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.
13 #include <patchlevel.h> /* Perl's one, needed since 5.6 */
18 #define DEBUGME /* Debug mode, turns assertions on as well */
19 #define DASSERT /* Assertion mode */
22 #if 0 /* On NetWare USE_PERLIO is not used */
23 #define DEBUGME /* Debug mode, turns assertions on as well */
24 #define DASSERT /* Assertion mode */
29 * Pre PerlIO time when none of USE_PERLIO and PERLIO_IS_STDIO is defined
30 * Provide them with the necessary defines so they can build with pre-5.004.
33 #ifndef PERLIO_IS_STDIO
35 #define PerlIO_getc(x) getc(x)
36 #define PerlIO_putc(f,x) putc(x,f)
37 #define PerlIO_read(x,y,z) fread(y,1,z,x)
38 #define PerlIO_write(x,y,z) fwrite(y,1,z,x)
39 #define PerlIO_stdoutf printf
40 #endif /* PERLIO_IS_STDIO */
41 #endif /* USE_PERLIO */
44 * Earlier versions of perl might be used, we can't assume they have the latest!
47 #ifndef PERL_VERSION /* For perls < 5.6 */
48 #define PERL_VERSION PATCHLEVEL
50 #define newRV_noinc(sv) ((Sv = newRV(sv)), --SvREFCNT(SvRV(Sv)), Sv)
52 #if (PATCHLEVEL <= 4) /* Older perls (<= 5.004) lack PL_ namespace */
53 #define PL_sv_yes sv_yes
54 #define PL_sv_no sv_no
55 #define PL_sv_undef sv_undef
56 #if (SUBVERSION <= 4) /* 5.004_04 has been reported to lack newSVpvn */
57 #define newSVpvn newSVpv
59 #endif /* PATCHLEVEL <= 4 */
60 #ifndef HvSHAREKEYS_off
61 #define HvSHAREKEYS_off(hv) /* Ignore */
63 #ifndef AvFILLp /* Older perls (<=5.003) lack AvFILLp */
64 #define AvFILLp AvFILL
66 typedef double NV; /* Older perls lack the NV type */
67 #define IVdf "ld" /* Various printf formats for Perl types */
71 #define INT2PTR(t,v) (t)(IV)(v)
72 #define PTR2UV(v) (unsigned long)(v)
73 #endif /* PERL_VERSION -- perls < 5.6 */
75 #ifndef NVef /* The following were not part of perl 5.6 */
76 #if defined(USE_LONG_DOUBLE) && \
77 defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
78 #define NVef PERL_PRIeldbl
79 #define NVff PERL_PRIfldbl
80 #define NVgf PERL_PRIgldbl
95 * TRACEME() will only output things when the $Storable::DEBUGME is true.
100 if (SvTRUE(perl_get_sv("Storable::DEBUGME", TRUE))) \
101 { PerlIO_stdoutf x; PerlIO_stdoutf("\n"); } \
108 #define ASSERT(x,y) \
111 PerlIO_stdoutf("ASSERT FAILED (\"%s\", line %d): ", \
112 __FILE__, __LINE__); \
113 PerlIO_stdoutf y; PerlIO_stdoutf("\n"); \
124 #define C(x) ((char) (x)) /* For markers with dynamic retrieval handling */
126 #define SX_OBJECT C(0) /* Already stored object */
127 #define SX_LSCALAR C(1) /* Scalar (large binary) follows (length, data) */
128 #define SX_ARRAY C(2) /* Array forthcominng (size, item list) */
129 #define SX_HASH C(3) /* Hash forthcoming (size, key/value pair list) */
130 #define SX_REF C(4) /* Reference to object forthcoming */
131 #define SX_UNDEF C(5) /* Undefined scalar */
132 #define SX_INTEGER C(6) /* Integer forthcoming */
133 #define SX_DOUBLE C(7) /* Double forthcoming */
134 #define SX_BYTE C(8) /* (signed) byte forthcoming */
135 #define SX_NETINT C(9) /* Integer in network order forthcoming */
136 #define SX_SCALAR C(10) /* Scalar (binary, small) follows (length, data) */
137 #define SX_TIED_ARRAY C(11) /* Tied array forthcoming */
138 #define SX_TIED_HASH C(12) /* Tied hash forthcoming */
139 #define SX_TIED_SCALAR C(13) /* Tied scalar forthcoming */
140 #define SX_SV_UNDEF C(14) /* Perl's immortal PL_sv_undef */
141 #define SX_SV_YES C(15) /* Perl's immortal PL_sv_yes */
142 #define SX_SV_NO C(16) /* Perl's immortal PL_sv_no */
143 #define SX_BLESS C(17) /* Object is blessed */
144 #define SX_IX_BLESS C(18) /* Object is blessed, classname given by index */
145 #define SX_HOOK C(19) /* Stored via hook, user-defined */
146 #define SX_OVERLOAD C(20) /* Overloaded reference */
147 #define SX_TIED_KEY C(21) /* Tied magic key forthcoming */
148 #define SX_TIED_IDX C(22) /* Tied magic index forthcoming */
149 #define SX_UTF8STR C(23) /* UTF-8 string forthcoming (small) */
150 #define SX_LUTF8STR C(24) /* UTF-8 string forthcoming (large) */
151 #define SX_FLAG_HASH C(25) /* Hash with flags forthcoming (size, flags, key/flags/value triplet list) */
152 #define SX_CODE C(26) /* Code references as perl source code */
153 #define SX_ERROR C(27) /* Error */
156 * Those are only used to retrieve "old" pre-0.6 binary images.
158 #define SX_ITEM 'i' /* An array item introducer */
159 #define SX_IT_UNDEF 'I' /* Undefined array item */
160 #define SX_KEY 'k' /* A hash key introducer */
161 #define SX_VALUE 'v' /* A hash value introducer */
162 #define SX_VL_UNDEF 'V' /* Undefined hash value */
165 * Those are only used to retrieve "old" pre-0.7 binary images
168 #define SX_CLASS 'b' /* Object is blessed, class name length <255 */
169 #define SX_LG_CLASS 'B' /* Object is blessed, class name length >255 */
170 #define SX_STORED 'X' /* End of object */
173 * Limits between short/long length representation.
176 #define LG_SCALAR 255 /* Large scalar length limit */
177 #define LG_BLESS 127 /* Large classname bless limit */
183 #define ST_STORE 0x1 /* Store operation */
184 #define ST_RETRIEVE 0x2 /* Retrieval operation */
185 #define ST_CLONE 0x4 /* Deep cloning operation */
188 * The following structure is used for hash table key retrieval. Since, when
189 * retrieving objects, we'll be facing blessed hash references, it's best
190 * to pre-allocate that buffer once and resize it as the need arises, never
191 * freeing it (keys will be saved away someplace else anyway, so even large
192 * keys are not enough a motivation to reclaim that space).
194 * This structure is also used for memory store/retrieve operations which
195 * happen in a fixed place before being malloc'ed elsewhere if persistency
196 * is required. Hence the aptr pointer.
199 char *arena; /* Will hold hash key strings, resized as needed */
200 STRLEN asiz; /* Size of aforementionned buffer */
201 char *aptr; /* Arena pointer, for in-place read/write ops */
202 char *aend; /* First invalid address */
207 * A hash table records the objects which have already been stored.
208 * Those are referred to as SX_OBJECT in the file, and their "tag" (i.e.
209 * an arbitrary sequence number) is used to identify them.
212 * An array table records the objects which have already been retrieved,
213 * as seen by the tag determind by counting the objects themselves. The
214 * reference to that retrieved object is kept in the table, and is returned
215 * when an SX_OBJECT is found bearing that same tag.
217 * The same processing is used to record "classname" for blessed objects:
218 * indexing by a hash at store time, and via an array at retrieve time.
221 typedef unsigned long stag_t; /* Used by pre-0.6 binary format */
224 * The following "thread-safe" related defines were contributed by
225 * Murray Nesbitt <murray@activestate.com> and integrated by RAM, who
226 * only renamed things a little bit to ensure consistency with surrounding
227 * code. -- RAM, 14/09/1999
229 * The original patch suffered from the fact that the stcxt_t structure
230 * was global. Murray tried to minimize the impact on the code as much as
233 * Starting with 0.7, Storable can be re-entrant, via the STORABLE_xxx hooks
234 * on objects. Therefore, the notion of context needs to be generalized,
238 #define MY_VERSION "Storable(" XS_VERSION ")"
242 * Conditional UTF8 support.
246 #define STORE_UTF8STR(pv, len) STORE_PV_LEN(pv, len, SX_UTF8STR, SX_LUTF8STR)
247 #define HAS_UTF8_SCALARS
249 #define HAS_UTF8_HASHES
252 /* 5.6 perl has utf8 scalars but not hashes */
256 #define STORE_UTF8STR(pv, len) CROAK(("panic: storing UTF8 in non-UTF8 perl"))
259 #define UTF8_CROAK() CROAK(("Cannot retrieve UTF8 data in non-UTF8 perl"))
262 #ifdef HvPLACEHOLDERS
263 #define HAS_RESTRICTED_HASHES
265 #define HVhek_PLACEHOLD 0x200
266 #define RESTRICTED_HASH_CROAK() CROAK(("Cannot retrieve restricted hash"))
270 #define HAS_HASH_KEY_FLAGS
274 * Fields s_tainted and s_dirty are prefixed with s_ because Perl's include
275 * files remap tainted and dirty when threading is enabled. That's bad for
276 * perl to remap such common words. -- RAM, 29/09/00
279 typedef struct stcxt {
280 int entry; /* flags recursion */
281 int optype; /* type of traversal operation */
282 HV *hseen; /* which objects have been seen, store time */
283 AV *hook_seen; /* which SVs were returned by STORABLE_freeze() */
284 AV *aseen; /* which objects have been seen, retrieve time */
285 HV *hclass; /* which classnames have been seen, store time */
286 AV *aclass; /* which classnames have been seen, retrieve time */
287 HV *hook; /* cache for hook methods per class name */
288 IV tagnum; /* incremented at store time for each seen object */
289 IV classnum; /* incremented at store time for each seen classname */
290 int netorder; /* true if network order used */
291 int s_tainted; /* true if input source is tainted, at retrieve time */
292 int forgive_me; /* whether to be forgiving... */
293 int deparse; /* whether to deparse code refs */
294 SV *eval; /* whether to eval source code */
295 int canonical; /* whether to store hashes sorted by key */
296 #ifndef HAS_RESTRICTED_HASHES
297 int derestrict; /* whether to downgrade restrcted hashes */
300 int use_bytes; /* whether to bytes-ify utf8 */
302 int accept_future_minor; /* croak immediately on future minor versions? */
303 int s_dirty; /* context is dirty due to CROAK() -- can be cleaned */
304 int membuf_ro; /* true means membuf is read-only and msaved is rw */
305 struct extendable keybuf; /* for hash key retrieval */
306 struct extendable membuf; /* for memory store/retrieve operations */
307 struct extendable msaved; /* where potentially valid mbuf is saved */
308 PerlIO *fio; /* where I/O are performed, NULL for memory */
309 int ver_major; /* major of version for retrieved object */
310 int ver_minor; /* minor of version for retrieved object */
311 SV *(**retrieve_vtbl)(); /* retrieve dispatch table */
312 SV *prev; /* contexts chained backwards in real recursion */
313 SV *my_sv; /* the blessed scalar who's SvPVX() I am */
316 #define NEW_STORABLE_CXT_OBJ(cxt) \
318 SV *self = newSV(sizeof(stcxt_t) - 1); \
319 SV *my_sv = newRV_noinc(self); \
320 sv_bless(my_sv, gv_stashpv("Storable::Cxt", TRUE)); \
321 cxt = (stcxt_t *)SvPVX(self); \
322 Zero(cxt, 1, stcxt_t); \
323 cxt->my_sv = my_sv; \
326 #if defined(MULTIPLICITY) || defined(PERL_OBJECT) || defined(PERL_CAPI)
328 #if (PATCHLEVEL <= 4) && (SUBVERSION < 68)
330 SV *perinterp_sv = perl_get_sv(MY_VERSION, FALSE)
331 #else /* >= perl5.004_68 */
333 SV *perinterp_sv = *hv_fetch(PL_modglobal, \
334 MY_VERSION, sizeof(MY_VERSION)-1, TRUE)
335 #endif /* < perl5.004_68 */
337 #define dSTCXT_PTR(T,name) \
338 T name = ((perinterp_sv && SvIOK(perinterp_sv) && SvIVX(perinterp_sv) \
339 ? (T)SvPVX(SvRV(INT2PTR(SV*,SvIVX(perinterp_sv)))) : (T) 0))
342 dSTCXT_PTR(stcxt_t *, cxt)
346 NEW_STORABLE_CXT_OBJ(cxt); \
347 sv_setiv(perinterp_sv, PTR2IV(cxt->my_sv))
349 #define SET_STCXT(x) \
352 sv_setiv(perinterp_sv, PTR2IV(x->my_sv)); \
355 #else /* !MULTIPLICITY && !PERL_OBJECT && !PERL_CAPI */
357 static stcxt_t *Context_ptr = NULL;
358 #define dSTCXT stcxt_t *cxt = Context_ptr
359 #define SET_STCXT(x) Context_ptr = x
362 NEW_STORABLE_CXT_OBJ(cxt); \
366 #endif /* MULTIPLICITY || PERL_OBJECT || PERL_CAPI */
370 * Croaking implies a memory leak, since we don't use setjmp/longjmp
371 * to catch the exit and free memory used during store or retrieve
372 * operations. This is not too difficult to fix, but I need to understand
373 * how Perl does it, and croaking is exceptional anyway, so I lack the
374 * motivation to do it.
376 * The current workaround is to mark the context as dirty when croaking,
377 * so that data structures can be freed whenever we renter Storable code
378 * (but only *then*: it's a workaround, not a fix).
380 * This is also imperfect, because we don't really know how far they trapped
381 * the croak(), and when we were recursing, we won't be able to clean anything
382 * but the topmost context stacked.
385 #define CROAK(x) STMT_START { cxt->s_dirty = 1; croak x; } STMT_END
388 * End of "thread-safe" related definitions.
394 * Keep only the low 32 bits of a pointer (used for tags, which are not
399 #define LOW_32BITS(x) ((I32) (x))
401 #define LOW_32BITS(x) ((I32) ((unsigned long) (x) & 0xffffffffUL))
407 * Hack for Crays, where sizeof(I32) == 8, and which are big-endians.
408 * Used in the WLEN and RLEN macros.
412 #define oI(x) ((I32 *) ((char *) (x) + 4))
413 #define oS(x) ((x) - 4)
414 #define oC(x) (x = 0)
423 * key buffer handling
425 #define kbuf (cxt->keybuf).arena
426 #define ksiz (cxt->keybuf).asiz
430 TRACEME(("** allocating kbuf of 128 bytes")); \
431 New(10003, kbuf, 128, char); \
438 TRACEME(("** extending kbuf to %d bytes (had %d)", x+1, ksiz)); \
439 Renew(kbuf, x+1, char); \
445 * memory buffer handling
447 #define mbase (cxt->membuf).arena
448 #define msiz (cxt->membuf).asiz
449 #define mptr (cxt->membuf).aptr
450 #define mend (cxt->membuf).aend
452 #define MGROW (1 << 13)
453 #define MMASK (MGROW - 1)
455 #define round_mgrow(x) \
456 ((unsigned long) (((unsigned long) (x) + MMASK) & ~MMASK))
457 #define trunc_int(x) \
458 ((unsigned long) ((unsigned long) (x) & ~(sizeof(int)-1)))
459 #define int_aligned(x) \
460 ((unsigned long) (x) == trunc_int(x))
462 #define MBUF_INIT(x) \
465 TRACEME(("** allocating mbase of %d bytes", MGROW)); \
466 New(10003, mbase, MGROW, char); \
473 mend = mbase + msiz; \
476 #define MBUF_TRUNC(x) mptr = mbase + x
477 #define MBUF_SIZE() (mptr - mbase)
483 * Those macros are used in do_retrieve() to save the current memory
484 * buffer into cxt->msaved, before MBUF_LOAD() can be used to retrieve
485 * data from a string.
487 #define MBUF_SAVE_AND_LOAD(in) \
489 ASSERT(!cxt->membuf_ro, ("mbase not already saved")); \
490 cxt->membuf_ro = 1; \
491 TRACEME(("saving mbuf")); \
492 StructCopy(&cxt->membuf, &cxt->msaved, struct extendable); \
496 #define MBUF_RESTORE() \
498 ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
499 cxt->membuf_ro = 0; \
500 TRACEME(("restoring mbuf")); \
501 StructCopy(&cxt->msaved, &cxt->membuf, struct extendable); \
505 * Use SvPOKp(), because SvPOK() fails on tainted scalars.
506 * See store_scalar() for other usage of this workaround.
508 #define MBUF_LOAD(v) \
510 ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
512 CROAK(("Not a scalar string")); \
513 mptr = mbase = SvPV(v, msiz); \
514 mend = mbase + msiz; \
517 #define MBUF_XTEND(x) \
519 int nsz = (int) round_mgrow((x)+msiz); \
520 int offset = mptr - mbase; \
521 ASSERT(!cxt->membuf_ro, ("mbase is not read-only")); \
522 TRACEME(("** extending mbase from %d to %d bytes (wants %d new)", \
524 Renew(mbase, nsz, char); \
526 mptr = mbase + offset; \
527 mend = mbase + nsz; \
530 #define MBUF_CHK(x) \
532 if ((mptr + (x)) > mend) \
536 #define MBUF_GETC(x) \
539 x = (int) (unsigned char) *mptr++; \
545 #define MBUF_GETINT(x) \
548 if ((mptr + 4) <= mend) { \
549 memcpy(oI(&x), mptr, 4); \
555 #define MBUF_GETINT(x) \
557 if ((mptr + sizeof(int)) <= mend) { \
558 if (int_aligned(mptr)) \
561 memcpy(&x, mptr, sizeof(int)); \
562 mptr += sizeof(int); \
568 #define MBUF_READ(x,s) \
570 if ((mptr + (s)) <= mend) { \
571 memcpy(x, mptr, s); \
577 #define MBUF_SAFEREAD(x,s,z) \
579 if ((mptr + (s)) <= mend) { \
580 memcpy(x, mptr, s); \
588 #define MBUF_PUTC(c) \
591 *mptr++ = (char) c; \
594 *mptr++ = (char) c; \
599 #define MBUF_PUTINT(i) \
602 memcpy(mptr, oI(&i), 4); \
606 #define MBUF_PUTINT(i) \
608 MBUF_CHK(sizeof(int)); \
609 if (int_aligned(mptr)) \
612 memcpy(mptr, &i, sizeof(int)); \
613 mptr += sizeof(int); \
617 #define MBUF_WRITE(x,s) \
620 memcpy(mptr, x, s); \
625 * Possible return values for sv_type().
629 #define svis_SCALAR 1
633 #define svis_TIED_ITEM 5
641 #define SHF_TYPE_MASK 0x03
642 #define SHF_LARGE_CLASSLEN 0x04
643 #define SHF_LARGE_STRLEN 0x08
644 #define SHF_LARGE_LISTLEN 0x10
645 #define SHF_IDX_CLASSNAME 0x20
646 #define SHF_NEED_RECURSE 0x40
647 #define SHF_HAS_LIST 0x80
650 * Types for SX_HOOK (last 2 bits in flags).
656 #define SHT_EXTRA 3 /* Read extra byte for type */
659 * The following are held in the "extra byte"...
662 #define SHT_TSCALAR 4 /* 4 + 0 -- tied scalar */
663 #define SHT_TARRAY 5 /* 4 + 1 -- tied array */
664 #define SHT_THASH 6 /* 4 + 2 -- tied hash */
667 * per hash flags for flagged hashes
670 #define SHV_RESTRICTED 0x01
673 * per key flags for flagged hashes
676 #define SHV_K_UTF8 0x01
677 #define SHV_K_WASUTF8 0x02
678 #define SHV_K_LOCKED 0x04
679 #define SHV_K_ISSV 0x08
680 #define SHV_K_PLACEHOLDER 0x10
683 * Before 0.6, the magic string was "perl-store" (binary version number 0).
685 * Since 0.6 introduced many binary incompatibilities, the magic string has
686 * been changed to "pst0" to allow an old image to be properly retrieved by
687 * a newer Storable, but ensure a newer image cannot be retrieved with an
690 * At 0.7, objects are given the ability to serialize themselves, and the
691 * set of markers is extended, backward compatibility is not jeopardized,
692 * so the binary version number could have remained unchanged. To correctly
693 * spot errors if a file making use of 0.7-specific extensions is given to
694 * 0.6 for retrieval, the binary version was moved to "2". And I'm introducing
695 * a "minor" version, to better track this kind of evolution from now on.
698 static const char old_magicstr[] = "perl-store"; /* Magic number before 0.6 */
699 static const char magicstr[] = "pst0"; /* Used as a magic number */
701 #define MAGICSTR_BYTES 'p','s','t','0'
702 #define OLDMAGICSTR_BYTES 'p','e','r','l','-','s','t','o','r','e'
704 /* 5.6.x introduced the ability to have IVs as long long.
705 However, Configure still defined BYTEORDER based on the size of a long.
706 Storable uses the BYTEORDER value as part of the header, but doesn't
707 explicity store sizeof(IV) anywhere in the header. Hence on 5.6.x built
708 with IV as long long on a platform that uses Configure (ie most things
709 except VMS and Windows) headers are identical for the different IV sizes,
710 despite the files containing some fields based on sizeof(IV)
712 5.8 is consistent - the following redifinition kludge is only needed on
713 5.6.x, but the interwork is needed on 5.8 while data survives in files
718 #if defined (IVSIZE) && (IVSIZE == 8) && (LONGSIZE == 4)
719 #ifndef NO_56_INTERWORK_KLUDGE
720 #define USE_56_INTERWORK_KLUDGE
722 #if BYTEORDER == 0x1234
724 #define BYTEORDER 0x12345678
726 #if BYTEORDER == 0x4321
728 #define BYTEORDER 0x87654321
733 #if BYTEORDER == 0x1234
734 #define BYTEORDER_BYTES '1','2','3','4'
736 #if BYTEORDER == 0x12345678
737 #define BYTEORDER_BYTES '1','2','3','4','5','6','7','8'
738 #ifdef USE_56_INTERWORK_KLUDGE
739 #define BYTEORDER_BYTES_56 '1','2','3','4'
742 #if BYTEORDER == 0x87654321
743 #define BYTEORDER_BYTES '8','7','6','5','4','3','2','1'
744 #ifdef USE_56_INTERWORK_KLUDGE
745 #define BYTEORDER_BYTES_56 '4','3','2','1'
748 #if BYTEORDER == 0x4321
749 #define BYTEORDER_BYTES '4','3','2','1'
751 #error Unknown byteoder. Please append your byteorder to Storable.xs
757 static const char byteorderstr[] = {BYTEORDER_BYTES, 0};
758 #ifdef USE_56_INTERWORK_KLUDGE
759 static const char byteorderstr_56[] = {BYTEORDER_BYTES_56, 0};
762 #define STORABLE_BIN_MAJOR 2 /* Binary major "version" */
763 #define STORABLE_BIN_MINOR 6 /* Binary minor "version" */
765 /* If we aren't 5.7.3 or later, we won't be writing out files that use the
766 * new flagged hash introdued in 2.5, so put 2.4 in the binary header to
767 * maximise ease of interoperation with older Storables.
768 * Could we write 2.3s if we're on 5.005_03? NWC
770 #if (PATCHLEVEL <= 6)
771 #define STORABLE_BIN_WRITE_MINOR 4
774 * As of perl 5.7.3, utf8 hash key is introduced.
775 * So this must change -- dankogai
777 #define STORABLE_BIN_WRITE_MINOR 6
778 #endif /* (PATCHLEVEL <= 6) */
781 * Useful store shortcuts...
788 else if (PerlIO_putc(cxt->fio, x) == EOF) \
792 #define WRITE_I32(x) \
794 ASSERT(sizeof(x) == sizeof(I32), ("writing an I32")); \
797 else if (PerlIO_write(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
804 if (cxt->netorder) { \
805 int y = (int) htonl(x); \
808 else if (PerlIO_write(cxt->fio,oI(&y),oS(sizeof(y))) != oS(sizeof(y))) \
813 else if (PerlIO_write(cxt->fio,oI(&x),oS(sizeof(x))) != oS(sizeof(x))) \
818 #define WLEN(x) WRITE_I32(x)
825 else if (PerlIO_write(cxt->fio, x, y) != y) \
829 #define STORE_PV_LEN(pv, len, small, large) \
831 if (len <= LG_SCALAR) { \
832 unsigned char clen = (unsigned char) len; \
844 #define STORE_SCALAR(pv, len) STORE_PV_LEN(pv, len, SX_SCALAR, SX_LSCALAR)
847 * Store undef in arrays and hashes without recursing through store().
849 #define STORE_UNDEF() \
856 * Useful retrieve shortcuts...
860 (cxt->fio ? PerlIO_getc(cxt->fio) : (mptr >= mend ? EOF : (int) *mptr++))
866 else if ((int) (x = PerlIO_getc(cxt->fio)) == EOF) \
870 #define READ_I32(x) \
872 ASSERT(sizeof(x) == sizeof(I32), ("reading an I32")); \
876 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
886 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
889 x = (int) ntohl(x); \
892 #define RLEN(x) READ_I32(x)
899 else if (PerlIO_read(cxt->fio, x, y) != y) \
903 #define SAFEREAD(x,y,z) \
906 MBUF_SAFEREAD(x,y,z); \
907 else if (PerlIO_read(cxt->fio, x, y) != y) { \
914 * This macro is used at retrieve time, to remember where object 'y', bearing a
915 * given tag 'tagnum', has been retrieved. Next time we see an SX_OBJECT marker,
916 * we'll therefore know where it has been retrieved and will be able to
917 * share the same reference, as in the original stored memory image.
919 * We also need to bless objects ASAP for hooks (which may compute "ref $x"
920 * on the objects given to STORABLE_thaw and expect that to be defined), and
921 * also for overloaded objects (for which we might not find the stash if the
922 * object is not blessed yet--this might occur for overloaded objects that
923 * refer to themselves indirectly: if we blessed upon return from a sub
924 * retrieve(), the SX_OBJECT marker we'd found could not have overloading
925 * restored on it because the underlying object would not be blessed yet!).
927 * To achieve that, the class name of the last retrieved object is passed down
928 * recursively, and the first SEEN() call for which the class name is not NULL
929 * will bless the object.
935 if (av_store(cxt->aseen, cxt->tagnum++, SvREFCNT_inc(y)) == 0) \
937 TRACEME(("aseen(#%d) = 0x%"UVxf" (refcnt=%d)", cxt->tagnum-1, \
938 PTR2UV(y), SvREFCNT(y)-1)); \
940 BLESS((SV *) (y), c); \
944 * Bless `s' in `p', via a temporary reference, required by sv_bless().
950 TRACEME(("blessing 0x%"UVxf" in %s", PTR2UV(s), (p))); \
951 stash = gv_stashpv((p), TRUE); \
952 ref = newRV_noinc(s); \
953 (void) sv_bless(ref, stash); \
959 static SV *retrieve(stcxt_t *cxt, char *cname);
962 * Dynamic dispatching table for SV store.
965 static int store_ref(stcxt_t *cxt, SV *sv);
966 static int store_scalar(stcxt_t *cxt, SV *sv);
967 static int store_array(stcxt_t *cxt, AV *av);
968 static int store_hash(stcxt_t *cxt, HV *hv);
969 static int store_tied(stcxt_t *cxt, SV *sv);
970 static int store_tied_item(stcxt_t *cxt, SV *sv);
971 static int store_code(stcxt_t *cxt, CV *cv);
972 static int store_other(stcxt_t *cxt, SV *sv);
973 static int store_blessed(stcxt_t *cxt, SV *sv, int type, HV *pkg);
975 static int (*sv_store[])(stcxt_t *cxt, SV *sv) = {
976 store_ref, /* svis_REF */
977 store_scalar, /* svis_SCALAR */
978 (int (*)(stcxt_t *cxt, SV *sv)) store_array, /* svis_ARRAY */
979 (int (*)(stcxt_t *cxt, SV *sv)) store_hash, /* svis_HASH */
980 store_tied, /* svis_TIED */
981 store_tied_item, /* svis_TIED_ITEM */
982 (int (*)(stcxt_t *cxt, SV *sv)) store_code, /* svis_CODE */
983 store_other, /* svis_OTHER */
986 #define SV_STORE(x) (*sv_store[x])
989 * Dynamic dispatching tables for SV retrieval.
992 static SV *retrieve_lscalar(stcxt_t *cxt, char *cname);
993 static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname);
994 static SV *old_retrieve_array(stcxt_t *cxt, char *cname);
995 static SV *old_retrieve_hash(stcxt_t *cxt, char *cname);
996 static SV *retrieve_ref(stcxt_t *cxt, char *cname);
997 static SV *retrieve_undef(stcxt_t *cxt, char *cname);
998 static SV *retrieve_integer(stcxt_t *cxt, char *cname);
999 static SV *retrieve_double(stcxt_t *cxt, char *cname);
1000 static SV *retrieve_byte(stcxt_t *cxt, char *cname);
1001 static SV *retrieve_netint(stcxt_t *cxt, char *cname);
1002 static SV *retrieve_scalar(stcxt_t *cxt, char *cname);
1003 static SV *retrieve_utf8str(stcxt_t *cxt, char *cname);
1004 static SV *retrieve_tied_array(stcxt_t *cxt, char *cname);
1005 static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname);
1006 static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname);
1007 static SV *retrieve_other(stcxt_t *cxt, char *cname);
1009 static SV *(*sv_old_retrieve[])(stcxt_t *cxt, char *cname) = {
1010 0, /* SX_OBJECT -- entry unused dynamically */
1011 retrieve_lscalar, /* SX_LSCALAR */
1012 old_retrieve_array, /* SX_ARRAY -- for pre-0.6 binaries */
1013 old_retrieve_hash, /* SX_HASH -- for pre-0.6 binaries */
1014 retrieve_ref, /* SX_REF */
1015 retrieve_undef, /* SX_UNDEF */
1016 retrieve_integer, /* SX_INTEGER */
1017 retrieve_double, /* SX_DOUBLE */
1018 retrieve_byte, /* SX_BYTE */
1019 retrieve_netint, /* SX_NETINT */
1020 retrieve_scalar, /* SX_SCALAR */
1021 retrieve_tied_array, /* SX_ARRAY */
1022 retrieve_tied_hash, /* SX_HASH */
1023 retrieve_tied_scalar, /* SX_SCALAR */
1024 retrieve_other, /* SX_SV_UNDEF not supported */
1025 retrieve_other, /* SX_SV_YES not supported */
1026 retrieve_other, /* SX_SV_NO not supported */
1027 retrieve_other, /* SX_BLESS not supported */
1028 retrieve_other, /* SX_IX_BLESS not supported */
1029 retrieve_other, /* SX_HOOK not supported */
1030 retrieve_other, /* SX_OVERLOADED not supported */
1031 retrieve_other, /* SX_TIED_KEY not supported */
1032 retrieve_other, /* SX_TIED_IDX not supported */
1033 retrieve_other, /* SX_UTF8STR not supported */
1034 retrieve_other, /* SX_LUTF8STR not supported */
1035 retrieve_other, /* SX_FLAG_HASH not supported */
1036 retrieve_other, /* SX_CODE not supported */
1037 retrieve_other, /* SX_ERROR */
1040 static SV *retrieve_array(stcxt_t *cxt, char *cname);
1041 static SV *retrieve_hash(stcxt_t *cxt, char *cname);
1042 static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname);
1043 static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname);
1044 static SV *retrieve_sv_no(stcxt_t *cxt, char *cname);
1045 static SV *retrieve_blessed(stcxt_t *cxt, char *cname);
1046 static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname);
1047 static SV *retrieve_hook(stcxt_t *cxt, char *cname);
1048 static SV *retrieve_overloaded(stcxt_t *cxt, char *cname);
1049 static SV *retrieve_tied_key(stcxt_t *cxt, char *cname);
1050 static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname);
1051 static SV *retrieve_flag_hash(stcxt_t *cxt, char *cname);
1052 static SV *retrieve_code(stcxt_t *cxt, char *cname);
1054 static SV *(*sv_retrieve[])(stcxt_t *cxt, char *cname) = {
1055 0, /* SX_OBJECT -- entry unused dynamically */
1056 retrieve_lscalar, /* SX_LSCALAR */
1057 retrieve_array, /* SX_ARRAY */
1058 retrieve_hash, /* SX_HASH */
1059 retrieve_ref, /* SX_REF */
1060 retrieve_undef, /* SX_UNDEF */
1061 retrieve_integer, /* SX_INTEGER */
1062 retrieve_double, /* SX_DOUBLE */
1063 retrieve_byte, /* SX_BYTE */
1064 retrieve_netint, /* SX_NETINT */
1065 retrieve_scalar, /* SX_SCALAR */
1066 retrieve_tied_array, /* SX_ARRAY */
1067 retrieve_tied_hash, /* SX_HASH */
1068 retrieve_tied_scalar, /* SX_SCALAR */
1069 retrieve_sv_undef, /* SX_SV_UNDEF */
1070 retrieve_sv_yes, /* SX_SV_YES */
1071 retrieve_sv_no, /* SX_SV_NO */
1072 retrieve_blessed, /* SX_BLESS */
1073 retrieve_idx_blessed, /* SX_IX_BLESS */
1074 retrieve_hook, /* SX_HOOK */
1075 retrieve_overloaded, /* SX_OVERLOAD */
1076 retrieve_tied_key, /* SX_TIED_KEY */
1077 retrieve_tied_idx, /* SX_TIED_IDX */
1078 retrieve_utf8str, /* SX_UTF8STR */
1079 retrieve_lutf8str, /* SX_LUTF8STR */
1080 retrieve_flag_hash, /* SX_HASH */
1081 retrieve_code, /* SX_CODE */
1082 retrieve_other, /* SX_ERROR */
1085 #define RETRIEVE(c,x) (*(c)->retrieve_vtbl[(x) >= SX_ERROR ? SX_ERROR : (x)])
1087 static SV *mbuf2sv(void);
1090 *** Context management.
1096 * Called once per "thread" (interpreter) to initialize some global context.
1098 static void init_perinterp(void)
1102 cxt->netorder = 0; /* true if network order used */
1103 cxt->forgive_me = -1; /* whether to be forgiving... */
1109 * Called at the end of every context cleaning, to perform common reset
1112 static void reset_context(stcxt_t *cxt)
1116 cxt->optype &= ~(ST_STORE|ST_RETRIEVE); /* Leave ST_CLONE alone */
1120 * init_store_context
1122 * Initialize a new store context for real recursion.
1124 static void init_store_context(
1130 TRACEME(("init_store_context"));
1132 cxt->netorder = network_order;
1133 cxt->forgive_me = -1; /* Fetched from perl if needed */
1134 cxt->deparse = -1; /* Idem */
1135 cxt->eval = NULL; /* Idem */
1136 cxt->canonical = -1; /* Idem */
1137 cxt->tagnum = -1; /* Reset tag numbers */
1138 cxt->classnum = -1; /* Reset class numbers */
1139 cxt->fio = f; /* Where I/O are performed */
1140 cxt->optype = optype; /* A store, or a deep clone */
1141 cxt->entry = 1; /* No recursion yet */
1144 * The `hseen' table is used to keep track of each SV stored and their
1145 * associated tag numbers is special. It is "abused" because the
1146 * values stored are not real SV, just integers cast to (SV *),
1147 * which explains the freeing below.
1149 * It is also one possible bottlneck to achieve good storing speed,
1150 * so the "shared keys" optimization is turned off (unlikely to be
1151 * of any use here), and the hash table is "pre-extended". Together,
1152 * those optimizations increase the throughput by 12%.
1155 cxt->hseen = newHV(); /* Table where seen objects are stored */
1156 HvSHAREKEYS_off(cxt->hseen);
1159 * The following does not work well with perl5.004_04, and causes
1160 * a core dump later on, in a completely unrelated spot, which
1161 * makes me think there is a memory corruption going on.
1163 * Calling hv_ksplit(hseen, HBUCKETS) instead of manually hacking
1164 * it below does not make any difference. It seems to work fine
1165 * with perl5.004_68 but given the probable nature of the bug,
1166 * that does not prove anything.
1168 * It's a shame because increasing the amount of buckets raises
1169 * store() throughput by 5%, but until I figure this out, I can't
1170 * allow for this to go into production.
1172 * It is reported fixed in 5.005, hence the #if.
1174 #if PERL_VERSION >= 5
1175 #define HBUCKETS 4096 /* Buckets for %hseen */
1176 HvMAX(cxt->hseen) = HBUCKETS - 1; /* keys %hseen = $HBUCKETS; */
1180 * The `hclass' hash uses the same settings as `hseen' above, but it is
1181 * used to assign sequential tags (numbers) to class names for blessed
1184 * We turn the shared key optimization on.
1187 cxt->hclass = newHV(); /* Where seen classnames are stored */
1189 #if PERL_VERSION >= 5
1190 HvMAX(cxt->hclass) = HBUCKETS - 1; /* keys %hclass = $HBUCKETS; */
1194 * The `hook' hash table is used to keep track of the references on
1195 * the STORABLE_freeze hook routines, when found in some class name.
1197 * It is assumed that the inheritance tree will not be changed during
1198 * storing, and that no new method will be dynamically created by the
1202 cxt->hook = newHV(); /* Table where hooks are cached */
1205 * The `hook_seen' array keeps track of all the SVs returned by
1206 * STORABLE_freeze hooks for us to serialize, so that they are not
1207 * reclaimed until the end of the serialization process. Each SV is
1208 * only stored once, the first time it is seen.
1211 cxt->hook_seen = newAV(); /* Lists SVs returned by STORABLE_freeze */
1215 * clean_store_context
1217 * Clean store context by
1219 static void clean_store_context(stcxt_t *cxt)
1223 TRACEME(("clean_store_context"));
1225 ASSERT(cxt->optype & ST_STORE, ("was performing a store()"));
1228 * Insert real values into hashes where we stored faked pointers.
1232 hv_iterinit(cxt->hseen);
1233 while ((he = hv_iternext(cxt->hseen))) /* Extra () for -Wall, grr.. */
1234 HeVAL(he) = &PL_sv_undef;
1238 hv_iterinit(cxt->hclass);
1239 while ((he = hv_iternext(cxt->hclass))) /* Extra () for -Wall, grr.. */
1240 HeVAL(he) = &PL_sv_undef;
1244 * And now dispose of them...
1246 * The surrounding if() protection has been added because there might be
1247 * some cases where this routine is called more than once, during
1248 * exceptionnal events. This was reported by Marc Lehmann when Storable
1249 * is executed from mod_perl, and the fix was suggested by him.
1250 * -- RAM, 20/12/2000
1254 HV *hseen = cxt->hseen;
1257 sv_free((SV *) hseen);
1261 HV *hclass = cxt->hclass;
1264 sv_free((SV *) hclass);
1268 HV *hook = cxt->hook;
1271 sv_free((SV *) hook);
1274 if (cxt->hook_seen) {
1275 AV *hook_seen = cxt->hook_seen;
1277 av_undef(hook_seen);
1278 sv_free((SV *) hook_seen);
1281 cxt->forgive_me = -1; /* Fetched from perl if needed */
1282 cxt->deparse = -1; /* Idem */
1284 SvREFCNT_dec(cxt->eval);
1286 cxt->eval = NULL; /* Idem */
1287 cxt->canonical = -1; /* Idem */
1293 * init_retrieve_context
1295 * Initialize a new retrieve context for real recursion.
1297 static void init_retrieve_context(stcxt_t *cxt, int optype, int is_tainted)
1299 TRACEME(("init_retrieve_context"));
1302 * The hook hash table is used to keep track of the references on
1303 * the STORABLE_thaw hook routines, when found in some class name.
1305 * It is assumed that the inheritance tree will not be changed during
1306 * storing, and that no new method will be dynamically created by the
1310 cxt->hook = newHV(); /* Caches STORABLE_thaw */
1313 * If retrieving an old binary version, the cxt->retrieve_vtbl variable
1314 * was set to sv_old_retrieve. We'll need a hash table to keep track of
1315 * the correspondance between the tags and the tag number used by the
1316 * new retrieve routines.
1319 cxt->hseen = ((cxt->retrieve_vtbl == sv_old_retrieve) ? newHV() : 0);
1321 cxt->aseen = newAV(); /* Where retrieved objects are kept */
1322 cxt->aclass = newAV(); /* Where seen classnames are kept */
1323 cxt->tagnum = 0; /* Have to count objects... */
1324 cxt->classnum = 0; /* ...and class names as well */
1325 cxt->optype = optype;
1326 cxt->s_tainted = is_tainted;
1327 cxt->entry = 1; /* No recursion yet */
1328 #ifndef HAS_RESTRICTED_HASHES
1329 cxt->derestrict = -1; /* Fetched from perl if needed */
1331 #ifndef HAS_UTF8_ALL
1332 cxt->use_bytes = -1; /* Fetched from perl if needed */
1334 cxt->accept_future_minor = -1; /* Fetched from perl if needed */
1338 * clean_retrieve_context
1340 * Clean retrieve context by
1342 static void clean_retrieve_context(stcxt_t *cxt)
1344 TRACEME(("clean_retrieve_context"));
1346 ASSERT(cxt->optype & ST_RETRIEVE, ("was performing a retrieve()"));
1349 AV *aseen = cxt->aseen;
1352 sv_free((SV *) aseen);
1356 AV *aclass = cxt->aclass;
1359 sv_free((SV *) aclass);
1363 HV *hook = cxt->hook;
1366 sv_free((SV *) hook);
1370 HV *hseen = cxt->hseen;
1373 sv_free((SV *) hseen); /* optional HV, for backward compat. */
1376 #ifndef HAS_RESTRICTED_HASHES
1377 cxt->derestrict = -1; /* Fetched from perl if needed */
1379 #ifndef HAS_UTF8_ALL
1380 cxt->use_bytes = -1; /* Fetched from perl if needed */
1382 cxt->accept_future_minor = -1; /* Fetched from perl if needed */
1390 * A workaround for the CROAK bug: cleanup the last context.
1392 static void clean_context(stcxt_t *cxt)
1394 TRACEME(("clean_context"));
1396 ASSERT(cxt->s_dirty, ("dirty context"));
1401 ASSERT(!cxt->membuf_ro, ("mbase is not read-only"));
1403 if (cxt->optype & ST_RETRIEVE)
1404 clean_retrieve_context(cxt);
1405 else if (cxt->optype & ST_STORE)
1406 clean_store_context(cxt);
1410 ASSERT(!cxt->s_dirty, ("context is clean"));
1411 ASSERT(cxt->entry == 0, ("context is reset"));
1417 * Allocate a new context and push it on top of the parent one.
1418 * This new context is made globally visible via SET_STCXT().
1420 static stcxt_t *allocate_context(parent_cxt)
1421 stcxt_t *parent_cxt;
1425 TRACEME(("allocate_context"));
1427 ASSERT(!parent_cxt->s_dirty, ("parent context clean"));
1429 NEW_STORABLE_CXT_OBJ(cxt);
1430 cxt->prev = parent_cxt->my_sv;
1433 ASSERT(!cxt->s_dirty, ("clean context"));
1441 * Free current context, which cannot be the "root" one.
1442 * Make the context underneath globally visible via SET_STCXT().
1444 static void free_context(cxt)
1447 stcxt_t *prev = (stcxt_t *)(cxt->prev ? SvPVX(SvRV(cxt->prev)) : 0);
1449 TRACEME(("free_context"));
1451 ASSERT(!cxt->s_dirty, ("clean context"));
1452 ASSERT(prev, ("not freeing root context"));
1454 SvREFCNT_dec(cxt->my_sv);
1457 ASSERT(cxt, ("context not void"));
1467 * Tells whether we're in the middle of a store operation.
1469 int is_storing(void)
1473 return cxt->entry && (cxt->optype & ST_STORE);
1479 * Tells whether we're in the middle of a retrieve operation.
1481 int is_retrieving(void)
1485 return cxt->entry && (cxt->optype & ST_RETRIEVE);
1489 * last_op_in_netorder
1491 * Returns whether last operation was made using network order.
1493 * This is typically out-of-band information that might prove useful
1494 * to people wishing to convert native to network order data when used.
1496 int last_op_in_netorder(void)
1500 return cxt->netorder;
1504 *** Hook lookup and calling routines.
1510 * A wrapper on gv_fetchmethod_autoload() which caches results.
1512 * Returns the routine reference as an SV*, or null if neither the package
1513 * nor its ancestors know about the method.
1515 static SV *pkg_fetchmeth(
1524 * The following code is the same as the one performed by UNIVERSAL::can
1528 gv = gv_fetchmethod_autoload(pkg, method, FALSE);
1529 if (gv && isGV(gv)) {
1530 sv = newRV((SV*) GvCV(gv));
1531 TRACEME(("%s->%s: 0x%"UVxf, HvNAME(pkg), method, PTR2UV(sv)));
1533 sv = newSVsv(&PL_sv_undef);
1534 TRACEME(("%s->%s: not found", HvNAME(pkg), method));
1538 * Cache the result, ignoring failure: if we can't store the value,
1539 * it just won't be cached.
1542 (void) hv_store(cache, HvNAME(pkg), strlen(HvNAME(pkg)), sv, 0);
1544 return SvOK(sv) ? sv : (SV *) 0;
1550 * Force cached value to be undef: hook ignored even if present.
1552 static void pkg_hide(
1557 (void) hv_store(cache,
1558 HvNAME(pkg), strlen(HvNAME(pkg)), newSVsv(&PL_sv_undef), 0);
1564 * Discard cached value: a whole fetch loop will be retried at next lookup.
1566 static void pkg_uncache(
1571 (void) hv_delete(cache, HvNAME(pkg), strlen(HvNAME(pkg)), G_DISCARD);
1577 * Our own "UNIVERSAL::can", which caches results.
1579 * Returns the routine reference as an SV*, or null if the object does not
1580 * know about the method.
1590 TRACEME(("pkg_can for %s->%s", HvNAME(pkg), method));
1593 * Look into the cache to see whether we already have determined
1594 * where the routine was, if any.
1596 * NOTA BENE: we don't use `method' at all in our lookup, since we know
1597 * that only one hook (i.e. always the same) is cached in a given cache.
1600 svh = hv_fetch(cache, HvNAME(pkg), strlen(HvNAME(pkg)), FALSE);
1604 TRACEME(("cached %s->%s: not found", HvNAME(pkg), method));
1607 TRACEME(("cached %s->%s: 0x%"UVxf,
1608 HvNAME(pkg), method, PTR2UV(sv)));
1613 TRACEME(("not cached yet"));
1614 return pkg_fetchmeth(cache, pkg, method); /* Fetch and cache */
1620 * Call routine as obj->hook(av) in scalar context.
1621 * Propagates the single returned value if not called in void context.
1623 static SV *scalar_call(
1634 TRACEME(("scalar_call (cloning=%d)", cloning));
1641 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1643 SV **ary = AvARRAY(av);
1644 int cnt = AvFILLp(av) + 1;
1646 XPUSHs(ary[0]); /* Frozen string */
1647 for (i = 1; i < cnt; i++) {
1648 TRACEME(("pushing arg #%d (0x%"UVxf")...",
1649 i, PTR2UV(ary[i])));
1650 XPUSHs(sv_2mortal(newRV(ary[i])));
1655 TRACEME(("calling..."));
1656 count = perl_call_sv(hook, flags); /* Go back to Perl code */
1657 TRACEME(("count = %d", count));
1663 SvREFCNT_inc(sv); /* We're returning it, must stay alive! */
1676 * Call routine obj->hook(cloning) in list context.
1677 * Returns the list of returned values in an array.
1679 static AV *array_call(
1689 TRACEME(("array_call (cloning=%d)", cloning));
1695 XPUSHs(obj); /* Target object */
1696 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1699 count = perl_call_sv(hook, G_ARRAY); /* Go back to Perl code */
1704 for (i = count - 1; i >= 0; i--) {
1706 av_store(av, i, SvREFCNT_inc(sv));
1719 * Lookup the class name in the `hclass' table and either assign it a new ID
1720 * or return the existing one, by filling in `classnum'.
1722 * Return true if the class was known, false if the ID was just generated.
1724 static int known_class(
1726 char *name, /* Class name */
1727 int len, /* Name length */
1731 HV *hclass = cxt->hclass;
1733 TRACEME(("known_class (%s)", name));
1736 * Recall that we don't store pointers in this hash table, but tags.
1737 * Therefore, we need LOW_32BITS() to extract the relevant parts.
1740 svh = hv_fetch(hclass, name, len, FALSE);
1742 *classnum = LOW_32BITS(*svh);
1747 * Unknown classname, we need to record it.
1751 if (!hv_store(hclass, name, len, INT2PTR(SV*, cxt->classnum), 0))
1752 CROAK(("Unable to record new classname"));
1754 *classnum = cxt->classnum;
1759 *** Sepcific store routines.
1765 * Store a reference.
1766 * Layout is SX_REF <object> or SX_OVERLOAD <object>.
1768 static int store_ref(stcxt_t *cxt, SV *sv)
1770 TRACEME(("store_ref (0x%"UVxf")", PTR2UV(sv)));
1773 * Follow reference, and check if target is overloaded.
1779 HV *stash = (HV *) SvSTASH(sv);
1780 if (stash && Gv_AMG(stash)) {
1781 TRACEME(("ref (0x%"UVxf") is overloaded", PTR2UV(sv)));
1782 PUTMARK(SX_OVERLOAD);
1788 return store(cxt, sv);
1796 * Layout is SX_LSCALAR <length> <data>, SX_SCALAR <length> <data> or SX_UNDEF.
1797 * The <data> section is omitted if <length> is 0.
1799 * If integer or double, the layout is SX_INTEGER <data> or SX_DOUBLE <data>.
1800 * Small integers (within [-127, +127]) are stored as SX_BYTE <byte>.
1802 static int store_scalar(stcxt_t *cxt, SV *sv)
1807 U32 flags = SvFLAGS(sv); /* "cc -O" may put it in register */
1809 TRACEME(("store_scalar (0x%"UVxf")", PTR2UV(sv)));
1812 * For efficiency, break the SV encapsulation by peaking at the flags
1813 * directly without using the Perl macros to avoid dereferencing
1814 * sv->sv_flags each time we wish to check the flags.
1817 if (!(flags & SVf_OK)) { /* !SvOK(sv) */
1818 if (sv == &PL_sv_undef) {
1819 TRACEME(("immortal undef"));
1820 PUTMARK(SX_SV_UNDEF);
1822 TRACEME(("undef at 0x%"UVxf, PTR2UV(sv)));
1829 * Always store the string representation of a scalar if it exists.
1830 * Gisle Aas provided me with this test case, better than a long speach:
1832 * perl -MDevel::Peek -le '$a="abc"; $a+0; Dump($a)'
1833 * SV = PVNV(0x80c8520)
1835 * FLAGS = (NOK,POK,pNOK,pPOK)
1838 * PV = 0x80c83d0 "abc"\0
1842 * Write SX_SCALAR, length, followed by the actual data.
1844 * Otherwise, write an SX_BYTE, SX_INTEGER or an SX_DOUBLE as
1845 * appropriate, followed by the actual (binary) data. A double
1846 * is written as a string if network order, for portability.
1848 * NOTE: instead of using SvNOK(sv), we test for SvNOKp(sv).
1849 * The reason is that when the scalar value is tainted, the SvNOK(sv)
1852 * The test for a read-only scalar with both POK and NOK set is meant
1853 * to quickly detect &PL_sv_yes and &PL_sv_no without having to pay the
1854 * address comparison for each scalar we store.
1857 #define SV_MAYBE_IMMORTAL (SVf_READONLY|SVf_POK|SVf_NOK)
1859 if ((flags & SV_MAYBE_IMMORTAL) == SV_MAYBE_IMMORTAL) {
1860 if (sv == &PL_sv_yes) {
1861 TRACEME(("immortal yes"));
1863 } else if (sv == &PL_sv_no) {
1864 TRACEME(("immortal no"));
1867 pv = SvPV(sv, len); /* We know it's SvPOK */
1868 goto string; /* Share code below */
1870 } else if (flags & SVf_POK) {
1871 /* public string - go direct to string read. */
1872 goto string_readlen;
1874 #if (PATCHLEVEL <= 6)
1875 /* For 5.6 and earlier NV flag trumps IV flag, so only use integer
1876 direct if NV flag is off. */
1877 (flags & (SVf_NOK | SVf_IOK)) == SVf_IOK
1879 /* 5.7 rules are that if IV public flag is set, IV value is as
1880 good, if not better, than NV value. */
1886 * Will come here from below with iv set if double is an integer.
1890 /* Sorry. This isn't in 5.005_56 (IIRC) or earlier. */
1892 /* Need to do this out here, else 0xFFFFFFFF becomes iv of -1
1893 * (for example) and that ends up in the optimised small integer
1896 if ((flags & SVf_IVisUV) && SvUV(sv) > IV_MAX) {
1897 TRACEME(("large unsigned integer as string, value = %"UVuf, SvUV(sv)));
1898 goto string_readlen;
1902 * Optimize small integers into a single byte, otherwise store as
1903 * a real integer (converted into network order if they asked).
1906 if (iv >= -128 && iv <= 127) {
1907 unsigned char siv = (unsigned char) (iv + 128); /* [0,255] */
1910 TRACEME(("small integer stored as %d", siv));
1911 } else if (cxt->netorder) {
1913 TRACEME(("no htonl, fall back to string for integer"));
1914 goto string_readlen;
1922 /* Sorry. This isn't in 5.005_56 (IIRC) or earlier. */
1923 ((flags & SVf_IVisUV) && SvUV(sv) > 0x7FFFFFFF) ||
1925 (iv > 0x7FFFFFFF) || (iv < -0x80000000)) {
1926 /* Bigger than 32 bits. */
1927 TRACEME(("large network order integer as string, value = %"IVdf, iv));
1928 goto string_readlen;
1932 niv = (I32) htonl((I32) iv);
1933 TRACEME(("using network order"));
1938 PUTMARK(SX_INTEGER);
1939 WRITE(&iv, sizeof(iv));
1942 TRACEME(("ok (integer 0x%"UVxf", value = %"IVdf")", PTR2UV(sv), iv));
1943 } else if (flags & SVf_NOK) {
1945 #if (PATCHLEVEL <= 6)
1948 * Watch for number being an integer in disguise.
1950 if (nv == (NV) (iv = I_V(nv))) {
1951 TRACEME(("double %"NVff" is actually integer %"IVdf, nv, iv));
1952 goto integer; /* Share code above */
1959 goto integer; /* Share code above */
1964 if (cxt->netorder) {
1965 TRACEME(("double %"NVff" stored as string", nv));
1966 goto string_readlen; /* Share code below */
1970 WRITE(&nv, sizeof(nv));
1972 TRACEME(("ok (double 0x%"UVxf", value = %"NVff")", PTR2UV(sv), nv));
1974 } else if (flags & (SVp_POK | SVp_NOK | SVp_IOK)) {
1975 I32 wlen; /* For 64-bit machines */
1981 * Will come here from above if it was readonly, POK and NOK but
1982 * neither &PL_sv_yes nor &PL_sv_no.
1986 wlen = (I32) len; /* WLEN via STORE_SCALAR expects I32 */
1988 STORE_UTF8STR(pv, wlen);
1990 STORE_SCALAR(pv, wlen);
1991 TRACEME(("ok (scalar 0x%"UVxf" '%s', length = %"IVdf")",
1992 PTR2UV(sv), SvPVX(sv), (IV)len));
1994 CROAK(("Can't determine type of %s(0x%"UVxf")",
1995 sv_reftype(sv, FALSE),
1997 return 0; /* Ok, no recursion on scalars */
2005 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
2006 * Each item is stored as <object>.
2008 static int store_array(stcxt_t *cxt, AV *av)
2011 I32 len = av_len(av) + 1;
2015 TRACEME(("store_array (0x%"UVxf")", PTR2UV(av)));
2018 * Signal array by emitting SX_ARRAY, followed by the array length.
2023 TRACEME(("size = %d", len));
2026 * Now store each item recursively.
2029 for (i = 0; i < len; i++) {
2030 sav = av_fetch(av, i, 0);
2032 TRACEME(("(#%d) undef item", i));
2036 TRACEME(("(#%d) item", i));
2037 if ((ret = store(cxt, *sav))) /* Extra () for -Wall, grr... */
2041 TRACEME(("ok (array)"));
2050 * Borrowed from perl source file pp_ctl.c, where it is used by pp_sort.
2053 sortcmp(const void *a, const void *b)
2055 return sv_cmp(*(SV * const *) a, *(SV * const *) b);
2062 * Store a hash table.
2064 * For a "normal" hash (not restricted, no utf8 keys):
2066 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
2067 * Values are stored as <object>.
2068 * Keys are stored as <length> <data>, the <data> section being omitted
2071 * For a "fancy" hash (restricted or utf8 keys):
2073 * Layout is SX_FLAG_HASH <size> <hash flags> followed by each key/value pair,
2075 * Values are stored as <object>.
2076 * Keys are stored as <flags> <length> <data>, the <data> section being omitted
2078 * Currently the only hash flag is "restriced"
2079 * Key flags are as for hv.h
2081 static int store_hash(stcxt_t *cxt, HV *hv)
2084 #ifdef HAS_RESTRICTED_HASHES
2093 int flagged_hash = ((SvREADONLY(hv)
2094 #ifdef HAS_HASH_KEY_FLAGS
2098 unsigned char hash_flags = (SvREADONLY(hv) ? SHV_RESTRICTED : 0);
2101 /* needs int cast for C++ compilers, doesn't it? */
2102 TRACEME(("store_hash (0x%"UVxf") (flags %x)", PTR2UV(hv),
2105 TRACEME(("store_hash (0x%"UVxf")", PTR2UV(hv)));
2109 * Signal hash by emitting SX_HASH, followed by the table length.
2113 PUTMARK(SX_FLAG_HASH);
2114 PUTMARK(hash_flags);
2119 TRACEME(("size = %d", len));
2122 * Save possible iteration state via each() on that table.
2125 riter = HvRITER(hv);
2126 eiter = HvEITER(hv);
2130 * Now store each item recursively.
2132 * If canonical is defined to some true value then store each
2133 * key/value pair in sorted order otherwise the order is random.
2134 * Canonical order is irrelevant when a deep clone operation is performed.
2136 * Fetch the value from perl only once per store() operation, and only
2141 !(cxt->optype & ST_CLONE) && (cxt->canonical == 1 ||
2142 (cxt->canonical < 0 && (cxt->canonical =
2143 (SvTRUE(perl_get_sv("Storable::canonical", TRUE)) ? 1 : 0))))
2146 * Storing in order, sorted by key.
2147 * Run through the hash, building up an array of keys in a
2148 * mortal array, sort the array and then run through the
2154 /*av_extend (av, len);*/
2156 TRACEME(("using canonical order"));
2158 for (i = 0; i < len; i++) {
2159 #ifdef HAS_RESTRICTED_HASHES
2160 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
2162 HE *he = hv_iternext(hv);
2164 SV *key = hv_iterkeysv(he);
2165 av_store(av, AvFILLp(av)+1, key); /* av_push(), really */
2168 qsort((char *) AvARRAY(av), len, sizeof(SV *), sortcmp);
2170 for (i = 0; i < len; i++) {
2171 unsigned char flags;
2175 SV *key = av_shift(av);
2176 HE *he = hv_fetch_ent(hv, key, 0, 0);
2177 SV *val = HeVAL(he);
2179 return 1; /* Internal error, not I/O error */
2182 * Store value first.
2185 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
2187 if ((ret = store(cxt, val))) /* Extra () for -Wall, grr... */
2192 * Keys are written after values to make sure retrieval
2193 * can be optimal in terms of memory usage, where keys are
2194 * read into a fixed unique buffer called kbuf.
2195 * See retrieve_hash() for details.
2198 /* Implementation of restricted hashes isn't nicely
2201 = (((hash_flags & SHV_RESTRICTED)
2203 ? SHV_K_LOCKED : 0);
2204 if (val == &PL_sv_undef)
2205 flags |= SHV_K_PLACEHOLDER;
2207 keyval = SvPV(key, keylen_tmp);
2208 keylen = keylen_tmp;
2209 #ifdef HAS_UTF8_HASHES
2210 /* If you build without optimisation on pre 5.6
2211 then nothing spots that SvUTF8(key) is always 0,
2212 so the block isn't optimised away, at which point
2213 the linker dislikes the reference to
2216 const char *keysave = keyval;
2217 bool is_utf8 = TRUE;
2219 /* Just casting the &klen to (STRLEN) won't work
2220 well if STRLEN and I32 are of different widths.
2222 keyval = (char*)bytes_from_utf8((U8*)keyval,
2226 /* If we were able to downgrade here, then than
2227 means that we have a key which only had chars
2228 0-255, but was utf8 encoded. */
2230 if (keyval != keysave) {
2231 keylen = keylen_tmp;
2232 flags |= SHV_K_WASUTF8;
2234 /* keylen_tmp can't have changed, so no need
2235 to assign back to keylen. */
2236 flags |= SHV_K_UTF8;
2243 TRACEME(("(#%d) key '%s' flags %x %u", i, keyval, flags, *keyval));
2245 assert (flags == 0);
2246 TRACEME(("(#%d) key '%s'", i, keyval));
2250 WRITE(keyval, keylen);
2251 if (flags & SHV_K_WASUTF8)
2256 * Free up the temporary array
2265 * Storing in "random" order (in the order the keys are stored
2266 * within the the hash). This is the default and will be faster!
2269 for (i = 0; i < len; i++) {
2272 unsigned char flags;
2273 #ifdef HV_ITERNEXT_WANTPLACEHOLDERS
2274 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
2276 HE *he = hv_iternext(hv);
2278 SV *val = (he ? hv_iterval(hv, he) : 0);
2283 return 1; /* Internal error, not I/O error */
2286 * Store value first.
2289 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
2291 if ((ret = store(cxt, val))) /* Extra () for -Wall, grr... */
2294 /* Implementation of restricted hashes isn't nicely
2297 = (((hash_flags & SHV_RESTRICTED)
2299 ? SHV_K_LOCKED : 0);
2300 if (val == &PL_sv_undef)
2301 flags |= SHV_K_PLACEHOLDER;
2303 hek = HeKEY_hek(he);
2305 if (len == HEf_SVKEY) {
2306 /* This is somewhat sick, but the internal APIs are
2307 * such that XS code could put one of these in in
2309 * Maybe we should be capable of storing one if
2312 key_sv = HeKEY_sv(he);
2313 flags |= SHV_K_ISSV;
2315 /* Regular string key. */
2316 #ifdef HAS_HASH_KEY_FLAGS
2318 flags |= SHV_K_UTF8;
2319 if (HEK_WASUTF8(hek))
2320 flags |= SHV_K_WASUTF8;
2326 * Keys are written after values to make sure retrieval
2327 * can be optimal in terms of memory usage, where keys are
2328 * read into a fixed unique buffer called kbuf.
2329 * See retrieve_hash() for details.
2334 TRACEME(("(#%d) key '%s' flags %x", i, key, flags));
2336 assert (flags == 0);
2337 TRACEME(("(#%d) key '%s'", i, key));
2339 if (flags & SHV_K_ISSV) {
2349 TRACEME(("ok (hash 0x%"UVxf")", PTR2UV(hv)));
2352 HvRITER(hv) = riter; /* Restore hash iterator state */
2353 HvEITER(hv) = eiter;
2361 * Store a code reference.
2363 * Layout is SX_CODE <length> followed by a scalar containing the perl
2364 * source code of the code reference.
2366 static int store_code(stcxt_t *cxt, CV *cv)
2368 #if PERL_VERSION < 6
2370 * retrieve_code does not work with perl 5.005 or less
2372 return store_other(cxt, (SV*)cv);
2376 int ret, count, reallen;
2377 SV *text, *bdeparse;
2379 TRACEME(("store_code (0x%"UVxf")", PTR2UV(cv)));
2382 cxt->deparse == 0 ||
2383 (cxt->deparse < 0 && !(cxt->deparse =
2384 SvTRUE(perl_get_sv("Storable::Deparse", TRUE)) ? 1 : 0))
2386 return store_other(cxt, (SV*)cv);
2390 * Require B::Deparse. At least B::Deparse 0.61 is needed for
2391 * blessed code references.
2393 /* XXX sv_2mortal seems to be evil here. why? */
2394 load_module(PERL_LOADMOD_NOIMPORT, newSVpvn("B::Deparse",10), newSVnv(0.61));
2400 * create the B::Deparse object
2404 XPUSHs(sv_2mortal(newSVpvn("B::Deparse",10)));
2406 count = call_method("new", G_SCALAR);
2409 CROAK(("Unexpected return value from B::Deparse::new\n"));
2413 * call the coderef2text method
2417 XPUSHs(bdeparse); /* XXX is this already mortal? */
2418 XPUSHs(sv_2mortal(newRV_inc((SV*)cv)));
2420 count = call_method("coderef2text", G_SCALAR);
2423 CROAK(("Unexpected return value from B::Deparse::coderef2text\n"));
2427 reallen = strlen(SvPV(text,PL_na));
2430 * Empty code references or XS functions are deparsed as
2431 * "(prototype) ;" or ";".
2434 if (len == 0 || *(SvPV(text,PL_na)+reallen-1) == ';') {
2435 CROAK(("The result of B::Deparse::coderef2text was empty - maybe you're trying to serialize an XS function?\n"));
2439 * Signal code by emitting SX_CODE.
2443 TRACEME(("size = %d", len));
2444 TRACEME(("code = %s", SvPV(text,PL_na)));
2447 * Now store the source code.
2450 STORE_SCALAR(SvPV(text,PL_na), len);
2455 TRACEME(("ok (code)"));
2464 * When storing a tied object (be it a tied scalar, array or hash), we lay out
2465 * a special mark, followed by the underlying tied object. For instance, when
2466 * dealing with a tied hash, we store SX_TIED_HASH <hash object>, where
2467 * <hash object> stands for the serialization of the tied hash.
2469 static int store_tied(stcxt_t *cxt, SV *sv)
2473 int svt = SvTYPE(sv);
2476 TRACEME(("store_tied (0x%"UVxf")", PTR2UV(sv)));
2479 * We have a small run-time penalty here because we chose to factorise
2480 * all tieds objects into the same routine, and not have a store_tied_hash,
2481 * a store_tied_array, etc...
2483 * Don't use a switch() statement, as most compilers don't optimize that
2484 * well for 2/3 values. An if() else if() cascade is just fine. We put
2485 * tied hashes first, as they are the most likely beasts.
2488 if (svt == SVt_PVHV) {
2489 TRACEME(("tied hash"));
2490 PUTMARK(SX_TIED_HASH); /* Introduces tied hash */
2491 } else if (svt == SVt_PVAV) {
2492 TRACEME(("tied array"));
2493 PUTMARK(SX_TIED_ARRAY); /* Introduces tied array */
2495 TRACEME(("tied scalar"));
2496 PUTMARK(SX_TIED_SCALAR); /* Introduces tied scalar */
2500 if (!(mg = mg_find(sv, mtype)))
2501 CROAK(("No magic '%c' found while storing tied %s", mtype,
2502 (svt == SVt_PVHV) ? "hash" :
2503 (svt == SVt_PVAV) ? "array" : "scalar"));
2506 * The mg->mg_obj found by mg_find() above actually points to the
2507 * underlying tied Perl object implementation. For instance, if the
2508 * original SV was that of a tied array, then mg->mg_obj is an AV.
2510 * Note that we store the Perl object as-is. We don't call its FETCH
2511 * method along the way. At retrieval time, we won't call its STORE
2512 * method either, but the tieing magic will be re-installed. In itself,
2513 * that ensures that the tieing semantics are preserved since futher
2514 * accesses on the retrieved object will indeed call the magic methods...
2517 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
2520 TRACEME(("ok (tied)"));
2528 * Stores a reference to an item within a tied structure:
2530 * . \$h{key}, stores both the (tied %h) object and 'key'.
2531 * . \$a[idx], stores both the (tied @a) object and 'idx'.
2533 * Layout is therefore either:
2534 * SX_TIED_KEY <object> <key>
2535 * SX_TIED_IDX <object> <index>
2537 static int store_tied_item(stcxt_t *cxt, SV *sv)
2542 TRACEME(("store_tied_item (0x%"UVxf")", PTR2UV(sv)));
2544 if (!(mg = mg_find(sv, 'p')))
2545 CROAK(("No magic 'p' found while storing reference to tied item"));
2548 * We discriminate between \$h{key} and \$a[idx] via mg_ptr.
2552 TRACEME(("store_tied_item: storing a ref to a tied hash item"));
2553 PUTMARK(SX_TIED_KEY);
2554 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2556 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
2559 TRACEME(("store_tied_item: storing PTR 0x%"UVxf, PTR2UV(mg->mg_ptr)));
2561 if ((ret = store(cxt, (SV *) mg->mg_ptr))) /* Idem, for -Wall */
2564 I32 idx = mg->mg_len;
2566 TRACEME(("store_tied_item: storing a ref to a tied array item "));
2567 PUTMARK(SX_TIED_IDX);
2568 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2570 if ((ret = store(cxt, mg->mg_obj))) /* Idem, for -Wall */
2573 TRACEME(("store_tied_item: storing IDX %d", idx));
2578 TRACEME(("ok (tied item)"));
2584 * store_hook -- dispatched manually, not via sv_store[]
2586 * The blessed SV is serialized by a hook.
2590 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2592 * where <flags> indicates how long <len>, <len2> and <len3> are, whether
2593 * the trailing part [] is present, the type of object (scalar, array or hash).
2594 * There is also a bit which says how the classname is stored between:
2599 * and when the <index> form is used (classname already seen), the "large
2600 * classname" bit in <flags> indicates how large the <index> is.
2602 * The serialized string returned by the hook is of length <len2> and comes
2603 * next. It is an opaque string for us.
2605 * Those <len3> object IDs which are listed last represent the extra references
2606 * not directly serialized by the hook, but which are linked to the object.
2608 * When recursion is mandated to resolve object-IDs not yet seen, we have
2609 * instead, with <header> being flags with bits set to indicate the object type
2610 * and that recursion was indeed needed:
2612 * SX_HOOK <header> <object> <header> <object> <flags>
2614 * that same header being repeated between serialized objects obtained through
2615 * recursion, until we reach flags indicating no recursion, at which point
2616 * we know we've resynchronized with a single layout, after <flags>.
2618 * When storing a blessed ref to a tied variable, the following format is
2621 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
2623 * The first <flags> indication carries an object of type SHT_EXTRA, and the
2624 * real object type is held in the <extra> flag. At the very end of the
2625 * serialization stream, the underlying magic object is serialized, just like
2626 * any other tied variable.
2628 static int store_hook(
2641 int count; /* really len3 + 1 */
2642 unsigned char flags;
2645 int recursed = 0; /* counts recursion */
2646 int obj_type; /* object type, on 2 bits */
2649 int clone = cxt->optype & ST_CLONE;
2650 char mtype = '\0'; /* for blessed ref to tied structures */
2651 unsigned char eflags = '\0'; /* used when object type is SHT_EXTRA */
2653 TRACEME(("store_hook, class \"%s\", tagged #%d", HvNAME(pkg), cxt->tagnum));
2656 * Determine object type on 2 bits.
2661 obj_type = SHT_SCALAR;
2664 obj_type = SHT_ARRAY;
2667 obj_type = SHT_HASH;
2671 * Produced by a blessed ref to a tied data structure, $o in the
2672 * following Perl code.
2676 * my $o = bless \%h, 'BAR';
2678 * Signal the tie-ing magic by setting the object type as SHT_EXTRA
2679 * (since we have only 2 bits in <flags> to store the type), and an
2680 * <extra> byte flag will be emitted after the FIRST <flags> in the
2681 * stream, carrying what we put in `eflags'.
2683 obj_type = SHT_EXTRA;
2684 switch (SvTYPE(sv)) {
2686 eflags = (unsigned char) SHT_THASH;
2690 eflags = (unsigned char) SHT_TARRAY;
2694 eflags = (unsigned char) SHT_TSCALAR;
2700 CROAK(("Unexpected object type (%d) in store_hook()", type));
2702 flags = SHF_NEED_RECURSE | obj_type;
2704 class = HvNAME(pkg);
2705 len = strlen(class);
2708 * To call the hook, we need to fake a call like:
2710 * $object->STORABLE_freeze($cloning);
2712 * but we don't have the $object here. For instance, if $object is
2713 * a blessed array, what we have in `sv' is the array, and we can't
2714 * call a method on those.
2716 * Therefore, we need to create a temporary reference to the object and
2717 * make the call on that reference.
2720 TRACEME(("about to call STORABLE_freeze on class %s", class));
2722 ref = newRV_noinc(sv); /* Temporary reference */
2723 av = array_call(ref, hook, clone); /* @a = $object->STORABLE_freeze($c) */
2725 SvREFCNT_dec(ref); /* Reclaim temporary reference */
2727 count = AvFILLp(av) + 1;
2728 TRACEME(("store_hook, array holds %d items", count));
2731 * If they return an empty list, it means they wish to ignore the
2732 * hook for this class (and not just this instance -- that's for them
2733 * to handle if they so wish).
2735 * Simply disable the cached entry for the hook (it won't be recomputed
2736 * since it's present in the cache) and recurse to store_blessed().
2741 * They must not change their mind in the middle of a serialization.
2744 if (hv_fetch(cxt->hclass, class, len, FALSE))
2745 CROAK(("Too late to ignore hooks for %s class \"%s\"",
2746 (cxt->optype & ST_CLONE) ? "cloning" : "storing", class));
2748 pkg_hide(cxt->hook, pkg, "STORABLE_freeze");
2750 ASSERT(!pkg_can(cxt->hook, pkg, "STORABLE_freeze"), ("hook invisible"));
2751 TRACEME(("ignoring STORABLE_freeze in class \"%s\"", class));
2753 return store_blessed(cxt, sv, type, pkg);
2757 * Get frozen string.
2761 pv = SvPV(ary[0], len2);
2764 * If they returned more than one item, we need to serialize some
2765 * extra references if not already done.
2767 * Loop over the array, starting at position #1, and for each item,
2768 * ensure it is a reference, serialize it if not already done, and
2769 * replace the entry with the tag ID of the corresponding serialized
2772 * We CHEAT by not calling av_fetch() and read directly within the
2776 for (i = 1; i < count; i++) {
2780 AV *av_hook = cxt->hook_seen;
2783 CROAK(("Item #%d returned by STORABLE_freeze "
2784 "for %s is not a reference", i, class));
2785 xsv = SvRV(rsv); /* Follow ref to know what to look for */
2788 * Look in hseen and see if we have a tag already.
2789 * Serialize entry if not done already, and get its tag.
2792 if ((svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE)))
2793 goto sv_seen; /* Avoid moving code too far to the right */
2795 TRACEME(("listed object %d at 0x%"UVxf" is unknown", i-1, PTR2UV(xsv)));
2798 * We need to recurse to store that object and get it to be known
2799 * so that we can resolve the list of object-IDs at retrieve time.
2801 * The first time we do this, we need to emit the proper header
2802 * indicating that we recursed, and what the type of object is (the
2803 * object we're storing via a user-hook). Indeed, during retrieval,
2804 * we'll have to create the object before recursing to retrieve the
2805 * others, in case those would point back at that object.
2808 /* [SX_HOOK] <flags> [<extra>] <object>*/
2812 if (obj_type == SHT_EXTRA)
2817 if ((ret = store(cxt, xsv))) /* Given by hook for us to store */
2820 svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE);
2822 CROAK(("Could not serialize item #%d from hook in %s", i, class));
2825 * It was the first time we serialized `xsv'.
2827 * Keep this SV alive until the end of the serialization: if we
2828 * disposed of it right now by decrementing its refcount, and it was
2829 * a temporary value, some next temporary value allocated during
2830 * another STORABLE_freeze might take its place, and we'd wrongly
2831 * assume that new SV was already serialized, based on its presence
2834 * Therefore, push it away in cxt->hook_seen.
2837 av_store(av_hook, AvFILLp(av_hook)+1, SvREFCNT_inc(xsv));
2841 * Dispose of the REF they returned. If we saved the `xsv' away
2842 * in the array of returned SVs, that will not cause the underlying
2843 * referenced SV to be reclaimed.
2846 ASSERT(SvREFCNT(xsv) > 1, ("SV will survive disposal of its REF"));
2847 SvREFCNT_dec(rsv); /* Dispose of reference */
2850 * Replace entry with its tag (not a real SV, so no refcnt increment)
2854 TRACEME(("listed object %d at 0x%"UVxf" is tag #%"UVuf,
2855 i-1, PTR2UV(xsv), PTR2UV(*svh)));
2859 * Allocate a class ID if not already done.
2861 * This needs to be done after the recursion above, since at retrieval
2862 * time, we'll see the inner objects first. Many thanks to
2863 * Salvador Ortiz Garcia <sog@msg.com.mx> who spot that bug and
2864 * proposed the right fix. -- RAM, 15/09/2000
2867 if (!known_class(cxt, class, len, &classnum)) {
2868 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2869 classnum = -1; /* Mark: we must store classname */
2871 TRACEME(("already seen class %s, ID = %d", class, classnum));
2875 * Compute leading flags.
2879 if (((classnum == -1) ? len : classnum) > LG_SCALAR)
2880 flags |= SHF_LARGE_CLASSLEN;
2882 flags |= SHF_IDX_CLASSNAME;
2883 if (len2 > LG_SCALAR)
2884 flags |= SHF_LARGE_STRLEN;
2886 flags |= SHF_HAS_LIST;
2887 if (count > (LG_SCALAR + 1))
2888 flags |= SHF_LARGE_LISTLEN;
2891 * We're ready to emit either serialized form:
2893 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2894 * SX_HOOK <flags> <index> <len2> <str> [<len3> <object-IDs>]
2896 * If we recursed, the SX_HOOK has already been emitted.
2899 TRACEME(("SX_HOOK (recursed=%d) flags=0x%x "
2900 "class=%"IVdf" len=%"IVdf" len2=%"IVdf" len3=%d",
2901 recursed, flags, (IV)classnum, (IV)len, (IV)len2, count-1));
2903 /* SX_HOOK <flags> [<extra>] */
2907 if (obj_type == SHT_EXTRA)
2912 /* <len> <classname> or <index> */
2913 if (flags & SHF_IDX_CLASSNAME) {
2914 if (flags & SHF_LARGE_CLASSLEN)
2917 unsigned char cnum = (unsigned char) classnum;
2921 if (flags & SHF_LARGE_CLASSLEN)
2924 unsigned char clen = (unsigned char) len;
2927 WRITE(class, len); /* Final \0 is omitted */
2930 /* <len2> <frozen-str> */
2931 if (flags & SHF_LARGE_STRLEN) {
2932 I32 wlen2 = len2; /* STRLEN might be 8 bytes */
2933 WLEN(wlen2); /* Must write an I32 for 64-bit machines */
2935 unsigned char clen = (unsigned char) len2;
2939 WRITE(pv, (SSize_t)len2); /* Final \0 is omitted */
2941 /* [<len3> <object-IDs>] */
2942 if (flags & SHF_HAS_LIST) {
2943 int len3 = count - 1;
2944 if (flags & SHF_LARGE_LISTLEN)
2947 unsigned char clen = (unsigned char) len3;
2952 * NOTA BENE, for 64-bit machines: the ary[i] below does not yield a
2953 * real pointer, rather a tag number, well under the 32-bit limit.
2956 for (i = 1; i < count; i++) {
2957 I32 tagval = htonl(LOW_32BITS(ary[i]));
2959 TRACEME(("object %d, tag #%d", i-1, ntohl(tagval)));
2964 * Free the array. We need extra care for indices after 0, since they
2965 * don't hold real SVs but integers cast.
2969 AvFILLp(av) = 0; /* Cheat, nothing after 0 interests us */
2974 * If object was tied, need to insert serialization of the magic object.
2977 if (obj_type == SHT_EXTRA) {
2980 if (!(mg = mg_find(sv, mtype))) {
2981 int svt = SvTYPE(sv);
2982 CROAK(("No magic '%c' found while storing ref to tied %s with hook",
2983 mtype, (svt == SVt_PVHV) ? "hash" :
2984 (svt == SVt_PVAV) ? "array" : "scalar"));
2987 TRACEME(("handling the magic object 0x%"UVxf" part of 0x%"UVxf,
2988 PTR2UV(mg->mg_obj), PTR2UV(sv)));
2994 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
3002 * store_blessed -- dispatched manually, not via sv_store[]
3004 * Check whether there is a STORABLE_xxx hook defined in the class or in one
3005 * of its ancestors. If there is, then redispatch to store_hook();
3007 * Otherwise, the blessed SV is stored using the following layout:
3009 * SX_BLESS <flag> <len> <classname> <object>
3011 * where <flag> indicates whether <len> is stored on 0 or 4 bytes, depending
3012 * on the high-order bit in flag: if 1, then length follows on 4 bytes.
3013 * Otherwise, the low order bits give the length, thereby giving a compact
3014 * representation for class names less than 127 chars long.
3016 * Each <classname> seen is remembered and indexed, so that the next time
3017 * an object in the blessed in the same <classname> is stored, the following
3020 * SX_IX_BLESS <flag> <index> <object>
3022 * where <index> is the classname index, stored on 0 or 4 bytes depending
3023 * on the high-order bit in flag (same encoding as above for <len>).
3025 static int store_blessed(
3036 TRACEME(("store_blessed, type %d, class \"%s\"", type, HvNAME(pkg)));
3039 * Look for a hook for this blessed SV and redirect to store_hook()
3043 hook = pkg_can(cxt->hook, pkg, "STORABLE_freeze");
3045 return store_hook(cxt, sv, type, pkg, hook);
3048 * This is a blessed SV without any serialization hook.
3051 class = HvNAME(pkg);
3052 len = strlen(class);
3054 TRACEME(("blessed 0x%"UVxf" in %s, no hook: tagged #%d",
3055 PTR2UV(sv), class, cxt->tagnum));
3058 * Determine whether it is the first time we see that class name (in which
3059 * case it will be stored in the SX_BLESS form), or whether we already
3060 * saw that class name before (in which case the SX_IX_BLESS form will be
3064 if (known_class(cxt, class, len, &classnum)) {
3065 TRACEME(("already seen class %s, ID = %d", class, classnum));
3066 PUTMARK(SX_IX_BLESS);
3067 if (classnum <= LG_BLESS) {
3068 unsigned char cnum = (unsigned char) classnum;
3071 unsigned char flag = (unsigned char) 0x80;
3076 TRACEME(("first time we see class %s, ID = %d", class, classnum));
3078 if (len <= LG_BLESS) {
3079 unsigned char clen = (unsigned char) len;
3082 unsigned char flag = (unsigned char) 0x80;
3084 WLEN(len); /* Don't BER-encode, this should be rare */
3086 WRITE(class, len); /* Final \0 is omitted */
3090 * Now emit the <object> part.
3093 return SV_STORE(type)(cxt, sv);
3099 * We don't know how to store the item we reached, so return an error condition.
3100 * (it's probably a GLOB, some CODE reference, etc...)
3102 * If they defined the `forgive_me' variable at the Perl level to some
3103 * true value, then don't croak, just warn, and store a placeholder string
3106 static int store_other(stcxt_t *cxt, SV *sv)
3109 static char buf[80];
3111 TRACEME(("store_other"));
3114 * Fetch the value from perl only once per store() operation.
3118 cxt->forgive_me == 0 ||
3119 (cxt->forgive_me < 0 && !(cxt->forgive_me =
3120 SvTRUE(perl_get_sv("Storable::forgive_me", TRUE)) ? 1 : 0))
3122 CROAK(("Can't store %s items", sv_reftype(sv, FALSE)));
3124 warn("Can't store item %s(0x%"UVxf")",
3125 sv_reftype(sv, FALSE), PTR2UV(sv));
3128 * Store placeholder string as a scalar instead...
3131 (void) sprintf(buf, "You lost %s(0x%"UVxf")%c", sv_reftype(sv, FALSE),
3132 PTR2UV(sv), (char) 0);
3135 STORE_SCALAR(buf, len);
3136 TRACEME(("ok (dummy \"%s\", length = %"IVdf")", buf, (IV) len));
3142 *** Store driving routines
3148 * WARNING: partially duplicates Perl's sv_reftype for speed.
3150 * Returns the type of the SV, identified by an integer. That integer
3151 * may then be used to index the dynamic routine dispatch table.
3153 static int sv_type(SV *sv)
3155 switch (SvTYPE(sv)) {
3160 * No need to check for ROK, that can't be set here since there
3161 * is no field capable of hodling the xrv_rv reference.
3169 * Starting from SVt_PV, it is possible to have the ROK flag
3170 * set, the pointer to the other SV being either stored in
3171 * the xrv_rv (in the case of a pure SVt_RV), or as the
3172 * xpv_pv field of an SVt_PV and its heirs.
3174 * However, those SV cannot be magical or they would be an
3175 * SVt_PVMG at least.
3177 return SvROK(sv) ? svis_REF : svis_SCALAR;
3179 case SVt_PVLV: /* Workaround for perl5.004_04 "LVALUE" bug */
3180 if (SvRMAGICAL(sv) && (mg_find(sv, 'p')))
3181 return svis_TIED_ITEM;
3184 if (SvRMAGICAL(sv) && (mg_find(sv, 'q')))
3186 return SvROK(sv) ? svis_REF : svis_SCALAR;
3188 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
3192 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
3207 * Recursively store objects pointed to by the sv to the specified file.
3209 * Layout is <content> or SX_OBJECT <tagnum> if we reach an already stored
3210 * object (one for which storage has started -- it may not be over if we have
3211 * a self-referenced structure). This data set forms a stored <object>.
3213 static int store(stcxt_t *cxt, SV *sv)
3218 HV *hseen = cxt->hseen;
3220 TRACEME(("store (0x%"UVxf")", PTR2UV(sv)));
3223 * If object has already been stored, do not duplicate data.
3224 * Simply emit the SX_OBJECT marker followed by its tag data.
3225 * The tag is always written in network order.
3227 * NOTA BENE, for 64-bit machines: the "*svh" below does not yield a
3228 * real pointer, rather a tag number (watch the insertion code below).
3229 * That means it probably safe to assume it is well under the 32-bit limit,
3230 * and makes the truncation safe.
3231 * -- RAM, 14/09/1999
3234 svh = hv_fetch(hseen, (char *) &sv, sizeof(sv), FALSE);
3236 I32 tagval = htonl(LOW_32BITS(*svh));
3238 TRACEME(("object 0x%"UVxf" seen as #%d", PTR2UV(sv), ntohl(tagval)));
3246 * Allocate a new tag and associate it with the address of the sv being
3247 * stored, before recursing...
3249 * In order to avoid creating new SvIVs to hold the tagnum we just
3250 * cast the tagnum to an SV pointer and store that in the hash. This
3251 * means that we must clean up the hash manually afterwards, but gives
3252 * us a 15% throughput increase.
3257 if (!hv_store(hseen,
3258 (char *) &sv, sizeof(sv), INT2PTR(SV*, cxt->tagnum), 0))
3262 * Store `sv' and everything beneath it, using appropriate routine.
3263 * Abort immediately if we get a non-zero status back.
3268 TRACEME(("storing 0x%"UVxf" tag #%d, type %d...",
3269 PTR2UV(sv), cxt->tagnum, type));
3272 HV *pkg = SvSTASH(sv);
3273 ret = store_blessed(cxt, sv, type, pkg);
3275 ret = SV_STORE(type)(cxt, sv);
3277 TRACEME(("%s (stored 0x%"UVxf", refcnt=%d, %s)",
3278 ret ? "FAILED" : "ok", PTR2UV(sv),
3279 SvREFCNT(sv), sv_reftype(sv, FALSE)));
3287 * Write magic number and system information into the file.
3288 * Layout is <magic> <network> [<len> <byteorder> <sizeof int> <sizeof long>
3289 * <sizeof ptr>] where <len> is the length of the byteorder hexa string.
3290 * All size and lenghts are written as single characters here.
3292 * Note that no byte ordering info is emitted when <network> is true, since
3293 * integers will be emitted in network order in that case.
3295 static int magic_write(stcxt_t *cxt)
3298 * Starting with 0.6, the "use_network_order" byte flag is also used to
3299 * indicate the version number of the binary image, encoded in the upper
3300 * bits. The bit 0 is always used to indicate network order.
3303 * Starting with 0.7, a full byte is dedicated to the minor version of
3304 * the binary format, which is incremented only when new markers are
3305 * introduced, for instance, but when backward compatibility is preserved.
3308 /* Make these at compile time. The WRITE() macro is sufficiently complex
3309 that it saves about 200 bytes doing it this way and only using it
3311 static const unsigned char network_file_header[] = {
3313 (STORABLE_BIN_MAJOR << 1) | 1,
3314 STORABLE_BIN_WRITE_MINOR
3316 static const unsigned char file_header[] = {
3318 (STORABLE_BIN_MAJOR << 1) | 0,
3319 STORABLE_BIN_WRITE_MINOR,
3320 /* sizeof the array includes the 0 byte at the end: */
3321 (char) sizeof (byteorderstr) - 1,
3323 (unsigned char) sizeof(int),
3324 (unsigned char) sizeof(long),
3325 (unsigned char) sizeof(char *),
3326 (unsigned char) sizeof(NV)
3328 #ifdef USE_56_INTERWORK_KLUDGE
3329 static const unsigned char file_header_56[] = {
3331 (STORABLE_BIN_MAJOR << 1) | 0,
3332 STORABLE_BIN_WRITE_MINOR,
3333 /* sizeof the array includes the 0 byte at the end: */
3334 (char) sizeof (byteorderstr_56) - 1,
3336 (unsigned char) sizeof(int),
3337 (unsigned char) sizeof(long),
3338 (unsigned char) sizeof(char *),
3339 (unsigned char) sizeof(NV)
3342 const unsigned char *header;
3345 TRACEME(("magic_write on fd=%d", cxt->fio ? PerlIO_fileno(cxt->fio) : -1));
3347 if (cxt->netorder) {
3348 header = network_file_header;
3349 length = sizeof (network_file_header);
3351 #ifdef USE_56_INTERWORK_KLUDGE
3352 if (SvTRUE(perl_get_sv("Storable::interwork_56_64bit", TRUE))) {
3353 header = file_header_56;
3354 length = sizeof (file_header_56);
3358 header = file_header;
3359 length = sizeof (file_header);
3364 /* sizeof the array includes the 0 byte at the end. */
3365 header += sizeof (magicstr) - 1;
3366 length -= sizeof (magicstr) - 1;
3369 WRITE(header, length);
3371 if (!cxt->netorder) {
3372 TRACEME(("ok (magic_write byteorder = 0x%lx [%d], I%d L%d P%d D%d)",
3373 (unsigned long) BYTEORDER, (int) sizeof (byteorderstr) - 1,
3374 (int) sizeof(int), (int) sizeof(long),
3375 (int) sizeof(char *), (int) sizeof(NV)));
3383 * Common code for store operations.
3385 * When memory store is requested (f = NULL) and a non null SV* is given in
3386 * `res', it is filled with a new SV created out of the memory buffer.
3388 * It is required to provide a non-null `res' when the operation type is not
3389 * dclone() and store() is performed to memory.
3391 static int do_store(
3401 ASSERT(!(f == 0 && !(optype & ST_CLONE)) || res,
3402 ("must supply result SV pointer for real recursion to memory"));
3404 TRACEME(("do_store (optype=%d, netorder=%d)",
3405 optype, network_order));
3410 * Workaround for CROAK leak: if they enter with a "dirty" context,
3411 * free up memory for them now.
3418 * Now that STORABLE_xxx hooks exist, it is possible that they try to
3419 * re-enter store() via the hooks. We need to stack contexts.
3423 cxt = allocate_context(cxt);
3427 ASSERT(cxt->entry == 1, ("starting new recursion"));
3428 ASSERT(!cxt->s_dirty, ("clean context"));
3431 * Ensure sv is actually a reference. From perl, we called something
3433 * pstore(FILE, \@array);
3434 * so we must get the scalar value behing that reference.
3438 CROAK(("Not a reference"));
3439 sv = SvRV(sv); /* So follow it to know what to store */
3442 * If we're going to store to memory, reset the buffer.
3449 * Prepare context and emit headers.
3452 init_store_context(cxt, f, optype, network_order);
3454 if (-1 == magic_write(cxt)) /* Emit magic and ILP info */
3455 return 0; /* Error */
3458 * Recursively store object...
3461 ASSERT(is_storing(), ("within store operation"));
3463 status = store(cxt, sv); /* Just do it! */
3466 * If they asked for a memory store and they provided an SV pointer,
3467 * make an SV string out of the buffer and fill their pointer.
3469 * When asking for ST_REAL, it's MANDATORY for the caller to provide
3470 * an SV, since context cleanup might free the buffer if we did recurse.
3471 * (unless caller is dclone(), which is aware of that).
3474 if (!cxt->fio && res)
3480 * The "root" context is never freed, since it is meant to be always
3481 * handy for the common case where no recursion occurs at all (i.e.
3482 * we enter store() outside of any Storable code and leave it, period).
3483 * We know it's the "root" context because there's nothing stacked
3488 * When deep cloning, we don't free the context: doing so would force
3489 * us to copy the data in the memory buffer. Sicne we know we're
3490 * about to enter do_retrieve...
3493 clean_store_context(cxt);
3494 if (cxt->prev && !(cxt->optype & ST_CLONE))
3497 TRACEME(("do_store returns %d", status));
3505 * Store the transitive data closure of given object to disk.
3506 * Returns 0 on error, a true value otherwise.
3508 int pstore(PerlIO *f, SV *sv)
3510 TRACEME(("pstore"));
3511 return do_store(f, sv, 0, FALSE, (SV**) 0);
3518 * Same as pstore(), but network order is used for integers and doubles are
3519 * emitted as strings.
3521 int net_pstore(PerlIO *f, SV *sv)
3523 TRACEME(("net_pstore"));
3524 return do_store(f, sv, 0, TRUE, (SV**) 0);
3534 * Build a new SV out of the content of the internal memory buffer.
3536 static SV *mbuf2sv(void)
3540 return newSVpv(mbase, MBUF_SIZE());
3546 * Store the transitive data closure of given object to memory.
3547 * Returns undef on error, a scalar value containing the data otherwise.
3553 TRACEME(("mstore"));
3555 if (!do_store((PerlIO*) 0, sv, 0, FALSE, &out))
3556 return &PL_sv_undef;
3564 * Same as mstore(), but network order is used for integers and doubles are
3565 * emitted as strings.
3567 SV *net_mstore(SV *sv)
3571 TRACEME(("net_mstore"));
3573 if (!do_store((PerlIO*) 0, sv, 0, TRUE, &out))
3574 return &PL_sv_undef;
3580 *** Specific retrieve callbacks.
3586 * Return an error via croak, since it is not possible that we get here
3587 * under normal conditions, when facing a file produced via pstore().
3589 static SV *retrieve_other(stcxt_t *cxt, char *cname)
3592 cxt->ver_major != STORABLE_BIN_MAJOR &&
3593 cxt->ver_minor != STORABLE_BIN_MINOR
3595 CROAK(("Corrupted storable %s (binary v%d.%d), current is v%d.%d",
3596 cxt->fio ? "file" : "string",
3597 cxt->ver_major, cxt->ver_minor,
3598 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
3600 CROAK(("Corrupted storable %s (binary v%d.%d)",
3601 cxt->fio ? "file" : "string",
3602 cxt->ver_major, cxt->ver_minor));
3605 return (SV *) 0; /* Just in case */
3609 * retrieve_idx_blessed
3611 * Layout is SX_IX_BLESS <index> <object> with SX_IX_BLESS already read.
3612 * <index> can be coded on either 1 or 5 bytes.
3614 static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname)
3621 TRACEME(("retrieve_idx_blessed (#%d)", cxt->tagnum));
3622 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3624 GETMARK(idx); /* Index coded on a single char? */
3629 * Fetch classname in `aclass'
3632 sva = av_fetch(cxt->aclass, idx, FALSE);
3634 CROAK(("Class name #%"IVdf" should have been seen already", (IV) idx));
3636 class = SvPVX(*sva); /* We know it's a PV, by construction */
3638 TRACEME(("class ID %d => %s", idx, class));
3641 * Retrieve object and bless it.
3644 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
3652 * Layout is SX_BLESS <len> <classname> <object> with SX_BLESS already read.
3653 * <len> can be coded on either 1 or 5 bytes.
3655 static SV *retrieve_blessed(stcxt_t *cxt, char *cname)
3659 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3662 TRACEME(("retrieve_blessed (#%d)", cxt->tagnum));
3663 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3666 * Decode class name length and read that name.
3668 * Short classnames have two advantages: their length is stored on one
3669 * single byte, and the string can be read on the stack.
3672 GETMARK(len); /* Length coded on a single char? */
3675 TRACEME(("** allocating %d bytes for class name", len+1));
3676 New(10003, class, len+1, char);
3679 class[len] = '\0'; /* Mark string end */
3682 * It's a new classname, otherwise it would have been an SX_IX_BLESS.
3685 TRACEME(("new class name \"%s\" will bear ID = %d", class, cxt->classnum));
3687 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3691 * Retrieve object and bless it.
3694 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
3704 * Layout: SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
3705 * with leading mark already read, as usual.
3707 * When recursion was involved during serialization of the object, there
3708 * is an unknown amount of serialized objects after the SX_HOOK mark. Until
3709 * we reach a <flags> marker with the recursion bit cleared.
3711 * If the first <flags> byte contains a type of SHT_EXTRA, then the real type
3712 * is held in the <extra> byte, and if the object is tied, the serialized
3713 * magic object comes at the very end:
3715 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
3717 * This means the STORABLE_thaw hook will NOT get a tied variable during its
3718 * processing (since we won't have seen the magic object by the time the hook
3719 * is called). See comments below for why it was done that way.
3721 static SV *retrieve_hook(stcxt_t *cxt, char *cname)
3724 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3735 int clone = cxt->optype & ST_CLONE;
3737 unsigned int extra_type = 0;
3739 TRACEME(("retrieve_hook (#%d)", cxt->tagnum));
3740 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3743 * Read flags, which tell us about the type, and whether we need to recurse.
3749 * Create the (empty) object, and mark it as seen.
3751 * This must be done now, because tags are incremented, and during
3752 * serialization, the object tag was affected before recursion could
3756 obj_type = flags & SHF_TYPE_MASK;
3762 sv = (SV *) newAV();
3765 sv = (SV *) newHV();
3769 * Read <extra> flag to know the type of the object.
3770 * Record associated magic type for later.
3772 GETMARK(extra_type);
3773 switch (extra_type) {
3779 sv = (SV *) newAV();
3783 sv = (SV *) newHV();
3787 return retrieve_other(cxt, 0); /* Let it croak */
3791 return retrieve_other(cxt, 0); /* Let it croak */
3793 SEEN(sv, 0); /* Don't bless yet */
3796 * Whilst flags tell us to recurse, do so.
3798 * We don't need to remember the addresses returned by retrieval, because
3799 * all the references will be obtained through indirection via the object
3800 * tags in the object-ID list.
3802 * We need to decrement the reference count for these objects
3803 * because, if the user doesn't save a reference to them in the hook,
3804 * they must be freed when this context is cleaned.
3807 while (flags & SHF_NEED_RECURSE) {
3808 TRACEME(("retrieve_hook recursing..."));
3809 rv = retrieve(cxt, 0);
3813 TRACEME(("retrieve_hook back with rv=0x%"UVxf,
3818 if (flags & SHF_IDX_CLASSNAME) {
3823 * Fetch index from `aclass'
3826 if (flags & SHF_LARGE_CLASSLEN)
3831 sva = av_fetch(cxt->aclass, idx, FALSE);
3833 CROAK(("Class name #%"IVdf" should have been seen already",
3836 class = SvPVX(*sva); /* We know it's a PV, by construction */
3837 TRACEME(("class ID %d => %s", idx, class));
3841 * Decode class name length and read that name.
3843 * NOTA BENE: even if the length is stored on one byte, we don't read
3844 * on the stack. Just like retrieve_blessed(), we limit the name to
3845 * LG_BLESS bytes. This is an arbitrary decision.
3848 if (flags & SHF_LARGE_CLASSLEN)
3853 if (len > LG_BLESS) {
3854 TRACEME(("** allocating %d bytes for class name", len+1));
3855 New(10003, class, len+1, char);
3859 class[len] = '\0'; /* Mark string end */
3862 * Record new classname.
3865 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3869 TRACEME(("class name: %s", class));
3872 * Decode user-frozen string length and read it in an SV.
3874 * For efficiency reasons, we read data directly into the SV buffer.
3875 * To understand that code, read retrieve_scalar()
3878 if (flags & SHF_LARGE_STRLEN)
3883 frozen = NEWSV(10002, len2);
3885 SAFEREAD(SvPVX(frozen), len2, frozen);
3886 SvCUR_set(frozen, len2);
3887 *SvEND(frozen) = '\0';
3889 (void) SvPOK_only(frozen); /* Validates string pointer */
3890 if (cxt->s_tainted) /* Is input source tainted? */
3893 TRACEME(("frozen string: %d bytes", len2));
3896 * Decode object-ID list length, if present.
3899 if (flags & SHF_HAS_LIST) {
3900 if (flags & SHF_LARGE_LISTLEN)
3906 av_extend(av, len3 + 1); /* Leave room for [0] */
3907 AvFILLp(av) = len3; /* About to be filled anyway */
3911 TRACEME(("has %d object IDs to link", len3));
3914 * Read object-ID list into array.
3915 * Because we pre-extended it, we can cheat and fill it manually.
3917 * We read object tags and we can convert them into SV* on the fly
3918 * because we know all the references listed in there (as tags)
3919 * have been already serialized, hence we have a valid correspondance
3920 * between each of those tags and the recreated SV.
3924 SV **ary = AvARRAY(av);
3926 for (i = 1; i <= len3; i++) { /* We leave [0] alone */
3933 svh = av_fetch(cxt->aseen, tag, FALSE);
3935 CROAK(("Object #%"IVdf" should have been retrieved already",
3938 ary[i] = SvREFCNT_inc(xsv);
3943 * Bless the object and look up the STORABLE_thaw hook.
3947 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3950 * Hook not found. Maybe they did not require the module where this
3951 * hook is defined yet?
3953 * If the require below succeeds, we'll be able to find the hook.
3954 * Still, it only works reliably when each class is defined in a
3958 SV *psv = newSVpvn("require ", 8);
3959 sv_catpv(psv, class);
3961 TRACEME(("No STORABLE_thaw defined for objects of class %s", class));
3962 TRACEME(("Going to require module '%s' with '%s'", class, SvPVX(psv)));
3964 perl_eval_sv(psv, G_DISCARD);
3968 * We cache results of pkg_can, so we need to uncache before attempting
3972 pkg_uncache(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3973 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3976 CROAK(("No STORABLE_thaw defined for objects of class %s "
3977 "(even after a \"require %s;\")", class, class));
3981 * If we don't have an `av' yet, prepare one.
3982 * Then insert the frozen string as item [0].
3990 AvARRAY(av)[0] = SvREFCNT_inc(frozen);
3995 * $object->STORABLE_thaw($cloning, $frozen, @refs);
3997 * where $object is our blessed (empty) object, $cloning is a boolean
3998 * telling whether we're running a deep clone, $frozen is the frozen
3999 * string the user gave us in his serializing hook, and @refs, which may
4000 * be empty, is the list of extra references he returned along for us
4003 * In effect, the hook is an alternate creation routine for the class,
4004 * the object itself being already created by the runtime.
4007 TRACEME(("calling STORABLE_thaw on %s at 0x%"UVxf" (%"IVdf" args)",
4008 class, PTR2UV(sv), (IV) AvFILLp(av) + 1));
4011 (void) scalar_call(rv, hook, clone, av, G_SCALAR|G_DISCARD);
4018 SvREFCNT_dec(frozen);
4021 if (!(flags & SHF_IDX_CLASSNAME) && class != buf)
4025 * If we had an <extra> type, then the object was not as simple, and
4026 * we need to restore extra magic now.
4032 TRACEME(("retrieving magic object for 0x%"UVxf"...", PTR2UV(sv)));
4034 rv = retrieve(cxt, 0); /* Retrieve <magic object> */
4036 TRACEME(("restoring the magic object 0x%"UVxf" part of 0x%"UVxf,
4037 PTR2UV(rv), PTR2UV(sv)));
4039 switch (extra_type) {
4041 sv_upgrade(sv, SVt_PVMG);
4044 sv_upgrade(sv, SVt_PVAV);
4045 AvREAL_off((AV *)sv);
4048 sv_upgrade(sv, SVt_PVHV);
4051 CROAK(("Forgot to deal with extra type %d", extra_type));
4056 * Adding the magic only now, well after the STORABLE_thaw hook was called
4057 * means the hook cannot know it deals with an object whose variable is
4058 * tied. But this is happening when retrieving $o in the following case:
4062 * my $o = bless \%h, 'BAR';
4064 * The 'BAR' class is NOT the one where %h is tied into. Therefore, as
4065 * far as the 'BAR' class is concerned, the fact that %h is not a REAL
4066 * hash but a tied one should not matter at all, and remain transparent.
4067 * This means the magic must be restored by Storable AFTER the hook is
4070 * That looks very reasonable to me, but then I've come up with this
4071 * after a bug report from David Nesting, who was trying to store such
4072 * an object and caused Storable to fail. And unfortunately, it was
4073 * also the easiest way to retrofit support for blessed ref to tied objects
4074 * into the existing design. -- RAM, 17/02/2001
4077 sv_magic(sv, rv, mtype, Nullch, 0);
4078 SvREFCNT_dec(rv); /* Undo refcnt inc from sv_magic() */
4086 * Retrieve reference to some other scalar.
4087 * Layout is SX_REF <object>, with SX_REF already read.
4089 static SV *retrieve_ref(stcxt_t *cxt, char *cname)
4094 TRACEME(("retrieve_ref (#%d)", cxt->tagnum));
4097 * We need to create the SV that holds the reference to the yet-to-retrieve
4098 * object now, so that we may record the address in the seen table.
4099 * Otherwise, if the object to retrieve references us, we won't be able
4100 * to resolve the SX_OBJECT we'll see at that point! Hence we cannot
4101 * do the retrieve first and use rv = newRV(sv) since it will be too late
4102 * for SEEN() recording.
4105 rv = NEWSV(10002, 0);
4106 SEEN(rv, cname); /* Will return if rv is null */
4107 sv = retrieve(cxt, 0); /* Retrieve <object> */
4109 return (SV *) 0; /* Failed */
4112 * WARNING: breaks RV encapsulation.
4114 * Now for the tricky part. We have to upgrade our existing SV, so that
4115 * it is now an RV on sv... Again, we cheat by duplicating the code
4116 * held in newSVrv(), since we already got our SV from retrieve().
4120 * SvRV(rv) = SvREFCNT_inc(sv);
4122 * here because the reference count we got from retrieve() above is
4123 * already correct: if the object was retrieved from the file, then
4124 * its reference count is one. Otherwise, if it was retrieved via
4125 * an SX_OBJECT indication, a ref count increment was done.
4128 sv_upgrade(rv, SVt_RV);
4129 SvRV(rv) = sv; /* $rv = \$sv */
4132 TRACEME(("ok (retrieve_ref at 0x%"UVxf")", PTR2UV(rv)));
4138 * retrieve_overloaded
4140 * Retrieve reference to some other scalar with overloading.
4141 * Layout is SX_OVERLOAD <object>, with SX_OVERLOAD already read.
4143 static SV *retrieve_overloaded(stcxt_t *cxt, char *cname)
4149 TRACEME(("retrieve_overloaded (#%d)", cxt->tagnum));
4152 * Same code as retrieve_ref(), duplicated to avoid extra call.
4155 rv = NEWSV(10002, 0);
4156 SEEN(rv, cname); /* Will return if rv is null */
4157 sv = retrieve(cxt, 0); /* Retrieve <object> */
4159 return (SV *) 0; /* Failed */
4162 * WARNING: breaks RV encapsulation.
4165 sv_upgrade(rv, SVt_RV);
4166 SvRV(rv) = sv; /* $rv = \$sv */
4170 * Restore overloading magic.
4173 stash = (HV *) SvSTASH (sv);
4174 if (!stash || !Gv_AMG(stash))
4175 CROAK(("Cannot restore overloading on %s(0x%"UVxf") (package %s)",
4176 sv_reftype(sv, FALSE),
4178 stash ? HvNAME(stash) : "<unknown>"));
4182 TRACEME(("ok (retrieve_overloaded at 0x%"UVxf")", PTR2UV(rv)));
4188 * retrieve_tied_array
4190 * Retrieve tied array
4191 * Layout is SX_TIED_ARRAY <object>, with SX_TIED_ARRAY already read.
4193 static SV *retrieve_tied_array(stcxt_t *cxt, char *cname)
4198 TRACEME(("retrieve_tied_array (#%d)", cxt->tagnum));
4200 tv = NEWSV(10002, 0);
4201 SEEN(tv, cname); /* Will return if tv is null */
4202 sv = retrieve(cxt, 0); /* Retrieve <object> */
4204 return (SV *) 0; /* Failed */
4206 sv_upgrade(tv, SVt_PVAV);
4207 AvREAL_off((AV *)tv);
4208 sv_magic(tv, sv, 'P', Nullch, 0);
4209 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4211 TRACEME(("ok (retrieve_tied_array at 0x%"UVxf")", PTR2UV(tv)));
4217 * retrieve_tied_hash
4219 * Retrieve tied hash
4220 * Layout is SX_TIED_HASH <object>, with SX_TIED_HASH already read.
4222 static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname)
4227 TRACEME(("retrieve_tied_hash (#%d)", cxt->tagnum));
4229 tv = NEWSV(10002, 0);
4230 SEEN(tv, cname); /* Will return if tv is null */
4231 sv = retrieve(cxt, 0); /* Retrieve <object> */
4233 return (SV *) 0; /* Failed */
4235 sv_upgrade(tv, SVt_PVHV);
4236 sv_magic(tv, sv, 'P', Nullch, 0);
4237 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4239 TRACEME(("ok (retrieve_tied_hash at 0x%"UVxf")", PTR2UV(tv)));
4245 * retrieve_tied_scalar
4247 * Retrieve tied scalar
4248 * Layout is SX_TIED_SCALAR <object>, with SX_TIED_SCALAR already read.
4250 static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname)
4255 TRACEME(("retrieve_tied_scalar (#%d)", cxt->tagnum));
4257 tv = NEWSV(10002, 0);
4258 SEEN(tv, cname); /* Will return if rv is null */
4259 sv = retrieve(cxt, 0); /* Retrieve <object> */
4261 return (SV *) 0; /* Failed */
4263 sv_upgrade(tv, SVt_PVMG);
4264 sv_magic(tv, sv, 'q', Nullch, 0);
4265 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4267 TRACEME(("ok (retrieve_tied_scalar at 0x%"UVxf")", PTR2UV(tv)));
4275 * Retrieve reference to value in a tied hash.
4276 * Layout is SX_TIED_KEY <object> <key>, with SX_TIED_KEY already read.
4278 static SV *retrieve_tied_key(stcxt_t *cxt, char *cname)
4284 TRACEME(("retrieve_tied_key (#%d)", cxt->tagnum));
4286 tv = NEWSV(10002, 0);
4287 SEEN(tv, cname); /* Will return if tv is null */
4288 sv = retrieve(cxt, 0); /* Retrieve <object> */
4290 return (SV *) 0; /* Failed */
4292 key = retrieve(cxt, 0); /* Retrieve <key> */
4294 return (SV *) 0; /* Failed */
4296 sv_upgrade(tv, SVt_PVMG);
4297 sv_magic(tv, sv, 'p', (char *)key, HEf_SVKEY);
4298 SvREFCNT_dec(key); /* Undo refcnt inc from sv_magic() */
4299 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4307 * Retrieve reference to value in a tied array.
4308 * Layout is SX_TIED_IDX <object> <idx>, with SX_TIED_IDX already read.
4310 static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname)
4316 TRACEME(("retrieve_tied_idx (#%d)", cxt->tagnum));
4318 tv = NEWSV(10002, 0);
4319 SEEN(tv, cname); /* Will return if tv is null */
4320 sv = retrieve(cxt, 0); /* Retrieve <object> */
4322 return (SV *) 0; /* Failed */
4324 RLEN(idx); /* Retrieve <idx> */
4326 sv_upgrade(tv, SVt_PVMG);
4327 sv_magic(tv, sv, 'p', Nullch, idx);
4328 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4337 * Retrieve defined long (string) scalar.
4339 * Layout is SX_LSCALAR <length> <data>, with SX_LSCALAR already read.
4340 * The scalar is "long" in that <length> is larger than LG_SCALAR so it
4341 * was not stored on a single byte.
4343 static SV *retrieve_lscalar(stcxt_t *cxt, char *cname)
4349 TRACEME(("retrieve_lscalar (#%d), len = %"IVdf, cxt->tagnum, (IV) len));
4352 * Allocate an empty scalar of the suitable length.
4355 sv = NEWSV(10002, len);
4356 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4359 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
4361 * Now, for efficiency reasons, read data directly inside the SV buffer,
4362 * and perform the SV final settings directly by duplicating the final
4363 * work done by sv_setpv. Since we're going to allocate lots of scalars
4364 * this way, it's worth the hassle and risk.
4367 SAFEREAD(SvPVX(sv), len, sv);
4368 SvCUR_set(sv, len); /* Record C string length */
4369 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
4370 (void) SvPOK_only(sv); /* Validate string pointer */
4371 if (cxt->s_tainted) /* Is input source tainted? */
4372 SvTAINT(sv); /* External data cannot be trusted */
4374 TRACEME(("large scalar len %"IVdf" '%s'", (IV) len, SvPVX(sv)));
4375 TRACEME(("ok (retrieve_lscalar at 0x%"UVxf")", PTR2UV(sv)));
4383 * Retrieve defined short (string) scalar.
4385 * Layout is SX_SCALAR <length> <data>, with SX_SCALAR already read.
4386 * The scalar is "short" so <length> is single byte. If it is 0, there
4387 * is no <data> section.
4389 static SV *retrieve_scalar(stcxt_t *cxt, char *cname)
4395 TRACEME(("retrieve_scalar (#%d), len = %d", cxt->tagnum, len));
4398 * Allocate an empty scalar of the suitable length.
4401 sv = NEWSV(10002, len);
4402 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4405 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
4410 * newSV did not upgrade to SVt_PV so the scalar is undefined.
4411 * To make it defined with an empty length, upgrade it now...
4412 * Don't upgrade to a PV if the original type contains more
4413 * information than a scalar.
4415 if (SvTYPE(sv) <= SVt_PV) {
4416 sv_upgrade(sv, SVt_PV);
4419 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
4420 TRACEME(("ok (retrieve_scalar empty at 0x%"UVxf")", PTR2UV(sv)));
4423 * Now, for efficiency reasons, read data directly inside the SV buffer,
4424 * and perform the SV final settings directly by duplicating the final
4425 * work done by sv_setpv. Since we're going to allocate lots of scalars
4426 * this way, it's worth the hassle and risk.
4428 SAFEREAD(SvPVX(sv), len, sv);
4429 SvCUR_set(sv, len); /* Record C string length */
4430 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
4431 TRACEME(("small scalar len %d '%s'", len, SvPVX(sv)));
4434 (void) SvPOK_only(sv); /* Validate string pointer */
4435 if (cxt->s_tainted) /* Is input source tainted? */
4436 SvTAINT(sv); /* External data cannot be trusted */
4438 TRACEME(("ok (retrieve_scalar at 0x%"UVxf")", PTR2UV(sv)));
4445 * Like retrieve_scalar(), but tag result as utf8.
4446 * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
4448 static SV *retrieve_utf8str(stcxt_t *cxt, char *cname)
4452 TRACEME(("retrieve_utf8str"));
4454 sv = retrieve_scalar(cxt, cname);
4456 #ifdef HAS_UTF8_SCALARS
4459 if (cxt->use_bytes < 0)
4461 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4463 if (cxt->use_bytes == 0)
4474 * Like retrieve_lscalar(), but tag result as utf8.
4475 * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
4477 static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname)
4481 TRACEME(("retrieve_lutf8str"));
4483 sv = retrieve_lscalar(cxt, cname);
4485 #ifdef HAS_UTF8_SCALARS
4488 if (cxt->use_bytes < 0)
4490 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4492 if (cxt->use_bytes == 0)
4502 * Retrieve defined integer.
4503 * Layout is SX_INTEGER <data>, whith SX_INTEGER already read.
4505 static SV *retrieve_integer(stcxt_t *cxt, char *cname)
4510 TRACEME(("retrieve_integer (#%d)", cxt->tagnum));
4512 READ(&iv, sizeof(iv));
4514 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4516 TRACEME(("integer %"IVdf, iv));
4517 TRACEME(("ok (retrieve_integer at 0x%"UVxf")", PTR2UV(sv)));
4525 * Retrieve defined integer in network order.
4526 * Layout is SX_NETINT <data>, whith SX_NETINT already read.
4528 static SV *retrieve_netint(stcxt_t *cxt, char *cname)
4533 TRACEME(("retrieve_netint (#%d)", cxt->tagnum));
4537 sv = newSViv((int) ntohl(iv));
4538 TRACEME(("network integer %d", (int) ntohl(iv)));
4541 TRACEME(("network integer (as-is) %d", iv));
4543 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4545 TRACEME(("ok (retrieve_netint at 0x%"UVxf")", PTR2UV(sv)));
4553 * Retrieve defined double.
4554 * Layout is SX_DOUBLE <data>, whith SX_DOUBLE already read.
4556 static SV *retrieve_double(stcxt_t *cxt, char *cname)
4561 TRACEME(("retrieve_double (#%d)", cxt->tagnum));
4563 READ(&nv, sizeof(nv));
4565 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4567 TRACEME(("double %"NVff, nv));
4568 TRACEME(("ok (retrieve_double at 0x%"UVxf")", PTR2UV(sv)));
4576 * Retrieve defined byte (small integer within the [-128, +127] range).
4577 * Layout is SX_BYTE <data>, whith SX_BYTE already read.
4579 static SV *retrieve_byte(stcxt_t *cxt, char *cname)
4583 signed char tmp; /* Workaround for AIX cc bug --H.Merijn Brand */
4585 TRACEME(("retrieve_byte (#%d)", cxt->tagnum));
4588 TRACEME(("small integer read as %d", (unsigned char) siv));
4589 tmp = (unsigned char) siv - 128;
4591 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4593 TRACEME(("byte %d", tmp));
4594 TRACEME(("ok (retrieve_byte at 0x%"UVxf")", PTR2UV(sv)));
4602 * Return the undefined value.
4604 static SV *retrieve_undef(stcxt_t *cxt, char *cname)
4608 TRACEME(("retrieve_undef"));
4619 * Return the immortal undefined value.
4621 static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname)
4623 SV *sv = &PL_sv_undef;
4625 TRACEME(("retrieve_sv_undef"));
4634 * Return the immortal yes value.
4636 static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname)
4638 SV *sv = &PL_sv_yes;
4640 TRACEME(("retrieve_sv_yes"));
4649 * Return the immortal no value.
4651 static SV *retrieve_sv_no(stcxt_t *cxt, char *cname)
4655 TRACEME(("retrieve_sv_no"));
4664 * Retrieve a whole array.
4665 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
4666 * Each item is stored as <object>.
4668 * When we come here, SX_ARRAY has been read already.
4670 static SV *retrieve_array(stcxt_t *cxt, char *cname)
4677 TRACEME(("retrieve_array (#%d)", cxt->tagnum));
4680 * Read length, and allocate array, then pre-extend it.
4684 TRACEME(("size = %d", len));
4686 SEEN(av, cname); /* Will return if array not allocated nicely */
4690 return (SV *) av; /* No data follow if array is empty */
4693 * Now get each item in turn...
4696 for (i = 0; i < len; i++) {
4697 TRACEME(("(#%d) item", i));
4698 sv = retrieve(cxt, 0); /* Retrieve item */
4701 if (av_store(av, i, sv) == 0)
4705 TRACEME(("ok (retrieve_array at 0x%"UVxf")", PTR2UV(av)));
4713 * Retrieve a whole hash table.
4714 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4715 * Keys are stored as <length> <data>, the <data> section being omitted
4717 * Values are stored as <object>.
4719 * When we come here, SX_HASH has been read already.
4721 static SV *retrieve_hash(stcxt_t *cxt, char *cname)
4729 TRACEME(("retrieve_hash (#%d)", cxt->tagnum));
4732 * Read length, allocate table.
4736 TRACEME(("size = %d", len));
4738 SEEN(hv, cname); /* Will return if table not allocated properly */
4740 return (SV *) hv; /* No data follow if table empty */
4741 hv_ksplit(hv, len); /* pre-extend hash to save multiple splits */
4744 * Now get each key/value pair in turn...
4747 for (i = 0; i < len; i++) {
4752 TRACEME(("(#%d) value", i));
4753 sv = retrieve(cxt, 0);
4759 * Since we're reading into kbuf, we must ensure we're not
4760 * recursing between the read and the hv_store() where it's used.
4761 * Hence the key comes after the value.
4764 RLEN(size); /* Get key size */
4765 KBUFCHK((STRLEN)size); /* Grow hash key read pool if needed */
4768 kbuf[size] = '\0'; /* Mark string end, just in case */
4769 TRACEME(("(#%d) key '%s'", i, kbuf));
4772 * Enter key/value pair into hash table.
4775 if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
4779 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4787 * Retrieve a whole hash table.
4788 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4789 * Keys are stored as <length> <data>, the <data> section being omitted
4791 * Values are stored as <object>.
4793 * When we come here, SX_HASH has been read already.
4795 static SV *retrieve_flag_hash(stcxt_t *cxt, char *cname)
4804 GETMARK(hash_flags);
4805 TRACEME(("retrieve_flag_hash (#%d)", cxt->tagnum));
4807 * Read length, allocate table.
4810 #ifndef HAS_RESTRICTED_HASHES
4811 if (hash_flags & SHV_RESTRICTED) {
4812 if (cxt->derestrict < 0)
4814 = (SvTRUE(perl_get_sv("Storable::downgrade_restricted", TRUE))
4816 if (cxt->derestrict == 0)
4817 RESTRICTED_HASH_CROAK();
4822 TRACEME(("size = %d, flags = %d", len, hash_flags));
4824 SEEN(hv, cname); /* Will return if table not allocated properly */
4826 return (SV *) hv; /* No data follow if table empty */
4827 hv_ksplit(hv, len); /* pre-extend hash to save multiple splits */
4830 * Now get each key/value pair in turn...
4833 for (i = 0; i < len; i++) {
4835 int store_flags = 0;
4840 TRACEME(("(#%d) value", i));
4841 sv = retrieve(cxt, 0);
4846 #ifdef HAS_RESTRICTED_HASHES
4847 if ((hash_flags & SHV_RESTRICTED) && (flags & SHV_K_LOCKED))
4851 if (flags & SHV_K_ISSV) {
4852 /* XXX you can't set a placeholder with an SV key.
4853 Then again, you can't get an SV key.
4854 Without messing around beyond what the API is supposed to do.
4857 TRACEME(("(#%d) keysv, flags=%d", i, flags));
4858 keysv = retrieve(cxt, 0);
4862 if (!hv_store_ent(hv, keysv, sv, 0))
4867 * Since we're reading into kbuf, we must ensure we're not
4868 * recursing between the read and the hv_store() where it's used.
4869 * Hence the key comes after the value.
4872 if (flags & SHV_K_PLACEHOLDER) {
4875 store_flags |= HVhek_PLACEHOLD;
4877 if (flags & SHV_K_UTF8) {
4878 #ifdef HAS_UTF8_HASHES
4879 store_flags |= HVhek_UTF8;
4881 if (cxt->use_bytes < 0)
4883 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4885 if (cxt->use_bytes == 0)
4889 #ifdef HAS_UTF8_HASHES
4890 if (flags & SHV_K_WASUTF8)
4891 store_flags |= HVhek_WASUTF8;
4894 RLEN(size); /* Get key size */
4895 KBUFCHK((STRLEN)size); /* Grow hash key read pool if needed */
4898 kbuf[size] = '\0'; /* Mark string end, just in case */
4899 TRACEME(("(#%d) key '%s' flags %X store_flags %X", i, kbuf,
4900 flags, store_flags));
4903 * Enter key/value pair into hash table.
4906 #ifdef HAS_RESTRICTED_HASHES
4907 if (hv_store_flags(hv, kbuf, size, sv, 0, flags) == 0)
4910 if (!(store_flags & HVhek_PLACEHOLD))
4911 if (hv_store(hv, kbuf, size, sv, 0) == 0)
4916 #ifdef HAS_RESTRICTED_HASHES
4917 if (hash_flags & SHV_RESTRICTED)
4921 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4929 * Return a code reference.
4931 static SV *retrieve_code(stcxt_t *cxt, char *cname)
4933 #if PERL_VERSION < 6
4934 CROAK(("retrieve_code does not work with perl 5.005 or less\n"));
4939 SV *sv, *text, *sub, *errsv;
4941 TRACEME(("retrieve_code (#%d)", cxt->tagnum));
4944 * Retrieve the source of the code reference
4945 * as a small or large scalar
4951 text = retrieve_scalar(cxt, cname);
4954 text = retrieve_lscalar(cxt, cname);
4957 CROAK(("Unexpected type %d in retrieve_code\n", type));
4961 * prepend "sub " to the source
4964 sub = newSVpvn("sub ", 4);
4965 sv_catpv(sub, SvPV(text, PL_na)); /* XXX no sv_catsv! */
4969 * evaluate the source to a code reference and use the CV value
4972 if (cxt->eval == NULL) {
4973 cxt->eval = perl_get_sv("Storable::Eval", TRUE);
4974 SvREFCNT_inc(cxt->eval);
4976 if (!SvTRUE(cxt->eval)) {
4978 cxt->forgive_me == 0 ||
4979 (cxt->forgive_me < 0 && !(cxt->forgive_me =
4980 SvTRUE(perl_get_sv("Storable::forgive_me", TRUE)) ? 1 : 0))
4982 CROAK(("Can't eval, please set $Storable::Eval to a true value"));
4992 if (SvROK(cxt->eval) && SvTYPE(SvRV(cxt->eval)) == SVt_PVCV) {
4993 SV* errsv = get_sv("@", TRUE);
4994 sv_setpv(errsv, ""); /* clear $@ */
4996 XPUSHs(sv_2mortal(newSVsv(sub)));
4998 count = call_sv(cxt->eval, G_SCALAR);
5001 CROAK(("Unexpected return value from $Storable::Eval callback\n"));
5003 if (SvTRUE(errsv)) {
5004 CROAK(("code %s caused an error: %s", SvPV(sub, PL_na), SvPV(errsv, PL_na)));
5008 cv = eval_pv(SvPV(sub, PL_na), TRUE);
5010 if (cv && SvROK(cv) && SvTYPE(SvRV(cv)) == SVt_PVCV) {
5013 CROAK(("code %s did not evaluate to a subroutine reference\n", SvPV(sub, PL_na)));
5016 SvREFCNT_inc(sv); /* XXX seems to be necessary */
5028 * old_retrieve_array
5030 * Retrieve a whole array in pre-0.6 binary format.
5032 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
5033 * Each item is stored as SX_ITEM <object> or SX_IT_UNDEF for "holes".
5035 * When we come here, SX_ARRAY has been read already.
5037 static SV *old_retrieve_array(stcxt_t *cxt, char *cname)
5045 TRACEME(("old_retrieve_array (#%d)", cxt->tagnum));
5048 * Read length, and allocate array, then pre-extend it.
5052 TRACEME(("size = %d", len));
5054 SEEN(av, 0); /* Will return if array not allocated nicely */
5058 return (SV *) av; /* No data follow if array is empty */
5061 * Now get each item in turn...
5064 for (i = 0; i < len; i++) {
5066 if (c == SX_IT_UNDEF) {
5067 TRACEME(("(#%d) undef item", i));
5068 continue; /* av_extend() already filled us with undef */
5071 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
5072 TRACEME(("(#%d) item", i));
5073 sv = retrieve(cxt, 0); /* Retrieve item */
5076 if (av_store(av, i, sv) == 0)
5080 TRACEME(("ok (old_retrieve_array at 0x%"UVxf")", PTR2UV(av)));
5088 * Retrieve a whole hash table in pre-0.6 binary format.
5090 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
5091 * Keys are stored as SX_KEY <length> <data>, the <data> section being omitted
5093 * Values are stored as SX_VALUE <object> or SX_VL_UNDEF for "holes".
5095 * When we come here, SX_HASH has been read already.
5097 static SV *old_retrieve_hash(stcxt_t *cxt, char *cname)
5105 static SV *sv_h_undef = (SV *) 0; /* hv_store() bug */
5107 TRACEME(("old_retrieve_hash (#%d)", cxt->tagnum));
5110 * Read length, allocate table.
5114 TRACEME(("size = %d", len));
5116 SEEN(hv, 0); /* Will return if table not allocated properly */
5118 return (SV *) hv; /* No data follow if table empty */
5119 hv_ksplit(hv, len); /* pre-extend hash to save multiple splits */
5122 * Now get each key/value pair in turn...
5125 for (i = 0; i < len; i++) {
5131 if (c == SX_VL_UNDEF) {
5132 TRACEME(("(#%d) undef value", i));
5134 * Due to a bug in hv_store(), it's not possible to pass
5135 * &PL_sv_undef to hv_store() as a value, otherwise the
5136 * associated key will not be creatable any more. -- RAM, 14/01/97
5139 sv_h_undef = newSVsv(&PL_sv_undef);
5140 sv = SvREFCNT_inc(sv_h_undef);
5141 } else if (c == SX_VALUE) {
5142 TRACEME(("(#%d) value", i));
5143 sv = retrieve(cxt, 0);
5147 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
5151 * Since we're reading into kbuf, we must ensure we're not
5152 * recursing between the read and the hv_store() where it's used.
5153 * Hence the key comes after the value.
5158 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
5159 RLEN(size); /* Get key size */
5160 KBUFCHK((STRLEN)size); /* Grow hash key read pool if needed */
5163 kbuf[size] = '\0'; /* Mark string end, just in case */
5164 TRACEME(("(#%d) key '%s'", i, kbuf));
5167 * Enter key/value pair into hash table.
5170 if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
5174 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
5180 *** Retrieval engine.
5186 * Make sure the stored data we're trying to retrieve has been produced
5187 * on an ILP compatible system with the same byteorder. It croaks out in
5188 * case an error is detected. [ILP = integer-long-pointer sizes]
5189 * Returns null if error is detected, &PL_sv_undef otherwise.
5191 * Note that there's no byte ordering info emitted when network order was
5192 * used at store time.
5194 static SV *magic_check(stcxt_t *cxt)
5196 /* The worst case for a malicious header would be old magic (which is
5197 longer), major, minor, byteorder length byte of 255, 255 bytes of
5198 garbage, sizeof int, long, pointer, NV.
5199 So the worse of that we can read is 255 bytes of garbage plus 4.
5200 Err, I am assuming 8 bit bytes here. Please file a bug report if you're
5201 compiling perl on a system with chars that are larger than 8 bits.
5202 (Even Crays aren't *that* perverse).
5204 unsigned char buf[4 + 255];
5205 unsigned char *current;
5208 int use_network_order;
5211 int version_minor = 0;
5213 TRACEME(("magic_check"));
5216 * The "magic number" is only for files, not when freezing in memory.
5220 /* This includes the '\0' at the end. I want to read the extra byte,
5221 which is usually going to be the major version number. */
5222 STRLEN len = sizeof(magicstr);
5225 READ(buf, (SSize_t)(len)); /* Not null-terminated */
5227 /* Point at the byte after the byte we read. */
5228 current = buf + --len; /* Do the -- outside of macros. */
5230 if (memNE(buf, magicstr, len)) {
5232 * Try to read more bytes to check for the old magic number, which
5236 TRACEME(("trying for old magic number"));
5238 old_len = sizeof(old_magicstr) - 1;
5239 READ(current + 1, (SSize_t)(old_len - len));
5241 if (memNE(buf, old_magicstr, old_len))
5242 CROAK(("File is not a perl storable"));
5243 current = buf + old_len;
5245 use_network_order = *current;
5247 GETMARK(use_network_order);
5250 * Starting with 0.6, the "use_network_order" byte flag is also used to
5251 * indicate the version number of the binary, and therefore governs the
5252 * setting of sv_retrieve_vtbl. See magic_write().
5255 version_major = use_network_order >> 1;
5256 cxt->retrieve_vtbl = version_major ? sv_retrieve : sv_old_retrieve;
5258 TRACEME(("magic_check: netorder = 0x%x", use_network_order));
5262 * Starting with 0.7 (binary major 2), a full byte is dedicated to the
5263 * minor version of the protocol. See magic_write().
5266 if (version_major > 1)
5267 GETMARK(version_minor);
5269 cxt->ver_major = version_major;
5270 cxt->ver_minor = version_minor;
5272 TRACEME(("binary image version is %d.%d", version_major, version_minor));
5275 * Inter-operability sanity check: we can't retrieve something stored
5276 * using a format more recent than ours, because we have no way to
5277 * know what has changed, and letting retrieval go would mean a probable
5278 * failure reporting a "corrupted" storable file.
5282 version_major > STORABLE_BIN_MAJOR ||
5283 (version_major == STORABLE_BIN_MAJOR &&
5284 version_minor > STORABLE_BIN_MINOR)
5287 TRACEME(("but I am version is %d.%d", STORABLE_BIN_MAJOR,
5288 STORABLE_BIN_MINOR));
5290 if (version_major == STORABLE_BIN_MAJOR) {
5291 TRACEME(("cxt->accept_future_minor is %d",
5292 cxt->accept_future_minor));
5293 if (cxt->accept_future_minor < 0)
5294 cxt->accept_future_minor
5295 = (SvTRUE(perl_get_sv("Storable::accept_future_minor",
5298 if (cxt->accept_future_minor == 1)
5299 croak_now = 0; /* Don't croak yet. */
5302 CROAK(("Storable binary image v%d.%d more recent than I am (v%d.%d)",
5303 version_major, version_minor,
5304 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
5309 * If they stored using network order, there's no byte ordering
5310 * information to check.
5313 if ((cxt->netorder = (use_network_order & 0x1))) /* Extra () for -Wall */
5314 return &PL_sv_undef; /* No byte ordering info */
5316 /* In C truth is 1, falsehood is 0. Very convienient. */
5317 use_NV_size = version_major >= 2 && version_minor >= 2;
5320 length = c + 3 + use_NV_size;
5321 READ(buf, length); /* Not null-terminated */
5323 TRACEME(("byte order '%.*s' %d", c, buf, c));
5325 #ifdef USE_56_INTERWORK_KLUDGE
5326 /* No point in caching this in the context as we only need it once per
5327 retrieve, and we need to recheck it each read. */
5328 if (SvTRUE(perl_get_sv("Storable::interwork_56_64bit", TRUE))) {
5329 if ((c != (sizeof (byteorderstr_56) - 1))
5330 || memNE(buf, byteorderstr_56, c))
5331 CROAK(("Byte order is not compatible"));
5335 if ((c != (sizeof (byteorderstr) - 1)) || memNE(buf, byteorderstr, c))
5336 CROAK(("Byte order is not compatible"));
5342 if ((int) *current++ != sizeof(int))
5343 CROAK(("Integer size is not compatible"));
5346 if ((int) *current++ != sizeof(long))
5347 CROAK(("Long integer size is not compatible"));
5349 /* sizeof(char *) */
5350 if ((int) *current != sizeof(char *))
5351 CROAK(("Pointer integer size is not compatible"));
5355 if ((int) *++current != sizeof(NV))
5356 CROAK(("Double size is not compatible"));
5359 return &PL_sv_undef; /* OK */
5365 * Recursively retrieve objects from the specified file and return their
5366 * root SV (which may be an AV or an HV for what we care).
5367 * Returns null if there is a problem.
5369 static SV *retrieve(stcxt_t *cxt, char *cname)
5375 TRACEME(("retrieve"));
5378 * Grab address tag which identifies the object if we are retrieving
5379 * an older format. Since the new binary format counts objects and no
5380 * longer explicitely tags them, we must keep track of the correspondance
5383 * The following section will disappear one day when the old format is
5384 * no longer supported, hence the final "goto" in the "if" block.
5387 if (cxt->hseen) { /* Retrieving old binary */
5389 if (cxt->netorder) {
5391 READ(&nettag, sizeof(I32)); /* Ordered sequence of I32 */
5392 tag = (stag_t) nettag;
5394 READ(&tag, sizeof(stag_t)); /* Original address of the SV */
5397 if (type == SX_OBJECT) {
5399 svh = hv_fetch(cxt->hseen, (char *) &tag, sizeof(tag), FALSE);
5401 CROAK(("Old tag 0x%"UVxf" should have been mapped already",
5403 tagn = SvIV(*svh); /* Mapped tag number computed earlier below */
5406 * The following code is common with the SX_OBJECT case below.
5409 svh = av_fetch(cxt->aseen, tagn, FALSE);
5411 CROAK(("Object #%"IVdf" should have been retrieved already",
5414 TRACEME(("has retrieved #%d at 0x%"UVxf, tagn, PTR2UV(sv)));
5415 SvREFCNT_inc(sv); /* One more reference to this same sv */
5416 return sv; /* The SV pointer where object was retrieved */
5420 * Map new object, but don't increase tagnum. This will be done
5421 * by each of the retrieve_* functions when they call SEEN().
5423 * The mapping associates the "tag" initially present with a unique
5424 * tag number. See test for SX_OBJECT above to see how this is perused.
5427 if (!hv_store(cxt->hseen, (char *) &tag, sizeof(tag),
5428 newSViv(cxt->tagnum), 0))
5435 * Regular post-0.6 binary format.
5440 TRACEME(("retrieve type = %d", type));
5443 * Are we dealing with an object we should have already retrieved?
5446 if (type == SX_OBJECT) {
5450 svh = av_fetch(cxt->aseen, tag, FALSE);
5452 CROAK(("Object #%"IVdf" should have been retrieved already",
5455 TRACEME(("had retrieved #%d at 0x%"UVxf, tag, PTR2UV(sv)));
5456 SvREFCNT_inc(sv); /* One more reference to this same sv */
5457 return sv; /* The SV pointer where object was retrieved */
5458 } else if (type >= SX_ERROR && cxt->ver_minor > STORABLE_BIN_MINOR) {
5459 if (cxt->accept_future_minor < 0)
5460 cxt->accept_future_minor
5461 = (SvTRUE(perl_get_sv("Storable::accept_future_minor",
5464 if (cxt->accept_future_minor == 1) {
5465 CROAK(("Storable binary image v%d.%d contains data of type %d. "
5466 "This Storable is v%d.%d and can only handle data types up to %d",
5467 cxt->ver_major, cxt->ver_minor, type,
5468 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR, SX_ERROR - 1));
5472 first_time: /* Will disappear when support for old format is dropped */
5475 * Okay, first time through for this one.
5478 sv = RETRIEVE(cxt, type)(cxt, cname);
5480 return (SV *) 0; /* Failed */
5483 * Old binary formats (pre-0.7).
5485 * Final notifications, ended by SX_STORED may now follow.
5486 * Currently, the only pertinent notification to apply on the
5487 * freshly retrieved object is either:
5488 * SX_CLASS <char-len> <classname> for short classnames.
5489 * SX_LG_CLASS <int-len> <classname> for larger one (rare!).
5490 * Class name is then read into the key buffer pool used by
5491 * hash table key retrieval.
5494 if (cxt->ver_major < 2) {
5495 while ((type = GETCHAR()) != SX_STORED) {
5499 GETMARK(len); /* Length coded on a single char */
5501 case SX_LG_CLASS: /* Length coded on a regular integer */
5506 return (SV *) 0; /* Failed */
5508 KBUFCHK((STRLEN)len); /* Grow buffer as necessary */
5511 kbuf[len] = '\0'; /* Mark string end */
5516 TRACEME(("ok (retrieved 0x%"UVxf", refcnt=%d, %s)", PTR2UV(sv),
5517 SvREFCNT(sv) - 1, sv_reftype(sv, FALSE)));
5525 * Retrieve data held in file and return the root object.
5526 * Common routine for pretrieve and mretrieve.
5528 static SV *do_retrieve(
5535 int is_tainted; /* Is input source tainted? */
5536 int pre_06_fmt = 0; /* True with pre Storable 0.6 formats */
5538 TRACEME(("do_retrieve (optype = 0x%x)", optype));
5540 optype |= ST_RETRIEVE;
5543 * Sanity assertions for retrieve dispatch tables.
5546 ASSERT(sizeof(sv_old_retrieve) == sizeof(sv_retrieve),
5547 ("old and new retrieve dispatch table have same size"));
5548 ASSERT(sv_old_retrieve[SX_ERROR] == retrieve_other,
5549 ("SX_ERROR entry correctly initialized in old dispatch table"));
5550 ASSERT(sv_retrieve[SX_ERROR] == retrieve_other,
5551 ("SX_ERROR entry correctly initialized in new dispatch table"));
5554 * Workaround for CROAK leak: if they enter with a "dirty" context,
5555 * free up memory for them now.
5562 * Now that STORABLE_xxx hooks exist, it is possible that they try to
5563 * re-enter retrieve() via the hooks.
5567 cxt = allocate_context(cxt);
5571 ASSERT(cxt->entry == 1, ("starting new recursion"));
5572 ASSERT(!cxt->s_dirty, ("clean context"));
5577 * Data is loaded into the memory buffer when f is NULL, unless `in' is
5578 * also NULL, in which case we're expecting the data to already lie
5579 * in the buffer (dclone case).
5582 KBUFINIT(); /* Allocate hash key reading pool once */
5585 MBUF_SAVE_AND_LOAD(in);
5588 * Magic number verifications.
5590 * This needs to be done before calling init_retrieve_context()
5591 * since the format indication in the file are necessary to conduct
5592 * some of the initializations.
5595 cxt->fio = f; /* Where I/O are performed */
5597 if (!magic_check(cxt))
5598 CROAK(("Magic number checking on storable %s failed",
5599 cxt->fio ? "file" : "string"));
5601 TRACEME(("data stored in %s format",
5602 cxt->netorder ? "net order" : "native"));
5605 * Check whether input source is tainted, so that we don't wrongly
5606 * taint perfectly good values...
5608 * We assume file input is always tainted. If both `f' and `in' are
5609 * NULL, then we come from dclone, and tainted is already filled in
5610 * the context. That's a kludge, but the whole dclone() thing is
5611 * already quite a kludge anyway! -- RAM, 15/09/2000.
5614 is_tainted = f ? 1 : (in ? SvTAINTED(in) : cxt->s_tainted);
5615 TRACEME(("input source is %s", is_tainted ? "tainted" : "trusted"));
5616 init_retrieve_context(cxt, optype, is_tainted);
5618 ASSERT(is_retrieving(), ("within retrieve operation"));
5620 sv = retrieve(cxt, 0); /* Recursively retrieve object, get root SV */
5629 pre_06_fmt = cxt->hseen != NULL; /* Before we clean context */
5632 * The "root" context is never freed.
5635 clean_retrieve_context(cxt);
5636 if (cxt->prev) /* This context was stacked */
5637 free_context(cxt); /* It was not the "root" context */
5640 * Prepare returned value.
5644 TRACEME(("retrieve ERROR"));
5645 return &PL_sv_undef; /* Something went wrong, return undef */
5648 TRACEME(("retrieve got %s(0x%"UVxf")",
5649 sv_reftype(sv, FALSE), PTR2UV(sv)));
5652 * Backward compatibility with Storable-0.5@9 (which we know we
5653 * are retrieving if hseen is non-null): don't create an extra RV
5654 * for objects since we special-cased it at store time.
5656 * Build a reference to the SV returned by pretrieve even if it is
5657 * already one and not a scalar, for consistency reasons.
5660 if (pre_06_fmt) { /* Was not handling overloading by then */
5662 TRACEME(("fixing for old formats -- pre 0.6"));
5663 if (sv_type(sv) == svis_REF && (rv = SvRV(sv)) && SvOBJECT(rv)) {
5664 TRACEME(("ended do_retrieve() with an object -- pre 0.6"));
5670 * If reference is overloaded, restore behaviour.
5672 * NB: minor glitch here: normally, overloaded refs are stored specially
5673 * so that we can croak when behaviour cannot be re-installed, and also
5674 * avoid testing for overloading magic at each reference retrieval.
5676 * Unfortunately, the root reference is implicitely stored, so we must
5677 * check for possible overloading now. Furthermore, if we don't restore
5678 * overloading, we cannot croak as if the original ref was, because we
5679 * have no way to determine whether it was an overloaded ref or not in
5682 * It's a pity that overloading magic is attached to the rv, and not to
5683 * the underlying sv as blessing is.
5687 HV *stash = (HV *) SvSTASH(sv);
5688 SV *rv = newRV_noinc(sv);
5689 if (stash && Gv_AMG(stash)) {
5691 TRACEME(("restored overloading on root reference"));
5693 TRACEME(("ended do_retrieve() with an object"));
5697 TRACEME(("regular do_retrieve() end"));
5699 return newRV_noinc(sv);
5705 * Retrieve data held in file and return the root object, undef on error.
5707 SV *pretrieve(PerlIO *f)
5709 TRACEME(("pretrieve"));
5710 return do_retrieve(f, Nullsv, 0);
5716 * Retrieve data held in scalar and return the root object, undef on error.
5718 SV *mretrieve(SV *sv)
5720 TRACEME(("mretrieve"));
5721 return do_retrieve((PerlIO*) 0, sv, 0);
5731 * Deep clone: returns a fresh copy of the original referenced SV tree.
5733 * This is achieved by storing the object in memory and restoring from
5734 * there. Not that efficient, but it should be faster than doing it from
5741 stcxt_t *real_context;
5744 TRACEME(("dclone"));
5747 * Workaround for CROAK leak: if they enter with a "dirty" context,
5748 * free up memory for them now.
5755 * do_store() optimizes for dclone by not freeing its context, should
5756 * we need to allocate one because we're deep cloning from a hook.
5759 if (!do_store((PerlIO*) 0, sv, ST_CLONE, FALSE, (SV**) 0))
5760 return &PL_sv_undef; /* Error during store */
5763 * Because of the above optimization, we have to refresh the context,
5764 * since a new one could have been allocated and stacked by do_store().
5767 { dSTCXT; real_context = cxt; } /* Sub-block needed for macro */
5768 cxt = real_context; /* And we need this temporary... */
5771 * Now, `cxt' may refer to a new context.
5774 ASSERT(!cxt->s_dirty, ("clean context"));
5775 ASSERT(!cxt->entry, ("entry will not cause new context allocation"));
5778 TRACEME(("dclone stored %d bytes", size));
5782 * Since we're passing do_retrieve() both a NULL file and sv, we need
5783 * to pre-compute the taintedness of the input by setting cxt->tainted
5784 * to whatever state our own input string was. -- RAM, 15/09/2000
5786 * do_retrieve() will free non-root context.
5789 cxt->s_tainted = SvTAINTED(sv);
5790 out = do_retrieve((PerlIO*) 0, Nullsv, ST_CLONE);
5792 TRACEME(("dclone returns 0x%"UVxf, PTR2UV(out)));
5802 * The Perl IO GV object distinguishes between input and output for sockets
5803 * but not for plain files. To allow Storable to transparently work on
5804 * plain files and sockets transparently, we have to ask xsubpp to fetch the
5805 * right object for us. Hence the OutputStream and InputStream declarations.
5807 * Before perl 5.004_05, those entries in the standard typemap are not
5808 * defined in perl include files, so we do that here.
5811 #ifndef OutputStream
5812 #define OutputStream PerlIO *
5813 #define InputStream PerlIO *
5814 #endif /* !OutputStream */
5816 MODULE = Storable PACKAGE = Storable::Cxt
5822 stcxt_t *cxt = (stcxt_t *)SvPVX(SvRV(self));
5826 if (!cxt->membuf_ro && mbase)
5828 if (cxt->membuf_ro && (cxt->msaved).arena)
5829 Safefree((cxt->msaved).arena);
5832 MODULE = Storable PACKAGE = Storable
5838 gv_fetchpv("Storable::drop_utf8", GV_ADDMULTI, SVt_PV);
5840 /* Only disable the used only once warning if we are in debugging mode. */
5841 gv_fetchpv("Storable::DEBUGME", GV_ADDMULTI, SVt_PV);
5843 #ifdef USE_56_INTERWORK_KLUDGE
5844 gv_fetchpv("Storable::interwork_56_64bit", GV_ADDMULTI, SVt_PV);
5878 last_op_in_netorder()