This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Storable.xs: Add comments to store_scalar concerning utf8
[perl5.git] / dist / Storable / Storable.xs
CommitLineData
7a6a85bf 1/*
6ad89a2f 2 * Store and retrieve mechanism.
7a6a85bf
RG
3 *
4 * Copyright (c) 1995-2000, Raphael Manfredi
5 *
9e21b3d0
JH
6 * You may redistribute only under the same terms as Perl 5, as specified
7 * in the README file that comes with the distribution.
7a6a85bf 8 *
7a6a85bf
RG
9 */
10
138ec36d 11#define PERL_NO_GET_CONTEXT /* we want efficiency */
7a6a85bf
RG
12#include <EXTERN.h>
13#include <perl.h>
a3bf621f
JH
14#include <XSUB.h>
15
eadddfac 16#ifndef PATCHLEVEL
fac63a07 17#include <patchlevel.h> /* Perl's one, needed since 5.6 */
069d7f71 18#endif
7a6a85bf 19
9bdaae78 20#if !defined(PERL_VERSION) || PERL_VERSION < 10 || (PERL_VERSION == 10 && PERL_SUBVERSION < 1)
80f4f327
NC
21#define NEED_load_module
22#define NEED_vload_module
9111ec31 23#define NEED_newCONSTSUB
afce0a13 24#define NEED_newSVpvn_flags
6797988b 25#define NEED_newRV_noinc
3f575d8d
NC
26#include "ppport.h" /* handle old perls */
27#endif
28
e8189732 29#if 0
9e21b3d0
JH
30#define DEBUGME /* Debug mode, turns assertions on as well */
31#define DASSERT /* Assertion mode */
32#endif
7a6a85bf
RG
33
34/*
35 * Pre PerlIO time when none of USE_PERLIO and PERLIO_IS_STDIO is defined
36 * Provide them with the necessary defines so they can build with pre-5.004.
37 */
38#ifndef USE_PERLIO
39#ifndef PERLIO_IS_STDIO
40#define PerlIO FILE
41#define PerlIO_getc(x) getc(x)
42#define PerlIO_putc(f,x) putc(x,f)
43#define PerlIO_read(x,y,z) fread(y,1,z,x)
44#define PerlIO_write(x,y,z) fwrite(y,1,z,x)
45#define PerlIO_stdoutf printf
46#endif /* PERLIO_IS_STDIO */
47#endif /* USE_PERLIO */
48
49/*
50 * Earlier versions of perl might be used, we can't assume they have the latest!
51 */
f0ffaed8 52
7a6a85bf
RG
53#ifndef HvSHAREKEYS_off
54#define HvSHAREKEYS_off(hv) /* Ignore */
55#endif
cc964657 56
e88d61e9
FC
57/* perl <= 5.8.2 needs this */
58#ifndef SvIsCOW
59# define SvIsCOW(sv) 0
60#endif
61
bfcb3514 62#ifndef HvRITER_set
0bb78401 63# define HvRITER_set(hv,r) (HvRITER(hv) = r)
bfcb3514
NC
64#endif
65#ifndef HvEITER_set
0bb78401 66# define HvEITER_set(hv,r) (HvEITER(hv) = r)
bfcb3514
NC
67#endif
68
69#ifndef HvRITER_get
70# define HvRITER_get HvRITER
71#endif
72#ifndef HvEITER_get
73# define HvEITER_get HvEITER
74#endif
75
ca732855
NC
76#ifndef HvPLACEHOLDERS_get
77# define HvPLACEHOLDERS_get HvPLACEHOLDERS
78#endif
79
1c4fe6e3
NC
80#ifndef HvTOTALKEYS
81# define HvTOTALKEYS(hv) HvKEYS(hv)
82#endif
83
7a6a85bf 84#ifdef DEBUGME
8be2b38b
JH
85
86#ifndef DASSERT
87#define DASSERT
88#endif
89
90826881
JH
90/*
91 * TRACEME() will only output things when the $Storable::DEBUGME is true.
92 */
93
111e03c1
RG
94#define TRACEME(x) \
95 STMT_START { \
3509f647 96 if (SvTRUE(perl_get_sv("Storable::DEBUGME", GV_ADD))) \
111e03c1
RG
97 { PerlIO_stdoutf x; PerlIO_stdoutf("\n"); } \
98 } STMT_END
7a6a85bf
RG
99#else
100#define TRACEME(x)
8be2b38b 101#endif /* DEBUGME */
7a6a85bf
RG
102
103#ifdef DASSERT
111e03c1
RG
104#define ASSERT(x,y) \
105 STMT_START { \
7a6a85bf
RG
106 if (!(x)) { \
107 PerlIO_stdoutf("ASSERT FAILED (\"%s\", line %d): ", \
108 __FILE__, __LINE__); \
109 PerlIO_stdoutf y; PerlIO_stdoutf("\n"); \
110 } \
111e03c1 111 } STMT_END
7a6a85bf
RG
112#else
113#define ASSERT(x,y)
114#endif
115
116/*
117 * Type markers.
118 */
119
120#define C(x) ((char) (x)) /* For markers with dynamic retrieval handling */
121
122#define SX_OBJECT C(0) /* Already stored object */
dd19458b 123#define SX_LSCALAR C(1) /* Scalar (large binary) follows (length, data) */
c4a6f826 124#define SX_ARRAY C(2) /* Array forthcoming (size, item list) */
7a6a85bf
RG
125#define SX_HASH C(3) /* Hash forthcoming (size, key/value pair list) */
126#define SX_REF C(4) /* Reference to object forthcoming */
127#define SX_UNDEF C(5) /* Undefined scalar */
128#define SX_INTEGER C(6) /* Integer forthcoming */
129#define SX_DOUBLE C(7) /* Double forthcoming */
130#define SX_BYTE C(8) /* (signed) byte forthcoming */
131#define SX_NETINT C(9) /* Integer in network order forthcoming */
dd19458b 132#define SX_SCALAR C(10) /* Scalar (binary, small) follows (length, data) */
f062ea6c
PN
133#define SX_TIED_ARRAY C(11) /* Tied array forthcoming */
134#define SX_TIED_HASH C(12) /* Tied hash forthcoming */
135#define SX_TIED_SCALAR C(13) /* Tied scalar forthcoming */
7a6a85bf
RG
136#define SX_SV_UNDEF C(14) /* Perl's immortal PL_sv_undef */
137#define SX_SV_YES C(15) /* Perl's immortal PL_sv_yes */
138#define SX_SV_NO C(16) /* Perl's immortal PL_sv_no */
139#define SX_BLESS C(17) /* Object is blessed */
140#define SX_IX_BLESS C(18) /* Object is blessed, classname given by index */
141#define SX_HOOK C(19) /* Stored via hook, user-defined */
142#define SX_OVERLOAD C(20) /* Overloaded reference */
f062ea6c
PN
143#define SX_TIED_KEY C(21) /* Tied magic key forthcoming */
144#define SX_TIED_IDX C(22) /* Tied magic index forthcoming */
145#define SX_UTF8STR C(23) /* UTF-8 string forthcoming (small) */
146#define SX_LUTF8STR C(24) /* UTF-8 string forthcoming (large) */
147#define SX_FLAG_HASH C(25) /* Hash with flags forthcoming (size, flags, key/flags/value triplet list) */
464b080a 148#define SX_CODE C(26) /* Code references as perl source code */
c3c53033
NC
149#define SX_WEAKREF C(27) /* Weak reference to object forthcoming */
150#define SX_WEAKOVERLOAD C(28) /* Overloaded weak reference */
151#define SX_ERROR C(29) /* Error */
7a6a85bf
RG
152
153/*
154 * Those are only used to retrieve "old" pre-0.6 binary images.
155 */
156#define SX_ITEM 'i' /* An array item introducer */
157#define SX_IT_UNDEF 'I' /* Undefined array item */
d1be9408
JF
158#define SX_KEY 'k' /* A hash key introducer */
159#define SX_VALUE 'v' /* A hash value introducer */
7a6a85bf
RG
160#define SX_VL_UNDEF 'V' /* Undefined hash value */
161
162/*
163 * Those are only used to retrieve "old" pre-0.7 binary images
164 */
165
166#define SX_CLASS 'b' /* Object is blessed, class name length <255 */
f062ea6c 167#define SX_LG_CLASS 'B' /* Object is blessed, class name length >255 */
7a6a85bf
RG
168#define SX_STORED 'X' /* End of object */
169
170/*
171 * Limits between short/long length representation.
172 */
173
174#define LG_SCALAR 255 /* Large scalar length limit */
175#define LG_BLESS 127 /* Large classname bless limit */
176
177/*
178 * Operation types
179 */
180
181#define ST_STORE 0x1 /* Store operation */
182#define ST_RETRIEVE 0x2 /* Retrieval operation */
183#define ST_CLONE 0x4 /* Deep cloning operation */
184
185/*
186 * The following structure is used for hash table key retrieval. Since, when
187 * retrieving objects, we'll be facing blessed hash references, it's best
188 * to pre-allocate that buffer once and resize it as the need arises, never
189 * freeing it (keys will be saved away someplace else anyway, so even large
190 * keys are not enough a motivation to reclaim that space).
191 *
192 * This structure is also used for memory store/retrieve operations which
c4a6f826 193 * happen in a fixed place before being malloc'ed elsewhere if persistence
7a6a85bf
RG
194 * is required. Hence the aptr pointer.
195 */
196struct extendable {
197 char *arena; /* Will hold hash key strings, resized as needed */
c4a6f826 198 STRLEN asiz; /* Size of aforementioned buffer */
7a6a85bf
RG
199 char *aptr; /* Arena pointer, for in-place read/write ops */
200 char *aend; /* First invalid address */
201};
202
203/*
204 * At store time:
d1be9408 205 * A hash table records the objects which have already been stored.
7a6a85bf
RG
206 * Those are referred to as SX_OBJECT in the file, and their "tag" (i.e.
207 * an arbitrary sequence number) is used to identify them.
208 *
209 * At retrieve time:
210 * An array table records the objects which have already been retrieved,
c4a6f826 211 * as seen by the tag determined by counting the objects themselves. The
7a6a85bf
RG
212 * reference to that retrieved object is kept in the table, and is returned
213 * when an SX_OBJECT is found bearing that same tag.
214 *
215 * The same processing is used to record "classname" for blessed objects:
216 * indexing by a hash at store time, and via an array at retrieve time.
217 */
218
219typedef unsigned long stag_t; /* Used by pre-0.6 binary format */
220
221/*
222 * The following "thread-safe" related defines were contributed by
223 * Murray Nesbitt <murray@activestate.com> and integrated by RAM, who
224 * only renamed things a little bit to ensure consistency with surrounding
225 * code. -- RAM, 14/09/1999
226 *
227 * The original patch suffered from the fact that the stcxt_t structure
228 * was global. Murray tried to minimize the impact on the code as much as
229 * possible.
230 *
231 * Starting with 0.7, Storable can be re-entrant, via the STORABLE_xxx hooks
232 * on objects. Therefore, the notion of context needs to be generalized,
233 * threading or not.
234 */
235
236#define MY_VERSION "Storable(" XS_VERSION ")"
237
530b72ba
NC
238
239/*
240 * Conditional UTF8 support.
241 *
242 */
243#ifdef SvUTF8_on
244#define STORE_UTF8STR(pv, len) STORE_PV_LEN(pv, len, SX_UTF8STR, SX_LUTF8STR)
245#define HAS_UTF8_SCALARS
246#ifdef HeKUTF8
247#define HAS_UTF8_HASHES
248#define HAS_UTF8_ALL
249#else
250/* 5.6 perl has utf8 scalars but not hashes */
251#endif
252#else
253#define SvUTF8(sv) 0
254#define STORE_UTF8STR(pv, len) CROAK(("panic: storing UTF8 in non-UTF8 perl"))
255#endif
256#ifndef HAS_UTF8_ALL
257#define UTF8_CROAK() CROAK(("Cannot retrieve UTF8 data in non-UTF8 perl"))
258#endif
c3c53033
NC
259#ifndef SvWEAKREF
260#define WEAKREF_CROAK() CROAK(("Cannot retrieve weak references in this perl"))
261#endif
530b72ba
NC
262
263#ifdef HvPLACEHOLDERS
264#define HAS_RESTRICTED_HASHES
265#else
266#define HVhek_PLACEHOLD 0x200
267#define RESTRICTED_HASH_CROAK() CROAK(("Cannot retrieve restricted hash"))
268#endif
269
270#ifdef HvHASKFLAGS
271#define HAS_HASH_KEY_FLAGS
272#endif
273
ab923da1
NC
274#ifdef ptr_table_new
275#define USE_PTR_TABLE
276#endif
277
dd19458b
JH
278/*
279 * Fields s_tainted and s_dirty are prefixed with s_ because Perl's include
280 * files remap tainted and dirty when threading is enabled. That's bad for
281 * perl to remap such common words. -- RAM, 29/09/00
282 */
283
0723351e 284struct stcxt;
7a6a85bf
RG
285typedef struct stcxt {
286 int entry; /* flags recursion */
287 int optype; /* type of traversal operation */
ab923da1
NC
288 /* which objects have been seen, store time.
289 tags are numbers, which are cast to (SV *) and stored directly */
290#ifdef USE_PTR_TABLE
291 /* use pseen if we have ptr_tables. We have to store tag+1, because
292 tag numbers start at 0, and we can't store (SV *) 0 in a ptr_table
293 without it being confused for a fetch lookup failure. */
294 struct ptr_tbl *pseen;
295 /* Still need hseen for the 0.6 file format code. */
296#endif
297 HV *hseen;
e993d95c
JH
298 AV *hook_seen; /* which SVs were returned by STORABLE_freeze() */
299 AV *aseen; /* which objects have been seen, retrieve time */
dfd91409 300 IV where_is_undef; /* index in aseen of PL_sv_undef */
e993d95c
JH
301 HV *hclass; /* which classnames have been seen, store time */
302 AV *aclass; /* which classnames have been seen, retrieve time */
303 HV *hook; /* cache for hook methods per class name */
304 IV tagnum; /* incremented at store time for each seen object */
305 IV classnum; /* incremented at store time for each seen classname */
306 int netorder; /* true if network order used */
307 int s_tainted; /* true if input source is tainted, at retrieve time */
308 int forgive_me; /* whether to be forgiving... */
464b080a
SR
309 int deparse; /* whether to deparse code refs */
310 SV *eval; /* whether to eval source code */
e993d95c 311 int canonical; /* whether to store hashes sorted by key */
530b72ba 312#ifndef HAS_RESTRICTED_HASHES
c4a6f826 313 int derestrict; /* whether to downgrade restricted hashes */
530b72ba
NC
314#endif
315#ifndef HAS_UTF8_ALL
316 int use_bytes; /* whether to bytes-ify utf8 */
317#endif
e8189732 318 int accept_future_minor; /* croak immediately on future minor versions? */
dd19458b 319 int s_dirty; /* context is dirty due to CROAK() -- can be cleaned */
e993d95c
JH
320 int membuf_ro; /* true means membuf is read-only and msaved is rw */
321 struct extendable keybuf; /* for hash key retrieval */
322 struct extendable membuf; /* for memory store/retrieve operations */
323 struct extendable msaved; /* where potentially valid mbuf is saved */
7a6a85bf
RG
324 PerlIO *fio; /* where I/O are performed, NULL for memory */
325 int ver_major; /* major of version for retrieved object */
326 int ver_minor; /* minor of version for retrieved object */
aa07b2f6 327 SV *(**retrieve_vtbl)(pTHX_ struct stcxt *, const char *); /* retrieve dispatch table */
111e03c1
RG
328 SV *prev; /* contexts chained backwards in real recursion */
329 SV *my_sv; /* the blessed scalar who's SvPVX() I am */
51f77169 330 int in_retrieve_overloaded; /* performance hack for retrieving overloaded objects */
7a6a85bf
RG
331} stcxt_t;
332
111e03c1
RG
333#define NEW_STORABLE_CXT_OBJ(cxt) \
334 STMT_START { \
335 SV *self = newSV(sizeof(stcxt_t) - 1); \
336 SV *my_sv = newRV_noinc(self); \
da51bb9b 337 sv_bless(my_sv, gv_stashpv("Storable::Cxt", GV_ADD)); \
111e03c1
RG
338 cxt = (stcxt_t *)SvPVX(self); \
339 Zero(cxt, 1, stcxt_t); \
340 cxt->my_sv = my_sv; \
341 } STMT_END
342
7a6a85bf
RG
343#if defined(MULTIPLICITY) || defined(PERL_OBJECT) || defined(PERL_CAPI)
344
e993d95c 345#if (PATCHLEVEL <= 4) && (SUBVERSION < 68)
7a6a85bf 346#define dSTCXT_SV \
3509f647 347 SV *perinterp_sv = perl_get_sv(MY_VERSION, 0)
7a6a85bf
RG
348#else /* >= perl5.004_68 */
349#define dSTCXT_SV \
350 SV *perinterp_sv = *hv_fetch(PL_modglobal, \
351 MY_VERSION, sizeof(MY_VERSION)-1, TRUE)
352#endif /* < perl5.004_68 */
353
354#define dSTCXT_PTR(T,name) \
111e03c1 355 T name = ((perinterp_sv && SvIOK(perinterp_sv) && SvIVX(perinterp_sv) \
436c6dd3 356 ? (T)SvPVX(SvRV(INT2PTR(SV*,SvIVX(perinterp_sv)))) : (T) 0))
7a6a85bf
RG
357#define dSTCXT \
358 dSTCXT_SV; \
359 dSTCXT_PTR(stcxt_t *, cxt)
360
111e03c1
RG
361#define INIT_STCXT \
362 dSTCXT; \
363 NEW_STORABLE_CXT_OBJ(cxt); \
364 sv_setiv(perinterp_sv, PTR2IV(cxt->my_sv))
7a6a85bf 365
111e03c1
RG
366#define SET_STCXT(x) \
367 STMT_START { \
7a6a85bf 368 dSTCXT_SV; \
111e03c1
RG
369 sv_setiv(perinterp_sv, PTR2IV(x->my_sv)); \
370 } STMT_END
7a6a85bf
RG
371
372#else /* !MULTIPLICITY && !PERL_OBJECT && !PERL_CAPI */
373
85535365 374static stcxt_t *Context_ptr = NULL;
7a6a85bf 375#define dSTCXT stcxt_t *cxt = Context_ptr
85535365 376#define SET_STCXT(x) Context_ptr = x
111e03c1
RG
377#define INIT_STCXT \
378 dSTCXT; \
85535365
RG
379 NEW_STORABLE_CXT_OBJ(cxt); \
380 SET_STCXT(cxt)
111e03c1 381
7a6a85bf
RG
382
383#endif /* MULTIPLICITY || PERL_OBJECT || PERL_CAPI */
384
385/*
386 * KNOWN BUG:
387 * Croaking implies a memory leak, since we don't use setjmp/longjmp
388 * to catch the exit and free memory used during store or retrieve
389 * operations. This is not too difficult to fix, but I need to understand
390 * how Perl does it, and croaking is exceptional anyway, so I lack the
391 * motivation to do it.
392 *
393 * The current workaround is to mark the context as dirty when croaking,
394 * so that data structures can be freed whenever we renter Storable code
395 * (but only *then*: it's a workaround, not a fix).
396 *
397 * This is also imperfect, because we don't really know how far they trapped
398 * the croak(), and when we were recursing, we won't be able to clean anything
399 * but the topmost context stacked.
400 */
401
111e03c1 402#define CROAK(x) STMT_START { cxt->s_dirty = 1; croak x; } STMT_END
7a6a85bf
RG
403
404/*
405 * End of "thread-safe" related definitions.
406 */
407
408/*
9e21b3d0
JH
409 * LOW_32BITS
410 *
411 * Keep only the low 32 bits of a pointer (used for tags, which are not
412 * really pointers).
413 */
414
415#if PTRSIZE <= 4
416#define LOW_32BITS(x) ((I32) (x))
417#else
418#define LOW_32BITS(x) ((I32) ((unsigned long) (x) & 0xffffffffUL))
419#endif
420
421/*
422 * oI, oS, oC
423 *
424 * Hack for Crays, where sizeof(I32) == 8, and which are big-endians.
425 * Used in the WLEN and RLEN macros.
426 */
427
428#if INTSIZE > 4
429#define oI(x) ((I32 *) ((char *) (x) + 4))
430#define oS(x) ((x) - 4)
431#define oC(x) (x = 0)
432#define CRAY_HACK
433#else
434#define oI(x) (x)
435#define oS(x) (x)
436#define oC(x)
437#endif
438
439/*
7a6a85bf
RG
440 * key buffer handling
441 */
442#define kbuf (cxt->keybuf).arena
443#define ksiz (cxt->keybuf).asiz
111e03c1
RG
444#define KBUFINIT() \
445 STMT_START { \
7a6a85bf
RG
446 if (!kbuf) { \
447 TRACEME(("** allocating kbuf of 128 bytes")); \
448 New(10003, kbuf, 128, char); \
449 ksiz = 128; \
450 } \
111e03c1
RG
451 } STMT_END
452#define KBUFCHK(x) \
453 STMT_START { \
7a6a85bf 454 if (x >= ksiz) { \
e993d95c 455 TRACEME(("** extending kbuf to %d bytes (had %d)", x+1, ksiz)); \
7a6a85bf
RG
456 Renew(kbuf, x+1, char); \
457 ksiz = x+1; \
458 } \
111e03c1 459 } STMT_END
7a6a85bf
RG
460
461/*
462 * memory buffer handling
463 */
464#define mbase (cxt->membuf).arena
465#define msiz (cxt->membuf).asiz
466#define mptr (cxt->membuf).aptr
467#define mend (cxt->membuf).aend
468
469#define MGROW (1 << 13)
470#define MMASK (MGROW - 1)
471
472#define round_mgrow(x) \
473 ((unsigned long) (((unsigned long) (x) + MMASK) & ~MMASK))
474#define trunc_int(x) \
475 ((unsigned long) ((unsigned long) (x) & ~(sizeof(int)-1)))
476#define int_aligned(x) \
477 ((unsigned long) (x) == trunc_int(x))
478
111e03c1
RG
479#define MBUF_INIT(x) \
480 STMT_START { \
7a6a85bf
RG
481 if (!mbase) { \
482 TRACEME(("** allocating mbase of %d bytes", MGROW)); \
483 New(10003, mbase, MGROW, char); \
2cc1b180 484 msiz = (STRLEN)MGROW; \
7a6a85bf
RG
485 } \
486 mptr = mbase; \
487 if (x) \
488 mend = mbase + x; \
489 else \
490 mend = mbase + msiz; \
111e03c1 491 } STMT_END
7a6a85bf
RG
492
493#define MBUF_TRUNC(x) mptr = mbase + x
494#define MBUF_SIZE() (mptr - mbase)
495
496/*
e993d95c
JH
497 * MBUF_SAVE_AND_LOAD
498 * MBUF_RESTORE
499 *
500 * Those macros are used in do_retrieve() to save the current memory
501 * buffer into cxt->msaved, before MBUF_LOAD() can be used to retrieve
502 * data from a string.
503 */
111e03c1
RG
504#define MBUF_SAVE_AND_LOAD(in) \
505 STMT_START { \
e993d95c
JH
506 ASSERT(!cxt->membuf_ro, ("mbase not already saved")); \
507 cxt->membuf_ro = 1; \
508 TRACEME(("saving mbuf")); \
509 StructCopy(&cxt->membuf, &cxt->msaved, struct extendable); \
510 MBUF_LOAD(in); \
111e03c1 511 } STMT_END
e993d95c 512
111e03c1
RG
513#define MBUF_RESTORE() \
514 STMT_START { \
e993d95c
JH
515 ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
516 cxt->membuf_ro = 0; \
517 TRACEME(("restoring mbuf")); \
518 StructCopy(&cxt->msaved, &cxt->membuf, struct extendable); \
111e03c1 519 } STMT_END
e993d95c
JH
520
521/*
7a6a85bf
RG
522 * Use SvPOKp(), because SvPOK() fails on tainted scalars.
523 * See store_scalar() for other usage of this workaround.
524 */
111e03c1
RG
525#define MBUF_LOAD(v) \
526 STMT_START { \
e993d95c 527 ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
7a6a85bf
RG
528 if (!SvPOKp(v)) \
529 CROAK(("Not a scalar string")); \
530 mptr = mbase = SvPV(v, msiz); \
531 mend = mbase + msiz; \
111e03c1 532 } STMT_END
7a6a85bf 533
111e03c1
RG
534#define MBUF_XTEND(x) \
535 STMT_START { \
7a6a85bf
RG
536 int nsz = (int) round_mgrow((x)+msiz); \
537 int offset = mptr - mbase; \
e993d95c
JH
538 ASSERT(!cxt->membuf_ro, ("mbase is not read-only")); \
539 TRACEME(("** extending mbase from %d to %d bytes (wants %d new)", \
540 msiz, nsz, (x))); \
7a6a85bf
RG
541 Renew(mbase, nsz, char); \
542 msiz = nsz; \
543 mptr = mbase + offset; \
544 mend = mbase + nsz; \
111e03c1 545 } STMT_END
7a6a85bf 546
111e03c1
RG
547#define MBUF_CHK(x) \
548 STMT_START { \
7a6a85bf
RG
549 if ((mptr + (x)) > mend) \
550 MBUF_XTEND(x); \
111e03c1 551 } STMT_END
7a6a85bf 552
111e03c1
RG
553#define MBUF_GETC(x) \
554 STMT_START { \
7a6a85bf
RG
555 if (mptr < mend) \
556 x = (int) (unsigned char) *mptr++; \
557 else \
558 return (SV *) 0; \
111e03c1 559 } STMT_END
7a6a85bf 560
9e21b3d0 561#ifdef CRAY_HACK
111e03c1
RG
562#define MBUF_GETINT(x) \
563 STMT_START { \
9e21b3d0
JH
564 oC(x); \
565 if ((mptr + 4) <= mend) { \
566 memcpy(oI(&x), mptr, 4); \
567 mptr += 4; \
568 } else \
569 return (SV *) 0; \
111e03c1 570 } STMT_END
9e21b3d0 571#else
111e03c1
RG
572#define MBUF_GETINT(x) \
573 STMT_START { \
7a6a85bf
RG
574 if ((mptr + sizeof(int)) <= mend) { \
575 if (int_aligned(mptr)) \
576 x = *(int *) mptr; \
577 else \
578 memcpy(&x, mptr, sizeof(int)); \
579 mptr += sizeof(int); \
580 } else \
581 return (SV *) 0; \
111e03c1 582 } STMT_END
9e21b3d0 583#endif
7a6a85bf 584
111e03c1
RG
585#define MBUF_READ(x,s) \
586 STMT_START { \
7a6a85bf
RG
587 if ((mptr + (s)) <= mend) { \
588 memcpy(x, mptr, s); \
589 mptr += s; \
590 } else \
591 return (SV *) 0; \
111e03c1 592 } STMT_END
7a6a85bf 593
111e03c1
RG
594#define MBUF_SAFEREAD(x,s,z) \
595 STMT_START { \
7a6a85bf
RG
596 if ((mptr + (s)) <= mend) { \
597 memcpy(x, mptr, s); \
598 mptr += s; \
599 } else { \
600 sv_free(z); \
601 return (SV *) 0; \
602 } \
111e03c1 603 } STMT_END
7a6a85bf 604
dd57a815
NC
605#define MBUF_SAFEPVREAD(x,s,z) \
606 STMT_START { \
607 if ((mptr + (s)) <= mend) { \
608 memcpy(x, mptr, s); \
609 mptr += s; \
610 } else { \
611 Safefree(z); \
612 return (SV *) 0; \
613 } \
614 } STMT_END
615
111e03c1
RG
616#define MBUF_PUTC(c) \
617 STMT_START { \
7a6a85bf
RG
618 if (mptr < mend) \
619 *mptr++ = (char) c; \
620 else { \
621 MBUF_XTEND(1); \
622 *mptr++ = (char) c; \
623 } \
111e03c1 624 } STMT_END
7a6a85bf 625
9e21b3d0 626#ifdef CRAY_HACK
111e03c1
RG
627#define MBUF_PUTINT(i) \
628 STMT_START { \
9e21b3d0
JH
629 MBUF_CHK(4); \
630 memcpy(mptr, oI(&i), 4); \
631 mptr += 4; \
111e03c1 632 } STMT_END
9e21b3d0 633#else
111e03c1
RG
634#define MBUF_PUTINT(i) \
635 STMT_START { \
7a6a85bf
RG
636 MBUF_CHK(sizeof(int)); \
637 if (int_aligned(mptr)) \
638 *(int *) mptr = i; \
639 else \
640 memcpy(mptr, &i, sizeof(int)); \
641 mptr += sizeof(int); \
111e03c1 642 } STMT_END
9e21b3d0 643#endif
7a6a85bf 644
111e03c1
RG
645#define MBUF_WRITE(x,s) \
646 STMT_START { \
7a6a85bf
RG
647 MBUF_CHK(s); \
648 memcpy(mptr, x, s); \
649 mptr += s; \
111e03c1 650 } STMT_END
7a6a85bf
RG
651
652/*
7a6a85bf
RG
653 * Possible return values for sv_type().
654 */
655
656#define svis_REF 0
657#define svis_SCALAR 1
658#define svis_ARRAY 2
659#define svis_HASH 3
660#define svis_TIED 4
661#define svis_TIED_ITEM 5
464b080a
SR
662#define svis_CODE 6
663#define svis_OTHER 7
7a6a85bf
RG
664
665/*
666 * Flags for SX_HOOK.
667 */
668
669#define SHF_TYPE_MASK 0x03
670#define SHF_LARGE_CLASSLEN 0x04
671#define SHF_LARGE_STRLEN 0x08
672#define SHF_LARGE_LISTLEN 0x10
673#define SHF_IDX_CLASSNAME 0x20
674#define SHF_NEED_RECURSE 0x40
675#define SHF_HAS_LIST 0x80
676
677/*
b12202d0 678 * Types for SX_HOOK (last 2 bits in flags).
7a6a85bf
RG
679 */
680
681#define SHT_SCALAR 0
682#define SHT_ARRAY 1
683#define SHT_HASH 2
b12202d0
JH
684#define SHT_EXTRA 3 /* Read extra byte for type */
685
686/*
687 * The following are held in the "extra byte"...
688 */
689
690#define SHT_TSCALAR 4 /* 4 + 0 -- tied scalar */
691#define SHT_TARRAY 5 /* 4 + 1 -- tied array */
692#define SHT_THASH 6 /* 4 + 2 -- tied hash */
7a6a85bf
RG
693
694/*
e16e2ff8
NC
695 * per hash flags for flagged hashes
696 */
697
698#define SHV_RESTRICTED 0x01
699
700/*
701 * per key flags for flagged hashes
702 */
703
704#define SHV_K_UTF8 0x01
705#define SHV_K_WASUTF8 0x02
706#define SHV_K_LOCKED 0x04
707#define SHV_K_ISSV 0x08
708#define SHV_K_PLACEHOLDER 0x10
709
710/*
7a6a85bf
RG
711 * Before 0.6, the magic string was "perl-store" (binary version number 0).
712 *
713 * Since 0.6 introduced many binary incompatibilities, the magic string has
714 * been changed to "pst0" to allow an old image to be properly retrieved by
715 * a newer Storable, but ensure a newer image cannot be retrieved with an
716 * older version.
717 *
718 * At 0.7, objects are given the ability to serialize themselves, and the
719 * set of markers is extended, backward compatibility is not jeopardized,
720 * so the binary version number could have remained unchanged. To correctly
721 * spot errors if a file making use of 0.7-specific extensions is given to
722 * 0.6 for retrieval, the binary version was moved to "2". And I'm introducing
723 * a "minor" version, to better track this kind of evolution from now on.
724 *
725 */
2aeb6432
NC
726static const char old_magicstr[] = "perl-store"; /* Magic number before 0.6 */
727static const char magicstr[] = "pst0"; /* Used as a magic number */
7a6a85bf 728
2aeb6432
NC
729#define MAGICSTR_BYTES 'p','s','t','0'
730#define OLDMAGICSTR_BYTES 'p','e','r','l','-','s','t','o','r','e'
731
ee0f7aac
NC
732/* 5.6.x introduced the ability to have IVs as long long.
733 However, Configure still defined BYTEORDER based on the size of a long.
734 Storable uses the BYTEORDER value as part of the header, but doesn't
c4a6f826 735 explicitly store sizeof(IV) anywhere in the header. Hence on 5.6.x built
ee0f7aac
NC
736 with IV as long long on a platform that uses Configure (ie most things
737 except VMS and Windows) headers are identical for the different IV sizes,
738 despite the files containing some fields based on sizeof(IV)
739 Erk. Broken-ness.
c4a6f826 740 5.8 is consistent - the following redefinition kludge is only needed on
ee0f7aac
NC
741 5.6.x, but the interwork is needed on 5.8 while data survives in files
742 with the 5.6 header.
743
744*/
745
746#if defined (IVSIZE) && (IVSIZE == 8) && (LONGSIZE == 4)
747#ifndef NO_56_INTERWORK_KLUDGE
748#define USE_56_INTERWORK_KLUDGE
749#endif
750#if BYTEORDER == 0x1234
751#undef BYTEORDER
752#define BYTEORDER 0x12345678
753#else
754#if BYTEORDER == 0x4321
755#undef BYTEORDER
756#define BYTEORDER 0x87654321
757#endif
758#endif
759#endif
760
2aeb6432
NC
761#if BYTEORDER == 0x1234
762#define BYTEORDER_BYTES '1','2','3','4'
763#else
764#if BYTEORDER == 0x12345678
765#define BYTEORDER_BYTES '1','2','3','4','5','6','7','8'
ee0f7aac
NC
766#ifdef USE_56_INTERWORK_KLUDGE
767#define BYTEORDER_BYTES_56 '1','2','3','4'
768#endif
2aeb6432
NC
769#else
770#if BYTEORDER == 0x87654321
771#define BYTEORDER_BYTES '8','7','6','5','4','3','2','1'
ee0f7aac
NC
772#ifdef USE_56_INTERWORK_KLUDGE
773#define BYTEORDER_BYTES_56 '4','3','2','1'
774#endif
2aeb6432
NC
775#else
776#if BYTEORDER == 0x4321
777#define BYTEORDER_BYTES '4','3','2','1'
778#else
c597ea9d 779#error Unknown byteorder. Please append your byteorder to Storable.xs
2aeb6432
NC
780#endif
781#endif
782#endif
783#endif
784
785static const char byteorderstr[] = {BYTEORDER_BYTES, 0};
ee0f7aac
NC
786#ifdef USE_56_INTERWORK_KLUDGE
787static const char byteorderstr_56[] = {BYTEORDER_BYTES_56, 0};
788#endif
530b72ba 789
e16e2ff8 790#define STORABLE_BIN_MAJOR 2 /* Binary major "version" */
be7c46f2 791#define STORABLE_BIN_MINOR 8 /* Binary minor "version" */
530b72ba 792
c3c53033 793#if (PATCHLEVEL <= 5)
530b72ba 794#define STORABLE_BIN_WRITE_MINOR 4
e16e2ff8 795#else
c3c53033
NC
796/*
797 * Perl 5.6.0 onwards can do weak references.
e16e2ff8 798*/
be7c46f2 799#define STORABLE_BIN_WRITE_MINOR 8
c3c53033 800#endif /* (PATCHLEVEL <= 5) */
7a6a85bf 801
e9822705 802#if (PATCHLEVEL < 8 || (PATCHLEVEL == 8 && SUBVERSION < 1))
fcaa57e7
AMS
803#define PL_sv_placeholder PL_sv_undef
804#endif
805
7a6a85bf
RG
806/*
807 * Useful store shortcuts...
808 */
809
a8b7ef86
AMS
810/*
811 * Note that if you put more than one mark for storing a particular
812 * type of thing, *and* in the retrieve_foo() function you mark both
813 * the thingy's you get off with SEEN(), you *must* increase the
814 * tagnum with cxt->tagnum++ along with this macro!
815 * - samv 20Jan04
816 */
111e03c1
RG
817#define PUTMARK(x) \
818 STMT_START { \
7a6a85bf
RG
819 if (!cxt->fio) \
820 MBUF_PUTC(x); \
821 else if (PerlIO_putc(cxt->fio, x) == EOF) \
822 return -1; \
111e03c1 823 } STMT_END
7a6a85bf 824
111e03c1
RG
825#define WRITE_I32(x) \
826 STMT_START { \
9e21b3d0
JH
827 ASSERT(sizeof(x) == sizeof(I32), ("writing an I32")); \
828 if (!cxt->fio) \
829 MBUF_PUTINT(x); \
830 else if (PerlIO_write(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
831 return -1; \
111e03c1 832 } STMT_END
9e21b3d0 833
7a6a85bf 834#ifdef HAS_HTONL
111e03c1
RG
835#define WLEN(x) \
836 STMT_START { \
7a6a85bf
RG
837 if (cxt->netorder) { \
838 int y = (int) htonl(x); \
839 if (!cxt->fio) \
840 MBUF_PUTINT(y); \
9e21b3d0 841 else if (PerlIO_write(cxt->fio,oI(&y),oS(sizeof(y))) != oS(sizeof(y))) \
7a6a85bf
RG
842 return -1; \
843 } else { \
844 if (!cxt->fio) \
845 MBUF_PUTINT(x); \
9e21b3d0 846 else if (PerlIO_write(cxt->fio,oI(&x),oS(sizeof(x))) != oS(sizeof(x))) \
7a6a85bf
RG
847 return -1; \
848 } \
111e03c1 849 } STMT_END
7a6a85bf 850#else
9e21b3d0 851#define WLEN(x) WRITE_I32(x)
7a6a85bf
RG
852#endif
853
111e03c1
RG
854#define WRITE(x,y) \
855 STMT_START { \
7a6a85bf
RG
856 if (!cxt->fio) \
857 MBUF_WRITE(x,y); \
858 else if (PerlIO_write(cxt->fio, x, y) != y) \
859 return -1; \
111e03c1 860 } STMT_END
7a6a85bf 861
111e03c1
RG
862#define STORE_PV_LEN(pv, len, small, large) \
863 STMT_START { \
7a6a85bf
RG
864 if (len <= LG_SCALAR) { \
865 unsigned char clen = (unsigned char) len; \
dd19458b 866 PUTMARK(small); \
7a6a85bf
RG
867 PUTMARK(clen); \
868 if (len) \
869 WRITE(pv, len); \
870 } else { \
dd19458b 871 PUTMARK(large); \
7a6a85bf
RG
872 WLEN(len); \
873 WRITE(pv, len); \
874 } \
111e03c1 875 } STMT_END
7a6a85bf 876
dd19458b
JH
877#define STORE_SCALAR(pv, len) STORE_PV_LEN(pv, len, SX_SCALAR, SX_LSCALAR)
878
879/*
20bb3f55 880 * Store &PL_sv_undef in arrays without recursing through store().
7a6a85bf 881 */
20bb3f55 882#define STORE_SV_UNDEF() \
111e03c1 883 STMT_START { \
7a6a85bf 884 cxt->tagnum++; \
20bb3f55 885 PUTMARK(SX_SV_UNDEF); \
111e03c1 886 } STMT_END
7a6a85bf
RG
887
888/*
889 * Useful retrieve shortcuts...
890 */
891
892#define GETCHAR() \
893 (cxt->fio ? PerlIO_getc(cxt->fio) : (mptr >= mend ? EOF : (int) *mptr++))
894
111e03c1
RG
895#define GETMARK(x) \
896 STMT_START { \
7a6a85bf
RG
897 if (!cxt->fio) \
898 MBUF_GETC(x); \
76df4757 899 else if ((int) (x = PerlIO_getc(cxt->fio)) == EOF) \
7a6a85bf 900 return (SV *) 0; \
111e03c1 901 } STMT_END
7a6a85bf 902
111e03c1
RG
903#define READ_I32(x) \
904 STMT_START { \
9e21b3d0
JH
905 ASSERT(sizeof(x) == sizeof(I32), ("reading an I32")); \
906 oC(x); \
7a6a85bf
RG
907 if (!cxt->fio) \
908 MBUF_GETINT(x); \
9e21b3d0 909 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
7a6a85bf 910 return (SV *) 0; \
111e03c1 911 } STMT_END
9e21b3d0
JH
912
913#ifdef HAS_NTOHL
111e03c1
RG
914#define RLEN(x) \
915 STMT_START { \
9e21b3d0 916 oC(x); \
7a6a85bf
RG
917 if (!cxt->fio) \
918 MBUF_GETINT(x); \
9e21b3d0 919 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
7a6a85bf 920 return (SV *) 0; \
9e21b3d0
JH
921 if (cxt->netorder) \
922 x = (int) ntohl(x); \
111e03c1 923 } STMT_END
9e21b3d0
JH
924#else
925#define RLEN(x) READ_I32(x)
7a6a85bf
RG
926#endif
927
111e03c1
RG
928#define READ(x,y) \
929 STMT_START { \
7a6a85bf
RG
930 if (!cxt->fio) \
931 MBUF_READ(x, y); \
932 else if (PerlIO_read(cxt->fio, x, y) != y) \
933 return (SV *) 0; \
111e03c1 934 } STMT_END
7a6a85bf 935
111e03c1
RG
936#define SAFEREAD(x,y,z) \
937 STMT_START { \
7a6a85bf
RG
938 if (!cxt->fio) \
939 MBUF_SAFEREAD(x,y,z); \
940 else if (PerlIO_read(cxt->fio, x, y) != y) { \
941 sv_free(z); \
942 return (SV *) 0; \
943 } \
111e03c1 944 } STMT_END
7a6a85bf 945
dd57a815
NC
946#define SAFEPVREAD(x,y,z) \
947 STMT_START { \
948 if (!cxt->fio) \
949 MBUF_SAFEPVREAD(x,y,z); \
950 else if (PerlIO_read(cxt->fio, x, y) != y) { \
951 Safefree(z); \
952 return (SV *) 0; \
953 } \
954 } STMT_END
955
7a6a85bf
RG
956/*
957 * This macro is used at retrieve time, to remember where object 'y', bearing a
958 * given tag 'tagnum', has been retrieved. Next time we see an SX_OBJECT marker,
959 * we'll therefore know where it has been retrieved and will be able to
960 * share the same reference, as in the original stored memory image.
b12202d0
JH
961 *
962 * We also need to bless objects ASAP for hooks (which may compute "ref $x"
963 * on the objects given to STORABLE_thaw and expect that to be defined), and
964 * also for overloaded objects (for which we might not find the stash if the
965 * object is not blessed yet--this might occur for overloaded objects that
966 * refer to themselves indirectly: if we blessed upon return from a sub
967 * retrieve(), the SX_OBJECT marker we'd found could not have overloading
968 * restored on it because the underlying object would not be blessed yet!).
969 *
970 * To achieve that, the class name of the last retrieved object is passed down
971 * recursively, and the first SEEN() call for which the class name is not NULL
972 * will bless the object.
dfd91409
NC
973 *
974 * i should be true iff sv is immortal (ie PL_sv_yes, PL_sv_no or PL_sv_undef)
7a6a85bf 975 */
dfd91409 976#define SEEN(y,c,i) \
111e03c1 977 STMT_START { \
7a6a85bf
RG
978 if (!y) \
979 return (SV *) 0; \
dfd91409 980 if (av_store(cxt->aseen, cxt->tagnum++, i ? (SV*)(y) : SvREFCNT_inc(y)) == 0) \
7a6a85bf 981 return (SV *) 0; \
43d061fe 982 TRACEME(("aseen(#%d) = 0x%"UVxf" (refcnt=%d)", cxt->tagnum-1, \
b12202d0
JH
983 PTR2UV(y), SvREFCNT(y)-1)); \
984 if (c) \
985 BLESS((SV *) (y), c); \
111e03c1 986 } STMT_END
7a6a85bf
RG
987
988/*
6dfee1ec 989 * Bless 's' in 'p', via a temporary reference, required by sv_bless().
51f77169
AMS
990 * "A" magic is added before the sv_bless for overloaded classes, this avoids
991 * an expensive call to S_reset_amagic in sv_bless.
7a6a85bf 992 */
111e03c1
RG
993#define BLESS(s,p) \
994 STMT_START { \
7a6a85bf
RG
995 SV *ref; \
996 HV *stash; \
43d061fe 997 TRACEME(("blessing 0x%"UVxf" in %s", PTR2UV(s), (p))); \
da51bb9b 998 stash = gv_stashpv((p), GV_ADD); \
7a6a85bf 999 ref = newRV_noinc(s); \
51f77169
AMS
1000 if (cxt->in_retrieve_overloaded && Gv_AMG(stash)) \
1001 { \
1002 cxt->in_retrieve_overloaded = 0; \
1003 SvAMAGIC_on(ref); \
1004 } \
7a6a85bf 1005 (void) sv_bless(ref, stash); \
b162af07 1006 SvRV_set(ref, NULL); \
7a6a85bf 1007 SvREFCNT_dec(ref); \
111e03c1 1008 } STMT_END
138ec36d 1009/*
1010 * sort (used in store_hash) - conditionally use qsort when
1011 * sortsv is not available ( <= 5.6.1 ).
1012 */
1013
1014#if (PATCHLEVEL <= 6)
1015
1016#if defined(USE_ITHREADS)
1017
1018#define STORE_HASH_SORT \
1019 ENTER; { \
1020 PerlInterpreter *orig_perl = PERL_GET_CONTEXT; \
1021 SAVESPTR(orig_perl); \
1022 PERL_SET_CONTEXT(aTHX); \
1023 qsort((char *) AvARRAY(av), len, sizeof(SV *), sortcmp); \
1024 } LEAVE;
1025
1026#else /* ! USE_ITHREADS */
7a6a85bf 1027
138ec36d 1028#define STORE_HASH_SORT \
1029 qsort((char *) AvARRAY(av), len, sizeof(SV *), sortcmp);
1030
1031#endif /* USE_ITHREADS */
1032
1033#else /* PATCHLEVEL > 6 */
1034
1035#define STORE_HASH_SORT \
1036 sortsv(AvARRAY(av), len, Perl_sv_cmp);
1037
1038#endif /* PATCHLEVEL <= 6 */
1039
1040static int store(pTHX_ stcxt_t *cxt, SV *sv);
aa07b2f6 1041static SV *retrieve(pTHX_ stcxt_t *cxt, const char *cname);
7a6a85bf 1042
ecc6a8ca
IZ
1043#define UNSEE() \
1044 STMT_START { \
1045 av_pop(cxt->aseen); \
1046 cxt->tagnum--; \
1047 } STMT_END
1048
7a6a85bf
RG
1049/*
1050 * Dynamic dispatching table for SV store.
1051 */
1052
138ec36d 1053static int store_ref(pTHX_ stcxt_t *cxt, SV *sv);
1054static int store_scalar(pTHX_ stcxt_t *cxt, SV *sv);
1055static int store_array(pTHX_ stcxt_t *cxt, AV *av);
1056static int store_hash(pTHX_ stcxt_t *cxt, HV *hv);
1057static int store_tied(pTHX_ stcxt_t *cxt, SV *sv);
1058static int store_tied_item(pTHX_ stcxt_t *cxt, SV *sv);
1059static int store_code(pTHX_ stcxt_t *cxt, CV *cv);
1060static int store_other(pTHX_ stcxt_t *cxt, SV *sv);
1061static int store_blessed(pTHX_ stcxt_t *cxt, SV *sv, int type, HV *pkg);
1062
93ad979b
MB
1063typedef int (*sv_store_t)(pTHX_ stcxt_t *cxt, SV *sv);
1064
5c271e25 1065static const sv_store_t sv_store[] = {
93ad979b
MB
1066 (sv_store_t)store_ref, /* svis_REF */
1067 (sv_store_t)store_scalar, /* svis_SCALAR */
1068 (sv_store_t)store_array, /* svis_ARRAY */
1069 (sv_store_t)store_hash, /* svis_HASH */
1070 (sv_store_t)store_tied, /* svis_TIED */
1071 (sv_store_t)store_tied_item, /* svis_TIED_ITEM */
1072 (sv_store_t)store_code, /* svis_CODE */
1073 (sv_store_t)store_other, /* svis_OTHER */
7a6a85bf
RG
1074};
1075
1076#define SV_STORE(x) (*sv_store[x])
1077
1078/*
1079 * Dynamic dispatching tables for SV retrieval.
1080 */
1081
aa07b2f6
SP
1082static SV *retrieve_lscalar(pTHX_ stcxt_t *cxt, const char *cname);
1083static SV *retrieve_lutf8str(pTHX_ stcxt_t *cxt, const char *cname);
1084static SV *old_retrieve_array(pTHX_ stcxt_t *cxt, const char *cname);
1085static SV *old_retrieve_hash(pTHX_ stcxt_t *cxt, const char *cname);
1086static SV *retrieve_ref(pTHX_ stcxt_t *cxt, const char *cname);
1087static SV *retrieve_undef(pTHX_ stcxt_t *cxt, const char *cname);
1088static SV *retrieve_integer(pTHX_ stcxt_t *cxt, const char *cname);
1089static SV *retrieve_double(pTHX_ stcxt_t *cxt, const char *cname);
1090static SV *retrieve_byte(pTHX_ stcxt_t *cxt, const char *cname);
1091static SV *retrieve_netint(pTHX_ stcxt_t *cxt, const char *cname);
1092static SV *retrieve_scalar(pTHX_ stcxt_t *cxt, const char *cname);
1093static SV *retrieve_utf8str(pTHX_ stcxt_t *cxt, const char *cname);
1094static SV *retrieve_tied_array(pTHX_ stcxt_t *cxt, const char *cname);
1095static SV *retrieve_tied_hash(pTHX_ stcxt_t *cxt, const char *cname);
1096static SV *retrieve_tied_scalar(pTHX_ stcxt_t *cxt, const char *cname);
1097static SV *retrieve_other(pTHX_ stcxt_t *cxt, const char *cname);
1098
1099typedef SV* (*sv_retrieve_t)(pTHX_ stcxt_t *cxt, const char *name);
93ad979b
MB
1100
1101static const sv_retrieve_t sv_old_retrieve[] = {
1102 0, /* SX_OBJECT -- entry unused dynamically */
1103 (sv_retrieve_t)retrieve_lscalar, /* SX_LSCALAR */
1104 (sv_retrieve_t)old_retrieve_array, /* SX_ARRAY -- for pre-0.6 binaries */
1105 (sv_retrieve_t)old_retrieve_hash, /* SX_HASH -- for pre-0.6 binaries */
1106 (sv_retrieve_t)retrieve_ref, /* SX_REF */
1107 (sv_retrieve_t)retrieve_undef, /* SX_UNDEF */
1108 (sv_retrieve_t)retrieve_integer, /* SX_INTEGER */
1109 (sv_retrieve_t)retrieve_double, /* SX_DOUBLE */
1110 (sv_retrieve_t)retrieve_byte, /* SX_BYTE */
1111 (sv_retrieve_t)retrieve_netint, /* SX_NETINT */
1112 (sv_retrieve_t)retrieve_scalar, /* SX_SCALAR */
1113 (sv_retrieve_t)retrieve_tied_array, /* SX_ARRAY */
1114 (sv_retrieve_t)retrieve_tied_hash, /* SX_HASH */
1115 (sv_retrieve_t)retrieve_tied_scalar, /* SX_SCALAR */
1116 (sv_retrieve_t)retrieve_other, /* SX_SV_UNDEF not supported */
1117 (sv_retrieve_t)retrieve_other, /* SX_SV_YES not supported */
1118 (sv_retrieve_t)retrieve_other, /* SX_SV_NO not supported */
1119 (sv_retrieve_t)retrieve_other, /* SX_BLESS not supported */
1120 (sv_retrieve_t)retrieve_other, /* SX_IX_BLESS not supported */
1121 (sv_retrieve_t)retrieve_other, /* SX_HOOK not supported */
1122 (sv_retrieve_t)retrieve_other, /* SX_OVERLOADED not supported */
1123 (sv_retrieve_t)retrieve_other, /* SX_TIED_KEY not supported */
1124 (sv_retrieve_t)retrieve_other, /* SX_TIED_IDX not supported */
1125 (sv_retrieve_t)retrieve_other, /* SX_UTF8STR not supported */
1126 (sv_retrieve_t)retrieve_other, /* SX_LUTF8STR not supported */
1127 (sv_retrieve_t)retrieve_other, /* SX_FLAG_HASH not supported */
1128 (sv_retrieve_t)retrieve_other, /* SX_CODE not supported */
1129 (sv_retrieve_t)retrieve_other, /* SX_WEAKREF not supported */
1130 (sv_retrieve_t)retrieve_other, /* SX_WEAKOVERLOAD not supported */
1131 (sv_retrieve_t)retrieve_other, /* SX_ERROR */
7a6a85bf
RG
1132};
1133
aa07b2f6
SP
1134static SV *retrieve_array(pTHX_ stcxt_t *cxt, const char *cname);
1135static SV *retrieve_hash(pTHX_ stcxt_t *cxt, const char *cname);
1136static SV *retrieve_sv_undef(pTHX_ stcxt_t *cxt, const char *cname);
1137static SV *retrieve_sv_yes(pTHX_ stcxt_t *cxt, const char *cname);
1138static SV *retrieve_sv_no(pTHX_ stcxt_t *cxt, const char *cname);
1139static SV *retrieve_blessed(pTHX_ stcxt_t *cxt, const char *cname);
1140static SV *retrieve_idx_blessed(pTHX_ stcxt_t *cxt, const char *cname);
1141static SV *retrieve_hook(pTHX_ stcxt_t *cxt, const char *cname);
1142static SV *retrieve_overloaded(pTHX_ stcxt_t *cxt, const char *cname);
1143static SV *retrieve_tied_key(pTHX_ stcxt_t *cxt, const char *cname);
1144static SV *retrieve_tied_idx(pTHX_ stcxt_t *cxt, const char *cname);
1145static SV *retrieve_flag_hash(pTHX_ stcxt_t *cxt, const char *cname);
1146static SV *retrieve_code(pTHX_ stcxt_t *cxt, const char *cname);
1147static SV *retrieve_weakref(pTHX_ stcxt_t *cxt, const char *cname);
1148static SV *retrieve_weakoverloaded(pTHX_ stcxt_t *cxt, const char *cname);
138ec36d 1149
93ad979b 1150static const sv_retrieve_t sv_retrieve[] = {
7a6a85bf 1151 0, /* SX_OBJECT -- entry unused dynamically */
93ad979b
MB
1152 (sv_retrieve_t)retrieve_lscalar, /* SX_LSCALAR */
1153 (sv_retrieve_t)retrieve_array, /* SX_ARRAY */
1154 (sv_retrieve_t)retrieve_hash, /* SX_HASH */
1155 (sv_retrieve_t)retrieve_ref, /* SX_REF */
1156 (sv_retrieve_t)retrieve_undef, /* SX_UNDEF */
1157 (sv_retrieve_t)retrieve_integer, /* SX_INTEGER */
1158 (sv_retrieve_t)retrieve_double, /* SX_DOUBLE */
1159 (sv_retrieve_t)retrieve_byte, /* SX_BYTE */
1160 (sv_retrieve_t)retrieve_netint, /* SX_NETINT */
1161 (sv_retrieve_t)retrieve_scalar, /* SX_SCALAR */
1162 (sv_retrieve_t)retrieve_tied_array, /* SX_ARRAY */
1163 (sv_retrieve_t)retrieve_tied_hash, /* SX_HASH */
1164 (sv_retrieve_t)retrieve_tied_scalar, /* SX_SCALAR */
1165 (sv_retrieve_t)retrieve_sv_undef, /* SX_SV_UNDEF */
1166 (sv_retrieve_t)retrieve_sv_yes, /* SX_SV_YES */
1167 (sv_retrieve_t)retrieve_sv_no, /* SX_SV_NO */
1168 (sv_retrieve_t)retrieve_blessed, /* SX_BLESS */
1169 (sv_retrieve_t)retrieve_idx_blessed, /* SX_IX_BLESS */
1170 (sv_retrieve_t)retrieve_hook, /* SX_HOOK */
1171 (sv_retrieve_t)retrieve_overloaded, /* SX_OVERLOAD */
1172 (sv_retrieve_t)retrieve_tied_key, /* SX_TIED_KEY */
1173 (sv_retrieve_t)retrieve_tied_idx, /* SX_TIED_IDX */
1174 (sv_retrieve_t)retrieve_utf8str, /* SX_UTF8STR */
1175 (sv_retrieve_t)retrieve_lutf8str, /* SX_LUTF8STR */
1176 (sv_retrieve_t)retrieve_flag_hash, /* SX_HASH */
1177 (sv_retrieve_t)retrieve_code, /* SX_CODE */
1178 (sv_retrieve_t)retrieve_weakref, /* SX_WEAKREF */
1179 (sv_retrieve_t)retrieve_weakoverloaded, /* SX_WEAKOVERLOAD */
1180 (sv_retrieve_t)retrieve_other, /* SX_ERROR */
7a6a85bf
RG
1181};
1182
1183#define RETRIEVE(c,x) (*(c)->retrieve_vtbl[(x) >= SX_ERROR ? SX_ERROR : (x)])
1184
138ec36d 1185static SV *mbuf2sv(pTHX);
7a6a85bf
RG
1186
1187/***
1188 *** Context management.
1189 ***/
1190
1191/*
1192 * init_perinterp
1193 *
1194 * Called once per "thread" (interpreter) to initialize some global context.
1195 */
138ec36d 1196static void init_perinterp(pTHX)
f0ffaed8 1197{
7a6a85bf
RG
1198 INIT_STCXT;
1199
1200 cxt->netorder = 0; /* true if network order used */
1201 cxt->forgive_me = -1; /* whether to be forgiving... */
0b6a08b2 1202 cxt->accept_future_minor = -1; /* would otherwise occur too late */
7a6a85bf
RG
1203}
1204
1205/*
e993d95c
JH
1206 * reset_context
1207 *
1208 * Called at the end of every context cleaning, to perform common reset
1209 * operations.
1210 */
1211static void reset_context(stcxt_t *cxt)
1212{
1213 cxt->entry = 0;
1214 cxt->s_dirty = 0;
1215 cxt->optype &= ~(ST_STORE|ST_RETRIEVE); /* Leave ST_CLONE alone */
1216}
1217
1218/*
7a6a85bf
RG
1219 * init_store_context
1220 *
1221 * Initialize a new store context for real recursion.
1222 */
f0ffaed8 1223static void init_store_context(
138ec36d 1224 pTHX_
f0ffaed8
JH
1225 stcxt_t *cxt,
1226 PerlIO *f,
1227 int optype,
1228 int network_order)
7a6a85bf
RG
1229{
1230 TRACEME(("init_store_context"));
1231
1232 cxt->netorder = network_order;
1233 cxt->forgive_me = -1; /* Fetched from perl if needed */
464b080a
SR
1234 cxt->deparse = -1; /* Idem */
1235 cxt->eval = NULL; /* Idem */
7a6a85bf
RG
1236 cxt->canonical = -1; /* Idem */
1237 cxt->tagnum = -1; /* Reset tag numbers */
1238 cxt->classnum = -1; /* Reset class numbers */
1239 cxt->fio = f; /* Where I/O are performed */
1240 cxt->optype = optype; /* A store, or a deep clone */
1241 cxt->entry = 1; /* No recursion yet */
1242
1243 /*
6dfee1ec 1244 * The 'hseen' table is used to keep track of each SV stored and their
7a6a85bf
RG
1245 * associated tag numbers is special. It is "abused" because the
1246 * values stored are not real SV, just integers cast to (SV *),
1247 * which explains the freeing below.
1248 *
c4a6f826 1249 * It is also one possible bottleneck to achieve good storing speed,
7a6a85bf
RG
1250 * so the "shared keys" optimization is turned off (unlikely to be
1251 * of any use here), and the hash table is "pre-extended". Together,
1252 * those optimizations increase the throughput by 12%.
1253 */
1254
ab923da1
NC
1255#ifdef USE_PTR_TABLE
1256 cxt->pseen = ptr_table_new();
1257 cxt->hseen = 0;
1258#else
7a6a85bf
RG
1259 cxt->hseen = newHV(); /* Table where seen objects are stored */
1260 HvSHAREKEYS_off(cxt->hseen);
ab923da1 1261#endif
7a6a85bf
RG
1262 /*
1263 * The following does not work well with perl5.004_04, and causes
1264 * a core dump later on, in a completely unrelated spot, which
1265 * makes me think there is a memory corruption going on.
1266 *
1267 * Calling hv_ksplit(hseen, HBUCKETS) instead of manually hacking
1268 * it below does not make any difference. It seems to work fine
1269 * with perl5.004_68 but given the probable nature of the bug,
1270 * that does not prove anything.
1271 *
1272 * It's a shame because increasing the amount of buckets raises
1273 * store() throughput by 5%, but until I figure this out, I can't
1274 * allow for this to go into production.
1275 *
1276 * It is reported fixed in 5.005, hence the #if.
1277 */
f0ffaed8 1278#if PERL_VERSION >= 5
7a6a85bf 1279#define HBUCKETS 4096 /* Buckets for %hseen */
ab923da1 1280#ifndef USE_PTR_TABLE
7a6a85bf
RG
1281 HvMAX(cxt->hseen) = HBUCKETS - 1; /* keys %hseen = $HBUCKETS; */
1282#endif
ab923da1 1283#endif
7a6a85bf
RG
1284
1285 /*
6dfee1ec 1286 * The 'hclass' hash uses the same settings as 'hseen' above, but it is
7a6a85bf
RG
1287 * used to assign sequential tags (numbers) to class names for blessed
1288 * objects.
1289 *
1290 * We turn the shared key optimization on.
1291 */
1292
1293 cxt->hclass = newHV(); /* Where seen classnames are stored */
1294
f0ffaed8 1295#if PERL_VERSION >= 5
7a6a85bf
RG
1296 HvMAX(cxt->hclass) = HBUCKETS - 1; /* keys %hclass = $HBUCKETS; */
1297#endif
1298
1299 /*
6dfee1ec 1300 * The 'hook' hash table is used to keep track of the references on
7a6a85bf
RG
1301 * the STORABLE_freeze hook routines, when found in some class name.
1302 *
1303 * It is assumed that the inheritance tree will not be changed during
1304 * storing, and that no new method will be dynamically created by the
1305 * hooks.
1306 */
1307
1308 cxt->hook = newHV(); /* Table where hooks are cached */
90826881
JH
1309
1310 /*
6dfee1ec 1311 * The 'hook_seen' array keeps track of all the SVs returned by
90826881
JH
1312 * STORABLE_freeze hooks for us to serialize, so that they are not
1313 * reclaimed until the end of the serialization process. Each SV is
1314 * only stored once, the first time it is seen.
1315 */
1316
1317 cxt->hook_seen = newAV(); /* Lists SVs returned by STORABLE_freeze */
7a6a85bf
RG
1318}
1319
1320/*
1321 * clean_store_context
1322 *
1323 * Clean store context by
1324 */
138ec36d 1325static void clean_store_context(pTHX_ stcxt_t *cxt)
7a6a85bf
RG
1326{
1327 HE *he;
1328
1329 TRACEME(("clean_store_context"));
1330
1331 ASSERT(cxt->optype & ST_STORE, ("was performing a store()"));
1332
1333 /*
1334 * Insert real values into hashes where we stored faked pointers.
1335 */
1336
ab923da1 1337#ifndef USE_PTR_TABLE
e993d95c
JH
1338 if (cxt->hseen) {
1339 hv_iterinit(cxt->hseen);
1340 while ((he = hv_iternext(cxt->hseen))) /* Extra () for -Wall, grr.. */
da5add9b 1341 HeVAL(he) = &PL_sv_undef;
e993d95c 1342 }
ab923da1 1343#endif
7a6a85bf 1344
e993d95c
JH
1345 if (cxt->hclass) {
1346 hv_iterinit(cxt->hclass);
1347 while ((he = hv_iternext(cxt->hclass))) /* Extra () for -Wall, grr.. */
da5add9b 1348 HeVAL(he) = &PL_sv_undef;
e993d95c 1349 }
7a6a85bf
RG
1350
1351 /*
1352 * And now dispose of them...
862382c7
JH
1353 *
1354 * The surrounding if() protection has been added because there might be
1355 * some cases where this routine is called more than once, during
c4a6f826 1356 * exceptional events. This was reported by Marc Lehmann when Storable
862382c7
JH
1357 * is executed from mod_perl, and the fix was suggested by him.
1358 * -- RAM, 20/12/2000
1359 */
1360
ab923da1
NC
1361#ifdef USE_PTR_TABLE
1362 if (cxt->pseen) {
1363 struct ptr_tbl *pseen = cxt->pseen;
1364 cxt->pseen = 0;
1365 ptr_table_free(pseen);
1366 }
1367 assert(!cxt->hseen);
1368#else
862382c7
JH
1369 if (cxt->hseen) {
1370 HV *hseen = cxt->hseen;
1371 cxt->hseen = 0;
1372 hv_undef(hseen);
1373 sv_free((SV *) hseen);
1374 }
ab923da1 1375#endif
7a6a85bf 1376
862382c7
JH
1377 if (cxt->hclass) {
1378 HV *hclass = cxt->hclass;
1379 cxt->hclass = 0;
1380 hv_undef(hclass);
1381 sv_free((SV *) hclass);
1382 }
7a6a85bf 1383
862382c7
JH
1384 if (cxt->hook) {
1385 HV *hook = cxt->hook;
1386 cxt->hook = 0;
1387 hv_undef(hook);
1388 sv_free((SV *) hook);
1389 }
7a6a85bf 1390
862382c7
JH
1391 if (cxt->hook_seen) {
1392 AV *hook_seen = cxt->hook_seen;
1393 cxt->hook_seen = 0;
1394 av_undef(hook_seen);
1395 sv_free((SV *) hook_seen);
1396 }
90826881 1397
e8189732 1398 cxt->forgive_me = -1; /* Fetched from perl if needed */
464b080a
SR
1399 cxt->deparse = -1; /* Idem */
1400 if (cxt->eval) {
1401 SvREFCNT_dec(cxt->eval);
1402 }
1403 cxt->eval = NULL; /* Idem */
e8189732
NC
1404 cxt->canonical = -1; /* Idem */
1405
e993d95c 1406 reset_context(cxt);
7a6a85bf
RG
1407}
1408
1409/*
1410 * init_retrieve_context
1411 *
1412 * Initialize a new retrieve context for real recursion.
1413 */
138ec36d 1414static void init_retrieve_context(pTHX_ stcxt_t *cxt, int optype, int is_tainted)
7a6a85bf
RG
1415{
1416 TRACEME(("init_retrieve_context"));
1417
1418 /*
1419 * The hook hash table is used to keep track of the references on
1420 * the STORABLE_thaw hook routines, when found in some class name.
1421 *
1422 * It is assumed that the inheritance tree will not be changed during
1423 * storing, and that no new method will be dynamically created by the
1424 * hooks.
1425 */
1426
1427 cxt->hook = newHV(); /* Caches STORABLE_thaw */
1428
ab923da1
NC
1429#ifdef USE_PTR_TABLE
1430 cxt->pseen = 0;
1431#endif
1432
7a6a85bf
RG
1433 /*
1434 * If retrieving an old binary version, the cxt->retrieve_vtbl variable
1435 * was set to sv_old_retrieve. We'll need a hash table to keep track of
c4a6f826 1436 * the correspondence between the tags and the tag number used by the
7a6a85bf
RG
1437 * new retrieve routines.
1438 */
1439
2cc1b180
JH
1440 cxt->hseen = (((void*)cxt->retrieve_vtbl == (void*)sv_old_retrieve)
1441 ? newHV() : 0);
7a6a85bf
RG
1442
1443 cxt->aseen = newAV(); /* Where retrieved objects are kept */
dfd91409 1444 cxt->where_is_undef = -1; /* Special case for PL_sv_undef */
7a6a85bf
RG
1445 cxt->aclass = newAV(); /* Where seen classnames are kept */
1446 cxt->tagnum = 0; /* Have to count objects... */
1447 cxt->classnum = 0; /* ...and class names as well */
1448 cxt->optype = optype;
dd19458b 1449 cxt->s_tainted = is_tainted;
7a6a85bf 1450 cxt->entry = 1; /* No recursion yet */
530b72ba
NC
1451#ifndef HAS_RESTRICTED_HASHES
1452 cxt->derestrict = -1; /* Fetched from perl if needed */
1453#endif
1454#ifndef HAS_UTF8_ALL
1455 cxt->use_bytes = -1; /* Fetched from perl if needed */
1456#endif
e8189732 1457 cxt->accept_future_minor = -1; /* Fetched from perl if needed */
51f77169 1458 cxt->in_retrieve_overloaded = 0;
7a6a85bf
RG
1459}
1460
1461/*
1462 * clean_retrieve_context
1463 *
1464 * Clean retrieve context by
1465 */
138ec36d 1466static void clean_retrieve_context(pTHX_ stcxt_t *cxt)
7a6a85bf
RG
1467{
1468 TRACEME(("clean_retrieve_context"));
1469
1470 ASSERT(cxt->optype & ST_RETRIEVE, ("was performing a retrieve()"));
1471
862382c7
JH
1472 if (cxt->aseen) {
1473 AV *aseen = cxt->aseen;
1474 cxt->aseen = 0;
1475 av_undef(aseen);
1476 sv_free((SV *) aseen);
1477 }
dfd91409 1478 cxt->where_is_undef = -1;
7a6a85bf 1479
862382c7
JH
1480 if (cxt->aclass) {
1481 AV *aclass = cxt->aclass;
1482 cxt->aclass = 0;
1483 av_undef(aclass);
1484 sv_free((SV *) aclass);
1485 }
7a6a85bf 1486
862382c7
JH
1487 if (cxt->hook) {
1488 HV *hook = cxt->hook;
1489 cxt->hook = 0;
1490 hv_undef(hook);
1491 sv_free((SV *) hook);
1492 }
7a6a85bf 1493
862382c7
JH
1494 if (cxt->hseen) {
1495 HV *hseen = cxt->hseen;
1496 cxt->hseen = 0;
1497 hv_undef(hseen);
1498 sv_free((SV *) hseen); /* optional HV, for backward compat. */
1499 }
7a6a85bf 1500
e8189732
NC
1501#ifndef HAS_RESTRICTED_HASHES
1502 cxt->derestrict = -1; /* Fetched from perl if needed */
1503#endif
1504#ifndef HAS_UTF8_ALL
1505 cxt->use_bytes = -1; /* Fetched from perl if needed */
1506#endif
1507 cxt->accept_future_minor = -1; /* Fetched from perl if needed */
1508
51f77169 1509 cxt->in_retrieve_overloaded = 0;
e993d95c 1510 reset_context(cxt);
7a6a85bf
RG
1511}
1512
1513/*
1514 * clean_context
1515 *
1516 * A workaround for the CROAK bug: cleanup the last context.
1517 */
138ec36d 1518static void clean_context(pTHX_ stcxt_t *cxt)
7a6a85bf
RG
1519{
1520 TRACEME(("clean_context"));
1521
dd19458b 1522 ASSERT(cxt->s_dirty, ("dirty context"));
7a6a85bf 1523
e993d95c
JH
1524 if (cxt->membuf_ro)
1525 MBUF_RESTORE();
1526
1527 ASSERT(!cxt->membuf_ro, ("mbase is not read-only"));
1528
7a6a85bf 1529 if (cxt->optype & ST_RETRIEVE)
138ec36d 1530 clean_retrieve_context(aTHX_ cxt);
e993d95c 1531 else if (cxt->optype & ST_STORE)
138ec36d 1532 clean_store_context(aTHX_ cxt);
e993d95c
JH
1533 else
1534 reset_context(cxt);
862382c7
JH
1535
1536 ASSERT(!cxt->s_dirty, ("context is clean"));
e993d95c 1537 ASSERT(cxt->entry == 0, ("context is reset"));
7a6a85bf
RG
1538}
1539
1540/*
1541 * allocate_context
1542 *
1543 * Allocate a new context and push it on top of the parent one.
1544 * This new context is made globally visible via SET_STCXT().
1545 */
138ec36d 1546static stcxt_t *allocate_context(pTHX_ stcxt_t *parent_cxt)
7a6a85bf
RG
1547{
1548 stcxt_t *cxt;
1549
1550 TRACEME(("allocate_context"));
1551
dd19458b 1552 ASSERT(!parent_cxt->s_dirty, ("parent context clean"));
7a6a85bf 1553
111e03c1
RG
1554 NEW_STORABLE_CXT_OBJ(cxt);
1555 cxt->prev = parent_cxt->my_sv;
7a6a85bf
RG
1556 SET_STCXT(cxt);
1557
e993d95c
JH
1558 ASSERT(!cxt->s_dirty, ("clean context"));
1559
7a6a85bf
RG
1560 return cxt;
1561}
1562
1563/*
1564 * free_context
1565 *
1566 * Free current context, which cannot be the "root" one.
1567 * Make the context underneath globally visible via SET_STCXT().
1568 */
138ec36d 1569static void free_context(pTHX_ stcxt_t *cxt)
7a6a85bf 1570{
111e03c1 1571 stcxt_t *prev = (stcxt_t *)(cxt->prev ? SvPVX(SvRV(cxt->prev)) : 0);
7a6a85bf
RG
1572
1573 TRACEME(("free_context"));
1574
dd19458b 1575 ASSERT(!cxt->s_dirty, ("clean context"));
7a6a85bf
RG
1576 ASSERT(prev, ("not freeing root context"));
1577
111e03c1 1578 SvREFCNT_dec(cxt->my_sv);
7a6a85bf 1579 SET_STCXT(prev);
e993d95c
JH
1580
1581 ASSERT(cxt, ("context not void"));
7a6a85bf
RG
1582}
1583
1584/***
1585 *** Predicates.
1586 ***/
1587
1588/*
1589 * is_storing
1590 *
1591 * Tells whether we're in the middle of a store operation.
1592 */
c3551ae4 1593static int is_storing(pTHX)
7a6a85bf
RG
1594{
1595 dSTCXT;
1596
1597 return cxt->entry && (cxt->optype & ST_STORE);
1598}
1599
1600/*
1601 * is_retrieving
1602 *
1603 * Tells whether we're in the middle of a retrieve operation.
1604 */
c3551ae4 1605static int is_retrieving(pTHX)
7a6a85bf
RG
1606{
1607 dSTCXT;
1608
1609 return cxt->entry && (cxt->optype & ST_RETRIEVE);
1610}
1611
1612/*
1613 * last_op_in_netorder
1614 *
1615 * Returns whether last operation was made using network order.
1616 *
1617 * This is typically out-of-band information that might prove useful
1618 * to people wishing to convert native to network order data when used.
1619 */
c3551ae4 1620static int last_op_in_netorder(pTHX)
7a6a85bf
RG
1621{
1622 dSTCXT;
1623
1624 return cxt->netorder;
1625}
1626
1627/***
1628 *** Hook lookup and calling routines.
1629 ***/
1630
1631/*
1632 * pkg_fetchmeth
1633 *
1634 * A wrapper on gv_fetchmethod_autoload() which caches results.
1635 *
1636 * Returns the routine reference as an SV*, or null if neither the package
1637 * nor its ancestors know about the method.
1638 */
f0ffaed8 1639static SV *pkg_fetchmeth(
138ec36d 1640 pTHX_
f0ffaed8
JH
1641 HV *cache,
1642 HV *pkg,
a9eee89a 1643 const char *method)
7a6a85bf
RG
1644{
1645 GV *gv;
1646 SV *sv;
bfcb3514
NC
1647 const char *hvname = HvNAME_get(pkg);
1648
7a6a85bf
RG
1649
1650 /*
1651 * The following code is the same as the one performed by UNIVERSAL::can
1652 * in the Perl core.
1653 */
1654
1655 gv = gv_fetchmethod_autoload(pkg, method, FALSE);
1656 if (gv && isGV(gv)) {
1657 sv = newRV((SV*) GvCV(gv));
bfcb3514 1658 TRACEME(("%s->%s: 0x%"UVxf, hvname, method, PTR2UV(sv)));
7a6a85bf
RG
1659 } else {
1660 sv = newSVsv(&PL_sv_undef);
bfcb3514 1661 TRACEME(("%s->%s: not found", hvname, method));
7a6a85bf
RG
1662 }
1663
1664 /*
1665 * Cache the result, ignoring failure: if we can't store the value,
1666 * it just won't be cached.
1667 */
1668
bfcb3514 1669 (void) hv_store(cache, hvname, strlen(hvname), sv, 0);
7a6a85bf
RG
1670
1671 return SvOK(sv) ? sv : (SV *) 0;
1672}
1673
1674/*
1675 * pkg_hide
1676 *
1677 * Force cached value to be undef: hook ignored even if present.
1678 */
f0ffaed8 1679static void pkg_hide(
138ec36d 1680 pTHX_
f0ffaed8
JH
1681 HV *cache,
1682 HV *pkg,
a9eee89a 1683 const char *method)
7a6a85bf 1684{
bfcb3514 1685 const char *hvname = HvNAME_get(pkg);
c33e8be1 1686 PERL_UNUSED_ARG(method);
7a6a85bf 1687 (void) hv_store(cache,
bfcb3514 1688 hvname, strlen(hvname), newSVsv(&PL_sv_undef), 0);
7a6a85bf
RG
1689}
1690
1691/*
212e9bde
JH
1692 * pkg_uncache
1693 *
1694 * Discard cached value: a whole fetch loop will be retried at next lookup.
1695 */
1696static void pkg_uncache(
138ec36d 1697 pTHX_
212e9bde
JH
1698 HV *cache,
1699 HV *pkg,
a9eee89a 1700 const char *method)
212e9bde 1701{
bfcb3514 1702 const char *hvname = HvNAME_get(pkg);
c33e8be1 1703 PERL_UNUSED_ARG(method);
bfcb3514 1704 (void) hv_delete(cache, hvname, strlen(hvname), G_DISCARD);
212e9bde
JH
1705}
1706
1707/*
7a6a85bf
RG
1708 * pkg_can
1709 *
1710 * Our own "UNIVERSAL::can", which caches results.
1711 *
1712 * Returns the routine reference as an SV*, or null if the object does not
1713 * know about the method.
1714 */
f0ffaed8 1715static SV *pkg_can(
138ec36d 1716 pTHX_
f0ffaed8
JH
1717 HV *cache,
1718 HV *pkg,
a9eee89a 1719 const char *method)
7a6a85bf
RG
1720{
1721 SV **svh;
1722 SV *sv;
bfcb3514 1723 const char *hvname = HvNAME_get(pkg);
7a6a85bf 1724
bfcb3514 1725 TRACEME(("pkg_can for %s->%s", hvname, method));
7a6a85bf
RG
1726
1727 /*
1728 * Look into the cache to see whether we already have determined
1729 * where the routine was, if any.
1730 *
6dfee1ec 1731 * NOTA BENE: we don't use 'method' at all in our lookup, since we know
7a6a85bf
RG
1732 * that only one hook (i.e. always the same) is cached in a given cache.
1733 */
1734
bfcb3514 1735 svh = hv_fetch(cache, hvname, strlen(hvname), FALSE);
7a6a85bf
RG
1736 if (svh) {
1737 sv = *svh;
1738 if (!SvOK(sv)) {
bfcb3514 1739 TRACEME(("cached %s->%s: not found", hvname, method));
7a6a85bf
RG
1740 return (SV *) 0;
1741 } else {
43d061fe 1742 TRACEME(("cached %s->%s: 0x%"UVxf,
bfcb3514 1743 hvname, method, PTR2UV(sv)));
7a6a85bf
RG
1744 return sv;
1745 }
1746 }
1747
1748 TRACEME(("not cached yet"));
138ec36d 1749 return pkg_fetchmeth(aTHX_ cache, pkg, method); /* Fetch and cache */
7a6a85bf
RG
1750}
1751
1752/*
1753 * scalar_call
1754 *
1755 * Call routine as obj->hook(av) in scalar context.
1756 * Propagates the single returned value if not called in void context.
1757 */
f0ffaed8 1758static SV *scalar_call(
138ec36d 1759 pTHX_
f0ffaed8
JH
1760 SV *obj,
1761 SV *hook,
1762 int cloning,
1763 AV *av,
1764 I32 flags)
7a6a85bf
RG
1765{
1766 dSP;
1767 int count;
1768 SV *sv = 0;
1769
1770 TRACEME(("scalar_call (cloning=%d)", cloning));
1771
1772 ENTER;
1773 SAVETMPS;
1774
1775 PUSHMARK(sp);
1776 XPUSHs(obj);
1777 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1778 if (av) {
1779 SV **ary = AvARRAY(av);
1780 int cnt = AvFILLp(av) + 1;
1781 int i;
1782 XPUSHs(ary[0]); /* Frozen string */
1783 for (i = 1; i < cnt; i++) {
43d061fe
JH
1784 TRACEME(("pushing arg #%d (0x%"UVxf")...",
1785 i, PTR2UV(ary[i])));
7a6a85bf
RG
1786 XPUSHs(sv_2mortal(newRV(ary[i])));
1787 }
1788 }
1789 PUTBACK;
1790
1791 TRACEME(("calling..."));
1792 count = perl_call_sv(hook, flags); /* Go back to Perl code */
1793 TRACEME(("count = %d", count));
1794
1795 SPAGAIN;
1796
1797 if (count) {
1798 sv = POPs;
1799 SvREFCNT_inc(sv); /* We're returning it, must stay alive! */
1800 }
1801
1802 PUTBACK;
1803 FREETMPS;
1804 LEAVE;
1805
1806 return sv;
1807}
1808
1809/*
1810 * array_call
1811 *
f9a1036d 1812 * Call routine obj->hook(cloning) in list context.
7a6a85bf
RG
1813 * Returns the list of returned values in an array.
1814 */
f0ffaed8 1815static AV *array_call(
138ec36d 1816 pTHX_
f0ffaed8
JH
1817 SV *obj,
1818 SV *hook,
1819 int cloning)
7a6a85bf
RG
1820{
1821 dSP;
1822 int count;
1823 AV *av;
1824 int i;
1825
f0ffaed8 1826 TRACEME(("array_call (cloning=%d)", cloning));
7a6a85bf
RG
1827
1828 ENTER;
1829 SAVETMPS;
1830
1831 PUSHMARK(sp);
1832 XPUSHs(obj); /* Target object */
1833 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1834 PUTBACK;
1835
1836 count = perl_call_sv(hook, G_ARRAY); /* Go back to Perl code */
1837
1838 SPAGAIN;
1839
1840 av = newAV();
1841 for (i = count - 1; i >= 0; i--) {
1842 SV *sv = POPs;
1843 av_store(av, i, SvREFCNT_inc(sv));
1844 }
1845
1846 PUTBACK;
1847 FREETMPS;
1848 LEAVE;
1849
1850 return av;
1851}
1852
1853/*
1854 * known_class
1855 *
6dfee1ec
JK
1856 * Lookup the class name in the 'hclass' table and either assign it a new ID
1857 * or return the existing one, by filling in 'classnum'.
7a6a85bf
RG
1858 *
1859 * Return true if the class was known, false if the ID was just generated.
1860 */
f0ffaed8 1861static int known_class(
138ec36d 1862 pTHX_
f0ffaed8
JH
1863 stcxt_t *cxt,
1864 char *name, /* Class name */
1865 int len, /* Name length */
1866 I32 *classnum)
7a6a85bf
RG
1867{
1868 SV **svh;
1869 HV *hclass = cxt->hclass;
1870
1871 TRACEME(("known_class (%s)", name));
1872
1873 /*
1874 * Recall that we don't store pointers in this hash table, but tags.
1875 * Therefore, we need LOW_32BITS() to extract the relevant parts.
1876 */
1877
1878 svh = hv_fetch(hclass, name, len, FALSE);
1879 if (svh) {
1880 *classnum = LOW_32BITS(*svh);
1881 return TRUE;
1882 }
1883
1884 /*
1885 * Unknown classname, we need to record it.
7a6a85bf
RG
1886 */
1887
1888 cxt->classnum++;
3341c981 1889 if (!hv_store(hclass, name, len, INT2PTR(SV*, cxt->classnum), 0))
7a6a85bf
RG
1890 CROAK(("Unable to record new classname"));
1891
1892 *classnum = cxt->classnum;
1893 return FALSE;
1894}
1895
1896/***
c4a6f826 1897 *** Specific store routines.
7a6a85bf
RG
1898 ***/
1899
1900/*
1901 * store_ref
1902 *
1903 * Store a reference.
1904 * Layout is SX_REF <object> or SX_OVERLOAD <object>.
1905 */
138ec36d 1906static int store_ref(pTHX_ stcxt_t *cxt, SV *sv)
7a6a85bf 1907{
c3c53033 1908 int is_weak = 0;
43d061fe 1909 TRACEME(("store_ref (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
1910
1911 /*
1912 * Follow reference, and check if target is overloaded.
1913 */
1914
96466a21 1915#ifdef SvWEAKREF
c3c53033
NC
1916 if (SvWEAKREF(sv))
1917 is_weak = 1;
1918 TRACEME(("ref (0x%"UVxf") is%s weak", PTR2UV(sv), is_weak ? "" : "n't"));
1919#endif
7a6a85bf
RG
1920 sv = SvRV(sv);
1921
1922 if (SvOBJECT(sv)) {
1923 HV *stash = (HV *) SvSTASH(sv);
1924 if (stash && Gv_AMG(stash)) {
9e21b3d0 1925 TRACEME(("ref (0x%"UVxf") is overloaded", PTR2UV(sv)));
c3c53033 1926 PUTMARK(is_weak ? SX_WEAKOVERLOAD : SX_OVERLOAD);
7a6a85bf 1927 } else
c3c53033 1928 PUTMARK(is_weak ? SX_WEAKREF : SX_REF);
7a6a85bf 1929 } else
c3c53033 1930 PUTMARK(is_weak ? SX_WEAKREF : SX_REF);
7a6a85bf 1931
138ec36d 1932 return store(aTHX_ cxt, sv);
7a6a85bf
RG
1933}
1934
1935/*
1936 * store_scalar
1937 *
1938 * Store a scalar.
1939 *
e16e2ff8 1940 * Layout is SX_LSCALAR <length> <data>, SX_SCALAR <length> <data> or SX_UNDEF.
a137b8e5 1941 * SX_LUTF8STR and SX_UTF8STR are used for UTF-8 strings.
7a6a85bf
RG
1942 * The <data> section is omitted if <length> is 0.
1943 *
1944 * If integer or double, the layout is SX_INTEGER <data> or SX_DOUBLE <data>.
1945 * Small integers (within [-127, +127]) are stored as SX_BYTE <byte>.
1946 */
138ec36d 1947static int store_scalar(pTHX_ stcxt_t *cxt, SV *sv)
7a6a85bf
RG
1948{
1949 IV iv;
1950 char *pv;
1951 STRLEN len;
1952 U32 flags = SvFLAGS(sv); /* "cc -O" may put it in register */
1953
43d061fe 1954 TRACEME(("store_scalar (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
1955
1956 /*
1957 * For efficiency, break the SV encapsulation by peaking at the flags
1958 * directly without using the Perl macros to avoid dereferencing
1959 * sv->sv_flags each time we wish to check the flags.
1960 */
1961
1962 if (!(flags & SVf_OK)) { /* !SvOK(sv) */
1963 if (sv == &PL_sv_undef) {
1964 TRACEME(("immortal undef"));
1965 PUTMARK(SX_SV_UNDEF);
1966 } else {
86bbd6dc 1967 TRACEME(("undef at 0x%"UVxf, PTR2UV(sv)));
7a6a85bf
RG
1968 PUTMARK(SX_UNDEF);
1969 }
1970 return 0;
1971 }
1972
1973 /*
1974 * Always store the string representation of a scalar if it exists.
1975 * Gisle Aas provided me with this test case, better than a long speach:
1976 *
1977 * perl -MDevel::Peek -le '$a="abc"; $a+0; Dump($a)'
1978 * SV = PVNV(0x80c8520)
1979 * REFCNT = 1
1980 * FLAGS = (NOK,POK,pNOK,pPOK)
1981 * IV = 0
1982 * NV = 0
1983 * PV = 0x80c83d0 "abc"\0
1984 * CUR = 3
1985 * LEN = 4
1986 *
1987 * Write SX_SCALAR, length, followed by the actual data.
1988 *
1989 * Otherwise, write an SX_BYTE, SX_INTEGER or an SX_DOUBLE as
1990 * appropriate, followed by the actual (binary) data. A double
1991 * is written as a string if network order, for portability.
1992 *
1993 * NOTE: instead of using SvNOK(sv), we test for SvNOKp(sv).
1994 * The reason is that when the scalar value is tainted, the SvNOK(sv)
1995 * value is false.
1996 *
1997 * The test for a read-only scalar with both POK and NOK set is meant
1998 * to quickly detect &PL_sv_yes and &PL_sv_no without having to pay the
1999 * address comparison for each scalar we store.
2000 */
2001
2002#define SV_MAYBE_IMMORTAL (SVf_READONLY|SVf_POK|SVf_NOK)
2003
2004 if ((flags & SV_MAYBE_IMMORTAL) == SV_MAYBE_IMMORTAL) {
2005 if (sv == &PL_sv_yes) {
2006 TRACEME(("immortal yes"));
2007 PUTMARK(SX_SV_YES);
2008 } else if (sv == &PL_sv_no) {
2009 TRACEME(("immortal no"));
2010 PUTMARK(SX_SV_NO);
2011 } else {
2012 pv = SvPV(sv, len); /* We know it's SvPOK */
2013 goto string; /* Share code below */
2014 }
db670f21
NC
2015 } else if (flags & SVf_POK) {
2016 /* public string - go direct to string read. */
2017 goto string_readlen;
2018 } else if (
2019#if (PATCHLEVEL <= 6)
2020 /* For 5.6 and earlier NV flag trumps IV flag, so only use integer
2021 direct if NV flag is off. */
2022 (flags & (SVf_NOK | SVf_IOK)) == SVf_IOK
2023#else
2024 /* 5.7 rules are that if IV public flag is set, IV value is as
2025 good, if not better, than NV value. */
2026 flags & SVf_IOK
2027#endif
2028 ) {
2029 iv = SvIV(sv);
2030 /*
2031 * Will come here from below with iv set if double is an integer.
2032 */
2033 integer:
7a6a85bf 2034
db670f21
NC
2035 /* Sorry. This isn't in 5.005_56 (IIRC) or earlier. */
2036#ifdef SVf_IVisUV
2037 /* Need to do this out here, else 0xFFFFFFFF becomes iv of -1
2038 * (for example) and that ends up in the optimised small integer
2039 * case.
2040 */
2041 if ((flags & SVf_IVisUV) && SvUV(sv) > IV_MAX) {
2042 TRACEME(("large unsigned integer as string, value = %"UVuf, SvUV(sv)));
2043 goto string_readlen;
2044 }
2045#endif
2046 /*
2047 * Optimize small integers into a single byte, otherwise store as
2048 * a real integer (converted into network order if they asked).
2049 */
7a6a85bf 2050
db670f21
NC
2051 if (iv >= -128 && iv <= 127) {
2052 unsigned char siv = (unsigned char) (iv + 128); /* [0,255] */
2053 PUTMARK(SX_BYTE);
2054 PUTMARK(siv);
2055 TRACEME(("small integer stored as %d", siv));
2056 } else if (cxt->netorder) {
2057#ifndef HAS_HTONL
2058 TRACEME(("no htonl, fall back to string for integer"));
2059 goto string_readlen;
2060#else
2061 I32 niv;
7a6a85bf 2062
7a6a85bf 2063
db670f21
NC
2064#if IVSIZE > 4
2065 if (
2066#ifdef SVf_IVisUV
2067 /* Sorry. This isn't in 5.005_56 (IIRC) or earlier. */
41c44503 2068 ((flags & SVf_IVisUV) && SvUV(sv) > (UV)0x7FFFFFFF) ||
db670f21 2069#endif
41c44503 2070 (iv > (IV)0x7FFFFFFF) || (iv < -(IV)0x80000000)) {
db670f21
NC
2071 /* Bigger than 32 bits. */
2072 TRACEME(("large network order integer as string, value = %"IVdf, iv));
2073 goto string_readlen;
2074 }
2075#endif
7a6a85bf 2076
db670f21
NC
2077 niv = (I32) htonl((I32) iv);
2078 TRACEME(("using network order"));
2079 PUTMARK(SX_NETINT);
2080 WRITE_I32(niv);
2081#endif
2082 } else {
2083 PUTMARK(SX_INTEGER);
2084 WRITE(&iv, sizeof(iv));
2085 }
2086
2087 TRACEME(("ok (integer 0x%"UVxf", value = %"IVdf")", PTR2UV(sv), iv));
2088 } else if (flags & SVf_NOK) {
2089 NV nv;
2090#if (PATCHLEVEL <= 6)
2091 nv = SvNV(sv);
2092 /*
2093 * Watch for number being an integer in disguise.
2094 */
2095 if (nv == (NV) (iv = I_V(nv))) {
2096 TRACEME(("double %"NVff" is actually integer %"IVdf, nv, iv));
2097 goto integer; /* Share code above */
2098 }
2099#else
7a6a85bf 2100
db670f21 2101 SvIV_please(sv);
3ddd445a 2102 if (SvIOK_notUV(sv)) {
db670f21
NC
2103 iv = SvIV(sv);
2104 goto integer; /* Share code above */
2105 }
2106 nv = SvNV(sv);
2107#endif
7a6a85bf 2108
db670f21
NC
2109 if (cxt->netorder) {
2110 TRACEME(("double %"NVff" stored as string", nv));
2111 goto string_readlen; /* Share code below */
2112 }
7a6a85bf 2113
db670f21
NC
2114 PUTMARK(SX_DOUBLE);
2115 WRITE(&nv, sizeof(nv));
7a6a85bf 2116
db670f21 2117 TRACEME(("ok (double 0x%"UVxf", value = %"NVff")", PTR2UV(sv), nv));
7a6a85bf 2118
db670f21
NC
2119 } else if (flags & (SVp_POK | SVp_NOK | SVp_IOK)) {
2120 I32 wlen; /* For 64-bit machines */
7a6a85bf 2121
db670f21
NC
2122 string_readlen:
2123 pv = SvPV(sv, len);
7a6a85bf 2124
db670f21
NC
2125 /*
2126 * Will come here from above if it was readonly, POK and NOK but
2127 * neither &PL_sv_yes nor &PL_sv_no.
2128 */
2129 string:
2130
2131 wlen = (I32) len; /* WLEN via STORE_SCALAR expects I32 */
2132 if (SvUTF8 (sv))
2133 STORE_UTF8STR(pv, wlen);
2134 else
2135 STORE_SCALAR(pv, wlen);
2136 TRACEME(("ok (scalar 0x%"UVxf" '%s', length = %"IVdf")",
2137 PTR2UV(sv), SvPVX(sv), (IV)len));
7a6a85bf 2138 } else
db670f21
NC
2139 CROAK(("Can't determine type of %s(0x%"UVxf")",
2140 sv_reftype(sv, FALSE),
2141 PTR2UV(sv)));
2142 return 0; /* Ok, no recursion on scalars */
7a6a85bf
RG
2143}
2144
2145/*
2146 * store_array
2147 *
2148 * Store an array.
2149 *
c4a6f826 2150 * Layout is SX_ARRAY <size> followed by each item, in increasing index order.
7a6a85bf
RG
2151 * Each item is stored as <object>.
2152 */
138ec36d 2153static int store_array(pTHX_ stcxt_t *cxt, AV *av)
7a6a85bf
RG
2154{
2155 SV **sav;
2156 I32 len = av_len(av) + 1;
2157 I32 i;
2158 int ret;
2159
43d061fe 2160 TRACEME(("store_array (0x%"UVxf")", PTR2UV(av)));
7a6a85bf
RG
2161
2162 /*
2163 * Signal array by emitting SX_ARRAY, followed by the array length.
2164 */
2165
2166 PUTMARK(SX_ARRAY);
2167 WLEN(len);
2168 TRACEME(("size = %d", len));
2169
2170 /*
2171 * Now store each item recursively.
2172 */
2173
2174 for (i = 0; i < len; i++) {
2175 sav = av_fetch(av, i, 0);
2176 if (!sav) {
2177 TRACEME(("(#%d) undef item", i));
20bb3f55 2178 STORE_SV_UNDEF();
7a6a85bf
RG
2179 continue;
2180 }
2181 TRACEME(("(#%d) item", i));
138ec36d 2182 if ((ret = store(aTHX_ cxt, *sav))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2183 return ret;
2184 }
2185
2186 TRACEME(("ok (array)"));
2187
2188 return 0;
2189}
2190
138ec36d 2191
2192#if (PATCHLEVEL <= 6)
2193
7a6a85bf
RG
2194/*
2195 * sortcmp
2196 *
2197 * Sort two SVs
2198 * Borrowed from perl source file pp_ctl.c, where it is used by pp_sort.
2199 */
2200static int
f0ffaed8 2201sortcmp(const void *a, const void *b)
7a6a85bf 2202{
138ec36d 2203#if defined(USE_ITHREADS)
2204 dTHX;
2205#endif /* USE_ITHREADS */
2206 return sv_cmp(*(SV * const *) a, *(SV * const *) b);
7a6a85bf
RG
2207}
2208
138ec36d 2209#endif /* PATCHLEVEL <= 6 */
7a6a85bf
RG
2210
2211/*
2212 * store_hash
2213 *
d1be9408 2214 * Store a hash table.
7a6a85bf 2215 *
e16e2ff8
NC
2216 * For a "normal" hash (not restricted, no utf8 keys):
2217 *
7a6a85bf
RG
2218 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
2219 * Values are stored as <object>.
2220 * Keys are stored as <length> <data>, the <data> section being omitted
2221 * if length is 0.
c194a0a3
TB
2222 *
2223 * For a "fancy" hash (restricted or utf8 keys):
2224 *
2225 * Layout is SX_FLAG_HASH <size> <hash flags> followed by each key/value pair,
e16e2ff8
NC
2226 * in random order.
2227 * Values are stored as <object>.
2228 * Keys are stored as <flags> <length> <data>, the <data> section being omitted
2229 * if length is 0.
c4a6f826 2230 * Currently the only hash flag is "restricted"
e16e2ff8 2231 * Key flags are as for hv.h
7a6a85bf 2232 */
138ec36d 2233static int store_hash(pTHX_ stcxt_t *cxt, HV *hv)
7a6a85bf 2234{
27da23d5 2235 dVAR;
1c4fe6e3 2236 I32 len = HvTOTALKEYS(hv);
7a6a85bf
RG
2237 I32 i;
2238 int ret = 0;
2239 I32 riter;
2240 HE *eiter;
530b72ba
NC
2241 int flagged_hash = ((SvREADONLY(hv)
2242#ifdef HAS_HASH_KEY_FLAGS
2243 || HvHASKFLAGS(hv)
2244#endif
2245 ) ? 1 : 0);
e16e2ff8 2246 unsigned char hash_flags = (SvREADONLY(hv) ? SHV_RESTRICTED : 0);
7a6a85bf 2247
e16e2ff8
NC
2248 if (flagged_hash) {
2249 /* needs int cast for C++ compilers, doesn't it? */
2250 TRACEME(("store_hash (0x%"UVxf") (flags %x)", PTR2UV(hv),
2251 (int) hash_flags));
2252 } else {
2253 TRACEME(("store_hash (0x%"UVxf")", PTR2UV(hv)));
2254 }
7a6a85bf
RG
2255
2256 /*
2257 * Signal hash by emitting SX_HASH, followed by the table length.
2258 */
2259
e16e2ff8
NC
2260 if (flagged_hash) {
2261 PUTMARK(SX_FLAG_HASH);
2262 PUTMARK(hash_flags);
2263 } else {
2264 PUTMARK(SX_HASH);
2265 }
7a6a85bf
RG
2266 WLEN(len);
2267 TRACEME(("size = %d", len));
2268
2269 /*
2270 * Save possible iteration state via each() on that table.
2271 */
2272
bfcb3514
NC
2273 riter = HvRITER_get(hv);
2274 eiter = HvEITER_get(hv);
7a6a85bf
RG
2275 hv_iterinit(hv);
2276
2277 /*
2278 * Now store each item recursively.
2279 *
2280 * If canonical is defined to some true value then store each
2281 * key/value pair in sorted order otherwise the order is random.
2282 * Canonical order is irrelevant when a deep clone operation is performed.
2283 *
2284 * Fetch the value from perl only once per store() operation, and only
2285 * when needed.
2286 */
2287
2288 if (
2289 !(cxt->optype & ST_CLONE) && (cxt->canonical == 1 ||
2290 (cxt->canonical < 0 && (cxt->canonical =
3509f647 2291 (SvTRUE(perl_get_sv("Storable::canonical", GV_ADD)) ? 1 : 0))))
7a6a85bf
RG
2292 ) {
2293 /*
2294 * Storing in order, sorted by key.
2295 * Run through the hash, building up an array of keys in a
2296 * mortal array, sort the array and then run through the
2297 * array.
2298 */
2299
2300 AV *av = newAV();
2301
e16e2ff8
NC
2302 /*av_extend (av, len);*/
2303
7a6a85bf
RG
2304 TRACEME(("using canonical order"));
2305
2306 for (i = 0; i < len; i++) {
530b72ba 2307#ifdef HAS_RESTRICTED_HASHES
e16e2ff8 2308 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
530b72ba
NC
2309#else
2310 HE *he = hv_iternext(hv);
2311#endif
0d326098
NC
2312 SV *key;
2313
2314 if (!he)
c33e8be1 2315 CROAK(("Hash %p inconsistent - expected %d keys, %dth is NULL", hv, (int)len, (int)i));
0d326098 2316 key = hv_iterkeysv(he);
7a6a85bf
RG
2317 av_store(av, AvFILLp(av)+1, key); /* av_push(), really */
2318 }
2319
138ec36d 2320 STORE_HASH_SORT;
7a6a85bf
RG
2321
2322 for (i = 0; i < len; i++) {
dfd91409 2323#ifdef HAS_RESTRICTED_HASHES
ca732855 2324 int placeholders = (int)HvPLACEHOLDERS_get(hv);
dfd91409
NC
2325#endif
2326 unsigned char flags = 0;
7a6a85bf 2327 char *keyval;
e16e2ff8
NC
2328 STRLEN keylen_tmp;
2329 I32 keylen;
7a6a85bf 2330 SV *key = av_shift(av);
dfd91409
NC
2331 /* This will fail if key is a placeholder.
2332 Track how many placeholders we have, and error if we
2333 "see" too many. */
7a6a85bf 2334 HE *he = hv_fetch_ent(hv, key, 0, 0);
dfd91409
NC
2335 SV *val;
2336
2337 if (he) {
2338 if (!(val = HeVAL(he))) {
2339 /* Internal error, not I/O error */
2340 return 1;
2341 }
2342 } else {
2343#ifdef HAS_RESTRICTED_HASHES
2344 /* Should be a placeholder. */
2345 if (placeholders-- < 0) {
2346 /* This should not happen - number of
2347 retrieves should be identical to
2348 number of placeholders. */
2349 return 1;
2350 }
2351 /* Value is never needed, and PL_sv_undef is
2352 more space efficient to store. */
2353 val = &PL_sv_undef;
2354 ASSERT (flags == 0,
2355 ("Flags not 0 but %d", flags));
2356 flags = SHV_K_PLACEHOLDER;
2357#else
2358 return 1;
2359#endif
2360 }
7a6a85bf
RG
2361
2362 /*
2363 * Store value first.
2364 */
2365
9e21b3d0 2366 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
7a6a85bf 2367
138ec36d 2368 if ((ret = store(aTHX_ cxt, val))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2369 goto out;
2370
2371 /*
2372 * Write key string.
2373 * Keys are written after values to make sure retrieval
2374 * can be optimal in terms of memory usage, where keys are
2375 * read into a fixed unique buffer called kbuf.
2376 * See retrieve_hash() for details.
2377 */
2378
e16e2ff8
NC
2379 /* Implementation of restricted hashes isn't nicely
2380 abstracted: */
a991bd3b
FC
2381 if ((hash_flags & SHV_RESTRICTED)
2382 && SvREADONLY(val) && !SvIsCOW(val)) {
dfd91409
NC
2383 flags |= SHV_K_LOCKED;
2384 }
e16e2ff8
NC
2385
2386 keyval = SvPV(key, keylen_tmp);
2387 keylen = keylen_tmp;
530b72ba
NC
2388#ifdef HAS_UTF8_HASHES
2389 /* If you build without optimisation on pre 5.6
2390 then nothing spots that SvUTF8(key) is always 0,
2391 so the block isn't optimised away, at which point
2392 the linker dislikes the reference to
2393 bytes_from_utf8. */
e16e2ff8
NC
2394 if (SvUTF8(key)) {
2395 const char *keysave = keyval;
2396 bool is_utf8 = TRUE;
2397
2398 /* Just casting the &klen to (STRLEN) won't work
2399 well if STRLEN and I32 are of different widths.
2400 --jhi */
2401 keyval = (char*)bytes_from_utf8((U8*)keyval,
2402 &keylen_tmp,
2403 &is_utf8);
2404
2405 /* If we were able to downgrade here, then than
2406 means that we have a key which only had chars
2407 0-255, but was utf8 encoded. */
2408
2409 if (keyval != keysave) {
2410 keylen = keylen_tmp;
2411 flags |= SHV_K_WASUTF8;
2412 } else {
2413 /* keylen_tmp can't have changed, so no need
2414 to assign back to keylen. */
2415 flags |= SHV_K_UTF8;
2416 }
2417 }
530b72ba 2418#endif
e16e2ff8
NC
2419
2420 if (flagged_hash) {
2421 PUTMARK(flags);
2422 TRACEME(("(#%d) key '%s' flags %x %u", i, keyval, flags, *keyval));
2423 } else {
fcaa57e7
AMS
2424 /* This is a workaround for a bug in 5.8.0
2425 that causes the HEK_WASUTF8 flag to be
2426 set on an HEK without the hash being
2427 marked as having key flags. We just
2428 cross our fingers and drop the flag.
2429 AMS 20030901 */
2430 assert (flags == 0 || flags == SHV_K_WASUTF8);
e16e2ff8
NC
2431 TRACEME(("(#%d) key '%s'", i, keyval));
2432 }
7a6a85bf
RG
2433 WLEN(keylen);
2434 if (keylen)
2435 WRITE(keyval, keylen);
e16e2ff8
NC
2436 if (flags & SHV_K_WASUTF8)
2437 Safefree (keyval);
7a6a85bf
RG
2438 }
2439
2440 /*
2441 * Free up the temporary array
2442 */
2443
2444 av_undef(av);
2445 sv_free((SV *) av);
2446
2447 } else {
2448
2449 /*
2450 * Storing in "random" order (in the order the keys are stored
a6d05634 2451 * within the hash). This is the default and will be faster!
7a6a85bf
RG
2452 */
2453
2454 for (i = 0; i < len; i++) {
0bb78401 2455 char *key = 0;
7a6a85bf 2456 I32 len;
e16e2ff8 2457 unsigned char flags;
530b72ba 2458#ifdef HV_ITERNEXT_WANTPLACEHOLDERS
e16e2ff8 2459 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
530b72ba
NC
2460#else
2461 HE *he = hv_iternext(hv);
2462#endif
e16e2ff8
NC
2463 SV *val = (he ? hv_iterval(hv, he) : 0);
2464 SV *key_sv = NULL;
2465 HEK *hek;
7a6a85bf
RG
2466
2467 if (val == 0)
2468 return 1; /* Internal error, not I/O error */
2469
dfd91409
NC
2470 /* Implementation of restricted hashes isn't nicely
2471 abstracted: */
2472 flags
2473 = (((hash_flags & SHV_RESTRICTED)
a991bd3b 2474 && SvREADONLY(val) && !SvIsCOW(val))
dfd91409
NC
2475 ? SHV_K_LOCKED : 0);
2476
2477 if (val == &PL_sv_placeholder) {
2478 flags |= SHV_K_PLACEHOLDER;
2479 val = &PL_sv_undef;
2480 }
2481
7a6a85bf
RG
2482 /*
2483 * Store value first.
2484 */
2485
9e21b3d0 2486 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
7a6a85bf 2487
138ec36d 2488 if ((ret = store(aTHX_ cxt, val))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2489 goto out;
2490
e16e2ff8
NC
2491
2492 hek = HeKEY_hek(he);
2493 len = HEK_LEN(hek);
2494 if (len == HEf_SVKEY) {
2495 /* This is somewhat sick, but the internal APIs are
2496 * such that XS code could put one of these in in
2497 * a regular hash.
2498 * Maybe we should be capable of storing one if
2499 * found.
2500 */
2501 key_sv = HeKEY_sv(he);
2502 flags |= SHV_K_ISSV;
2503 } else {
2504 /* Regular string key. */
530b72ba 2505#ifdef HAS_HASH_KEY_FLAGS
e16e2ff8
NC
2506 if (HEK_UTF8(hek))
2507 flags |= SHV_K_UTF8;
2508 if (HEK_WASUTF8(hek))
2509 flags |= SHV_K_WASUTF8;
530b72ba 2510#endif
e16e2ff8
NC
2511 key = HEK_KEY(hek);
2512 }
7a6a85bf
RG
2513 /*
2514 * Write key string.
2515 * Keys are written after values to make sure retrieval
2516 * can be optimal in terms of memory usage, where keys are
2517 * read into a fixed unique buffer called kbuf.
2518 * See retrieve_hash() for details.
2519 */
2520
e16e2ff8
NC
2521 if (flagged_hash) {
2522 PUTMARK(flags);
2523 TRACEME(("(#%d) key '%s' flags %x", i, key, flags));
2524 } else {
fcaa57e7
AMS
2525 /* This is a workaround for a bug in 5.8.0
2526 that causes the HEK_WASUTF8 flag to be
2527 set on an HEK without the hash being
2528 marked as having key flags. We just
2529 cross our fingers and drop the flag.
2530 AMS 20030901 */
2531 assert (flags == 0 || flags == SHV_K_WASUTF8);
e16e2ff8
NC
2532 TRACEME(("(#%d) key '%s'", i, key));
2533 }
2534 if (flags & SHV_K_ISSV) {
138ec36d 2535 store(aTHX_ cxt, key_sv);
e16e2ff8
NC
2536 } else {
2537 WLEN(len);
2538 if (len)
7a6a85bf 2539 WRITE(key, len);
e16e2ff8 2540 }
7a6a85bf
RG
2541 }
2542 }
2543
43d061fe 2544 TRACEME(("ok (hash 0x%"UVxf")", PTR2UV(hv)));
7a6a85bf
RG
2545
2546out:
bfcb3514
NC
2547 HvRITER_set(hv, riter); /* Restore hash iterator state */
2548 HvEITER_set(hv, eiter);
7a6a85bf
RG
2549
2550 return ret;
2551}
2552
2553/*
464b080a
SR
2554 * store_code
2555 *
2556 * Store a code reference.
2557 *
2558 * Layout is SX_CODE <length> followed by a scalar containing the perl
2559 * source code of the code reference.
2560 */
138ec36d 2561static int store_code(pTHX_ stcxt_t *cxt, CV *cv)
464b080a
SR
2562{
2563#if PERL_VERSION < 6
2564 /*
2565 * retrieve_code does not work with perl 5.005 or less
2566 */
138ec36d 2567 return store_other(aTHX_ cxt, (SV*)cv);
464b080a
SR
2568#else
2569 dSP;
2570 I32 len;
c5661c80 2571 int count, reallen;
464b080a
SR
2572 SV *text, *bdeparse;
2573
2574 TRACEME(("store_code (0x%"UVxf")", PTR2UV(cv)));
2575
2576 if (
2577 cxt->deparse == 0 ||
2578 (cxt->deparse < 0 && !(cxt->deparse =
3509f647 2579 SvTRUE(perl_get_sv("Storable::Deparse", GV_ADD)) ? 1 : 0))
464b080a 2580 ) {
138ec36d 2581 return store_other(aTHX_ cxt, (SV*)cv);
464b080a
SR
2582 }
2583
2584 /*
2585 * Require B::Deparse. At least B::Deparse 0.61 is needed for
2586 * blessed code references.
2587 */
17625bd2 2588 /* Ownership of both SVs is passed to load_module, which frees them. */
464b080a 2589 load_module(PERL_LOADMOD_NOIMPORT, newSVpvn("B::Deparse",10), newSVnv(0.61));
85472d4f 2590 SPAGAIN;
464b080a
SR
2591
2592 ENTER;
2593 SAVETMPS;
2594
2595 /*
2596 * create the B::Deparse object
2597 */
2598
2599 PUSHMARK(sp);
afce0a13 2600 XPUSHs(newSVpvs_flags("B::Deparse", SVs_TEMP));
464b080a
SR
2601 PUTBACK;
2602 count = call_method("new", G_SCALAR);
2603 SPAGAIN;
2604 if (count != 1)
2605 CROAK(("Unexpected return value from B::Deparse::new\n"));
2606 bdeparse = POPs;
2607
2608 /*
2609 * call the coderef2text method
2610 */
2611
2612 PUSHMARK(sp);
2613 XPUSHs(bdeparse); /* XXX is this already mortal? */
2614 XPUSHs(sv_2mortal(newRV_inc((SV*)cv)));
2615 PUTBACK;
2616 count = call_method("coderef2text", G_SCALAR);
2617 SPAGAIN;
2618 if (count != 1)
2619 CROAK(("Unexpected return value from B::Deparse::coderef2text\n"));
2620
2621 text = POPs;
dfe4365a 2622 len = SvCUR(text);
e3feee4e 2623 reallen = strlen(SvPV_nolen(text));
464b080a
SR
2624
2625 /*
2626 * Empty code references or XS functions are deparsed as
2627 * "(prototype) ;" or ";".
2628 */
2629
e3feee4e 2630 if (len == 0 || *(SvPV_nolen(text)+reallen-1) == ';') {
464b080a
SR
2631 CROAK(("The result of B::Deparse::coderef2text was empty - maybe you're trying to serialize an XS function?\n"));
2632 }
2633
2634 /*
2635 * Signal code by emitting SX_CODE.
2636 */
2637
2638 PUTMARK(SX_CODE);
a8b7ef86 2639 cxt->tagnum++; /* necessary, as SX_CODE is a SEEN() candidate */
464b080a 2640 TRACEME(("size = %d", len));
e3feee4e 2641 TRACEME(("code = %s", SvPV_nolen(text)));
464b080a
SR
2642
2643 /*
2644 * Now store the source code.
2645 */
2646
70b88f41
DL
2647 if(SvUTF8 (text))
2648 STORE_UTF8STR(SvPV_nolen(text), len);
2649 else
2650 STORE_SCALAR(SvPV_nolen(text), len);
464b080a
SR
2651
2652 FREETMPS;
2653 LEAVE;
2654
2655 TRACEME(("ok (code)"));
2656
2657 return 0;
2658#endif
2659}
2660
2661/*
7a6a85bf
RG
2662 * store_tied
2663 *
2664 * When storing a tied object (be it a tied scalar, array or hash), we lay out
2665 * a special mark, followed by the underlying tied object. For instance, when
2666 * dealing with a tied hash, we store SX_TIED_HASH <hash object>, where
2667 * <hash object> stands for the serialization of the tied hash.
2668 */
138ec36d 2669static int store_tied(pTHX_ stcxt_t *cxt, SV *sv)
7a6a85bf
RG
2670{
2671 MAGIC *mg;
72edffd8 2672 SV *obj = NULL;
7a6a85bf
RG
2673 int ret = 0;
2674 int svt = SvTYPE(sv);
2675 char mtype = 'P';
2676
43d061fe 2677 TRACEME(("store_tied (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
2678
2679 /*
2680 * We have a small run-time penalty here because we chose to factorise
2681 * all tieds objects into the same routine, and not have a store_tied_hash,
2682 * a store_tied_array, etc...
2683 *
2684 * Don't use a switch() statement, as most compilers don't optimize that
2685 * well for 2/3 values. An if() else if() cascade is just fine. We put
2686 * tied hashes first, as they are the most likely beasts.
2687 */
2688
2689 if (svt == SVt_PVHV) {
2690 TRACEME(("tied hash"));
2691 PUTMARK(SX_TIED_HASH); /* Introduces tied hash */
2692 } else if (svt == SVt_PVAV) {
2693 TRACEME(("tied array"));
2694 PUTMARK(SX_TIED_ARRAY); /* Introduces tied array */
2695 } else {
2696 TRACEME(("tied scalar"));
2697 PUTMARK(SX_TIED_SCALAR); /* Introduces tied scalar */
2698 mtype = 'q';
2699 }
2700
2701 if (!(mg = mg_find(sv, mtype)))
2702 CROAK(("No magic '%c' found while storing tied %s", mtype,
2703 (svt == SVt_PVHV) ? "hash" :
2704 (svt == SVt_PVAV) ? "array" : "scalar"));
2705
2706 /*
2707 * The mg->mg_obj found by mg_find() above actually points to the
2708 * underlying tied Perl object implementation. For instance, if the
2709 * original SV was that of a tied array, then mg->mg_obj is an AV.
2710 *
2711 * Note that we store the Perl object as-is. We don't call its FETCH
2712 * method along the way. At retrieval time, we won't call its STORE
2713 * method either, but the tieing magic will be re-installed. In itself,
c4a6f826 2714 * that ensures that the tieing semantics are preserved since further
7a6a85bf
RG
2715 * accesses on the retrieved object will indeed call the magic methods...
2716 */
2717
72edffd8
AMS
2718 /* [#17040] mg_obj is NULL for scalar self-ties. AMS 20030416 */
2719 obj = mg->mg_obj ? mg->mg_obj : newSV(0);
138ec36d 2720 if ((ret = store(aTHX_ cxt, obj)))
7a6a85bf
RG
2721 return ret;
2722
2723 TRACEME(("ok (tied)"));
2724
2725 return 0;
2726}
2727
2728/*
2729 * store_tied_item
2730 *
2731 * Stores a reference to an item within a tied structure:
2732 *
2733 * . \$h{key}, stores both the (tied %h) object and 'key'.
2734 * . \$a[idx], stores both the (tied @a) object and 'idx'.
2735 *
2736 * Layout is therefore either:
2737 * SX_TIED_KEY <object> <key>
2738 * SX_TIED_IDX <object> <index>
2739 */
138ec36d 2740static int store_tied_item(pTHX_ stcxt_t *cxt, SV *sv)
7a6a85bf
RG
2741{
2742 MAGIC *mg;
2743 int ret;
2744
43d061fe 2745 TRACEME(("store_tied_item (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
2746
2747 if (!(mg = mg_find(sv, 'p')))
2748 CROAK(("No magic 'p' found while storing reference to tied item"));
2749
2750 /*
2751 * We discriminate between \$h{key} and \$a[idx] via mg_ptr.
2752 */
2753
2754 if (mg->mg_ptr) {
2755 TRACEME(("store_tied_item: storing a ref to a tied hash item"));
2756 PUTMARK(SX_TIED_KEY);
9e21b3d0 2757 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
7a6a85bf 2758
138ec36d 2759 if ((ret = store(aTHX_ cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2760 return ret;
2761
9e21b3d0 2762 TRACEME(("store_tied_item: storing PTR 0x%"UVxf, PTR2UV(mg->mg_ptr)));
7a6a85bf 2763
138ec36d 2764 if ((ret = store(aTHX_ cxt, (SV *) mg->mg_ptr))) /* Idem, for -Wall */
7a6a85bf
RG
2765 return ret;
2766 } else {
2767 I32 idx = mg->mg_len;
2768
2769 TRACEME(("store_tied_item: storing a ref to a tied array item "));
2770 PUTMARK(SX_TIED_IDX);
9e21b3d0 2771 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
7a6a85bf 2772
138ec36d 2773 if ((ret = store(aTHX_ cxt, mg->mg_obj))) /* Idem, for -Wall */
7a6a85bf
RG
2774 return ret;
2775
2776 TRACEME(("store_tied_item: storing IDX %d", idx));
2777
2778 WLEN(idx);
2779 }
2780
2781 TRACEME(("ok (tied item)"));
2782
2783 return 0;
2784}
2785
2786/*
2787 * store_hook -- dispatched manually, not via sv_store[]
2788 *
2789 * The blessed SV is serialized by a hook.
2790 *
2791 * Simple Layout is:
2792 *
2793 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2794 *
2795 * where <flags> indicates how long <len>, <len2> and <len3> are, whether
2796 * the trailing part [] is present, the type of object (scalar, array or hash).
2797 * There is also a bit which says how the classname is stored between:
2798 *
2799 * <len> <classname>
2800 * <index>
2801 *
2802 * and when the <index> form is used (classname already seen), the "large
2803 * classname" bit in <flags> indicates how large the <index> is.
2804 *
2805 * The serialized string returned by the hook is of length <len2> and comes
2806 * next. It is an opaque string for us.
2807 *
2808 * Those <len3> object IDs which are listed last represent the extra references
2809 * not directly serialized by the hook, but which are linked to the object.
2810 *
2811 * When recursion is mandated to resolve object-IDs not yet seen, we have
2812 * instead, with <header> being flags with bits set to indicate the object type
2813 * and that recursion was indeed needed:
2814 *
2815 * SX_HOOK <header> <object> <header> <object> <flags>
2816 *
2817 * that same header being repeated between serialized objects obtained through
2818 * recursion, until we reach flags indicating no recursion, at which point
2819 * we know we've resynchronized with a single layout, after <flags>.
b12202d0
JH
2820 *
2821 * When storing a blessed ref to a tied variable, the following format is
2822 * used:
2823 *
2824 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
2825 *
2826 * The first <flags> indication carries an object of type SHT_EXTRA, and the
2827 * real object type is held in the <extra> flag. At the very end of the
2828 * serialization stream, the underlying magic object is serialized, just like
2829 * any other tied variable.
7a6a85bf 2830 */
f0ffaed8 2831static int store_hook(
138ec36d 2832 pTHX_
f0ffaed8
JH
2833 stcxt_t *cxt,
2834 SV *sv,
2835 int type,
2836 HV *pkg,
2837 SV *hook)
7a6a85bf
RG
2838{
2839 I32 len;
0723351e 2840 char *classname;
7a6a85bf
RG
2841 STRLEN len2;
2842 SV *ref;
2843 AV *av;
2844 SV **ary;
2845 int count; /* really len3 + 1 */
2846 unsigned char flags;
2847 char *pv;
2848 int i;
2849 int recursed = 0; /* counts recursion */
2850 int obj_type; /* object type, on 2 bits */
2851 I32 classnum;
2852 int ret;
2853 int clone = cxt->optype & ST_CLONE;
e993d95c
JH
2854 char mtype = '\0'; /* for blessed ref to tied structures */
2855 unsigned char eflags = '\0'; /* used when object type is SHT_EXTRA */
7a6a85bf 2856
bfcb3514 2857 TRACEME(("store_hook, classname \"%s\", tagged #%d", HvNAME_get(pkg), cxt->tagnum));
7a6a85bf
RG
2858
2859 /*
2860 * Determine object type on 2 bits.
2861 */
2862
2863 switch (type) {
cc4aa37c 2864 case svis_REF:
7a6a85bf
RG
2865 case svis_SCALAR:
2866 obj_type = SHT_SCALAR;
2867 break;
2868 case svis_ARRAY:
2869 obj_type = SHT_ARRAY;
2870 break;
2871 case svis_HASH:
2872 obj_type = SHT_HASH;
2873 break;
b12202d0
JH
2874 case svis_TIED:
2875 /*
2876 * Produced by a blessed ref to a tied data structure, $o in the
2877 * following Perl code.
2878 *
2879 * my %h;
2880 * tie %h, 'FOO';
2881 * my $o = bless \%h, 'BAR';
2882 *
2883 * Signal the tie-ing magic by setting the object type as SHT_EXTRA
2884 * (since we have only 2 bits in <flags> to store the type), and an
2885 * <extra> byte flag will be emitted after the FIRST <flags> in the
6dfee1ec 2886 * stream, carrying what we put in 'eflags'.
b12202d0
JH
2887 */
2888 obj_type = SHT_EXTRA;
2889 switch (SvTYPE(sv)) {
2890 case SVt_PVHV:
2891 eflags = (unsigned char) SHT_THASH;
2892 mtype = 'P';
2893 break;
2894 case SVt_PVAV:
2895 eflags = (unsigned char) SHT_TARRAY;
2896 mtype = 'P';
2897 break;
2898 default:
2899 eflags = (unsigned char) SHT_TSCALAR;
2900 mtype = 'q';
2901 break;
2902 }
2903 break;
7a6a85bf
RG
2904 default:
2905 CROAK(("Unexpected object type (%d) in store_hook()", type));
2906 }
2907 flags = SHF_NEED_RECURSE | obj_type;
2908
bfcb3514 2909 classname = HvNAME_get(pkg);
0723351e 2910 len = strlen(classname);
7a6a85bf
RG
2911
2912 /*
2913 * To call the hook, we need to fake a call like:
2914 *
2915 * $object->STORABLE_freeze($cloning);
2916 *
2917 * but we don't have the $object here. For instance, if $object is
6dfee1ec 2918 * a blessed array, what we have in 'sv' is the array, and we can't
7a6a85bf
RG
2919 * call a method on those.
2920 *
2921 * Therefore, we need to create a temporary reference to the object and
2922 * make the call on that reference.
2923 */
2924
0723351e 2925 TRACEME(("about to call STORABLE_freeze on class %s", classname));
7a6a85bf 2926
27cc3b5a 2927 ref = newRV_inc(sv); /* Temporary reference */
138ec36d 2928 av = array_call(aTHX_ ref, hook, clone); /* @a = $object->STORABLE_freeze($c) */
7a6a85bf
RG
2929 SvREFCNT_dec(ref); /* Reclaim temporary reference */
2930
2931 count = AvFILLp(av) + 1;
2932 TRACEME(("store_hook, array holds %d items", count));
2933
2934 /*
2935 * If they return an empty list, it means they wish to ignore the
2936 * hook for this class (and not just this instance -- that's for them
2937 * to handle if they so wish).
2938 *
2939 * Simply disable the cached entry for the hook (it won't be recomputed
2940 * since it's present in the cache) and recurse to store_blessed().
2941 */
2942
2943 if (!count) {
2944 /*
2945 * They must not change their mind in the middle of a serialization.
2946 */
2947
0723351e 2948 if (hv_fetch(cxt->hclass, classname, len, FALSE))
7a6a85bf 2949 CROAK(("Too late to ignore hooks for %s class \"%s\"",
0723351e 2950 (cxt->optype & ST_CLONE) ? "cloning" : "storing", classname));
7a6a85bf 2951
138ec36d 2952 pkg_hide(aTHX_ cxt->hook, pkg, "STORABLE_freeze");
7a6a85bf 2953
138ec36d 2954 ASSERT(!pkg_can(aTHX_ cxt->hook, pkg, "STORABLE_freeze"), ("hook invisible"));
0723351e 2955 TRACEME(("ignoring STORABLE_freeze in class \"%s\"", classname));
7a6a85bf 2956
138ec36d 2957 return store_blessed(aTHX_ cxt, sv, type, pkg);
7a6a85bf
RG
2958 }
2959
2960 /*
2961 * Get frozen string.
2962 */
2963
2964 ary = AvARRAY(av);
2965 pv = SvPV(ary[0], len2);
2f796f32
AMS
2966 /* We can't use pkg_can here because it only caches one method per
2967 * package */
2968 {
2969 GV* gv = gv_fetchmethod_autoload(pkg, "STORABLE_attach", FALSE);
2970 if (gv && isGV(gv)) {
2971 if (count > 1)
2972 CROAK(("Freeze cannot return references if %s class is using STORABLE_attach", classname));
2973 goto check_done;
2974 }
2975 }
7a6a85bf
RG
2976
2977 /*
7a6a85bf
RG
2978 * If they returned more than one item, we need to serialize some
2979 * extra references if not already done.
2980 *
10ffa93f 2981 * Loop over the array, starting at position #1, and for each item,
7a6a85bf
RG
2982 * ensure it is a reference, serialize it if not already done, and
2983 * replace the entry with the tag ID of the corresponding serialized
2984 * object.
2985 *
2986 * We CHEAT by not calling av_fetch() and read directly within the
2987 * array, for speed.
2988 */
2989
2990 for (i = 1; i < count; i++) {
ab923da1
NC
2991#ifdef USE_PTR_TABLE
2992 char *fake_tag;
2993#else
7a6a85bf 2994 SV **svh;
ab923da1 2995#endif
90826881
JH
2996 SV *rsv = ary[i];
2997 SV *xsv;
ab923da1 2998 SV *tag;
90826881 2999 AV *av_hook = cxt->hook_seen;
7a6a85bf 3000
90826881
JH
3001 if (!SvROK(rsv))
3002 CROAK(("Item #%d returned by STORABLE_freeze "
0723351e 3003 "for %s is not a reference", i, classname));
90826881 3004 xsv = SvRV(rsv); /* Follow ref to know what to look for */
7a6a85bf
RG
3005
3006 /*
3007 * Look in hseen and see if we have a tag already.
3008 * Serialize entry if not done already, and get its tag.
3009 */
ab923da1
NC
3010
3011#ifdef USE_PTR_TABLE
3012 /* Fakery needed because ptr_table_fetch returns zero for a
3013 failure, whereas the existing code assumes that it can
3014 safely store a tag zero. So for ptr_tables we store tag+1
3015 */
ea17c9b6 3016 if ((fake_tag = (char *)ptr_table_fetch(cxt->pseen, xsv)))
ab923da1
NC
3017 goto sv_seen; /* Avoid moving code too far to the right */
3018#else
13689cfe 3019 if ((svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE)))
7a6a85bf 3020 goto sv_seen; /* Avoid moving code too far to the right */
ab923da1 3021#endif
7a6a85bf 3022
9e21b3d0 3023 TRACEME(("listed object %d at 0x%"UVxf" is unknown", i-1, PTR2UV(xsv)));
7a6a85bf
RG
3024
3025 /*
3026 * We need to recurse to store that object and get it to be known
3027 * so that we can resolve the list of object-IDs at retrieve time.
3028 *
3029 * The first time we do this, we need to emit the proper header
3030 * indicating that we recursed, and what the type of object is (the
3031 * object we're storing via a user-hook). Indeed, during retrieval,
3032 * we'll have to create the object before recursing to retrieve the
3033 * others, in case those would point back at that object.
3034 */
3035
b12202d0
JH
3036 /* [SX_HOOK] <flags> [<extra>] <object>*/
3037 if (!recursed++) {
7a6a85bf 3038 PUTMARK(SX_HOOK);
b12202d0
JH
3039 PUTMARK(flags);
3040 if (obj_type == SHT_EXTRA)
3041 PUTMARK(eflags);
3042 } else
3043 PUTMARK(flags);
7a6a85bf 3044
138ec36d 3045 if ((ret = store(aTHX_ cxt, xsv))) /* Given by hook for us to store */
7a6a85bf
RG
3046 return ret;
3047
ab923da1 3048#ifdef USE_PTR_TABLE
ea17c9b6 3049 fake_tag = (char *)ptr_table_fetch(cxt->pseen, xsv);
ab923da1
NC
3050 if (!sv)
3051 CROAK(("Could not serialize item #%d from hook in %s", i, classname));
3052#else
7a6a85bf
RG
3053 svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE);
3054 if (!svh)
0723351e 3055 CROAK(("Could not serialize item #%d from hook in %s", i, classname));
ab923da1 3056#endif
7a6a85bf 3057 /*
6dfee1ec 3058 * It was the first time we serialized 'xsv'.
90826881
JH
3059 *
3060 * Keep this SV alive until the end of the serialization: if we
3061 * disposed of it right now by decrementing its refcount, and it was
3062 * a temporary value, some next temporary value allocated during
3063 * another STORABLE_freeze might take its place, and we'd wrongly
3064 * assume that new SV was already serialized, based on its presence
3065 * in cxt->hseen.
3066 *
3067 * Therefore, push it away in cxt->hook_seen.
7a6a85bf
RG
3068 */
3069
90826881
JH
3070 av_store(av_hook, AvFILLp(av_hook)+1, SvREFCNT_inc(xsv));
3071
7a6a85bf 3072 sv_seen:
90826881 3073 /*
6dfee1ec 3074 * Dispose of the REF they returned. If we saved the 'xsv' away
90826881
JH
3075 * in the array of returned SVs, that will not cause the underlying
3076 * referenced SV to be reclaimed.
3077 */
3078
3079 ASSERT(SvREFCNT(xsv) > 1, ("SV will survive disposal of its REF"));
3080 SvREFCNT_dec(rsv); /* Dispose of reference */
3081
3082 /*
3083 * Replace entry with its tag (not a real SV, so no refcnt increment)
3084 */
3085
ab923da1
NC
3086#ifdef USE_PTR_TABLE
3087 tag = (SV *)--fake_tag;
3088#else
3089 tag = *svh;
3090#endif
672ac946 3091 ary[i] = tag;
76edffbb 3092 TRACEME(("listed object %d at 0x%"UVxf" is tag #%"UVuf,
ab923da1 3093 i-1, PTR2UV(xsv), PTR2UV(tag)));
7a6a85bf
RG
3094 }
3095
3096 /*
dd19458b
JH
3097 * Allocate a class ID if not already done.
3098 *
3099 * This needs to be done after the recursion above, since at retrieval
3100 * time, we'll see the inner objects first. Many thanks to
3101 * Salvador Ortiz Garcia <sog@msg.com.mx> who spot that bug and
3102 * proposed the right fix. -- RAM, 15/09/2000
3103 */
3104
2f796f32 3105check_done:
0723351e
NC
3106 if (!known_class(aTHX_ cxt, classname, len, &classnum)) {
3107 TRACEME(("first time we see class %s, ID = %d", classname, classnum));
dd19458b
JH
3108 classnum = -1; /* Mark: we must store classname */
3109 } else {
0723351e 3110 TRACEME(("already seen class %s, ID = %d", classname, classnum));
dd19458b
JH
3111 }
3112
3113 /*
7a6a85bf
RG
3114 * Compute leading flags.
3115 */
3116
3117 flags = obj_type;
3118 if (((classnum == -1) ? len : classnum) > LG_SCALAR)
3119 flags |= SHF_LARGE_CLASSLEN;
3120 if (classnum != -1)
3121 flags |= SHF_IDX_CLASSNAME;
3122 if (len2 > LG_SCALAR)
3123 flags |= SHF_LARGE_STRLEN;
3124 if (count > 1)
3125 flags |= SHF_HAS_LIST;
3126 if (count > (LG_SCALAR + 1))
3127 flags |= SHF_LARGE_LISTLEN;
3128
3129 /*
3130 * We're ready to emit either serialized form:
3131 *
3132 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
3133 * SX_HOOK <flags> <index> <len2> <str> [<len3> <object-IDs>]
3134 *
3135 * If we recursed, the SX_HOOK has already been emitted.
3136 */
3137
9e21b3d0
JH
3138 TRACEME(("SX_HOOK (recursed=%d) flags=0x%x "
3139 "class=%"IVdf" len=%"IVdf" len2=%"IVdf" len3=%d",
d67b2c17 3140 recursed, flags, (IV)classnum, (IV)len, (IV)len2, count-1));
7a6a85bf 3141
b12202d0
JH
3142 /* SX_HOOK <flags> [<extra>] */
3143 if (!recursed) {
7a6a85bf 3144 PUTMARK(SX_HOOK);
b12202d0
JH
3145 PUTMARK(flags);
3146 if (obj_type == SHT_EXTRA)
3147 PUTMARK(eflags);
3148 } else
3149 PUTMARK(flags);
7a6a85bf
RG
3150
3151 /* <len> <classname> or <index> */
3152 if (flags & SHF_IDX_CLASSNAME) {
3153 if (flags & SHF_LARGE_CLASSLEN)
3154 WLEN(classnum);
3155 else {
3156 unsigned char cnum = (unsigned char) classnum;
3157 PUTMARK(cnum);
3158 }
3159 } else {
3160 if (flags & SHF_LARGE_CLASSLEN)
3161 WLEN(len);
3162 else {
3163 unsigned char clen = (unsigned char) len;
3164 PUTMARK(clen);
3165 }
0723351e 3166 WRITE(classname, len); /* Final \0 is omitted */
7a6a85bf
RG
3167 }
3168
3169 /* <len2> <frozen-str> */
cc964657
JH
3170 if (flags & SHF_LARGE_STRLEN) {
3171 I32 wlen2 = len2; /* STRLEN might be 8 bytes */
3172 WLEN(wlen2); /* Must write an I32 for 64-bit machines */
3173 } else {
7a6a85bf
RG
3174 unsigned char clen = (unsigned char) len2;
3175 PUTMARK(clen);
3176 }
3177 if (len2)
7c436af3 3178 WRITE(pv, (SSize_t)len2); /* Final \0 is omitted */
7a6a85bf
RG
3179
3180 /* [<len3> <object-IDs>] */
3181 if (flags & SHF_HAS_LIST) {
3182 int len3 = count - 1;
3183 if (flags & SHF_LARGE_LISTLEN)
3184 WLEN(len3);
3185 else {
3186 unsigned char clen = (unsigned char) len3;
3187 PUTMARK(clen);
3188 }
3189
3190 /*
3191 * NOTA BENE, for 64-bit machines: the ary[i] below does not yield a
3192 * real pointer, rather a tag number, well under the 32-bit limit.
3193 */
3194
3195 for (i = 1; i < count; i++) {
3196 I32 tagval = htonl(LOW_32BITS(ary[i]));
9e21b3d0 3197 WRITE_I32(tagval);
7a6a85bf
RG
3198 TRACEME(("object %d, tag #%d", i-1, ntohl(tagval)));
3199 }
3200 }
3201
3202 /*
3203 * Free the array. We need extra care for indices after 0, since they
3204 * don't hold real SVs but integers cast.
3205 */
3206
3207 if (count > 1)
3208 AvFILLp(av) = 0; /* Cheat, nothing after 0 interests us */
3209 av_undef(av);
3210 sv_free((SV *) av);
3211
b12202d0
JH
3212 /*
3213 * If object was tied, need to insert serialization of the magic object.
3214 */
3215
3216 if (obj_type == SHT_EXTRA) {
3217 MAGIC *mg;
3218
3219 if (!(mg = mg_find(sv, mtype))) {
3220 int svt = SvTYPE(sv);
3221 CROAK(("No magic '%c' found while storing ref to tied %s with hook",
3222 mtype, (svt == SVt_PVHV) ? "hash" :
3223 (svt == SVt_PVAV) ? "array" : "scalar"));
3224 }
3225
3226 TRACEME(("handling the magic object 0x%"UVxf" part of 0x%"UVxf,
3227 PTR2UV(mg->mg_obj), PTR2UV(sv)));
3228
3229 /*
3230 * [<magic object>]
3231 */
3232
138ec36d 3233 if ((ret = store(aTHX_ cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
b12202d0
JH
3234 return ret;
3235 }
3236
7a6a85bf
RG
3237 return 0;
3238}
3239
3240/*
3241 * store_blessed -- dispatched manually, not via sv_store[]
3242 *
3243 * Check whether there is a STORABLE_xxx hook defined in the class or in one
3244 * of its ancestors. If there is, then redispatch to store_hook();
3245 *
3246 * Otherwise, the blessed SV is stored using the following layout:
3247 *
3248 * SX_BLESS <flag> <len> <classname> <object>
3249 *
3250 * where <flag> indicates whether <len> is stored on 0 or 4 bytes, depending
3251 * on the high-order bit in flag: if 1, then length follows on 4 bytes.
3252 * Otherwise, the low order bits give the length, thereby giving a compact
3253 * representation for class names less than 127 chars long.
3254 *
3255 * Each <classname> seen is remembered and indexed, so that the next time
3256 * an object in the blessed in the same <classname> is stored, the following
3257 * will be emitted:
3258 *
3259 * SX_IX_BLESS <flag> <index> <object>
3260 *
3261 * where <index> is the classname index, stored on 0 or 4 bytes depending
3262 * on the high-order bit in flag (same encoding as above for <len>).
3263 */
f0ffaed8 3264static int store_blessed(
138ec36d 3265 pTHX_
f0ffaed8
JH
3266 stcxt_t *cxt,
3267 SV *sv,
3268 int type,
3269 HV *pkg)
7a6a85bf
RG
3270{
3271 SV *hook;
3272 I32 len;
0723351e 3273 char *classname;
7a6a85bf
RG
3274 I32 classnum;
3275
bfcb3514 3276 TRACEME(("store_blessed, type %d, class \"%s\"", type, HvNAME_get(pkg)));
7a6a85bf
RG
3277
3278 /*
3279 * Look for a hook for this blessed SV and redirect to store_hook()
3280 * if needed.
3281 */
3282
138ec36d 3283 hook = pkg_can(aTHX_ cxt->hook, pkg, "STORABLE_freeze");
7a6a85bf 3284 if (hook)
138ec36d 3285 return store_hook(aTHX_ cxt, sv, type, pkg, hook);
7a6a85bf
RG
3286
3287 /*
3288 * This is a blessed SV without any serialization hook.
3289 */
3290
bfcb3514 3291 classname = HvNAME_get(pkg);
0723351e 3292 len = strlen(classname);
7a6a85bf 3293
43d061fe 3294 TRACEME(("blessed 0x%"UVxf" in %s, no hook: tagged #%d",
5e081687 3295 PTR2UV(sv), classname, cxt->tagnum));
7a6a85bf
RG
3296
3297 /*
3298 * Determine whether it is the first time we see that class name (in which
3299 * case it will be stored in the SX_BLESS form), or whether we already
3300 * saw that class name before (in which case the SX_IX_BLESS form will be
3301 * used).
3302 */
3303
0723351e
NC
3304 if (known_class(aTHX_ cxt, classname, len, &classnum)) {
3305 TRACEME(("already seen class %s, ID = %d", classname, classnum));
7a6a85bf
RG
3306 PUTMARK(SX_IX_BLESS);
3307 if (classnum <= LG_BLESS) {
3308 unsigned char cnum = (unsigned char) classnum;
3309 PUTMARK(cnum);
3310 } else {
3311 unsigned char flag = (unsigned char) 0x80;
3312 PUTMARK(flag);
3313 WLEN(classnum);
3314 }
3315 } else {
0723351e 3316 TRACEME(("first time we see class %s, ID = %d", classname, classnum));
7a6a85bf
RG
3317 PUTMARK(SX_BLESS);
3318 if (len <= LG_BLESS) {
3319 unsigned char clen = (unsigned char) len;
3320 PUTMARK(clen);
3321 } else {
3322 unsigned char flag = (unsigned char) 0x80;
3323 PUTMARK(flag);
3324 WLEN(len); /* Don't BER-encode, this should be rare */
3325 }
0723351e 3326 WRITE(classname, len); /* Final \0 is omitted */
7a6a85bf
RG
3327 }
3328
3329 /*
3330 * Now emit the <object> part.
3331 */
3332
138ec36d 3333 return SV_STORE(type)(aTHX_ cxt, sv);
7a6a85bf
RG
3334}
3335
3336/*
3337 * store_other
3338 *
3339 * We don't know how to store the item we reached, so return an error condition.
3340 * (it's probably a GLOB, some CODE reference, etc...)
3341 *
6dfee1ec 3342 * If they defined the 'forgive_me' variable at the Perl level to some
7a6a85bf
RG
3343 * true value, then don't croak, just warn, and store a placeholder string
3344 * instead.
3345 */
138ec36d 3346static int store_other(pTHX_ stcxt_t *cxt, SV *sv)
7a6a85bf 3347{
cc964657 3348 I32 len;
27da23d5 3349 char buf[80];
7a6a85bf
RG
3350
3351 TRACEME(("store_other"));
3352
3353 /*
3354 * Fetch the value from perl only once per store() operation.
3355 */
3356
3357 if (
3358 cxt->forgive_me == 0 ||
3359 (cxt->forgive_me < 0 && !(cxt->forgive_me =
3509f647 3360 SvTRUE(perl_get_sv("Storable::forgive_me", GV_ADD)) ? 1 : 0))
7a6a85bf
RG
3361 )
3362 CROAK(("Can't store %s items", sv_reftype(sv, FALSE)));
3363
43d061fe
JH
3364 warn("Can't store item %s(0x%"UVxf")",
3365 sv_reftype(sv, FALSE), PTR2UV(sv));
7a6a85bf
RG
3366
3367 /*
3368 * Store placeholder string as a scalar instead...
3369 */
3370
13689cfe 3371 (void) sprintf(buf, "You lost %s(0x%"UVxf")%c", sv_reftype(sv, FALSE),
e993d95c 3372 PTR2UV(sv), (char) 0);
7a6a85bf
RG
3373
3374 len = strlen(buf);
3375 STORE_SCALAR(buf, len);
1cf92b12 3376 TRACEME(("ok (dummy \"%s\", length = %"IVdf")", buf, (IV) len));
7a6a85bf
RG
3377
3378 return 0;
3379}
3380
3381/***
3382 *** Store driving routines
3383 ***/
3384
3385/*
3386 * sv_type
3387 *
3388 * WARNING: partially duplicates Perl's sv_reftype for speed.
3389 *
3390 * Returns the type of the SV, identified by an integer. That integer
3391 * may then be used to index the dynamic routine dispatch table.
3392 */
138ec36d 3393static int sv_type(pTHX_ SV *sv)
7a6a85bf
RG
3394{
3395 switch (SvTYPE(sv)) {
3396 case SVt_NULL:
4df7f6af 3397#if PERL_VERSION <= 10
7a6a85bf 3398 case SVt_IV:
4df7f6af 3399#endif
7a6a85bf
RG
3400 case SVt_NV:
3401 /*
3402 * No need to check for ROK, that can't be set here since there
3403 * is no field capable of hodling the xrv_rv reference.
3404 */
3405 return svis_SCALAR;
3406 case SVt_PV:
4df7f6af 3407#if PERL_VERSION <= 10
7a6a85bf 3408 case SVt_RV:
4df7f6af
NC
3409#else
3410 case SVt_IV:
3411#endif
7a6a85bf
RG
3412 case SVt_PVIV:
3413 case SVt_PVNV:
3414 /*
3415 * Starting from SVt_PV, it is possible to have the ROK flag
3416 * set, the pointer to the other SV being either stored in
3417 * the xrv_rv (in the case of a pure SVt_RV), or as the
3418 * xpv_pv field of an SVt_PV and its heirs.
3419 *
3420 * However, those SV cannot be magical or they would be an
3421 * SVt_PVMG at least.
3422 */
3423 return SvROK(sv) ? svis_REF : svis_SCALAR;
3424 case SVt_PVMG:
3425 case SVt_PVLV: /* Workaround for perl5.004_04 "LVALUE" bug */
3426 if (SvRMAGICAL(sv) && (mg_find(sv, 'p')))
3427 return svis_TIED_ITEM;
3428 /* FALL THROUGH */
cecf5685 3429#if PERL_VERSION < 9
7a6a85bf 3430 case SVt_PVBM:
cecf5685 3431#endif
7a6a85bf
RG
3432 if (SvRMAGICAL(sv) && (mg_find(sv, 'q')))
3433 return svis_TIED;
3434 return SvROK(sv) ? svis_REF : svis_SCALAR;
3435 case SVt_PVAV:
3436 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
3437 return svis_TIED;
3438 return svis_ARRAY;
3439 case SVt_PVHV:
3440 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
3441 return svis_TIED;
3442 return svis_HASH;
464b080a
SR
3443 case SVt_PVCV:
3444 return svis_CODE;
cecf5685
NC
3445#if PERL_VERSION > 8
3446 /* case SVt_BIND: */
3447#endif
7a6a85bf
RG
3448 default:
3449 break;
3450 }
3451
3452 return svis_OTHER;
3453}
3454
3455/*
3456 * store
3457 *
3458 * Recursively store objects pointed to by the sv to the specified file.
3459 *
3460 * Layout is <content> or SX_OBJECT <tagnum> if we reach an already stored
3461 * object (one for which storage has started -- it may not be over if we have
3462 * a self-referenced structure). This data set forms a stored <object>.
3463 */
138ec36d 3464static int store(pTHX_ stcxt_t *cxt, SV *sv)
7a6a85bf
RG
3465{
3466 SV **svh;
3467 int ret;
7a6a85bf 3468 int type;
ab923da1
NC
3469#ifdef USE_PTR_TABLE
3470 struct ptr_tbl *pseen = cxt->pseen;
3471#else
43d061fe 3472 HV *hseen = cxt->hseen;
ab923da1 3473#endif
7a6a85bf 3474
43d061fe 3475 TRACEME(("store (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
3476
3477 /*
3478 * If object has already been stored, do not duplicate data.
3479 * Simply emit the SX_OBJECT marker followed by its tag data.
3480 * The tag is always written in network order.
3481 *
3482 * NOTA BENE, for 64-bit machines: the "*svh" below does not yield a
3483 * real pointer, rather a tag number (watch the insertion code below).
464b080a 3484 * That means it probably safe to assume it is well under the 32-bit limit,
7a6a85bf
RG
3485 * and makes the truncation safe.
3486 * -- RAM, 14/09/1999
3487 */
3488
ab923da1 3489#ifdef USE_PTR_TABLE
ea17c9b6 3490 svh = (SV **)ptr_table_fetch(pseen, sv);
ab923da1 3491#else
7a6a85bf 3492 svh = hv_fetch(hseen, (char *) &sv, sizeof(sv), FALSE);
ab923da1 3493#endif
7a6a85bf 3494 if (svh) {
dfd91409
NC
3495 I32 tagval;
3496
3497 if (sv == &PL_sv_undef) {
3498 /* We have seen PL_sv_undef before, but fake it as
3499 if we have not.
3500
3501 Not the simplest solution to making restricted
3502 hashes work on 5.8.0, but it does mean that
3503 repeated references to the one true undef will
3504 take up less space in the output file.
3505 */
3506 /* Need to jump past the next hv_store, because on the
3507 second store of undef the old hash value will be
17625bd2 3508 SvREFCNT_dec()ed, and as Storable cheats horribly
dfd91409
NC
3509 by storing non-SVs in the hash a SEGV will ensure.
3510 Need to increase the tag number so that the
3511 receiver has no idea what games we're up to. This
3512 special casing doesn't affect hooks that store
3513 undef, as the hook routine does its own lookup into
3514 hseen. Also this means that any references back
3515 to PL_sv_undef (from the pathological case of hooks
3516 storing references to it) will find the seen hash
3517 entry for the first time, as if we didn't have this
3518 hackery here. (That hseen lookup works even on 5.8.0
3519 because it's a key of &PL_sv_undef and a value
3520 which is a tag number, not a value which is
3521 PL_sv_undef.) */
3522 cxt->tagnum++;
3523 type = svis_SCALAR;
3524 goto undef_special_case;
3525 }
3526
ab923da1
NC
3527#ifdef USE_PTR_TABLE
3528 tagval = htonl(LOW_32BITS(((char *)svh)-1));
3529#else
dfd91409 3530 tagval = htonl(LOW_32BITS(*svh));
ab923da1 3531#endif
7a6a85bf 3532
9e21b3d0 3533 TRACEME(("object 0x%"UVxf" seen as #%d", PTR2UV(sv), ntohl(tagval)));
7a6a85bf
RG
3534
3535 PUTMARK(SX_OBJECT);
9e21b3d0 3536 WRITE_I32(tagval);
7a6a85bf
RG
3537 return 0;
3538 }
3539
3540 /*
3541 * Allocate a new tag and associate it with the address of the sv being
3542 * stored, before recursing...
3543 *
3544 * In order to avoid creating new SvIVs to hold the tagnum we just
d1be9408 3545 * cast the tagnum to an SV pointer and store that in the hash. This
7a6a85bf
RG
3546 * means that we must clean up the hash manually afterwards, but gives
3547 * us a 15% throughput increase.
3548 *
7a6a85bf
RG
3549 */
3550
3551 cxt->tagnum++;
ab923da1
NC
3552#ifdef USE_PTR_TABLE
3553 ptr_table_store(pseen, sv, INT2PTR(SV*, 1 + cxt->tagnum));
3554#else
7a6a85bf 3555 if (!hv_store(hseen,
3341c981 3556 (char *) &sv, sizeof(sv), INT2PTR(SV*, cxt->tagnum), 0))
7a6a85bf 3557 return -1;
ab923da1 3558#endif
7a6a85bf
RG
3559
3560 /*
6dfee1ec 3561 * Store 'sv' and everything beneath it, using appropriate routine.
7a6a85bf
RG
3562 * Abort immediately if we get a non-zero status back.
3563 */
3564
138ec36d 3565 type = sv_type(aTHX_ sv);
7a6a85bf 3566