This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
bench.pl: fix --sort and --compact options
[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
354eabfa 133 if (PL_op->op_private & OPpASSIGN_BACKWARDS) { /* {or,and,dor}assign */
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 . $r */
289 && ckWARN(WARN_UNINITIALIZED)
290 )
291 report_uninit(left);
adf14ec6 292 SvPVCLEAR(left);
c75ab21a 293 }
18ea7bf2
SM
294 else {
295 SvPV_force_nomg_nolen(left);
296 }
583a5589 297 lbyte = !DO_UTF8(left);
90f5826e 298 if (IN_BYTES)
18ea7bf2 299 SvUTF8_off(left);
8d6d96c1 300 }
a12c0f56 301
c75ab21a 302 if (!rcopied) {
6f1401dc 303 rpv = SvPV_nomg_const(right, rlen);
c75ab21a
RH
304 rbyte = !DO_UTF8(right);
305 }
8d6d96c1
HS
306 if (lbyte != rbyte) {
307 if (lbyte)
308 sv_utf8_upgrade_nomg(TARG);
309 else {
db79b45b 310 if (!rcopied)
59cd0e26 311 right = newSVpvn_flags(rpv, rlen, SVs_TEMP);
8d6d96c1 312 sv_utf8_upgrade_nomg(right);
6f1401dc 313 rpv = SvPV_nomg_const(right, rlen);
69b47968 314 }
a0d0e21e 315 }
8d6d96c1 316 sv_catpvn_nomg(TARG, rpv, rlen);
43ebc500 317
a0d0e21e
LW
318 SETTARG;
319 RETURN;
748a9306 320 }
a0d0e21e
LW
321}
322
d5524600
DM
323/* push the elements of av onto the stack.
324 * XXX Note that padav has similar code but without the mg_get().
325 * I suspect that the mg_get is no longer needed, but while padav
326 * differs, it can't share this function */
327
f9ae8fb6 328STATIC void
d5524600
DM
329S_pushav(pTHX_ AV* const av)
330{
331 dSP;
c70927a6 332 const SSize_t maxarg = AvFILL(av) + 1;
d5524600 333 EXTEND(SP, maxarg);
5d9574c1 334 if (UNLIKELY(SvRMAGICAL(av))) {
c70927a6
FC
335 PADOFFSET i;
336 for (i=0; i < (PADOFFSET)maxarg; i++) {
d5524600
DM
337 SV ** const svp = av_fetch(av, i, FALSE);
338 /* See note in pp_helem, and bug id #27839 */
339 SP[i+1] = svp
340 ? SvGMAGICAL(*svp) ? (mg_get(*svp), *svp) : *svp
341 : &PL_sv_undef;
342 }
343 }
344 else {
c70927a6
FC
345 PADOFFSET i;
346 for (i=0; i < (PADOFFSET)maxarg; i++) {
ce0d59fd 347 SV * const sv = AvARRAY(av)[i];
5d9574c1 348 SP[i+1] = LIKELY(sv) ? sv : &PL_sv_undef;
ce0d59fd 349 }
d5524600
DM
350 }
351 SP += maxarg;
352 PUTBACK;
353}
354
355
a7fd8ef6
DM
356/* ($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
a0d0e21e
LW
863/* Oversized hot code. */
864
b1c05ba5
DM
865/* also used for: pp_say() */
866
a0d0e21e
LW
867PP(pp_print)
868{
20b7effb 869 dSP; dMARK; dORIGMARK;
eb578fdb 870 PerlIO *fp;
236988e4 871 MAGIC *mg;
159b6efe
NC
872 GV * const gv
873 = (PL_op->op_flags & OPf_STACKED) ? MUTABLE_GV(*++MARK) : PL_defoutgv;
9c9f25b8 874 IO *io = GvIO(gv);
5b468f54 875
9c9f25b8 876 if (io
ad64d0ec 877 && (mg = SvTIED_mg((const SV *)io, PERL_MAGIC_tiedscalar)))
5b468f54 878 {
01bb7c6d 879 had_magic:
68dc0745 880 if (MARK == ORIGMARK) {
1c846c1f 881 /* If using default handle then we need to make space to
a60c0954
NIS
882 * pass object as 1st arg, so move other args up ...
883 */
4352c267 884 MEXTEND(SP, 1);
68dc0745
PP
885 ++MARK;
886 Move(MARK, MARK + 1, (SP - MARK) + 1, SV*);
887 ++SP;
888 }
3e0cb5de 889 return Perl_tied_method(aTHX_ SV_CONST(PRINT), mark - 1, MUTABLE_SV(io),
94bc412f
NC
890 mg,
891 (G_SCALAR | TIED_METHOD_ARGUMENTS_ON_STACK
892 | (PL_op->op_type == OP_SAY
893 ? TIED_METHOD_SAY : 0)), sp - mark);
236988e4 894 }
9c9f25b8 895 if (!io) {
68b590d9 896 if ( gv && GvEGVx(gv) && (io = GvIO(GvEGV(gv)))
ad64d0ec 897 && (mg = SvTIED_mg((const SV *)io, PERL_MAGIC_tiedscalar)))
01bb7c6d 898 goto had_magic;
51087808 899 report_evil_fh(gv);
93189314 900 SETERRNO(EBADF,RMS_IFI);
a0d0e21e
LW
901 goto just_say_no;
902 }
903 else if (!(fp = IoOFP(io))) {
7716c5c5
NC
904 if (IoIFP(io))
905 report_wrongway_fh(gv, '<');
51087808 906 else
7716c5c5 907 report_evil_fh(gv);
93189314 908 SETERRNO(EBADF,IoIFP(io)?RMS_FAC:RMS_IFI);
a0d0e21e
LW
909 goto just_say_no;
910 }
911 else {
e23d9e2f 912 SV * const ofs = GvSV(PL_ofsgv); /* $, */
a0d0e21e 913 MARK++;
e23d9e2f 914 if (ofs && (SvGMAGICAL(ofs) || SvOK(ofs))) {
a0d0e21e
LW
915 while (MARK <= SP) {
916 if (!do_print(*MARK, fp))
917 break;
918 MARK++;
919 if (MARK <= SP) {
e23d9e2f
CS
920 /* don't use 'ofs' here - it may be invalidated by magic callbacks */
921 if (!do_print(GvSV(PL_ofsgv), fp)) {
a0d0e21e
LW
922 MARK--;
923 break;
924 }
925 }
926 }
927 }
928 else {
929 while (MARK <= SP) {
930 if (!do_print(*MARK, fp))
931 break;
932 MARK++;
933 }
934 }
935 if (MARK <= SP)
936 goto just_say_no;
937 else {
cfc4a7da
GA
938 if (PL_op->op_type == OP_SAY) {
939 if (PerlIO_write(fp, "\n", 1) == 0 || PerlIO_error(fp))
940 goto just_say_no;
941 }
942 else if (PL_ors_sv && SvOK(PL_ors_sv))
7889fe52 943 if (!do_print(PL_ors_sv, fp)) /* $\ */
a0d0e21e
LW
944 goto just_say_no;
945
946 if (IoFLAGS(io) & IOf_FLUSH)
760ac839 947 if (PerlIO_flush(fp) == EOF)
a0d0e21e
LW
948 goto just_say_no;
949 }
950 }
951 SP = ORIGMARK;
e52fd6f4 952 XPUSHs(&PL_sv_yes);
a0d0e21e
LW
953 RETURN;
954
955 just_say_no:
956 SP = ORIGMARK;
e52fd6f4 957 XPUSHs(&PL_sv_undef);
a0d0e21e
LW
958 RETURN;
959}
960
b1c05ba5
DM
961
962/* also used for: pp_rv2hv() */
bdaf10a5 963/* also called directly by pp_lvavref */
b1c05ba5 964
a0d0e21e
LW
965PP(pp_rv2av)
966{
20b7effb 967 dSP; dTOPss;
1c23e2bd 968 const U8 gimme = GIMME_V;
13c59d41
MH
969 static const char an_array[] = "an ARRAY";
970 static const char a_hash[] = "a HASH";
bdaf10a5
FC
971 const bool is_pp_rv2av = PL_op->op_type == OP_RV2AV
972 || PL_op->op_type == OP_LVAVREF;
d83b45b8 973 const svtype type = is_pp_rv2av ? SVt_PVAV : SVt_PVHV;
a0d0e21e 974
9026059d 975 SvGETMAGIC(sv);
a0d0e21e 976 if (SvROK(sv)) {
5d9574c1 977 if (UNLIKELY(SvAMAGIC(sv))) {
93d7320b 978 sv = amagic_deref_call(sv, is_pp_rv2av ? to_av_amg : to_hv_amg);
93d7320b 979 }
17ab7946 980 sv = SvRV(sv);
5d9574c1 981 if (UNLIKELY(SvTYPE(sv) != type))
dcbac5bb 982 /* diag_listed_as: Not an ARRAY reference */
13c59d41 983 DIE(aTHX_ "Not %s reference", is_pp_rv2av ? an_array : a_hash);
5d9574c1
DM
984 else if (UNLIKELY(PL_op->op_flags & OPf_MOD
985 && PL_op->op_private & OPpLVAL_INTRO))
3da99855 986 Perl_croak(aTHX_ "%s", PL_no_localize_ref);
a0d0e21e 987 }
5d9574c1 988 else if (UNLIKELY(SvTYPE(sv) != type)) {
67955e0c 989 GV *gv;
1c846c1f 990
6e592b3a 991 if (!isGV_with_GP(sv)) {
13c59d41 992 gv = Perl_softref2xv(aTHX_ sv, is_pp_rv2av ? an_array : a_hash,
dc3c76f8
NC
993 type, &sp);
994 if (!gv)
995 RETURN;
35cd451c
GS
996 }
997 else {
159b6efe 998 gv = MUTABLE_GV(sv);
a0d0e21e 999 }
ad64d0ec 1000 sv = is_pp_rv2av ? MUTABLE_SV(GvAVn(gv)) : MUTABLE_SV(GvHVn(gv));
533c011a 1001 if (PL_op->op_private & OPpLVAL_INTRO)
ad64d0ec 1002 sv = is_pp_rv2av ? MUTABLE_SV(save_ary(gv)) : MUTABLE_SV(save_hash(gv));
9f527363
FC
1003 }
1004 if (PL_op->op_flags & OPf_REF) {
17ab7946 1005 SETs(sv);
a0d0e21e 1006 RETURN;
9f527363 1007 }
5d9574c1 1008 else if (UNLIKELY(PL_op->op_private & OPpMAYBE_LVSUB)) {
40c94d11
FC
1009 const I32 flags = is_lvalue_sub();
1010 if (flags && !(flags & OPpENTERSUB_INARGS)) {
cde874ca 1011 if (gimme != G_ARRAY)
042560a6 1012 goto croak_cant_return;
17ab7946 1013 SETs(sv);
78f9721b 1014 RETURN;
40c94d11 1015 }
a0d0e21e
LW
1016 }
1017
17ab7946 1018 if (is_pp_rv2av) {
502c6561 1019 AV *const av = MUTABLE_AV(sv);
636fe681 1020 /* The guts of pp_rv2av */
96913b52 1021 if (gimme == G_ARRAY) {
d5524600
DM
1022 SP--;
1023 PUTBACK;
1024 S_pushav(aTHX_ av);
1025 SPAGAIN;
1c846c1f 1026 }
96913b52
VP
1027 else if (gimme == G_SCALAR) {
1028 dTARGET;
c70927a6 1029 const SSize_t maxarg = AvFILL(av) + 1;
96913b52 1030 SETi(maxarg);
93965878 1031 }
17ab7946
NC
1032 } else {
1033 /* The guts of pp_rv2hv */
96913b52
VP
1034 if (gimme == G_ARRAY) { /* array wanted */
1035 *PL_stack_sp = sv;
981b7185 1036 return Perl_do_kv(aTHX);
96913b52 1037 }
c8fe3bdf 1038 else if ((PL_op->op_private & OPpTRUEBOOL
adc42c31 1039 || ( PL_op->op_private & OPpMAYBE_TRUEBOOL
c8fe3bdf
FC
1040 && block_gimme() == G_VOID ))
1041 && (!SvRMAGICAL(sv) || !mg_find(sv, PERL_MAGIC_tied)))
1042 SETs(HvUSEDKEYS(sv) ? &PL_sv_yes : sv_2mortal(newSViv(0)));
96913b52 1043 else if (gimme == G_SCALAR) {
1a8bdda9 1044 dTARG;
96913b52 1045 TARG = Perl_hv_scalar(aTHX_ MUTABLE_HV(sv));
96913b52
VP
1046 SETTARG;
1047 }
17ab7946 1048 }
be85d344 1049 RETURN;
042560a6
NC
1050
1051 croak_cant_return:
1052 Perl_croak(aTHX_ "Can't return %s to lvalue scalar context",
1053 is_pp_rv2av ? "array" : "hash");
77e217c6 1054 RETURN;
a0d0e21e
LW
1055}
1056
10c8fecd 1057STATIC void
fb8f4cf8 1058S_do_oddball(pTHX_ SV **oddkey, SV **firstkey)
10c8fecd 1059{
7918f24d
NC
1060 PERL_ARGS_ASSERT_DO_ODDBALL;
1061
fb8f4cf8 1062 if (*oddkey) {
6d822dc4 1063 if (ckWARN(WARN_MISC)) {
a3b680e6 1064 const char *err;
fb8f4cf8
RZ
1065 if (oddkey == firstkey &&
1066 SvROK(*oddkey) &&
1067 (SvTYPE(SvRV(*oddkey)) == SVt_PVAV ||
1068 SvTYPE(SvRV(*oddkey)) == SVt_PVHV))
10c8fecd 1069 {
a3b680e6 1070 err = "Reference found where even-sized list expected";
10c8fecd
GS
1071 }
1072 else
a3b680e6 1073 err = "Odd number of elements in hash assignment";
f1f66076 1074 Perl_warner(aTHX_ packWARN(WARN_MISC), "%s", err);
10c8fecd 1075 }
6d822dc4 1076
10c8fecd
GS
1077 }
1078}
1079
a5f48505
DM
1080
1081/* Do a mark and sweep with the SVf_BREAK flag to detect elements which
1082 * are common to both the LHS and RHS of an aassign, and replace them
1083 * with copies. All these copies are made before the actual list assign is
1084 * done.
1085 *
1086 * For example in ($a,$b) = ($b,$a), assigning the value of the first RHS
1087 * element ($b) to the first LH element ($a), modifies $a; when the
1088 * second assignment is done, the second RH element now has the wrong
1089 * value. So we initially replace the RHS with ($b, mortalcopy($a)).
1090 * Note that we don't need to make a mortal copy of $b.
1091 *
1092 * The algorithm below works by, for every RHS element, mark the
1093 * corresponding LHS target element with SVf_BREAK. Then if the RHS
1094 * element is found with SVf_BREAK set, it means it would have been
1095 * modified, so make a copy.
1096 * Note that by scanning both LHS and RHS in lockstep, we avoid
1097 * unnecessary copies (like $b above) compared with a naive
1098 * "mark all LHS; copy all marked RHS; unmark all LHS".
1099 *
1100 * If the LHS element is a 'my' declaration' and has a refcount of 1, then
1101 * it can't be common and can be skipped.
ebc643ce
DM
1102 *
1103 * On DEBUGGING builds it takes an extra boolean, fake. If true, it means
1104 * that we thought we didn't need to call S_aassign_copy_common(), but we
1105 * have anyway for sanity checking. If we find we need to copy, then panic.
a5f48505
DM
1106 */
1107
1108PERL_STATIC_INLINE void
1109S_aassign_copy_common(pTHX_ SV **firstlelem, SV **lastlelem,
ebc643ce
DM
1110 SV **firstrelem, SV **lastrelem
1111#ifdef DEBUGGING
1112 , bool fake
1113#endif
1114)
a5f48505
DM
1115{
1116 dVAR;
1117 SV **relem;
1118 SV **lelem;
1119 SSize_t lcount = lastlelem - firstlelem + 1;
1120 bool marked = FALSE; /* have we marked any LHS with SVf_BREAK ? */
1121 bool const do_rc1 = cBOOL(PL_op->op_private & OPpASSIGN_COMMON_RC1);
beb08a1e 1122 bool copy_all = FALSE;
a5f48505
DM
1123
1124 assert(!PL_in_clean_all); /* SVf_BREAK not already in use */
1125 assert(firstlelem < lastlelem); /* at least 2 LH elements */
1126 assert(firstrelem < lastrelem); /* at least 2 RH elements */
1127
ebc643ce
DM
1128
1129 lelem = firstlelem;
a5f48505
DM
1130 /* we never have to copy the first RH element; it can't be corrupted
1131 * by assigning something to the corresponding first LH element.
1132 * So this scan does in a loop: mark LHS[N]; test RHS[N+1]
1133 */
ebc643ce 1134 relem = firstrelem + 1;
a5f48505
DM
1135
1136 for (; relem <= lastrelem; relem++) {
1137 SV *svr;
1138
1139 /* mark next LH element */
1140
1141 if (--lcount >= 0) {
1142 SV *svl = *lelem++;
1143
1144 if (UNLIKELY(!svl)) {/* skip AV alias marker */
1145 assert (lelem <= lastlelem);
1146 svl = *lelem++;
1147 lcount--;
1148 }
1149
1150 assert(svl);
beb08a1e
TC
1151 if (SvSMAGICAL(svl)) {
1152 copy_all = TRUE;
1153 }
a5f48505
DM
1154 if (SvTYPE(svl) == SVt_PVAV || SvTYPE(svl) == SVt_PVHV) {
1155 if (!marked)
1156 return;
1157 /* this LH element will consume all further args;
1158 * no need to mark any further LH elements (if any).
1159 * But we still need to scan any remaining RHS elements;
1160 * set lcount negative to distinguish from lcount == 0,
1161 * so the loop condition continues being true
1162 */
1163 lcount = -1;
1164 lelem--; /* no need to unmark this element */
1165 }
1166 else if (!(do_rc1 && SvREFCNT(svl) == 1) && svl != &PL_sv_undef) {
1167 assert(!SvIMMORTAL(svl));
1168 SvFLAGS(svl) |= SVf_BREAK;
1169 marked = TRUE;
1170 }
1171 else if (!marked) {
1172 /* don't check RH element if no SVf_BREAK flags set yet */
1173 if (!lcount)
1174 break;
1175 continue;
1176 }
1177 }
1178
1179 /* see if corresponding RH element needs copying */
1180
1181 assert(marked);
1182 svr = *relem;
1183 assert(svr);
1184
5c1db569 1185 if (UNLIKELY(SvFLAGS(svr) & (SVf_BREAK|SVs_GMG) || copy_all)) {
a5f48505 1186
ebc643ce
DM
1187#ifdef DEBUGGING
1188 if (fake) {
9ae0115f 1189 /* op_dump(PL_op); */
ebc643ce
DM
1190 Perl_croak(aTHX_
1191 "panic: aassign skipped needed copy of common RH elem %"
1192 UVuf, (UV)(relem - firstrelem));
1193 }
1194#endif
1195
a5f48505
DM
1196 TAINT_NOT; /* Each item is independent */
1197
1198 /* Dear TODO test in t/op/sort.t, I love you.
1199 (It's relying on a panic, not a "semi-panic" from newSVsv()
1200 and then an assertion failure below.) */
1201 if (UNLIKELY(SvIS_FREED(svr))) {
1202 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p",
1203 (void*)svr);
1204 }
1205 /* avoid break flag while copying; otherwise COW etc
1206 * disabled... */
1207 SvFLAGS(svr) &= ~SVf_BREAK;
1208 /* Not newSVsv(), as it does not allow copy-on-write,
8c1e192f
DM
1209 resulting in wasteful copies.
1210 Also, we use SV_NOSTEAL in case the SV is used more than
1211 once, e.g. (...) = (f())[0,0]
1212 Where the same SV appears twice on the RHS without a ref
1213 count bump. (Although I suspect that the SV won't be
1214 stealable here anyway - DAPM).
1215 */
a5f48505
DM
1216 *relem = sv_mortalcopy_flags(svr,
1217 SV_GMAGIC|SV_DO_COW_SVSETSV|SV_NOSTEAL);
1218 /* ... but restore afterwards in case it's needed again,
1219 * e.g. ($a,$b,$c) = (1,$a,$a)
1220 */
1221 SvFLAGS(svr) |= SVf_BREAK;
1222 }
1223
1224 if (!lcount)
1225 break;
1226 }
1227
1228 if (!marked)
1229 return;
1230
1231 /*unmark LHS */
1232
1233 while (lelem > firstlelem) {
1234 SV * const svl = *(--lelem);
1235 if (svl)
1236 SvFLAGS(svl) &= ~SVf_BREAK;
1237 }
1238}
1239
1240
1241
a0d0e21e
LW
1242PP(pp_aassign)
1243{
27da23d5 1244 dVAR; dSP;
3280af22
NIS
1245 SV **lastlelem = PL_stack_sp;
1246 SV **lastrelem = PL_stack_base + POPMARK;
1247 SV **firstrelem = PL_stack_base + POPMARK + 1;
a0d0e21e
LW
1248 SV **firstlelem = lastrelem + 1;
1249
eb578fdb
KW
1250 SV **relem;
1251 SV **lelem;
a0d0e21e 1252
eb578fdb
KW
1253 SV *sv;
1254 AV *ary;
a0d0e21e 1255
1c23e2bd 1256 U8 gimme;
a0d0e21e 1257 HV *hash;
c70927a6 1258 SSize_t i;
a0d0e21e 1259 int magic;
a5f48505 1260 U32 lval;
a68090fe
DM
1261 /* PL_delaymagic is restored by JUMPENV_POP on dieing, so we
1262 * only need to save locally, not on the save stack */
1263 U16 old_delaymagic = PL_delaymagic;
ebc643ce
DM
1264#ifdef DEBUGGING
1265 bool fake = 0;
1266#endif
5637b936 1267
3280af22 1268 PL_delaymagic = DM_DELAY; /* catch simultaneous items */
a0d0e21e
LW
1269
1270 /* If there's a common identifier on both sides we have to take
1271 * special care that assigning the identifier on the left doesn't
1272 * clobber a value on the right that's used later in the list.
1273 */
acdea6f0 1274
beb08a1e
TC
1275 /* at least 2 LH and RH elements, or commonality isn't an issue */
1276 if (firstlelem < lastlelem && firstrelem < lastrelem) {
5c1db569
TC
1277 for (relem = firstrelem+1; relem <= lastrelem; relem++) {
1278 if (SvGMAGICAL(*relem))
1279 goto do_scan;
1280 }
beb08a1e
TC
1281 for (lelem = firstlelem; lelem <= lastlelem; lelem++) {
1282 if (*lelem && SvSMAGICAL(*lelem))
1283 goto do_scan;
a5f48505 1284 }
beb08a1e
TC
1285 if ( PL_op->op_private & (OPpASSIGN_COMMON_SCALAR|OPpASSIGN_COMMON_RC1) ) {
1286 if (PL_op->op_private & OPpASSIGN_COMMON_RC1) {
1287 /* skip the scan if all scalars have a ref count of 1 */
1288 for (lelem = firstlelem; lelem <= lastlelem; lelem++) {
1289 sv = *lelem;
1290 if (!sv || SvREFCNT(sv) == 1)
1291 continue;
1292 if (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVAV)
1293 goto do_scan;
1294 break;
1295 }
1296 }
1297 else {
1298 do_scan:
1299 S_aassign_copy_common(aTHX_
1300 firstlelem, lastlelem, firstrelem, lastrelem
ebc643ce 1301#ifdef DEBUGGING
beb08a1e 1302 , fake
ebc643ce 1303#endif
beb08a1e
TC
1304 );
1305 }
a5f48505 1306 }
a0d0e21e 1307 }
ebc643ce
DM
1308#ifdef DEBUGGING
1309 else {
1310 /* on debugging builds, do the scan even if we've concluded we
1311 * don't need to, then panic if we find commonality. Note that the
1312 * scanner assumes at least 2 elements */
1313 if (firstlelem < lastlelem && firstrelem < lastrelem) {
1314 fake = 1;
1315 goto do_scan;
1316 }
1317 }
1318#endif
a0d0e21e 1319
a5f48505
DM
1320 gimme = GIMME_V;
1321 lval = (gimme == G_ARRAY) ? (PL_op->op_flags & OPf_MOD || LVRET) : 0;
1322
a0d0e21e
LW
1323 relem = firstrelem;
1324 lelem = firstlelem;
4608196e
RGS
1325 ary = NULL;
1326 hash = NULL;
10c8fecd 1327
5d9574c1 1328 while (LIKELY(lelem <= lastlelem)) {
bdaf10a5 1329 bool alias = FALSE;
bbce6d69 1330 TAINT_NOT; /* Each item stands on its own, taintwise. */
a0d0e21e 1331 sv = *lelem++;
bdaf10a5
FC
1332 if (UNLIKELY(!sv)) {
1333 alias = TRUE;
1334 sv = *lelem++;
1335 ASSUME(SvTYPE(sv) == SVt_PVAV);
1336 }
a0d0e21e 1337 switch (SvTYPE(sv)) {
a5f48505
DM
1338 case SVt_PVAV: {
1339 bool already_copied = FALSE;
60edcf09 1340 ary = MUTABLE_AV(sv);
748a9306 1341 magic = SvMAGICAL(ary) != 0;
60edcf09
FC
1342 ENTER;
1343 SAVEFREESV(SvREFCNT_inc_simple_NN(sv));
a5f48505
DM
1344
1345 /* We need to clear ary. The is a danger that if we do this,
1346 * elements on the RHS may be prematurely freed, e.g.
1347 * @a = ($a[0]);
1348 * In the case of possible commonality, make a copy of each
1349 * RHS SV *before* clearing the array, and add a reference
1350 * from the tmps stack, so that it doesn't leak on death.
1351 * Otherwise, make a copy of each RHS SV only as we're storing
1352 * it into the array - that way we don't have to worry about
1353 * it being leaked if we die, but don't incur the cost of
1354 * mortalising everything.
1355 */
1356
1357 if ( (PL_op->op_private & OPpASSIGN_COMMON_AGG)
1358 && (relem <= lastrelem)
1359 && (magic || AvFILL(ary) != -1))
1360 {
1361 SV **svp;
1362 EXTEND_MORTAL(lastrelem - relem + 1);
1363 for (svp = relem; svp <= lastrelem; svp++) {
8c1e192f 1364 /* see comment in S_aassign_copy_common about SV_NOSTEAL */
a5f48505
DM
1365 *svp = sv_mortalcopy_flags(*svp,
1366 SV_GMAGIC|SV_DO_COW_SVSETSV|SV_NOSTEAL);
1367 TAINT_NOT;
1368 }
1369 already_copied = TRUE;
1370 }
1371
1372 av_clear(ary);
6768377c
DM
1373 if (relem <= lastrelem)
1374 av_extend(ary, lastrelem - relem);
1375
a0d0e21e
LW
1376 i = 0;
1377 while (relem <= lastrelem) { /* gobble up all the rest */
5117ca91 1378 SV **didstore;
bdaf10a5 1379 if (LIKELY(!alias)) {
a5f48505
DM
1380 if (already_copied)
1381 sv = *relem;
1382 else {
1383 if (LIKELY(*relem))
1384 /* before newSV, in case it dies */
1385 SvGETMAGIC(*relem);
1386 sv = newSV(0);
8c1e192f
DM
1387 /* see comment in S_aassign_copy_common about
1388 * SV_NOSTEAL */
1389 sv_setsv_flags(sv, *relem,
1390 (SV_DO_COW_SVSETSV|SV_NOSTEAL));
a5f48505
DM
1391 *relem = sv;
1392 }
bdaf10a5
FC
1393 }
1394 else {
a5f48505
DM
1395 if (!already_copied)
1396 SvGETMAGIC(*relem);
bdaf10a5
FC
1397 if (!SvROK(*relem))
1398 DIE(aTHX_ "Assigned value is not a reference");
1399 if (SvTYPE(SvRV(*relem)) > SVt_PVLV)
1400 /* diag_listed_as: Assigned value is not %s reference */
1401 DIE(aTHX_
1402 "Assigned value is not a SCALAR reference");
a5f48505 1403 if (lval && !already_copied)
bdaf10a5
FC
1404 *relem = sv_mortalcopy(*relem);
1405 /* XXX else check for weak refs? */
0ad694a7 1406 sv = SvREFCNT_inc_NN(SvRV(*relem));
bdaf10a5
FC
1407 }
1408 relem++;
a5f48505 1409 if (already_copied)
0ad694a7 1410 SvREFCNT_inc_simple_void_NN(sv); /* undo mortal free */
5117ca91
GS
1411 didstore = av_store(ary,i++,sv);
1412 if (magic) {
18024492
FC
1413 if (!didstore)
1414 sv_2mortal(sv);
8ef24240 1415 if (SvSMAGICAL(sv))
fb73857a 1416 mg_set(sv);
5117ca91 1417 }
bbce6d69 1418 TAINT_NOT;
a0d0e21e 1419 }
5d9574c1 1420 if (UNLIKELY(PL_delaymagic & DM_ARRAY_ISA))
ad64d0ec 1421 SvSETMAGIC(MUTABLE_SV(ary));
60edcf09 1422 LEAVE;
a0d0e21e 1423 break;
a5f48505
DM
1424 }
1425
10c8fecd 1426 case SVt_PVHV: { /* normal hash */
a0d0e21e 1427 SV *tmpstr;
1c4ea384
RZ
1428 int odd;
1429 int duplicates = 0;
45960564 1430 SV** topelem = relem;
1c4ea384 1431 SV **firsthashrelem = relem;
a5f48505 1432 bool already_copied = FALSE;
a0d0e21e 1433
60edcf09 1434 hash = MUTABLE_HV(sv);
748a9306 1435 magic = SvMAGICAL(hash) != 0;
1c4ea384
RZ
1436
1437 odd = ((lastrelem - firsthashrelem)&1)? 0 : 1;
5d9574c1 1438 if (UNLIKELY(odd)) {
fb8f4cf8 1439 do_oddball(lastrelem, firsthashrelem);
1d2b3927
HS
1440 /* we have firstlelem to reuse, it's not needed anymore
1441 */
1c4ea384
RZ
1442 *(lastrelem+1) = &PL_sv_undef;
1443 }
1444
60edcf09
FC
1445 ENTER;
1446 SAVEFREESV(SvREFCNT_inc_simple_NN(sv));
a5f48505
DM
1447
1448 /* We need to clear hash. The is a danger that if we do this,
1449 * elements on the RHS may be prematurely freed, e.g.
1450 * %h = (foo => $h{bar});
1451 * In the case of possible commonality, make a copy of each
1452 * RHS SV *before* clearing the hash, and add a reference
1453 * from the tmps stack, so that it doesn't leak on death.
1454 */
1455
1456 if ( (PL_op->op_private & OPpASSIGN_COMMON_AGG)
1457 && (relem <= lastrelem)
1458 && (magic || HvUSEDKEYS(hash)))
1459 {
1460 SV **svp;
1461 EXTEND_MORTAL(lastrelem - relem + 1);
1462 for (svp = relem; svp <= lastrelem; svp++) {
1463 *svp = sv_mortalcopy_flags(*svp,
1464 SV_GMAGIC|SV_DO_COW_SVSETSV|SV_NOSTEAL);
1465 TAINT_NOT;
1466 }
1467 already_copied = TRUE;
1468 }
1469
a0d0e21e 1470 hv_clear(hash);
a5f48505 1471
5d9574c1 1472 while (LIKELY(relem < lastrelem+odd)) { /* gobble up all the rest */
5117ca91 1473 HE *didstore;
1c4ea384 1474 assert(*relem);
632b9d6f
FC
1475 /* Copy the key if aassign is called in lvalue context,
1476 to avoid having the next op modify our rhs. Copy
1477 it also if it is gmagical, lest it make the
1478 hv_store_ent call below croak, leaking the value. */
a5f48505 1479 sv = (lval || SvGMAGICAL(*relem)) && !already_copied
632b9d6f
FC
1480 ? sv_mortalcopy(*relem)
1481 : *relem;
45960564 1482 relem++;
1c4ea384 1483 assert(*relem);
a5f48505
DM
1484 if (already_copied)
1485 tmpstr = *relem++;
1486 else {
1487 SvGETMAGIC(*relem);
1488 tmpstr = newSV(0);
1489 sv_setsv_nomg(tmpstr,*relem++); /* value */
1490 }
1491
a88bf2bc 1492 if (gimme == G_ARRAY) {
45960564
DM
1493 if (hv_exists_ent(hash, sv, 0))
1494 /* key overwrites an existing entry */
1495 duplicates += 2;
a88bf2bc 1496 else {
45960564 1497 /* copy element back: possibly to an earlier
1d2b3927
HS
1498 * stack location if we encountered dups earlier,
1499 * possibly to a later stack location if odd */
45960564
DM
1500 *topelem++ = sv;
1501 *topelem++ = tmpstr;
1502 }
1503 }
a5f48505 1504 if (already_copied)
0ad694a7 1505 SvREFCNT_inc_simple_void_NN(tmpstr); /* undo mortal free */
5117ca91 1506 didstore = hv_store_ent(hash,sv,tmpstr,0);
632b9d6f
FC
1507 if (magic) {
1508 if (!didstore) sv_2mortal(tmpstr);
1509 SvSETMAGIC(tmpstr);
1510 }
bbce6d69 1511 TAINT_NOT;
8e07c86e 1512 }
60edcf09 1513 LEAVE;
1c4ea384
RZ
1514 if (duplicates && gimme == G_ARRAY) {
1515 /* at this point we have removed the duplicate key/value
1516 * pairs from the stack, but the remaining values may be
1517 * wrong; i.e. with (a 1 a 2 b 3) on the stack we've removed
1518 * the (a 2), but the stack now probably contains
1519 * (a <freed> b 3), because { hv_save(a,1); hv_save(a,2) }
1520 * obliterates the earlier key. So refresh all values. */
1521 lastrelem -= duplicates;
1522 relem = firsthashrelem;
1523 while (relem < lastrelem+odd) {
1524 HE *he;
1525 he = hv_fetch_ent(hash, *relem++, 0, 0);
1526 *relem++ = (he ? HeVAL(he) : &PL_sv_undef);
1527 }
1528 }
1529 if (odd && gimme == G_ARRAY) lastrelem++;
a0d0e21e
LW
1530 }
1531 break;
1532 default:
6fc92669
GS
1533 if (SvIMMORTAL(sv)) {
1534 if (relem <= lastrelem)
1535 relem++;
1536 break;
a0d0e21e
LW
1537 }
1538 if (relem <= lastrelem) {
5d9574c1 1539 if (UNLIKELY(
1c70fb82
FC
1540 SvTEMP(sv) && !SvSMAGICAL(sv) && SvREFCNT(sv) == 1 &&
1541 (!isGV_with_GP(sv) || SvFAKE(sv)) && ckWARN(WARN_MISC)
5d9574c1 1542 ))
1c70fb82
FC
1543 Perl_warner(aTHX_
1544 packWARN(WARN_MISC),
1545 "Useless assignment to a temporary"
1546 );
a0d0e21e
LW
1547 sv_setsv(sv, *relem);
1548 *(relem++) = sv;
1549 }
1550 else
3280af22 1551 sv_setsv(sv, &PL_sv_undef);
8ef24240 1552 SvSETMAGIC(sv);
a0d0e21e
LW
1553 break;
1554 }
1555 }
5d9574c1 1556 if (UNLIKELY(PL_delaymagic & ~DM_DELAY)) {
985213f2 1557 /* Will be used to set PL_tainting below */
dfff4baf
BF
1558 Uid_t tmp_uid = PerlProc_getuid();
1559 Uid_t tmp_euid = PerlProc_geteuid();
1560 Gid_t tmp_gid = PerlProc_getgid();
1561 Gid_t tmp_egid = PerlProc_getegid();
985213f2 1562
b469f1e0 1563 /* XXX $> et al currently silently ignore failures */
3280af22 1564 if (PL_delaymagic & DM_UID) {
a0d0e21e 1565#ifdef HAS_SETRESUID
b469f1e0
JH
1566 PERL_UNUSED_RESULT(
1567 setresuid((PL_delaymagic & DM_RUID) ? PL_delaymagic_uid : (Uid_t)-1,
1568 (PL_delaymagic & DM_EUID) ? PL_delaymagic_euid : (Uid_t)-1,
1569 (Uid_t)-1));
56febc5e
AD
1570#else
1571# ifdef HAS_SETREUID
b469f1e0
JH
1572 PERL_UNUSED_RESULT(
1573 setreuid((PL_delaymagic & DM_RUID) ? PL_delaymagic_uid : (Uid_t)-1,
1574 (PL_delaymagic & DM_EUID) ? PL_delaymagic_euid : (Uid_t)-1));
56febc5e
AD
1575# else
1576# ifdef HAS_SETRUID
b28d0864 1577 if ((PL_delaymagic & DM_UID) == DM_RUID) {
b469f1e0 1578 PERL_UNUSED_RESULT(setruid(PL_delaymagic_uid));
b28d0864 1579 PL_delaymagic &= ~DM_RUID;
a0d0e21e 1580 }
56febc5e
AD
1581# endif /* HAS_SETRUID */
1582# ifdef HAS_SETEUID
b28d0864 1583 if ((PL_delaymagic & DM_UID) == DM_EUID) {
b469f1e0 1584 PERL_UNUSED_RESULT(seteuid(PL_delaymagic_euid));
b28d0864 1585 PL_delaymagic &= ~DM_EUID;
a0d0e21e 1586 }
56febc5e 1587# endif /* HAS_SETEUID */
b28d0864 1588 if (PL_delaymagic & DM_UID) {
985213f2 1589 if (PL_delaymagic_uid != PL_delaymagic_euid)
cea2e8a9 1590 DIE(aTHX_ "No setreuid available");
b469f1e0 1591 PERL_UNUSED_RESULT(PerlProc_setuid(PL_delaymagic_uid));
a0d0e21e 1592 }
56febc5e
AD
1593# endif /* HAS_SETREUID */
1594#endif /* HAS_SETRESUID */
04783dc7 1595
985213f2
AB
1596 tmp_uid = PerlProc_getuid();
1597 tmp_euid = PerlProc_geteuid();
a0d0e21e 1598 }
b469f1e0 1599 /* XXX $> et al currently silently ignore failures */
3280af22 1600 if (PL_delaymagic & DM_GID) {
a0d0e21e 1601#ifdef HAS_SETRESGID
b469f1e0
JH
1602 PERL_UNUSED_RESULT(
1603 setresgid((PL_delaymagic & DM_RGID) ? PL_delaymagic_gid : (Gid_t)-1,
1604 (PL_delaymagic & DM_EGID) ? PL_delaymagic_egid : (Gid_t)-1,
1605 (Gid_t)-1));
56febc5e
AD
1606#else
1607# ifdef HAS_SETREGID
b469f1e0
JH
1608 PERL_UNUSED_RESULT(
1609 setregid((PL_delaymagic & DM_RGID) ? PL_delaymagic_gid : (Gid_t)-1,
1610 (PL_delaymagic & DM_EGID) ? PL_delaymagic_egid : (Gid_t)-1));
56febc5e
AD
1611# else
1612# ifdef HAS_SETRGID
b28d0864 1613 if ((PL_delaymagic & DM_GID) == DM_RGID) {
b469f1e0 1614 PERL_UNUSED_RESULT(setrgid(PL_delaymagic_gid));
b28d0864 1615 PL_delaymagic &= ~DM_RGID;
a0d0e21e 1616 }
56febc5e
AD
1617# endif /* HAS_SETRGID */
1618# ifdef HAS_SETEGID
b28d0864 1619 if ((PL_delaymagic & DM_GID) == DM_EGID) {
b469f1e0 1620 PERL_UNUSED_RESULT(setegid(PL_delaymagic_egid));
b28d0864 1621 PL_delaymagic &= ~DM_EGID;
a0d0e21e 1622 }
56febc5e 1623# endif /* HAS_SETEGID */
b28d0864 1624 if (PL_delaymagic & DM_GID) {
985213f2 1625 if (PL_delaymagic_gid != PL_delaymagic_egid)
cea2e8a9 1626 DIE(aTHX_ "No setregid available");
b469f1e0 1627 PERL_UNUSED_RESULT(PerlProc_setgid(PL_delaymagic_gid));
a0d0e21e 1628 }
56febc5e
AD
1629# endif /* HAS_SETREGID */
1630#endif /* HAS_SETRESGID */
04783dc7 1631
985213f2
AB
1632 tmp_gid = PerlProc_getgid();
1633 tmp_egid = PerlProc_getegid();
a0d0e21e 1634 }
284167a5 1635 TAINTING_set( TAINTING_get | (tmp_uid && (tmp_euid != tmp_uid || tmp_egid != tmp_gid)) );
9a9b5ec9
DM
1636#ifdef NO_TAINT_SUPPORT
1637 PERL_UNUSED_VAR(tmp_uid);
1638 PERL_UNUSED_VAR(tmp_euid);
1639 PERL_UNUSED_VAR(tmp_gid);
1640 PERL_UNUSED_VAR(tmp_egid);
1641#endif
a0d0e21e 1642 }
a68090fe 1643 PL_delaymagic = old_delaymagic;
54310121 1644
54310121
PP
1645 if (gimme == G_VOID)
1646 SP = firstrelem - 1;
1647 else if (gimme == G_SCALAR) {
1648 dTARGET;
1649 SP = firstrelem;
231cbeb2 1650 SETi(lastrelem - firstrelem + 1);
54310121
PP
1651 }
1652 else {
1c4ea384 1653 if (ary || hash)
1d2b3927
HS
1654 /* note that in this case *firstlelem may have been overwritten
1655 by sv_undef in the odd hash case */
a0d0e21e 1656 SP = lastrelem;
1c4ea384 1657 else {
a0d0e21e 1658 SP = firstrelem + (lastlelem - firstlelem);
1c4ea384
RZ
1659 lelem = firstlelem + (relem - firstrelem);
1660 while (relem <= SP)
1661 *relem++ = (lelem <= lastlelem) ? *lelem++ : &PL_sv_undef;
1662 }
a0d0e21e 1663 }
08aeb9f7 1664
54310121 1665 RETURN;
a0d0e21e
LW
1666}
1667
8782bef2
GB
1668PP(pp_qr)
1669{
20b7effb 1670 dSP;
eb578fdb 1671 PMOP * const pm = cPMOP;
fe578d7f 1672 REGEXP * rx = PM_GETRE(pm);
10599a69 1673 SV * const pkg = rx ? CALLREG_PACKAGE(rx) : NULL;
c4420975 1674 SV * const rv = sv_newmortal();
d63c20f2
DM
1675 CV **cvp;
1676 CV *cv;
288b8c02
NC
1677
1678 SvUPGRADE(rv, SVt_IV);
c2123ae3
NC
1679 /* For a subroutine describing itself as "This is a hacky workaround" I'm
1680 loathe to use it here, but it seems to be the right fix. Or close.
1681 The key part appears to be that it's essential for pp_qr to return a new
1682 object (SV), which implies that there needs to be an effective way to
1683 generate a new SV from the existing SV that is pre-compiled in the
1684 optree. */
1685 SvRV_set(rv, MUTABLE_SV(reg_temp_copy(NULL, rx)));
288b8c02
NC
1686 SvROK_on(rv);
1687
8d919b0a 1688 cvp = &( ReANY((REGEXP *)SvRV(rv))->qr_anoncv);
5d9574c1 1689 if (UNLIKELY((cv = *cvp) && CvCLONE(*cvp))) {
d63c20f2 1690 *cvp = cv_clone(cv);
fc2b2dca 1691 SvREFCNT_dec_NN(cv);
d63c20f2
DM
1692 }
1693
288b8c02 1694 if (pkg) {
f815daf2 1695 HV *const stash = gv_stashsv(pkg, GV_ADD);
fc2b2dca 1696 SvREFCNT_dec_NN(pkg);
288b8c02
NC
1697 (void)sv_bless(rv, stash);
1698 }
1699
5d9574c1 1700 if (UNLIKELY(RX_ISTAINTED(rx))) {
e08e52cf 1701 SvTAINTED_on(rv);
9274aefd
DM
1702 SvTAINTED_on(SvRV(rv));
1703 }
c8c13c22 1704 XPUSHs(rv);
1705 RETURN;
8782bef2
GB
1706}
1707
a0d0e21e
LW
1708PP(pp_match)
1709{
20b7effb 1710 dSP; dTARG;
eb578fdb 1711 PMOP *pm = cPMOP;
d65afb4b 1712 PMOP *dynpm = pm;
eb578fdb 1713 const char *s;
5c144d81 1714 const char *strend;
99a90e59 1715 SSize_t curpos = 0; /* initial pos() or current $+[0] */
a0d0e21e 1716 I32 global;
7fadf4a7 1717 U8 r_flags = 0;
5c144d81 1718 const char *truebase; /* Start of string */
eb578fdb 1719 REGEXP *rx = PM_GETRE(pm);
b3eb6a9b 1720 bool rxtainted;
1c23e2bd 1721 const U8 gimme = GIMME_V;
a0d0e21e 1722 STRLEN len;
a3b680e6 1723 const I32 oldsave = PL_savestack_ix;
e60df1fa 1724 I32 had_zerolen = 0;
b1422d62 1725 MAGIC *mg = NULL;
a0d0e21e 1726
533c011a 1727 if (PL_op->op_flags & OPf_STACKED)
a0d0e21e 1728 TARG = POPs;
6ffceeb7 1729 else if (ARGTARG)
59f00321 1730 GETTARGET;
a0d0e21e 1731 else {
54b9620d 1732 TARG = DEFSV;
a0d0e21e
LW
1733 EXTEND(SP,1);
1734 }
d9f424b2 1735
c277df42 1736 PUTBACK; /* EVAL blocks need stack_sp. */
69dc4b30
FC
1737 /* Skip get-magic if this is a qr// clone, because regcomp has
1738 already done it. */
f1d31338 1739 truebase = ReANY(rx)->mother_re
69dc4b30
FC
1740 ? SvPV_nomg_const(TARG, len)
1741 : SvPV_const(TARG, len);
f1d31338 1742 if (!truebase)
2269b42e 1743 DIE(aTHX_ "panic: pp_match");
f1d31338 1744 strend = truebase + len;
284167a5
SM
1745 rxtainted = (RX_ISTAINTED(rx) ||
1746 (TAINT_get && (pm->op_pmflags & PMf_RETAINT)));
9212bbba 1747 TAINT_NOT;
a0d0e21e 1748
6c864ec2 1749 /* We need to know this in case we fail out early - pos() must be reset */
de0df3c0
MH
1750 global = dynpm->op_pmflags & PMf_GLOBAL;
1751
d65afb4b 1752 /* PMdf_USED is set after a ?? matches once */
c737faaf
YO
1753 if (
1754#ifdef USE_ITHREADS
1755 SvREADONLY(PL_regex_pad[pm->op_pmoffset])
1756#else
1757 pm->op_pmflags & PMf_USED
1758#endif
1759 ) {
e5dc5375 1760 DEBUG_r(PerlIO_printf(Perl_debug_log, "?? already matched once"));
de0df3c0 1761 goto nope;
a0d0e21e
LW
1762 }
1763
7e313637
FC
1764 /* empty pattern special-cased to use last successful pattern if
1765 possible, except for qr// */
794826f4
YO
1766 if (!ReANY(rx)->mother_re && !RX_PRELEN(rx) && PL_curpm) {
1767 if (PL_curpm == PL_reg_curpm)
31775478 1768 Perl_croak(aTHX_ "Use of the empty pattern inside of "
794826f4 1769 "a regex code block is forbidden");
3280af22 1770 pm = PL_curpm;
aaa362c4 1771 rx = PM_GETRE(pm);
a0d0e21e 1772 }
d65afb4b 1773
389ecb56 1774 if (RX_MINLEN(rx) >= 0 && (STRLEN)RX_MINLEN(rx) > len) {
75d43e96
FC
1775 DEBUG_r(PerlIO_printf(Perl_debug_log, "String shorter than min possible regex match (%"
1776 UVuf" < %"IVdf")\n",
1777 (UV)len, (IV)RX_MINLEN(rx)));
de0df3c0 1778 goto nope;
e5dc5375 1779 }
c277df42 1780
8ef97b0e 1781 /* get pos() if //g */
de0df3c0 1782 if (global) {
b1422d62 1783 mg = mg_find_mglob(TARG);
8ef97b0e 1784 if (mg && mg->mg_len >= 0) {
25fdce4a 1785 curpos = MgBYTEPOS(mg, TARG, truebase, len);
8ef97b0e
DM
1786 /* last time pos() was set, it was zero-length match */
1787 if (mg->mg_flags & MGf_MINMATCH)
1788 had_zerolen = 1;
1789 }
a0d0e21e 1790 }
8ef97b0e 1791
6e240d0b 1792#ifdef PERL_SAWAMPERSAND
a41aa44d 1793 if ( RX_NPARENS(rx)
6502e081 1794 || PL_sawampersand
6502e081 1795 || (RX_EXTFLAGS(rx) & (RXf_EVAL_SEEN|RXf_PMf_KEEPCOPY))
5b0e71e9 1796 || (dynpm->op_pmflags & PMf_KEEPCOPY)
6e240d0b
FC
1797 )
1798#endif
1799 {
6502e081
DM
1800 r_flags |= (REXEC_COPY_STR|REXEC_COPY_SKIP_PRE);
1801 /* in @a =~ /(.)/g, we iterate multiple times, but copy the buffer
1802 * only on the first iteration. Therefore we need to copy $' as well
1803 * as $&, to make the rest of the string available for captures in
1804 * subsequent iterations */
1805 if (! (global && gimme == G_ARRAY))
1806 r_flags |= REXEC_COPY_SKIP_POST;
1807 };
5b0e71e9
DM
1808#ifdef PERL_SAWAMPERSAND
1809 if (dynpm->op_pmflags & PMf_KEEPCOPY)
1810 /* handle KEEPCOPY in pmop but not rx, eg $r=qr/a/; /$r/p */
1811 r_flags &= ~(REXEC_COPY_SKIP_PRE|REXEC_COPY_SKIP_POST);
1812#endif
22e551b9 1813
f1d31338
DM
1814 s = truebase;
1815
d7be1480 1816 play_it_again:
985afbc1 1817 if (global)
03c83e26 1818 s = truebase + curpos;
f722798b 1819
77da2310 1820 if (!CALLREGEXEC(rx, (char*)s, (char *)strend, (char*)truebase,
03c83e26 1821 had_zerolen, TARG, NULL, r_flags))
03b6c93d 1822 goto nope;
77da2310
NC
1823
1824 PL_curpm = pm;
985afbc1 1825 if (dynpm->op_pmflags & PMf_ONCE)
c737faaf 1826#ifdef USE_ITHREADS
77da2310 1827 SvREADONLY_on(PL_regex_pad[dynpm->op_pmoffset]);
c737faaf 1828#else
77da2310 1829 dynpm->op_pmflags |= PMf_USED;
c737faaf 1830#endif
a0d0e21e 1831
72311751
GS
1832 if (rxtainted)
1833 RX_MATCH_TAINTED_on(rx);
1834 TAINT_IF(RX_MATCH_TAINTED(rx));
35c2ccc3
DM
1835
1836 /* update pos */
1837
1838 if (global && (gimme != G_ARRAY || (dynpm->op_pmflags & PMf_CONTINUE))) {
b1422d62 1839 if (!mg)
35c2ccc3 1840 mg = sv_magicext_mglob(TARG);
25fdce4a 1841 MgBYTEPOS_set(mg, TARG, truebase, RX_OFFS(rx)[0].end);
adf51885
DM
1842 if (RX_ZERO_LEN(rx))
1843 mg->mg_flags |= MGf_MINMATCH;
1844 else
1845 mg->mg_flags &= ~MGf_MINMATCH;
35c2ccc3
DM
1846 }
1847
bf9dff51
DM
1848 if ((!RX_NPARENS(rx) && !global) || gimme != G_ARRAY) {
1849 LEAVE_SCOPE(oldsave);
1850 RETPUSHYES;
1851 }
1852
88ab22af
DM
1853 /* push captures on stack */
1854
bf9dff51 1855 {
07bc277f 1856 const I32 nparens = RX_NPARENS(rx);
a3b680e6 1857 I32 i = (global && !nparens) ? 1 : 0;
a0d0e21e 1858
c277df42 1859 SPAGAIN; /* EVAL blocks could move the stack. */
ffc61ed2
JH
1860 EXTEND(SP, nparens + i);
1861 EXTEND_MORTAL(nparens + i);
1862 for (i = !i; i <= nparens; i++) {
a0d0e21e 1863 PUSHs(sv_newmortal());
5d9574c1
DM
1864 if (LIKELY((RX_OFFS(rx)[i].start != -1)
1865 && RX_OFFS(rx)[i].end != -1 ))
1866 {
07bc277f 1867 const I32 len = RX_OFFS(rx)[i].end - RX_OFFS(rx)[i].start;
f1d31338 1868 const char * const s = RX_OFFS(rx)[i].start + truebase;
5d9574c1
DM
1869 if (UNLIKELY(RX_OFFS(rx)[i].end < 0 || RX_OFFS(rx)[i].start < 0
1870 || len < 0 || len > strend - s))
5637ef5b
NC
1871 DIE(aTHX_ "panic: pp_match start/end pointers, i=%ld, "
1872 "start=%ld, end=%ld, s=%p, strend=%p, len=%"UVuf,
1873 (long) i, (long) RX_OFFS(rx)[i].start,
1874 (long)RX_OFFS(rx)[i].end, s, strend, (UV) len);
a0d0e21e 1875 sv_setpvn(*SP, s, len);
cce850e4 1876 if (DO_UTF8(TARG) && is_utf8_string((U8*)s, len))
a197cbdd 1877 SvUTF8_on(*SP);
a0d0e21e
LW
1878 }
1879 }
1880 if (global) {
0e0b3e82 1881 curpos = (UV)RX_OFFS(rx)[0].end;
03c83e26 1882 had_zerolen = RX_ZERO_LEN(rx);
c277df42 1883 PUTBACK; /* EVAL blocks may use stack */
cf93c79d 1884 r_flags |= REXEC_IGNOREPOS | REXEC_NOT_FIRST;
a0d0e21e
LW
1885 goto play_it_again;
1886 }
4633a7c4 1887 LEAVE_SCOPE(oldsave);
a0d0e21e
LW
1888 RETURN;
1889 }
e5964223 1890 NOT_REACHED; /* NOTREACHED */
a0d0e21e 1891
7b52d656 1892 nope:
d65afb4b 1893 if (global && !(dynpm->op_pmflags & PMf_CONTINUE)) {
b1422d62
DM
1894 if (!mg)
1895 mg = mg_find_mglob(TARG);
1896 if (mg)
1897 mg->mg_len = -1;
a0d0e21e 1898 }
4633a7c4 1899 LEAVE_SCOPE(oldsave);
a0d0e21e
LW
1900 if (gimme == G_ARRAY)
1901 RETURN;
1902 RETPUSHNO;
1903}
1904
1905OP *
864dbfa3 1906Perl_do_readline(pTHX)
a0d0e21e 1907{
20b7effb 1908 dSP; dTARGETSTACKED;
eb578fdb 1909 SV *sv;
a0d0e21e
LW
1910 STRLEN tmplen = 0;
1911 STRLEN offset;
760ac839 1912 PerlIO *fp;
eb578fdb
KW
1913 IO * const io = GvIO(PL_last_in_gv);
1914 const I32 type = PL_op->op_type;
1c23e2bd 1915 const U8 gimme = GIMME_V;
a0d0e21e 1916
6136c704 1917 if (io) {
50db69d8 1918 const MAGIC *const mg = SvTIED_mg((const SV *)io, PERL_MAGIC_tiedscalar);
6136c704 1919 if (mg) {
3e0cb5de 1920 Perl_tied_method(aTHX_ SV_CONST(READLINE), SP, MUTABLE_SV(io), mg, gimme, 0);
6136c704 1921 if (gimme == G_SCALAR) {
50db69d8
NC
1922 SPAGAIN;
1923 SvSetSV_nosteal(TARG, TOPs);
1924 SETTARG;
6136c704 1925 }
50db69d8 1926 return NORMAL;
0b7c7b4f 1927 }
e79b0511 1928 }
4608196e 1929 fp = NULL;
a0d0e21e
LW
1930 if (io) {
1931 fp = IoIFP(io);
1932 if (!fp) {
1933 if (IoFLAGS(io) & IOf_ARGV) {
1934 if (IoFLAGS(io) & IOf_START) {
a0d0e21e 1935 IoLINES(io) = 0;
b9f2b683 1936 if (av_tindex(GvAVn(PL_last_in_gv)) < 0) {
1d7c1841 1937 IoFLAGS(io) &= ~IOf_START;
d5eb9a46 1938 do_open6(PL_last_in_gv, "-", 1, NULL, NULL, 0);
4bac9ae4 1939 SvTAINTED_off(GvSVn(PL_last_in_gv)); /* previous tainting irrelevant */
76f68e9b 1940 sv_setpvs(GvSVn(PL_last_in_gv), "-");
3280af22 1941 SvSETMAGIC(GvSV(PL_last_in_gv));
a2008d6d
GS
1942 fp = IoIFP(io);
1943 goto have_fp;
a0d0e21e
LW
1944 }
1945 }
157fb5a1 1946 fp = nextargv(PL_last_in_gv, PL_op->op_flags & OPf_SPECIAL);
a0d0e21e 1947 if (!fp) { /* Note: fp != IoIFP(io) */
3280af22 1948 (void)do_close(PL_last_in_gv, FALSE); /* now it does*/
a0d0e21e
LW
1949 }
1950 }
0d44d22b
NC
1951 else if (type == OP_GLOB)
1952 fp = Perl_start_glob(aTHX_ POPs, io);
a0d0e21e
LW
1953 }
1954 else if (type == OP_GLOB)
1955 SP--;
7716c5c5 1956 else if (IoTYPE(io) == IoTYPE_WRONLY) {
a5390457 1957 report_wrongway_fh(PL_last_in_gv, '>');
a00b5bd3 1958 }
a0d0e21e
LW
1959 }
1960 if (!fp) {
041457d9 1961 if ((!io || !(IoFLAGS(io) & IOf_START))
de7dabb6
TC
1962 && ckWARN(WARN_CLOSED)
1963 && type != OP_GLOB)
041457d9 1964 {
de7dabb6 1965 report_evil_fh(PL_last_in_gv);
3f4520fe 1966 }
54310121 1967 if (gimme == G_SCALAR) {
79628082 1968 /* undef TARG, and push that undefined value */
ba92458f 1969 if (type != OP_RCATLINE) {
aab1202a 1970 sv_setsv(TARG,NULL);
ba92458f 1971 }
a0d0e21e
LW
1972 PUSHTARG;
1973 }
1974 RETURN;
1975 }
a2008d6d 1976 have_fp:
54310121 1977 if (gimme == G_SCALAR) {
a0d0e21e 1978 sv = TARG;
0f722b55
RGS
1979 if (type == OP_RCATLINE && SvGMAGICAL(sv))
1980 mg_get(sv);
48de12d9
RGS
1981 if (SvROK(sv)) {
1982 if (type == OP_RCATLINE)
5668452f 1983 SvPV_force_nomg_nolen(sv);
48de12d9
RGS
1984 else
1985 sv_unref(sv);
1986 }
f7877b28 1987 else if (isGV_with_GP(sv)) {
5668452f 1988 SvPV_force_nomg_nolen(sv);
f7877b28 1989 }
862a34c6 1990 SvUPGRADE(sv, SVt_PV);
a0d0e21e 1991 tmplen = SvLEN(sv); /* remember if already alloced */
e3918bb7 1992 if (!tmplen && !SvREADONLY(sv) && !SvIsCOW(sv)) {
f72e8700
JJ
1993 /* try short-buffering it. Please update t/op/readline.t
1994 * if you change the growth length.
1995 */
1996 Sv_Grow(sv, 80);
1997 }
2b5e58c4
AMS
1998 offset = 0;
1999 if (type == OP_RCATLINE && SvOK(sv)) {
2000 if (!SvPOK(sv)) {
5668452f 2001 SvPV_force_nomg_nolen(sv);
2b5e58c4 2002 }
a0d0e21e 2003 offset = SvCUR(sv);
2b5e58c4 2004 }
a0d0e21e 2005 }
54310121 2006 else {
561b68a9 2007 sv = sv_2mortal(newSV(80));
54310121
PP
2008 offset = 0;
2009 }
fbad3eb5 2010
3887d568
AP
2011 /* This should not be marked tainted if the fp is marked clean */
2012#define MAYBE_TAINT_LINE(io, sv) \
2013 if (!(IoFLAGS(io) & IOf_UNTAINT)) { \
2014 TAINT; \
2015 SvTAINTED_on(sv); \
2016 }
2017
684bef36 2018/* delay EOF state for a snarfed empty file */
fbad3eb5 2019#define SNARF_EOF(gimme,rs,io,sv) \
684bef36 2020 (gimme != G_SCALAR || SvCUR(sv) \
b9fee9ba 2021 || (IoFLAGS(io) & IOf_NOLINE) || !RsSNARF(rs))
fbad3eb5 2022
a0d0e21e 2023 for (;;) {
09e8efcc 2024 PUTBACK;
fbad3eb5 2025 if (!sv_gets(sv, fp, offset)
2d726892
TF
2026 && (type == OP_GLOB
2027 || SNARF_EOF(gimme, PL_rs, io, sv)
2028 || PerlIO_error(fp)))
fbad3eb5 2029 {
760ac839 2030 PerlIO_clearerr(fp);
a0d0e21e 2031 if (IoFLAGS(io) & IOf_ARGV) {
157fb5a1 2032 fp = nextargv(PL_last_in_gv, PL_op->op_flags & OPf_SPECIAL);
a0d0e21e
LW
2033 if (fp)
2034 continue;
3280af22 2035 (void)do_close(PL_last_in_gv, FALSE);
a0d0e21e
LW
2036 }
2037 else if (type == OP_GLOB) {
a2a5de95
NC
2038 if (!do_close(PL_last_in_gv, FALSE)) {
2039 Perl_ck_warner(aTHX_ packWARN(WARN_GLOB),
2040 "glob failed (child exited with status %d%s)",
2041 (int)(STATUS_CURRENT >> 8),
2042 (STATUS_CURRENT & 0x80) ? ", core dumped" : "");
4eb79ab5 2043 }
a0d0e21e 2044 }
54310121 2045 if (gimme == G_SCALAR) {
ba92458f
AE
2046 if (type != OP_RCATLINE) {
2047 SV_CHECK_THINKFIRST_COW_DROP(TARG);
0c34ef67 2048 SvOK_off(TARG);
ba92458f 2049 }
09e8efcc 2050 SPAGAIN;
a0d0e21e
LW
2051 PUSHTARG;
2052 }
3887d568 2053 MAYBE_TAINT_LINE(io, sv);
a0d0e21e
LW
2054 RETURN;
2055 }
3887d568 2056 MAYBE_TAINT_LINE(io, sv);
a0d0e21e 2057 IoLINES(io)++;
b9fee9ba 2058 IoFLAGS(io) |= IOf_NOLINE;
71be2cbc 2059 SvSETMAGIC(sv);
09e8efcc 2060 SPAGAIN;
a0d0e21e 2061 XPUSHs(sv);
a0d0e21e 2062 if (type == OP_GLOB) {
349d4f2f 2063 const char *t1;
45a23732 2064 Stat_t statbuf;
a0d0e21e 2065
3280af22 2066 if (SvCUR(sv) > 0 && SvCUR(PL_rs) > 0) {
6136c704 2067 char * const tmps = SvEND(sv) - 1;
aa07b2f6 2068 if (*tmps == *SvPVX_const(PL_rs)) {
c07a80fd 2069 *tmps = '\0';
b162af07 2070 SvCUR_set(sv, SvCUR(sv) - 1);
c07a80fd
PP
2071 }
2072 }
349d4f2f 2073 for (t1 = SvPVX_const(sv); *t1; t1++)
b51c3e77
CB
2074#ifdef __VMS
2075 if (strchr("*%?", *t1))
2076#else
7ad1e72d 2077 if (strchr("$&*(){}[]'\";\\|?<>~`", *t1))
b51c3e77 2078#endif
a0d0e21e 2079 break;
45a23732 2080 if (*t1 && PerlLIO_lstat(SvPVX_const(sv), &statbuf) < 0) {
a0d0e21e
LW
2081 (void)POPs; /* Unmatched wildcard? Chuck it... */
2082 continue;
2083 }
2d79bf7f 2084 } else if (SvUTF8(sv)) { /* OP_READLINE, OP_RCATLINE */
d4c19fe8
AL
2085 if (ckWARN(WARN_UTF8)) {
2086 const U8 * const s = (const U8*)SvPVX_const(sv) + offset;
2087 const STRLEN len = SvCUR(sv) - offset;
2088 const U8 *f;
2089
2090 if (!is_utf8_string_loc(s, len, &f))
2091 /* Emulate :encoding(utf8) warning in the same case. */
2092 Perl_warner(aTHX_ packWARN(WARN_UTF8),
2093 "utf8 \"\\x%02X\" does not map to Unicode",
2094 f < (U8*)SvEND(sv) ? *f : 0);
2095 }
a0d0e21e 2096 }
54310121 2097 if (gimme == G_ARRAY) {
a0d0e21e 2098 if (SvLEN(sv) - SvCUR(sv) > 20) {
1da4ca5f 2099 SvPV_shrink_to_cur(sv);
a0d0e21e 2100 }
561b68a9 2101 sv = sv_2mortal(newSV(80));
a0d0e21e
LW
2102 continue;
2103 }
54310121 2104 else if (gimme == G_SCALAR && !tmplen && SvLEN(sv) - SvCUR(sv) > 80) {
a0d0e21e 2105 /* try to reclaim a bit of scalar space (only on 1st alloc) */
d5b5861b
NC
2106 const STRLEN new_len
2107 = SvCUR(sv) < 60 ? 80 : SvCUR(sv)+40; /* allow some slop */
1da4ca5f 2108 SvPV_renew(sv, new_len);
a0d0e21e
LW
2109 }
2110 RETURN;
2111 }
2112}
2113
a0d0e21e
LW
2114PP(pp_helem)
2115{
20b7effb 2116 dSP;
760ac839 2117 HE* he;
ae77835f 2118 SV **svp;
c445ea15 2119 SV * const keysv = POPs;
85fbaab2 2120 HV * const hv = MUTABLE_HV(POPs);
a3b680e6
AL
2121 const U32 lval = PL_op->op_flags & OPf_MOD || LVRET;
2122 const U32 defer = PL_op->op_private & OPpLVAL_DEFER;
be6c24e0 2123 SV *sv;
92970b93 2124 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
d30e492c 2125 bool preeminent = TRUE;
a0d0e21e 2126
6dfc73ea
SM
2127 if (SvTYPE(hv) != SVt_PVHV)
2128 RETPUSHUNDEF;
d4c19fe8 2129
92970b93 2130 if (localizing) {
d4c19fe8
AL
2131 MAGIC *mg;
2132 HV *stash;
d30e492c
VP
2133
2134 /* If we can determine whether the element exist,
2135 * Try to preserve the existenceness of a tied hash
2136 * element by using EXISTS and DELETE if possible.
2137 * Fallback to FETCH and STORE otherwise. */
2c5f48c2 2138 if (SvCANEXISTDELETE(hv))
d30e492c 2139 preeminent = hv_exists_ent(hv, keysv, 0);
d4c19fe8 2140 }
d30e492c 2141
5f9d7e2b 2142 he = hv_fetch_ent(hv, keysv, lval && !defer, 0);
d4c19fe8 2143 svp = he ? &HeVAL(he) : NULL;
a0d0e21e 2144 if (lval) {
746f6409 2145 if (!svp || !*svp || *svp == &PL_sv_undef) {
68dc0745
PP
2146 SV* lv;
2147 SV* key2;
2d8e6c8d 2148 if (!defer) {
be2597df 2149 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
2d8e6c8d 2150 }
68dc0745
PP
2151 lv = sv_newmortal();
2152 sv_upgrade(lv, SVt_PVLV);
2153 LvTYPE(lv) = 'y';
6136c704 2154 sv_magic(lv, key2 = newSVsv(keysv), PERL_MAGIC_defelem, NULL, 0);
fc2b2dca 2155 SvREFCNT_dec_NN(key2); /* sv_magic() increments refcount */
0ad694a7 2156 LvTARG(lv) = SvREFCNT_inc_simple_NN(hv);
68dc0745
PP
2157 LvTARGLEN(lv) = 1;
2158 PUSHs(lv);
2159 RETURN;
2160 }
92970b93 2161 if (localizing) {
bfcb3514 2162 if (HvNAME_get(hv) && isGV(*svp))
159b6efe 2163 save_gp(MUTABLE_GV(*svp), !(PL_op->op_flags & OPf_SPECIAL));
47cfc530
VP
2164 else if (preeminent)
2165 save_helem_flags(hv, keysv, svp,
2166 (PL_op->op_flags & OPf_SPECIAL) ? 0 : SAVEf_SETMAGIC);
2167 else
2168 SAVEHDELETE(hv, keysv);
5f05dabc 2169 }
9026059d
GG
2170 else if (PL_op->op_private & OPpDEREF) {
2171 PUSHs(vivify_ref(*svp, PL_op->op_private & OPpDEREF));
2172 RETURN;
2173 }
a0d0e21e 2174 }
746f6409 2175 sv = (svp && *svp ? *svp : &PL_sv_undef);
fd69380d
DM
2176 /* Originally this did a conditional C<sv = sv_mortalcopy(sv)>; this
2177 * was to make C<local $tied{foo} = $tied{foo}> possible.
2178 * However, it seems no longer to be needed for that purpose, and
2179 * introduced a new bug: stuff like C<while ($hash{taintedval} =~ /.../g>
2180 * would loop endlessly since the pos magic is getting set on the
2181 * mortal copy and lost. However, the copy has the effect of
2182 * triggering the get magic, and losing it altogether made things like
2183 * c<$tied{foo};> in void context no longer do get magic, which some
2184 * code relied on. Also, delayed triggering of magic on @+ and friends
2185 * meant the original regex may be out of scope by now. So as a
2186 * compromise, do the get magic here. (The MGf_GSKIP flag will stop it
2187 * being called too many times). */
39cf747a 2188 if (!lval && SvRMAGICAL(hv) && SvGMAGICAL(sv))
fd69380d 2189 mg_get(sv);
be6c24e0 2190 PUSHs(sv);
a0d0e21e
LW
2191 RETURN;
2192}
2193
fedf30e1
DM
2194
2195/* a stripped-down version of Perl_softref2xv() for use by
2196 * pp_multideref(), which doesn't use PL_op->op_flags */
2197
f9db5646 2198STATIC GV *
fedf30e1
DM
2199S_softref2xv_lite(pTHX_ SV *const sv, const char *const what,
2200 const svtype type)
2201{
2202 if (PL_op->op_private & HINT_STRICT_REFS) {
2203 if (SvOK(sv))
2204 Perl_die(aTHX_ PL_no_symref_sv, sv,
2205 (SvPOKp(sv) && SvCUR(sv)>32 ? "..." : ""), what);
2206 else
2207 Perl_die(aTHX_ PL_no_usym, what);
2208 }
2209 if (!SvOK(sv))
2210 Perl_die(aTHX_ PL_no_usym, what);
2211 return gv_fetchsv_nomg(sv, GV_ADD, type);
2212}
2213
2214
79815f56
DM
2215/* Handle one or more aggregate derefs and array/hash indexings, e.g.
2216 * $h->{foo} or $a[0]{$key}[$i] or f()->[1]
fedf30e1
DM
2217 *
2218 * op_aux points to an array of unions of UV / IV / SV* / PADOFFSET.
79815f56
DM
2219 * Each of these either contains a set of actions, or an argument, such as
2220 * an IV to use as an array index, or a lexical var to retrieve.
2221 * Several actions re stored per UV; we keep shifting new actions off the
2222 * one UV, and only reload when it becomes zero.
fedf30e1
DM
2223 */
2224
2225PP(pp_multideref)
2226{
2227 SV *sv = NULL; /* init to avoid spurious 'may be used uninitialized' */
2228 UNOP_AUX_item *items = cUNOP_AUXx(PL_op)->op_aux;
2229 UV actions = items->uv;
2230
2231 assert(actions);
2232 /* this tells find_uninit_var() where we're up to */
2233 PL_multideref_pc = items;
2234
2235 while (1) {
2236 /* there are three main classes of action; the first retrieve
2237 * the initial AV or HV from a variable or the stack; the second
2238 * does the equivalent of an unrolled (/DREFAV, rv2av, aelem),
2239 * the third an unrolled (/DREFHV, rv2hv, helem).
2240 */
2241 switch (actions & MDEREF_ACTION_MASK) {
2242
2243 case MDEREF_reload:
2244 actions = (++items)->uv;
2245 continue;
2246
2247 case MDEREF_AV_padav_aelem: /* $lex[...] */
2248 sv = PAD_SVl((++items)->pad_offset);
2249 goto do_AV_aelem;
2250
2251 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
2252 sv = UNOP_AUX_item_sv(++items);
2253 assert(isGV_with_GP(sv));
2254 sv = (SV*)GvAVn((GV*)sv);
2255 goto do_AV_aelem;
2256
2257 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
2258 {
2259 dSP;
2260 sv = POPs;
2261 PUTBACK;
2262 goto do_AV_rv2av_aelem;
2263 }
2264
2265 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
2266 sv = UNOP_AUX_item_sv(++items);
2267 assert(isGV_with_GP(sv));
2268 sv = GvSVn((GV*)sv);
2269 goto do_AV_vivify_rv2av_aelem;
2270
2271 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
2272 sv = PAD_SVl((++items)->pad_offset);
2273 /* FALLTHROUGH */
2274
2275 do_AV_vivify_rv2av_aelem:
2276 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
2277 /* this is the OPpDEREF action normally found at the end of
2278 * ops like aelem, helem, rv2sv */
2279 sv = vivify_ref(sv, OPpDEREF_AV);
2280 /* FALLTHROUGH */
2281
2282 do_AV_rv2av_aelem:
2283 /* this is basically a copy of pp_rv2av when it just has the
2284 * sKR/1 flags */
2285 SvGETMAGIC(sv);
2286 if (LIKELY(SvROK(sv))) {
2287 if (UNLIKELY(SvAMAGIC(sv))) {
2288 sv = amagic_deref_call(sv, to_av_amg);
2289 }
2290 sv = SvRV(sv);
2291 if (UNLIKELY(SvTYPE(sv) != SVt_PVAV))
2292 DIE(aTHX_ "Not an ARRAY reference");
2293 }
2294 else if (SvTYPE(sv) != SVt_PVAV) {
2295 if (!isGV_with_GP(sv))
2296 sv = (SV*)S_softref2xv_lite(aTHX_ sv, "an ARRAY", SVt_PVAV);
2297 sv = MUTABLE_SV(GvAVn((GV*)sv));
2298 }
2299 /* FALLTHROUGH */
2300
2301 do_AV_aelem:
2302 {
2303 /* retrieve the key; this may be either a lexical or package
2304 * var (whose index/ptr is stored as an item) or a signed
2305 * integer constant stored as an item.
2306 */
2307 SV *elemsv;
2308 IV elem = 0; /* to shut up stupid compiler warnings */
2309
2310
2311 assert(SvTYPE(sv) == SVt_PVAV);
2312
2313 switch (actions & MDEREF_INDEX_MASK) {
2314 case MDEREF_INDEX_none:
2315 goto finish;
2316 case MDEREF_INDEX_const:
2317 elem = (++items)->iv;
2318 break;
2319 case MDEREF_INDEX_padsv:
2320 elemsv = PAD_SVl((++items)->pad_offset);
2321 goto check_elem;
2322 case MDEREF_INDEX_gvsv:
2323 elemsv = UNOP_AUX_item_sv(++items);
2324 assert(isGV_with_GP(elemsv));
2325 elemsv = GvSVn((GV*)elemsv);
2326 check_elem:
2327 if (UNLIKELY(SvROK(elemsv) && !SvGAMAGIC(elemsv)
2328 && ckWARN(WARN_MISC)))
2329 Perl_warner(aTHX_ packWARN(WARN_MISC),
2330 "Use of reference \"%"SVf"\" as array index",
2331 SVfARG(elemsv));
2332 /* the only time that S_find_uninit_var() needs this
2333 * is to determine which index value triggered the
2334 * undef warning. So just update it here. Note that
2335 * since we don't save and restore this var (e.g. for
2336 * tie or overload execution), its value will be
2337 * meaningless apart from just here */
2338 PL_multideref_pc = items;
2339 elem = SvIV(elemsv);
2340 break;
2341 }
2342
2343
2344 /* this is basically a copy of pp_aelem with OPpDEREF skipped */
2345
2346 if (!(actions & MDEREF_FLAG_last)) {
2347 SV** svp = av_fetch((AV*)sv, elem, 1);
2348 if (!svp || ! (sv=*svp))
2349 DIE(aTHX_ PL_no_aelem, elem);
2350 break;
2351 }
2352
2353 if (PL_op->op_private &
2354 (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE))
2355 {
2356 if (PL_op->op_private & OPpMULTIDEREF_EXISTS) {
2357 sv = av_exists((AV*)sv, elem) ? &PL_sv_yes : &PL_sv_no;
2358 }
2359 else {
2360 I32 discard = (GIMME_V == G_VOID) ? G_DISCARD : 0;
2361 sv = av_delete((AV*)sv, elem, discard);
2362 if (discard)
2363 return NORMAL;
2364 if (!sv)
2365 sv = &PL_sv_undef;
2366 }
2367 }
2368 else {
2369 const U32 lval = PL_op->op_flags & OPf_MOD || LVRET;
2370 const U32 defer = PL_op->op_private & OPpLVAL_DEFER;
2371 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
2372 bool preeminent = TRUE;
2373 AV *const av = (AV*)sv;
2374 SV** svp;
2375
2376 if (UNLIKELY(localizing)) {
2377 MAGIC *mg;
2378 HV *stash;
2379
2380 /* If we can determine whether the element exist,
2381 * Try to preserve the existenceness of a tied array
2382 * element by using EXISTS and DELETE if possible.
2383 * Fallback to FETCH and STORE otherwise. */
2384 if (SvCANEXISTDELETE(av))
2385 preeminent = av_exists(av, elem);
2386 }
2387
2388 svp = av_fetch(av, elem, lval && !defer);
2389
2390 if (lval) {
2391 if (!svp || !(sv = *svp)) {
2392 IV len;
2393 if (!defer)
2394 DIE(aTHX_ PL_no_aelem, elem);
2395 len = av_tindex(av);
2396 sv = sv_2mortal(newSVavdefelem(av,
2397 /* Resolve a negative index now, unless it points
2398 * before the beginning of the array, in which
2399 * case record it for error reporting in
2400 * magic_setdefelem. */
2401 elem < 0 && len + elem >= 0
2402 ? len + elem : elem, 1));
2403 }
2404 else {
2405 if (UNLIKELY(localizing)) {
2406 if (preeminent) {
2407 save_aelem(av, elem, svp);
2408 sv = *svp; /* may have changed */
2409 }
2410 else
2411 SAVEADELETE(av, elem);
2412 }
2413 }
2414 }
2415 else {
2416 sv = (svp ? *svp : &PL_sv_undef);
2417 /* see note in pp_helem() */
2418 if (SvRMAGICAL(av) && SvGMAGICAL(sv))
2419 mg_get(sv);
2420 }
2421 }
2422
2423 }
2424 finish:
2425 {
2426 dSP;
2427 XPUSHs(sv);
2428 RETURN;
2429 }
2430 /* NOTREACHED */
2431
2432
2433
2434
2435 case MDEREF_HV_padhv_helem: /* $lex{...} */
2436 sv = PAD_SVl((++items)->pad_offset);
2437 goto do_HV_helem;
2438
2439 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
2440 sv = UNOP_AUX_item_sv(++items);
2441 assert(isGV_with_GP(sv));
2442 sv = (SV*)GvHVn((GV*)sv);
2443 goto do_HV_helem;
2444
2445 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
2446 {
2447 dSP;
2448 sv = POPs;
2449 PUTBACK;
2450 goto do_HV_rv2hv_helem;
2451 }
2452
2453 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
2454 sv = UNOP_AUX_item_sv(++items);
2455 assert(isGV_with_GP(sv));
2456 sv = GvSVn((GV*)sv);
2457 goto do_HV_vivify_rv2hv_helem;
2458
2459 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
2460 sv = PAD_SVl((++items)->pad_offset);
2461 /* FALLTHROUGH */
2462
2463 do_HV_vivify_rv2hv_helem:
2464 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
2465 /* this is the OPpDEREF action normally found at the end of
2466 * ops like aelem, helem, rv2sv */
2467 sv = vivify_ref(sv, OPpDEREF_HV);
2468 /* FALLTHROUGH */
2469
2470 do_HV_rv2hv_helem:
2471 /* this is basically a copy of pp_rv2hv when it just has the
2472 * sKR/1 flags (and pp_rv2hv is aliased to pp_rv2av) */
2473
2474 SvGETMAGIC(sv);
2475 if (LIKELY(SvROK(sv))) {
2476 if (UNLIKELY(SvAMAGIC(sv))) {
2477 sv = amagic_deref_call(sv, to_hv_amg);
2478 }
2479 sv = SvRV(sv);
2480 if (UNLIKELY(SvTYPE(sv) != SVt_PVHV))
2481 DIE(aTHX_ "Not a HASH reference");
2482 }
2483 else if (SvTYPE(sv) != SVt_PVHV) {
2484 if (!isGV_with_GP(sv))
2485 sv = (SV*)S_softref2xv_lite(aTHX_ sv, "a HASH", SVt_PVHV);
2486 sv = MUTABLE_SV(GvHVn((GV*)sv));
2487 }
2488 /* FALLTHROUGH */
2489
2490 do_HV_helem:
2491 {
2492 /* retrieve the key; this may be either a lexical / package
2493 * var or a string constant, whose index/ptr is stored as an
2494 * item
2495 */
2496 SV *keysv = NULL; /* to shut up stupid compiler warnings */
2497
2498 assert(SvTYPE(sv) == SVt_PVHV);
2499
2500 switch (actions & MDEREF_INDEX_MASK) {
2501 case MDEREF_INDEX_none:
2502 goto finish;
2503
2504 case MDEREF_INDEX_const:
2505 keysv = UNOP_AUX_item_sv(++items);
2506 break;
2507
2508 case MDEREF_INDEX_padsv:
2509 keysv = PAD_SVl((++items)->pad_offset);
2510 break;
2511
2512 case MDEREF_INDEX_gvsv:
2513 keysv = UNOP_AUX_item_sv(++items);
2514 keysv = GvSVn((GV*)keysv);
2515 break;
2516 }
2517
2518 /* see comment above about setting this var */
2519 PL_multideref_pc = items;
2520
2521
2522 /* ensure that candidate CONSTs have been HEKified */
2523 assert( ((actions & MDEREF_INDEX_MASK) != MDEREF_INDEX_const)
2524 || SvTYPE(keysv) >= SVt_PVMG
2525 || !SvOK(keysv)
2526 || SvROK(keysv)
2527 || SvIsCOW_shared_hash(keysv));
2528
2529 /* this is basically a copy of pp_helem with OPpDEREF skipped */
2530
2531 if (!(actions & MDEREF_FLAG_last)) {
2532 HE *he = hv_fetch_ent((HV*)sv, keysv, 1, 0);
2533 if (!he || !(sv=HeVAL(he)) || sv == &PL_sv_undef)
2534 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
2535 break;
2536 }
2537
2538 if (PL_op->op_private &
2539 (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE))
2540 {
2541 if (PL_op->op_private & OPpMULTIDEREF_EXISTS) {
2542 sv = hv_exists_ent((HV*)sv, keysv, 0)
2543 ? &PL_sv_yes : &PL_sv_no;
2544 }
2545 else {
2546 I32 discard = (GIMME_V == G_VOID) ? G_DISCARD : 0;
2547 sv = hv_delete_ent((HV*)sv, keysv, discard, 0);
2548 if (discard)
2549 return NORMAL;
2550 if (!sv)
2551 sv = &PL_sv_undef;
2552 }
2553 }
2554 else {
2555 const U32 lval = PL_op->op_flags & OPf_MOD || LVRET;
2556 const U32 defer = PL_op->op_private & OPpLVAL_DEFER;
2557 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
2558 bool preeminent = TRUE;
2559 SV **svp;
2560 HV * const hv = (HV*)sv;
2561 HE* he;
2562
2563 if (UNLIKELY(localizing)) {
2564 MAGIC *mg;
2565 HV *stash;
2566
2567 /* If we can determine whether the element exist,
2568 * Try to preserve the existenceness of a tied hash
2569 * element by using EXISTS and DELETE if possible.
2570 * Fallback to FETCH and STORE otherwise. */
2571 if (SvCANEXISTDELETE(hv))
2572 preeminent = hv_exists_ent(hv, keysv, 0);
2573 }
2574
2575 he = hv_fetch_ent(hv, keysv, lval && !defer, 0);
2576 svp = he ? &HeVAL(he) : NULL;
2577
2578
2579 if (lval) {
2580 if (!svp || !(sv = *svp) || sv == &PL_sv_undef) {
2581 SV* lv;
2582 SV* key2;
2583 if (!defer)
2584 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
2585 lv = sv_newmortal();
2586 sv_upgrade(lv, SVt_PVLV);
2587 LvTYPE(lv) = 'y';
2588 sv_magic(lv, key2 = newSVsv(keysv),
2589 PERL_MAGIC_defelem, NULL, 0);
2590 /* sv_magic() increments refcount */
2591 SvREFCNT_dec_NN(key2);
0ad694a7 2592 LvTARG(lv) = SvREFCNT_inc_simple_NN(hv);
fedf30e1
DM
2593 LvTARGLEN(lv) = 1;
2594 sv = lv;
2595 }
2596 else {
2597 if (localizing) {
2598 if (HvNAME_get(hv) && isGV(sv))
2599 save_gp(MUTABLE_GV(sv),
2600 !(PL_op->op_flags & OPf_SPECIAL));
2601 else if (preeminent) {
2602 save_helem_flags(hv, keysv, svp,
2603 (PL_op->op_flags & OPf_SPECIAL)
2604 ? 0 : SAVEf_SETMAGIC);
2605 sv = *svp; /* may have changed */
2606 }
2607 else
2608 SAVEHDELETE(hv, keysv);
2609 }
2610 }
2611 }
2612 else {
2613 sv = (svp && *svp ? *svp : &PL_sv_undef);
2614 /* see note in pp_helem() */
2615 if (SvRMAGICAL(hv) && SvGMAGICAL(sv))
2616 mg_get(sv);
2617 }
2618 }
2619 goto finish;
2620 }
2621
2622 } /* switch */
2623
2624 actions >>= MDEREF_SHIFT;
2625 } /* while */
2626 /* NOTREACHED */
2627}
2628
2629
a0d0e21e
LW
2630PP(pp_iter)
2631{
eb578fdb 2632 PERL_CONTEXT *cx;
7d6c2cef 2633 SV *oldsv;
1d7c1841 2634 SV **itersvp;
8a1f10dd 2635 SV *retsv;
a0d0e21e 2636
84f05d57
JH
2637 SV *sv;
2638 AV *av;
2639 IV ix;
2640 IV inc;
2641
4ebe6e95 2642 cx = CX_CUR();
1d7c1841 2643 itersvp = CxITERVAR(cx);
4b5c941e 2644 assert(itersvp);
a48ce6be
DM
2645
2646 switch (CxTYPE(cx)) {
17c91640 2647
b552b52c
DM
2648 case CXt_LOOP_LAZYSV: /* string increment */
2649 {
2650 SV* cur = cx->blk_loop.state_u.lazysv.cur;
2651 SV *end = cx->blk_loop.state_u.lazysv.end;
2652 /* If the maximum is !SvOK(), pp_enteriter substitutes PL_sv_no.
2653 It has SvPVX of "" and SvCUR of 0, which is what we want. */
2654 STRLEN maxlen = 0;
2655 const char *max = SvPV_const(end, maxlen);
5d9574c1 2656 if (UNLIKELY(SvNIOK(cur) || SvCUR(cur) > maxlen))
8a1f10dd 2657 goto retno;
b552b52c
DM
2658
2659 oldsv = *itersvp;
6d3ca00e
DM
2660 /* NB: on the first iteration, oldsv will have a ref count of at
2661 * least 2 (one extra from blk_loop.itersave), so the GV or pad
2662 * slot will get localised; on subsequent iterations the RC==1
2663 * optimisation may kick in and the SV will be reused. */
2664 if (oldsv && LIKELY(SvREFCNT(oldsv) == 1 && !SvMAGICAL(oldsv))) {
b552b52c
DM
2665 /* safe to reuse old SV */
2666 sv_setsv(oldsv, cur);
a48ce6be 2667 }
b552b52c
DM
2668 else
2669 {
2670 /* we need a fresh SV every time so that loop body sees a
2671 * completely new SV for closures/references to work as
2672 * they used to */
2673 *itersvp = newSVsv(cur);
6d3ca00e 2674 SvREFCNT_dec(oldsv);
b552b52c
DM
2675 }
2676 if (strEQ(SvPVX_const(cur), max))
2677 sv_setiv(cur, 0); /* terminate next time */
2678 else
2679 sv_inc(cur);
2680 break;
2681 }
a48ce6be 2682
fcef60b4
DM
2683 case CXt_LOOP_LAZYIV: /* integer increment */
2684 {
2685 IV cur = cx->blk_loop.state_u.lazyiv.cur;
5d9574c1 2686 if (UNLIKELY(cur > cx->blk_loop.state_u.lazyiv.end))
8a1f10dd 2687 goto retno;
7f61b687 2688
fcef60b4 2689 oldsv = *itersvp;
6d3ca00e
DM
2690 /* see NB comment above */
2691 if (oldsv && LIKELY(SvREFCNT(oldsv) == 1 && !SvMAGICAL(oldsv))) {
eaa5c2d6 2692 /* safe to reuse old SV */
47b96a1e
DM
2693
2694 if ( (SvFLAGS(oldsv) & (SVTYPEMASK|SVf_THINKFIRST|SVf_IVisUV))
2695 == SVt_IV)
2696 {
2697 /* Cheap SvIOK_only().
2698 * Assert that flags which SvIOK_only() would test or
2699 * clear can't be set, because we're SVt_IV */
2700 assert(!(SvFLAGS(oldsv) &
2701 (SVf_OOK|SVf_UTF8|(SVf_OK & ~(SVf_IOK|SVp_IOK)))));
2702 SvFLAGS(oldsv) |= (SVf_IOK|SVp_IOK);
2703 /* SvIV_set() where sv_any points to head */
2704 oldsv->sv_u.svu_iv = cur;
2705
2706 }
2707 else
2708 sv_setiv(oldsv, cur);
eaa5c2d6 2709 }
1c846c1f 2710 else
eaa5c2d6
GA
2711 {
2712 /* we need a fresh SV every time so that loop body sees a
2713 * completely new SV for closures/references to work as they
2714 * used to */
fcef60b4 2715 *itersvp = newSViv(cur);
6d3ca00e 2716 SvREFCNT_dec(oldsv);
eaa5c2d6 2717 }
a2309040 2718
5d9574c1 2719 if (UNLIKELY(cur == IV_MAX)) {
cdc1aa42
NC
2720 /* Handle end of range at IV_MAX */
2721 cx->blk_loop.state_u.lazyiv.end = IV_MIN;
2722 } else
2723 ++cx->blk_loop.state_u.lazyiv.cur;
a48ce6be 2724 break;
fcef60b4 2725 }
a48ce6be 2726
93661e56
DM
2727 case CXt_LOOP_LIST: /* for (1,2,3) */
2728
2729 assert(OPpITER_REVERSED == 2); /* so inc becomes -1 or 1 */
2730 inc = 1 - (PL_op->op_private & OPpITER_REVERSED);
2731 ix = (cx->blk_loop.state_u.stack.ix += inc);
2732 if (UNLIKELY(inc > 0
2733 ? ix > cx->blk_oldsp
2734 : ix <= cx->blk_loop.state_u.stack.basesp)
2735 )
8a1f10dd 2736 goto retno;
93661e56
DM
2737
2738 sv = PL_stack_base[ix];
2739 av = NULL;
2740 goto loop_ary_common;
2741
2742 case CXt_LOOP_ARY: /* for (@ary) */
2743
2744 av = cx->blk_loop.state_u.ary.ary;
2745 inc = 1 - (PL_op->op_private & OPpITER_REVERSED);
2746 ix = (cx->blk_loop.state_u.ary.ix += inc);
2747 if (UNLIKELY(inc > 0
2748 ? ix > AvFILL(av)
2749 : ix < 0)
2750 )
8a1f10dd 2751 goto retno;
de080daa 2752
9d1ee8e0 2753 if (UNLIKELY(SvRMAGICAL(av))) {
a8a20bb6
DM
2754 SV * const * const svp = av_fetch(av, ix, FALSE);
2755 sv = svp ? *svp : NULL;
2756 }
2757 else {
2758 sv = AvARRAY(av)[ix];
de080daa 2759 }
ef3e5ea9 2760
93661e56
DM
2761 loop_ary_common:
2762
d39c26a6
FC
2763 if (UNLIKELY(cx->cx_type & CXp_FOR_LVREF)) {
2764 SvSetMagicSV(*itersvp, sv);
2765 break;
2766 }
2767
5d9574c1
DM
2768 if (LIKELY(sv)) {
2769 if (UNLIKELY(SvIS_FREED(sv))) {
f38aa882
DM
2770 *itersvp = NULL;
2771 Perl_croak(aTHX_ "Use of freed value in iteration");
2772 }
60779a30 2773 if (SvPADTMP(sv)) {
8e079c2a 2774 sv = newSVsv(sv);
60779a30 2775 }
8e079c2a
FC
2776 else {
2777 SvTEMP_off(sv);
2778 SvREFCNT_inc_simple_void_NN(sv);
2779 }
de080daa 2780 }
93661e56 2781 else if (av) {
199f858d 2782 sv = newSVavdefelem(av, ix, 0);
de080daa 2783 }
a600f7e6
FC
2784 else
2785 sv = &PL_sv_undef;
a0d0e21e 2786
de080daa
DM
2787 oldsv = *itersvp;
2788 *itersvp = sv;
2789 SvREFCNT_dec(oldsv);
de080daa 2790 break;
a48ce6be
DM
2791
2792 default:
2793 DIE(aTHX_ "panic: pp_iter, type=%u", CxTYPE(cx));
2794 }
8a1f10dd
DM
2795
2796 retsv = &PL_sv_yes;
2797 if (0) {
2798 retno:
2799 retsv = &PL_sv_no;
2800 }
2801 /* pp_enteriter should have pre-extended the stack */
2802 assert(PL_stack_sp < PL_stack_max);
2803 *++PL_stack_sp =retsv;
2804
2805 return PL_op->op_next;
a0d0e21e
LW
2806}
2807
ef07e810
DM
2808/*
2809A description of how taint works in pattern matching and substitution.
2810
284167a5
SM
2811This is all conditional on NO_TAINT_SUPPORT not being defined. Under
2812NO_TAINT_SUPPORT, taint-related operations should become no-ops.
2813
4e19c54b 2814While the pattern is being assembled/concatenated and then compiled,
284167a5
SM
2815PL_tainted will get set (via TAINT_set) if any component of the pattern
2816is tainted, e.g. /.*$tainted/. At the end of pattern compilation,
2817the RXf_TAINTED flag is set on the pattern if PL_tainted is set (via
1738e041
DM
2818TAINT_get). It will also be set if any component of the pattern matches
2819based on locale-dependent behavior.
ef07e810 2820
0ab462a6
DM
2821When the pattern is copied, e.g. $r = qr/..../, the SV holding the ref to
2822the pattern is marked as tainted. This means that subsequent usage, such
284167a5
SM
2823as /x$r/, will set PL_tainted using TAINT_set, and thus RXf_TAINTED,
2824on the new pattern too.
ef07e810 2825
272d35c9 2826RXf_TAINTED_SEEN is used post-execution by the get magic code
ef07e810
DM
2827of $1 et al to indicate whether the returned value should be tainted.
2828It is the responsibility of the caller of the pattern (i.e. pp_match,
2829pp_subst etc) to set this flag for any other circumstances where $1 needs
2830to be tainted.
2831
2832The taint behaviour of pp_subst (and pp_substcont) is quite complex.
2833
2834There are three possible sources of taint
2835 * the source string
2836 * the pattern (both compile- and run-time, RXf_TAINTED / RXf_TAINTED_SEEN)
2837 * the replacement string (or expression under /e)
2838
2839There are four destinations of taint and they are affected by the sources
2840according to the rules below:
2841
2842 * the return value (not including /r):
2843 tainted by the source string and pattern, but only for the
2844 number-of-iterations case; boolean returns aren't tainted;
2845 * the modified string (or modified copy under /r):
2846 tainted by the source string, pattern, and replacement strings;
2847 * $1 et al:
2848 tainted by the pattern, and under 'use re "taint"', by the source
2849 string too;
2850 * PL_taint - i.e. whether subsequent code (e.g. in a /e block) is tainted:
2851 should always be unset before executing subsequent code.
2852
2853The overall action of pp_subst is:
2854
2855 * at the start, set bits in rxtainted indicating the taint status of
2856 the various sources.
2857
2858 * After each pattern execution, update the SUBST_TAINT_PAT bit in
2859 rxtainted if RXf_TAINTED_SEEN has been set, to indicate that the
2860 pattern has subsequently become tainted via locale ops.
2861
2862 * If control is being passed to pp_substcont to execute a /e block,
2863 save rxtainted in the CXt_SUBST block, for future use by
2864 pp_substcont.
2865
2866 * Whenever control is being returned to perl code (either by falling
2867 off the "end" of pp_subst/pp_substcont, or by entering a /e block),
2868 use the flag bits in rxtainted to make all the appropriate types of
0ab462a6
DM
2869 destination taint visible; e.g. set RXf_TAINTED_SEEN so that $1
2870 et al will appear tainted.
ef07e810
DM
2871
2872pp_match is just a simpler version of the above.
2873
2874*/
2875
a0d0e21e
LW
2876PP(pp_subst)
2877{
20b7effb 2878 dSP; dTARG;
eb578fdb 2879 PMOP *pm = cPMOP;
a0d0e21e 2880 PMOP *rpm = pm;
eb578fdb 2881 char *s;
a0d0e21e 2882 char *strend;
5c144d81 2883 const char *c;
a0d0e21e 2884 STRLEN clen;
3c6ef0a5
FC
2885 SSize_t iters = 0;
2886 SSize_t maxiters;
a0d0e21e 2887 bool once;
ef07e810
DM
2888 U8 rxtainted = 0; /* holds various SUBST_TAINT_* flag bits.
2889 See "how taint works" above */
a0d0e21e 2890 char *orig;
1ed74d04 2891 U8 r_flags;
eb578fdb 2892 REGEXP *rx = PM_GETRE(pm);
a0d0e21e
LW
2893 STRLEN len;
2894 int force_on_match = 0;
0bcc34c2 2895 const I32 oldsave = PL_savestack_ix;
792b2c16 2896 STRLEN slen;
26a74523 2897 bool doutf8 = FALSE; /* whether replacement is in utf8 */
db2c6cb3 2898#ifdef PERL_ANY_COW
106d9a13 2899 bool was_cow;
ed252734 2900#endif
a0714e2c 2901 SV *nsv = NULL;
b770e143 2902 /* known replacement string? */
eb578fdb 2903 SV *dstr = (pm->op_pmflags & PMf_CONST) ? POPs : NULL;
a0d0e21e 2904
f410a211
NC
2905 PERL_ASYNC_CHECK();
2906
533c011a 2907 if (PL_op->op_flags & OPf_STACKED)
a0d0e21e 2908 TARG = POPs;
6ffceeb7 2909 else if (ARGTARG)
59f00321 2910 GETTARGET;
a0d0e21e 2911 else {
54b9620d 2912 TARG = DEFSV;
a0d0e21e 2913 EXTEND(SP,1);
1c846c1f 2914 }
d9f424b2 2915
64534138 2916 SvGETMAGIC(TARG); /* must come before cow check */
db2c6cb3 2917#ifdef PERL_ANY_COW
106d9a13
DM
2918 /* note that a string might get converted to COW during matching */
2919 was_cow = cBOOL(SvIsCOW(TARG));
ed252734 2920#endif
d13a5d3b
TC
2921 if (!(rpm->op_pmflags & PMf_NONDESTRUCT)) {
2922#ifndef PERL_ANY_COW
2923 if (SvIsCOW(TARG))
2924 sv_force_normal_flags(TARG,0);
2925#endif
2926 if ((SvREADONLY(TARG)
2927 || ( ((SvTYPE(TARG) == SVt_PVGV && isGV_with_GP(TARG))
2928 || SvTYPE(TARG) > SVt_PVLV)
2929 && !(SvTYPE(TARG) == SVt_PVGV && SvFAKE(TARG)))))
2930 Perl_croak_no_modify();
2931 }
8ec5e241
NIS
2932 PUTBACK;
2933
6ac6605d
DM
2934 orig = SvPV_nomg(TARG, len);
2935 /* note we don't (yet) force the var into being a string; if we fail
92711104 2936 * to match, we leave as-is; on successful match however, we *will*
6ac6605d 2937 * coerce into a string, then repeat the match */
4499db73 2938 if (!SvPOKp(TARG) || SvTYPE(TARG) == SVt_PVGV || SvVOK(TARG))
a0d0e21e 2939 force_on_match = 1;
20be6587
DM
2940
2941 /* only replace once? */
2942 once = !(rpm->op_pmflags & PMf_GLOBAL);
2943
ef07e810 2944 /* See "how taint works" above */
284167a5 2945 if (TAINTING_get) {
20be6587
DM
2946 rxtainted = (
2947 (SvTAINTED(TARG) ? SUBST_TAINT_STR : 0)
284167a5 2948 | (RX_ISTAINTED(rx) ? SUBST_TAINT_PAT : 0)
20be6587
DM
2949 | ((pm->op_pmflags & PMf_RETAINT) ? SUBST_TAINT_RETAINT : 0)
2950 | ((once && !(rpm->op_pmflags & PMf_NONDESTRUCT))
2951 ? SUBST_TAINT_BOOLRET : 0));
2952 TAINT_NOT;
2953 }
a12c0f56 2954
a0d0e21e 2955 force_it:
6ac6605d
DM
2956 if (!pm || !orig)
2957 DIE(aTHX_ "panic: pp_subst, pm=%p, orig=%p", pm, orig);
a0d0e21e 2958
6ac6605d
DM
2959 strend = orig + len;
2960 slen = DO_UTF8(TARG) ? utf8_length((U8*)orig, (U8*)strend) : len;
792b2c16
JH
2961 maxiters = 2 * slen + 10; /* We can match twice at each
2962 position, once with zero-length,
2963 second time with non-zero. */
a0d0e21e 2964
794826f4
YO
2965 /* handle the empty pattern */
2966 if (!RX_PRELEN(rx) && PL_curpm && !ReANY(rx)->mother_re) {
2967 if (PL_curpm == PL_reg_curpm)
31775478 2968 Perl_croak(aTHX_ "Use of the empty pattern inside of "
794826f4 2969 "a regex code block is forbidden");
3280af22 2970 pm = PL_curpm;
aaa362c4 2971 rx = PM_GETRE(pm);
a0d0e21e 2972 }
6502e081 2973
6e240d0b 2974#ifdef PERL_SAWAMPERSAND
6502e081
DM
2975 r_flags = ( RX_NPARENS(rx)
2976 || PL_sawampersand
6502e081 2977 || (RX_EXTFLAGS(rx) & (RXf_EVAL_SEEN|RXf_PMf_KEEPCOPY))
5b0e71e9 2978 || (rpm->op_pmflags & PMf_KEEPCOPY)
6502e081
DM
2979 )
2980 ? REXEC_COPY_STR
2981 : 0;
6e240d0b
FC
2982#else
2983 r_flags = REXEC_COPY_STR;
2984#endif
7fba1cd6 2985
0395280b 2986 if (!CALLREGEXEC(rx, orig, strend, orig, 0, TARG, NULL, r_flags))
8b64c330 2987 {
5e79dfb9
DM
2988 SPAGAIN;
2989 PUSHs(rpm->op_pmflags & PMf_NONDESTRUCT ? TARG : &PL_sv_no);
2990 LEAVE_SCOPE(oldsave);
2991 RETURN;
2992 }
1754320d
FC
2993 PL_curpm = pm;
2994
71be2cbc 2995 /* known replacement string? */
f272994b 2996 if (dstr) {
8514a05a
JH
2997 /* replacement needing upgrading? */
2998 if (DO_UTF8(TARG) && !doutf8) {
db79b45b 2999 nsv = sv_newmortal();
4a176938 3000 SvSetSV(nsv, dstr);
8df0e7a2 3001 sv_utf8_upgrade(nsv);
5c144d81 3002 c = SvPV_const(nsv, clen);
4a176938
JH
3003 doutf8 = TRUE;
3004 }
3005 else {
5c144d81 3006 c = SvPV_const(dstr, clen);
4a176938 3007 doutf8 = DO_UTF8(dstr);
8514a05a 3008 }
bb933b9b
FC
3009
3010 if (SvTAINTED(dstr))
3011 rxtainted |= SUBST_TAINT_REPL;
f272994b
A
3012 }
3013 else {
6136c704 3014 c = NULL;
f272994b
A
3015 doutf8 = FALSE;
3016 }
3017
71be2cbc 3018 /* can do inplace substitution? */
ed252734 3019 if (c
db2c6cb3 3020#ifdef PERL_ANY_COW
106d9a13 3021 && !was_cow
ed252734 3022#endif
fbfb1899 3023 && (I32)clen <= RX_MINLENRET(rx)
9cefd268
FC
3024 && ( once
3025 || !(r_flags & REXEC_COPY_STR)
3026 || (!SvGMAGICAL(dstr) && !(RX_EXTFLAGS(rx) & RXf_EVAL_SEEN))
3027 )
dbc200c5 3028 && !(RX_EXTFLAGS(rx) & RXf_NO_INPLACE_SUBST)
8ca8a454
NC
3029 && (!doutf8 || SvUTF8(TARG))
3030 && !(rpm->op_pmflags & PMf_NONDESTRUCT))
8b030b38 3031 {
ec911639 3032
db2c6cb3 3033#ifdef PERL_ANY_COW
106d9a13 3034 /* string might have got converted to COW since we set was_cow */
ed252734 3035 if (SvIsCOW(TARG)) {
f7a8268c 3036 if (!force_on_match)
ed252734 3037 goto have_a_cow;
f7a8268c 3038 assert(SvVOK(TARG));
ed252734
NC
3039 }
3040#endif
71be2cbc 3041 if (force_on_match) {
6ac6605d
DM
3042 /* redo the first match, this time with the orig var
3043 * forced into being a string */
71be2cbc 3044 force_on_match = 0;
6ac6605d 3045 orig = SvPV_force_nomg(TARG, len);
71be2cbc
PP
3046 goto force_it;
3047 }
39b40493 3048
71be2cbc 3049 if (once) {
c67ab8f2 3050 char *d, *m;
20be6587
DM
3051 if (RX_MATCH_TAINTED(rx)) /* run time pattern taint, eg locale */
3052 rxtainted |= SUBST_TAINT_PAT;
07bc277f
NC
3053 m = orig + RX_OFFS(rx)[0].start;
3054 d = orig + RX_OFFS(rx)[0].end;
71be2cbc
PP
3055 s = orig;
3056 if (m - s > strend - d) { /* faster to shorten from end */
2ec7214c 3057 I32 i;
71be2cbc
PP
3058 if (clen) {
3059 Copy(c, m, clen, char);
3060 m += clen;
a0d0e21e 3061 }
71be2cbc
PP
3062 i = strend - d;
3063 if (i > 0) {
3064 Move(d, m, i, char);
3065 m += i;
a0d0e21e 3066 }
71be2cbc
PP
3067 *m = '\0';
3068 SvCUR_set(TARG, m - s);
3069 }
2ec7214c
DM
3070 else { /* faster from front */
3071 I32 i = m - s;
71be2cbc 3072 d -= clen;
2ec7214c
DM
3073 if (i > 0)
3074 Move(s, d - i, i, char);
71be2cbc 3075 sv_chop(TARG, d-i);
71be2cbc 3076 if (clen)
c947cd8d 3077 Copy(c, d, clen, char);
71be2cbc 3078 }
8ec5e241 3079 SPAGAIN;
8ca8a454 3080 PUSHs(&PL_sv_yes);
71be2cbc
PP
3081 }
3082 else {
c67ab8f2 3083 char *d, *m;
0395280b 3084 d = s = RX_OFFS(rx)[0].start + orig;
71be2cbc 3085 do {
2b25edcf 3086 I32 i;
5d9574c1 3087 if (UNLIKELY(iters++ > maxiters))
cea2e8a9 3088 DIE(aTHX_ "Substitution loop");
5d9574c1 3089 if (UNLIKELY(RX_MATCH_TAINTED(rx))) /* run time pattern taint, eg locale */
20be6587 3090 rxtainted |= SUBST_TAINT_PAT;
07bc277f 3091 m = RX_OFFS(rx)[0].start + orig;
155aba94 3092 if ((i = m - s)) {
71be2cbc
PP
3093 if (s != d)
3094 Move(s, d, i, char);
3095 d += i;
a0d0e21e 3096 }
71be2cbc
PP
3097 if (clen) {
3098 Copy(c, d, clen, char);
3099 d += clen;
3100 }
07bc277f 3101 s = RX_OFFS(rx)[0].end + orig;
7ce41e5c
FC
3102 } while (CALLREGEXEC(rx, s, strend, orig,
3103 s == m, /* don't match same null twice */
f722798b 3104 TARG, NULL,
d5e7783a 3105 REXEC_NOT_FIRST|REXEC_IGNOREPOS|REXEC_FAIL_ON_UNDERFLOW));
71be2cbc 3106 if (s != d) {
2b25edcf 3107 I32 i = strend - s;
aa07b2f6 3108 SvCUR_set(TARG, d - SvPVX_const(TARG) + i);
71be2cbc 3109 Move(s, d, i+1, char); /* include the NUL */
a0d0e21e 3110 }
8ec5e241 3111 SPAGAIN;
3c6ef0a5 3112 mPUSHi(iters);
a0d0e21e
LW
3113 }
3114 }
ff6e92e8 3115 else {
1754320d 3116 bool first;
c67ab8f2 3117 char *m;
1754320d 3118 SV *repl;
a0d0e21e 3119 if (force_on_match) {
6ac6605d
DM
3120 /* redo the first match, this time with the orig var
3121 * forced into being a string */
a0d0e21e 3122 force_on_match = 0;
0c1438a1
NC
3123 if (rpm->op_pmflags & PMf_NONDESTRUCT) {
3124 /* I feel that it should be possible to avoid this mortal copy
3125 given that the code below copies into a new destination.
3126 However, I suspect it isn't worth the complexity of
3127 unravelling the C<goto force_it> for the small number of
3128 cases where it would be viable to drop into the copy code. */
3129 TARG = sv_2mortal(newSVsv(TARG));
3130 }
6ac6605d 3131 orig = SvPV_force_nomg(TARG, len);
a0d0e21e
LW
3132 goto force_it;
3133 }
db2c6cb3 3134#ifdef PERL_ANY_COW
ed252734
NC
3135 have_a_cow:
3136#endif
20be6587
DM
3137 if (RX_MATCH_TAINTED(rx)) /* run time pattern taint, eg locale */
3138 rxtainted |= SUBST_TAINT_PAT;
1754320d 3139 repl = dstr;
0395280b
DM
3140 s = RX_OFFS(rx)[0].start + orig;
3141 dstr = newSVpvn_flags(orig, s-orig,
3142 SVs_TEMP | (DO_UTF8(TARG) ? SVf_UTF8 : 0));
a0d0e21e 3143 if (!c) {
eb578fdb 3144 PERL_CONTEXT *cx;
8ec5e241 3145 SPAGAIN;
0395280b 3146 m = orig;
20be6587
DM
3147 /* note that a whole bunch of local vars are saved here for
3148 * use by pp_substcont: here's a list of them in case you're
3149 * searching for places in this sub that uses a particular var:
3150 * iters maxiters r_flags oldsave rxtainted orig dstr targ
3151 * s m strend rx once */
490576d1 3152 CX_PUSHSUBST(cx);
20e98b0f 3153 RETURNOP(cPMOP->op_pmreplrootu.op_pmreplroot);
a0d0e21e 3154 }
1754320d 3155 first = TRUE;
a0d0e21e 3156 do {
5d9574c1 3157 if (UNLIKELY(iters++ > maxiters))
cea2e8a9 3158 DIE(aTHX_ "Substitution loop");
5d9574c1 3159 if (UNLIKELY(RX_MATCH_TAINTED(rx)))
20be6587 3160 rxtainted |= SUBST_TAINT_PAT;
07bc277f 3161 if (RX_MATCH_COPIED(rx) && RX_SUBBEG(rx) != orig) {
c67ab8f2
DM
3162 char *old_s = s;
3163 char *old_orig = orig;
6502e081 3164 assert(RX_SUBOFFSET(rx) == 0);
c67ab8f2 3165
07bc277f 3166 orig = RX_SUBBEG(rx);
c67ab8f2
DM
3167 s = orig + (old_s - old_orig);
3168 strend = s + (strend - old_s);
a0d0e21e 3169 }
07bc277f 3170 m = RX_OFFS(rx)[0].start + orig;
64534138 3171 sv_catpvn_nomg_maybeutf8(dstr, s, m - s, DO_UTF8(TARG));
07bc277f 3172 s = RX_OFFS(rx)[0].end + orig;
1754320d
FC
3173 if (first) {
3174 /* replacement already stringified */
3175 if (clen)
64534138 3176 sv_catpvn_nomg_maybeutf8(dstr, c, clen, doutf8);
1754320d
FC
3177 first = FALSE;
3178 }
3179 else {
8df0e7a2 3180 sv_catsv(dstr, repl);
5d9574c1 3181 if (UNLIKELY(SvTAINTED(repl)))
bb933b9b 3182 rxtainted |= SUBST_TAINT_REPL;
1754320d 3183 }
a0d0e21e
LW
3184 if (once)
3185 break;
ff27773b
KW
3186 } while (CALLREGEXEC(rx, s, strend, orig,
3187 s == m, /* Yields minend of 0 or 1 */
d5e7783a
DM
3188 TARG, NULL,
3189 REXEC_NOT_FIRST|REXEC_IGNOREPOS|REXEC_FAIL_ON_UNDERFLOW));
aba224f7 3190 assert(strend >= s);
64534138 3191 sv_catpvn_nomg_maybeutf8(dstr, s, strend - s, DO_UTF8(TARG));
748a9306 3192
8ca8a454
NC
3193 if (rpm->op_pmflags & PMf_NONDESTRUCT) {
3194 /* From here on down we're using the copy, and leaving the original
3195 untouched. */
3196 TARG = dstr;
3197 SPAGAIN;
3198 PUSHs(dstr);
3199 } else {
db2c6cb3 3200#ifdef PERL_ANY_COW
8ca8a454
NC
3201 /* The match may make the string COW. If so, brilliant, because
3202 that's just saved us one malloc, copy and free - the regexp has
3203 donated the old buffer, and we malloc an entirely new one, rather
3204 than the regexp malloc()ing a buffer and copying our original,
3205 only for us to throw it away here during the substitution. */
3206 if (SvIsCOW(TARG)) {
3207 sv_force_normal_flags(TARG, SV_COW_DROP_PV);
3208 } else
ed252734 3209#endif
8ca8a454
NC
3210 {
3211 SvPV_free(TARG);
3212 }
3213 SvPV_set(TARG, SvPVX(dstr));
3214 SvCUR_set(TARG, SvCUR(dstr));
3215 SvLEN_set(TARG, SvLEN(dstr));
64534138 3216 SvFLAGS(TARG) |= SvUTF8(dstr);
8ca8a454 3217 SvPV_set(dstr, NULL);
748a9306 3218
8ca8a454 3219 SPAGAIN;
3c6ef0a5 3220 mPUSHi(iters);
8ca8a454
NC
3221 }
3222 }
3223
3224 if (!(rpm->op_pmflags & PMf_NONDESTRUCT)) {
3225 (void)SvPOK_only_UTF8(TARG);
a0d0e21e 3226 }
20be6587 3227
ef07e810 3228 /* See "how taint works" above */
284167a5 3229 if (TAINTING_get) {
20be6587
DM
3230 if ((rxtainted & SUBST_TAINT_PAT) ||
3231 ((rxtainted & (SUBST_TAINT_STR|SUBST_TAINT_RETAINT)) ==
3232 (SUBST_TAINT_STR|SUBST_TAINT_RETAINT))
3233 )
3234 (RX_MATCH_TAINTED_on(rx)); /* taint $1 et al */
3235
3236 if (!(rxtainted & SUBST_TAINT_BOOLRET)
3237 && (rxtainted & (SUBST_TAINT_STR|SUBST_TAINT_PAT))
3238 )
3239 SvTAINTED_on(TOPs); /* taint return value */
3240 else
3241 SvTAINTED_off(TOPs); /* may have got tainted earlier */
3242
3243 /* needed for mg_set below */
284167a5
SM
3244 TAINT_set(
3245 cBOOL(rxtainted & (SUBST_TAINT_STR|SUBST_TAINT_PAT|SUBST_TAINT_REPL))
3246 );
20be6587
DM
3247 SvTAINT(TARG);
3248 }
3249 SvSETMAGIC(TARG); /* PL_tainted must be correctly set for this mg_set */
3250 TAINT_NOT;
f1a76097
DM
3251 LEAVE_SCOPE(oldsave);
3252 RETURN;
a0d0e21e
LW
3253}
3254
3255PP(pp_grepwhile)
3256{
20b7effb 3257 dSP;
a0d0e21e
LW
3258
3259 if (SvTRUEx(POPs))
3280af22
NIS
3260 PL_stack_base[PL_markstack_ptr[-1]++] = PL_stack_base[*PL_markstack_ptr];
3261 ++*PL_markstack_ptr;
b2a2a901 3262 FREETMPS;
d343c3ef 3263 LEAVE_with_name("grep_item"); /* exit inner scope */
a0d0e21e
LW
3264
3265 /* All done yet? */
5d9574c1 3266 if (UNLIKELY(PL_stack_base + *PL_markstack_ptr > SP)) {
a0d0e21e 3267 I32 items;
1c23e2bd 3268 const U8 gimme = GIMME_V;
a0d0e21e 3269
d343c3ef 3270 LEAVE_with_name("grep"); /* exit outer scope */
a0d0e21e 3271 (void)POPMARK; /* pop src */
3280af22 3272 items = --*PL_markstack_ptr - PL_markstack_ptr[-1];
a0d0e21e 3273 (void)POPMARK; /* pop dst */
3280af22 3274 SP = PL_stack_base + POPMARK; /* pop original mark */
54310121 3275 if (gimme == G_SCALAR) {
7cc47870
RGS
3276 dTARGET;
3277 XPUSHi(items);
a0d0e21e 3278 }
54310121
PP
3279 else if (gimme == G_ARRAY)
3280 SP += items;
a0d0e21e
LW
3281 RETURN;
3282 }
3283 else {
3284 SV *src;
3285
d343c3ef 3286 ENTER_with_name("grep_item"); /* enter inner scope */
1d7c1841 3287 SAVEVPTR(PL_curpm);
a0d0e21e 3288
6cae08a8 3289 src = PL_stack_base[TOPMARK];
60779a30 3290 if (SvPADTMP(src)) {
6cae08a8 3291 src = PL_stack_base[TOPMARK] = sv_mortalcopy(src);
a0ed822e
FC
3292 PL_tmps_floor++;
3293 }
a0d0e21e 3294 SvTEMP_off(src);
ffd49c98 3295 DEFSV_set(src);
a0d0e21e
LW
3296
3297 RETURNOP(cLOGOP->op_other);
3298 }
3299}
3300
799da9d7 3301/* leave_adjust_stacks():
f7a874b8 3302 *
e02ce34b
DM
3303 * Process a scope's return args (in the range from_sp+1 .. PL_stack_sp),
3304 * positioning them at to_sp+1 onwards, and do the equivalent of a
3305 * FREEMPS and TAINT_NOT.
3306 *
f7a874b8
DM
3307 * Not intended to be called in void context.
3308 *
799da9d7
DM
3309 * When leaving a sub, eval, do{} or other scope, the things that need
3310 * doing to process the return args are:
f7a874b8 3311 * * in scalar context, only return the last arg (or PL_sv_undef if none);
799da9d7
DM
3312 * * for the types of return that return copies of their args (such
3313 * as rvalue sub return), make a mortal copy of every return arg,
3314 * except where we can optimise the copy away without it being
3315 * semantically visible;
3316 * * make sure that the arg isn't prematurely freed; in the case of an
3317 * arg not copied, this may involve mortalising it. For example, in
f7a874b8
DM
3318 * C<sub f { my $x = ...; $x }>, $x would be freed when we do
3319 * CX_LEAVE_SCOPE(cx) unless it's protected or copied.
3320 *
799da9d7
DM
3321 * What condition to use when deciding whether to pass the arg through
3322 * or make a copy, is determined by the 'pass' arg; its valid values are:
3323 * 0: rvalue sub/eval exit
3324 * 1: other rvalue scope exit
3325 * 2: :lvalue sub exit in rvalue context
3326 * 3: :lvalue sub exit in lvalue context and other lvalue scope exits
3327 *
f7a874b8 3328 * There is a big issue with doing a FREETMPS. We would like to free any
799da9d7 3329 * temps created by the last statement which the sub executed, rather than
f7a874b8
DM
3330 * leaving them for the caller. In a situation where a sub call isn't
3331 * soon followed by a nextstate (e.g. nested recursive calls, a la
3332 * fibonacci()), temps can accumulate, causing memory and performance
3333 * issues.
3334 *
3335 * On the other hand, we don't want to free any TEMPs which are keeping
799da9d7
DM
3336 * alive any return args that we skipped copying; nor do we wish to undo
3337 * any mortalising done here.
f7a874b8
DM
3338 *
3339 * The solution is to split the temps stack frame into two, with a cut
3340 * point delineating the two halves. We arrange that by the end of this
3341 * function, all the temps stack frame entries we wish to keep are in the
799da9d7 3342 * range PL_tmps_floor+1.. tmps_base-1, while the ones to free now are in
f7a874b8
DM
3343 * the range tmps_base .. PL_tmps_ix. During the course of this
3344 * function, tmps_base starts off as PL_tmps_floor+1, then increases
3345 * whenever we find or create a temp that we know should be kept. In
3346 * general the stuff above tmps_base is undecided until we reach the end,
3347 * and we may need a sort stage for that.
3348 *
3349 * To determine whether a TEMP is keeping a return arg alive, every
3350 * arg that is kept rather than copied and which has the SvTEMP flag
3351 * set, has the flag temporarily unset, to mark it. At the end we scan
799da9d7 3352 * the temps stack frame above the cut for entries without SvTEMP and
f7a874b8 3353 * keep them, while turning SvTEMP on again. Note that if we die before
799da9d7 3354 * the SvTEMPs flags are set again, its safe: at worst, subsequent use of
f7a874b8
DM
3355 * those SVs may be slightly less efficient.
3356 *
3357 * In practice various optimisations for some common cases mean we can
3358 * avoid most of the scanning and swapping about with the temps stack.
3359 */
3360
799da9d7 3361void
1c23e2bd 3362Perl_leave_adjust_stacks(pTHX_ SV **from_sp, SV **to_sp, U8 gimme, int pass)
a0d0e21e 3363{
263e0548 3364 dVAR;
20b7effb 3365 dSP;
f7a874b8
DM
3366 SSize_t tmps_base; /* lowest index into tmps stack that needs freeing now */
3367 SSize_t nargs;
3368
799da9d7
DM
3369 PERL_ARGS_ASSERT_LEAVE_ADJUST_STACKS;
3370
f7a874b8
DM
3371 TAINT_NOT;
3372
3373 if (gimme == G_ARRAY) {
e02ce34b
DM
3374 nargs = SP - from_sp;
3375 from_sp++;
f7a874b8
DM
3376 }
3377 else {
3378 assert(gimme == G_SCALAR);
e02ce34b 3379 if (UNLIKELY(from_sp >= SP)) {
f7a874b8 3380 /* no return args */
e02ce34b 3381 assert(from_sp == SP);
f7a874b8
DM
3382 EXTEND(SP, 1);
3383 *++SP = &PL_sv_undef;
e02ce34b 3384 to_sp = SP;
f7a874b8
DM
3385 nargs = 0;
3386 }
3387 else {
3388 from_sp = SP;
3389 nargs = 1;
3390 }
3391 }
3392
3393 /* common code for G_SCALAR and G_ARRAY */
3394
3395 tmps_base = PL_tmps_floor + 1;
3396
3397 assert(nargs >= 0);
3398 if (nargs) {
3399 /* pointer version of tmps_base. Not safe across temp stack
3400 * reallocs. */
3401 SV **tmps_basep;
3402
3403 EXTEND_MORTAL(nargs); /* one big extend for worst-case scenario */
3404 tmps_basep = PL_tmps_stack + tmps_base;
f7a874b8
DM
3405
3406 /* process each return arg */
3407
3408 do {
3409 SV *sv = *from_sp++;
3410
3411 assert(PL_tmps_ix + nargs < PL_tmps_max);
3645bb38
DM
3412#ifdef DEBUGGING
3413 /* PADTMPs with container set magic shouldn't appear in the
3414 * wild. This assert is more important for pp_leavesublv(),
3415 * but by testing for it here, we're more likely to catch
3416 * bad cases (what with :lvalue subs not being widely
3417 * deployed). The two issues are that for something like
3418 * sub :lvalue { $tied{foo} }
3419 * or
3420 * sub :lvalue { substr($foo,1,2) }
3421 * pp_leavesublv() will croak if the sub returns a PADTMP,
3422 * and currently functions like pp_substr() return a mortal
3423 * rather than using their PADTMP when returning a PVLV.
3424 * This is because the PVLV will hold a ref to $foo,
3425 * so $foo would get delayed in being freed while
3426 * the PADTMP SV remained in the PAD.
3427 * So if this assert fails it means either:
3428 * 1) there is pp code similar to pp_substr that is
3429 * returning a PADTMP instead of a mortal, and probably
3430 * needs fixing, or
5d9c1c9a 3431 * 2) pp_leavesublv is making unwarranted assumptions
3645bb38
DM
3432 * about always croaking on a PADTMP
3433 */
3434 if (SvPADTMP(sv) && SvSMAGICAL(sv)) {
3435 MAGIC *mg;
3436 for (mg = SvMAGIC(sv); mg; mg = mg->mg_moremagic) {
3437 assert(PERL_MAGIC_TYPE_IS_VALUE_MAGIC(mg->mg_type));
3438 }
3439 }
3440#endif
f7a874b8 3441
799da9d7
DM
3442 if (
3443 pass == 0 ? (SvTEMP(sv) && !SvMAGICAL(sv) && SvREFCNT(sv) == 1)
3444 : pass == 1 ? ((SvTEMP(sv) || SvPADTMP(sv)) && !SvMAGICAL(sv) && SvREFCNT(sv) == 1)
3445 : pass == 2 ? (!SvPADTMP(sv))
3446 : 1)
3447 {
3448 /* pass through: skip copy for logic or optimisation
3449 * reasons; instead mortalise it, except that ... */
e02ce34b 3450 *++to_sp = sv;
f7a874b8 3451
799da9d7
DM
3452 if (SvTEMP(sv)) {
3453 /* ... since this SV is an SvTEMP , we don't need to
3454 * re-mortalise it; instead we just need to ensure
3455 * that its existing entry in the temps stack frame
3456 * ends up below the cut and so avoids being freed
3457 * this time round. We mark it as needing to be kept
3458 * by temporarily unsetting SvTEMP; then at the end,
3459 * we shuffle any !SvTEMP entries on the tmps stack
3460 * back below the cut.
3461 * However, there's a significant chance that there's
3462 * a 1:1 correspondence between the first few (or all)
3463 * elements in the return args stack frame and those
3464 * in the temps stack frame; e,g.:
3465 * sub f { ....; map {...} .... },
3466 * or if we're exiting multiple scopes and one of the
3467 * inner scopes has already made mortal copies of each
3468 * return arg.
3469 *
3470 * If so, this arg sv will correspond to the next item
3471 * on the tmps stack above the cut, and so can be kept
3472 * merely by moving the cut boundary up one, rather
3473 * than messing with SvTEMP. If all args are 1:1 then
3474 * we can avoid the sorting stage below completely.
977d0c81
DM
3475 *
3476 * If there are no items above the cut on the tmps
3477 * stack, then the SvTEMP must comne from an item
3478 * below the cut, so there's nothing to do.
799da9d7 3479 */
977d0c81
DM
3480 if (tmps_basep <= &PL_tmps_stack[PL_tmps_ix]) {
3481 if (sv == *tmps_basep)
3482 tmps_basep++;
3483 else
3484 SvTEMP_off(sv);
3485 }
799da9d7 3486 }
75bc488d 3487 else if (!SvPADTMP(sv)) {
799da9d7 3488 /* mortalise arg to avoid it being freed during save
75bc488d 3489 * stack unwinding. Pad tmps don't need mortalising as
977d0c81
DM
3490 * they're never freed. This is the equivalent of
3491 * sv_2mortal(SvREFCNT_inc(sv)), except that:
799da9d7
DM
3492 * * it assumes that the temps stack has already been
3493 * extended;
3494 * * it puts the new item at the cut rather than at
3495 * ++PL_tmps_ix, moving the previous occupant there
3496 * instead.
3497 */
3498 if (!SvIMMORTAL(sv)) {
977d0c81 3499 SvREFCNT_inc_simple_void_NN(sv);
799da9d7 3500 SvTEMP_on(sv);
977d0c81
DM
3501 /* Note that if there's nothing above the cut,
3502 * this copies the garbage one slot above
3503 * PL_tmps_ix onto itself. This is harmless (the
3504 * stack's already been extended), but might in
3505 * theory trigger warnings from tools like ASan
3506 */
799da9d7
DM
3507 PL_tmps_stack[++PL_tmps_ix] = *tmps_basep;
3508 *tmps_basep++ = sv;
3509 }
3510 }
f7a874b8
DM
3511 }
3512 else {
3513 /* Make a mortal copy of the SV.
3514 * The following code is the equivalent of sv_mortalcopy()
3515 * except that:
3516 * * it assumes the temps stack has already been extended;
3517 * * it optimises the copying for some simple SV types;
3518 * * it puts the new item at the cut rather than at
3519 * ++PL_tmps_ix, moving the previous occupant there
3520 * instead.
3521 */
3522 SV *newsv = newSV(0);
3523
3524 PL_tmps_stack[++PL_tmps_ix] = *tmps_basep;
3525 /* put it on the tmps stack early so it gets freed if we die */
3526 *tmps_basep++ = newsv;
e02ce34b 3527 *++to_sp = newsv;
f7a874b8
DM
3528
3529 if (SvTYPE(sv) <= SVt_IV) {
3530 /* arg must be one of undef, IV/UV, or RV: skip
3531 * sv_setsv_flags() and do the copy directly */
3532 U32 dstflags;
3533 U32 srcflags = SvFLAGS(sv);
3534
3535 assert(!SvGMAGICAL(sv));
3536 if (srcflags & (SVf_IOK|SVf_ROK)) {
3537 SET_SVANY_FOR_BODYLESS_IV(newsv);
3538
3539 if (srcflags & SVf_ROK) {
3540 newsv->sv_u.svu_rv = SvREFCNT_inc(SvRV(sv));
3541 /* SV type plus flags */
3542 dstflags = (SVt_IV|SVf_ROK|SVs_TEMP);
3543 }
3544 else {
3545 /* both src and dst are <= SVt_IV, so sv_any
3546 * points to the head; so access the heads
3547 * directly rather than going via sv_any.
3548 */
3549 assert( &(sv->sv_u.svu_iv)
3550 == &(((XPVIV*) SvANY(sv))->xiv_iv));
3551 assert( &(newsv->sv_u.svu_iv)
3552 == &(((XPVIV*) SvANY(newsv))->xiv_iv));
3553 newsv->sv_u.svu_iv = sv->sv_u.svu_iv;
3554 /* SV type plus flags */
3555 dstflags = (SVt_IV|SVf_IOK|SVp_IOK|SVs_TEMP
3556 |(srcflags & SVf_IVisUV));
3557 }
3558 }
3559 else {
3560 assert(!(srcflags & SVf_OK));
3561 dstflags = (SVt_NULL|SVs_TEMP); /* SV type plus flags */
3562 }
3563 SvFLAGS(newsv) = dstflags;
3564
3565 }
3566 else {
3567 /* do the full sv_setsv() */
3568 SSize_t old_base;
3569
3570 SvTEMP_on(newsv);
3571 old_base = tmps_basep - PL_tmps_stack;
3572 SvGETMAGIC(sv);
3573 sv_setsv_flags(newsv, sv, SV_DO_COW_SVSETSV);
799da9d7 3574 /* the mg_get or sv_setsv might have created new temps
f7a874b8
DM
3575 * or realloced the tmps stack; regrow and reload */
3576 EXTEND_MORTAL(nargs);
3577 tmps_basep = PL_tmps_stack + old_base;
3578 TAINT_NOT; /* Each item is independent */
3579 }
3580
3581 }
3582 } while (--nargs);
3583
3584 /* If there are any temps left above the cut, we need to sort
3585 * them into those to keep and those to free. The only ones to
3586 * keep are those for which we've temporarily unset SvTEMP.
3587 * Work inwards from the two ends at tmps_basep .. PL_tmps_ix,
3588 * swapping pairs as necessary. Stop when we meet in the middle.
3589 */
3590 {
3591 SV **top = PL_tmps_stack + PL_tmps_ix;
3592 while (tmps_basep <= top) {
3593 SV *sv = *top;
3594 if (SvTEMP(sv))
3595 top--;
3596 else {
3597 SvTEMP_on(sv);
3598 *top = *tmps_basep;