This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
[perl #127635] s///r with -DPERL_NO_COW attempts to modify source SV
[perl5.git] / pp_hot.c
CommitLineData
a0d0e21e
LW
1/* pp_hot.c
2 *
1129b882
NC
3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 by Larry Wall and others
a0d0e21e
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
12 * Then he heard Merry change the note, and up went the Horn-cry of Buckland,
13 * shaking the air.
14 *
4ac71550
TC
15 * Awake! Awake! Fear, Fire, Foes! Awake!
16 * Fire, Foes! Awake!
17 *
18 * [p.1007 of _The Lord of the Rings_, VI/viii: "The Scouring of the Shire"]
a0d0e21e
LW
19 */
20
166f8a29
DM
21/* This file contains 'hot' pp ("push/pop") functions that
22 * execute the opcodes that make up a perl program. A typical pp function
23 * expects to find its arguments on the stack, and usually pushes its
24 * results onto the stack, hence the 'pp' terminology. Each OP structure
25 * contains a pointer to the relevant pp_foo() function.
26 *
27 * By 'hot', we mean common ops whose execution speed is critical.
28 * By gathering them together into a single file, we encourage
29 * CPU cache hits on hot code. Also it could be taken as a warning not to
30 * change any code in this file unless you're sure it won't affect
31 * performance.
32 */
33
a0d0e21e 34#include "EXTERN.h"
864dbfa3 35#define PERL_IN_PP_HOT_C
a0d0e21e
LW
36#include "perl.h"
37
38/* Hot code. */
39
40PP(pp_const)
41{
39644a26 42 dSP;
996c9baa 43 XPUSHs(cSVOP_sv);
a0d0e21e
LW
44 RETURN;
45}
46
47PP(pp_nextstate)
48{
533c011a 49 PL_curcop = (COP*)PL_op;
a0d0e21e 50 TAINT_NOT; /* Each statement is presumed innocent */
4ebe6e95 51 PL_stack_sp = PL_stack_base + CX_CUR()->blk_oldsp;
a0d0e21e 52 FREETMPS;
f410a211 53 PERL_ASYNC_CHECK();
a0d0e21e
LW
54 return NORMAL;
55}
56
57PP(pp_gvsv)
58{
39644a26 59 dSP;
924508f0 60 EXTEND(SP,1);
5d9574c1 61 if (UNLIKELY(PL_op->op_private & OPpLVAL_INTRO))
1d7c1841 62 PUSHs(save_scalar(cGVOP_gv));
a0d0e21e 63 else
c69033f2 64 PUSHs(GvSVn(cGVOP_gv));
a0d0e21e
LW
65 RETURN;
66}
67
b1c05ba5
DM
68
69/* also used for: pp_lineseq() pp_regcmaybe() pp_scalar() pp_scope() */
70
a0d0e21e
LW
71PP(pp_null)
72{
73 return NORMAL;
74}
75
3dd9d4e4
FC
76/* This is sometimes called directly by pp_coreargs, pp_grepstart and
77 amagic_call. */
a0d0e21e
LW
78PP(pp_pushmark)
79{
3280af22 80 PUSHMARK(PL_stack_sp);
a0d0e21e
LW
81 return NORMAL;
82}
83
84PP(pp_stringify)
85{
20b7effb 86 dSP; dTARGET;
4cc783ef
DD
87 SV * const sv = TOPs;
88 SETs(TARG);
89 sv_copypv(TARG, sv);
90 SvSETMAGIC(TARG);
91 /* no PUTBACK, SETs doesn't inc/dec SP */
92 return NORMAL;
a0d0e21e
LW
93}
94
95PP(pp_gv)
96{
20b7effb 97 dSP;
ad64d0ec 98 XPUSHs(MUTABLE_SV(cGVOP_gv));
a0d0e21e
LW
99 RETURN;
100}
101
b1c05ba5
DM
102
103/* also used for: pp_andassign() */
104
a0d0e21e
LW
105PP(pp_and)
106{
f410a211 107 PERL_ASYNC_CHECK();
4cc783ef
DD
108 {
109 /* SP is not used to remove a variable that is saved across the
110 sv_2bool_flags call in SvTRUE_NN, if a RISC/CISC or low/high machine
111 register or load/store vs direct mem ops macro is introduced, this
112 should be a define block between direct PL_stack_sp and dSP operations,
113 presently, using PL_stack_sp is bias towards CISC cpus */
114 SV * const sv = *PL_stack_sp;
115 if (!SvTRUE_NN(sv))
116 return NORMAL;
117 else {
118 if (PL_op->op_type == OP_AND)
119 --PL_stack_sp;
120 return cLOGOP->op_other;
121 }
a0d0e21e
LW
122 }
123}
124
125PP(pp_sassign)
126{
20b7effb 127 dSP;
3e75a3c4
RU
128 /* sassign keeps its args in the optree traditionally backwards.
129 So we pop them differently.
130 */
131 SV *left = POPs; SV *right = TOPs;
748a9306 132
533c011a 133 if (PL_op->op_private & OPpASSIGN_BACKWARDS) {
0bd48802
AL
134 SV * const temp = left;
135 left = right; right = temp;
a0d0e21e 136 }
d48c660d
DM
137 assert(TAINTING_get || !TAINT_get);
138 if (UNLIKELY(TAINT_get) && !SvTAINTED(right))
a0d0e21e 139 TAINT_NOT;
5d9574c1
DM
140 if (UNLIKELY(PL_op->op_private & OPpASSIGN_CV_TO_GV)) {
141 /* *foo =\&bar */
3e75a3c4 142 SV * const cv = SvRV(right);
e26df76a 143 const U32 cv_type = SvTYPE(cv);
3e75a3c4 144 const bool is_gv = isGV_with_GP(left);
6136c704 145 const bool got_coderef = cv_type == SVt_PVCV || cv_type == SVt_PVFM;
e26df76a
NC
146
147 if (!got_coderef) {
148 assert(SvROK(cv));
149 }
150
3e75a3c4
RU
151 /* Can do the optimisation if left (LVALUE) is not a typeglob,
152 right (RVALUE) is a reference to something, and we're in void
e26df76a 153 context. */
13be902c 154 if (!got_coderef && !is_gv && GIMME_V == G_VOID) {
e26df76a 155 /* Is the target symbol table currently empty? */
3e75a3c4 156 GV * const gv = gv_fetchsv_nomg(left, GV_NOINIT, SVt_PVGV);
bb112e5a 157 if (SvTYPE(gv) != SVt_PVGV && !SvOK(gv)) {
e26df76a
NC
158 /* Good. Create a new proxy constant subroutine in the target.
159 The gv becomes a(nother) reference to the constant. */
160 SV *const value = SvRV(cv);
161
ad64d0ec 162 SvUPGRADE(MUTABLE_SV(gv), SVt_IV);
1ccdb730 163 SvPCS_IMPORTED_on(gv);
e26df76a 164 SvRV_set(gv, value);
b37c2d43 165 SvREFCNT_inc_simple_void(value);
3e75a3c4 166 SETs(left);
e26df76a
NC
167 RETURN;
168 }
169 }
170
171 /* Need to fix things up. */
13be902c 172 if (!is_gv) {
e26df76a 173 /* Need to fix GV. */
3e75a3c4 174 left = MUTABLE_SV(gv_fetchsv_nomg(left,GV_ADD, SVt_PVGV));
e26df76a
NC
175 }
176
177 if (!got_coderef) {
178 /* We've been returned a constant rather than a full subroutine,
179 but they expect a subroutine reference to apply. */
53a42478 180 if (SvROK(cv)) {
d343c3ef 181 ENTER_with_name("sassign_coderef");
53a42478
NC
182 SvREFCNT_inc_void(SvRV(cv));
183 /* newCONSTSUB takes a reference count on the passed in SV
184 from us. We set the name to NULL, otherwise we get into
185 all sorts of fun as the reference to our new sub is
186 donated to the GV that we're about to assign to.
187 */
3e75a3c4 188 SvRV_set(right, MUTABLE_SV(newCONSTSUB(GvSTASH(left), NULL,
ad64d0ec 189 SvRV(cv))));
fc2b2dca 190 SvREFCNT_dec_NN(cv);
d343c3ef 191 LEAVE_with_name("sassign_coderef");
53a42478
NC
192 } else {
193 /* What can happen for the corner case *{"BONK"} = \&{"BONK"};
194 is that
195 First: ops for \&{"BONK"}; return us the constant in the
196 symbol table
197 Second: ops for *{"BONK"} cause that symbol table entry
198 (and our reference to it) to be upgraded from RV
199 to typeblob)
200 Thirdly: We get here. cv is actually PVGV now, and its
201 GvCV() is actually the subroutine we're looking for
202
203 So change the reference so that it points to the subroutine
204 of that typeglob, as that's what they were after all along.
205 */
159b6efe 206 GV *const upgraded = MUTABLE_GV(cv);
53a42478
NC
207 CV *const source = GvCV(upgraded);
208
209 assert(source);
210 assert(CvFLAGS(source) & CVf_CONST);
211
0ad694a7 212 SvREFCNT_inc_simple_void_NN(source);
fc2b2dca 213 SvREFCNT_dec_NN(upgraded);
3e75a3c4 214 SvRV_set(right, MUTABLE_SV(source));
53a42478 215 }
e26df76a 216 }
53a42478 217
e26df76a 218 }
8fe85e3f 219 if (
5d9574c1 220 UNLIKELY(SvTEMP(left)) && !SvSMAGICAL(left) && SvREFCNT(left) == 1 &&
3e75a3c4 221 (!isGV_with_GP(left) || SvFAKE(left)) && ckWARN(WARN_MISC)
8fe85e3f
FC
222 )
223 Perl_warner(aTHX_
224 packWARN(WARN_MISC), "Useless assignment to a temporary"
225 );
3e75a3c4
RU
226 SvSetMagicSV(left, right);
227 SETs(left);
a0d0e21e
LW
228 RETURN;
229}
230
231PP(pp_cond_expr)
232{
20b7effb 233 dSP;
f410a211 234 PERL_ASYNC_CHECK();
a0d0e21e 235 if (SvTRUEx(POPs))
1a67a97c 236 RETURNOP(cLOGOP->op_other);
a0d0e21e 237 else
1a67a97c 238 RETURNOP(cLOGOP->op_next);
a0d0e21e
LW
239}
240
241PP(pp_unstack)
242{
f5319de9 243 PERL_CONTEXT *cx;
8f3964af 244 PERL_ASYNC_CHECK();
a0d0e21e 245 TAINT_NOT; /* Each statement is presumed innocent */
4ebe6e95 246 cx = CX_CUR();
f5319de9 247 PL_stack_sp = PL_stack_base + cx->blk_oldsp;
a0d0e21e 248 FREETMPS;
eae48c89 249 if (!(PL_op->op_flags & OPf_SPECIAL)) {
93661e56 250 assert(CxTYPE(cx) == CXt_BLOCK || CxTYPE_is_LOOP(cx));
dfe0f39b 251 CX_LEAVE_SCOPE(cx);
eae48c89 252 }
a0d0e21e
LW
253 return NORMAL;
254}
255
a0d0e21e
LW
256PP(pp_concat)
257{
20b7effb 258 dSP; dATARGET; tryAMAGICbin_MG(concat_amg, AMGf_assign);
748a9306
LW
259 {
260 dPOPTOPssrl;
8d6d96c1
HS
261 bool lbyte;
262 STRLEN rlen;
d4c19fe8 263 const char *rpv = NULL;
a6b599c7 264 bool rbyte = FALSE;
a9c4fd4e 265 bool rcopied = FALSE;
8d6d96c1 266
6f1401dc
DM
267 if (TARG == right && right != left) { /* $r = $l.$r */
268 rpv = SvPV_nomg_const(right, rlen);
c75ab21a 269 rbyte = !DO_UTF8(right);
59cd0e26 270 right = newSVpvn_flags(rpv, rlen, SVs_TEMP);
349d4f2f 271 rpv = SvPV_const(right, rlen); /* no point setting UTF-8 here */
db79b45b 272 rcopied = TRUE;
8d6d96c1 273 }
7889fe52 274
89734059 275 if (TARG != left) { /* not $l .= $r */
a9c4fd4e 276 STRLEN llen;
6f1401dc 277 const char* const lpv = SvPV_nomg_const(left, llen);
90f5826e 278 lbyte = !DO_UTF8(left);
8d6d96c1
HS
279 sv_setpvn(TARG, lpv, llen);
280 if (!lbyte)
281 SvUTF8_on(TARG);
282 else
283 SvUTF8_off(TARG);
284 }
18ea7bf2
SM
285 else { /* $l .= $r and left == TARG */
286 if (!SvOK(left)) {
89734059 287 if (left == right && ckWARN(WARN_UNINITIALIZED)) /* $l .= $l */
c75ab21a 288 report_uninit(right);
76f68e9b 289 sv_setpvs(left, "");
c75ab21a 290 }
18ea7bf2
SM
291 else {
292 SvPV_force_nomg_nolen(left);
293 }
583a5589 294 lbyte = !DO_UTF8(left);
90f5826e 295 if (IN_BYTES)
18ea7bf2 296 SvUTF8_off(left);
8d6d96c1 297 }
a12c0f56 298
c75ab21a 299 if (!rcopied) {
6f1401dc 300 rpv = SvPV_nomg_const(right, rlen);
c75ab21a
RH
301 rbyte = !DO_UTF8(right);
302 }
8d6d96c1
HS
303 if (lbyte != rbyte) {
304 if (lbyte)
305 sv_utf8_upgrade_nomg(TARG);
306 else {
db79b45b 307 if (!rcopied)
59cd0e26 308 right = newSVpvn_flags(rpv, rlen, SVs_TEMP);
8d6d96c1 309 sv_utf8_upgrade_nomg(right);
6f1401dc 310 rpv = SvPV_nomg_const(right, rlen);
69b47968 311 }
a0d0e21e 312 }
8d6d96c1 313 sv_catpvn_nomg(TARG, rpv, rlen);
43ebc500 314
a0d0e21e
LW
315 SETTARG;
316 RETURN;
748a9306 317 }
a0d0e21e
LW
318}
319
d5524600
DM
320/* push the elements of av onto the stack.
321 * XXX Note that padav has similar code but without the mg_get().
322 * I suspect that the mg_get is no longer needed, but while padav
323 * differs, it can't share this function */
324
f9ae8fb6 325STATIC void
d5524600
DM
326S_pushav(pTHX_ AV* const av)
327{
328 dSP;
c70927a6 329 const SSize_t maxarg = AvFILL(av) + 1;
d5524600 330 EXTEND(SP, maxarg);
5d9574c1 331 if (UNLIKELY(SvRMAGICAL(av))) {
c70927a6
FC
332 PADOFFSET i;
333 for (i=0; i < (PADOFFSET)maxarg; i++) {
d5524600
DM
334 SV ** const svp = av_fetch(av, i, FALSE);
335 /* See note in pp_helem, and bug id #27839 */
336 SP[i+1] = svp
337 ? SvGMAGICAL(*svp) ? (mg_get(*svp), *svp) : *svp
338 : &PL_sv_undef;
339 }
340 }
341 else {
c70927a6
FC
342 PADOFFSET i;
343 for (i=0; i < (PADOFFSET)maxarg; i++) {
ce0d59fd 344 SV * const sv = AvARRAY(av)[i];
5d9574c1 345 SP[i+1] = LIKELY(sv) ? sv : &PL_sv_undef;
ce0d59fd 346 }
d5524600
DM
347 }
348 SP += maxarg;
349 PUTBACK;
350}
351
352
a7fd8ef6
DM
353/* ($lex1,@lex2,...) or my ($lex1,@lex2,...) */
354
355PP(pp_padrange)
356{
20b7effb 357 dSP;
a7fd8ef6
DM
358 PADOFFSET base = PL_op->op_targ;
359 int count = (int)(PL_op->op_private) & OPpPADRANGE_COUNTMASK;
360 int i;
d5524600
DM
361 if (PL_op->op_flags & OPf_SPECIAL) {
362 /* fake the RHS of my ($x,$y,..) = @_ */
363 PUSHMARK(SP);
364 S_pushav(aTHX_ GvAVn(PL_defgv));
365 SPAGAIN;
366 }
367
a7fd8ef6
DM
368 /* note, this is only skipped for compile-time-known void cxt */
369 if ((PL_op->op_flags & OPf_WANT) != OPf_WANT_VOID) {
370 EXTEND(SP, count);
371 PUSHMARK(SP);
372 for (i = 0; i <count; i++)
373 *++SP = PAD_SV(base+i);
374 }
375 if (PL_op->op_private & OPpLVAL_INTRO) {
4e09461c
DM
376 SV **svp = &(PAD_SVl(base));
377 const UV payload = (UV)(
378 (base << (OPpPADRANGE_COUNTSHIFT + SAVE_TIGHT_SHIFT))
379 | (count << SAVE_TIGHT_SHIFT)
380 | SAVEt_CLEARPADRANGE);
6d59e610 381 STATIC_ASSERT_STMT(OPpPADRANGE_COUNTMASK + 1 == (1 << OPpPADRANGE_COUNTSHIFT));
4e09461c 382 assert((payload >> (OPpPADRANGE_COUNTSHIFT+SAVE_TIGHT_SHIFT)) == base);
a3444cc5
DM
383 {
384 dSS_ADD;
385 SS_ADD_UV(payload);
386 SS_ADD_END(1);
387 }
4e09461c 388
a7fd8ef6 389 for (i = 0; i <count; i++)
4e09461c 390 SvPADSTALE_off(*svp++); /* mark lexical as active */
a7fd8ef6
DM
391 }
392 RETURN;
393}
394
395
a0d0e21e
LW
396PP(pp_padsv)
397{
20b7effb 398 dSP;
6c28b496
DD
399 EXTEND(SP, 1);
400 {
401 OP * const op = PL_op;
402 /* access PL_curpad once */
403 SV ** const padentry = &(PAD_SVl(op->op_targ));
404 {
405 dTARG;
406 TARG = *padentry;
407 PUSHs(TARG);
408 PUTBACK; /* no pop/push after this, TOPs ok */
8ec5e241 409 }
6c28b496
DD
410 if (op->op_flags & OPf_MOD) {
411 if (op->op_private & OPpLVAL_INTRO)
412 if (!(op->op_private & OPpPAD_STATE))
413 save_clearsv(padentry);
414 if (op->op_private & OPpDEREF) {
8f90a16d
FC
415 /* TOPs is equivalent to TARG here. Using TOPs (SP) rather
416 than TARG reduces the scope of TARG, so it does not
417 span the call to save_clearsv, resulting in smaller
418 machine code. */
6c28b496
DD
419 TOPs = vivify_ref(TOPs, op->op_private & OPpDEREF);
420 }
421 }
422 return op->op_next;
4633a7c4 423 }
a0d0e21e
LW
424}
425
426PP(pp_readline)
427{
30901a8a
FC
428 dSP;
429 if (TOPs) {
430 SvGETMAGIC(TOPs);
fc99edcf 431 tryAMAGICunTARGETlist(iter_amg, 0);
30901a8a
FC
432 PL_last_in_gv = MUTABLE_GV(*PL_stack_sp--);
433 }
434 else PL_last_in_gv = PL_argvgv, PL_stack_sp--;
6e592b3a
BM
435 if (!isGV_with_GP(PL_last_in_gv)) {
436 if (SvROK(PL_last_in_gv) && isGV_with_GP(SvRV(PL_last_in_gv)))
159b6efe 437 PL_last_in_gv = MUTABLE_GV(SvRV(PL_last_in_gv));
8efb3254 438 else {
f5284f61 439 dSP;
ad64d0ec 440 XPUSHs(MUTABLE_SV(PL_last_in_gv));
f5284f61 441 PUTBACK;
897d3989 442 Perl_pp_rv2gv(aTHX);
159b6efe 443 PL_last_in_gv = MUTABLE_GV(*PL_stack_sp--);
84ee769f
FC
444 if (PL_last_in_gv == (GV *)&PL_sv_undef)
445 PL_last_in_gv = NULL;
446 else
447 assert(isGV_with_GP(PL_last_in_gv));
f5284f61
IZ
448 }
449 }
a0d0e21e
LW
450 return do_readline();
451}
452
453PP(pp_eq)
454{
20b7effb 455 dSP;
33efebe6
DM
456 SV *left, *right;
457
a42d0242 458 tryAMAGICbin_MG(eq_amg, AMGf_set|AMGf_numeric);
33efebe6
DM
459 right = POPs;
460 left = TOPs;
461 SETs(boolSV(
462 (SvIOK_notUV(left) && SvIOK_notUV(right))
463 ? (SvIVX(left) == SvIVX(right))
464 : ( do_ncmp(left, right) == 0)
465 ));
466 RETURN;
a0d0e21e
LW
467}
468
b1c05ba5 469
4c2c3128 470/* also used for: pp_i_preinc() */
b1c05ba5 471
a0d0e21e
LW
472PP(pp_preinc)
473{
4c2c3128
DM
474 SV *sv = *PL_stack_sp;
475
476 if (LIKELY(((sv->sv_flags &
477 (SVf_THINKFIRST|SVs_GMG|SVf_IVisUV|
478 SVf_IOK|SVf_NOK|SVf_POK|SVp_NOK|SVp_POK|SVf_ROK))
479 == SVf_IOK))
480 && SvIVX(sv) != IV_MAX)
481 {
482 SvIV_set(sv, SvIVX(sv) + 1);
483 }
484 else /* Do all the PERL_PRESERVE_IVUV and hard cases in sv_inc */
485 sv_inc(sv);
486 SvSETMAGIC(sv);
487 return NORMAL;
488}
489
490
491/* also used for: pp_i_predec() */
492
493PP(pp_predec)
494{
495 SV *sv = *PL_stack_sp;
496
497 if (LIKELY(((sv->sv_flags &
498 (SVf_THINKFIRST|SVs_GMG|SVf_IVisUV|
499 SVf_IOK|SVf_NOK|SVf_POK|SVp_NOK|SVp_POK|SVf_ROK))
500 == SVf_IOK))
501 && SvIVX(sv) != IV_MIN)
55497cff 502 {
4c2c3128 503 SvIV_set(sv, SvIVX(sv) - 1);
748a9306 504 }
4c2c3128
DM
505 else /* Do all the PERL_PRESERVE_IVUV and hard cases in sv_dec */
506 sv_dec(sv);
507 SvSETMAGIC(sv);
a0d0e21e
LW
508 return NORMAL;
509}
510
b1c05ba5
DM
511
512/* also used for: pp_orassign() */
513
a0d0e21e
LW
514PP(pp_or)
515{
20b7effb 516 dSP;
f410a211 517 PERL_ASYNC_CHECK();
a0d0e21e
LW
518 if (SvTRUE(TOPs))
519 RETURN;
520 else {
c960fc3b
SP
521 if (PL_op->op_type == OP_OR)
522 --SP;
a0d0e21e
LW
523 RETURNOP(cLOGOP->op_other);
524 }
525}
526
b1c05ba5
DM
527
528/* also used for: pp_dor() pp_dorassign() */
529
25a55bd7 530PP(pp_defined)
c963b151 531{
20b7effb 532 dSP;
eb578fdb 533 SV* sv;
6136c704 534 bool defined;
25a55bd7 535 const int op_type = PL_op->op_type;
ea5195b7 536 const bool is_dor = (op_type == OP_DOR || op_type == OP_DORASSIGN);
c963b151 537
6136c704 538 if (is_dor) {
f410a211 539 PERL_ASYNC_CHECK();
25a55bd7 540 sv = TOPs;
5d9574c1 541 if (UNLIKELY(!sv || !SvANY(sv))) {
2bd49cfc
NC
542 if (op_type == OP_DOR)
543 --SP;
25a55bd7
SP
544 RETURNOP(cLOGOP->op_other);
545 }
b7c44293
RGS
546 }
547 else {
548 /* OP_DEFINED */
25a55bd7 549 sv = POPs;
5d9574c1 550 if (UNLIKELY(!sv || !SvANY(sv)))
25a55bd7 551 RETPUSHNO;
b7c44293 552 }
25a55bd7 553
6136c704 554 defined = FALSE;
c963b151
BD
555 switch (SvTYPE(sv)) {
556 case SVt_PVAV:
557 if (AvMAX(sv) >= 0 || SvGMAGICAL(sv) || (SvRMAGICAL(sv) && mg_find(sv, PERL_MAGIC_tied)))
25a55bd7 558 defined = TRUE;
c963b151
BD
559 break;
560 case SVt_PVHV:
561 if (HvARRAY(sv) || SvGMAGICAL(sv) || (SvRMAGICAL(sv) && mg_find(sv, PERL_MAGIC_tied)))
25a55bd7 562 defined = TRUE;
c963b151
BD
563 break;
564 case SVt_PVCV:
565 if (CvROOT(sv) || CvXSUB(sv))
25a55bd7 566 defined = TRUE;
c963b151
BD
567 break;
568 default:
5b295bef 569 SvGETMAGIC(sv);
c963b151 570 if (SvOK(sv))
25a55bd7 571 defined = TRUE;
6136c704 572 break;
c963b151 573 }
6136c704
AL
574
575 if (is_dor) {
c960fc3b
SP
576 if(defined)
577 RETURN;
578 if(op_type == OP_DOR)
579 --SP;
25a55bd7 580 RETURNOP(cLOGOP->op_other);
25a55bd7 581 }
d9aa96a4
SP
582 /* assuming OP_DEFINED */
583 if(defined)
584 RETPUSHYES;
585 RETPUSHNO;
c963b151
BD
586}
587
230ee21f
DM
588
589
a0d0e21e
LW
590PP(pp_add)
591{
20b7effb 592 dSP; dATARGET; bool useleft; SV *svl, *svr;
230ee21f 593
6f1401dc
DM
594 tryAMAGICbin_MG(add_amg, AMGf_assign|AMGf_numeric);
595 svr = TOPs;
596 svl = TOPm1s;
597
28e5dec8 598#ifdef PERL_PRESERVE_IVUV
230ee21f
DM
599
600 /* special-case some simple common cases */
601 if (!((svl->sv_flags|svr->sv_flags) & (SVf_IVisUV|SVs_GMG))) {
602 IV il, ir;
603 U32 flags = (svl->sv_flags & svr->sv_flags);
604 if (flags & SVf_IOK) {
605 /* both args are simple IVs */
606 UV topl, topr;
607 il = SvIVX(svl);
608 ir = SvIVX(svr);
609 do_iv:
610 topl = ((UV)il) >> (UVSIZE * 8 - 2);
611 topr = ((UV)ir) >> (UVSIZE * 8 - 2);
612
613 /* if both are in a range that can't under/overflow, do a
614 * simple integer add: if the top of both numbers
615 * are 00 or 11, then it's safe */
616 if (!( ((topl+1) | (topr+1)) & 2)) {
617 SP--;
618 TARGi(il + ir, 0); /* args not GMG, so can't be tainted */
619 SETs(TARG);
620 RETURN;
621 }
622 goto generic;
623 }
624 else if (flags & SVf_NOK) {
625 /* both args are NVs */
626 NV nl = SvNVX(svl);
627 NV nr = SvNVX(svr);
628
3336af0b
DD
629 if (
630#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
631 !Perl_isnan(nl) && nl == (NV)(il = (IV)nl)
632 && !Perl_isnan(nr) && nr == (NV)(ir = (IV)nr)
633#else
634 nl == (NV)(il = (IV)nl) && nr == (NV)(ir = (IV)nr)
635#endif
636 )
230ee21f
DM
637 /* nothing was lost by converting to IVs */
638 goto do_iv;
639 SP--;
640 TARGn(nl + nr, 0); /* args not GMG, so can't be tainted */
641 SETs(TARG);
642 RETURN;
643 }
644 }
645
646 generic:
647
648 useleft = USE_LEFT(svl);
28e5dec8
JH
649 /* We must see if we can perform the addition with integers if possible,
650 as the integer code detects overflow while the NV code doesn't.
651 If either argument hasn't had a numeric conversion yet attempt to get
652 the IV. It's important to do this now, rather than just assuming that
653 it's not IOK as a PV of "9223372036854775806" may not take well to NV
654 addition, and an SV which is NOK, NV=6.0 ought to be coerced to
655 integer in case the second argument is IV=9223372036854775806
656 We can (now) rely on sv_2iv to do the right thing, only setting the
657 public IOK flag if the value in the NV (or PV) slot is truly integer.
658
659 A side effect is that this also aggressively prefers integer maths over
7dca457a
NC
660 fp maths for integer values.
661
a00b5bd3 662 How to detect overflow?
7dca457a
NC
663
664 C 99 section 6.2.6.1 says
665
666 The range of nonnegative values of a signed integer type is a subrange
667 of the corresponding unsigned integer type, and the representation of
668 the same value in each type is the same. A computation involving
669 unsigned operands can never overflow, because a result that cannot be
670 represented by the resulting unsigned integer type is reduced modulo
671 the number that is one greater than the largest value that can be
672 represented by the resulting type.
673
674 (the 9th paragraph)
675
676 which I read as "unsigned ints wrap."
677
678 signed integer overflow seems to be classed as "exception condition"
679
680 If an exceptional condition occurs during the evaluation of an
681 expression (that is, if the result is not mathematically defined or not
682 in the range of representable values for its type), the behavior is
683 undefined.
684
685 (6.5, the 5th paragraph)
686
687 I had assumed that on 2s complement machines signed arithmetic would
688 wrap, hence coded pp_add and pp_subtract on the assumption that
689 everything perl builds on would be happy. After much wailing and
690 gnashing of teeth it would seem that irix64 knows its ANSI spec well,
691 knows that it doesn't need to, and doesn't. Bah. Anyway, the all-
692 unsigned code below is actually shorter than the old code. :-)
693 */
694
01f91bf2 695 if (SvIV_please_nomg(svr)) {
28e5dec8
JH
696 /* Unless the left argument is integer in range we are going to have to
697 use NV maths. Hence only attempt to coerce the right argument if
698 we know the left is integer. */
eb578fdb 699 UV auv = 0;
9c5ffd7c 700 bool auvok = FALSE;
7dca457a
NC
701 bool a_valid = 0;
702
28e5dec8 703 if (!useleft) {
7dca457a
NC
704 auv = 0;
705 a_valid = auvok = 1;
706 /* left operand is undef, treat as zero. + 0 is identity,
707 Could SETi or SETu right now, but space optimise by not adding
708 lots of code to speed up what is probably a rarish case. */
709 } else {
710 /* Left operand is defined, so is it IV? */
01f91bf2 711 if (SvIV_please_nomg(svl)) {
800401ee
JH
712 if ((auvok = SvUOK(svl)))
713 auv = SvUVX(svl);
7dca457a 714 else {
eb578fdb 715 const IV aiv = SvIVX(svl);
7dca457a
NC
716 if (aiv >= 0) {
717 auv = aiv;
718 auvok = 1; /* Now acting as a sign flag. */
53e2bfb7
DM
719 } else {
720 auv = (aiv == IV_MIN) ? (UV)aiv : (UV)(-aiv);
7dca457a
NC
721 }
722 }
723 a_valid = 1;
28e5dec8
JH
724 }
725 }
7dca457a
NC
726 if (a_valid) {
727 bool result_good = 0;
728 UV result;
eb578fdb 729 UV buv;
800401ee 730 bool buvok = SvUOK(svr);
a00b5bd3 731
7dca457a 732 if (buvok)
800401ee 733 buv = SvUVX(svr);
7dca457a 734 else {
eb578fdb 735 const IV biv = SvIVX(svr);
7dca457a
NC
736 if (biv >= 0) {
737 buv = biv;
738 buvok = 1;
739 } else
53e2bfb7 740 buv = (biv == IV_MIN) ? (UV)biv : (UV)(-biv);
7dca457a
NC
741 }
742 /* ?uvok if value is >= 0. basically, flagged as UV if it's +ve,
602f51c4 743 else "IV" now, independent of how it came in.
7dca457a
NC
744 if a, b represents positive, A, B negative, a maps to -A etc
745 a + b => (a + b)
746 A + b => -(a - b)
747 a + B => (a - b)
748 A + B => -(a + b)
749 all UV maths. negate result if A negative.
750 add if signs same, subtract if signs differ. */
751
752 if (auvok ^ buvok) {
753 /* Signs differ. */
754 if (auv >= buv) {
755 result = auv - buv;
756 /* Must get smaller */
757 if (result <= auv)
758 result_good = 1;
759 } else {
760 result = buv - auv;
761 if (result <= buv) {
762 /* result really should be -(auv-buv). as its negation
763 of true value, need to swap our result flag */
764 auvok = !auvok;
765 result_good = 1;
28e5dec8
JH
766 }
767 }
7dca457a
NC
768 } else {
769 /* Signs same */
770 result = auv + buv;
771 if (result >= auv)
772 result_good = 1;
773 }
774 if (result_good) {
775 SP--;
776 if (auvok)
28e5dec8 777 SETu( result );
7dca457a
NC
778 else {
779 /* Negate result */
780 if (result <= (UV)IV_MIN)
53e2bfb7
DM
781 SETi(result == (UV)IV_MIN
782 ? IV_MIN : -(IV)result);
7dca457a
NC
783 else {
784 /* result valid, but out of range for IV. */
785 SETn( -(NV)result );
28e5dec8
JH
786 }
787 }
7dca457a
NC
788 RETURN;
789 } /* Overflow, drop through to NVs. */
28e5dec8
JH
790 }
791 }
230ee21f
DM
792
793#else
794 useleft = USE_LEFT(svl);
28e5dec8 795#endif
230ee21f 796
a0d0e21e 797 {
6f1401dc 798 NV value = SvNV_nomg(svr);
4efa5a16 799 (void)POPs;
28e5dec8
JH
800 if (!useleft) {
801 /* left operand is undef, treat as zero. + 0.0 is identity. */
802 SETn(value);
803 RETURN;
804 }
6f1401dc 805 SETn( value + SvNV_nomg(svl) );
28e5dec8 806 RETURN;
a0d0e21e
LW
807 }
808}
809
b1c05ba5
DM
810
811/* also used for: pp_aelemfast_lex() */
812
a0d0e21e
LW
813PP(pp_aelemfast)
814{
20b7effb 815 dSP;
93bad3fd 816 AV * const av = PL_op->op_type == OP_AELEMFAST_LEX
8f878375 817 ? MUTABLE_AV(PAD_SV(PL_op->op_targ)) : GvAVn(cGVOP_gv);
a3b680e6 818 const U32 lval = PL_op->op_flags & OPf_MOD;
b024352e 819 SV** const svp = av_fetch(av, (I8)PL_op->op_private, lval);
3280af22 820 SV *sv = (svp ? *svp : &PL_sv_undef);
b024352e
DM
821
822 if (UNLIKELY(!svp && lval))
823 DIE(aTHX_ PL_no_aelem, (int)(I8)PL_op->op_private);
824
6ff81951 825 EXTEND(SP, 1);
39cf747a 826 if (!lval && SvRMAGICAL(av) && SvGMAGICAL(sv)) /* see note in pp_helem() */
fd69380d 827 mg_get(sv);
be6c24e0 828 PUSHs(sv);
a0d0e21e
LW
829 RETURN;
830}
831
832PP(pp_join)
833{
20b7effb 834 dSP; dMARK; dTARGET;
a0d0e21e
LW
835 MARK++;
836 do_join(TARG, *MARK, MARK, SP);
837 SP = MARK;
838 SETs(TARG);
839 RETURN;
840}
841
842PP(pp_pushre)
843{
20b7effb 844 dSP;
44a8e56a
PP
845#ifdef DEBUGGING
846 /*
847 * We ass_u_me that LvTARGOFF() comes first, and that two STRLENs
848 * will be enough to hold an OP*.
849 */
c4420975 850 SV* const sv = sv_newmortal();
44a8e56a
PP
851 sv_upgrade(sv, SVt_PVLV);
852 LvTYPE(sv) = '/';
533c011a 853 Copy(&PL_op, &LvTARGOFF(sv), 1, OP*);
44a8e56a
PP
854 XPUSHs(sv);
855#else
ad64d0ec 856 XPUSHs(MUTABLE_SV(PL_op));
44a8e56a 857#endif
a0d0e21e
LW
858 RETURN;
859}
860
861/* Oversized hot code. */
862
b1c05ba5
DM
863/* also used for: pp_say() */
864
a0d0e21e
LW
865PP(pp_print)
866{
20b7effb 867 dSP; dMARK; dORIGMARK;
eb578fdb 868 PerlIO *fp;
236988e4 869 MAGIC *mg;
159b6efe
NC
870 GV * const gv
871 = (PL_op->op_flags & OPf_STACKED) ? MUTABLE_GV(*++MARK) : PL_defoutgv;
9c9f25b8 872 IO *io = GvIO(gv);
5b468f54 873
9c9f25b8 874 if (io
ad64d0ec 875 && (mg = SvTIED_mg((const SV *)io, PERL_MAGIC_tiedscalar)))
5b468f54 876 {
01bb7c6d 877 had_magic:
68dc0745 878 if (MARK == ORIGMARK) {
1c846c1f 879 /* If using default handle then we need to make space to
a60c0954
NIS
880 * pass object as 1st arg, so move other args up ...
881 */
4352c267 882 MEXTEND(SP, 1);
68dc0745
PP
883 ++MARK;
884 Move(MARK, MARK + 1, (SP - MARK) + 1, SV*);
885 ++SP;
886 }
3e0cb5de 887 return Perl_tied_method(aTHX_ SV_CONST(PRINT), mark - 1, MUTABLE_SV(io),
94bc412f
NC
888 mg,
889 (G_SCALAR | TIED_METHOD_ARGUMENTS_ON_STACK
890 | (PL_op->op_type == OP_SAY
891 ? TIED_METHOD_SAY : 0)), sp - mark);
236988e4 892 }
9c9f25b8 893 if (!io) {
68b590d9 894 if ( gv && GvEGVx(gv) && (io = GvIO(GvEGV(gv)))
ad64d0ec 895 && (mg = SvTIED_mg((const SV *)io, PERL_MAGIC_tiedscalar)))
01bb7c6d 896 goto had_magic;
51087808 897 report_evil_fh(gv);
93189314 898 SETERRNO(EBADF,RMS_IFI);
a0d0e21e
LW
899 goto just_say_no;
900 }
901 else if (!(fp = IoOFP(io))) {
7716c5c5
NC
902 if (IoIFP(io))
903 report_wrongway_fh(gv, '<');
51087808 904 else
7716c5c5 905 report_evil_fh(gv);
93189314 906 SETERRNO(EBADF,IoIFP(io)?RMS_FAC:RMS_IFI);
a0d0e21e
LW
907 goto just_say_no;
908 }
909 else {
e23d9e2f 910 SV * const ofs = GvSV(PL_ofsgv); /* $, */
a0d0e21e 911 MARK++;
e23d9e2f 912 if (ofs && (SvGMAGICAL(ofs) || SvOK(ofs))) {
a0d0e21e
LW
913 while (MARK <= SP) {
914 if (!do_print(*MARK, fp))
915 break;
916 MARK++;
917 if (MARK <= SP) {
e23d9e2f
CS
918 /* don't use 'ofs' here - it may be invalidated by magic callbacks */
919 if (!do_print(GvSV(PL_ofsgv), fp)) {
a0d0e21e
LW
920 MARK--;
921 break;
922 }
923 }
924 }
925 }
926 else {
927 while (MARK <= SP) {
928 if (!do_print(*MARK, fp))
929 break;
930 MARK++;
931 }
932 }
933 if (MARK <= SP)
934 goto just_say_no;
935 else {
cfc4a7da
GA
936 if (PL_op->op_type == OP_SAY) {
937 if (PerlIO_write(fp, "\n", 1) == 0 || PerlIO_error(fp))
938 goto just_say_no;
939 }
940 else if (PL_ors_sv && SvOK(PL_ors_sv))
7889fe52 941 if (!do_print(PL_ors_sv, fp)) /* $\ */
a0d0e21e
LW
942 goto just_say_no;
943
944 if (IoFLAGS(io) & IOf_FLUSH)
760ac839 945 if (PerlIO_flush(fp) == EOF)
a0d0e21e
LW
946 goto just_say_no;
947 }
948 }
949 SP = ORIGMARK;
e52fd6f4 950 XPUSHs(&PL_sv_yes);
a0d0e21e
LW
951 RETURN;
952
953 just_say_no:
954 SP = ORIGMARK;
e52fd6f4 955 XPUSHs(&PL_sv_undef);
a0d0e21e
LW
956 RETURN;
957}
958
b1c05ba5
DM
959
960/* also used for: pp_rv2hv() */
bdaf10a5 961/* also called directly by pp_lvavref */
b1c05ba5 962
a0d0e21e
LW
963PP(pp_rv2av)
964{
20b7effb 965 dSP; dTOPss;
1c23e2bd 966 const U8 gimme = GIMME_V;
13c59d41
MH
967 static const char an_array[] = "an ARRAY";
968 static const char a_hash[] = "a HASH";
bdaf10a5
FC
969 const bool is_pp_rv2av = PL_op->op_type == OP_RV2AV
970 || PL_op->op_type == OP_LVAVREF;
d83b45b8 971 const svtype type = is_pp_rv2av ? SVt_PVAV : SVt_PVHV;
a0d0e21e 972
9026059d 973 SvGETMAGIC(sv);
a0d0e21e 974 if (SvROK(sv)) {
5d9574c1 975 if (UNLIKELY(SvAMAGIC(sv))) {
93d7320b 976 sv = amagic_deref_call(sv, is_pp_rv2av ? to_av_amg : to_hv_amg);
93d7320b 977 }
17ab7946 978 sv = SvRV(sv);
5d9574c1 979 if (UNLIKELY(SvTYPE(sv) != type))
dcbac5bb 980 /* diag_listed_as: Not an ARRAY reference */
13c59d41 981 DIE(aTHX_ "Not %s reference", is_pp_rv2av ? an_array : a_hash);
5d9574c1
DM
982 else if (UNLIKELY(PL_op->op_flags & OPf_MOD
983 && PL_op->op_private & OPpLVAL_INTRO))
3da99855 984 Perl_croak(aTHX_ "%s", PL_no_localize_ref);
a0d0e21e 985 }
5d9574c1 986 else if (UNLIKELY(SvTYPE(sv) != type)) {
67955e0c 987 GV *gv;
1c846c1f 988
6e592b3a 989 if (!isGV_with_GP(sv)) {
13c59d41 990 gv = Perl_softref2xv(aTHX_ sv, is_pp_rv2av ? an_array : a_hash,
dc3c76f8
NC
991 type, &sp);
992 if (!gv)
993 RETURN;
35cd451c
GS
994 }
995 else {
159b6efe 996 gv = MUTABLE_GV(sv);
a0d0e21e 997 }
ad64d0ec 998 sv = is_pp_rv2av ? MUTABLE_SV(GvAVn(gv)) : MUTABLE_SV(GvHVn(gv));
533c011a 999 if (PL_op->op_private & OPpLVAL_INTRO)
ad64d0ec 1000 sv = is_pp_rv2av ? MUTABLE_SV(save_ary(gv)) : MUTABLE_SV(save_hash(gv));
9f527363
FC
1001 }
1002 if (PL_op->op_flags & OPf_REF) {
17ab7946 1003 SETs(sv);
a0d0e21e 1004 RETURN;
9f527363 1005 }
5d9574c1 1006 else if (UNLIKELY(PL_op->op_private & OPpMAYBE_LVSUB)) {
40c94d11
FC
1007 const I32 flags = is_lvalue_sub();
1008 if (flags && !(flags & OPpENTERSUB_INARGS)) {
cde874ca 1009 if (gimme != G_ARRAY)
042560a6 1010 goto croak_cant_return;
17ab7946 1011 SETs(sv);
78f9721b 1012 RETURN;
40c94d11 1013 }
a0d0e21e
LW
1014 }
1015
17ab7946 1016 if (is_pp_rv2av) {
502c6561 1017 AV *const av = MUTABLE_AV(sv);
636fe681 1018 /* The guts of pp_rv2av */
96913b52 1019 if (gimme == G_ARRAY) {
d5524600
DM
1020 SP--;
1021 PUTBACK;
1022 S_pushav(aTHX_ av);
1023 SPAGAIN;
1c846c1f 1024 }
96913b52
VP
1025 else if (gimme == G_SCALAR) {
1026 dTARGET;
c70927a6 1027 const SSize_t maxarg = AvFILL(av) + 1;
96913b52 1028 SETi(maxarg);
93965878 1029 }
17ab7946
NC
1030 } else {
1031 /* The guts of pp_rv2hv */
96913b52
VP
1032 if (gimme == G_ARRAY) { /* array wanted */
1033 *PL_stack_sp = sv;
981b7185 1034 return Perl_do_kv(aTHX);
96913b52 1035 }
c8fe3bdf 1036 else if ((PL_op->op_private & OPpTRUEBOOL
adc42c31 1037 || ( PL_op->op_private & OPpMAYBE_TRUEBOOL
c8fe3bdf
FC
1038 && block_gimme() == G_VOID ))
1039 && (!SvRMAGICAL(sv) || !mg_find(sv, PERL_MAGIC_tied)))
1040 SETs(HvUSEDKEYS(sv) ? &PL_sv_yes : sv_2mortal(newSViv(0)));
96913b52 1041 else if (gimme == G_SCALAR) {
1a8bdda9 1042 dTARG;
96913b52 1043 TARG = Perl_hv_scalar(aTHX_ MUTABLE_HV(sv));
96913b52
VP
1044 SETTARG;
1045 }
17ab7946 1046 }
be85d344 1047 RETURN;
042560a6
NC
1048
1049 croak_cant_return:
1050 Perl_croak(aTHX_ "Can't return %s to lvalue scalar context",
1051 is_pp_rv2av ? "array" : "hash");
77e217c6 1052 RETURN;
a0d0e21e
LW
1053}
1054
10c8fecd 1055STATIC void
fb8f4cf8 1056S_do_oddball(pTHX_ SV **oddkey, SV **firstkey)
10c8fecd 1057{
7918f24d
NC
1058 PERL_ARGS_ASSERT_DO_ODDBALL;
1059
fb8f4cf8 1060 if (*oddkey) {
6d822dc4 1061 if (ckWARN(WARN_MISC)) {
a3b680e6 1062 const char *err;
fb8f4cf8
RZ
1063 if (oddkey == firstkey &&
1064 SvROK(*oddkey) &&
1065 (SvTYPE(SvRV(*oddkey)) == SVt_PVAV ||
1066 SvTYPE(SvRV(*oddkey)) == SVt_PVHV))
10c8fecd 1067 {
a3b680e6 1068 err = "Reference found where even-sized list expected";
10c8fecd
GS
1069 }
1070 else
a3b680e6 1071 err = "Odd number of elements in hash assignment";
f1f66076 1072 Perl_warner(aTHX_ packWARN(WARN_MISC), "%s", err);
10c8fecd 1073 }
6d822dc4 1074
10c8fecd
GS
1075 }
1076}
1077
a5f48505
DM
1078
1079/* Do a mark and sweep with the SVf_BREAK flag to detect elements which
1080 * are common to both the LHS and RHS of an aassign, and replace them
1081 * with copies. All these copies are made before the actual list assign is
1082 * done.
1083 *
1084 * For example in ($a,$b) = ($b,$a), assigning the value of the first RHS
1085 * element ($b) to the first LH element ($a), modifies $a; when the
1086 * second assignment is done, the second RH element now has the wrong
1087 * value. So we initially replace the RHS with ($b, mortalcopy($a)).
1088 * Note that we don't need to make a mortal copy of $b.
1089 *
1090 * The algorithm below works by, for every RHS element, mark the
1091 * corresponding LHS target element with SVf_BREAK. Then if the RHS
1092 * element is found with SVf_BREAK set, it means it would have been
1093 * modified, so make a copy.
1094 * Note that by scanning both LHS and RHS in lockstep, we avoid
1095 * unnecessary copies (like $b above) compared with a naive
1096 * "mark all LHS; copy all marked RHS; unmark all LHS".
1097 *
1098 * If the LHS element is a 'my' declaration' and has a refcount of 1, then
1099 * it can't be common and can be skipped.
ebc643ce
DM
1100 *
1101 * On DEBUGGING builds it takes an extra boolean, fake. If true, it means
1102 * that we thought we didn't need to call S_aassign_copy_common(), but we
1103 * have anyway for sanity checking. If we find we need to copy, then panic.
a5f48505
DM
1104 */
1105
1106PERL_STATIC_INLINE void
1107S_aassign_copy_common(pTHX_ SV **firstlelem, SV **lastlelem,
ebc643ce
DM
1108 SV **firstrelem, SV **lastrelem
1109#ifdef DEBUGGING
1110 , bool fake
1111#endif
1112)
a5f48505
DM
1113{
1114 dVAR;
1115 SV **relem;
1116 SV **lelem;
1117 SSize_t lcount = lastlelem - firstlelem + 1;
1118 bool marked = FALSE; /* have we marked any LHS with SVf_BREAK ? */
1119 bool const do_rc1 = cBOOL(PL_op->op_private & OPpASSIGN_COMMON_RC1);
beb08a1e 1120 bool copy_all = FALSE;
a5f48505
DM
1121
1122 assert(!PL_in_clean_all); /* SVf_BREAK not already in use */
1123 assert(firstlelem < lastlelem); /* at least 2 LH elements */
1124 assert(firstrelem < lastrelem); /* at least 2 RH elements */
1125
ebc643ce
DM
1126
1127 lelem = firstlelem;
a5f48505
DM
1128 /* we never have to copy the first RH element; it can't be corrupted
1129 * by assigning something to the corresponding first LH element.
1130 * So this scan does in a loop: mark LHS[N]; test RHS[N+1]
1131 */
ebc643ce 1132 relem = firstrelem + 1;
a5f48505
DM
1133
1134 for (; relem <= lastrelem; relem++) {
1135 SV *svr;
1136
1137 /* mark next LH element */
1138
1139 if (--lcount >= 0) {
1140 SV *svl = *lelem++;
1141
1142 if (UNLIKELY(!svl)) {/* skip AV alias marker */
1143 assert (lelem <= lastlelem);
1144 svl = *lelem++;
1145 lcount--;
1146 }
1147
1148 assert(svl);
beb08a1e
TC
1149 if (SvSMAGICAL(svl)) {
1150 copy_all = TRUE;
1151 }
a5f48505
DM
1152 if (SvTYPE(svl) == SVt_PVAV || SvTYPE(svl) == SVt_PVHV) {
1153 if (!marked)
1154 return;
1155 /* this LH element will consume all further args;
1156 * no need to mark any further LH elements (if any).
1157 * But we still need to scan any remaining RHS elements;
1158 * set lcount negative to distinguish from lcount == 0,
1159 * so the loop condition continues being true
1160 */
1161 lcount = -1;
1162 lelem--; /* no need to unmark this element */
1163 }
1164 else if (!(do_rc1 && SvREFCNT(svl) == 1) && svl != &PL_sv_undef) {
1165 assert(!SvIMMORTAL(svl));
1166 SvFLAGS(svl) |= SVf_BREAK;
1167 marked = TRUE;
1168 }
1169 else if (!marked) {
1170 /* don't check RH element if no SVf_BREAK flags set yet */
1171 if (!lcount)
1172 break;
1173 continue;
1174 }
1175 }
1176
1177 /* see if corresponding RH element needs copying */
1178
1179 assert(marked);
1180 svr = *relem;
1181 assert(svr);
1182
5c1db569 1183 if (UNLIKELY(SvFLAGS(svr) & (SVf_BREAK|SVs_GMG) || copy_all)) {
a5f48505 1184
ebc643ce
DM
1185#ifdef DEBUGGING
1186 if (fake) {
9ae0115f 1187 /* op_dump(PL_op); */
ebc643ce
DM
1188 Perl_croak(aTHX_
1189 "panic: aassign skipped needed copy of common RH elem %"
1190 UVuf, (UV)(relem - firstrelem));
1191 }
1192#endif
1193
a5f48505
DM
1194 TAINT_NOT; /* Each item is independent */
1195
1196 /* Dear TODO test in t/op/sort.t, I love you.
1197 (It's relying on a panic, not a "semi-panic" from newSVsv()
1198 and then an assertion failure below.) */
1199 if (UNLIKELY(SvIS_FREED(svr))) {
1200 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p",
1201 (void*)svr);
1202 }
1203 /* avoid break flag while copying; otherwise COW etc
1204 * disabled... */
1205 SvFLAGS(svr) &= ~SVf_BREAK;
1206 /* Not newSVsv(), as it does not allow copy-on-write,
8c1e192f
DM
1207 resulting in wasteful copies.
1208 Also, we use SV_NOSTEAL in case the SV is used more than
1209 once, e.g. (...) = (f())[0,0]
1210 Where the same SV appears twice on the RHS without a ref
1211 count bump. (Although I suspect that the SV won't be
1212 stealable here anyway - DAPM).
1213 */
a5f48505
DM
1214 *relem = sv_mortalcopy_flags(svr,
1215 SV_GMAGIC|SV_DO_COW_SVSETSV|SV_NOSTEAL);
1216 /* ... but restore afterwards in case it's needed again,
1217 * e.g. ($a,$b,$c) = (1,$a,$a)
1218 */
1219 SvFLAGS(svr) |= SVf_BREAK;
1220 }
1221
1222 if (!lcount)
1223 break;
1224 }
1225
1226 if (!marked)
1227 return;
1228
1229 /*unmark LHS */
1230
1231 while (lelem > firstlelem) {
1232 SV * const svl = *(--lelem);
1233 if (svl)
1234 SvFLAGS(svl) &= ~SVf_BREAK;
1235 }
1236}
1237
1238
1239
a0d0e21e
LW
1240PP(pp_aassign)
1241{
27da23d5 1242 dVAR; dSP;
3280af22
NIS
1243 SV **lastlelem = PL_stack_sp;
1244 SV **lastrelem = PL_stack_base + POPMARK;
1245 SV **firstrelem = PL_stack_base + POPMARK + 1;
a0d0e21e
LW
1246 SV **firstlelem = lastrelem + 1;
1247
eb578fdb
KW
1248 SV **relem;
1249 SV **lelem;
a0d0e21e 1250
eb578fdb
KW
1251 SV *sv;
1252 AV *ary;
a0d0e21e 1253
1c23e2bd 1254 U8 gimme;
a0d0e21e 1255 HV *hash;
c70927a6 1256 SSize_t i;
a0d0e21e 1257 int magic;
a5f48505 1258 U32 lval;
a68090fe
DM
1259 /* PL_delaymagic is restored by JUMPENV_POP on dieing, so we
1260 * only need to save locally, not on the save stack */
1261 U16 old_delaymagic = PL_delaymagic;
ebc643ce
DM
1262#ifdef DEBUGGING
1263 bool fake = 0;
1264#endif
5637b936 1265
3280af22 1266 PL_delaymagic = DM_DELAY; /* catch simultaneous items */
a0d0e21e
LW
1267
1268 /* If there's a common identifier on both sides we have to take
1269 * special care that assigning the identifier on the left doesn't
1270 * clobber a value on the right that's used later in the list.
1271 */
acdea6f0 1272
beb08a1e
TC
1273 /* at least 2 LH and RH elements, or commonality isn't an issue */
1274 if (firstlelem < lastlelem && firstrelem < lastrelem) {
5c1db569
TC
1275 for (relem = firstrelem+1; relem <= lastrelem; relem++) {
1276 if (SvGMAGICAL(*relem))
1277 goto do_scan;
1278 }
beb08a1e
TC
1279 for (lelem = firstlelem; lelem <= lastlelem; lelem++) {
1280 if (*lelem && SvSMAGICAL(*lelem))
1281 goto do_scan;
a5f48505 1282 }
beb08a1e
TC
1283 if ( PL_op->op_private & (OPpASSIGN_COMMON_SCALAR|OPpASSIGN_COMMON_RC1) ) {
1284 if (PL_op->op_private & OPpASSIGN_COMMON_RC1) {
1285 /* skip the scan if all scalars have a ref count of 1 */
1286 for (lelem = firstlelem; lelem <= lastlelem; lelem++) {
1287 sv = *lelem;
1288 if (!sv || SvREFCNT(sv) == 1)
1289 continue;
1290 if (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVAV)
1291 goto do_scan;
1292 break;
1293 }
1294 }
1295 else {
1296 do_scan:
1297 S_aassign_copy_common(aTHX_
1298 firstlelem, lastlelem, firstrelem, lastrelem
ebc643ce 1299#ifdef DEBUGGING
beb08a1e 1300 , fake
ebc643ce 1301#endif
beb08a1e
TC
1302 );
1303 }
a5f48505 1304 }
a0d0e21e 1305 }
ebc643ce
DM
1306#ifdef DEBUGGING
1307 else {
1308 /* on debugging builds, do the scan even if we've concluded we
1309 * don't need to, then panic if we find commonality. Note that the
1310 * scanner assumes at least 2 elements */
1311 if (firstlelem < lastlelem && firstrelem < lastrelem) {
1312 fake = 1;
1313 goto do_scan;
1314 }
1315 }
1316#endif
a0d0e21e 1317
a5f48505
DM
1318 gimme = GIMME_V;
1319 lval = (gimme == G_ARRAY) ? (PL_op->op_flags & OPf_MOD || LVRET) : 0;
1320
a0d0e21e
LW
1321 relem = firstrelem;
1322 lelem = firstlelem;
4608196e
RGS
1323 ary = NULL;
1324 hash = NULL;
10c8fecd 1325
5d9574c1 1326 while (LIKELY(lelem <= lastlelem)) {
bdaf10a5 1327 bool alias = FALSE;
bbce6d69 1328 TAINT_NOT; /* Each item stands on its own, taintwise. */
a0d0e21e 1329 sv = *lelem++;
bdaf10a5
FC
1330 if (UNLIKELY(!sv)) {
1331 alias = TRUE;
1332 sv = *lelem++;
1333 ASSUME(SvTYPE(sv) == SVt_PVAV);
1334 }
a0d0e21e 1335 switch (SvTYPE(sv)) {
a5f48505
DM
1336 case SVt_PVAV: {
1337 bool already_copied = FALSE;
60edcf09 1338 ary = MUTABLE_AV(sv);
748a9306 1339 magic = SvMAGICAL(ary) != 0;
60edcf09
FC
1340 ENTER;
1341 SAVEFREESV(SvREFCNT_inc_simple_NN(sv));
a5f48505
DM
1342
1343 /* We need to clear ary. The is a danger that if we do this,
1344 * elements on the RHS may be prematurely freed, e.g.
1345 * @a = ($a[0]);
1346 * In the case of possible commonality, make a copy of each
1347 * RHS SV *before* clearing the array, and add a reference
1348 * from the tmps stack, so that it doesn't leak on death.
1349 * Otherwise, make a copy of each RHS SV only as we're storing
1350 * it into the array - that way we don't have to worry about
1351 * it being leaked if we die, but don't incur the cost of
1352 * mortalising everything.
1353 */
1354
1355 if ( (PL_op->op_private & OPpASSIGN_COMMON_AGG)
1356 && (relem <= lastrelem)
1357 && (magic || AvFILL(ary) != -1))
1358 {
1359 SV **svp;
1360 EXTEND_MORTAL(lastrelem - relem + 1);
1361 for (svp = relem; svp <= lastrelem; svp++) {
8c1e192f 1362 /* see comment in S_aassign_copy_common about SV_NOSTEAL */
a5f48505
DM
1363 *svp = sv_mortalcopy_flags(*svp,
1364 SV_GMAGIC|SV_DO_COW_SVSETSV|SV_NOSTEAL);
1365 TAINT_NOT;
1366 }
1367 already_copied = TRUE;
1368 }
1369
1370 av_clear(ary);
6768377c
DM
1371 if (relem <= lastrelem)
1372 av_extend(ary, lastrelem - relem);
1373
a0d0e21e
LW
1374 i = 0;
1375 while (relem <= lastrelem) { /* gobble up all the rest */
5117ca91 1376 SV **didstore;
bdaf10a5 1377 if (LIKELY(!alias)) {
a5f48505
DM
1378 if (already_copied)
1379 sv = *relem;
1380 else {
1381 if (LIKELY(*relem))
1382 /* before newSV, in case it dies */
1383 SvGETMAGIC(*relem);
1384 sv = newSV(0);
8c1e192f
DM
1385 /* see comment in S_aassign_copy_common about
1386 * SV_NOSTEAL */
1387 sv_setsv_flags(sv, *relem,
1388 (SV_DO_COW_SVSETSV|SV_NOSTEAL));
a5f48505
DM
1389 *relem = sv;
1390 }
bdaf10a5
FC
1391 }
1392 else {
a5f48505
DM
1393 if (!already_copied)
1394 SvGETMAGIC(*relem);
bdaf10a5
FC
1395 if (!SvROK(*relem))
1396 DIE(aTHX_ "Assigned value is not a reference");
1397 if (SvTYPE(SvRV(*relem)) > SVt_PVLV)
1398 /* diag_listed_as: Assigned value is not %s reference */
1399 DIE(aTHX_
1400 "Assigned value is not a SCALAR reference");
a5f48505 1401 if (lval && !already_copied)
bdaf10a5
FC
1402 *relem = sv_mortalcopy(*relem);
1403 /* XXX else check for weak refs? */
0ad694a7 1404 sv = SvREFCNT_inc_NN(SvRV(*relem));
bdaf10a5
FC
1405 }
1406 relem++;
a5f48505 1407 if (already_copied)
0ad694a7 1408 SvREFCNT_inc_simple_void_NN(sv); /* undo mortal free */
5117ca91
GS
1409 didstore = av_store(ary,i++,sv);
1410 if (magic) {
18024492
FC
1411 if (!didstore)
1412 sv_2mortal(sv);
8ef24240 1413 if (SvSMAGICAL(sv))
fb73857a 1414 mg_set(sv);
5117ca91 1415 }
bbce6d69 1416 TAINT_NOT;
a0d0e21e 1417 }
5d9574c1 1418 if (UNLIKELY(PL_delaymagic & DM_ARRAY_ISA))
ad64d0ec 1419 SvSETMAGIC(MUTABLE_SV(ary));
60edcf09 1420 LEAVE;
a0d0e21e 1421 break;
a5f48505
DM
1422 }
1423
10c8fecd 1424 case SVt_PVHV: { /* normal hash */
a0d0e21e 1425 SV *tmpstr;
1c4ea384
RZ
1426 int odd;
1427 int duplicates = 0;
45960564 1428 SV** topelem = relem;
1c4ea384 1429 SV **firsthashrelem = relem;
a5f48505 1430 bool already_copied = FALSE;
a0d0e21e 1431
60edcf09 1432 hash = MUTABLE_HV(sv);
748a9306 1433 magic = SvMAGICAL(hash) != 0;
1c4ea384
RZ
1434
1435 odd = ((lastrelem - firsthashrelem)&1)? 0 : 1;
5d9574c1 1436 if (UNLIKELY(odd)) {
fb8f4cf8 1437 do_oddball(lastrelem, firsthashrelem);
1d2b3927
HS
1438 /* we have firstlelem to reuse, it's not needed anymore
1439 */
1c4ea384
RZ
1440 *(lastrelem+1) = &PL_sv_undef;
1441 }
1442
60edcf09
FC
1443 ENTER;
1444 SAVEFREESV(SvREFCNT_inc_simple_NN(sv));
a5f48505
DM
1445
1446 /* We need to clear hash. The is a danger that if we do this,
1447 * elements on the RHS may be prematurely freed, e.g.
1448 * %h = (foo => $h{bar});
1449 * In the case of possible commonality, make a copy of each
1450 * RHS SV *before* clearing the hash, and add a reference
1451 * from the tmps stack, so that it doesn't leak on death.
1452 */
1453
1454 if ( (PL_op->op_private & OPpASSIGN_COMMON_AGG)
1455 && (relem <= lastrelem)
1456 && (magic || HvUSEDKEYS(hash)))
1457 {
1458 SV **svp;
1459 EXTEND_MORTAL(lastrelem - relem + 1);
1460 for (svp = relem; svp <= lastrelem; svp++) {
1461 *svp = sv_mortalcopy_flags(*svp,
1462 SV_GMAGIC|SV_DO_COW_SVSETSV|SV_NOSTEAL);
1463 TAINT_NOT;
1464 }
1465 already_copied = TRUE;
1466 }
1467
a0d0e21e 1468 hv_clear(hash);
a5f48505 1469
5d9574c1 1470 while (LIKELY(relem < lastrelem+odd)) { /* gobble up all the rest */
5117ca91 1471 HE *didstore;
1c4ea384 1472 assert(*relem);
632b9d6f
FC
1473 /* Copy the key if aassign is called in lvalue context,
1474 to avoid having the next op modify our rhs. Copy
1475 it also if it is gmagical, lest it make the
1476 hv_store_ent call below croak, leaking the value. */
a5f48505 1477 sv = (lval || SvGMAGICAL(*relem)) && !already_copied
632b9d6f
FC
1478 ? sv_mortalcopy(*relem)
1479 : *relem;
45960564 1480 relem++;
1c4ea384 1481 assert(*relem);
a5f48505
DM
1482 if (already_copied)
1483 tmpstr = *relem++;
1484 else {
1485 SvGETMAGIC(*relem);
1486 tmpstr = newSV(0);
1487 sv_setsv_nomg(tmpstr,*relem++); /* value */
1488 }
1489
a88bf2bc 1490 if (gimme == G_ARRAY) {
45960564
DM
1491 if (hv_exists_ent(hash, sv, 0))
1492 /* key overwrites an existing entry */
1493 duplicates += 2;
a88bf2bc 1494 else {
45960564 1495 /* copy element back: possibly to an earlier
1d2b3927
HS
1496 * stack location if we encountered dups earlier,
1497 * possibly to a later stack location if odd */
45960564
DM
1498 *topelem++ = sv;
1499 *topelem++ = tmpstr;
1500 }
1501 }
a5f48505 1502 if (already_copied)
0ad694a7 1503 SvREFCNT_inc_simple_void_NN(tmpstr); /* undo mortal free */
5117ca91 1504 didstore = hv_store_ent(hash,sv,tmpstr,0);
632b9d6f
FC
1505 if (magic) {
1506 if (!didstore) sv_2mortal(tmpstr);
1507 SvSETMAGIC(tmpstr);
1508 }
bbce6d69 1509 TAINT_NOT;
8e07c86e 1510 }
60edcf09 1511 LEAVE;
1c4ea384
RZ
1512 if (duplicates && gimme == G_ARRAY) {
1513 /* at this point we have removed the duplicate key/value
1514 * pairs from the stack, but the remaining values may be
1515 * wrong; i.e. with (a 1 a 2 b 3) on the stack we've removed
1516 * the (a 2), but the stack now probably contains
1517 * (a <freed> b 3), because { hv_save(a,1); hv_save(a,2) }
1518 * obliterates the earlier key. So refresh all values. */
1519 lastrelem -= duplicates;
1520 relem = firsthashrelem;
1521 while (relem < lastrelem+odd) {
1522 HE *he;
1523 he = hv_fetch_ent(hash, *relem++, 0, 0);
1524 *relem++ = (he ? HeVAL(he) : &PL_sv_undef);
1525 }
1526 }
1527 if (odd && gimme == G_ARRAY) lastrelem++;
a0d0e21e
LW
1528 }
1529 break;
1530 default:
6fc92669
GS
1531 if (SvIMMORTAL(sv)) {
1532 if (relem <= lastrelem)
1533 relem++;
1534 break;
a0d0e21e
LW
1535 }
1536 if (relem <= lastrelem) {
5d9574c1 1537 if (UNLIKELY(
1c70fb82
FC
1538 SvTEMP(sv) && !SvSMAGICAL(sv) && SvREFCNT(sv) == 1 &&
1539 (!isGV_with_GP(sv) || SvFAKE(sv)) && ckWARN(WARN_MISC)
5d9574c1 1540 ))
1c70fb82
FC
1541 Perl_warner(aTHX_
1542 packWARN(WARN_MISC),
1543 "Useless assignment to a temporary"
1544 );
a0d0e21e
LW
1545 sv_setsv(sv, *relem);
1546 *(relem++) = sv;
1547 }
1548 else
3280af22 1549 sv_setsv(sv, &PL_sv_undef);
8ef24240 1550 SvSETMAGIC(sv);
a0d0e21e
LW
1551 break;
1552 }
1553 }
5d9574c1 1554 if (UNLIKELY(PL_delaymagic & ~DM_DELAY)) {
985213f2 1555 /* Will be used to set PL_tainting below */
dfff4baf
BF
1556 Uid_t tmp_uid = PerlProc_getuid();
1557 Uid_t tmp_euid = PerlProc_geteuid();
1558 Gid_t tmp_gid = PerlProc_getgid();
1559 Gid_t tmp_egid = PerlProc_getegid();
985213f2 1560
b469f1e0 1561 /* XXX $> et al currently silently ignore failures */
3280af22 1562 if (PL_delaymagic & DM_UID) {
a0d0e21e 1563#ifdef HAS_SETRESUID
b469f1e0
JH
1564 PERL_UNUSED_RESULT(
1565 setresuid((PL_delaymagic & DM_RUID) ? PL_delaymagic_uid : (Uid_t)-1,
1566 (PL_delaymagic & DM_EUID) ? PL_delaymagic_euid : (Uid_t)-1,
1567 (Uid_t)-1));
56febc5e
AD
1568#else
1569# ifdef HAS_SETREUID
b469f1e0
JH
1570 PERL_UNUSED_RESULT(
1571 setreuid((PL_delaymagic & DM_RUID) ? PL_delaymagic_uid : (Uid_t)-1,
1572 (PL_delaymagic & DM_EUID) ? PL_delaymagic_euid : (Uid_t)-1));
56febc5e
AD
1573# else
1574# ifdef HAS_SETRUID
b28d0864 1575 if ((PL_delaymagic & DM_UID) == DM_RUID) {
b469f1e0 1576 PERL_UNUSED_RESULT(setruid(PL_delaymagic_uid));
b28d0864 1577 PL_delaymagic &= ~DM_RUID;
a0d0e21e 1578 }
56febc5e
AD
1579# endif /* HAS_SETRUID */
1580# ifdef HAS_SETEUID
b28d0864 1581 if ((PL_delaymagic & DM_UID) == DM_EUID) {
b469f1e0 1582 PERL_UNUSED_RESULT(seteuid(PL_delaymagic_euid));
b28d0864 1583 PL_delaymagic &= ~DM_EUID;
a0d0e21e 1584 }
56febc5e 1585# endif /* HAS_SETEUID */
b28d0864 1586 if (PL_delaymagic & DM_UID) {
985213f2 1587 if (PL_delaymagic_uid != PL_delaymagic_euid)
cea2e8a9 1588 DIE(aTHX_ "No setreuid available");
b469f1e0 1589 PERL_UNUSED_RESULT(PerlProc_setuid(PL_delaymagic_uid));
a0d0e21e 1590 }
56febc5e
AD
1591# endif /* HAS_SETREUID */
1592#endif /* HAS_SETRESUID */
04783dc7 1593
985213f2
AB
1594 tmp_uid = PerlProc_getuid();
1595 tmp_euid = PerlProc_geteuid();
a0d0e21e 1596 }
b469f1e0 1597 /* XXX $> et al currently silently ignore failures */
3280af22 1598 if (PL_delaymagic & DM_GID) {
a0d0e21e 1599#ifdef HAS_SETRESGID
b469f1e0
JH
1600 PERL_UNUSED_RESULT(
1601 setresgid((PL_delaymagic & DM_RGID) ? PL_delaymagic_gid : (Gid_t)-1,
1602 (PL_delaymagic & DM_EGID) ? PL_delaymagic_egid : (Gid_t)-1,
1603 (Gid_t)-1));
56febc5e
AD
1604#else
1605# ifdef HAS_SETREGID
b469f1e0
JH
1606 PERL_UNUSED_RESULT(
1607 setregid((PL_delaymagic & DM_RGID) ? PL_delaymagic_gid : (Gid_t)-1,
1608 (PL_delaymagic & DM_EGID) ? PL_delaymagic_egid : (Gid_t)-1));
56febc5e
AD
1609# else
1610# ifdef HAS_SETRGID
b28d0864 1611 if ((PL_delaymagic & DM_GID) == DM_RGID) {
b469f1e0 1612 PERL_UNUSED_RESULT(setrgid(PL_delaymagic_gid));
b28d0864 1613 PL_delaymagic &= ~DM_RGID;
a0d0e21e 1614 }
56febc5e
AD
1615# endif /* HAS_SETRGID */
1616# ifdef HAS_SETEGID
b28d0864 1617 if ((PL_delaymagic & DM_GID) == DM_EGID) {
b469f1e0 1618 PERL_UNUSED_RESULT(setegid(PL_delaymagic_egid));
b28d0864 1619 PL_delaymagic &= ~DM_EGID;
a0d0e21e 1620 }
56febc5e 1621# endif /* HAS_SETEGID */
b28d0864 1622 if (PL_delaymagic & DM_GID) {
985213f2 1623 if (PL_delaymagic_gid != PL_delaymagic_egid)
cea2e8a9 1624 DIE(aTHX_ "No setregid available");
b469f1e0 1625 PERL_UNUSED_RESULT(PerlProc_setgid(PL_delaymagic_gid));
a0d0e21e 1626 }
56febc5e
AD
1627# endif /* HAS_SETREGID */
1628#endif /* HAS_SETRESGID */
04783dc7 1629
985213f2
AB
1630 tmp_gid = PerlProc_getgid();
1631 tmp_egid = PerlProc_getegid();
a0d0e21e 1632 }
284167a5 1633 TAINTING_set( TAINTING_get | (tmp_uid && (tmp_euid != tmp_uid || tmp_egid != tmp_gid)) );
9a9b5ec9
DM
1634#ifdef NO_TAINT_SUPPORT
1635 PERL_UNUSED_VAR(tmp_uid);
1636 PERL_UNUSED_VAR(tmp_euid);
1637 PERL_UNUSED_VAR(tmp_gid);
1638 PERL_UNUSED_VAR(tmp_egid);
1639#endif
a0d0e21e 1640 }
a68090fe 1641 PL_delaymagic = old_delaymagic;
54310121 1642
54310121
PP
1643 if (gimme == G_VOID)
1644 SP = firstrelem - 1;
1645 else if (gimme == G_SCALAR) {
1646 dTARGET;
1647 SP = firstrelem;
231cbeb2 1648 SETi(lastrelem - firstrelem + 1);
54310121
PP
1649 }
1650 else {
1c4ea384 1651 if (ary || hash)
1d2b3927
HS
1652 /* note that in this case *firstlelem may have been overwritten
1653 by sv_undef in the odd hash case */
a0d0e21e 1654 SP = lastrelem;
1c4ea384 1655 else {
a0d0e21e 1656 SP = firstrelem + (lastlelem - firstlelem);
1c4ea384
RZ
1657 lelem = firstlelem + (relem - firstrelem);
1658 while (relem <= SP)
1659 *relem++ = (lelem <= lastlelem) ? *lelem++ : &PL_sv_undef;
1660 }
a0d0e21e 1661 }
08aeb9f7 1662
54310121 1663 RETURN;
a0d0e21e
LW
1664}
1665
8782bef2
GB
1666PP(pp_qr)
1667{
20b7effb 1668 dSP;
eb578fdb 1669 PMOP * const pm = cPMOP;
fe578d7f 1670 REGEXP * rx = PM_GETRE(pm);
10599a69 1671 SV * const pkg = rx ? CALLREG_PACKAGE(rx) : NULL;
c4420975 1672 SV * const rv = sv_newmortal();
d63c20f2
DM
1673 CV **cvp;
1674 CV *cv;
288b8c02
NC
1675
1676 SvUPGRADE(rv, SVt_IV);
c2123ae3
NC
1677 /* For a subroutine describing itself as "This is a hacky workaround" I'm
1678 loathe to use it here, but it seems to be the right fix. Or close.
1679 The key part appears to be that it's essential for pp_qr to return a new
1680 object (SV), which implies that there needs to be an effective way to
1681 generate a new SV from the existing SV that is pre-compiled in the
1682 optree. */
1683 SvRV_set(rv, MUTABLE_SV(reg_temp_copy(NULL, rx)));
288b8c02
NC
1684 SvROK_on(rv);
1685
8d919b0a 1686 cvp = &( ReANY((REGEXP *)SvRV(rv))->qr_anoncv);
5d9574c1 1687 if (UNLIKELY((cv = *cvp) && CvCLONE(*cvp))) {
d63c20f2 1688 *cvp = cv_clone(cv);
fc2b2dca 1689 SvREFCNT_dec_NN(cv);
d63c20f2
DM
1690 }
1691
288b8c02 1692 if (pkg) {
f815daf2 1693 HV *const stash = gv_stashsv(pkg, GV_ADD);
fc2b2dca 1694 SvREFCNT_dec_NN(pkg);
288b8c02
NC
1695 (void)sv_bless(rv, stash);
1696 }
1697
5d9574c1 1698 if (UNLIKELY(RX_ISTAINTED(rx))) {
e08e52cf 1699 SvTAINTED_on(rv);
9274aefd
DM
1700 SvTAINTED_on(SvRV(rv));
1701 }
c8c13c22 1702 XPUSHs(rv);
1703 RETURN;
8782bef2
GB
1704}
1705
a0d0e21e
LW
1706PP(pp_match)
1707{
20b7effb 1708 dSP; dTARG;
eb578fdb 1709 PMOP *pm = cPMOP;
d65afb4b 1710 PMOP *dynpm = pm;
eb578fdb 1711 const char *s;
5c144d81 1712 const char *strend;
99a90e59 1713 SSize_t curpos = 0; /* initial pos() or current $+[0] */
a0d0e21e 1714 I32 global;
7fadf4a7 1715 U8 r_flags = 0;
5c144d81 1716 const char *truebase; /* Start of string */
eb578fdb 1717 REGEXP *rx = PM_GETRE(pm);
b3eb6a9b 1718 bool rxtainted;
1c23e2bd 1719 const U8 gimme = GIMME_V;
a0d0e21e 1720 STRLEN len;
a3b680e6 1721 const I32 oldsave = PL_savestack_ix;
e60df1fa 1722 I32 had_zerolen = 0;
b1422d62 1723 MAGIC *mg = NULL;
a0d0e21e 1724
533c011a 1725 if (PL_op->op_flags & OPf_STACKED)
a0d0e21e 1726 TARG = POPs;
6ffceeb7 1727 else if (ARGTARG)
59f00321 1728 GETTARGET;
a0d0e21e 1729 else {
54b9620d 1730 TARG = DEFSV;
a0d0e21e
LW
1731 EXTEND(SP,1);
1732 }
d9f424b2 1733
c277df42 1734 PUTBACK; /* EVAL blocks need stack_sp. */
69dc4b30
FC
1735 /* Skip get-magic if this is a qr// clone, because regcomp has
1736 already done it. */
f1d31338 1737 truebase = ReANY(rx)->mother_re
69dc4b30
FC
1738 ? SvPV_nomg_const(TARG, len)
1739 : SvPV_const(TARG, len);
f1d31338 1740 if (!truebase)
2269b42e 1741 DIE(aTHX_ "panic: pp_match");
f1d31338 1742 strend = truebase + len;
284167a5
SM
1743 rxtainted = (RX_ISTAINTED(rx) ||
1744 (TAINT_get && (pm->op_pmflags & PMf_RETAINT)));
9212bbba 1745 TAINT_NOT;
a0d0e21e 1746
6c864ec2 1747 /* We need to know this in case we fail out early - pos() must be reset */
de0df3c0
MH
1748 global = dynpm->op_pmflags & PMf_GLOBAL;
1749
d65afb4b 1750 /* PMdf_USED is set after a ?? matches once */
c737faaf
YO
1751 if (
1752#ifdef USE_ITHREADS
1753 SvREADONLY(PL_regex_pad[pm->op_pmoffset])
1754#else
1755 pm->op_pmflags & PMf_USED
1756#endif
1757 ) {
e5dc5375 1758 DEBUG_r(PerlIO_printf(Perl_debug_log, "?? already matched once"));
de0df3c0 1759 goto nope;
a0d0e21e
LW
1760 }
1761
7e313637
FC
1762 /* empty pattern special-cased to use last successful pattern if
1763 possible, except for qr// */
8d919b0a 1764 if (!ReANY(rx)->mother_re && !RX_PRELEN(rx)
7e313637 1765 && PL_curpm) {
3280af22 1766 pm = PL_curpm;
aaa362c4 1767 rx = PM_GETRE(pm);
a0d0e21e 1768 }
d65afb4b 1769
389ecb56 1770 if (RX_MINLEN(rx) >= 0 && (STRLEN)RX_MINLEN(rx) > len) {
75d43e96
FC
1771 DEBUG_r(PerlIO_printf(Perl_debug_log, "String shorter than min possible regex match (%"
1772 UVuf" < %"IVdf")\n",
1773 (UV)len, (IV)RX_MINLEN(rx)));
de0df3c0 1774 goto nope;
e5dc5375 1775 }
c277df42 1776
8ef97b0e 1777 /* get pos() if //g */
de0df3c0 1778 if (global) {
b1422d62 1779 mg = mg_find_mglob(TARG);
8ef97b0e 1780 if (mg && mg->mg_len >= 0) {
25fdce4a 1781 curpos = MgBYTEPOS(mg, TARG, truebase, len);
8ef97b0e
DM
1782 /* last time pos() was set, it was zero-length match */
1783 if (mg->mg_flags & MGf_MINMATCH)
1784 had_zerolen = 1;
1785 }
a0d0e21e 1786 }
8ef97b0e 1787
6e240d0b 1788#ifdef PERL_SAWAMPERSAND
a41aa44d 1789 if ( RX_NPARENS(rx)
6502e081 1790 || PL_sawampersand
6502e081 1791 || (RX_EXTFLAGS(rx) & (RXf_EVAL_SEEN|RXf_PMf_KEEPCOPY))
5b0e71e9 1792 || (dynpm->op_pmflags & PMf_KEEPCOPY)
6e240d0b
FC
1793 )
1794#endif
1795 {
6502e081
DM
1796 r_flags |= (REXEC_COPY_STR|REXEC_COPY_SKIP_PRE);
1797 /* in @a =~ /(.)/g, we iterate multiple times, but copy the buffer
1798 * only on the first iteration. Therefore we need to copy $' as well
1799 * as $&, to make the rest of the string available for captures in
1800 * subsequent iterations */
1801 if (! (global && gimme == G_ARRAY))
1802 r_flags |= REXEC_COPY_SKIP_POST;
1803 };
5b0e71e9
DM
1804#ifdef PERL_SAWAMPERSAND
1805 if (dynpm->op_pmflags & PMf_KEEPCOPY)
1806 /* handle KEEPCOPY in pmop but not rx, eg $r=qr/a/; /$r/p */
1807 r_flags &= ~(REXEC_COPY_SKIP_PRE|REXEC_COPY_SKIP_POST);
1808#endif
22e551b9 1809
f1d31338
DM
1810 s = truebase;
1811
d7be1480 1812 play_it_again:
985afbc1 1813 if (global)
03c83e26 1814 s = truebase + curpos;
f722798b 1815
77da2310 1816 if (!CALLREGEXEC(rx, (char*)s, (char *)strend, (char*)truebase,
03c83e26 1817 had_zerolen, TARG, NULL, r_flags))
03b6c93d 1818 goto nope;
77da2310
NC
1819
1820 PL_curpm = pm;
985afbc1 1821 if (dynpm->op_pmflags & PMf_ONCE)
c737faaf 1822#ifdef USE_ITHREADS
77da2310 1823 SvREADONLY_on(PL_regex_pad[dynpm->op_pmoffset]);
c737faaf 1824#else
77da2310 1825 dynpm->op_pmflags |= PMf_USED;
c737faaf 1826#endif
a0d0e21e 1827
72311751
GS
1828 if (rxtainted)
1829 RX_MATCH_TAINTED_on(rx);
1830 TAINT_IF(RX_MATCH_TAINTED(rx));
35c2ccc3
DM
1831
1832 /* update pos */
1833
1834 if (global && (gimme != G_ARRAY || (dynpm->op_pmflags & PMf_CONTINUE))) {
b1422d62 1835 if (!mg)
35c2ccc3 1836 mg = sv_magicext_mglob(TARG);
25fdce4a 1837 MgBYTEPOS_set(mg, TARG, truebase, RX_OFFS(rx)[0].end);
adf51885
DM
1838 if (RX_ZERO_LEN(rx))
1839 mg->mg_flags |= MGf_MINMATCH;
1840 else
1841 mg->mg_flags &= ~MGf_MINMATCH;
35c2ccc3
DM
1842 }
1843
bf9dff51
DM
1844 if ((!RX_NPARENS(rx) && !global) || gimme != G_ARRAY) {
1845 LEAVE_SCOPE(oldsave);
1846 RETPUSHYES;
1847 }
1848
88ab22af
DM
1849 /* push captures on stack */
1850
bf9dff51 1851 {
07bc277f 1852 const I32 nparens = RX_NPARENS(rx);
a3b680e6 1853 I32 i = (global && !nparens) ? 1 : 0;
a0d0e21e 1854
c277df42 1855 SPAGAIN; /* EVAL blocks could move the stack. */
ffc61ed2
JH
1856 EXTEND(SP, nparens + i);
1857 EXTEND_MORTAL(nparens + i);
1858 for (i = !i; i <= nparens; i++) {
a0d0e21e 1859 PUSHs(sv_newmortal());
5d9574c1
DM
1860 if (LIKELY((RX_OFFS(rx)[i].start != -1)
1861 && RX_OFFS(rx)[i].end != -1 ))
1862 {
07bc277f 1863 const I32 len = RX_OFFS(rx)[i].end - RX_OFFS(rx)[i].start;
f1d31338 1864 const char * const s = RX_OFFS(rx)[i].start + truebase;
5d9574c1
DM
1865 if (UNLIKELY(RX_OFFS(rx)[i].end < 0 || RX_OFFS(rx)[i].start < 0
1866 || len < 0 || len > strend - s))
5637ef5b
NC
1867 DIE(aTHX_ "panic: pp_match start/end pointers, i=%ld, "
1868 "start=%ld, end=%ld, s=%p, strend=%p, len=%"UVuf,
1869 (long) i, (long) RX_OFFS(rx)[i].start,
1870 (long)RX_OFFS(rx)[i].end, s, strend, (UV) len);
a0d0e21e 1871 sv_setpvn(*SP, s, len);
cce850e4 1872 if (DO_UTF8(TARG) && is_utf8_string((U8*)s, len))
a197cbdd 1873 SvUTF8_on(*SP);
a0d0e21e
LW
1874 }
1875 }
1876 if (global) {
0e0b3e82 1877 curpos = (UV)RX_OFFS(rx)[0].end;
03c83e26 1878 had_zerolen = RX_ZERO_LEN(rx);
c277df42 1879 PUTBACK; /* EVAL blocks may use stack */
cf93c79d 1880 r_flags |= REXEC_IGNOREPOS | REXEC_NOT_FIRST;
a0d0e21e
LW
1881 goto play_it_again;
1882 }
4633a7c4 1883 LEAVE_SCOPE(oldsave);
a0d0e21e
LW
1884 RETURN;
1885 }
e5964223 1886 NOT_REACHED; /* NOTREACHED */
a0d0e21e 1887
7b52d656 1888 nope:
d65afb4b 1889 if (global && !(dynpm->op_pmflags & PMf_CONTINUE)) {
b1422d62
DM
1890 if (!mg)
1891 mg = mg_find_mglob(TARG);
1892 if (mg)
1893 mg->mg_len = -1;
a0d0e21e 1894 }
4633a7c4 1895 LEAVE_SCOPE(oldsave);
a0d0e21e
LW
1896 if (gimme == G_ARRAY)
1897 RETURN;
1898 RETPUSHNO;
1899}
1900
1901OP *
864dbfa3 1902Perl_do_readline(pTHX)
a0d0e21e 1903{
20b7effb 1904 dSP; dTARGETSTACKED;
eb578fdb 1905 SV *sv;
a0d0e21e
LW
1906 STRLEN tmplen = 0;
1907 STRLEN offset;
760ac839 1908 PerlIO *fp;
eb578fdb
KW
1909 IO * const io = GvIO(PL_last_in_gv);
1910 const I32 type = PL_op->op_type;
1c23e2bd 1911 const U8 gimme = GIMME_V;
a0d0e21e 1912
6136c704 1913 if (io) {
50db69d8 1914 const MAGIC *const mg = SvTIED_mg((const SV *)io, PERL_MAGIC_tiedscalar);
6136c704 1915 if (mg) {
3e0cb5de 1916 Perl_tied_method(aTHX_ SV_CONST(READLINE), SP, MUTABLE_SV(io), mg, gimme, 0);
6136c704 1917 if (gimme == G_SCALAR) {
50db69d8
NC
1918 SPAGAIN;
1919 SvSetSV_nosteal(TARG, TOPs);
1920 SETTARG;
6136c704 1921 }
50db69d8 1922 return NORMAL;
0b7c7b4f 1923 }
e79b0511 1924 }
4608196e 1925 fp = NULL;
a0d0e21e
LW
1926 if (io) {
1927 fp = IoIFP(io);
1928 if (!fp) {
1929 if (IoFLAGS(io) & IOf_ARGV) {
1930 if (IoFLAGS(io) & IOf_START) {
a0d0e21e 1931 IoLINES(io) = 0;
b9f2b683 1932 if (av_tindex(GvAVn(PL_last_in_gv)) < 0) {
1d7c1841 1933 IoFLAGS(io) &= ~IOf_START;
d5eb9a46 1934 do_open6(PL_last_in_gv, "-", 1, NULL, NULL, 0);
4bac9ae4 1935 SvTAINTED_off(GvSVn(PL_last_in_gv)); /* previous tainting irrelevant */
76f68e9b 1936 sv_setpvs(GvSVn(PL_last_in_gv), "-");
3280af22 1937 SvSETMAGIC(GvSV(PL_last_in_gv));
a2008d6d
GS
1938 fp = IoIFP(io);
1939 goto have_fp;
a0d0e21e
LW
1940 }
1941 }
157fb5a1 1942 fp = nextargv(PL_last_in_gv, PL_op->op_flags & OPf_SPECIAL);
a0d0e21e 1943 if (!fp) { /* Note: fp != IoIFP(io) */
3280af22 1944 (void)do_close(PL_last_in_gv, FALSE); /* now it does*/
a0d0e21e
LW
1945 }
1946 }
0d44d22b
NC
1947 else if (type == OP_GLOB)
1948 fp = Perl_start_glob(aTHX_ POPs, io);
a0d0e21e
LW
1949 }
1950 else if (type == OP_GLOB)
1951 SP--;
7716c5c5 1952 else if (IoTYPE(io) == IoTYPE_WRONLY) {
a5390457 1953 report_wrongway_fh(PL_last_in_gv, '>');
a00b5bd3 1954 }
a0d0e21e
LW
1955 }
1956 if (!fp) {
041457d9 1957 if ((!io || !(IoFLAGS(io) & IOf_START))
de7dabb6
TC
1958 && ckWARN(WARN_CLOSED)
1959 && type != OP_GLOB)
041457d9 1960 {
de7dabb6 1961 report_evil_fh(PL_last_in_gv);
3f4520fe 1962 }
54310121 1963 if (gimme == G_SCALAR) {
79628082 1964 /* undef TARG, and push that undefined value */
ba92458f 1965 if (type != OP_RCATLINE) {
aab1202a 1966 sv_setsv(TARG,NULL);
ba92458f 1967 }
a0d0e21e
LW
1968 PUSHTARG;
1969 }
1970 RETURN;
1971 }
a2008d6d 1972 have_fp:
54310121 1973 if (gimme == G_SCALAR) {
a0d0e21e 1974 sv = TARG;
0f722b55
RGS
1975 if (type == OP_RCATLINE && SvGMAGICAL(sv))
1976 mg_get(sv);
48de12d9
RGS
1977 if (SvROK(sv)) {
1978 if (type == OP_RCATLINE)
5668452f 1979 SvPV_force_nomg_nolen(sv);
48de12d9
RGS
1980 else
1981 sv_unref(sv);
1982 }
f7877b28 1983 else if (isGV_with_GP(sv)) {
5668452f 1984 SvPV_force_nomg_nolen(sv);
f7877b28 1985 }
862a34c6 1986 SvUPGRADE(sv, SVt_PV);
a0d0e21e 1987 tmplen = SvLEN(sv); /* remember if already alloced */
e3918bb7 1988 if (!tmplen && !SvREADONLY(sv) && !SvIsCOW(sv)) {
f72e8700
JJ
1989 /* try short-buffering it. Please update t/op/readline.t
1990 * if you change the growth length.
1991 */
1992 Sv_Grow(sv, 80);
1993 }
2b5e58c4
AMS
1994 offset = 0;
1995 if (type == OP_RCATLINE && SvOK(sv)) {
1996 if (!SvPOK(sv)) {
5668452f 1997 SvPV_force_nomg_nolen(sv);
2b5e58c4 1998 }
a0d0e21e 1999 offset = SvCUR(sv);
2b5e58c4 2000 }
a0d0e21e 2001 }
54310121 2002 else {
561b68a9 2003 sv = sv_2mortal(newSV(80));
54310121
PP
2004 offset = 0;
2005 }
fbad3eb5 2006
3887d568
AP
2007 /* This should not be marked tainted if the fp is marked clean */
2008#define MAYBE_TAINT_LINE(io, sv) \
2009 if (!(IoFLAGS(io) & IOf_UNTAINT)) { \
2010 TAINT; \
2011 SvTAINTED_on(sv); \
2012 }
2013
684bef36 2014/* delay EOF state for a snarfed empty file */
fbad3eb5 2015#define SNARF_EOF(gimme,rs,io,sv) \
684bef36 2016 (gimme != G_SCALAR || SvCUR(sv) \
b9fee9ba 2017 || (IoFLAGS(io) & IOf_NOLINE) || !RsSNARF(rs))
fbad3eb5 2018
a0d0e21e 2019 for (;;) {
09e8efcc 2020 PUTBACK;
fbad3eb5 2021 if (!sv_gets(sv, fp, offset)
2d726892
TF
2022 && (type == OP_GLOB
2023 || SNARF_EOF(gimme, PL_rs, io, sv)
2024 || PerlIO_error(fp)))
fbad3eb5 2025 {
760ac839 2026 PerlIO_clearerr(fp);
a0d0e21e 2027 if (IoFLAGS(io) & IOf_ARGV) {
157fb5a1 2028 fp = nextargv(PL_last_in_gv, PL_op->op_flags & OPf_SPECIAL);
a0d0e21e
LW
2029 if (fp)
2030 continue;
3280af22 2031 (void)do_close(PL_last_in_gv, FALSE);
a0d0e21e
LW
2032 }
2033 else if (type == OP_GLOB) {
a2a5de95
NC
2034 if (!do_close(PL_last_in_gv, FALSE)) {
2035 Perl_ck_warner(aTHX_ packWARN(WARN_GLOB),
2036 "glob failed (child exited with status %d%s)",
2037 (int)(STATUS_CURRENT >> 8),
2038 (STATUS_CURRENT & 0x80) ? ", core dumped" : "");
4eb79ab5 2039 }
a0d0e21e 2040 }
54310121 2041 if (gimme == G_SCALAR) {
ba92458f
AE
2042 if (type != OP_RCATLINE) {
2043 SV_CHECK_THINKFIRST_COW_DROP(TARG);
0c34ef67 2044 SvOK_off(TARG);
ba92458f 2045 }
09e8efcc 2046 SPAGAIN;
a0d0e21e
LW
2047 PUSHTARG;
2048 }
3887d568 2049 MAYBE_TAINT_LINE(io, sv);
a0d0e21e
LW
2050 RETURN;
2051 }
3887d568 2052 MAYBE_TAINT_LINE(io, sv);
a0d0e21e 2053 IoLINES(io)++;
b9fee9ba 2054 IoFLAGS(io) |= IOf_NOLINE;
71be2cbc 2055 SvSETMAGIC(sv);
09e8efcc 2056 SPAGAIN;
a0d0e21e 2057 XPUSHs(sv);
a0d0e21e 2058 if (type == OP_GLOB) {
349d4f2f 2059 const char *t1;
45a23732 2060 Stat_t statbuf;
a0d0e21e 2061
3280af22 2062 if (SvCUR(sv) > 0 && SvCUR(PL_rs) > 0) {
6136c704 2063 char * const tmps = SvEND(sv) - 1;
aa07b2f6 2064 if (*tmps == *SvPVX_const(PL_rs)) {
c07a80fd 2065 *tmps = '\0';
b162af07 2066 SvCUR_set(sv, SvCUR(sv) - 1);
c07a80fd
PP
2067 }
2068 }
349d4f2f 2069 for (t1 = SvPVX_const(sv); *t1; t1++)
b51c3e77
CB
2070#ifdef __VMS
2071 if (strchr("*%?", *t1))
2072#else
7ad1e72d 2073 if (strchr("$&*(){}[]'\";\\|?<>~`", *t1))
b51c3e77 2074#endif
a0d0e21e 2075 break;
45a23732 2076 if (*t1 && PerlLIO_lstat(SvPVX_const(sv), &statbuf) < 0) {
a0d0e21e
LW
2077 (void)POPs; /* Unmatched wildcard? Chuck it... */
2078 continue;
2079 }
2d79bf7f 2080 } else if (SvUTF8(sv)) { /* OP_READLINE, OP_RCATLINE */
d4c19fe8
AL
2081 if (ckWARN(WARN_UTF8)) {
2082 const U8 * const s = (const U8*)SvPVX_const(sv) + offset;
2083 const STRLEN len = SvCUR(sv) - offset;
2084 const U8 *f;
2085
2086 if (!is_utf8_string_loc(s, len, &f))
2087 /* Emulate :encoding(utf8) warning in the same case. */
2088 Perl_warner(aTHX_ packWARN(WARN_UTF8),
2089 "utf8 \"\\x%02X\" does not map to Unicode",
2090 f < (U8*)SvEND(sv) ? *f : 0);
2091 }
a0d0e21e 2092 }
54310121 2093 if (gimme == G_ARRAY) {
a0d0e21e 2094 if (SvLEN(sv) - SvCUR(sv) > 20) {
1da4ca5f 2095 SvPV_shrink_to_cur(sv);
a0d0e21e 2096 }
561b68a9 2097 sv = sv_2mortal(newSV(80));
a0d0e21e
LW
2098 continue;
2099 }
54310121 2100 else if (gimme == G_SCALAR && !tmplen && SvLEN(sv) - SvCUR(sv) > 80) {
a0d0e21e 2101 /* try to reclaim a bit of scalar space (only on 1st alloc) */
d5b5861b
NC
2102 const STRLEN new_len
2103 = SvCUR(sv) < 60 ? 80 : SvCUR(sv)+40; /* allow some slop */
1da4ca5f 2104 SvPV_renew(sv, new_len);
a0d0e21e
LW
2105 }
2106 RETURN;
2107 }
2108}
2109
a0d0e21e
LW
2110PP(pp_helem)
2111{
20b7effb 2112 dSP;
760ac839 2113 HE* he;
ae77835f 2114 SV **svp;
c445ea15 2115 SV * const keysv = POPs;
85fbaab2 2116 HV * const hv = MUTABLE_HV(POPs);
a3b680e6
AL
2117 const U32 lval = PL_op->op_flags & OPf_MOD || LVRET;
2118 const U32 defer = PL_op->op_private & OPpLVAL_DEFER;
be6c24e0 2119 SV *sv;
92970b93 2120 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
d30e492c 2121 bool preeminent = TRUE;
a0d0e21e 2122
6dfc73ea
SM
2123 if (SvTYPE(hv) != SVt_PVHV)
2124 RETPUSHUNDEF;
d4c19fe8 2125
92970b93 2126 if (localizing) {
d4c19fe8
AL
2127 MAGIC *mg;
2128 HV *stash;
d30e492c
VP
2129
2130 /* If we can determine whether the element exist,
2131 * Try to preserve the existenceness of a tied hash
2132 * element by using EXISTS and DELETE if possible.
2133 * Fallback to FETCH and STORE otherwise. */
2c5f48c2 2134 if (SvCANEXISTDELETE(hv))
d30e492c 2135 preeminent = hv_exists_ent(hv, keysv, 0);
d4c19fe8 2136 }
d30e492c 2137
5f9d7e2b 2138 he = hv_fetch_ent(hv, keysv, lval && !defer, 0);
d4c19fe8 2139 svp = he ? &HeVAL(he) : NULL;
a0d0e21e 2140 if (lval) {
746f6409 2141 if (!svp || !*svp || *svp == &PL_sv_undef) {
68dc0745
PP
2142 SV* lv;
2143 SV* key2;
2d8e6c8d 2144 if (!defer) {
be2597df 2145 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
2d8e6c8d 2146 }
68dc0745
PP
2147 lv = sv_newmortal();
2148 sv_upgrade(lv, SVt_PVLV);
2149 LvTYPE(lv) = 'y';
6136c704 2150 sv_magic(lv, key2 = newSVsv(keysv), PERL_MAGIC_defelem, NULL, 0);
fc2b2dca 2151 SvREFCNT_dec_NN(key2); /* sv_magic() increments refcount */
0ad694a7 2152 LvTARG(lv) = SvREFCNT_inc_simple_NN(hv);
68dc0745
PP
2153 LvTARGLEN(lv) = 1;
2154 PUSHs(lv);
2155 RETURN;
2156 }
92970b93 2157 if (localizing) {
bfcb3514 2158 if (HvNAME_get(hv) && isGV(*svp))
159b6efe 2159 save_gp(MUTABLE_GV(*svp), !(PL_op->op_flags & OPf_SPECIAL));
47cfc530
VP
2160 else if (preeminent)
2161 save_helem_flags(hv, keysv, svp,
2162 (PL_op->op_flags & OPf_SPECIAL) ? 0 : SAVEf_SETMAGIC);
2163 else
2164 SAVEHDELETE(hv, keysv);
5f05dabc 2165 }
9026059d
GG
2166 else if (PL_op->op_private & OPpDEREF) {
2167 PUSHs(vivify_ref(*svp, PL_op->op_private & OPpDEREF));
2168 RETURN;
2169 }
a0d0e21e 2170 }
746f6409 2171 sv = (svp && *svp ? *svp : &PL_sv_undef);
fd69380d
DM
2172 /* Originally this did a conditional C<sv = sv_mortalcopy(sv)>; this
2173 * was to make C<local $tied{foo} = $tied{foo}> possible.
2174 * However, it seems no longer to be needed for that purpose, and
2175 * introduced a new bug: stuff like C<while ($hash{taintedval} =~ /.../g>
2176 * would loop endlessly since the pos magic is getting set on the
2177 * mortal copy and lost. However, the copy has the effect of
2178 * triggering the get magic, and losing it altogether made things like
2179 * c<$tied{foo};> in void context no longer do get magic, which some
2180 * code relied on. Also, delayed triggering of magic on @+ and friends
2181 * meant the original regex may be out of scope by now. So as a
2182 * compromise, do the get magic here. (The MGf_GSKIP flag will stop it
2183 * being called too many times). */
39cf747a 2184 if (!lval && SvRMAGICAL(hv) && SvGMAGICAL(sv))
fd69380d 2185 mg_get(sv);
be6c24e0 2186 PUSHs(sv);
a0d0e21e
LW
2187 RETURN;
2188}
2189
fedf30e1
DM
2190
2191/* a stripped-down version of Perl_softref2xv() for use by
2192 * pp_multideref(), which doesn't use PL_op->op_flags */
2193
f9db5646 2194STATIC GV *
fedf30e1
DM
2195S_softref2xv_lite(pTHX_ SV *const sv, const char *const what,
2196 const svtype type)
2197{
2198 if (PL_op->op_private & HINT_STRICT_REFS) {
2199 if (SvOK(sv))
2200 Perl_die(aTHX_ PL_no_symref_sv, sv,
2201 (SvPOKp(sv) && SvCUR(sv)>32 ? "..." : ""), what);
2202 else
2203 Perl_die(aTHX_ PL_no_usym, what);
2204 }
2205 if (!SvOK(sv))
2206 Perl_die(aTHX_ PL_no_usym, what);
2207 return gv_fetchsv_nomg(sv, GV_ADD, type);
2208}
2209
2210
79815f56
DM
2211/* Handle one or more aggregate derefs and array/hash indexings, e.g.
2212 * $h->{foo} or $a[0]{$key}[$i] or f()->[1]
fedf30e1
DM
2213 *
2214 * op_aux points to an array of unions of UV / IV / SV* / PADOFFSET.
79815f56
DM
2215 * Each of these either contains a set of actions, or an argument, such as
2216 * an IV to use as an array index, or a lexical var to retrieve.
2217 * Several actions re stored per UV; we keep shifting new actions off the
2218 * one UV, and only reload when it becomes zero.
fedf30e1
DM
2219 */
2220
2221PP(pp_multideref)
2222{
2223 SV *sv = NULL; /* init to avoid spurious 'may be used uninitialized' */
2224 UNOP_AUX_item *items = cUNOP_AUXx(PL_op)->op_aux;
2225 UV actions = items->uv;
2226
2227 assert(actions);
2228 /* this tells find_uninit_var() where we're up to */
2229 PL_multideref_pc = items;
2230
2231 while (1) {
2232 /* there are three main classes of action; the first retrieve
2233 * the initial AV or HV from a variable or the stack; the second
2234 * does the equivalent of an unrolled (/DREFAV, rv2av, aelem),
2235 * the third an unrolled (/DREFHV, rv2hv, helem).
2236 */
2237 switch (actions & MDEREF_ACTION_MASK) {
2238
2239 case MDEREF_reload:
2240 actions = (++items)->uv;
2241 continue;
2242
2243 case MDEREF_AV_padav_aelem: /* $lex[...] */
2244 sv = PAD_SVl((++items)->pad_offset);
2245 goto do_AV_aelem;
2246
2247 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
2248 sv = UNOP_AUX_item_sv(++items);
2249 assert(isGV_with_GP(sv));
2250 sv = (SV*)GvAVn((GV*)sv);
2251 goto do_AV_aelem;
2252
2253 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
2254 {
2255 dSP;
2256 sv = POPs;
2257 PUTBACK;
2258 goto do_AV_rv2av_aelem;
2259 }
2260
2261 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
2262 sv = UNOP_AUX_item_sv(++items);
2263 assert(isGV_with_GP(sv));
2264 sv = GvSVn((GV*)sv);
2265 goto do_AV_vivify_rv2av_aelem;
2266
2267 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
2268 sv = PAD_SVl((++items)->pad_offset);
2269 /* FALLTHROUGH */
2270
2271 do_AV_vivify_rv2av_aelem:
2272 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
2273 /* this is the OPpDEREF action normally found at the end of
2274 * ops like aelem, helem, rv2sv */
2275 sv = vivify_ref(sv, OPpDEREF_AV);
2276 /* FALLTHROUGH */
2277
2278 do_AV_rv2av_aelem:
2279 /* this is basically a copy of pp_rv2av when it just has the
2280 * sKR/1 flags */
2281 SvGETMAGIC(sv);
2282 if (LIKELY(SvROK(sv))) {
2283 if (UNLIKELY(SvAMAGIC(sv))) {
2284 sv = amagic_deref_call(sv, to_av_amg);
2285 }
2286 sv = SvRV(sv);
2287 if (UNLIKELY(SvTYPE(sv) != SVt_PVAV))
2288 DIE(aTHX_ "Not an ARRAY reference");
2289 }
2290 else if (SvTYPE(sv) != SVt_PVAV) {
2291 if (!isGV_with_GP(sv))
2292 sv = (SV*)S_softref2xv_lite(aTHX_ sv, "an ARRAY", SVt_PVAV);
2293 sv = MUTABLE_SV(GvAVn((GV*)sv));
2294 }
2295 /* FALLTHROUGH */
2296
2297 do_AV_aelem:
2298 {
2299 /* retrieve the key; this may be either a lexical or package
2300 * var (whose index/ptr is stored as an item) or a signed
2301 * integer constant stored as an item.
2302 */
2303 SV *elemsv;
2304 IV elem = 0; /* to shut up stupid compiler warnings */
2305
2306
2307 assert(SvTYPE(sv) == SVt_PVAV);
2308
2309 switch (actions & MDEREF_INDEX_MASK) {
2310 case MDEREF_INDEX_none:
2311 goto finish;
2312 case MDEREF_INDEX_const:
2313 elem = (++items)->iv;
2314 break;
2315 case MDEREF_INDEX_padsv:
2316 elemsv = PAD_SVl((++items)->pad_offset);
2317 goto check_elem;
2318 case MDEREF_INDEX_gvsv:
2319 elemsv = UNOP_AUX_item_sv(++items);
2320 assert(isGV_with_GP(elemsv));
2321 elemsv = GvSVn((GV*)elemsv);
2322 check_elem:
2323 if (UNLIKELY(SvROK(elemsv) && !SvGAMAGIC(elemsv)
2324 && ckWARN(WARN_MISC)))
2325 Perl_warner(aTHX_ packWARN(WARN_MISC),
2326 "Use of reference \"%"SVf"\" as array index",
2327 SVfARG(elemsv));
2328 /* the only time that S_find_uninit_var() needs this
2329 * is to determine which index value triggered the
2330 * undef warning. So just update it here. Note that
2331 * since we don't save and restore this var (e.g. for
2332 * tie or overload execution), its value will be
2333 * meaningless apart from just here */
2334 PL_multideref_pc = items;
2335 elem = SvIV(elemsv);
2336 break;
2337 }
2338
2339
2340 /* this is basically a copy of pp_aelem with OPpDEREF skipped */
2341
2342 if (!(actions & MDEREF_FLAG_last)) {
2343 SV** svp = av_fetch((AV*)sv, elem, 1);
2344 if (!svp || ! (sv=*svp))
2345 DIE(aTHX_ PL_no_aelem, elem);
2346 break;
2347 }
2348
2349 if (PL_op->op_private &
2350 (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE))
2351 {
2352 if (PL_op->op_private & OPpMULTIDEREF_EXISTS) {
2353 sv = av_exists((AV*)sv, elem) ? &PL_sv_yes : &PL_sv_no;
2354 }
2355 else {
2356 I32 discard = (GIMME_V == G_VOID) ? G_DISCARD : 0;
2357 sv = av_delete((AV*)sv, elem, discard);
2358 if (discard)
2359 return NORMAL;
2360 if (!sv)
2361 sv = &PL_sv_undef;
2362 }
2363 }
2364 else {
2365 const U32 lval = PL_op->op_flags & OPf_MOD || LVRET;
2366 const U32 defer = PL_op->op_private & OPpLVAL_DEFER;
2367 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
2368 bool preeminent = TRUE;
2369 AV *const av = (AV*)sv;
2370 SV** svp;
2371
2372 if (UNLIKELY(localizing)) {
2373 MAGIC *mg;
2374 HV *stash;
2375
2376 /* If we can determine whether the element exist,
2377 * Try to preserve the existenceness of a tied array
2378 * element by using EXISTS and DELETE if possible.
2379 * Fallback to FETCH and STORE otherwise. */
2380 if (SvCANEXISTDELETE(av))
2381 preeminent = av_exists(av, elem);
2382 }
2383
2384 svp = av_fetch(av, elem, lval && !defer);
2385
2386 if (lval) {
2387 if (!svp || !(sv = *svp)) {
2388 IV len;
2389 if (!defer)
2390 DIE(aTHX_ PL_no_aelem, elem);
2391 len = av_tindex(av);
2392 sv = sv_2mortal(newSVavdefelem(av,
2393 /* Resolve a negative index now, unless it points
2394 * before the beginning of the array, in which
2395 * case record it for error reporting in
2396 * magic_setdefelem. */
2397 elem < 0 && len + elem >= 0
2398 ? len + elem : elem, 1));
2399 }
2400 else {
2401 if (UNLIKELY(localizing)) {
2402 if (preeminent) {
2403 save_aelem(av, elem, svp);
2404 sv = *svp; /* may have changed */
2405 }
2406 else
2407 SAVEADELETE(av, elem);
2408 }
2409 }
2410 }
2411 else {
2412 sv = (svp ? *svp : &PL_sv_undef);
2413 /* see note in pp_helem() */
2414 if (SvRMAGICAL(av) && SvGMAGICAL(sv))
2415 mg_get(sv);
2416 }
2417 }
2418
2419 }
2420 finish:
2421 {
2422 dSP;
2423 XPUSHs(sv);
2424 RETURN;
2425 }
2426 /* NOTREACHED */
2427
2428
2429
2430
2431 case MDEREF_HV_padhv_helem: /* $lex{...} */
2432 sv = PAD_SVl((++items)->pad_offset);
2433 goto do_HV_helem;
2434
2435 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
2436 sv = UNOP_AUX_item_sv(++items);
2437 assert(isGV_with_GP(sv));
2438 sv = (SV*)GvHVn((GV*)sv);
2439 goto do_HV_helem;
2440
2441 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
2442 {
2443 dSP;
2444 sv = POPs;
2445 PUTBACK;
2446 goto do_HV_rv2hv_helem;
2447 }
2448
2449 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
2450 sv = UNOP_AUX_item_sv(++items);
2451 assert(isGV_with_GP(sv));
2452 sv = GvSVn((GV*)sv);
2453 goto do_HV_vivify_rv2hv_helem;
2454
2455 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
2456 sv = PAD_SVl((++items)->pad_offset);
2457 /* FALLTHROUGH */
2458
2459 do_HV_vivify_rv2hv_helem:
2460 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
2461 /* this is the OPpDEREF action normally found at the end of
2462 * ops like aelem, helem, rv2sv */
2463 sv = vivify_ref(sv, OPpDEREF_HV);
2464 /* FALLTHROUGH */
2465
2466 do_HV_rv2hv_helem:
2467 /* this is basically a copy of pp_rv2hv when it just has the
2468 * sKR/1 flags (and pp_rv2hv is aliased to pp_rv2av) */
2469
2470 SvGETMAGIC(sv);
2471 if (LIKELY(SvROK(sv))) {
2472 if (UNLIKELY(SvAMAGIC(sv))) {
2473 sv = amagic_deref_call(sv, to_hv_amg);
2474 }
2475 sv = SvRV(sv);
2476 if (UNLIKELY(SvTYPE(sv) != SVt_PVHV))
2477 DIE(aTHX_ "Not a HASH reference");
2478 }
2479 else if (SvTYPE(sv) != SVt_PVHV) {
2480 if (!isGV_with_GP(sv))
2481 sv = (SV*)S_softref2xv_lite(aTHX_ sv, "a HASH", SVt_PVHV);
2482 sv = MUTABLE_SV(GvHVn((GV*)sv));
2483 }
2484 /* FALLTHROUGH */
2485
2486 do_HV_helem:
2487 {
2488 /* retrieve the key; this may be either a lexical / package
2489 * var or a string constant, whose index/ptr is stored as an
2490 * item
2491 */
2492 SV *keysv = NULL; /* to shut up stupid compiler warnings */
2493
2494 assert(SvTYPE(sv) == SVt_PVHV);
2495
2496 switch (actions & MDEREF_INDEX_MASK) {
2497 case MDEREF_INDEX_none:
2498 goto finish;
2499
2500 case MDEREF_INDEX_const:
2501 keysv = UNOP_AUX_item_sv(++items);
2502 break;
2503
2504 case MDEREF_INDEX_padsv:
2505 keysv = PAD_SVl((++items)->pad_offset);
2506 break;
2507
2508 case MDEREF_INDEX_gvsv:
2509 keysv = UNOP_AUX_item_sv(++items);
2510 keysv = GvSVn((GV*)keysv);
2511 break;
2512 }
2513
2514 /* see comment above about setting this var */
2515 PL_multideref_pc = items;
2516
2517
2518 /* ensure that candidate CONSTs have been HEKified */
2519 assert( ((actions & MDEREF_INDEX_MASK) != MDEREF_INDEX_const)
2520 || SvTYPE(keysv) >= SVt_PVMG
2521 || !SvOK(keysv)
2522 || SvROK(keysv)
2523 || SvIsCOW_shared_hash(keysv));
2524
2525 /* this is basically a copy of pp_helem with OPpDEREF skipped */
2526
2527 if (!(actions & MDEREF_FLAG_last)) {
2528 HE *he = hv_fetch_ent((HV*)sv, keysv, 1, 0);
2529 if (!he || !(sv=HeVAL(he)) || sv == &PL_sv_undef)
2530 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
2531 break;
2532 }
2533
2534 if (PL_op->op_private &
2535 (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE))
2536 {
2537 if (PL_op->op_private & OPpMULTIDEREF_EXISTS) {
2538 sv = hv_exists_ent((HV*)sv, keysv, 0)
2539 ? &PL_sv_yes : &PL_sv_no;
2540 }
2541 else {
2542 I32 discard = (GIMME_V == G_VOID) ? G_DISCARD : 0;
2543 sv = hv_delete_ent((HV*)sv, keysv, discard, 0);
2544 if (discard)
2545 return NORMAL;
2546 if (!sv)
2547 sv = &PL_sv_undef;
2548 }
2549 }
2550 else {
2551 const U32 lval = PL_op->op_flags & OPf_MOD || LVRET;
2552 const U32 defer = PL_op->op_private & OPpLVAL_DEFER;
2553 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
2554 bool preeminent = TRUE;
2555 SV **svp;
2556 HV * const hv = (HV*)sv;
2557 HE* he;
2558
2559 if (UNLIKELY(localizing)) {
2560 MAGIC *mg;
2561 HV *stash;
2562
2563 /* If we can determine whether the element exist,
2564 * Try to preserve the existenceness of a tied hash
2565 * element by using EXISTS and DELETE if possible.
2566 * Fallback to FETCH and STORE otherwise. */
2567 if (SvCANEXISTDELETE(hv))
2568 preeminent = hv_exists_ent(hv, keysv, 0);
2569 }
2570
2571 he = hv_fetch_ent(hv, keysv, lval && !defer, 0);
2572 svp = he ? &HeVAL(he) : NULL;
2573
2574
2575 if (lval) {
2576 if (!svp || !(sv = *svp) || sv == &PL_sv_undef) {
2577 SV* lv;
2578 SV* key2;
2579 if (!defer)
2580 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
2581 lv = sv_newmortal();
2582 sv_upgrade(lv, SVt_PVLV);
2583 LvTYPE(lv) = 'y';
2584 sv_magic(lv, key2 = newSVsv(keysv),
2585 PERL_MAGIC_defelem, NULL, 0);
2586 /* sv_magic() increments refcount */
2587 SvREFCNT_dec_NN(key2);
0ad694a7 2588 LvTARG(lv) = SvREFCNT_inc_simple_NN(hv);
fedf30e1
DM
2589 LvTARGLEN(lv) = 1;
2590 sv = lv;
2591 }
2592 else {
2593 if (localizing) {
2594 if (HvNAME_get(hv) && isGV(sv))
2595 save_gp(MUTABLE_GV(sv),
2596 !(PL_op->op_flags & OPf_SPECIAL));
2597 else if (preeminent) {
2598 save_helem_flags(hv, keysv, svp,
2599 (PL_op->op_flags & OPf_SPECIAL)
2600 ? 0 : SAVEf_SETMAGIC);
2601 sv = *svp; /* may have changed */
2602 }
2603 else
2604 SAVEHDELETE(hv, keysv);
2605 }
2606 }
2607 }
2608 else {
2609 sv = (svp && *svp ? *svp : &PL_sv_undef);
2610 /* see note in pp_helem() */
2611 if (SvRMAGICAL(hv) && SvGMAGICAL(sv))
2612 mg_get(sv);
2613 }
2614 }
2615 goto finish;
2616 }
2617
2618 } /* switch */
2619
2620 actions >>= MDEREF_SHIFT;
2621 } /* while */
2622 /* NOTREACHED */
2623}
2624
2625
a0d0e21e
LW
2626PP(pp_iter)
2627{
eb578fdb 2628 PERL_CONTEXT *cx;
7d6c2cef 2629 SV *oldsv;
1d7c1841 2630 SV **itersvp;
8a1f10dd 2631 SV *retsv;
a0d0e21e 2632
84f05d57
JH
2633 SV *sv;
2634 AV *av;
2635 IV ix;
2636 IV inc;
2637
4ebe6e95 2638 cx = CX_CUR();
1d7c1841 2639 itersvp = CxITERVAR(cx);
4b5c941e 2640 assert(itersvp);
a48ce6be
DM
2641
2642 switch (CxTYPE(cx)) {
17c91640 2643
b552b52c
DM
2644 case CXt_LOOP_LAZYSV: /* string increment */
2645 {
2646 SV* cur = cx->blk_loop.state_u.lazysv.cur;
2647 SV *end = cx->blk_loop.state_u.lazysv.end;
2648 /* If the maximum is !SvOK(), pp_enteriter substitutes PL_sv_no.
2649 It has SvPVX of "" and SvCUR of 0, which is what we want. */
2650 STRLEN maxlen = 0;
2651 const char *max = SvPV_const(end, maxlen);
5d9574c1 2652 if (UNLIKELY(SvNIOK(cur) || SvCUR(cur) > maxlen))
8a1f10dd 2653 goto retno;
b552b52c
DM
2654
2655 oldsv = *itersvp;
6d3ca00e
DM
2656 /* NB: on the first iteration, oldsv will have a ref count of at
2657 * least 2 (one extra from blk_loop.itersave), so the GV or pad
2658 * slot will get localised; on subsequent iterations the RC==1
2659 * optimisation may kick in and the SV will be reused. */
2660 if (oldsv && LIKELY(SvREFCNT(oldsv) == 1 && !SvMAGICAL(oldsv))) {
b552b52c
DM
2661 /* safe to reuse old SV */
2662 sv_setsv(oldsv, cur);
a48ce6be 2663 }
b552b52c
DM
2664 else
2665 {
2666 /* we need a fresh SV every time so that loop body sees a
2667 * completely new SV for closures/references to work as
2668 * they used to */
2669 *itersvp = newSVsv(cur);
6d3ca00e 2670 SvREFCNT_dec(oldsv);
b552b52c
DM
2671 }
2672 if (strEQ(SvPVX_const(cur), max))
2673 sv_setiv(cur, 0); /* terminate next time */
2674 else
2675 sv_inc(cur);
2676 break;
2677 }
a48ce6be 2678
fcef60b4
DM
2679 case CXt_LOOP_LAZYIV: /* integer increment */
2680 {
2681 IV cur = cx->blk_loop.state_u.lazyiv.cur;
5d9574c1 2682 if (UNLIKELY(cur > cx->blk_loop.state_u.lazyiv.end))
8a1f10dd 2683 goto retno;
7f61b687 2684
fcef60b4 2685 oldsv = *itersvp;
6d3ca00e
DM
2686 /* see NB comment above */
2687 if (oldsv && LIKELY(SvREFCNT(oldsv) == 1 && !SvMAGICAL(oldsv))) {
eaa5c2d6 2688 /* safe to reuse old SV */
47b96a1e
DM
2689
2690 if ( (SvFLAGS(oldsv) & (SVTYPEMASK|SVf_THINKFIRST|SVf_IVisUV))
2691 == SVt_IV)
2692 {
2693 /* Cheap SvIOK_only().
2694 * Assert that flags which SvIOK_only() would test or
2695 * clear can't be set, because we're SVt_IV */
2696 assert(!(SvFLAGS(oldsv) &
2697 (SVf_OOK|SVf_UTF8|(SVf_OK & ~(SVf_IOK|SVp_IOK)))));
2698 SvFLAGS(oldsv) |= (SVf_IOK|SVp_IOK);
2699 /* SvIV_set() where sv_any points to head */
2700 oldsv->sv_u.svu_iv = cur;
2701
2702 }
2703 else
2704 sv_setiv(oldsv, cur);
eaa5c2d6 2705 }
1c846c1f 2706 else
eaa5c2d6
GA
2707 {
2708 /* we need a fresh SV every time so that loop body sees a
2709 * completely new SV for closures/references to work as they
2710 * used to */
fcef60b4 2711 *itersvp = newSViv(cur);
6d3ca00e 2712 SvREFCNT_dec(oldsv);
eaa5c2d6 2713 }
a2309040 2714
5d9574c1 2715 if (UNLIKELY(cur == IV_MAX)) {
cdc1aa42
NC
2716 /* Handle end of range at IV_MAX */
2717 cx->blk_loop.state_u.lazyiv.end = IV_MIN;
2718 } else
2719 ++cx->blk_loop.state_u.lazyiv.cur;
a48ce6be 2720 break;
fcef60b4 2721 }
a48ce6be 2722
93661e56
DM
2723 case CXt_LOOP_LIST: /* for (1,2,3) */
2724
2725 assert(OPpITER_REVERSED == 2); /* so inc becomes -1 or 1 */
2726 inc = 1 - (PL_op->op_private & OPpITER_REVERSED);
2727 ix = (cx->blk_loop.state_u.stack.ix += inc);
2728 if (UNLIKELY(inc > 0
2729 ? ix > cx->blk_oldsp
2730 : ix <= cx->blk_loop.state_u.stack.basesp)
2731 )
8a1f10dd 2732 goto retno;
93661e56
DM
2733
2734 sv = PL_stack_base[ix];
2735 av = NULL;
2736 goto loop_ary_common;
2737
2738 case CXt_LOOP_ARY: /* for (@ary) */
2739
2740 av = cx->blk_loop.state_u.ary.ary;
2741 inc = 1 - (PL_op->op_private & OPpITER_REVERSED);
2742 ix = (cx->blk_loop.state_u.ary.ix += inc);
2743 if (UNLIKELY(inc > 0
2744 ? ix > AvFILL(av)
2745 : ix < 0)
2746 )
8a1f10dd 2747 goto retno;
de080daa 2748
9d1ee8e0 2749 if (UNLIKELY(SvRMAGICAL(av))) {
a8a20bb6
DM
2750 SV * const * const svp = av_fetch(av, ix, FALSE);
2751 sv = svp ? *svp : NULL;
2752 }
2753 else {
2754 sv = AvARRAY(av)[ix];
de080daa 2755 }
ef3e5ea9 2756
93661e56
DM
2757 loop_ary_common:
2758
d39c26a6
FC
2759 if (UNLIKELY(cx->cx_type & CXp_FOR_LVREF)) {
2760 SvSetMagicSV(*itersvp, sv);
2761 break;
2762 }
2763
5d9574c1
DM
2764 if (LIKELY(sv)) {
2765 if (UNLIKELY(SvIS_FREED(sv))) {
f38aa882
DM
2766 *itersvp = NULL;
2767 Perl_croak(aTHX_ "Use of freed value in iteration");
2768 }
60779a30 2769 if (SvPADTMP(sv)) {
8e079c2a 2770 sv = newSVsv(sv);
60779a30 2771 }
8e079c2a
FC
2772 else {
2773 SvTEMP_off(sv);
2774 SvREFCNT_inc_simple_void_NN(sv);
2775 }
de080daa 2776 }
93661e56 2777 else if (av) {
199f858d 2778 sv = newSVavdefelem(av, ix, 0);
de080daa 2779 }
a600f7e6
FC
2780 else
2781 sv = &PL_sv_undef;
a0d0e21e 2782
de080daa
DM
2783 oldsv = *itersvp;
2784 *itersvp = sv;
2785 SvREFCNT_dec(oldsv);
de080daa 2786 break;
a48ce6be
DM
2787
2788 default:
2789 DIE(aTHX_ "panic: pp_iter, type=%u", CxTYPE(cx));
2790 }
8a1f10dd
DM
2791
2792 retsv = &PL_sv_yes;
2793 if (0) {
2794 retno:
2795 retsv = &PL_sv_no;
2796 }
2797 /* pp_enteriter should have pre-extended the stack */
2798 assert(PL_stack_sp < PL_stack_max);
2799 *++PL_stack_sp =retsv;
2800
2801 return PL_op->op_next;
a0d0e21e
LW
2802}
2803
ef07e810
DM
2804/*
2805A description of how taint works in pattern matching and substitution.
2806
284167a5
SM
2807This is all conditional on NO_TAINT_SUPPORT not being defined. Under
2808NO_TAINT_SUPPORT, taint-related operations should become no-ops.
2809
4e19c54b 2810While the pattern is being assembled/concatenated and then compiled,
284167a5
SM
2811PL_tainted will get set (via TAINT_set) if any component of the pattern
2812is tainted, e.g. /.*$tainted/. At the end of pattern compilation,
2813the RXf_TAINTED flag is set on the pattern if PL_tainted is set (via
1738e041
DM
2814TAINT_get). It will also be set if any component of the pattern matches
2815based on locale-dependent behavior.
ef07e810 2816
0ab462a6
DM
2817When the pattern is copied, e.g. $r = qr/..../, the SV holding the ref to
2818the pattern is marked as tainted. This means that subsequent usage, such
284167a5
SM
2819as /x$r/, will set PL_tainted using TAINT_set, and thus RXf_TAINTED,
2820on the new pattern too.
ef07e810 2821
272d35c9 2822RXf_TAINTED_SEEN is used post-execution by the get magic code
ef07e810
DM
2823of $1 et al to indicate whether the returned value should be tainted.
2824It is the responsibility of the caller of the pattern (i.e. pp_match,
2825pp_subst etc) to set this flag for any other circumstances where $1 needs
2826to be tainted.
2827
2828The taint behaviour of pp_subst (and pp_substcont) is quite complex.
2829
2830There are three possible sources of taint
2831 * the source string
2832 * the pattern (both compile- and run-time, RXf_TAINTED / RXf_TAINTED_SEEN)
2833 * the replacement string (or expression under /e)
2834
2835There are four destinations of taint and they are affected by the sources
2836according to the rules below:
2837
2838 * the return value (not including /r):
2839 tainted by the source string and pattern, but only for the
2840 number-of-iterations case; boolean returns aren't tainted;
2841 * the modified string (or modified copy under /r):
2842 tainted by the source string, pattern, and replacement strings;
2843 * $1 et al:
2844 tainted by the pattern, and under 'use re "taint"', by the source
2845 string too;
2846 * PL_taint - i.e. whether subsequent code (e.g. in a /e block) is tainted:
2847 should always be unset before executing subsequent code.
2848
2849The overall action of pp_subst is:
2850
2851 * at the start, set bits in rxtainted indicating the taint status of
2852 the various sources.
2853
2854 * After each pattern execution, update the SUBST_TAINT_PAT bit in
2855 rxtainted if RXf_TAINTED_SEEN has been set, to indicate that the
2856 pattern has subsequently become tainted via locale ops.
2857
2858 * If control is being passed to pp_substcont to execute a /e block,
2859 save rxtainted in the CXt_SUBST block, for future use by
2860 pp_substcont.
2861
2862 * Whenever control is being returned to perl code (either by falling
2863 off the "end" of pp_subst/pp_substcont, or by entering a /e block),
2864 use the flag bits in rxtainted to make all the appropriate types of
0ab462a6
DM
2865 destination taint visible; e.g. set RXf_TAINTED_SEEN so that $1
2866 et al will appear tainted.
ef07e810
DM
2867
2868pp_match is just a simpler version of the above.
2869
2870*/
2871
a0d0e21e
LW
2872PP(pp_subst)
2873{
20b7effb 2874 dSP; dTARG;
eb578fdb 2875 PMOP *pm = cPMOP;
a0d0e21e 2876 PMOP *rpm = pm;
eb578fdb 2877 char *s;
a0d0e21e 2878 char *strend;
5c144d81 2879 const char *c;
a0d0e21e 2880 STRLEN clen;
3c6ef0a5
FC
2881 SSize_t iters = 0;
2882 SSize_t maxiters;
a0d0e21e 2883 bool once;
ef07e810
DM
2884 U8 rxtainted = 0; /* holds various SUBST_TAINT_* flag bits.
2885 See "how taint works" above */
a0d0e21e 2886 char *orig;
1ed74d04 2887 U8 r_flags;
eb578fdb 2888 REGEXP *rx = PM_GETRE(pm);
a0d0e21e
LW
2889 STRLEN len;
2890 int force_on_match = 0;
0bcc34c2 2891 const I32 oldsave = PL_savestack_ix;
792b2c16 2892 STRLEN slen;
26a74523 2893 bool doutf8 = FALSE; /* whether replacement is in utf8 */
db2c6cb3 2894#ifdef PERL_ANY_COW
ed252734
NC
2895 bool is_cow;
2896#endif
a0714e2c 2897 SV *nsv = NULL;
b770e143 2898 /* known replacement string? */
eb578fdb 2899 SV *dstr = (pm->op_pmflags & PMf_CONST) ? POPs : NULL;
a0d0e21e 2900
f410a211
NC
2901 PERL_ASYNC_CHECK();
2902
533c011a 2903 if (PL_op->op_flags & OPf_STACKED)
a0d0e21e 2904 TARG = POPs;
6ffceeb7 2905 else if (ARGTARG)
59f00321 2906 GETTARGET;
a0d0e21e 2907 else {
54b9620d 2908 TARG = DEFSV;
a0d0e21e 2909 EXTEND(SP,1);
1c846c1f 2910 }
d9f424b2 2911
64534138 2912 SvGETMAGIC(TARG); /* must come before cow check */
db2c6cb3 2913#ifdef PERL_ANY_COW
ed252734
NC
2914 /* Awooga. Awooga. "bool" types that are actually char are dangerous,
2915 because they make integers such as 256 "false". */
2916 is_cow = SvIsCOW(TARG) ? TRUE : FALSE;
ed252734 2917#endif
d13a5d3b
TC
2918 if (!(rpm->op_pmflags & PMf_NONDESTRUCT)) {
2919#ifndef PERL_ANY_COW
2920 if (SvIsCOW(TARG))
2921 sv_force_normal_flags(TARG,0);
2922#endif
2923 if ((SvREADONLY(TARG)
2924 || ( ((SvTYPE(TARG) == SVt_PVGV && isGV_with_GP(TARG))
2925 || SvTYPE(TARG) > SVt_PVLV)
2926 && !(SvTYPE(TARG) == SVt_PVGV && SvFAKE(TARG)))))
2927 Perl_croak_no_modify();
2928 }
8ec5e241
NIS
2929 PUTBACK;
2930
6ac6605d
DM
2931 orig = SvPV_nomg(TARG, len);
2932 /* note we don't (yet) force the var into being a string; if we fail
2933 * to match, we leave as-is; on successful match howeverm, we *will*
2934 * coerce into a string, then repeat the match */
4499db73 2935 if (!SvPOKp(TARG) || SvTYPE(TARG) == SVt_PVGV || SvVOK(TARG))
a0d0e21e 2936 force_on_match = 1;
20be6587
DM
2937
2938 /* only replace once? */
2939 once = !(rpm->op_pmflags & PMf_GLOBAL);
2940
ef07e810 2941 /* See "how taint works" above */
284167a5 2942 if (TAINTING_get) {
20be6587
DM
2943 rxtainted = (
2944 (SvTAINTED(TARG) ? SUBST_TAINT_STR : 0)
284167a5 2945 | (RX_ISTAINTED(rx) ? SUBST_TAINT_PAT : 0)
20be6587
DM
2946 | ((pm->op_pmflags & PMf_RETAINT) ? SUBST_TAINT_RETAINT : 0)
2947 | ((once && !(rpm->op_pmflags & PMf_NONDESTRUCT))
2948 ? SUBST_TAINT_BOOLRET : 0));
2949 TAINT_NOT;
2950 }
a12c0f56 2951
a0d0e21e 2952 force_it:
6ac6605d
DM
2953 if (!pm || !orig)
2954 DIE(aTHX_ "panic: pp_subst, pm=%p, orig=%p", pm, orig);
a0d0e21e 2955
6ac6605d
DM
2956 strend = orig + len;
2957 slen = DO_UTF8(TARG) ? utf8_length((U8*)orig, (U8*)strend) : len;
792b2c16
JH
2958 maxiters = 2 * slen + 10; /* We can match twice at each
2959 position, once with zero-length,
2960 second time with non-zero. */
a0d0e21e 2961
6a97c51d 2962 if (!RX_PRELEN(rx) && PL_curpm
8d919b0a 2963 && !ReANY(rx)->mother_re) {
3280af22 2964 pm = PL_curpm;
aaa362c4 2965 rx = PM_GETRE(pm);
a0d0e21e 2966 }
6502e081 2967
6e240d0b 2968#ifdef PERL_SAWAMPERSAND
6502e081
DM
2969 r_flags = ( RX_NPARENS(rx)
2970 || PL_sawampersand
6502e081 2971 || (RX_EXTFLAGS(rx) & (RXf_EVAL_SEEN|RXf_PMf_KEEPCOPY))
5b0e71e9 2972 || (rpm->op_pmflags & PMf_KEEPCOPY)
6502e081
DM
2973 )
2974 ? REXEC_COPY_STR
2975 : 0;
6e240d0b
FC
2976#else
2977 r_flags = REXEC_COPY_STR;
2978#endif
7fba1cd6 2979
0395280b 2980 if (!CALLREGEXEC(rx, orig, strend, orig, 0, TARG, NULL, r_flags))
8b64c330 2981 {
5e79dfb9
DM
2982 SPAGAIN;
2983 PUSHs(rpm->op_pmflags & PMf_NONDESTRUCT ? TARG : &PL_sv_no);
2984 LEAVE_SCOPE(oldsave);
2985 RETURN;
2986 }
1754320d
FC
2987 PL_curpm = pm;
2988
71be2cbc 2989 /* known replacement string? */
f272994b 2990 if (dstr) {
8514a05a
JH
2991 /* replacement needing upgrading? */
2992 if (DO_UTF8(TARG) && !doutf8) {
db79b45b 2993 nsv = sv_newmortal();
4a176938 2994 SvSetSV(nsv, dstr);
47e13f24 2995 if (IN_ENCODING)
ad2de1b2 2996 sv_recode_to_utf8(nsv, _get_encoding());
8514a05a
JH
2997 else
2998 sv_utf8_upgrade(nsv);
5c144d81 2999 c = SvPV_const(nsv, clen);
4a176938
JH
3000 doutf8 = TRUE;
3001 }
3002 else {
5c144d81 3003 c = SvPV_const(dstr, clen);
4a176938 3004 doutf8 = DO_UTF8(dstr);
8514a05a 3005 }
bb933b9b
FC
3006
3007 if (SvTAINTED(dstr))
3008 rxtainted |= SUBST_TAINT_REPL;
f272994b
A
3009 }
3010 else {
6136c704 3011 c = NULL;
f272994b
A
3012 doutf8 = FALSE;
3013 }
3014
71be2cbc 3015 /* can do inplace substitution? */
ed252734 3016 if (c
db2c6cb3 3017#ifdef PERL_ANY_COW
ed252734
NC
3018 && !is_cow
3019#endif
fbfb1899 3020 && (I32)clen <= RX_MINLENRET(rx)
9cefd268
FC
3021 && ( once
3022 || !(r_flags & REXEC_COPY_STR)
3023 || (!SvGMAGICAL(dstr) && !(RX_EXTFLAGS(rx) & RXf_EVAL_SEEN))
3024 )
dbc200c5 3025 && !(RX_EXTFLAGS(rx) & RXf_NO_INPLACE_SUBST)
8ca8a454
NC
3026 && (!doutf8 || SvUTF8(TARG))
3027 && !(rpm->op_pmflags & PMf_NONDESTRUCT))
8b030b38 3028 {
ec911639 3029
db2c6cb3 3030#ifdef PERL_ANY_COW
ed252734 3031 if (SvIsCOW(TARG)) {
f7a8268c 3032 if (!force_on_match)
ed252734 3033 goto have_a_cow;
f7a8268c 3034 assert(SvVOK(TARG));
ed252734
NC
3035 }
3036#endif
71be2cbc 3037 if (force_on_match) {
6ac6605d
DM
3038 /* redo the first match, this time with the orig var
3039 * forced into being a string */
71be2cbc 3040 force_on_match = 0;
6ac6605d 3041 orig = SvPV_force_nomg(TARG, len);
71be2cbc
PP
3042 goto force_it;
3043 }
39b40493 3044
71be2cbc 3045 if (once) {
c67ab8f2 3046 char *d, *m;
20be6587
DM
3047 if (RX_MATCH_TAINTED(rx)) /* run time pattern taint, eg locale */
3048 rxtainted |= SUBST_TAINT_PAT;
07bc277f
NC
3049 m = orig + RX_OFFS(rx)[0].start;
3050 d = orig + RX_OFFS(rx)[0].end;
71be2cbc
PP
3051 s = orig;
3052 if (m - s > strend - d) { /* faster to shorten from end */
2ec7214c 3053 I32 i;
71be2cbc
PP
3054 if (clen) {
3055 Copy(c, m, clen, char);
3056 m += clen;
a0d0e21e 3057 }
71be2cbc
PP
3058 i = strend - d;
3059 if (i > 0) {
3060 Move(d, m, i, char);
3061 m += i;
a0d0e21e 3062 }
71be2cbc
PP
3063 *m = '\0';
3064 SvCUR_set(TARG, m - s);
3065 }
2ec7214c
DM
3066 else { /* faster from front */
3067 I32 i = m - s;
71be2cbc 3068 d -= clen;
2ec7214c
DM
3069 if (i > 0)
3070 Move(s, d - i, i, char);
71be2cbc 3071 sv_chop(TARG, d-i);
71be2cbc 3072 if (clen)
c947cd8d 3073 Copy(c, d, clen, char);
71be2cbc 3074 }
8ec5e241 3075 SPAGAIN;
8ca8a454 3076 PUSHs(&PL_sv_yes);
71be2cbc
PP
3077 }
3078 else {
c67ab8f2 3079 char *d, *m;
0395280b 3080 d = s = RX_OFFS(rx)[0].start + orig;
71be2cbc 3081 do {
2b25edcf 3082 I32 i;
5d9574c1 3083 if (UNLIKELY(iters++ > maxiters))
cea2e8a9 3084 DIE(aTHX_ "Substitution loop");
5d9574c1 3085 if (UNLIKELY(RX_MATCH_TAINTED(rx))) /* run time pattern taint, eg locale */
20be6587 3086 rxtainted |= SUBST_TAINT_PAT;
07bc277f 3087 m = RX_OFFS(rx)[0].start + orig;
155aba94 3088 if ((i = m - s)) {
71be2cbc
PP
3089 if (s != d)
3090 Move(s, d, i, char);
3091 d += i;
a0d0e21e 3092 }
71be2cbc
PP
3093 if (clen) {
3094 Copy(c, d, clen, char);
3095 d += clen;
3096 }
07bc277f 3097 s = RX_OFFS(rx)[0].end + orig;
7ce41e5c
FC
3098 } while (CALLREGEXEC(rx, s, strend, orig,
3099 s == m, /* don't match same null twice */
f722798b 3100 TARG, NULL,
d5e7783a 3101 REXEC_NOT_FIRST|REXEC_IGNOREPOS|REXEC_FAIL_ON_UNDERFLOW));
71be2cbc 3102 if (s != d) {
2b25edcf 3103 I32 i = strend - s;
aa07b2f6 3104 SvCUR_set(TARG, d - SvPVX_const(TARG) + i);
71be2cbc 3105 Move(s, d, i+1, char); /* include the NUL */
a0d0e21e 3106 }
8ec5e241 3107 SPAGAIN;
3c6ef0a5 3108 mPUSHi(iters);
a0d0e21e
LW
3109 }
3110 }
ff6e92e8 3111 else {
1754320d 3112 bool first;
c67ab8f2 3113 char *m;
1754320d 3114 SV *repl;
a0d0e21e 3115 if (force_on_match) {
6ac6605d
DM
3116 /* redo the first match, this time with the orig var
3117 * forced into being a string */
a0d0e21e 3118 force_on_match = 0;
0c1438a1
NC
3119 if (rpm->op_pmflags & PMf_NONDESTRUCT) {
3120 /* I feel that it should be possible to avoid this mortal copy
3121 given that the code below copies into a new destination.
3122 However, I suspect it isn't worth the complexity of
3123 unravelling the C<goto force_it> for the small number of
3124 cases where it would be viable to drop into the copy code. */
3125 TARG = sv_2mortal(newSVsv(TARG));
3126 }
6ac6605d 3127 orig = SvPV_force_nomg(TARG, len);
a0d0e21e
LW
3128 goto force_it;
3129 }
db2c6cb3 3130#ifdef PERL_ANY_COW
ed252734
NC
3131 have_a_cow:
3132#endif
20be6587
DM
3133 if (RX_MATCH_TAINTED(rx)) /* run time pattern taint, eg locale */
3134 rxtainted |= SUBST_TAINT_PAT;
1754320d 3135 repl = dstr;
0395280b
DM
3136 s = RX_OFFS(rx)[0].start + orig;
3137 dstr = newSVpvn_flags(orig, s-orig,
3138 SVs_TEMP | (DO_UTF8(TARG) ? SVf_UTF8 : 0));
a0d0e21e 3139 if (!c) {
eb578fdb 3140 PERL_CONTEXT *cx;
8ec5e241 3141 SPAGAIN;
0395280b 3142 m = orig;
20be6587
DM
3143 /* note that a whole bunch of local vars are saved here for
3144 * use by pp_substcont: here's a list of them in case you're
3145 * searching for places in this sub that uses a particular var:
3146 * iters maxiters r_flags oldsave rxtainted orig dstr targ
3147 * s m strend rx once */
490576d1 3148 CX_PUSHSUBST(cx);
20e98b0f 3149 RETURNOP(cPMOP->op_pmreplrootu.op_pmreplroot);
a0d0e21e 3150 }
1754320d 3151 first = TRUE;
a0d0e21e 3152 do {
5d9574c1 3153 if (UNLIKELY(iters++ > maxiters))
cea2e8a9 3154 DIE(aTHX_ "Substitution loop");
5d9574c1 3155 if (UNLIKELY(RX_MATCH_TAINTED(rx)))
20be6587 3156 rxtainted |= SUBST_TAINT_PAT;
07bc277f 3157 if (RX_MATCH_COPIED(rx) && RX_SUBBEG(rx) != orig) {
c67ab8f2
DM
3158 char *old_s = s;
3159 char *old_orig = orig;
6502e081 3160 assert(RX_SUBOFFSET(rx) == 0);
c67ab8f2 3161
07bc277f 3162 orig = RX_SUBBEG(rx);
c67ab8f2
DM
3163 s = orig + (old_s - old_orig);
3164 strend = s + (strend - old_s);
a0d0e21e 3165 }
07bc277f 3166 m = RX_OFFS(rx)[0].start + orig;
64534138 3167 sv_catpvn_nomg_maybeutf8(dstr, s, m - s, DO_UTF8(TARG));
07bc277f 3168 s = RX_OFFS(rx)[0].end + orig;
1754320d
FC
3169 if (first) {
3170 /* replacement already stringified */
3171 if (clen)
64534138 3172 sv_catpvn_nomg_maybeutf8(dstr, c, clen, doutf8);
1754320d
FC
3173 first = FALSE;
3174 }
3175 else {
47e13f24 3176 if (IN_ENCODING) {
1754320d
FC
3177 if (!nsv) nsv = sv_newmortal();
3178 sv_copypv(nsv, repl);
ad2de1b2 3179 if (!DO_UTF8(nsv)) sv_recode_to_utf8(nsv, _get_encoding());
1754320d
FC
3180 sv_catsv(dstr, nsv);
3181 }
3182 else sv_catsv(dstr, repl);
5d9574c1 3183 if (UNLIKELY(SvTAINTED(repl)))
bb933b9b 3184 rxtainted |= SUBST_TAINT_REPL;
1754320d 3185 }
a0d0e21e
LW
3186 if (once)
3187 break;
ff27773b
KW
3188 } while (CALLREGEXEC(rx, s, strend, orig,
3189 s == m, /* Yields minend of 0 or 1 */
d5e7783a
DM
3190 TARG, NULL,
3191 REXEC_NOT_FIRST|REXEC_IGNOREPOS|REXEC_FAIL_ON_UNDERFLOW));
aba224f7 3192 assert(strend >= s);
64534138 3193 sv_catpvn_nomg_maybeutf8(dstr, s, strend - s, DO_UTF8(TARG));
748a9306 3194
8ca8a454
NC
3195 if (rpm->op_pmflags & PMf_NONDESTRUCT) {
3196 /* From here on down we're using the copy, and leaving the original
3197 untouched. */
3198 TARG = dstr;
3199 SPAGAIN;
3200 PUSHs(dstr);
3201 } else {
db2c6cb3 3202#ifdef PERL_ANY_COW
8ca8a454
NC
3203 /* The match may make the string COW. If so, brilliant, because
3204 that's just saved us one malloc, copy and free - the regexp has
3205 donated the old buffer, and we malloc an entirely new one, rather
3206 than the regexp malloc()ing a buffer and copying our original,
3207 only for us to throw it away here during the substitution. */
3208 if (SvIsCOW(TARG)) {
3209 sv_force_normal_flags(TARG, SV_COW_DROP_PV);
3210 } else
ed252734 3211#endif
8ca8a454
NC
3212 {
3213 SvPV_free(TARG);
3214 }
3215 SvPV_set(TARG, SvPVX(dstr));
3216 SvCUR_set(TARG, SvCUR(dstr));
3217 SvLEN_set(TARG, SvLEN(dstr));
64534138 3218 SvFLAGS(TARG) |= SvUTF8(dstr);
8ca8a454 3219 SvPV_set(dstr, NULL);
748a9306 3220
8ca8a454 3221 SPAGAIN;
3c6ef0a5 3222 mPUSHi(iters);
8ca8a454
NC
3223 }
3224 }
3225
3226 if (!(rpm->op_pmflags & PMf_NONDESTRUCT)) {
3227 (void)SvPOK_only_UTF8(TARG);
a0d0e21e 3228 }
20be6587 3229
ef07e810 3230 /* See "how taint works" above */
284167a5 3231 if (TAINTING_get) {
20be6587
DM
3232 if ((rxtainted & SUBST_TAINT_PAT) ||
3233 ((rxtainted & (SUBST_TAINT_STR|SUBST_TAINT_RETAINT)) ==
3234 (SUBST_TAINT_STR|SUBST_TAINT_RETAINT))
3235 )
3236 (RX_MATCH_TAINTED_on(rx)); /* taint $1 et al */
3237
3238 if (!(rxtainted & SUBST_TAINT_BOOLRET)
3239 && (rxtainted & (SUBST_TAINT_STR|SUBST_TAINT_PAT))
3240 )
3241 SvTAINTED_on(TOPs); /* taint return value */
3242 else
3243 SvTAINTED_off(TOPs); /* may have got tainted earlier */
3244
3245 /* needed for mg_set below */
284167a5
SM
3246 TAINT_set(
3247 cBOOL(rxtainted & (SUBST_TAINT_STR|SUBST_TAINT_PAT|SUBST_TAINT_REPL))
3248 );
20be6587
DM
3249 SvTAINT(TARG);
3250 }
3251 SvSETMAGIC(TARG); /* PL_tainted must be correctly set for this mg_set */
3252 TAINT_NOT;
f1a76097
DM
3253 LEAVE_SCOPE(oldsave);
3254 RETURN;
a0d0e21e
LW
3255}
3256
3257PP(pp_grepwhile)
3258{
20b7effb 3259 dSP;
a0d0e21e
LW
3260
3261 if (SvTRUEx(POPs))
3280af22
NIS
3262 PL_stack_base[PL_markstack_ptr[-1]++] = PL_stack_base[*PL_markstack_ptr];
3263 ++*PL_markstack_ptr;
b2a2a901 3264 FREETMPS;
d343c3ef 3265 LEAVE_with_name("grep_item"); /* exit inner scope */
a0d0e21e
LW
3266
3267 /* All done yet? */
5d9574c1 3268 if (UNLIKELY(PL_stack_base + *PL_markstack_ptr > SP)) {
a0d0e21e 3269 I32 items;
1c23e2bd 3270 const U8 gimme = GIMME_V;
a0d0e21e 3271
d343c3ef 3272 LEAVE_with_name("grep"); /* exit outer scope */
a0d0e21e 3273 (void)POPMARK; /* pop src */
3280af22 3274 items = --*PL_markstack_ptr - PL_markstack_ptr[-1];
a0d0e21e 3275 (void)POPMARK; /* pop dst */
3280af22 3276 SP = PL_stack_base + POPMARK; /* pop original mark */
54310121 3277 if (gimme == G_SCALAR) {
7cc47870
RGS
3278 dTARGET;
3279 XPUSHi(items);
a0d0e21e 3280 }
54310121
PP
3281 else if (gimme == G_ARRAY)
3282 SP += items;
a0d0e21e
LW
3283 RETURN;
3284 }
3285 else {
3286 SV *src;
3287
d343c3ef 3288 ENTER_with_name("grep_item"); /* enter inner scope */
1d7c1841 3289 SAVEVPTR(PL_curpm);
a0d0e21e 3290
6cae08a8 3291 src = PL_stack_base[TOPMARK];
60779a30 3292 if (SvPADTMP(src)) {
6cae08a8 3293 src = PL_stack_base[TOPMARK] = sv_mortalcopy(src);
a0ed822e
FC
3294 PL_tmps_floor++;
3295 }
a0d0e21e 3296 SvTEMP_off(src);
ffd49c98 3297 DEFSV_set(src);
a0d0e21e
LW
3298
3299 RETURNOP(cLOGOP->op_other);
3300 }
3301}
3302
799da9d7 3303/* leave_adjust_stacks():
f7a874b8 3304 *
e02ce34b
DM
3305 * Process a scope's return args (in the range from_sp+1 .. PL_stack_sp),
3306 * positioning them at to_sp+1 onwards, and do the equivalent of a
3307 * FREEMPS and TAINT_NOT.
3308 *
f7a874b8
DM
3309 * Not intended to be called in void context.
3310 *
799da9d7
DM
3311 * When leaving a sub, eval, do{} or other scope, the things that need
3312 * doing to process the return args are:
f7a874b8 3313 * * in scalar context, only return the last arg (or PL_sv_undef if none);
799da9d7
DM
3314 * * for the types of return that return copies of their args (such
3315 * as rvalue sub return), make a mortal copy of every return arg,
3316 * except where we can optimise the copy away without it being
3317 * semantically visible;
3318 * * make sure that the arg isn't prematurely freed; in the case of an
3319 * arg not copied, this may involve mortalising it. For example, in
f7a874b8
DM
3320 * C<sub f { my $x = ...; $x }>, $x would be freed when we do
3321 * CX_LEAVE_SCOPE(cx) unless it's protected or copied.
3322 *
799da9d7
DM
3323 * What condition to use when deciding whether to pass the arg through
3324 * or make a copy, is determined by the 'pass' arg; its valid values are:
3325 * 0: rvalue sub/eval exit
3326 * 1: other rvalue scope exit
3327 * 2: :lvalue sub exit in rvalue context
3328 * 3: :lvalue sub exit in lvalue context and other lvalue scope exits
3329 *
f7a874b8 3330 * There is a big issue with doing a FREETMPS. We would like to free any
799da9d7 3331 * temps created by the last statement which the sub executed, rather than
f7a874b8
DM
3332 * leaving them for the caller. In a situation where a sub call isn't
3333 * soon followed by a nextstate (e.g. nested recursive calls, a la
3334 * fibonacci()), temps can accumulate, causing memory and performance
3335 * issues.
3336 *
3337 * On the other hand, we don't want to free any TEMPs which are keeping
799da9d7
DM
3338 * alive any return args that we skipped copying; nor do we wish to undo
3339 * any mortalising done here.
f7a874b8
DM
3340 *
3341 * The solution is to split the temps stack frame into two, with a cut
3342 * point delineating the two halves. We arrange that by the end of this
3343 * function, all the temps stack frame entries we wish to keep are in the
799da9d7 3344 * range PL_tmps_floor+1.. tmps_base-1, while the ones to free now are in
f7a874b8
DM
3345 * the range tmps_base .. PL_tmps_ix. During the course of this
3346 * function, tmps_base starts off as PL_tmps_floor+1, then increases
3347 * whenever we find or create a temp that we know should be kept. In
3348 * general the stuff above tmps_base is undecided until we reach the end,
3349 * and we may need a sort stage for that.
3350 *
3351 * To determine whether a TEMP is keeping a return arg alive, every
3352 * arg that is kept rather than copied and which has the SvTEMP flag
3353 * set, has the flag temporarily unset, to mark it. At the end we scan
799da9d7 3354 * the temps stack frame above the cut for entries without SvTEMP and
f7a874b8 3355 * keep them, while turning SvTEMP on again. Note that if we die before
799da9d7 3356 * the SvTEMPs flags are set again, its safe: at worst, subsequent use of
f7a874b8
DM
3357 * those SVs may be slightly less efficient.
3358 *
3359 * In practice various optimisations for some common cases mean we can
3360 * avoid most of the scanning and swapping about with the temps stack.
3361 */
3362
799da9d7 3363void
1c23e2bd 3364Perl_leave_adjust_stacks(pTHX_ SV **from_sp, SV **to_sp, U8 gimme, int pass)
a0d0e21e 3365{
263e0548 3366 dVAR;
20b7effb 3367 dSP;
f7a874b8
DM
3368 SSize_t tmps_base; /* lowest index into tmps stack that needs freeing now */
3369 SSize_t nargs;
3370
799da9d7
DM
3371 PERL_ARGS_ASSERT_LEAVE_ADJUST_STACKS;
3372
f7a874b8
DM
3373 TAINT_NOT;
3374
3375 if (gimme == G_ARRAY) {
e02ce34b
DM
3376 nargs = SP - from_sp;
3377 from_sp++;
f7a874b8
DM
3378 }
3379 else {
3380 assert(gimme == G_SCALAR);
e02ce34b 3381 if (UNLIKELY(from_sp >= SP)) {
f7a874b8 3382 /* no return args */
e02ce34b 3383 assert(from_sp == SP);
f7a874b8
DM
3384 EXTEND(SP, 1);
3385 *++SP = &PL_sv_undef;
e02ce34b 3386 to_sp = SP;
f7a874b8
DM
3387 nargs = 0;
3388 }
3389 else {
3390 from_sp = SP;
3391 nargs = 1;
3392 }
3393 }
3394
3395 /* common code for G_SCALAR and G_ARRAY */
3396
3397 tmps_base = PL_tmps_floor + 1;
3398
3399 assert(nargs >= 0);
3400 if (nargs) {
3401 /* pointer version of tmps_base. Not safe across temp stack
3402 * reallocs. */
3403 SV **tmps_basep;
3404
3405 EXTEND_MORTAL(nargs); /* one big extend for worst-case scenario */
3406 tmps_basep = PL_tmps_stack + tmps_base;
f7a874b8
DM
3407
3408 /* process each return arg */
3409
3410 do {
3411 SV *sv = *from_sp++;
3412
3413 assert(PL_tmps_ix + nargs < PL_tmps_max);
3645bb38
DM
3414#ifdef DEBUGGING
3415 /* PADTMPs with container set magic shouldn't appear in the
3416 * wild. This assert is more important for pp_leavesublv(),
3417 * but by testing for it here, we're more likely to catch
3418 * bad cases (what with :lvalue subs not being widely
3419 * deployed). The two issues are that for something like
3420 * sub :lvalue { $tied{foo} }
3421 * or
3422 * sub :lvalue { substr($foo,1,2) }
3423 * pp_leavesublv() will croak if the sub returns a PADTMP,
3424 * and currently functions like pp_substr() return a mortal
3425 * rather than using their PADTMP when returning a PVLV.
3426 * This is because the PVLV will hold a ref to $foo,
3427 * so $foo would get delayed in being freed while
3428 * the PADTMP SV remained in the PAD.
3429 * So if this assert fails it means either:
3430 * 1) there is pp code similar to pp_substr that is
3431 * returning a PADTMP instead of a mortal, and probably
3432 * needs fixing, or
5d9c1c9a 3433 * 2) pp_leavesublv is making unwarranted assumptions
3645bb38
DM
3434 * about always croaking on a PADTMP
3435 */
3436 if (SvPADTMP(sv) && SvSMAGICAL(sv)) {
3437 MAGIC *mg;
3438 for (mg = SvMAGIC(sv); mg; mg = mg->mg_moremagic) {
3439 assert(PERL_MAGIC_TYPE_IS_VALUE_MAGIC(mg->mg_type));
3440 }
3441 }
3442#endif
f7a874b8 3443
799da9d7
DM
3444 if (
3445 pass == 0 ? (SvTEMP(sv) && !SvMAGICAL(sv) && SvREFCNT(sv) == 1)
3446 : pass == 1 ? ((SvTEMP(sv) || SvPADTMP(sv)) && !SvMAGICAL(sv) && SvREFCNT(sv) == 1)
3447 : pass == 2 ? (!SvPADTMP(sv))
3448 : 1)
3449 {
3450 /* pass through: skip copy for logic or optimisation
3451 * reasons; instead mortalise it, except that ... */
e02ce34b 3452 *++to_sp = sv;
f7a874b8 3453
799da9d7
DM
3454 if (SvTEMP(sv)) {
3455 /* ... since this SV is an SvTEMP , we don't need to
3456 * re-mortalise it; instead we just need to ensure
3457 * that its existing entry in the temps stack frame
3458 * ends up below the cut and so avoids being freed
3459 * this time round. We mark it as needing to be kept
3460 * by temporarily unsetting SvTEMP; then at the end,
3461 * we shuffle any !SvTEMP entries on the tmps stack
3462 * back below the cut.
3463 * However, there's a significant chance that there's
3464 * a 1:1 correspondence between the first few (or all)
3465 * elements in the return args stack frame and those
3466 * in the temps stack frame; e,g.:
3467 * sub f { ....; map {...} .... },
3468 * or if we're exiting multiple scopes and one of the
3469 * inner scopes has already made mortal copies of each
3470 * return arg.
3471 *
3472 * If so, this arg sv will correspond to the next item
3473 * on the tmps stack above the cut, and so can be kept
3474 * merely by moving the cut boundary up one, rather
3475 * than messing with SvTEMP. If all args are 1:1 then
3476 * we can avoid the sorting stage below completely.
977d0c81
DM
3477 *
3478 * If there are no items above the cut on the tmps
3479 * stack, then the SvTEMP must comne from an item
3480 * below the cut, so there's nothing to do.
799da9d7 3481 */
977d0c81
DM
3482 if (tmps_basep <= &PL_tmps_stack[PL_tmps_ix]) {
3483 if (sv == *tmps_basep)
3484 tmps_basep++;
3485 else
3486 SvTEMP_off(sv);
3487 }
799da9d7 3488 }
75bc488d 3489 else if (!SvPADTMP(sv)) {
799da9d7 3490 /* mortalise arg to avoid it being freed during save
75bc488d 3491 * stack unwinding. Pad tmps don't need mortalising as
977d0c81
DM
3492 * they're never freed. This is the equivalent of
3493 * sv_2mortal(SvREFCNT_inc(sv)), except that:
799da9d7
DM
3494 * * it assumes that the temps stack has already been
3495 * extended;
3496 * * it puts the new item at the cut rather than at
3497 * ++PL_tmps_ix, moving the previous occupant there
3498 * instead.
3499 */
3500 if (!SvIMMORTAL(sv)) {
977d0c81 3501 SvREFCNT_inc_simple_void_NN(sv);
799da9d7 3502 SvTEMP_on(sv);
977d0c81
DM
3503 /* Note that if there's nothing above the cut,
3504 * this copies the garbage one slot above
3505 * PL_tmps_ix onto itself. This is harmless (the
3506 * stack's already been extended), but might in
3507 * theory trigger warnings from tools like ASan
3508 */
799da9d7
DM
3509 PL_tmps_stack[++PL_tmps_ix] = *tmps_basep;
3510 *tmps_basep++ = sv;
3511 }
3512 }
f7a874b8
DM
3513 }
3514 else {
3515 /* Make a mortal copy of the SV.
3516 * The following code is the equivalent of sv_mortalcopy()
3517 * except that:
3518 * * it assumes the temps stack has already been extended;
3519 * * it optimises the copying for some simple SV types;
3520 * * it puts the new item at the cut rather than at
3521 * ++PL_tmps_ix, moving the previous occupant there
3522 * instead.
3523 */
3524 SV *newsv = newSV(0);
3525
3526 PL_tmps_stack[++PL_tmps_ix] = *tmps_basep;
3527 /* put it on the tmps stack early so it gets freed if we die */
3528 *tmps_basep++ = newsv;
e02ce34b 3529 *++to_sp = newsv;
f7a874b8
DM
3530
3531 if (SvTYPE(sv) <= SVt_IV) {
3532 /* arg must be one of undef, IV/UV, or RV: skip
3533 * sv_setsv_flags() and do the copy directly */
3534 U32 dstflags;
3535 U32 srcflags = SvFLAGS(sv);
3536
3537 assert(!SvGMAGICAL(sv));
3538 if (srcflags & (SVf_IOK|SVf_ROK)) {
3539 SET_SVANY_FOR_BODYLESS_IV(newsv);
3540
3541 if (srcflags & SVf_ROK) {
3542 newsv->sv_u.svu_rv = SvREFCNT_inc(SvRV(sv));
3543 /* SV type plus flags */
3544 dstflags = (SVt_IV|SVf_ROK|SVs_TEMP);
3545 }
3546 else {
3547 /* both src and dst are <= SVt_IV, so sv_any
3548 * points to the head; so access the heads
3549 * directly rather than going via sv_any.
3550 */
3551 assert( &(sv->sv_u.svu_iv)
3552 == &(((XPVIV*) SvANY(sv))->xiv_iv));
3553 assert( &(newsv->sv_u.svu_iv)
3554 == &(((XPVIV*) SvANY(newsv))->xiv_iv));
3555 newsv->sv_u.svu_iv = sv->sv_u.svu_iv;
3556 /* SV type plus flags */
3557 dstflags = (SVt_IV|SVf_IOK|SVp_IOK|SVs_TEMP
3558 |(srcflags & SVf_IVisUV));
3559 }
3560 }
3561 else {
3562 assert(!(srcflags & SVf_OK));
3563 dstflags = (SVt_NULL|SVs_TEMP); /* SV type plus flags */
3564 }
3565 SvFLAGS(newsv) = dstflags;
3566
3567 }
3568 else {
3569 /* do the full sv_setsv() */
3570 SSize_t old_base;
3571
3572 SvTEMP_on(newsv);
3573 old_base = tmps_basep - PL_tmps_stack;
3574 SvGETMAGIC(sv);
3575 sv_setsv_flags(newsv, sv, SV_DO_COW_SVSETSV);
799da9d7 3576 /* the mg_get or sv_setsv might have created new temps
f7a874b8
DM
3577 * or realloced the tmps stack; regrow and reload */
3578 EXTEND_MORTAL(nargs);
3579 tmps_basep = PL_tmps_stack + old_base;
3580 TAINT_NOT; /* Each item is independent */
3581 }
3582
3583 }
3584 } while (--nargs);
3585
3586 /* If there are any temps left above the cut, we need to sort
3587 * them into those to keep and those to free. The only ones to
3588 * keep are those for which we've temporarily unset SvTEMP.
3589 * Work inwards from the two ends at tmps_basep .. PL_tmps_ix,
3590 * swapping pairs as necessary. Stop when we meet in the middle.
3591 */
3592 {
3593 SV **top = PL_tmps_stack + PL_tmps_ix;
3594 while (tmps_basep <= top) {
3595 SV *sv = *top;
3596 if (SvTEMP(sv))