This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Concise: improve hints decoding
[perl5.git] / pp_hot.c
CommitLineData
a0d0e21e
LW
1/* pp_hot.c
2 *
1129b882
NC
3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 by Larry Wall and others
a0d0e21e
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
12 * Then he heard Merry change the note, and up went the Horn-cry of Buckland,
13 * shaking the air.
14 *
4ac71550
TC
15 * Awake! Awake! Fear, Fire, Foes! Awake!
16 * Fire, Foes! Awake!
17 *
18 * [p.1007 of _The Lord of the Rings_, VI/viii: "The Scouring of the Shire"]
a0d0e21e
LW
19 */
20
166f8a29
DM
21/* This file contains 'hot' pp ("push/pop") functions that
22 * execute the opcodes that make up a perl program. A typical pp function
23 * expects to find its arguments on the stack, and usually pushes its
24 * results onto the stack, hence the 'pp' terminology. Each OP structure
25 * contains a pointer to the relevant pp_foo() function.
26 *
27 * By 'hot', we mean common ops whose execution speed is critical.
28 * By gathering them together into a single file, we encourage
29 * CPU cache hits on hot code. Also it could be taken as a warning not to
30 * change any code in this file unless you're sure it won't affect
31 * performance.
32 */
33
a0d0e21e 34#include "EXTERN.h"
864dbfa3 35#define PERL_IN_PP_HOT_C
a0d0e21e
LW
36#include "perl.h"
37
38/* Hot code. */
39
40PP(pp_const)
41{
39644a26 42 dSP;
996c9baa 43 XPUSHs(cSVOP_sv);
a0d0e21e
LW
44 RETURN;
45}
46
47PP(pp_nextstate)
48{
533c011a 49 PL_curcop = (COP*)PL_op;
a0d0e21e 50 TAINT_NOT; /* Each statement is presumed innocent */
4ebe6e95 51 PL_stack_sp = PL_stack_base + CX_CUR()->blk_oldsp;
a0d0e21e 52 FREETMPS;
f410a211 53 PERL_ASYNC_CHECK();
a0d0e21e
LW
54 return NORMAL;
55}
56
57PP(pp_gvsv)
58{
39644a26 59 dSP;
924508f0 60 EXTEND(SP,1);
5d9574c1 61 if (UNLIKELY(PL_op->op_private & OPpLVAL_INTRO))
1d7c1841 62 PUSHs(save_scalar(cGVOP_gv));
a0d0e21e 63 else
c69033f2 64 PUSHs(GvSVn(cGVOP_gv));
a0d0e21e
LW
65 RETURN;
66}
67
b1c05ba5
DM
68
69/* also used for: pp_lineseq() pp_regcmaybe() pp_scalar() pp_scope() */
70
a0d0e21e
LW
71PP(pp_null)
72{
73 return NORMAL;
74}
75
3dd9d4e4
FC
76/* This is sometimes called directly by pp_coreargs, pp_grepstart and
77 amagic_call. */
a0d0e21e
LW
78PP(pp_pushmark)
79{
3280af22 80 PUSHMARK(PL_stack_sp);
a0d0e21e
LW
81 return NORMAL;
82}
83
84PP(pp_stringify)
85{
20b7effb 86 dSP; dTARGET;
4cc783ef
DD
87 SV * const sv = TOPs;
88 SETs(TARG);
89 sv_copypv(TARG, sv);
90 SvSETMAGIC(TARG);
91 /* no PUTBACK, SETs doesn't inc/dec SP */
92 return NORMAL;
a0d0e21e
LW
93}
94
95PP(pp_gv)
96{
20b7effb 97 dSP;
ad64d0ec 98 XPUSHs(MUTABLE_SV(cGVOP_gv));
a0d0e21e
LW
99 RETURN;
100}
101
b1c05ba5
DM
102
103/* also used for: pp_andassign() */
104
a0d0e21e
LW
105PP(pp_and)
106{
f410a211 107 PERL_ASYNC_CHECK();
4cc783ef
DD
108 {
109 /* SP is not used to remove a variable that is saved across the
110 sv_2bool_flags call in SvTRUE_NN, if a RISC/CISC or low/high machine
111 register or load/store vs direct mem ops macro is introduced, this
112 should be a define block between direct PL_stack_sp and dSP operations,
113 presently, using PL_stack_sp is bias towards CISC cpus */
114 SV * const sv = *PL_stack_sp;
115 if (!SvTRUE_NN(sv))
116 return NORMAL;
117 else {
118 if (PL_op->op_type == OP_AND)
119 --PL_stack_sp;
120 return cLOGOP->op_other;
121 }
a0d0e21e
LW
122 }
123}
124
125PP(pp_sassign)
126{
20b7effb 127 dSP;
3e75a3c4
RU
128 /* sassign keeps its args in the optree traditionally backwards.
129 So we pop them differently.
130 */
131 SV *left = POPs; SV *right = TOPs;
748a9306 132
533c011a 133 if (PL_op->op_private & OPpASSIGN_BACKWARDS) {
0bd48802
AL
134 SV * const temp = left;
135 left = right; right = temp;
a0d0e21e 136 }
d48c660d
DM
137 assert(TAINTING_get || !TAINT_get);
138 if (UNLIKELY(TAINT_get) && !SvTAINTED(right))
a0d0e21e 139 TAINT_NOT;
5d9574c1
DM
140 if (UNLIKELY(PL_op->op_private & OPpASSIGN_CV_TO_GV)) {
141 /* *foo =\&bar */
3e75a3c4 142 SV * const cv = SvRV(right);
e26df76a 143 const U32 cv_type = SvTYPE(cv);
3e75a3c4 144 const bool is_gv = isGV_with_GP(left);
6136c704 145 const bool got_coderef = cv_type == SVt_PVCV || cv_type == SVt_PVFM;
e26df76a
NC
146
147 if (!got_coderef) {
148 assert(SvROK(cv));
149 }
150
3e75a3c4
RU
151 /* Can do the optimisation if left (LVALUE) is not a typeglob,
152 right (RVALUE) is a reference to something, and we're in void
e26df76a 153 context. */
13be902c 154 if (!got_coderef && !is_gv && GIMME_V == G_VOID) {
e26df76a 155 /* Is the target symbol table currently empty? */
3e75a3c4 156 GV * const gv = gv_fetchsv_nomg(left, GV_NOINIT, SVt_PVGV);
bb112e5a 157 if (SvTYPE(gv) != SVt_PVGV && !SvOK(gv)) {
e26df76a
NC
158 /* Good. Create a new proxy constant subroutine in the target.
159 The gv becomes a(nother) reference to the constant. */
160 SV *const value = SvRV(cv);
161
ad64d0ec 162 SvUPGRADE(MUTABLE_SV(gv), SVt_IV);
1ccdb730 163 SvPCS_IMPORTED_on(gv);
e26df76a 164 SvRV_set(gv, value);
b37c2d43 165 SvREFCNT_inc_simple_void(value);
3e75a3c4 166 SETs(left);
e26df76a
NC
167 RETURN;
168 }
169 }
170
171 /* Need to fix things up. */
13be902c 172 if (!is_gv) {
e26df76a 173 /* Need to fix GV. */
3e75a3c4 174 left = MUTABLE_SV(gv_fetchsv_nomg(left,GV_ADD, SVt_PVGV));
e26df76a
NC
175 }
176
177 if (!got_coderef) {
178 /* We've been returned a constant rather than a full subroutine,
179 but they expect a subroutine reference to apply. */
53a42478 180 if (SvROK(cv)) {
d343c3ef 181 ENTER_with_name("sassign_coderef");
53a42478
NC
182 SvREFCNT_inc_void(SvRV(cv));
183 /* newCONSTSUB takes a reference count on the passed in SV
184 from us. We set the name to NULL, otherwise we get into
185 all sorts of fun as the reference to our new sub is
186 donated to the GV that we're about to assign to.
187 */
3e75a3c4 188 SvRV_set(right, MUTABLE_SV(newCONSTSUB(GvSTASH(left), NULL,
ad64d0ec 189 SvRV(cv))));
fc2b2dca 190 SvREFCNT_dec_NN(cv);
d343c3ef 191 LEAVE_with_name("sassign_coderef");
53a42478
NC
192 } else {
193 /* What can happen for the corner case *{"BONK"} = \&{"BONK"};
194 is that
195 First: ops for \&{"BONK"}; return us the constant in the
196 symbol table
197 Second: ops for *{"BONK"} cause that symbol table entry
198 (and our reference to it) to be upgraded from RV
199 to typeblob)
200 Thirdly: We get here. cv is actually PVGV now, and its
201 GvCV() is actually the subroutine we're looking for
202
203 So change the reference so that it points to the subroutine
204 of that typeglob, as that's what they were after all along.
205 */
159b6efe 206 GV *const upgraded = MUTABLE_GV(cv);
53a42478
NC
207 CV *const source = GvCV(upgraded);
208
209 assert(source);
210 assert(CvFLAGS(source) & CVf_CONST);
211
0ad694a7 212 SvREFCNT_inc_simple_void_NN(source);
fc2b2dca 213 SvREFCNT_dec_NN(upgraded);
3e75a3c4 214 SvRV_set(right, MUTABLE_SV(source));
53a42478 215 }
e26df76a 216 }
53a42478 217
e26df76a 218 }
8fe85e3f 219 if (
5d9574c1 220 UNLIKELY(SvTEMP(left)) && !SvSMAGICAL(left) && SvREFCNT(left) == 1 &&
3e75a3c4 221 (!isGV_with_GP(left) || SvFAKE(left)) && ckWARN(WARN_MISC)
8fe85e3f
FC
222 )
223 Perl_warner(aTHX_
224 packWARN(WARN_MISC), "Useless assignment to a temporary"
225 );
3e75a3c4
RU
226 SvSetMagicSV(left, right);
227 SETs(left);
a0d0e21e
LW
228 RETURN;
229}
230
231PP(pp_cond_expr)
232{
20b7effb 233 dSP;
f410a211 234 PERL_ASYNC_CHECK();
a0d0e21e 235 if (SvTRUEx(POPs))
1a67a97c 236 RETURNOP(cLOGOP->op_other);
a0d0e21e 237 else
1a67a97c 238 RETURNOP(cLOGOP->op_next);
a0d0e21e
LW
239}
240
241PP(pp_unstack)
242{
f5319de9 243 PERL_CONTEXT *cx;
8f3964af 244 PERL_ASYNC_CHECK();
a0d0e21e 245 TAINT_NOT; /* Each statement is presumed innocent */
4ebe6e95 246 cx = CX_CUR();
f5319de9 247 PL_stack_sp = PL_stack_base + cx->blk_oldsp;
a0d0e21e 248 FREETMPS;
eae48c89 249 if (!(PL_op->op_flags & OPf_SPECIAL)) {
93661e56 250 assert(CxTYPE(cx) == CXt_BLOCK || CxTYPE_is_LOOP(cx));
dfe0f39b 251 CX_LEAVE_SCOPE(cx);
eae48c89 252 }
a0d0e21e
LW
253 return NORMAL;
254}
255
a0d0e21e
LW
256PP(pp_concat)
257{
20b7effb 258 dSP; dATARGET; tryAMAGICbin_MG(concat_amg, AMGf_assign);
748a9306
LW
259 {
260 dPOPTOPssrl;
8d6d96c1
HS
261 bool lbyte;
262 STRLEN rlen;
d4c19fe8 263 const char *rpv = NULL;
a6b599c7 264 bool rbyte = FALSE;
a9c4fd4e 265 bool rcopied = FALSE;
8d6d96c1 266
6f1401dc
DM
267 if (TARG == right && right != left) { /* $r = $l.$r */
268 rpv = SvPV_nomg_const(right, rlen);
c75ab21a 269 rbyte = !DO_UTF8(right);
59cd0e26 270 right = newSVpvn_flags(rpv, rlen, SVs_TEMP);
349d4f2f 271 rpv = SvPV_const(right, rlen); /* no point setting UTF-8 here */
db79b45b 272 rcopied = TRUE;
8d6d96c1 273 }
7889fe52 274
89734059 275 if (TARG != left) { /* not $l .= $r */
a9c4fd4e 276 STRLEN llen;
6f1401dc 277 const char* const lpv = SvPV_nomg_const(left, llen);
90f5826e 278 lbyte = !DO_UTF8(left);
8d6d96c1
HS
279 sv_setpvn(TARG, lpv, llen);
280 if (!lbyte)
281 SvUTF8_on(TARG);
282 else
283 SvUTF8_off(TARG);
284 }
18ea7bf2
SM
285 else { /* $l .= $r and left == TARG */
286 if (!SvOK(left)) {
51f69a24
AC
287 if ((left == right /* $l .= $l */
288 || (PL_op->op_private & OPpTARGET_MY)) /* $l = $l . $r */
289 && ckWARN(WARN_UNINITIALIZED)
290 )
291 report_uninit(left);
76f68e9b 292 sv_setpvs(left, "");
c75ab21a 293 }
18ea7bf2
SM
294 else {
295 SvPV_force_nomg_nolen(left);
296 }
583a5589 297 lbyte = !DO_UTF8(left);
90f5826e 298 if (IN_BYTES)
18ea7bf2 299 SvUTF8_off(left);
8d6d96c1 300 }
a12c0f56 301
c75ab21a 302 if (!rcopied) {
6f1401dc 303 rpv = SvPV_nomg_const(right, rlen);
c75ab21a
RH
304 rbyte = !DO_UTF8(right);
305 }
8d6d96c1
HS
306 if (lbyte != rbyte) {
307 if (lbyte)
308 sv_utf8_upgrade_nomg(TARG);
309 else {
db79b45b 310 if (!rcopied)
59cd0e26 311 right = newSVpvn_flags(rpv, rlen, SVs_TEMP);
8d6d96c1 312 sv_utf8_upgrade_nomg(right);
6f1401dc 313 rpv = SvPV_nomg_const(right, rlen);
69b47968 314 }
a0d0e21e 315 }
8d6d96c1 316 sv_catpvn_nomg(TARG, rpv, rlen);
43ebc500 317
a0d0e21e
LW
318 SETTARG;
319 RETURN;
748a9306 320 }
a0d0e21e
LW
321}
322
d5524600
DM
323/* push the elements of av onto the stack.
324 * XXX Note that padav has similar code but without the mg_get().
325 * I suspect that the mg_get is no longer needed, but while padav
326 * differs, it can't share this function */
327
f9ae8fb6 328STATIC void
d5524600
DM
329S_pushav(pTHX_ AV* const av)
330{
331 dSP;
c70927a6 332 const SSize_t maxarg = AvFILL(av) + 1;
d5524600 333 EXTEND(SP, maxarg);
5d9574c1 334 if (UNLIKELY(SvRMAGICAL(av))) {
c70927a6
FC
335 PADOFFSET i;
336 for (i=0; i < (PADOFFSET)maxarg; i++) {
d5524600
DM
337 SV ** const svp = av_fetch(av, i, FALSE);
338 /* See note in pp_helem, and bug id #27839 */
339 SP[i+1] = svp
340 ? SvGMAGICAL(*svp) ? (mg_get(*svp), *svp) : *svp
341 : &PL_sv_undef;
342 }
343 }
344 else {
c70927a6
FC
345 PADOFFSET i;
346 for (i=0; i < (PADOFFSET)maxarg; i++) {
ce0d59fd 347 SV * const sv = AvARRAY(av)[i];
5d9574c1 348 SP[i+1] = LIKELY(sv) ? sv : &PL_sv_undef;
ce0d59fd 349 }
d5524600
DM
350 }
351 SP += maxarg;
352 PUTBACK;
353}
354
355
a7fd8ef6
DM
356/* ($lex1,@lex2,...) or my ($lex1,@lex2,...) */
357
358PP(pp_padrange)
359{
20b7effb 360 dSP;
a7fd8ef6
DM
361 PADOFFSET base = PL_op->op_targ;
362 int count = (int)(PL_op->op_private) & OPpPADRANGE_COUNTMASK;
363 int i;
d5524600
DM
364 if (PL_op->op_flags & OPf_SPECIAL) {
365 /* fake the RHS of my ($x,$y,..) = @_ */
366 PUSHMARK(SP);
367 S_pushav(aTHX_ GvAVn(PL_defgv));
368 SPAGAIN;
369 }
370
a7fd8ef6
DM
371 /* note, this is only skipped for compile-time-known void cxt */
372 if ((PL_op->op_flags & OPf_WANT) != OPf_WANT_VOID) {
373 EXTEND(SP, count);
374 PUSHMARK(SP);
375 for (i = 0; i <count; i++)
376 *++SP = PAD_SV(base+i);
377 }
378 if (PL_op->op_private & OPpLVAL_INTRO) {
4e09461c
DM
379 SV **svp = &(PAD_SVl(base));
380 const UV payload = (UV)(
381 (base << (OPpPADRANGE_COUNTSHIFT + SAVE_TIGHT_SHIFT))
382 | (count << SAVE_TIGHT_SHIFT)
383 | SAVEt_CLEARPADRANGE);
6d59e610 384 STATIC_ASSERT_STMT(OPpPADRANGE_COUNTMASK + 1 == (1 << OPpPADRANGE_COUNTSHIFT));
4e09461c 385 assert((payload >> (OPpPADRANGE_COUNTSHIFT+SAVE_TIGHT_SHIFT)) == base);
a3444cc5
DM
386 {
387 dSS_ADD;
388 SS_ADD_UV(payload);
389 SS_ADD_END(1);
390 }
4e09461c 391
a7fd8ef6 392 for (i = 0; i <count; i++)
4e09461c 393 SvPADSTALE_off(*svp++); /* mark lexical as active */
a7fd8ef6
DM
394 }
395 RETURN;
396}
397
398
a0d0e21e
LW
399PP(pp_padsv)
400{
20b7effb 401 dSP;
6c28b496
DD
402 EXTEND(SP, 1);
403 {
404 OP * const op = PL_op;
405 /* access PL_curpad once */
406 SV ** const padentry = &(PAD_SVl(op->op_targ));
407 {
408 dTARG;
409 TARG = *padentry;
410 PUSHs(TARG);
411 PUTBACK; /* no pop/push after this, TOPs ok */
8ec5e241 412 }
6c28b496
DD
413 if (op->op_flags & OPf_MOD) {
414 if (op->op_private & OPpLVAL_INTRO)
415 if (!(op->op_private & OPpPAD_STATE))
416 save_clearsv(padentry);
417 if (op->op_private & OPpDEREF) {
8f90a16d
FC
418 /* TOPs is equivalent to TARG here. Using TOPs (SP) rather
419 than TARG reduces the scope of TARG, so it does not
420 span the call to save_clearsv, resulting in smaller
421 machine code. */
6c28b496
DD
422 TOPs = vivify_ref(TOPs, op->op_private & OPpDEREF);
423 }
424 }
425 return op->op_next;
4633a7c4 426 }
a0d0e21e
LW
427}
428
429PP(pp_readline)
430{
30901a8a
FC
431 dSP;
432 if (TOPs) {
433 SvGETMAGIC(TOPs);
fc99edcf 434 tryAMAGICunTARGETlist(iter_amg, 0);
30901a8a
FC
435 PL_last_in_gv = MUTABLE_GV(*PL_stack_sp--);
436 }
437 else PL_last_in_gv = PL_argvgv, PL_stack_sp--;
6e592b3a
BM
438 if (!isGV_with_GP(PL_last_in_gv)) {
439 if (SvROK(PL_last_in_gv) && isGV_with_GP(SvRV(PL_last_in_gv)))
159b6efe 440 PL_last_in_gv = MUTABLE_GV(SvRV(PL_last_in_gv));
8efb3254 441 else {
f5284f61 442 dSP;
ad64d0ec 443 XPUSHs(MUTABLE_SV(PL_last_in_gv));
f5284f61 444 PUTBACK;
897d3989 445 Perl_pp_rv2gv(aTHX);
159b6efe 446 PL_last_in_gv = MUTABLE_GV(*PL_stack_sp--);
84ee769f
FC
447 if (PL_last_in_gv == (GV *)&PL_sv_undef)
448 PL_last_in_gv = NULL;
449 else
450 assert(isGV_with_GP(PL_last_in_gv));
f5284f61
IZ
451 }
452 }
a0d0e21e
LW
453 return do_readline();
454}
455
456PP(pp_eq)
457{
20b7effb 458 dSP;
33efebe6
DM
459 SV *left, *right;
460
a42d0242 461 tryAMAGICbin_MG(eq_amg, AMGf_set|AMGf_numeric);
33efebe6
DM
462 right = POPs;
463 left = TOPs;
464 SETs(boolSV(
465 (SvIOK_notUV(left) && SvIOK_notUV(right))
466 ? (SvIVX(left) == SvIVX(right))
467 : ( do_ncmp(left, right) == 0)
468 ));
469 RETURN;
a0d0e21e
LW
470}
471
b1c05ba5 472
4c2c3128 473/* also used for: pp_i_preinc() */
b1c05ba5 474
a0d0e21e
LW
475PP(pp_preinc)
476{
4c2c3128
DM
477 SV *sv = *PL_stack_sp;
478
479 if (LIKELY(((sv->sv_flags &
480 (SVf_THINKFIRST|SVs_GMG|SVf_IVisUV|
481 SVf_IOK|SVf_NOK|SVf_POK|SVp_NOK|SVp_POK|SVf_ROK))
482 == SVf_IOK))
483 && SvIVX(sv) != IV_MAX)
484 {
485 SvIV_set(sv, SvIVX(sv) + 1);
486 }
487 else /* Do all the PERL_PRESERVE_IVUV and hard cases in sv_inc */
488 sv_inc(sv);
489 SvSETMAGIC(sv);
490 return NORMAL;
491}
492
493
494/* also used for: pp_i_predec() */
495
496PP(pp_predec)
497{
498 SV *sv = *PL_stack_sp;
499
500 if (LIKELY(((sv->sv_flags &
501 (SVf_THINKFIRST|SVs_GMG|SVf_IVisUV|
502 SVf_IOK|SVf_NOK|SVf_POK|SVp_NOK|SVp_POK|SVf_ROK))
503 == SVf_IOK))
504 && SvIVX(sv) != IV_MIN)
55497cff 505 {
4c2c3128 506 SvIV_set(sv, SvIVX(sv) - 1);
748a9306 507 }
4c2c3128
DM
508 else /* Do all the PERL_PRESERVE_IVUV and hard cases in sv_dec */
509 sv_dec(sv);
510 SvSETMAGIC(sv);
a0d0e21e
LW
511 return NORMAL;
512}
513
b1c05ba5
DM
514
515/* also used for: pp_orassign() */
516
a0d0e21e
LW
517PP(pp_or)
518{
20b7effb 519 dSP;
f410a211 520 PERL_ASYNC_CHECK();
a0d0e21e
LW
521 if (SvTRUE(TOPs))
522 RETURN;
523 else {
c960fc3b
SP
524 if (PL_op->op_type == OP_OR)
525 --SP;
a0d0e21e
LW
526 RETURNOP(cLOGOP->op_other);
527 }
528}
529
b1c05ba5
DM
530
531/* also used for: pp_dor() pp_dorassign() */
532
25a55bd7 533PP(pp_defined)
c963b151 534{
20b7effb 535 dSP;
eb578fdb 536 SV* sv;
6136c704 537 bool defined;
25a55bd7 538 const int op_type = PL_op->op_type;
ea5195b7 539 const bool is_dor = (op_type == OP_DOR || op_type == OP_DORASSIGN);
c963b151 540
6136c704 541 if (is_dor) {
f410a211 542 PERL_ASYNC_CHECK();
25a55bd7 543 sv = TOPs;
5d9574c1 544 if (UNLIKELY(!sv || !SvANY(sv))) {
2bd49cfc
NC
545 if (op_type == OP_DOR)
546 --SP;
25a55bd7
SP
547 RETURNOP(cLOGOP->op_other);
548 }
b7c44293
RGS
549 }
550 else {
551 /* OP_DEFINED */
25a55bd7 552 sv = POPs;
5d9574c1 553 if (UNLIKELY(!sv || !SvANY(sv)))
25a55bd7 554 RETPUSHNO;
b7c44293 555 }
25a55bd7 556
6136c704 557 defined = FALSE;
c963b151
BD
558 switch (SvTYPE(sv)) {
559 case SVt_PVAV:
560 if (AvMAX(sv) >= 0 || SvGMAGICAL(sv) || (SvRMAGICAL(sv) && mg_find(sv, PERL_MAGIC_tied)))
25a55bd7 561 defined = TRUE;
c963b151
BD
562 break;
563 case SVt_PVHV:
564 if (HvARRAY(sv) || SvGMAGICAL(sv) || (SvRMAGICAL(sv) && mg_find(sv, PERL_MAGIC_tied)))
25a55bd7 565 defined = TRUE;
c963b151
BD
566 break;
567 case SVt_PVCV:
568 if (CvROOT(sv) || CvXSUB(sv))
25a55bd7 569 defined = TRUE;
c963b151
BD
570 break;
571 default:
5b295bef 572 SvGETMAGIC(sv);
c963b151 573 if (SvOK(sv))
25a55bd7 574 defined = TRUE;
6136c704 575 break;
c963b151 576 }
6136c704
AL
577
578 if (is_dor) {
c960fc3b
SP
579 if(defined)
580 RETURN;
581 if(op_type == OP_DOR)
582 --SP;
25a55bd7 583 RETURNOP(cLOGOP->op_other);
25a55bd7 584 }
d9aa96a4
SP
585 /* assuming OP_DEFINED */
586 if(defined)
587 RETPUSHYES;
588 RETPUSHNO;
c963b151
BD
589}
590
230ee21f
DM
591
592
a0d0e21e
LW
593PP(pp_add)
594{
20b7effb 595 dSP; dATARGET; bool useleft; SV *svl, *svr;
230ee21f 596
6f1401dc
DM
597 tryAMAGICbin_MG(add_amg, AMGf_assign|AMGf_numeric);
598 svr = TOPs;
599 svl = TOPm1s;
600
28e5dec8 601#ifdef PERL_PRESERVE_IVUV
230ee21f
DM
602
603 /* special-case some simple common cases */
604 if (!((svl->sv_flags|svr->sv_flags) & (SVf_IVisUV|SVs_GMG))) {
605 IV il, ir;
606 U32 flags = (svl->sv_flags & svr->sv_flags);
607 if (flags & SVf_IOK) {
608 /* both args are simple IVs */
609 UV topl, topr;
610 il = SvIVX(svl);
611 ir = SvIVX(svr);
612 do_iv:
613 topl = ((UV)il) >> (UVSIZE * 8 - 2);
614 topr = ((UV)ir) >> (UVSIZE * 8 - 2);
615
616 /* if both are in a range that can't under/overflow, do a
617 * simple integer add: if the top of both numbers
618 * are 00 or 11, then it's safe */
619 if (!( ((topl+1) | (topr+1)) & 2)) {
620 SP--;
621 TARGi(il + ir, 0); /* args not GMG, so can't be tainted */
622 SETs(TARG);
623 RETURN;
624 }
625 goto generic;
626 }
627 else if (flags & SVf_NOK) {
628 /* both args are NVs */
629 NV nl = SvNVX(svl);
630 NV nr = SvNVX(svr);
631
3336af0b
DD
632 if (
633#if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
634 !Perl_isnan(nl) && nl == (NV)(il = (IV)nl)
635 && !Perl_isnan(nr) && nr == (NV)(ir = (IV)nr)
636#else
637 nl == (NV)(il = (IV)nl) && nr == (NV)(ir = (IV)nr)
638#endif
639 )
230ee21f
DM
640 /* nothing was lost by converting to IVs */
641 goto do_iv;
642 SP--;
643 TARGn(nl + nr, 0); /* args not GMG, so can't be tainted */
644 SETs(TARG);
645 RETURN;
646 }
647 }
648
649 generic:
650
651 useleft = USE_LEFT(svl);
28e5dec8
JH
652 /* We must see if we can perform the addition with integers if possible,
653 as the integer code detects overflow while the NV code doesn't.
654 If either argument hasn't had a numeric conversion yet attempt to get
655 the IV. It's important to do this now, rather than just assuming that
656 it's not IOK as a PV of "9223372036854775806" may not take well to NV
657 addition, and an SV which is NOK, NV=6.0 ought to be coerced to
658 integer in case the second argument is IV=9223372036854775806
659 We can (now) rely on sv_2iv to do the right thing, only setting the
660 public IOK flag if the value in the NV (or PV) slot is truly integer.
661
662 A side effect is that this also aggressively prefers integer maths over
7dca457a
NC
663 fp maths for integer values.
664
a00b5bd3 665 How to detect overflow?
7dca457a
NC
666
667 C 99 section 6.2.6.1 says
668
669 The range of nonnegative values of a signed integer type is a subrange
670 of the corresponding unsigned integer type, and the representation of
671 the same value in each type is the same. A computation involving
672 unsigned operands can never overflow, because a result that cannot be
673 represented by the resulting unsigned integer type is reduced modulo
674 the number that is one greater than the largest value that can be
675 represented by the resulting type.
676
677 (the 9th paragraph)
678
679 which I read as "unsigned ints wrap."
680
681 signed integer overflow seems to be classed as "exception condition"
682
683 If an exceptional condition occurs during the evaluation of an
684 expression (that is, if the result is not mathematically defined or not
685 in the range of representable values for its type), the behavior is
686 undefined.
687
688 (6.5, the 5th paragraph)
689
690 I had assumed that on 2s complement machines signed arithmetic would
691 wrap, hence coded pp_add and pp_subtract on the assumption that
692 everything perl builds on would be happy. After much wailing and
693 gnashing of teeth it would seem that irix64 knows its ANSI spec well,
694 knows that it doesn't need to, and doesn't. Bah. Anyway, the all-
695 unsigned code below is actually shorter than the old code. :-)
696 */
697
01f91bf2 698 if (SvIV_please_nomg(svr)) {
28e5dec8
JH
699 /* Unless the left argument is integer in range we are going to have to
700 use NV maths. Hence only attempt to coerce the right argument if
701 we know the left is integer. */
eb578fdb 702 UV auv = 0;
9c5ffd7c 703 bool auvok = FALSE;
7dca457a
NC
704 bool a_valid = 0;
705
28e5dec8 706 if (!useleft) {
7dca457a
NC
707 auv = 0;
708 a_valid = auvok = 1;
709 /* left operand is undef, treat as zero. + 0 is identity,
710 Could SETi or SETu right now, but space optimise by not adding
711 lots of code to speed up what is probably a rarish case. */
712 } else {
713 /* Left operand is defined, so is it IV? */
01f91bf2 714 if (SvIV_please_nomg(svl)) {
800401ee
JH
715 if ((auvok = SvUOK(svl)))
716 auv = SvUVX(svl);
7dca457a 717 else {
eb578fdb 718 const IV aiv = SvIVX(svl);
7dca457a
NC
719 if (aiv >= 0) {
720 auv = aiv;
721 auvok = 1; /* Now acting as a sign flag. */
53e2bfb7
DM
722 } else {
723 auv = (aiv == IV_MIN) ? (UV)aiv : (UV)(-aiv);
7dca457a
NC
724 }
725 }
726 a_valid = 1;
28e5dec8
JH
727 }
728 }
7dca457a
NC
729 if (a_valid) {
730 bool result_good = 0;
731 UV result;
eb578fdb 732 UV buv;
800401ee 733 bool buvok = SvUOK(svr);
a00b5bd3 734
7dca457a 735 if (buvok)
800401ee 736 buv = SvUVX(svr);
7dca457a 737 else {
eb578fdb 738 const IV biv = SvIVX(svr);
7dca457a
NC
739 if (biv >= 0) {
740 buv = biv;
741 buvok = 1;
742 } else
53e2bfb7 743 buv = (biv == IV_MIN) ? (UV)biv : (UV)(-biv);
7dca457a
NC
744 }
745 /* ?uvok if value is >= 0. basically, flagged as UV if it's +ve,
602f51c4 746 else "IV" now, independent of how it came in.
7dca457a
NC
747 if a, b represents positive, A, B negative, a maps to -A etc
748 a + b => (a + b)
749 A + b => -(a - b)
750 a + B => (a - b)
751 A + B => -(a + b)
752 all UV maths. negate result if A negative.
753 add if signs same, subtract if signs differ. */
754
755 if (auvok ^ buvok) {
756 /* Signs differ. */
757 if (auv >= buv) {
758 result = auv - buv;
759 /* Must get smaller */
760 if (result <= auv)
761 result_good = 1;
762 } else {
763 result = buv - auv;
764 if (result <= buv) {
765 /* result really should be -(auv-buv). as its negation
766 of true value, need to swap our result flag */
767 auvok = !auvok;
768 result_good = 1;
28e5dec8
JH
769 }
770 }
7dca457a
NC
771 } else {
772 /* Signs same */
773 result = auv + buv;
774 if (result >= auv)
775 result_good = 1;
776 }
777 if (result_good) {
778 SP--;
779 if (auvok)
28e5dec8 780 SETu( result );
7dca457a
NC
781 else {
782 /* Negate result */
783 if (result <= (UV)IV_MIN)
53e2bfb7
DM
784 SETi(result == (UV)IV_MIN
785 ? IV_MIN : -(IV)result);
7dca457a
NC
786 else {
787 /* result valid, but out of range for IV. */
788 SETn( -(NV)result );
28e5dec8
JH
789 }
790 }
7dca457a
NC
791 RETURN;
792 } /* Overflow, drop through to NVs. */
28e5dec8
JH
793 }
794 }
230ee21f
DM
795
796#else
797 useleft = USE_LEFT(svl);
28e5dec8 798#endif
230ee21f 799
a0d0e21e 800 {
6f1401dc 801 NV value = SvNV_nomg(svr);
4efa5a16 802 (void)POPs;
28e5dec8
JH
803 if (!useleft) {
804 /* left operand is undef, treat as zero. + 0.0 is identity. */
805 SETn(value);
806 RETURN;
807 }
6f1401dc 808 SETn( value + SvNV_nomg(svl) );
28e5dec8 809 RETURN;
a0d0e21e
LW
810 }
811}
812
b1c05ba5
DM
813
814/* also used for: pp_aelemfast_lex() */
815
a0d0e21e
LW
816PP(pp_aelemfast)
817{
20b7effb 818 dSP;
93bad3fd 819 AV * const av = PL_op->op_type == OP_AELEMFAST_LEX
8f878375 820 ? MUTABLE_AV(PAD_SV(PL_op->op_targ)) : GvAVn(cGVOP_gv);
a3b680e6 821 const U32 lval = PL_op->op_flags & OPf_MOD;
7e169e84
DM
822 const I8 key = (I8)PL_op->op_private;
823 SV** svp;
824 SV *sv;
825
826 assert(SvTYPE(av) == SVt_PVAV);
827
f4484b87
DM
828 EXTEND(SP, 1);
829
7e169e84
DM
830 /* inlined av_fetch() for simple cases ... */
831 if (!SvRMAGICAL(av) && key >= 0 && key <= AvFILLp(av)) {
832 sv = AvARRAY(av)[key];
833 if (sv && !SvIS_FREED(sv)) {
834 PUSHs(sv);
835 RETURN;
836 }
837 }
838
839 /* ... else do it the hard way */
840 svp = av_fetch(av, key, lval);
841 sv = (svp ? *svp : &PL_sv_undef);
b024352e
DM
842
843 if (UNLIKELY(!svp && lval))
7e169e84 844 DIE(aTHX_ PL_no_aelem, (int)key);
b024352e 845
39cf747a 846 if (!lval && SvRMAGICAL(av) && SvGMAGICAL(sv)) /* see note in pp_helem() */
fd69380d 847 mg_get(sv);
be6c24e0 848 PUSHs(sv);
a0d0e21e
LW
849 RETURN;
850}
851
852PP(pp_join)
853{
20b7effb 854 dSP; dMARK; dTARGET;
a0d0e21e
LW
855 MARK++;
856 do_join(TARG, *MARK, MARK, SP);
857 SP = MARK;
858 SETs(TARG);
859 RETURN;
860}
861
862PP(pp_pushre)
863{
20b7effb 864 dSP;
44a8e56a
PP
865#ifdef DEBUGGING
866 /*
867 * We ass_u_me that LvTARGOFF() comes first, and that two STRLENs
868 * will be enough to hold an OP*.
869 */
c4420975 870 SV* const sv = sv_newmortal();
44a8e56a
PP
871 sv_upgrade(sv, SVt_PVLV);
872 LvTYPE(sv) = '/';
533c011a 873 Copy(&PL_op, &LvTARGOFF(sv), 1, OP*);
44a8e56a
PP
874 XPUSHs(sv);
875#else
ad64d0ec 876 XPUSHs(MUTABLE_SV(PL_op));
44a8e56a 877#endif
a0d0e21e
LW
878 RETURN;
879}
880
881/* Oversized hot code. */
882
b1c05ba5
DM
883/* also used for: pp_say() */
884
a0d0e21e
LW
885PP(pp_print)
886{
20b7effb 887 dSP; dMARK; dORIGMARK;
eb578fdb 888 PerlIO *fp;
236988e4 889 MAGIC *mg;
159b6efe
NC
890 GV * const gv
891 = (PL_op->op_flags & OPf_STACKED) ? MUTABLE_GV(*++MARK) : PL_defoutgv;
9c9f25b8 892 IO *io = GvIO(gv);
5b468f54 893
9c9f25b8 894 if (io
ad64d0ec 895 && (mg = SvTIED_mg((const SV *)io, PERL_MAGIC_tiedscalar)))
5b468f54 896 {
01bb7c6d 897 had_magic:
68dc0745 898 if (MARK == ORIGMARK) {
1c846c1f 899 /* If using default handle then we need to make space to
a60c0954
NIS
900 * pass object as 1st arg, so move other args up ...
901 */
4352c267 902 MEXTEND(SP, 1);
68dc0745
PP
903 ++MARK;
904 Move(MARK, MARK + 1, (SP - MARK) + 1, SV*);
905 ++SP;
906 }
3e0cb5de 907 return Perl_tied_method(aTHX_ SV_CONST(PRINT), mark - 1, MUTABLE_SV(io),
94bc412f
NC
908 mg,
909 (G_SCALAR | TIED_METHOD_ARGUMENTS_ON_STACK
910 | (PL_op->op_type == OP_SAY
911 ? TIED_METHOD_SAY : 0)), sp - mark);
236988e4 912 }
9c9f25b8 913 if (!io) {
68b590d9 914 if ( gv && GvEGVx(gv) && (io = GvIO(GvEGV(gv)))
ad64d0ec 915 && (mg = SvTIED_mg((const SV *)io, PERL_MAGIC_tiedscalar)))
01bb7c6d 916 goto had_magic;
51087808 917 report_evil_fh(gv);
93189314 918 SETERRNO(EBADF,RMS_IFI);
a0d0e21e
LW
919 goto just_say_no;
920 }
921 else if (!(fp = IoOFP(io))) {
7716c5c5
NC
922 if (IoIFP(io))
923 report_wrongway_fh(gv, '<');
51087808 924 else
7716c5c5 925 report_evil_fh(gv);
93189314 926 SETERRNO(EBADF,IoIFP(io)?RMS_FAC:RMS_IFI);
a0d0e21e
LW
927 goto just_say_no;
928 }
929 else {
e23d9e2f 930 SV * const ofs = GvSV(PL_ofsgv); /* $, */
a0d0e21e 931 MARK++;
e23d9e2f 932 if (ofs && (SvGMAGICAL(ofs) || SvOK(ofs))) {
a0d0e21e
LW
933 while (MARK <= SP) {
934 if (!do_print(*MARK, fp))
935 break;
936 MARK++;
937 if (MARK <= SP) {
e23d9e2f
CS
938 /* don't use 'ofs' here - it may be invalidated by magic callbacks */
939 if (!do_print(GvSV(PL_ofsgv), fp)) {
a0d0e21e
LW
940 MARK--;
941 break;
942 }
943 }
944 }
945 }
946 else {
947 while (MARK <= SP) {
948 if (!do_print(*MARK, fp))
949 break;
950 MARK++;
951 }
952 }
953 if (MARK <= SP)
954 goto just_say_no;
955 else {
cfc4a7da
GA
956 if (PL_op->op_type == OP_SAY) {
957 if (PerlIO_write(fp, "\n", 1) == 0 || PerlIO_error(fp))
958 goto just_say_no;
959 }
960 else if (PL_ors_sv && SvOK(PL_ors_sv))
7889fe52 961 if (!do_print(PL_ors_sv, fp)) /* $\ */
a0d0e21e
LW
962 goto just_say_no;
963
964 if (IoFLAGS(io) & IOf_FLUSH)
760ac839 965 if (PerlIO_flush(fp) == EOF)
a0d0e21e
LW
966 goto just_say_no;
967 }
968 }
969 SP = ORIGMARK;
e52fd6f4 970 XPUSHs(&PL_sv_yes);
a0d0e21e
LW
971 RETURN;
972
973 just_say_no:
974 SP = ORIGMARK;
e52fd6f4 975 XPUSHs(&PL_sv_undef);
a0d0e21e
LW
976 RETURN;
977}
978
b1c05ba5
DM
979
980/* also used for: pp_rv2hv() */
bdaf10a5 981/* also called directly by pp_lvavref */
b1c05ba5 982
a0d0e21e
LW
983PP(pp_rv2av)
984{
20b7effb 985 dSP; dTOPss;
1c23e2bd 986 const U8 gimme = GIMME_V;
13c59d41
MH
987 static const char an_array[] = "an ARRAY";
988 static const char a_hash[] = "a HASH";
bdaf10a5
FC
989 const bool is_pp_rv2av = PL_op->op_type == OP_RV2AV
990 || PL_op->op_type == OP_LVAVREF;
d83b45b8 991 const svtype type = is_pp_rv2av ? SVt_PVAV : SVt_PVHV;
a0d0e21e 992
9026059d 993 SvGETMAGIC(sv);
a0d0e21e 994 if (SvROK(sv)) {
5d9574c1 995 if (UNLIKELY(SvAMAGIC(sv))) {
93d7320b 996 sv = amagic_deref_call(sv, is_pp_rv2av ? to_av_amg : to_hv_amg);
93d7320b 997 }
17ab7946 998 sv = SvRV(sv);
5d9574c1 999 if (UNLIKELY(SvTYPE(sv) != type))
dcbac5bb 1000 /* diag_listed_as: Not an ARRAY reference */
13c59d41 1001 DIE(aTHX_ "Not %s reference", is_pp_rv2av ? an_array : a_hash);
5d9574c1
DM
1002 else if (UNLIKELY(PL_op->op_flags & OPf_MOD
1003 && PL_op->op_private & OPpLVAL_INTRO))
3da99855 1004 Perl_croak(aTHX_ "%s", PL_no_localize_ref);
a0d0e21e 1005 }
5d9574c1 1006 else if (UNLIKELY(SvTYPE(sv) != type)) {
67955e0c 1007 GV *gv;
1c846c1f 1008
6e592b3a 1009 if (!isGV_with_GP(sv)) {
13c59d41 1010 gv = Perl_softref2xv(aTHX_ sv, is_pp_rv2av ? an_array : a_hash,
dc3c76f8
NC
1011 type, &sp);
1012 if (!gv)
1013 RETURN;
35cd451c
GS
1014 }
1015 else {
159b6efe 1016 gv = MUTABLE_GV(sv);
a0d0e21e 1017 }
ad64d0ec 1018 sv = is_pp_rv2av ? MUTABLE_SV(GvAVn(gv)) : MUTABLE_SV(GvHVn(gv));
533c011a 1019 if (PL_op->op_private & OPpLVAL_INTRO)
ad64d0ec 1020 sv = is_pp_rv2av ? MUTABLE_SV(save_ary(gv)) : MUTABLE_SV(save_hash(gv));
9f527363
FC
1021 }
1022 if (PL_op->op_flags & OPf_REF) {
17ab7946 1023 SETs(sv);
a0d0e21e 1024 RETURN;
9f527363 1025 }
5d9574c1 1026 else if (UNLIKELY(PL_op->op_private & OPpMAYBE_LVSUB)) {
40c94d11
FC
1027 const I32 flags = is_lvalue_sub();
1028 if (flags && !(flags & OPpENTERSUB_INARGS)) {
cde874ca 1029 if (gimme != G_ARRAY)
042560a6 1030 goto croak_cant_return;
17ab7946 1031 SETs(sv);
78f9721b 1032 RETURN;
40c94d11 1033 }
a0d0e21e
LW
1034 }
1035
17ab7946 1036 if (is_pp_rv2av) {
502c6561 1037 AV *const av = MUTABLE_AV(sv);
636fe681 1038 /* The guts of pp_rv2av */
96913b52 1039 if (gimme == G_ARRAY) {
d5524600
DM
1040 SP--;
1041 PUTBACK;
1042 S_pushav(aTHX_ av);
1043 SPAGAIN;
1c846c1f 1044 }
96913b52
VP
1045 else if (gimme == G_SCALAR) {
1046 dTARGET;
c70927a6 1047 const SSize_t maxarg = AvFILL(av) + 1;
96913b52 1048 SETi(maxarg);
93965878 1049 }
17ab7946
NC
1050 } else {
1051 /* The guts of pp_rv2hv */
96913b52
VP
1052 if (gimme == G_ARRAY) { /* array wanted */
1053 *PL_stack_sp = sv;
981b7185 1054 return Perl_do_kv(aTHX);
96913b52 1055 }
c8fe3bdf 1056 else if ((PL_op->op_private & OPpTRUEBOOL
adc42c31 1057 || ( PL_op->op_private & OPpMAYBE_TRUEBOOL
c8fe3bdf
FC
1058 && block_gimme() == G_VOID ))
1059 && (!SvRMAGICAL(sv) || !mg_find(sv, PERL_MAGIC_tied)))
1060 SETs(HvUSEDKEYS(sv) ? &PL_sv_yes : sv_2mortal(newSViv(0)));
96913b52 1061 else if (gimme == G_SCALAR) {
1a8bdda9 1062 dTARG;
96913b52 1063 TARG = Perl_hv_scalar(aTHX_ MUTABLE_HV(sv));
96913b52
VP
1064 SETTARG;
1065 }
17ab7946 1066 }
be85d344 1067 RETURN;
042560a6
NC
1068
1069 croak_cant_return:
1070 Perl_croak(aTHX_ "Can't return %s to lvalue scalar context",
1071 is_pp_rv2av ? "array" : "hash");
77e217c6 1072 RETURN;
a0d0e21e
LW
1073}
1074
10c8fecd 1075STATIC void
fb8f4cf8 1076S_do_oddball(pTHX_ SV **oddkey, SV **firstkey)
10c8fecd 1077{
7918f24d
NC
1078 PERL_ARGS_ASSERT_DO_ODDBALL;
1079
fb8f4cf8 1080 if (*oddkey) {
6d822dc4 1081 if (ckWARN(WARN_MISC)) {
a3b680e6 1082 const char *err;
fb8f4cf8
RZ
1083 if (oddkey == firstkey &&
1084 SvROK(*oddkey) &&
1085 (SvTYPE(SvRV(*oddkey)) == SVt_PVAV ||
1086 SvTYPE(SvRV(*oddkey)) == SVt_PVHV))
10c8fecd 1087 {
a3b680e6 1088 err = "Reference found where even-sized list expected";
10c8fecd
GS
1089 }
1090 else
a3b680e6 1091 err = "Odd number of elements in hash assignment";
f1f66076 1092 Perl_warner(aTHX_ packWARN(WARN_MISC), "%s", err);
10c8fecd 1093 }
6d822dc4 1094
10c8fecd
GS
1095 }
1096}
1097
a5f48505
DM
1098
1099/* Do a mark and sweep with the SVf_BREAK flag to detect elements which
1100 * are common to both the LHS and RHS of an aassign, and replace them
1101 * with copies. All these copies are made before the actual list assign is
1102 * done.
1103 *
1104 * For example in ($a,$b) = ($b,$a), assigning the value of the first RHS
1105 * element ($b) to the first LH element ($a), modifies $a; when the
1106 * second assignment is done, the second RH element now has the wrong
1107 * value. So we initially replace the RHS with ($b, mortalcopy($a)).
1108 * Note that we don't need to make a mortal copy of $b.
1109 *
1110 * The algorithm below works by, for every RHS element, mark the
1111 * corresponding LHS target element with SVf_BREAK. Then if the RHS
1112 * element is found with SVf_BREAK set, it means it would have been
1113 * modified, so make a copy.
1114 * Note that by scanning both LHS and RHS in lockstep, we avoid
1115 * unnecessary copies (like $b above) compared with a naive
1116 * "mark all LHS; copy all marked RHS; unmark all LHS".
1117 *
1118 * If the LHS element is a 'my' declaration' and has a refcount of 1, then
1119 * it can't be common and can be skipped.
ebc643ce
DM
1120 *
1121 * On DEBUGGING builds it takes an extra boolean, fake. If true, it means
1122 * that we thought we didn't need to call S_aassign_copy_common(), but we
1123 * have anyway for sanity checking. If we find we need to copy, then panic.
a5f48505
DM
1124 */
1125
1126PERL_STATIC_INLINE void
1127S_aassign_copy_common(pTHX_ SV **firstlelem, SV **lastlelem,
ebc643ce
DM
1128 SV **firstrelem, SV **lastrelem
1129#ifdef DEBUGGING
1130 , bool fake
1131#endif
1132)
a5f48505
DM
1133{
1134 dVAR;
1135 SV **relem;
1136 SV **lelem;
1137 SSize_t lcount = lastlelem - firstlelem + 1;
1138 bool marked = FALSE; /* have we marked any LHS with SVf_BREAK ? */
1139 bool const do_rc1 = cBOOL(PL_op->op_private & OPpASSIGN_COMMON_RC1);
beb08a1e 1140 bool copy_all = FALSE;
a5f48505
DM
1141
1142 assert(!PL_in_clean_all); /* SVf_BREAK not already in use */
1143 assert(firstlelem < lastlelem); /* at least 2 LH elements */
1144 assert(firstrelem < lastrelem); /* at least 2 RH elements */
1145
ebc643ce
DM
1146
1147 lelem = firstlelem;
a5f48505
DM
1148 /* we never have to copy the first RH element; it can't be corrupted
1149 * by assigning something to the corresponding first LH element.
1150 * So this scan does in a loop: mark LHS[N]; test RHS[N+1]
1151 */
ebc643ce 1152 relem = firstrelem + 1;
a5f48505
DM
1153
1154 for (; relem <= lastrelem; relem++) {
1155 SV *svr;
1156
1157 /* mark next LH element */
1158
1159 if (--lcount >= 0) {
1160 SV *svl = *lelem++;
1161
1162 if (UNLIKELY(!svl)) {/* skip AV alias marker */
1163 assert (lelem <= lastlelem);
1164 svl = *lelem++;
1165 lcount--;
1166 }
1167
1168 assert(svl);
beb08a1e
TC
1169 if (SvSMAGICAL(svl)) {
1170 copy_all = TRUE;
1171 }
a5f48505
DM
1172 if (SvTYPE(svl) == SVt_PVAV || SvTYPE(svl) == SVt_PVHV) {
1173 if (!marked)
1174 return;
1175 /* this LH element will consume all further args;
1176 * no need to mark any further LH elements (if any).
1177 * But we still need to scan any remaining RHS elements;
1178 * set lcount negative to distinguish from lcount == 0,
1179 * so the loop condition continues being true
1180 */
1181 lcount = -1;
1182 lelem--; /* no need to unmark this element */
1183 }
1184 else if (!(do_rc1 && SvREFCNT(svl) == 1) && svl != &PL_sv_undef) {
1185 assert(!SvIMMORTAL(svl));
1186 SvFLAGS(svl) |= SVf_BREAK;
1187 marked = TRUE;
1188 }
1189 else if (!marked) {
1190 /* don't check RH element if no SVf_BREAK flags set yet */
1191 if (!lcount)
1192 break;
1193 continue;
1194 }
1195 }
1196
1197 /* see if corresponding RH element needs copying */
1198
1199 assert(marked);
1200 svr = *relem;
1201 assert(svr);
1202
5c1db569 1203 if (UNLIKELY(SvFLAGS(svr) & (SVf_BREAK|SVs_GMG) || copy_all)) {
a5f48505 1204
ebc643ce
DM
1205#ifdef DEBUGGING
1206 if (fake) {
9ae0115f 1207 /* op_dump(PL_op); */
ebc643ce
DM
1208 Perl_croak(aTHX_
1209 "panic: aassign skipped needed copy of common RH elem %"
1210 UVuf, (UV)(relem - firstrelem));
1211 }
1212#endif
1213
a5f48505
DM
1214 TAINT_NOT; /* Each item is independent */
1215
1216 /* Dear TODO test in t/op/sort.t, I love you.
1217 (It's relying on a panic, not a "semi-panic" from newSVsv()
1218 and then an assertion failure below.) */
1219 if (UNLIKELY(SvIS_FREED(svr))) {
1220 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p",
1221 (void*)svr);
1222 }
1223 /* avoid break flag while copying; otherwise COW etc
1224 * disabled... */
1225 SvFLAGS(svr) &= ~SVf_BREAK;
1226 /* Not newSVsv(), as it does not allow copy-on-write,
8c1e192f
DM
1227 resulting in wasteful copies.
1228 Also, we use SV_NOSTEAL in case the SV is used more than
1229 once, e.g. (...) = (f())[0,0]
1230 Where the same SV appears twice on the RHS without a ref
1231 count bump. (Although I suspect that the SV won't be
1232 stealable here anyway - DAPM).
1233 */
a5f48505
DM
1234 *relem = sv_mortalcopy_flags(svr,
1235 SV_GMAGIC|SV_DO_COW_SVSETSV|SV_NOSTEAL);
1236 /* ... but restore afterwards in case it's needed again,
1237 * e.g. ($a,$b,$c) = (1,$a,$a)
1238 */
1239 SvFLAGS(svr) |= SVf_BREAK;
1240 }
1241
1242 if (!lcount)
1243 break;
1244 }
1245
1246 if (!marked)
1247 return;
1248
1249 /*unmark LHS */
1250
1251 while (lelem > firstlelem) {
1252 SV * const svl = *(--lelem);
1253 if (svl)
1254 SvFLAGS(svl) &= ~SVf_BREAK;
1255 }
1256}
1257
1258
1259
a0d0e21e
LW
1260PP(pp_aassign)
1261{
27da23d5 1262 dVAR; dSP;
3280af22
NIS
1263 SV **lastlelem = PL_stack_sp;
1264 SV **lastrelem = PL_stack_base + POPMARK;
1265 SV **firstrelem = PL_stack_base + POPMARK + 1;
a0d0e21e
LW
1266 SV **firstlelem = lastrelem + 1;
1267
eb578fdb
KW
1268 SV **relem;
1269 SV **lelem;
a0d0e21e 1270
eb578fdb
KW
1271 SV *sv;
1272 AV *ary;
a0d0e21e 1273
1c23e2bd 1274 U8 gimme;
a0d0e21e 1275 HV *hash;
c70927a6 1276 SSize_t i;
a0d0e21e 1277 int magic;
a5f48505 1278 U32 lval;
a68090fe
DM
1279 /* PL_delaymagic is restored by JUMPENV_POP on dieing, so we
1280 * only need to save locally, not on the save stack */
1281 U16 old_delaymagic = PL_delaymagic;
ebc643ce
DM
1282#ifdef DEBUGGING
1283 bool fake = 0;
1284#endif
5637b936 1285
3280af22 1286 PL_delaymagic = DM_DELAY; /* catch simultaneous items */
a0d0e21e
LW
1287
1288 /* If there's a common identifier on both sides we have to take
1289 * special care that assigning the identifier on the left doesn't
1290 * clobber a value on the right that's used later in the list.
1291 */
acdea6f0 1292
beb08a1e
TC
1293 /* at least 2 LH and RH elements, or commonality isn't an issue */
1294 if (firstlelem < lastlelem && firstrelem < lastrelem) {
5c1db569
TC
1295 for (relem = firstrelem+1; relem <= lastrelem; relem++) {
1296 if (SvGMAGICAL(*relem))
1297 goto do_scan;
1298 }
beb08a1e
TC
1299 for (lelem = firstlelem; lelem <= lastlelem; lelem++) {
1300 if (*lelem && SvSMAGICAL(*lelem))
1301 goto do_scan;
a5f48505 1302 }
beb08a1e
TC
1303 if ( PL_op->op_private & (OPpASSIGN_COMMON_SCALAR|OPpASSIGN_COMMON_RC1) ) {
1304 if (PL_op->op_private & OPpASSIGN_COMMON_RC1) {
1305 /* skip the scan if all scalars have a ref count of 1 */
1306 for (lelem = firstlelem; lelem <= lastlelem; lelem++) {
1307 sv = *lelem;
1308 if (!sv || SvREFCNT(sv) == 1)
1309 continue;
1310 if (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVAV)
1311 goto do_scan;
1312 break;
1313 }
1314 }
1315 else {
1316 do_scan:
1317 S_aassign_copy_common(aTHX_
1318 firstlelem, lastlelem, firstrelem, lastrelem
ebc643ce 1319#ifdef DEBUGGING
beb08a1e 1320 , fake
ebc643ce 1321#endif
beb08a1e
TC
1322 );
1323 }
a5f48505 1324 }
a0d0e21e 1325 }
ebc643ce
DM
1326#ifdef DEBUGGING
1327 else {
1328 /* on debugging builds, do the scan even if we've concluded we
1329 * don't need to, then panic if we find commonality. Note that the
1330 * scanner assumes at least 2 elements */
1331 if (firstlelem < lastlelem && firstrelem < lastrelem) {
1332 fake = 1;
1333 goto do_scan;
1334 }
1335 }
1336#endif
a0d0e21e 1337
a5f48505
DM
1338 gimme = GIMME_V;
1339 lval = (gimme == G_ARRAY) ? (PL_op->op_flags & OPf_MOD || LVRET) : 0;
1340
a0d0e21e
LW
1341 relem = firstrelem;
1342 lelem = firstlelem;
4608196e
RGS
1343 ary = NULL;
1344 hash = NULL;
10c8fecd 1345
5d9574c1 1346 while (LIKELY(lelem <= lastlelem)) {
bdaf10a5 1347 bool alias = FALSE;
bbce6d69 1348 TAINT_NOT; /* Each item stands on its own, taintwise. */
a0d0e21e 1349 sv = *lelem++;
bdaf10a5
FC
1350 if (UNLIKELY(!sv)) {
1351 alias = TRUE;
1352 sv = *lelem++;
1353 ASSUME(SvTYPE(sv) == SVt_PVAV);
1354 }
a0d0e21e 1355 switch (SvTYPE(sv)) {
a5f48505
DM
1356 case SVt_PVAV: {
1357 bool already_copied = FALSE;
60edcf09 1358 ary = MUTABLE_AV(sv);
748a9306 1359 magic = SvMAGICAL(ary) != 0;
60edcf09
FC
1360 ENTER;
1361 SAVEFREESV(SvREFCNT_inc_simple_NN(sv));
a5f48505
DM
1362
1363 /* We need to clear ary. The is a danger that if we do this,
1364 * elements on the RHS may be prematurely freed, e.g.
1365 * @a = ($a[0]);
1366 * In the case of possible commonality, make a copy of each
1367 * RHS SV *before* clearing the array, and add a reference
1368 * from the tmps stack, so that it doesn't leak on death.
1369 * Otherwise, make a copy of each RHS SV only as we're storing
1370 * it into the array - that way we don't have to worry about
1371 * it being leaked if we die, but don't incur the cost of
1372 * mortalising everything.
1373 */
1374
1375 if ( (PL_op->op_private & OPpASSIGN_COMMON_AGG)
1376 && (relem <= lastrelem)
1377 && (magic || AvFILL(ary) != -1))
1378 {
1379 SV **svp;
1380 EXTEND_MORTAL(lastrelem - relem + 1);
1381 for (svp = relem; svp <= lastrelem; svp++) {
8c1e192f 1382 /* see comment in S_aassign_copy_common about SV_NOSTEAL */
a5f48505
DM
1383 *svp = sv_mortalcopy_flags(*svp,
1384 SV_GMAGIC|SV_DO_COW_SVSETSV|SV_NOSTEAL);
1385 TAINT_NOT;
1386 }
1387 already_copied = TRUE;
1388 }
1389
1390 av_clear(ary);
6768377c
DM
1391 if (relem <= lastrelem)
1392 av_extend(ary, lastrelem - relem);
1393
a0d0e21e
LW
1394 i = 0;
1395 while (relem <= lastrelem) { /* gobble up all the rest */
5117ca91 1396 SV **didstore;
bdaf10a5 1397 if (LIKELY(!alias)) {
a5f48505
DM
1398 if (already_copied)
1399 sv = *relem;
1400 else {
1401 if (LIKELY(*relem))
1402 /* before newSV, in case it dies */
1403 SvGETMAGIC(*relem);
1404 sv = newSV(0);
8c1e192f
DM
1405 /* see comment in S_aassign_copy_common about
1406 * SV_NOSTEAL */
1407 sv_setsv_flags(sv, *relem,
1408 (SV_DO_COW_SVSETSV|SV_NOSTEAL));
a5f48505
DM
1409 *relem = sv;
1410 }
bdaf10a5
FC
1411 }
1412 else {
a5f48505
DM
1413 if (!already_copied)
1414 SvGETMAGIC(*relem);
bdaf10a5
FC
1415 if (!SvROK(*relem))
1416 DIE(aTHX_ "Assigned value is not a reference");
1417 if (SvTYPE(SvRV(*relem)) > SVt_PVLV)
1418 /* diag_listed_as: Assigned value is not %s reference */
1419 DIE(aTHX_
1420 "Assigned value is not a SCALAR reference");
a5f48505 1421 if (lval && !already_copied)
bdaf10a5
FC
1422 *relem = sv_mortalcopy(*relem);
1423 /* XXX else check for weak refs? */
0ad694a7 1424 sv = SvREFCNT_inc_NN(SvRV(*relem));
bdaf10a5
FC
1425 }
1426 relem++;
a5f48505 1427 if (already_copied)
0ad694a7 1428 SvREFCNT_inc_simple_void_NN(sv); /* undo mortal free */
5117ca91
GS
1429 didstore = av_store(ary,i++,sv);
1430 if (magic) {
18024492
FC
1431 if (!didstore)
1432 sv_2mortal(sv);
8ef24240 1433 if (SvSMAGICAL(sv))
fb73857a 1434 mg_set(sv);
5117ca91 1435 }
bbce6d69 1436 TAINT_NOT;
a0d0e21e 1437 }
5d9574c1 1438 if (UNLIKELY(PL_delaymagic & DM_ARRAY_ISA))
ad64d0ec 1439 SvSETMAGIC(MUTABLE_SV(ary));
60edcf09 1440 LEAVE;
a0d0e21e 1441 break;
a5f48505
DM
1442 }
1443
10c8fecd 1444 case SVt_PVHV: { /* normal hash */
a0d0e21e 1445 SV *tmpstr;
1c4ea384
RZ
1446 int odd;
1447 int duplicates = 0;
45960564 1448 SV** topelem = relem;
1c4ea384 1449 SV **firsthashrelem = relem;
a5f48505 1450 bool already_copied = FALSE;
a0d0e21e 1451
60edcf09 1452 hash = MUTABLE_HV(sv);
748a9306 1453 magic = SvMAGICAL(hash) != 0;
1c4ea384
RZ
1454
1455 odd = ((lastrelem - firsthashrelem)&1)? 0 : 1;
5d9574c1 1456 if (UNLIKELY(odd)) {
fb8f4cf8 1457 do_oddball(lastrelem, firsthashrelem);
1d2b3927
HS
1458 /* we have firstlelem to reuse, it's not needed anymore
1459 */
1c4ea384
RZ
1460 *(lastrelem+1) = &PL_sv_undef;
1461 }
1462
60edcf09
FC
1463 ENTER;
1464 SAVEFREESV(SvREFCNT_inc_simple_NN(sv));
a5f48505
DM
1465
1466 /* We need to clear hash. The is a danger that if we do this,
1467 * elements on the RHS may be prematurely freed, e.g.
1468 * %h = (foo => $h{bar});
1469 * In the case of possible commonality, make a copy of each
1470 * RHS SV *before* clearing the hash, and add a reference
1471 * from the tmps stack, so that it doesn't leak on death.
1472 */
1473
1474 if ( (PL_op->op_private & OPpASSIGN_COMMON_AGG)
1475 && (relem <= lastrelem)
1476 && (magic || HvUSEDKEYS(hash)))
1477 {
1478 SV **svp;
1479 EXTEND_MORTAL(lastrelem - relem + 1);
1480 for (svp = relem; svp <= lastrelem; svp++) {
1481 *svp = sv_mortalcopy_flags(*svp,
1482 SV_GMAGIC|SV_DO_COW_SVSETSV|SV_NOSTEAL);
1483 TAINT_NOT;
1484 }
1485 already_copied = TRUE;
1486 }
1487
a0d0e21e 1488 hv_clear(hash);
a5f48505 1489
5d9574c1 1490 while (LIKELY(relem < lastrelem+odd)) { /* gobble up all the rest */
5117ca91 1491 HE *didstore;
1c4ea384 1492 assert(*relem);
632b9d6f
FC
1493 /* Copy the key if aassign is called in lvalue context,
1494 to avoid having the next op modify our rhs. Copy
1495 it also if it is gmagical, lest it make the
1496 hv_store_ent call below croak, leaking the value. */
a5f48505 1497 sv = (lval || SvGMAGICAL(*relem)) && !already_copied
632b9d6f
FC
1498 ? sv_mortalcopy(*relem)
1499 : *relem;
45960564 1500 relem++;
1c4ea384 1501 assert(*relem);
a5f48505
DM
1502 if (already_copied)
1503 tmpstr = *relem++;
1504 else {
1505 SvGETMAGIC(*relem);
1506 tmpstr = newSV(0);
1507 sv_setsv_nomg(tmpstr,*relem++); /* value */
1508 }
1509
a88bf2bc 1510 if (gimme == G_ARRAY) {
45960564
DM
1511 if (hv_exists_ent(hash, sv, 0))
1512 /* key overwrites an existing entry */
1513 duplicates += 2;
a88bf2bc 1514 else {
45960564 1515 /* copy element back: possibly to an earlier
1d2b3927
HS
1516 * stack location if we encountered dups earlier,
1517 * possibly to a later stack location if odd */
45960564
DM
1518 *topelem++ = sv;
1519 *topelem++ = tmpstr;
1520 }
1521 }
a5f48505 1522 if (already_copied)
0ad694a7 1523 SvREFCNT_inc_simple_void_NN(tmpstr); /* undo mortal free */
5117ca91 1524 didstore = hv_store_ent(hash,sv,tmpstr,0);
632b9d6f
FC
1525 if (magic) {
1526 if (!didstore) sv_2mortal(tmpstr);
1527 SvSETMAGIC(tmpstr);
1528 }
bbce6d69 1529 TAINT_NOT;
8e07c86e 1530 }
60edcf09 1531 LEAVE;
1c4ea384
RZ
1532 if (duplicates && gimme == G_ARRAY) {
1533 /* at this point we have removed the duplicate key/value
1534 * pairs from the stack, but the remaining values may be
1535 * wrong; i.e. with (a 1 a 2 b 3) on the stack we've removed
1536 * the (a 2), but the stack now probably contains
1537 * (a <freed> b 3), because { hv_save(a,1); hv_save(a,2) }
1538 * obliterates the earlier key. So refresh all values. */
1539 lastrelem -= duplicates;
1540 relem = firsthashrelem;
1541 while (relem < lastrelem+odd) {
1542 HE *he;
1543 he = hv_fetch_ent(hash, *relem++, 0, 0);
1544 *relem++ = (he ? HeVAL(he) : &PL_sv_undef);
1545 }
1546 }
1547 if (odd && gimme == G_ARRAY) lastrelem++;
a0d0e21e
LW
1548 }
1549 break;
1550 default:
6fc92669
GS
1551 if (SvIMMORTAL(sv)) {
1552 if (relem <= lastrelem)
1553 relem++;
1554 break;
a0d0e21e
LW
1555 }
1556 if (relem <= lastrelem) {
5d9574c1 1557 if (UNLIKELY(
1c70fb82
FC
1558 SvTEMP(sv) && !SvSMAGICAL(sv) && SvREFCNT(sv) == 1 &&
1559 (!isGV_with_GP(sv) || SvFAKE(sv)) && ckWARN(WARN_MISC)
5d9574c1 1560 ))
1c70fb82
FC
1561 Perl_warner(aTHX_
1562 packWARN(WARN_MISC),
1563 "Useless assignment to a temporary"
1564 );
a0d0e21e
LW
1565 sv_setsv(sv, *relem);
1566 *(relem++) = sv;
1567 }
1568 else
3280af22 1569 sv_setsv(sv, &PL_sv_undef);
8ef24240 1570 SvSETMAGIC(sv);
a0d0e21e
LW
1571 break;
1572 }
1573 }
5d9574c1 1574 if (UNLIKELY(PL_delaymagic & ~DM_DELAY)) {
985213f2 1575 /* Will be used to set PL_tainting below */
dfff4baf
BF
1576 Uid_t tmp_uid = PerlProc_getuid();
1577 Uid_t tmp_euid = PerlProc_geteuid();
1578 Gid_t tmp_gid = PerlProc_getgid();
1579 Gid_t tmp_egid = PerlProc_getegid();
985213f2 1580
b469f1e0 1581 /* XXX $> et al currently silently ignore failures */
3280af22 1582 if (PL_delaymagic & DM_UID) {
a0d0e21e 1583#ifdef HAS_SETRESUID
b469f1e0
JH
1584 PERL_UNUSED_RESULT(
1585 setresuid((PL_delaymagic & DM_RUID) ? PL_delaymagic_uid : (Uid_t)-1,
1586 (PL_delaymagic & DM_EUID) ? PL_delaymagic_euid : (Uid_t)-1,
1587 (Uid_t)-1));
56febc5e
AD
1588#else
1589# ifdef HAS_SETREUID
b469f1e0
JH
1590 PERL_UNUSED_RESULT(
1591 setreuid((PL_delaymagic & DM_RUID) ? PL_delaymagic_uid : (Uid_t)-1,
1592 (PL_delaymagic & DM_EUID) ? PL_delaymagic_euid : (Uid_t)-1));
56febc5e
AD
1593# else
1594# ifdef HAS_SETRUID
b28d0864 1595 if ((PL_delaymagic & DM_UID) == DM_RUID) {
b469f1e0 1596 PERL_UNUSED_RESULT(setruid(PL_delaymagic_uid));
b28d0864 1597 PL_delaymagic &= ~DM_RUID;
a0d0e21e 1598 }
56febc5e
AD
1599# endif /* HAS_SETRUID */
1600# ifdef HAS_SETEUID
b28d0864 1601 if ((PL_delaymagic & DM_UID) == DM_EUID) {
b469f1e0 1602 PERL_UNUSED_RESULT(seteuid(PL_delaymagic_euid));
b28d0864 1603 PL_delaymagic &= ~DM_EUID;
a0d0e21e 1604 }
56febc5e 1605# endif /* HAS_SETEUID */
b28d0864 1606 if (PL_delaymagic & DM_UID) {
985213f2 1607 if (PL_delaymagic_uid != PL_delaymagic_euid)
cea2e8a9 1608 DIE(aTHX_ "No setreuid available");
b469f1e0 1609 PERL_UNUSED_RESULT(PerlProc_setuid(PL_delaymagic_uid));
a0d0e21e 1610 }
56febc5e
AD
1611# endif /* HAS_SETREUID */
1612#endif /* HAS_SETRESUID */
04783dc7 1613
985213f2
AB
1614 tmp_uid = PerlProc_getuid();
1615 tmp_euid = PerlProc_geteuid();
a0d0e21e 1616 }
b469f1e0 1617 /* XXX $> et al currently silently ignore failures */
3280af22 1618 if (PL_delaymagic & DM_GID) {
a0d0e21e 1619#ifdef HAS_SETRESGID
b469f1e0
JH
1620 PERL_UNUSED_RESULT(
1621 setresgid((PL_delaymagic & DM_RGID) ? PL_delaymagic_gid : (Gid_t)-1,
1622 (PL_delaymagic & DM_EGID) ? PL_delaymagic_egid : (Gid_t)-1,
1623 (Gid_t)-1));
56febc5e
AD
1624#else
1625# ifdef HAS_SETREGID
b469f1e0
JH
1626 PERL_UNUSED_RESULT(
1627 setregid((PL_delaymagic & DM_RGID) ? PL_delaymagic_gid : (Gid_t)-1,
1628 (PL_delaymagic & DM_EGID) ? PL_delaymagic_egid : (Gid_t)-1));
56febc5e
AD
1629# else
1630# ifdef HAS_SETRGID
b28d0864 1631 if ((PL_delaymagic & DM_GID) == DM_RGID) {
b469f1e0 1632 PERL_UNUSED_RESULT(setrgid(PL_delaymagic_gid));
b28d0864 1633 PL_delaymagic &= ~DM_RGID;
a0d0e21e 1634 }
56febc5e
AD
1635# endif /* HAS_SETRGID */
1636# ifdef HAS_SETEGID
b28d0864 1637 if ((PL_delaymagic & DM_GID) == DM_EGID) {
b469f1e0 1638 PERL_UNUSED_RESULT(setegid(PL_delaymagic_egid));
b28d0864 1639 PL_delaymagic &= ~DM_EGID;
a0d0e21e 1640 }
56febc5e 1641# endif /* HAS_SETEGID */
b28d0864 1642 if (PL_delaymagic & DM_GID) {
985213f2 1643 if (PL_delaymagic_gid != PL_delaymagic_egid)
cea2e8a9 1644 DIE(aTHX_ "No setregid available");
b469f1e0 1645 PERL_UNUSED_RESULT(PerlProc_setgid(PL_delaymagic_gid));
a0d0e21e 1646 }
56febc5e
AD
1647# endif /* HAS_SETREGID */
1648#endif /* HAS_SETRESGID */
04783dc7 1649
985213f2
AB
1650 tmp_gid = PerlProc_getgid();
1651 tmp_egid = PerlProc_getegid();
a0d0e21e 1652 }
284167a5 1653 TAINTING_set( TAINTING_get | (tmp_uid && (tmp_euid != tmp_uid || tmp_egid != tmp_gid)) );
9a9b5ec9
DM
1654#ifdef NO_TAINT_SUPPORT
1655 PERL_UNUSED_VAR(tmp_uid);
1656 PERL_UNUSED_VAR(tmp_euid);
1657 PERL_UNUSED_VAR(tmp_gid);
1658 PERL_UNUSED_VAR(tmp_egid);
1659#endif
a0d0e21e 1660 }
a68090fe 1661 PL_delaymagic = old_delaymagic;
54310121 1662
54310121
PP
1663 if (gimme == G_VOID)
1664 SP = firstrelem - 1;
1665 else if (gimme == G_SCALAR) {
1666 dTARGET;
1667 SP = firstrelem;
231cbeb2 1668 SETi(lastrelem - firstrelem + 1);
54310121
PP
1669 }
1670 else {
1c4ea384 1671 if (ary || hash)
1d2b3927
HS
1672 /* note that in this case *firstlelem may have been overwritten
1673 by sv_undef in the odd hash case */
a0d0e21e 1674 SP = lastrelem;
1c4ea384 1675 else {
a0d0e21e 1676 SP = firstrelem + (lastlelem - firstlelem);
1c4ea384
RZ
1677 lelem = firstlelem + (relem - firstrelem);
1678 while (relem <= SP)
1679 *relem++ = (lelem <= lastlelem) ? *lelem++ : &PL_sv_undef;
1680 }
a0d0e21e 1681 }
08aeb9f7 1682
54310121 1683 RETURN;
a0d0e21e
LW
1684}
1685
8782bef2
GB
1686PP(pp_qr)
1687{
20b7effb 1688 dSP;
eb578fdb 1689 PMOP * const pm = cPMOP;
fe578d7f 1690 REGEXP * rx = PM_GETRE(pm);
10599a69 1691 SV * const pkg = rx ? CALLREG_PACKAGE(rx) : NULL;
c4420975 1692 SV * const rv = sv_newmortal();
d63c20f2
DM
1693 CV **cvp;
1694 CV *cv;
288b8c02
NC
1695
1696 SvUPGRADE(rv, SVt_IV);
c2123ae3
NC
1697 /* For a subroutine describing itself as "This is a hacky workaround" I'm
1698 loathe to use it here, but it seems to be the right fix. Or close.
1699 The key part appears to be that it's essential for pp_qr to return a new
1700 object (SV), which implies that there needs to be an effective way to
1701 generate a new SV from the existing SV that is pre-compiled in the
1702 optree. */
1703 SvRV_set(rv, MUTABLE_SV(reg_temp_copy(NULL, rx)));
288b8c02
NC
1704 SvROK_on(rv);
1705
8d919b0a 1706 cvp = &( ReANY((REGEXP *)SvRV(rv))->qr_anoncv);
5d9574c1 1707 if (UNLIKELY((cv = *cvp) && CvCLONE(*cvp))) {
d63c20f2 1708 *cvp = cv_clone(cv);
fc2b2dca 1709 SvREFCNT_dec_NN(cv);
d63c20f2
DM
1710 }
1711
288b8c02 1712 if (pkg) {
f815daf2 1713 HV *const stash = gv_stashsv(pkg, GV_ADD);
fc2b2dca 1714 SvREFCNT_dec_NN(pkg);
288b8c02
NC
1715 (void)sv_bless(rv, stash);
1716 }
1717
5d9574c1 1718 if (UNLIKELY(RX_ISTAINTED(rx))) {
e08e52cf 1719 SvTAINTED_on(rv);
9274aefd
DM
1720 SvTAINTED_on(SvRV(rv));
1721 }
c8c13c22 1722 XPUSHs(rv);
1723 RETURN;
8782bef2
GB
1724}
1725
a0d0e21e
LW
1726PP(pp_match)
1727{
20b7effb 1728 dSP; dTARG;
eb578fdb 1729 PMOP *pm = cPMOP;
d65afb4b 1730 PMOP *dynpm = pm;
eb578fdb 1731 const char *s;
5c144d81 1732 const char *strend;
99a90e59 1733 SSize_t curpos = 0; /* initial pos() or current $+[0] */
a0d0e21e 1734 I32 global;
7fadf4a7 1735 U8 r_flags = 0;
5c144d81 1736 const char *truebase; /* Start of string */
eb578fdb 1737 REGEXP *rx = PM_GETRE(pm);
b3eb6a9b 1738 bool rxtainted;
1c23e2bd 1739 const U8 gimme = GIMME_V;
a0d0e21e 1740 STRLEN len;
a3b680e6 1741 const I32 oldsave = PL_savestack_ix;
e60df1fa 1742 I32 had_zerolen = 0;
b1422d62 1743 MAGIC *mg = NULL;
a0d0e21e 1744
533c011a 1745 if (PL_op->op_flags & OPf_STACKED)
a0d0e21e 1746 TARG = POPs;
6ffceeb7 1747 else if (ARGTARG)
59f00321 1748 GETTARGET;
a0d0e21e 1749 else {
54b9620d 1750 TARG = DEFSV;
a0d0e21e
LW
1751 EXTEND(SP,1);
1752 }
d9f424b2 1753
c277df42 1754 PUTBACK; /* EVAL blocks need stack_sp. */
69dc4b30
FC
1755 /* Skip get-magic if this is a qr// clone, because regcomp has
1756 already done it. */
f1d31338 1757 truebase = ReANY(rx)->mother_re
69dc4b30
FC
1758 ? SvPV_nomg_const(TARG, len)
1759 : SvPV_const(TARG, len);
f1d31338 1760 if (!truebase)
2269b42e 1761 DIE(aTHX_ "panic: pp_match");
f1d31338 1762 strend = truebase + len;
284167a5
SM
1763 rxtainted = (RX_ISTAINTED(rx) ||
1764 (TAINT_get && (pm->op_pmflags & PMf_RETAINT)));
9212bbba 1765 TAINT_NOT;
a0d0e21e 1766
6c864ec2 1767 /* We need to know this in case we fail out early - pos() must be reset */
de0df3c0
MH
1768 global = dynpm->op_pmflags & PMf_GLOBAL;
1769
d65afb4b 1770 /* PMdf_USED is set after a ?? matches once */
c737faaf
YO
1771 if (
1772#ifdef USE_ITHREADS
1773 SvREADONLY(PL_regex_pad[pm->op_pmoffset])
1774#else
1775 pm->op_pmflags & PMf_USED
1776#endif
1777 ) {
e5dc5375 1778 DEBUG_r(PerlIO_printf(Perl_debug_log, "?? already matched once"));
de0df3c0 1779 goto nope;
a0d0e21e
LW
1780 }
1781
7e313637
FC
1782 /* empty pattern special-cased to use last successful pattern if
1783 possible, except for qr// */
8d919b0a 1784 if (!ReANY(rx)->mother_re && !RX_PRELEN(rx)
7e313637 1785 && PL_curpm) {
3280af22 1786 pm = PL_curpm;
aaa362c4 1787 rx = PM_GETRE(pm);
a0d0e21e 1788 }
d65afb4b 1789
389ecb56 1790 if (RX_MINLEN(rx) >= 0 && (STRLEN)RX_MINLEN(rx) > len) {
75d43e96
FC
1791 DEBUG_r(PerlIO_printf(Perl_debug_log, "String shorter than min possible regex match (%"
1792 UVuf" < %"IVdf")\n",
1793 (UV)len, (IV)RX_MINLEN(rx)));
de0df3c0 1794 goto nope;
e5dc5375 1795 }
c277df42 1796
8ef97b0e 1797 /* get pos() if //g */
de0df3c0 1798 if (global) {
b1422d62 1799 mg = mg_find_mglob(TARG);
8ef97b0e 1800 if (mg && mg->mg_len >= 0) {
25fdce4a 1801 curpos = MgBYTEPOS(mg, TARG, truebase, len);
8ef97b0e
DM
1802 /* last time pos() was set, it was zero-length match */
1803 if (mg->mg_flags & MGf_MINMATCH)
1804 had_zerolen = 1;
1805 }
a0d0e21e 1806 }
8ef97b0e 1807
6e240d0b 1808#ifdef PERL_SAWAMPERSAND
a41aa44d 1809 if ( RX_NPARENS(rx)
6502e081 1810 || PL_sawampersand
6502e081 1811 || (RX_EXTFLAGS(rx) & (RXf_EVAL_SEEN|RXf_PMf_KEEPCOPY))
5b0e71e9 1812 || (dynpm->op_pmflags & PMf_KEEPCOPY)
6e240d0b
FC
1813 )
1814#endif
1815 {
6502e081
DM
1816 r_flags |= (REXEC_COPY_STR|REXEC_COPY_SKIP_PRE);
1817 /* in @a =~ /(.)/g, we iterate multiple times, but copy the buffer
1818 * only on the first iteration. Therefore we need to copy $' as well
1819 * as $&, to make the rest of the string available for captures in
1820 * subsequent iterations */
1821 if (! (global && gimme == G_ARRAY))
1822 r_flags |= REXEC_COPY_SKIP_POST;
1823 };
5b0e71e9
DM
1824#ifdef PERL_SAWAMPERSAND
1825 if (dynpm->op_pmflags & PMf_KEEPCOPY)
1826 /* handle KEEPCOPY in pmop but not rx, eg $r=qr/a/; /$r/p */
1827 r_flags &= ~(REXEC_COPY_SKIP_PRE|REXEC_COPY_SKIP_POST);
1828#endif
22e551b9 1829
f1d31338
DM
1830 s = truebase;
1831
d7be1480 1832 play_it_again:
985afbc1 1833 if (global)
03c83e26 1834 s = truebase + curpos;
f722798b 1835
77da2310 1836 if (!CALLREGEXEC(rx, (char*)s, (char *)strend, (char*)truebase,
03c83e26 1837 had_zerolen, TARG, NULL, r_flags))
03b6c93d 1838 goto nope;
77da2310
NC
1839
1840 PL_curpm = pm;
985afbc1 1841 if (dynpm->op_pmflags & PMf_ONCE)
c737faaf 1842#ifdef USE_ITHREADS
77da2310 1843 SvREADONLY_on(PL_regex_pad[dynpm->op_pmoffset]);
c737faaf 1844#else
77da2310 1845 dynpm->op_pmflags |= PMf_USED;
c737faaf 1846#endif
a0d0e21e 1847
72311751
GS
1848 if (rxtainted)
1849 RX_MATCH_TAINTED_on(rx);
1850 TAINT_IF(RX_MATCH_TAINTED(rx));
35c2ccc3
DM
1851
1852 /* update pos */
1853
1854 if (global && (gimme != G_ARRAY || (dynpm->op_pmflags & PMf_CONTINUE))) {
b1422d62 1855 if (!mg)
35c2ccc3 1856 mg = sv_magicext_mglob(TARG);
25fdce4a 1857 MgBYTEPOS_set(mg, TARG, truebase, RX_OFFS(rx)[0].end);
adf51885
DM
1858 if (RX_ZERO_LEN(rx))
1859 mg->mg_flags |= MGf_MINMATCH;
1860 else
1861 mg->mg_flags &= ~MGf_MINMATCH;
35c2ccc3
DM
1862 }
1863
bf9dff51
DM
1864 if ((!RX_NPARENS(rx) && !global) || gimme != G_ARRAY) {
1865 LEAVE_SCOPE(oldsave);
1866 RETPUSHYES;
1867 }
1868
88ab22af
DM
1869 /* push captures on stack */
1870
bf9dff51 1871 {
07bc277f 1872 const I32 nparens = RX_NPARENS(rx);
a3b680e6 1873 I32 i = (global && !nparens) ? 1 : 0;
a0d0e21e 1874
c277df42 1875 SPAGAIN; /* EVAL blocks could move the stack. */
ffc61ed2
JH
1876 EXTEND(SP, nparens + i);
1877 EXTEND_MORTAL(nparens + i);
1878 for (i = !i; i <= nparens; i++) {
a0d0e21e 1879 PUSHs(sv_newmortal());
5d9574c1
DM
1880 if (LIKELY((RX_OFFS(rx)[i].start != -1)
1881 && RX_OFFS(rx)[i].end != -1 ))
1882 {
07bc277f 1883 const I32 len = RX_OFFS(rx)[i].end - RX_OFFS(rx)[i].start;
f1d31338 1884 const char * const s = RX_OFFS(rx)[i].start + truebase;
5d9574c1
DM
1885 if (UNLIKELY(RX_OFFS(rx)[i].end < 0 || RX_OFFS(rx)[i].start < 0
1886 || len < 0 || len > strend - s))
5637ef5b
NC
1887 DIE(aTHX_ "panic: pp_match start/end pointers, i=%ld, "
1888 "start=%ld, end=%ld, s=%p, strend=%p, len=%"UVuf,
1889 (long) i, (long) RX_OFFS(rx)[i].start,
1890 (long)RX_OFFS(rx)[i].end, s, strend, (UV) len);
a0d0e21e 1891 sv_setpvn(*SP, s, len);
cce850e4 1892 if (DO_UTF8(TARG) && is_utf8_string((U8*)s, len))
a197cbdd 1893 SvUTF8_on(*SP);
a0d0e21e
LW
1894 }
1895 }
1896 if (global) {
0e0b3e82 1897 curpos = (UV)RX_OFFS(rx)[0].end;
03c83e26 1898 had_zerolen = RX_ZERO_LEN(rx);
c277df42 1899 PUTBACK; /* EVAL blocks may use stack */
cf93c79d 1900 r_flags |= REXEC_IGNOREPOS | REXEC_NOT_FIRST;
a0d0e21e
LW
1901 goto play_it_again;
1902 }
4633a7c4 1903 LEAVE_SCOPE(oldsave);
a0d0e21e
LW
1904 RETURN;
1905 }
e5964223 1906 NOT_REACHED; /* NOTREACHED */
a0d0e21e 1907
7b52d656 1908 nope:
d65afb4b 1909 if (global && !(dynpm->op_pmflags & PMf_CONTINUE)) {
b1422d62
DM
1910 if (!mg)
1911 mg = mg_find_mglob(TARG);
1912 if (mg)
1913 mg->mg_len = -1;
a0d0e21e 1914 }
4633a7c4 1915 LEAVE_SCOPE(oldsave);
a0d0e21e
LW
1916 if (gimme == G_ARRAY)
1917 RETURN;
1918 RETPUSHNO;
1919}
1920
1921OP *
864dbfa3 1922Perl_do_readline(pTHX)
a0d0e21e 1923{
20b7effb 1924 dSP; dTARGETSTACKED;
eb578fdb 1925 SV *sv;
a0d0e21e
LW
1926 STRLEN tmplen = 0;
1927 STRLEN offset;
760ac839 1928 PerlIO *fp;
eb578fdb
KW
1929 IO * const io = GvIO(PL_last_in_gv);
1930 const I32 type = PL_op->op_type;
1c23e2bd 1931 const U8 gimme = GIMME_V;
a0d0e21e 1932
6136c704 1933 if (io) {
50db69d8 1934 const MAGIC *const mg = SvTIED_mg((const SV *)io, PERL_MAGIC_tiedscalar);
6136c704 1935 if (mg) {
3e0cb5de 1936 Perl_tied_method(aTHX_ SV_CONST(READLINE), SP, MUTABLE_SV(io), mg, gimme, 0);
6136c704 1937 if (gimme == G_SCALAR) {
50db69d8
NC
1938 SPAGAIN;
1939 SvSetSV_nosteal(TARG, TOPs);
1940 SETTARG;
6136c704 1941 }
50db69d8 1942 return NORMAL;
0b7c7b4f 1943 }
e79b0511 1944 }
4608196e 1945 fp = NULL;
a0d0e21e
LW
1946 if (io) {
1947 fp = IoIFP(io);
1948 if (!fp) {
1949 if (IoFLAGS(io) & IOf_ARGV) {
1950 if (IoFLAGS(io) & IOf_START) {
a0d0e21e 1951 IoLINES(io) = 0;
b9f2b683 1952 if (av_tindex(GvAVn(PL_last_in_gv)) < 0) {
1d7c1841 1953 IoFLAGS(io) &= ~IOf_START;
d5eb9a46 1954 do_open6(PL_last_in_gv, "-", 1, NULL, NULL, 0);
4bac9ae4 1955 SvTAINTED_off(GvSVn(PL_last_in_gv)); /* previous tainting irrelevant */
76f68e9b 1956 sv_setpvs(GvSVn(PL_last_in_gv), "-");
3280af22 1957 SvSETMAGIC(GvSV(PL_last_in_gv));
a2008d6d
GS
1958 fp = IoIFP(io);
1959 goto have_fp;
a0d0e21e
LW
1960 }
1961 }
157fb5a1 1962 fp = nextargv(PL_last_in_gv, PL_op->op_flags & OPf_SPECIAL);
a0d0e21e 1963 if (!fp) { /* Note: fp != IoIFP(io) */
3280af22 1964 (void)do_close(PL_last_in_gv, FALSE); /* now it does*/
a0d0e21e
LW
1965 }
1966 }
0d44d22b
NC
1967 else if (type == OP_GLOB)
1968 fp = Perl_start_glob(aTHX_ POPs, io);
a0d0e21e
LW
1969 }
1970 else if (type == OP_GLOB)
1971 SP--;
7716c5c5 1972 else if (IoTYPE(io) == IoTYPE_WRONLY) {
a5390457 1973 report_wrongway_fh(PL_last_in_gv, '>');
a00b5bd3 1974 }
a0d0e21e
LW
1975 }
1976 if (!fp) {
041457d9 1977 if ((!io || !(IoFLAGS(io) & IOf_START))
de7dabb6
TC
1978 && ckWARN(WARN_CLOSED)
1979 && type != OP_GLOB)
041457d9 1980 {
de7dabb6 1981 report_evil_fh(PL_last_in_gv);
3f4520fe 1982 }
54310121 1983 if (gimme == G_SCALAR) {
79628082 1984 /* undef TARG, and push that undefined value */
ba92458f 1985 if (type != OP_RCATLINE) {
aab1202a 1986 sv_setsv(TARG,NULL);
ba92458f 1987 }
a0d0e21e
LW
1988 PUSHTARG;
1989 }
1990 RETURN;
1991 }
a2008d6d 1992 have_fp:
54310121 1993 if (gimme == G_SCALAR) {
a0d0e21e 1994 sv = TARG;
0f722b55
RGS
1995 if (type == OP_RCATLINE && SvGMAGICAL(sv))
1996 mg_get(sv);
48de12d9
RGS
1997 if (SvROK(sv)) {
1998 if (type == OP_RCATLINE)
5668452f 1999 SvPV_force_nomg_nolen(sv);
48de12d9
RGS
2000 else
2001 sv_unref(sv);
2002 }
f7877b28 2003 else if (isGV_with_GP(sv)) {
5668452f 2004 SvPV_force_nomg_nolen(sv);
f7877b28 2005 }
862a34c6 2006 SvUPGRADE(sv, SVt_PV);
a0d0e21e 2007 tmplen = SvLEN(sv); /* remember if already alloced */
e3918bb7 2008 if (!tmplen && !SvREADONLY(sv) && !SvIsCOW(sv)) {
f72e8700
JJ
2009 /* try short-buffering it. Please update t/op/readline.t
2010 * if you change the growth length.
2011 */
2012 Sv_Grow(sv, 80);
2013 }
2b5e58c4
AMS
2014 offset = 0;
2015 if (type == OP_RCATLINE && SvOK(sv)) {
2016 if (!SvPOK(sv)) {
5668452f 2017 SvPV_force_nomg_nolen(sv);
2b5e58c4 2018 }
a0d0e21e 2019 offset = SvCUR(sv);
2b5e58c4 2020 }
a0d0e21e 2021 }
54310121 2022 else {
561b68a9 2023 sv = sv_2mortal(newSV(80));
54310121
PP
2024 offset = 0;
2025 }
fbad3eb5 2026
3887d568
AP
2027 /* This should not be marked tainted if the fp is marked clean */
2028#define MAYBE_TAINT_LINE(io, sv) \
2029 if (!(IoFLAGS(io) & IOf_UNTAINT)) { \
2030 TAINT; \
2031 SvTAINTED_on(sv); \
2032 }
2033
684bef36 2034/* delay EOF state for a snarfed empty file */
fbad3eb5 2035#define SNARF_EOF(gimme,rs,io,sv) \
684bef36 2036 (gimme != G_SCALAR || SvCUR(sv) \
b9fee9ba 2037 || (IoFLAGS(io) & IOf_NOLINE) || !RsSNARF(rs))
fbad3eb5 2038
a0d0e21e 2039 for (;;) {
09e8efcc 2040 PUTBACK;
fbad3eb5 2041 if (!sv_gets(sv, fp, offset)
2d726892
TF
2042 && (type == OP_GLOB
2043 || SNARF_EOF(gimme, PL_rs, io, sv)
2044 || PerlIO_error(fp)))
fbad3eb5 2045 {
760ac839 2046 PerlIO_clearerr(fp);
a0d0e21e 2047 if (IoFLAGS(io) & IOf_ARGV) {
157fb5a1 2048 fp = nextargv(PL_last_in_gv, PL_op->op_flags & OPf_SPECIAL);
a0d0e21e
LW
2049 if (fp)
2050 continue;
3280af22 2051 (void)do_close(PL_last_in_gv, FALSE);
a0d0e21e
LW
2052 }
2053 else if (type == OP_GLOB) {
a2a5de95
NC
2054 if (!do_close(PL_last_in_gv, FALSE)) {
2055 Perl_ck_warner(aTHX_ packWARN(WARN_GLOB),
2056 "glob failed (child exited with status %d%s)",
2057 (int)(STATUS_CURRENT >> 8),
2058 (STATUS_CURRENT & 0x80) ? ", core dumped" : "");
4eb79ab5 2059 }
a0d0e21e 2060 }
54310121 2061 if (gimme == G_SCALAR) {
ba92458f
AE
2062 if (type != OP_RCATLINE) {
2063 SV_CHECK_THINKFIRST_COW_DROP(TARG);
0c34ef67 2064 SvOK_off(TARG);
ba92458f 2065 }
09e8efcc 2066 SPAGAIN;
a0d0e21e
LW
2067 PUSHTARG;
2068 }
3887d568 2069 MAYBE_TAINT_LINE(io, sv);
a0d0e21e
LW
2070 RETURN;
2071 }
3887d568 2072 MAYBE_TAINT_LINE(io, sv);
a0d0e21e 2073 IoLINES(io)++;
b9fee9ba 2074 IoFLAGS(io) |= IOf_NOLINE;
71be2cbc 2075 SvSETMAGIC(sv);
09e8efcc 2076 SPAGAIN;
a0d0e21e 2077 XPUSHs(sv);
a0d0e21e 2078 if (type == OP_GLOB) {
349d4f2f 2079 const char *t1;
45a23732 2080 Stat_t statbuf;
a0d0e21e 2081
3280af22 2082 if (SvCUR(sv) > 0 && SvCUR(PL_rs) > 0) {
6136c704 2083 char * const tmps = SvEND(sv) - 1;
aa07b2f6 2084 if (*tmps == *SvPVX_const(PL_rs)) {
c07a80fd 2085 *tmps = '\0';
b162af07 2086 SvCUR_set(sv, SvCUR(sv) - 1);
c07a80fd
PP
2087 }
2088 }
349d4f2f 2089 for (t1 = SvPVX_const(sv); *t1; t1++)
b51c3e77
CB
2090#ifdef __VMS
2091 if (strchr("*%?", *t1))
2092#else
7ad1e72d 2093 if (strchr("$&*(){}[]'\";\\|?<>~`", *t1))
b51c3e77 2094#endif
a0d0e21e 2095 break;
45a23732 2096 if (*t1 && PerlLIO_lstat(SvPVX_const(sv), &statbuf) < 0) {
a0d0e21e
LW
2097 (void)POPs; /* Unmatched wildcard? Chuck it... */
2098 continue;
2099 }
2d79bf7f 2100 } else if (SvUTF8(sv)) { /* OP_READLINE, OP_RCATLINE */
d4c19fe8
AL
2101 if (ckWARN(WARN_UTF8)) {
2102 const U8 * const s = (const U8*)SvPVX_const(sv) + offset;
2103 const STRLEN len = SvCUR(sv) - offset;
2104 const U8 *f;
2105
2106 if (!is_utf8_string_loc(s, len, &f))
2107 /* Emulate :encoding(utf8) warning in the same case. */
2108 Perl_warner(aTHX_ packWARN(WARN_UTF8),
2109 "utf8 \"\\x%02X\" does not map to Unicode",
2110 f < (U8*)SvEND(sv) ? *f : 0);
2111 }
a0d0e21e 2112 }
54310121 2113 if (gimme == G_ARRAY) {
a0d0e21e 2114 if (SvLEN(sv) - SvCUR(sv) > 20) {
1da4ca5f 2115 SvPV_shrink_to_cur(sv);
a0d0e21e 2116 }
561b68a9 2117 sv = sv_2mortal(newSV(80));
a0d0e21e
LW
2118 continue;
2119 }
54310121 2120 else if (gimme == G_SCALAR && !tmplen && SvLEN(sv) - SvCUR(sv) > 80) {
a0d0e21e 2121 /* try to reclaim a bit of scalar space (only on 1st alloc) */
d5b5861b
NC
2122 const STRLEN new_len
2123 = SvCUR(sv) < 60 ? 80 : SvCUR(sv)+40; /* allow some slop */
1da4ca5f 2124 SvPV_renew(sv, new_len);
a0d0e21e
LW
2125 }
2126 RETURN;
2127 }
2128}
2129
a0d0e21e
LW
2130PP(pp_helem)
2131{
20b7effb 2132 dSP;
760ac839 2133 HE* he;
ae77835f 2134 SV **svp;
c445ea15 2135 SV * const keysv = POPs;
85fbaab2 2136 HV * const hv = MUTABLE_HV(POPs);
a3b680e6
AL
2137 const U32 lval = PL_op->op_flags & OPf_MOD || LVRET;
2138 const U32 defer = PL_op->op_private & OPpLVAL_DEFER;
be6c24e0 2139 SV *sv;
92970b93 2140 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
d30e492c 2141 bool preeminent = TRUE;
a0d0e21e 2142
6dfc73ea
SM
2143 if (SvTYPE(hv) != SVt_PVHV)
2144 RETPUSHUNDEF;
d4c19fe8 2145
92970b93 2146 if (localizing) {
d4c19fe8
AL
2147 MAGIC *mg;
2148 HV *stash;
d30e492c
VP
2149
2150 /* If we can determine whether the element exist,
2151 * Try to preserve the existenceness of a tied hash
2152 * element by using EXISTS and DELETE if possible.
2153 * Fallback to FETCH and STORE otherwise. */
2c5f48c2 2154 if (SvCANEXISTDELETE(hv))
d30e492c 2155 preeminent = hv_exists_ent(hv, keysv, 0);
d4c19fe8 2156 }
d30e492c 2157
5f9d7e2b 2158 he = hv_fetch_ent(hv, keysv, lval && !defer, 0);
d4c19fe8 2159 svp = he ? &HeVAL(he) : NULL;
a0d0e21e 2160 if (lval) {
746f6409 2161 if (!svp || !*svp || *svp == &PL_sv_undef) {
68dc0745
PP
2162 SV* lv;
2163 SV* key2;
2d8e6c8d 2164 if (!defer) {
be2597df 2165 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
2d8e6c8d 2166 }
68dc0745
PP
2167 lv = sv_newmortal();
2168 sv_upgrade(lv, SVt_PVLV);
2169 LvTYPE(lv) = 'y';
6136c704 2170 sv_magic(lv, key2 = newSVsv(keysv), PERL_MAGIC_defelem, NULL, 0);
fc2b2dca 2171 SvREFCNT_dec_NN(key2); /* sv_magic() increments refcount */
0ad694a7 2172 LvTARG(lv) = SvREFCNT_inc_simple_NN(hv);
68dc0745
PP
2173 LvTARGLEN(lv) = 1;
2174 PUSHs(lv);
2175 RETURN;
2176 }
92970b93 2177 if (localizing) {
bfcb3514 2178 if (HvNAME_get(hv) && isGV(*svp))
159b6efe 2179 save_gp(MUTABLE_GV(*svp), !(PL_op->op_flags & OPf_SPECIAL));
47cfc530
VP
2180 else if (preeminent)
2181 save_helem_flags(hv, keysv, svp,
2182 (PL_op->op_flags & OPf_SPECIAL) ? 0 : SAVEf_SETMAGIC);
2183 else
2184 SAVEHDELETE(hv, keysv);
5f05dabc 2185 }
9026059d
GG
2186 else if (PL_op->op_private & OPpDEREF) {
2187 PUSHs(vivify_ref(*svp, PL_op->op_private & OPpDEREF));
2188 RETURN;
2189 }
a0d0e21e 2190 }
746f6409 2191 sv = (svp && *svp ? *svp : &PL_sv_undef);
fd69380d
DM
2192 /* Originally this did a conditional C<sv = sv_mortalcopy(sv)>; this
2193 * was to make C<local $tied{foo} = $tied{foo}> possible.
2194 * However, it seems no longer to be needed for that purpose, and
2195 * introduced a new bug: stuff like C<while ($hash{taintedval} =~ /.../g>
2196 * would loop endlessly since the pos magic is getting set on the
2197 * mortal copy and lost. However, the copy has the effect of
2198 * triggering the get magic, and losing it altogether made things like
2199 * c<$tied{foo};> in void context no longer do get magic, which some
2200 * code relied on. Also, delayed triggering of magic on @+ and friends
2201 * meant the original regex may be out of scope by now. So as a
2202 * compromise, do the get magic here. (The MGf_GSKIP flag will stop it
2203 * being called too many times). */
39cf747a 2204 if (!lval && SvRMAGICAL(hv) && SvGMAGICAL(sv))
fd69380d 2205 mg_get(sv);
be6c24e0 2206 PUSHs(sv);
a0d0e21e
LW
2207 RETURN;
2208}
2209
fedf30e1
DM
2210
2211/* a stripped-down version of Perl_softref2xv() for use by
2212 * pp_multideref(), which doesn't use PL_op->op_flags */
2213
f9db5646 2214STATIC GV *
fedf30e1
DM
2215S_softref2xv_lite(pTHX_ SV *const sv, const char *const what,
2216 const svtype type)
2217{
2218 if (PL_op->op_private & HINT_STRICT_REFS) {
2219 if (SvOK(sv))
2220 Perl_die(aTHX_ PL_no_symref_sv, sv,
2221 (SvPOKp(sv) && SvCUR(sv)>32 ? "..." : ""), what);
2222 else
2223 Perl_die(aTHX_ PL_no_usym, what);
2224 }
2225 if (!SvOK(sv))
2226 Perl_die(aTHX_ PL_no_usym, what);
2227 return gv_fetchsv_nomg(sv, GV_ADD, type);
2228}
2229
2230
79815f56
DM
2231/* Handle one or more aggregate derefs and array/hash indexings, e.g.
2232 * $h->{foo} or $a[0]{$key}[$i] or f()->[1]
fedf30e1
DM
2233 *
2234 * op_aux points to an array of unions of UV / IV / SV* / PADOFFSET.
79815f56
DM
2235 * Each of these either contains a set of actions, or an argument, such as
2236 * an IV to use as an array index, or a lexical var to retrieve.
2237 * Several actions re stored per UV; we keep shifting new actions off the
2238 * one UV, and only reload when it becomes zero.
fedf30e1
DM
2239 */
2240
2241PP(pp_multideref)
2242{
2243 SV *sv = NULL; /* init to avoid spurious 'may be used uninitialized' */
2244 UNOP_AUX_item *items = cUNOP_AUXx(PL_op)->op_aux;
2245 UV actions = items->uv;
2246
2247 assert(actions);
2248 /* this tells find_uninit_var() where we're up to */
2249 PL_multideref_pc = items;
2250
2251 while (1) {
2252 /* there are three main classes of action; the first retrieve
2253 * the initial AV or HV from a variable or the stack; the second
2254 * does the equivalent of an unrolled (/DREFAV, rv2av, aelem),
2255 * the third an unrolled (/DREFHV, rv2hv, helem).
2256 */
2257 switch (actions & MDEREF_ACTION_MASK) {
2258
2259 case MDEREF_reload:
2260 actions = (++items)->uv;
2261 continue;
2262
2263 case MDEREF_AV_padav_aelem: /* $lex[...] */
2264 sv = PAD_SVl((++items)->pad_offset);
2265 goto do_AV_aelem;
2266
2267 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
2268 sv = UNOP_AUX_item_sv(++items);
2269 assert(isGV_with_GP(sv));
2270 sv = (SV*)GvAVn((GV*)sv);
2271 goto do_AV_aelem;
2272
2273 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
2274 {
2275 dSP;
2276 sv = POPs;
2277 PUTBACK;
2278 goto do_AV_rv2av_aelem;
2279 }
2280
2281 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
2282 sv = UNOP_AUX_item_sv(++items);
2283 assert(isGV_with_GP(sv));
2284 sv = GvSVn((GV*)sv);
2285 goto do_AV_vivify_rv2av_aelem;
2286
2287 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
2288 sv = PAD_SVl((++items)->pad_offset);
2289 /* FALLTHROUGH */
2290
2291 do_AV_vivify_rv2av_aelem:
2292 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
2293 /* this is the OPpDEREF action normally found at the end of
2294 * ops like aelem, helem, rv2sv */
2295 sv = vivify_ref(sv, OPpDEREF_AV);
2296 /* FALLTHROUGH */
2297
2298 do_AV_rv2av_aelem:
2299 /* this is basically a copy of pp_rv2av when it just has the
2300 * sKR/1 flags */
2301 SvGETMAGIC(sv);
2302 if (LIKELY(SvROK(sv))) {
2303 if (UNLIKELY(SvAMAGIC(sv))) {
2304 sv = amagic_deref_call(sv, to_av_amg);
2305 }
2306 sv = SvRV(sv);
2307 if (UNLIKELY(SvTYPE(sv) != SVt_PVAV))
2308 DIE(aTHX_ "Not an ARRAY reference");
2309 }
2310 else if (SvTYPE(sv) != SVt_PVAV) {
2311 if (!isGV_with_GP(sv))
2312 sv = (SV*)S_softref2xv_lite(aTHX_ sv, "an ARRAY", SVt_PVAV);
2313 sv = MUTABLE_SV(GvAVn((GV*)sv));
2314 }
2315 /* FALLTHROUGH */
2316
2317 do_AV_aelem:
2318 {
2319 /* retrieve the key; this may be either a lexical or package
2320 * var (whose index/ptr is stored as an item) or a signed
2321 * integer constant stored as an item.
2322 */
2323 SV *elemsv;
2324 IV elem = 0; /* to shut up stupid compiler warnings */
2325
2326
2327 assert(SvTYPE(sv) == SVt_PVAV);
2328
2329 switch (actions & MDEREF_INDEX_MASK) {
2330 case MDEREF_INDEX_none:
2331 goto finish;
2332 case MDEREF_INDEX_const:
2333 elem = (++items)->iv;
2334 break;
2335 case MDEREF_INDEX_padsv:
2336 elemsv = PAD_SVl((++items)->pad_offset);
2337 goto check_elem;
2338 case MDEREF_INDEX_gvsv:
2339 elemsv = UNOP_AUX_item_sv(++items);
2340 assert(isGV_with_GP(elemsv));
2341 elemsv = GvSVn((GV*)elemsv);
2342 check_elem:
2343 if (UNLIKELY(SvROK(elemsv) && !SvGAMAGIC(elemsv)
2344 && ckWARN(WARN_MISC)))
2345 Perl_warner(aTHX_ packWARN(WARN_MISC),
2346 "Use of reference \"%"SVf"\" as array index",
2347 SVfARG(elemsv));
2348 /* the only time that S_find_uninit_var() needs this
2349 * is to determine which index value triggered the
2350 * undef warning. So just update it here. Note that
2351 * since we don't save and restore this var (e.g. for
2352 * tie or overload execution), its value will be
2353 * meaningless apart from just here */
2354 PL_multideref_pc = items;
2355 elem = SvIV(elemsv);
2356 break;
2357 }
2358
2359
2360 /* this is basically a copy of pp_aelem with OPpDEREF skipped */
2361
2362 if (!(actions & MDEREF_FLAG_last)) {
2363 SV** svp = av_fetch((AV*)sv, elem, 1);
2364 if (!svp || ! (sv=*svp))
2365 DIE(aTHX_ PL_no_aelem, elem);
2366 break;
2367 }
2368
2369 if (PL_op->op_private &
2370 (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE))
2371 {
2372 if (PL_op->op_private & OPpMULTIDEREF_EXISTS) {
2373 sv = av_exists((AV*)sv, elem) ? &PL_sv_yes : &PL_sv_no;
2374 }
2375 else {
2376 I32 discard = (GIMME_V == G_VOID) ? G_DISCARD : 0;
2377 sv = av_delete((AV*)sv, elem, discard);
2378 if (discard)
2379 return NORMAL;
2380 if (!sv)
2381 sv = &PL_sv_undef;
2382 }
2383 }
2384 else {
2385 const U32 lval = PL_op->op_flags & OPf_MOD || LVRET;
2386 const U32 defer = PL_op->op_private & OPpLVAL_DEFER;
2387 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
2388 bool preeminent = TRUE;
2389 AV *const av = (AV*)sv;
2390 SV** svp;
2391
2392 if (UNLIKELY(localizing)) {
2393 MAGIC *mg;
2394 HV *stash;
2395
2396 /* If we can determine whether the element exist,
2397 * Try to preserve the existenceness of a tied array
2398 * element by using EXISTS and DELETE if possible.
2399 * Fallback to FETCH and STORE otherwise. */
2400 if (SvCANEXISTDELETE(av))
2401 preeminent = av_exists(av, elem);
2402 }
2403
2404 svp = av_fetch(av, elem, lval && !defer);
2405
2406 if (lval) {
2407 if (!svp || !(sv = *svp)) {
2408 IV len;
2409 if (!defer)
2410 DIE(aTHX_ PL_no_aelem, elem);
2411 len = av_tindex(av);
2412 sv = sv_2mortal(newSVavdefelem(av,
2413 /* Resolve a negative index now, unless it points
2414 * before the beginning of the array, in which
2415 * case record it for error reporting in
2416 * magic_setdefelem. */
2417 elem < 0 && len + elem >= 0
2418 ? len + elem : elem, 1));
2419 }
2420 else {
2421 if (UNLIKELY(localizing)) {
2422 if (preeminent) {
2423 save_aelem(av, elem, svp);
2424 sv = *svp; /* may have changed */
2425 }
2426 else
2427 SAVEADELETE(av, elem);
2428 }
2429 }
2430 }
2431 else {
2432 sv = (svp ? *svp : &PL_sv_undef);
2433 /* see note in pp_helem() */
2434 if (SvRMAGICAL(av) && SvGMAGICAL(sv))
2435 mg_get(sv);
2436 }
2437 }
2438
2439 }
2440 finish:
2441 {
2442 dSP;
2443 XPUSHs(sv);
2444 RETURN;
2445 }
2446 /* NOTREACHED */
2447
2448
2449
2450
2451 case MDEREF_HV_padhv_helem: /* $lex{...} */
2452 sv = PAD_SVl((++items)->pad_offset);
2453 goto do_HV_helem;
2454
2455 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
2456 sv = UNOP_AUX_item_sv(++items);
2457 assert(isGV_with_GP(sv));
2458 sv = (SV*)GvHVn((GV*)sv);
2459 goto do_HV_helem;
2460
2461 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
2462 {
2463 dSP;
2464 sv = POPs;
2465 PUTBACK;
2466 goto do_HV_rv2hv_helem;
2467 }
2468
2469 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
2470 sv = UNOP_AUX_item_sv(++items);
2471 assert(isGV_with_GP(sv));
2472 sv = GvSVn((GV*)sv);
2473 goto do_HV_vivify_rv2hv_helem;
2474
2475 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
2476 sv = PAD_SVl((++items)->pad_offset);
2477 /* FALLTHROUGH */
2478
2479 do_HV_vivify_rv2hv_helem:
2480 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
2481 /* this is the OPpDEREF action normally found at the end of
2482 * ops like aelem, helem, rv2sv */
2483 sv = vivify_ref(sv, OPpDEREF_HV);
2484 /* FALLTHROUGH */
2485
2486 do_HV_rv2hv_helem:
2487 /* this is basically a copy of pp_rv2hv when it just has the
2488 * sKR/1 flags (and pp_rv2hv is aliased to pp_rv2av) */
2489
2490 SvGETMAGIC(sv);
2491 if (LIKELY(SvROK(sv))) {
2492 if (UNLIKELY(SvAMAGIC(sv))) {
2493 sv = amagic_deref_call(sv, to_hv_amg);
2494 }
2495 sv = SvRV(sv);
2496 if (UNLIKELY(SvTYPE(sv) != SVt_PVHV))
2497 DIE(aTHX_ "Not a HASH reference");
2498 }
2499 else if (SvTYPE(sv) != SVt_PVHV) {
2500 if (!isGV_with_GP(sv))
2501 sv = (SV*)S_softref2xv_lite(aTHX_ sv, "a HASH", SVt_PVHV);
2502 sv = MUTABLE_SV(GvHVn((GV*)sv));
2503 }
2504 /* FALLTHROUGH */
2505
2506 do_HV_helem:
2507 {
2508 /* retrieve the key; this may be either a lexical / package
2509 * var or a string constant, whose index/ptr is stored as an
2510 * item
2511 */
2512 SV *keysv = NULL; /* to shut up stupid compiler warnings */
2513
2514 assert(SvTYPE(sv) == SVt_PVHV);
2515
2516 switch (actions & MDEREF_INDEX_MASK) {
2517 case MDEREF_INDEX_none:
2518 goto finish;
2519
2520 case MDEREF_INDEX_const:
2521 keysv = UNOP_AUX_item_sv(++items);
2522 break;
2523
2524 case MDEREF_INDEX_padsv:
2525 keysv = PAD_SVl((++items)->pad_offset);
2526 break;
2527
2528 case MDEREF_INDEX_gvsv:
2529 keysv = UNOP_AUX_item_sv(++items);
2530 keysv = GvSVn((GV*)keysv);
2531 break;
2532 }
2533
2534 /* see comment above about setting this var */
2535 PL_multideref_pc = items;
2536
2537
2538 /* ensure that candidate CONSTs have been HEKified */
2539 assert( ((actions & MDEREF_INDEX_MASK) != MDEREF_INDEX_const)
2540 || SvTYPE(keysv) >= SVt_PVMG
2541 || !SvOK(keysv)
2542 || SvROK(keysv)
2543 || SvIsCOW_shared_hash(keysv));
2544
2545 /* this is basically a copy of pp_helem with OPpDEREF skipped */
2546
2547 if (!(actions & MDEREF_FLAG_last)) {
2548 HE *he = hv_fetch_ent((HV*)sv, keysv, 1, 0);
2549 if (!he || !(sv=HeVAL(he)) || sv == &PL_sv_undef)
2550 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
2551 break;
2552 }
2553
2554 if (PL_op->op_private &
2555 (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE))
2556 {
2557 if (PL_op->op_private & OPpMULTIDEREF_EXISTS) {
2558 sv = hv_exists_ent((HV*)sv, keysv, 0)
2559 ? &PL_sv_yes : &PL_sv_no;
2560 }
2561 else {
2562 I32 discard = (GIMME_V == G_VOID) ? G_DISCARD : 0;
2563 sv = hv_delete_ent((HV*)sv, keysv, discard, 0);
2564 if (discard)
2565 return NORMAL;
2566 if (!sv)
2567 sv = &PL_sv_undef;
2568 }
2569 }
2570 else {
2571 const U32 lval = PL_op->op_flags & OPf_MOD || LVRET;
2572 const U32 defer = PL_op->op_private & OPpLVAL_DEFER;
2573 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
2574 bool preeminent = TRUE;
2575 SV **svp;
2576 HV * const hv = (HV*)sv;
2577 HE* he;
2578
2579 if (UNLIKELY(localizing)) {
2580 MAGIC *mg;
2581 HV *stash;
2582
2583 /* If we can determine whether the element exist,
2584 * Try to preserve the existenceness of a tied hash
2585 * element by using EXISTS and DELETE if possible.
2586 * Fallback to FETCH and STORE otherwise. */
2587 if (SvCANEXISTDELETE(hv))
2588 preeminent = hv_exists_ent(hv, keysv, 0);
2589 }
2590
2591 he = hv_fetch_ent(hv, keysv, lval && !defer, 0);
2592 svp = he ? &HeVAL(he) : NULL;
2593
2594
2595 if (lval) {
2596 if (!svp || !(sv = *svp) || sv == &PL_sv_undef) {
2597 SV* lv;
2598 SV* key2;
2599 if (!defer)
2600 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
2601 lv = sv_newmortal();
2602 sv_upgrade(lv, SVt_PVLV);
2603 LvTYPE(lv) = 'y';
2604 sv_magic(lv, key2 = newSVsv(keysv),
2605 PERL_MAGIC_defelem, NULL, 0);
2606 /* sv_magic() increments refcount */
2607 SvREFCNT_dec_NN(key2);
0ad694a7 2608 LvTARG(lv) = SvREFCNT_inc_simple_NN(hv);
fedf30e1
DM
2609 LvTARGLEN(lv) = 1;
2610 sv = lv;
2611 }
2612 else {
2613 if (localizing) {
2614 if (HvNAME_get(hv) && isGV(sv))
2615 save_gp(MUTABLE_GV(sv),
2616 !(PL_op->op_flags & OPf_SPECIAL));
2617 else if (preeminent) {
2618 save_helem_flags(hv, keysv, svp,
2619 (PL_op->op_flags & OPf_SPECIAL)
2620 ? 0 : SAVEf_SETMAGIC);
2621 sv = *svp; /* may have changed */
2622 }
2623 else
2624 SAVEHDELETE(hv, keysv);
2625 }
2626 }
2627 }
2628 else {
2629 sv = (svp && *svp ? *svp : &PL_sv_undef);
2630 /* see note in pp_helem() */
2631 if (SvRMAGICAL(hv) && SvGMAGICAL(sv))
2632 mg_get(sv);
2633 }
2634 }
2635 goto finish;
2636 }
2637
2638 } /* switch */
2639
2640 actions >>= MDEREF_SHIFT;
2641 } /* while */
2642 /* NOTREACHED */
2643}
2644
2645
a0d0e21e
LW
2646PP(pp_iter)
2647{
eb578fdb 2648 PERL_CONTEXT *cx;
7d6c2cef 2649 SV *oldsv;
1d7c1841 2650 SV **itersvp;
8a1f10dd 2651 SV *retsv;
a0d0e21e 2652
84f05d57
JH
2653 SV *sv;
2654 AV *av;
2655 IV ix;
2656 IV inc;
2657
4ebe6e95 2658 cx = CX_CUR();
1d7c1841 2659 itersvp = CxITERVAR(cx);
4b5c941e 2660 assert(itersvp);
a48ce6be
DM
2661
2662 switch (CxTYPE(cx)) {
17c91640 2663
b552b52c
DM
2664 case CXt_LOOP_LAZYSV: /* string increment */
2665 {
2666 SV* cur = cx->blk_loop.state_u.lazysv.cur;
2667 SV *end = cx->blk_loop.state_u.lazysv.end;
2668 /* If the maximum is !SvOK(), pp_enteriter substitutes PL_sv_no.
2669 It has SvPVX of "" and SvCUR of 0, which is what we want. */
2670 STRLEN maxlen = 0;
2671 const char *max = SvPV_const(end, maxlen);
5d9574c1 2672 if (UNLIKELY(SvNIOK(cur) || SvCUR(cur) > maxlen))
8a1f10dd 2673 goto retno;
b552b52c
DM
2674
2675 oldsv = *itersvp;
6d3ca00e
DM
2676 /* NB: on the first iteration, oldsv will have a ref count of at
2677 * least 2 (one extra from blk_loop.itersave), so the GV or pad
2678 * slot will get localised; on subsequent iterations the RC==1
2679 * optimisation may kick in and the SV will be reused. */
2680 if (oldsv && LIKELY(SvREFCNT(oldsv) == 1 && !SvMAGICAL(oldsv))) {
b552b52c
DM
2681 /* safe to reuse old SV */
2682 sv_setsv(oldsv, cur);
a48ce6be 2683 }
b552b52c
DM
2684 else
2685 {
2686 /* we need a fresh SV every time so that loop body sees a
2687 * completely new SV for closures/references to work as
2688 * they used to */
2689 *itersvp = newSVsv(cur);
6d3ca00e 2690 SvREFCNT_dec(oldsv);
b552b52c
DM
2691 }
2692 if (strEQ(SvPVX_const(cur), max))
2693 sv_setiv(cur, 0); /* terminate next time */
2694 else
2695 sv_inc(cur);
2696 break;
2697 }
a48ce6be 2698
fcef60b4
DM
2699 case CXt_LOOP_LAZYIV: /* integer increment */
2700 {
2701 IV cur = cx->blk_loop.state_u.lazyiv.cur;
5d9574c1 2702 if (UNLIKELY(cur > cx->blk_loop.state_u.lazyiv.end))
8a1f10dd 2703 goto retno;
7f61b687 2704
fcef60b4 2705 oldsv = *itersvp;
6d3ca00e
DM
2706 /* see NB comment above */
2707 if (oldsv && LIKELY(SvREFCNT(oldsv) == 1 && !SvMAGICAL(oldsv))) {
eaa5c2d6 2708 /* safe to reuse old SV */
47b96a1e
DM
2709
2710 if ( (SvFLAGS(oldsv) & (SVTYPEMASK|SVf_THINKFIRST|SVf_IVisUV))
2711 == SVt_IV)
2712 {
2713 /* Cheap SvIOK_only().
2714 * Assert that flags which SvIOK_only() would test or
2715 * clear can't be set, because we're SVt_IV */
2716 assert(!(SvFLAGS(oldsv) &
2717 (SVf_OOK|SVf_UTF8|(SVf_OK & ~(SVf_IOK|SVp_IOK)))));
2718 SvFLAGS(oldsv) |= (SVf_IOK|SVp_IOK);
2719 /* SvIV_set() where sv_any points to head */
2720 oldsv->sv_u.svu_iv = cur;
2721
2722 }
2723 else
2724 sv_setiv(oldsv, cur);
eaa5c2d6 2725 }
1c846c1f 2726 else
eaa5c2d6
GA
2727 {
2728 /* we need a fresh SV every time so that loop body sees a
2729 * completely new SV for closures/references to work as they
2730 * used to */
fcef60b4 2731 *itersvp = newSViv(cur);
6d3ca00e 2732 SvREFCNT_dec(oldsv);
eaa5c2d6 2733 }
a2309040 2734
5d9574c1 2735 if (UNLIKELY(cur == IV_MAX)) {
cdc1aa42
NC
2736 /* Handle end of range at IV_MAX */
2737 cx->blk_loop.state_u.lazyiv.end = IV_MIN;
2738 } else
2739 ++cx->blk_loop.state_u.lazyiv.cur;
a48ce6be 2740 break;
fcef60b4 2741 }
a48ce6be 2742
93661e56
DM
2743 case CXt_LOOP_LIST: /* for (1,2,3) */
2744
2745 assert(OPpITER_REVERSED == 2); /* so inc becomes -1 or 1 */
2746 inc = 1 - (PL_op->op_private & OPpITER_REVERSED);
2747 ix = (cx->blk_loop.state_u.stack.ix += inc);
2748 if (UNLIKELY(inc > 0
2749 ? ix > cx->blk_oldsp
2750 : ix <= cx->blk_loop.state_u.stack.basesp)
2751 )
8a1f10dd 2752 goto retno;
93661e56
DM
2753
2754 sv = PL_stack_base[ix];
2755 av = NULL;
2756 goto loop_ary_common;
2757
2758 case CXt_LOOP_ARY: /* for (@ary) */
2759
2760 av = cx->blk_loop.state_u.ary.ary;
2761 inc = 1 - (PL_op->op_private & OPpITER_REVERSED);
2762 ix = (cx->blk_loop.state_u.ary.ix += inc);
2763 if (UNLIKELY(inc > 0
2764 ? ix > AvFILL(av)
2765 : ix < 0)
2766 )
8a1f10dd 2767 goto retno;
de080daa 2768
9d1ee8e0 2769 if (UNLIKELY(SvRMAGICAL(av))) {
a8a20bb6
DM
2770 SV * const * const svp = av_fetch(av, ix, FALSE);
2771 sv = svp ? *svp : NULL;
2772 }
2773 else {
2774 sv = AvARRAY(av)[ix];
de080daa 2775 }
ef3e5ea9 2776
93661e56
DM
2777 loop_ary_common:
2778
d39c26a6
FC
2779 if (UNLIKELY(cx->cx_type & CXp_FOR_LVREF)) {
2780 SvSetMagicSV(*itersvp, sv);
2781 break;
2782 }
2783
5d9574c1
DM
2784 if (LIKELY(sv)) {
2785 if (UNLIKELY(SvIS_FREED(sv))) {
f38aa882
DM
2786 *itersvp = NULL;
2787 Perl_croak(aTHX_ "Use of freed value in iteration");
2788 }
60779a30 2789 if (SvPADTMP(sv)) {
8e079c2a 2790 sv = newSVsv(sv);
60779a30 2791 }
8e079c2a
FC
2792 else {
2793 SvTEMP_off(sv);
2794 SvREFCNT_inc_simple_void_NN(sv);
2795 }
de080daa 2796 }
93661e56 2797 else if (av) {
199f858d 2798 sv = newSVavdefelem(av, ix, 0);
de080daa 2799 }
a600f7e6
FC
2800 else
2801 sv = &PL_sv_undef;
a0d0e21e 2802
de080daa
DM
2803 oldsv = *itersvp;
2804 *itersvp = sv;
2805 SvREFCNT_dec(oldsv);
de080daa 2806 break;
a48ce6be
DM
2807
2808 default:
2809 DIE(aTHX_ "panic: pp_iter, type=%u", CxTYPE(cx));
2810 }
8a1f10dd
DM
2811
2812 retsv = &PL_sv_yes;
2813 if (0) {
2814 retno:
2815 retsv = &PL_sv_no;
2816 }
2817 /* pp_enteriter should have pre-extended the stack */
2818 assert(PL_stack_sp < PL_stack_max);
2819 *++PL_stack_sp =retsv;
2820
2821 return PL_op->op_next;
a0d0e21e
LW
2822}
2823
ef07e810
DM
2824/*
2825A description of how taint works in pattern matching and substitution.
2826
284167a5
SM
2827This is all conditional on NO_TAINT_SUPPORT not being defined. Under
2828NO_TAINT_SUPPORT, taint-related operations should become no-ops.
2829
4e19c54b 2830While the pattern is being assembled/concatenated and then compiled,
284167a5
SM
2831PL_tainted will get set (via TAINT_set) if any component of the pattern
2832is tainted, e.g. /.*$tainted/. At the end of pattern compilation,
2833the RXf_TAINTED flag is set on the pattern if PL_tainted is set (via
1738e041
DM
2834TAINT_get). It will also be set if any component of the pattern matches
2835based on locale-dependent behavior.
ef07e810 2836
0ab462a6
DM
2837When the pattern is copied, e.g. $r = qr/..../, the SV holding the ref to
2838the pattern is marked as tainted. This means that subsequent usage, such
284167a5
SM
2839as /x$r/, will set PL_tainted using TAINT_set, and thus RXf_TAINTED,
2840on the new pattern too.
ef07e810 2841
272d35c9 2842RXf_TAINTED_SEEN is used post-execution by the get magic code
ef07e810
DM
2843of $1 et al to indicate whether the returned value should be tainted.
2844It is the responsibility of the caller of the pattern (i.e. pp_match,
2845pp_subst etc) to set this flag for any other circumstances where $1 needs
2846to be tainted.
2847
2848The taint behaviour of pp_subst (and pp_substcont) is quite complex.
2849
2850There are three possible sources of taint
2851 * the source string
2852 * the pattern (both compile- and run-time, RXf_TAINTED / RXf_TAINTED_SEEN)
2853 * the replacement string (or expression under /e)
2854
2855There are four destinations of taint and they are affected by the sources
2856according to the rules below:
2857
2858 * the return value (not including /r):
2859 tainted by the source string and pattern, but only for the
2860 number-of-iterations case; boolean returns aren't tainted;
2861 * the modified string (or modified copy under /r):
2862 tainted by the source string, pattern, and replacement strings;
2863 * $1 et al:
2864 tainted by the pattern, and under 'use re "taint"', by the source
2865 string too;
2866 * PL_taint - i.e. whether subsequent code (e.g. in a /e block) is tainted:
2867 should always be unset before executing subsequent code.
2868
2869The overall action of pp_subst is:
2870
2871 * at the start, set bits in rxtainted indicating the taint status of
2872 the various sources.
2873
2874 * After each pattern execution, update the SUBST_TAINT_PAT bit in
2875 rxtainted if RXf_TAINTED_SEEN has been set, to indicate that the
2876 pattern has subsequently become tainted via locale ops.
2877
2878 * If control is being passed to pp_substcont to execute a /e block,
2879 save rxtainted in the CXt_SUBST block, for future use by
2880 pp_substcont.
2881
2882 * Whenever control is being returned to perl code (either by falling
2883 off the "end" of pp_subst/pp_substcont, or by entering a /e block),
2884 use the flag bits in rxtainted to make all the appropriate types of
0ab462a6
DM
2885 destination taint visible; e.g. set RXf_TAINTED_SEEN so that $1
2886 et al will appear tainted.
ef07e810
DM
2887
2888pp_match is just a simpler version of the above.
2889
2890*/
2891
a0d0e21e
LW
2892PP(pp_subst)
2893{
20b7effb 2894 dSP; dTARG;
eb578fdb 2895 PMOP *pm = cPMOP;
a0d0e21e 2896 PMOP *rpm = pm;
eb578fdb 2897 char *s;
a0d0e21e 2898 char *strend;
5c144d81 2899 const char *c;
a0d0e21e 2900 STRLEN clen;
3c6ef0a5
FC
2901 SSize_t iters = 0;
2902 SSize_t maxiters;
a0d0e21e 2903 bool once;
ef07e810
DM
2904 U8 rxtainted = 0; /* holds various SUBST_TAINT_* flag bits.
2905 See "how taint works" above */
a0d0e21e 2906 char *orig;
1ed74d04 2907 U8 r_flags;
eb578fdb 2908 REGEXP *rx = PM_GETRE(pm);
a0d0e21e
LW
2909 STRLEN len;
2910 int force_on_match = 0;
0bcc34c2 2911 const I32 oldsave = PL_savestack_ix;
792b2c16 2912 STRLEN slen;
26a74523 2913 bool doutf8 = FALSE; /* whether replacement is in utf8 */
db2c6cb3 2914#ifdef PERL_ANY_COW
106d9a13 2915 bool was_cow;
ed252734 2916#endif
a0714e2c 2917 SV *nsv = NULL;
b770e143 2918 /* known replacement string? */
eb578fdb 2919 SV *dstr = (pm->op_pmflags & PMf_CONST) ? POPs : NULL;
a0d0e21e 2920
f410a211
NC
2921 PERL_ASYNC_CHECK();
2922
533c011a 2923 if (PL_op->op_flags & OPf_STACKED)
a0d0e21e 2924 TARG = POPs;
6ffceeb7 2925 else if (ARGTARG)
59f00321 2926 GETTARGET;
a0d0e21e 2927 else {
54b9620d 2928 TARG = DEFSV;
a0d0e21e 2929 EXTEND(SP,1);
1c846c1f 2930 }
d9f424b2 2931
64534138 2932 SvGETMAGIC(TARG); /* must come before cow check */
db2c6cb3 2933#ifdef PERL_ANY_COW
106d9a13
DM
2934 /* note that a string might get converted to COW during matching */
2935 was_cow = cBOOL(SvIsCOW(TARG));
ed252734 2936#endif
d13a5d3b
TC
2937 if (!(rpm->op_pmflags & PMf_NONDESTRUCT)) {
2938#ifndef PERL_ANY_COW
2939 if (SvIsCOW(TARG))
2940 sv_force_normal_flags(TARG,0);
2941#endif
2942 if ((SvREADONLY(TARG)
2943 || ( ((SvTYPE(TARG) == SVt_PVGV && isGV_with_GP(TARG))
2944 || SvTYPE(TARG) > SVt_PVLV)
2945 && !(SvTYPE(TARG) == SVt_PVGV && SvFAKE(TARG)))))
2946 Perl_croak_no_modify();
2947 }
8ec5e241
NIS
2948 PUTBACK;
2949
6ac6605d
DM
2950 orig = SvPV_nomg(TARG, len);
2951 /* note we don't (yet) force the var into being a string; if we fail
92711104 2952 * to match, we leave as-is; on successful match however, we *will*
6ac6605d 2953 * coerce into a string, then repeat the match */
4499db73 2954 if (!SvPOKp(TARG) || SvTYPE(TARG) == SVt_PVGV || SvVOK(TARG))
a0d0e21e 2955 force_on_match = 1;
20be6587
DM
2956
2957 /* only replace once? */
2958 once = !(rpm->op_pmflags & PMf_GLOBAL);
2959
ef07e810 2960 /* See "how taint works" above */
284167a5 2961 if (TAINTING_get) {
20be6587
DM
2962 rxtainted = (
2963 (SvTAINTED(TARG) ? SUBST_TAINT_STR : 0)
284167a5 2964 | (RX_ISTAINTED(rx) ? SUBST_TAINT_PAT : 0)
20be6587
DM
2965 | ((pm->op_pmflags & PMf_RETAINT) ? SUBST_TAINT_RETAINT : 0)
2966 | ((once && !(rpm->op_pmflags & PMf_NONDESTRUCT))
2967 ? SUBST_TAINT_BOOLRET : 0));
2968 TAINT_NOT;
2969 }
a12c0f56 2970
a0d0e21e 2971 force_it:
6ac6605d
DM
2972 if (!pm || !orig)
2973 DIE(aTHX_ "panic: pp_subst, pm=%p, orig=%p", pm, orig);
a0d0e21e 2974
6ac6605d
DM
2975 strend = orig + len;
2976 slen = DO_UTF8(TARG) ? utf8_length((U8*)orig, (U8*)strend) : len;
792b2c16
JH
2977 maxiters = 2 * slen + 10; /* We can match twice at each
2978 position, once with zero-length,
2979 second time with non-zero. */
a0d0e21e 2980
6a97c51d 2981 if (!RX_PRELEN(rx) && PL_curpm
8d919b0a 2982 && !ReANY(rx)->mother_re) {
3280af22 2983 pm = PL_curpm;
aaa362c4 2984 rx = PM_GETRE(pm);
a0d0e21e 2985 }
6502e081 2986
6e240d0b 2987#ifdef PERL_SAWAMPERSAND
6502e081
DM
2988 r_flags = ( RX_NPARENS(rx)
2989 || PL_sawampersand
6502e081 2990 || (RX_EXTFLAGS(rx) & (RXf_EVAL_SEEN|RXf_PMf_KEEPCOPY))
5b0e71e9 2991 || (rpm->op_pmflags & PMf_KEEPCOPY)
6502e081
DM
2992 )
2993 ? REXEC_COPY_STR
2994 : 0;
6e240d0b
FC
2995#else
2996 r_flags = REXEC_COPY_STR;
2997#endif
7fba1cd6 2998
0395280b 2999 if (!CALLREGEXEC(rx, orig, strend, orig, 0, TARG, NULL, r_flags))
8b64c330 3000 {
5e79dfb9
DM
3001 SPAGAIN;
3002 PUSHs(rpm->op_pmflags & PMf_NONDESTRUCT ? TARG : &PL_sv_no);
3003 LEAVE_SCOPE(oldsave);
3004 RETURN;
3005 }
1754320d
FC
3006 PL_curpm = pm;
3007
71be2cbc 3008 /* known replacement string? */
f272994b 3009 if (dstr) {
8514a05a
JH
3010 /* replacement needing upgrading? */
3011 if (DO_UTF8(TARG) && !doutf8) {
db79b45b 3012 nsv = sv_newmortal();
4a176938 3013 SvSetSV(nsv, dstr);
8df0e7a2 3014 sv_utf8_upgrade(nsv);
5c144d81 3015 c = SvPV_const(nsv, clen);
4a176938
JH
3016 doutf8 = TRUE;
3017 }
3018 else {
5c144d81 3019 c = SvPV_const(dstr, clen);
4a176938 3020 doutf8 = DO_UTF8(dstr);
8514a05a 3021 }
bb933b9b
FC
3022
3023 if (SvTAINTED(dstr))
3024 rxtainted |= SUBST_TAINT_REPL;
f272994b
A
3025 }
3026 else {
6136c704 3027 c = NULL;
f272994b
A
3028 doutf8 = FALSE;
3029 }
3030
71be2cbc 3031 /* can do inplace substitution? */
ed252734 3032 if (c
db2c6cb3 3033#ifdef PERL_ANY_COW
106d9a13 3034 && !was_cow
ed252734 3035#endif
fbfb1899 3036 && (I32)clen <= RX_MINLENRET(rx)
9cefd268
FC
3037 && ( once
3038 || !(r_flags & REXEC_COPY_STR)
3039 || (!SvGMAGICAL(dstr) && !(RX_EXTFLAGS(rx) & RXf_EVAL_SEEN))
3040 )
dbc200c5 3041 && !(RX_EXTFLAGS(rx) & RXf_NO_INPLACE_SUBST)
8ca8a454
NC
3042 && (!doutf8 || SvUTF8(TARG))
3043 && !(rpm->op_pmflags & PMf_NONDESTRUCT))
8b030b38 3044 {
ec911639 3045
db2c6cb3 3046#ifdef PERL_ANY_COW
106d9a13 3047 /* string might have got converted to COW since we set was_cow */
ed252734 3048 if (SvIsCOW(TARG)) {
f7a8268c 3049 if (!force_on_match)
ed252734 3050 goto have_a_cow;
f7a8268c 3051 assert(SvVOK(TARG));
ed252734
NC
3052 }
3053#endif
71be2cbc 3054 if (force_on_match) {
6ac6605d
DM
3055 /* redo the first match, this time with the orig var
3056 * forced into being a string */
71be2cbc 3057 force_on_match = 0;
6ac6605d 3058 orig = SvPV_force_nomg(TARG, len);
71be2cbc
PP
3059 goto force_it;
3060 }
39b40493 3061
71be2cbc 3062 if (once) {
c67ab8f2 3063 char *d, *m;
20be6587
DM
3064 if (RX_MATCH_TAINTED(rx)) /* run time pattern taint, eg locale */
3065 rxtainted |= SUBST_TAINT_PAT;
07bc277f
NC
3066 m = orig + RX_OFFS(rx)[0].start;
3067 d = orig + RX_OFFS(rx)[0].end;
71be2cbc
PP
3068 s = orig;
3069 if (m - s > strend - d) { /* faster to shorten from end */
2ec7214c 3070 I32 i;
71be2cbc
PP
3071 if (clen) {
3072 Copy(c, m, clen, char);
3073 m += clen;
a0d0e21e 3074 }
71be2cbc
PP
3075 i = strend - d;
3076 if (i > 0) {
3077 Move(d, m, i, char);
3078 m += i;
a0d0e21e 3079 }
71be2cbc
PP
3080 *m = '\0';
3081 SvCUR_set(TARG, m - s);
3082 }
2ec7214c
DM
3083 else { /* faster from front */
3084 I32 i = m - s;
71be2cbc 3085 d -= clen;
2ec7214c
DM
3086 if (i > 0)
3087 Move(s, d - i, i, char);
71be2cbc 3088 sv_chop(TARG, d-i);
71be2cbc 3089 if (clen)
c947cd8d 3090 Copy(c, d, clen, char);
71be2cbc 3091 }
8ec5e241 3092 SPAGAIN;
8ca8a454 3093 PUSHs(&PL_sv_yes);
71be2cbc
PP
3094 }
3095 else {
c67ab8f2 3096 char *d, *m;
0395280b 3097 d = s = RX_OFFS(rx)[0].start + orig;
71be2cbc 3098 do {
2b25edcf 3099 I32 i;
5d9574c1 3100 if (UNLIKELY(iters++ > maxiters))
cea2e8a9 3101 DIE(aTHX_ "Substitution loop");
5d9574c1 3102 if (UNLIKELY(RX_MATCH_TAINTED(rx))) /* run time pattern taint, eg locale */
20be6587 3103 rxtainted |= SUBST_TAINT_PAT;
07bc277f 3104 m = RX_OFFS(rx)[0].start + orig;
155aba94 3105 if ((i = m - s)) {
71be2cbc
PP
3106 if (s != d)
3107 Move(s, d, i, char);
3108 d += i;
a0d0e21e 3109 }
71be2cbc
PP
3110 if (clen) {
3111 Copy(c, d, clen, char);
3112 d += clen;
3113 }
07bc277f 3114 s = RX_OFFS(rx)[0].end + orig;
7ce41e5c
FC
3115 } while (CALLREGEXEC(rx, s, strend, orig,
3116 s == m, /* don't match same null twice */
f722798b 3117 TARG, NULL,
d5e7783a 3118 REXEC_NOT_FIRST|REXEC_IGNOREPOS|REXEC_FAIL_ON_UNDERFLOW));
71be2cbc 3119 if (s != d) {
2b25edcf 3120 I32 i = strend - s;
aa07b2f6 3121 SvCUR_set(TARG, d - SvPVX_const(TARG) + i);
71be2cbc 3122 Move(s, d, i+1, char); /* include the NUL */
a0d0e21e 3123 }
8ec5e241 3124 SPAGAIN;
3c6ef0a5 3125 mPUSHi(iters);
a0d0e21e
LW
3126 }
3127 }
ff6e92e8 3128 else {
1754320d 3129 bool first;
c67ab8f2 3130 char *m;
1754320d 3131 SV *repl;
a0d0e21e 3132 if (force_on_match) {
6ac6605d
DM
3133 /* redo the first match, this time with the orig var
3134 * forced into being a string */
a0d0e21e 3135 force_on_match = 0;
0c1438a1
NC
3136 if (rpm->op_pmflags & PMf_NONDESTRUCT) {
3137 /* I feel that it should be possible to avoid this mortal copy
3138 given that the code below copies into a new destination.
3139 However, I suspect it isn't worth the complexity of
3140 unravelling the C<goto force_it> for the small number of
3141 cases where it would be viable to drop into the copy code. */
3142 TARG = sv_2mortal(newSVsv(TARG));
3143 }
6ac6605d 3144 orig = SvPV_force_nomg(TARG, len);
a0d0e21e
LW
3145 goto force_it;
3146 }
db2c6cb3 3147#ifdef PERL_ANY_COW
ed252734
NC
3148 have_a_cow:
3149#endif
20be6587
DM
3150 if (RX_MATCH_TAINTED(rx)) /* run time pattern taint, eg locale */
3151 rxtainted |= SUBST_TAINT_PAT;
1754320d 3152 repl = dstr;
0395280b
DM
3153 s = RX_OFFS(rx)[0].start + orig;
3154 dstr = newSVpvn_flags(orig, s-orig,
3155 SVs_TEMP | (DO_UTF8(TARG) ? SVf_UTF8 : 0));
a0d0e21e 3156 if (!c) {
eb578fdb 3157 PERL_CONTEXT *cx;
8ec5e241 3158 SPAGAIN;
0395280b 3159 m = orig;
20be6587
DM
3160 /* note that a whole bunch of local vars are saved here for
3161 * use by pp_substcont: here's a list of them in case you're
3162 * searching for places in this sub that uses a particular var:
3163 * iters maxiters r_flags oldsave rxtainted orig dstr targ
3164 * s m strend rx once */
490576d1 3165 CX_PUSHSUBST(cx);
20e98b0f 3166 RETURNOP(cPMOP->op_pmreplrootu.op_pmreplroot);
a0d0e21e 3167 }
1754320d 3168 first = TRUE;
a0d0e21e 3169 do {
5d9574c1 3170 if (UNLIKELY(iters++ > maxiters))
cea2e8a9 3171 DIE(aTHX_ "Substitution loop");
5d9574c1 3172 if (UNLIKELY(RX_MATCH_TAINTED(rx)))
20be6587 3173 rxtainted |= SUBST_TAINT_PAT;
07bc277f 3174 if (RX_MATCH_COPIED(rx) && RX_SUBBEG(rx) != orig) {
c67ab8f2
DM
3175 char *old_s = s;
3176 char *old_orig = orig;
6502e081 3177 assert(RX_SUBOFFSET(rx) == 0);
c67ab8f2 3178
07bc277f 3179 orig = RX_SUBBEG(rx);
c67ab8f2
DM
3180 s = orig + (old_s - old_orig);
3181 strend = s + (strend - old_s);
a0d0e21e 3182 }
07bc277f 3183 m = RX_OFFS(rx)[0].start + orig;
64534138 3184 sv_catpvn_nomg_maybeutf8(dstr, s, m - s, DO_UTF8(TARG));
07bc277f 3185 s = RX_OFFS(rx)[0].end + orig;
1754320d
FC
3186 if (first) {
3187 /* replacement already stringified */
3188 if (clen)
64534138 3189 sv_catpvn_nomg_maybeutf8(dstr, c, clen, doutf8);
1754320d
FC
3190 first = FALSE;
3191 }
3192 else {
8df0e7a2 3193 sv_catsv(dstr, repl);
5d9574c1 3194 if (UNLIKELY(SvTAINTED(repl)))
bb933b9b 3195 rxtainted |= SUBST_TAINT_REPL;
1754320d 3196 }
a0d0e21e
LW
3197 if (once)
3198 break;
ff27773b
KW
3199 } while (CALLREGEXEC(rx, s, strend, orig,
3200 s == m, /* Yields minend of 0 or 1 */
d5e7783a
DM
3201 TARG, NULL,
3202 REXEC_NOT_FIRST|REXEC_IGNOREPOS|REXEC_FAIL_ON_UNDERFLOW));
aba224f7 3203 assert(strend >= s);
64534138 3204 sv_catpvn_nomg_maybeutf8(dstr, s, strend - s, DO_UTF8(TARG));
748a9306 3205
8ca8a454
NC
3206 if (rpm->op_pmflags & PMf_NONDESTRUCT) {
3207 /* From here on down we're using the copy, and leaving the original
3208 untouched. */
3209 TARG = dstr;
3210 SPAGAIN;
3211 PUSHs(dstr);
3212 } else {
db2c6cb3 3213#ifdef PERL_ANY_COW
8ca8a454
NC
3214 /* The match may make the string COW. If so, brilliant, because
3215 that's just saved us one malloc, copy and free - the regexp has
3216 donated the old buffer, and we malloc an entirely new one, rather
3217 than the regexp malloc()ing a buffer and copying our original,
3218 only for us to throw it away here during the substitution. */
3219 if (SvIsCOW(TARG)) {
3220 sv_force_normal_flags(TARG, SV_COW_DROP_PV);
3221 } else
ed252734 3222#endif
8ca8a454
NC
3223 {
3224 SvPV_free(TARG);
3225 }
3226 SvPV_set(TARG, SvPVX(dstr));
3227 SvCUR_set(TARG, SvCUR(dstr));
3228 SvLEN_set(TARG, SvLEN(dstr));
64534138 3229 SvFLAGS(TARG) |= SvUTF8(dstr);
8ca8a454 3230 SvPV_set(dstr, NULL);
748a9306 3231
8ca8a454 3232 SPAGAIN;
3c6ef0a5 3233 mPUSHi(iters);
8ca8a454
NC
3234 }
3235 }
3236
3237 if (!(rpm->op_pmflags & PMf_NONDESTRUCT)) {
3238 (void)SvPOK_only_UTF8(TARG);
a0d0e21e 3239 }
20be6587 3240
ef07e810 3241 /* See "how taint works" above */
284167a5 3242 if (TAINTING_get) {
20be6587
DM
3243 if ((rxtainted & SUBST_TAINT_PAT) ||
3244 ((rxtainted & (SUBST_TAINT_STR|SUBST_TAINT_RETAINT)) ==
3245 (SUBST_TAINT_STR|SUBST_TAINT_RETAINT))
3246 )
3247 (RX_MATCH_TAINTED_on(rx)); /* taint $1 et al */
3248
3249 if (!(rxtainted & SUBST_TAINT_BOOLRET)
3250 && (rxtainted & (SUBST_TAINT_STR|SUBST_TAINT_PAT))
3251 )
3252 SvTAINTED_on(TOPs); /* taint return value */
3253 else
3254 SvTAINTED_off(TOPs); /* may have got tainted earlier */
3255
3256 /* needed for mg_set below */
284167a5
SM
3257 TAINT_set(
3258 cBOOL(rxtainted & (SUBST_TAINT_STR|SUBST_TAINT_PAT|SUBST_TAINT_REPL))
3259 );
20be6587
DM
3260 SvTAINT(TARG);
3261 }
3262 SvSETMAGIC(TARG); /* PL_tainted must be correctly set for this mg_set */
3263 TAINT_NOT;
f1a76097
DM
3264 LEAVE_SCOPE(oldsave);
3265 RETURN;
a0d0e21e
LW
3266}
3267
3268PP(pp_grepwhile)
3269{
20b7effb 3270 dSP;
a0d0e21e
LW
3271
3272 if (SvTRUEx(POPs))
3280af22
NIS
3273 PL_stack_base[PL_markstack_ptr[-1]++] = PL_stack_base[*PL_markstack_ptr];
3274 ++*PL_markstack_ptr;
b2a2a901 3275 FREETMPS;
d343c3ef 3276 LEAVE_with_name("grep_item"); /* exit inner scope */
a0d0e21e
LW
3277
3278 /* All done yet? */
5d9574c1 3279 if (UNLIKELY(PL_stack_base + *PL_markstack_ptr > SP)) {
a0d0e21e 3280 I32 items;
1c23e2bd 3281 const U8 gimme = GIMME_V;
a0d0e21e 3282
d343c3ef 3283 LEAVE_with_name("grep"); /* exit outer scope */
a0d0e21e 3284 (void)POPMARK; /* pop src */
3280af22 3285 items = --*PL_markstack_ptr - PL_markstack_ptr[-1];
a0d0e21e 3286 (void)POPMARK; /* pop dst */
3280af22 3287 SP = PL_stack_base + POPMARK; /* pop original mark */
54310121 3288 if (gimme == G_SCALAR) {
7cc47870
RGS
3289 dTARGET;
3290 XPUSHi(items);
a0d0e21e 3291 }
54310121
PP
3292 else if (gimme == G_ARRAY)
3293 SP += items;
a0d0e21e
LW
3294 RETURN;
3295 }
3296 else {
3297 SV *src;
3298
d343c3ef 3299 ENTER_with_name("grep_item"); /* enter inner scope */
1d7c1841 3300 SAVEVPTR(PL_curpm);
a0d0e21e 3301
6cae08a8 3302 src = PL_stack_base[TOPMARK];
60779a30 3303 if (SvPADTMP(src)) {
6cae08a8 3304 src = PL_stack_base[TOPMARK] = sv_mortalcopy(src);
a0ed822e
FC
3305 PL_tmps_floor++;
3306 }
a0d0e21e 3307 SvTEMP_off(src);
ffd49c98 3308 DEFSV_set(src);
a0d0e21e
LW
3309
3310 RETURNOP(cLOGOP->op_other);
3311 }
3312}
3313
799da9d7 3314/* leave_adjust_stacks():
f7a874b8 3315 *
e02ce34b
DM
3316 * Process a scope's return args (in the range from_sp+1 .. PL_stack_sp),
3317 * positioning them at to_sp+1 onwards, and do the equivalent of a
3318 * FREEMPS and TAINT_NOT.
3319 *
f7a874b8
DM
3320 * Not intended to be called in void context.
3321 *
799da9d7
DM
3322 * When leaving a sub, eval, do{} or other scope, the things that need
3323 * doing to process the return args are:
f7a874b8 3324 * * in scalar context, only return the last arg (or PL_sv_undef if none);
799da9d7
DM
3325 * * for the types of return that return copies of their args (such
3326 * as rvalue sub return), make a mortal copy of every return arg,
3327 * except where we can optimise the copy away without it being
3328 * semantically visible;
3329 * * make sure that the arg isn't prematurely freed; in the case of an
3330 * arg not copied, this may involve mortalising it. For example, in
f7a874b8
DM
3331 * C<sub f { my $x = ...; $x }>, $x would be freed when we do
3332 * CX_LEAVE_SCOPE(cx) unless it's protected or copied.
3333 *
799da9d7
DM
3334 * What condition to use when deciding whether to pass the arg through
3335 * or make a copy, is determined by the 'pass' arg; its valid values are:
3336 * 0: rvalue sub/eval exit
3337 * 1: other rvalue scope exit
3338 * 2: :lvalue sub exit in rvalue context
3339 * 3: :lvalue sub exit in lvalue context and other lvalue scope exits
3340 *
f7a874b8 3341 * There is a big issue with doing a FREETMPS. We would like to free any
799da9d7 3342 * temps created by the last statement which the sub executed, rather than
f7a874b8
DM
3343 * leaving them for the caller. In a situation where a sub call isn't
3344 * soon followed by a nextstate (e.g. nested recursive calls, a la
3345 * fibonacci()), temps can accumulate, causing memory and performance
3346 * issues.
3347 *
3348 * On the other hand, we don't want to free any TEMPs which are keeping
799da9d7
DM
3349 * alive any return args that we skipped copying; nor do we wish to undo
3350 * any mortalising done here.
f7a874b8
DM
3351 *
3352 * The solution is to split the temps stack frame into two, with a cut
3353 * point delineating the two halves. We arrange that by the end of this
3354 * function, all the temps stack frame entries we wish to keep are in the
799da9d7 3355 * range PL_tmps_floor+1.. tmps_base-1, while the ones to free now are in
f7a874b8
DM
3356 * the range tmps_base .. PL_tmps_ix. During the course of this
3357 * function, tmps_base starts off as PL_tmps_floor+1, then increases
3358 * whenever we find or create a temp that we know should be kept. In
3359 * general the stuff above tmps_base is undecided until we reach the end,
3360 * and we may need a sort stage for that.
3361 *
3362 * To determine whether a TEMP is keeping a return arg alive, every
3363 * arg that is kept rather than copied and which has the SvTEMP flag
3364 * set, has the flag temporarily unset, to mark it. At the end we scan
799da9d7 3365 * the temps stack frame above the cut for entries without SvTEMP and
f7a874b8 3366 * keep them, while turning SvTEMP on again. Note that if we die before
799da9d7 3367 * the SvTEMPs flags are set again, its safe: at worst, subsequent use of
f7a874b8
DM
3368 * those SVs may be slightly less efficient.
3369 *
3370 * In practice various optimisations for some common cases mean we can
3371 * avoid most of the scanning and swapping about with the temps stack.
3372 */
3373
799da9d7 3374void
1c23e2bd 3375Perl_leave_adjust_stacks(pTHX_ SV **from_sp, SV **to_sp, U8 gimme, int pass)
a0d0e21e 3376{
263e0548 3377 dVAR;
20b7effb 3378 dSP;
f7a874b8
DM
3379 SSize_t tmps_base; /* lowest index into tmps stack that needs freeing now */
3380 SSize_t nargs;
3381
799da9d7
DM
3382 PERL_ARGS_ASSERT_LEAVE_ADJUST_STACKS;
3383
f7a874b8
DM
3384 TAINT_NOT;
3385
3386 if (gimme == G_ARRAY) {
e02ce34b
DM
3387 nargs = SP - from_sp;
3388 from_sp++;
f7a874b8
DM
3389 }
3390 else {
3391 assert(gimme == G_SCALAR);
e02ce34b 3392 if (UNLIKELY(from_sp >= SP)) {
f7a874b8 3393 /* no return args */
e02ce34b 3394 assert(from_sp == SP);
f7a874b8
DM
3395 EXTEND(SP, 1);
3396 *++SP = &PL_sv_undef;
e02ce34b 3397 to_sp = SP;
f7a874b8
DM
3398 nargs = 0;
3399 }
3400 else {
3401 from_sp = SP;
3402 nargs = 1;
3403 }
3404 }
3405
3406 /* common code for G_SCALAR and G_ARRAY */
3407
3408 tmps_base = PL_tmps_floor + 1;
3409
3410 assert(nargs >= 0);
3411 if (nargs) {
3412 /* pointer version of tmps_base. Not safe across temp stack
3413 * reallocs. */
3414 SV **tmps_basep;
3415
3416 EXTEND_MORTAL(nargs); /* one big extend for worst-case scenario */
3417 tmps_basep = PL_tmps_stack + tmps_base;
f7a874b8
DM
3418
3419 /* process each return arg */
3420
3421 do {
3422 SV *sv = *from_sp++;
3423
3424 assert(PL_tmps_ix + nargs < PL_tmps_max);
3645bb38
DM
3425#ifdef DEBUGGING
3426 /* PADTMPs with container set magic shouldn't appear in the
3427 * wild. This assert is more important for pp_leavesublv(),
3428 * but by testing for it here, we're more likely to catch
3429 * bad cases (what with :lvalue subs not being widely
3430 * deployed). The two issues are that for something like
3431 * sub :lvalue { $tied{foo} }
3432 * or
3433 * sub :lvalue { substr($foo,1,2) }
3434 * pp_leavesublv() will croak if the sub returns a PADTMP,
3435 * and currently functions like pp_substr() return a mortal
3436 * rather than using their PADTMP when returning a PVLV.
3437 * This is because the PVLV will hold a ref to $foo,
3438 * so $foo would get delayed in being freed while
3439 * the PADTMP SV remained in the PAD.
3440 * So if this assert fails it means either:
3441 * 1) there is pp code similar to pp_substr that is
3442 * returning a PADTMP instead of a mortal, and probably
3443 * needs fixing, or
5d9c1c9a 3444 * 2) pp_leavesublv is making unwarranted assumptions
3645bb38
DM
3445 * about always croaking on a PADTMP
3446 */
3447 if (SvPADTMP(sv) && SvSMAGICAL(sv)) {
3448 MAGIC *mg;
3449 for (mg = SvMAGIC(sv); mg; mg = mg->mg_moremagic) {
3450 assert(PERL_MAGIC_TYPE_IS_VALUE_MAGIC(mg->mg_type));
3451 }
3452 }
3453#endif
f7a874b8 3454
799da9d7
DM
3455 if (
3456 pass == 0 ? (SvTEMP(sv) && !SvMAGICAL(sv) && SvREFCNT(sv) == 1)
3457 : pass == 1 ? ((SvTEMP(sv) || SvPADTMP(sv)) && !SvMAGICAL(sv) && SvREFCNT(sv) == 1)
3458 : pass == 2 ? (!SvPADTMP(sv))
3459 : 1)
3460 {
3461 /* pass through: skip copy for logic or optimisation
3462 * reasons; instead mortalise it, except that ... */
e02ce34b 3463 *++to_sp = sv;
f7a874b8 3464
799da9d7
DM
3465 if (SvTEMP(sv)) {
3466 /* ... since this SV is an SvTEMP , we don't need to
3467 * re-mortalise it; instead we just need to ensure
3468 * that its existing entry in the temps stack frame
3469 * ends up below the cut and so avoids being freed
3470 * this time round. We mark it as needing to be kept
3471 * by temporarily unsetting SvTEMP; then at the end,
3472 * we shuffle any !SvTEMP entries on the tmps stack
3473 * back below the cut.
3474 * However, there's a significant chance that there's
3475 * a 1:1 correspondence between the first few (or all)
3476 * elements in the return args stack frame and those
3477 * in the temps stack frame; e,g.:
3478 * sub f { ....; map {...} .... },
3479 * or if we're exiting multiple scopes and one of the
3480 * inner scopes has already made mortal copies of each
3481 * return arg.
3482 *
3483 * If so, this arg sv will correspond to the next item
3484 * on the tmps stack above the cut, and so can be kept
3485 * merely by moving the cut boundary up one, rather
3486 * than messing with SvTEMP. If all args are 1:1 then
3487 * we can avoid the sorting stage below completely.
977d0c81
DM
3488 *
3489 * If there are no items above the cut on the tmps
3490 * stack, then the SvTEMP must comne from an item
3491 * below the cut, so there's nothing to do.
799da9d7 3492 */
977d0c81
DM
3493 if (tmps_basep <= &PL_tmps_stack[PL_tmps_ix]) {
3494 if (sv == *tmps_basep)
3495 tmps_basep++;
3496 else
3497 SvTEMP_off(sv);
3498 }
799da9d7 3499 }
75bc488d 3500 else if (!SvPADTMP(sv)) {
799da9d7 3501 /* mortalise arg to avoid it being freed during save
75bc488d 3502 * stack unwinding. Pad tmps don't need mortalising as
977d0c81
DM
3503 * they're never freed. This is the equivalent of
3504 * sv_2mortal(SvREFCNT_inc(sv)), except that:
799da9d7
DM
3505 * * it assumes that the temps stack has already been
3506 * extended;
3507 * * it puts the new item at the cut rather than at
3508 * ++PL_tmps_ix, moving the previous occupant there
3509 * instead.
3510 */
3511 if (!SvIMMORTAL(sv)) {
977d0c81 3512 SvREFCNT_inc_simple_void_NN(sv);
799da9d7 3513 SvTEMP_on(sv);
977d0c81
DM
3514 /* Note that if there's nothing above the cut,
3515 * this copies the garbage one slot above
3516 * PL_tmps_ix onto itself. This is harmless (the
3517 * stack's already been extended), but might in
3518 * theory trigger warnings from tools like ASan
3519 */
799da9d7
DM
3520 PL_tmps_stack[++PL_tmps_ix] = *tmps_basep;
3521 *tmps_basep++ = sv;
3522 }
3523 }
f7a874b8
DM
3524 }
3525 else {
3526 /* Make a mortal copy of the SV.
3527 * The following code is the equivalent of sv_mortalcopy()
3528 * except that:
3529 * * it assumes the temps stack has already been extended;
3530 * * it optimises the copying for some simple SV types;
3531 * * it puts the new item at the cut rather than at
3532 * ++PL_tmps_ix, moving the previous occupant there
3533 * instead.
3534 */
3535 SV *newsv = newSV(0);
3536
3537 PL_tmps_stack[++PL_tmps_ix] = *tmps_basep;
3538 /* put it on the tmps stack early so it gets freed if we die */
3539 *tmps_basep++ = newsv;
e02ce34b 3540 *++to_sp = newsv;
f7a874b8
DM
3541
3542 if (SvTYPE(sv) <= SVt_IV) {
3543 /* arg must be one of undef, IV/UV, or RV: skip
3544 * sv_setsv_flags() and do the copy directly */
3545 U32 dstflags;
3546 U32 srcflags = SvFLAGS(sv);
3547
3548 assert(!SvGMAGICAL(sv));
3549 if (srcflags & (SVf_IOK|SVf_ROK)) {
3550 SET_SVANY_FOR_BODYLESS_IV(newsv);
3551
3552 if (srcflags & SVf_ROK) {
3553 newsv->sv_u.svu_rv = SvREFCNT_inc(SvRV(sv));
3554 /* SV type plus flags */
3555 dstflags = (SVt_IV|SVf_ROK|SVs_TEMP);
3556 }
3557 else {
3558 /* both src and dst are <= SVt_IV, so sv_any
3559 * points to the head; so access the heads
3560 * directly rather than going via sv_any.
3561 */
3562 assert( &(sv->sv_u.svu_iv)
3563 == &(((XPVIV*) SvANY(sv))->xiv_iv));
3564 assert( &(newsv->sv_u.svu_iv)
3565 == &(((XPVIV*) SvANY(newsv))->xiv_iv));
3566 newsv->sv_u.svu_iv = sv->sv_u.svu_iv;
3567 /* SV type plus flags */
3568 dstflags = (SVt_IV|SVf_IOK|SVp_IOK|SVs_TEMP
3569 |(srcflags & SVf_IVisUV));
3570 }
3571 }
3572 else {
3573 assert(!(srcflags & SVf_OK));
3574 dstflags = (SVt_NULL|SVs_TEMP); /* SV type plus flags */
3575 }
3576 SvFLAGS(newsv) = dstflags;
3577
3578 }
3579 else {
3580 /* do the full sv_setsv() */
3581 SSize_t old_base;
3582
3583 SvTEMP_on(newsv);
3584 old_base = tmps_basep - PL_tmps_stack;
3585 SvGETMAGIC(sv);
3586 sv_setsv_flags(newsv, sv, SV_DO_COW_SVSETSV);
799da9d7 3587 /* the mg_get or sv_setsv might have created new temps
f7a874b8
DM
3588 * or realloced the tmps stack; regrow and reload */
3589 EXTEND_MORTAL(nargs);
3590 tmps_basep = PL_tmps_stack + old_base;
3591 TAINT_NOT; /* Each item is independent */
3592 }
3593
3594 }
3595 } while (--nargs);
3596
3597 /* If there are any temps left above the cut, we need to sort
3598 * them into those to keep and those to free. The only ones to
3599 * keep are those for which we've temporarily unset SvTEMP.
36