2 * Store and retrieve mechanism.
4 * Copyright (c) 1995-2000, Raphael Manfredi
6 * You may redistribute only under the same terms as Perl 5, as specified
7 * in the README file that comes with the distribution.
16 # include <patchlevel.h> /* Perl's one, needed since 5.6 */
17 # if !(defined(PERL_VERSION) || (SUBVERSION > 0 && defined(PATCHLEVEL)))
18 # include <could_not_find_Perl_patchlevel.h>
24 #define DEBUGME /* Debug mode, turns assertions on as well */
25 #define DASSERT /* Assertion mode */
28 #if 0 /* On NetWare USE_PERLIO is not used */
29 #define DEBUGME /* Debug mode, turns assertions on as well */
30 #define DASSERT /* Assertion mode */
35 * Pre PerlIO time when none of USE_PERLIO and PERLIO_IS_STDIO is defined
36 * Provide them with the necessary defines so they can build with pre-5.004.
39 #ifndef PERLIO_IS_STDIO
41 #define PerlIO_getc(x) getc(x)
42 #define PerlIO_putc(f,x) putc(x,f)
43 #define PerlIO_read(x,y,z) fread(y,1,z,x)
44 #define PerlIO_write(x,y,z) fwrite(y,1,z,x)
45 #define PerlIO_stdoutf printf
46 #endif /* PERLIO_IS_STDIO */
47 #endif /* USE_PERLIO */
50 * Earlier versions of perl might be used, we can't assume they have the latest!
53 #ifndef PERL_VERSION /* For perls < 5.6 */
54 #define PERL_VERSION PATCHLEVEL
56 #define newRV_noinc(sv) ((Sv = newRV(sv)), --SvREFCNT(SvRV(Sv)), Sv)
58 #if (PATCHLEVEL <= 4) /* Older perls (<= 5.004) lack PL_ namespace */
59 #define PL_sv_yes sv_yes
60 #define PL_sv_no sv_no
61 #define PL_sv_undef sv_undef
62 #if (SUBVERSION <= 4) /* 5.004_04 has been reported to lack newSVpvn */
63 #define newSVpvn newSVpv
65 #endif /* PATCHLEVEL <= 4 */
66 #ifndef HvSHAREKEYS_off
67 #define HvSHAREKEYS_off(hv) /* Ignore */
69 #ifndef AvFILLp /* Older perls (<=5.003) lack AvFILLp */
70 #define AvFILLp AvFILL
72 typedef double NV; /* Older perls lack the NV type */
73 #define IVdf "ld" /* Various printf formats for Perl types */
77 #define INT2PTR(t,v) (t)(IV)(v)
78 #define PTR2UV(v) (unsigned long)(v)
79 #endif /* PERL_VERSION -- perls < 5.6 */
81 #ifndef NVef /* The following were not part of perl 5.6 */
82 #if defined(USE_LONG_DOUBLE) && \
83 defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
84 #define NVef PERL_PRIeldbl
85 #define NVff PERL_PRIfldbl
86 #define NVgf PERL_PRIgldbl
101 * TRACEME() will only output things when the $Storable::DEBUGME is true.
106 if (SvTRUE(perl_get_sv("Storable::DEBUGME", TRUE))) \
107 { PerlIO_stdoutf x; PerlIO_stdoutf("\n"); } \
114 #define ASSERT(x,y) \
117 PerlIO_stdoutf("ASSERT FAILED (\"%s\", line %d): ", \
118 __FILE__, __LINE__); \
119 PerlIO_stdoutf y; PerlIO_stdoutf("\n"); \
130 #define C(x) ((char) (x)) /* For markers with dynamic retrieval handling */
132 #define SX_OBJECT C(0) /* Already stored object */
133 #define SX_LSCALAR C(1) /* Scalar (large binary) follows (length, data) */
134 #define SX_ARRAY C(2) /* Array forthcominng (size, item list) */
135 #define SX_HASH C(3) /* Hash forthcoming (size, key/value pair list) */
136 #define SX_REF C(4) /* Reference to object forthcoming */
137 #define SX_UNDEF C(5) /* Undefined scalar */
138 #define SX_INTEGER C(6) /* Integer forthcoming */
139 #define SX_DOUBLE C(7) /* Double forthcoming */
140 #define SX_BYTE C(8) /* (signed) byte forthcoming */
141 #define SX_NETINT C(9) /* Integer in network order forthcoming */
142 #define SX_SCALAR C(10) /* Scalar (binary, small) follows (length, data) */
143 #define SX_TIED_ARRAY C(11) /* Tied array forthcoming */
144 #define SX_TIED_HASH C(12) /* Tied hash forthcoming */
145 #define SX_TIED_SCALAR C(13) /* Tied scalar forthcoming */
146 #define SX_SV_UNDEF C(14) /* Perl's immortal PL_sv_undef */
147 #define SX_SV_YES C(15) /* Perl's immortal PL_sv_yes */
148 #define SX_SV_NO C(16) /* Perl's immortal PL_sv_no */
149 #define SX_BLESS C(17) /* Object is blessed */
150 #define SX_IX_BLESS C(18) /* Object is blessed, classname given by index */
151 #define SX_HOOK C(19) /* Stored via hook, user-defined */
152 #define SX_OVERLOAD C(20) /* Overloaded reference */
153 #define SX_TIED_KEY C(21) /* Tied magic key forthcoming */
154 #define SX_TIED_IDX C(22) /* Tied magic index forthcoming */
155 #define SX_UTF8STR C(23) /* UTF-8 string forthcoming (small) */
156 #define SX_LUTF8STR C(24) /* UTF-8 string forthcoming (large) */
157 #define SX_FLAG_HASH C(25) /* Hash with flags forthcoming (size, flags, key/flags/value triplet list) */
158 #define SX_CODE C(26) /* Code references as perl source code */
159 #define SX_ERROR C(27) /* Error */
162 * Those are only used to retrieve "old" pre-0.6 binary images.
164 #define SX_ITEM 'i' /* An array item introducer */
165 #define SX_IT_UNDEF 'I' /* Undefined array item */
166 #define SX_KEY 'k' /* A hash key introducer */
167 #define SX_VALUE 'v' /* A hash value introducer */
168 #define SX_VL_UNDEF 'V' /* Undefined hash value */
171 * Those are only used to retrieve "old" pre-0.7 binary images
174 #define SX_CLASS 'b' /* Object is blessed, class name length <255 */
175 #define SX_LG_CLASS 'B' /* Object is blessed, class name length >255 */
176 #define SX_STORED 'X' /* End of object */
179 * Limits between short/long length representation.
182 #define LG_SCALAR 255 /* Large scalar length limit */
183 #define LG_BLESS 127 /* Large classname bless limit */
189 #define ST_STORE 0x1 /* Store operation */
190 #define ST_RETRIEVE 0x2 /* Retrieval operation */
191 #define ST_CLONE 0x4 /* Deep cloning operation */
194 * The following structure is used for hash table key retrieval. Since, when
195 * retrieving objects, we'll be facing blessed hash references, it's best
196 * to pre-allocate that buffer once and resize it as the need arises, never
197 * freeing it (keys will be saved away someplace else anyway, so even large
198 * keys are not enough a motivation to reclaim that space).
200 * This structure is also used for memory store/retrieve operations which
201 * happen in a fixed place before being malloc'ed elsewhere if persistency
202 * is required. Hence the aptr pointer.
205 char *arena; /* Will hold hash key strings, resized as needed */
206 STRLEN asiz; /* Size of aforementionned buffer */
207 char *aptr; /* Arena pointer, for in-place read/write ops */
208 char *aend; /* First invalid address */
213 * A hash table records the objects which have already been stored.
214 * Those are referred to as SX_OBJECT in the file, and their "tag" (i.e.
215 * an arbitrary sequence number) is used to identify them.
218 * An array table records the objects which have already been retrieved,
219 * as seen by the tag determind by counting the objects themselves. The
220 * reference to that retrieved object is kept in the table, and is returned
221 * when an SX_OBJECT is found bearing that same tag.
223 * The same processing is used to record "classname" for blessed objects:
224 * indexing by a hash at store time, and via an array at retrieve time.
227 typedef unsigned long stag_t; /* Used by pre-0.6 binary format */
230 * The following "thread-safe" related defines were contributed by
231 * Murray Nesbitt <murray@activestate.com> and integrated by RAM, who
232 * only renamed things a little bit to ensure consistency with surrounding
233 * code. -- RAM, 14/09/1999
235 * The original patch suffered from the fact that the stcxt_t structure
236 * was global. Murray tried to minimize the impact on the code as much as
239 * Starting with 0.7, Storable can be re-entrant, via the STORABLE_xxx hooks
240 * on objects. Therefore, the notion of context needs to be generalized,
244 #define MY_VERSION "Storable(" XS_VERSION ")"
248 * Conditional UTF8 support.
252 #define STORE_UTF8STR(pv, len) STORE_PV_LEN(pv, len, SX_UTF8STR, SX_LUTF8STR)
253 #define HAS_UTF8_SCALARS
255 #define HAS_UTF8_HASHES
258 /* 5.6 perl has utf8 scalars but not hashes */
262 #define STORE_UTF8STR(pv, len) CROAK(("panic: storing UTF8 in non-UTF8 perl"))
265 #define UTF8_CROAK() CROAK(("Cannot retrieve UTF8 data in non-UTF8 perl"))
268 #ifdef HvPLACEHOLDERS
269 #define HAS_RESTRICTED_HASHES
271 #define HVhek_PLACEHOLD 0x200
272 #define RESTRICTED_HASH_CROAK() CROAK(("Cannot retrieve restricted hash"))
276 #define HAS_HASH_KEY_FLAGS
280 * Fields s_tainted and s_dirty are prefixed with s_ because Perl's include
281 * files remap tainted and dirty when threading is enabled. That's bad for
282 * perl to remap such common words. -- RAM, 29/09/00
285 typedef struct stcxt {
286 int entry; /* flags recursion */
287 int optype; /* type of traversal operation */
288 HV *hseen; /* which objects have been seen, store time */
289 AV *hook_seen; /* which SVs were returned by STORABLE_freeze() */
290 AV *aseen; /* which objects have been seen, retrieve time */
291 HV *hclass; /* which classnames have been seen, store time */
292 AV *aclass; /* which classnames have been seen, retrieve time */
293 HV *hook; /* cache for hook methods per class name */
294 IV tagnum; /* incremented at store time for each seen object */
295 IV classnum; /* incremented at store time for each seen classname */
296 int netorder; /* true if network order used */
297 int s_tainted; /* true if input source is tainted, at retrieve time */
298 int forgive_me; /* whether to be forgiving... */
299 int deparse; /* whether to deparse code refs */
300 SV *eval; /* whether to eval source code */
301 int canonical; /* whether to store hashes sorted by key */
302 #ifndef HAS_RESTRICTED_HASHES
303 int derestrict; /* whether to downgrade restrcted hashes */
306 int use_bytes; /* whether to bytes-ify utf8 */
308 int accept_future_minor; /* croak immediately on future minor versions? */
309 int s_dirty; /* context is dirty due to CROAK() -- can be cleaned */
310 int membuf_ro; /* true means membuf is read-only and msaved is rw */
311 struct extendable keybuf; /* for hash key retrieval */
312 struct extendable membuf; /* for memory store/retrieve operations */
313 struct extendable msaved; /* where potentially valid mbuf is saved */
314 PerlIO *fio; /* where I/O are performed, NULL for memory */
315 int ver_major; /* major of version for retrieved object */
316 int ver_minor; /* minor of version for retrieved object */
317 SV *(**retrieve_vtbl)(); /* retrieve dispatch table */
318 SV *prev; /* contexts chained backwards in real recursion */
319 SV *my_sv; /* the blessed scalar who's SvPVX() I am */
322 #define NEW_STORABLE_CXT_OBJ(cxt) \
324 SV *self = newSV(sizeof(stcxt_t) - 1); \
325 SV *my_sv = newRV_noinc(self); \
326 sv_bless(my_sv, gv_stashpv("Storable::Cxt", TRUE)); \
327 cxt = (stcxt_t *)SvPVX(self); \
328 Zero(cxt, 1, stcxt_t); \
329 cxt->my_sv = my_sv; \
332 #if defined(MULTIPLICITY) || defined(PERL_OBJECT) || defined(PERL_CAPI)
334 #if (PATCHLEVEL <= 4) && (SUBVERSION < 68)
336 SV *perinterp_sv = perl_get_sv(MY_VERSION, FALSE)
337 #else /* >= perl5.004_68 */
339 SV *perinterp_sv = *hv_fetch(PL_modglobal, \
340 MY_VERSION, sizeof(MY_VERSION)-1, TRUE)
341 #endif /* < perl5.004_68 */
343 #define dSTCXT_PTR(T,name) \
344 T name = ((perinterp_sv && SvIOK(perinterp_sv) && SvIVX(perinterp_sv) \
345 ? (T)SvPVX(SvRV(INT2PTR(SV*,SvIVX(perinterp_sv)))) : (T) 0))
348 dSTCXT_PTR(stcxt_t *, cxt)
352 NEW_STORABLE_CXT_OBJ(cxt); \
353 sv_setiv(perinterp_sv, PTR2IV(cxt->my_sv))
355 #define SET_STCXT(x) \
358 sv_setiv(perinterp_sv, PTR2IV(x->my_sv)); \
361 #else /* !MULTIPLICITY && !PERL_OBJECT && !PERL_CAPI */
363 static stcxt_t *Context_ptr = NULL;
364 #define dSTCXT stcxt_t *cxt = Context_ptr
365 #define SET_STCXT(x) Context_ptr = x
368 NEW_STORABLE_CXT_OBJ(cxt); \
372 #endif /* MULTIPLICITY || PERL_OBJECT || PERL_CAPI */
376 * Croaking implies a memory leak, since we don't use setjmp/longjmp
377 * to catch the exit and free memory used during store or retrieve
378 * operations. This is not too difficult to fix, but I need to understand
379 * how Perl does it, and croaking is exceptional anyway, so I lack the
380 * motivation to do it.
382 * The current workaround is to mark the context as dirty when croaking,
383 * so that data structures can be freed whenever we renter Storable code
384 * (but only *then*: it's a workaround, not a fix).
386 * This is also imperfect, because we don't really know how far they trapped
387 * the croak(), and when we were recursing, we won't be able to clean anything
388 * but the topmost context stacked.
391 #define CROAK(x) STMT_START { cxt->s_dirty = 1; croak x; } STMT_END
394 * End of "thread-safe" related definitions.
400 * Keep only the low 32 bits of a pointer (used for tags, which are not
405 #define LOW_32BITS(x) ((I32) (x))
407 #define LOW_32BITS(x) ((I32) ((unsigned long) (x) & 0xffffffffUL))
413 * Hack for Crays, where sizeof(I32) == 8, and which are big-endians.
414 * Used in the WLEN and RLEN macros.
418 #define oI(x) ((I32 *) ((char *) (x) + 4))
419 #define oS(x) ((x) - 4)
420 #define oC(x) (x = 0)
429 * key buffer handling
431 #define kbuf (cxt->keybuf).arena
432 #define ksiz (cxt->keybuf).asiz
436 TRACEME(("** allocating kbuf of 128 bytes")); \
437 New(10003, kbuf, 128, char); \
444 TRACEME(("** extending kbuf to %d bytes (had %d)", x+1, ksiz)); \
445 Renew(kbuf, x+1, char); \
451 * memory buffer handling
453 #define mbase (cxt->membuf).arena
454 #define msiz (cxt->membuf).asiz
455 #define mptr (cxt->membuf).aptr
456 #define mend (cxt->membuf).aend
458 #define MGROW (1 << 13)
459 #define MMASK (MGROW - 1)
461 #define round_mgrow(x) \
462 ((unsigned long) (((unsigned long) (x) + MMASK) & ~MMASK))
463 #define trunc_int(x) \
464 ((unsigned long) ((unsigned long) (x) & ~(sizeof(int)-1)))
465 #define int_aligned(x) \
466 ((unsigned long) (x) == trunc_int(x))
468 #define MBUF_INIT(x) \
471 TRACEME(("** allocating mbase of %d bytes", MGROW)); \
472 New(10003, mbase, MGROW, char); \
479 mend = mbase + msiz; \
482 #define MBUF_TRUNC(x) mptr = mbase + x
483 #define MBUF_SIZE() (mptr - mbase)
489 * Those macros are used in do_retrieve() to save the current memory
490 * buffer into cxt->msaved, before MBUF_LOAD() can be used to retrieve
491 * data from a string.
493 #define MBUF_SAVE_AND_LOAD(in) \
495 ASSERT(!cxt->membuf_ro, ("mbase not already saved")); \
496 cxt->membuf_ro = 1; \
497 TRACEME(("saving mbuf")); \
498 StructCopy(&cxt->membuf, &cxt->msaved, struct extendable); \
502 #define MBUF_RESTORE() \
504 ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
505 cxt->membuf_ro = 0; \
506 TRACEME(("restoring mbuf")); \
507 StructCopy(&cxt->msaved, &cxt->membuf, struct extendable); \
511 * Use SvPOKp(), because SvPOK() fails on tainted scalars.
512 * See store_scalar() for other usage of this workaround.
514 #define MBUF_LOAD(v) \
516 ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
518 CROAK(("Not a scalar string")); \
519 mptr = mbase = SvPV(v, msiz); \
520 mend = mbase + msiz; \
523 #define MBUF_XTEND(x) \
525 int nsz = (int) round_mgrow((x)+msiz); \
526 int offset = mptr - mbase; \
527 ASSERT(!cxt->membuf_ro, ("mbase is not read-only")); \
528 TRACEME(("** extending mbase from %d to %d bytes (wants %d new)", \
530 Renew(mbase, nsz, char); \
532 mptr = mbase + offset; \
533 mend = mbase + nsz; \
536 #define MBUF_CHK(x) \
538 if ((mptr + (x)) > mend) \
542 #define MBUF_GETC(x) \
545 x = (int) (unsigned char) *mptr++; \
551 #define MBUF_GETINT(x) \
554 if ((mptr + 4) <= mend) { \
555 memcpy(oI(&x), mptr, 4); \
561 #define MBUF_GETINT(x) \
563 if ((mptr + sizeof(int)) <= mend) { \
564 if (int_aligned(mptr)) \
567 memcpy(&x, mptr, sizeof(int)); \
568 mptr += sizeof(int); \
574 #define MBUF_READ(x,s) \
576 if ((mptr + (s)) <= mend) { \
577 memcpy(x, mptr, s); \
583 #define MBUF_SAFEREAD(x,s,z) \
585 if ((mptr + (s)) <= mend) { \
586 memcpy(x, mptr, s); \
594 #define MBUF_PUTC(c) \
597 *mptr++ = (char) c; \
600 *mptr++ = (char) c; \
605 #define MBUF_PUTINT(i) \
608 memcpy(mptr, oI(&i), 4); \
612 #define MBUF_PUTINT(i) \
614 MBUF_CHK(sizeof(int)); \
615 if (int_aligned(mptr)) \
618 memcpy(mptr, &i, sizeof(int)); \
619 mptr += sizeof(int); \
623 #define MBUF_WRITE(x,s) \
626 memcpy(mptr, x, s); \
631 * Possible return values for sv_type().
635 #define svis_SCALAR 1
639 #define svis_TIED_ITEM 5
647 #define SHF_TYPE_MASK 0x03
648 #define SHF_LARGE_CLASSLEN 0x04
649 #define SHF_LARGE_STRLEN 0x08
650 #define SHF_LARGE_LISTLEN 0x10
651 #define SHF_IDX_CLASSNAME 0x20
652 #define SHF_NEED_RECURSE 0x40
653 #define SHF_HAS_LIST 0x80
656 * Types for SX_HOOK (last 2 bits in flags).
662 #define SHT_EXTRA 3 /* Read extra byte for type */
665 * The following are held in the "extra byte"...
668 #define SHT_TSCALAR 4 /* 4 + 0 -- tied scalar */
669 #define SHT_TARRAY 5 /* 4 + 1 -- tied array */
670 #define SHT_THASH 6 /* 4 + 2 -- tied hash */
673 * per hash flags for flagged hashes
676 #define SHV_RESTRICTED 0x01
679 * per key flags for flagged hashes
682 #define SHV_K_UTF8 0x01
683 #define SHV_K_WASUTF8 0x02
684 #define SHV_K_LOCKED 0x04
685 #define SHV_K_ISSV 0x08
686 #define SHV_K_PLACEHOLDER 0x10
689 * Before 0.6, the magic string was "perl-store" (binary version number 0).
691 * Since 0.6 introduced many binary incompatibilities, the magic string has
692 * been changed to "pst0" to allow an old image to be properly retrieved by
693 * a newer Storable, but ensure a newer image cannot be retrieved with an
696 * At 0.7, objects are given the ability to serialize themselves, and the
697 * set of markers is extended, backward compatibility is not jeopardized,
698 * so the binary version number could have remained unchanged. To correctly
699 * spot errors if a file making use of 0.7-specific extensions is given to
700 * 0.6 for retrieval, the binary version was moved to "2". And I'm introducing
701 * a "minor" version, to better track this kind of evolution from now on.
704 static const char old_magicstr[] = "perl-store"; /* Magic number before 0.6 */
705 static const char magicstr[] = "pst0"; /* Used as a magic number */
707 #define MAGICSTR_BYTES 'p','s','t','0'
708 #define OLDMAGICSTR_BYTES 'p','e','r','l','-','s','t','o','r','e'
710 /* 5.6.x introduced the ability to have IVs as long long.
711 However, Configure still defined BYTEORDER based on the size of a long.
712 Storable uses the BYTEORDER value as part of the header, but doesn't
713 explicity store sizeof(IV) anywhere in the header. Hence on 5.6.x built
714 with IV as long long on a platform that uses Configure (ie most things
715 except VMS and Windows) headers are identical for the different IV sizes,
716 despite the files containing some fields based on sizeof(IV)
718 5.8 is consistent - the following redifinition kludge is only needed on
719 5.6.x, but the interwork is needed on 5.8 while data survives in files
724 #if defined (IVSIZE) && (IVSIZE == 8) && (LONGSIZE == 4)
725 #ifndef NO_56_INTERWORK_KLUDGE
726 #define USE_56_INTERWORK_KLUDGE
728 #if BYTEORDER == 0x1234
730 #define BYTEORDER 0x12345678
732 #if BYTEORDER == 0x4321
734 #define BYTEORDER 0x87654321
739 #if BYTEORDER == 0x1234
740 #define BYTEORDER_BYTES '1','2','3','4'
742 #if BYTEORDER == 0x12345678
743 #define BYTEORDER_BYTES '1','2','3','4','5','6','7','8'
744 #ifdef USE_56_INTERWORK_KLUDGE
745 #define BYTEORDER_BYTES_56 '1','2','3','4'
748 #if BYTEORDER == 0x87654321
749 #define BYTEORDER_BYTES '8','7','6','5','4','3','2','1'
750 #ifdef USE_56_INTERWORK_KLUDGE
751 #define BYTEORDER_BYTES_56 '4','3','2','1'
754 #if BYTEORDER == 0x4321
755 #define BYTEORDER_BYTES '4','3','2','1'
757 #error Unknown byteoder. Please append your byteorder to Storable.xs
763 static const char byteorderstr[] = {BYTEORDER_BYTES, 0};
764 #ifdef USE_56_INTERWORK_KLUDGE
765 static const char byteorderstr_56[] = {BYTEORDER_BYTES_56, 0};
768 #define STORABLE_BIN_MAJOR 2 /* Binary major "version" */
769 #define STORABLE_BIN_MINOR 6 /* Binary minor "version" */
771 /* If we aren't 5.7.3 or later, we won't be writing out files that use the
772 * new flagged hash introdued in 2.5, so put 2.4 in the binary header to
773 * maximise ease of interoperation with older Storables.
774 * Could we write 2.3s if we're on 5.005_03? NWC
776 #if (PATCHLEVEL <= 6)
777 #define STORABLE_BIN_WRITE_MINOR 4
780 * As of perl 5.7.3, utf8 hash key is introduced.
781 * So this must change -- dankogai
783 #define STORABLE_BIN_WRITE_MINOR 6
784 #endif /* (PATCHLEVEL <= 6) */
787 * Useful store shortcuts...
794 else if (PerlIO_putc(cxt->fio, x) == EOF) \
798 #define WRITE_I32(x) \
800 ASSERT(sizeof(x) == sizeof(I32), ("writing an I32")); \
803 else if (PerlIO_write(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
810 if (cxt->netorder) { \
811 int y = (int) htonl(x); \
814 else if (PerlIO_write(cxt->fio,oI(&y),oS(sizeof(y))) != oS(sizeof(y))) \
819 else if (PerlIO_write(cxt->fio,oI(&x),oS(sizeof(x))) != oS(sizeof(x))) \
824 #define WLEN(x) WRITE_I32(x)
831 else if (PerlIO_write(cxt->fio, x, y) != y) \
835 #define STORE_PV_LEN(pv, len, small, large) \
837 if (len <= LG_SCALAR) { \
838 unsigned char clen = (unsigned char) len; \
850 #define STORE_SCALAR(pv, len) STORE_PV_LEN(pv, len, SX_SCALAR, SX_LSCALAR)
853 * Store undef in arrays and hashes without recursing through store().
855 #define STORE_UNDEF() \
862 * Useful retrieve shortcuts...
866 (cxt->fio ? PerlIO_getc(cxt->fio) : (mptr >= mend ? EOF : (int) *mptr++))
872 else if ((int) (x = PerlIO_getc(cxt->fio)) == EOF) \
876 #define READ_I32(x) \
878 ASSERT(sizeof(x) == sizeof(I32), ("reading an I32")); \
882 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
892 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
895 x = (int) ntohl(x); \
898 #define RLEN(x) READ_I32(x)
905 else if (PerlIO_read(cxt->fio, x, y) != y) \
909 #define SAFEREAD(x,y,z) \
912 MBUF_SAFEREAD(x,y,z); \
913 else if (PerlIO_read(cxt->fio, x, y) != y) { \
920 * This macro is used at retrieve time, to remember where object 'y', bearing a
921 * given tag 'tagnum', has been retrieved. Next time we see an SX_OBJECT marker,
922 * we'll therefore know where it has been retrieved and will be able to
923 * share the same reference, as in the original stored memory image.
925 * We also need to bless objects ASAP for hooks (which may compute "ref $x"
926 * on the objects given to STORABLE_thaw and expect that to be defined), and
927 * also for overloaded objects (for which we might not find the stash if the
928 * object is not blessed yet--this might occur for overloaded objects that
929 * refer to themselves indirectly: if we blessed upon return from a sub
930 * retrieve(), the SX_OBJECT marker we'd found could not have overloading
931 * restored on it because the underlying object would not be blessed yet!).
933 * To achieve that, the class name of the last retrieved object is passed down
934 * recursively, and the first SEEN() call for which the class name is not NULL
935 * will bless the object.
941 if (av_store(cxt->aseen, cxt->tagnum++, SvREFCNT_inc(y)) == 0) \
943 TRACEME(("aseen(#%d) = 0x%"UVxf" (refcnt=%d)", cxt->tagnum-1, \
944 PTR2UV(y), SvREFCNT(y)-1)); \
946 BLESS((SV *) (y), c); \
950 * Bless `s' in `p', via a temporary reference, required by sv_bless().
956 TRACEME(("blessing 0x%"UVxf" in %s", PTR2UV(s), (p))); \
957 stash = gv_stashpv((p), TRUE); \
958 ref = newRV_noinc(s); \
959 (void) sv_bless(ref, stash); \
965 static SV *retrieve(stcxt_t *cxt, char *cname);
968 * Dynamic dispatching table for SV store.
971 static int store_ref(stcxt_t *cxt, SV *sv);
972 static int store_scalar(stcxt_t *cxt, SV *sv);
973 static int store_array(stcxt_t *cxt, AV *av);
974 static int store_hash(stcxt_t *cxt, HV *hv);
975 static int store_tied(stcxt_t *cxt, SV *sv);
976 static int store_tied_item(stcxt_t *cxt, SV *sv);
977 static int store_code(stcxt_t *cxt, CV *cv);
978 static int store_other(stcxt_t *cxt, SV *sv);
979 static int store_blessed(stcxt_t *cxt, SV *sv, int type, HV *pkg);
981 static int (*sv_store[])(stcxt_t *cxt, SV *sv) = {
982 store_ref, /* svis_REF */
983 store_scalar, /* svis_SCALAR */
984 (int (*)(stcxt_t *cxt, SV *sv)) store_array, /* svis_ARRAY */
985 (int (*)(stcxt_t *cxt, SV *sv)) store_hash, /* svis_HASH */
986 store_tied, /* svis_TIED */
987 store_tied_item, /* svis_TIED_ITEM */
988 (int (*)(stcxt_t *cxt, SV *sv)) store_code, /* svis_CODE */
989 store_other, /* svis_OTHER */
992 #define SV_STORE(x) (*sv_store[x])
995 * Dynamic dispatching tables for SV retrieval.
998 static SV *retrieve_lscalar(stcxt_t *cxt, char *cname);
999 static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname);
1000 static SV *old_retrieve_array(stcxt_t *cxt, char *cname);
1001 static SV *old_retrieve_hash(stcxt_t *cxt, char *cname);
1002 static SV *retrieve_ref(stcxt_t *cxt, char *cname);
1003 static SV *retrieve_undef(stcxt_t *cxt, char *cname);
1004 static SV *retrieve_integer(stcxt_t *cxt, char *cname);
1005 static SV *retrieve_double(stcxt_t *cxt, char *cname);
1006 static SV *retrieve_byte(stcxt_t *cxt, char *cname);
1007 static SV *retrieve_netint(stcxt_t *cxt, char *cname);
1008 static SV *retrieve_scalar(stcxt_t *cxt, char *cname);
1009 static SV *retrieve_utf8str(stcxt_t *cxt, char *cname);
1010 static SV *retrieve_tied_array(stcxt_t *cxt, char *cname);
1011 static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname);
1012 static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname);
1013 static SV *retrieve_other(stcxt_t *cxt, char *cname);
1015 static SV *(*sv_old_retrieve[])(stcxt_t *cxt, char *cname) = {
1016 0, /* SX_OBJECT -- entry unused dynamically */
1017 retrieve_lscalar, /* SX_LSCALAR */
1018 old_retrieve_array, /* SX_ARRAY -- for pre-0.6 binaries */
1019 old_retrieve_hash, /* SX_HASH -- for pre-0.6 binaries */
1020 retrieve_ref, /* SX_REF */
1021 retrieve_undef, /* SX_UNDEF */
1022 retrieve_integer, /* SX_INTEGER */
1023 retrieve_double, /* SX_DOUBLE */
1024 retrieve_byte, /* SX_BYTE */
1025 retrieve_netint, /* SX_NETINT */
1026 retrieve_scalar, /* SX_SCALAR */
1027 retrieve_tied_array, /* SX_ARRAY */
1028 retrieve_tied_hash, /* SX_HASH */
1029 retrieve_tied_scalar, /* SX_SCALAR */
1030 retrieve_other, /* SX_SV_UNDEF not supported */
1031 retrieve_other, /* SX_SV_YES not supported */
1032 retrieve_other, /* SX_SV_NO not supported */
1033 retrieve_other, /* SX_BLESS not supported */
1034 retrieve_other, /* SX_IX_BLESS not supported */
1035 retrieve_other, /* SX_HOOK not supported */
1036 retrieve_other, /* SX_OVERLOADED not supported */
1037 retrieve_other, /* SX_TIED_KEY not supported */
1038 retrieve_other, /* SX_TIED_IDX not supported */
1039 retrieve_other, /* SX_UTF8STR not supported */
1040 retrieve_other, /* SX_LUTF8STR not supported */
1041 retrieve_other, /* SX_FLAG_HASH not supported */
1042 retrieve_other, /* SX_CODE not supported */
1043 retrieve_other, /* SX_ERROR */
1046 static SV *retrieve_array(stcxt_t *cxt, char *cname);
1047 static SV *retrieve_hash(stcxt_t *cxt, char *cname);
1048 static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname);
1049 static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname);
1050 static SV *retrieve_sv_no(stcxt_t *cxt, char *cname);
1051 static SV *retrieve_blessed(stcxt_t *cxt, char *cname);
1052 static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname);
1053 static SV *retrieve_hook(stcxt_t *cxt, char *cname);
1054 static SV *retrieve_overloaded(stcxt_t *cxt, char *cname);
1055 static SV *retrieve_tied_key(stcxt_t *cxt, char *cname);
1056 static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname);
1057 static SV *retrieve_flag_hash(stcxt_t *cxt, char *cname);
1058 static SV *retrieve_code(stcxt_t *cxt, char *cname);
1060 static SV *(*sv_retrieve[])(stcxt_t *cxt, char *cname) = {
1061 0, /* SX_OBJECT -- entry unused dynamically */
1062 retrieve_lscalar, /* SX_LSCALAR */
1063 retrieve_array, /* SX_ARRAY */
1064 retrieve_hash, /* SX_HASH */
1065 retrieve_ref, /* SX_REF */
1066 retrieve_undef, /* SX_UNDEF */
1067 retrieve_integer, /* SX_INTEGER */
1068 retrieve_double, /* SX_DOUBLE */
1069 retrieve_byte, /* SX_BYTE */
1070 retrieve_netint, /* SX_NETINT */
1071 retrieve_scalar, /* SX_SCALAR */
1072 retrieve_tied_array, /* SX_ARRAY */
1073 retrieve_tied_hash, /* SX_HASH */
1074 retrieve_tied_scalar, /* SX_SCALAR */
1075 retrieve_sv_undef, /* SX_SV_UNDEF */
1076 retrieve_sv_yes, /* SX_SV_YES */
1077 retrieve_sv_no, /* SX_SV_NO */
1078 retrieve_blessed, /* SX_BLESS */
1079 retrieve_idx_blessed, /* SX_IX_BLESS */
1080 retrieve_hook, /* SX_HOOK */
1081 retrieve_overloaded, /* SX_OVERLOAD */
1082 retrieve_tied_key, /* SX_TIED_KEY */
1083 retrieve_tied_idx, /* SX_TIED_IDX */
1084 retrieve_utf8str, /* SX_UTF8STR */
1085 retrieve_lutf8str, /* SX_LUTF8STR */
1086 retrieve_flag_hash, /* SX_HASH */
1087 retrieve_code, /* SX_CODE */
1088 retrieve_other, /* SX_ERROR */
1091 #define RETRIEVE(c,x) (*(c)->retrieve_vtbl[(x) >= SX_ERROR ? SX_ERROR : (x)])
1093 static SV *mbuf2sv(void);
1096 *** Context management.
1102 * Called once per "thread" (interpreter) to initialize some global context.
1104 static void init_perinterp(void)
1108 cxt->netorder = 0; /* true if network order used */
1109 cxt->forgive_me = -1; /* whether to be forgiving... */
1115 * Called at the end of every context cleaning, to perform common reset
1118 static void reset_context(stcxt_t *cxt)
1122 cxt->optype &= ~(ST_STORE|ST_RETRIEVE); /* Leave ST_CLONE alone */
1126 * init_store_context
1128 * Initialize a new store context for real recursion.
1130 static void init_store_context(
1136 TRACEME(("init_store_context"));
1138 cxt->netorder = network_order;
1139 cxt->forgive_me = -1; /* Fetched from perl if needed */
1140 cxt->deparse = -1; /* Idem */
1141 cxt->eval = NULL; /* Idem */
1142 cxt->canonical = -1; /* Idem */
1143 cxt->tagnum = -1; /* Reset tag numbers */
1144 cxt->classnum = -1; /* Reset class numbers */
1145 cxt->fio = f; /* Where I/O are performed */
1146 cxt->optype = optype; /* A store, or a deep clone */
1147 cxt->entry = 1; /* No recursion yet */
1150 * The `hseen' table is used to keep track of each SV stored and their
1151 * associated tag numbers is special. It is "abused" because the
1152 * values stored are not real SV, just integers cast to (SV *),
1153 * which explains the freeing below.
1155 * It is also one possible bottlneck to achieve good storing speed,
1156 * so the "shared keys" optimization is turned off (unlikely to be
1157 * of any use here), and the hash table is "pre-extended". Together,
1158 * those optimizations increase the throughput by 12%.
1161 cxt->hseen = newHV(); /* Table where seen objects are stored */
1162 HvSHAREKEYS_off(cxt->hseen);
1165 * The following does not work well with perl5.004_04, and causes
1166 * a core dump later on, in a completely unrelated spot, which
1167 * makes me think there is a memory corruption going on.
1169 * Calling hv_ksplit(hseen, HBUCKETS) instead of manually hacking
1170 * it below does not make any difference. It seems to work fine
1171 * with perl5.004_68 but given the probable nature of the bug,
1172 * that does not prove anything.
1174 * It's a shame because increasing the amount of buckets raises
1175 * store() throughput by 5%, but until I figure this out, I can't
1176 * allow for this to go into production.
1178 * It is reported fixed in 5.005, hence the #if.
1180 #if PERL_VERSION >= 5
1181 #define HBUCKETS 4096 /* Buckets for %hseen */
1182 HvMAX(cxt->hseen) = HBUCKETS - 1; /* keys %hseen = $HBUCKETS; */
1186 * The `hclass' hash uses the same settings as `hseen' above, but it is
1187 * used to assign sequential tags (numbers) to class names for blessed
1190 * We turn the shared key optimization on.
1193 cxt->hclass = newHV(); /* Where seen classnames are stored */
1195 #if PERL_VERSION >= 5
1196 HvMAX(cxt->hclass) = HBUCKETS - 1; /* keys %hclass = $HBUCKETS; */
1200 * The `hook' hash table is used to keep track of the references on
1201 * the STORABLE_freeze hook routines, when found in some class name.
1203 * It is assumed that the inheritance tree will not be changed during
1204 * storing, and that no new method will be dynamically created by the
1208 cxt->hook = newHV(); /* Table where hooks are cached */
1211 * The `hook_seen' array keeps track of all the SVs returned by
1212 * STORABLE_freeze hooks for us to serialize, so that they are not
1213 * reclaimed until the end of the serialization process. Each SV is
1214 * only stored once, the first time it is seen.
1217 cxt->hook_seen = newAV(); /* Lists SVs returned by STORABLE_freeze */
1221 * clean_store_context
1223 * Clean store context by
1225 static void clean_store_context(stcxt_t *cxt)
1229 TRACEME(("clean_store_context"));
1231 ASSERT(cxt->optype & ST_STORE, ("was performing a store()"));
1234 * Insert real values into hashes where we stored faked pointers.
1238 hv_iterinit(cxt->hseen);
1239 while ((he = hv_iternext(cxt->hseen))) /* Extra () for -Wall, grr.. */
1240 HeVAL(he) = &PL_sv_undef;
1244 hv_iterinit(cxt->hclass);
1245 while ((he = hv_iternext(cxt->hclass))) /* Extra () for -Wall, grr.. */
1246 HeVAL(he) = &PL_sv_undef;
1250 * And now dispose of them...
1252 * The surrounding if() protection has been added because there might be
1253 * some cases where this routine is called more than once, during
1254 * exceptionnal events. This was reported by Marc Lehmann when Storable
1255 * is executed from mod_perl, and the fix was suggested by him.
1256 * -- RAM, 20/12/2000
1260 HV *hseen = cxt->hseen;
1263 sv_free((SV *) hseen);
1267 HV *hclass = cxt->hclass;
1270 sv_free((SV *) hclass);
1274 HV *hook = cxt->hook;
1277 sv_free((SV *) hook);
1280 if (cxt->hook_seen) {
1281 AV *hook_seen = cxt->hook_seen;
1283 av_undef(hook_seen);
1284 sv_free((SV *) hook_seen);
1287 cxt->forgive_me = -1; /* Fetched from perl if needed */
1288 cxt->deparse = -1; /* Idem */
1290 SvREFCNT_dec(cxt->eval);
1292 cxt->eval = NULL; /* Idem */
1293 cxt->canonical = -1; /* Idem */
1299 * init_retrieve_context
1301 * Initialize a new retrieve context for real recursion.
1303 static void init_retrieve_context(stcxt_t *cxt, int optype, int is_tainted)
1305 TRACEME(("init_retrieve_context"));
1308 * The hook hash table is used to keep track of the references on
1309 * the STORABLE_thaw hook routines, when found in some class name.
1311 * It is assumed that the inheritance tree will not be changed during
1312 * storing, and that no new method will be dynamically created by the
1316 cxt->hook = newHV(); /* Caches STORABLE_thaw */
1319 * If retrieving an old binary version, the cxt->retrieve_vtbl variable
1320 * was set to sv_old_retrieve. We'll need a hash table to keep track of
1321 * the correspondance between the tags and the tag number used by the
1322 * new retrieve routines.
1325 cxt->hseen = ((cxt->retrieve_vtbl == sv_old_retrieve) ? newHV() : 0);
1327 cxt->aseen = newAV(); /* Where retrieved objects are kept */
1328 cxt->aclass = newAV(); /* Where seen classnames are kept */
1329 cxt->tagnum = 0; /* Have to count objects... */
1330 cxt->classnum = 0; /* ...and class names as well */
1331 cxt->optype = optype;
1332 cxt->s_tainted = is_tainted;
1333 cxt->entry = 1; /* No recursion yet */
1334 #ifndef HAS_RESTRICTED_HASHES
1335 cxt->derestrict = -1; /* Fetched from perl if needed */
1337 #ifndef HAS_UTF8_ALL
1338 cxt->use_bytes = -1; /* Fetched from perl if needed */
1340 cxt->accept_future_minor = -1; /* Fetched from perl if needed */
1344 * clean_retrieve_context
1346 * Clean retrieve context by
1348 static void clean_retrieve_context(stcxt_t *cxt)
1350 TRACEME(("clean_retrieve_context"));
1352 ASSERT(cxt->optype & ST_RETRIEVE, ("was performing a retrieve()"));
1355 AV *aseen = cxt->aseen;
1358 sv_free((SV *) aseen);
1362 AV *aclass = cxt->aclass;
1365 sv_free((SV *) aclass);
1369 HV *hook = cxt->hook;
1372 sv_free((SV *) hook);
1376 HV *hseen = cxt->hseen;
1379 sv_free((SV *) hseen); /* optional HV, for backward compat. */
1382 #ifndef HAS_RESTRICTED_HASHES
1383 cxt->derestrict = -1; /* Fetched from perl if needed */
1385 #ifndef HAS_UTF8_ALL
1386 cxt->use_bytes = -1; /* Fetched from perl if needed */
1388 cxt->accept_future_minor = -1; /* Fetched from perl if needed */
1396 * A workaround for the CROAK bug: cleanup the last context.
1398 static void clean_context(stcxt_t *cxt)
1400 TRACEME(("clean_context"));
1402 ASSERT(cxt->s_dirty, ("dirty context"));
1407 ASSERT(!cxt->membuf_ro, ("mbase is not read-only"));
1409 if (cxt->optype & ST_RETRIEVE)
1410 clean_retrieve_context(cxt);
1411 else if (cxt->optype & ST_STORE)
1412 clean_store_context(cxt);
1416 ASSERT(!cxt->s_dirty, ("context is clean"));
1417 ASSERT(cxt->entry == 0, ("context is reset"));
1423 * Allocate a new context and push it on top of the parent one.
1424 * This new context is made globally visible via SET_STCXT().
1426 static stcxt_t *allocate_context(parent_cxt)
1427 stcxt_t *parent_cxt;
1431 TRACEME(("allocate_context"));
1433 ASSERT(!parent_cxt->s_dirty, ("parent context clean"));
1435 NEW_STORABLE_CXT_OBJ(cxt);
1436 cxt->prev = parent_cxt->my_sv;
1439 ASSERT(!cxt->s_dirty, ("clean context"));
1447 * Free current context, which cannot be the "root" one.
1448 * Make the context underneath globally visible via SET_STCXT().
1450 static void free_context(cxt)
1453 stcxt_t *prev = (stcxt_t *)(cxt->prev ? SvPVX(SvRV(cxt->prev)) : 0);
1455 TRACEME(("free_context"));
1457 ASSERT(!cxt->s_dirty, ("clean context"));
1458 ASSERT(prev, ("not freeing root context"));
1460 SvREFCNT_dec(cxt->my_sv);
1463 ASSERT(cxt, ("context not void"));
1473 * Tells whether we're in the middle of a store operation.
1475 int is_storing(void)
1479 return cxt->entry && (cxt->optype & ST_STORE);
1485 * Tells whether we're in the middle of a retrieve operation.
1487 int is_retrieving(void)
1491 return cxt->entry && (cxt->optype & ST_RETRIEVE);
1495 * last_op_in_netorder
1497 * Returns whether last operation was made using network order.
1499 * This is typically out-of-band information that might prove useful
1500 * to people wishing to convert native to network order data when used.
1502 int last_op_in_netorder(void)
1506 return cxt->netorder;
1510 *** Hook lookup and calling routines.
1516 * A wrapper on gv_fetchmethod_autoload() which caches results.
1518 * Returns the routine reference as an SV*, or null if neither the package
1519 * nor its ancestors know about the method.
1521 static SV *pkg_fetchmeth(
1530 * The following code is the same as the one performed by UNIVERSAL::can
1534 gv = gv_fetchmethod_autoload(pkg, method, FALSE);
1535 if (gv && isGV(gv)) {
1536 sv = newRV((SV*) GvCV(gv));
1537 TRACEME(("%s->%s: 0x%"UVxf, HvNAME(pkg), method, PTR2UV(sv)));
1539 sv = newSVsv(&PL_sv_undef);
1540 TRACEME(("%s->%s: not found", HvNAME(pkg), method));
1544 * Cache the result, ignoring failure: if we can't store the value,
1545 * it just won't be cached.
1548 (void) hv_store(cache, HvNAME(pkg), strlen(HvNAME(pkg)), sv, 0);
1550 return SvOK(sv) ? sv : (SV *) 0;
1556 * Force cached value to be undef: hook ignored even if present.
1558 static void pkg_hide(
1563 (void) hv_store(cache,
1564 HvNAME(pkg), strlen(HvNAME(pkg)), newSVsv(&PL_sv_undef), 0);
1570 * Discard cached value: a whole fetch loop will be retried at next lookup.
1572 static void pkg_uncache(
1577 (void) hv_delete(cache, HvNAME(pkg), strlen(HvNAME(pkg)), G_DISCARD);
1583 * Our own "UNIVERSAL::can", which caches results.
1585 * Returns the routine reference as an SV*, or null if the object does not
1586 * know about the method.
1596 TRACEME(("pkg_can for %s->%s", HvNAME(pkg), method));
1599 * Look into the cache to see whether we already have determined
1600 * where the routine was, if any.
1602 * NOTA BENE: we don't use `method' at all in our lookup, since we know
1603 * that only one hook (i.e. always the same) is cached in a given cache.
1606 svh = hv_fetch(cache, HvNAME(pkg), strlen(HvNAME(pkg)), FALSE);
1610 TRACEME(("cached %s->%s: not found", HvNAME(pkg), method));
1613 TRACEME(("cached %s->%s: 0x%"UVxf,
1614 HvNAME(pkg), method, PTR2UV(sv)));
1619 TRACEME(("not cached yet"));
1620 return pkg_fetchmeth(cache, pkg, method); /* Fetch and cache */
1626 * Call routine as obj->hook(av) in scalar context.
1627 * Propagates the single returned value if not called in void context.
1629 static SV *scalar_call(
1640 TRACEME(("scalar_call (cloning=%d)", cloning));
1647 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1649 SV **ary = AvARRAY(av);
1650 int cnt = AvFILLp(av) + 1;
1652 XPUSHs(ary[0]); /* Frozen string */
1653 for (i = 1; i < cnt; i++) {
1654 TRACEME(("pushing arg #%d (0x%"UVxf")...",
1655 i, PTR2UV(ary[i])));
1656 XPUSHs(sv_2mortal(newRV(ary[i])));
1661 TRACEME(("calling..."));
1662 count = perl_call_sv(hook, flags); /* Go back to Perl code */
1663 TRACEME(("count = %d", count));
1669 SvREFCNT_inc(sv); /* We're returning it, must stay alive! */
1682 * Call routine obj->hook(cloning) in list context.
1683 * Returns the list of returned values in an array.
1685 static AV *array_call(
1695 TRACEME(("array_call (cloning=%d)", cloning));
1701 XPUSHs(obj); /* Target object */
1702 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1705 count = perl_call_sv(hook, G_ARRAY); /* Go back to Perl code */
1710 for (i = count - 1; i >= 0; i--) {
1712 av_store(av, i, SvREFCNT_inc(sv));
1725 * Lookup the class name in the `hclass' table and either assign it a new ID
1726 * or return the existing one, by filling in `classnum'.
1728 * Return true if the class was known, false if the ID was just generated.
1730 static int known_class(
1732 char *name, /* Class name */
1733 int len, /* Name length */
1737 HV *hclass = cxt->hclass;
1739 TRACEME(("known_class (%s)", name));
1742 * Recall that we don't store pointers in this hash table, but tags.
1743 * Therefore, we need LOW_32BITS() to extract the relevant parts.
1746 svh = hv_fetch(hclass, name, len, FALSE);
1748 *classnum = LOW_32BITS(*svh);
1753 * Unknown classname, we need to record it.
1757 if (!hv_store(hclass, name, len, INT2PTR(SV*, cxt->classnum), 0))
1758 CROAK(("Unable to record new classname"));
1760 *classnum = cxt->classnum;
1765 *** Sepcific store routines.
1771 * Store a reference.
1772 * Layout is SX_REF <object> or SX_OVERLOAD <object>.
1774 static int store_ref(stcxt_t *cxt, SV *sv)
1776 TRACEME(("store_ref (0x%"UVxf")", PTR2UV(sv)));
1779 * Follow reference, and check if target is overloaded.
1785 HV *stash = (HV *) SvSTASH(sv);
1786 if (stash && Gv_AMG(stash)) {
1787 TRACEME(("ref (0x%"UVxf") is overloaded", PTR2UV(sv)));
1788 PUTMARK(SX_OVERLOAD);
1794 return store(cxt, sv);
1802 * Layout is SX_LSCALAR <length> <data>, SX_SCALAR <length> <data> or SX_UNDEF.
1803 * The <data> section is omitted if <length> is 0.
1805 * If integer or double, the layout is SX_INTEGER <data> or SX_DOUBLE <data>.
1806 * Small integers (within [-127, +127]) are stored as SX_BYTE <byte>.
1808 static int store_scalar(stcxt_t *cxt, SV *sv)
1813 U32 flags = SvFLAGS(sv); /* "cc -O" may put it in register */
1815 TRACEME(("store_scalar (0x%"UVxf")", PTR2UV(sv)));
1818 * For efficiency, break the SV encapsulation by peaking at the flags
1819 * directly without using the Perl macros to avoid dereferencing
1820 * sv->sv_flags each time we wish to check the flags.
1823 if (!(flags & SVf_OK)) { /* !SvOK(sv) */
1824 if (sv == &PL_sv_undef) {
1825 TRACEME(("immortal undef"));
1826 PUTMARK(SX_SV_UNDEF);
1828 TRACEME(("undef at 0x%"UVxf, PTR2UV(sv)));
1835 * Always store the string representation of a scalar if it exists.
1836 * Gisle Aas provided me with this test case, better than a long speach:
1838 * perl -MDevel::Peek -le '$a="abc"; $a+0; Dump($a)'
1839 * SV = PVNV(0x80c8520)
1841 * FLAGS = (NOK,POK,pNOK,pPOK)
1844 * PV = 0x80c83d0 "abc"\0
1848 * Write SX_SCALAR, length, followed by the actual data.
1850 * Otherwise, write an SX_BYTE, SX_INTEGER or an SX_DOUBLE as
1851 * appropriate, followed by the actual (binary) data. A double
1852 * is written as a string if network order, for portability.
1854 * NOTE: instead of using SvNOK(sv), we test for SvNOKp(sv).
1855 * The reason is that when the scalar value is tainted, the SvNOK(sv)
1858 * The test for a read-only scalar with both POK and NOK set is meant
1859 * to quickly detect &PL_sv_yes and &PL_sv_no without having to pay the
1860 * address comparison for each scalar we store.
1863 #define SV_MAYBE_IMMORTAL (SVf_READONLY|SVf_POK|SVf_NOK)
1865 if ((flags & SV_MAYBE_IMMORTAL) == SV_MAYBE_IMMORTAL) {
1866 if (sv == &PL_sv_yes) {
1867 TRACEME(("immortal yes"));
1869 } else if (sv == &PL_sv_no) {
1870 TRACEME(("immortal no"));
1873 pv = SvPV(sv, len); /* We know it's SvPOK */
1874 goto string; /* Share code below */
1876 } else if (flags & SVf_POK) {
1877 /* public string - go direct to string read. */
1878 goto string_readlen;
1880 #if (PATCHLEVEL <= 6)
1881 /* For 5.6 and earlier NV flag trumps IV flag, so only use integer
1882 direct if NV flag is off. */
1883 (flags & (SVf_NOK | SVf_IOK)) == SVf_IOK
1885 /* 5.7 rules are that if IV public flag is set, IV value is as
1886 good, if not better, than NV value. */
1892 * Will come here from below with iv set if double is an integer.
1896 /* Sorry. This isn't in 5.005_56 (IIRC) or earlier. */
1898 /* Need to do this out here, else 0xFFFFFFFF becomes iv of -1
1899 * (for example) and that ends up in the optimised small integer
1902 if ((flags & SVf_IVisUV) && SvUV(sv) > IV_MAX) {
1903 TRACEME(("large unsigned integer as string, value = %"UVuf, SvUV(sv)));
1904 goto string_readlen;
1908 * Optimize small integers into a single byte, otherwise store as
1909 * a real integer (converted into network order if they asked).
1912 if (iv >= -128 && iv <= 127) {
1913 unsigned char siv = (unsigned char) (iv + 128); /* [0,255] */
1916 TRACEME(("small integer stored as %d", siv));
1917 } else if (cxt->netorder) {
1919 TRACEME(("no htonl, fall back to string for integer"));
1920 goto string_readlen;
1928 /* Sorry. This isn't in 5.005_56 (IIRC) or earlier. */
1929 ((flags & SVf_IVisUV) && SvUV(sv) > 0x7FFFFFFF) ||
1931 (iv > 0x7FFFFFFF) || (iv < -0x80000000)) {
1932 /* Bigger than 32 bits. */
1933 TRACEME(("large network order integer as string, value = %"IVdf, iv));
1934 goto string_readlen;
1938 niv = (I32) htonl((I32) iv);
1939 TRACEME(("using network order"));
1944 PUTMARK(SX_INTEGER);
1945 WRITE(&iv, sizeof(iv));
1948 TRACEME(("ok (integer 0x%"UVxf", value = %"IVdf")", PTR2UV(sv), iv));
1949 } else if (flags & SVf_NOK) {
1951 #if (PATCHLEVEL <= 6)
1954 * Watch for number being an integer in disguise.
1956 if (nv == (NV) (iv = I_V(nv))) {
1957 TRACEME(("double %"NVff" is actually integer %"IVdf, nv, iv));
1958 goto integer; /* Share code above */
1965 goto integer; /* Share code above */
1970 if (cxt->netorder) {
1971 TRACEME(("double %"NVff" stored as string", nv));
1972 goto string_readlen; /* Share code below */
1976 WRITE(&nv, sizeof(nv));
1978 TRACEME(("ok (double 0x%"UVxf", value = %"NVff")", PTR2UV(sv), nv));
1980 } else if (flags & (SVp_POK | SVp_NOK | SVp_IOK)) {
1981 I32 wlen; /* For 64-bit machines */
1987 * Will come here from above if it was readonly, POK and NOK but
1988 * neither &PL_sv_yes nor &PL_sv_no.
1992 wlen = (I32) len; /* WLEN via STORE_SCALAR expects I32 */
1994 STORE_UTF8STR(pv, wlen);
1996 STORE_SCALAR(pv, wlen);
1997 TRACEME(("ok (scalar 0x%"UVxf" '%s', length = %"IVdf")",
1998 PTR2UV(sv), SvPVX(sv), (IV)len));
2000 CROAK(("Can't determine type of %s(0x%"UVxf")",
2001 sv_reftype(sv, FALSE),
2003 return 0; /* Ok, no recursion on scalars */
2011 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
2012 * Each item is stored as <object>.
2014 static int store_array(stcxt_t *cxt, AV *av)
2017 I32 len = av_len(av) + 1;
2021 TRACEME(("store_array (0x%"UVxf")", PTR2UV(av)));
2024 * Signal array by emitting SX_ARRAY, followed by the array length.
2029 TRACEME(("size = %d", len));
2032 * Now store each item recursively.
2035 for (i = 0; i < len; i++) {
2036 sav = av_fetch(av, i, 0);
2038 TRACEME(("(#%d) undef item", i));
2042 TRACEME(("(#%d) item", i));
2043 if ((ret = store(cxt, *sav))) /* Extra () for -Wall, grr... */
2047 TRACEME(("ok (array)"));
2056 * Borrowed from perl source file pp_ctl.c, where it is used by pp_sort.
2059 sortcmp(const void *a, const void *b)
2061 return sv_cmp(*(SV * const *) a, *(SV * const *) b);
2068 * Store a hash table.
2070 * For a "normal" hash (not restricted, no utf8 keys):
2072 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
2073 * Values are stored as <object>.
2074 * Keys are stored as <length> <data>, the <data> section being omitted
2077 * For a "fancy" hash (restricted or utf8 keys):
2079 * Layout is SX_FLAG_HASH <size> <hash flags> followed by each key/value pair,
2081 * Values are stored as <object>.
2082 * Keys are stored as <flags> <length> <data>, the <data> section being omitted
2084 * Currently the only hash flag is "restriced"
2085 * Key flags are as for hv.h
2087 static int store_hash(stcxt_t *cxt, HV *hv)
2090 #ifdef HAS_RESTRICTED_HASHES
2099 int flagged_hash = ((SvREADONLY(hv)
2100 #ifdef HAS_HASH_KEY_FLAGS
2104 unsigned char hash_flags = (SvREADONLY(hv) ? SHV_RESTRICTED : 0);
2107 /* needs int cast for C++ compilers, doesn't it? */
2108 TRACEME(("store_hash (0x%"UVxf") (flags %x)", PTR2UV(hv),
2111 TRACEME(("store_hash (0x%"UVxf")", PTR2UV(hv)));
2115 * Signal hash by emitting SX_HASH, followed by the table length.
2119 PUTMARK(SX_FLAG_HASH);
2120 PUTMARK(hash_flags);
2125 TRACEME(("size = %d", len));
2128 * Save possible iteration state via each() on that table.
2131 riter = HvRITER(hv);
2132 eiter = HvEITER(hv);
2136 * Now store each item recursively.
2138 * If canonical is defined to some true value then store each
2139 * key/value pair in sorted order otherwise the order is random.
2140 * Canonical order is irrelevant when a deep clone operation is performed.
2142 * Fetch the value from perl only once per store() operation, and only
2147 !(cxt->optype & ST_CLONE) && (cxt->canonical == 1 ||
2148 (cxt->canonical < 0 && (cxt->canonical =
2149 (SvTRUE(perl_get_sv("Storable::canonical", TRUE)) ? 1 : 0))))
2152 * Storing in order, sorted by key.
2153 * Run through the hash, building up an array of keys in a
2154 * mortal array, sort the array and then run through the
2160 /*av_extend (av, len);*/
2162 TRACEME(("using canonical order"));
2164 for (i = 0; i < len; i++) {
2165 #ifdef HAS_RESTRICTED_HASHES
2166 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
2168 HE *he = hv_iternext(hv);
2170 SV *key = hv_iterkeysv(he);
2171 av_store(av, AvFILLp(av)+1, key); /* av_push(), really */
2174 qsort((char *) AvARRAY(av), len, sizeof(SV *), sortcmp);
2176 for (i = 0; i < len; i++) {
2177 unsigned char flags;
2181 SV *key = av_shift(av);
2182 HE *he = hv_fetch_ent(hv, key, 0, 0);
2183 SV *val = HeVAL(he);
2185 return 1; /* Internal error, not I/O error */
2188 * Store value first.
2191 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
2193 if ((ret = store(cxt, val))) /* Extra () for -Wall, grr... */
2198 * Keys are written after values to make sure retrieval
2199 * can be optimal in terms of memory usage, where keys are
2200 * read into a fixed unique buffer called kbuf.
2201 * See retrieve_hash() for details.
2204 /* Implementation of restricted hashes isn't nicely
2207 = (((hash_flags & SHV_RESTRICTED)
2209 ? SHV_K_LOCKED : 0);
2210 if (val == &PL_sv_undef)
2211 flags |= SHV_K_PLACEHOLDER;
2213 keyval = SvPV(key, keylen_tmp);
2214 keylen = keylen_tmp;
2215 #ifdef HAS_UTF8_HASHES
2216 /* If you build without optimisation on pre 5.6
2217 then nothing spots that SvUTF8(key) is always 0,
2218 so the block isn't optimised away, at which point
2219 the linker dislikes the reference to
2222 const char *keysave = keyval;
2223 bool is_utf8 = TRUE;
2225 /* Just casting the &klen to (STRLEN) won't work
2226 well if STRLEN and I32 are of different widths.
2228 keyval = (char*)bytes_from_utf8((U8*)keyval,
2232 /* If we were able to downgrade here, then than
2233 means that we have a key which only had chars
2234 0-255, but was utf8 encoded. */
2236 if (keyval != keysave) {
2237 keylen = keylen_tmp;
2238 flags |= SHV_K_WASUTF8;
2240 /* keylen_tmp can't have changed, so no need
2241 to assign back to keylen. */
2242 flags |= SHV_K_UTF8;
2249 TRACEME(("(#%d) key '%s' flags %x %u", i, keyval, flags, *keyval));
2251 assert (flags == 0);
2252 TRACEME(("(#%d) key '%s'", i, keyval));
2256 WRITE(keyval, keylen);
2257 if (flags & SHV_K_WASUTF8)
2262 * Free up the temporary array
2271 * Storing in "random" order (in the order the keys are stored
2272 * within the hash). This is the default and will be faster!
2275 for (i = 0; i < len; i++) {
2278 unsigned char flags;
2279 #ifdef HV_ITERNEXT_WANTPLACEHOLDERS
2280 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
2282 HE *he = hv_iternext(hv);
2284 SV *val = (he ? hv_iterval(hv, he) : 0);
2289 return 1; /* Internal error, not I/O error */
2292 * Store value first.
2295 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
2297 if ((ret = store(cxt, val))) /* Extra () for -Wall, grr... */
2300 /* Implementation of restricted hashes isn't nicely
2303 = (((hash_flags & SHV_RESTRICTED)
2305 ? SHV_K_LOCKED : 0);
2306 if (val == &PL_sv_undef)
2307 flags |= SHV_K_PLACEHOLDER;
2309 hek = HeKEY_hek(he);
2311 if (len == HEf_SVKEY) {
2312 /* This is somewhat sick, but the internal APIs are
2313 * such that XS code could put one of these in in
2315 * Maybe we should be capable of storing one if
2318 key_sv = HeKEY_sv(he);
2319 flags |= SHV_K_ISSV;
2321 /* Regular string key. */
2322 #ifdef HAS_HASH_KEY_FLAGS
2324 flags |= SHV_K_UTF8;
2325 if (HEK_WASUTF8(hek))
2326 flags |= SHV_K_WASUTF8;
2332 * Keys are written after values to make sure retrieval
2333 * can be optimal in terms of memory usage, where keys are
2334 * read into a fixed unique buffer called kbuf.
2335 * See retrieve_hash() for details.
2340 TRACEME(("(#%d) key '%s' flags %x", i, key, flags));
2342 assert (flags == 0);
2343 TRACEME(("(#%d) key '%s'", i, key));
2345 if (flags & SHV_K_ISSV) {
2355 TRACEME(("ok (hash 0x%"UVxf")", PTR2UV(hv)));
2358 HvRITER(hv) = riter; /* Restore hash iterator state */
2359 HvEITER(hv) = eiter;
2367 * Store a code reference.
2369 * Layout is SX_CODE <length> followed by a scalar containing the perl
2370 * source code of the code reference.
2372 static int store_code(stcxt_t *cxt, CV *cv)
2374 #if PERL_VERSION < 6
2376 * retrieve_code does not work with perl 5.005 or less
2378 return store_other(cxt, (SV*)cv);
2382 int ret, count, reallen;
2383 SV *text, *bdeparse;
2385 TRACEME(("store_code (0x%"UVxf")", PTR2UV(cv)));
2388 cxt->deparse == 0 ||
2389 (cxt->deparse < 0 && !(cxt->deparse =
2390 SvTRUE(perl_get_sv("Storable::Deparse", TRUE)) ? 1 : 0))
2392 return store_other(cxt, (SV*)cv);
2396 * Require B::Deparse. At least B::Deparse 0.61 is needed for
2397 * blessed code references.
2399 /* XXX sv_2mortal seems to be evil here. why? */
2400 load_module(PERL_LOADMOD_NOIMPORT, newSVpvn("B::Deparse",10), newSVnv(0.61));
2406 * create the B::Deparse object
2410 XPUSHs(sv_2mortal(newSVpvn("B::Deparse",10)));
2412 count = call_method("new", G_SCALAR);
2415 CROAK(("Unexpected return value from B::Deparse::new\n"));
2419 * call the coderef2text method
2423 XPUSHs(bdeparse); /* XXX is this already mortal? */
2424 XPUSHs(sv_2mortal(newRV_inc((SV*)cv)));
2426 count = call_method("coderef2text", G_SCALAR);
2429 CROAK(("Unexpected return value from B::Deparse::coderef2text\n"));
2433 reallen = strlen(SvPV_nolen(text));
2436 * Empty code references or XS functions are deparsed as
2437 * "(prototype) ;" or ";".
2440 if (len == 0 || *(SvPV_nolen(text)+reallen-1) == ';') {
2441 CROAK(("The result of B::Deparse::coderef2text was empty - maybe you're trying to serialize an XS function?\n"));
2445 * Signal code by emitting SX_CODE.
2449 TRACEME(("size = %d", len));
2450 TRACEME(("code = %s", SvPV_nolen(text)));
2453 * Now store the source code.
2456 STORE_SCALAR(SvPV_nolen(text), len);
2461 TRACEME(("ok (code)"));
2470 * When storing a tied object (be it a tied scalar, array or hash), we lay out
2471 * a special mark, followed by the underlying tied object. For instance, when
2472 * dealing with a tied hash, we store SX_TIED_HASH <hash object>, where
2473 * <hash object> stands for the serialization of the tied hash.
2475 static int store_tied(stcxt_t *cxt, SV *sv)
2480 int svt = SvTYPE(sv);
2483 TRACEME(("store_tied (0x%"UVxf")", PTR2UV(sv)));
2486 * We have a small run-time penalty here because we chose to factorise
2487 * all tieds objects into the same routine, and not have a store_tied_hash,
2488 * a store_tied_array, etc...
2490 * Don't use a switch() statement, as most compilers don't optimize that
2491 * well for 2/3 values. An if() else if() cascade is just fine. We put
2492 * tied hashes first, as they are the most likely beasts.
2495 if (svt == SVt_PVHV) {
2496 TRACEME(("tied hash"));
2497 PUTMARK(SX_TIED_HASH); /* Introduces tied hash */
2498 } else if (svt == SVt_PVAV) {
2499 TRACEME(("tied array"));
2500 PUTMARK(SX_TIED_ARRAY); /* Introduces tied array */
2502 TRACEME(("tied scalar"));
2503 PUTMARK(SX_TIED_SCALAR); /* Introduces tied scalar */
2507 if (!(mg = mg_find(sv, mtype)))
2508 CROAK(("No magic '%c' found while storing tied %s", mtype,
2509 (svt == SVt_PVHV) ? "hash" :
2510 (svt == SVt_PVAV) ? "array" : "scalar"));
2513 * The mg->mg_obj found by mg_find() above actually points to the
2514 * underlying tied Perl object implementation. For instance, if the
2515 * original SV was that of a tied array, then mg->mg_obj is an AV.
2517 * Note that we store the Perl object as-is. We don't call its FETCH
2518 * method along the way. At retrieval time, we won't call its STORE
2519 * method either, but the tieing magic will be re-installed. In itself,
2520 * that ensures that the tieing semantics are preserved since futher
2521 * accesses on the retrieved object will indeed call the magic methods...
2524 /* [#17040] mg_obj is NULL for scalar self-ties. AMS 20030416 */
2525 obj = mg->mg_obj ? mg->mg_obj : newSV(0);
2526 if ((ret = store(cxt, obj)))
2529 TRACEME(("ok (tied)"));
2537 * Stores a reference to an item within a tied structure:
2539 * . \$h{key}, stores both the (tied %h) object and 'key'.
2540 * . \$a[idx], stores both the (tied @a) object and 'idx'.
2542 * Layout is therefore either:
2543 * SX_TIED_KEY <object> <key>
2544 * SX_TIED_IDX <object> <index>
2546 static int store_tied_item(stcxt_t *cxt, SV *sv)
2551 TRACEME(("store_tied_item (0x%"UVxf")", PTR2UV(sv)));
2553 if (!(mg = mg_find(sv, 'p')))
2554 CROAK(("No magic 'p' found while storing reference to tied item"));
2557 * We discriminate between \$h{key} and \$a[idx] via mg_ptr.
2561 TRACEME(("store_tied_item: storing a ref to a tied hash item"));
2562 PUTMARK(SX_TIED_KEY);
2563 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2565 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
2568 TRACEME(("store_tied_item: storing PTR 0x%"UVxf, PTR2UV(mg->mg_ptr)));
2570 if ((ret = store(cxt, (SV *) mg->mg_ptr))) /* Idem, for -Wall */
2573 I32 idx = mg->mg_len;
2575 TRACEME(("store_tied_item: storing a ref to a tied array item "));
2576 PUTMARK(SX_TIED_IDX);
2577 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2579 if ((ret = store(cxt, mg->mg_obj))) /* Idem, for -Wall */
2582 TRACEME(("store_tied_item: storing IDX %d", idx));
2587 TRACEME(("ok (tied item)"));
2593 * store_hook -- dispatched manually, not via sv_store[]
2595 * The blessed SV is serialized by a hook.
2599 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2601 * where <flags> indicates how long <len>, <len2> and <len3> are, whether
2602 * the trailing part [] is present, the type of object (scalar, array or hash).
2603 * There is also a bit which says how the classname is stored between:
2608 * and when the <index> form is used (classname already seen), the "large
2609 * classname" bit in <flags> indicates how large the <index> is.
2611 * The serialized string returned by the hook is of length <len2> and comes
2612 * next. It is an opaque string for us.
2614 * Those <len3> object IDs which are listed last represent the extra references
2615 * not directly serialized by the hook, but which are linked to the object.
2617 * When recursion is mandated to resolve object-IDs not yet seen, we have
2618 * instead, with <header> being flags with bits set to indicate the object type
2619 * and that recursion was indeed needed:
2621 * SX_HOOK <header> <object> <header> <object> <flags>
2623 * that same header being repeated between serialized objects obtained through
2624 * recursion, until we reach flags indicating no recursion, at which point
2625 * we know we've resynchronized with a single layout, after <flags>.
2627 * When storing a blessed ref to a tied variable, the following format is
2630 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
2632 * The first <flags> indication carries an object of type SHT_EXTRA, and the
2633 * real object type is held in the <extra> flag. At the very end of the
2634 * serialization stream, the underlying magic object is serialized, just like
2635 * any other tied variable.
2637 static int store_hook(
2650 int count; /* really len3 + 1 */
2651 unsigned char flags;
2654 int recursed = 0; /* counts recursion */
2655 int obj_type; /* object type, on 2 bits */
2658 int clone = cxt->optype & ST_CLONE;
2659 char mtype = '\0'; /* for blessed ref to tied structures */
2660 unsigned char eflags = '\0'; /* used when object type is SHT_EXTRA */
2662 TRACEME(("store_hook, class \"%s\", tagged #%d", HvNAME(pkg), cxt->tagnum));
2665 * Determine object type on 2 bits.
2670 obj_type = SHT_SCALAR;
2673 obj_type = SHT_ARRAY;
2676 obj_type = SHT_HASH;
2680 * Produced by a blessed ref to a tied data structure, $o in the
2681 * following Perl code.
2685 * my $o = bless \%h, 'BAR';
2687 * Signal the tie-ing magic by setting the object type as SHT_EXTRA
2688 * (since we have only 2 bits in <flags> to store the type), and an
2689 * <extra> byte flag will be emitted after the FIRST <flags> in the
2690 * stream, carrying what we put in `eflags'.
2692 obj_type = SHT_EXTRA;
2693 switch (SvTYPE(sv)) {
2695 eflags = (unsigned char) SHT_THASH;
2699 eflags = (unsigned char) SHT_TARRAY;
2703 eflags = (unsigned char) SHT_TSCALAR;
2709 CROAK(("Unexpected object type (%d) in store_hook()", type));
2711 flags = SHF_NEED_RECURSE | obj_type;
2713 class = HvNAME(pkg);
2714 len = strlen(class);
2717 * To call the hook, we need to fake a call like:
2719 * $object->STORABLE_freeze($cloning);
2721 * but we don't have the $object here. For instance, if $object is
2722 * a blessed array, what we have in `sv' is the array, and we can't
2723 * call a method on those.
2725 * Therefore, we need to create a temporary reference to the object and
2726 * make the call on that reference.
2729 TRACEME(("about to call STORABLE_freeze on class %s", class));
2731 ref = newRV_noinc(sv); /* Temporary reference */
2732 av = array_call(ref, hook, clone); /* @a = $object->STORABLE_freeze($c) */
2734 SvREFCNT_dec(ref); /* Reclaim temporary reference */
2736 count = AvFILLp(av) + 1;
2737 TRACEME(("store_hook, array holds %d items", count));
2740 * If they return an empty list, it means they wish to ignore the
2741 * hook for this class (and not just this instance -- that's for them
2742 * to handle if they so wish).
2744 * Simply disable the cached entry for the hook (it won't be recomputed
2745 * since it's present in the cache) and recurse to store_blessed().
2750 * They must not change their mind in the middle of a serialization.
2753 if (hv_fetch(cxt->hclass, class, len, FALSE))
2754 CROAK(("Too late to ignore hooks for %s class \"%s\"",
2755 (cxt->optype & ST_CLONE) ? "cloning" : "storing", class));
2757 pkg_hide(cxt->hook, pkg, "STORABLE_freeze");
2759 ASSERT(!pkg_can(cxt->hook, pkg, "STORABLE_freeze"), ("hook invisible"));
2760 TRACEME(("ignoring STORABLE_freeze in class \"%s\"", class));
2762 return store_blessed(cxt, sv, type, pkg);
2766 * Get frozen string.
2770 pv = SvPV(ary[0], len2);
2773 * If they returned more than one item, we need to serialize some
2774 * extra references if not already done.
2776 * Loop over the array, starting at position #1, and for each item,
2777 * ensure it is a reference, serialize it if not already done, and
2778 * replace the entry with the tag ID of the corresponding serialized
2781 * We CHEAT by not calling av_fetch() and read directly within the
2785 for (i = 1; i < count; i++) {
2789 AV *av_hook = cxt->hook_seen;
2792 CROAK(("Item #%d returned by STORABLE_freeze "
2793 "for %s is not a reference", i, class));
2794 xsv = SvRV(rsv); /* Follow ref to know what to look for */
2797 * Look in hseen and see if we have a tag already.
2798 * Serialize entry if not done already, and get its tag.
2801 if ((svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE)))
2802 goto sv_seen; /* Avoid moving code too far to the right */
2804 TRACEME(("listed object %d at 0x%"UVxf" is unknown", i-1, PTR2UV(xsv)));
2807 * We need to recurse to store that object and get it to be known
2808 * so that we can resolve the list of object-IDs at retrieve time.
2810 * The first time we do this, we need to emit the proper header
2811 * indicating that we recursed, and what the type of object is (the
2812 * object we're storing via a user-hook). Indeed, during retrieval,
2813 * we'll have to create the object before recursing to retrieve the
2814 * others, in case those would point back at that object.
2817 /* [SX_HOOK] <flags> [<extra>] <object>*/
2821 if (obj_type == SHT_EXTRA)
2826 if ((ret = store(cxt, xsv))) /* Given by hook for us to store */
2829 svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE);
2831 CROAK(("Could not serialize item #%d from hook in %s", i, class));
2834 * It was the first time we serialized `xsv'.
2836 * Keep this SV alive until the end of the serialization: if we
2837 * disposed of it right now by decrementing its refcount, and it was
2838 * a temporary value, some next temporary value allocated during
2839 * another STORABLE_freeze might take its place, and we'd wrongly
2840 * assume that new SV was already serialized, based on its presence
2843 * Therefore, push it away in cxt->hook_seen.
2846 av_store(av_hook, AvFILLp(av_hook)+1, SvREFCNT_inc(xsv));
2850 * Dispose of the REF they returned. If we saved the `xsv' away
2851 * in the array of returned SVs, that will not cause the underlying
2852 * referenced SV to be reclaimed.
2855 ASSERT(SvREFCNT(xsv) > 1, ("SV will survive disposal of its REF"));
2856 SvREFCNT_dec(rsv); /* Dispose of reference */
2859 * Replace entry with its tag (not a real SV, so no refcnt increment)
2863 TRACEME(("listed object %d at 0x%"UVxf" is tag #%"UVuf,
2864 i-1, PTR2UV(xsv), PTR2UV(*svh)));
2868 * Allocate a class ID if not already done.
2870 * This needs to be done after the recursion above, since at retrieval
2871 * time, we'll see the inner objects first. Many thanks to
2872 * Salvador Ortiz Garcia <sog@msg.com.mx> who spot that bug and
2873 * proposed the right fix. -- RAM, 15/09/2000
2876 if (!known_class(cxt, class, len, &classnum)) {
2877 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2878 classnum = -1; /* Mark: we must store classname */
2880 TRACEME(("already seen class %s, ID = %d", class, classnum));
2884 * Compute leading flags.
2888 if (((classnum == -1) ? len : classnum) > LG_SCALAR)
2889 flags |= SHF_LARGE_CLASSLEN;
2891 flags |= SHF_IDX_CLASSNAME;
2892 if (len2 > LG_SCALAR)
2893 flags |= SHF_LARGE_STRLEN;
2895 flags |= SHF_HAS_LIST;
2896 if (count > (LG_SCALAR + 1))
2897 flags |= SHF_LARGE_LISTLEN;
2900 * We're ready to emit either serialized form:
2902 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2903 * SX_HOOK <flags> <index> <len2> <str> [<len3> <object-IDs>]
2905 * If we recursed, the SX_HOOK has already been emitted.
2908 TRACEME(("SX_HOOK (recursed=%d) flags=0x%x "
2909 "class=%"IVdf" len=%"IVdf" len2=%"IVdf" len3=%d",
2910 recursed, flags, (IV)classnum, (IV)len, (IV)len2, count-1));
2912 /* SX_HOOK <flags> [<extra>] */
2916 if (obj_type == SHT_EXTRA)
2921 /* <len> <classname> or <index> */
2922 if (flags & SHF_IDX_CLASSNAME) {
2923 if (flags & SHF_LARGE_CLASSLEN)
2926 unsigned char cnum = (unsigned char) classnum;
2930 if (flags & SHF_LARGE_CLASSLEN)
2933 unsigned char clen = (unsigned char) len;
2936 WRITE(class, len); /* Final \0 is omitted */
2939 /* <len2> <frozen-str> */
2940 if (flags & SHF_LARGE_STRLEN) {
2941 I32 wlen2 = len2; /* STRLEN might be 8 bytes */
2942 WLEN(wlen2); /* Must write an I32 for 64-bit machines */
2944 unsigned char clen = (unsigned char) len2;
2948 WRITE(pv, (SSize_t)len2); /* Final \0 is omitted */
2950 /* [<len3> <object-IDs>] */
2951 if (flags & SHF_HAS_LIST) {
2952 int len3 = count - 1;
2953 if (flags & SHF_LARGE_LISTLEN)
2956 unsigned char clen = (unsigned char) len3;
2961 * NOTA BENE, for 64-bit machines: the ary[i] below does not yield a
2962 * real pointer, rather a tag number, well under the 32-bit limit.
2965 for (i = 1; i < count; i++) {
2966 I32 tagval = htonl(LOW_32BITS(ary[i]));
2968 TRACEME(("object %d, tag #%d", i-1, ntohl(tagval)));
2973 * Free the array. We need extra care for indices after 0, since they
2974 * don't hold real SVs but integers cast.
2978 AvFILLp(av) = 0; /* Cheat, nothing after 0 interests us */
2983 * If object was tied, need to insert serialization of the magic object.
2986 if (obj_type == SHT_EXTRA) {
2989 if (!(mg = mg_find(sv, mtype))) {
2990 int svt = SvTYPE(sv);
2991 CROAK(("No magic '%c' found while storing ref to tied %s with hook",
2992 mtype, (svt == SVt_PVHV) ? "hash" :
2993 (svt == SVt_PVAV) ? "array" : "scalar"));
2996 TRACEME(("handling the magic object 0x%"UVxf" part of 0x%"UVxf,
2997 PTR2UV(mg->mg_obj), PTR2UV(sv)));
3003 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
3011 * store_blessed -- dispatched manually, not via sv_store[]
3013 * Check whether there is a STORABLE_xxx hook defined in the class or in one
3014 * of its ancestors. If there is, then redispatch to store_hook();
3016 * Otherwise, the blessed SV is stored using the following layout:
3018 * SX_BLESS <flag> <len> <classname> <object>
3020 * where <flag> indicates whether <len> is stored on 0 or 4 bytes, depending
3021 * on the high-order bit in flag: if 1, then length follows on 4 bytes.
3022 * Otherwise, the low order bits give the length, thereby giving a compact
3023 * representation for class names less than 127 chars long.
3025 * Each <classname> seen is remembered and indexed, so that the next time
3026 * an object in the blessed in the same <classname> is stored, the following
3029 * SX_IX_BLESS <flag> <index> <object>
3031 * where <index> is the classname index, stored on 0 or 4 bytes depending
3032 * on the high-order bit in flag (same encoding as above for <len>).
3034 static int store_blessed(
3045 TRACEME(("store_blessed, type %d, class \"%s\"", type, HvNAME(pkg)));
3048 * Look for a hook for this blessed SV and redirect to store_hook()
3052 hook = pkg_can(cxt->hook, pkg, "STORABLE_freeze");
3054 return store_hook(cxt, sv, type, pkg, hook);
3057 * This is a blessed SV without any serialization hook.
3060 class = HvNAME(pkg);
3061 len = strlen(class);
3063 TRACEME(("blessed 0x%"UVxf" in %s, no hook: tagged #%d",
3064 PTR2UV(sv), class, cxt->tagnum));
3067 * Determine whether it is the first time we see that class name (in which
3068 * case it will be stored in the SX_BLESS form), or whether we already
3069 * saw that class name before (in which case the SX_IX_BLESS form will be
3073 if (known_class(cxt, class, len, &classnum)) {
3074 TRACEME(("already seen class %s, ID = %d", class, classnum));
3075 PUTMARK(SX_IX_BLESS);
3076 if (classnum <= LG_BLESS) {
3077 unsigned char cnum = (unsigned char) classnum;
3080 unsigned char flag = (unsigned char) 0x80;
3085 TRACEME(("first time we see class %s, ID = %d", class, classnum));
3087 if (len <= LG_BLESS) {
3088 unsigned char clen = (unsigned char) len;
3091 unsigned char flag = (unsigned char) 0x80;
3093 WLEN(len); /* Don't BER-encode, this should be rare */
3095 WRITE(class, len); /* Final \0 is omitted */
3099 * Now emit the <object> part.
3102 return SV_STORE(type)(cxt, sv);
3108 * We don't know how to store the item we reached, so return an error condition.
3109 * (it's probably a GLOB, some CODE reference, etc...)
3111 * If they defined the `forgive_me' variable at the Perl level to some
3112 * true value, then don't croak, just warn, and store a placeholder string
3115 static int store_other(stcxt_t *cxt, SV *sv)
3118 static char buf[80];
3120 TRACEME(("store_other"));
3123 * Fetch the value from perl only once per store() operation.
3127 cxt->forgive_me == 0 ||
3128 (cxt->forgive_me < 0 && !(cxt->forgive_me =
3129 SvTRUE(perl_get_sv("Storable::forgive_me", TRUE)) ? 1 : 0))
3131 CROAK(("Can't store %s items", sv_reftype(sv, FALSE)));
3133 warn("Can't store item %s(0x%"UVxf")",
3134 sv_reftype(sv, FALSE), PTR2UV(sv));
3137 * Store placeholder string as a scalar instead...
3140 (void) sprintf(buf, "You lost %s(0x%"UVxf")%c", sv_reftype(sv, FALSE),
3141 PTR2UV(sv), (char) 0);
3144 STORE_SCALAR(buf, len);
3145 TRACEME(("ok (dummy \"%s\", length = %"IVdf")", buf, (IV) len));
3151 *** Store driving routines
3157 * WARNING: partially duplicates Perl's sv_reftype for speed.
3159 * Returns the type of the SV, identified by an integer. That integer
3160 * may then be used to index the dynamic routine dispatch table.
3162 static int sv_type(SV *sv)
3164 switch (SvTYPE(sv)) {
3169 * No need to check for ROK, that can't be set here since there
3170 * is no field capable of hodling the xrv_rv reference.
3178 * Starting from SVt_PV, it is possible to have the ROK flag
3179 * set, the pointer to the other SV being either stored in
3180 * the xrv_rv (in the case of a pure SVt_RV), or as the
3181 * xpv_pv field of an SVt_PV and its heirs.
3183 * However, those SV cannot be magical or they would be an
3184 * SVt_PVMG at least.
3186 return SvROK(sv) ? svis_REF : svis_SCALAR;
3188 case SVt_PVLV: /* Workaround for perl5.004_04 "LVALUE" bug */
3189 if (SvRMAGICAL(sv) && (mg_find(sv, 'p')))
3190 return svis_TIED_ITEM;
3193 if (SvRMAGICAL(sv) && (mg_find(sv, 'q')))
3195 return SvROK(sv) ? svis_REF : svis_SCALAR;
3197 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
3201 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
3216 * Recursively store objects pointed to by the sv to the specified file.
3218 * Layout is <content> or SX_OBJECT <tagnum> if we reach an already stored
3219 * object (one for which storage has started -- it may not be over if we have
3220 * a self-referenced structure). This data set forms a stored <object>.
3222 static int store(stcxt_t *cxt, SV *sv)
3227 HV *hseen = cxt->hseen;
3229 TRACEME(("store (0x%"UVxf")", PTR2UV(sv)));
3232 * If object has already been stored, do not duplicate data.
3233 * Simply emit the SX_OBJECT marker followed by its tag data.
3234 * The tag is always written in network order.
3236 * NOTA BENE, for 64-bit machines: the "*svh" below does not yield a
3237 * real pointer, rather a tag number (watch the insertion code below).
3238 * That means it probably safe to assume it is well under the 32-bit limit,
3239 * and makes the truncation safe.
3240 * -- RAM, 14/09/1999
3243 svh = hv_fetch(hseen, (char *) &sv, sizeof(sv), FALSE);
3245 I32 tagval = htonl(LOW_32BITS(*svh));
3247 TRACEME(("object 0x%"UVxf" seen as #%d", PTR2UV(sv), ntohl(tagval)));
3255 * Allocate a new tag and associate it with the address of the sv being
3256 * stored, before recursing...
3258 * In order to avoid creating new SvIVs to hold the tagnum we just
3259 * cast the tagnum to an SV pointer and store that in the hash. This
3260 * means that we must clean up the hash manually afterwards, but gives
3261 * us a 15% throughput increase.
3266 if (!hv_store(hseen,
3267 (char *) &sv, sizeof(sv), INT2PTR(SV*, cxt->tagnum), 0))
3271 * Store `sv' and everything beneath it, using appropriate routine.
3272 * Abort immediately if we get a non-zero status back.
3277 TRACEME(("storing 0x%"UVxf" tag #%d, type %d...",
3278 PTR2UV(sv), cxt->tagnum, type));
3281 HV *pkg = SvSTASH(sv);
3282 ret = store_blessed(cxt, sv, type, pkg);
3284 ret = SV_STORE(type)(cxt, sv);
3286 TRACEME(("%s (stored 0x%"UVxf", refcnt=%d, %s)",
3287 ret ? "FAILED" : "ok", PTR2UV(sv),
3288 SvREFCNT(sv), sv_reftype(sv, FALSE)));
3296 * Write magic number and system information into the file.
3297 * Layout is <magic> <network> [<len> <byteorder> <sizeof int> <sizeof long>
3298 * <sizeof ptr>] where <len> is the length of the byteorder hexa string.
3299 * All size and lenghts are written as single characters here.
3301 * Note that no byte ordering info is emitted when <network> is true, since
3302 * integers will be emitted in network order in that case.
3304 static int magic_write(stcxt_t *cxt)
3307 * Starting with 0.6, the "use_network_order" byte flag is also used to
3308 * indicate the version number of the binary image, encoded in the upper
3309 * bits. The bit 0 is always used to indicate network order.
3312 * Starting with 0.7, a full byte is dedicated to the minor version of
3313 * the binary format, which is incremented only when new markers are
3314 * introduced, for instance, but when backward compatibility is preserved.
3317 /* Make these at compile time. The WRITE() macro is sufficiently complex
3318 that it saves about 200 bytes doing it this way and only using it
3320 static const unsigned char network_file_header[] = {
3322 (STORABLE_BIN_MAJOR << 1) | 1,
3323 STORABLE_BIN_WRITE_MINOR
3325 static const unsigned char file_header[] = {
3327 (STORABLE_BIN_MAJOR << 1) | 0,
3328 STORABLE_BIN_WRITE_MINOR,
3329 /* sizeof the array includes the 0 byte at the end: */
3330 (char) sizeof (byteorderstr) - 1,
3332 (unsigned char) sizeof(int),
3333 (unsigned char) sizeof(long),
3334 (unsigned char) sizeof(char *),
3335 (unsigned char) sizeof(NV)
3337 #ifdef USE_56_INTERWORK_KLUDGE
3338 static const unsigned char file_header_56[] = {
3340 (STORABLE_BIN_MAJOR << 1) | 0,
3341 STORABLE_BIN_WRITE_MINOR,
3342 /* sizeof the array includes the 0 byte at the end: */
3343 (char) sizeof (byteorderstr_56) - 1,
3345 (unsigned char) sizeof(int),
3346 (unsigned char) sizeof(long),
3347 (unsigned char) sizeof(char *),
3348 (unsigned char) sizeof(NV)
3351 const unsigned char *header;
3354 TRACEME(("magic_write on fd=%d", cxt->fio ? PerlIO_fileno(cxt->fio) : -1));
3356 if (cxt->netorder) {
3357 header = network_file_header;
3358 length = sizeof (network_file_header);
3360 #ifdef USE_56_INTERWORK_KLUDGE
3361 if (SvTRUE(perl_get_sv("Storable::interwork_56_64bit", TRUE))) {
3362 header = file_header_56;
3363 length = sizeof (file_header_56);
3367 header = file_header;
3368 length = sizeof (file_header);
3373 /* sizeof the array includes the 0 byte at the end. */
3374 header += sizeof (magicstr) - 1;
3375 length -= sizeof (magicstr) - 1;
3378 WRITE( (unsigned char*) header, length);
3380 if (!cxt->netorder) {
3381 TRACEME(("ok (magic_write byteorder = 0x%lx [%d], I%d L%d P%d D%d)",
3382 (unsigned long) BYTEORDER, (int) sizeof (byteorderstr) - 1,
3383 (int) sizeof(int), (int) sizeof(long),
3384 (int) sizeof(char *), (int) sizeof(NV)));
3392 * Common code for store operations.
3394 * When memory store is requested (f = NULL) and a non null SV* is given in
3395 * `res', it is filled with a new SV created out of the memory buffer.
3397 * It is required to provide a non-null `res' when the operation type is not
3398 * dclone() and store() is performed to memory.
3400 static int do_store(
3410 ASSERT(!(f == 0 && !(optype & ST_CLONE)) || res,
3411 ("must supply result SV pointer for real recursion to memory"));
3413 TRACEME(("do_store (optype=%d, netorder=%d)",
3414 optype, network_order));
3419 * Workaround for CROAK leak: if they enter with a "dirty" context,
3420 * free up memory for them now.
3427 * Now that STORABLE_xxx hooks exist, it is possible that they try to
3428 * re-enter store() via the hooks. We need to stack contexts.
3432 cxt = allocate_context(cxt);
3436 ASSERT(cxt->entry == 1, ("starting new recursion"));
3437 ASSERT(!cxt->s_dirty, ("clean context"));
3440 * Ensure sv is actually a reference. From perl, we called something
3442 * pstore(FILE, \@array);
3443 * so we must get the scalar value behing that reference.
3447 CROAK(("Not a reference"));
3448 sv = SvRV(sv); /* So follow it to know what to store */
3451 * If we're going to store to memory, reset the buffer.
3458 * Prepare context and emit headers.
3461 init_store_context(cxt, f, optype, network_order);
3463 if (-1 == magic_write(cxt)) /* Emit magic and ILP info */
3464 return 0; /* Error */
3467 * Recursively store object...
3470 ASSERT(is_storing(), ("within store operation"));
3472 status = store(cxt, sv); /* Just do it! */
3475 * If they asked for a memory store and they provided an SV pointer,
3476 * make an SV string out of the buffer and fill their pointer.
3478 * When asking for ST_REAL, it's MANDATORY for the caller to provide
3479 * an SV, since context cleanup might free the buffer if we did recurse.
3480 * (unless caller is dclone(), which is aware of that).
3483 if (!cxt->fio && res)
3489 * The "root" context is never freed, since it is meant to be always
3490 * handy for the common case where no recursion occurs at all (i.e.
3491 * we enter store() outside of any Storable code and leave it, period).
3492 * We know it's the "root" context because there's nothing stacked
3497 * When deep cloning, we don't free the context: doing so would force
3498 * us to copy the data in the memory buffer. Sicne we know we're
3499 * about to enter do_retrieve...
3502 clean_store_context(cxt);
3503 if (cxt->prev && !(cxt->optype & ST_CLONE))
3506 TRACEME(("do_store returns %d", status));
3514 * Store the transitive data closure of given object to disk.
3515 * Returns 0 on error, a true value otherwise.
3517 int pstore(PerlIO *f, SV *sv)
3519 TRACEME(("pstore"));
3520 return do_store(f, sv, 0, FALSE, (SV**) 0);
3527 * Same as pstore(), but network order is used for integers and doubles are
3528 * emitted as strings.
3530 int net_pstore(PerlIO *f, SV *sv)
3532 TRACEME(("net_pstore"));
3533 return do_store(f, sv, 0, TRUE, (SV**) 0);
3543 * Build a new SV out of the content of the internal memory buffer.
3545 static SV *mbuf2sv(void)
3549 return newSVpv(mbase, MBUF_SIZE());
3555 * Store the transitive data closure of given object to memory.
3556 * Returns undef on error, a scalar value containing the data otherwise.
3562 TRACEME(("mstore"));
3564 if (!do_store((PerlIO*) 0, sv, 0, FALSE, &out))
3565 return &PL_sv_undef;
3573 * Same as mstore(), but network order is used for integers and doubles are
3574 * emitted as strings.
3576 SV *net_mstore(SV *sv)
3580 TRACEME(("net_mstore"));
3582 if (!do_store((PerlIO*) 0, sv, 0, TRUE, &out))
3583 return &PL_sv_undef;
3589 *** Specific retrieve callbacks.
3595 * Return an error via croak, since it is not possible that we get here
3596 * under normal conditions, when facing a file produced via pstore().
3598 static SV *retrieve_other(stcxt_t *cxt, char *cname)
3601 cxt->ver_major != STORABLE_BIN_MAJOR &&
3602 cxt->ver_minor != STORABLE_BIN_MINOR
3604 CROAK(("Corrupted storable %s (binary v%d.%d), current is v%d.%d",
3605 cxt->fio ? "file" : "string",
3606 cxt->ver_major, cxt->ver_minor,
3607 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
3609 CROAK(("Corrupted storable %s (binary v%d.%d)",
3610 cxt->fio ? "file" : "string",
3611 cxt->ver_major, cxt->ver_minor));
3614 return (SV *) 0; /* Just in case */
3618 * retrieve_idx_blessed
3620 * Layout is SX_IX_BLESS <index> <object> with SX_IX_BLESS already read.
3621 * <index> can be coded on either 1 or 5 bytes.
3623 static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname)
3630 TRACEME(("retrieve_idx_blessed (#%d)", cxt->tagnum));
3631 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3633 GETMARK(idx); /* Index coded on a single char? */
3638 * Fetch classname in `aclass'
3641 sva = av_fetch(cxt->aclass, idx, FALSE);
3643 CROAK(("Class name #%"IVdf" should have been seen already", (IV) idx));
3645 class = SvPVX(*sva); /* We know it's a PV, by construction */
3647 TRACEME(("class ID %d => %s", idx, class));
3650 * Retrieve object and bless it.
3653 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
3661 * Layout is SX_BLESS <len> <classname> <object> with SX_BLESS already read.
3662 * <len> can be coded on either 1 or 5 bytes.
3664 static SV *retrieve_blessed(stcxt_t *cxt, char *cname)
3668 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3671 TRACEME(("retrieve_blessed (#%d)", cxt->tagnum));
3672 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3675 * Decode class name length and read that name.
3677 * Short classnames have two advantages: their length is stored on one
3678 * single byte, and the string can be read on the stack.
3681 GETMARK(len); /* Length coded on a single char? */
3684 TRACEME(("** allocating %d bytes for class name", len+1));
3685 New(10003, class, len+1, char);
3688 class[len] = '\0'; /* Mark string end */
3691 * It's a new classname, otherwise it would have been an SX_IX_BLESS.
3694 TRACEME(("new class name \"%s\" will bear ID = %d", class, cxt->classnum));
3696 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3700 * Retrieve object and bless it.
3703 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
3713 * Layout: SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
3714 * with leading mark already read, as usual.
3716 * When recursion was involved during serialization of the object, there
3717 * is an unknown amount of serialized objects after the SX_HOOK mark. Until
3718 * we reach a <flags> marker with the recursion bit cleared.
3720 * If the first <flags> byte contains a type of SHT_EXTRA, then the real type
3721 * is held in the <extra> byte, and if the object is tied, the serialized
3722 * magic object comes at the very end:
3724 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
3726 * This means the STORABLE_thaw hook will NOT get a tied variable during its
3727 * processing (since we won't have seen the magic object by the time the hook
3728 * is called). See comments below for why it was done that way.
3730 static SV *retrieve_hook(stcxt_t *cxt, char *cname)
3733 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3744 int clone = cxt->optype & ST_CLONE;
3746 unsigned int extra_type = 0;
3748 TRACEME(("retrieve_hook (#%d)", cxt->tagnum));
3749 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3752 * Read flags, which tell us about the type, and whether we need to recurse.
3758 * Create the (empty) object, and mark it as seen.
3760 * This must be done now, because tags are incremented, and during
3761 * serialization, the object tag was affected before recursion could
3765 obj_type = flags & SHF_TYPE_MASK;
3771 sv = (SV *) newAV();
3774 sv = (SV *) newHV();
3778 * Read <extra> flag to know the type of the object.
3779 * Record associated magic type for later.
3781 GETMARK(extra_type);
3782 switch (extra_type) {
3788 sv = (SV *) newAV();
3792 sv = (SV *) newHV();
3796 return retrieve_other(cxt, 0); /* Let it croak */
3800 return retrieve_other(cxt, 0); /* Let it croak */
3802 SEEN(sv, 0); /* Don't bless yet */
3805 * Whilst flags tell us to recurse, do so.
3807 * We don't need to remember the addresses returned by retrieval, because
3808 * all the references will be obtained through indirection via the object
3809 * tags in the object-ID list.
3811 * We need to decrement the reference count for these objects
3812 * because, if the user doesn't save a reference to them in the hook,
3813 * they must be freed when this context is cleaned.
3816 while (flags & SHF_NEED_RECURSE) {
3817 TRACEME(("retrieve_hook recursing..."));
3818 rv = retrieve(cxt, 0);
3822 TRACEME(("retrieve_hook back with rv=0x%"UVxf,
3827 if (flags & SHF_IDX_CLASSNAME) {
3832 * Fetch index from `aclass'
3835 if (flags & SHF_LARGE_CLASSLEN)
3840 sva = av_fetch(cxt->aclass, idx, FALSE);
3842 CROAK(("Class name #%"IVdf" should have been seen already",
3845 class = SvPVX(*sva); /* We know it's a PV, by construction */
3846 TRACEME(("class ID %d => %s", idx, class));
3850 * Decode class name length and read that name.
3852 * NOTA BENE: even if the length is stored on one byte, we don't read
3853 * on the stack. Just like retrieve_blessed(), we limit the name to
3854 * LG_BLESS bytes. This is an arbitrary decision.
3857 if (flags & SHF_LARGE_CLASSLEN)
3862 if (len > LG_BLESS) {
3863 TRACEME(("** allocating %d bytes for class name", len+1));
3864 New(10003, class, len+1, char);
3868 class[len] = '\0'; /* Mark string end */
3871 * Record new classname.
3874 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3878 TRACEME(("class name: %s", class));
3881 * Decode user-frozen string length and read it in an SV.
3883 * For efficiency reasons, we read data directly into the SV buffer.
3884 * To understand that code, read retrieve_scalar()
3887 if (flags & SHF_LARGE_STRLEN)
3892 frozen = NEWSV(10002, len2);
3894 SAFEREAD(SvPVX(frozen), len2, frozen);
3895 SvCUR_set(frozen, len2);
3896 *SvEND(frozen) = '\0';
3898 (void) SvPOK_only(frozen); /* Validates string pointer */
3899 if (cxt->s_tainted) /* Is input source tainted? */
3902 TRACEME(("frozen string: %d bytes", len2));
3905 * Decode object-ID list length, if present.
3908 if (flags & SHF_HAS_LIST) {
3909 if (flags & SHF_LARGE_LISTLEN)
3915 av_extend(av, len3 + 1); /* Leave room for [0] */
3916 AvFILLp(av) = len3; /* About to be filled anyway */
3920 TRACEME(("has %d object IDs to link", len3));
3923 * Read object-ID list into array.
3924 * Because we pre-extended it, we can cheat and fill it manually.
3926 * We read object tags and we can convert them into SV* on the fly
3927 * because we know all the references listed in there (as tags)
3928 * have been already serialized, hence we have a valid correspondance
3929 * between each of those tags and the recreated SV.
3933 SV **ary = AvARRAY(av);
3935 for (i = 1; i <= len3; i++) { /* We leave [0] alone */
3942 svh = av_fetch(cxt->aseen, tag, FALSE);
3944 CROAK(("Object #%"IVdf" should have been retrieved already",
3947 ary[i] = SvREFCNT_inc(xsv);
3952 * Bless the object and look up the STORABLE_thaw hook.
3956 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3959 * Hook not found. Maybe they did not require the module where this
3960 * hook is defined yet?
3962 * If the require below succeeds, we'll be able to find the hook.
3963 * Still, it only works reliably when each class is defined in a
3967 SV *psv = newSVpvn("require ", 8);
3968 sv_catpv(psv, class);
3970 TRACEME(("No STORABLE_thaw defined for objects of class %s", class));
3971 TRACEME(("Going to require module '%s' with '%s'", class, SvPVX(psv)));
3973 perl_eval_sv(psv, G_DISCARD);
3977 * We cache results of pkg_can, so we need to uncache before attempting
3981 pkg_uncache(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3982 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3985 CROAK(("No STORABLE_thaw defined for objects of class %s "
3986 "(even after a \"require %s;\")", class, class));
3990 * If we don't have an `av' yet, prepare one.
3991 * Then insert the frozen string as item [0].
3999 AvARRAY(av)[0] = SvREFCNT_inc(frozen);
4004 * $object->STORABLE_thaw($cloning, $frozen, @refs);
4006 * where $object is our blessed (empty) object, $cloning is a boolean
4007 * telling whether we're running a deep clone, $frozen is the frozen
4008 * string the user gave us in his serializing hook, and @refs, which may
4009 * be empty, is the list of extra references he returned along for us
4012 * In effect, the hook is an alternate creation routine for the class,
4013 * the object itself being already created by the runtime.
4016 TRACEME(("calling STORABLE_thaw on %s at 0x%"UVxf" (%"IVdf" args)",
4017 class, PTR2UV(sv), (IV) AvFILLp(av) + 1));
4020 (void) scalar_call(rv, hook, clone, av, G_SCALAR|G_DISCARD);
4027 SvREFCNT_dec(frozen);
4030 if (!(flags & SHF_IDX_CLASSNAME) && class != buf)
4034 * If we had an <extra> type, then the object was not as simple, and
4035 * we need to restore extra magic now.
4041 TRACEME(("retrieving magic object for 0x%"UVxf"...", PTR2UV(sv)));
4043 rv = retrieve(cxt, 0); /* Retrieve <magic object> */
4045 TRACEME(("restoring the magic object 0x%"UVxf" part of 0x%"UVxf,
4046 PTR2UV(rv), PTR2UV(sv)));
4048 switch (extra_type) {
4050 sv_upgrade(sv, SVt_PVMG);
4053 sv_upgrade(sv, SVt_PVAV);
4054 AvREAL_off((AV *)sv);
4057 sv_upgrade(sv, SVt_PVHV);
4060 CROAK(("Forgot to deal with extra type %d", extra_type));
4065 * Adding the magic only now, well after the STORABLE_thaw hook was called
4066 * means the hook cannot know it deals with an object whose variable is
4067 * tied. But this is happening when retrieving $o in the following case:
4071 * my $o = bless \%h, 'BAR';
4073 * The 'BAR' class is NOT the one where %h is tied into. Therefore, as
4074 * far as the 'BAR' class is concerned, the fact that %h is not a REAL
4075 * hash but a tied one should not matter at all, and remain transparent.
4076 * This means the magic must be restored by Storable AFTER the hook is
4079 * That looks very reasonable to me, but then I've come up with this
4080 * after a bug report from David Nesting, who was trying to store such
4081 * an object and caused Storable to fail. And unfortunately, it was
4082 * also the easiest way to retrofit support for blessed ref to tied objects
4083 * into the existing design. -- RAM, 17/02/2001
4086 sv_magic(sv, rv, mtype, Nullch, 0);
4087 SvREFCNT_dec(rv); /* Undo refcnt inc from sv_magic() */
4095 * Retrieve reference to some other scalar.
4096 * Layout is SX_REF <object>, with SX_REF already read.
4098 static SV *retrieve_ref(stcxt_t *cxt, char *cname)
4103 TRACEME(("retrieve_ref (#%d)", cxt->tagnum));
4106 * We need to create the SV that holds the reference to the yet-to-retrieve
4107 * object now, so that we may record the address in the seen table.
4108 * Otherwise, if the object to retrieve references us, we won't be able
4109 * to resolve the SX_OBJECT we'll see at that point! Hence we cannot
4110 * do the retrieve first and use rv = newRV(sv) since it will be too late
4111 * for SEEN() recording.
4114 rv = NEWSV(10002, 0);
4115 SEEN(rv, cname); /* Will return if rv is null */
4116 sv = retrieve(cxt, 0); /* Retrieve <object> */
4118 return (SV *) 0; /* Failed */
4121 * WARNING: breaks RV encapsulation.
4123 * Now for the tricky part. We have to upgrade our existing SV, so that
4124 * it is now an RV on sv... Again, we cheat by duplicating the code
4125 * held in newSVrv(), since we already got our SV from retrieve().
4129 * SvRV(rv) = SvREFCNT_inc(sv);
4131 * here because the reference count we got from retrieve() above is
4132 * already correct: if the object was retrieved from the file, then
4133 * its reference count is one. Otherwise, if it was retrieved via
4134 * an SX_OBJECT indication, a ref count increment was done.
4138 /* Do not use sv_upgrade to preserve STASH */
4139 SvFLAGS(rv) &= ~SVTYPEMASK;
4140 SvFLAGS(rv) |= SVt_RV;
4142 sv_upgrade(rv, SVt_RV);
4145 SvRV(rv) = sv; /* $rv = \$sv */
4148 TRACEME(("ok (retrieve_ref at 0x%"UVxf")", PTR2UV(rv)));
4154 * retrieve_overloaded
4156 * Retrieve reference to some other scalar with overloading.
4157 * Layout is SX_OVERLOAD <object>, with SX_OVERLOAD already read.
4159 static SV *retrieve_overloaded(stcxt_t *cxt, char *cname)
4165 TRACEME(("retrieve_overloaded (#%d)", cxt->tagnum));
4168 * Same code as retrieve_ref(), duplicated to avoid extra call.
4171 rv = NEWSV(10002, 0);
4172 SEEN(rv, cname); /* Will return if rv is null */
4173 sv = retrieve(cxt, 0); /* Retrieve <object> */
4175 return (SV *) 0; /* Failed */
4178 * WARNING: breaks RV encapsulation.
4181 sv_upgrade(rv, SVt_RV);
4182 SvRV(rv) = sv; /* $rv = \$sv */
4186 * Restore overloading magic.
4189 stash = (HV *) SvSTASH (sv);
4190 if (!stash || !Gv_AMG(stash))
4191 CROAK(("Cannot restore overloading on %s(0x%"UVxf") (package %s)",
4192 sv_reftype(sv, FALSE),
4194 stash ? HvNAME(stash) : "<unknown>"));
4198 TRACEME(("ok (retrieve_overloaded at 0x%"UVxf")", PTR2UV(rv)));
4204 * retrieve_tied_array
4206 * Retrieve tied array
4207 * Layout is SX_TIED_ARRAY <object>, with SX_TIED_ARRAY already read.
4209 static SV *retrieve_tied_array(stcxt_t *cxt, char *cname)
4214 TRACEME(("retrieve_tied_array (#%d)", cxt->tagnum));
4216 tv = NEWSV(10002, 0);
4217 SEEN(tv, cname); /* Will return if tv is null */
4218 sv = retrieve(cxt, 0); /* Retrieve <object> */
4220 return (SV *) 0; /* Failed */
4222 sv_upgrade(tv, SVt_PVAV);
4223 AvREAL_off((AV *)tv);
4224 sv_magic(tv, sv, 'P', Nullch, 0);
4225 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4227 TRACEME(("ok (retrieve_tied_array at 0x%"UVxf")", PTR2UV(tv)));
4233 * retrieve_tied_hash
4235 * Retrieve tied hash
4236 * Layout is SX_TIED_HASH <object>, with SX_TIED_HASH already read.
4238 static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname)
4243 TRACEME(("retrieve_tied_hash (#%d)", cxt->tagnum));
4245 tv = NEWSV(10002, 0);
4246 SEEN(tv, cname); /* Will return if tv is null */
4247 sv = retrieve(cxt, 0); /* Retrieve <object> */
4249 return (SV *) 0; /* Failed */
4251 sv_upgrade(tv, SVt_PVHV);
4252 sv_magic(tv, sv, 'P', Nullch, 0);
4253 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4255 TRACEME(("ok (retrieve_tied_hash at 0x%"UVxf")", PTR2UV(tv)));
4261 * retrieve_tied_scalar
4263 * Retrieve tied scalar
4264 * Layout is SX_TIED_SCALAR <object>, with SX_TIED_SCALAR already read.
4266 static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname)
4269 SV *sv, *obj = NULL;
4271 TRACEME(("retrieve_tied_scalar (#%d)", cxt->tagnum));
4273 tv = NEWSV(10002, 0);
4274 SEEN(tv, cname); /* Will return if rv is null */
4275 sv = retrieve(cxt, 0); /* Retrieve <object> */
4277 return (SV *) 0; /* Failed */
4279 else if (SvTYPE(sv) != SVt_NULL) {
4283 sv_upgrade(tv, SVt_PVMG);
4284 sv_magic(tv, obj, 'q', Nullch, 0);
4287 /* Undo refcnt inc from sv_magic() */
4291 TRACEME(("ok (retrieve_tied_scalar at 0x%"UVxf")", PTR2UV(tv)));
4299 * Retrieve reference to value in a tied hash.
4300 * Layout is SX_TIED_KEY <object> <key>, with SX_TIED_KEY already read.
4302 static SV *retrieve_tied_key(stcxt_t *cxt, char *cname)
4308 TRACEME(("retrieve_tied_key (#%d)", cxt->tagnum));
4310 tv = NEWSV(10002, 0);
4311 SEEN(tv, cname); /* Will return if tv is null */
4312 sv = retrieve(cxt, 0); /* Retrieve <object> */
4314 return (SV *) 0; /* Failed */
4316 key = retrieve(cxt, 0); /* Retrieve <key> */
4318 return (SV *) 0; /* Failed */
4320 sv_upgrade(tv, SVt_PVMG);
4321 sv_magic(tv, sv, 'p', (char *)key, HEf_SVKEY);
4322 SvREFCNT_dec(key); /* Undo refcnt inc from sv_magic() */
4323 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4331 * Retrieve reference to value in a tied array.
4332 * Layout is SX_TIED_IDX <object> <idx>, with SX_TIED_IDX already read.
4334 static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname)
4340 TRACEME(("retrieve_tied_idx (#%d)", cxt->tagnum));
4342 tv = NEWSV(10002, 0);
4343 SEEN(tv, cname); /* Will return if tv is null */
4344 sv = retrieve(cxt, 0); /* Retrieve <object> */
4346 return (SV *) 0; /* Failed */
4348 RLEN(idx); /* Retrieve <idx> */
4350 sv_upgrade(tv, SVt_PVMG);
4351 sv_magic(tv, sv, 'p', Nullch, idx);
4352 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4361 * Retrieve defined long (string) scalar.
4363 * Layout is SX_LSCALAR <length> <data>, with SX_LSCALAR already read.
4364 * The scalar is "long" in that <length> is larger than LG_SCALAR so it
4365 * was not stored on a single byte.
4367 static SV *retrieve_lscalar(stcxt_t *cxt, char *cname)
4373 TRACEME(("retrieve_lscalar (#%d), len = %"IVdf, cxt->tagnum, (IV) len));
4376 * Allocate an empty scalar of the suitable length.
4379 sv = NEWSV(10002, len);
4380 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4383 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
4385 * Now, for efficiency reasons, read data directly inside the SV buffer,
4386 * and perform the SV final settings directly by duplicating the final
4387 * work done by sv_setpv. Since we're going to allocate lots of scalars
4388 * this way, it's worth the hassle and risk.
4391 SAFEREAD(SvPVX(sv), len, sv);
4392 SvCUR_set(sv, len); /* Record C string length */
4393 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
4394 (void) SvPOK_only(sv); /* Validate string pointer */
4395 if (cxt->s_tainted) /* Is input source tainted? */
4396 SvTAINT(sv); /* External data cannot be trusted */
4398 TRACEME(("large scalar len %"IVdf" '%s'", (IV) len, SvPVX(sv)));
4399 TRACEME(("ok (retrieve_lscalar at 0x%"UVxf")", PTR2UV(sv)));
4407 * Retrieve defined short (string) scalar.
4409 * Layout is SX_SCALAR <length> <data>, with SX_SCALAR already read.
4410 * The scalar is "short" so <length> is single byte. If it is 0, there
4411 * is no <data> section.
4413 static SV *retrieve_scalar(stcxt_t *cxt, char *cname)
4419 TRACEME(("retrieve_scalar (#%d), len = %d", cxt->tagnum, len));
4422 * Allocate an empty scalar of the suitable length.
4425 sv = NEWSV(10002, len);
4426 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4429 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
4434 * newSV did not upgrade to SVt_PV so the scalar is undefined.
4435 * To make it defined with an empty length, upgrade it now...
4436 * Don't upgrade to a PV if the original type contains more
4437 * information than a scalar.
4439 if (SvTYPE(sv) <= SVt_PV) {
4440 sv_upgrade(sv, SVt_PV);
4443 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
4444 TRACEME(("ok (retrieve_scalar empty at 0x%"UVxf")", PTR2UV(sv)));
4447 * Now, for efficiency reasons, read data directly inside the SV buffer,
4448 * and perform the SV final settings directly by duplicating the final
4449 * work done by sv_setpv. Since we're going to allocate lots of scalars
4450 * this way, it's worth the hassle and risk.
4452 SAFEREAD(SvPVX(sv), len, sv);
4453 SvCUR_set(sv, len); /* Record C string length */
4454 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
4455 TRACEME(("small scalar len %d '%s'", len, SvPVX(sv)));
4458 (void) SvPOK_only(sv); /* Validate string pointer */
4459 if (cxt->s_tainted) /* Is input source tainted? */
4460 SvTAINT(sv); /* External data cannot be trusted */
4462 TRACEME(("ok (retrieve_scalar at 0x%"UVxf")", PTR2UV(sv)));
4469 * Like retrieve_scalar(), but tag result as utf8.
4470 * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
4472 static SV *retrieve_utf8str(stcxt_t *cxt, char *cname)
4476 TRACEME(("retrieve_utf8str"));
4478 sv = retrieve_scalar(cxt, cname);
4480 #ifdef HAS_UTF8_SCALARS
4483 if (cxt->use_bytes < 0)
4485 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4487 if (cxt->use_bytes == 0)
4498 * Like retrieve_lscalar(), but tag result as utf8.
4499 * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
4501 static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname)
4505 TRACEME(("retrieve_lutf8str"));
4507 sv = retrieve_lscalar(cxt, cname);
4509 #ifdef HAS_UTF8_SCALARS
4512 if (cxt->use_bytes < 0)
4514 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4516 if (cxt->use_bytes == 0)
4526 * Retrieve defined integer.
4527 * Layout is SX_INTEGER <data>, whith SX_INTEGER already read.
4529 static SV *retrieve_integer(stcxt_t *cxt, char *cname)
4534 TRACEME(("retrieve_integer (#%d)", cxt->tagnum));
4536 READ(&iv, sizeof(iv));
4538 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4540 TRACEME(("integer %"IVdf, iv));
4541 TRACEME(("ok (retrieve_integer at 0x%"UVxf")", PTR2UV(sv)));
4549 * Retrieve defined integer in network order.
4550 * Layout is SX_NETINT <data>, whith SX_NETINT already read.
4552 static SV *retrieve_netint(stcxt_t *cxt, char *cname)
4557 TRACEME(("retrieve_netint (#%d)", cxt->tagnum));
4561 sv = newSViv((int) ntohl(iv));
4562 TRACEME(("network integer %d", (int) ntohl(iv)));
4565 TRACEME(("network integer (as-is) %d", iv));
4567 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4569 TRACEME(("ok (retrieve_netint at 0x%"UVxf")", PTR2UV(sv)));
4577 * Retrieve defined double.
4578 * Layout is SX_DOUBLE <data>, whith SX_DOUBLE already read.
4580 static SV *retrieve_double(stcxt_t *cxt, char *cname)
4585 TRACEME(("retrieve_double (#%d)", cxt->tagnum));
4587 READ(&nv, sizeof(nv));
4589 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4591 TRACEME(("double %"NVff, nv));
4592 TRACEME(("ok (retrieve_double at 0x%"UVxf")", PTR2UV(sv)));
4600 * Retrieve defined byte (small integer within the [-128, +127] range).
4601 * Layout is SX_BYTE <data>, whith SX_BYTE already read.
4603 static SV *retrieve_byte(stcxt_t *cxt, char *cname)
4607 signed char tmp; /* Workaround for AIX cc bug --H.Merijn Brand */
4609 TRACEME(("retrieve_byte (#%d)", cxt->tagnum));
4612 TRACEME(("small integer read as %d", (unsigned char) siv));
4613 tmp = (unsigned char) siv - 128;
4615 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4617 TRACEME(("byte %d", tmp));
4618 TRACEME(("ok (retrieve_byte at 0x%"UVxf")", PTR2UV(sv)));
4626 * Return the undefined value.
4628 static SV *retrieve_undef(stcxt_t *cxt, char *cname)
4632 TRACEME(("retrieve_undef"));
4643 * Return the immortal undefined value.
4645 static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname)
4647 SV *sv = &PL_sv_undef;
4649 TRACEME(("retrieve_sv_undef"));
4658 * Return the immortal yes value.
4660 static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname)
4662 SV *sv = &PL_sv_yes;
4664 TRACEME(("retrieve_sv_yes"));
4673 * Return the immortal no value.
4675 static SV *retrieve_sv_no(stcxt_t *cxt, char *cname)
4679 TRACEME(("retrieve_sv_no"));
4688 * Retrieve a whole array.
4689 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
4690 * Each item is stored as <object>.
4692 * When we come here, SX_ARRAY has been read already.
4694 static SV *retrieve_array(stcxt_t *cxt, char *cname)
4701 TRACEME(("retrieve_array (#%d)", cxt->tagnum));
4704 * Read length, and allocate array, then pre-extend it.
4708 TRACEME(("size = %d", len));
4710 SEEN(av, cname); /* Will return if array not allocated nicely */
4714 return (SV *) av; /* No data follow if array is empty */
4717 * Now get each item in turn...
4720 for (i = 0; i < len; i++) {
4721 TRACEME(("(#%d) item", i));
4722 sv = retrieve(cxt, 0); /* Retrieve item */
4725 if (av_store(av, i, sv) == 0)
4729 TRACEME(("ok (retrieve_array at 0x%"UVxf")", PTR2UV(av)));
4737 * Retrieve a whole hash table.
4738 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4739 * Keys are stored as <length> <data>, the <data> section being omitted
4741 * Values are stored as <object>.
4743 * When we come here, SX_HASH has been read already.
4745 static SV *retrieve_hash(stcxt_t *cxt, char *cname)
4753 TRACEME(("retrieve_hash (#%d)", cxt->tagnum));
4756 * Read length, allocate table.
4760 TRACEME(("size = %d", len));
4762 SEEN(hv, cname); /* Will return if table not allocated properly */
4764 return (SV *) hv; /* No data follow if table empty */
4765 hv_ksplit(hv, len); /* pre-extend hash to save multiple splits */
4768 * Now get each key/value pair in turn...
4771 for (i = 0; i < len; i++) {
4776 TRACEME(("(#%d) value", i));
4777 sv = retrieve(cxt, 0);
4783 * Since we're reading into kbuf, we must ensure we're not
4784 * recursing between the read and the hv_store() where it's used.
4785 * Hence the key comes after the value.
4788 RLEN(size); /* Get key size */
4789 KBUFCHK((STRLEN)size); /* Grow hash key read pool if needed */
4792 kbuf[size] = '\0'; /* Mark string end, just in case */
4793 TRACEME(("(#%d) key '%s'", i, kbuf));
4796 * Enter key/value pair into hash table.
4799 if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
4803 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4811 * Retrieve a whole hash table.
4812 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4813 * Keys are stored as <length> <data>, the <data> section being omitted
4815 * Values are stored as <object>.
4817 * When we come here, SX_HASH has been read already.
4819 static SV *retrieve_flag_hash(stcxt_t *cxt, char *cname)
4828 GETMARK(hash_flags);
4829 TRACEME(("retrieve_flag_hash (#%d)", cxt->tagnum));
4831 * Read length, allocate table.
4834 #ifndef HAS_RESTRICTED_HASHES
4835 if (hash_flags & SHV_RESTRICTED) {
4836 if (cxt->derestrict < 0)
4838 = (SvTRUE(perl_get_sv("Storable::downgrade_restricted", TRUE))
4840 if (cxt->derestrict == 0)
4841 RESTRICTED_HASH_CROAK();
4846 TRACEME(("size = %d, flags = %d", len, hash_flags));
4848 SEEN(hv, cname); /* Will return if table not allocated properly */
4850 return (SV *) hv; /* No data follow if table empty */
4851 hv_ksplit(hv, len); /* pre-extend hash to save multiple splits */
4854 * Now get each key/value pair in turn...
4857 for (i = 0; i < len; i++) {
4859 int store_flags = 0;
4864 TRACEME(("(#%d) value", i));
4865 sv = retrieve(cxt, 0);
4870 #ifdef HAS_RESTRICTED_HASHES
4871 if ((hash_flags & SHV_RESTRICTED) && (flags & SHV_K_LOCKED))
4875 if (flags & SHV_K_ISSV) {
4876 /* XXX you can't set a placeholder with an SV key.
4877 Then again, you can't get an SV key.
4878 Without messing around beyond what the API is supposed to do.
4881 TRACEME(("(#%d) keysv, flags=%d", i, flags));
4882 keysv = retrieve(cxt, 0);
4886 if (!hv_store_ent(hv, keysv, sv, 0))
4891 * Since we're reading into kbuf, we must ensure we're not
4892 * recursing between the read and the hv_store() where it's used.
4893 * Hence the key comes after the value.
4896 if (flags & SHV_K_PLACEHOLDER) {
4899 store_flags |= HVhek_PLACEHOLD;
4901 if (flags & SHV_K_UTF8) {
4902 #ifdef HAS_UTF8_HASHES
4903 store_flags |= HVhek_UTF8;
4905 if (cxt->use_bytes < 0)
4907 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4909 if (cxt->use_bytes == 0)
4913 #ifdef HAS_UTF8_HASHES
4914 if (flags & SHV_K_WASUTF8)
4915 store_flags |= HVhek_WASUTF8;
4918 RLEN(size); /* Get key size */
4919 KBUFCHK((STRLEN)size); /* Grow hash key read pool if needed */
4922 kbuf[size] = '\0'; /* Mark string end, just in case */
4923 TRACEME(("(#%d) key '%s' flags %X store_flags %X", i, kbuf,
4924 flags, store_flags));
4927 * Enter key/value pair into hash table.
4930 #ifdef HAS_RESTRICTED_HASHES
4931 if (hv_store_flags(hv, kbuf, size, sv, 0, flags) == 0)
4934 if (!(store_flags & HVhek_PLACEHOLD))
4935 if (hv_store(hv, kbuf, size, sv, 0) == 0)
4940 #ifdef HAS_RESTRICTED_HASHES
4941 if (hash_flags & SHV_RESTRICTED)
4945 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4953 * Return a code reference.
4955 static SV *retrieve_code(stcxt_t *cxt, char *cname)
4957 #if PERL_VERSION < 6
4958 CROAK(("retrieve_code does not work with perl 5.005 or less\n"));
4963 SV *sv, *text, *sub, *errsv;
4965 TRACEME(("retrieve_code (#%d)", cxt->tagnum));
4968 * Retrieve the source of the code reference
4969 * as a small or large scalar
4975 text = retrieve_scalar(cxt, cname);
4978 text = retrieve_lscalar(cxt, cname);
4981 CROAK(("Unexpected type %d in retrieve_code\n", type));
4985 * prepend "sub " to the source
4988 sub = newSVpvn("sub ", 4);
4989 sv_catpv(sub, SvPV_nolen(text)); /* XXX no sv_catsv! */
4993 * evaluate the source to a code reference and use the CV value
4996 if (cxt->eval == NULL) {
4997 cxt->eval = perl_get_sv("Storable::Eval", TRUE);
4998 SvREFCNT_inc(cxt->eval);
5000 if (!SvTRUE(cxt->eval)) {
5002 cxt->forgive_me == 0 ||
5003 (cxt->forgive_me < 0 && !(cxt->forgive_me =
5004 SvTRUE(perl_get_sv("Storable::forgive_me", TRUE)) ? 1 : 0))
5006 CROAK(("Can't eval, please set $Storable::Eval to a true value"));
5016 if (SvROK(cxt->eval) && SvTYPE(SvRV(cxt->eval)) == SVt_PVCV) {
5017 SV* errsv = get_sv("@", TRUE);
5018 sv_setpv(errsv, ""); /* clear $@ */
5020 XPUSHs(sv_2mortal(newSVsv(sub)));
5022 count = call_sv(cxt->eval, G_SCALAR);
5025 CROAK(("Unexpected return value from $Storable::Eval callback\n"));
5027 if (SvTRUE(errsv)) {
5028 CROAK(("code %s caused an error: %s",
5029 SvPV_nolen(sub), SvPV_nolen(errsv)));
5033 cv = eval_pv(SvPV_nolen(sub), TRUE);
5035 if (cv && SvROK(cv) && SvTYPE(SvRV(cv)) == SVt_PVCV) {
5038 CROAK(("code %s did not evaluate to a subroutine reference\n", SvPV_nolen(sub)));
5041 SvREFCNT_inc(sv); /* XXX seems to be necessary */
5053 * old_retrieve_array
5055 * Retrieve a whole array in pre-0.6 binary format.
5057 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
5058 * Each item is stored as SX_ITEM <object> or SX_IT_UNDEF for "holes".
5060 * When we come here, SX_ARRAY has been read already.
5062 static SV *old_retrieve_array(stcxt_t *cxt, char *cname)
5070 TRACEME(("old_retrieve_array (#%d)", cxt->tagnum));
5073 * Read length, and allocate array, then pre-extend it.
5077 TRACEME(("size = %d", len));
5079 SEEN(av, 0); /* Will return if array not allocated nicely */
5083 return (SV *) av; /* No data follow if array is empty */
5086 * Now get each item in turn...
5089 for (i = 0; i < len; i++) {
5091 if (c == SX_IT_UNDEF) {
5092 TRACEME(("(#%d) undef item", i));
5093 continue; /* av_extend() already filled us with undef */
5096 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
5097 TRACEME(("(#%d) item", i));
5098 sv = retrieve(cxt, 0); /* Retrieve item */
5101 if (av_store(av, i, sv) == 0)
5105 TRACEME(("ok (old_retrieve_array at 0x%"UVxf")", PTR2UV(av)));
5113 * Retrieve a whole hash table in pre-0.6 binary format.
5115 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
5116 * Keys are stored as SX_KEY <length> <data>, the <data> section being omitted
5118 * Values are stored as SX_VALUE <object> or SX_VL_UNDEF for "holes".
5120 * When we come here, SX_HASH has been read already.
5122 static SV *old_retrieve_hash(stcxt_t *cxt, char *cname)
5130 static SV *sv_h_undef = (SV *) 0; /* hv_store() bug */
5132 TRACEME(("old_retrieve_hash (#%d)", cxt->tagnum));
5135 * Read length, allocate table.
5139 TRACEME(("size = %d", len));
5141 SEEN(hv, 0); /* Will return if table not allocated properly */
5143 return (SV *) hv; /* No data follow if table empty */
5144 hv_ksplit(hv, len); /* pre-extend hash to save multiple splits */
5147 * Now get each key/value pair in turn...
5150 for (i = 0; i < len; i++) {
5156 if (c == SX_VL_UNDEF) {
5157 TRACEME(("(#%d) undef value", i));
5159 * Due to a bug in hv_store(), it's not possible to pass
5160 * &PL_sv_undef to hv_store() as a value, otherwise the
5161 * associated key will not be creatable any more. -- RAM, 14/01/97
5164 sv_h_undef = newSVsv(&PL_sv_undef);
5165 sv = SvREFCNT_inc(sv_h_undef);
5166 } else if (c == SX_VALUE) {
5167 TRACEME(("(#%d) value", i));
5168 sv = retrieve(cxt, 0);
5172 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
5176 * Since we're reading into kbuf, we must ensure we're not
5177 * recursing between the read and the hv_store() where it's used.
5178 * Hence the key comes after the value.
5183 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
5184 RLEN(size); /* Get key size */
5185 KBUFCHK((STRLEN)size); /* Grow hash key read pool if needed */
5188 kbuf[size] = '\0'; /* Mark string end, just in case */
5189 TRACEME(("(#%d) key '%s'", i, kbuf));
5192 * Enter key/value pair into hash table.
5195 if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
5199 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
5205 *** Retrieval engine.
5211 * Make sure the stored data we're trying to retrieve has been produced
5212 * on an ILP compatible system with the same byteorder. It croaks out in
5213 * case an error is detected. [ILP = integer-long-pointer sizes]
5214 * Returns null if error is detected, &PL_sv_undef otherwise.
5216 * Note that there's no byte ordering info emitted when network order was
5217 * used at store time.
5219 static SV *magic_check(stcxt_t *cxt)
5221 /* The worst case for a malicious header would be old magic (which is
5222 longer), major, minor, byteorder length byte of 255, 255 bytes of
5223 garbage, sizeof int, long, pointer, NV.
5224 So the worse of that we can read is 255 bytes of garbage plus 4.
5225 Err, I am assuming 8 bit bytes here. Please file a bug report if you're
5226 compiling perl on a system with chars that are larger than 8 bits.
5227 (Even Crays aren't *that* perverse).
5229 unsigned char buf[4 + 255];
5230 unsigned char *current;
5233 int use_network_order;
5236 int version_minor = 0;
5238 TRACEME(("magic_check"));
5241 * The "magic number" is only for files, not when freezing in memory.
5245 /* This includes the '\0' at the end. I want to read the extra byte,
5246 which is usually going to be the major version number. */
5247 STRLEN len = sizeof(magicstr);
5250 READ(buf, (SSize_t)(len)); /* Not null-terminated */
5252 /* Point at the byte after the byte we read. */
5253 current = buf + --len; /* Do the -- outside of macros. */
5255 if (memNE(buf, magicstr, len)) {
5257 * Try to read more bytes to check for the old magic number, which
5261 TRACEME(("trying for old magic number"));
5263 old_len = sizeof(old_magicstr) - 1;
5264 READ(current + 1, (SSize_t)(old_len - len));
5266 if (memNE(buf, old_magicstr, old_len))
5267 CROAK(("File is not a perl storable"));
5268 current = buf + old_len;
5270 use_network_order = *current;
5272 GETMARK(use_network_order);
5275 * Starting with 0.6, the "use_network_order" byte flag is also used to
5276 * indicate the version number of the binary, and therefore governs the
5277 * setting of sv_retrieve_vtbl. See magic_write().
5280 version_major = use_network_order >> 1;
5281 cxt->retrieve_vtbl = version_major ? sv_retrieve : sv_old_retrieve;
5283 TRACEME(("magic_check: netorder = 0x%x", use_network_order));
5287 * Starting with 0.7 (binary major 2), a full byte is dedicated to the
5288 * minor version of the protocol. See magic_write().
5291 if (version_major > 1)
5292 GETMARK(version_minor);
5294 cxt->ver_major = version_major;
5295 cxt->ver_minor = version_minor;
5297 TRACEME(("binary image version is %d.%d", version_major, version_minor));
5300 * Inter-operability sanity check: we can't retrieve something stored
5301 * using a format more recent than ours, because we have no way to
5302 * know what has changed, and letting retrieval go would mean a probable
5303 * failure reporting a "corrupted" storable file.
5307 version_major > STORABLE_BIN_MAJOR ||
5308 (version_major == STORABLE_BIN_MAJOR &&
5309 version_minor > STORABLE_BIN_MINOR)
5312 TRACEME(("but I am version is %d.%d", STORABLE_BIN_MAJOR,
5313 STORABLE_BIN_MINOR));
5315 if (version_major == STORABLE_BIN_MAJOR) {
5316 TRACEME(("cxt->accept_future_minor is %d",
5317 cxt->accept_future_minor));
5318 if (cxt->accept_future_minor < 0)
5319 cxt->accept_future_minor
5320 = (SvTRUE(perl_get_sv("Storable::accept_future_minor",
5323 if (cxt->accept_future_minor == 1)
5324 croak_now = 0; /* Don't croak yet. */
5327 CROAK(("Storable binary image v%d.%d more recent than I am (v%d.%d)",
5328 version_major, version_minor,
5329 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
5334 * If they stored using network order, there's no byte ordering
5335 * information to check.
5338 if ((cxt->netorder = (use_network_order & 0x1))) /* Extra () for -Wall */
5339 return &PL_sv_undef; /* No byte ordering info */
5341 /* In C truth is 1, falsehood is 0. Very convienient. */
5342 use_NV_size = version_major >= 2 && version_minor >= 2;
5345 length = c + 3 + use_NV_size;
5346 READ(buf, length); /* Not null-terminated */
5348 TRACEME(("byte order '%.*s' %d", c, buf, c));
5350 #ifdef USE_56_INTERWORK_KLUDGE
5351 /* No point in caching this in the context as we only need it once per
5352 retrieve, and we need to recheck it each read. */
5353 if (SvTRUE(perl_get_sv("Storable::interwork_56_64bit", TRUE))) {
5354 if ((c != (sizeof (byteorderstr_56) - 1))
5355 || memNE(buf, byteorderstr_56, c))
5356 CROAK(("Byte order is not compatible"));
5360 if ((c != (sizeof (byteorderstr) - 1)) || memNE(buf, byteorderstr, c))
5361 CROAK(("Byte order is not compatible"));
5367 if ((int) *current++ != sizeof(int))
5368 CROAK(("Integer size is not compatible"));
5371 if ((int) *current++ != sizeof(long))
5372 CROAK(("Long integer size is not compatible"));
5374 /* sizeof(char *) */
5375 if ((int) *current != sizeof(char *))
5376 CROAK(("Pointer size is not compatible"));
5380 if ((int) *++current != sizeof(NV))
5381 CROAK(("Double size is not compatible"));
5384 return &PL_sv_undef; /* OK */
5390 * Recursively retrieve objects from the specified file and return their
5391 * root SV (which may be an AV or an HV for what we care).
5392 * Returns null if there is a problem.
5394 static SV *retrieve(stcxt_t *cxt, char *cname)
5400 TRACEME(("retrieve"));
5403 * Grab address tag which identifies the object if we are retrieving
5404 * an older format. Since the new binary format counts objects and no
5405 * longer explicitely tags them, we must keep track of the correspondance
5408 * The following section will disappear one day when the old format is
5409 * no longer supported, hence the final "goto" in the "if" block.
5412 if (cxt->hseen) { /* Retrieving old binary */
5414 if (cxt->netorder) {
5416 READ(&nettag, sizeof(I32)); /* Ordered sequence of I32 */
5417 tag = (stag_t) nettag;
5419 READ(&tag, sizeof(stag_t)); /* Original address of the SV */
5422 if (type == SX_OBJECT) {
5424 svh = hv_fetch(cxt->hseen, (char *) &tag, sizeof(tag), FALSE);
5426 CROAK(("Old tag 0x%"UVxf" should have been mapped already",
5428 tagn = SvIV(*svh); /* Mapped tag number computed earlier below */
5431 * The following code is common with the SX_OBJECT case below.
5434 svh = av_fetch(cxt->aseen, tagn, FALSE);
5436 CROAK(("Object #%"IVdf" should have been retrieved already",
5439 TRACEME(("has retrieved #%d at 0x%"UVxf, tagn, PTR2UV(sv)));
5440 SvREFCNT_inc(sv); /* One more reference to this same sv */
5441 return sv; /* The SV pointer where object was retrieved */
5445 * Map new object, but don't increase tagnum. This will be done
5446 * by each of the retrieve_* functions when they call SEEN().
5448 * The mapping associates the "tag" initially present with a unique
5449 * tag number. See test for SX_OBJECT above to see how this is perused.
5452 if (!hv_store(cxt->hseen, (char *) &tag, sizeof(tag),
5453 newSViv(cxt->tagnum), 0))
5460 * Regular post-0.6 binary format.
5465 TRACEME(("retrieve type = %d", type));
5468 * Are we dealing with an object we should have already retrieved?
5471 if (type == SX_OBJECT) {
5475 svh = av_fetch(cxt->aseen, tag, FALSE);
5477 CROAK(("Object #%"IVdf" should have been retrieved already",
5480 TRACEME(("had retrieved #%d at 0x%"UVxf, tag, PTR2UV(sv)));
5481 SvREFCNT_inc(sv); /* One more reference to this same sv */
5482 return sv; /* The SV pointer where object was retrieved */
5483 } else if (type >= SX_ERROR && cxt->ver_minor > STORABLE_BIN_MINOR) {
5484 if (cxt->accept_future_minor < 0)
5485 cxt->accept_future_minor
5486 = (SvTRUE(perl_get_sv("Storable::accept_future_minor",
5489 if (cxt->accept_future_minor == 1) {
5490 CROAK(("Storable binary image v%d.%d contains data of type %d. "
5491 "This Storable is v%d.%d and can only handle data types up to %d",
5492 cxt->ver_major, cxt->ver_minor, type,
5493 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR, SX_ERROR - 1));
5497 first_time: /* Will disappear when support for old format is dropped */
5500 * Okay, first time through for this one.
5503 sv = RETRIEVE(cxt, type)(cxt, cname);
5505 return (SV *) 0; /* Failed */
5508 * Old binary formats (pre-0.7).
5510 * Final notifications, ended by SX_STORED may now follow.
5511 * Currently, the only pertinent notification to apply on the
5512 * freshly retrieved object is either:
5513 * SX_CLASS <char-len> <classname> for short classnames.
5514 * SX_LG_CLASS <int-len> <classname> for larger one (rare!).
5515 * Class name is then read into the key buffer pool used by
5516 * hash table key retrieval.
5519 if (cxt->ver_major < 2) {
5520 while ((type = GETCHAR()) != SX_STORED) {
5524 GETMARK(len); /* Length coded on a single char */
5526 case SX_LG_CLASS: /* Length coded on a regular integer */
5531 return (SV *) 0; /* Failed */
5533 KBUFCHK((STRLEN)len); /* Grow buffer as necessary */
5536 kbuf[len] = '\0'; /* Mark string end */
5541 TRACEME(("ok (retrieved 0x%"UVxf", refcnt=%d, %s)", PTR2UV(sv),
5542 SvREFCNT(sv) - 1, sv_reftype(sv, FALSE)));
5550 * Retrieve data held in file and return the root object.
5551 * Common routine for pretrieve and mretrieve.
5553 static SV *do_retrieve(
5560 int is_tainted; /* Is input source tainted? */
5561 int pre_06_fmt = 0; /* True with pre Storable 0.6 formats */
5563 TRACEME(("do_retrieve (optype = 0x%x)", optype));
5565 optype |= ST_RETRIEVE;
5568 * Sanity assertions for retrieve dispatch tables.
5571 ASSERT(sizeof(sv_old_retrieve) == sizeof(sv_retrieve),
5572 ("old and new retrieve dispatch table have same size"));
5573 ASSERT(sv_old_retrieve[SX_ERROR] == retrieve_other,
5574 ("SX_ERROR entry correctly initialized in old dispatch table"));
5575 ASSERT(sv_retrieve[SX_ERROR] == retrieve_other,
5576 ("SX_ERROR entry correctly initialized in new dispatch table"));
5579 * Workaround for CROAK leak: if they enter with a "dirty" context,
5580 * free up memory for them now.
5587 * Now that STORABLE_xxx hooks exist, it is possible that they try to
5588 * re-enter retrieve() via the hooks.
5592 cxt = allocate_context(cxt);
5596 ASSERT(cxt->entry == 1, ("starting new recursion"));
5597 ASSERT(!cxt->s_dirty, ("clean context"));
5602 * Data is loaded into the memory buffer when f is NULL, unless `in' is
5603 * also NULL, in which case we're expecting the data to already lie
5604 * in the buffer (dclone case).
5607 KBUFINIT(); /* Allocate hash key reading pool once */
5610 MBUF_SAVE_AND_LOAD(in);
5613 * Magic number verifications.
5615 * This needs to be done before calling init_retrieve_context()
5616 * since the format indication in the file are necessary to conduct
5617 * some of the initializations.
5620 cxt->fio = f; /* Where I/O are performed */
5622 if (!magic_check(cxt))
5623 CROAK(("Magic number checking on storable %s failed",
5624 cxt->fio ? "file" : "string"));
5626 TRACEME(("data stored in %s format",
5627 cxt->netorder ? "net order" : "native"));
5630 * Check whether input source is tainted, so that we don't wrongly
5631 * taint perfectly good values...
5633 * We assume file input is always tainted. If both `f' and `in' are
5634 * NULL, then we come from dclone, and tainted is already filled in
5635 * the context. That's a kludge, but the whole dclone() thing is
5636 * already quite a kludge anyway! -- RAM, 15/09/2000.
5639 is_tainted = f ? 1 : (in ? SvTAINTED(in) : cxt->s_tainted);
5640 TRACEME(("input source is %s", is_tainted ? "tainted" : "trusted"));
5641 init_retrieve_context(cxt, optype, is_tainted);
5643 ASSERT(is_retrieving(), ("within retrieve operation"));
5645 sv = retrieve(cxt, 0); /* Recursively retrieve object, get root SV */
5654 pre_06_fmt = cxt->hseen != NULL; /* Before we clean context */
5657 * The "root" context is never freed.
5660 clean_retrieve_context(cxt);
5661 if (cxt->prev) /* This context was stacked */
5662 free_context(cxt); /* It was not the "root" context */
5665 * Prepare returned value.
5669 TRACEME(("retrieve ERROR"));
5670 #if (PATCHLEVEL <= 4)
5671 /* perl 5.00405 seems to screw up at this point with an
5672 'attempt to modify a read only value' error reported in the
5673 eval { $self = pretrieve(*FILE) } in _retrieve.
5674 I can't see what the cause of this error is, but I suspect a
5675 bug in 5.004, as it seems to be capable of issuing spurious
5676 errors or core dumping with matches on $@. I'm not going to
5677 spend time on what could be a fruitless search for the cause,
5678 so here's a bodge. If you're running 5.004 and don't like
5679 this inefficiency, either upgrade to a newer perl, or you are
5680 welcome to find the problem and send in a patch.
5684 return &PL_sv_undef; /* Something went wrong, return undef */
5688 TRACEME(("retrieve got %s(0x%"UVxf")",
5689 sv_reftype(sv, FALSE), PTR2UV(sv)));
5692 * Backward compatibility with Storable-0.5@9 (which we know we
5693 * are retrieving if hseen is non-null): don't create an extra RV
5694 * for objects since we special-cased it at store time.
5696 * Build a reference to the SV returned by pretrieve even if it is
5697 * already one and not a scalar, for consistency reasons.
5700 if (pre_06_fmt) { /* Was not handling overloading by then */
5702 TRACEME(("fixing for old formats -- pre 0.6"));
5703 if (sv_type(sv) == svis_REF && (rv = SvRV(sv)) && SvOBJECT(rv)) {
5704 TRACEME(("ended do_retrieve() with an object -- pre 0.6"));
5710 * If reference is overloaded, restore behaviour.
5712 * NB: minor glitch here: normally, overloaded refs are stored specially
5713 * so that we can croak when behaviour cannot be re-installed, and also
5714 * avoid testing for overloading magic at each reference retrieval.
5716 * Unfortunately, the root reference is implicitely stored, so we must
5717 * check for possible overloading now. Furthermore, if we don't restore
5718 * overloading, we cannot croak as if the original ref was, because we
5719 * have no way to determine whether it was an overloaded ref or not in
5722 * It's a pity that overloading magic is attached to the rv, and not to
5723 * the underlying sv as blessing is.
5727 HV *stash = (HV *) SvSTASH(sv);
5728 SV *rv = newRV_noinc(sv);
5729 if (stash && Gv_AMG(stash)) {
5731 TRACEME(("restored overloading on root reference"));
5733 TRACEME(("ended do_retrieve() with an object"));
5737 TRACEME(("regular do_retrieve() end"));
5739 return newRV_noinc(sv);
5745 * Retrieve data held in file and return the root object, undef on error.
5747 SV *pretrieve(PerlIO *f)
5749 TRACEME(("pretrieve"));
5750 return do_retrieve(f, Nullsv, 0);
5756 * Retrieve data held in scalar and return the root object, undef on error.
5758 SV *mretrieve(SV *sv)
5760 TRACEME(("mretrieve"));
5761 return do_retrieve((PerlIO*) 0, sv, 0);
5771 * Deep clone: returns a fresh copy of the original referenced SV tree.
5773 * This is achieved by storing the object in memory and restoring from
5774 * there. Not that efficient, but it should be faster than doing it from
5781 stcxt_t *real_context;
5784 TRACEME(("dclone"));
5787 * Workaround for CROAK leak: if they enter with a "dirty" context,
5788 * free up memory for them now.
5795 * do_store() optimizes for dclone by not freeing its context, should
5796 * we need to allocate one because we're deep cloning from a hook.
5799 if (!do_store((PerlIO*) 0, sv, ST_CLONE, FALSE, (SV**) 0))
5800 return &PL_sv_undef; /* Error during store */
5803 * Because of the above optimization, we have to refresh the context,
5804 * since a new one could have been allocated and stacked by do_store().
5807 { dSTCXT; real_context = cxt; } /* Sub-block needed for macro */
5808 cxt = real_context; /* And we need this temporary... */
5811 * Now, `cxt' may refer to a new context.
5814 ASSERT(!cxt->s_dirty, ("clean context"));
5815 ASSERT(!cxt->entry, ("entry will not cause new context allocation"));
5818 TRACEME(("dclone stored %d bytes", size));
5822 * Since we're passing do_retrieve() both a NULL file and sv, we need
5823 * to pre-compute the taintedness of the input by setting cxt->tainted
5824 * to whatever state our own input string was. -- RAM, 15/09/2000
5826 * do_retrieve() will free non-root context.
5829 cxt->s_tainted = SvTAINTED(sv);
5830 out = do_retrieve((PerlIO*) 0, Nullsv, ST_CLONE);
5832 TRACEME(("dclone returns 0x%"UVxf, PTR2UV(out)));
5842 * The Perl IO GV object distinguishes between input and output for sockets
5843 * but not for plain files. To allow Storable to transparently work on
5844 * plain files and sockets transparently, we have to ask xsubpp to fetch the
5845 * right object for us. Hence the OutputStream and InputStream declarations.
5847 * Before perl 5.004_05, those entries in the standard typemap are not
5848 * defined in perl include files, so we do that here.
5851 #ifndef OutputStream
5852 #define OutputStream PerlIO *
5853 #define InputStream PerlIO *
5854 #endif /* !OutputStream */
5856 MODULE = Storable PACKAGE = Storable::Cxt
5862 stcxt_t *cxt = (stcxt_t *)SvPVX(SvRV(self));
5866 if (!cxt->membuf_ro && mbase)
5868 if (cxt->membuf_ro && (cxt->msaved).arena)
5869 Safefree((cxt->msaved).arena);
5872 MODULE = Storable PACKAGE = Storable
5878 gv_fetchpv("Storable::drop_utf8", GV_ADDMULTI, SVt_PV);
5880 /* Only disable the used only once warning if we are in debugging mode. */
5881 gv_fetchpv("Storable::DEBUGME", GV_ADDMULTI, SVt_PV);
5883 #ifdef USE_56_INTERWORK_KLUDGE
5884 gv_fetchpv("Storable::interwork_56_64bit", GV_ADDMULTI, SVt_PV);
5918 last_op_in_netorder()