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); \
473 msiz = (STRLEN)MGROW; \
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_placeholder;
1244 hv_iterinit(cxt->hclass);
1245 while ((he = hv_iternext(cxt->hclass))) /* Extra () for -Wall, grr.. */
1246 HeVAL(he) = &PL_sv_placeholder;
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 = (((void*)cxt->retrieve_vtbl == (void*)sv_old_retrieve)
1328 cxt->aseen = newAV(); /* Where retrieved objects are kept */
1329 cxt->aclass = newAV(); /* Where seen classnames are kept */
1330 cxt->tagnum = 0; /* Have to count objects... */
1331 cxt->classnum = 0; /* ...and class names as well */
1332 cxt->optype = optype;
1333 cxt->s_tainted = is_tainted;
1334 cxt->entry = 1; /* No recursion yet */
1335 #ifndef HAS_RESTRICTED_HASHES
1336 cxt->derestrict = -1; /* Fetched from perl if needed */
1338 #ifndef HAS_UTF8_ALL
1339 cxt->use_bytes = -1; /* Fetched from perl if needed */
1341 cxt->accept_future_minor = -1; /* Fetched from perl if needed */
1345 * clean_retrieve_context
1347 * Clean retrieve context by
1349 static void clean_retrieve_context(stcxt_t *cxt)
1351 TRACEME(("clean_retrieve_context"));
1353 ASSERT(cxt->optype & ST_RETRIEVE, ("was performing a retrieve()"));
1356 AV *aseen = cxt->aseen;
1359 sv_free((SV *) aseen);
1363 AV *aclass = cxt->aclass;
1366 sv_free((SV *) aclass);
1370 HV *hook = cxt->hook;
1373 sv_free((SV *) hook);
1377 HV *hseen = cxt->hseen;
1380 sv_free((SV *) hseen); /* optional HV, for backward compat. */
1383 #ifndef HAS_RESTRICTED_HASHES
1384 cxt->derestrict = -1; /* Fetched from perl if needed */
1386 #ifndef HAS_UTF8_ALL
1387 cxt->use_bytes = -1; /* Fetched from perl if needed */
1389 cxt->accept_future_minor = -1; /* Fetched from perl if needed */
1397 * A workaround for the CROAK bug: cleanup the last context.
1399 static void clean_context(stcxt_t *cxt)
1401 TRACEME(("clean_context"));
1403 ASSERT(cxt->s_dirty, ("dirty context"));
1408 ASSERT(!cxt->membuf_ro, ("mbase is not read-only"));
1410 if (cxt->optype & ST_RETRIEVE)
1411 clean_retrieve_context(cxt);
1412 else if (cxt->optype & ST_STORE)
1413 clean_store_context(cxt);
1417 ASSERT(!cxt->s_dirty, ("context is clean"));
1418 ASSERT(cxt->entry == 0, ("context is reset"));
1424 * Allocate a new context and push it on top of the parent one.
1425 * This new context is made globally visible via SET_STCXT().
1427 static stcxt_t *allocate_context(parent_cxt)
1428 stcxt_t *parent_cxt;
1432 TRACEME(("allocate_context"));
1434 ASSERT(!parent_cxt->s_dirty, ("parent context clean"));
1436 NEW_STORABLE_CXT_OBJ(cxt);
1437 cxt->prev = parent_cxt->my_sv;
1440 ASSERT(!cxt->s_dirty, ("clean context"));
1448 * Free current context, which cannot be the "root" one.
1449 * Make the context underneath globally visible via SET_STCXT().
1451 static void free_context(cxt)
1454 stcxt_t *prev = (stcxt_t *)(cxt->prev ? SvPVX(SvRV(cxt->prev)) : 0);
1456 TRACEME(("free_context"));
1458 ASSERT(!cxt->s_dirty, ("clean context"));
1459 ASSERT(prev, ("not freeing root context"));
1461 SvREFCNT_dec(cxt->my_sv);
1464 ASSERT(cxt, ("context not void"));
1474 * Tells whether we're in the middle of a store operation.
1476 int is_storing(void)
1480 return cxt->entry && (cxt->optype & ST_STORE);
1486 * Tells whether we're in the middle of a retrieve operation.
1488 int is_retrieving(void)
1492 return cxt->entry && (cxt->optype & ST_RETRIEVE);
1496 * last_op_in_netorder
1498 * Returns whether last operation was made using network order.
1500 * This is typically out-of-band information that might prove useful
1501 * to people wishing to convert native to network order data when used.
1503 int last_op_in_netorder(void)
1507 return cxt->netorder;
1511 *** Hook lookup and calling routines.
1517 * A wrapper on gv_fetchmethod_autoload() which caches results.
1519 * Returns the routine reference as an SV*, or null if neither the package
1520 * nor its ancestors know about the method.
1522 static SV *pkg_fetchmeth(
1531 * The following code is the same as the one performed by UNIVERSAL::can
1535 gv = gv_fetchmethod_autoload(pkg, method, FALSE);
1536 if (gv && isGV(gv)) {
1537 sv = newRV((SV*) GvCV(gv));
1538 TRACEME(("%s->%s: 0x%"UVxf, HvNAME(pkg), method, PTR2UV(sv)));
1540 sv = newSVsv(&PL_sv_undef);
1541 TRACEME(("%s->%s: not found", HvNAME(pkg), method));
1545 * Cache the result, ignoring failure: if we can't store the value,
1546 * it just won't be cached.
1549 (void) hv_store(cache, HvNAME(pkg), strlen(HvNAME(pkg)), sv, 0);
1551 return SvOK(sv) ? sv : (SV *) 0;
1557 * Force cached value to be undef: hook ignored even if present.
1559 static void pkg_hide(
1564 (void) hv_store(cache,
1565 HvNAME(pkg), strlen(HvNAME(pkg)), newSVsv(&PL_sv_undef), 0);
1571 * Discard cached value: a whole fetch loop will be retried at next lookup.
1573 static void pkg_uncache(
1578 (void) hv_delete(cache, HvNAME(pkg), strlen(HvNAME(pkg)), G_DISCARD);
1584 * Our own "UNIVERSAL::can", which caches results.
1586 * Returns the routine reference as an SV*, or null if the object does not
1587 * know about the method.
1597 TRACEME(("pkg_can for %s->%s", HvNAME(pkg), method));
1600 * Look into the cache to see whether we already have determined
1601 * where the routine was, if any.
1603 * NOTA BENE: we don't use `method' at all in our lookup, since we know
1604 * that only one hook (i.e. always the same) is cached in a given cache.
1607 svh = hv_fetch(cache, HvNAME(pkg), strlen(HvNAME(pkg)), FALSE);
1611 TRACEME(("cached %s->%s: not found", HvNAME(pkg), method));
1614 TRACEME(("cached %s->%s: 0x%"UVxf,
1615 HvNAME(pkg), method, PTR2UV(sv)));
1620 TRACEME(("not cached yet"));
1621 return pkg_fetchmeth(cache, pkg, method); /* Fetch and cache */
1627 * Call routine as obj->hook(av) in scalar context.
1628 * Propagates the single returned value if not called in void context.
1630 static SV *scalar_call(
1641 TRACEME(("scalar_call (cloning=%d)", cloning));
1648 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1650 SV **ary = AvARRAY(av);
1651 int cnt = AvFILLp(av) + 1;
1653 XPUSHs(ary[0]); /* Frozen string */
1654 for (i = 1; i < cnt; i++) {
1655 TRACEME(("pushing arg #%d (0x%"UVxf")...",
1656 i, PTR2UV(ary[i])));
1657 XPUSHs(sv_2mortal(newRV(ary[i])));
1662 TRACEME(("calling..."));
1663 count = perl_call_sv(hook, flags); /* Go back to Perl code */
1664 TRACEME(("count = %d", count));
1670 SvREFCNT_inc(sv); /* We're returning it, must stay alive! */
1683 * Call routine obj->hook(cloning) in list context.
1684 * Returns the list of returned values in an array.
1686 static AV *array_call(
1696 TRACEME(("array_call (cloning=%d)", cloning));
1702 XPUSHs(obj); /* Target object */
1703 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1706 count = perl_call_sv(hook, G_ARRAY); /* Go back to Perl code */
1711 for (i = count - 1; i >= 0; i--) {
1713 av_store(av, i, SvREFCNT_inc(sv));
1726 * Lookup the class name in the `hclass' table and either assign it a new ID
1727 * or return the existing one, by filling in `classnum'.
1729 * Return true if the class was known, false if the ID was just generated.
1731 static int known_class(
1733 char *name, /* Class name */
1734 int len, /* Name length */
1738 HV *hclass = cxt->hclass;
1740 TRACEME(("known_class (%s)", name));
1743 * Recall that we don't store pointers in this hash table, but tags.
1744 * Therefore, we need LOW_32BITS() to extract the relevant parts.
1747 svh = hv_fetch(hclass, name, len, FALSE);
1749 *classnum = LOW_32BITS(*svh);
1754 * Unknown classname, we need to record it.
1758 if (!hv_store(hclass, name, len, INT2PTR(SV*, cxt->classnum), 0))
1759 CROAK(("Unable to record new classname"));
1761 *classnum = cxt->classnum;
1766 *** Sepcific store routines.
1772 * Store a reference.
1773 * Layout is SX_REF <object> or SX_OVERLOAD <object>.
1775 static int store_ref(stcxt_t *cxt, SV *sv)
1777 TRACEME(("store_ref (0x%"UVxf")", PTR2UV(sv)));
1780 * Follow reference, and check if target is overloaded.
1786 HV *stash = (HV *) SvSTASH(sv);
1787 if (stash && Gv_AMG(stash)) {
1788 TRACEME(("ref (0x%"UVxf") is overloaded", PTR2UV(sv)));
1789 PUTMARK(SX_OVERLOAD);
1795 return store(cxt, sv);
1803 * Layout is SX_LSCALAR <length> <data>, SX_SCALAR <length> <data> or SX_UNDEF.
1804 * The <data> section is omitted if <length> is 0.
1806 * If integer or double, the layout is SX_INTEGER <data> or SX_DOUBLE <data>.
1807 * Small integers (within [-127, +127]) are stored as SX_BYTE <byte>.
1809 static int store_scalar(stcxt_t *cxt, SV *sv)
1814 U32 flags = SvFLAGS(sv); /* "cc -O" may put it in register */
1816 TRACEME(("store_scalar (0x%"UVxf")", PTR2UV(sv)));
1819 * For efficiency, break the SV encapsulation by peaking at the flags
1820 * directly without using the Perl macros to avoid dereferencing
1821 * sv->sv_flags each time we wish to check the flags.
1824 if (!(flags & SVf_OK)) { /* !SvOK(sv) */
1825 if (sv == &PL_sv_undef) {
1826 TRACEME(("immortal undef"));
1827 PUTMARK(SX_SV_UNDEF);
1829 TRACEME(("undef at 0x%"UVxf, PTR2UV(sv)));
1836 * Always store the string representation of a scalar if it exists.
1837 * Gisle Aas provided me with this test case, better than a long speach:
1839 * perl -MDevel::Peek -le '$a="abc"; $a+0; Dump($a)'
1840 * SV = PVNV(0x80c8520)
1842 * FLAGS = (NOK,POK,pNOK,pPOK)
1845 * PV = 0x80c83d0 "abc"\0
1849 * Write SX_SCALAR, length, followed by the actual data.
1851 * Otherwise, write an SX_BYTE, SX_INTEGER or an SX_DOUBLE as
1852 * appropriate, followed by the actual (binary) data. A double
1853 * is written as a string if network order, for portability.
1855 * NOTE: instead of using SvNOK(sv), we test for SvNOKp(sv).
1856 * The reason is that when the scalar value is tainted, the SvNOK(sv)
1859 * The test for a read-only scalar with both POK and NOK set is meant
1860 * to quickly detect &PL_sv_yes and &PL_sv_no without having to pay the
1861 * address comparison for each scalar we store.
1864 #define SV_MAYBE_IMMORTAL (SVf_READONLY|SVf_POK|SVf_NOK)
1866 if ((flags & SV_MAYBE_IMMORTAL) == SV_MAYBE_IMMORTAL) {
1867 if (sv == &PL_sv_yes) {
1868 TRACEME(("immortal yes"));
1870 } else if (sv == &PL_sv_no) {
1871 TRACEME(("immortal no"));
1874 pv = SvPV(sv, len); /* We know it's SvPOK */
1875 goto string; /* Share code below */
1877 } else if (flags & SVf_POK) {
1878 /* public string - go direct to string read. */
1879 goto string_readlen;
1881 #if (PATCHLEVEL <= 6)
1882 /* For 5.6 and earlier NV flag trumps IV flag, so only use integer
1883 direct if NV flag is off. */
1884 (flags & (SVf_NOK | SVf_IOK)) == SVf_IOK
1886 /* 5.7 rules are that if IV public flag is set, IV value is as
1887 good, if not better, than NV value. */
1893 * Will come here from below with iv set if double is an integer.
1897 /* Sorry. This isn't in 5.005_56 (IIRC) or earlier. */
1899 /* Need to do this out here, else 0xFFFFFFFF becomes iv of -1
1900 * (for example) and that ends up in the optimised small integer
1903 if ((flags & SVf_IVisUV) && SvUV(sv) > IV_MAX) {
1904 TRACEME(("large unsigned integer as string, value = %"UVuf, SvUV(sv)));
1905 goto string_readlen;
1909 * Optimize small integers into a single byte, otherwise store as
1910 * a real integer (converted into network order if they asked).
1913 if (iv >= -128 && iv <= 127) {
1914 unsigned char siv = (unsigned char) (iv + 128); /* [0,255] */
1917 TRACEME(("small integer stored as %d", siv));
1918 } else if (cxt->netorder) {
1920 TRACEME(("no htonl, fall back to string for integer"));
1921 goto string_readlen;
1929 /* Sorry. This isn't in 5.005_56 (IIRC) or earlier. */
1930 ((flags & SVf_IVisUV) && SvUV(sv) > 0x7FFFFFFF) ||
1932 (iv > 0x7FFFFFFF) || (iv < -0x80000000)) {
1933 /* Bigger than 32 bits. */
1934 TRACEME(("large network order integer as string, value = %"IVdf, iv));
1935 goto string_readlen;
1939 niv = (I32) htonl((I32) iv);
1940 TRACEME(("using network order"));
1945 PUTMARK(SX_INTEGER);
1946 WRITE(&iv, sizeof(iv));
1949 TRACEME(("ok (integer 0x%"UVxf", value = %"IVdf")", PTR2UV(sv), iv));
1950 } else if (flags & SVf_NOK) {
1952 #if (PATCHLEVEL <= 6)
1955 * Watch for number being an integer in disguise.
1957 if (nv == (NV) (iv = I_V(nv))) {
1958 TRACEME(("double %"NVff" is actually integer %"IVdf, nv, iv));
1959 goto integer; /* Share code above */
1964 if (SvIOK_notUV(sv)) {
1966 goto integer; /* Share code above */
1971 if (cxt->netorder) {
1972 TRACEME(("double %"NVff" stored as string", nv));
1973 goto string_readlen; /* Share code below */
1977 WRITE(&nv, sizeof(nv));
1979 TRACEME(("ok (double 0x%"UVxf", value = %"NVff")", PTR2UV(sv), nv));
1981 } else if (flags & (SVp_POK | SVp_NOK | SVp_IOK)) {
1982 I32 wlen; /* For 64-bit machines */
1988 * Will come here from above if it was readonly, POK and NOK but
1989 * neither &PL_sv_yes nor &PL_sv_no.
1993 wlen = (I32) len; /* WLEN via STORE_SCALAR expects I32 */
1995 STORE_UTF8STR(pv, wlen);
1997 STORE_SCALAR(pv, wlen);
1998 TRACEME(("ok (scalar 0x%"UVxf" '%s', length = %"IVdf")",
1999 PTR2UV(sv), SvPVX(sv), (IV)len));
2001 CROAK(("Can't determine type of %s(0x%"UVxf")",
2002 sv_reftype(sv, FALSE),
2004 return 0; /* Ok, no recursion on scalars */
2012 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
2013 * Each item is stored as <object>.
2015 static int store_array(stcxt_t *cxt, AV *av)
2018 I32 len = av_len(av) + 1;
2022 TRACEME(("store_array (0x%"UVxf")", PTR2UV(av)));
2025 * Signal array by emitting SX_ARRAY, followed by the array length.
2030 TRACEME(("size = %d", len));
2033 * Now store each item recursively.
2036 for (i = 0; i < len; i++) {
2037 sav = av_fetch(av, i, 0);
2039 TRACEME(("(#%d) undef item", i));
2043 TRACEME(("(#%d) item", i));
2044 if ((ret = store(cxt, *sav))) /* Extra () for -Wall, grr... */
2048 TRACEME(("ok (array)"));
2057 * Borrowed from perl source file pp_ctl.c, where it is used by pp_sort.
2060 sortcmp(const void *a, const void *b)
2062 return sv_cmp(*(SV * const *) a, *(SV * const *) b);
2069 * Store a hash table.
2071 * For a "normal" hash (not restricted, no utf8 keys):
2073 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
2074 * Values are stored as <object>.
2075 * Keys are stored as <length> <data>, the <data> section being omitted
2078 * For a "fancy" hash (restricted or utf8 keys):
2080 * Layout is SX_FLAG_HASH <size> <hash flags> followed by each key/value pair,
2082 * Values are stored as <object>.
2083 * Keys are stored as <flags> <length> <data>, the <data> section being omitted
2085 * Currently the only hash flag is "restriced"
2086 * Key flags are as for hv.h
2088 static int store_hash(stcxt_t *cxt, HV *hv)
2091 #ifdef HAS_RESTRICTED_HASHES
2100 int flagged_hash = ((SvREADONLY(hv)
2101 #ifdef HAS_HASH_KEY_FLAGS
2105 unsigned char hash_flags = (SvREADONLY(hv) ? SHV_RESTRICTED : 0);
2108 /* needs int cast for C++ compilers, doesn't it? */
2109 TRACEME(("store_hash (0x%"UVxf") (flags %x)", PTR2UV(hv),
2112 TRACEME(("store_hash (0x%"UVxf")", PTR2UV(hv)));
2116 * Signal hash by emitting SX_HASH, followed by the table length.
2120 PUTMARK(SX_FLAG_HASH);
2121 PUTMARK(hash_flags);
2126 TRACEME(("size = %d", len));
2129 * Save possible iteration state via each() on that table.
2132 riter = HvRITER(hv);
2133 eiter = HvEITER(hv);
2137 * Now store each item recursively.
2139 * If canonical is defined to some true value then store each
2140 * key/value pair in sorted order otherwise the order is random.
2141 * Canonical order is irrelevant when a deep clone operation is performed.
2143 * Fetch the value from perl only once per store() operation, and only
2148 !(cxt->optype & ST_CLONE) && (cxt->canonical == 1 ||
2149 (cxt->canonical < 0 && (cxt->canonical =
2150 (SvTRUE(perl_get_sv("Storable::canonical", TRUE)) ? 1 : 0))))
2153 * Storing in order, sorted by key.
2154 * Run through the hash, building up an array of keys in a
2155 * mortal array, sort the array and then run through the
2161 /*av_extend (av, len);*/
2163 TRACEME(("using canonical order"));
2165 for (i = 0; i < len; i++) {
2166 #ifdef HAS_RESTRICTED_HASHES
2167 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
2169 HE *he = hv_iternext(hv);
2171 SV *key = hv_iterkeysv(he);
2172 av_store(av, AvFILLp(av)+1, key); /* av_push(), really */
2175 qsort((char *) AvARRAY(av), len, sizeof(SV *), sortcmp);
2177 for (i = 0; i < len; i++) {
2178 unsigned char flags;
2182 SV *key = av_shift(av);
2183 HE *he = hv_fetch_ent(hv, key, 0, 0);
2184 SV *val = HeVAL(he);
2186 return 1; /* Internal error, not I/O error */
2189 * Store value first.
2192 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
2194 if ((ret = store(cxt, val))) /* Extra () for -Wall, grr... */
2199 * Keys are written after values to make sure retrieval
2200 * can be optimal in terms of memory usage, where keys are
2201 * read into a fixed unique buffer called kbuf.
2202 * See retrieve_hash() for details.
2205 /* Implementation of restricted hashes isn't nicely
2208 = (((hash_flags & SHV_RESTRICTED)
2210 ? SHV_K_LOCKED : 0);
2211 #ifdef PL_sv_placeholder
2212 if (val == &PL_sv_placeholder)
2214 if (val == &PL_sv_undef)
2216 flags |= SHV_K_PLACEHOLDER;
2218 keyval = SvPV(key, keylen_tmp);
2219 keylen = keylen_tmp;
2220 #ifdef HAS_UTF8_HASHES
2221 /* If you build without optimisation on pre 5.6
2222 then nothing spots that SvUTF8(key) is always 0,
2223 so the block isn't optimised away, at which point
2224 the linker dislikes the reference to
2227 const char *keysave = keyval;
2228 bool is_utf8 = TRUE;
2230 /* Just casting the &klen to (STRLEN) won't work
2231 well if STRLEN and I32 are of different widths.
2233 keyval = (char*)bytes_from_utf8((U8*)keyval,
2237 /* If we were able to downgrade here, then than
2238 means that we have a key which only had chars
2239 0-255, but was utf8 encoded. */
2241 if (keyval != keysave) {
2242 keylen = keylen_tmp;
2243 flags |= SHV_K_WASUTF8;
2245 /* keylen_tmp can't have changed, so no need
2246 to assign back to keylen. */
2247 flags |= SHV_K_UTF8;
2254 TRACEME(("(#%d) key '%s' flags %x %u", i, keyval, flags, *keyval));
2256 assert (flags == 0);
2257 TRACEME(("(#%d) key '%s'", i, keyval));
2261 WRITE(keyval, keylen);
2262 if (flags & SHV_K_WASUTF8)
2267 * Free up the temporary array
2276 * Storing in "random" order (in the order the keys are stored
2277 * within the hash). This is the default and will be faster!
2280 for (i = 0; i < len; i++) {
2283 unsigned char flags;
2284 #ifdef HV_ITERNEXT_WANTPLACEHOLDERS
2285 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
2287 HE *he = hv_iternext(hv);
2289 SV *val = (he ? hv_iterval(hv, he) : 0);
2294 return 1; /* Internal error, not I/O error */
2297 * Store value first.
2300 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
2302 if ((ret = store(cxt, val))) /* Extra () for -Wall, grr... */
2305 /* Implementation of restricted hashes isn't nicely
2308 = (((hash_flags & SHV_RESTRICTED)
2310 ? SHV_K_LOCKED : 0);
2311 #ifdef PL_sv_placeholder
2312 if (val == &PL_sv_placeholder)
2314 if (val == &PL_sv_undef)
2316 flags |= SHV_K_PLACEHOLDER;
2318 hek = HeKEY_hek(he);
2320 if (len == HEf_SVKEY) {
2321 /* This is somewhat sick, but the internal APIs are
2322 * such that XS code could put one of these in in
2324 * Maybe we should be capable of storing one if
2327 key_sv = HeKEY_sv(he);
2328 flags |= SHV_K_ISSV;
2330 /* Regular string key. */
2331 #ifdef HAS_HASH_KEY_FLAGS
2333 flags |= SHV_K_UTF8;
2334 if (HEK_WASUTF8(hek))
2335 flags |= SHV_K_WASUTF8;
2341 * Keys are written after values to make sure retrieval
2342 * can be optimal in terms of memory usage, where keys are
2343 * read into a fixed unique buffer called kbuf.
2344 * See retrieve_hash() for details.
2349 TRACEME(("(#%d) key '%s' flags %x", i, key, flags));
2351 assert (flags == 0);
2352 TRACEME(("(#%d) key '%s'", i, key));
2354 if (flags & SHV_K_ISSV) {
2364 TRACEME(("ok (hash 0x%"UVxf")", PTR2UV(hv)));
2367 HvRITER(hv) = riter; /* Restore hash iterator state */
2368 HvEITER(hv) = eiter;
2376 * Store a code reference.
2378 * Layout is SX_CODE <length> followed by a scalar containing the perl
2379 * source code of the code reference.
2381 static int store_code(stcxt_t *cxt, CV *cv)
2383 #if PERL_VERSION < 6
2385 * retrieve_code does not work with perl 5.005 or less
2387 return store_other(cxt, (SV*)cv);
2392 SV *text, *bdeparse;
2394 TRACEME(("store_code (0x%"UVxf")", PTR2UV(cv)));
2397 cxt->deparse == 0 ||
2398 (cxt->deparse < 0 && !(cxt->deparse =
2399 SvTRUE(perl_get_sv("Storable::Deparse", TRUE)) ? 1 : 0))
2401 return store_other(cxt, (SV*)cv);
2405 * Require B::Deparse. At least B::Deparse 0.61 is needed for
2406 * blessed code references.
2408 /* XXX sv_2mortal seems to be evil here. why? */
2409 load_module(PERL_LOADMOD_NOIMPORT, newSVpvn("B::Deparse",10), newSVnv(0.61));
2415 * create the B::Deparse object
2419 XPUSHs(sv_2mortal(newSVpvn("B::Deparse",10)));
2421 count = call_method("new", G_SCALAR);
2424 CROAK(("Unexpected return value from B::Deparse::new\n"));
2428 * call the coderef2text method
2432 XPUSHs(bdeparse); /* XXX is this already mortal? */
2433 XPUSHs(sv_2mortal(newRV_inc((SV*)cv)));
2435 count = call_method("coderef2text", G_SCALAR);
2438 CROAK(("Unexpected return value from B::Deparse::coderef2text\n"));
2442 reallen = strlen(SvPV_nolen(text));
2445 * Empty code references or XS functions are deparsed as
2446 * "(prototype) ;" or ";".
2449 if (len == 0 || *(SvPV_nolen(text)+reallen-1) == ';') {
2450 CROAK(("The result of B::Deparse::coderef2text was empty - maybe you're trying to serialize an XS function?\n"));
2454 * Signal code by emitting SX_CODE.
2458 TRACEME(("size = %d", len));
2459 TRACEME(("code = %s", SvPV_nolen(text)));
2462 * Now store the source code.
2465 STORE_SCALAR(SvPV_nolen(text), len);
2470 TRACEME(("ok (code)"));
2479 * When storing a tied object (be it a tied scalar, array or hash), we lay out
2480 * a special mark, followed by the underlying tied object. For instance, when
2481 * dealing with a tied hash, we store SX_TIED_HASH <hash object>, where
2482 * <hash object> stands for the serialization of the tied hash.
2484 static int store_tied(stcxt_t *cxt, SV *sv)
2489 int svt = SvTYPE(sv);
2492 TRACEME(("store_tied (0x%"UVxf")", PTR2UV(sv)));
2495 * We have a small run-time penalty here because we chose to factorise
2496 * all tieds objects into the same routine, and not have a store_tied_hash,
2497 * a store_tied_array, etc...
2499 * Don't use a switch() statement, as most compilers don't optimize that
2500 * well for 2/3 values. An if() else if() cascade is just fine. We put
2501 * tied hashes first, as they are the most likely beasts.
2504 if (svt == SVt_PVHV) {
2505 TRACEME(("tied hash"));
2506 PUTMARK(SX_TIED_HASH); /* Introduces tied hash */
2507 } else if (svt == SVt_PVAV) {
2508 TRACEME(("tied array"));
2509 PUTMARK(SX_TIED_ARRAY); /* Introduces tied array */
2511 TRACEME(("tied scalar"));
2512 PUTMARK(SX_TIED_SCALAR); /* Introduces tied scalar */
2516 if (!(mg = mg_find(sv, mtype)))
2517 CROAK(("No magic '%c' found while storing tied %s", mtype,
2518 (svt == SVt_PVHV) ? "hash" :
2519 (svt == SVt_PVAV) ? "array" : "scalar"));
2522 * The mg->mg_obj found by mg_find() above actually points to the
2523 * underlying tied Perl object implementation. For instance, if the
2524 * original SV was that of a tied array, then mg->mg_obj is an AV.
2526 * Note that we store the Perl object as-is. We don't call its FETCH
2527 * method along the way. At retrieval time, we won't call its STORE
2528 * method either, but the tieing magic will be re-installed. In itself,
2529 * that ensures that the tieing semantics are preserved since futher
2530 * accesses on the retrieved object will indeed call the magic methods...
2533 /* [#17040] mg_obj is NULL for scalar self-ties. AMS 20030416 */
2534 obj = mg->mg_obj ? mg->mg_obj : newSV(0);
2535 if ((ret = store(cxt, obj)))
2538 TRACEME(("ok (tied)"));
2546 * Stores a reference to an item within a tied structure:
2548 * . \$h{key}, stores both the (tied %h) object and 'key'.
2549 * . \$a[idx], stores both the (tied @a) object and 'idx'.
2551 * Layout is therefore either:
2552 * SX_TIED_KEY <object> <key>
2553 * SX_TIED_IDX <object> <index>
2555 static int store_tied_item(stcxt_t *cxt, SV *sv)
2560 TRACEME(("store_tied_item (0x%"UVxf")", PTR2UV(sv)));
2562 if (!(mg = mg_find(sv, 'p')))
2563 CROAK(("No magic 'p' found while storing reference to tied item"));
2566 * We discriminate between \$h{key} and \$a[idx] via mg_ptr.
2570 TRACEME(("store_tied_item: storing a ref to a tied hash item"));
2571 PUTMARK(SX_TIED_KEY);
2572 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2574 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
2577 TRACEME(("store_tied_item: storing PTR 0x%"UVxf, PTR2UV(mg->mg_ptr)));
2579 if ((ret = store(cxt, (SV *) mg->mg_ptr))) /* Idem, for -Wall */
2582 I32 idx = mg->mg_len;
2584 TRACEME(("store_tied_item: storing a ref to a tied array item "));
2585 PUTMARK(SX_TIED_IDX);
2586 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2588 if ((ret = store(cxt, mg->mg_obj))) /* Idem, for -Wall */
2591 TRACEME(("store_tied_item: storing IDX %d", idx));
2596 TRACEME(("ok (tied item)"));
2602 * store_hook -- dispatched manually, not via sv_store[]
2604 * The blessed SV is serialized by a hook.
2608 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2610 * where <flags> indicates how long <len>, <len2> and <len3> are, whether
2611 * the trailing part [] is present, the type of object (scalar, array or hash).
2612 * There is also a bit which says how the classname is stored between:
2617 * and when the <index> form is used (classname already seen), the "large
2618 * classname" bit in <flags> indicates how large the <index> is.
2620 * The serialized string returned by the hook is of length <len2> and comes
2621 * next. It is an opaque string for us.
2623 * Those <len3> object IDs which are listed last represent the extra references
2624 * not directly serialized by the hook, but which are linked to the object.
2626 * When recursion is mandated to resolve object-IDs not yet seen, we have
2627 * instead, with <header> being flags with bits set to indicate the object type
2628 * and that recursion was indeed needed:
2630 * SX_HOOK <header> <object> <header> <object> <flags>
2632 * that same header being repeated between serialized objects obtained through
2633 * recursion, until we reach flags indicating no recursion, at which point
2634 * we know we've resynchronized with a single layout, after <flags>.
2636 * When storing a blessed ref to a tied variable, the following format is
2639 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
2641 * The first <flags> indication carries an object of type SHT_EXTRA, and the
2642 * real object type is held in the <extra> flag. At the very end of the
2643 * serialization stream, the underlying magic object is serialized, just like
2644 * any other tied variable.
2646 static int store_hook(
2659 int count; /* really len3 + 1 */
2660 unsigned char flags;
2663 int recursed = 0; /* counts recursion */
2664 int obj_type; /* object type, on 2 bits */
2667 int clone = cxt->optype & ST_CLONE;
2668 char mtype = '\0'; /* for blessed ref to tied structures */
2669 unsigned char eflags = '\0'; /* used when object type is SHT_EXTRA */
2671 TRACEME(("store_hook, class \"%s\", tagged #%d", HvNAME(pkg), cxt->tagnum));
2674 * Determine object type on 2 bits.
2679 obj_type = SHT_SCALAR;
2682 obj_type = SHT_ARRAY;
2685 obj_type = SHT_HASH;
2689 * Produced by a blessed ref to a tied data structure, $o in the
2690 * following Perl code.
2694 * my $o = bless \%h, 'BAR';
2696 * Signal the tie-ing magic by setting the object type as SHT_EXTRA
2697 * (since we have only 2 bits in <flags> to store the type), and an
2698 * <extra> byte flag will be emitted after the FIRST <flags> in the
2699 * stream, carrying what we put in `eflags'.
2701 obj_type = SHT_EXTRA;
2702 switch (SvTYPE(sv)) {
2704 eflags = (unsigned char) SHT_THASH;
2708 eflags = (unsigned char) SHT_TARRAY;
2712 eflags = (unsigned char) SHT_TSCALAR;
2718 CROAK(("Unexpected object type (%d) in store_hook()", type));
2720 flags = SHF_NEED_RECURSE | obj_type;
2722 class = HvNAME(pkg);
2723 len = strlen(class);
2726 * To call the hook, we need to fake a call like:
2728 * $object->STORABLE_freeze($cloning);
2730 * but we don't have the $object here. For instance, if $object is
2731 * a blessed array, what we have in `sv' is the array, and we can't
2732 * call a method on those.
2734 * Therefore, we need to create a temporary reference to the object and
2735 * make the call on that reference.
2738 TRACEME(("about to call STORABLE_freeze on class %s", class));
2740 ref = newRV_noinc(sv); /* Temporary reference */
2741 av = array_call(ref, hook, clone); /* @a = $object->STORABLE_freeze($c) */
2743 SvREFCNT_dec(ref); /* Reclaim temporary reference */
2745 count = AvFILLp(av) + 1;
2746 TRACEME(("store_hook, array holds %d items", count));
2749 * If they return an empty list, it means they wish to ignore the
2750 * hook for this class (and not just this instance -- that's for them
2751 * to handle if they so wish).
2753 * Simply disable the cached entry for the hook (it won't be recomputed
2754 * since it's present in the cache) and recurse to store_blessed().
2759 * They must not change their mind in the middle of a serialization.
2762 if (hv_fetch(cxt->hclass, class, len, FALSE))
2763 CROAK(("Too late to ignore hooks for %s class \"%s\"",
2764 (cxt->optype & ST_CLONE) ? "cloning" : "storing", class));
2766 pkg_hide(cxt->hook, pkg, "STORABLE_freeze");
2768 ASSERT(!pkg_can(cxt->hook, pkg, "STORABLE_freeze"), ("hook invisible"));
2769 TRACEME(("ignoring STORABLE_freeze in class \"%s\"", class));
2771 return store_blessed(cxt, sv, type, pkg);
2775 * Get frozen string.
2779 pv = SvPV(ary[0], len2);
2782 * If they returned more than one item, we need to serialize some
2783 * extra references if not already done.
2785 * Loop over the array, starting at position #1, and for each item,
2786 * ensure it is a reference, serialize it if not already done, and
2787 * replace the entry with the tag ID of the corresponding serialized
2790 * We CHEAT by not calling av_fetch() and read directly within the
2794 for (i = 1; i < count; i++) {
2798 AV *av_hook = cxt->hook_seen;
2801 CROAK(("Item #%d returned by STORABLE_freeze "
2802 "for %s is not a reference", i, class));
2803 xsv = SvRV(rsv); /* Follow ref to know what to look for */
2806 * Look in hseen and see if we have a tag already.
2807 * Serialize entry if not done already, and get its tag.
2810 if ((svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE)))
2811 goto sv_seen; /* Avoid moving code too far to the right */
2813 TRACEME(("listed object %d at 0x%"UVxf" is unknown", i-1, PTR2UV(xsv)));
2816 * We need to recurse to store that object and get it to be known
2817 * so that we can resolve the list of object-IDs at retrieve time.
2819 * The first time we do this, we need to emit the proper header
2820 * indicating that we recursed, and what the type of object is (the
2821 * object we're storing via a user-hook). Indeed, during retrieval,
2822 * we'll have to create the object before recursing to retrieve the
2823 * others, in case those would point back at that object.
2826 /* [SX_HOOK] <flags> [<extra>] <object>*/
2830 if (obj_type == SHT_EXTRA)
2835 if ((ret = store(cxt, xsv))) /* Given by hook for us to store */
2838 svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE);
2840 CROAK(("Could not serialize item #%d from hook in %s", i, class));
2843 * It was the first time we serialized `xsv'.
2845 * Keep this SV alive until the end of the serialization: if we
2846 * disposed of it right now by decrementing its refcount, and it was
2847 * a temporary value, some next temporary value allocated during
2848 * another STORABLE_freeze might take its place, and we'd wrongly
2849 * assume that new SV was already serialized, based on its presence
2852 * Therefore, push it away in cxt->hook_seen.
2855 av_store(av_hook, AvFILLp(av_hook)+1, SvREFCNT_inc(xsv));
2859 * Dispose of the REF they returned. If we saved the `xsv' away
2860 * in the array of returned SVs, that will not cause the underlying
2861 * referenced SV to be reclaimed.
2864 ASSERT(SvREFCNT(xsv) > 1, ("SV will survive disposal of its REF"));
2865 SvREFCNT_dec(rsv); /* Dispose of reference */
2868 * Replace entry with its tag (not a real SV, so no refcnt increment)
2872 TRACEME(("listed object %d at 0x%"UVxf" is tag #%"UVuf,
2873 i-1, PTR2UV(xsv), PTR2UV(*svh)));
2877 * Allocate a class ID if not already done.
2879 * This needs to be done after the recursion above, since at retrieval
2880 * time, we'll see the inner objects first. Many thanks to
2881 * Salvador Ortiz Garcia <sog@msg.com.mx> who spot that bug and
2882 * proposed the right fix. -- RAM, 15/09/2000
2885 if (!known_class(cxt, class, len, &classnum)) {
2886 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2887 classnum = -1; /* Mark: we must store classname */
2889 TRACEME(("already seen class %s, ID = %d", class, classnum));
2893 * Compute leading flags.
2897 if (((classnum == -1) ? len : classnum) > LG_SCALAR)
2898 flags |= SHF_LARGE_CLASSLEN;
2900 flags |= SHF_IDX_CLASSNAME;
2901 if (len2 > LG_SCALAR)
2902 flags |= SHF_LARGE_STRLEN;
2904 flags |= SHF_HAS_LIST;
2905 if (count > (LG_SCALAR + 1))
2906 flags |= SHF_LARGE_LISTLEN;
2909 * We're ready to emit either serialized form:
2911 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2912 * SX_HOOK <flags> <index> <len2> <str> [<len3> <object-IDs>]
2914 * If we recursed, the SX_HOOK has already been emitted.
2917 TRACEME(("SX_HOOK (recursed=%d) flags=0x%x "
2918 "class=%"IVdf" len=%"IVdf" len2=%"IVdf" len3=%d",
2919 recursed, flags, (IV)classnum, (IV)len, (IV)len2, count-1));
2921 /* SX_HOOK <flags> [<extra>] */
2925 if (obj_type == SHT_EXTRA)
2930 /* <len> <classname> or <index> */
2931 if (flags & SHF_IDX_CLASSNAME) {
2932 if (flags & SHF_LARGE_CLASSLEN)
2935 unsigned char cnum = (unsigned char) classnum;
2939 if (flags & SHF_LARGE_CLASSLEN)
2942 unsigned char clen = (unsigned char) len;
2945 WRITE(class, len); /* Final \0 is omitted */
2948 /* <len2> <frozen-str> */
2949 if (flags & SHF_LARGE_STRLEN) {
2950 I32 wlen2 = len2; /* STRLEN might be 8 bytes */
2951 WLEN(wlen2); /* Must write an I32 for 64-bit machines */
2953 unsigned char clen = (unsigned char) len2;
2957 WRITE(pv, (SSize_t)len2); /* Final \0 is omitted */
2959 /* [<len3> <object-IDs>] */
2960 if (flags & SHF_HAS_LIST) {
2961 int len3 = count - 1;
2962 if (flags & SHF_LARGE_LISTLEN)
2965 unsigned char clen = (unsigned char) len3;
2970 * NOTA BENE, for 64-bit machines: the ary[i] below does not yield a
2971 * real pointer, rather a tag number, well under the 32-bit limit.
2974 for (i = 1; i < count; i++) {
2975 I32 tagval = htonl(LOW_32BITS(ary[i]));
2977 TRACEME(("object %d, tag #%d", i-1, ntohl(tagval)));
2982 * Free the array. We need extra care for indices after 0, since they
2983 * don't hold real SVs but integers cast.
2987 AvFILLp(av) = 0; /* Cheat, nothing after 0 interests us */
2992 * If object was tied, need to insert serialization of the magic object.
2995 if (obj_type == SHT_EXTRA) {
2998 if (!(mg = mg_find(sv, mtype))) {
2999 int svt = SvTYPE(sv);
3000 CROAK(("No magic '%c' found while storing ref to tied %s with hook",
3001 mtype, (svt == SVt_PVHV) ? "hash" :
3002 (svt == SVt_PVAV) ? "array" : "scalar"));
3005 TRACEME(("handling the magic object 0x%"UVxf" part of 0x%"UVxf,
3006 PTR2UV(mg->mg_obj), PTR2UV(sv)));
3012 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
3020 * store_blessed -- dispatched manually, not via sv_store[]
3022 * Check whether there is a STORABLE_xxx hook defined in the class or in one
3023 * of its ancestors. If there is, then redispatch to store_hook();
3025 * Otherwise, the blessed SV is stored using the following layout:
3027 * SX_BLESS <flag> <len> <classname> <object>
3029 * where <flag> indicates whether <len> is stored on 0 or 4 bytes, depending
3030 * on the high-order bit in flag: if 1, then length follows on 4 bytes.
3031 * Otherwise, the low order bits give the length, thereby giving a compact
3032 * representation for class names less than 127 chars long.
3034 * Each <classname> seen is remembered and indexed, so that the next time
3035 * an object in the blessed in the same <classname> is stored, the following
3038 * SX_IX_BLESS <flag> <index> <object>
3040 * where <index> is the classname index, stored on 0 or 4 bytes depending
3041 * on the high-order bit in flag (same encoding as above for <len>).
3043 static int store_blessed(
3054 TRACEME(("store_blessed, type %d, class \"%s\"", type, HvNAME(pkg)));
3057 * Look for a hook for this blessed SV and redirect to store_hook()
3061 hook = pkg_can(cxt->hook, pkg, "STORABLE_freeze");
3063 return store_hook(cxt, sv, type, pkg, hook);
3066 * This is a blessed SV without any serialization hook.
3069 class = HvNAME(pkg);
3070 len = strlen(class);
3072 TRACEME(("blessed 0x%"UVxf" in %s, no hook: tagged #%d",
3073 PTR2UV(sv), class, cxt->tagnum));
3076 * Determine whether it is the first time we see that class name (in which
3077 * case it will be stored in the SX_BLESS form), or whether we already
3078 * saw that class name before (in which case the SX_IX_BLESS form will be
3082 if (known_class(cxt, class, len, &classnum)) {
3083 TRACEME(("already seen class %s, ID = %d", class, classnum));
3084 PUTMARK(SX_IX_BLESS);
3085 if (classnum <= LG_BLESS) {
3086 unsigned char cnum = (unsigned char) classnum;
3089 unsigned char flag = (unsigned char) 0x80;
3094 TRACEME(("first time we see class %s, ID = %d", class, classnum));
3096 if (len <= LG_BLESS) {
3097 unsigned char clen = (unsigned char) len;
3100 unsigned char flag = (unsigned char) 0x80;
3102 WLEN(len); /* Don't BER-encode, this should be rare */
3104 WRITE(class, len); /* Final \0 is omitted */
3108 * Now emit the <object> part.
3111 return SV_STORE(type)(cxt, sv);
3117 * We don't know how to store the item we reached, so return an error condition.
3118 * (it's probably a GLOB, some CODE reference, etc...)
3120 * If they defined the `forgive_me' variable at the Perl level to some
3121 * true value, then don't croak, just warn, and store a placeholder string
3124 static int store_other(stcxt_t *cxt, SV *sv)
3127 static char buf[80];
3129 TRACEME(("store_other"));
3132 * Fetch the value from perl only once per store() operation.
3136 cxt->forgive_me == 0 ||
3137 (cxt->forgive_me < 0 && !(cxt->forgive_me =
3138 SvTRUE(perl_get_sv("Storable::forgive_me", TRUE)) ? 1 : 0))
3140 CROAK(("Can't store %s items", sv_reftype(sv, FALSE)));
3142 warn("Can't store item %s(0x%"UVxf")",
3143 sv_reftype(sv, FALSE), PTR2UV(sv));
3146 * Store placeholder string as a scalar instead...
3149 (void) sprintf(buf, "You lost %s(0x%"UVxf")%c", sv_reftype(sv, FALSE),
3150 PTR2UV(sv), (char) 0);
3153 STORE_SCALAR(buf, len);
3154 TRACEME(("ok (dummy \"%s\", length = %"IVdf")", buf, (IV) len));
3160 *** Store driving routines
3166 * WARNING: partially duplicates Perl's sv_reftype for speed.
3168 * Returns the type of the SV, identified by an integer. That integer
3169 * may then be used to index the dynamic routine dispatch table.
3171 static int sv_type(SV *sv)
3173 switch (SvTYPE(sv)) {
3178 * No need to check for ROK, that can't be set here since there
3179 * is no field capable of hodling the xrv_rv reference.
3187 * Starting from SVt_PV, it is possible to have the ROK flag
3188 * set, the pointer to the other SV being either stored in
3189 * the xrv_rv (in the case of a pure SVt_RV), or as the
3190 * xpv_pv field of an SVt_PV and its heirs.
3192 * However, those SV cannot be magical or they would be an
3193 * SVt_PVMG at least.
3195 return SvROK(sv) ? svis_REF : svis_SCALAR;
3197 case SVt_PVLV: /* Workaround for perl5.004_04 "LVALUE" bug */
3198 if (SvRMAGICAL(sv) && (mg_find(sv, 'p')))
3199 return svis_TIED_ITEM;
3202 if (SvRMAGICAL(sv) && (mg_find(sv, 'q')))
3204 return SvROK(sv) ? svis_REF : svis_SCALAR;
3206 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
3210 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
3225 * Recursively store objects pointed to by the sv to the specified file.
3227 * Layout is <content> or SX_OBJECT <tagnum> if we reach an already stored
3228 * object (one for which storage has started -- it may not be over if we have
3229 * a self-referenced structure). This data set forms a stored <object>.
3231 static int store(stcxt_t *cxt, SV *sv)
3236 HV *hseen = cxt->hseen;
3238 TRACEME(("store (0x%"UVxf")", PTR2UV(sv)));
3241 * If object has already been stored, do not duplicate data.
3242 * Simply emit the SX_OBJECT marker followed by its tag data.
3243 * The tag is always written in network order.
3245 * NOTA BENE, for 64-bit machines: the "*svh" below does not yield a
3246 * real pointer, rather a tag number (watch the insertion code below).
3247 * That means it probably safe to assume it is well under the 32-bit limit,
3248 * and makes the truncation safe.
3249 * -- RAM, 14/09/1999
3252 svh = hv_fetch(hseen, (char *) &sv, sizeof(sv), FALSE);
3254 I32 tagval = htonl(LOW_32BITS(*svh));
3256 TRACEME(("object 0x%"UVxf" seen as #%d", PTR2UV(sv), ntohl(tagval)));
3264 * Allocate a new tag and associate it with the address of the sv being
3265 * stored, before recursing...
3267 * In order to avoid creating new SvIVs to hold the tagnum we just
3268 * cast the tagnum to an SV pointer and store that in the hash. This
3269 * means that we must clean up the hash manually afterwards, but gives
3270 * us a 15% throughput increase.
3275 if (!hv_store(hseen,
3276 (char *) &sv, sizeof(sv), INT2PTR(SV*, cxt->tagnum), 0))
3280 * Store `sv' and everything beneath it, using appropriate routine.
3281 * Abort immediately if we get a non-zero status back.
3286 TRACEME(("storing 0x%"UVxf" tag #%d, type %d...",
3287 PTR2UV(sv), cxt->tagnum, type));
3290 HV *pkg = SvSTASH(sv);
3291 ret = store_blessed(cxt, sv, type, pkg);
3293 ret = SV_STORE(type)(cxt, sv);
3295 TRACEME(("%s (stored 0x%"UVxf", refcnt=%d, %s)",
3296 ret ? "FAILED" : "ok", PTR2UV(sv),
3297 SvREFCNT(sv), sv_reftype(sv, FALSE)));
3305 * Write magic number and system information into the file.
3306 * Layout is <magic> <network> [<len> <byteorder> <sizeof int> <sizeof long>
3307 * <sizeof ptr>] where <len> is the length of the byteorder hexa string.
3308 * All size and lenghts are written as single characters here.
3310 * Note that no byte ordering info is emitted when <network> is true, since
3311 * integers will be emitted in network order in that case.
3313 static int magic_write(stcxt_t *cxt)
3316 * Starting with 0.6, the "use_network_order" byte flag is also used to
3317 * indicate the version number of the binary image, encoded in the upper
3318 * bits. The bit 0 is always used to indicate network order.
3321 * Starting with 0.7, a full byte is dedicated to the minor version of
3322 * the binary format, which is incremented only when new markers are
3323 * introduced, for instance, but when backward compatibility is preserved.
3326 /* Make these at compile time. The WRITE() macro is sufficiently complex
3327 that it saves about 200 bytes doing it this way and only using it
3329 static const unsigned char network_file_header[] = {
3331 (STORABLE_BIN_MAJOR << 1) | 1,
3332 STORABLE_BIN_WRITE_MINOR
3334 static const unsigned char file_header[] = {
3336 (STORABLE_BIN_MAJOR << 1) | 0,
3337 STORABLE_BIN_WRITE_MINOR,
3338 /* sizeof the array includes the 0 byte at the end: */
3339 (char) sizeof (byteorderstr) - 1,
3341 (unsigned char) sizeof(int),
3342 (unsigned char) sizeof(long),
3343 (unsigned char) sizeof(char *),
3344 (unsigned char) sizeof(NV)
3346 #ifdef USE_56_INTERWORK_KLUDGE
3347 static const unsigned char file_header_56[] = {
3349 (STORABLE_BIN_MAJOR << 1) | 0,
3350 STORABLE_BIN_WRITE_MINOR,
3351 /* sizeof the array includes the 0 byte at the end: */
3352 (char) sizeof (byteorderstr_56) - 1,
3354 (unsigned char) sizeof(int),
3355 (unsigned char) sizeof(long),
3356 (unsigned char) sizeof(char *),
3357 (unsigned char) sizeof(NV)
3360 const unsigned char *header;
3363 TRACEME(("magic_write on fd=%d", cxt->fio ? PerlIO_fileno(cxt->fio) : -1));
3365 if (cxt->netorder) {
3366 header = network_file_header;
3367 length = sizeof (network_file_header);
3369 #ifdef USE_56_INTERWORK_KLUDGE
3370 if (SvTRUE(perl_get_sv("Storable::interwork_56_64bit", TRUE))) {
3371 header = file_header_56;
3372 length = sizeof (file_header_56);
3376 header = file_header;
3377 length = sizeof (file_header);
3382 /* sizeof the array includes the 0 byte at the end. */
3383 header += sizeof (magicstr) - 1;
3384 length -= sizeof (magicstr) - 1;
3387 WRITE( (unsigned char*) header, length);
3389 if (!cxt->netorder) {
3390 TRACEME(("ok (magic_write byteorder = 0x%lx [%d], I%d L%d P%d D%d)",
3391 (unsigned long) BYTEORDER, (int) sizeof (byteorderstr) - 1,
3392 (int) sizeof(int), (int) sizeof(long),
3393 (int) sizeof(char *), (int) sizeof(NV)));
3401 * Common code for store operations.
3403 * When memory store is requested (f = NULL) and a non null SV* is given in
3404 * `res', it is filled with a new SV created out of the memory buffer.
3406 * It is required to provide a non-null `res' when the operation type is not
3407 * dclone() and store() is performed to memory.
3409 static int do_store(
3419 ASSERT(!(f == 0 && !(optype & ST_CLONE)) || res,
3420 ("must supply result SV pointer for real recursion to memory"));
3422 TRACEME(("do_store (optype=%d, netorder=%d)",
3423 optype, network_order));
3428 * Workaround for CROAK leak: if they enter with a "dirty" context,
3429 * free up memory for them now.
3436 * Now that STORABLE_xxx hooks exist, it is possible that they try to
3437 * re-enter store() via the hooks. We need to stack contexts.
3441 cxt = allocate_context(cxt);
3445 ASSERT(cxt->entry == 1, ("starting new recursion"));
3446 ASSERT(!cxt->s_dirty, ("clean context"));
3449 * Ensure sv is actually a reference. From perl, we called something
3451 * pstore(FILE, \@array);
3452 * so we must get the scalar value behing that reference.
3456 CROAK(("Not a reference"));
3457 sv = SvRV(sv); /* So follow it to know what to store */
3460 * If we're going to store to memory, reset the buffer.
3467 * Prepare context and emit headers.
3470 init_store_context(cxt, f, optype, network_order);
3472 if (-1 == magic_write(cxt)) /* Emit magic and ILP info */
3473 return 0; /* Error */
3476 * Recursively store object...
3479 ASSERT(is_storing(), ("within store operation"));
3481 status = store(cxt, sv); /* Just do it! */
3484 * If they asked for a memory store and they provided an SV pointer,
3485 * make an SV string out of the buffer and fill their pointer.
3487 * When asking for ST_REAL, it's MANDATORY for the caller to provide
3488 * an SV, since context cleanup might free the buffer if we did recurse.
3489 * (unless caller is dclone(), which is aware of that).
3492 if (!cxt->fio && res)
3498 * The "root" context is never freed, since it is meant to be always
3499 * handy for the common case where no recursion occurs at all (i.e.
3500 * we enter store() outside of any Storable code and leave it, period).
3501 * We know it's the "root" context because there's nothing stacked
3506 * When deep cloning, we don't free the context: doing so would force
3507 * us to copy the data in the memory buffer. Sicne we know we're
3508 * about to enter do_retrieve...
3511 clean_store_context(cxt);
3512 if (cxt->prev && !(cxt->optype & ST_CLONE))
3515 TRACEME(("do_store returns %d", status));
3523 * Store the transitive data closure of given object to disk.
3524 * Returns 0 on error, a true value otherwise.
3526 int pstore(PerlIO *f, SV *sv)
3528 TRACEME(("pstore"));
3529 return do_store(f, sv, 0, FALSE, (SV**) 0);
3536 * Same as pstore(), but network order is used for integers and doubles are
3537 * emitted as strings.
3539 int net_pstore(PerlIO *f, SV *sv)
3541 TRACEME(("net_pstore"));
3542 return do_store(f, sv, 0, TRUE, (SV**) 0);
3552 * Build a new SV out of the content of the internal memory buffer.
3554 static SV *mbuf2sv(void)
3558 return newSVpv(mbase, MBUF_SIZE());
3564 * Store the transitive data closure of given object to memory.
3565 * Returns undef on error, a scalar value containing the data otherwise.
3571 TRACEME(("mstore"));
3573 if (!do_store((PerlIO*) 0, sv, 0, FALSE, &out))
3574 return &PL_sv_undef;
3582 * Same as mstore(), but network order is used for integers and doubles are
3583 * emitted as strings.
3585 SV *net_mstore(SV *sv)
3589 TRACEME(("net_mstore"));
3591 if (!do_store((PerlIO*) 0, sv, 0, TRUE, &out))
3592 return &PL_sv_undef;
3598 *** Specific retrieve callbacks.
3604 * Return an error via croak, since it is not possible that we get here
3605 * under normal conditions, when facing a file produced via pstore().
3607 static SV *retrieve_other(stcxt_t *cxt, char *cname)
3610 cxt->ver_major != STORABLE_BIN_MAJOR &&
3611 cxt->ver_minor != STORABLE_BIN_MINOR
3613 CROAK(("Corrupted storable %s (binary v%d.%d), current is v%d.%d",
3614 cxt->fio ? "file" : "string",
3615 cxt->ver_major, cxt->ver_minor,
3616 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
3618 CROAK(("Corrupted storable %s (binary v%d.%d)",
3619 cxt->fio ? "file" : "string",
3620 cxt->ver_major, cxt->ver_minor));
3623 return (SV *) 0; /* Just in case */
3627 * retrieve_idx_blessed
3629 * Layout is SX_IX_BLESS <index> <object> with SX_IX_BLESS already read.
3630 * <index> can be coded on either 1 or 5 bytes.
3632 static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname)
3639 TRACEME(("retrieve_idx_blessed (#%d)", cxt->tagnum));
3640 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3642 GETMARK(idx); /* Index coded on a single char? */
3647 * Fetch classname in `aclass'
3650 sva = av_fetch(cxt->aclass, idx, FALSE);
3652 CROAK(("Class name #%"IVdf" should have been seen already", (IV) idx));
3654 class = SvPVX(*sva); /* We know it's a PV, by construction */
3656 TRACEME(("class ID %d => %s", idx, class));
3659 * Retrieve object and bless it.
3662 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
3670 * Layout is SX_BLESS <len> <classname> <object> with SX_BLESS already read.
3671 * <len> can be coded on either 1 or 5 bytes.
3673 static SV *retrieve_blessed(stcxt_t *cxt, char *cname)
3677 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3680 TRACEME(("retrieve_blessed (#%d)", cxt->tagnum));
3681 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3684 * Decode class name length and read that name.
3686 * Short classnames have two advantages: their length is stored on one
3687 * single byte, and the string can be read on the stack.
3690 GETMARK(len); /* Length coded on a single char? */
3693 TRACEME(("** allocating %d bytes for class name", len+1));
3694 New(10003, class, len+1, char);
3697 class[len] = '\0'; /* Mark string end */
3700 * It's a new classname, otherwise it would have been an SX_IX_BLESS.
3703 TRACEME(("new class name \"%s\" will bear ID = %d", class, cxt->classnum));
3705 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3709 * Retrieve object and bless it.
3712 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
3722 * Layout: SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
3723 * with leading mark already read, as usual.
3725 * When recursion was involved during serialization of the object, there
3726 * is an unknown amount of serialized objects after the SX_HOOK mark. Until
3727 * we reach a <flags> marker with the recursion bit cleared.
3729 * If the first <flags> byte contains a type of SHT_EXTRA, then the real type
3730 * is held in the <extra> byte, and if the object is tied, the serialized
3731 * magic object comes at the very end:
3733 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
3735 * This means the STORABLE_thaw hook will NOT get a tied variable during its
3736 * processing (since we won't have seen the magic object by the time the hook
3737 * is called). See comments below for why it was done that way.
3739 static SV *retrieve_hook(stcxt_t *cxt, char *cname)
3742 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3753 int clone = cxt->optype & ST_CLONE;
3755 unsigned int extra_type = 0;
3757 TRACEME(("retrieve_hook (#%d)", cxt->tagnum));
3758 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3761 * Read flags, which tell us about the type, and whether we need to recurse.
3767 * Create the (empty) object, and mark it as seen.
3769 * This must be done now, because tags are incremented, and during
3770 * serialization, the object tag was affected before recursion could
3774 obj_type = flags & SHF_TYPE_MASK;
3780 sv = (SV *) newAV();
3783 sv = (SV *) newHV();
3787 * Read <extra> flag to know the type of the object.
3788 * Record associated magic type for later.
3790 GETMARK(extra_type);
3791 switch (extra_type) {
3797 sv = (SV *) newAV();
3801 sv = (SV *) newHV();
3805 return retrieve_other(cxt, 0); /* Let it croak */
3809 return retrieve_other(cxt, 0); /* Let it croak */
3811 SEEN(sv, 0); /* Don't bless yet */
3814 * Whilst flags tell us to recurse, do so.
3816 * We don't need to remember the addresses returned by retrieval, because
3817 * all the references will be obtained through indirection via the object
3818 * tags in the object-ID list.
3820 * We need to decrement the reference count for these objects
3821 * because, if the user doesn't save a reference to them in the hook,
3822 * they must be freed when this context is cleaned.
3825 while (flags & SHF_NEED_RECURSE) {
3826 TRACEME(("retrieve_hook recursing..."));
3827 rv = retrieve(cxt, 0);
3831 TRACEME(("retrieve_hook back with rv=0x%"UVxf,
3836 if (flags & SHF_IDX_CLASSNAME) {
3841 * Fetch index from `aclass'
3844 if (flags & SHF_LARGE_CLASSLEN)
3849 sva = av_fetch(cxt->aclass, idx, FALSE);
3851 CROAK(("Class name #%"IVdf" should have been seen already",
3854 class = SvPVX(*sva); /* We know it's a PV, by construction */
3855 TRACEME(("class ID %d => %s", idx, class));
3859 * Decode class name length and read that name.
3861 * NOTA BENE: even if the length is stored on one byte, we don't read
3862 * on the stack. Just like retrieve_blessed(), we limit the name to
3863 * LG_BLESS bytes. This is an arbitrary decision.
3866 if (flags & SHF_LARGE_CLASSLEN)
3871 if (len > LG_BLESS) {
3872 TRACEME(("** allocating %d bytes for class name", len+1));
3873 New(10003, class, len+1, char);
3877 class[len] = '\0'; /* Mark string end */
3880 * Record new classname.
3883 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3887 TRACEME(("class name: %s", class));
3890 * Decode user-frozen string length and read it in an SV.
3892 * For efficiency reasons, we read data directly into the SV buffer.
3893 * To understand that code, read retrieve_scalar()
3896 if (flags & SHF_LARGE_STRLEN)
3901 frozen = NEWSV(10002, len2);
3903 SAFEREAD(SvPVX(frozen), len2, frozen);
3904 SvCUR_set(frozen, len2);
3905 *SvEND(frozen) = '\0';
3907 (void) SvPOK_only(frozen); /* Validates string pointer */
3908 if (cxt->s_tainted) /* Is input source tainted? */
3911 TRACEME(("frozen string: %d bytes", len2));
3914 * Decode object-ID list length, if present.
3917 if (flags & SHF_HAS_LIST) {
3918 if (flags & SHF_LARGE_LISTLEN)
3924 av_extend(av, len3 + 1); /* Leave room for [0] */
3925 AvFILLp(av) = len3; /* About to be filled anyway */
3929 TRACEME(("has %d object IDs to link", len3));
3932 * Read object-ID list into array.
3933 * Because we pre-extended it, we can cheat and fill it manually.
3935 * We read object tags and we can convert them into SV* on the fly
3936 * because we know all the references listed in there (as tags)
3937 * have been already serialized, hence we have a valid correspondance
3938 * between each of those tags and the recreated SV.
3942 SV **ary = AvARRAY(av);
3944 for (i = 1; i <= len3; i++) { /* We leave [0] alone */
3951 svh = av_fetch(cxt->aseen, tag, FALSE);
3953 CROAK(("Object #%"IVdf" should have been retrieved already",
3956 ary[i] = SvREFCNT_inc(xsv);
3961 * Bless the object and look up the STORABLE_thaw hook.
3965 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3968 * Hook not found. Maybe they did not require the module where this
3969 * hook is defined yet?
3971 * If the require below succeeds, we'll be able to find the hook.
3972 * Still, it only works reliably when each class is defined in a
3976 SV *psv = newSVpvn("require ", 8);
3977 sv_catpv(psv, class);
3979 TRACEME(("No STORABLE_thaw defined for objects of class %s", class));
3980 TRACEME(("Going to require module '%s' with '%s'", class, SvPVX(psv)));
3982 perl_eval_sv(psv, G_DISCARD);
3986 * We cache results of pkg_can, so we need to uncache before attempting
3990 pkg_uncache(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3991 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3994 CROAK(("No STORABLE_thaw defined for objects of class %s "
3995 "(even after a \"require %s;\")", class, class));
3999 * If we don't have an `av' yet, prepare one.
4000 * Then insert the frozen string as item [0].
4008 AvARRAY(av)[0] = SvREFCNT_inc(frozen);
4013 * $object->STORABLE_thaw($cloning, $frozen, @refs);
4015 * where $object is our blessed (empty) object, $cloning is a boolean
4016 * telling whether we're running a deep clone, $frozen is the frozen
4017 * string the user gave us in his serializing hook, and @refs, which may
4018 * be empty, is the list of extra references he returned along for us
4021 * In effect, the hook is an alternate creation routine for the class,
4022 * the object itself being already created by the runtime.
4025 TRACEME(("calling STORABLE_thaw on %s at 0x%"UVxf" (%"IVdf" args)",
4026 class, PTR2UV(sv), (IV) AvFILLp(av) + 1));
4029 (void) scalar_call(rv, hook, clone, av, G_SCALAR|G_DISCARD);
4036 SvREFCNT_dec(frozen);
4039 if (!(flags & SHF_IDX_CLASSNAME) && class != buf)
4043 * If we had an <extra> type, then the object was not as simple, and
4044 * we need to restore extra magic now.
4050 TRACEME(("retrieving magic object for 0x%"UVxf"...", PTR2UV(sv)));
4052 rv = retrieve(cxt, 0); /* Retrieve <magic object> */
4054 TRACEME(("restoring the magic object 0x%"UVxf" part of 0x%"UVxf,
4055 PTR2UV(rv), PTR2UV(sv)));
4057 switch (extra_type) {
4059 sv_upgrade(sv, SVt_PVMG);
4062 sv_upgrade(sv, SVt_PVAV);
4063 AvREAL_off((AV *)sv);
4066 sv_upgrade(sv, SVt_PVHV);
4069 CROAK(("Forgot to deal with extra type %d", extra_type));
4074 * Adding the magic only now, well after the STORABLE_thaw hook was called
4075 * means the hook cannot know it deals with an object whose variable is
4076 * tied. But this is happening when retrieving $o in the following case:
4080 * my $o = bless \%h, 'BAR';
4082 * The 'BAR' class is NOT the one where %h is tied into. Therefore, as
4083 * far as the 'BAR' class is concerned, the fact that %h is not a REAL
4084 * hash but a tied one should not matter at all, and remain transparent.
4085 * This means the magic must be restored by Storable AFTER the hook is
4088 * That looks very reasonable to me, but then I've come up with this
4089 * after a bug report from David Nesting, who was trying to store such
4090 * an object and caused Storable to fail. And unfortunately, it was
4091 * also the easiest way to retrofit support for blessed ref to tied objects
4092 * into the existing design. -- RAM, 17/02/2001
4095 sv_magic(sv, rv, mtype, Nullch, 0);
4096 SvREFCNT_dec(rv); /* Undo refcnt inc from sv_magic() */
4104 * Retrieve reference to some other scalar.
4105 * Layout is SX_REF <object>, with SX_REF already read.
4107 static SV *retrieve_ref(stcxt_t *cxt, char *cname)
4112 TRACEME(("retrieve_ref (#%d)", cxt->tagnum));
4115 * We need to create the SV that holds the reference to the yet-to-retrieve
4116 * object now, so that we may record the address in the seen table.
4117 * Otherwise, if the object to retrieve references us, we won't be able
4118 * to resolve the SX_OBJECT we'll see at that point! Hence we cannot
4119 * do the retrieve first and use rv = newRV(sv) since it will be too late
4120 * for SEEN() recording.
4123 rv = NEWSV(10002, 0);
4124 SEEN(rv, cname); /* Will return if rv is null */
4125 sv = retrieve(cxt, 0); /* Retrieve <object> */
4127 return (SV *) 0; /* Failed */
4130 * WARNING: breaks RV encapsulation.
4132 * Now for the tricky part. We have to upgrade our existing SV, so that
4133 * it is now an RV on sv... Again, we cheat by duplicating the code
4134 * held in newSVrv(), since we already got our SV from retrieve().
4138 * SvRV(rv) = SvREFCNT_inc(sv);
4140 * here because the reference count we got from retrieve() above is
4141 * already correct: if the object was retrieved from the file, then
4142 * its reference count is one. Otherwise, if it was retrieved via
4143 * an SX_OBJECT indication, a ref count increment was done.
4147 /* Do not use sv_upgrade to preserve STASH */
4148 SvFLAGS(rv) &= ~SVTYPEMASK;
4149 SvFLAGS(rv) |= SVt_RV;
4151 sv_upgrade(rv, SVt_RV);
4154 SvRV(rv) = sv; /* $rv = \$sv */
4157 TRACEME(("ok (retrieve_ref at 0x%"UVxf")", PTR2UV(rv)));
4163 * retrieve_overloaded
4165 * Retrieve reference to some other scalar with overloading.
4166 * Layout is SX_OVERLOAD <object>, with SX_OVERLOAD already read.
4168 static SV *retrieve_overloaded(stcxt_t *cxt, char *cname)
4174 TRACEME(("retrieve_overloaded (#%d)", cxt->tagnum));
4177 * Same code as retrieve_ref(), duplicated to avoid extra call.
4180 rv = NEWSV(10002, 0);
4181 SEEN(rv, cname); /* Will return if rv is null */
4182 sv = retrieve(cxt, 0); /* Retrieve <object> */
4184 return (SV *) 0; /* Failed */
4187 * WARNING: breaks RV encapsulation.
4190 sv_upgrade(rv, SVt_RV);
4191 SvRV(rv) = sv; /* $rv = \$sv */
4195 * Restore overloading magic.
4198 stash = (HV *) SvSTASH (sv);
4199 if (!stash || !Gv_AMG(stash))
4200 CROAK(("Cannot restore overloading on %s(0x%"UVxf") (package %s)",
4201 sv_reftype(sv, FALSE),
4203 stash ? HvNAME(stash) : "<unknown>"));
4207 TRACEME(("ok (retrieve_overloaded at 0x%"UVxf")", PTR2UV(rv)));
4213 * retrieve_tied_array
4215 * Retrieve tied array
4216 * Layout is SX_TIED_ARRAY <object>, with SX_TIED_ARRAY already read.
4218 static SV *retrieve_tied_array(stcxt_t *cxt, char *cname)
4223 TRACEME(("retrieve_tied_array (#%d)", cxt->tagnum));
4225 tv = NEWSV(10002, 0);
4226 SEEN(tv, cname); /* Will return if tv is null */
4227 sv = retrieve(cxt, 0); /* Retrieve <object> */
4229 return (SV *) 0; /* Failed */
4231 sv_upgrade(tv, SVt_PVAV);
4232 AvREAL_off((AV *)tv);
4233 sv_magic(tv, sv, 'P', Nullch, 0);
4234 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4236 TRACEME(("ok (retrieve_tied_array at 0x%"UVxf")", PTR2UV(tv)));
4242 * retrieve_tied_hash
4244 * Retrieve tied hash
4245 * Layout is SX_TIED_HASH <object>, with SX_TIED_HASH already read.
4247 static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname)
4252 TRACEME(("retrieve_tied_hash (#%d)", cxt->tagnum));
4254 tv = NEWSV(10002, 0);
4255 SEEN(tv, cname); /* Will return if tv is null */
4256 sv = retrieve(cxt, 0); /* Retrieve <object> */
4258 return (SV *) 0; /* Failed */
4260 sv_upgrade(tv, SVt_PVHV);
4261 sv_magic(tv, sv, 'P', Nullch, 0);
4262 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4264 TRACEME(("ok (retrieve_tied_hash at 0x%"UVxf")", PTR2UV(tv)));
4270 * retrieve_tied_scalar
4272 * Retrieve tied scalar
4273 * Layout is SX_TIED_SCALAR <object>, with SX_TIED_SCALAR already read.
4275 static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname)
4278 SV *sv, *obj = NULL;
4280 TRACEME(("retrieve_tied_scalar (#%d)", cxt->tagnum));
4282 tv = NEWSV(10002, 0);
4283 SEEN(tv, cname); /* Will return if rv is null */
4284 sv = retrieve(cxt, 0); /* Retrieve <object> */
4286 return (SV *) 0; /* Failed */
4288 else if (SvTYPE(sv) != SVt_NULL) {
4292 sv_upgrade(tv, SVt_PVMG);
4293 sv_magic(tv, obj, 'q', Nullch, 0);
4296 /* Undo refcnt inc from sv_magic() */
4300 TRACEME(("ok (retrieve_tied_scalar at 0x%"UVxf")", PTR2UV(tv)));
4308 * Retrieve reference to value in a tied hash.
4309 * Layout is SX_TIED_KEY <object> <key>, with SX_TIED_KEY already read.
4311 static SV *retrieve_tied_key(stcxt_t *cxt, char *cname)
4317 TRACEME(("retrieve_tied_key (#%d)", cxt->tagnum));
4319 tv = NEWSV(10002, 0);
4320 SEEN(tv, cname); /* Will return if tv is null */
4321 sv = retrieve(cxt, 0); /* Retrieve <object> */
4323 return (SV *) 0; /* Failed */
4325 key = retrieve(cxt, 0); /* Retrieve <key> */
4327 return (SV *) 0; /* Failed */
4329 sv_upgrade(tv, SVt_PVMG);
4330 sv_magic(tv, sv, 'p', (char *)key, HEf_SVKEY);
4331 SvREFCNT_dec(key); /* Undo refcnt inc from sv_magic() */
4332 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4340 * Retrieve reference to value in a tied array.
4341 * Layout is SX_TIED_IDX <object> <idx>, with SX_TIED_IDX already read.
4343 static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname)
4349 TRACEME(("retrieve_tied_idx (#%d)", cxt->tagnum));
4351 tv = NEWSV(10002, 0);
4352 SEEN(tv, cname); /* Will return if tv is null */
4353 sv = retrieve(cxt, 0); /* Retrieve <object> */
4355 return (SV *) 0; /* Failed */
4357 RLEN(idx); /* Retrieve <idx> */
4359 sv_upgrade(tv, SVt_PVMG);
4360 sv_magic(tv, sv, 'p', Nullch, idx);
4361 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4370 * Retrieve defined long (string) scalar.
4372 * Layout is SX_LSCALAR <length> <data>, with SX_LSCALAR already read.
4373 * The scalar is "long" in that <length> is larger than LG_SCALAR so it
4374 * was not stored on a single byte.
4376 static SV *retrieve_lscalar(stcxt_t *cxt, char *cname)
4382 TRACEME(("retrieve_lscalar (#%d), len = %"IVdf, cxt->tagnum, (IV) len));
4385 * Allocate an empty scalar of the suitable length.
4388 sv = NEWSV(10002, len);
4389 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4392 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
4394 * Now, for efficiency reasons, read data directly inside the SV buffer,
4395 * and perform the SV final settings directly by duplicating the final
4396 * work done by sv_setpv. Since we're going to allocate lots of scalars
4397 * this way, it's worth the hassle and risk.
4400 SAFEREAD(SvPVX(sv), len, sv);
4401 SvCUR_set(sv, len); /* Record C string length */
4402 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
4403 (void) SvPOK_only(sv); /* Validate string pointer */
4404 if (cxt->s_tainted) /* Is input source tainted? */
4405 SvTAINT(sv); /* External data cannot be trusted */
4407 TRACEME(("large scalar len %"IVdf" '%s'", (IV) len, SvPVX(sv)));
4408 TRACEME(("ok (retrieve_lscalar at 0x%"UVxf")", PTR2UV(sv)));
4416 * Retrieve defined short (string) scalar.
4418 * Layout is SX_SCALAR <length> <data>, with SX_SCALAR already read.
4419 * The scalar is "short" so <length> is single byte. If it is 0, there
4420 * is no <data> section.
4422 static SV *retrieve_scalar(stcxt_t *cxt, char *cname)
4428 TRACEME(("retrieve_scalar (#%d), len = %d", cxt->tagnum, len));
4431 * Allocate an empty scalar of the suitable length.
4434 sv = NEWSV(10002, len);
4435 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4438 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
4443 * newSV did not upgrade to SVt_PV so the scalar is undefined.
4444 * To make it defined with an empty length, upgrade it now...
4445 * Don't upgrade to a PV if the original type contains more
4446 * information than a scalar.
4448 if (SvTYPE(sv) <= SVt_PV) {
4449 sv_upgrade(sv, SVt_PV);
4452 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
4453 TRACEME(("ok (retrieve_scalar empty at 0x%"UVxf")", PTR2UV(sv)));
4456 * Now, for efficiency reasons, read data directly inside the SV buffer,
4457 * and perform the SV final settings directly by duplicating the final
4458 * work done by sv_setpv. Since we're going to allocate lots of scalars
4459 * this way, it's worth the hassle and risk.
4461 SAFEREAD(SvPVX(sv), len, sv);
4462 SvCUR_set(sv, len); /* Record C string length */
4463 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
4464 TRACEME(("small scalar len %d '%s'", len, SvPVX(sv)));
4467 (void) SvPOK_only(sv); /* Validate string pointer */
4468 if (cxt->s_tainted) /* Is input source tainted? */
4469 SvTAINT(sv); /* External data cannot be trusted */
4471 TRACEME(("ok (retrieve_scalar at 0x%"UVxf")", PTR2UV(sv)));
4478 * Like retrieve_scalar(), but tag result as utf8.
4479 * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
4481 static SV *retrieve_utf8str(stcxt_t *cxt, char *cname)
4485 TRACEME(("retrieve_utf8str"));
4487 sv = retrieve_scalar(cxt, cname);
4489 #ifdef HAS_UTF8_SCALARS
4492 if (cxt->use_bytes < 0)
4494 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4496 if (cxt->use_bytes == 0)
4507 * Like retrieve_lscalar(), but tag result as utf8.
4508 * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
4510 static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname)
4514 TRACEME(("retrieve_lutf8str"));
4516 sv = retrieve_lscalar(cxt, cname);
4518 #ifdef HAS_UTF8_SCALARS
4521 if (cxt->use_bytes < 0)
4523 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4525 if (cxt->use_bytes == 0)
4535 * Retrieve defined integer.
4536 * Layout is SX_INTEGER <data>, whith SX_INTEGER already read.
4538 static SV *retrieve_integer(stcxt_t *cxt, char *cname)
4543 TRACEME(("retrieve_integer (#%d)", cxt->tagnum));
4545 READ(&iv, sizeof(iv));
4547 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4549 TRACEME(("integer %"IVdf, iv));
4550 TRACEME(("ok (retrieve_integer at 0x%"UVxf")", PTR2UV(sv)));
4558 * Retrieve defined integer in network order.
4559 * Layout is SX_NETINT <data>, whith SX_NETINT already read.
4561 static SV *retrieve_netint(stcxt_t *cxt, char *cname)
4566 TRACEME(("retrieve_netint (#%d)", cxt->tagnum));
4570 sv = newSViv((int) ntohl(iv));
4571 TRACEME(("network integer %d", (int) ntohl(iv)));
4574 TRACEME(("network integer (as-is) %d", iv));
4576 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4578 TRACEME(("ok (retrieve_netint at 0x%"UVxf")", PTR2UV(sv)));
4586 * Retrieve defined double.
4587 * Layout is SX_DOUBLE <data>, whith SX_DOUBLE already read.
4589 static SV *retrieve_double(stcxt_t *cxt, char *cname)
4594 TRACEME(("retrieve_double (#%d)", cxt->tagnum));
4596 READ(&nv, sizeof(nv));
4598 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4600 TRACEME(("double %"NVff, nv));
4601 TRACEME(("ok (retrieve_double at 0x%"UVxf")", PTR2UV(sv)));
4609 * Retrieve defined byte (small integer within the [-128, +127] range).
4610 * Layout is SX_BYTE <data>, whith SX_BYTE already read.
4612 static SV *retrieve_byte(stcxt_t *cxt, char *cname)
4616 signed char tmp; /* Workaround for AIX cc bug --H.Merijn Brand */
4618 TRACEME(("retrieve_byte (#%d)", cxt->tagnum));
4621 TRACEME(("small integer read as %d", (unsigned char) siv));
4622 tmp = (unsigned char) siv - 128;
4624 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4626 TRACEME(("byte %d", tmp));
4627 TRACEME(("ok (retrieve_byte at 0x%"UVxf")", PTR2UV(sv)));
4635 * Return the undefined value.
4637 static SV *retrieve_undef(stcxt_t *cxt, char *cname)
4641 TRACEME(("retrieve_undef"));
4652 * Return the immortal undefined value.
4654 static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname)
4656 SV *sv = &PL_sv_undef;
4658 TRACEME(("retrieve_sv_undef"));
4667 * Return the immortal yes value.
4669 static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname)
4671 SV *sv = &PL_sv_yes;
4673 TRACEME(("retrieve_sv_yes"));
4682 * Return the immortal no value.
4684 static SV *retrieve_sv_no(stcxt_t *cxt, char *cname)
4688 TRACEME(("retrieve_sv_no"));
4697 * Retrieve a whole array.
4698 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
4699 * Each item is stored as <object>.
4701 * When we come here, SX_ARRAY has been read already.
4703 static SV *retrieve_array(stcxt_t *cxt, char *cname)
4710 TRACEME(("retrieve_array (#%d)", cxt->tagnum));
4713 * Read length, and allocate array, then pre-extend it.
4717 TRACEME(("size = %d", len));
4719 SEEN(av, cname); /* Will return if array not allocated nicely */
4723 return (SV *) av; /* No data follow if array is empty */
4726 * Now get each item in turn...
4729 for (i = 0; i < len; i++) {
4730 TRACEME(("(#%d) item", i));
4731 sv = retrieve(cxt, 0); /* Retrieve item */
4734 if (av_store(av, i, sv) == 0)
4738 TRACEME(("ok (retrieve_array at 0x%"UVxf")", PTR2UV(av)));
4746 * Retrieve a whole hash table.
4747 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4748 * Keys are stored as <length> <data>, the <data> section being omitted
4750 * Values are stored as <object>.
4752 * When we come here, SX_HASH has been read already.
4754 static SV *retrieve_hash(stcxt_t *cxt, char *cname)
4762 TRACEME(("retrieve_hash (#%d)", cxt->tagnum));
4765 * Read length, allocate table.
4769 TRACEME(("size = %d", len));
4771 SEEN(hv, cname); /* Will return if table not allocated properly */
4773 return (SV *) hv; /* No data follow if table empty */
4774 hv_ksplit(hv, len); /* pre-extend hash to save multiple splits */
4777 * Now get each key/value pair in turn...
4780 for (i = 0; i < len; i++) {
4785 TRACEME(("(#%d) value", i));
4786 sv = retrieve(cxt, 0);
4792 * Since we're reading into kbuf, we must ensure we're not
4793 * recursing between the read and the hv_store() where it's used.
4794 * Hence the key comes after the value.
4797 RLEN(size); /* Get key size */
4798 KBUFCHK((STRLEN)size); /* Grow hash key read pool if needed */
4801 kbuf[size] = '\0'; /* Mark string end, just in case */
4802 TRACEME(("(#%d) key '%s'", i, kbuf));
4805 * Enter key/value pair into hash table.
4808 if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
4812 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4820 * Retrieve a whole hash table.
4821 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4822 * Keys are stored as <length> <data>, the <data> section being omitted
4824 * Values are stored as <object>.
4826 * When we come here, SX_HASH has been read already.
4828 static SV *retrieve_flag_hash(stcxt_t *cxt, char *cname)
4837 GETMARK(hash_flags);
4838 TRACEME(("retrieve_flag_hash (#%d)", cxt->tagnum));
4840 * Read length, allocate table.
4843 #ifndef HAS_RESTRICTED_HASHES
4844 if (hash_flags & SHV_RESTRICTED) {
4845 if (cxt->derestrict < 0)
4847 = (SvTRUE(perl_get_sv("Storable::downgrade_restricted", TRUE))
4849 if (cxt->derestrict == 0)
4850 RESTRICTED_HASH_CROAK();
4855 TRACEME(("size = %d, flags = %d", len, hash_flags));
4857 SEEN(hv, cname); /* Will return if table not allocated properly */
4859 return (SV *) hv; /* No data follow if table empty */
4860 hv_ksplit(hv, len); /* pre-extend hash to save multiple splits */
4863 * Now get each key/value pair in turn...
4866 for (i = 0; i < len; i++) {
4868 int store_flags = 0;
4873 TRACEME(("(#%d) value", i));
4874 sv = retrieve(cxt, 0);
4879 #ifdef HAS_RESTRICTED_HASHES
4880 if ((hash_flags & SHV_RESTRICTED) && (flags & SHV_K_LOCKED))
4884 if (flags & SHV_K_ISSV) {
4885 /* XXX you can't set a placeholder with an SV key.
4886 Then again, you can't get an SV key.
4887 Without messing around beyond what the API is supposed to do.
4890 TRACEME(("(#%d) keysv, flags=%d", i, flags));
4891 keysv = retrieve(cxt, 0);
4895 if (!hv_store_ent(hv, keysv, sv, 0))
4900 * Since we're reading into kbuf, we must ensure we're not
4901 * recursing between the read and the hv_store() where it's used.
4902 * Hence the key comes after the value.
4905 if (flags & SHV_K_PLACEHOLDER) {
4907 sv = &PL_sv_placeholder;
4908 store_flags |= HVhek_PLACEHOLD;
4910 if (flags & SHV_K_UTF8) {
4911 #ifdef HAS_UTF8_HASHES
4912 store_flags |= HVhek_UTF8;
4914 if (cxt->use_bytes < 0)
4916 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4918 if (cxt->use_bytes == 0)
4922 #ifdef HAS_UTF8_HASHES
4923 if (flags & SHV_K_WASUTF8)
4924 store_flags |= HVhek_WASUTF8;
4927 RLEN(size); /* Get key size */
4928 KBUFCHK((STRLEN)size); /* Grow hash key read pool if needed */
4931 kbuf[size] = '\0'; /* Mark string end, just in case */
4932 TRACEME(("(#%d) key '%s' flags %X store_flags %X", i, kbuf,
4933 flags, store_flags));
4936 * Enter key/value pair into hash table.
4939 #ifdef HAS_RESTRICTED_HASHES
4940 if (hv_store_flags(hv, kbuf, size, sv, 0, flags) == 0)
4943 if (!(store_flags & HVhek_PLACEHOLD))
4944 if (hv_store(hv, kbuf, size, sv, 0) == 0)
4949 #ifdef HAS_RESTRICTED_HASHES
4950 if (hash_flags & SHV_RESTRICTED)
4954 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4962 * Return a code reference.
4964 static SV *retrieve_code(stcxt_t *cxt, char *cname)
4966 #if PERL_VERSION < 6
4967 CROAK(("retrieve_code does not work with perl 5.005 or less\n"));
4972 SV *sv, *text, *sub;
4974 TRACEME(("retrieve_code (#%d)", cxt->tagnum));
4977 * Retrieve the source of the code reference
4978 * as a small or large scalar
4984 text = retrieve_scalar(cxt, cname);
4987 text = retrieve_lscalar(cxt, cname);
4990 CROAK(("Unexpected type %d in retrieve_code\n", type));
4994 * prepend "sub " to the source
4997 sub = newSVpvn("sub ", 4);
4998 sv_catpv(sub, SvPV_nolen(text)); /* XXX no sv_catsv! */
5002 * evaluate the source to a code reference and use the CV value
5005 if (cxt->eval == NULL) {
5006 cxt->eval = perl_get_sv("Storable::Eval", TRUE);
5007 SvREFCNT_inc(cxt->eval);
5009 if (!SvTRUE(cxt->eval)) {
5011 cxt->forgive_me == 0 ||
5012 (cxt->forgive_me < 0 && !(cxt->forgive_me =
5013 SvTRUE(perl_get_sv("Storable::forgive_me", TRUE)) ? 1 : 0))
5015 CROAK(("Can't eval, please set $Storable::Eval to a true value"));
5025 if (SvROK(cxt->eval) && SvTYPE(SvRV(cxt->eval)) == SVt_PVCV) {
5026 SV* errsv = get_sv("@", TRUE);
5027 sv_setpv(errsv, ""); /* clear $@ */
5029 XPUSHs(sv_2mortal(newSVsv(sub)));
5031 count = call_sv(cxt->eval, G_SCALAR);
5034 CROAK(("Unexpected return value from $Storable::Eval callback\n"));
5036 if (SvTRUE(errsv)) {
5037 CROAK(("code %s caused an error: %s",
5038 SvPV_nolen(sub), SvPV_nolen(errsv)));
5042 cv = eval_pv(SvPV_nolen(sub), TRUE);
5044 if (cv && SvROK(cv) && SvTYPE(SvRV(cv)) == SVt_PVCV) {
5047 CROAK(("code %s did not evaluate to a subroutine reference\n", SvPV_nolen(sub)));
5050 SvREFCNT_inc(sv); /* XXX seems to be necessary */
5062 * old_retrieve_array
5064 * Retrieve a whole array in pre-0.6 binary format.
5066 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
5067 * Each item is stored as SX_ITEM <object> or SX_IT_UNDEF for "holes".
5069 * When we come here, SX_ARRAY has been read already.
5071 static SV *old_retrieve_array(stcxt_t *cxt, char *cname)
5079 TRACEME(("old_retrieve_array (#%d)", cxt->tagnum));
5082 * Read length, and allocate array, then pre-extend it.
5086 TRACEME(("size = %d", len));
5088 SEEN(av, 0); /* Will return if array not allocated nicely */
5092 return (SV *) av; /* No data follow if array is empty */
5095 * Now get each item in turn...
5098 for (i = 0; i < len; i++) {
5100 if (c == SX_IT_UNDEF) {
5101 TRACEME(("(#%d) undef item", i));
5102 continue; /* av_extend() already filled us with undef */
5105 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
5106 TRACEME(("(#%d) item", i));
5107 sv = retrieve(cxt, 0); /* Retrieve item */
5110 if (av_store(av, i, sv) == 0)
5114 TRACEME(("ok (old_retrieve_array at 0x%"UVxf")", PTR2UV(av)));
5122 * Retrieve a whole hash table in pre-0.6 binary format.
5124 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
5125 * Keys are stored as SX_KEY <length> <data>, the <data> section being omitted
5127 * Values are stored as SX_VALUE <object> or SX_VL_UNDEF for "holes".
5129 * When we come here, SX_HASH has been read already.
5131 static SV *old_retrieve_hash(stcxt_t *cxt, char *cname)
5139 static SV *sv_h_undef = (SV *) 0; /* hv_store() bug */
5141 TRACEME(("old_retrieve_hash (#%d)", cxt->tagnum));
5144 * Read length, allocate table.
5148 TRACEME(("size = %d", len));
5150 SEEN(hv, 0); /* Will return if table not allocated properly */
5152 return (SV *) hv; /* No data follow if table empty */
5153 hv_ksplit(hv, len); /* pre-extend hash to save multiple splits */
5156 * Now get each key/value pair in turn...
5159 for (i = 0; i < len; i++) {
5165 if (c == SX_VL_UNDEF) {
5166 TRACEME(("(#%d) undef value", i));
5168 * Due to a bug in hv_store(), it's not possible to pass
5169 * &PL_sv_undef to hv_store() as a value, otherwise the
5170 * associated key will not be creatable any more. -- RAM, 14/01/97
5173 sv_h_undef = newSVsv(&PL_sv_undef);
5174 sv = SvREFCNT_inc(sv_h_undef);
5175 } else if (c == SX_VALUE) {
5176 TRACEME(("(#%d) value", i));
5177 sv = retrieve(cxt, 0);
5181 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
5185 * Since we're reading into kbuf, we must ensure we're not
5186 * recursing between the read and the hv_store() where it's used.
5187 * Hence the key comes after the value.
5192 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
5193 RLEN(size); /* Get key size */
5194 KBUFCHK((STRLEN)size); /* Grow hash key read pool if needed */
5197 kbuf[size] = '\0'; /* Mark string end, just in case */
5198 TRACEME(("(#%d) key '%s'", i, kbuf));
5201 * Enter key/value pair into hash table.
5204 if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
5208 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
5214 *** Retrieval engine.
5220 * Make sure the stored data we're trying to retrieve has been produced
5221 * on an ILP compatible system with the same byteorder. It croaks out in
5222 * case an error is detected. [ILP = integer-long-pointer sizes]
5223 * Returns null if error is detected, &PL_sv_undef otherwise.
5225 * Note that there's no byte ordering info emitted when network order was
5226 * used at store time.
5228 static SV *magic_check(stcxt_t *cxt)
5230 /* The worst case for a malicious header would be old magic (which is
5231 longer), major, minor, byteorder length byte of 255, 255 bytes of
5232 garbage, sizeof int, long, pointer, NV.
5233 So the worse of that we can read is 255 bytes of garbage plus 4.
5234 Err, I am assuming 8 bit bytes here. Please file a bug report if you're
5235 compiling perl on a system with chars that are larger than 8 bits.
5236 (Even Crays aren't *that* perverse).
5238 unsigned char buf[4 + 255];
5239 unsigned char *current;
5242 int use_network_order;
5245 int version_minor = 0;
5247 TRACEME(("magic_check"));
5250 * The "magic number" is only for files, not when freezing in memory.
5254 /* This includes the '\0' at the end. I want to read the extra byte,
5255 which is usually going to be the major version number. */
5256 STRLEN len = sizeof(magicstr);
5259 READ(buf, (SSize_t)(len)); /* Not null-terminated */
5261 /* Point at the byte after the byte we read. */
5262 current = buf + --len; /* Do the -- outside of macros. */
5264 if (memNE(buf, magicstr, len)) {
5266 * Try to read more bytes to check for the old magic number, which
5270 TRACEME(("trying for old magic number"));
5272 old_len = sizeof(old_magicstr) - 1;
5273 READ(current + 1, (SSize_t)(old_len - len));
5275 if (memNE(buf, old_magicstr, old_len))
5276 CROAK(("File is not a perl storable"));
5277 current = buf + old_len;
5279 use_network_order = *current;
5281 GETMARK(use_network_order);
5284 * Starting with 0.6, the "use_network_order" byte flag is also used to
5285 * indicate the version number of the binary, and therefore governs the
5286 * setting of sv_retrieve_vtbl. See magic_write().
5289 version_major = use_network_order >> 1;
5290 cxt->retrieve_vtbl = version_major ? sv_retrieve : sv_old_retrieve;
5292 TRACEME(("magic_check: netorder = 0x%x", use_network_order));
5296 * Starting with 0.7 (binary major 2), a full byte is dedicated to the
5297 * minor version of the protocol. See magic_write().
5300 if (version_major > 1)
5301 GETMARK(version_minor);
5303 cxt->ver_major = version_major;
5304 cxt->ver_minor = version_minor;
5306 TRACEME(("binary image version is %d.%d", version_major, version_minor));
5309 * Inter-operability sanity check: we can't retrieve something stored
5310 * using a format more recent than ours, because we have no way to
5311 * know what has changed, and letting retrieval go would mean a probable
5312 * failure reporting a "corrupted" storable file.
5316 version_major > STORABLE_BIN_MAJOR ||
5317 (version_major == STORABLE_BIN_MAJOR &&
5318 version_minor > STORABLE_BIN_MINOR)
5321 TRACEME(("but I am version is %d.%d", STORABLE_BIN_MAJOR,
5322 STORABLE_BIN_MINOR));
5324 if (version_major == STORABLE_BIN_MAJOR) {
5325 TRACEME(("cxt->accept_future_minor is %d",
5326 cxt->accept_future_minor));
5327 if (cxt->accept_future_minor < 0)
5328 cxt->accept_future_minor
5329 = (SvTRUE(perl_get_sv("Storable::accept_future_minor",
5332 if (cxt->accept_future_minor == 1)
5333 croak_now = 0; /* Don't croak yet. */
5336 CROAK(("Storable binary image v%d.%d more recent than I am (v%d.%d)",
5337 version_major, version_minor,
5338 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
5343 * If they stored using network order, there's no byte ordering
5344 * information to check.
5347 if ((cxt->netorder = (use_network_order & 0x1))) /* Extra () for -Wall */
5348 return &PL_sv_undef; /* No byte ordering info */
5350 /* In C truth is 1, falsehood is 0. Very convienient. */
5351 use_NV_size = version_major >= 2 && version_minor >= 2;
5354 length = c + 3 + use_NV_size;
5355 READ(buf, length); /* Not null-terminated */
5357 TRACEME(("byte order '%.*s' %d", c, buf, c));
5359 #ifdef USE_56_INTERWORK_KLUDGE
5360 /* No point in caching this in the context as we only need it once per
5361 retrieve, and we need to recheck it each read. */
5362 if (SvTRUE(perl_get_sv("Storable::interwork_56_64bit", TRUE))) {
5363 if ((c != (sizeof (byteorderstr_56) - 1))
5364 || memNE(buf, byteorderstr_56, c))
5365 CROAK(("Byte order is not compatible"));
5369 if ((c != (sizeof (byteorderstr) - 1)) || memNE(buf, byteorderstr, c))
5370 CROAK(("Byte order is not compatible"));
5376 if ((int) *current++ != sizeof(int))
5377 CROAK(("Integer size is not compatible"));
5380 if ((int) *current++ != sizeof(long))
5381 CROAK(("Long integer size is not compatible"));
5383 /* sizeof(char *) */
5384 if ((int) *current != sizeof(char *))
5385 CROAK(("Pointer size is not compatible"));
5389 if ((int) *++current != sizeof(NV))
5390 CROAK(("Double size is not compatible"));
5393 return &PL_sv_undef; /* OK */
5399 * Recursively retrieve objects from the specified file and return their
5400 * root SV (which may be an AV or an HV for what we care).
5401 * Returns null if there is a problem.
5403 static SV *retrieve(stcxt_t *cxt, char *cname)
5409 TRACEME(("retrieve"));
5412 * Grab address tag which identifies the object if we are retrieving
5413 * an older format. Since the new binary format counts objects and no
5414 * longer explicitely tags them, we must keep track of the correspondance
5417 * The following section will disappear one day when the old format is
5418 * no longer supported, hence the final "goto" in the "if" block.
5421 if (cxt->hseen) { /* Retrieving old binary */
5423 if (cxt->netorder) {
5425 READ(&nettag, sizeof(I32)); /* Ordered sequence of I32 */
5426 tag = (stag_t) nettag;
5428 READ(&tag, sizeof(stag_t)); /* Original address of the SV */
5431 if (type == SX_OBJECT) {
5433 svh = hv_fetch(cxt->hseen, (char *) &tag, sizeof(tag), FALSE);
5435 CROAK(("Old tag 0x%"UVxf" should have been mapped already",
5437 tagn = SvIV(*svh); /* Mapped tag number computed earlier below */
5440 * The following code is common with the SX_OBJECT case below.
5443 svh = av_fetch(cxt->aseen, tagn, FALSE);
5445 CROAK(("Object #%"IVdf" should have been retrieved already",
5448 TRACEME(("has retrieved #%d at 0x%"UVxf, tagn, PTR2UV(sv)));
5449 SvREFCNT_inc(sv); /* One more reference to this same sv */
5450 return sv; /* The SV pointer where object was retrieved */
5454 * Map new object, but don't increase tagnum. This will be done
5455 * by each of the retrieve_* functions when they call SEEN().
5457 * The mapping associates the "tag" initially present with a unique
5458 * tag number. See test for SX_OBJECT above to see how this is perused.
5461 if (!hv_store(cxt->hseen, (char *) &tag, sizeof(tag),
5462 newSViv(cxt->tagnum), 0))
5469 * Regular post-0.6 binary format.
5474 TRACEME(("retrieve type = %d", type));
5477 * Are we dealing with an object we should have already retrieved?
5480 if (type == SX_OBJECT) {
5484 svh = av_fetch(cxt->aseen, tag, FALSE);
5486 CROAK(("Object #%"IVdf" should have been retrieved already",
5489 TRACEME(("had retrieved #%d at 0x%"UVxf, tag, PTR2UV(sv)));
5490 SvREFCNT_inc(sv); /* One more reference to this same sv */
5491 return sv; /* The SV pointer where object was retrieved */
5492 } else if (type >= SX_ERROR && cxt->ver_minor > STORABLE_BIN_MINOR) {
5493 if (cxt->accept_future_minor < 0)
5494 cxt->accept_future_minor
5495 = (SvTRUE(perl_get_sv("Storable::accept_future_minor",
5498 if (cxt->accept_future_minor == 1) {
5499 CROAK(("Storable binary image v%d.%d contains data of type %d. "
5500 "This Storable is v%d.%d and can only handle data types up to %d",
5501 cxt->ver_major, cxt->ver_minor, type,
5502 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR, SX_ERROR - 1));
5506 first_time: /* Will disappear when support for old format is dropped */
5509 * Okay, first time through for this one.
5512 sv = RETRIEVE(cxt, type)(cxt, cname);
5514 return (SV *) 0; /* Failed */
5517 * Old binary formats (pre-0.7).
5519 * Final notifications, ended by SX_STORED may now follow.
5520 * Currently, the only pertinent notification to apply on the
5521 * freshly retrieved object is either:
5522 * SX_CLASS <char-len> <classname> for short classnames.
5523 * SX_LG_CLASS <int-len> <classname> for larger one (rare!).
5524 * Class name is then read into the key buffer pool used by
5525 * hash table key retrieval.
5528 if (cxt->ver_major < 2) {
5529 while ((type = GETCHAR()) != SX_STORED) {
5533 GETMARK(len); /* Length coded on a single char */
5535 case SX_LG_CLASS: /* Length coded on a regular integer */
5540 return (SV *) 0; /* Failed */
5542 KBUFCHK((STRLEN)len); /* Grow buffer as necessary */
5545 kbuf[len] = '\0'; /* Mark string end */
5550 TRACEME(("ok (retrieved 0x%"UVxf", refcnt=%d, %s)", PTR2UV(sv),
5551 SvREFCNT(sv) - 1, sv_reftype(sv, FALSE)));
5559 * Retrieve data held in file and return the root object.
5560 * Common routine for pretrieve and mretrieve.
5562 static SV *do_retrieve(
5569 int is_tainted; /* Is input source tainted? */
5570 int pre_06_fmt = 0; /* True with pre Storable 0.6 formats */
5572 TRACEME(("do_retrieve (optype = 0x%x)", optype));
5574 optype |= ST_RETRIEVE;
5577 * Sanity assertions for retrieve dispatch tables.
5580 ASSERT(sizeof(sv_old_retrieve) == sizeof(sv_retrieve),
5581 ("old and new retrieve dispatch table have same size"));
5582 ASSERT(sv_old_retrieve[SX_ERROR] == retrieve_other,
5583 ("SX_ERROR entry correctly initialized in old dispatch table"));
5584 ASSERT(sv_retrieve[SX_ERROR] == retrieve_other,
5585 ("SX_ERROR entry correctly initialized in new dispatch table"));
5588 * Workaround for CROAK leak: if they enter with a "dirty" context,
5589 * free up memory for them now.
5596 * Now that STORABLE_xxx hooks exist, it is possible that they try to
5597 * re-enter retrieve() via the hooks.
5601 cxt = allocate_context(cxt);
5605 ASSERT(cxt->entry == 1, ("starting new recursion"));
5606 ASSERT(!cxt->s_dirty, ("clean context"));
5611 * Data is loaded into the memory buffer when f is NULL, unless `in' is
5612 * also NULL, in which case we're expecting the data to already lie
5613 * in the buffer (dclone case).
5616 KBUFINIT(); /* Allocate hash key reading pool once */
5619 MBUF_SAVE_AND_LOAD(in);
5622 * Magic number verifications.
5624 * This needs to be done before calling init_retrieve_context()
5625 * since the format indication in the file are necessary to conduct
5626 * some of the initializations.
5629 cxt->fio = f; /* Where I/O are performed */
5631 if (!magic_check(cxt))
5632 CROAK(("Magic number checking on storable %s failed",
5633 cxt->fio ? "file" : "string"));
5635 TRACEME(("data stored in %s format",
5636 cxt->netorder ? "net order" : "native"));
5639 * Check whether input source is tainted, so that we don't wrongly
5640 * taint perfectly good values...
5642 * We assume file input is always tainted. If both `f' and `in' are
5643 * NULL, then we come from dclone, and tainted is already filled in
5644 * the context. That's a kludge, but the whole dclone() thing is
5645 * already quite a kludge anyway! -- RAM, 15/09/2000.
5648 is_tainted = f ? 1 : (in ? SvTAINTED(in) : cxt->s_tainted);
5649 TRACEME(("input source is %s", is_tainted ? "tainted" : "trusted"));
5650 init_retrieve_context(cxt, optype, is_tainted);
5652 ASSERT(is_retrieving(), ("within retrieve operation"));
5654 sv = retrieve(cxt, 0); /* Recursively retrieve object, get root SV */
5663 pre_06_fmt = cxt->hseen != NULL; /* Before we clean context */
5666 * The "root" context is never freed.
5669 clean_retrieve_context(cxt);
5670 if (cxt->prev) /* This context was stacked */
5671 free_context(cxt); /* It was not the "root" context */
5674 * Prepare returned value.
5678 TRACEME(("retrieve ERROR"));
5679 #if (PATCHLEVEL <= 4)
5680 /* perl 5.00405 seems to screw up at this point with an
5681 'attempt to modify a read only value' error reported in the
5682 eval { $self = pretrieve(*FILE) } in _retrieve.
5683 I can't see what the cause of this error is, but I suspect a
5684 bug in 5.004, as it seems to be capable of issuing spurious
5685 errors or core dumping with matches on $@. I'm not going to
5686 spend time on what could be a fruitless search for the cause,
5687 so here's a bodge. If you're running 5.004 and don't like
5688 this inefficiency, either upgrade to a newer perl, or you are
5689 welcome to find the problem and send in a patch.
5693 return &PL_sv_undef; /* Something went wrong, return undef */
5697 TRACEME(("retrieve got %s(0x%"UVxf")",
5698 sv_reftype(sv, FALSE), PTR2UV(sv)));
5701 * Backward compatibility with Storable-0.5@9 (which we know we
5702 * are retrieving if hseen is non-null): don't create an extra RV
5703 * for objects since we special-cased it at store time.
5705 * Build a reference to the SV returned by pretrieve even if it is
5706 * already one and not a scalar, for consistency reasons.
5709 if (pre_06_fmt) { /* Was not handling overloading by then */
5711 TRACEME(("fixing for old formats -- pre 0.6"));
5712 if (sv_type(sv) == svis_REF && (rv = SvRV(sv)) && SvOBJECT(rv)) {
5713 TRACEME(("ended do_retrieve() with an object -- pre 0.6"));
5719 * If reference is overloaded, restore behaviour.
5721 * NB: minor glitch here: normally, overloaded refs are stored specially
5722 * so that we can croak when behaviour cannot be re-installed, and also
5723 * avoid testing for overloading magic at each reference retrieval.
5725 * Unfortunately, the root reference is implicitely stored, so we must
5726 * check for possible overloading now. Furthermore, if we don't restore
5727 * overloading, we cannot croak as if the original ref was, because we
5728 * have no way to determine whether it was an overloaded ref or not in
5731 * It's a pity that overloading magic is attached to the rv, and not to
5732 * the underlying sv as blessing is.
5736 HV *stash = (HV *) SvSTASH(sv);
5737 SV *rv = newRV_noinc(sv);
5738 if (stash && Gv_AMG(stash)) {
5740 TRACEME(("restored overloading on root reference"));
5742 TRACEME(("ended do_retrieve() with an object"));
5746 TRACEME(("regular do_retrieve() end"));
5748 return newRV_noinc(sv);
5754 * Retrieve data held in file and return the root object, undef on error.
5756 SV *pretrieve(PerlIO *f)
5758 TRACEME(("pretrieve"));
5759 return do_retrieve(f, Nullsv, 0);
5765 * Retrieve data held in scalar and return the root object, undef on error.
5767 SV *mretrieve(SV *sv)
5769 TRACEME(("mretrieve"));
5770 return do_retrieve((PerlIO*) 0, sv, 0);
5780 * Deep clone: returns a fresh copy of the original referenced SV tree.
5782 * This is achieved by storing the object in memory and restoring from
5783 * there. Not that efficient, but it should be faster than doing it from
5790 stcxt_t *real_context;
5793 TRACEME(("dclone"));
5796 * Workaround for CROAK leak: if they enter with a "dirty" context,
5797 * free up memory for them now.
5804 * do_store() optimizes for dclone by not freeing its context, should
5805 * we need to allocate one because we're deep cloning from a hook.
5808 if (!do_store((PerlIO*) 0, sv, ST_CLONE, FALSE, (SV**) 0))
5809 return &PL_sv_undef; /* Error during store */
5812 * Because of the above optimization, we have to refresh the context,
5813 * since a new one could have been allocated and stacked by do_store().
5816 { dSTCXT; real_context = cxt; } /* Sub-block needed for macro */
5817 cxt = real_context; /* And we need this temporary... */
5820 * Now, `cxt' may refer to a new context.
5823 ASSERT(!cxt->s_dirty, ("clean context"));
5824 ASSERT(!cxt->entry, ("entry will not cause new context allocation"));
5827 TRACEME(("dclone stored %d bytes", size));
5831 * Since we're passing do_retrieve() both a NULL file and sv, we need
5832 * to pre-compute the taintedness of the input by setting cxt->tainted
5833 * to whatever state our own input string was. -- RAM, 15/09/2000
5835 * do_retrieve() will free non-root context.
5838 cxt->s_tainted = SvTAINTED(sv);
5839 out = do_retrieve((PerlIO*) 0, Nullsv, ST_CLONE);
5841 TRACEME(("dclone returns 0x%"UVxf, PTR2UV(out)));
5851 * The Perl IO GV object distinguishes between input and output for sockets
5852 * but not for plain files. To allow Storable to transparently work on
5853 * plain files and sockets transparently, we have to ask xsubpp to fetch the
5854 * right object for us. Hence the OutputStream and InputStream declarations.
5856 * Before perl 5.004_05, those entries in the standard typemap are not
5857 * defined in perl include files, so we do that here.
5860 #ifndef OutputStream
5861 #define OutputStream PerlIO *
5862 #define InputStream PerlIO *
5863 #endif /* !OutputStream */
5865 MODULE = Storable PACKAGE = Storable::Cxt
5871 stcxt_t *cxt = (stcxt_t *)SvPVX(SvRV(self));
5875 if (!cxt->membuf_ro && mbase)
5877 if (cxt->membuf_ro && (cxt->msaved).arena)
5878 Safefree((cxt->msaved).arena);
5881 MODULE = Storable PACKAGE = Storable
5887 gv_fetchpv("Storable::drop_utf8", GV_ADDMULTI, SVt_PV);
5889 /* Only disable the used only once warning if we are in debugging mode. */
5890 gv_fetchpv("Storable::DEBUGME", GV_ADDMULTI, SVt_PV);
5892 #ifdef USE_56_INTERWORK_KLUDGE
5893 gv_fetchpv("Storable::interwork_56_64bit", GV_ADDMULTI, SVt_PV);
5927 last_op_in_netorder()