2 * Store and retrieve mechanism.
6 * $Id: Storable.xs,v 1.0.1.10 2001/08/28 21:52:14 ram Exp $
8 * Copyright (c) 1995-2000, Raphael Manfredi
10 * You may redistribute only under the same terms as Perl 5, as specified
11 * in the README file that comes with the distribution.
13 * $Log: Storable.xs,v $
14 * Revision 1.0.1.10 2001/08/28 21:52:14 ram
15 * patch13: removed spurious debugging messages
17 * Revision 1.0.1.9 2001/07/01 11:25:02 ram
18 * patch12: fixed memory corruption on croaks during thaw()
19 * patch12: made code compile cleanly with -Wall (Jarkko Hietaniemi)
20 * patch12: changed tagnum and classnum from I32 to IV in context
22 * Revision 1.0.1.8 2001/03/15 00:20:55 ram
23 * patch11: last version was wrongly compiling with assertions on
25 * Revision 1.0.1.7 2001/02/17 12:25:26 ram
26 * patch8: now bless objects ASAP at retrieve time
27 * patch8: added support for blessed ref to tied structures
29 * Revision 1.0.1.6 2001/01/03 09:40:40 ram
30 * patch7: prototype and casting cleanup
31 * patch7: trace offending package when overloading cannot be restored
32 * patch7: made context cleanup safer to avoid dup freeing
34 * Revision 1.0.1.5 2000/11/05 17:21:24 ram
35 * patch6: fixed severe "object lost" bug for STORABLE_freeze returns
37 * Revision 1.0.1.4 2000/10/26 17:11:04 ram
38 * patch5: auto requires module of blessed ref when STORABLE_thaw misses
40 * Revision 1.0.1.3 2000/09/29 19:49:57 ram
41 * patch3: avoid using "tainted" and "dirty" since Perl remaps them via cpp
43 * Revision 1.0.1.2 2000/09/28 21:43:10 ram
44 * patch2: perls before 5.004_04 lack newSVpvn
46 * Revision 1.0.1.1 2000/09/17 16:47:49 ram
47 * patch1: now only taint retrieved data when source was tainted
48 * patch1: added support for UTF-8 strings
49 * patch1: fixed store hook bug: was allocating class id too soon
51 * Revision 1.0 2000/09/01 19:40:41 ram
52 * Baseline for first official release.
58 #include <patchlevel.h> /* Perl's one, needed since 5.6 */
63 #define DEBUGME /* Debug mode, turns assertions on as well */
64 #define DASSERT /* Assertion mode */
67 #if 0 /* On NetWare USE_PERLIO is not used */
68 #define DEBUGME /* Debug mode, turns assertions on as well */
69 #define DASSERT /* Assertion mode */
74 * Pre PerlIO time when none of USE_PERLIO and PERLIO_IS_STDIO is defined
75 * Provide them with the necessary defines so they can build with pre-5.004.
78 #ifndef PERLIO_IS_STDIO
80 #define PerlIO_getc(x) getc(x)
81 #define PerlIO_putc(f,x) putc(x,f)
82 #define PerlIO_read(x,y,z) fread(y,1,z,x)
83 #define PerlIO_write(x,y,z) fwrite(y,1,z,x)
84 #define PerlIO_stdoutf printf
85 #endif /* PERLIO_IS_STDIO */
86 #endif /* USE_PERLIO */
89 * Earlier versions of perl might be used, we can't assume they have the latest!
92 #ifndef PERL_VERSION /* For perls < 5.6 */
93 #define PERL_VERSION PATCHLEVEL
95 #define newRV_noinc(sv) ((Sv = newRV(sv)), --SvREFCNT(SvRV(Sv)), Sv)
97 #if (PATCHLEVEL <= 4) /* Older perls (<= 5.004) lack PL_ namespace */
98 #define PL_sv_yes sv_yes
99 #define PL_sv_no sv_no
100 #define PL_sv_undef sv_undef
101 #if (SUBVERSION <= 4) /* 5.004_04 has been reported to lack newSVpvn */
102 #define newSVpvn newSVpv
104 #endif /* PATCHLEVEL <= 4 */
105 #ifndef HvSHAREKEYS_off
106 #define HvSHAREKEYS_off(hv) /* Ignore */
108 #ifndef AvFILLp /* Older perls (<=5.003) lack AvFILLp */
109 #define AvFILLp AvFILL
111 typedef double NV; /* Older perls lack the NV type */
112 #define IVdf "ld" /* Various printf formats for Perl types */
116 #define INT2PTR(t,v) (t)(IV)(v)
117 #define PTR2UV(v) (unsigned long)(v)
118 #endif /* PERL_VERSION -- perls < 5.6 */
120 #ifndef NVef /* The following were not part of perl 5.6 */
121 #if defined(USE_LONG_DOUBLE) && \
122 defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
123 #define NVef PERL_PRIeldbl
124 #define NVff PERL_PRIfldbl
125 #define NVgf PERL_PRIgldbl
140 * TRACEME() will only output things when the $Storable::DEBUGME is true.
145 if (SvTRUE(perl_get_sv("Storable::DEBUGME", TRUE))) \
146 { PerlIO_stdoutf x; PerlIO_stdoutf("\n"); } \
153 #define ASSERT(x,y) \
156 PerlIO_stdoutf("ASSERT FAILED (\"%s\", line %d): ", \
157 __FILE__, __LINE__); \
158 PerlIO_stdoutf y; PerlIO_stdoutf("\n"); \
169 #define C(x) ((char) (x)) /* For markers with dynamic retrieval handling */
171 #define SX_OBJECT C(0) /* Already stored object */
172 #define SX_LSCALAR C(1) /* Scalar (large binary) follows (length, data) */
173 #define SX_ARRAY C(2) /* Array forthcominng (size, item list) */
174 #define SX_HASH C(3) /* Hash forthcoming (size, key/value pair list) */
175 #define SX_REF C(4) /* Reference to object forthcoming */
176 #define SX_UNDEF C(5) /* Undefined scalar */
177 #define SX_INTEGER C(6) /* Integer forthcoming */
178 #define SX_DOUBLE C(7) /* Double forthcoming */
179 #define SX_BYTE C(8) /* (signed) byte forthcoming */
180 #define SX_NETINT C(9) /* Integer in network order forthcoming */
181 #define SX_SCALAR C(10) /* Scalar (binary, small) follows (length, data) */
182 #define SX_TIED_ARRAY C(11) /* Tied array forthcoming */
183 #define SX_TIED_HASH C(12) /* Tied hash forthcoming */
184 #define SX_TIED_SCALAR C(13) /* Tied scalar forthcoming */
185 #define SX_SV_UNDEF C(14) /* Perl's immortal PL_sv_undef */
186 #define SX_SV_YES C(15) /* Perl's immortal PL_sv_yes */
187 #define SX_SV_NO C(16) /* Perl's immortal PL_sv_no */
188 #define SX_BLESS C(17) /* Object is blessed */
189 #define SX_IX_BLESS C(18) /* Object is blessed, classname given by index */
190 #define SX_HOOK C(19) /* Stored via hook, user-defined */
191 #define SX_OVERLOAD C(20) /* Overloaded reference */
192 #define SX_TIED_KEY C(21) /* Tied magic key forthcoming */
193 #define SX_TIED_IDX C(22) /* Tied magic index forthcoming */
194 #define SX_UTF8STR C(23) /* UTF-8 string forthcoming (small) */
195 #define SX_LUTF8STR C(24) /* UTF-8 string forthcoming (large) */
196 #define SX_FLAG_HASH C(25) /* Hash with flags forthcoming (size, flags, key/flags/value triplet list) */
197 #define SX_ERROR C(26) /* Error */
200 * Those are only used to retrieve "old" pre-0.6 binary images.
202 #define SX_ITEM 'i' /* An array item introducer */
203 #define SX_IT_UNDEF 'I' /* Undefined array item */
204 #define SX_KEY 'k' /* A hash key introducer */
205 #define SX_VALUE 'v' /* A hash value introducer */
206 #define SX_VL_UNDEF 'V' /* Undefined hash value */
209 * Those are only used to retrieve "old" pre-0.7 binary images
212 #define SX_CLASS 'b' /* Object is blessed, class name length <255 */
213 #define SX_LG_CLASS 'B' /* Object is blessed, class name length >255 */
214 #define SX_STORED 'X' /* End of object */
217 * Limits between short/long length representation.
220 #define LG_SCALAR 255 /* Large scalar length limit */
221 #define LG_BLESS 127 /* Large classname bless limit */
227 #define ST_STORE 0x1 /* Store operation */
228 #define ST_RETRIEVE 0x2 /* Retrieval operation */
229 #define ST_CLONE 0x4 /* Deep cloning operation */
232 * The following structure is used for hash table key retrieval. Since, when
233 * retrieving objects, we'll be facing blessed hash references, it's best
234 * to pre-allocate that buffer once and resize it as the need arises, never
235 * freeing it (keys will be saved away someplace else anyway, so even large
236 * keys are not enough a motivation to reclaim that space).
238 * This structure is also used for memory store/retrieve operations which
239 * happen in a fixed place before being malloc'ed elsewhere if persistency
240 * is required. Hence the aptr pointer.
243 char *arena; /* Will hold hash key strings, resized as needed */
244 STRLEN asiz; /* Size of aforementionned buffer */
245 char *aptr; /* Arena pointer, for in-place read/write ops */
246 char *aend; /* First invalid address */
251 * A hash table records the objects which have already been stored.
252 * Those are referred to as SX_OBJECT in the file, and their "tag" (i.e.
253 * an arbitrary sequence number) is used to identify them.
256 * An array table records the objects which have already been retrieved,
257 * as seen by the tag determind by counting the objects themselves. The
258 * reference to that retrieved object is kept in the table, and is returned
259 * when an SX_OBJECT is found bearing that same tag.
261 * The same processing is used to record "classname" for blessed objects:
262 * indexing by a hash at store time, and via an array at retrieve time.
265 typedef unsigned long stag_t; /* Used by pre-0.6 binary format */
268 * The following "thread-safe" related defines were contributed by
269 * Murray Nesbitt <murray@activestate.com> and integrated by RAM, who
270 * only renamed things a little bit to ensure consistency with surrounding
271 * code. -- RAM, 14/09/1999
273 * The original patch suffered from the fact that the stcxt_t structure
274 * was global. Murray tried to minimize the impact on the code as much as
277 * Starting with 0.7, Storable can be re-entrant, via the STORABLE_xxx hooks
278 * on objects. Therefore, the notion of context needs to be generalized,
282 #define MY_VERSION "Storable(" XS_VERSION ")"
286 * Conditional UTF8 support.
290 #define STORE_UTF8STR(pv, len) STORE_PV_LEN(pv, len, SX_UTF8STR, SX_LUTF8STR)
291 #define HAS_UTF8_SCALARS
293 #define HAS_UTF8_HASHES
296 /* 5.6 perl has utf8 scalars but not hashes */
300 #define STORE_UTF8STR(pv, len) CROAK(("panic: storing UTF8 in non-UTF8 perl"))
303 #define UTF8_CROAK() CROAK(("Cannot retrieve UTF8 data in non-UTF8 perl"))
306 #ifdef HvPLACEHOLDERS
307 #define HAS_RESTRICTED_HASHES
309 #define HVhek_PLACEHOLD 0x200
310 #define RESTRICTED_HASH_CROAK() CROAK(("Cannot retrieve restricted hash"))
314 #define HAS_HASH_KEY_FLAGS
318 * Fields s_tainted and s_dirty are prefixed with s_ because Perl's include
319 * files remap tainted and dirty when threading is enabled. That's bad for
320 * perl to remap such common words. -- RAM, 29/09/00
323 typedef struct stcxt {
324 int entry; /* flags recursion */
325 int optype; /* type of traversal operation */
326 HV *hseen; /* which objects have been seen, store time */
327 AV *hook_seen; /* which SVs were returned by STORABLE_freeze() */
328 AV *aseen; /* which objects have been seen, retrieve time */
329 HV *hclass; /* which classnames have been seen, store time */
330 AV *aclass; /* which classnames have been seen, retrieve time */
331 HV *hook; /* cache for hook methods per class name */
332 IV tagnum; /* incremented at store time for each seen object */
333 IV classnum; /* incremented at store time for each seen classname */
334 int netorder; /* true if network order used */
335 int s_tainted; /* true if input source is tainted, at retrieve time */
336 int forgive_me; /* whether to be forgiving... */
337 int canonical; /* whether to store hashes sorted by key */
338 #ifndef HAS_RESTRICTED_HASHES
339 int derestrict; /* whether to downgrade restrcted hashes */
342 int use_bytes; /* whether to bytes-ify utf8 */
344 int accept_future_minor; /* croak immediately on future minor versions? */
345 int s_dirty; /* context is dirty due to CROAK() -- can be cleaned */
346 int membuf_ro; /* true means membuf is read-only and msaved is rw */
347 struct extendable keybuf; /* for hash key retrieval */
348 struct extendable membuf; /* for memory store/retrieve operations */
349 struct extendable msaved; /* where potentially valid mbuf is saved */
350 PerlIO *fio; /* where I/O are performed, NULL for memory */
351 int ver_major; /* major of version for retrieved object */
352 int ver_minor; /* minor of version for retrieved object */
353 SV *(**retrieve_vtbl)(); /* retrieve dispatch table */
354 SV *prev; /* contexts chained backwards in real recursion */
355 SV *my_sv; /* the blessed scalar who's SvPVX() I am */
358 #define NEW_STORABLE_CXT_OBJ(cxt) \
360 SV *self = newSV(sizeof(stcxt_t) - 1); \
361 SV *my_sv = newRV_noinc(self); \
362 sv_bless(my_sv, gv_stashpv("Storable::Cxt", TRUE)); \
363 cxt = (stcxt_t *)SvPVX(self); \
364 Zero(cxt, 1, stcxt_t); \
365 cxt->my_sv = my_sv; \
368 #if defined(MULTIPLICITY) || defined(PERL_OBJECT) || defined(PERL_CAPI)
370 #if (PATCHLEVEL <= 4) && (SUBVERSION < 68)
372 SV *perinterp_sv = perl_get_sv(MY_VERSION, FALSE)
373 #else /* >= perl5.004_68 */
375 SV *perinterp_sv = *hv_fetch(PL_modglobal, \
376 MY_VERSION, sizeof(MY_VERSION)-1, TRUE)
377 #endif /* < perl5.004_68 */
379 #define dSTCXT_PTR(T,name) \
380 T name = ((perinterp_sv && SvIOK(perinterp_sv) && SvIVX(perinterp_sv) \
381 ? (T)SvPVX(SvRV((SV*)SvIVX(perinterp_sv))) : (T) 0))
384 dSTCXT_PTR(stcxt_t *, cxt)
388 NEW_STORABLE_CXT_OBJ(cxt); \
389 sv_setiv(perinterp_sv, PTR2IV(cxt->my_sv))
391 #define SET_STCXT(x) \
394 sv_setiv(perinterp_sv, PTR2IV(x->my_sv)); \
397 #else /* !MULTIPLICITY && !PERL_OBJECT && !PERL_CAPI */
399 static stcxt_t Context;
400 static stcxt_t *Context_ptr = &Context;
401 #define dSTCXT stcxt_t *cxt = Context_ptr
404 NEW_STORABLE_CXT_OBJ(cxt)
406 #define SET_STCXT(x) Context_ptr = x
408 #endif /* MULTIPLICITY || PERL_OBJECT || PERL_CAPI */
412 * Croaking implies a memory leak, since we don't use setjmp/longjmp
413 * to catch the exit and free memory used during store or retrieve
414 * operations. This is not too difficult to fix, but I need to understand
415 * how Perl does it, and croaking is exceptional anyway, so I lack the
416 * motivation to do it.
418 * The current workaround is to mark the context as dirty when croaking,
419 * so that data structures can be freed whenever we renter Storable code
420 * (but only *then*: it's a workaround, not a fix).
422 * This is also imperfect, because we don't really know how far they trapped
423 * the croak(), and when we were recursing, we won't be able to clean anything
424 * but the topmost context stacked.
427 #define CROAK(x) STMT_START { cxt->s_dirty = 1; croak x; } STMT_END
430 * End of "thread-safe" related definitions.
436 * Keep only the low 32 bits of a pointer (used for tags, which are not
441 #define LOW_32BITS(x) ((I32) (x))
443 #define LOW_32BITS(x) ((I32) ((unsigned long) (x) & 0xffffffffUL))
449 * Hack for Crays, where sizeof(I32) == 8, and which are big-endians.
450 * Used in the WLEN and RLEN macros.
454 #define oI(x) ((I32 *) ((char *) (x) + 4))
455 #define oS(x) ((x) - 4)
456 #define oC(x) (x = 0)
465 * key buffer handling
467 #define kbuf (cxt->keybuf).arena
468 #define ksiz (cxt->keybuf).asiz
472 TRACEME(("** allocating kbuf of 128 bytes")); \
473 New(10003, kbuf, 128, char); \
480 TRACEME(("** extending kbuf to %d bytes (had %d)", x+1, ksiz)); \
481 Renew(kbuf, x+1, char); \
487 * memory buffer handling
489 #define mbase (cxt->membuf).arena
490 #define msiz (cxt->membuf).asiz
491 #define mptr (cxt->membuf).aptr
492 #define mend (cxt->membuf).aend
494 #define MGROW (1 << 13)
495 #define MMASK (MGROW - 1)
497 #define round_mgrow(x) \
498 ((unsigned long) (((unsigned long) (x) + MMASK) & ~MMASK))
499 #define trunc_int(x) \
500 ((unsigned long) ((unsigned long) (x) & ~(sizeof(int)-1)))
501 #define int_aligned(x) \
502 ((unsigned long) (x) == trunc_int(x))
504 #define MBUF_INIT(x) \
507 TRACEME(("** allocating mbase of %d bytes", MGROW)); \
508 New(10003, mbase, MGROW, char); \
515 mend = mbase + msiz; \
518 #define MBUF_TRUNC(x) mptr = mbase + x
519 #define MBUF_SIZE() (mptr - mbase)
525 * Those macros are used in do_retrieve() to save the current memory
526 * buffer into cxt->msaved, before MBUF_LOAD() can be used to retrieve
527 * data from a string.
529 #define MBUF_SAVE_AND_LOAD(in) \
531 ASSERT(!cxt->membuf_ro, ("mbase not already saved")); \
532 cxt->membuf_ro = 1; \
533 TRACEME(("saving mbuf")); \
534 StructCopy(&cxt->membuf, &cxt->msaved, struct extendable); \
538 #define MBUF_RESTORE() \
540 ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
541 cxt->membuf_ro = 0; \
542 TRACEME(("restoring mbuf")); \
543 StructCopy(&cxt->msaved, &cxt->membuf, struct extendable); \
547 * Use SvPOKp(), because SvPOK() fails on tainted scalars.
548 * See store_scalar() for other usage of this workaround.
550 #define MBUF_LOAD(v) \
552 ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
554 CROAK(("Not a scalar string")); \
555 mptr = mbase = SvPV(v, msiz); \
556 mend = mbase + msiz; \
559 #define MBUF_XTEND(x) \
561 int nsz = (int) round_mgrow((x)+msiz); \
562 int offset = mptr - mbase; \
563 ASSERT(!cxt->membuf_ro, ("mbase is not read-only")); \
564 TRACEME(("** extending mbase from %d to %d bytes (wants %d new)", \
566 Renew(mbase, nsz, char); \
568 mptr = mbase + offset; \
569 mend = mbase + nsz; \
572 #define MBUF_CHK(x) \
574 if ((mptr + (x)) > mend) \
578 #define MBUF_GETC(x) \
581 x = (int) (unsigned char) *mptr++; \
587 #define MBUF_GETINT(x) \
590 if ((mptr + 4) <= mend) { \
591 memcpy(oI(&x), mptr, 4); \
597 #define MBUF_GETINT(x) \
599 if ((mptr + sizeof(int)) <= mend) { \
600 if (int_aligned(mptr)) \
603 memcpy(&x, mptr, sizeof(int)); \
604 mptr += sizeof(int); \
610 #define MBUF_READ(x,s) \
612 if ((mptr + (s)) <= mend) { \
613 memcpy(x, mptr, s); \
619 #define MBUF_SAFEREAD(x,s,z) \
621 if ((mptr + (s)) <= mend) { \
622 memcpy(x, mptr, s); \
630 #define MBUF_PUTC(c) \
633 *mptr++ = (char) c; \
636 *mptr++ = (char) c; \
641 #define MBUF_PUTINT(i) \
644 memcpy(mptr, oI(&i), 4); \
648 #define MBUF_PUTINT(i) \
650 MBUF_CHK(sizeof(int)); \
651 if (int_aligned(mptr)) \
654 memcpy(mptr, &i, sizeof(int)); \
655 mptr += sizeof(int); \
659 #define MBUF_WRITE(x,s) \
662 memcpy(mptr, x, s); \
667 * Possible return values for sv_type().
671 #define svis_SCALAR 1
675 #define svis_TIED_ITEM 5
682 #define SHF_TYPE_MASK 0x03
683 #define SHF_LARGE_CLASSLEN 0x04
684 #define SHF_LARGE_STRLEN 0x08
685 #define SHF_LARGE_LISTLEN 0x10
686 #define SHF_IDX_CLASSNAME 0x20
687 #define SHF_NEED_RECURSE 0x40
688 #define SHF_HAS_LIST 0x80
691 * Types for SX_HOOK (last 2 bits in flags).
697 #define SHT_EXTRA 3 /* Read extra byte for type */
700 * The following are held in the "extra byte"...
703 #define SHT_TSCALAR 4 /* 4 + 0 -- tied scalar */
704 #define SHT_TARRAY 5 /* 4 + 1 -- tied array */
705 #define SHT_THASH 6 /* 4 + 2 -- tied hash */
708 * per hash flags for flagged hashes
711 #define SHV_RESTRICTED 0x01
714 * per key flags for flagged hashes
717 #define SHV_K_UTF8 0x01
718 #define SHV_K_WASUTF8 0x02
719 #define SHV_K_LOCKED 0x04
720 #define SHV_K_ISSV 0x08
721 #define SHV_K_PLACEHOLDER 0x10
724 * Before 0.6, the magic string was "perl-store" (binary version number 0).
726 * Since 0.6 introduced many binary incompatibilities, the magic string has
727 * been changed to "pst0" to allow an old image to be properly retrieved by
728 * a newer Storable, but ensure a newer image cannot be retrieved with an
731 * At 0.7, objects are given the ability to serialize themselves, and the
732 * set of markers is extended, backward compatibility is not jeopardized,
733 * so the binary version number could have remained unchanged. To correctly
734 * spot errors if a file making use of 0.7-specific extensions is given to
735 * 0.6 for retrieval, the binary version was moved to "2". And I'm introducing
736 * a "minor" version, to better track this kind of evolution from now on.
739 static char old_magicstr[] = "perl-store"; /* Magic number before 0.6 */
740 static char magicstr[] = "pst0"; /* Used as a magic number */
743 #define STORABLE_BIN_MAJOR 2 /* Binary major "version" */
744 #define STORABLE_BIN_MINOR 5 /* Binary minor "version" */
746 /* If we aren't 5.7.3 or later, we won't be writing out files that use the
747 * new flagged hash introdued in 2.5, so put 2.4 in the binary header to
748 * maximise ease of interoperation with older Storables.
749 * Could we write 2.3s if we're on 5.005_03? NWC
751 #if (PATCHLEVEL <= 6)
752 #define STORABLE_BIN_WRITE_MINOR 4
755 * As of perl 5.7.3, utf8 hash key is introduced.
756 * So this must change -- dankogai
758 #define STORABLE_BIN_WRITE_MINOR 5
759 #endif /* (PATCHLEVEL <= 6) */
762 * Useful store shortcuts...
769 else if (PerlIO_putc(cxt->fio, x) == EOF) \
773 #define WRITE_I32(x) \
775 ASSERT(sizeof(x) == sizeof(I32), ("writing an I32")); \
778 else if (PerlIO_write(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
785 if (cxt->netorder) { \
786 int y = (int) htonl(x); \
789 else if (PerlIO_write(cxt->fio,oI(&y),oS(sizeof(y))) != oS(sizeof(y))) \
794 else if (PerlIO_write(cxt->fio,oI(&x),oS(sizeof(x))) != oS(sizeof(x))) \
799 #define WLEN(x) WRITE_I32(x)
806 else if (PerlIO_write(cxt->fio, x, y) != y) \
810 #define STORE_PV_LEN(pv, len, small, large) \
812 if (len <= LG_SCALAR) { \
813 unsigned char clen = (unsigned char) len; \
825 #define STORE_SCALAR(pv, len) STORE_PV_LEN(pv, len, SX_SCALAR, SX_LSCALAR)
828 * Store undef in arrays and hashes without recursing through store().
830 #define STORE_UNDEF() \
837 * Useful retrieve shortcuts...
841 (cxt->fio ? PerlIO_getc(cxt->fio) : (mptr >= mend ? EOF : (int) *mptr++))
847 else if ((int) (x = PerlIO_getc(cxt->fio)) == EOF) \
851 #define READ_I32(x) \
853 ASSERT(sizeof(x) == sizeof(I32), ("reading an I32")); \
857 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
867 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
870 x = (int) ntohl(x); \
873 #define RLEN(x) READ_I32(x)
880 else if (PerlIO_read(cxt->fio, x, y) != y) \
884 #define SAFEREAD(x,y,z) \
887 MBUF_SAFEREAD(x,y,z); \
888 else if (PerlIO_read(cxt->fio, x, y) != y) { \
895 * This macro is used at retrieve time, to remember where object 'y', bearing a
896 * given tag 'tagnum', has been retrieved. Next time we see an SX_OBJECT marker,
897 * we'll therefore know where it has been retrieved and will be able to
898 * share the same reference, as in the original stored memory image.
900 * We also need to bless objects ASAP for hooks (which may compute "ref $x"
901 * on the objects given to STORABLE_thaw and expect that to be defined), and
902 * also for overloaded objects (for which we might not find the stash if the
903 * object is not blessed yet--this might occur for overloaded objects that
904 * refer to themselves indirectly: if we blessed upon return from a sub
905 * retrieve(), the SX_OBJECT marker we'd found could not have overloading
906 * restored on it because the underlying object would not be blessed yet!).
908 * To achieve that, the class name of the last retrieved object is passed down
909 * recursively, and the first SEEN() call for which the class name is not NULL
910 * will bless the object.
916 if (av_store(cxt->aseen, cxt->tagnum++, SvREFCNT_inc(y)) == 0) \
918 TRACEME(("aseen(#%d) = 0x%"UVxf" (refcnt=%d)", cxt->tagnum-1, \
919 PTR2UV(y), SvREFCNT(y)-1)); \
921 BLESS((SV *) (y), c); \
925 * Bless `s' in `p', via a temporary reference, required by sv_bless().
931 TRACEME(("blessing 0x%"UVxf" in %s", PTR2UV(s), (p))); \
932 stash = gv_stashpv((p), TRUE); \
933 ref = newRV_noinc(s); \
934 (void) sv_bless(ref, stash); \
940 static SV *retrieve(stcxt_t *cxt, char *cname);
943 * Dynamic dispatching table for SV store.
946 static int store_ref(stcxt_t *cxt, SV *sv);
947 static int store_scalar(stcxt_t *cxt, SV *sv);
948 static int store_array(stcxt_t *cxt, AV *av);
949 static int store_hash(stcxt_t *cxt, HV *hv);
950 static int store_tied(stcxt_t *cxt, SV *sv);
951 static int store_tied_item(stcxt_t *cxt, SV *sv);
952 static int store_other(stcxt_t *cxt, SV *sv);
953 static int store_blessed(stcxt_t *cxt, SV *sv, int type, HV *pkg);
955 static int (*sv_store[])(stcxt_t *cxt, SV *sv) = {
956 store_ref, /* svis_REF */
957 store_scalar, /* svis_SCALAR */
958 (int (*)(stcxt_t *cxt, SV *sv)) store_array, /* svis_ARRAY */
959 (int (*)(stcxt_t *cxt, SV *sv)) store_hash, /* svis_HASH */
960 store_tied, /* svis_TIED */
961 store_tied_item, /* svis_TIED_ITEM */
962 store_other, /* svis_OTHER */
965 #define SV_STORE(x) (*sv_store[x])
968 * Dynamic dispatching tables for SV retrieval.
971 static SV *retrieve_lscalar(stcxt_t *cxt, char *cname);
972 static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname);
973 static SV *old_retrieve_array(stcxt_t *cxt, char *cname);
974 static SV *old_retrieve_hash(stcxt_t *cxt, char *cname);
975 static SV *retrieve_ref(stcxt_t *cxt, char *cname);
976 static SV *retrieve_undef(stcxt_t *cxt, char *cname);
977 static SV *retrieve_integer(stcxt_t *cxt, char *cname);
978 static SV *retrieve_double(stcxt_t *cxt, char *cname);
979 static SV *retrieve_byte(stcxt_t *cxt, char *cname);
980 static SV *retrieve_netint(stcxt_t *cxt, char *cname);
981 static SV *retrieve_scalar(stcxt_t *cxt, char *cname);
982 static SV *retrieve_utf8str(stcxt_t *cxt, char *cname);
983 static SV *retrieve_tied_array(stcxt_t *cxt, char *cname);
984 static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname);
985 static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname);
986 static SV *retrieve_other(stcxt_t *cxt, char *cname);
988 static SV *(*sv_old_retrieve[])(stcxt_t *cxt, char *cname) = {
989 0, /* SX_OBJECT -- entry unused dynamically */
990 retrieve_lscalar, /* SX_LSCALAR */
991 old_retrieve_array, /* SX_ARRAY -- for pre-0.6 binaries */
992 old_retrieve_hash, /* SX_HASH -- for pre-0.6 binaries */
993 retrieve_ref, /* SX_REF */
994 retrieve_undef, /* SX_UNDEF */
995 retrieve_integer, /* SX_INTEGER */
996 retrieve_double, /* SX_DOUBLE */
997 retrieve_byte, /* SX_BYTE */
998 retrieve_netint, /* SX_NETINT */
999 retrieve_scalar, /* SX_SCALAR */
1000 retrieve_tied_array, /* SX_ARRAY */
1001 retrieve_tied_hash, /* SX_HASH */
1002 retrieve_tied_scalar, /* SX_SCALAR */
1003 retrieve_other, /* SX_SV_UNDEF not supported */
1004 retrieve_other, /* SX_SV_YES not supported */
1005 retrieve_other, /* SX_SV_NO not supported */
1006 retrieve_other, /* SX_BLESS not supported */
1007 retrieve_other, /* SX_IX_BLESS not supported */
1008 retrieve_other, /* SX_HOOK not supported */
1009 retrieve_other, /* SX_OVERLOADED not supported */
1010 retrieve_other, /* SX_TIED_KEY not supported */
1011 retrieve_other, /* SX_TIED_IDX not supported */
1012 retrieve_other, /* SX_UTF8STR not supported */
1013 retrieve_other, /* SX_LUTF8STR not supported */
1014 retrieve_other, /* SX_FLAG_HASH not supported */
1015 retrieve_other, /* SX_ERROR */
1018 static SV *retrieve_array(stcxt_t *cxt, char *cname);
1019 static SV *retrieve_hash(stcxt_t *cxt, char *cname);
1020 static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname);
1021 static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname);
1022 static SV *retrieve_sv_no(stcxt_t *cxt, char *cname);
1023 static SV *retrieve_blessed(stcxt_t *cxt, char *cname);
1024 static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname);
1025 static SV *retrieve_hook(stcxt_t *cxt, char *cname);
1026 static SV *retrieve_overloaded(stcxt_t *cxt, char *cname);
1027 static SV *retrieve_tied_key(stcxt_t *cxt, char *cname);
1028 static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname);
1029 static SV *retrieve_flag_hash(stcxt_t *cxt, char *cname);
1031 static SV *(*sv_retrieve[])(stcxt_t *cxt, char *cname) = {
1032 0, /* SX_OBJECT -- entry unused dynamically */
1033 retrieve_lscalar, /* SX_LSCALAR */
1034 retrieve_array, /* SX_ARRAY */
1035 retrieve_hash, /* SX_HASH */
1036 retrieve_ref, /* SX_REF */
1037 retrieve_undef, /* SX_UNDEF */
1038 retrieve_integer, /* SX_INTEGER */
1039 retrieve_double, /* SX_DOUBLE */
1040 retrieve_byte, /* SX_BYTE */
1041 retrieve_netint, /* SX_NETINT */
1042 retrieve_scalar, /* SX_SCALAR */
1043 retrieve_tied_array, /* SX_ARRAY */
1044 retrieve_tied_hash, /* SX_HASH */
1045 retrieve_tied_scalar, /* SX_SCALAR */
1046 retrieve_sv_undef, /* SX_SV_UNDEF */
1047 retrieve_sv_yes, /* SX_SV_YES */
1048 retrieve_sv_no, /* SX_SV_NO */
1049 retrieve_blessed, /* SX_BLESS */
1050 retrieve_idx_blessed, /* SX_IX_BLESS */
1051 retrieve_hook, /* SX_HOOK */
1052 retrieve_overloaded, /* SX_OVERLOAD */
1053 retrieve_tied_key, /* SX_TIED_KEY */
1054 retrieve_tied_idx, /* SX_TIED_IDX */
1055 retrieve_utf8str, /* SX_UTF8STR */
1056 retrieve_lutf8str, /* SX_LUTF8STR */
1057 retrieve_flag_hash, /* SX_HASH */
1058 retrieve_other, /* SX_ERROR */
1061 #define RETRIEVE(c,x) (*(c)->retrieve_vtbl[(x) >= SX_ERROR ? SX_ERROR : (x)])
1063 static SV *mbuf2sv(void);
1066 *** Context management.
1072 * Called once per "thread" (interpreter) to initialize some global context.
1074 static void init_perinterp(void)
1078 cxt->netorder = 0; /* true if network order used */
1079 cxt->forgive_me = -1; /* whether to be forgiving... */
1085 * Called at the end of every context cleaning, to perform common reset
1088 static void reset_context(stcxt_t *cxt)
1092 cxt->optype &= ~(ST_STORE|ST_RETRIEVE); /* Leave ST_CLONE alone */
1096 * init_store_context
1098 * Initialize a new store context for real recursion.
1100 static void init_store_context(
1106 TRACEME(("init_store_context"));
1108 cxt->netorder = network_order;
1109 cxt->forgive_me = -1; /* Fetched from perl if needed */
1110 cxt->canonical = -1; /* Idem */
1111 cxt->tagnum = -1; /* Reset tag numbers */
1112 cxt->classnum = -1; /* Reset class numbers */
1113 cxt->fio = f; /* Where I/O are performed */
1114 cxt->optype = optype; /* A store, or a deep clone */
1115 cxt->entry = 1; /* No recursion yet */
1118 * The `hseen' table is used to keep track of each SV stored and their
1119 * associated tag numbers is special. It is "abused" because the
1120 * values stored are not real SV, just integers cast to (SV *),
1121 * which explains the freeing below.
1123 * It is also one possible bottlneck to achieve good storing speed,
1124 * so the "shared keys" optimization is turned off (unlikely to be
1125 * of any use here), and the hash table is "pre-extended". Together,
1126 * those optimizations increase the throughput by 12%.
1129 cxt->hseen = newHV(); /* Table where seen objects are stored */
1130 HvSHAREKEYS_off(cxt->hseen);
1133 * The following does not work well with perl5.004_04, and causes
1134 * a core dump later on, in a completely unrelated spot, which
1135 * makes me think there is a memory corruption going on.
1137 * Calling hv_ksplit(hseen, HBUCKETS) instead of manually hacking
1138 * it below does not make any difference. It seems to work fine
1139 * with perl5.004_68 but given the probable nature of the bug,
1140 * that does not prove anything.
1142 * It's a shame because increasing the amount of buckets raises
1143 * store() throughput by 5%, but until I figure this out, I can't
1144 * allow for this to go into production.
1146 * It is reported fixed in 5.005, hence the #if.
1148 #if PERL_VERSION >= 5
1149 #define HBUCKETS 4096 /* Buckets for %hseen */
1150 HvMAX(cxt->hseen) = HBUCKETS - 1; /* keys %hseen = $HBUCKETS; */
1154 * The `hclass' hash uses the same settings as `hseen' above, but it is
1155 * used to assign sequential tags (numbers) to class names for blessed
1158 * We turn the shared key optimization on.
1161 cxt->hclass = newHV(); /* Where seen classnames are stored */
1163 #if PERL_VERSION >= 5
1164 HvMAX(cxt->hclass) = HBUCKETS - 1; /* keys %hclass = $HBUCKETS; */
1168 * The `hook' hash table is used to keep track of the references on
1169 * the STORABLE_freeze hook routines, when found in some class name.
1171 * It is assumed that the inheritance tree will not be changed during
1172 * storing, and that no new method will be dynamically created by the
1176 cxt->hook = newHV(); /* Table where hooks are cached */
1179 * The `hook_seen' array keeps track of all the SVs returned by
1180 * STORABLE_freeze hooks for us to serialize, so that they are not
1181 * reclaimed until the end of the serialization process. Each SV is
1182 * only stored once, the first time it is seen.
1185 cxt->hook_seen = newAV(); /* Lists SVs returned by STORABLE_freeze */
1189 * clean_store_context
1191 * Clean store context by
1193 static void clean_store_context(stcxt_t *cxt)
1197 TRACEME(("clean_store_context"));
1199 ASSERT(cxt->optype & ST_STORE, ("was performing a store()"));
1202 * Insert real values into hashes where we stored faked pointers.
1206 hv_iterinit(cxt->hseen);
1207 while ((he = hv_iternext(cxt->hseen))) /* Extra () for -Wall, grr.. */
1208 HeVAL(he) = &PL_sv_undef;
1212 hv_iterinit(cxt->hclass);
1213 while ((he = hv_iternext(cxt->hclass))) /* Extra () for -Wall, grr.. */
1214 HeVAL(he) = &PL_sv_undef;
1218 * And now dispose of them...
1220 * The surrounding if() protection has been added because there might be
1221 * some cases where this routine is called more than once, during
1222 * exceptionnal events. This was reported by Marc Lehmann when Storable
1223 * is executed from mod_perl, and the fix was suggested by him.
1224 * -- RAM, 20/12/2000
1228 HV *hseen = cxt->hseen;
1231 sv_free((SV *) hseen);
1235 HV *hclass = cxt->hclass;
1238 sv_free((SV *) hclass);
1242 HV *hook = cxt->hook;
1245 sv_free((SV *) hook);
1248 if (cxt->hook_seen) {
1249 AV *hook_seen = cxt->hook_seen;
1251 av_undef(hook_seen);
1252 sv_free((SV *) hook_seen);
1255 cxt->forgive_me = -1; /* Fetched from perl if needed */
1256 cxt->canonical = -1; /* Idem */
1262 * init_retrieve_context
1264 * Initialize a new retrieve context for real recursion.
1266 static void init_retrieve_context(stcxt_t *cxt, int optype, int is_tainted)
1268 TRACEME(("init_retrieve_context"));
1271 * The hook hash table is used to keep track of the references on
1272 * the STORABLE_thaw hook routines, when found in some class name.
1274 * It is assumed that the inheritance tree will not be changed during
1275 * storing, and that no new method will be dynamically created by the
1279 cxt->hook = newHV(); /* Caches STORABLE_thaw */
1282 * If retrieving an old binary version, the cxt->retrieve_vtbl variable
1283 * was set to sv_old_retrieve. We'll need a hash table to keep track of
1284 * the correspondance between the tags and the tag number used by the
1285 * new retrieve routines.
1288 cxt->hseen = ((cxt->retrieve_vtbl == sv_old_retrieve) ? newHV() : 0);
1290 cxt->aseen = newAV(); /* Where retrieved objects are kept */
1291 cxt->aclass = newAV(); /* Where seen classnames are kept */
1292 cxt->tagnum = 0; /* Have to count objects... */
1293 cxt->classnum = 0; /* ...and class names as well */
1294 cxt->optype = optype;
1295 cxt->s_tainted = is_tainted;
1296 cxt->entry = 1; /* No recursion yet */
1297 #ifndef HAS_RESTRICTED_HASHES
1298 cxt->derestrict = -1; /* Fetched from perl if needed */
1300 #ifndef HAS_UTF8_ALL
1301 cxt->use_bytes = -1; /* Fetched from perl if needed */
1303 cxt->accept_future_minor = -1; /* Fetched from perl if needed */
1307 * clean_retrieve_context
1309 * Clean retrieve context by
1311 static void clean_retrieve_context(stcxt_t *cxt)
1313 TRACEME(("clean_retrieve_context"));
1315 ASSERT(cxt->optype & ST_RETRIEVE, ("was performing a retrieve()"));
1318 AV *aseen = cxt->aseen;
1321 sv_free((SV *) aseen);
1325 AV *aclass = cxt->aclass;
1328 sv_free((SV *) aclass);
1332 HV *hook = cxt->hook;
1335 sv_free((SV *) hook);
1339 HV *hseen = cxt->hseen;
1342 sv_free((SV *) hseen); /* optional HV, for backward compat. */
1345 #ifndef HAS_RESTRICTED_HASHES
1346 cxt->derestrict = -1; /* Fetched from perl if needed */
1348 #ifndef HAS_UTF8_ALL
1349 cxt->use_bytes = -1; /* Fetched from perl if needed */
1351 cxt->accept_future_minor = -1; /* Fetched from perl if needed */
1359 * A workaround for the CROAK bug: cleanup the last context.
1361 static void clean_context(stcxt_t *cxt)
1363 TRACEME(("clean_context"));
1365 ASSERT(cxt->s_dirty, ("dirty context"));
1370 ASSERT(!cxt->membuf_ro, ("mbase is not read-only"));
1372 if (cxt->optype & ST_RETRIEVE)
1373 clean_retrieve_context(cxt);
1374 else if (cxt->optype & ST_STORE)
1375 clean_store_context(cxt);
1379 ASSERT(!cxt->s_dirty, ("context is clean"));
1380 ASSERT(cxt->entry == 0, ("context is reset"));
1386 * Allocate a new context and push it on top of the parent one.
1387 * This new context is made globally visible via SET_STCXT().
1389 static stcxt_t *allocate_context(parent_cxt)
1390 stcxt_t *parent_cxt;
1394 TRACEME(("allocate_context"));
1396 ASSERT(!parent_cxt->s_dirty, ("parent context clean"));
1398 NEW_STORABLE_CXT_OBJ(cxt);
1399 cxt->prev = parent_cxt->my_sv;
1402 ASSERT(!cxt->s_dirty, ("clean context"));
1410 * Free current context, which cannot be the "root" one.
1411 * Make the context underneath globally visible via SET_STCXT().
1413 static void free_context(cxt)
1416 stcxt_t *prev = (stcxt_t *)(cxt->prev ? SvPVX(SvRV(cxt->prev)) : 0);
1418 TRACEME(("free_context"));
1420 ASSERT(!cxt->s_dirty, ("clean context"));
1421 ASSERT(prev, ("not freeing root context"));
1423 SvREFCNT_dec(cxt->my_sv);
1426 ASSERT(cxt, ("context not void"));
1436 * Tells whether we're in the middle of a store operation.
1438 int is_storing(void)
1442 return cxt->entry && (cxt->optype & ST_STORE);
1448 * Tells whether we're in the middle of a retrieve operation.
1450 int is_retrieving(void)
1454 return cxt->entry && (cxt->optype & ST_RETRIEVE);
1458 * last_op_in_netorder
1460 * Returns whether last operation was made using network order.
1462 * This is typically out-of-band information that might prove useful
1463 * to people wishing to convert native to network order data when used.
1465 int last_op_in_netorder(void)
1469 return cxt->netorder;
1473 *** Hook lookup and calling routines.
1479 * A wrapper on gv_fetchmethod_autoload() which caches results.
1481 * Returns the routine reference as an SV*, or null if neither the package
1482 * nor its ancestors know about the method.
1484 static SV *pkg_fetchmeth(
1493 * The following code is the same as the one performed by UNIVERSAL::can
1497 gv = gv_fetchmethod_autoload(pkg, method, FALSE);
1498 if (gv && isGV(gv)) {
1499 sv = newRV((SV*) GvCV(gv));
1500 TRACEME(("%s->%s: 0x%"UVxf, HvNAME(pkg), method, PTR2UV(sv)));
1502 sv = newSVsv(&PL_sv_undef);
1503 TRACEME(("%s->%s: not found", HvNAME(pkg), method));
1507 * Cache the result, ignoring failure: if we can't store the value,
1508 * it just won't be cached.
1511 (void) hv_store(cache, HvNAME(pkg), strlen(HvNAME(pkg)), sv, 0);
1513 return SvOK(sv) ? sv : (SV *) 0;
1519 * Force cached value to be undef: hook ignored even if present.
1521 static void pkg_hide(
1526 (void) hv_store(cache,
1527 HvNAME(pkg), strlen(HvNAME(pkg)), newSVsv(&PL_sv_undef), 0);
1533 * Discard cached value: a whole fetch loop will be retried at next lookup.
1535 static void pkg_uncache(
1540 (void) hv_delete(cache, HvNAME(pkg), strlen(HvNAME(pkg)), G_DISCARD);
1546 * Our own "UNIVERSAL::can", which caches results.
1548 * Returns the routine reference as an SV*, or null if the object does not
1549 * know about the method.
1559 TRACEME(("pkg_can for %s->%s", HvNAME(pkg), method));
1562 * Look into the cache to see whether we already have determined
1563 * where the routine was, if any.
1565 * NOTA BENE: we don't use `method' at all in our lookup, since we know
1566 * that only one hook (i.e. always the same) is cached in a given cache.
1569 svh = hv_fetch(cache, HvNAME(pkg), strlen(HvNAME(pkg)), FALSE);
1573 TRACEME(("cached %s->%s: not found", HvNAME(pkg), method));
1576 TRACEME(("cached %s->%s: 0x%"UVxf,
1577 HvNAME(pkg), method, PTR2UV(sv)));
1582 TRACEME(("not cached yet"));
1583 return pkg_fetchmeth(cache, pkg, method); /* Fetch and cache */
1589 * Call routine as obj->hook(av) in scalar context.
1590 * Propagates the single returned value if not called in void context.
1592 static SV *scalar_call(
1603 TRACEME(("scalar_call (cloning=%d)", cloning));
1610 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1612 SV **ary = AvARRAY(av);
1613 int cnt = AvFILLp(av) + 1;
1615 XPUSHs(ary[0]); /* Frozen string */
1616 for (i = 1; i < cnt; i++) {
1617 TRACEME(("pushing arg #%d (0x%"UVxf")...",
1618 i, PTR2UV(ary[i])));
1619 XPUSHs(sv_2mortal(newRV(ary[i])));
1624 TRACEME(("calling..."));
1625 count = perl_call_sv(hook, flags); /* Go back to Perl code */
1626 TRACEME(("count = %d", count));
1632 SvREFCNT_inc(sv); /* We're returning it, must stay alive! */
1645 * Call routine obj->hook(cloning) in list context.
1646 * Returns the list of returned values in an array.
1648 static AV *array_call(
1658 TRACEME(("array_call (cloning=%d)", cloning));
1664 XPUSHs(obj); /* Target object */
1665 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1668 count = perl_call_sv(hook, G_ARRAY); /* Go back to Perl code */
1673 for (i = count - 1; i >= 0; i--) {
1675 av_store(av, i, SvREFCNT_inc(sv));
1688 * Lookup the class name in the `hclass' table and either assign it a new ID
1689 * or return the existing one, by filling in `classnum'.
1691 * Return true if the class was known, false if the ID was just generated.
1693 static int known_class(
1695 char *name, /* Class name */
1696 int len, /* Name length */
1700 HV *hclass = cxt->hclass;
1702 TRACEME(("known_class (%s)", name));
1705 * Recall that we don't store pointers in this hash table, but tags.
1706 * Therefore, we need LOW_32BITS() to extract the relevant parts.
1709 svh = hv_fetch(hclass, name, len, FALSE);
1711 *classnum = LOW_32BITS(*svh);
1716 * Unknown classname, we need to record it.
1720 if (!hv_store(hclass, name, len, INT2PTR(SV*, cxt->classnum), 0))
1721 CROAK(("Unable to record new classname"));
1723 *classnum = cxt->classnum;
1728 *** Sepcific store routines.
1734 * Store a reference.
1735 * Layout is SX_REF <object> or SX_OVERLOAD <object>.
1737 static int store_ref(stcxt_t *cxt, SV *sv)
1739 TRACEME(("store_ref (0x%"UVxf")", PTR2UV(sv)));
1742 * Follow reference, and check if target is overloaded.
1748 HV *stash = (HV *) SvSTASH(sv);
1749 if (stash && Gv_AMG(stash)) {
1750 TRACEME(("ref (0x%"UVxf") is overloaded", PTR2UV(sv)));
1751 PUTMARK(SX_OVERLOAD);
1757 return store(cxt, sv);
1765 * Layout is SX_LSCALAR <length> <data>, SX_SCALAR <length> <data> or SX_UNDEF.
1766 * The <data> section is omitted if <length> is 0.
1768 * If integer or double, the layout is SX_INTEGER <data> or SX_DOUBLE <data>.
1769 * Small integers (within [-127, +127]) are stored as SX_BYTE <byte>.
1771 static int store_scalar(stcxt_t *cxt, SV *sv)
1776 U32 flags = SvFLAGS(sv); /* "cc -O" may put it in register */
1778 TRACEME(("store_scalar (0x%"UVxf")", PTR2UV(sv)));
1781 * For efficiency, break the SV encapsulation by peaking at the flags
1782 * directly without using the Perl macros to avoid dereferencing
1783 * sv->sv_flags each time we wish to check the flags.
1786 if (!(flags & SVf_OK)) { /* !SvOK(sv) */
1787 if (sv == &PL_sv_undef) {
1788 TRACEME(("immortal undef"));
1789 PUTMARK(SX_SV_UNDEF);
1791 TRACEME(("undef at 0x%"UVxf, PTR2UV(sv)));
1798 * Always store the string representation of a scalar if it exists.
1799 * Gisle Aas provided me with this test case, better than a long speach:
1801 * perl -MDevel::Peek -le '$a="abc"; $a+0; Dump($a)'
1802 * SV = PVNV(0x80c8520)
1804 * FLAGS = (NOK,POK,pNOK,pPOK)
1807 * PV = 0x80c83d0 "abc"\0
1811 * Write SX_SCALAR, length, followed by the actual data.
1813 * Otherwise, write an SX_BYTE, SX_INTEGER or an SX_DOUBLE as
1814 * appropriate, followed by the actual (binary) data. A double
1815 * is written as a string if network order, for portability.
1817 * NOTE: instead of using SvNOK(sv), we test for SvNOKp(sv).
1818 * The reason is that when the scalar value is tainted, the SvNOK(sv)
1821 * The test for a read-only scalar with both POK and NOK set is meant
1822 * to quickly detect &PL_sv_yes and &PL_sv_no without having to pay the
1823 * address comparison for each scalar we store.
1826 #define SV_MAYBE_IMMORTAL (SVf_READONLY|SVf_POK|SVf_NOK)
1828 if ((flags & SV_MAYBE_IMMORTAL) == SV_MAYBE_IMMORTAL) {
1829 if (sv == &PL_sv_yes) {
1830 TRACEME(("immortal yes"));
1832 } else if (sv == &PL_sv_no) {
1833 TRACEME(("immortal no"));
1836 pv = SvPV(sv, len); /* We know it's SvPOK */
1837 goto string; /* Share code below */
1839 } else if (flags & SVp_POK) { /* SvPOKp(sv) => string */
1840 I32 wlen; /* For 64-bit machines */
1844 * Will come here from below with pv and len set if double & netorder,
1845 * or from above if it was readonly, POK and NOK but neither &PL_sv_yes
1850 wlen = (I32) len; /* WLEN via STORE_SCALAR expects I32 */
1852 STORE_UTF8STR(pv, wlen);
1854 STORE_SCALAR(pv, wlen);
1855 TRACEME(("ok (scalar 0x%"UVxf" '%s', length = %"IVdf")",
1856 PTR2UV(sv), SvPVX(sv), (IV)len));
1858 } else if (flags & SVp_NOK) { /* SvNOKp(sv) => double */
1862 * Watch for number being an integer in disguise.
1864 if (nv == (NV) (iv = I_V(nv))) {
1865 TRACEME(("double %"NVff" is actually integer %"IVdf, nv, iv));
1866 goto integer; /* Share code below */
1869 if (cxt->netorder) {
1870 TRACEME(("double %"NVff" stored as string", nv));
1872 goto string; /* Share code above */
1876 WRITE(&nv, sizeof(nv));
1878 TRACEME(("ok (double 0x%"UVxf", value = %"NVff")", PTR2UV(sv), nv));
1880 } else if (flags & SVp_IOK) { /* SvIOKp(sv) => integer */
1884 * Will come here from above with iv set if double is an integer.
1889 * Optimize small integers into a single byte, otherwise store as
1890 * a real integer (converted into network order if they asked).
1893 if (iv >= -128 && iv <= 127) {
1894 unsigned char siv = (unsigned char) (iv + 128); /* [0,255] */
1897 TRACEME(("small integer stored as %d", siv));
1898 } else if (cxt->netorder) {
1901 niv = (I32) htonl(iv);
1902 TRACEME(("using network order"));
1905 TRACEME(("as-is for network order"));
1910 PUTMARK(SX_INTEGER);
1911 WRITE(&iv, sizeof(iv));
1914 TRACEME(("ok (integer 0x%"UVxf", value = %"IVdf")", PTR2UV(sv), iv));
1917 CROAK(("Can't determine type of %s(0x%"UVxf")",
1918 sv_reftype(sv, FALSE),
1921 return 0; /* Ok, no recursion on scalars */
1929 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
1930 * Each item is stored as <object>.
1932 static int store_array(stcxt_t *cxt, AV *av)
1935 I32 len = av_len(av) + 1;
1939 TRACEME(("store_array (0x%"UVxf")", PTR2UV(av)));
1942 * Signal array by emitting SX_ARRAY, followed by the array length.
1947 TRACEME(("size = %d", len));
1950 * Now store each item recursively.
1953 for (i = 0; i < len; i++) {
1954 sav = av_fetch(av, i, 0);
1956 TRACEME(("(#%d) undef item", i));
1960 TRACEME(("(#%d) item", i));
1961 if ((ret = store(cxt, *sav))) /* Extra () for -Wall, grr... */
1965 TRACEME(("ok (array)"));
1974 * Borrowed from perl source file pp_ctl.c, where it is used by pp_sort.
1977 sortcmp(const void *a, const void *b)
1979 return sv_cmp(*(SV * const *) a, *(SV * const *) b);
1986 * Store a hash table.
1988 * For a "normal" hash (not restricted, no utf8 keys):
1990 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
1991 * Values are stored as <object>.
1992 * Keys are stored as <length> <data>, the <data> section being omitted
1995 * For a "fancy" hash (restricted or utf8 keys):
1997 * Layout is SX_FLAG_HASH <size> <hash flags> followed by each key/value pair,
1999 * Values are stored as <object>.
2000 * Keys are stored as <flags> <length> <data>, the <data> section being omitted
2002 * Currently the only hash flag is "restriced"
2003 * Key flags are as for hv.h
2005 static int store_hash(stcxt_t *cxt, HV *hv)
2008 #ifdef HAS_RESTRICTED_HASHES
2017 int flagged_hash = ((SvREADONLY(hv)
2018 #ifdef HAS_HASH_KEY_FLAGS
2022 unsigned char hash_flags = (SvREADONLY(hv) ? SHV_RESTRICTED : 0);
2025 /* needs int cast for C++ compilers, doesn't it? */
2026 TRACEME(("store_hash (0x%"UVxf") (flags %x)", PTR2UV(hv),
2029 TRACEME(("store_hash (0x%"UVxf")", PTR2UV(hv)));
2033 * Signal hash by emitting SX_HASH, followed by the table length.
2037 PUTMARK(SX_FLAG_HASH);
2038 PUTMARK(hash_flags);
2043 TRACEME(("size = %d", len));
2046 * Save possible iteration state via each() on that table.
2049 riter = HvRITER(hv);
2050 eiter = HvEITER(hv);
2054 * Now store each item recursively.
2056 * If canonical is defined to some true value then store each
2057 * key/value pair in sorted order otherwise the order is random.
2058 * Canonical order is irrelevant when a deep clone operation is performed.
2060 * Fetch the value from perl only once per store() operation, and only
2065 !(cxt->optype & ST_CLONE) && (cxt->canonical == 1 ||
2066 (cxt->canonical < 0 && (cxt->canonical =
2067 (SvTRUE(perl_get_sv("Storable::canonical", TRUE)) ? 1 : 0))))
2070 * Storing in order, sorted by key.
2071 * Run through the hash, building up an array of keys in a
2072 * mortal array, sort the array and then run through the
2078 /*av_extend (av, len);*/
2080 TRACEME(("using canonical order"));
2082 for (i = 0; i < len; i++) {
2083 #ifdef HAS_RESTRICTED_HASHES
2084 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
2086 HE *he = hv_iternext(hv);
2088 SV *key = hv_iterkeysv(he);
2089 av_store(av, AvFILLp(av)+1, key); /* av_push(), really */
2092 qsort((char *) AvARRAY(av), len, sizeof(SV *), sortcmp);
2094 for (i = 0; i < len; i++) {
2095 unsigned char flags;
2099 SV *key = av_shift(av);
2100 HE *he = hv_fetch_ent(hv, key, 0, 0);
2101 SV *val = HeVAL(he);
2103 return 1; /* Internal error, not I/O error */
2106 * Store value first.
2109 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
2111 if ((ret = store(cxt, val))) /* Extra () for -Wall, grr... */
2116 * Keys are written after values to make sure retrieval
2117 * can be optimal in terms of memory usage, where keys are
2118 * read into a fixed unique buffer called kbuf.
2119 * See retrieve_hash() for details.
2122 /* Implementation of restricted hashes isn't nicely
2125 = (((hash_flags & SHV_RESTRICTED)
2127 ? SHV_K_LOCKED : 0);
2128 if (val == &PL_sv_undef)
2129 flags |= SHV_K_PLACEHOLDER;
2131 keyval = SvPV(key, keylen_tmp);
2132 keylen = keylen_tmp;
2133 #ifdef HAS_UTF8_HASHES
2134 /* If you build without optimisation on pre 5.6
2135 then nothing spots that SvUTF8(key) is always 0,
2136 so the block isn't optimised away, at which point
2137 the linker dislikes the reference to
2140 const char *keysave = keyval;
2141 bool is_utf8 = TRUE;
2143 /* Just casting the &klen to (STRLEN) won't work
2144 well if STRLEN and I32 are of different widths.
2146 keyval = (char*)bytes_from_utf8((U8*)keyval,
2150 /* If we were able to downgrade here, then than
2151 means that we have a key which only had chars
2152 0-255, but was utf8 encoded. */
2154 if (keyval != keysave) {
2155 keylen = keylen_tmp;
2156 flags |= SHV_K_WASUTF8;
2158 /* keylen_tmp can't have changed, so no need
2159 to assign back to keylen. */
2160 flags |= SHV_K_UTF8;
2167 TRACEME(("(#%d) key '%s' flags %x %u", i, keyval, flags, *keyval));
2169 assert (flags == 0);
2170 TRACEME(("(#%d) key '%s'", i, keyval));
2174 WRITE(keyval, keylen);
2175 if (flags & SHV_K_WASUTF8)
2180 * Free up the temporary array
2189 * Storing in "random" order (in the order the keys are stored
2190 * within the the hash). This is the default and will be faster!
2193 for (i = 0; i < len; i++) {
2196 unsigned char flags;
2197 #ifdef HV_ITERNEXT_WANTPLACEHOLDERS
2198 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
2200 HE *he = hv_iternext(hv);
2202 SV *val = (he ? hv_iterval(hv, he) : 0);
2207 return 1; /* Internal error, not I/O error */
2210 * Store value first.
2213 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
2215 if ((ret = store(cxt, val))) /* Extra () for -Wall, grr... */
2218 /* Implementation of restricted hashes isn't nicely
2221 = (((hash_flags & SHV_RESTRICTED)
2223 ? SHV_K_LOCKED : 0);
2224 if (val == &PL_sv_undef)
2225 flags |= SHV_K_PLACEHOLDER;
2227 hek = HeKEY_hek(he);
2229 if (len == HEf_SVKEY) {
2230 /* This is somewhat sick, but the internal APIs are
2231 * such that XS code could put one of these in in
2233 * Maybe we should be capable of storing one if
2236 key_sv = HeKEY_sv(he);
2237 flags |= SHV_K_ISSV;
2239 /* Regular string key. */
2240 #ifdef HAS_HASH_KEY_FLAGS
2242 flags |= SHV_K_UTF8;
2243 if (HEK_WASUTF8(hek))
2244 flags |= SHV_K_WASUTF8;
2250 * Keys are written after values to make sure retrieval
2251 * can be optimal in terms of memory usage, where keys are
2252 * read into a fixed unique buffer called kbuf.
2253 * See retrieve_hash() for details.
2258 TRACEME(("(#%d) key '%s' flags %x", i, key, flags));
2260 assert (flags == 0);
2261 TRACEME(("(#%d) key '%s'", i, key));
2263 if (flags & SHV_K_ISSV) {
2273 TRACEME(("ok (hash 0x%"UVxf")", PTR2UV(hv)));
2276 HvRITER(hv) = riter; /* Restore hash iterator state */
2277 HvEITER(hv) = eiter;
2285 * When storing a tied object (be it a tied scalar, array or hash), we lay out
2286 * a special mark, followed by the underlying tied object. For instance, when
2287 * dealing with a tied hash, we store SX_TIED_HASH <hash object>, where
2288 * <hash object> stands for the serialization of the tied hash.
2290 static int store_tied(stcxt_t *cxt, SV *sv)
2294 int svt = SvTYPE(sv);
2297 TRACEME(("store_tied (0x%"UVxf")", PTR2UV(sv)));
2300 * We have a small run-time penalty here because we chose to factorise
2301 * all tieds objects into the same routine, and not have a store_tied_hash,
2302 * a store_tied_array, etc...
2304 * Don't use a switch() statement, as most compilers don't optimize that
2305 * well for 2/3 values. An if() else if() cascade is just fine. We put
2306 * tied hashes first, as they are the most likely beasts.
2309 if (svt == SVt_PVHV) {
2310 TRACEME(("tied hash"));
2311 PUTMARK(SX_TIED_HASH); /* Introduces tied hash */
2312 } else if (svt == SVt_PVAV) {
2313 TRACEME(("tied array"));
2314 PUTMARK(SX_TIED_ARRAY); /* Introduces tied array */
2316 TRACEME(("tied scalar"));
2317 PUTMARK(SX_TIED_SCALAR); /* Introduces tied scalar */
2321 if (!(mg = mg_find(sv, mtype)))
2322 CROAK(("No magic '%c' found while storing tied %s", mtype,
2323 (svt == SVt_PVHV) ? "hash" :
2324 (svt == SVt_PVAV) ? "array" : "scalar"));
2327 * The mg->mg_obj found by mg_find() above actually points to the
2328 * underlying tied Perl object implementation. For instance, if the
2329 * original SV was that of a tied array, then mg->mg_obj is an AV.
2331 * Note that we store the Perl object as-is. We don't call its FETCH
2332 * method along the way. At retrieval time, we won't call its STORE
2333 * method either, but the tieing magic will be re-installed. In itself,
2334 * that ensures that the tieing semantics are preserved since futher
2335 * accesses on the retrieved object will indeed call the magic methods...
2338 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
2341 TRACEME(("ok (tied)"));
2349 * Stores a reference to an item within a tied structure:
2351 * . \$h{key}, stores both the (tied %h) object and 'key'.
2352 * . \$a[idx], stores both the (tied @a) object and 'idx'.
2354 * Layout is therefore either:
2355 * SX_TIED_KEY <object> <key>
2356 * SX_TIED_IDX <object> <index>
2358 static int store_tied_item(stcxt_t *cxt, SV *sv)
2363 TRACEME(("store_tied_item (0x%"UVxf")", PTR2UV(sv)));
2365 if (!(mg = mg_find(sv, 'p')))
2366 CROAK(("No magic 'p' found while storing reference to tied item"));
2369 * We discriminate between \$h{key} and \$a[idx] via mg_ptr.
2373 TRACEME(("store_tied_item: storing a ref to a tied hash item"));
2374 PUTMARK(SX_TIED_KEY);
2375 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2377 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
2380 TRACEME(("store_tied_item: storing PTR 0x%"UVxf, PTR2UV(mg->mg_ptr)));
2382 if ((ret = store(cxt, (SV *) mg->mg_ptr))) /* Idem, for -Wall */
2385 I32 idx = mg->mg_len;
2387 TRACEME(("store_tied_item: storing a ref to a tied array item "));
2388 PUTMARK(SX_TIED_IDX);
2389 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2391 if ((ret = store(cxt, mg->mg_obj))) /* Idem, for -Wall */
2394 TRACEME(("store_tied_item: storing IDX %d", idx));
2399 TRACEME(("ok (tied item)"));
2405 * store_hook -- dispatched manually, not via sv_store[]
2407 * The blessed SV is serialized by a hook.
2411 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2413 * where <flags> indicates how long <len>, <len2> and <len3> are, whether
2414 * the trailing part [] is present, the type of object (scalar, array or hash).
2415 * There is also a bit which says how the classname is stored between:
2420 * and when the <index> form is used (classname already seen), the "large
2421 * classname" bit in <flags> indicates how large the <index> is.
2423 * The serialized string returned by the hook is of length <len2> and comes
2424 * next. It is an opaque string for us.
2426 * Those <len3> object IDs which are listed last represent the extra references
2427 * not directly serialized by the hook, but which are linked to the object.
2429 * When recursion is mandated to resolve object-IDs not yet seen, we have
2430 * instead, with <header> being flags with bits set to indicate the object type
2431 * and that recursion was indeed needed:
2433 * SX_HOOK <header> <object> <header> <object> <flags>
2435 * that same header being repeated between serialized objects obtained through
2436 * recursion, until we reach flags indicating no recursion, at which point
2437 * we know we've resynchronized with a single layout, after <flags>.
2439 * When storing a blessed ref to a tied variable, the following format is
2442 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
2444 * The first <flags> indication carries an object of type SHT_EXTRA, and the
2445 * real object type is held in the <extra> flag. At the very end of the
2446 * serialization stream, the underlying magic object is serialized, just like
2447 * any other tied variable.
2449 static int store_hook(
2462 int count; /* really len3 + 1 */
2463 unsigned char flags;
2466 int recursed = 0; /* counts recursion */
2467 int obj_type; /* object type, on 2 bits */
2470 int clone = cxt->optype & ST_CLONE;
2471 char mtype = '\0'; /* for blessed ref to tied structures */
2472 unsigned char eflags = '\0'; /* used when object type is SHT_EXTRA */
2474 TRACEME(("store_hook, class \"%s\", tagged #%d", HvNAME(pkg), cxt->tagnum));
2477 * Determine object type on 2 bits.
2482 obj_type = SHT_SCALAR;
2485 obj_type = SHT_ARRAY;
2488 obj_type = SHT_HASH;
2492 * Produced by a blessed ref to a tied data structure, $o in the
2493 * following Perl code.
2497 * my $o = bless \%h, 'BAR';
2499 * Signal the tie-ing magic by setting the object type as SHT_EXTRA
2500 * (since we have only 2 bits in <flags> to store the type), and an
2501 * <extra> byte flag will be emitted after the FIRST <flags> in the
2502 * stream, carrying what we put in `eflags'.
2504 obj_type = SHT_EXTRA;
2505 switch (SvTYPE(sv)) {
2507 eflags = (unsigned char) SHT_THASH;
2511 eflags = (unsigned char) SHT_TARRAY;
2515 eflags = (unsigned char) SHT_TSCALAR;
2521 CROAK(("Unexpected object type (%d) in store_hook()", type));
2523 flags = SHF_NEED_RECURSE | obj_type;
2525 class = HvNAME(pkg);
2526 len = strlen(class);
2529 * To call the hook, we need to fake a call like:
2531 * $object->STORABLE_freeze($cloning);
2533 * but we don't have the $object here. For instance, if $object is
2534 * a blessed array, what we have in `sv' is the array, and we can't
2535 * call a method on those.
2537 * Therefore, we need to create a temporary reference to the object and
2538 * make the call on that reference.
2541 TRACEME(("about to call STORABLE_freeze on class %s", class));
2543 ref = newRV_noinc(sv); /* Temporary reference */
2544 av = array_call(ref, hook, clone); /* @a = $object->STORABLE_freeze($c) */
2546 SvREFCNT_dec(ref); /* Reclaim temporary reference */
2548 count = AvFILLp(av) + 1;
2549 TRACEME(("store_hook, array holds %d items", count));
2552 * If they return an empty list, it means they wish to ignore the
2553 * hook for this class (and not just this instance -- that's for them
2554 * to handle if they so wish).
2556 * Simply disable the cached entry for the hook (it won't be recomputed
2557 * since it's present in the cache) and recurse to store_blessed().
2562 * They must not change their mind in the middle of a serialization.
2565 if (hv_fetch(cxt->hclass, class, len, FALSE))
2566 CROAK(("Too late to ignore hooks for %s class \"%s\"",
2567 (cxt->optype & ST_CLONE) ? "cloning" : "storing", class));
2569 pkg_hide(cxt->hook, pkg, "STORABLE_freeze");
2571 ASSERT(!pkg_can(cxt->hook, pkg, "STORABLE_freeze"), ("hook invisible"));
2572 TRACEME(("ignoring STORABLE_freeze in class \"%s\"", class));
2574 return store_blessed(cxt, sv, type, pkg);
2578 * Get frozen string.
2582 pv = SvPV(ary[0], len2);
2585 * If they returned more than one item, we need to serialize some
2586 * extra references if not already done.
2588 * Loop over the array, starting at postion #1, and for each item,
2589 * ensure it is a reference, serialize it if not already done, and
2590 * replace the entry with the tag ID of the corresponding serialized
2593 * We CHEAT by not calling av_fetch() and read directly within the
2597 for (i = 1; i < count; i++) {
2601 AV *av_hook = cxt->hook_seen;
2604 CROAK(("Item #%d returned by STORABLE_freeze "
2605 "for %s is not a reference", i, class));
2606 xsv = SvRV(rsv); /* Follow ref to know what to look for */
2609 * Look in hseen and see if we have a tag already.
2610 * Serialize entry if not done already, and get its tag.
2613 if ((svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE)))
2614 goto sv_seen; /* Avoid moving code too far to the right */
2616 TRACEME(("listed object %d at 0x%"UVxf" is unknown", i-1, PTR2UV(xsv)));
2619 * We need to recurse to store that object and get it to be known
2620 * so that we can resolve the list of object-IDs at retrieve time.
2622 * The first time we do this, we need to emit the proper header
2623 * indicating that we recursed, and what the type of object is (the
2624 * object we're storing via a user-hook). Indeed, during retrieval,
2625 * we'll have to create the object before recursing to retrieve the
2626 * others, in case those would point back at that object.
2629 /* [SX_HOOK] <flags> [<extra>] <object>*/
2633 if (obj_type == SHT_EXTRA)
2638 if ((ret = store(cxt, xsv))) /* Given by hook for us to store */
2641 svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE);
2643 CROAK(("Could not serialize item #%d from hook in %s", i, class));
2646 * It was the first time we serialized `xsv'.
2648 * Keep this SV alive until the end of the serialization: if we
2649 * disposed of it right now by decrementing its refcount, and it was
2650 * a temporary value, some next temporary value allocated during
2651 * another STORABLE_freeze might take its place, and we'd wrongly
2652 * assume that new SV was already serialized, based on its presence
2655 * Therefore, push it away in cxt->hook_seen.
2658 av_store(av_hook, AvFILLp(av_hook)+1, SvREFCNT_inc(xsv));
2662 * Dispose of the REF they returned. If we saved the `xsv' away
2663 * in the array of returned SVs, that will not cause the underlying
2664 * referenced SV to be reclaimed.
2667 ASSERT(SvREFCNT(xsv) > 1, ("SV will survive disposal of its REF"));
2668 SvREFCNT_dec(rsv); /* Dispose of reference */
2671 * Replace entry with its tag (not a real SV, so no refcnt increment)
2675 TRACEME(("listed object %d at 0x%"UVxf" is tag #%"UVuf,
2676 i-1, PTR2UV(xsv), PTR2UV(*svh)));
2680 * Allocate a class ID if not already done.
2682 * This needs to be done after the recursion above, since at retrieval
2683 * time, we'll see the inner objects first. Many thanks to
2684 * Salvador Ortiz Garcia <sog@msg.com.mx> who spot that bug and
2685 * proposed the right fix. -- RAM, 15/09/2000
2688 if (!known_class(cxt, class, len, &classnum)) {
2689 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2690 classnum = -1; /* Mark: we must store classname */
2692 TRACEME(("already seen class %s, ID = %d", class, classnum));
2696 * Compute leading flags.
2700 if (((classnum == -1) ? len : classnum) > LG_SCALAR)
2701 flags |= SHF_LARGE_CLASSLEN;
2703 flags |= SHF_IDX_CLASSNAME;
2704 if (len2 > LG_SCALAR)
2705 flags |= SHF_LARGE_STRLEN;
2707 flags |= SHF_HAS_LIST;
2708 if (count > (LG_SCALAR + 1))
2709 flags |= SHF_LARGE_LISTLEN;
2712 * We're ready to emit either serialized form:
2714 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2715 * SX_HOOK <flags> <index> <len2> <str> [<len3> <object-IDs>]
2717 * If we recursed, the SX_HOOK has already been emitted.
2720 TRACEME(("SX_HOOK (recursed=%d) flags=0x%x "
2721 "class=%"IVdf" len=%"IVdf" len2=%"IVdf" len3=%d",
2722 recursed, flags, (IV)classnum, (IV)len, (IV)len2, count-1));
2724 /* SX_HOOK <flags> [<extra>] */
2728 if (obj_type == SHT_EXTRA)
2733 /* <len> <classname> or <index> */
2734 if (flags & SHF_IDX_CLASSNAME) {
2735 if (flags & SHF_LARGE_CLASSLEN)
2738 unsigned char cnum = (unsigned char) classnum;
2742 if (flags & SHF_LARGE_CLASSLEN)
2745 unsigned char clen = (unsigned char) len;
2748 WRITE(class, len); /* Final \0 is omitted */
2751 /* <len2> <frozen-str> */
2752 if (flags & SHF_LARGE_STRLEN) {
2753 I32 wlen2 = len2; /* STRLEN might be 8 bytes */
2754 WLEN(wlen2); /* Must write an I32 for 64-bit machines */
2756 unsigned char clen = (unsigned char) len2;
2760 WRITE(pv, (SSize_t)len2); /* Final \0 is omitted */
2762 /* [<len3> <object-IDs>] */
2763 if (flags & SHF_HAS_LIST) {
2764 int len3 = count - 1;
2765 if (flags & SHF_LARGE_LISTLEN)
2768 unsigned char clen = (unsigned char) len3;
2773 * NOTA BENE, for 64-bit machines: the ary[i] below does not yield a
2774 * real pointer, rather a tag number, well under the 32-bit limit.
2777 for (i = 1; i < count; i++) {
2778 I32 tagval = htonl(LOW_32BITS(ary[i]));
2780 TRACEME(("object %d, tag #%d", i-1, ntohl(tagval)));
2785 * Free the array. We need extra care for indices after 0, since they
2786 * don't hold real SVs but integers cast.
2790 AvFILLp(av) = 0; /* Cheat, nothing after 0 interests us */
2795 * If object was tied, need to insert serialization of the magic object.
2798 if (obj_type == SHT_EXTRA) {
2801 if (!(mg = mg_find(sv, mtype))) {
2802 int svt = SvTYPE(sv);
2803 CROAK(("No magic '%c' found while storing ref to tied %s with hook",
2804 mtype, (svt == SVt_PVHV) ? "hash" :
2805 (svt == SVt_PVAV) ? "array" : "scalar"));
2808 TRACEME(("handling the magic object 0x%"UVxf" part of 0x%"UVxf,
2809 PTR2UV(mg->mg_obj), PTR2UV(sv)));
2815 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
2823 * store_blessed -- dispatched manually, not via sv_store[]
2825 * Check whether there is a STORABLE_xxx hook defined in the class or in one
2826 * of its ancestors. If there is, then redispatch to store_hook();
2828 * Otherwise, the blessed SV is stored using the following layout:
2830 * SX_BLESS <flag> <len> <classname> <object>
2832 * where <flag> indicates whether <len> is stored on 0 or 4 bytes, depending
2833 * on the high-order bit in flag: if 1, then length follows on 4 bytes.
2834 * Otherwise, the low order bits give the length, thereby giving a compact
2835 * representation for class names less than 127 chars long.
2837 * Each <classname> seen is remembered and indexed, so that the next time
2838 * an object in the blessed in the same <classname> is stored, the following
2841 * SX_IX_BLESS <flag> <index> <object>
2843 * where <index> is the classname index, stored on 0 or 4 bytes depending
2844 * on the high-order bit in flag (same encoding as above for <len>).
2846 static int store_blessed(
2857 TRACEME(("store_blessed, type %d, class \"%s\"", type, HvNAME(pkg)));
2860 * Look for a hook for this blessed SV and redirect to store_hook()
2864 hook = pkg_can(cxt->hook, pkg, "STORABLE_freeze");
2866 return store_hook(cxt, sv, type, pkg, hook);
2869 * This is a blessed SV without any serialization hook.
2872 class = HvNAME(pkg);
2873 len = strlen(class);
2875 TRACEME(("blessed 0x%"UVxf" in %s, no hook: tagged #%d",
2876 PTR2UV(sv), class, cxt->tagnum));
2879 * Determine whether it is the first time we see that class name (in which
2880 * case it will be stored in the SX_BLESS form), or whether we already
2881 * saw that class name before (in which case the SX_IX_BLESS form will be
2885 if (known_class(cxt, class, len, &classnum)) {
2886 TRACEME(("already seen class %s, ID = %d", class, classnum));
2887 PUTMARK(SX_IX_BLESS);
2888 if (classnum <= LG_BLESS) {
2889 unsigned char cnum = (unsigned char) classnum;
2892 unsigned char flag = (unsigned char) 0x80;
2897 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2899 if (len <= LG_BLESS) {
2900 unsigned char clen = (unsigned char) len;
2903 unsigned char flag = (unsigned char) 0x80;
2905 WLEN(len); /* Don't BER-encode, this should be rare */
2907 WRITE(class, len); /* Final \0 is omitted */
2911 * Now emit the <object> part.
2914 return SV_STORE(type)(cxt, sv);
2920 * We don't know how to store the item we reached, so return an error condition.
2921 * (it's probably a GLOB, some CODE reference, etc...)
2923 * If they defined the `forgive_me' variable at the Perl level to some
2924 * true value, then don't croak, just warn, and store a placeholder string
2927 static int store_other(stcxt_t *cxt, SV *sv)
2930 static char buf[80];
2932 TRACEME(("store_other"));
2935 * Fetch the value from perl only once per store() operation.
2939 cxt->forgive_me == 0 ||
2940 (cxt->forgive_me < 0 && !(cxt->forgive_me =
2941 SvTRUE(perl_get_sv("Storable::forgive_me", TRUE)) ? 1 : 0))
2943 CROAK(("Can't store %s items", sv_reftype(sv, FALSE)));
2945 warn("Can't store item %s(0x%"UVxf")",
2946 sv_reftype(sv, FALSE), PTR2UV(sv));
2949 * Store placeholder string as a scalar instead...
2952 (void) sprintf(buf, "You lost %s(0x%"UVxf")%c", sv_reftype(sv, FALSE),
2953 PTR2UV(sv), (char) 0);
2956 STORE_SCALAR(buf, len);
2957 TRACEME(("ok (dummy \"%s\", length = %"IVdf")", buf, len));
2963 *** Store driving routines
2969 * WARNING: partially duplicates Perl's sv_reftype for speed.
2971 * Returns the type of the SV, identified by an integer. That integer
2972 * may then be used to index the dynamic routine dispatch table.
2974 static int sv_type(SV *sv)
2976 switch (SvTYPE(sv)) {
2981 * No need to check for ROK, that can't be set here since there
2982 * is no field capable of hodling the xrv_rv reference.
2990 * Starting from SVt_PV, it is possible to have the ROK flag
2991 * set, the pointer to the other SV being either stored in
2992 * the xrv_rv (in the case of a pure SVt_RV), or as the
2993 * xpv_pv field of an SVt_PV and its heirs.
2995 * However, those SV cannot be magical or they would be an
2996 * SVt_PVMG at least.
2998 return SvROK(sv) ? svis_REF : svis_SCALAR;
3000 case SVt_PVLV: /* Workaround for perl5.004_04 "LVALUE" bug */
3001 if (SvRMAGICAL(sv) && (mg_find(sv, 'p')))
3002 return svis_TIED_ITEM;
3005 if (SvRMAGICAL(sv) && (mg_find(sv, 'q')))
3007 return SvROK(sv) ? svis_REF : svis_SCALAR;
3009 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
3013 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
3026 * Recursively store objects pointed to by the sv to the specified file.
3028 * Layout is <content> or SX_OBJECT <tagnum> if we reach an already stored
3029 * object (one for which storage has started -- it may not be over if we have
3030 * a self-referenced structure). This data set forms a stored <object>.
3032 static int store(stcxt_t *cxt, SV *sv)
3037 HV *hseen = cxt->hseen;
3039 TRACEME(("store (0x%"UVxf")", PTR2UV(sv)));
3042 * If object has already been stored, do not duplicate data.
3043 * Simply emit the SX_OBJECT marker followed by its tag data.
3044 * The tag is always written in network order.
3046 * NOTA BENE, for 64-bit machines: the "*svh" below does not yield a
3047 * real pointer, rather a tag number (watch the insertion code below).
3048 * That means it pobably safe to assume it is well under the 32-bit limit,
3049 * and makes the truncation safe.
3050 * -- RAM, 14/09/1999
3053 svh = hv_fetch(hseen, (char *) &sv, sizeof(sv), FALSE);
3055 I32 tagval = htonl(LOW_32BITS(*svh));
3057 TRACEME(("object 0x%"UVxf" seen as #%d", PTR2UV(sv), ntohl(tagval)));
3065 * Allocate a new tag and associate it with the address of the sv being
3066 * stored, before recursing...
3068 * In order to avoid creating new SvIVs to hold the tagnum we just
3069 * cast the tagnum to an SV pointer and store that in the hash. This
3070 * means that we must clean up the hash manually afterwards, but gives
3071 * us a 15% throughput increase.
3076 if (!hv_store(hseen,
3077 (char *) &sv, sizeof(sv), INT2PTR(SV*, cxt->tagnum), 0))
3081 * Store `sv' and everything beneath it, using appropriate routine.
3082 * Abort immediately if we get a non-zero status back.
3087 TRACEME(("storing 0x%"UVxf" tag #%d, type %d...",
3088 PTR2UV(sv), cxt->tagnum, type));
3091 HV *pkg = SvSTASH(sv);
3092 ret = store_blessed(cxt, sv, type, pkg);
3094 ret = SV_STORE(type)(cxt, sv);
3096 TRACEME(("%s (stored 0x%"UVxf", refcnt=%d, %s)",
3097 ret ? "FAILED" : "ok", PTR2UV(sv),
3098 SvREFCNT(sv), sv_reftype(sv, FALSE)));
3106 * Write magic number and system information into the file.
3107 * Layout is <magic> <network> [<len> <byteorder> <sizeof int> <sizeof long>
3108 * <sizeof ptr>] where <len> is the length of the byteorder hexa string.
3109 * All size and lenghts are written as single characters here.
3111 * Note that no byte ordering info is emitted when <network> is true, since
3112 * integers will be emitted in network order in that case.
3114 static int magic_write(stcxt_t *cxt)
3116 char buf[256]; /* Enough room for 256 hexa digits */
3118 int use_network_order = cxt->netorder;
3120 TRACEME(("magic_write on fd=%d", cxt->fio ? PerlIO_fileno(cxt->fio)
3124 WRITE(magicstr, (SSize_t)strlen(magicstr)); /* Don't write final \0 */
3127 * Starting with 0.6, the "use_network_order" byte flag is also used to
3128 * indicate the version number of the binary image, encoded in the upper
3129 * bits. The bit 0 is always used to indicate network order.
3133 ((use_network_order ? 0x1 : 0x0) | (STORABLE_BIN_MAJOR << 1));
3137 * Starting with 0.7, a full byte is dedicated to the minor version of
3138 * the binary format, which is incremented only when new markers are
3139 * introduced, for instance, but when backward compatibility is preserved.
3142 PUTMARK((unsigned char) STORABLE_BIN_WRITE_MINOR);
3144 if (use_network_order)
3145 return 0; /* Don't bother with byte ordering */
3147 sprintf(buf, "%lx", (unsigned long) BYTEORDER);
3148 c = (unsigned char) strlen(buf);
3150 WRITE(buf, (SSize_t)c); /* Don't write final \0 */
3151 PUTMARK((unsigned char) sizeof(int));
3152 PUTMARK((unsigned char) sizeof(long));
3153 PUTMARK((unsigned char) sizeof(char *));
3154 PUTMARK((unsigned char) sizeof(NV));
3156 TRACEME(("ok (magic_write byteorder = 0x%lx [%d], I%d L%d P%d D%d)",
3157 (unsigned long) BYTEORDER, (int) c,
3158 (int) sizeof(int), (int) sizeof(long),
3159 (int) sizeof(char *), (int) sizeof(NV)));
3167 * Common code for store operations.
3169 * When memory store is requested (f = NULL) and a non null SV* is given in
3170 * `res', it is filled with a new SV created out of the memory buffer.
3172 * It is required to provide a non-null `res' when the operation type is not
3173 * dclone() and store() is performed to memory.
3175 static int do_store(
3185 ASSERT(!(f == 0 && !(optype & ST_CLONE)) || res,
3186 ("must supply result SV pointer for real recursion to memory"));
3188 TRACEME(("do_store (optype=%d, netorder=%d)",
3189 optype, network_order));
3194 * Workaround for CROAK leak: if they enter with a "dirty" context,
3195 * free up memory for them now.
3202 * Now that STORABLE_xxx hooks exist, it is possible that they try to
3203 * re-enter store() via the hooks. We need to stack contexts.
3207 cxt = allocate_context(cxt);
3211 ASSERT(cxt->entry == 1, ("starting new recursion"));
3212 ASSERT(!cxt->s_dirty, ("clean context"));
3215 * Ensure sv is actually a reference. From perl, we called something
3217 * pstore(FILE, \@array);
3218 * so we must get the scalar value behing that reference.
3222 CROAK(("Not a reference"));
3223 sv = SvRV(sv); /* So follow it to know what to store */
3226 * If we're going to store to memory, reset the buffer.
3233 * Prepare context and emit headers.
3236 init_store_context(cxt, f, optype, network_order);
3238 if (-1 == magic_write(cxt)) /* Emit magic and ILP info */
3239 return 0; /* Error */
3242 * Recursively store object...
3245 ASSERT(is_storing(), ("within store operation"));
3247 status = store(cxt, sv); /* Just do it! */
3250 * If they asked for a memory store and they provided an SV pointer,
3251 * make an SV string out of the buffer and fill their pointer.
3253 * When asking for ST_REAL, it's MANDATORY for the caller to provide
3254 * an SV, since context cleanup might free the buffer if we did recurse.
3255 * (unless caller is dclone(), which is aware of that).
3258 if (!cxt->fio && res)
3264 * The "root" context is never freed, since it is meant to be always
3265 * handy for the common case where no recursion occurs at all (i.e.
3266 * we enter store() outside of any Storable code and leave it, period).
3267 * We know it's the "root" context because there's nothing stacked
3272 * When deep cloning, we don't free the context: doing so would force
3273 * us to copy the data in the memory buffer. Sicne we know we're
3274 * about to enter do_retrieve...
3277 clean_store_context(cxt);
3278 if (cxt->prev && !(cxt->optype & ST_CLONE))
3281 TRACEME(("do_store returns %d", status));
3289 * Store the transitive data closure of given object to disk.
3290 * Returns 0 on error, a true value otherwise.
3292 int pstore(PerlIO *f, SV *sv)
3294 TRACEME(("pstore"));
3295 return do_store(f, sv, 0, FALSE, (SV**) 0);
3302 * Same as pstore(), but network order is used for integers and doubles are
3303 * emitted as strings.
3305 int net_pstore(PerlIO *f, SV *sv)
3307 TRACEME(("net_pstore"));
3308 return do_store(f, sv, 0, TRUE, (SV**) 0);
3318 * Build a new SV out of the content of the internal memory buffer.
3320 static SV *mbuf2sv(void)
3324 return newSVpv(mbase, MBUF_SIZE());
3330 * Store the transitive data closure of given object to memory.
3331 * Returns undef on error, a scalar value containing the data otherwise.
3337 TRACEME(("mstore"));
3339 if (!do_store((PerlIO*) 0, sv, 0, FALSE, &out))
3340 return &PL_sv_undef;
3348 * Same as mstore(), but network order is used for integers and doubles are
3349 * emitted as strings.
3351 SV *net_mstore(SV *sv)
3355 TRACEME(("net_mstore"));
3357 if (!do_store((PerlIO*) 0, sv, 0, TRUE, &out))
3358 return &PL_sv_undef;
3364 *** Specific retrieve callbacks.
3370 * Return an error via croak, since it is not possible that we get here
3371 * under normal conditions, when facing a file produced via pstore().
3373 static SV *retrieve_other(stcxt_t *cxt, char *cname)
3376 cxt->ver_major != STORABLE_BIN_MAJOR &&
3377 cxt->ver_minor != STORABLE_BIN_MINOR
3379 CROAK(("Corrupted storable %s (binary v%d.%d), current is v%d.%d",
3380 cxt->fio ? "file" : "string",
3381 cxt->ver_major, cxt->ver_minor,
3382 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
3384 CROAK(("Corrupted storable %s (binary v%d.%d)",
3385 cxt->fio ? "file" : "string",
3386 cxt->ver_major, cxt->ver_minor));
3389 return (SV *) 0; /* Just in case */
3393 * retrieve_idx_blessed
3395 * Layout is SX_IX_BLESS <index> <object> with SX_IX_BLESS already read.
3396 * <index> can be coded on either 1 or 5 bytes.
3398 static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname)
3405 TRACEME(("retrieve_idx_blessed (#%d)", cxt->tagnum));
3406 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3408 GETMARK(idx); /* Index coded on a single char? */
3413 * Fetch classname in `aclass'
3416 sva = av_fetch(cxt->aclass, idx, FALSE);
3418 CROAK(("Class name #%"IVdf" should have been seen already", (IV) idx));
3420 class = SvPVX(*sva); /* We know it's a PV, by construction */
3422 TRACEME(("class ID %d => %s", idx, class));
3425 * Retrieve object and bless it.
3428 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
3436 * Layout is SX_BLESS <len> <classname> <object> with SX_BLESS already read.
3437 * <len> can be coded on either 1 or 5 bytes.
3439 static SV *retrieve_blessed(stcxt_t *cxt, char *cname)
3443 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3446 TRACEME(("retrieve_blessed (#%d)", cxt->tagnum));
3447 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3450 * Decode class name length and read that name.
3452 * Short classnames have two advantages: their length is stored on one
3453 * single byte, and the string can be read on the stack.
3456 GETMARK(len); /* Length coded on a single char? */
3459 TRACEME(("** allocating %d bytes for class name", len+1));
3460 New(10003, class, len+1, char);
3463 class[len] = '\0'; /* Mark string end */
3466 * It's a new classname, otherwise it would have been an SX_IX_BLESS.
3469 TRACEME(("new class name \"%s\" will bear ID = %d", class, cxt->classnum));
3471 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3475 * Retrieve object and bless it.
3478 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
3488 * Layout: SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
3489 * with leading mark already read, as usual.
3491 * When recursion was involved during serialization of the object, there
3492 * is an unknown amount of serialized objects after the SX_HOOK mark. Until
3493 * we reach a <flags> marker with the recursion bit cleared.
3495 * If the first <flags> byte contains a type of SHT_EXTRA, then the real type
3496 * is held in the <extra> byte, and if the object is tied, the serialized
3497 * magic object comes at the very end:
3499 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
3501 * This means the STORABLE_thaw hook will NOT get a tied variable during its
3502 * processing (since we won't have seen the magic object by the time the hook
3503 * is called). See comments below for why it was done that way.
3505 static SV *retrieve_hook(stcxt_t *cxt, char *cname)
3508 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3519 int clone = cxt->optype & ST_CLONE;
3521 unsigned int extra_type = 0;
3523 TRACEME(("retrieve_hook (#%d)", cxt->tagnum));
3524 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3527 * Read flags, which tell us about the type, and whether we need to recurse.
3533 * Create the (empty) object, and mark it as seen.
3535 * This must be done now, because tags are incremented, and during
3536 * serialization, the object tag was affected before recursion could
3540 obj_type = flags & SHF_TYPE_MASK;
3546 sv = (SV *) newAV();
3549 sv = (SV *) newHV();
3553 * Read <extra> flag to know the type of the object.
3554 * Record associated magic type for later.
3556 GETMARK(extra_type);
3557 switch (extra_type) {
3563 sv = (SV *) newAV();
3567 sv = (SV *) newHV();
3571 return retrieve_other(cxt, 0); /* Let it croak */
3575 return retrieve_other(cxt, 0); /* Let it croak */
3577 SEEN(sv, 0); /* Don't bless yet */
3580 * Whilst flags tell us to recurse, do so.
3582 * We don't need to remember the addresses returned by retrieval, because
3583 * all the references will be obtained through indirection via the object
3584 * tags in the object-ID list.
3587 while (flags & SHF_NEED_RECURSE) {
3588 TRACEME(("retrieve_hook recursing..."));
3589 rv = retrieve(cxt, 0);
3592 TRACEME(("retrieve_hook back with rv=0x%"UVxf,
3597 if (flags & SHF_IDX_CLASSNAME) {
3602 * Fetch index from `aclass'
3605 if (flags & SHF_LARGE_CLASSLEN)
3610 sva = av_fetch(cxt->aclass, idx, FALSE);
3612 CROAK(("Class name #%"IVdf" should have been seen already",
3615 class = SvPVX(*sva); /* We know it's a PV, by construction */
3616 TRACEME(("class ID %d => %s", idx, class));
3620 * Decode class name length and read that name.
3622 * NOTA BENE: even if the length is stored on one byte, we don't read
3623 * on the stack. Just like retrieve_blessed(), we limit the name to
3624 * LG_BLESS bytes. This is an arbitrary decision.
3627 if (flags & SHF_LARGE_CLASSLEN)
3632 if (len > LG_BLESS) {
3633 TRACEME(("** allocating %d bytes for class name", len+1));
3634 New(10003, class, len+1, char);
3638 class[len] = '\0'; /* Mark string end */
3641 * Record new classname.
3644 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3648 TRACEME(("class name: %s", class));
3651 * Decode user-frozen string length and read it in an SV.
3653 * For efficiency reasons, we read data directly into the SV buffer.
3654 * To understand that code, read retrieve_scalar()
3657 if (flags & SHF_LARGE_STRLEN)
3662 frozen = NEWSV(10002, len2);
3664 SAFEREAD(SvPVX(frozen), len2, frozen);
3665 SvCUR_set(frozen, len2);
3666 *SvEND(frozen) = '\0';
3668 (void) SvPOK_only(frozen); /* Validates string pointer */
3669 if (cxt->s_tainted) /* Is input source tainted? */
3672 TRACEME(("frozen string: %d bytes", len2));
3675 * Decode object-ID list length, if present.
3678 if (flags & SHF_HAS_LIST) {
3679 if (flags & SHF_LARGE_LISTLEN)
3685 av_extend(av, len3 + 1); /* Leave room for [0] */
3686 AvFILLp(av) = len3; /* About to be filled anyway */
3690 TRACEME(("has %d object IDs to link", len3));
3693 * Read object-ID list into array.
3694 * Because we pre-extended it, we can cheat and fill it manually.
3696 * We read object tags and we can convert them into SV* on the fly
3697 * because we know all the references listed in there (as tags)
3698 * have been already serialized, hence we have a valid correspondance
3699 * between each of those tags and the recreated SV.
3703 SV **ary = AvARRAY(av);
3705 for (i = 1; i <= len3; i++) { /* We leave [0] alone */
3712 svh = av_fetch(cxt->aseen, tag, FALSE);
3714 CROAK(("Object #%"IVdf" should have been retrieved already",
3717 ary[i] = SvREFCNT_inc(xsv);
3722 * Bless the object and look up the STORABLE_thaw hook.
3726 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3729 * Hook not found. Maybe they did not require the module where this
3730 * hook is defined yet?
3732 * If the require below succeeds, we'll be able to find the hook.
3733 * Still, it only works reliably when each class is defined in a
3737 SV *psv = newSVpvn("require ", 8);
3738 sv_catpv(psv, class);
3740 TRACEME(("No STORABLE_thaw defined for objects of class %s", class));
3741 TRACEME(("Going to require module '%s' with '%s'", class, SvPVX(psv)));
3743 perl_eval_sv(psv, G_DISCARD);
3747 * We cache results of pkg_can, so we need to uncache before attempting
3751 pkg_uncache(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3752 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3755 CROAK(("No STORABLE_thaw defined for objects of class %s "
3756 "(even after a \"require %s;\")", class, class));
3760 * If we don't have an `av' yet, prepare one.
3761 * Then insert the frozen string as item [0].
3769 AvARRAY(av)[0] = SvREFCNT_inc(frozen);
3774 * $object->STORABLE_thaw($cloning, $frozen, @refs);
3776 * where $object is our blessed (empty) object, $cloning is a boolean
3777 * telling whether we're running a deep clone, $frozen is the frozen
3778 * string the user gave us in his serializing hook, and @refs, which may
3779 * be empty, is the list of extra references he returned along for us
3782 * In effect, the hook is an alternate creation routine for the class,
3783 * the object itself being already created by the runtime.
3786 TRACEME(("calling STORABLE_thaw on %s at 0x%"UVxf" (%"IVdf" args)",
3787 class, PTR2UV(sv), AvFILLp(av) + 1));
3790 (void) scalar_call(rv, hook, clone, av, G_SCALAR|G_DISCARD);
3797 SvREFCNT_dec(frozen);
3800 if (!(flags & SHF_IDX_CLASSNAME) && class != buf)
3804 * If we had an <extra> type, then the object was not as simple, and
3805 * we need to restore extra magic now.
3811 TRACEME(("retrieving magic object for 0x%"UVxf"...", PTR2UV(sv)));
3813 rv = retrieve(cxt, 0); /* Retrieve <magic object> */
3815 TRACEME(("restoring the magic object 0x%"UVxf" part of 0x%"UVxf,
3816 PTR2UV(rv), PTR2UV(sv)));
3818 switch (extra_type) {
3820 sv_upgrade(sv, SVt_PVMG);
3823 sv_upgrade(sv, SVt_PVAV);
3824 AvREAL_off((AV *)sv);
3827 sv_upgrade(sv, SVt_PVHV);
3830 CROAK(("Forgot to deal with extra type %d", extra_type));
3835 * Adding the magic only now, well after the STORABLE_thaw hook was called
3836 * means the hook cannot know it deals with an object whose variable is
3837 * tied. But this is happening when retrieving $o in the following case:
3841 * my $o = bless \%h, 'BAR';
3843 * The 'BAR' class is NOT the one where %h is tied into. Therefore, as
3844 * far as the 'BAR' class is concerned, the fact that %h is not a REAL
3845 * hash but a tied one should not matter at all, and remain transparent.
3846 * This means the magic must be restored by Storable AFTER the hook is
3849 * That looks very reasonable to me, but then I've come up with this
3850 * after a bug report from David Nesting, who was trying to store such
3851 * an object and caused Storable to fail. And unfortunately, it was
3852 * also the easiest way to retrofit support for blessed ref to tied objects
3853 * into the existing design. -- RAM, 17/02/2001
3856 sv_magic(sv, rv, mtype, Nullch, 0);
3857 SvREFCNT_dec(rv); /* Undo refcnt inc from sv_magic() */
3865 * Retrieve reference to some other scalar.
3866 * Layout is SX_REF <object>, with SX_REF already read.
3868 static SV *retrieve_ref(stcxt_t *cxt, char *cname)
3873 TRACEME(("retrieve_ref (#%d)", cxt->tagnum));
3876 * We need to create the SV that holds the reference to the yet-to-retrieve
3877 * object now, so that we may record the address in the seen table.
3878 * Otherwise, if the object to retrieve references us, we won't be able
3879 * to resolve the SX_OBJECT we'll see at that point! Hence we cannot
3880 * do the retrieve first and use rv = newRV(sv) since it will be too late
3881 * for SEEN() recording.
3884 rv = NEWSV(10002, 0);
3885 SEEN(rv, cname); /* Will return if rv is null */
3886 sv = retrieve(cxt, 0); /* Retrieve <object> */
3888 return (SV *) 0; /* Failed */
3891 * WARNING: breaks RV encapsulation.
3893 * Now for the tricky part. We have to upgrade our existing SV, so that
3894 * it is now an RV on sv... Again, we cheat by duplicating the code
3895 * held in newSVrv(), since we already got our SV from retrieve().
3899 * SvRV(rv) = SvREFCNT_inc(sv);
3901 * here because the reference count we got from retrieve() above is
3902 * already correct: if the object was retrieved from the file, then
3903 * its reference count is one. Otherwise, if it was retrieved via
3904 * an SX_OBJECT indication, a ref count increment was done.
3907 sv_upgrade(rv, SVt_RV);
3908 SvRV(rv) = sv; /* $rv = \$sv */
3911 TRACEME(("ok (retrieve_ref at 0x%"UVxf")", PTR2UV(rv)));
3917 * retrieve_overloaded
3919 * Retrieve reference to some other scalar with overloading.
3920 * Layout is SX_OVERLOAD <object>, with SX_OVERLOAD already read.
3922 static SV *retrieve_overloaded(stcxt_t *cxt, char *cname)
3928 TRACEME(("retrieve_overloaded (#%d)", cxt->tagnum));
3931 * Same code as retrieve_ref(), duplicated to avoid extra call.
3934 rv = NEWSV(10002, 0);
3935 SEEN(rv, cname); /* Will return if rv is null */
3936 sv = retrieve(cxt, 0); /* Retrieve <object> */
3938 return (SV *) 0; /* Failed */
3941 * WARNING: breaks RV encapsulation.
3944 sv_upgrade(rv, SVt_RV);
3945 SvRV(rv) = sv; /* $rv = \$sv */
3949 * Restore overloading magic.
3952 stash = (HV *) SvSTASH (sv);
3953 if (!stash || !Gv_AMG(stash))
3954 CROAK(("Cannot restore overloading on %s(0x%"UVxf") (package %s)",
3955 sv_reftype(sv, FALSE),
3957 stash ? HvNAME(stash) : "<unknown>"));
3961 TRACEME(("ok (retrieve_overloaded at 0x%"UVxf")", PTR2UV(rv)));
3967 * retrieve_tied_array
3969 * Retrieve tied array
3970 * Layout is SX_TIED_ARRAY <object>, with SX_TIED_ARRAY already read.
3972 static SV *retrieve_tied_array(stcxt_t *cxt, char *cname)
3977 TRACEME(("retrieve_tied_array (#%d)", cxt->tagnum));
3979 tv = NEWSV(10002, 0);
3980 SEEN(tv, cname); /* Will return if tv is null */
3981 sv = retrieve(cxt, 0); /* Retrieve <object> */
3983 return (SV *) 0; /* Failed */
3985 sv_upgrade(tv, SVt_PVAV);
3986 AvREAL_off((AV *)tv);
3987 sv_magic(tv, sv, 'P', Nullch, 0);
3988 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3990 TRACEME(("ok (retrieve_tied_array at 0x%"UVxf")", PTR2UV(tv)));
3996 * retrieve_tied_hash
3998 * Retrieve tied hash
3999 * Layout is SX_TIED_HASH <object>, with SX_TIED_HASH already read.
4001 static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname)
4006 TRACEME(("retrieve_tied_hash (#%d)", cxt->tagnum));
4008 tv = NEWSV(10002, 0);
4009 SEEN(tv, cname); /* Will return if tv is null */
4010 sv = retrieve(cxt, 0); /* Retrieve <object> */
4012 return (SV *) 0; /* Failed */
4014 sv_upgrade(tv, SVt_PVHV);
4015 sv_magic(tv, sv, 'P', Nullch, 0);
4016 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4018 TRACEME(("ok (retrieve_tied_hash at 0x%"UVxf")", PTR2UV(tv)));
4024 * retrieve_tied_scalar
4026 * Retrieve tied scalar
4027 * Layout is SX_TIED_SCALAR <object>, with SX_TIED_SCALAR already read.
4029 static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname)
4034 TRACEME(("retrieve_tied_scalar (#%d)", cxt->tagnum));
4036 tv = NEWSV(10002, 0);
4037 SEEN(tv, cname); /* Will return if rv is null */
4038 sv = retrieve(cxt, 0); /* Retrieve <object> */
4040 return (SV *) 0; /* Failed */
4042 sv_upgrade(tv, SVt_PVMG);
4043 sv_magic(tv, sv, 'q', Nullch, 0);
4044 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4046 TRACEME(("ok (retrieve_tied_scalar at 0x%"UVxf")", PTR2UV(tv)));
4054 * Retrieve reference to value in a tied hash.
4055 * Layout is SX_TIED_KEY <object> <key>, with SX_TIED_KEY already read.
4057 static SV *retrieve_tied_key(stcxt_t *cxt, char *cname)
4063 TRACEME(("retrieve_tied_key (#%d)", cxt->tagnum));
4065 tv = NEWSV(10002, 0);
4066 SEEN(tv, cname); /* Will return if tv is null */
4067 sv = retrieve(cxt, 0); /* Retrieve <object> */
4069 return (SV *) 0; /* Failed */
4071 key = retrieve(cxt, 0); /* Retrieve <key> */
4073 return (SV *) 0; /* Failed */
4075 sv_upgrade(tv, SVt_PVMG);
4076 sv_magic(tv, sv, 'p', (char *)key, HEf_SVKEY);
4077 SvREFCNT_dec(key); /* Undo refcnt inc from sv_magic() */
4078 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4086 * Retrieve reference to value in a tied array.
4087 * Layout is SX_TIED_IDX <object> <idx>, with SX_TIED_IDX already read.
4089 static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname)
4095 TRACEME(("retrieve_tied_idx (#%d)", cxt->tagnum));
4097 tv = NEWSV(10002, 0);
4098 SEEN(tv, cname); /* Will return if tv is null */
4099 sv = retrieve(cxt, 0); /* Retrieve <object> */
4101 return (SV *) 0; /* Failed */
4103 RLEN(idx); /* Retrieve <idx> */
4105 sv_upgrade(tv, SVt_PVMG);
4106 sv_magic(tv, sv, 'p', Nullch, idx);
4107 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4116 * Retrieve defined long (string) scalar.
4118 * Layout is SX_LSCALAR <length> <data>, with SX_LSCALAR already read.
4119 * The scalar is "long" in that <length> is larger than LG_SCALAR so it
4120 * was not stored on a single byte.
4122 static SV *retrieve_lscalar(stcxt_t *cxt, char *cname)
4128 TRACEME(("retrieve_lscalar (#%d), len = %"IVdf, cxt->tagnum, len));
4131 * Allocate an empty scalar of the suitable length.
4134 sv = NEWSV(10002, len);
4135 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4138 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
4140 * Now, for efficiency reasons, read data directly inside the SV buffer,
4141 * and perform the SV final settings directly by duplicating the final
4142 * work done by sv_setpv. Since we're going to allocate lots of scalars
4143 * this way, it's worth the hassle and risk.
4146 SAFEREAD(SvPVX(sv), len, sv);
4147 SvCUR_set(sv, len); /* Record C string length */
4148 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
4149 (void) SvPOK_only(sv); /* Validate string pointer */
4150 if (cxt->s_tainted) /* Is input source tainted? */
4151 SvTAINT(sv); /* External data cannot be trusted */
4153 TRACEME(("large scalar len %"IVdf" '%s'", len, SvPVX(sv)));
4154 TRACEME(("ok (retrieve_lscalar at 0x%"UVxf")", PTR2UV(sv)));
4162 * Retrieve defined short (string) scalar.
4164 * Layout is SX_SCALAR <length> <data>, with SX_SCALAR already read.
4165 * The scalar is "short" so <length> is single byte. If it is 0, there
4166 * is no <data> section.
4168 static SV *retrieve_scalar(stcxt_t *cxt, char *cname)
4174 TRACEME(("retrieve_scalar (#%d), len = %d", cxt->tagnum, len));
4177 * Allocate an empty scalar of the suitable length.
4180 sv = NEWSV(10002, len);
4181 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4184 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
4189 * newSV did not upgrade to SVt_PV so the scalar is undefined.
4190 * To make it defined with an empty length, upgrade it now...
4191 * Don't upgrade to a PV if the original type contains more
4192 * information than a scalar.
4194 if (SvTYPE(sv) <= SVt_PV) {
4195 sv_upgrade(sv, SVt_PV);
4198 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
4199 TRACEME(("ok (retrieve_scalar empty at 0x%"UVxf")", PTR2UV(sv)));
4202 * Now, for efficiency reasons, read data directly inside the SV buffer,
4203 * and perform the SV final settings directly by duplicating the final
4204 * work done by sv_setpv. Since we're going to allocate lots of scalars
4205 * this way, it's worth the hassle and risk.
4207 SAFEREAD(SvPVX(sv), len, sv);
4208 SvCUR_set(sv, len); /* Record C string length */
4209 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
4210 TRACEME(("small scalar len %d '%s'", len, SvPVX(sv)));
4213 (void) SvPOK_only(sv); /* Validate string pointer */
4214 if (cxt->s_tainted) /* Is input source tainted? */
4215 SvTAINT(sv); /* External data cannot be trusted */
4217 TRACEME(("ok (retrieve_scalar at 0x%"UVxf")", PTR2UV(sv)));
4224 * Like retrieve_scalar(), but tag result as utf8.
4225 * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
4227 static SV *retrieve_utf8str(stcxt_t *cxt, char *cname)
4231 TRACEME(("retrieve_utf8str"));
4233 sv = retrieve_scalar(cxt, cname);
4235 #ifdef HAS_UTF8_SCALARS
4238 if (cxt->use_bytes < 0)
4240 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4242 if (cxt->use_bytes == 0)
4253 * Like retrieve_lscalar(), but tag result as utf8.
4254 * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
4256 static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname)
4260 TRACEME(("retrieve_lutf8str"));
4262 sv = retrieve_lscalar(cxt, cname);
4264 #ifdef HAS_UTF8_SCALARS
4267 if (cxt->use_bytes < 0)
4269 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4271 if (cxt->use_bytes == 0)
4281 * Retrieve defined integer.
4282 * Layout is SX_INTEGER <data>, whith SX_INTEGER already read.
4284 static SV *retrieve_integer(stcxt_t *cxt, char *cname)
4289 TRACEME(("retrieve_integer (#%d)", cxt->tagnum));
4291 READ(&iv, sizeof(iv));
4293 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4295 TRACEME(("integer %"IVdf, iv));
4296 TRACEME(("ok (retrieve_integer at 0x%"UVxf")", PTR2UV(sv)));
4304 * Retrieve defined integer in network order.
4305 * Layout is SX_NETINT <data>, whith SX_NETINT already read.
4307 static SV *retrieve_netint(stcxt_t *cxt, char *cname)
4312 TRACEME(("retrieve_netint (#%d)", cxt->tagnum));
4316 sv = newSViv((int) ntohl(iv));
4317 TRACEME(("network integer %d", (int) ntohl(iv)));
4320 TRACEME(("network integer (as-is) %d", iv));
4322 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4324 TRACEME(("ok (retrieve_netint at 0x%"UVxf")", PTR2UV(sv)));
4332 * Retrieve defined double.
4333 * Layout is SX_DOUBLE <data>, whith SX_DOUBLE already read.
4335 static SV *retrieve_double(stcxt_t *cxt, char *cname)
4340 TRACEME(("retrieve_double (#%d)", cxt->tagnum));
4342 READ(&nv, sizeof(nv));
4344 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4346 TRACEME(("double %"NVff, nv));
4347 TRACEME(("ok (retrieve_double at 0x%"UVxf")", PTR2UV(sv)));
4355 * Retrieve defined byte (small integer within the [-128, +127] range).
4356 * Layout is SX_BYTE <data>, whith SX_BYTE already read.
4358 static SV *retrieve_byte(stcxt_t *cxt, char *cname)
4362 signed char tmp; /* Workaround for AIX cc bug --H.Merijn Brand */
4364 TRACEME(("retrieve_byte (#%d)", cxt->tagnum));
4367 TRACEME(("small integer read as %d", (unsigned char) siv));
4368 tmp = (unsigned char) siv - 128;
4370 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4372 TRACEME(("byte %d", tmp));
4373 TRACEME(("ok (retrieve_byte at 0x%"UVxf")", PTR2UV(sv)));
4381 * Return the undefined value.
4383 static SV *retrieve_undef(stcxt_t *cxt, char *cname)
4387 TRACEME(("retrieve_undef"));
4398 * Return the immortal undefined value.
4400 static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname)
4402 SV *sv = &PL_sv_undef;
4404 TRACEME(("retrieve_sv_undef"));
4413 * Return the immortal yes value.
4415 static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname)
4417 SV *sv = &PL_sv_yes;
4419 TRACEME(("retrieve_sv_yes"));
4428 * Return the immortal no value.
4430 static SV *retrieve_sv_no(stcxt_t *cxt, char *cname)
4434 TRACEME(("retrieve_sv_no"));
4443 * Retrieve a whole array.
4444 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
4445 * Each item is stored as <object>.
4447 * When we come here, SX_ARRAY has been read already.
4449 static SV *retrieve_array(stcxt_t *cxt, char *cname)
4456 TRACEME(("retrieve_array (#%d)", cxt->tagnum));
4459 * Read length, and allocate array, then pre-extend it.
4463 TRACEME(("size = %d", len));
4465 SEEN(av, cname); /* Will return if array not allocated nicely */
4469 return (SV *) av; /* No data follow if array is empty */
4472 * Now get each item in turn...
4475 for (i = 0; i < len; i++) {
4476 TRACEME(("(#%d) item", i));
4477 sv = retrieve(cxt, 0); /* Retrieve item */
4480 if (av_store(av, i, sv) == 0)
4484 TRACEME(("ok (retrieve_array at 0x%"UVxf")", PTR2UV(av)));
4492 * Retrieve a whole hash table.
4493 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4494 * Keys are stored as <length> <data>, the <data> section being omitted
4496 * Values are stored as <object>.
4498 * When we come here, SX_HASH has been read already.
4500 static SV *retrieve_hash(stcxt_t *cxt, char *cname)
4508 TRACEME(("retrieve_hash (#%d)", cxt->tagnum));
4511 * Read length, allocate table.
4515 TRACEME(("size = %d", len));
4517 SEEN(hv, cname); /* Will return if table not allocated properly */
4519 return (SV *) hv; /* No data follow if table empty */
4520 hv_ksplit(hv, len); /* pre-extend hash to save multiple splits */
4523 * Now get each key/value pair in turn...
4526 for (i = 0; i < len; i++) {
4531 TRACEME(("(#%d) value", i));
4532 sv = retrieve(cxt, 0);
4538 * Since we're reading into kbuf, we must ensure we're not
4539 * recursing between the read and the hv_store() where it's used.
4540 * Hence the key comes after the value.
4543 RLEN(size); /* Get key size */
4544 KBUFCHK((STRLEN)size); /* Grow hash key read pool if needed */
4547 kbuf[size] = '\0'; /* Mark string end, just in case */
4548 TRACEME(("(#%d) key '%s'", i, kbuf));
4551 * Enter key/value pair into hash table.
4554 if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
4558 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4566 * Retrieve a whole hash table.
4567 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4568 * Keys are stored as <length> <data>, the <data> section being omitted
4570 * Values are stored as <object>.
4572 * When we come here, SX_HASH has been read already.
4574 static SV *retrieve_flag_hash(stcxt_t *cxt, char *cname)
4583 GETMARK(hash_flags);
4584 TRACEME(("retrieve_flag_hash (#%d)", cxt->tagnum));
4586 * Read length, allocate table.
4589 #ifndef HAS_RESTRICTED_HASHES
4590 if (hash_flags & SHV_RESTRICTED) {
4591 if (cxt->derestrict < 0)
4593 = (SvTRUE(perl_get_sv("Storable::downgrade_restricted", TRUE))
4595 if (cxt->derestrict == 0)
4596 RESTRICTED_HASH_CROAK();
4601 TRACEME(("size = %d, flags = %d", len, hash_flags));
4603 SEEN(hv, cname); /* Will return if table not allocated properly */
4605 return (SV *) hv; /* No data follow if table empty */
4606 hv_ksplit(hv, len); /* pre-extend hash to save multiple splits */
4609 * Now get each key/value pair in turn...
4612 for (i = 0; i < len; i++) {
4614 int store_flags = 0;
4619 TRACEME(("(#%d) value", i));
4620 sv = retrieve(cxt, 0);
4625 #ifdef HAS_RESTRICTED_HASHES
4626 if ((hash_flags & SHV_RESTRICTED) && (flags & SHV_K_LOCKED))
4630 if (flags & SHV_K_ISSV) {
4631 /* XXX you can't set a placeholder with an SV key.
4632 Then again, you can't get an SV key.
4633 Without messing around beyond what the API is supposed to do.
4636 TRACEME(("(#%d) keysv, flags=%d", i, flags));
4637 keysv = retrieve(cxt, 0);
4641 if (!hv_store_ent(hv, keysv, sv, 0))
4646 * Since we're reading into kbuf, we must ensure we're not
4647 * recursing between the read and the hv_store() where it's used.
4648 * Hence the key comes after the value.
4651 if (flags & SHV_K_PLACEHOLDER) {
4654 store_flags |= HVhek_PLACEHOLD;
4656 if (flags & SHV_K_UTF8) {
4657 #ifdef HAS_UTF8_HASHES
4658 store_flags |= HVhek_UTF8;
4660 if (cxt->use_bytes < 0)
4662 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4664 if (cxt->use_bytes == 0)
4668 #ifdef HAS_UTF8_HASHES
4669 if (flags & SHV_K_WASUTF8)
4670 store_flags |= HVhek_WASUTF8;
4673 RLEN(size); /* Get key size */
4674 KBUFCHK((STRLEN)size); /* Grow hash key read pool if needed */
4677 kbuf[size] = '\0'; /* Mark string end, just in case */
4678 TRACEME(("(#%d) key '%s' flags %X store_flags %X", i, kbuf,
4679 flags, store_flags));
4682 * Enter key/value pair into hash table.
4685 #ifdef HAS_RESTRICTED_HASHES
4686 if (hv_store_flags(hv, kbuf, size, sv, 0, flags) == 0)
4689 if (!(store_flags & HVhek_PLACEHOLD))
4690 if (hv_store(hv, kbuf, size, sv, 0) == 0)
4695 #ifdef HAS_RESTRICTED_HASHES
4696 if (hash_flags & SHV_RESTRICTED)
4700 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4706 * old_retrieve_array
4708 * Retrieve a whole array in pre-0.6 binary format.
4710 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
4711 * Each item is stored as SX_ITEM <object> or SX_IT_UNDEF for "holes".
4713 * When we come here, SX_ARRAY has been read already.
4715 static SV *old_retrieve_array(stcxt_t *cxt, char *cname)
4723 TRACEME(("old_retrieve_array (#%d)", cxt->tagnum));
4726 * Read length, and allocate array, then pre-extend it.
4730 TRACEME(("size = %d", len));
4732 SEEN(av, 0); /* Will return if array not allocated nicely */
4736 return (SV *) av; /* No data follow if array is empty */
4739 * Now get each item in turn...
4742 for (i = 0; i < len; i++) {
4744 if (c == SX_IT_UNDEF) {
4745 TRACEME(("(#%d) undef item", i));
4746 continue; /* av_extend() already filled us with undef */
4749 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
4750 TRACEME(("(#%d) item", i));
4751 sv = retrieve(cxt, 0); /* Retrieve item */
4754 if (av_store(av, i, sv) == 0)
4758 TRACEME(("ok (old_retrieve_array at 0x%"UVxf")", PTR2UV(av)));
4766 * Retrieve a whole hash table in pre-0.6 binary format.
4768 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4769 * Keys are stored as SX_KEY <length> <data>, the <data> section being omitted
4771 * Values are stored as SX_VALUE <object> or SX_VL_UNDEF for "holes".
4773 * When we come here, SX_HASH has been read already.
4775 static SV *old_retrieve_hash(stcxt_t *cxt, char *cname)
4783 static SV *sv_h_undef = (SV *) 0; /* hv_store() bug */
4785 TRACEME(("old_retrieve_hash (#%d)", cxt->tagnum));
4788 * Read length, allocate table.
4792 TRACEME(("size = %d", len));
4794 SEEN(hv, 0); /* Will return if table not allocated properly */
4796 return (SV *) hv; /* No data follow if table empty */
4797 hv_ksplit(hv, len); /* pre-extend hash to save multiple splits */
4800 * Now get each key/value pair in turn...
4803 for (i = 0; i < len; i++) {
4809 if (c == SX_VL_UNDEF) {
4810 TRACEME(("(#%d) undef value", i));
4812 * Due to a bug in hv_store(), it's not possible to pass
4813 * &PL_sv_undef to hv_store() as a value, otherwise the
4814 * associated key will not be creatable any more. -- RAM, 14/01/97
4817 sv_h_undef = newSVsv(&PL_sv_undef);
4818 sv = SvREFCNT_inc(sv_h_undef);
4819 } else if (c == SX_VALUE) {
4820 TRACEME(("(#%d) value", i));
4821 sv = retrieve(cxt, 0);
4825 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
4829 * Since we're reading into kbuf, we must ensure we're not
4830 * recursing between the read and the hv_store() where it's used.
4831 * Hence the key comes after the value.
4836 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
4837 RLEN(size); /* Get key size */
4838 KBUFCHK((STRLEN)size); /* Grow hash key read pool if needed */
4841 kbuf[size] = '\0'; /* Mark string end, just in case */
4842 TRACEME(("(#%d) key '%s'", i, kbuf));
4845 * Enter key/value pair into hash table.
4848 if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
4852 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4858 *** Retrieval engine.
4864 * Make sure the stored data we're trying to retrieve has been produced
4865 * on an ILP compatible system with the same byteorder. It croaks out in
4866 * case an error is detected. [ILP = integer-long-pointer sizes]
4867 * Returns null if error is detected, &PL_sv_undef otherwise.
4869 * Note that there's no byte ordering info emitted when network order was
4870 * used at store time.
4872 static SV *magic_check(stcxt_t *cxt)
4875 char byteorder[256];
4877 int use_network_order;
4879 int version_minor = 0;
4881 TRACEME(("magic_check"));
4884 * The "magic number" is only for files, not when freezing in memory.
4888 STRLEN len = sizeof(magicstr) - 1;
4891 READ(buf, (SSize_t)len); /* Not null-terminated */
4892 buf[len] = '\0'; /* Is now */
4894 if (0 == strcmp(buf, magicstr))
4898 * Try to read more bytes to check for the old magic number, which
4902 old_len = sizeof(old_magicstr) - 1;
4903 READ(&buf[len], (SSize_t)(old_len - len));
4904 buf[old_len] = '\0'; /* Is now null-terminated */
4906 if (strcmp(buf, old_magicstr))
4907 CROAK(("File is not a perl storable"));
4912 * Starting with 0.6, the "use_network_order" byte flag is also used to
4913 * indicate the version number of the binary, and therefore governs the
4914 * setting of sv_retrieve_vtbl. See magic_write().
4917 GETMARK(use_network_order);
4918 version_major = use_network_order >> 1;
4919 cxt->retrieve_vtbl = version_major ? sv_retrieve : sv_old_retrieve;
4921 TRACEME(("magic_check: netorder = 0x%x", use_network_order));
4925 * Starting with 0.7 (binary major 2), a full byte is dedicated to the
4926 * minor version of the protocol. See magic_write().
4929 if (version_major > 1)
4930 GETMARK(version_minor);
4932 cxt->ver_major = version_major;
4933 cxt->ver_minor = version_minor;
4935 TRACEME(("binary image version is %d.%d", version_major, version_minor));
4938 * Inter-operability sanity check: we can't retrieve something stored
4939 * using a format more recent than ours, because we have no way to
4940 * know what has changed, and letting retrieval go would mean a probable
4941 * failure reporting a "corrupted" storable file.
4945 version_major > STORABLE_BIN_MAJOR ||
4946 (version_major == STORABLE_BIN_MAJOR &&
4947 version_minor > STORABLE_BIN_MINOR)
4950 TRACEME(("but I am version is %d.%d", STORABLE_BIN_MAJOR,
4951 STORABLE_BIN_MINOR));
4953 if (version_major == STORABLE_BIN_MAJOR) {
4954 TRACEME(("cxt->accept_future_minor is %d",
4955 cxt->accept_future_minor));
4956 if (cxt->accept_future_minor < 0)
4957 cxt->accept_future_minor
4958 = (SvTRUE(perl_get_sv("Storable::accept_future_minor",
4961 if (cxt->accept_future_minor == 1)
4962 croak_now = 0; /* Don't croak yet. */
4965 CROAK(("Storable binary image v%d.%d more recent than I am (v%d.%d)",
4966 version_major, version_minor,
4967 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
4972 * If they stored using network order, there's no byte ordering
4973 * information to check.
4976 if ((cxt->netorder = (use_network_order & 0x1))) /* Extra () for -Wall */
4977 return &PL_sv_undef; /* No byte ordering info */
4979 sprintf(byteorder, "%lx", (unsigned long) BYTEORDER);
4981 READ(buf, c); /* Not null-terminated */
4982 buf[c] = '\0'; /* Is now */
4984 TRACEME(("byte order '%s'", buf));
4986 if (strcmp(buf, byteorder))
4987 CROAK(("Byte order is not compatible"));
4989 GETMARK(c); /* sizeof(int) */
4990 if ((int) c != sizeof(int))
4991 CROAK(("Integer size is not compatible"));
4993 GETMARK(c); /* sizeof(long) */
4994 if ((int) c != sizeof(long))
4995 CROAK(("Long integer size is not compatible"));
4997 GETMARK(c); /* sizeof(char *) */
4998 if ((int) c != sizeof(char *))
4999 CROAK(("Pointer integer size is not compatible"));
5001 if (version_major >= 2 && version_minor >= 2) {
5002 GETMARK(c); /* sizeof(NV) */
5003 if ((int) c != sizeof(NV))
5004 CROAK(("Double size is not compatible"));
5007 return &PL_sv_undef; /* OK */
5013 * Recursively retrieve objects from the specified file and return their
5014 * root SV (which may be an AV or an HV for what we care).
5015 * Returns null if there is a problem.
5017 static SV *retrieve(stcxt_t *cxt, char *cname)
5023 TRACEME(("retrieve"));
5026 * Grab address tag which identifies the object if we are retrieving
5027 * an older format. Since the new binary format counts objects and no
5028 * longer explicitely tags them, we must keep track of the correspondance
5031 * The following section will disappear one day when the old format is
5032 * no longer supported, hence the final "goto" in the "if" block.
5035 if (cxt->hseen) { /* Retrieving old binary */
5037 if (cxt->netorder) {
5039 READ(&nettag, sizeof(I32)); /* Ordered sequence of I32 */
5040 tag = (stag_t) nettag;
5042 READ(&tag, sizeof(stag_t)); /* Original address of the SV */
5045 if (type == SX_OBJECT) {
5047 svh = hv_fetch(cxt->hseen, (char *) &tag, sizeof(tag), FALSE);
5049 CROAK(("Old tag 0x%"UVxf" should have been mapped already",
5051 tagn = SvIV(*svh); /* Mapped tag number computed earlier below */
5054 * The following code is common with the SX_OBJECT case below.
5057 svh = av_fetch(cxt->aseen, tagn, FALSE);
5059 CROAK(("Object #%"IVdf" should have been retrieved already",
5062 TRACEME(("has retrieved #%d at 0x%"UVxf, tagn, PTR2UV(sv)));
5063 SvREFCNT_inc(sv); /* One more reference to this same sv */
5064 return sv; /* The SV pointer where object was retrieved */
5068 * Map new object, but don't increase tagnum. This will be done
5069 * by each of the retrieve_* functions when they call SEEN().
5071 * The mapping associates the "tag" initially present with a unique
5072 * tag number. See test for SX_OBJECT above to see how this is perused.
5075 if (!hv_store(cxt->hseen, (char *) &tag, sizeof(tag),
5076 newSViv(cxt->tagnum), 0))
5083 * Regular post-0.6 binary format.
5088 TRACEME(("retrieve type = %d", type));
5091 * Are we dealing with an object we should have already retrieved?
5094 if (type == SX_OBJECT) {
5098 svh = av_fetch(cxt->aseen, tag, FALSE);
5100 CROAK(("Object #%"IVdf" should have been retrieved already",
5103 TRACEME(("had retrieved #%d at 0x%"UVxf, tag, PTR2UV(sv)));
5104 SvREFCNT_inc(sv); /* One more reference to this same sv */
5105 return sv; /* The SV pointer where object was retrieved */
5106 } else if (type >= SX_ERROR && cxt->ver_minor > STORABLE_BIN_MINOR) {
5107 if (cxt->accept_future_minor < 0)
5108 cxt->accept_future_minor
5109 = (SvTRUE(perl_get_sv("Storable::accept_future_minor",
5112 if (cxt->accept_future_minor == 1) {
5113 CROAK(("Storable binary image v%d.%d contains data of type %d. "
5114 "This Storable is v%d.%d and can only handle data types up to %d",
5115 cxt->ver_major, cxt->ver_minor, type,
5116 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR, SX_ERROR - 1));
5120 first_time: /* Will disappear when support for old format is dropped */
5123 * Okay, first time through for this one.
5126 sv = RETRIEVE(cxt, type)(cxt, cname);
5128 return (SV *) 0; /* Failed */
5131 * Old binary formats (pre-0.7).
5133 * Final notifications, ended by SX_STORED may now follow.
5134 * Currently, the only pertinent notification to apply on the
5135 * freshly retrieved object is either:
5136 * SX_CLASS <char-len> <classname> for short classnames.
5137 * SX_LG_CLASS <int-len> <classname> for larger one (rare!).
5138 * Class name is then read into the key buffer pool used by
5139 * hash table key retrieval.
5142 if (cxt->ver_major < 2) {
5143 while ((type = GETCHAR()) != SX_STORED) {
5147 GETMARK(len); /* Length coded on a single char */
5149 case SX_LG_CLASS: /* Length coded on a regular integer */
5154 return (SV *) 0; /* Failed */
5156 KBUFCHK((STRLEN)len); /* Grow buffer as necessary */
5159 kbuf[len] = '\0'; /* Mark string end */
5164 TRACEME(("ok (retrieved 0x%"UVxf", refcnt=%d, %s)", PTR2UV(sv),
5165 SvREFCNT(sv) - 1, sv_reftype(sv, FALSE)));
5173 * Retrieve data held in file and return the root object.
5174 * Common routine for pretrieve and mretrieve.
5176 static SV *do_retrieve(
5183 int is_tainted; /* Is input source tainted? */
5184 int pre_06_fmt = 0; /* True with pre Storable 0.6 formats */
5186 TRACEME(("do_retrieve (optype = 0x%x)", optype));
5188 optype |= ST_RETRIEVE;
5191 * Sanity assertions for retrieve dispatch tables.
5194 ASSERT(sizeof(sv_old_retrieve) == sizeof(sv_retrieve),
5195 ("old and new retrieve dispatch table have same size"));
5196 ASSERT(sv_old_retrieve[SX_ERROR] == retrieve_other,
5197 ("SX_ERROR entry correctly initialized in old dispatch table"));
5198 ASSERT(sv_retrieve[SX_ERROR] == retrieve_other,
5199 ("SX_ERROR entry correctly initialized in new dispatch table"));
5202 * Workaround for CROAK leak: if they enter with a "dirty" context,
5203 * free up memory for them now.
5210 * Now that STORABLE_xxx hooks exist, it is possible that they try to
5211 * re-enter retrieve() via the hooks.
5215 cxt = allocate_context(cxt);
5219 ASSERT(cxt->entry == 1, ("starting new recursion"));
5220 ASSERT(!cxt->s_dirty, ("clean context"));
5225 * Data is loaded into the memory buffer when f is NULL, unless `in' is
5226 * also NULL, in which case we're expecting the data to already lie
5227 * in the buffer (dclone case).
5230 KBUFINIT(); /* Allocate hash key reading pool once */
5233 MBUF_SAVE_AND_LOAD(in);
5236 * Magic number verifications.
5238 * This needs to be done before calling init_retrieve_context()
5239 * since the format indication in the file are necessary to conduct
5240 * some of the initializations.
5243 cxt->fio = f; /* Where I/O are performed */
5245 if (!magic_check(cxt))
5246 CROAK(("Magic number checking on storable %s failed",
5247 cxt->fio ? "file" : "string"));
5249 TRACEME(("data stored in %s format",
5250 cxt->netorder ? "net order" : "native"));
5253 * Check whether input source is tainted, so that we don't wrongly
5254 * taint perfectly good values...
5256 * We assume file input is always tainted. If both `f' and `in' are
5257 * NULL, then we come from dclone, and tainted is already filled in
5258 * the context. That's a kludge, but the whole dclone() thing is
5259 * already quite a kludge anyway! -- RAM, 15/09/2000.
5262 is_tainted = f ? 1 : (in ? SvTAINTED(in) : cxt->s_tainted);
5263 TRACEME(("input source is %s", is_tainted ? "tainted" : "trusted"));
5264 init_retrieve_context(cxt, optype, is_tainted);
5266 ASSERT(is_retrieving(), ("within retrieve operation"));
5268 sv = retrieve(cxt, 0); /* Recursively retrieve object, get root SV */
5277 pre_06_fmt = cxt->hseen != NULL; /* Before we clean context */
5280 * The "root" context is never freed.
5283 clean_retrieve_context(cxt);
5284 if (cxt->prev) /* This context was stacked */
5285 free_context(cxt); /* It was not the "root" context */
5288 * Prepare returned value.
5292 TRACEME(("retrieve ERROR"));
5293 return &PL_sv_undef; /* Something went wrong, return undef */
5296 TRACEME(("retrieve got %s(0x%"UVxf")",
5297 sv_reftype(sv, FALSE), PTR2UV(sv)));
5300 * Backward compatibility with Storable-0.5@9 (which we know we
5301 * are retrieving if hseen is non-null): don't create an extra RV
5302 * for objects since we special-cased it at store time.
5304 * Build a reference to the SV returned by pretrieve even if it is
5305 * already one and not a scalar, for consistency reasons.
5308 if (pre_06_fmt) { /* Was not handling overloading by then */
5310 TRACEME(("fixing for old formats -- pre 0.6"));
5311 if (sv_type(sv) == svis_REF && (rv = SvRV(sv)) && SvOBJECT(rv)) {
5312 TRACEME(("ended do_retrieve() with an object -- pre 0.6"));
5318 * If reference is overloaded, restore behaviour.
5320 * NB: minor glitch here: normally, overloaded refs are stored specially
5321 * so that we can croak when behaviour cannot be re-installed, and also
5322 * avoid testing for overloading magic at each reference retrieval.
5324 * Unfortunately, the root reference is implicitely stored, so we must
5325 * check for possible overloading now. Furthermore, if we don't restore
5326 * overloading, we cannot croak as if the original ref was, because we
5327 * have no way to determine whether it was an overloaded ref or not in
5330 * It's a pity that overloading magic is attached to the rv, and not to
5331 * the underlying sv as blessing is.
5335 HV *stash = (HV *) SvSTASH(sv);
5336 SV *rv = newRV_noinc(sv);
5337 if (stash && Gv_AMG(stash)) {
5339 TRACEME(("restored overloading on root reference"));
5341 TRACEME(("ended do_retrieve() with an object"));
5345 TRACEME(("regular do_retrieve() end"));
5347 return newRV_noinc(sv);
5353 * Retrieve data held in file and return the root object, undef on error.
5355 SV *pretrieve(PerlIO *f)
5357 TRACEME(("pretrieve"));
5358 return do_retrieve(f, Nullsv, 0);
5364 * Retrieve data held in scalar and return the root object, undef on error.
5366 SV *mretrieve(SV *sv)
5368 TRACEME(("mretrieve"));
5369 return do_retrieve((PerlIO*) 0, sv, 0);
5379 * Deep clone: returns a fresh copy of the original referenced SV tree.
5381 * This is achieved by storing the object in memory and restoring from
5382 * there. Not that efficient, but it should be faster than doing it from
5389 stcxt_t *real_context;
5392 TRACEME(("dclone"));
5395 * Workaround for CROAK leak: if they enter with a "dirty" context,
5396 * free up memory for them now.
5403 * do_store() optimizes for dclone by not freeing its context, should
5404 * we need to allocate one because we're deep cloning from a hook.
5407 if (!do_store((PerlIO*) 0, sv, ST_CLONE, FALSE, (SV**) 0))
5408 return &PL_sv_undef; /* Error during store */
5411 * Because of the above optimization, we have to refresh the context,
5412 * since a new one could have been allocated and stacked by do_store().
5415 { dSTCXT; real_context = cxt; } /* Sub-block needed for macro */
5416 cxt = real_context; /* And we need this temporary... */
5419 * Now, `cxt' may refer to a new context.
5422 ASSERT(!cxt->s_dirty, ("clean context"));
5423 ASSERT(!cxt->entry, ("entry will not cause new context allocation"));
5426 TRACEME(("dclone stored %d bytes", size));
5430 * Since we're passing do_retrieve() both a NULL file and sv, we need
5431 * to pre-compute the taintedness of the input by setting cxt->tainted
5432 * to whatever state our own input string was. -- RAM, 15/09/2000
5434 * do_retrieve() will free non-root context.
5437 cxt->s_tainted = SvTAINTED(sv);
5438 out = do_retrieve((PerlIO*) 0, Nullsv, ST_CLONE);
5440 TRACEME(("dclone returns 0x%"UVxf, PTR2UV(out)));
5450 * The Perl IO GV object distinguishes between input and output for sockets
5451 * but not for plain files. To allow Storable to transparently work on
5452 * plain files and sockets transparently, we have to ask xsubpp to fetch the
5453 * right object for us. Hence the OutputStream and InputStream declarations.
5455 * Before perl 5.004_05, those entries in the standard typemap are not
5456 * defined in perl include files, so we do that here.
5459 #ifndef OutputStream
5460 #define OutputStream PerlIO *
5461 #define InputStream PerlIO *
5462 #endif /* !OutputStream */
5464 MODULE = Storable PACKAGE = Storable::Cxt
5470 stcxt_t *cxt = (stcxt_t *)SvPVX(SvRV(self));
5474 if (!cxt->membuf_ro && mbase)
5476 if (cxt->membuf_ro && (cxt->msaved).arena)
5477 Safefree((cxt->msaved).arena);
5480 MODULE = Storable PACKAGE = Storable
5518 last_op_in_netorder()