
2698 }
a48ce6be 2699
fcef60b4
DM
2700 case CXt_LOOP_LAZYIV: /* integer increment */
2701 {
2702 IV cur = cx->blk_loop.state_u.lazyiv.cur;
5d9574c1 2703 if (UNLIKELY(cur > cx->blk_loop.state_u.lazyiv.end))
8a1f10dd 2704 goto retno;
7f61b687 2705
fcef60b4 2706 oldsv = *itersvp;
6d3ca00e
DM
2707 /* see NB comment above */
2708 if (oldsv && LIKELY(SvREFCNT(oldsv) == 1 && !SvMAGICAL(oldsv))) {
eaa5c2d6 2709 /* safe to reuse old SV */
47b96a1e
DM
2710
2711 if ( (SvFLAGS(oldsv) & (SVTYPEMASK|SVf_THINKFIRST|SVf_IVisUV))
2712 == SVt_IV)
2713 {
2714 /* Cheap SvIOK_only().
2715 * Assert that flags which SvIOK_only() would test or
2716 * clear can't be set, because we're SVt_IV */
2717 assert(!(SvFLAGS(oldsv) &
2718 (SVf_OOK|SVf_UTF8|(SVf_OK & ~(SVf_IOK|SVp_IOK)))));
2719 SvFLAGS(oldsv) |= (SVf_IOK|SVp_IOK);
2720 /* SvIV_set() where sv_any points to head */
2721 oldsv->sv_u.svu_iv = cur;
2722
2723 }
2724 else
2725 sv_setiv(oldsv, cur);
eaa5c2d6 2726 }
1c846c1f 2727 else
eaa5c2d6
GA
2728 {
2729 /* we need a fresh SV every time so that loop body sees a
2730 * completely new SV for closures/references to work as they
2731 * used to */
fcef60b4 2732 *itersvp = newSViv(cur);
6d3ca00e 2733 SvREFCNT_dec(oldsv);
eaa5c2d6 2734 }
a2309040 2735
5d9574c1 2736 if (UNLIKELY(cur == IV_MAX)) {
cdc1aa42
NC
2737 /* Handle end of range at IV_MAX */
2738 cx->blk_loop.state_u.lazyiv.end = IV_MIN;
2739 } else
2740 ++cx->blk_loop.state_u.lazyiv.cur;
a48ce6be 2741 break;
fcef60b4 2742 }
a48ce6be 2743
93661e56
DM
2744 case CXt_LOOP_LIST: /* for (1,2,3) */
2745
2746 assert(OPpITER_REVERSED == 2); /* so inc becomes -1 or 1 */
2747 inc = 1 - (PL_op->op_private & OPpITER_REVERSED);
2748 ix = (cx->blk_loop.state_u.stack.ix += inc);
2749 if (UNLIKELY(inc > 0
2750 ? ix > cx->blk_oldsp
2751 : ix <= cx->blk_loop.state_u.stack.basesp)
2752 )
8a1f10dd 2753 goto retno;
93661e56
DM
2754
2755 sv = PL_stack_base[ix];
2756 av = NULL;
2757 goto loop_ary_common;
2758
2759 case CXt_LOOP_ARY: /* for (@ary) */
2760
2761 av = cx->blk_loop.state_u.ary.ary;
2762 inc = 1 - (PL_op->op_private & OPpITER_REVERSED);
2763 ix = (cx->blk_loop.state_u.ary.ix += inc);
2764 if (UNLIKELY(inc > 0
2765 ? ix > AvFILL(av)
2766 : ix < 0)
2767 )
8a1f10dd 2768 goto retno;
de080daa 2769
9d1ee8e0 2770 if (UNLIKELY(SvRMAGICAL(av))) {
a8a20bb6
DM
2771 SV * const * const svp = av_fetch(av, ix, FALSE);
2772 sv = svp ? *svp : NULL;
2773 }
2774 else {
2775 sv = AvARRAY(av)[ix];
de080daa 2776 }
ef3e5ea9 2777
93661e56
DM
2778 loop_ary_common:
2779
d39c26a6
FC
2780 if (UNLIKELY(cx->cx_type & CXp_FOR_LVREF)) {
2781 SvSetMagicSV(*itersvp, sv);
2782 break;
2783 }
2784
5d9574c1
DM
2785 if (LIKELY(sv)) {
2786 if (UNLIKELY(SvIS_FREED(sv))) {
f38aa882
DM
2787 *itersvp = NULL;
2788 Perl_croak(aTHX_ "Use of freed value in iteration");
2789 }
60779a30 2790 if (SvPADTMP(sv)) {
8e079c2a 2791 sv = newSVsv(sv);
60779a30 2792 }
8e079c2a
FC
2793 else {
2794 SvTEMP_off(sv);
2795 SvREFCNT_inc_simple_void_NN(sv);
2796 }
de080daa 2797 }
93661e56 2798 else if (av) {
199f858d 2799 sv = newSVavdefelem(av, ix, 0);
de080daa 2800 }
a600f7e6
FC
2801 else
2802 sv = &PL_sv_undef;
a0d0e21e 2803
de080daa
DM
2804 oldsv = *itersvp;
2805 *itersvp = sv;
2806 SvREFCNT_dec(oldsv);
de080daa 2807 break;
a48ce6be
DM
2808
2809 default:
2810 DIE(aTHX_ "panic: pp_iter, type=%u", CxTYPE(cx));
2811 }
8a1f10dd
DM
2812
2813 retsv = &PL_sv_yes;
2814 if (0) {
2815 retno:
2816 retsv = &PL_sv_no;
2817 }
2818 /* pp_enteriter should have pre-extended the stack */
2819 assert(PL_stack_sp < PL_stack_max);
2820 *++PL_stack_sp =retsv;
2821
2822 return PL_op->op_next;
a0d0e21e
LW
2823}
2824
ef07e810
DM
2825/*
2826A description of how taint works in pattern matching and substitution.
2827
284167a5
SM
2828This is all conditional on NO_TAINT_SUPPORT not being defined. Under
2829NO_TAINT_SUPPORT, taint-related operations should become no-ops.
2830
4e19c54b 2831While the pattern is being assembled/concatenated and then compiled,
284167a5
SM
2832PL_tainted will get set (via TAINT_set) if any component of the pattern
2833is tainted, e.g. /.*$tainted/. At the end of pattern compilation,
2834the RXf_TAINTED flag is set on the pattern if PL_tainted is set (via
1738e041
DM
2835TAINT_get). It will also be set if any component of the pattern matches
2836based on locale-dependent behavior.
ef07e810 2837
0ab462a6
DM
2838When the pattern is copied, e.g. $r = qr/..../, the SV holding the ref to
2839the pattern is marked as tainted. This means that subsequent usage, such
284167a5
SM
2840as /x$r/, will set PL_tainted using TAINT_set, and thus RXf_TAINTED,
2841on the new pattern too.
ef07e810 2842
272d35c9 2843RXf_TAINTED_SEEN is used post-execution by the get magic code
ef07e810
DM
2844of $1 et al to indicate whether the returned value should be tainted.
2845It is the responsibility of the caller of the pattern (i.e. pp_match,
2846pp_subst etc) to set this flag for any other circumstances where $1 needs
2847to be tainted.
2848
2849The taint behaviour of pp_subst (and pp_substcont) is quite complex.
2850
2851There are three possible sources of taint
2852 * the source string
2853 * the pattern (both compile- and run-time, RXf_TAINTED / RXf_TAINTED_SEEN)
2854 * the replacement string (or expression under /e)
2855
2856There are four destinations of taint and they are affected by the sources
2857according to the rules below:
2858
2859 * the return value (not including /r):
2860 tainted by the source string and pattern, but only for the
2861 number-of-iterations case; boolean returns aren't tainted;
2862 * the modified string (or modified copy under /r):
2863 tainted by the source string, pattern, and replacement strings;
2864 * $1 et al:
2865 tainted by the pattern, and under 'use re "taint"', by the source
2866 string too;
2867 * PL_taint - i.e. whether subsequent code (e.g. in a /e block) is tainted:
2868 should always be unset before executing subsequent code.
2869
2870The overall action of pp_subst is:
2871
2872 * at the start, set bits in rxtainted indicating the taint status of
2873 the various sources.
2874
2875 * After each pattern execution, update the SUBST_TAINT_PAT bit in
2876 rxtainted if RXf_TAINTED_SEEN has been set, to indicate that the
2877 pattern has subsequently become tainted via locale ops.
2878
2879 * If control is being passed to pp_substcont to execute a /e block,
2880 save rxtainted in the CXt_SUBST block, for future use by
2881 pp_substcont.
2882
2883 * Whenever control is being returned to perl code (either by falling
2884 off the "end" of pp_subst/pp_substcont, or by entering a /e block),
2885 use the flag bits in rxtainted to make all the appropriate types of
0ab462a6
DM
2886 destination taint visible; e.g. set RXf_TAINTED_SEEN so that $1
2887 et al will appear tainted.
ef07e810
DM
2888
2889pp_match is just a simpler version of the above.
2890
2891*/
2892
a0d0e21e
LW
2893PP(pp_subst)
2894{
20b7effb 2895 dSP; dTARG;
eb578fdb 2896 PMOP *pm = cPMOP;
a0d0e21e 2897 PMOP *rpm = pm;
eb578fdb 2898 char *s;
a0d0e21e 2899 char *strend;
5c144d81 2900 const char *c;
a0d0e21e 2901 STRLEN clen;
3c6ef0a5
FC
2902 SSize_t iters = 0;
2903 SSize_t maxiters;
a0d0e21e 2904 bool once;
ef07e810
DM
2905 U8 rxtainted = 0; /* holds various SUBST_TAINT_* flag bits.
2906 See "how taint works" above */
a0d0e21e 2907 char *orig;
1ed74d04 2908 U8 r_flags;
eb578fdb 2909 REGEXP *rx = PM_GETRE(pm);
a0d0e21e
LW
2910 STRLEN len;
2911 int force_on_match = 0;
0bcc34c2 2912 const I32 oldsave = PL_savestack_ix;
792b2c16 2913 STRLEN slen;
26a74523 2914 bool doutf8 = FALSE; /* whether replacement is in utf8 */
db2c6cb3 2915#ifdef PERL_ANY_COW
106d9a13 2916 bool was_cow;
ed252734 2917#endif
a0714e2c 2918 SV *nsv = NULL;
b770e143 2919 /* known replacement string? */
eb578fdb 2920 SV *dstr = (pm->op_pmflags & PMf_CONST) ? POPs : NULL;
a0d0e21e 2921
f410a211
NC
2922 PERL_ASYNC_CHECK();
2923
533c011a 2924 if (PL_op->op_flags & OPf_STACKED)
a0d0e21e 2925 TARG = POPs;
6ffceeb7 2926 else if (ARGTARG)
59f00321 2927 GETTARGET;
a0d0e21e 2928 else {
54b9620d 2929 TARG = DEFSV;
a0d0e21e 2930 EXTEND(SP,1);
1c846c1f 2931 }
d9f424b2 2932
64534138 2933 SvGETMAGIC(TARG); /* must come before cow check */
db2c6cb3 2934#ifdef PERL_ANY_COW
106d9a13
DM
2935 /* note that a string might get converted to COW during matching */
2936 was_cow = cBOOL(SvIsCOW(TARG));
ed252734 2937#endif
d13a5d3b
TC
2938 if (!(rpm->op_pmflags & PMf_NONDESTRUCT)) {
2939#ifndef PERL_ANY_COW
2940 if (SvIsCOW(TARG))
2941 sv_force_normal_flags(TARG,0);
2942#endif
2943 if ((SvREADONLY(TARG)
2944 || ( ((SvTYPE(TARG) == SVt_PVGV && isGV_with_GP(TARG))
2945 || SvTYPE(TARG) > SVt_PVLV)
2946 && !(SvTYPE(TARG) == SVt_PVGV && SvFAKE(TARG)))))
2947 Perl_croak_no_modify();
2948 }
8ec5e241
NIS
2949 PUTBACK;
2950
6ac6605d
DM
2951 orig = SvPV_nomg(TARG, len);
2952 /* note we don't (yet) force the var into being a string; if we fail
92711104 2953 * to match, we leave as-is; on successful match however, we *will*
6ac6605d 2954 * coerce into a string, then repeat the match */
4499db73 2955 if (!SvPOKp(TARG) || SvTYPE(TARG) == SVt_PVGV || SvVOK(TARG))
a0d0e21e 2956 force_on_match = 1;
20be6587
DM
2957
2958 /* only replace once? */
2959 once = !(rpm->op_pmflags & PMf_GLOBAL);
2960
ef07e810 2961 /* See "how taint works" above */
284167a5 2962 if (TAINTING_get) {
20be6587
DM
2963 rxtainted = (
2964 (SvTAINTED(TARG) ? SUBST_TAINT_STR : 0)
284167a5 2965 | (RX_ISTAINTED(rx) ? SUBST_TAINT_PAT : 0)
20be6587
DM
2966 | ((pm->op_pmflags & PMf_RETAINT) ? SUBST_TAINT_RETAINT : 0)
2967 | ((once && !(rpm->op_pmflags & PMf_NONDESTRUCT))
2968 ? SUBST_TAINT_BOOLRET : 0));
2969 TAINT_NOT;
2970 }
a12c0f56 2971
a0d0e21e 2972 force_it:
6ac6605d
DM
2973 if (!pm || !orig)
2974 DIE(aTHX_ "panic: pp_subst, pm=%p, orig=%p", pm, orig);
a0d0e21e 2975
6ac6605d
DM
2976 strend = orig + len;
2977 slen = DO_UTF8(TARG) ? utf8_length((U8*)orig, (U8*)strend) : len;
792b2c16
JH
2978 maxiters = 2 * slen + 10; /* We can match twice at each
2979 position, once with zero-length,
2980 second time with non-zero. */
a0d0e21e 2981
6a97c51d 2982 if (!RX_PRELEN(rx) && PL_curpm
8d919b0a 2983 && !ReANY(rx)->mother_re) {
3280af22 2984 pm = PL_curpm;
aaa362c4 2985 rx = PM_GETRE(pm);
a0d0e21e 2986 }
6502e081 2987
6e240d0b 2988#ifdef PERL_SAWAMPERSAND
6502e081
DM
2989 r_flags = ( RX_NPARENS(rx)
2990 || PL_sawampersand
6502e081 2991 || (RX_EXTFLAGS(rx) & (RXf_EVAL_SEEN|RXf_PMf_KEEPCOPY))
5b0e71e9 2992 || (rpm->op_pmflags & PMf_KEEPCOPY)
6502e081
DM
2993 )
2994 ? REXEC_COPY_STR
2995 : 0;
6e240d0b
FC
2996#else
2997 r_flags = REXEC_COPY_STR;
2998#endif
7fba1cd6 2999
0395280b 3000 if (!CALLREGEXEC(rx, orig, strend, orig, 0, TARG, NULL, r_flags))
8b64c330 3001 {
5e79dfb9
DM
3002 SPAGAIN;
3003 PUSHs(rpm->op_pmflags & PMf_NONDESTRUCT ? TARG : &PL_sv_no);
3004 LEAVE_SCOPE(oldsave);
3005 RETURN;
3006 }
1754320d
FC
3007 PL_curpm = pm;
3008
71be2cbc 3009 /* known replacement string? */
f272994b 3010 if (dstr) {
8514a05a
JH
3011 /* replacement needing upgrading? */
3012 if (DO_UTF8(TARG) && !doutf8) {
db79b45b 3013 nsv = sv_newmortal();
4a176938 3014 SvSetSV(nsv, dstr);
8df0e7a2 3015 sv_utf8_upgrade(nsv);
5c144d81 3016 c = SvPV_const(nsv, clen);
4a176938
JH
3017 doutf8 = TRUE;
3018 }
3019 else {
5c144d81 3020 c = SvPV_const(dstr, clen);
4a176938 3021 doutf8 = DO_UTF8(dstr);
8514a05a 3022 }
bb933b9b
FC
3023
3024 if (SvTAINTED(dstr))
3025 rxtainted |= SUBST_TAINT_REPL;
f272994b
A
3026 }
3027 else {
6136c704 3028 c = NULL;
f272994b
A
3029 doutf8 = FALSE;
3030 }
3031
71be2cbc 3032 /* can do inplace substitution? */
ed252734 3033 if (c
db2c6cb3 3034#ifdef PERL_ANY_COW
106d9a13 3035 && !was_cow
ed252734 3036#endif
fbfb1899 3037 && (I32)clen <= RX_MINLENRET(rx)
9cefd268
FC
3038 && ( once
3039 || !(r_flags & REXEC_COPY_STR)
3040 || (!SvGMAGICAL(dstr) && !(RX_EXTFLAGS(rx) & RXf_EVAL_SEEN))
3041 )
dbc200c5 3042 && !(RX_EXTFLAGS(rx) & RXf_NO_INPLACE_SUBST)
8ca8a454
NC
3043 && (!doutf8 || SvUTF8(TARG))
3044 && !(rpm->op_pmflags & PMf_NONDESTRUCT))
8b030b38 3045 {
ec911639 3046
db2c6cb3 3047#ifdef PERL_ANY_COW
106d9a13 3048 /* string might have got converted to COW since we set was_cow */
ed252734 3049 if (SvIsCOW(TARG)) {
f7a8268c 3050 if (!force_on_match)
ed252734 3051 goto have_a_cow;
f7a8268c 3052 assert(SvVOK(TARG));
ed252734
NC
3053 }
3054#endif
71be2cbc 3055 if (force_on_match) {
6ac6605d
DM
3056 /* redo the first match, this time with the orig var
3057 * forced into being a string */
71be2cbc 3058 force_on_match = 0;
6ac6605d 3059 orig = SvPV_force_nomg(TARG, len);
71be2cbc
PP
3060 goto force_it;
3061 }
39b40493 3062
71be2cbc 3063 if (once) {
c67ab8f2 3064 char *d, *m;
20be6587
DM
3065 if (RX_MATCH_TAINTED(rx)) /* run time pattern taint, eg locale */
3066 rxtainted |= SUBST_TAINT_PAT;
07bc277f
NC
3067 m = orig + RX_OFFS(rx)[0].start;
3068 d = orig + RX_OFFS(rx)[0].end;
71be2cbc
PP
3069 s = orig;
3070 if (m - s > strend - d) { /* faster to shorten from end */
2ec7214c 3071 I32 i;
71be2cbc
PP
3072 if (clen) {
3073 Copy(c, m, clen, char);
3074 m += clen;
a0d0e21e 3075 }
71be2cbc
PP
3076 i = strend - d;
3077 if (i > 0) {
3078 Move(d, m, i, char);
3079 m += i;
a0d0e21e 3080 }
71be2cbc
PP
3081 *m = '\0';
3082 SvCUR_set(TARG, m - s);
3083 }
2ec7214c
DM
3084 else { /* faster from front */
3085 I32 i = m - s;
71be2cbc 3086 d -= clen;
2ec7214c
DM
3087 if (i > 0)
3088 Move(s, d - i, i, char);
71be2cbc 3089 sv_chop(TARG, d-i);
71be2cbc 3090 if (clen)
c947cd8d 3091 Copy(c, d, clen, char);
71be2cbc 3092 }
8ec5e241 3093 SPAGAIN;
8ca8a454 3094 PUSHs(&PL_sv_yes);
71be2cbc
PP
3095 }
3096 else {
c67ab8f2 3097 char *d, *m;
0395280b 3098 d = s = RX_OFFS(rx)[0].start + orig;
71be2cbc 3099 do {
2b25edcf 3100 I32 i;
5d9574c1 3101 if (UNLIKELY(iters++ > maxiters))
cea2e8a9 3102 DIE(aTHX_ "Substitution loop");
5d9574c1 3103 if (UNLIKELY(RX_MATCH_TAINTED(rx))) /* run time pattern taint, eg locale */
20be6587 3104 rxtainted |= SUBST_TAINT_PAT;
07bc277f 3105 m = RX_OFFS(rx)[0].start + orig;
155aba94 3106 if ((i = m - s)) {
71be2cbc
PP
3107 if (s != d)
3108 Move(s, d, i, char);
3109 d += i;
a0d0e21e 3110 }
71be2cbc
PP
3111 if (clen) {
3112 Copy(c, d, clen, char);
3113 d += clen;
3114 }
07bc277f 3115 s = RX_OFFS(rx)[0].end + orig;
7ce41e5c
FC
3116 } while (CALLREGEXEC(rx, s, strend, orig,
3117 s == m, /* don't match same null twice */
f722798b 3118 TARG, NULL,
d5e7783a 3119 REXEC_NOT_FIRST|REXEC_IGNOREPOS|REXEC_FAIL_ON_UNDERFLOW));
71be2cbc 3120 if (s != d) {
2b25edcf 3121 I32 i = strend - s;
aa07b2f6 3122 SvCUR_set(TARG, d - SvPVX_const(TARG) + i);
71be2cbc 3123 Move(s, d, i+1, char); /* include the NUL */
a0d0e21e 3124 }
8ec5e241 3125 SPAGAIN;
3c6ef0a5 3126 mPUSHi(iters);
a0d0e21e
LW
3127 }
3128 }
ff6e92e8 3129 else {
1754320d 3130 bool first;
c67ab8f2 3131 char *m;
1754320d 3132 SV *repl;
a0d0e21e 3133 if (force_on_match) {
6ac6605d
DM
3134 /* redo the first match, this time with the orig var
3135 * forced into being a string */
a0d0e21e 3136 force_on_match = 0;
0c1438a1
NC
3137 if (rpm->op_pmflags & PMf_NONDESTRUCT) {
3138 /* I feel that it should be possible to avoid this mortal copy
3139 given that the code below copies into a new destination.
3140 However, I suspect it isn't worth the complexity of
3141 unravelling the C<goto force_it> for the small number of
3142 cases where it would be viable to drop into the copy code. */
3143 TARG = sv_2mortal(newSVsv(TARG));
3144 }
6ac6605d 3145 orig = SvPV_force_nomg(TARG, len);
a0d0e21e
LW
3146 goto force_it;
3147 }
db2c6cb3 3148#ifdef PERL_ANY_COW
ed252734
NC
3149 have_a_cow:
3150#endif
20be6587
DM
3151 if (RX_MATCH_TAINTED(rx)) /* run time pattern taint, eg locale */
3152 rxtainted |= SUBST_TAINT_PAT;
1754320d 3153 repl = dstr;
0395280b
DM
3154 s = RX_OFFS(rx)[0].start + orig;
3155 dstr = newSVpvn_flags(orig, s-orig,
3156 SVs_TEMP | (DO_UTF8(TARG) ? SVf_UTF8 : 0));
a0d0e21e 3157 if (!c) {
eb578fdb 3158 PERL_CONTEXT *cx;
8ec5e241 3159 SPAGAIN;
0395280b 3160 m = orig;
20be6587
DM
3161 /* note that a whole bunch of local vars are saved here for
3162 * use by pp_substcont: here's a list of them in case you're
3163 * searching for places in this sub that uses a particular var:
3164 * iters maxiters r_flags oldsave rxtainted orig dstr targ
3165 * s m strend rx once */
490576d1 3166 CX_PUSHSUBST(cx);
20e98b0f 3167 RETURNOP(cPMOP->op_pmreplrootu.op_pmreplroot);
a0d0e21e 3168 }
1754320d 3169 first = TRUE;
a0d0e21e 3170 do {
5d9574c1 3171 if (UNLIKELY(iters++ > maxiters))
cea2e8a9 3172 DIE(aTHX_ "Substitution loop");
5d9574c1 3173 if (UNLIKELY(RX_MATCH_TAINTED(rx)))
20be6587 3174 rxtainted |= SUBST_TAINT_PAT;
07bc277f 3175 if (RX_MATCH_COPIED(rx) && RX_SUBBEG(rx) != orig) {
c67ab8f2
DM
3176 char *old_s = s;
3177 char *old_orig = orig;
6502e081 3178 assert(RX_SUBOFFSET(rx) == 0);
c67ab8f2 3179
07bc277f 3180 orig = RX_SUBBEG(rx);
c67ab8f2
DM
3181 s = orig + (old_s - old_orig);
3182 strend = s + (strend - old_s);
a0d0e21e 3183 }
07bc277f 3184 m = RX_OFFS(rx)[0].start + orig;
64534138 3185 sv_catpvn_nomg_maybeutf8(dstr, s, m - s, DO_UTF8(TARG));
07bc277f 3186 s = RX_OFFS(rx)[0].end + orig;
1754320d
FC
3187 if (first) {
3188 /* replacement already stringified */
3189 if (clen)
64534138 3190 sv_catpvn_nomg_maybeutf8(dstr, c, clen, doutf8);
1754320d
FC
3191 first = FALSE;
3192 }
3193 else {
8df0e7a2 3194 sv_catsv(dstr, repl);
5d9574c1 3195 if (UNLIKELY(SvTAINTED(repl)))
bb933b9b 3196 rxtainted |= SUBST_TAINT_REPL;
1754320d 3197 }
a0d0e21e
LW
3198 if (once)
3199 break;
ff27773b
KW
3200 } while (CALLREGEXEC(rx, s, strend, orig,
3201 s == m, /* Yields minend of 0 or 1 */
d5e7783a
DM
3202 TARG, NULL,
3203 REXEC_NOT_FIRST|REXEC_IGNOREPOS|REXEC_FAIL_ON_UNDERFLOW));
aba224f7 3204 assert(strend >= s);
64534138 3205 sv_catpvn_nomg_maybeutf8(dstr, s, strend - s, DO_UTF8(TARG));
748a9306 3206
8ca8a454
NC
3207 if (rpm->op_pmflags & PMf_NONDESTRUCT) {
3208 /* From here on down we're using the copy, and leaving the original
3209 untouched. */
3210 TARG = dstr;
3211 SPAGAIN;
3212 PUSHs(dstr);
3213 } else {
db2c6cb3 3214#ifdef PERL_ANY_COW
8ca8a454
NC
3215 /* The match may make the string COW. If so, brilliant, because
3216 that's just saved us one malloc, copy and free - the regexp has
3217 donated the old buffer, and we malloc an entirely new one, rather
3218 than the regexp malloc()ing a buffer and copying our original,
3219 only for us to throw it away here during the substitution. */
3220 if (SvIsCOW(TARG)) {
3221 sv_force_normal_flags(TARG, SV_COW_DROP_PV);
3222 } else
ed252734 3223#endif
8ca8a454
NC
3224 {
3225 SvPV_free(TARG);
3226 }
3227 SvPV_set(TARG, SvPVX(dstr));
3228 SvCUR_set(TARG, SvCUR(dstr));
3229 SvLEN_set(TARG, SvLEN(dstr));
64534138 3230 SvFLAGS(TARG) |= SvUTF8(dstr);
8ca8a454 3231 SvPV_set(dstr, NULL);
748a9306 3232
8ca8a454 3233 SPAGAIN;
3c6ef0a5 3234 mPUSHi(iters);
8ca8a454
NC
3235 }
3236 }
3237
3238 if (!(rpm->op_pmflags & PMf_NONDESTRUCT)) {
3239 (void)SvPOK_only_UTF8(TARG);
a0d0e21e 3240 }
20be6587 3241
ef07e810 3242 /* See "how taint works" above */
284167a5 3243 if (TAINTING_get) {
20be6587
DM
3244 if ((rxtainted & SUBST_TAINT_PAT) ||
3245 ((rxtainted & (SUBST_TAINT_STR|SUBST_TAINT_RETAINT)) ==
3246 (SUBST_TAINT_STR|SUBST_TAINT_RETAINT))
3247 )
3248 (RX_MATCH_TAINTED_on(rx)); /* taint $1 et al */
3249
3250 if (!(rxtainted & SUBST_TAINT_BOOLRET)
3251 && (rxtainted & (SUBST_TAINT_STR|SUBST_TAINT_PAT))
3252 )
3253 SvTAINTED_on(TOPs); /* taint return value */
3254 else
3255 SvTAINTED_off(TOPs); /* may have got tainted earlier */
3256
3257 /* needed for mg_set below */
284167a5
SM
3258 TAINT_set(
3259 cBOOL(rxtainted & (SUBST_TAINT_STR|SUBST_TAINT_PAT|SUBST_TAINT_REPL))
3260 );
20be6587
DM
3261 SvTAINT(TARG);
3262 }
3263 SvSETMAGIC(TARG); /* PL_tainted must be correctly set for this mg_set */
3264 TAINT_NOT;
f1a76097
DM
3265 LEAVE_SCOPE(oldsave);
3266 RETURN;
a0d0e21e
LW
3267}
3268
3269PP(pp_grepwhile)
3270{
20b7effb 3271 dSP;
a0d0e21e
LW
3272
3273 if (SvTRUEx(POPs))
3280af22
NIS
3274 PL_stack_base[PL_markstack_ptr[-1]++] = PL_stack_base[*PL_markstack_ptr];
3275 ++*PL_markstack_ptr;
b2a2a901 3276 FREETMPS;
d343c3ef 3277 LEAVE_with_name("grep_item"); /* exit inner scope */
a0d0e21e
LW
3278
3279 /* All done yet? */
5d9574c1 3280 if (UNLIKELY(PL_stack_base + *PL_markstack_ptr > SP)) {
a0d0e21e 3281 I32 items;
1c23e2bd 3282 const U8 gimme = GIMME_V;
a0d0e21e 3283
d343c3ef 3284 LEAVE_with_name("grep"); /* exit outer scope */
a0d0e21e 3285 (void)POPMARK; /* pop src */
3280af22 3286 items = --*PL_markstack_ptr - PL_markstack_ptr[-1];
a0d0e21e 3287 (void)POPMARK; /* pop dst */
3280af22 3288 SP = PL_stack_base + POPMARK; /* pop original mark */
54310121 3289 if (gimme == G_SCALAR) {
7cc47870
RGS
3290 dTARGET;
3291 XPUSHi(items);
a0d0e21e 3292 }
54310121
PP
3293 else if (gimme == G_ARRAY)
3294 SP += items;
a0d0e21e
LW
3295 RETURN;
3296 }
3297 else {
3298 SV *src;
3299
d343c3ef 3300 ENTER_with_name("grep_item"); /* enter inner scope */
1d7c1841 3301 SAVEVPTR(PL_curpm);
a0d0e21e 3302
6cae08a8 3303 src = PL_stack_base[TOPMARK];
60779a30 3304 if (SvPADTMP(src)) {
6cae08a8 3305 src = PL_stack_base[TOPMARK] = sv_mortalcopy(src);
a0ed822e
FC
3306 PL_tmps_floor++;
3307 }
a0d0e21e 3308 SvTEMP_off(src);
ffd49c98 3309 DEFSV_set(src);
a0d0e21e
LW
3310
3311 RETURNOP(cLOGOP->op_other);
3312 }
3313}
3314
799da9d7 3315/* leave_adjust_stacks():
f7a874b8 3316 *
e02ce34b
DM
3317 * Process a scope's return args (in the range from_sp+1 .. PL_stack_sp),
3318 * positioning them at to_sp+1 onwards, and do the equivalent of a
3319 * FREEMPS and TAINT_NOT.
3320 *
f7a874b8
DM
3321 * Not intended to be called in void context.
3322 *
799da9d7
DM
3323 * When leaving a sub, eval, do{} or other scope, the things that need
3324 * doing to process the return args are:
f7a874b8 3325 * * in scalar context, only return the last arg (or PL_sv_undef if none);
799da9d7
DM
3326 * * for the types of return that return copies of their args (such
3327 * as rvalue sub return), make a mortal copy of every return arg,
3328 * except where we can optimise the copy away without it being
3329 * semantically visible;
3330 * * make sure that the arg isn't prematurely freed; in the case of an
3331 * arg not copied, this may involve mortalising it. For example, in
f7a874b8
DM
3332 * C<sub f { my $x = ...; $x }>, $x would be freed when we do
3333 * CX_LEAVE_SCOPE(cx) unless it's protected or copied.
3334 *
799da9d7
DM
3335 * What condition to use when deciding whether to pass the arg through
3336 * or make a copy, is determined by the 'pass' arg; its valid values are:
3337 * 0: rvalue sub/eval exit
3338 * 1: other rvalue scope exit
3339 * 2: :lvalue sub exit in rvalue context
3340 * 3: :lvalue sub exit in lvalue context and other lvalue scope exits
3341 *
f7a874b8 3342 * There is a big issue with doing a FREETMPS. We would like to free any
799da9d7 3343 * temps created by the last statement which the sub executed, rather than
f7a874b8
DM
3344 * leaving them for the caller. In a situation where a sub call isn't
3345 * soon followed by a nextstate (e.g. nested recursive calls, a la
3346 * fibonacci()), temps can accumulate, causing memory and performance
3347 * issues.
3348 *
3349 * On the other hand, we don't want to free any TEMPs which are keeping
799da9d7
DM
3350 * alive any return args that we skipped copying; nor do we wish to undo
3351 * any mortalising done here.
f7a874b8
DM
3352 *
3353 * The solution is to split the temps stack frame into two, with a cut
3354 * point delineating the two halves. We arrange that by the end of this
3355 * function, all the temps stack frame entries we wish to keep are in the
799da9d7 3356 * range PL_tmps_floor+1.. tmps_base-1, while the ones to free now are in
f7a874b8
DM
3357 * the range tmps_base .. PL_tmps_ix. During the course of this
3358 * function, tmps_base starts off as PL_tmps_floor+1, then increases
3359 * whenever we find or create a temp that we know should be kept. In
3360 * general the stuff above tmps_base is undecided until we reach the end,
3361 * and we may need a sort stage for that.
3362 *
3363 * To determine whether a TEMP is keeping a return arg alive, every
3364 * arg that is kept rather than copied and which has the SvTEMP flag
3365 * set, has the flag temporarily unset, to mark it. At the end we scan
799da9d7 3366 * the temps stack frame above the cut for entries without SvTEMP and
f7a874b8 3367 * keep them, while turning SvTEMP on again. Note that if we die before
799da9d7 3368 * the SvTEMPs flags are set again, its safe: at worst, subsequent use of
f7a874b8
DM
3369 * those SVs may be slightly less efficient.
3370 *
3371 * In practice various optimisations for some common cases mean we can
3372 * avoid most of the scanning and swapping about with the temps stack.
3373 */
3374
799da9d7 3375void
1c23e2bd 3376Perl_leave_adjust_stacks(pTHX_ SV **from_sp, SV **to_sp, U8 gimme, int pass)
a0d0e21e 3377{
263e0548 3378 dVAR;
20b7effb 3379 dSP;
f7a874b8
DM
3380 SSize_t tmps_base; /* lowest index into tmps stack that needs freeing now */
3381 SSize_t nargs;
3382
799da9d7
DM
3383 PERL_ARGS_ASSERT_LEAVE_ADJUST_STACKS;
3384
f7a874b8
DM
3385 TAINT_NOT;
3386
3387 if (gimme == G_ARRAY) {
e02ce34b
DM
3388 nargs = SP - from_sp;
3389 from_sp++;
f7a874b8
DM
3390 }
3391 else {
3392 assert(gimme == G_SCALAR);
e02ce34b 3393 if (UNLIKELY(from_sp >= SP)) {
f7a874b8 3394 /* no return args */
e02ce34b 3395 assert(from_sp == SP);
f7a874b8
DM
3396 EXTEND(SP, 1);
3397 *++SP = &PL_sv_undef;
e02ce34b 3398 to_sp = SP;
f7a874b8
DM
3399 nargs = 0;
3400 }
3401 else {
3402 from_sp = SP;
3403 nargs = 1;
3404 }
3405 }
3406
3407 /* common code for G_SCALAR and G_ARRAY */
3408
3409 tmps_base = PL_tmps_floor + 1;
3410
3411 assert(nargs >= 0);
3412 if (nargs) {
3413 /* pointer version of tmps_base. Not safe across temp stack
3414 * reallocs. */
3415 SV **tmps_basep;
3416
3417 EXTEND_MORTAL(nargs); /* one big extend for worst-case scenario */
3418 tmps_basep = PL_tmps_stack + tmps_base;
f7a874b8
DM
3419
3420 /* process each return arg */
3421
3422 do {
3423 SV *sv = *from_sp++;
3424
3425 assert(PL_tmps_ix + nargs < PL_tmps_max);
3645bb38
DM
3426#ifdef DEBUGGING
3427 /* PADTMPs with container set magic shouldn't appear in the
3428 * wild. This assert is more important for pp_leavesublv(),
3429 * but by testing for it here, we're more likely to catch
3430 * bad cases (what with :lvalue subs not being widely
3431 * deployed). The two issues are that for something like
3432 * sub :lvalue { $tied{foo} }
3433 * or
3434 * sub :lvalue { substr($foo,1,2) }
3435 * pp_leavesublv() will croak if the sub returns a PADTMP,
3436 * and currently functions like pp_substr() return a mortal
3437 * rather than using their PADTMP when returning a PVLV.
3438 * This is because the PVLV will hold a ref to $foo,
3439 * so $foo would get delayed in being freed while
3440 * the PADTMP SV remained in the PAD.
3441 * So if this assert fails it means either:
3442 * 1) there is pp code similar to pp_substr that is
3443 * returning a PADTMP instead of a mortal, and probably
3444 * needs fixing, or
5d9c1c9a 3445 * 2) pp_leavesublv is making unwarranted assumptions
3645bb38
DM
3446 * about always croaking on a PADTMP
3447 */
3448 if (SvPADTMP(sv) && SvSMAGICAL(sv)) {
3449 MAGIC *mg;
3450 for (mg = SvMAGIC(sv); mg; mg = mg->mg_moremagic) {
3451 assert(PERL_MAGIC_TYPE_IS_VALUE_MAGIC(mg->mg_type));
3452 }
3453 }
3454#endif
f7a874b8 3455
799da9d7
DM
3456 if (
3457 pass == 0 ? (SvTEMP(sv) && !SvMAGICAL(sv) && SvREFCNT(sv) == 1)
3458 : pass == 1 ? ((SvTEMP(sv) || SvPADTMP(sv)) && !SvMAGICAL(sv) && SvREFCNT(sv) == 1)
3459 : pass == 2 ? (!SvPADTMP(sv))
3460 : 1)
3461 {
3462 /* pass through: skip copy for logic or optimisation
3463 * reasons; instead mortalise it, except that ... */
e02ce34b 3464 *++to_sp = sv;
f7a874b8 3465
799da9d7
DM
3466 if (SvTEMP(sv)) {
3467 /* ... since this SV is an SvTEMP , we don't need to
3468 * re-mortalise it; instead we just need to ensure
3469 * that its existing entry in the temps stack frame
3470 * ends up below the cut and so avoids being freed
3471 * this time round. We mark it as needing to be kept
3472 * by temporarily unsetting SvTEMP; then at the end,
3473 * we shuffle any !SvTEMP entries on the tmps stack
3474 * back below the cut.
3475 * However, there's a significant chance that there's
3476 * a 1:1 correspondence between the first few (or all)
3477 * elements in the return args stack frame and those
3478 * in the temps stack frame; e,g.:
3479 * sub f { ....; map {...} .... },
3480 * or if we're exiting multiple scopes and one of the
3481 * inner scopes has already made mortal copies of each
3482 * return arg.
3483 *
3484 * If so, this arg sv will correspond to the next item
3485 * on the tmps stack above the cut, and so can be kept
3486 * merely by moving the cut boundary up one, rather
3487 * than messing with SvTEMP. If all args are 1:1 then
3488 * we can avoid the sorting stage below completely.
977d0c81
DM
3489 *
3490 * If there are no items above the cut on the tmps
3491 * stack, then the SvTEMP must comne from an item
3492 * below the cut, so there's nothing to do.
799da9d7 3493 */
977d0c81
DM
3494 if (tmps_basep <= &PL_tmps_stack[PL_tmps_ix]) {
3495 if (sv == *tmps_basep)
3496 tmps_basep++;
3497 else
3498 SvTEMP_off(sv);
3499 }
799da9d7 3500 }
75bc488d 3501 else if (!SvPADTMP(sv)) {
799da9d7 3502 /* mortalise arg to avoid it being freed during save
75bc488d 3503 * stack unwinding. Pad tmps don't need mortalising as
977d0c81
DM
3504 * they're never freed. This is the equivalent of
3505 * sv_2mortal(SvREFCNT_inc(sv)), except that:
799da9d7
DM
3506 * * it assumes that the temps stack has already been
3507 * extended;
3508 * * it puts the new item at the cut rather than at
3509 * ++PL_tmps_ix, moving the previous occupant there
3510 * instead.
3511 */
3512 if (!SvIMMORTAL(sv)) {
977d0c81 3513 SvREFCNT_inc_simple_void_NN(sv);
799da9d7 3514 SvTEMP_on(sv);
977d0c81
DM
3515 /* Note that if there's nothing above the cut,
3516 * this copies the garbage one slot above
3517 * PL_tmps_ix onto itself. This is harmless (the
3518 * stack's already been extended), but might in
3519 * theory trigger warnings from tools like ASan
3520 */
799da9d7
DM
3521 PL_tmps_stack[++PL_tmps_ix] = *tmps_basep;
3522 *tmps_basep++ = sv;
3523 }
3524 }
f7a874b8
DM
3525 }
3526 else {
3527 /* Make a mortal copy of the SV.
3528 * The following code is the equivalent of sv_mortalcopy()
3529 * except that:
3530 * * it assumes the temps stack has already been extended;
3531 * * it optimises the copying for some simple SV types;
3532 * * it puts the new item at the cut rather than at
3533 * ++PL_tmps_ix, moving the previous occupant there
3534 * instead.
3535 */
3536 SV *newsv = newSV(0);
3537
3538 PL_tmps_stack[++PL_tmps_ix] = *tmps_basep;
3539 /* put it on the tmps stack early so it gets freed if we die */
3540 *tmps_basep++ = newsv;
e02ce34b 3541 *++to_sp = newsv;
f7a874b8
DM
3542
3543 if (SvTYPE(sv) <= SVt_IV) {
3544 /* arg must be one of undef, IV/UV, or RV: skip
3545 * sv_setsv_flags() and do the copy directly */
3546 U32 dstflags;
3547 U32 srcflags = SvFLAGS(sv);
3548
3549 assert(!SvGMAGICAL(sv));
3550 if (srcflags & (SVf_IOK|SVf_ROK)) {
3551 SET_SVANY_FOR_BODYLESS_IV(newsv);
3552
3553 if (srcflags & SVf_ROK) {
3554 newsv->sv_u.svu_rv = SvREFCNT_inc(SvRV(sv));
3555 /* SV type plus flags */
3556 dstflags = (SVt_IV|SVf_ROK|SVs_TEMP);
3557 }
3558 else {
3559 /* both src and dst are <= SVt_IV, so sv_any
3560 * points to the head; so access the heads
3561 * directly rather than going via sv_any.
3562 */
3563 assert( &(sv->sv_u.svu_iv)
3564 == &(((XPVIV*) SvANY(sv))->xiv_iv));
3565 assert( &(newsv->sv_u.svu_iv)
3566 == &(((XPVIV*) SvANY(newsv))->xiv_iv));
3567 newsv->sv_u.svu_iv = sv->sv_u.svu_iv;
3568 /* SV type plus flags */
3569 dstflags = (SVt_IV|SVf_IOK|SVp_IOK|SVs_TEMP
3570 |(srcflags & SVf_IVisUV));
3571 }
3572 }
3573 else {
3574 assert(!(srcflags & SVf_OK));
3575 dstflags = (SVt_NULL|SVs_TEMP); /* SV type plus flags */
3576 }
3577 SvFLAGS(newsv) = dstflags;
3578
3579 }
3580 else {
3581 /* do the full sv_setsv() */
3582 SSize_t old_base;
3583
3584 SvTEMP_on(newsv);
3585 old_base = tmps_basep - PL_tmps_stack;
3586 SvGETMAGIC(sv);
3587 sv_setsv_flags(newsv, sv, SV_DO_COW_SVSETSV);
799da9d7 3588 /* the mg_get or sv_setsv might have created new temps
f7a874b8
DM
3589 * or realloced the tmps stack; regrow and reload */
3590 EXTEND_MORTAL(nargs);
3591 tmps_basep = PL_tmps_stack + old_base;
3592 TAINT_NOT; /* Each item is independent */
3593 }
3594
3595 }
3596 } while (--nargs);
3597
3598 /* If there are any temps left above the cut, we need to sort