2 * Store and retrieve mechanism.
6 * $Id: Storable.xs,v 1.0.1.10 2001/08/28 21:52:14 ram Exp $
8 * Copyright (c) 1995-2000, Raphael Manfredi
10 * You may redistribute only under the same terms as Perl 5, as specified
11 * in the README file that comes with the distribution.
13 * $Log: Storable.xs,v $
14 * Revision 1.0.1.10 2001/08/28 21:52:14 ram
15 * patch13: removed spurious debugging messages
17 * Revision 1.0.1.9 2001/07/01 11:25:02 ram
18 * patch12: fixed memory corruption on croaks during thaw()
19 * patch12: made code compile cleanly with -Wall (Jarkko Hietaniemi)
20 * patch12: changed tagnum and classnum from I32 to IV in context
22 * Revision 1.0.1.8 2001/03/15 00:20:55 ram
23 * patch11: last version was wrongly compiling with assertions on
25 * Revision 1.0.1.7 2001/02/17 12:25:26 ram
26 * patch8: now bless objects ASAP at retrieve time
27 * patch8: added support for blessed ref to tied structures
29 * Revision 1.0.1.6 2001/01/03 09:40:40 ram
30 * patch7: prototype and casting cleanup
31 * patch7: trace offending package when overloading cannot be restored
32 * patch7: made context cleanup safer to avoid dup freeing
34 * Revision 1.0.1.5 2000/11/05 17:21:24 ram
35 * patch6: fixed severe "object lost" bug for STORABLE_freeze returns
37 * Revision 1.0.1.4 2000/10/26 17:11:04 ram
38 * patch5: auto requires module of blessed ref when STORABLE_thaw misses
40 * Revision 1.0.1.3 2000/09/29 19:49:57 ram
41 * patch3: avoid using "tainted" and "dirty" since Perl remaps them via cpp
43 * Revision 1.0.1.2 2000/09/28 21:43:10 ram
44 * patch2: perls before 5.004_04 lack newSVpvn
46 * Revision 1.0.1.1 2000/09/17 16:47:49 ram
47 * patch1: now only taint retrieved data when source was tainted
48 * patch1: added support for UTF-8 strings
49 * patch1: fixed store hook bug: was allocating class id too soon
51 * Revision 1.0 2000/09/01 19:40:41 ram
52 * Baseline for first official release.
58 #include <patchlevel.h> /* Perl's one, needed since 5.6 */
63 #define DEBUGME /* Debug mode, turns assertions on as well */
64 #define DASSERT /* Assertion mode */
67 #if 0 /* On NetWare USE_PERLIO is not used */
68 #define DEBUGME /* Debug mode, turns assertions on as well */
69 #define DASSERT /* Assertion mode */
74 * Pre PerlIO time when none of USE_PERLIO and PERLIO_IS_STDIO is defined
75 * Provide them with the necessary defines so they can build with pre-5.004.
78 #ifndef PERLIO_IS_STDIO
80 #define PerlIO_getc(x) getc(x)
81 #define PerlIO_putc(f,x) putc(x,f)
82 #define PerlIO_read(x,y,z) fread(y,1,z,x)
83 #define PerlIO_write(x,y,z) fwrite(y,1,z,x)
84 #define PerlIO_stdoutf printf
85 #endif /* PERLIO_IS_STDIO */
86 #endif /* USE_PERLIO */
89 * Earlier versions of perl might be used, we can't assume they have the latest!
92 #ifndef PERL_VERSION /* For perls < 5.6 */
93 #define PERL_VERSION PATCHLEVEL
95 #define newRV_noinc(sv) ((Sv = newRV(sv)), --SvREFCNT(SvRV(Sv)), Sv)
97 #if (PATCHLEVEL <= 4) /* Older perls (<= 5.004) lack PL_ namespace */
98 #define PL_sv_yes sv_yes
99 #define PL_sv_no sv_no
100 #define PL_sv_undef sv_undef
101 #if (SUBVERSION <= 4) /* 5.004_04 has been reported to lack newSVpvn */
102 #define newSVpvn newSVpv
104 #endif /* PATCHLEVEL <= 4 */
105 #ifndef HvSHAREKEYS_off
106 #define HvSHAREKEYS_off(hv) /* Ignore */
108 #ifndef AvFILLp /* Older perls (<=5.003) lack AvFILLp */
109 #define AvFILLp AvFILL
111 typedef double NV; /* Older perls lack the NV type */
112 #define IVdf "ld" /* Various printf formats for Perl types */
116 #define INT2PTR(t,v) (t)(IV)(v)
117 #define PTR2UV(v) (unsigned long)(v)
118 #endif /* PERL_VERSION -- perls < 5.6 */
120 #ifndef NVef /* The following were not part of perl 5.6 */
121 #if defined(USE_LONG_DOUBLE) && \
122 defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
123 #define NVef PERL_PRIeldbl
124 #define NVff PERL_PRIfldbl
125 #define NVgf PERL_PRIgldbl
140 * TRACEME() will only output things when the $Storable::DEBUGME is true.
143 #define TRACEME(x) do { \
144 if (SvTRUE(perl_get_sv("Storable::DEBUGME", TRUE))) \
145 { PerlIO_stdoutf x; PerlIO_stdoutf("\n"); } \
152 #define ASSERT(x,y) do { \
154 PerlIO_stdoutf("ASSERT FAILED (\"%s\", line %d): ", \
155 __FILE__, __LINE__); \
156 PerlIO_stdoutf y; PerlIO_stdoutf("\n"); \
167 #define C(x) ((char) (x)) /* For markers with dynamic retrieval handling */
169 #define SX_OBJECT C(0) /* Already stored object */
170 #define SX_LSCALAR C(1) /* Scalar (large binary) follows (length, data) */
171 #define SX_ARRAY C(2) /* Array forthcominng (size, item list) */
172 #define SX_HASH C(3) /* Hash forthcoming (size, key/value pair list) */
173 #define SX_REF C(4) /* Reference to object forthcoming */
174 #define SX_UNDEF C(5) /* Undefined scalar */
175 #define SX_INTEGER C(6) /* Integer forthcoming */
176 #define SX_DOUBLE C(7) /* Double forthcoming */
177 #define SX_BYTE C(8) /* (signed) byte forthcoming */
178 #define SX_NETINT C(9) /* Integer in network order forthcoming */
179 #define SX_SCALAR C(10) /* Scalar (binary, small) follows (length, data) */
180 #define SX_TIED_ARRAY C(11) /* Tied array forthcoming */
181 #define SX_TIED_HASH C(12) /* Tied hash forthcoming */
182 #define SX_TIED_SCALAR C(13) /* Tied scalar forthcoming */
183 #define SX_SV_UNDEF C(14) /* Perl's immortal PL_sv_undef */
184 #define SX_SV_YES C(15) /* Perl's immortal PL_sv_yes */
185 #define SX_SV_NO C(16) /* Perl's immortal PL_sv_no */
186 #define SX_BLESS C(17) /* Object is blessed */
187 #define SX_IX_BLESS C(18) /* Object is blessed, classname given by index */
188 #define SX_HOOK C(19) /* Stored via hook, user-defined */
189 #define SX_OVERLOAD C(20) /* Overloaded reference */
190 #define SX_TIED_KEY C(21) /* Tied magic key forthcoming */
191 #define SX_TIED_IDX C(22) /* Tied magic index forthcoming */
192 #define SX_UTF8STR C(23) /* UTF-8 string forthcoming (small) */
193 #define SX_LUTF8STR C(24) /* UTF-8 string forthcoming (large) */
194 #define SX_FLAG_HASH C(25) /* Hash with flags forthcoming (size, flags, key/flags/value triplet list) */
195 #define SX_ERROR C(26) /* Error */
198 * Those are only used to retrieve "old" pre-0.6 binary images.
200 #define SX_ITEM 'i' /* An array item introducer */
201 #define SX_IT_UNDEF 'I' /* Undefined array item */
202 #define SX_KEY 'k' /* A hash key introducer */
203 #define SX_VALUE 'v' /* A hash value introducer */
204 #define SX_VL_UNDEF 'V' /* Undefined hash value */
207 * Those are only used to retrieve "old" pre-0.7 binary images
210 #define SX_CLASS 'b' /* Object is blessed, class name length <255 */
211 #define SX_LG_CLASS 'B' /* Object is blessed, class name length >255 */
212 #define SX_STORED 'X' /* End of object */
215 * Limits between short/long length representation.
218 #define LG_SCALAR 255 /* Large scalar length limit */
219 #define LG_BLESS 127 /* Large classname bless limit */
225 #define ST_STORE 0x1 /* Store operation */
226 #define ST_RETRIEVE 0x2 /* Retrieval operation */
227 #define ST_CLONE 0x4 /* Deep cloning operation */
230 * The following structure is used for hash table key retrieval. Since, when
231 * retrieving objects, we'll be facing blessed hash references, it's best
232 * to pre-allocate that buffer once and resize it as the need arises, never
233 * freeing it (keys will be saved away someplace else anyway, so even large
234 * keys are not enough a motivation to reclaim that space).
236 * This structure is also used for memory store/retrieve operations which
237 * happen in a fixed place before being malloc'ed elsewhere if persistency
238 * is required. Hence the aptr pointer.
241 char *arena; /* Will hold hash key strings, resized as needed */
242 STRLEN asiz; /* Size of aforementionned buffer */
243 char *aptr; /* Arena pointer, for in-place read/write ops */
244 char *aend; /* First invalid address */
249 * A hash table records the objects which have already been stored.
250 * Those are referred to as SX_OBJECT in the file, and their "tag" (i.e.
251 * an arbitrary sequence number) is used to identify them.
254 * An array table records the objects which have already been retrieved,
255 * as seen by the tag determind by counting the objects themselves. The
256 * reference to that retrieved object is kept in the table, and is returned
257 * when an SX_OBJECT is found bearing that same tag.
259 * The same processing is used to record "classname" for blessed objects:
260 * indexing by a hash at store time, and via an array at retrieve time.
263 typedef unsigned long stag_t; /* Used by pre-0.6 binary format */
266 * The following "thread-safe" related defines were contributed by
267 * Murray Nesbitt <murray@activestate.com> and integrated by RAM, who
268 * only renamed things a little bit to ensure consistency with surrounding
269 * code. -- RAM, 14/09/1999
271 * The original patch suffered from the fact that the stcxt_t structure
272 * was global. Murray tried to minimize the impact on the code as much as
275 * Starting with 0.7, Storable can be re-entrant, via the STORABLE_xxx hooks
276 * on objects. Therefore, the notion of context needs to be generalized,
280 #define MY_VERSION "Storable(" XS_VERSION ")"
284 * Conditional UTF8 support.
288 #define STORE_UTF8STR(pv, len) STORE_PV_LEN(pv, len, SX_UTF8STR, SX_LUTF8STR)
289 #define HAS_UTF8_SCALARS
291 #define HAS_UTF8_HASHES
294 /* 5.6 perl has utf8 scalars but not hashes */
298 #define STORE_UTF8STR(pv, len) CROAK(("panic: storing UTF8 in non-UTF8 perl"))
301 #define UTF8_CROAK() CROAK(("Cannot retrieve UTF8 data in non-UTF8 perl"))
304 #ifdef HvPLACEHOLDERS
305 #define HAS_RESTRICTED_HASHES
307 #define HVhek_PLACEHOLD 0x200
308 #define RESTRICTED_HASH_CROAK() CROAK(("Cannot retrieve restricted hash"))
312 #define HAS_HASH_KEY_FLAGS
316 * Fields s_tainted and s_dirty are prefixed with s_ because Perl's include
317 * files remap tainted and dirty when threading is enabled. That's bad for
318 * perl to remap such common words. -- RAM, 29/09/00
321 typedef struct stcxt {
322 int entry; /* flags recursion */
323 int optype; /* type of traversal operation */
324 HV *hseen; /* which objects have been seen, store time */
325 AV *hook_seen; /* which SVs were returned by STORABLE_freeze() */
326 AV *aseen; /* which objects have been seen, retrieve time */
327 HV *hclass; /* which classnames have been seen, store time */
328 AV *aclass; /* which classnames have been seen, retrieve time */
329 HV *hook; /* cache for hook methods per class name */
330 IV tagnum; /* incremented at store time for each seen object */
331 IV classnum; /* incremented at store time for each seen classname */
332 int netorder; /* true if network order used */
333 int s_tainted; /* true if input source is tainted, at retrieve time */
334 int forgive_me; /* whether to be forgiving... */
335 int canonical; /* whether to store hashes sorted by key */
336 #ifndef HAS_RESTRICTED_HASHES
337 int derestrict; /* whether to downgrade restrcted hashes */
340 int use_bytes; /* whether to bytes-ify utf8 */
342 int s_dirty; /* context is dirty due to CROAK() -- can be cleaned */
343 int membuf_ro; /* true means membuf is read-only and msaved is rw */
344 struct extendable keybuf; /* for hash key retrieval */
345 struct extendable membuf; /* for memory store/retrieve operations */
346 struct extendable msaved; /* where potentially valid mbuf is saved */
347 PerlIO *fio; /* where I/O are performed, NULL for memory */
348 int ver_major; /* major of version for retrieved object */
349 int ver_minor; /* minor of version for retrieved object */
350 SV *(**retrieve_vtbl)(); /* retrieve dispatch table */
351 struct stcxt *prev; /* contexts chained backwards in real recursion */
354 #if defined(MULTIPLICITY) || defined(PERL_OBJECT) || defined(PERL_CAPI)
356 #if (PATCHLEVEL <= 4) && (SUBVERSION < 68)
358 SV *perinterp_sv = perl_get_sv(MY_VERSION, FALSE)
359 #else /* >= perl5.004_68 */
361 SV *perinterp_sv = *hv_fetch(PL_modglobal, \
362 MY_VERSION, sizeof(MY_VERSION)-1, TRUE)
363 #endif /* < perl5.004_68 */
365 #define dSTCXT_PTR(T,name) \
366 T name = ((perinterp_sv && SvIOK(perinterp_sv) \
367 ? INT2PTR(T, SvIVX(perinterp_sv)) : (T) 0))
370 dSTCXT_PTR(stcxt_t *, cxt)
374 Newz(0, cxt, 1, stcxt_t); \
375 sv_setiv(perinterp_sv, PTR2IV(cxt))
377 #define SET_STCXT(x) do { \
379 sv_setiv(perinterp_sv, PTR2IV(x)); \
382 #else /* !MULTIPLICITY && !PERL_OBJECT && !PERL_CAPI */
384 static stcxt_t Context;
385 static stcxt_t *Context_ptr = &Context;
386 #define dSTCXT stcxt_t *cxt = Context_ptr
387 #define INIT_STCXT dSTCXT
388 #define SET_STCXT(x) Context_ptr = x
390 #endif /* MULTIPLICITY || PERL_OBJECT || PERL_CAPI */
394 * Croaking implies a memory leak, since we don't use setjmp/longjmp
395 * to catch the exit and free memory used during store or retrieve
396 * operations. This is not too difficult to fix, but I need to understand
397 * how Perl does it, and croaking is exceptional anyway, so I lack the
398 * motivation to do it.
400 * The current workaround is to mark the context as dirty when croaking,
401 * so that data structures can be freed whenever we renter Storable code
402 * (but only *then*: it's a workaround, not a fix).
404 * This is also imperfect, because we don't really know how far they trapped
405 * the croak(), and when we were recursing, we won't be able to clean anything
406 * but the topmost context stacked.
409 #define CROAK(x) do { cxt->s_dirty = 1; croak x; } while (0)
412 * End of "thread-safe" related definitions.
418 * Keep only the low 32 bits of a pointer (used for tags, which are not
423 #define LOW_32BITS(x) ((I32) (x))
425 #define LOW_32BITS(x) ((I32) ((unsigned long) (x) & 0xffffffffUL))
431 * Hack for Crays, where sizeof(I32) == 8, and which are big-endians.
432 * Used in the WLEN and RLEN macros.
436 #define oI(x) ((I32 *) ((char *) (x) + 4))
437 #define oS(x) ((x) - 4)
438 #define oC(x) (x = 0)
447 * key buffer handling
449 #define kbuf (cxt->keybuf).arena
450 #define ksiz (cxt->keybuf).asiz
451 #define KBUFINIT() do { \
453 TRACEME(("** allocating kbuf of 128 bytes")); \
454 New(10003, kbuf, 128, char); \
458 #define KBUFCHK(x) do { \
460 TRACEME(("** extending kbuf to %d bytes (had %d)", x+1, ksiz)); \
461 Renew(kbuf, x+1, char); \
467 * memory buffer handling
469 #define mbase (cxt->membuf).arena
470 #define msiz (cxt->membuf).asiz
471 #define mptr (cxt->membuf).aptr
472 #define mend (cxt->membuf).aend
474 #define MGROW (1 << 13)
475 #define MMASK (MGROW - 1)
477 #define round_mgrow(x) \
478 ((unsigned long) (((unsigned long) (x) + MMASK) & ~MMASK))
479 #define trunc_int(x) \
480 ((unsigned long) ((unsigned long) (x) & ~(sizeof(int)-1)))
481 #define int_aligned(x) \
482 ((unsigned long) (x) == trunc_int(x))
484 #define MBUF_INIT(x) do { \
486 TRACEME(("** allocating mbase of %d bytes", MGROW)); \
487 New(10003, mbase, MGROW, char); \
494 mend = mbase + msiz; \
497 #define MBUF_TRUNC(x) mptr = mbase + x
498 #define MBUF_SIZE() (mptr - mbase)
504 * Those macros are used in do_retrieve() to save the current memory
505 * buffer into cxt->msaved, before MBUF_LOAD() can be used to retrieve
506 * data from a string.
508 #define MBUF_SAVE_AND_LOAD(in) do { \
509 ASSERT(!cxt->membuf_ro, ("mbase not already saved")); \
510 cxt->membuf_ro = 1; \
511 TRACEME(("saving mbuf")); \
512 StructCopy(&cxt->membuf, &cxt->msaved, struct extendable); \
516 #define MBUF_RESTORE() do { \
517 ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
518 cxt->membuf_ro = 0; \
519 TRACEME(("restoring mbuf")); \
520 StructCopy(&cxt->msaved, &cxt->membuf, struct extendable); \
524 * Use SvPOKp(), because SvPOK() fails on tainted scalars.
525 * See store_scalar() for other usage of this workaround.
527 #define MBUF_LOAD(v) do { \
528 ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
530 CROAK(("Not a scalar string")); \
531 mptr = mbase = SvPV(v, msiz); \
532 mend = mbase + msiz; \
535 #define MBUF_XTEND(x) do { \
536 int nsz = (int) round_mgrow((x)+msiz); \
537 int offset = mptr - mbase; \
538 ASSERT(!cxt->membuf_ro, ("mbase is not read-only")); \
539 TRACEME(("** extending mbase from %d to %d bytes (wants %d new)", \
541 Renew(mbase, nsz, char); \
543 mptr = mbase + offset; \
544 mend = mbase + nsz; \
547 #define MBUF_CHK(x) do { \
548 if ((mptr + (x)) > mend) \
552 #define MBUF_GETC(x) do { \
554 x = (int) (unsigned char) *mptr++; \
560 #define MBUF_GETINT(x) do { \
562 if ((mptr + 4) <= mend) { \
563 memcpy(oI(&x), mptr, 4); \
569 #define MBUF_GETINT(x) do { \
570 if ((mptr + sizeof(int)) <= mend) { \
571 if (int_aligned(mptr)) \
574 memcpy(&x, mptr, sizeof(int)); \
575 mptr += sizeof(int); \
581 #define MBUF_READ(x,s) do { \
582 if ((mptr + (s)) <= mend) { \
583 memcpy(x, mptr, s); \
589 #define MBUF_SAFEREAD(x,s,z) do { \
590 if ((mptr + (s)) <= mend) { \
591 memcpy(x, mptr, s); \
599 #define MBUF_PUTC(c) do { \
601 *mptr++ = (char) c; \
604 *mptr++ = (char) c; \
609 #define MBUF_PUTINT(i) do { \
611 memcpy(mptr, oI(&i), 4); \
615 #define MBUF_PUTINT(i) do { \
616 MBUF_CHK(sizeof(int)); \
617 if (int_aligned(mptr)) \
620 memcpy(mptr, &i, sizeof(int)); \
621 mptr += sizeof(int); \
625 #define MBUF_WRITE(x,s) do { \
627 memcpy(mptr, x, s); \
632 * Possible return values for sv_type().
636 #define svis_SCALAR 1
640 #define svis_TIED_ITEM 5
647 #define SHF_TYPE_MASK 0x03
648 #define SHF_LARGE_CLASSLEN 0x04
649 #define SHF_LARGE_STRLEN 0x08
650 #define SHF_LARGE_LISTLEN 0x10
651 #define SHF_IDX_CLASSNAME 0x20
652 #define SHF_NEED_RECURSE 0x40
653 #define SHF_HAS_LIST 0x80
656 * Types for SX_HOOK (last 2 bits in flags).
662 #define SHT_EXTRA 3 /* Read extra byte for type */
665 * The following are held in the "extra byte"...
668 #define SHT_TSCALAR 4 /* 4 + 0 -- tied scalar */
669 #define SHT_TARRAY 5 /* 4 + 1 -- tied array */
670 #define SHT_THASH 6 /* 4 + 2 -- tied hash */
673 * per hash flags for flagged hashes
676 #define SHV_RESTRICTED 0x01
679 * per key flags for flagged hashes
682 #define SHV_K_UTF8 0x01
683 #define SHV_K_WASUTF8 0x02
684 #define SHV_K_LOCKED 0x04
685 #define SHV_K_ISSV 0x08
686 #define SHV_K_PLACEHOLDER 0x10
689 * Before 0.6, the magic string was "perl-store" (binary version number 0).
691 * Since 0.6 introduced many binary incompatibilities, the magic string has
692 * been changed to "pst0" to allow an old image to be properly retrieved by
693 * a newer Storable, but ensure a newer image cannot be retrieved with an
696 * At 0.7, objects are given the ability to serialize themselves, and the
697 * set of markers is extended, backward compatibility is not jeopardized,
698 * so the binary version number could have remained unchanged. To correctly
699 * spot errors if a file making use of 0.7-specific extensions is given to
700 * 0.6 for retrieval, the binary version was moved to "2". And I'm introducing
701 * a "minor" version, to better track this kind of evolution from now on.
704 static char old_magicstr[] = "perl-store"; /* Magic number before 0.6 */
705 static char magicstr[] = "pst0"; /* Used as a magic number */
708 #define STORABLE_BIN_MAJOR 2 /* Binary major "version" */
709 #define STORABLE_BIN_MINOR 5 /* Binary minor "version" */
711 /* If we aren't 5.7.3 or later, we won't be writing out files that use the
712 * new flagged hash introdued in 2.5, so put 2.4 in the binary header to
713 * maximise ease of interoperation with older Storables.
714 * Could we write 2.3s if we're on 5.005_03? NWC
716 #if (PATCHLEVEL <= 6)
717 #define STORABLE_BIN_WRITE_MINOR 4
720 * As of perl 5.7.3, utf8 hash key is introduced.
721 * So this must change -- dankogai
723 #define STORABLE_BIN_WRITE_MINOR 5
724 #endif /* (PATCHLEVEL <= 6) */
727 * Useful store shortcuts...
730 #define PUTMARK(x) do { \
733 else if (PerlIO_putc(cxt->fio, x) == EOF) \
737 #define WRITE_I32(x) do { \
738 ASSERT(sizeof(x) == sizeof(I32), ("writing an I32")); \
741 else if (PerlIO_write(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
746 #define WLEN(x) do { \
747 if (cxt->netorder) { \
748 int y = (int) htonl(x); \
751 else if (PerlIO_write(cxt->fio,oI(&y),oS(sizeof(y))) != oS(sizeof(y))) \
756 else if (PerlIO_write(cxt->fio,oI(&x),oS(sizeof(x))) != oS(sizeof(x))) \
761 #define WLEN(x) WRITE_I32(x)
764 #define WRITE(x,y) do { \
767 else if (PerlIO_write(cxt->fio, x, y) != y) \
771 #define STORE_PV_LEN(pv, len, small, large) do { \
772 if (len <= LG_SCALAR) { \
773 unsigned char clen = (unsigned char) len; \
785 #define STORE_SCALAR(pv, len) STORE_PV_LEN(pv, len, SX_SCALAR, SX_LSCALAR)
788 * Store undef in arrays and hashes without recursing through store().
790 #define STORE_UNDEF() do { \
796 * Useful retrieve shortcuts...
800 (cxt->fio ? PerlIO_getc(cxt->fio) : (mptr >= mend ? EOF : (int) *mptr++))
802 #define GETMARK(x) do { \
805 else if ((int) (x = PerlIO_getc(cxt->fio)) == EOF) \
809 #define READ_I32(x) do { \
810 ASSERT(sizeof(x) == sizeof(I32), ("reading an I32")); \
814 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
819 #define RLEN(x) do { \
823 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
826 x = (int) ntohl(x); \
829 #define RLEN(x) READ_I32(x)
832 #define READ(x,y) do { \
835 else if (PerlIO_read(cxt->fio, x, y) != y) \
839 #define SAFEREAD(x,y,z) do { \
841 MBUF_SAFEREAD(x,y,z); \
842 else if (PerlIO_read(cxt->fio, x, y) != y) { \
849 * This macro is used at retrieve time, to remember where object 'y', bearing a
850 * given tag 'tagnum', has been retrieved. Next time we see an SX_OBJECT marker,
851 * we'll therefore know where it has been retrieved and will be able to
852 * share the same reference, as in the original stored memory image.
854 * We also need to bless objects ASAP for hooks (which may compute "ref $x"
855 * on the objects given to STORABLE_thaw and expect that to be defined), and
856 * also for overloaded objects (for which we might not find the stash if the
857 * object is not blessed yet--this might occur for overloaded objects that
858 * refer to themselves indirectly: if we blessed upon return from a sub
859 * retrieve(), the SX_OBJECT marker we'd found could not have overloading
860 * restored on it because the underlying object would not be blessed yet!).
862 * To achieve that, the class name of the last retrieved object is passed down
863 * recursively, and the first SEEN() call for which the class name is not NULL
864 * will bless the object.
866 #define SEEN(y,c) do { \
869 if (av_store(cxt->aseen, cxt->tagnum++, SvREFCNT_inc(y)) == 0) \
871 TRACEME(("aseen(#%d) = 0x%"UVxf" (refcnt=%d)", cxt->tagnum-1, \
872 PTR2UV(y), SvREFCNT(y)-1)); \
874 BLESS((SV *) (y), c); \
878 * Bless `s' in `p', via a temporary reference, required by sv_bless().
880 #define BLESS(s,p) do { \
883 TRACEME(("blessing 0x%"UVxf" in %s", PTR2UV(s), (p))); \
884 stash = gv_stashpv((p), TRUE); \
885 ref = newRV_noinc(s); \
886 (void) sv_bless(ref, stash); \
892 static SV *retrieve(stcxt_t *cxt, char *cname);
895 * Dynamic dispatching table for SV store.
898 static int store_ref(stcxt_t *cxt, SV *sv);
899 static int store_scalar(stcxt_t *cxt, SV *sv);
900 static int store_array(stcxt_t *cxt, AV *av);
901 static int store_hash(stcxt_t *cxt, HV *hv);
902 static int store_tied(stcxt_t *cxt, SV *sv);
903 static int store_tied_item(stcxt_t *cxt, SV *sv);
904 static int store_other(stcxt_t *cxt, SV *sv);
905 static int store_blessed(stcxt_t *cxt, SV *sv, int type, HV *pkg);
907 static int (*sv_store[])(stcxt_t *cxt, SV *sv) = {
908 store_ref, /* svis_REF */
909 store_scalar, /* svis_SCALAR */
910 (int (*)(stcxt_t *cxt, SV *sv)) store_array, /* svis_ARRAY */
911 (int (*)(stcxt_t *cxt, SV *sv)) store_hash, /* svis_HASH */
912 store_tied, /* svis_TIED */
913 store_tied_item, /* svis_TIED_ITEM */
914 store_other, /* svis_OTHER */
917 #define SV_STORE(x) (*sv_store[x])
920 * Dynamic dispatching tables for SV retrieval.
923 static SV *retrieve_lscalar(stcxt_t *cxt, char *cname);
924 static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname);
925 static SV *old_retrieve_array(stcxt_t *cxt, char *cname);
926 static SV *old_retrieve_hash(stcxt_t *cxt, char *cname);
927 static SV *retrieve_ref(stcxt_t *cxt, char *cname);
928 static SV *retrieve_undef(stcxt_t *cxt, char *cname);
929 static SV *retrieve_integer(stcxt_t *cxt, char *cname);
930 static SV *retrieve_double(stcxt_t *cxt, char *cname);
931 static SV *retrieve_byte(stcxt_t *cxt, char *cname);
932 static SV *retrieve_netint(stcxt_t *cxt, char *cname);
933 static SV *retrieve_scalar(stcxt_t *cxt, char *cname);
934 static SV *retrieve_utf8str(stcxt_t *cxt, char *cname);
935 static SV *retrieve_tied_array(stcxt_t *cxt, char *cname);
936 static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname);
937 static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname);
938 static SV *retrieve_other(stcxt_t *cxt, char *cname);
940 static SV *(*sv_old_retrieve[])(stcxt_t *cxt, char *cname) = {
941 0, /* SX_OBJECT -- entry unused dynamically */
942 retrieve_lscalar, /* SX_LSCALAR */
943 old_retrieve_array, /* SX_ARRAY -- for pre-0.6 binaries */
944 old_retrieve_hash, /* SX_HASH -- for pre-0.6 binaries */
945 retrieve_ref, /* SX_REF */
946 retrieve_undef, /* SX_UNDEF */
947 retrieve_integer, /* SX_INTEGER */
948 retrieve_double, /* SX_DOUBLE */
949 retrieve_byte, /* SX_BYTE */
950 retrieve_netint, /* SX_NETINT */
951 retrieve_scalar, /* SX_SCALAR */
952 retrieve_tied_array, /* SX_ARRAY */
953 retrieve_tied_hash, /* SX_HASH */
954 retrieve_tied_scalar, /* SX_SCALAR */
955 retrieve_other, /* SX_SV_UNDEF not supported */
956 retrieve_other, /* SX_SV_YES not supported */
957 retrieve_other, /* SX_SV_NO not supported */
958 retrieve_other, /* SX_BLESS not supported */
959 retrieve_other, /* SX_IX_BLESS not supported */
960 retrieve_other, /* SX_HOOK not supported */
961 retrieve_other, /* SX_OVERLOADED not supported */
962 retrieve_other, /* SX_TIED_KEY not supported */
963 retrieve_other, /* SX_TIED_IDX not supported */
964 retrieve_other, /* SX_UTF8STR not supported */
965 retrieve_other, /* SX_LUTF8STR not supported */
966 retrieve_other, /* SX_FLAG_HASH not supported */
967 retrieve_other, /* SX_ERROR */
970 static SV *retrieve_array(stcxt_t *cxt, char *cname);
971 static SV *retrieve_hash(stcxt_t *cxt, char *cname);
972 static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname);
973 static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname);
974 static SV *retrieve_sv_no(stcxt_t *cxt, char *cname);
975 static SV *retrieve_blessed(stcxt_t *cxt, char *cname);
976 static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname);
977 static SV *retrieve_hook(stcxt_t *cxt, char *cname);
978 static SV *retrieve_overloaded(stcxt_t *cxt, char *cname);
979 static SV *retrieve_tied_key(stcxt_t *cxt, char *cname);
980 static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname);
981 static SV *retrieve_flag_hash(stcxt_t *cxt, char *cname);
983 static SV *(*sv_retrieve[])(stcxt_t *cxt, char *cname) = {
984 0, /* SX_OBJECT -- entry unused dynamically */
985 retrieve_lscalar, /* SX_LSCALAR */
986 retrieve_array, /* SX_ARRAY */
987 retrieve_hash, /* SX_HASH */
988 retrieve_ref, /* SX_REF */
989 retrieve_undef, /* SX_UNDEF */
990 retrieve_integer, /* SX_INTEGER */
991 retrieve_double, /* SX_DOUBLE */
992 retrieve_byte, /* SX_BYTE */
993 retrieve_netint, /* SX_NETINT */
994 retrieve_scalar, /* SX_SCALAR */
995 retrieve_tied_array, /* SX_ARRAY */
996 retrieve_tied_hash, /* SX_HASH */
997 retrieve_tied_scalar, /* SX_SCALAR */
998 retrieve_sv_undef, /* SX_SV_UNDEF */
999 retrieve_sv_yes, /* SX_SV_YES */
1000 retrieve_sv_no, /* SX_SV_NO */
1001 retrieve_blessed, /* SX_BLESS */
1002 retrieve_idx_blessed, /* SX_IX_BLESS */
1003 retrieve_hook, /* SX_HOOK */
1004 retrieve_overloaded, /* SX_OVERLOAD */
1005 retrieve_tied_key, /* SX_TIED_KEY */
1006 retrieve_tied_idx, /* SX_TIED_IDX */
1007 retrieve_utf8str, /* SX_UTF8STR */
1008 retrieve_lutf8str, /* SX_LUTF8STR */
1009 retrieve_flag_hash, /* SX_HASH */
1010 retrieve_other, /* SX_ERROR */
1013 #define RETRIEVE(c,x) (*(c)->retrieve_vtbl[(x) >= SX_ERROR ? SX_ERROR : (x)])
1015 static SV *mbuf2sv(void);
1018 *** Context management.
1024 * Called once per "thread" (interpreter) to initialize some global context.
1026 static void init_perinterp(void)
1030 cxt->netorder = 0; /* true if network order used */
1031 cxt->forgive_me = -1; /* whether to be forgiving... */
1037 * Called at the end of every context cleaning, to perform common reset
1040 static void reset_context(stcxt_t *cxt)
1044 cxt->optype &= ~(ST_STORE|ST_RETRIEVE); /* Leave ST_CLONE alone */
1048 * init_store_context
1050 * Initialize a new store context for real recursion.
1052 static void init_store_context(
1058 TRACEME(("init_store_context"));
1060 cxt->netorder = network_order;
1061 cxt->forgive_me = -1; /* Fetched from perl if needed */
1062 cxt->canonical = -1; /* Idem */
1063 cxt->tagnum = -1; /* Reset tag numbers */
1064 cxt->classnum = -1; /* Reset class numbers */
1065 cxt->fio = f; /* Where I/O are performed */
1066 cxt->optype = optype; /* A store, or a deep clone */
1067 cxt->entry = 1; /* No recursion yet */
1070 * The `hseen' table is used to keep track of each SV stored and their
1071 * associated tag numbers is special. It is "abused" because the
1072 * values stored are not real SV, just integers cast to (SV *),
1073 * which explains the freeing below.
1075 * It is also one possible bottlneck to achieve good storing speed,
1076 * so the "shared keys" optimization is turned off (unlikely to be
1077 * of any use here), and the hash table is "pre-extended". Together,
1078 * those optimizations increase the throughput by 12%.
1081 cxt->hseen = newHV(); /* Table where seen objects are stored */
1082 HvSHAREKEYS_off(cxt->hseen);
1085 * The following does not work well with perl5.004_04, and causes
1086 * a core dump later on, in a completely unrelated spot, which
1087 * makes me think there is a memory corruption going on.
1089 * Calling hv_ksplit(hseen, HBUCKETS) instead of manually hacking
1090 * it below does not make any difference. It seems to work fine
1091 * with perl5.004_68 but given the probable nature of the bug,
1092 * that does not prove anything.
1094 * It's a shame because increasing the amount of buckets raises
1095 * store() throughput by 5%, but until I figure this out, I can't
1096 * allow for this to go into production.
1098 * It is reported fixed in 5.005, hence the #if.
1100 #if PERL_VERSION >= 5
1101 #define HBUCKETS 4096 /* Buckets for %hseen */
1102 HvMAX(cxt->hseen) = HBUCKETS - 1; /* keys %hseen = $HBUCKETS; */
1106 * The `hclass' hash uses the same settings as `hseen' above, but it is
1107 * used to assign sequential tags (numbers) to class names for blessed
1110 * We turn the shared key optimization on.
1113 cxt->hclass = newHV(); /* Where seen classnames are stored */
1115 #if PERL_VERSION >= 5
1116 HvMAX(cxt->hclass) = HBUCKETS - 1; /* keys %hclass = $HBUCKETS; */
1120 * The `hook' hash table is used to keep track of the references on
1121 * the STORABLE_freeze hook routines, when found in some class name.
1123 * It is assumed that the inheritance tree will not be changed during
1124 * storing, and that no new method will be dynamically created by the
1128 cxt->hook = newHV(); /* Table where hooks are cached */
1131 * The `hook_seen' array keeps track of all the SVs returned by
1132 * STORABLE_freeze hooks for us to serialize, so that they are not
1133 * reclaimed until the end of the serialization process. Each SV is
1134 * only stored once, the first time it is seen.
1137 cxt->hook_seen = newAV(); /* Lists SVs returned by STORABLE_freeze */
1141 * clean_store_context
1143 * Clean store context by
1145 static void clean_store_context(stcxt_t *cxt)
1149 TRACEME(("clean_store_context"));
1151 ASSERT(cxt->optype & ST_STORE, ("was performing a store()"));
1154 * Insert real values into hashes where we stored faked pointers.
1158 hv_iterinit(cxt->hseen);
1159 while ((he = hv_iternext(cxt->hseen))) /* Extra () for -Wall, grr.. */
1160 HeVAL(he) = &PL_sv_undef;
1164 hv_iterinit(cxt->hclass);
1165 while ((he = hv_iternext(cxt->hclass))) /* Extra () for -Wall, grr.. */
1166 HeVAL(he) = &PL_sv_undef;
1170 * And now dispose of them...
1172 * The surrounding if() protection has been added because there might be
1173 * some cases where this routine is called more than once, during
1174 * exceptionnal events. This was reported by Marc Lehmann when Storable
1175 * is executed from mod_perl, and the fix was suggested by him.
1176 * -- RAM, 20/12/2000
1180 HV *hseen = cxt->hseen;
1183 sv_free((SV *) hseen);
1187 HV *hclass = cxt->hclass;
1190 sv_free((SV *) hclass);
1194 HV *hook = cxt->hook;
1197 sv_free((SV *) hook);
1200 if (cxt->hook_seen) {
1201 AV *hook_seen = cxt->hook_seen;
1203 av_undef(hook_seen);
1204 sv_free((SV *) hook_seen);
1211 * init_retrieve_context
1213 * Initialize a new retrieve context for real recursion.
1215 static void init_retrieve_context(stcxt_t *cxt, int optype, int is_tainted)
1217 TRACEME(("init_retrieve_context"));
1220 * The hook hash table is used to keep track of the references on
1221 * the STORABLE_thaw hook routines, when found in some class name.
1223 * It is assumed that the inheritance tree will not be changed during
1224 * storing, and that no new method will be dynamically created by the
1228 cxt->hook = newHV(); /* Caches STORABLE_thaw */
1231 * If retrieving an old binary version, the cxt->retrieve_vtbl variable
1232 * was set to sv_old_retrieve. We'll need a hash table to keep track of
1233 * the correspondance between the tags and the tag number used by the
1234 * new retrieve routines.
1237 cxt->hseen = ((cxt->retrieve_vtbl == sv_old_retrieve) ? newHV() : 0);
1239 cxt->aseen = newAV(); /* Where retrieved objects are kept */
1240 cxt->aclass = newAV(); /* Where seen classnames are kept */
1241 cxt->tagnum = 0; /* Have to count objects... */
1242 cxt->classnum = 0; /* ...and class names as well */
1243 cxt->optype = optype;
1244 cxt->s_tainted = is_tainted;
1245 cxt->entry = 1; /* No recursion yet */
1246 #ifndef HAS_RESTRICTED_HASHES
1247 cxt->derestrict = -1; /* Fetched from perl if needed */
1249 #ifndef HAS_UTF8_ALL
1250 cxt->use_bytes = -1; /* Fetched from perl if needed */
1255 * clean_retrieve_context
1257 * Clean retrieve context by
1259 static void clean_retrieve_context(stcxt_t *cxt)
1261 TRACEME(("clean_retrieve_context"));
1263 ASSERT(cxt->optype & ST_RETRIEVE, ("was performing a retrieve()"));
1266 AV *aseen = cxt->aseen;
1269 sv_free((SV *) aseen);
1273 AV *aclass = cxt->aclass;
1276 sv_free((SV *) aclass);
1280 HV *hook = cxt->hook;
1283 sv_free((SV *) hook);
1287 HV *hseen = cxt->hseen;
1290 sv_free((SV *) hseen); /* optional HV, for backward compat. */
1299 * A workaround for the CROAK bug: cleanup the last context.
1301 static void clean_context(stcxt_t *cxt)
1303 TRACEME(("clean_context"));
1305 ASSERT(cxt->s_dirty, ("dirty context"));
1310 ASSERT(!cxt->membuf_ro, ("mbase is not read-only"));
1312 if (cxt->optype & ST_RETRIEVE)
1313 clean_retrieve_context(cxt);
1314 else if (cxt->optype & ST_STORE)
1315 clean_store_context(cxt);
1319 ASSERT(!cxt->s_dirty, ("context is clean"));
1320 ASSERT(cxt->entry == 0, ("context is reset"));
1326 * Allocate a new context and push it on top of the parent one.
1327 * This new context is made globally visible via SET_STCXT().
1329 static stcxt_t *allocate_context(parent_cxt)
1330 stcxt_t *parent_cxt;
1334 TRACEME(("allocate_context"));
1336 ASSERT(!parent_cxt->s_dirty, ("parent context clean"));
1338 Newz(0, cxt, 1, stcxt_t);
1339 cxt->prev = parent_cxt;
1342 ASSERT(!cxt->s_dirty, ("clean context"));
1350 * Free current context, which cannot be the "root" one.
1351 * Make the context underneath globally visible via SET_STCXT().
1353 static void free_context(cxt)
1356 stcxt_t *prev = cxt->prev;
1358 TRACEME(("free_context"));
1360 ASSERT(!cxt->s_dirty, ("clean context"));
1361 ASSERT(prev, ("not freeing root context"));
1371 ASSERT(cxt, ("context not void"));
1381 * Tells whether we're in the middle of a store operation.
1383 int is_storing(void)
1387 return cxt->entry && (cxt->optype & ST_STORE);
1393 * Tells whether we're in the middle of a retrieve operation.
1395 int is_retrieving(void)
1399 return cxt->entry && (cxt->optype & ST_RETRIEVE);
1403 * last_op_in_netorder
1405 * Returns whether last operation was made using network order.
1407 * This is typically out-of-band information that might prove useful
1408 * to people wishing to convert native to network order data when used.
1410 int last_op_in_netorder(void)
1414 return cxt->netorder;
1418 *** Hook lookup and calling routines.
1424 * A wrapper on gv_fetchmethod_autoload() which caches results.
1426 * Returns the routine reference as an SV*, or null if neither the package
1427 * nor its ancestors know about the method.
1429 static SV *pkg_fetchmeth(
1438 * The following code is the same as the one performed by UNIVERSAL::can
1442 gv = gv_fetchmethod_autoload(pkg, method, FALSE);
1443 if (gv && isGV(gv)) {
1444 sv = newRV((SV*) GvCV(gv));
1445 TRACEME(("%s->%s: 0x%"UVxf, HvNAME(pkg), method, PTR2UV(sv)));
1447 sv = newSVsv(&PL_sv_undef);
1448 TRACEME(("%s->%s: not found", HvNAME(pkg), method));
1452 * Cache the result, ignoring failure: if we can't store the value,
1453 * it just won't be cached.
1456 (void) hv_store(cache, HvNAME(pkg), strlen(HvNAME(pkg)), sv, 0);
1458 return SvOK(sv) ? sv : (SV *) 0;
1464 * Force cached value to be undef: hook ignored even if present.
1466 static void pkg_hide(
1471 (void) hv_store(cache,
1472 HvNAME(pkg), strlen(HvNAME(pkg)), newSVsv(&PL_sv_undef), 0);
1478 * Discard cached value: a whole fetch loop will be retried at next lookup.
1480 static void pkg_uncache(
1485 (void) hv_delete(cache, HvNAME(pkg), strlen(HvNAME(pkg)), G_DISCARD);
1491 * Our own "UNIVERSAL::can", which caches results.
1493 * Returns the routine reference as an SV*, or null if the object does not
1494 * know about the method.
1504 TRACEME(("pkg_can for %s->%s", HvNAME(pkg), method));
1507 * Look into the cache to see whether we already have determined
1508 * where the routine was, if any.
1510 * NOTA BENE: we don't use `method' at all in our lookup, since we know
1511 * that only one hook (i.e. always the same) is cached in a given cache.
1514 svh = hv_fetch(cache, HvNAME(pkg), strlen(HvNAME(pkg)), FALSE);
1518 TRACEME(("cached %s->%s: not found", HvNAME(pkg), method));
1521 TRACEME(("cached %s->%s: 0x%"UVxf,
1522 HvNAME(pkg), method, PTR2UV(sv)));
1527 TRACEME(("not cached yet"));
1528 return pkg_fetchmeth(cache, pkg, method); /* Fetch and cache */
1534 * Call routine as obj->hook(av) in scalar context.
1535 * Propagates the single returned value if not called in void context.
1537 static SV *scalar_call(
1548 TRACEME(("scalar_call (cloning=%d)", cloning));
1555 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1557 SV **ary = AvARRAY(av);
1558 int cnt = AvFILLp(av) + 1;
1560 XPUSHs(ary[0]); /* Frozen string */
1561 for (i = 1; i < cnt; i++) {
1562 TRACEME(("pushing arg #%d (0x%"UVxf")...",
1563 i, PTR2UV(ary[i])));
1564 XPUSHs(sv_2mortal(newRV(ary[i])));
1569 TRACEME(("calling..."));
1570 count = perl_call_sv(hook, flags); /* Go back to Perl code */
1571 TRACEME(("count = %d", count));
1577 SvREFCNT_inc(sv); /* We're returning it, must stay alive! */
1590 * Call routine obj->hook(cloning) in list context.
1591 * Returns the list of returned values in an array.
1593 static AV *array_call(
1603 TRACEME(("array_call (cloning=%d)", cloning));
1609 XPUSHs(obj); /* Target object */
1610 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1613 count = perl_call_sv(hook, G_ARRAY); /* Go back to Perl code */
1618 for (i = count - 1; i >= 0; i--) {
1620 av_store(av, i, SvREFCNT_inc(sv));
1633 * Lookup the class name in the `hclass' table and either assign it a new ID
1634 * or return the existing one, by filling in `classnum'.
1636 * Return true if the class was known, false if the ID was just generated.
1638 static int known_class(
1640 char *name, /* Class name */
1641 int len, /* Name length */
1645 HV *hclass = cxt->hclass;
1647 TRACEME(("known_class (%s)", name));
1650 * Recall that we don't store pointers in this hash table, but tags.
1651 * Therefore, we need LOW_32BITS() to extract the relevant parts.
1654 svh = hv_fetch(hclass, name, len, FALSE);
1656 *classnum = LOW_32BITS(*svh);
1661 * Unknown classname, we need to record it.
1665 if (!hv_store(hclass, name, len, INT2PTR(SV*, cxt->classnum), 0))
1666 CROAK(("Unable to record new classname"));
1668 *classnum = cxt->classnum;
1673 *** Sepcific store routines.
1679 * Store a reference.
1680 * Layout is SX_REF <object> or SX_OVERLOAD <object>.
1682 static int store_ref(stcxt_t *cxt, SV *sv)
1684 TRACEME(("store_ref (0x%"UVxf")", PTR2UV(sv)));
1687 * Follow reference, and check if target is overloaded.
1693 HV *stash = (HV *) SvSTASH(sv);
1694 if (stash && Gv_AMG(stash)) {
1695 TRACEME(("ref (0x%"UVxf") is overloaded", PTR2UV(sv)));
1696 PUTMARK(SX_OVERLOAD);
1702 return store(cxt, sv);
1710 * Layout is SX_LSCALAR <length> <data>, SX_SCALAR <length> <data> or SX_UNDEF.
1711 * The <data> section is omitted if <length> is 0.
1713 * If integer or double, the layout is SX_INTEGER <data> or SX_DOUBLE <data>.
1714 * Small integers (within [-127, +127]) are stored as SX_BYTE <byte>.
1716 static int store_scalar(stcxt_t *cxt, SV *sv)
1721 U32 flags = SvFLAGS(sv); /* "cc -O" may put it in register */
1723 TRACEME(("store_scalar (0x%"UVxf")", PTR2UV(sv)));
1726 * For efficiency, break the SV encapsulation by peaking at the flags
1727 * directly without using the Perl macros to avoid dereferencing
1728 * sv->sv_flags each time we wish to check the flags.
1731 if (!(flags & SVf_OK)) { /* !SvOK(sv) */
1732 if (sv == &PL_sv_undef) {
1733 TRACEME(("immortal undef"));
1734 PUTMARK(SX_SV_UNDEF);
1736 TRACEME(("undef at 0x%"UVxf, PTR2UV(sv)));
1743 * Always store the string representation of a scalar if it exists.
1744 * Gisle Aas provided me with this test case, better than a long speach:
1746 * perl -MDevel::Peek -le '$a="abc"; $a+0; Dump($a)'
1747 * SV = PVNV(0x80c8520)
1749 * FLAGS = (NOK,POK,pNOK,pPOK)
1752 * PV = 0x80c83d0 "abc"\0
1756 * Write SX_SCALAR, length, followed by the actual data.
1758 * Otherwise, write an SX_BYTE, SX_INTEGER or an SX_DOUBLE as
1759 * appropriate, followed by the actual (binary) data. A double
1760 * is written as a string if network order, for portability.
1762 * NOTE: instead of using SvNOK(sv), we test for SvNOKp(sv).
1763 * The reason is that when the scalar value is tainted, the SvNOK(sv)
1766 * The test for a read-only scalar with both POK and NOK set is meant
1767 * to quickly detect &PL_sv_yes and &PL_sv_no without having to pay the
1768 * address comparison for each scalar we store.
1771 #define SV_MAYBE_IMMORTAL (SVf_READONLY|SVf_POK|SVf_NOK)
1773 if ((flags & SV_MAYBE_IMMORTAL) == SV_MAYBE_IMMORTAL) {
1774 if (sv == &PL_sv_yes) {
1775 TRACEME(("immortal yes"));
1777 } else if (sv == &PL_sv_no) {
1778 TRACEME(("immortal no"));
1781 pv = SvPV(sv, len); /* We know it's SvPOK */
1782 goto string; /* Share code below */
1784 } else if (flags & SVp_POK) { /* SvPOKp(sv) => string */
1785 I32 wlen; /* For 64-bit machines */
1789 * Will come here from below with pv and len set if double & netorder,
1790 * or from above if it was readonly, POK and NOK but neither &PL_sv_yes
1795 wlen = (I32) len; /* WLEN via STORE_SCALAR expects I32 */
1797 STORE_UTF8STR(pv, wlen);
1799 STORE_SCALAR(pv, wlen);
1800 TRACEME(("ok (scalar 0x%"UVxf" '%s', length = %"IVdf")",
1801 PTR2UV(sv), SvPVX(sv), (IV)len));
1803 } else if (flags & SVp_NOK) { /* SvNOKp(sv) => double */
1807 * Watch for number being an integer in disguise.
1809 if (nv == (NV) (iv = I_V(nv))) {
1810 TRACEME(("double %"NVff" is actually integer %"IVdf, nv, iv));
1811 goto integer; /* Share code below */
1814 if (cxt->netorder) {
1815 TRACEME(("double %"NVff" stored as string", nv));
1817 goto string; /* Share code above */
1821 WRITE(&nv, sizeof(nv));
1823 TRACEME(("ok (double 0x%"UVxf", value = %"NVff")", PTR2UV(sv), nv));
1825 } else if (flags & SVp_IOK) { /* SvIOKp(sv) => integer */
1829 * Will come here from above with iv set if double is an integer.
1834 * Optimize small integers into a single byte, otherwise store as
1835 * a real integer (converted into network order if they asked).
1838 if (iv >= -128 && iv <= 127) {
1839 unsigned char siv = (unsigned char) (iv + 128); /* [0,255] */
1842 TRACEME(("small integer stored as %d", siv));
1843 } else if (cxt->netorder) {
1846 niv = (I32) htonl(iv);
1847 TRACEME(("using network order"));
1850 TRACEME(("as-is for network order"));
1855 PUTMARK(SX_INTEGER);
1856 WRITE(&iv, sizeof(iv));
1859 TRACEME(("ok (integer 0x%"UVxf", value = %"IVdf")", PTR2UV(sv), iv));
1862 CROAK(("Can't determine type of %s(0x%"UVxf")",
1863 sv_reftype(sv, FALSE),
1866 return 0; /* Ok, no recursion on scalars */
1874 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
1875 * Each item is stored as <object>.
1877 static int store_array(stcxt_t *cxt, AV *av)
1880 I32 len = av_len(av) + 1;
1884 TRACEME(("store_array (0x%"UVxf")", PTR2UV(av)));
1887 * Signal array by emitting SX_ARRAY, followed by the array length.
1892 TRACEME(("size = %d", len));
1895 * Now store each item recursively.
1898 for (i = 0; i < len; i++) {
1899 sav = av_fetch(av, i, 0);
1901 TRACEME(("(#%d) undef item", i));
1905 TRACEME(("(#%d) item", i));
1906 if ((ret = store(cxt, *sav))) /* Extra () for -Wall, grr... */
1910 TRACEME(("ok (array)"));
1919 * Borrowed from perl source file pp_ctl.c, where it is used by pp_sort.
1922 sortcmp(const void *a, const void *b)
1924 return sv_cmp(*(SV * const *) a, *(SV * const *) b);
1931 * Store a hash table.
1933 * For a "normal" hash (not restricted, no utf8 keys):
1935 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
1936 * Values are stored as <object>.
1937 * Keys are stored as <length> <data>, the <data> section being omitted
1940 * For a "fancy" hash (restricted or utf8 keys):
1942 * Layout is SX_FLAG_HASH <size> <hash flags> followed by each key/value pair,
1944 * Values are stored as <object>.
1945 * Keys are stored as <flags> <length> <data>, the <data> section being omitted
1947 * Currently the only hash flag is "restriced"
1948 * Key flags are as for hv.h
1950 static int store_hash(stcxt_t *cxt, HV *hv)
1953 #ifdef HAS_RESTRICTED_HASHES
1962 int flagged_hash = ((SvREADONLY(hv)
1963 #ifdef HAS_HASH_KEY_FLAGS
1967 unsigned char hash_flags = (SvREADONLY(hv) ? SHV_RESTRICTED : 0);
1970 /* needs int cast for C++ compilers, doesn't it? */
1971 TRACEME(("store_hash (0x%"UVxf") (flags %x)", PTR2UV(hv),
1974 TRACEME(("store_hash (0x%"UVxf")", PTR2UV(hv)));
1978 * Signal hash by emitting SX_HASH, followed by the table length.
1982 PUTMARK(SX_FLAG_HASH);
1983 PUTMARK(hash_flags);
1988 TRACEME(("size = %d", len));
1991 * Save possible iteration state via each() on that table.
1994 riter = HvRITER(hv);
1995 eiter = HvEITER(hv);
1999 * Now store each item recursively.
2001 * If canonical is defined to some true value then store each
2002 * key/value pair in sorted order otherwise the order is random.
2003 * Canonical order is irrelevant when a deep clone operation is performed.
2005 * Fetch the value from perl only once per store() operation, and only
2010 !(cxt->optype & ST_CLONE) && (cxt->canonical == 1 ||
2011 (cxt->canonical < 0 && (cxt->canonical =
2012 (SvTRUE(perl_get_sv("Storable::canonical", TRUE)) ? 1 : 0))))
2015 * Storing in order, sorted by key.
2016 * Run through the hash, building up an array of keys in a
2017 * mortal array, sort the array and then run through the
2023 /*av_extend (av, len);*/
2025 TRACEME(("using canonical order"));
2027 for (i = 0; i < len; i++) {
2028 #ifdef HAS_RESTRICTED_HASHES
2029 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
2031 HE *he = hv_iternext(hv);
2033 SV *key = hv_iterkeysv(he);
2034 av_store(av, AvFILLp(av)+1, key); /* av_push(), really */
2037 qsort((char *) AvARRAY(av), len, sizeof(SV *), sortcmp);
2039 for (i = 0; i < len; i++) {
2040 unsigned char flags;
2044 SV *key = av_shift(av);
2045 HE *he = hv_fetch_ent(hv, key, 0, 0);
2046 SV *val = HeVAL(he);
2048 return 1; /* Internal error, not I/O error */
2051 * Store value first.
2054 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
2056 if ((ret = store(cxt, val))) /* Extra () for -Wall, grr... */
2061 * Keys are written after values to make sure retrieval
2062 * can be optimal in terms of memory usage, where keys are
2063 * read into a fixed unique buffer called kbuf.
2064 * See retrieve_hash() for details.
2067 /* Implementation of restricted hashes isn't nicely
2070 = (((hash_flags & SHV_RESTRICTED)
2072 ? SHV_K_LOCKED : 0);
2073 if (val == &PL_sv_undef)
2074 flags |= SHV_K_PLACEHOLDER;
2076 keyval = SvPV(key, keylen_tmp);
2077 keylen = keylen_tmp;
2078 #ifdef HAS_UTF8_HASHES
2079 /* If you build without optimisation on pre 5.6
2080 then nothing spots that SvUTF8(key) is always 0,
2081 so the block isn't optimised away, at which point
2082 the linker dislikes the reference to
2085 const char *keysave = keyval;
2086 bool is_utf8 = TRUE;
2088 /* Just casting the &klen to (STRLEN) won't work
2089 well if STRLEN and I32 are of different widths.
2091 keyval = (char*)bytes_from_utf8((U8*)keyval,
2095 /* If we were able to downgrade here, then than
2096 means that we have a key which only had chars
2097 0-255, but was utf8 encoded. */
2099 if (keyval != keysave) {
2100 keylen = keylen_tmp;
2101 flags |= SHV_K_WASUTF8;
2103 /* keylen_tmp can't have changed, so no need
2104 to assign back to keylen. */
2105 flags |= SHV_K_UTF8;
2112 TRACEME(("(#%d) key '%s' flags %x %u", i, keyval, flags, *keyval));
2114 assert (flags == 0);
2115 TRACEME(("(#%d) key '%s'", i, keyval));
2119 WRITE(keyval, keylen);
2120 if (flags & SHV_K_WASUTF8)
2125 * Free up the temporary array
2134 * Storing in "random" order (in the order the keys are stored
2135 * within the the hash). This is the default and will be faster!
2138 for (i = 0; i < len; i++) {
2141 unsigned char flags;
2142 #ifdef HV_ITERNEXT_WANTPLACEHOLDERS
2143 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
2145 HE *he = hv_iternext(hv);
2147 SV *val = (he ? hv_iterval(hv, he) : 0);
2152 return 1; /* Internal error, not I/O error */
2155 * Store value first.
2158 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
2160 if ((ret = store(cxt, val))) /* Extra () for -Wall, grr... */
2163 /* Implementation of restricted hashes isn't nicely
2166 = (((hash_flags & SHV_RESTRICTED)
2168 ? SHV_K_LOCKED : 0);
2169 if (val == &PL_sv_undef)
2170 flags |= SHV_K_PLACEHOLDER;
2172 hek = HeKEY_hek(he);
2174 if (len == HEf_SVKEY) {
2175 /* This is somewhat sick, but the internal APIs are
2176 * such that XS code could put one of these in in
2178 * Maybe we should be capable of storing one if
2181 key_sv = HeKEY_sv(he);
2182 flags |= SHV_K_ISSV;
2184 /* Regular string key. */
2185 #ifdef HAS_HASH_KEY_FLAGS
2187 flags |= SHV_K_UTF8;
2188 if (HEK_WASUTF8(hek))
2189 flags |= SHV_K_WASUTF8;
2195 * Keys are written after values to make sure retrieval
2196 * can be optimal in terms of memory usage, where keys are
2197 * read into a fixed unique buffer called kbuf.
2198 * See retrieve_hash() for details.
2203 TRACEME(("(#%d) key '%s' flags %x", i, key, flags));
2205 assert (flags == 0);
2206 TRACEME(("(#%d) key '%s'", i, key));
2208 if (flags & SHV_K_ISSV) {
2218 TRACEME(("ok (hash 0x%"UVxf")", PTR2UV(hv)));
2221 HvRITER(hv) = riter; /* Restore hash iterator state */
2222 HvEITER(hv) = eiter;
2230 * When storing a tied object (be it a tied scalar, array or hash), we lay out
2231 * a special mark, followed by the underlying tied object. For instance, when
2232 * dealing with a tied hash, we store SX_TIED_HASH <hash object>, where
2233 * <hash object> stands for the serialization of the tied hash.
2235 static int store_tied(stcxt_t *cxt, SV *sv)
2239 int svt = SvTYPE(sv);
2242 TRACEME(("store_tied (0x%"UVxf")", PTR2UV(sv)));
2245 * We have a small run-time penalty here because we chose to factorise
2246 * all tieds objects into the same routine, and not have a store_tied_hash,
2247 * a store_tied_array, etc...
2249 * Don't use a switch() statement, as most compilers don't optimize that
2250 * well for 2/3 values. An if() else if() cascade is just fine. We put
2251 * tied hashes first, as they are the most likely beasts.
2254 if (svt == SVt_PVHV) {
2255 TRACEME(("tied hash"));
2256 PUTMARK(SX_TIED_HASH); /* Introduces tied hash */
2257 } else if (svt == SVt_PVAV) {
2258 TRACEME(("tied array"));
2259 PUTMARK(SX_TIED_ARRAY); /* Introduces tied array */
2261 TRACEME(("tied scalar"));
2262 PUTMARK(SX_TIED_SCALAR); /* Introduces tied scalar */
2266 if (!(mg = mg_find(sv, mtype)))
2267 CROAK(("No magic '%c' found while storing tied %s", mtype,
2268 (svt == SVt_PVHV) ? "hash" :
2269 (svt == SVt_PVAV) ? "array" : "scalar"));
2272 * The mg->mg_obj found by mg_find() above actually points to the
2273 * underlying tied Perl object implementation. For instance, if the
2274 * original SV was that of a tied array, then mg->mg_obj is an AV.
2276 * Note that we store the Perl object as-is. We don't call its FETCH
2277 * method along the way. At retrieval time, we won't call its STORE
2278 * method either, but the tieing magic will be re-installed. In itself,
2279 * that ensures that the tieing semantics are preserved since futher
2280 * accesses on the retrieved object will indeed call the magic methods...
2283 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
2286 TRACEME(("ok (tied)"));
2294 * Stores a reference to an item within a tied structure:
2296 * . \$h{key}, stores both the (tied %h) object and 'key'.
2297 * . \$a[idx], stores both the (tied @a) object and 'idx'.
2299 * Layout is therefore either:
2300 * SX_TIED_KEY <object> <key>
2301 * SX_TIED_IDX <object> <index>
2303 static int store_tied_item(stcxt_t *cxt, SV *sv)
2308 TRACEME(("store_tied_item (0x%"UVxf")", PTR2UV(sv)));
2310 if (!(mg = mg_find(sv, 'p')))
2311 CROAK(("No magic 'p' found while storing reference to tied item"));
2314 * We discriminate between \$h{key} and \$a[idx] via mg_ptr.
2318 TRACEME(("store_tied_item: storing a ref to a tied hash item"));
2319 PUTMARK(SX_TIED_KEY);
2320 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2322 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
2325 TRACEME(("store_tied_item: storing PTR 0x%"UVxf, PTR2UV(mg->mg_ptr)));
2327 if ((ret = store(cxt, (SV *) mg->mg_ptr))) /* Idem, for -Wall */
2330 I32 idx = mg->mg_len;
2332 TRACEME(("store_tied_item: storing a ref to a tied array item "));
2333 PUTMARK(SX_TIED_IDX);
2334 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2336 if ((ret = store(cxt, mg->mg_obj))) /* Idem, for -Wall */
2339 TRACEME(("store_tied_item: storing IDX %d", idx));
2344 TRACEME(("ok (tied item)"));
2350 * store_hook -- dispatched manually, not via sv_store[]
2352 * The blessed SV is serialized by a hook.
2356 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2358 * where <flags> indicates how long <len>, <len2> and <len3> are, whether
2359 * the trailing part [] is present, the type of object (scalar, array or hash).
2360 * There is also a bit which says how the classname is stored between:
2365 * and when the <index> form is used (classname already seen), the "large
2366 * classname" bit in <flags> indicates how large the <index> is.
2368 * The serialized string returned by the hook is of length <len2> and comes
2369 * next. It is an opaque string for us.
2371 * Those <len3> object IDs which are listed last represent the extra references
2372 * not directly serialized by the hook, but which are linked to the object.
2374 * When recursion is mandated to resolve object-IDs not yet seen, we have
2375 * instead, with <header> being flags with bits set to indicate the object type
2376 * and that recursion was indeed needed:
2378 * SX_HOOK <header> <object> <header> <object> <flags>
2380 * that same header being repeated between serialized objects obtained through
2381 * recursion, until we reach flags indicating no recursion, at which point
2382 * we know we've resynchronized with a single layout, after <flags>.
2384 * When storing a blessed ref to a tied variable, the following format is
2387 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
2389 * The first <flags> indication carries an object of type SHT_EXTRA, and the
2390 * real object type is held in the <extra> flag. At the very end of the
2391 * serialization stream, the underlying magic object is serialized, just like
2392 * any other tied variable.
2394 static int store_hook(
2407 int count; /* really len3 + 1 */
2408 unsigned char flags;
2411 int recursed = 0; /* counts recursion */
2412 int obj_type; /* object type, on 2 bits */
2415 int clone = cxt->optype & ST_CLONE;
2416 char mtype = '\0'; /* for blessed ref to tied structures */
2417 unsigned char eflags = '\0'; /* used when object type is SHT_EXTRA */
2419 TRACEME(("store_hook, class \"%s\", tagged #%d", HvNAME(pkg), cxt->tagnum));
2422 * Determine object type on 2 bits.
2427 obj_type = SHT_SCALAR;
2430 obj_type = SHT_ARRAY;
2433 obj_type = SHT_HASH;
2437 * Produced by a blessed ref to a tied data structure, $o in the
2438 * following Perl code.
2442 * my $o = bless \%h, 'BAR';
2444 * Signal the tie-ing magic by setting the object type as SHT_EXTRA
2445 * (since we have only 2 bits in <flags> to store the type), and an
2446 * <extra> byte flag will be emitted after the FIRST <flags> in the
2447 * stream, carrying what we put in `eflags'.
2449 obj_type = SHT_EXTRA;
2450 switch (SvTYPE(sv)) {
2452 eflags = (unsigned char) SHT_THASH;
2456 eflags = (unsigned char) SHT_TARRAY;
2460 eflags = (unsigned char) SHT_TSCALAR;
2466 CROAK(("Unexpected object type (%d) in store_hook()", type));
2468 flags = SHF_NEED_RECURSE | obj_type;
2470 class = HvNAME(pkg);
2471 len = strlen(class);
2474 * To call the hook, we need to fake a call like:
2476 * $object->STORABLE_freeze($cloning);
2478 * but we don't have the $object here. For instance, if $object is
2479 * a blessed array, what we have in `sv' is the array, and we can't
2480 * call a method on those.
2482 * Therefore, we need to create a temporary reference to the object and
2483 * make the call on that reference.
2486 TRACEME(("about to call STORABLE_freeze on class %s", class));
2488 ref = newRV_noinc(sv); /* Temporary reference */
2489 av = array_call(ref, hook, clone); /* @a = $object->STORABLE_freeze($c) */
2491 SvREFCNT_dec(ref); /* Reclaim temporary reference */
2493 count = AvFILLp(av) + 1;
2494 TRACEME(("store_hook, array holds %d items", count));
2497 * If they return an empty list, it means they wish to ignore the
2498 * hook for this class (and not just this instance -- that's for them
2499 * to handle if they so wish).
2501 * Simply disable the cached entry for the hook (it won't be recomputed
2502 * since it's present in the cache) and recurse to store_blessed().
2507 * They must not change their mind in the middle of a serialization.
2510 if (hv_fetch(cxt->hclass, class, len, FALSE))
2511 CROAK(("Too late to ignore hooks for %s class \"%s\"",
2512 (cxt->optype & ST_CLONE) ? "cloning" : "storing", class));
2514 pkg_hide(cxt->hook, pkg, "STORABLE_freeze");
2516 ASSERT(!pkg_can(cxt->hook, pkg, "STORABLE_freeze"), ("hook invisible"));
2517 TRACEME(("ignoring STORABLE_freeze in class \"%s\"", class));
2519 return store_blessed(cxt, sv, type, pkg);
2523 * Get frozen string.
2527 pv = SvPV(ary[0], len2);
2530 * If they returned more than one item, we need to serialize some
2531 * extra references if not already done.
2533 * Loop over the array, starting at postion #1, and for each item,
2534 * ensure it is a reference, serialize it if not already done, and
2535 * replace the entry with the tag ID of the corresponding serialized
2538 * We CHEAT by not calling av_fetch() and read directly within the
2542 for (i = 1; i < count; i++) {
2546 AV *av_hook = cxt->hook_seen;
2549 CROAK(("Item #%d returned by STORABLE_freeze "
2550 "for %s is not a reference", i, class));
2551 xsv = SvRV(rsv); /* Follow ref to know what to look for */
2554 * Look in hseen and see if we have a tag already.
2555 * Serialize entry if not done already, and get its tag.
2558 if ((svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE)))
2559 goto sv_seen; /* Avoid moving code too far to the right */
2561 TRACEME(("listed object %d at 0x%"UVxf" is unknown", i-1, PTR2UV(xsv)));
2564 * We need to recurse to store that object and get it to be known
2565 * so that we can resolve the list of object-IDs at retrieve time.
2567 * The first time we do this, we need to emit the proper header
2568 * indicating that we recursed, and what the type of object is (the
2569 * object we're storing via a user-hook). Indeed, during retrieval,
2570 * we'll have to create the object before recursing to retrieve the
2571 * others, in case those would point back at that object.
2574 /* [SX_HOOK] <flags> [<extra>] <object>*/
2578 if (obj_type == SHT_EXTRA)
2583 if ((ret = store(cxt, xsv))) /* Given by hook for us to store */
2586 svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE);
2588 CROAK(("Could not serialize item #%d from hook in %s", i, class));
2591 * It was the first time we serialized `xsv'.
2593 * Keep this SV alive until the end of the serialization: if we
2594 * disposed of it right now by decrementing its refcount, and it was
2595 * a temporary value, some next temporary value allocated during
2596 * another STORABLE_freeze might take its place, and we'd wrongly
2597 * assume that new SV was already serialized, based on its presence
2600 * Therefore, push it away in cxt->hook_seen.
2603 av_store(av_hook, AvFILLp(av_hook)+1, SvREFCNT_inc(xsv));
2607 * Dispose of the REF they returned. If we saved the `xsv' away
2608 * in the array of returned SVs, that will not cause the underlying
2609 * referenced SV to be reclaimed.
2612 ASSERT(SvREFCNT(xsv) > 1, ("SV will survive disposal of its REF"));
2613 SvREFCNT_dec(rsv); /* Dispose of reference */
2616 * Replace entry with its tag (not a real SV, so no refcnt increment)
2620 TRACEME(("listed object %d at 0x%"UVxf" is tag #%"UVuf,
2621 i-1, PTR2UV(xsv), PTR2UV(*svh)));
2625 * Allocate a class ID if not already done.
2627 * This needs to be done after the recursion above, since at retrieval
2628 * time, we'll see the inner objects first. Many thanks to
2629 * Salvador Ortiz Garcia <sog@msg.com.mx> who spot that bug and
2630 * proposed the right fix. -- RAM, 15/09/2000
2633 if (!known_class(cxt, class, len, &classnum)) {
2634 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2635 classnum = -1; /* Mark: we must store classname */
2637 TRACEME(("already seen class %s, ID = %d", class, classnum));
2641 * Compute leading flags.
2645 if (((classnum == -1) ? len : classnum) > LG_SCALAR)
2646 flags |= SHF_LARGE_CLASSLEN;
2648 flags |= SHF_IDX_CLASSNAME;
2649 if (len2 > LG_SCALAR)
2650 flags |= SHF_LARGE_STRLEN;
2652 flags |= SHF_HAS_LIST;
2653 if (count > (LG_SCALAR + 1))
2654 flags |= SHF_LARGE_LISTLEN;
2657 * We're ready to emit either serialized form:
2659 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2660 * SX_HOOK <flags> <index> <len2> <str> [<len3> <object-IDs>]
2662 * If we recursed, the SX_HOOK has already been emitted.
2665 TRACEME(("SX_HOOK (recursed=%d) flags=0x%x "
2666 "class=%"IVdf" len=%"IVdf" len2=%"IVdf" len3=%d",
2667 recursed, flags, (IV)classnum, (IV)len, (IV)len2, count-1));
2669 /* SX_HOOK <flags> [<extra>] */
2673 if (obj_type == SHT_EXTRA)
2678 /* <len> <classname> or <index> */
2679 if (flags & SHF_IDX_CLASSNAME) {
2680 if (flags & SHF_LARGE_CLASSLEN)
2683 unsigned char cnum = (unsigned char) classnum;
2687 if (flags & SHF_LARGE_CLASSLEN)
2690 unsigned char clen = (unsigned char) len;
2693 WRITE(class, len); /* Final \0 is omitted */
2696 /* <len2> <frozen-str> */
2697 if (flags & SHF_LARGE_STRLEN) {
2698 I32 wlen2 = len2; /* STRLEN might be 8 bytes */
2699 WLEN(wlen2); /* Must write an I32 for 64-bit machines */
2701 unsigned char clen = (unsigned char) len2;
2705 WRITE(pv, (SSize_t)len2); /* Final \0 is omitted */
2707 /* [<len3> <object-IDs>] */
2708 if (flags & SHF_HAS_LIST) {
2709 int len3 = count - 1;
2710 if (flags & SHF_LARGE_LISTLEN)
2713 unsigned char clen = (unsigned char) len3;
2718 * NOTA BENE, for 64-bit machines: the ary[i] below does not yield a
2719 * real pointer, rather a tag number, well under the 32-bit limit.
2722 for (i = 1; i < count; i++) {
2723 I32 tagval = htonl(LOW_32BITS(ary[i]));
2725 TRACEME(("object %d, tag #%d", i-1, ntohl(tagval)));
2730 * Free the array. We need extra care for indices after 0, since they
2731 * don't hold real SVs but integers cast.
2735 AvFILLp(av) = 0; /* Cheat, nothing after 0 interests us */
2740 * If object was tied, need to insert serialization of the magic object.
2743 if (obj_type == SHT_EXTRA) {
2746 if (!(mg = mg_find(sv, mtype))) {
2747 int svt = SvTYPE(sv);
2748 CROAK(("No magic '%c' found while storing ref to tied %s with hook",
2749 mtype, (svt == SVt_PVHV) ? "hash" :
2750 (svt == SVt_PVAV) ? "array" : "scalar"));
2753 TRACEME(("handling the magic object 0x%"UVxf" part of 0x%"UVxf,
2754 PTR2UV(mg->mg_obj), PTR2UV(sv)));
2760 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
2768 * store_blessed -- dispatched manually, not via sv_store[]
2770 * Check whether there is a STORABLE_xxx hook defined in the class or in one
2771 * of its ancestors. If there is, then redispatch to store_hook();
2773 * Otherwise, the blessed SV is stored using the following layout:
2775 * SX_BLESS <flag> <len> <classname> <object>
2777 * where <flag> indicates whether <len> is stored on 0 or 4 bytes, depending
2778 * on the high-order bit in flag: if 1, then length follows on 4 bytes.
2779 * Otherwise, the low order bits give the length, thereby giving a compact
2780 * representation for class names less than 127 chars long.
2782 * Each <classname> seen is remembered and indexed, so that the next time
2783 * an object in the blessed in the same <classname> is stored, the following
2786 * SX_IX_BLESS <flag> <index> <object>
2788 * where <index> is the classname index, stored on 0 or 4 bytes depending
2789 * on the high-order bit in flag (same encoding as above for <len>).
2791 static int store_blessed(
2802 TRACEME(("store_blessed, type %d, class \"%s\"", type, HvNAME(pkg)));
2805 * Look for a hook for this blessed SV and redirect to store_hook()
2809 hook = pkg_can(cxt->hook, pkg, "STORABLE_freeze");
2811 return store_hook(cxt, sv, type, pkg, hook);
2814 * This is a blessed SV without any serialization hook.
2817 class = HvNAME(pkg);
2818 len = strlen(class);
2820 TRACEME(("blessed 0x%"UVxf" in %s, no hook: tagged #%d",
2821 PTR2UV(sv), class, cxt->tagnum));
2824 * Determine whether it is the first time we see that class name (in which
2825 * case it will be stored in the SX_BLESS form), or whether we already
2826 * saw that class name before (in which case the SX_IX_BLESS form will be
2830 if (known_class(cxt, class, len, &classnum)) {
2831 TRACEME(("already seen class %s, ID = %d", class, classnum));
2832 PUTMARK(SX_IX_BLESS);
2833 if (classnum <= LG_BLESS) {
2834 unsigned char cnum = (unsigned char) classnum;
2837 unsigned char flag = (unsigned char) 0x80;
2842 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2844 if (len <= LG_BLESS) {
2845 unsigned char clen = (unsigned char) len;
2848 unsigned char flag = (unsigned char) 0x80;
2850 WLEN(len); /* Don't BER-encode, this should be rare */
2852 WRITE(class, len); /* Final \0 is omitted */
2856 * Now emit the <object> part.
2859 return SV_STORE(type)(cxt, sv);
2865 * We don't know how to store the item we reached, so return an error condition.
2866 * (it's probably a GLOB, some CODE reference, etc...)
2868 * If they defined the `forgive_me' variable at the Perl level to some
2869 * true value, then don't croak, just warn, and store a placeholder string
2872 static int store_other(stcxt_t *cxt, SV *sv)
2875 static char buf[80];
2877 TRACEME(("store_other"));
2880 * Fetch the value from perl only once per store() operation.
2884 cxt->forgive_me == 0 ||
2885 (cxt->forgive_me < 0 && !(cxt->forgive_me =
2886 SvTRUE(perl_get_sv("Storable::forgive_me", TRUE)) ? 1 : 0))
2888 CROAK(("Can't store %s items", sv_reftype(sv, FALSE)));
2890 warn("Can't store item %s(0x%"UVxf")",
2891 sv_reftype(sv, FALSE), PTR2UV(sv));
2894 * Store placeholder string as a scalar instead...
2897 (void) sprintf(buf, "You lost %s(0x%"UVxf")%c", sv_reftype(sv, FALSE),
2898 PTR2UV(sv), (char) 0);
2901 STORE_SCALAR(buf, len);
2902 TRACEME(("ok (dummy \"%s\", length = %"IVdf")", buf, len));
2908 *** Store driving routines
2914 * WARNING: partially duplicates Perl's sv_reftype for speed.
2916 * Returns the type of the SV, identified by an integer. That integer
2917 * may then be used to index the dynamic routine dispatch table.
2919 static int sv_type(SV *sv)
2921 switch (SvTYPE(sv)) {
2926 * No need to check for ROK, that can't be set here since there
2927 * is no field capable of hodling the xrv_rv reference.
2935 * Starting from SVt_PV, it is possible to have the ROK flag
2936 * set, the pointer to the other SV being either stored in
2937 * the xrv_rv (in the case of a pure SVt_RV), or as the
2938 * xpv_pv field of an SVt_PV and its heirs.
2940 * However, those SV cannot be magical or they would be an
2941 * SVt_PVMG at least.
2943 return SvROK(sv) ? svis_REF : svis_SCALAR;
2945 case SVt_PVLV: /* Workaround for perl5.004_04 "LVALUE" bug */
2946 if (SvRMAGICAL(sv) && (mg_find(sv, 'p')))
2947 return svis_TIED_ITEM;
2950 if (SvRMAGICAL(sv) && (mg_find(sv, 'q')))
2952 return SvROK(sv) ? svis_REF : svis_SCALAR;
2954 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
2958 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
2971 * Recursively store objects pointed to by the sv to the specified file.
2973 * Layout is <content> or SX_OBJECT <tagnum> if we reach an already stored
2974 * object (one for which storage has started -- it may not be over if we have
2975 * a self-referenced structure). This data set forms a stored <object>.
2977 static int store(stcxt_t *cxt, SV *sv)
2982 HV *hseen = cxt->hseen;
2984 TRACEME(("store (0x%"UVxf")", PTR2UV(sv)));
2987 * If object has already been stored, do not duplicate data.
2988 * Simply emit the SX_OBJECT marker followed by its tag data.
2989 * The tag is always written in network order.
2991 * NOTA BENE, for 64-bit machines: the "*svh" below does not yield a
2992 * real pointer, rather a tag number (watch the insertion code below).
2993 * That means it pobably safe to assume it is well under the 32-bit limit,
2994 * and makes the truncation safe.
2995 * -- RAM, 14/09/1999
2998 svh = hv_fetch(hseen, (char *) &sv, sizeof(sv), FALSE);
3000 I32 tagval = htonl(LOW_32BITS(*svh));
3002 TRACEME(("object 0x%"UVxf" seen as #%d", PTR2UV(sv), ntohl(tagval)));
3010 * Allocate a new tag and associate it with the address of the sv being
3011 * stored, before recursing...
3013 * In order to avoid creating new SvIVs to hold the tagnum we just
3014 * cast the tagnum to an SV pointer and store that in the hash. This
3015 * means that we must clean up the hash manually afterwards, but gives
3016 * us a 15% throughput increase.
3021 if (!hv_store(hseen,
3022 (char *) &sv, sizeof(sv), INT2PTR(SV*, cxt->tagnum), 0))
3026 * Store `sv' and everything beneath it, using appropriate routine.
3027 * Abort immediately if we get a non-zero status back.
3032 TRACEME(("storing 0x%"UVxf" tag #%d, type %d...",
3033 PTR2UV(sv), cxt->tagnum, type));
3036 HV *pkg = SvSTASH(sv);
3037 ret = store_blessed(cxt, sv, type, pkg);
3039 ret = SV_STORE(type)(cxt, sv);
3041 TRACEME(("%s (stored 0x%"UVxf", refcnt=%d, %s)",
3042 ret ? "FAILED" : "ok", PTR2UV(sv),
3043 SvREFCNT(sv), sv_reftype(sv, FALSE)));
3051 * Write magic number and system information into the file.
3052 * Layout is <magic> <network> [<len> <byteorder> <sizeof int> <sizeof long>
3053 * <sizeof ptr>] where <len> is the length of the byteorder hexa string.
3054 * All size and lenghts are written as single characters here.
3056 * Note that no byte ordering info is emitted when <network> is true, since
3057 * integers will be emitted in network order in that case.
3059 static int magic_write(stcxt_t *cxt)
3061 char buf[256]; /* Enough room for 256 hexa digits */
3063 int use_network_order = cxt->netorder;
3065 TRACEME(("magic_write on fd=%d", cxt->fio ? PerlIO_fileno(cxt->fio)
3069 WRITE(magicstr, (SSize_t)strlen(magicstr)); /* Don't write final \0 */
3072 * Starting with 0.6, the "use_network_order" byte flag is also used to
3073 * indicate the version number of the binary image, encoded in the upper
3074 * bits. The bit 0 is always used to indicate network order.
3078 ((use_network_order ? 0x1 : 0x0) | (STORABLE_BIN_MAJOR << 1));
3082 * Starting with 0.7, a full byte is dedicated to the minor version of
3083 * the binary format, which is incremented only when new markers are
3084 * introduced, for instance, but when backward compatibility is preserved.
3087 PUTMARK((unsigned char) STORABLE_BIN_WRITE_MINOR);
3089 if (use_network_order)
3090 return 0; /* Don't bother with byte ordering */
3092 sprintf(buf, "%lx", (unsigned long) BYTEORDER);
3093 c = (unsigned char) strlen(buf);
3095 WRITE(buf, (SSize_t)c); /* Don't write final \0 */
3096 PUTMARK((unsigned char) sizeof(int));
3097 PUTMARK((unsigned char) sizeof(long));
3098 PUTMARK((unsigned char) sizeof(char *));
3099 PUTMARK((unsigned char) sizeof(NV));
3101 TRACEME(("ok (magic_write byteorder = 0x%lx [%d], I%d L%d P%d D%d)",
3102 (unsigned long) BYTEORDER, (int) c,
3103 (int) sizeof(int), (int) sizeof(long),
3104 (int) sizeof(char *), (int) sizeof(NV)));
3112 * Common code for store operations.
3114 * When memory store is requested (f = NULL) and a non null SV* is given in
3115 * `res', it is filled with a new SV created out of the memory buffer.
3117 * It is required to provide a non-null `res' when the operation type is not
3118 * dclone() and store() is performed to memory.
3120 static int do_store(
3130 ASSERT(!(f == 0 && !(optype & ST_CLONE)) || res,
3131 ("must supply result SV pointer for real recursion to memory"));
3133 TRACEME(("do_store (optype=%d, netorder=%d)",
3134 optype, network_order));
3139 * Workaround for CROAK leak: if they enter with a "dirty" context,
3140 * free up memory for them now.
3147 * Now that STORABLE_xxx hooks exist, it is possible that they try to
3148 * re-enter store() via the hooks. We need to stack contexts.
3152 cxt = allocate_context(cxt);
3156 ASSERT(cxt->entry == 1, ("starting new recursion"));
3157 ASSERT(!cxt->s_dirty, ("clean context"));
3160 * Ensure sv is actually a reference. From perl, we called something
3162 * pstore(FILE, \@array);
3163 * so we must get the scalar value behing that reference.
3167 CROAK(("Not a reference"));
3168 sv = SvRV(sv); /* So follow it to know what to store */
3171 * If we're going to store to memory, reset the buffer.
3178 * Prepare context and emit headers.
3181 init_store_context(cxt, f, optype, network_order);
3183 if (-1 == magic_write(cxt)) /* Emit magic and ILP info */
3184 return 0; /* Error */
3187 * Recursively store object...
3190 ASSERT(is_storing(), ("within store operation"));
3192 status = store(cxt, sv); /* Just do it! */
3195 * If they asked for a memory store and they provided an SV pointer,
3196 * make an SV string out of the buffer and fill their pointer.
3198 * When asking for ST_REAL, it's MANDATORY for the caller to provide
3199 * an SV, since context cleanup might free the buffer if we did recurse.
3200 * (unless caller is dclone(), which is aware of that).
3203 if (!cxt->fio && res)
3209 * The "root" context is never freed, since it is meant to be always
3210 * handy for the common case where no recursion occurs at all (i.e.
3211 * we enter store() outside of any Storable code and leave it, period).
3212 * We know it's the "root" context because there's nothing stacked
3217 * When deep cloning, we don't free the context: doing so would force
3218 * us to copy the data in the memory buffer. Sicne we know we're
3219 * about to enter do_retrieve...
3222 clean_store_context(cxt);
3223 if (cxt->prev && !(cxt->optype & ST_CLONE))
3226 TRACEME(("do_store returns %d", status));
3234 * Store the transitive data closure of given object to disk.
3235 * Returns 0 on error, a true value otherwise.
3237 int pstore(PerlIO *f, SV *sv)
3239 TRACEME(("pstore"));
3240 return do_store(f, sv, 0, FALSE, (SV**) 0);
3247 * Same as pstore(), but network order is used for integers and doubles are
3248 * emitted as strings.
3250 int net_pstore(PerlIO *f, SV *sv)
3252 TRACEME(("net_pstore"));
3253 return do_store(f, sv, 0, TRUE, (SV**) 0);
3263 * Build a new SV out of the content of the internal memory buffer.
3265 static SV *mbuf2sv(void)
3269 return newSVpv(mbase, MBUF_SIZE());
3275 * Store the transitive data closure of given object to memory.
3276 * Returns undef on error, a scalar value containing the data otherwise.
3282 TRACEME(("mstore"));
3284 if (!do_store((PerlIO*) 0, sv, 0, FALSE, &out))
3285 return &PL_sv_undef;
3293 * Same as mstore(), but network order is used for integers and doubles are
3294 * emitted as strings.
3296 SV *net_mstore(SV *sv)
3300 TRACEME(("net_mstore"));
3302 if (!do_store((PerlIO*) 0, sv, 0, TRUE, &out))
3303 return &PL_sv_undef;
3309 *** Specific retrieve callbacks.
3315 * Return an error via croak, since it is not possible that we get here
3316 * under normal conditions, when facing a file produced via pstore().
3318 static SV *retrieve_other(stcxt_t *cxt, char *cname)
3321 cxt->ver_major != STORABLE_BIN_MAJOR &&
3322 cxt->ver_minor != STORABLE_BIN_MINOR
3324 CROAK(("Corrupted storable %s (binary v%d.%d), current is v%d.%d",
3325 cxt->fio ? "file" : "string",
3326 cxt->ver_major, cxt->ver_minor,
3327 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
3329 CROAK(("Corrupted storable %s (binary v%d.%d)",
3330 cxt->fio ? "file" : "string",
3331 cxt->ver_major, cxt->ver_minor));
3334 return (SV *) 0; /* Just in case */
3338 * retrieve_idx_blessed
3340 * Layout is SX_IX_BLESS <index> <object> with SX_IX_BLESS already read.
3341 * <index> can be coded on either 1 or 5 bytes.
3343 static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname)
3350 TRACEME(("retrieve_idx_blessed (#%d)", cxt->tagnum));
3351 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3353 GETMARK(idx); /* Index coded on a single char? */
3358 * Fetch classname in `aclass'
3361 sva = av_fetch(cxt->aclass, idx, FALSE);
3363 CROAK(("Class name #%"IVdf" should have been seen already", (IV) idx));
3365 class = SvPVX(*sva); /* We know it's a PV, by construction */
3367 TRACEME(("class ID %d => %s", idx, class));
3370 * Retrieve object and bless it.
3373 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
3381 * Layout is SX_BLESS <len> <classname> <object> with SX_BLESS already read.
3382 * <len> can be coded on either 1 or 5 bytes.
3384 static SV *retrieve_blessed(stcxt_t *cxt, char *cname)
3388 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3391 TRACEME(("retrieve_blessed (#%d)", cxt->tagnum));
3392 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3395 * Decode class name length and read that name.
3397 * Short classnames have two advantages: their length is stored on one
3398 * single byte, and the string can be read on the stack.
3401 GETMARK(len); /* Length coded on a single char? */
3404 TRACEME(("** allocating %d bytes for class name", len+1));
3405 New(10003, class, len+1, char);
3408 class[len] = '\0'; /* Mark string end */
3411 * It's a new classname, otherwise it would have been an SX_IX_BLESS.
3414 TRACEME(("new class name \"%s\" will bear ID = %d", class, cxt->classnum));
3416 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3420 * Retrieve object and bless it.
3423 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
3433 * Layout: SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
3434 * with leading mark already read, as usual.
3436 * When recursion was involved during serialization of the object, there
3437 * is an unknown amount of serialized objects after the SX_HOOK mark. Until
3438 * we reach a <flags> marker with the recursion bit cleared.
3440 * If the first <flags> byte contains a type of SHT_EXTRA, then the real type
3441 * is held in the <extra> byte, and if the object is tied, the serialized
3442 * magic object comes at the very end:
3444 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
3446 * This means the STORABLE_thaw hook will NOT get a tied variable during its
3447 * processing (since we won't have seen the magic object by the time the hook
3448 * is called). See comments below for why it was done that way.
3450 static SV *retrieve_hook(stcxt_t *cxt, char *cname)
3453 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3464 int clone = cxt->optype & ST_CLONE;
3466 unsigned int extra_type = 0;
3468 TRACEME(("retrieve_hook (#%d)", cxt->tagnum));
3469 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3472 * Read flags, which tell us about the type, and whether we need to recurse.
3478 * Create the (empty) object, and mark it as seen.
3480 * This must be done now, because tags are incremented, and during
3481 * serialization, the object tag was affected before recursion could
3485 obj_type = flags & SHF_TYPE_MASK;
3491 sv = (SV *) newAV();
3494 sv = (SV *) newHV();
3498 * Read <extra> flag to know the type of the object.
3499 * Record associated magic type for later.
3501 GETMARK(extra_type);
3502 switch (extra_type) {
3508 sv = (SV *) newAV();
3512 sv = (SV *) newHV();
3516 return retrieve_other(cxt, 0); /* Let it croak */
3520 return retrieve_other(cxt, 0); /* Let it croak */
3522 SEEN(sv, 0); /* Don't bless yet */
3525 * Whilst flags tell us to recurse, do so.
3527 * We don't need to remember the addresses returned by retrieval, because
3528 * all the references will be obtained through indirection via the object
3529 * tags in the object-ID list.
3532 while (flags & SHF_NEED_RECURSE) {
3533 TRACEME(("retrieve_hook recursing..."));
3534 rv = retrieve(cxt, 0);
3537 TRACEME(("retrieve_hook back with rv=0x%"UVxf,
3542 if (flags & SHF_IDX_CLASSNAME) {
3547 * Fetch index from `aclass'
3550 if (flags & SHF_LARGE_CLASSLEN)
3555 sva = av_fetch(cxt->aclass, idx, FALSE);
3557 CROAK(("Class name #%"IVdf" should have been seen already",
3560 class = SvPVX(*sva); /* We know it's a PV, by construction */
3561 TRACEME(("class ID %d => %s", idx, class));
3565 * Decode class name length and read that name.
3567 * NOTA BENE: even if the length is stored on one byte, we don't read
3568 * on the stack. Just like retrieve_blessed(), we limit the name to
3569 * LG_BLESS bytes. This is an arbitrary decision.
3572 if (flags & SHF_LARGE_CLASSLEN)
3577 if (len > LG_BLESS) {
3578 TRACEME(("** allocating %d bytes for class name", len+1));
3579 New(10003, class, len+1, char);
3583 class[len] = '\0'; /* Mark string end */
3586 * Record new classname.
3589 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3593 TRACEME(("class name: %s", class));
3596 * Decode user-frozen string length and read it in an SV.
3598 * For efficiency reasons, we read data directly into the SV buffer.
3599 * To understand that code, read retrieve_scalar()
3602 if (flags & SHF_LARGE_STRLEN)
3607 frozen = NEWSV(10002, len2);
3609 SAFEREAD(SvPVX(frozen), len2, frozen);
3610 SvCUR_set(frozen, len2);
3611 *SvEND(frozen) = '\0';
3613 (void) SvPOK_only(frozen); /* Validates string pointer */
3614 if (cxt->s_tainted) /* Is input source tainted? */
3617 TRACEME(("frozen string: %d bytes", len2));
3620 * Decode object-ID list length, if present.
3623 if (flags & SHF_HAS_LIST) {
3624 if (flags & SHF_LARGE_LISTLEN)
3630 av_extend(av, len3 + 1); /* Leave room for [0] */
3631 AvFILLp(av) = len3; /* About to be filled anyway */
3635 TRACEME(("has %d object IDs to link", len3));
3638 * Read object-ID list into array.
3639 * Because we pre-extended it, we can cheat and fill it manually.
3641 * We read object tags and we can convert them into SV* on the fly
3642 * because we know all the references listed in there (as tags)
3643 * have been already serialized, hence we have a valid correspondance
3644 * between each of those tags and the recreated SV.
3648 SV **ary = AvARRAY(av);
3650 for (i = 1; i <= len3; i++) { /* We leave [0] alone */
3657 svh = av_fetch(cxt->aseen, tag, FALSE);
3659 CROAK(("Object #%"IVdf" should have been retrieved already",
3662 ary[i] = SvREFCNT_inc(xsv);
3667 * Bless the object and look up the STORABLE_thaw hook.
3671 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3674 * Hook not found. Maybe they did not require the module where this
3675 * hook is defined yet?
3677 * If the require below succeeds, we'll be able to find the hook.
3678 * Still, it only works reliably when each class is defined in a
3682 SV *psv = newSVpvn("require ", 8);
3683 sv_catpv(psv, class);
3685 TRACEME(("No STORABLE_thaw defined for objects of class %s", class));
3686 TRACEME(("Going to require module '%s' with '%s'", class, SvPVX(psv)));
3688 perl_eval_sv(psv, G_DISCARD);
3692 * We cache results of pkg_can, so we need to uncache before attempting
3696 pkg_uncache(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3697 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3700 CROAK(("No STORABLE_thaw defined for objects of class %s "
3701 "(even after a \"require %s;\")", class, class));
3705 * If we don't have an `av' yet, prepare one.
3706 * Then insert the frozen string as item [0].
3714 AvARRAY(av)[0] = SvREFCNT_inc(frozen);
3719 * $object->STORABLE_thaw($cloning, $frozen, @refs);
3721 * where $object is our blessed (empty) object, $cloning is a boolean
3722 * telling whether we're running a deep clone, $frozen is the frozen
3723 * string the user gave us in his serializing hook, and @refs, which may
3724 * be empty, is the list of extra references he returned along for us
3727 * In effect, the hook is an alternate creation routine for the class,
3728 * the object itself being already created by the runtime.
3731 TRACEME(("calling STORABLE_thaw on %s at 0x%"UVxf" (%"IVdf" args)",
3732 class, PTR2UV(sv), AvFILLp(av) + 1));
3735 (void) scalar_call(rv, hook, clone, av, G_SCALAR|G_DISCARD);
3742 SvREFCNT_dec(frozen);
3745 if (!(flags & SHF_IDX_CLASSNAME) && class != buf)
3749 * If we had an <extra> type, then the object was not as simple, and
3750 * we need to restore extra magic now.
3756 TRACEME(("retrieving magic object for 0x%"UVxf"...", PTR2UV(sv)));
3758 rv = retrieve(cxt, 0); /* Retrieve <magic object> */
3760 TRACEME(("restoring the magic object 0x%"UVxf" part of 0x%"UVxf,
3761 PTR2UV(rv), PTR2UV(sv)));
3763 switch (extra_type) {
3765 sv_upgrade(sv, SVt_PVMG);
3768 sv_upgrade(sv, SVt_PVAV);
3769 AvREAL_off((AV *)sv);
3772 sv_upgrade(sv, SVt_PVHV);
3775 CROAK(("Forgot to deal with extra type %d", extra_type));
3780 * Adding the magic only now, well after the STORABLE_thaw hook was called
3781 * means the hook cannot know it deals with an object whose variable is
3782 * tied. But this is happening when retrieving $o in the following case:
3786 * my $o = bless \%h, 'BAR';
3788 * The 'BAR' class is NOT the one where %h is tied into. Therefore, as
3789 * far as the 'BAR' class is concerned, the fact that %h is not a REAL
3790 * hash but a tied one should not matter at all, and remain transparent.
3791 * This means the magic must be restored by Storable AFTER the hook is
3794 * That looks very reasonable to me, but then I've come up with this
3795 * after a bug report from David Nesting, who was trying to store such
3796 * an object and caused Storable to fail. And unfortunately, it was
3797 * also the easiest way to retrofit support for blessed ref to tied objects
3798 * into the existing design. -- RAM, 17/02/2001
3801 sv_magic(sv, rv, mtype, Nullch, 0);
3802 SvREFCNT_dec(rv); /* Undo refcnt inc from sv_magic() */
3810 * Retrieve reference to some other scalar.
3811 * Layout is SX_REF <object>, with SX_REF already read.
3813 static SV *retrieve_ref(stcxt_t *cxt, char *cname)
3818 TRACEME(("retrieve_ref (#%d)", cxt->tagnum));
3821 * We need to create the SV that holds the reference to the yet-to-retrieve
3822 * object now, so that we may record the address in the seen table.
3823 * Otherwise, if the object to retrieve references us, we won't be able
3824 * to resolve the SX_OBJECT we'll see at that point! Hence we cannot
3825 * do the retrieve first and use rv = newRV(sv) since it will be too late
3826 * for SEEN() recording.
3829 rv = NEWSV(10002, 0);
3830 SEEN(rv, cname); /* Will return if rv is null */
3831 sv = retrieve(cxt, 0); /* Retrieve <object> */
3833 return (SV *) 0; /* Failed */
3836 * WARNING: breaks RV encapsulation.
3838 * Now for the tricky part. We have to upgrade our existing SV, so that
3839 * it is now an RV on sv... Again, we cheat by duplicating the code
3840 * held in newSVrv(), since we already got our SV from retrieve().
3844 * SvRV(rv) = SvREFCNT_inc(sv);
3846 * here because the reference count we got from retrieve() above is
3847 * already correct: if the object was retrieved from the file, then
3848 * its reference count is one. Otherwise, if it was retrieved via
3849 * an SX_OBJECT indication, a ref count increment was done.
3852 sv_upgrade(rv, SVt_RV);
3853 SvRV(rv) = sv; /* $rv = \$sv */
3856 TRACEME(("ok (retrieve_ref at 0x%"UVxf")", PTR2UV(rv)));
3862 * retrieve_overloaded
3864 * Retrieve reference to some other scalar with overloading.
3865 * Layout is SX_OVERLOAD <object>, with SX_OVERLOAD already read.
3867 static SV *retrieve_overloaded(stcxt_t *cxt, char *cname)
3873 TRACEME(("retrieve_overloaded (#%d)", cxt->tagnum));
3876 * Same code as retrieve_ref(), duplicated to avoid extra call.
3879 rv = NEWSV(10002, 0);
3880 SEEN(rv, cname); /* Will return if rv is null */
3881 sv = retrieve(cxt, 0); /* Retrieve <object> */
3883 return (SV *) 0; /* Failed */
3886 * WARNING: breaks RV encapsulation.
3889 sv_upgrade(rv, SVt_RV);
3890 SvRV(rv) = sv; /* $rv = \$sv */
3894 * Restore overloading magic.
3897 stash = (HV *) SvSTASH (sv);
3898 if (!stash || !Gv_AMG(stash))
3899 CROAK(("Cannot restore overloading on %s(0x%"UVxf") (package %s)",
3900 sv_reftype(sv, FALSE),
3902 stash ? HvNAME(stash) : "<unknown>"));
3906 TRACEME(("ok (retrieve_overloaded at 0x%"UVxf")", PTR2UV(rv)));
3912 * retrieve_tied_array
3914 * Retrieve tied array
3915 * Layout is SX_TIED_ARRAY <object>, with SX_TIED_ARRAY already read.
3917 static SV *retrieve_tied_array(stcxt_t *cxt, char *cname)
3922 TRACEME(("retrieve_tied_array (#%d)", cxt->tagnum));
3924 tv = NEWSV(10002, 0);
3925 SEEN(tv, cname); /* Will return if tv is null */
3926 sv = retrieve(cxt, 0); /* Retrieve <object> */
3928 return (SV *) 0; /* Failed */
3930 sv_upgrade(tv, SVt_PVAV);
3931 AvREAL_off((AV *)tv);
3932 sv_magic(tv, sv, 'P', Nullch, 0);
3933 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3935 TRACEME(("ok (retrieve_tied_array at 0x%"UVxf")", PTR2UV(tv)));
3941 * retrieve_tied_hash
3943 * Retrieve tied hash
3944 * Layout is SX_TIED_HASH <object>, with SX_TIED_HASH already read.
3946 static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname)
3951 TRACEME(("retrieve_tied_hash (#%d)", cxt->tagnum));
3953 tv = NEWSV(10002, 0);
3954 SEEN(tv, cname); /* Will return if tv is null */
3955 sv = retrieve(cxt, 0); /* Retrieve <object> */
3957 return (SV *) 0; /* Failed */
3959 sv_upgrade(tv, SVt_PVHV);
3960 sv_magic(tv, sv, 'P', Nullch, 0);
3961 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3963 TRACEME(("ok (retrieve_tied_hash at 0x%"UVxf")", PTR2UV(tv)));
3969 * retrieve_tied_scalar
3971 * Retrieve tied scalar
3972 * Layout is SX_TIED_SCALAR <object>, with SX_TIED_SCALAR already read.
3974 static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname)
3979 TRACEME(("retrieve_tied_scalar (#%d)", cxt->tagnum));
3981 tv = NEWSV(10002, 0);
3982 SEEN(tv, cname); /* Will return if rv is null */
3983 sv = retrieve(cxt, 0); /* Retrieve <object> */
3985 return (SV *) 0; /* Failed */
3987 sv_upgrade(tv, SVt_PVMG);
3988 sv_magic(tv, sv, 'q', Nullch, 0);
3989 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3991 TRACEME(("ok (retrieve_tied_scalar at 0x%"UVxf")", PTR2UV(tv)));
3999 * Retrieve reference to value in a tied hash.
4000 * Layout is SX_TIED_KEY <object> <key>, with SX_TIED_KEY already read.
4002 static SV *retrieve_tied_key(stcxt_t *cxt, char *cname)
4008 TRACEME(("retrieve_tied_key (#%d)", cxt->tagnum));
4010 tv = NEWSV(10002, 0);
4011 SEEN(tv, cname); /* Will return if tv is null */
4012 sv = retrieve(cxt, 0); /* Retrieve <object> */
4014 return (SV *) 0; /* Failed */
4016 key = retrieve(cxt, 0); /* Retrieve <key> */
4018 return (SV *) 0; /* Failed */
4020 sv_upgrade(tv, SVt_PVMG);
4021 sv_magic(tv, sv, 'p', (char *)key, HEf_SVKEY);
4022 SvREFCNT_dec(key); /* Undo refcnt inc from sv_magic() */
4023 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4031 * Retrieve reference to value in a tied array.
4032 * Layout is SX_TIED_IDX <object> <idx>, with SX_TIED_IDX already read.
4034 static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname)
4040 TRACEME(("retrieve_tied_idx (#%d)", cxt->tagnum));
4042 tv = NEWSV(10002, 0);
4043 SEEN(tv, cname); /* Will return if tv is null */
4044 sv = retrieve(cxt, 0); /* Retrieve <object> */
4046 return (SV *) 0; /* Failed */
4048 RLEN(idx); /* Retrieve <idx> */
4050 sv_upgrade(tv, SVt_PVMG);
4051 sv_magic(tv, sv, 'p', Nullch, idx);
4052 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
4061 * Retrieve defined long (string) scalar.
4063 * Layout is SX_LSCALAR <length> <data>, with SX_LSCALAR already read.
4064 * The scalar is "long" in that <length> is larger than LG_SCALAR so it
4065 * was not stored on a single byte.
4067 static SV *retrieve_lscalar(stcxt_t *cxt, char *cname)
4073 TRACEME(("retrieve_lscalar (#%d), len = %"IVdf, cxt->tagnum, len));
4076 * Allocate an empty scalar of the suitable length.
4079 sv = NEWSV(10002, len);
4080 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4083 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
4085 * Now, for efficiency reasons, read data directly inside the SV buffer,
4086 * and perform the SV final settings directly by duplicating the final
4087 * work done by sv_setpv. Since we're going to allocate lots of scalars
4088 * this way, it's worth the hassle and risk.
4091 SAFEREAD(SvPVX(sv), len, sv);
4092 SvCUR_set(sv, len); /* Record C string length */
4093 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
4094 (void) SvPOK_only(sv); /* Validate string pointer */
4095 if (cxt->s_tainted) /* Is input source tainted? */
4096 SvTAINT(sv); /* External data cannot be trusted */
4098 TRACEME(("large scalar len %"IVdf" '%s'", len, SvPVX(sv)));
4099 TRACEME(("ok (retrieve_lscalar at 0x%"UVxf")", PTR2UV(sv)));
4107 * Retrieve defined short (string) scalar.
4109 * Layout is SX_SCALAR <length> <data>, with SX_SCALAR already read.
4110 * The scalar is "short" so <length> is single byte. If it is 0, there
4111 * is no <data> section.
4113 static SV *retrieve_scalar(stcxt_t *cxt, char *cname)
4119 TRACEME(("retrieve_scalar (#%d), len = %d", cxt->tagnum, len));
4122 * Allocate an empty scalar of the suitable length.
4125 sv = NEWSV(10002, len);
4126 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4129 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
4134 * newSV did not upgrade to SVt_PV so the scalar is undefined.
4135 * To make it defined with an empty length, upgrade it now...
4136 * Don't upgrade to a PV if the original type contains more
4137 * information than a scalar.
4139 if (SvTYPE(sv) <= SVt_PV) {
4140 sv_upgrade(sv, SVt_PV);
4143 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
4144 TRACEME(("ok (retrieve_scalar empty at 0x%"UVxf")", PTR2UV(sv)));
4147 * Now, for efficiency reasons, read data directly inside the SV buffer,
4148 * and perform the SV final settings directly by duplicating the final
4149 * work done by sv_setpv. Since we're going to allocate lots of scalars
4150 * this way, it's worth the hassle and risk.
4152 SAFEREAD(SvPVX(sv), len, sv);
4153 SvCUR_set(sv, len); /* Record C string length */
4154 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
4155 TRACEME(("small scalar len %d '%s'", len, SvPVX(sv)));
4158 (void) SvPOK_only(sv); /* Validate string pointer */
4159 if (cxt->s_tainted) /* Is input source tainted? */
4160 SvTAINT(sv); /* External data cannot be trusted */
4162 TRACEME(("ok (retrieve_scalar at 0x%"UVxf")", PTR2UV(sv)));
4169 * Like retrieve_scalar(), but tag result as utf8.
4170 * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
4172 static SV *retrieve_utf8str(stcxt_t *cxt, char *cname)
4176 TRACEME(("retrieve_utf8str"));
4178 sv = retrieve_scalar(cxt, cname);
4180 #ifdef HAS_UTF8_SCALARS
4183 if (cxt->use_bytes < 0)
4185 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4187 if (cxt->use_bytes == 0)
4198 * Like retrieve_lscalar(), but tag result as utf8.
4199 * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
4201 static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname)
4205 TRACEME(("retrieve_lutf8str"));
4207 sv = retrieve_lscalar(cxt, cname);
4209 #ifdef HAS_UTF8_SCALARS
4212 if (cxt->use_bytes < 0)
4214 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4216 if (cxt->use_bytes == 0)
4226 * Retrieve defined integer.
4227 * Layout is SX_INTEGER <data>, whith SX_INTEGER already read.
4229 static SV *retrieve_integer(stcxt_t *cxt, char *cname)
4234 TRACEME(("retrieve_integer (#%d)", cxt->tagnum));
4236 READ(&iv, sizeof(iv));
4238 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4240 TRACEME(("integer %"IVdf, iv));
4241 TRACEME(("ok (retrieve_integer at 0x%"UVxf")", PTR2UV(sv)));
4249 * Retrieve defined integer in network order.
4250 * Layout is SX_NETINT <data>, whith SX_NETINT already read.
4252 static SV *retrieve_netint(stcxt_t *cxt, char *cname)
4257 TRACEME(("retrieve_netint (#%d)", cxt->tagnum));
4261 sv = newSViv((int) ntohl(iv));
4262 TRACEME(("network integer %d", (int) ntohl(iv)));
4265 TRACEME(("network integer (as-is) %d", iv));
4267 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4269 TRACEME(("ok (retrieve_netint at 0x%"UVxf")", PTR2UV(sv)));
4277 * Retrieve defined double.
4278 * Layout is SX_DOUBLE <data>, whith SX_DOUBLE already read.
4280 static SV *retrieve_double(stcxt_t *cxt, char *cname)
4285 TRACEME(("retrieve_double (#%d)", cxt->tagnum));
4287 READ(&nv, sizeof(nv));
4289 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4291 TRACEME(("double %"NVff, nv));
4292 TRACEME(("ok (retrieve_double at 0x%"UVxf")", PTR2UV(sv)));
4300 * Retrieve defined byte (small integer within the [-128, +127] range).
4301 * Layout is SX_BYTE <data>, whith SX_BYTE already read.
4303 static SV *retrieve_byte(stcxt_t *cxt, char *cname)
4307 signed char tmp; /* Workaround for AIX cc bug --H.Merijn Brand */
4309 TRACEME(("retrieve_byte (#%d)", cxt->tagnum));
4312 TRACEME(("small integer read as %d", (unsigned char) siv));
4313 tmp = (unsigned char) siv - 128;
4315 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4317 TRACEME(("byte %d", tmp));
4318 TRACEME(("ok (retrieve_byte at 0x%"UVxf")", PTR2UV(sv)));
4326 * Return the undefined value.
4328 static SV *retrieve_undef(stcxt_t *cxt, char *cname)
4332 TRACEME(("retrieve_undef"));
4343 * Return the immortal undefined value.
4345 static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname)
4347 SV *sv = &PL_sv_undef;
4349 TRACEME(("retrieve_sv_undef"));
4358 * Return the immortal yes value.
4360 static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname)
4362 SV *sv = &PL_sv_yes;
4364 TRACEME(("retrieve_sv_yes"));
4373 * Return the immortal no value.
4375 static SV *retrieve_sv_no(stcxt_t *cxt, char *cname)
4379 TRACEME(("retrieve_sv_no"));
4388 * Retrieve a whole array.
4389 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
4390 * Each item is stored as <object>.
4392 * When we come here, SX_ARRAY has been read already.
4394 static SV *retrieve_array(stcxt_t *cxt, char *cname)
4401 TRACEME(("retrieve_array (#%d)", cxt->tagnum));
4404 * Read length, and allocate array, then pre-extend it.
4408 TRACEME(("size = %d", len));
4410 SEEN(av, cname); /* Will return if array not allocated nicely */
4414 return (SV *) av; /* No data follow if array is empty */
4417 * Now get each item in turn...
4420 for (i = 0; i < len; i++) {
4421 TRACEME(("(#%d) item", i));
4422 sv = retrieve(cxt, 0); /* Retrieve item */
4425 if (av_store(av, i, sv) == 0)
4429 TRACEME(("ok (retrieve_array at 0x%"UVxf")", PTR2UV(av)));
4437 * Retrieve a whole hash table.
4438 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4439 * Keys are stored as <length> <data>, the <data> section being omitted
4441 * Values are stored as <object>.
4443 * When we come here, SX_HASH has been read already.
4445 static SV *retrieve_hash(stcxt_t *cxt, char *cname)
4453 TRACEME(("retrieve_hash (#%d)", cxt->tagnum));
4456 * Read length, allocate table.
4460 TRACEME(("size = %d", len));
4462 SEEN(hv, cname); /* Will return if table not allocated properly */
4464 return (SV *) hv; /* No data follow if table empty */
4465 hv_ksplit(hv, len); /* pre-extend hash to save multiple splits */
4468 * Now get each key/value pair in turn...
4471 for (i = 0; i < len; i++) {
4476 TRACEME(("(#%d) value", i));
4477 sv = retrieve(cxt, 0);
4483 * Since we're reading into kbuf, we must ensure we're not
4484 * recursing between the read and the hv_store() where it's used.
4485 * Hence the key comes after the value.
4488 RLEN(size); /* Get key size */
4489 KBUFCHK((STRLEN)size); /* Grow hash key read pool if needed */
4492 kbuf[size] = '\0'; /* Mark string end, just in case */
4493 TRACEME(("(#%d) key '%s'", i, kbuf));
4496 * Enter key/value pair into hash table.
4499 if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
4503 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4511 * Retrieve a whole hash table.
4512 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4513 * Keys are stored as <length> <data>, the <data> section being omitted
4515 * Values are stored as <object>.
4517 * When we come here, SX_HASH has been read already.
4519 static SV *retrieve_flag_hash(stcxt_t *cxt, char *cname)
4528 GETMARK(hash_flags);
4529 TRACEME(("retrieve_flag_hash (#%d)", cxt->tagnum));
4531 * Read length, allocate table.
4534 #ifndef HAS_RESTRICTED_HASHES
4535 if (hash_flags & SHV_RESTRICTED) {
4536 if (cxt->derestrict < 0)
4538 = (SvTRUE(perl_get_sv("Storable::downgrade_restricted", TRUE))
4540 if (cxt->derestrict == 0)
4541 RESTRICTED_HASH_CROAK();
4546 TRACEME(("size = %d, flags = %d", len, hash_flags));
4548 SEEN(hv, cname); /* Will return if table not allocated properly */
4550 return (SV *) hv; /* No data follow if table empty */
4551 hv_ksplit(hv, len); /* pre-extend hash to save multiple splits */
4554 * Now get each key/value pair in turn...
4557 for (i = 0; i < len; i++) {
4559 int store_flags = 0;
4564 TRACEME(("(#%d) value", i));
4565 sv = retrieve(cxt, 0);
4570 #ifdef HAS_RESTRICTED_HASHES
4571 if ((hash_flags & SHV_RESTRICTED) && (flags & SHV_K_LOCKED))
4575 if (flags & SHV_K_ISSV) {
4576 /* XXX you can't set a placeholder with an SV key.
4577 Then again, you can't get an SV key.
4578 Without messing around beyond what the API is supposed to do.
4581 TRACEME(("(#%d) keysv, flags=%d", i, flags));
4582 keysv = retrieve(cxt, 0);
4586 if (!hv_store_ent(hv, keysv, sv, 0))
4591 * Since we're reading into kbuf, we must ensure we're not
4592 * recursing between the read and the hv_store() where it's used.
4593 * Hence the key comes after the value.
4596 if (flags & SHV_K_PLACEHOLDER) {
4599 store_flags |= HVhek_PLACEHOLD;
4601 if (flags & SHV_K_UTF8) {
4602 #ifdef HAS_UTF8_HASHES
4603 store_flags |= HVhek_UTF8;
4605 if (cxt->use_bytes < 0)
4607 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4609 if (cxt->use_bytes == 0)
4613 #ifdef HAS_UTF8_HASHES
4614 if (flags & SHV_K_WASUTF8)
4615 store_flags |= HVhek_WASUTF8;
4618 RLEN(size); /* Get key size */
4619 KBUFCHK((STRLEN)size); /* Grow hash key read pool if needed */
4622 kbuf[size] = '\0'; /* Mark string end, just in case */
4623 TRACEME(("(#%d) key '%s' flags %X store_flags %X", i, kbuf,
4624 flags, store_flags));
4627 * Enter key/value pair into hash table.
4630 #ifdef HAS_RESTRICTED_HASHES
4631 if (hv_store_flags(hv, kbuf, size, sv, 0, flags) == 0)
4634 if (!(store_flags & HVhek_PLACEHOLD))
4635 if (hv_store(hv, kbuf, size, sv, 0) == 0)
4640 #ifdef HAS_RESTRICTED_HASHES
4641 if (hash_flags & SHV_RESTRICTED)
4645 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4651 * old_retrieve_array
4653 * Retrieve a whole array in pre-0.6 binary format.
4655 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
4656 * Each item is stored as SX_ITEM <object> or SX_IT_UNDEF for "holes".
4658 * When we come here, SX_ARRAY has been read already.
4660 static SV *old_retrieve_array(stcxt_t *cxt, char *cname)
4668 TRACEME(("old_retrieve_array (#%d)", cxt->tagnum));
4671 * Read length, and allocate array, then pre-extend it.
4675 TRACEME(("size = %d", len));
4677 SEEN(av, 0); /* Will return if array not allocated nicely */
4681 return (SV *) av; /* No data follow if array is empty */
4684 * Now get each item in turn...
4687 for (i = 0; i < len; i++) {
4689 if (c == SX_IT_UNDEF) {
4690 TRACEME(("(#%d) undef item", i));
4691 continue; /* av_extend() already filled us with undef */
4694 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
4695 TRACEME(("(#%d) item", i));
4696 sv = retrieve(cxt, 0); /* Retrieve item */
4699 if (av_store(av, i, sv) == 0)
4703 TRACEME(("ok (old_retrieve_array at 0x%"UVxf")", PTR2UV(av)));
4711 * Retrieve a whole hash table in pre-0.6 binary format.
4713 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4714 * Keys are stored as SX_KEY <length> <data>, the <data> section being omitted
4716 * Values are stored as SX_VALUE <object> or SX_VL_UNDEF for "holes".
4718 * When we come here, SX_HASH has been read already.
4720 static SV *old_retrieve_hash(stcxt_t *cxt, char *cname)
4728 static SV *sv_h_undef = (SV *) 0; /* hv_store() bug */
4730 TRACEME(("old_retrieve_hash (#%d)", cxt->tagnum));
4733 * Read length, allocate table.
4737 TRACEME(("size = %d", len));
4739 SEEN(hv, 0); /* Will return if table not allocated properly */
4741 return (SV *) hv; /* No data follow if table empty */
4742 hv_ksplit(hv, len); /* pre-extend hash to save multiple splits */
4745 * Now get each key/value pair in turn...
4748 for (i = 0; i < len; i++) {
4754 if (c == SX_VL_UNDEF) {
4755 TRACEME(("(#%d) undef value", i));
4757 * Due to a bug in hv_store(), it's not possible to pass
4758 * &PL_sv_undef to hv_store() as a value, otherwise the
4759 * associated key will not be creatable any more. -- RAM, 14/01/97
4762 sv_h_undef = newSVsv(&PL_sv_undef);
4763 sv = SvREFCNT_inc(sv_h_undef);
4764 } else if (c == SX_VALUE) {
4765 TRACEME(("(#%d) value", i));
4766 sv = retrieve(cxt, 0);
4770 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
4774 * Since we're reading into kbuf, we must ensure we're not
4775 * recursing between the read and the hv_store() where it's used.
4776 * Hence the key comes after the value.
4781 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
4782 RLEN(size); /* Get key size */
4783 KBUFCHK((STRLEN)size); /* Grow hash key read pool if needed */
4786 kbuf[size] = '\0'; /* Mark string end, just in case */
4787 TRACEME(("(#%d) key '%s'", i, kbuf));
4790 * Enter key/value pair into hash table.
4793 if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
4797 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4803 *** Retrieval engine.
4809 * Make sure the stored data we're trying to retrieve has been produced
4810 * on an ILP compatible system with the same byteorder. It croaks out in
4811 * case an error is detected. [ILP = integer-long-pointer sizes]
4812 * Returns null if error is detected, &PL_sv_undef otherwise.
4814 * Note that there's no byte ordering info emitted when network order was
4815 * used at store time.
4817 static SV *magic_check(stcxt_t *cxt)
4820 char byteorder[256];
4822 int use_network_order;
4824 int version_minor = 0;
4826 TRACEME(("magic_check"));
4829 * The "magic number" is only for files, not when freezing in memory.
4833 STRLEN len = sizeof(magicstr) - 1;
4836 READ(buf, (SSize_t)len); /* Not null-terminated */
4837 buf[len] = '\0'; /* Is now */
4839 if (0 == strcmp(buf, magicstr))
4843 * Try to read more bytes to check for the old magic number, which
4847 old_len = sizeof(old_magicstr) - 1;
4848 READ(&buf[len], (SSize_t)(old_len - len));
4849 buf[old_len] = '\0'; /* Is now null-terminated */
4851 if (strcmp(buf, old_magicstr))
4852 CROAK(("File is not a perl storable"));
4857 * Starting with 0.6, the "use_network_order" byte flag is also used to
4858 * indicate the version number of the binary, and therefore governs the
4859 * setting of sv_retrieve_vtbl. See magic_write().
4862 GETMARK(use_network_order);
4863 version_major = use_network_order >> 1;
4864 cxt->retrieve_vtbl = version_major ? sv_retrieve : sv_old_retrieve;
4866 TRACEME(("magic_check: netorder = 0x%x", use_network_order));
4870 * Starting with 0.7 (binary major 2), a full byte is dedicated to the
4871 * minor version of the protocol. See magic_write().
4874 if (version_major > 1)
4875 GETMARK(version_minor);
4877 cxt->ver_major = version_major;
4878 cxt->ver_minor = version_minor;
4880 TRACEME(("binary image version is %d.%d", version_major, version_minor));
4883 * Inter-operability sanity check: we can't retrieve something stored
4884 * using a format more recent than ours, because we have no way to
4885 * know what has changed, and letting retrieval go would mean a probable
4886 * failure reporting a "corrupted" storable file.
4890 version_major > STORABLE_BIN_MAJOR ||
4891 (version_major == STORABLE_BIN_MAJOR &&
4892 version_minor > STORABLE_BIN_MINOR)
4894 TRACEME(("but I am version is %d.%d", STORABLE_BIN_MAJOR,
4895 STORABLE_BIN_MINOR));
4897 CROAK(("Storable binary image v%d.%d more recent than I am (v%d.%d)",
4898 version_major, version_minor,
4899 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
4903 * If they stored using network order, there's no byte ordering
4904 * information to check.
4907 if ((cxt->netorder = (use_network_order & 0x1))) /* Extra () for -Wall */
4908 return &PL_sv_undef; /* No byte ordering info */
4910 sprintf(byteorder, "%lx", (unsigned long) BYTEORDER);
4912 READ(buf, c); /* Not null-terminated */
4913 buf[c] = '\0'; /* Is now */
4915 TRACEME(("byte order '%s'", buf));
4917 if (strcmp(buf, byteorder))
4918 CROAK(("Byte order is not compatible"));
4920 GETMARK(c); /* sizeof(int) */
4921 if ((int) c != sizeof(int))
4922 CROAK(("Integer size is not compatible"));
4924 GETMARK(c); /* sizeof(long) */
4925 if ((int) c != sizeof(long))
4926 CROAK(("Long integer size is not compatible"));
4928 GETMARK(c); /* sizeof(char *) */
4929 if ((int) c != sizeof(char *))
4930 CROAK(("Pointer integer size is not compatible"));
4932 if (version_major >= 2 && version_minor >= 2) {
4933 GETMARK(c); /* sizeof(NV) */
4934 if ((int) c != sizeof(NV))
4935 CROAK(("Double size is not compatible"));
4938 return &PL_sv_undef; /* OK */
4944 * Recursively retrieve objects from the specified file and return their
4945 * root SV (which may be an AV or an HV for what we care).
4946 * Returns null if there is a problem.
4948 static SV *retrieve(stcxt_t *cxt, char *cname)
4954 TRACEME(("retrieve"));
4957 * Grab address tag which identifies the object if we are retrieving
4958 * an older format. Since the new binary format counts objects and no
4959 * longer explicitely tags them, we must keep track of the correspondance
4962 * The following section will disappear one day when the old format is
4963 * no longer supported, hence the final "goto" in the "if" block.
4966 if (cxt->hseen) { /* Retrieving old binary */
4968 if (cxt->netorder) {
4970 READ(&nettag, sizeof(I32)); /* Ordered sequence of I32 */
4971 tag = (stag_t) nettag;
4973 READ(&tag, sizeof(stag_t)); /* Original address of the SV */
4976 if (type == SX_OBJECT) {
4978 svh = hv_fetch(cxt->hseen, (char *) &tag, sizeof(tag), FALSE);
4980 CROAK(("Old tag 0x%"UVxf" should have been mapped already",
4982 tagn = SvIV(*svh); /* Mapped tag number computed earlier below */
4985 * The following code is common with the SX_OBJECT case below.
4988 svh = av_fetch(cxt->aseen, tagn, FALSE);
4990 CROAK(("Object #%"IVdf" should have been retrieved already",
4993 TRACEME(("has retrieved #%d at 0x%"UVxf, tagn, PTR2UV(sv)));
4994 SvREFCNT_inc(sv); /* One more reference to this same sv */
4995 return sv; /* The SV pointer where object was retrieved */
4999 * Map new object, but don't increase tagnum. This will be done
5000 * by each of the retrieve_* functions when they call SEEN().
5002 * The mapping associates the "tag" initially present with a unique
5003 * tag number. See test for SX_OBJECT above to see how this is perused.
5006 if (!hv_store(cxt->hseen, (char *) &tag, sizeof(tag),
5007 newSViv(cxt->tagnum), 0))
5014 * Regular post-0.6 binary format.
5019 TRACEME(("retrieve type = %d", type));
5022 * Are we dealing with an object we should have already retrieved?
5025 if (type == SX_OBJECT) {
5029 svh = av_fetch(cxt->aseen, tag, FALSE);
5031 CROAK(("Object #%"IVdf" should have been retrieved already",
5034 TRACEME(("had retrieved #%d at 0x%"UVxf, tag, PTR2UV(sv)));
5035 SvREFCNT_inc(sv); /* One more reference to this same sv */
5036 return sv; /* The SV pointer where object was retrieved */
5039 first_time: /* Will disappear when support for old format is dropped */
5042 * Okay, first time through for this one.
5045 sv = RETRIEVE(cxt, type)(cxt, cname);
5047 return (SV *) 0; /* Failed */
5050 * Old binary formats (pre-0.7).
5052 * Final notifications, ended by SX_STORED may now follow.
5053 * Currently, the only pertinent notification to apply on the
5054 * freshly retrieved object is either:
5055 * SX_CLASS <char-len> <classname> for short classnames.
5056 * SX_LG_CLASS <int-len> <classname> for larger one (rare!).
5057 * Class name is then read into the key buffer pool used by
5058 * hash table key retrieval.
5061 if (cxt->ver_major < 2) {
5062 while ((type = GETCHAR()) != SX_STORED) {
5066 GETMARK(len); /* Length coded on a single char */
5068 case SX_LG_CLASS: /* Length coded on a regular integer */
5073 return (SV *) 0; /* Failed */
5075 KBUFCHK((STRLEN)len); /* Grow buffer as necessary */
5078 kbuf[len] = '\0'; /* Mark string end */
5083 TRACEME(("ok (retrieved 0x%"UVxf", refcnt=%d, %s)", PTR2UV(sv),
5084 SvREFCNT(sv) - 1, sv_reftype(sv, FALSE)));
5092 * Retrieve data held in file and return the root object.
5093 * Common routine for pretrieve and mretrieve.
5095 static SV *do_retrieve(
5102 int is_tainted; /* Is input source tainted? */
5103 int pre_06_fmt = 0; /* True with pre Storable 0.6 formats */
5105 TRACEME(("do_retrieve (optype = 0x%x)", optype));
5107 optype |= ST_RETRIEVE;
5110 * Sanity assertions for retrieve dispatch tables.
5113 ASSERT(sizeof(sv_old_retrieve) == sizeof(sv_retrieve),
5114 ("old and new retrieve dispatch table have same size"));
5115 ASSERT(sv_old_retrieve[SX_ERROR] == retrieve_other,
5116 ("SX_ERROR entry correctly initialized in old dispatch table"));
5117 ASSERT(sv_retrieve[SX_ERROR] == retrieve_other,
5118 ("SX_ERROR entry correctly initialized in new dispatch table"));
5121 * Workaround for CROAK leak: if they enter with a "dirty" context,
5122 * free up memory for them now.
5129 * Now that STORABLE_xxx hooks exist, it is possible that they try to
5130 * re-enter retrieve() via the hooks.
5134 cxt = allocate_context(cxt);
5138 ASSERT(cxt->entry == 1, ("starting new recursion"));
5139 ASSERT(!cxt->s_dirty, ("clean context"));
5144 * Data is loaded into the memory buffer when f is NULL, unless `in' is
5145 * also NULL, in which case we're expecting the data to already lie
5146 * in the buffer (dclone case).
5149 KBUFINIT(); /* Allocate hash key reading pool once */
5152 MBUF_SAVE_AND_LOAD(in);
5155 * Magic number verifications.
5157 * This needs to be done before calling init_retrieve_context()
5158 * since the format indication in the file are necessary to conduct
5159 * some of the initializations.
5162 cxt->fio = f; /* Where I/O are performed */
5164 if (!magic_check(cxt))
5165 CROAK(("Magic number checking on storable %s failed",
5166 cxt->fio ? "file" : "string"));
5168 TRACEME(("data stored in %s format",
5169 cxt->netorder ? "net order" : "native"));
5172 * Check whether input source is tainted, so that we don't wrongly
5173 * taint perfectly good values...
5175 * We assume file input is always tainted. If both `f' and `in' are
5176 * NULL, then we come from dclone, and tainted is already filled in
5177 * the context. That's a kludge, but the whole dclone() thing is
5178 * already quite a kludge anyway! -- RAM, 15/09/2000.
5181 is_tainted = f ? 1 : (in ? SvTAINTED(in) : cxt->s_tainted);
5182 TRACEME(("input source is %s", is_tainted ? "tainted" : "trusted"));
5183 init_retrieve_context(cxt, optype, is_tainted);
5185 ASSERT(is_retrieving(), ("within retrieve operation"));
5187 sv = retrieve(cxt, 0); /* Recursively retrieve object, get root SV */
5196 pre_06_fmt = cxt->hseen != NULL; /* Before we clean context */
5199 * The "root" context is never freed.
5202 clean_retrieve_context(cxt);
5203 if (cxt->prev) /* This context was stacked */
5204 free_context(cxt); /* It was not the "root" context */
5207 * Prepare returned value.
5211 TRACEME(("retrieve ERROR"));
5212 return &PL_sv_undef; /* Something went wrong, return undef */
5215 TRACEME(("retrieve got %s(0x%"UVxf")",
5216 sv_reftype(sv, FALSE), PTR2UV(sv)));
5219 * Backward compatibility with Storable-0.5@9 (which we know we
5220 * are retrieving if hseen is non-null): don't create an extra RV
5221 * for objects since we special-cased it at store time.
5223 * Build a reference to the SV returned by pretrieve even if it is
5224 * already one and not a scalar, for consistency reasons.
5227 if (pre_06_fmt) { /* Was not handling overloading by then */
5229 TRACEME(("fixing for old formats -- pre 0.6"));
5230 if (sv_type(sv) == svis_REF && (rv = SvRV(sv)) && SvOBJECT(rv)) {
5231 TRACEME(("ended do_retrieve() with an object -- pre 0.6"));
5237 * If reference is overloaded, restore behaviour.
5239 * NB: minor glitch here: normally, overloaded refs are stored specially
5240 * so that we can croak when behaviour cannot be re-installed, and also
5241 * avoid testing for overloading magic at each reference retrieval.
5243 * Unfortunately, the root reference is implicitely stored, so we must
5244 * check for possible overloading now. Furthermore, if we don't restore
5245 * overloading, we cannot croak as if the original ref was, because we
5246 * have no way to determine whether it was an overloaded ref or not in
5249 * It's a pity that overloading magic is attached to the rv, and not to
5250 * the underlying sv as blessing is.
5254 HV *stash = (HV *) SvSTASH(sv);
5255 SV *rv = newRV_noinc(sv);
5256 if (stash && Gv_AMG(stash)) {
5258 TRACEME(("restored overloading on root reference"));
5260 TRACEME(("ended do_retrieve() with an object"));
5264 TRACEME(("regular do_retrieve() end"));
5266 return newRV_noinc(sv);
5272 * Retrieve data held in file and return the root object, undef on error.
5274 SV *pretrieve(PerlIO *f)
5276 TRACEME(("pretrieve"));
5277 return do_retrieve(f, Nullsv, 0);
5283 * Retrieve data held in scalar and return the root object, undef on error.
5285 SV *mretrieve(SV *sv)
5287 TRACEME(("mretrieve"));
5288 return do_retrieve((PerlIO*) 0, sv, 0);
5298 * Deep clone: returns a fresh copy of the original referenced SV tree.
5300 * This is achieved by storing the object in memory and restoring from
5301 * there. Not that efficient, but it should be faster than doing it from
5308 stcxt_t *real_context;
5311 TRACEME(("dclone"));
5314 * Workaround for CROAK leak: if they enter with a "dirty" context,
5315 * free up memory for them now.
5322 * do_store() optimizes for dclone by not freeing its context, should
5323 * we need to allocate one because we're deep cloning from a hook.
5326 if (!do_store((PerlIO*) 0, sv, ST_CLONE, FALSE, (SV**) 0))
5327 return &PL_sv_undef; /* Error during store */
5330 * Because of the above optimization, we have to refresh the context,
5331 * since a new one could have been allocated and stacked by do_store().
5334 { dSTCXT; real_context = cxt; } /* Sub-block needed for macro */
5335 cxt = real_context; /* And we need this temporary... */
5338 * Now, `cxt' may refer to a new context.
5341 ASSERT(!cxt->s_dirty, ("clean context"));
5342 ASSERT(!cxt->entry, ("entry will not cause new context allocation"));
5345 TRACEME(("dclone stored %d bytes", size));
5349 * Since we're passing do_retrieve() both a NULL file and sv, we need
5350 * to pre-compute the taintedness of the input by setting cxt->tainted
5351 * to whatever state our own input string was. -- RAM, 15/09/2000
5353 * do_retrieve() will free non-root context.
5356 cxt->s_tainted = SvTAINTED(sv);
5357 out = do_retrieve((PerlIO*) 0, Nullsv, ST_CLONE);
5359 TRACEME(("dclone returns 0x%"UVxf, PTR2UV(out)));
5369 * The Perl IO GV object distinguishes between input and output for sockets
5370 * but not for plain files. To allow Storable to transparently work on
5371 * plain files and sockets transparently, we have to ask xsubpp to fetch the
5372 * right object for us. Hence the OutputStream and InputStream declarations.
5374 * Before perl 5.004_05, those entries in the standard typemap are not
5375 * defined in perl include files, so we do that here.
5378 #ifndef OutputStream
5379 #define OutputStream PerlIO *
5380 #define InputStream PerlIO *
5381 #endif /* !OutputStream */
5383 MODULE = Storable PACKAGE = Storable
5421 last_op_in_netorder()