2 * Store and retrieve mechanism.
6 * $Id: Storable.xs,v 1.0.1.8 2001/03/15 00:20:55 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 1.0.1.8 2001/03/15 00:20:55 ram
15 * patch11: last version was wrongly compiling with assertions on
17 * Revision 1.0.1.7 2001/02/17 12:25:26 ram
18 * patch8: now bless objects ASAP at retrieve time
19 * patch8: added support for blessed ref to tied structures
21 * Revision 1.0.1.6 2001/01/03 09:40:40 ram
22 * patch7: prototype and casting cleanup
23 * patch7: trace offending package when overloading cannot be restored
24 * patch7: made context cleanup safer to avoid dup freeing
26 * Revision 1.0.1.5 2000/11/05 17:21:24 ram
27 * patch6: fixed severe "object lost" bug for STORABLE_freeze returns
29 * Revision 1.0.1.4 2000/10/26 17:11:04 ram
30 * patch5: auto requires module of blessed ref when STORABLE_thaw misses
32 * Revision 1.0.1.3 2000/09/29 19:49:57 ram
33 * patch3: avoid using "tainted" and "dirty" since Perl remaps them via cpp
35 * Revision 1.0.1.2 2000/09/28 21:43:10 ram
36 * patch2: perls before 5.004_04 lack newSVpvn
38 * Revision 1.0.1.1 2000/09/17 16:47:49 ram
39 * patch1: now only taint retrieved data when source was tainted
40 * patch1: added support for UTF-8 strings
41 * patch1: fixed store hook bug: was allocating class id too soon
43 * Revision 1.0 2000/09/01 19:40:41 ram
44 * Baseline for first official release.
50 #include <patchlevel.h> /* Perl's one, needed since 5.6 */
54 #define DEBUGME /* Debug mode, turns assertions on as well */
55 #define DASSERT /* Assertion mode */
59 * Pre PerlIO time when none of USE_PERLIO and PERLIO_IS_STDIO is defined
60 * Provide them with the necessary defines so they can build with pre-5.004.
63 #ifndef PERLIO_IS_STDIO
65 #define PerlIO_getc(x) getc(x)
66 #define PerlIO_putc(f,x) putc(x,f)
67 #define PerlIO_read(x,y,z) fread(y,1,z,x)
68 #define PerlIO_write(x,y,z) fwrite(y,1,z,x)
69 #define PerlIO_stdoutf printf
70 #endif /* PERLIO_IS_STDIO */
71 #endif /* USE_PERLIO */
74 * Earlier versions of perl might be used, we can't assume they have the latest!
77 #ifndef PERL_VERSION /* For perls < 5.6 */
78 #define PERL_VERSION PATCHLEVEL
80 #define newRV_noinc(sv) ((Sv = newRV(sv)), --SvREFCNT(SvRV(Sv)), Sv)
82 #if (PATCHLEVEL <= 4) /* Older perls (<= 5.004) lack PL_ namespace */
83 #define PL_sv_yes sv_yes
84 #define PL_sv_no sv_no
85 #define PL_sv_undef sv_undef
86 #if (SUBVERSION <= 4) /* 5.004_04 has been reported to lack newSVpvn */
87 #define newSVpvn newSVpv
89 #endif /* PATCHLEVEL <= 4 */
90 #ifndef HvSHAREKEYS_off
91 #define HvSHAREKEYS_off(hv) /* Ignore */
93 #ifndef AvFILLp /* Older perls (<=5.003) lack AvFILLp */
94 #define AvFILLp AvFILL
96 typedef double NV; /* Older perls lack the NV type */
97 #define IVdf "ld" /* Various printf formats for Perl types */
101 #define INT2PTR(t,v) (t)(IV)(v)
102 #define PTR2UV(v) (unsigned long)(v)
103 #endif /* PERL_VERSION -- perls < 5.6 */
105 #ifndef NVef /* The following were not part of perl 5.6 */
106 #if defined(USE_LONG_DOUBLE) && \
107 defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
108 #define NVef PERL_PRIeldbl
109 #define NVff PERL_PRIfldbl
110 #define NVgf PERL_PRIgldbl
125 * TRACEME() will only output things when the $Storable::DEBUGME is true.
128 #define TRACEME(x) do { \
129 if (SvTRUE(perl_get_sv("Storable::DEBUGME", TRUE))) \
130 { PerlIO_stdoutf x; PerlIO_stdoutf("\n"); } \
137 #define ASSERT(x,y) do { \
139 PerlIO_stdoutf("ASSERT FAILED (\"%s\", line %d): ", \
140 __FILE__, __LINE__); \
141 PerlIO_stdoutf y; PerlIO_stdoutf("\n"); \
152 #define C(x) ((char) (x)) /* For markers with dynamic retrieval handling */
154 #define SX_OBJECT C(0) /* Already stored object */
155 #define SX_LSCALAR C(1) /* Scalar (large binary) follows (length, data) */
156 #define SX_ARRAY C(2) /* Array forthcominng (size, item list) */
157 #define SX_HASH C(3) /* Hash forthcoming (size, key/value pair list) */
158 #define SX_REF C(4) /* Reference to object forthcoming */
159 #define SX_UNDEF C(5) /* Undefined scalar */
160 #define SX_INTEGER C(6) /* Integer forthcoming */
161 #define SX_DOUBLE C(7) /* Double forthcoming */
162 #define SX_BYTE C(8) /* (signed) byte forthcoming */
163 #define SX_NETINT C(9) /* Integer in network order forthcoming */
164 #define SX_SCALAR C(10) /* Scalar (binary, small) follows (length, data) */
165 #define SX_TIED_ARRAY C(11) /* Tied array forthcoming */
166 #define SX_TIED_HASH C(12) /* Tied hash forthcoming */
167 #define SX_TIED_SCALAR C(13) /* Tied scalar forthcoming */
168 #define SX_SV_UNDEF C(14) /* Perl's immortal PL_sv_undef */
169 #define SX_SV_YES C(15) /* Perl's immortal PL_sv_yes */
170 #define SX_SV_NO C(16) /* Perl's immortal PL_sv_no */
171 #define SX_BLESS C(17) /* Object is blessed */
172 #define SX_IX_BLESS C(18) /* Object is blessed, classname given by index */
173 #define SX_HOOK C(19) /* Stored via hook, user-defined */
174 #define SX_OVERLOAD C(20) /* Overloaded reference */
175 #define SX_TIED_KEY C(21) /* Tied magic key forthcoming */
176 #define SX_TIED_IDX C(22) /* Tied magic index forthcoming */
177 #define SX_UTF8STR C(23) /* UTF-8 string forthcoming (small) */
178 #define SX_LUTF8STR C(24) /* UTF-8 string forthcoming (large) */
179 #define SX_ERROR C(25) /* Error */
182 * Those are only used to retrieve "old" pre-0.6 binary images.
184 #define SX_ITEM 'i' /* An array item introducer */
185 #define SX_IT_UNDEF 'I' /* Undefined array item */
186 #define SX_KEY 'k' /* An hash key introducer */
187 #define SX_VALUE 'v' /* An hash value introducer */
188 #define SX_VL_UNDEF 'V' /* Undefined hash value */
191 * Those are only used to retrieve "old" pre-0.7 binary images
194 #define SX_CLASS 'b' /* Object is blessed, class name length <255 */
195 #define SX_LG_CLASS 'B' /* Object is blessed, class name length >255 */
196 #define SX_STORED 'X' /* End of object */
199 * Limits between short/long length representation.
202 #define LG_SCALAR 255 /* Large scalar length limit */
203 #define LG_BLESS 127 /* Large classname bless limit */
209 #define ST_STORE 0x1 /* Store operation */
210 #define ST_RETRIEVE 0x2 /* Retrieval operation */
211 #define ST_CLONE 0x4 /* Deep cloning operation */
214 * The following structure is used for hash table key retrieval. Since, when
215 * retrieving objects, we'll be facing blessed hash references, it's best
216 * to pre-allocate that buffer once and resize it as the need arises, never
217 * freeing it (keys will be saved away someplace else anyway, so even large
218 * keys are not enough a motivation to reclaim that space).
220 * This structure is also used for memory store/retrieve operations which
221 * happen in a fixed place before being malloc'ed elsewhere if persistency
222 * is required. Hence the aptr pointer.
225 char *arena; /* Will hold hash key strings, resized as needed */
226 STRLEN asiz; /* Size of aforementionned buffer */
227 char *aptr; /* Arena pointer, for in-place read/write ops */
228 char *aend; /* First invalid address */
233 * An hash table records the objects which have already been stored.
234 * Those are referred to as SX_OBJECT in the file, and their "tag" (i.e.
235 * an arbitrary sequence number) is used to identify them.
238 * An array table records the objects which have already been retrieved,
239 * as seen by the tag determind by counting the objects themselves. The
240 * reference to that retrieved object is kept in the table, and is returned
241 * when an SX_OBJECT is found bearing that same tag.
243 * The same processing is used to record "classname" for blessed objects:
244 * indexing by a hash at store time, and via an array at retrieve time.
247 typedef unsigned long stag_t; /* Used by pre-0.6 binary format */
250 * The following "thread-safe" related defines were contributed by
251 * Murray Nesbitt <murray@activestate.com> and integrated by RAM, who
252 * only renamed things a little bit to ensure consistency with surrounding
253 * code. -- RAM, 14/09/1999
255 * The original patch suffered from the fact that the stcxt_t structure
256 * was global. Murray tried to minimize the impact on the code as much as
259 * Starting with 0.7, Storable can be re-entrant, via the STORABLE_xxx hooks
260 * on objects. Therefore, the notion of context needs to be generalized,
264 #define MY_VERSION "Storable(" XS_VERSION ")"
267 * Fields s_tainted and s_dirty are prefixed with s_ because Perl's include
268 * files remap tainted and dirty when threading is enabled. That's bad for
269 * perl to remap such common words. -- RAM, 29/09/00
272 typedef struct stcxt {
273 int entry; /* flags recursion */
274 int optype; /* type of traversal operation */
275 HV *hseen; /* which objects have been seen, store time */
276 AV *hook_seen; /* which SVs were returned by STORABLE_freeze() */
277 AV *aseen; /* which objects have been seen, retrieve time */
278 HV *hclass; /* which classnames have been seen, store time */
279 AV *aclass; /* which classnames have been seen, retrieve time */
280 HV *hook; /* cache for hook methods per class name */
281 I32 tagnum; /* incremented at store time for each seen object */
282 I32 classnum; /* incremented at store time for each seen classname */
283 int netorder; /* true if network order used */
284 int s_tainted; /* true if input source is tainted, at retrieve time */
285 int forgive_me; /* whether to be forgiving... */
286 int canonical; /* whether to store hashes sorted by key */
287 int s_dirty; /* context is dirty due to CROAK() -- can be cleaned */
288 struct extendable keybuf; /* for hash key retrieval */
289 struct extendable membuf; /* for memory store/retrieve operations */
290 PerlIO *fio; /* where I/O are performed, NULL for memory */
291 int ver_major; /* major of version for retrieved object */
292 int ver_minor; /* minor of version for retrieved object */
293 SV *(**retrieve_vtbl)(); /* retrieve dispatch table */
294 struct stcxt *prev; /* contexts chained backwards in real recursion */
297 #if defined(MULTIPLICITY) || defined(PERL_OBJECT) || defined(PERL_CAPI)
299 #if (PATCHLEVEL <= 4) && (SUBVERSION < 68)
301 SV *perinterp_sv = perl_get_sv(MY_VERSION, FALSE)
302 #else /* >= perl5.004_68 */
304 SV *perinterp_sv = *hv_fetch(PL_modglobal, \
305 MY_VERSION, sizeof(MY_VERSION)-1, TRUE)
306 #endif /* < perl5.004_68 */
308 #define dSTCXT_PTR(T,name) \
309 T name = (perinterp_sv && SvIOK(perinterp_sv) \
310 ? INT2PTR(T, SvIVX(perinterp_sv)) : (T) 0)
313 dSTCXT_PTR(stcxt_t *, cxt)
317 Newz(0, cxt, 1, stcxt_t); \
318 sv_setiv(perinterp_sv, PTR2IV(cxt))
320 #define SET_STCXT(x) do { \
322 sv_setiv(perinterp_sv, PTR2IV(x)); \
325 #else /* !MULTIPLICITY && !PERL_OBJECT && !PERL_CAPI */
327 static stcxt_t Context;
328 static stcxt_t *Context_ptr = &Context;
329 #define dSTCXT stcxt_t *cxt = Context_ptr
330 #define INIT_STCXT dSTCXT
331 #define SET_STCXT(x) Context_ptr = x
333 #endif /* MULTIPLICITY || PERL_OBJECT || PERL_CAPI */
337 * Croaking implies a memory leak, since we don't use setjmp/longjmp
338 * to catch the exit and free memory used during store or retrieve
339 * operations. This is not too difficult to fix, but I need to understand
340 * how Perl does it, and croaking is exceptional anyway, so I lack the
341 * motivation to do it.
343 * The current workaround is to mark the context as dirty when croaking,
344 * so that data structures can be freed whenever we renter Storable code
345 * (but only *then*: it's a workaround, not a fix).
347 * This is also imperfect, because we don't really know how far they trapped
348 * the croak(), and when we were recursing, we won't be able to clean anything
349 * but the topmost context stacked.
352 #define CROAK(x) do { cxt->s_dirty = 1; croak x; } while (0)
355 * End of "thread-safe" related definitions.
361 * Keep only the low 32 bits of a pointer (used for tags, which are not
366 #define LOW_32BITS(x) ((I32) (x))
368 #define LOW_32BITS(x) ((I32) ((unsigned long) (x) & 0xffffffffUL))
374 * Hack for Crays, where sizeof(I32) == 8, and which are big-endians.
375 * Used in the WLEN and RLEN macros.
379 #define oI(x) ((I32 *) ((char *) (x) + 4))
380 #define oS(x) ((x) - 4)
381 #define oC(x) (x = 0)
390 * key buffer handling
392 #define kbuf (cxt->keybuf).arena
393 #define ksiz (cxt->keybuf).asiz
394 #define KBUFINIT() do { \
396 TRACEME(("** allocating kbuf of 128 bytes")); \
397 New(10003, kbuf, 128, char); \
401 #define KBUFCHK(x) do { \
403 TRACEME(("** extending kbuf to %d bytes", x+1)); \
404 Renew(kbuf, x+1, char); \
410 * memory buffer handling
412 #define mbase (cxt->membuf).arena
413 #define msiz (cxt->membuf).asiz
414 #define mptr (cxt->membuf).aptr
415 #define mend (cxt->membuf).aend
417 #define MGROW (1 << 13)
418 #define MMASK (MGROW - 1)
420 #define round_mgrow(x) \
421 ((unsigned long) (((unsigned long) (x) + MMASK) & ~MMASK))
422 #define trunc_int(x) \
423 ((unsigned long) ((unsigned long) (x) & ~(sizeof(int)-1)))
424 #define int_aligned(x) \
425 ((unsigned long) (x) == trunc_int(x))
427 #define MBUF_INIT(x) do { \
429 TRACEME(("** allocating mbase of %d bytes", MGROW)); \
430 New(10003, mbase, MGROW, char); \
437 mend = mbase + msiz; \
440 #define MBUF_TRUNC(x) mptr = mbase + x
441 #define MBUF_SIZE() (mptr - mbase)
444 * Use SvPOKp(), because SvPOK() fails on tainted scalars.
445 * See store_scalar() for other usage of this workaround.
447 #define MBUF_LOAD(v) do { \
449 CROAK(("Not a scalar string")); \
450 mptr = mbase = SvPV(v, msiz); \
451 mend = mbase + msiz; \
454 #define MBUF_XTEND(x) do { \
455 int nsz = (int) round_mgrow((x)+msiz); \
456 int offset = mptr - mbase; \
457 TRACEME(("** extending mbase to %d bytes", nsz)); \
458 Renew(mbase, nsz, char); \
460 mptr = mbase + offset; \
461 mend = mbase + nsz; \
464 #define MBUF_CHK(x) do { \
465 if ((mptr + (x)) > mend) \
469 #define MBUF_GETC(x) do { \
471 x = (int) (unsigned char) *mptr++; \
477 #define MBUF_GETINT(x) do { \
479 if ((mptr + 4) <= mend) { \
480 memcpy(oI(&x), mptr, 4); \
486 #define MBUF_GETINT(x) do { \
487 if ((mptr + sizeof(int)) <= mend) { \
488 if (int_aligned(mptr)) \
491 memcpy(&x, mptr, sizeof(int)); \
492 mptr += sizeof(int); \
498 #define MBUF_READ(x,s) do { \
499 if ((mptr + (s)) <= mend) { \
500 memcpy(x, mptr, s); \
506 #define MBUF_SAFEREAD(x,s,z) do { \
507 if ((mptr + (s)) <= mend) { \
508 memcpy(x, mptr, s); \
516 #define MBUF_PUTC(c) do { \
518 *mptr++ = (char) c; \
521 *mptr++ = (char) c; \
526 #define MBUF_PUTINT(i) do { \
528 memcpy(mptr, oI(&i), 4); \
532 #define MBUF_PUTINT(i) do { \
533 MBUF_CHK(sizeof(int)); \
534 if (int_aligned(mptr)) \
537 memcpy(mptr, &i, sizeof(int)); \
538 mptr += sizeof(int); \
542 #define MBUF_WRITE(x,s) do { \
544 memcpy(mptr, x, s); \
549 * Possible return values for sv_type().
553 #define svis_SCALAR 1
557 #define svis_TIED_ITEM 5
564 #define SHF_TYPE_MASK 0x03
565 #define SHF_LARGE_CLASSLEN 0x04
566 #define SHF_LARGE_STRLEN 0x08
567 #define SHF_LARGE_LISTLEN 0x10
568 #define SHF_IDX_CLASSNAME 0x20
569 #define SHF_NEED_RECURSE 0x40
570 #define SHF_HAS_LIST 0x80
573 * Types for SX_HOOK (last 2 bits in flags).
579 #define SHT_EXTRA 3 /* Read extra byte for type */
582 * The following are held in the "extra byte"...
585 #define SHT_TSCALAR 4 /* 4 + 0 -- tied scalar */
586 #define SHT_TARRAY 5 /* 4 + 1 -- tied array */
587 #define SHT_THASH 6 /* 4 + 2 -- tied hash */
590 * Before 0.6, the magic string was "perl-store" (binary version number 0).
592 * Since 0.6 introduced many binary incompatibilities, the magic string has
593 * been changed to "pst0" to allow an old image to be properly retrieved by
594 * a newer Storable, but ensure a newer image cannot be retrieved with an
597 * At 0.7, objects are given the ability to serialize themselves, and the
598 * set of markers is extended, backward compatibility is not jeopardized,
599 * so the binary version number could have remained unchanged. To correctly
600 * spot errors if a file making use of 0.7-specific extensions is given to
601 * 0.6 for retrieval, the binary version was moved to "2". And I'm introducing
602 * a "minor" version, to better track this kind of evolution from now on.
605 static char old_magicstr[] = "perl-store"; /* Magic number before 0.6 */
606 static char magicstr[] = "pst0"; /* Used as a magic number */
608 #define STORABLE_BIN_MAJOR 2 /* Binary major "version" */
609 #define STORABLE_BIN_MINOR 4 /* Binary minor "version" */
612 * Useful store shortcuts...
615 #define PUTMARK(x) do { \
618 else if (PerlIO_putc(cxt->fio, x) == EOF) \
622 #define WRITE_I32(x) do { \
623 ASSERT(sizeof(x) == sizeof(I32), ("writing an I32")); \
626 else if (PerlIO_write(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
631 #define WLEN(x) do { \
632 if (cxt->netorder) { \
633 int y = (int) htonl(x); \
636 else if (PerlIO_write(cxt->fio,oI(&y),oS(sizeof(y))) != oS(sizeof(y))) \
641 else if (PerlIO_write(cxt->fio,oI(&x),oS(sizeof(x))) != oS(sizeof(x))) \
646 #define WLEN(x) WRITE_I32(x)
649 #define WRITE(x,y) do { \
652 else if (PerlIO_write(cxt->fio, x, y) != y) \
656 #define STORE_PV_LEN(pv, len, small, large) do { \
657 if (len <= LG_SCALAR) { \
658 unsigned char clen = (unsigned char) len; \
670 #define STORE_SCALAR(pv, len) STORE_PV_LEN(pv, len, SX_SCALAR, SX_LSCALAR)
673 * Conditional UTF8 support.
674 * On non-UTF8 perls, UTF8 strings are returned as normal strings.
678 #define STORE_UTF8STR(pv, len) STORE_PV_LEN(pv, len, SX_UTF8STR, SX_LUTF8STR)
681 #define STORE_UTF8STR(pv, len) CROAK(("panic: storing UTF8 in non-UTF8 perl"))
682 #define SvUTF8_on(sv) CROAK(("Cannot retrieve UTF8 data in non-UTF8 perl"))
686 * Store undef in arrays and hashes without recursing through store().
688 #define STORE_UNDEF() do { \
694 * Useful retrieve shortcuts...
698 (cxt->fio ? PerlIO_getc(cxt->fio) : (mptr >= mend ? EOF : (int) *mptr++))
700 #define GETMARK(x) do { \
703 else if ((int) (x = PerlIO_getc(cxt->fio)) == EOF) \
707 #define READ_I32(x) do { \
708 ASSERT(sizeof(x) == sizeof(I32), ("reading an I32")); \
712 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
717 #define RLEN(x) do { \
721 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
724 x = (int) ntohl(x); \
727 #define RLEN(x) READ_I32(x)
730 #define READ(x,y) do { \
733 else if (PerlIO_read(cxt->fio, x, y) != y) \
737 #define SAFEREAD(x,y,z) do { \
739 MBUF_SAFEREAD(x,y,z); \
740 else if (PerlIO_read(cxt->fio, x, y) != y) { \
747 * This macro is used at retrieve time, to remember where object 'y', bearing a
748 * given tag 'tagnum', has been retrieved. Next time we see an SX_OBJECT marker,
749 * we'll therefore know where it has been retrieved and will be able to
750 * share the same reference, as in the original stored memory image.
752 * We also need to bless objects ASAP for hooks (which may compute "ref $x"
753 * on the objects given to STORABLE_thaw and expect that to be defined), and
754 * also for overloaded objects (for which we might not find the stash if the
755 * object is not blessed yet--this might occur for overloaded objects that
756 * refer to themselves indirectly: if we blessed upon return from a sub
757 * retrieve(), the SX_OBJECT marker we'd found could not have overloading
758 * restored on it because the underlying object would not be blessed yet!).
760 * To achieve that, the class name of the last retrieved object is passed down
761 * recursively, and the first SEEN() call for which the class name is not NULL
762 * will bless the object.
764 #define SEEN(y,c) do { \
767 if (av_store(cxt->aseen, cxt->tagnum++, SvREFCNT_inc(y)) == 0) \
769 TRACEME(("aseen(#%d) = 0x%"UVxf" (refcnt=%d)", cxt->tagnum-1, \
770 PTR2UV(y), SvREFCNT(y)-1)); \
772 BLESS((SV *) (y), c); \
776 * Bless `s' in `p', via a temporary reference, required by sv_bless().
778 #define BLESS(s,p) do { \
781 TRACEME(("blessing 0x%"UVxf" in %s", PTR2UV(s), (p))); \
782 stash = gv_stashpv((p), TRUE); \
783 ref = newRV_noinc(s); \
784 (void) sv_bless(ref, stash); \
790 static SV *retrieve(stcxt_t *cxt, char *cname);
793 * Dynamic dispatching table for SV store.
796 static int store_ref(stcxt_t *cxt, SV *sv);
797 static int store_scalar(stcxt_t *cxt, SV *sv);
798 static int store_array(stcxt_t *cxt, AV *av);
799 static int store_hash(stcxt_t *cxt, HV *hv);
800 static int store_tied(stcxt_t *cxt, SV *sv);
801 static int store_tied_item(stcxt_t *cxt, SV *sv);
802 static int store_other(stcxt_t *cxt, SV *sv);
803 static int store_blessed(stcxt_t *cxt, SV *sv, int type, HV *pkg);
805 static int (*sv_store[])(stcxt_t *cxt, SV *sv) = {
806 store_ref, /* svis_REF */
807 store_scalar, /* svis_SCALAR */
808 (int (*)(stcxt_t *cxt, SV *sv)) store_array, /* svis_ARRAY */
809 (int (*)(stcxt_t *cxt, SV *sv)) store_hash, /* svis_HASH */
810 store_tied, /* svis_TIED */
811 store_tied_item, /* svis_TIED_ITEM */
812 store_other, /* svis_OTHER */
815 #define SV_STORE(x) (*sv_store[x])
818 * Dynamic dispatching tables for SV retrieval.
821 static SV *retrieve_lscalar(stcxt_t *cxt, char *cname);
822 static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname);
823 static SV *old_retrieve_array(stcxt_t *cxt, char *cname);
824 static SV *old_retrieve_hash(stcxt_t *cxt, char *cname);
825 static SV *retrieve_ref(stcxt_t *cxt, char *cname);
826 static SV *retrieve_undef(stcxt_t *cxt, char *cname);
827 static SV *retrieve_integer(stcxt_t *cxt, char *cname);
828 static SV *retrieve_double(stcxt_t *cxt, char *cname);
829 static SV *retrieve_byte(stcxt_t *cxt, char *cname);
830 static SV *retrieve_netint(stcxt_t *cxt, char *cname);
831 static SV *retrieve_scalar(stcxt_t *cxt, char *cname);
832 static SV *retrieve_utf8str(stcxt_t *cxt, char *cname);
833 static SV *retrieve_tied_array(stcxt_t *cxt, char *cname);
834 static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname);
835 static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname);
836 static SV *retrieve_other(stcxt_t *cxt, char *cname);
838 static SV *(*sv_old_retrieve[])(stcxt_t *cxt, char *cname) = {
839 0, /* SX_OBJECT -- entry unused dynamically */
840 retrieve_lscalar, /* SX_LSCALAR */
841 old_retrieve_array, /* SX_ARRAY -- for pre-0.6 binaries */
842 old_retrieve_hash, /* SX_HASH -- for pre-0.6 binaries */
843 retrieve_ref, /* SX_REF */
844 retrieve_undef, /* SX_UNDEF */
845 retrieve_integer, /* SX_INTEGER */
846 retrieve_double, /* SX_DOUBLE */
847 retrieve_byte, /* SX_BYTE */
848 retrieve_netint, /* SX_NETINT */
849 retrieve_scalar, /* SX_SCALAR */
850 retrieve_tied_array, /* SX_ARRAY */
851 retrieve_tied_hash, /* SX_HASH */
852 retrieve_tied_scalar, /* SX_SCALAR */
853 retrieve_other, /* SX_SV_UNDEF not supported */
854 retrieve_other, /* SX_SV_YES not supported */
855 retrieve_other, /* SX_SV_NO not supported */
856 retrieve_other, /* SX_BLESS not supported */
857 retrieve_other, /* SX_IX_BLESS not supported */
858 retrieve_other, /* SX_HOOK not supported */
859 retrieve_other, /* SX_OVERLOADED not supported */
860 retrieve_other, /* SX_TIED_KEY not supported */
861 retrieve_other, /* SX_TIED_IDX not supported */
862 retrieve_other, /* SX_UTF8STR not supported */
863 retrieve_other, /* SX_LUTF8STR not supported */
864 retrieve_other, /* SX_ERROR */
867 static SV *retrieve_array(stcxt_t *cxt, char *cname);
868 static SV *retrieve_hash(stcxt_t *cxt, char *cname);
869 static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname);
870 static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname);
871 static SV *retrieve_sv_no(stcxt_t *cxt, char *cname);
872 static SV *retrieve_blessed(stcxt_t *cxt, char *cname);
873 static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname);
874 static SV *retrieve_hook(stcxt_t *cxt, char *cname);
875 static SV *retrieve_overloaded(stcxt_t *cxt, char *cname);
876 static SV *retrieve_tied_key(stcxt_t *cxt, char *cname);
877 static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname);
879 static SV *(*sv_retrieve[])(stcxt_t *cxt, char *cname) = {
880 0, /* SX_OBJECT -- entry unused dynamically */
881 retrieve_lscalar, /* SX_LSCALAR */
882 retrieve_array, /* SX_ARRAY */
883 retrieve_hash, /* SX_HASH */
884 retrieve_ref, /* SX_REF */
885 retrieve_undef, /* SX_UNDEF */
886 retrieve_integer, /* SX_INTEGER */
887 retrieve_double, /* SX_DOUBLE */
888 retrieve_byte, /* SX_BYTE */
889 retrieve_netint, /* SX_NETINT */
890 retrieve_scalar, /* SX_SCALAR */
891 retrieve_tied_array, /* SX_ARRAY */
892 retrieve_tied_hash, /* SX_HASH */
893 retrieve_tied_scalar, /* SX_SCALAR */
894 retrieve_sv_undef, /* SX_SV_UNDEF */
895 retrieve_sv_yes, /* SX_SV_YES */
896 retrieve_sv_no, /* SX_SV_NO */
897 retrieve_blessed, /* SX_BLESS */
898 retrieve_idx_blessed, /* SX_IX_BLESS */
899 retrieve_hook, /* SX_HOOK */
900 retrieve_overloaded, /* SX_OVERLOAD */
901 retrieve_tied_key, /* SX_TIED_KEY */
902 retrieve_tied_idx, /* SX_TIED_IDX */
903 retrieve_utf8str, /* SX_UTF8STR */
904 retrieve_lutf8str, /* SX_LUTF8STR */
905 retrieve_other, /* SX_ERROR */
908 #define RETRIEVE(c,x) (*(c)->retrieve_vtbl[(x) >= SX_ERROR ? SX_ERROR : (x)])
910 static SV *mbuf2sv(void);
913 *** Context management.
919 * Called once per "thread" (interpreter) to initialize some global context.
921 static void init_perinterp(void)
925 cxt->netorder = 0; /* true if network order used */
926 cxt->forgive_me = -1; /* whether to be forgiving... */
932 * Initialize a new store context for real recursion.
934 static void init_store_context(
940 TRACEME(("init_store_context"));
942 cxt->netorder = network_order;
943 cxt->forgive_me = -1; /* Fetched from perl if needed */
944 cxt->canonical = -1; /* Idem */
945 cxt->tagnum = -1; /* Reset tag numbers */
946 cxt->classnum = -1; /* Reset class numbers */
947 cxt->fio = f; /* Where I/O are performed */
948 cxt->optype = optype; /* A store, or a deep clone */
949 cxt->entry = 1; /* No recursion yet */
952 * The `hseen' table is used to keep track of each SV stored and their
953 * associated tag numbers is special. It is "abused" because the
954 * values stored are not real SV, just integers cast to (SV *),
955 * which explains the freeing below.
957 * It is also one possible bottlneck to achieve good storing speed,
958 * so the "shared keys" optimization is turned off (unlikely to be
959 * of any use here), and the hash table is "pre-extended". Together,
960 * those optimizations increase the throughput by 12%.
963 cxt->hseen = newHV(); /* Table where seen objects are stored */
964 HvSHAREKEYS_off(cxt->hseen);
967 * The following does not work well with perl5.004_04, and causes
968 * a core dump later on, in a completely unrelated spot, which
969 * makes me think there is a memory corruption going on.
971 * Calling hv_ksplit(hseen, HBUCKETS) instead of manually hacking
972 * it below does not make any difference. It seems to work fine
973 * with perl5.004_68 but given the probable nature of the bug,
974 * that does not prove anything.
976 * It's a shame because increasing the amount of buckets raises
977 * store() throughput by 5%, but until I figure this out, I can't
978 * allow for this to go into production.
980 * It is reported fixed in 5.005, hence the #if.
982 #if PERL_VERSION >= 5
983 #define HBUCKETS 4096 /* Buckets for %hseen */
984 HvMAX(cxt->hseen) = HBUCKETS - 1; /* keys %hseen = $HBUCKETS; */
988 * The `hclass' hash uses the same settings as `hseen' above, but it is
989 * used to assign sequential tags (numbers) to class names for blessed
992 * We turn the shared key optimization on.
995 cxt->hclass = newHV(); /* Where seen classnames are stored */
997 #if PERL_VERSION >= 5
998 HvMAX(cxt->hclass) = HBUCKETS - 1; /* keys %hclass = $HBUCKETS; */
1002 * The `hook' hash table is used to keep track of the references on
1003 * the STORABLE_freeze hook routines, when found in some class name.
1005 * It is assumed that the inheritance tree will not be changed during
1006 * storing, and that no new method will be dynamically created by the
1010 cxt->hook = newHV(); /* Table where hooks are cached */
1013 * The `hook_seen' array keeps track of all the SVs returned by
1014 * STORABLE_freeze hooks for us to serialize, so that they are not
1015 * reclaimed until the end of the serialization process. Each SV is
1016 * only stored once, the first time it is seen.
1019 cxt->hook_seen = newAV(); /* Lists SVs returned by STORABLE_freeze */
1023 * clean_store_context
1025 * Clean store context by
1027 static void clean_store_context(stcxt_t *cxt)
1031 TRACEME(("clean_store_context"));
1033 ASSERT(cxt->optype & ST_STORE, ("was performing a store()"));
1036 * Insert real values into hashes where we stored faked pointers.
1039 hv_iterinit(cxt->hseen);
1040 while (he = hv_iternext(cxt->hseen))
1041 HeVAL(he) = &PL_sv_undef;
1043 hv_iterinit(cxt->hclass);
1044 while (he = hv_iternext(cxt->hclass))
1045 HeVAL(he) = &PL_sv_undef;
1048 * And now dispose of them...
1050 * The surrounding if() protection has been added because there might be
1051 * some cases where this routine is called more than once, during
1052 * exceptionnal events. This was reported by Marc Lehmann when Storable
1053 * is executed from mod_perl, and the fix was suggested by him.
1054 * -- RAM, 20/12/2000
1058 HV *hseen = cxt->hseen;
1061 sv_free((SV *) hseen);
1065 HV *hclass = cxt->hclass;
1068 sv_free((SV *) hclass);
1072 HV *hook = cxt->hook;
1075 sv_free((SV *) hook);
1078 if (cxt->hook_seen) {
1079 AV *hook_seen = cxt->hook_seen;
1081 av_undef(hook_seen);
1082 sv_free((SV *) hook_seen);
1090 * init_retrieve_context
1092 * Initialize a new retrieve context for real recursion.
1094 static void init_retrieve_context(stcxt_t *cxt, int optype, int is_tainted)
1096 TRACEME(("init_retrieve_context"));
1099 * The hook hash table is used to keep track of the references on
1100 * the STORABLE_thaw hook routines, when found in some class name.
1102 * It is assumed that the inheritance tree will not be changed during
1103 * storing, and that no new method will be dynamically created by the
1107 cxt->hook = newHV(); /* Caches STORABLE_thaw */
1110 * If retrieving an old binary version, the cxt->retrieve_vtbl variable
1111 * was set to sv_old_retrieve. We'll need a hash table to keep track of
1112 * the correspondance between the tags and the tag number used by the
1113 * new retrieve routines.
1116 cxt->hseen = (cxt->retrieve_vtbl == sv_old_retrieve) ? newHV() : 0;
1118 cxt->aseen = newAV(); /* Where retrieved objects are kept */
1119 cxt->aclass = newAV(); /* Where seen classnames are kept */
1120 cxt->tagnum = 0; /* Have to count objects... */
1121 cxt->classnum = 0; /* ...and class names as well */
1122 cxt->optype = optype;
1123 cxt->s_tainted = is_tainted;
1124 cxt->entry = 1; /* No recursion yet */
1128 * clean_retrieve_context
1130 * Clean retrieve context by
1132 static void clean_retrieve_context(stcxt_t *cxt)
1134 TRACEME(("clean_retrieve_context"));
1136 ASSERT(cxt->optype & ST_RETRIEVE, ("was performing a retrieve()"));
1139 AV *aseen = cxt->aseen;
1142 sv_free((SV *) aseen);
1146 AV *aclass = cxt->aclass;
1149 sv_free((SV *) aclass);
1153 HV *hook = cxt->hook;
1156 sv_free((SV *) hook);
1160 HV *hseen = cxt->hseen;
1163 sv_free((SV *) hseen); /* optional HV, for backward compat. */
1173 * A workaround for the CROAK bug: cleanup the last context.
1175 static void clean_context(cxt)
1178 TRACEME(("clean_context"));
1180 ASSERT(cxt->s_dirty, ("dirty context"));
1182 if (cxt->optype & ST_RETRIEVE)
1183 clean_retrieve_context(cxt);
1185 clean_store_context(cxt);
1187 ASSERT(!cxt->s_dirty, ("context is clean"));
1193 * Allocate a new context and push it on top of the parent one.
1194 * This new context is made globally visible via SET_STCXT().
1196 static stcxt_t *allocate_context(parent_cxt)
1197 stcxt_t *parent_cxt;
1201 TRACEME(("allocate_context"));
1203 ASSERT(!parent_cxt->s_dirty, ("parent context clean"));
1205 Newz(0, cxt, 1, stcxt_t);
1206 cxt->prev = parent_cxt;
1215 * Free current context, which cannot be the "root" one.
1216 * Make the context underneath globally visible via SET_STCXT().
1218 static void free_context(cxt)
1221 stcxt_t *prev = cxt->prev;
1223 TRACEME(("free_context"));
1225 ASSERT(!cxt->s_dirty, ("clean context"));
1226 ASSERT(prev, ("not freeing root context"));
1244 * Tells whether we're in the middle of a store operation.
1246 int is_storing(void)
1250 return cxt->entry && (cxt->optype & ST_STORE);
1256 * Tells whether we're in the middle of a retrieve operation.
1258 int is_retrieving(void)
1262 return cxt->entry && (cxt->optype & ST_RETRIEVE);
1266 * last_op_in_netorder
1268 * Returns whether last operation was made using network order.
1270 * This is typically out-of-band information that might prove useful
1271 * to people wishing to convert native to network order data when used.
1273 int last_op_in_netorder(void)
1277 return cxt->netorder;
1281 *** Hook lookup and calling routines.
1287 * A wrapper on gv_fetchmethod_autoload() which caches results.
1289 * Returns the routine reference as an SV*, or null if neither the package
1290 * nor its ancestors know about the method.
1292 static SV *pkg_fetchmeth(
1302 * The following code is the same as the one performed by UNIVERSAL::can
1306 gv = gv_fetchmethod_autoload(pkg, method, FALSE);
1307 if (gv && isGV(gv)) {
1308 sv = newRV((SV*) GvCV(gv));
1309 TRACEME(("%s->%s: 0x%"UVxf, HvNAME(pkg), method, PTR2UV(sv)));
1311 sv = newSVsv(&PL_sv_undef);
1312 TRACEME(("%s->%s: not found", HvNAME(pkg), method));
1316 * Cache the result, ignoring failure: if we can't store the value,
1317 * it just won't be cached.
1320 (void) hv_store(cache, HvNAME(pkg), strlen(HvNAME(pkg)), sv, 0);
1322 return SvOK(sv) ? sv : (SV *) 0;
1328 * Force cached value to be undef: hook ignored even if present.
1330 static void pkg_hide(
1335 (void) hv_store(cache,
1336 HvNAME(pkg), strlen(HvNAME(pkg)), newSVsv(&PL_sv_undef), 0);
1342 * Discard cached value: a whole fetch loop will be retried at next lookup.
1344 static void pkg_uncache(
1349 (void) hv_delete(cache, HvNAME(pkg), strlen(HvNAME(pkg)), G_DISCARD);
1355 * Our own "UNIVERSAL::can", which caches results.
1357 * Returns the routine reference as an SV*, or null if the object does not
1358 * know about the method.
1368 TRACEME(("pkg_can for %s->%s", HvNAME(pkg), method));
1371 * Look into the cache to see whether we already have determined
1372 * where the routine was, if any.
1374 * NOTA BENE: we don't use `method' at all in our lookup, since we know
1375 * that only one hook (i.e. always the same) is cached in a given cache.
1378 svh = hv_fetch(cache, HvNAME(pkg), strlen(HvNAME(pkg)), FALSE);
1382 TRACEME(("cached %s->%s: not found", HvNAME(pkg), method));
1385 TRACEME(("cached %s->%s: 0x%"UVxf,
1386 HvNAME(pkg), method, PTR2UV(sv)));
1391 TRACEME(("not cached yet"));
1392 return pkg_fetchmeth(cache, pkg, method); /* Fetch and cache */
1398 * Call routine as obj->hook(av) in scalar context.
1399 * Propagates the single returned value if not called in void context.
1401 static SV *scalar_call(
1412 TRACEME(("scalar_call (cloning=%d)", cloning));
1419 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1421 SV **ary = AvARRAY(av);
1422 int cnt = AvFILLp(av) + 1;
1424 XPUSHs(ary[0]); /* Frozen string */
1425 for (i = 1; i < cnt; i++) {
1426 TRACEME(("pushing arg #%d (0x%"UVxf")...",
1427 i, PTR2UV(ary[i])));
1428 XPUSHs(sv_2mortal(newRV(ary[i])));
1433 TRACEME(("calling..."));
1434 count = perl_call_sv(hook, flags); /* Go back to Perl code */
1435 TRACEME(("count = %d", count));
1441 SvREFCNT_inc(sv); /* We're returning it, must stay alive! */
1454 * Call routine obj->hook(cloning) in list context.
1455 * Returns the list of returned values in an array.
1457 static AV *array_call(
1467 TRACEME(("array_call (cloning=%d)", cloning));
1473 XPUSHs(obj); /* Target object */
1474 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1477 count = perl_call_sv(hook, G_ARRAY); /* Go back to Perl code */
1482 for (i = count - 1; i >= 0; i--) {
1484 av_store(av, i, SvREFCNT_inc(sv));
1497 * Lookup the class name in the `hclass' table and either assign it a new ID
1498 * or return the existing one, by filling in `classnum'.
1500 * Return true if the class was known, false if the ID was just generated.
1502 static int known_class(
1504 char *name, /* Class name */
1505 int len, /* Name length */
1509 HV *hclass = cxt->hclass;
1511 TRACEME(("known_class (%s)", name));
1514 * Recall that we don't store pointers in this hash table, but tags.
1515 * Therefore, we need LOW_32BITS() to extract the relevant parts.
1518 svh = hv_fetch(hclass, name, len, FALSE);
1520 *classnum = LOW_32BITS(*svh);
1525 * Unknown classname, we need to record it.
1529 if (!hv_store(hclass, name, len, INT2PTR(SV*, cxt->classnum), 0))
1530 CROAK(("Unable to record new classname"));
1532 *classnum = cxt->classnum;
1537 *** Sepcific store routines.
1543 * Store a reference.
1544 * Layout is SX_REF <object> or SX_OVERLOAD <object>.
1546 static int store_ref(stcxt_t *cxt, SV *sv)
1548 TRACEME(("store_ref (0x%"UVxf")", PTR2UV(sv)));
1551 * Follow reference, and check if target is overloaded.
1557 HV *stash = (HV *) SvSTASH(sv);
1558 if (stash && Gv_AMG(stash)) {
1559 TRACEME(("ref (0x%"UVxf") is overloaded", PTR2UV(sv)));
1560 PUTMARK(SX_OVERLOAD);
1566 return store(cxt, sv);
1574 * Layout is SX_LSCALAR <length> <data>, SX_SCALAR <lenght> <data> or SX_UNDEF.
1575 * The <data> section is omitted if <length> is 0.
1577 * If integer or double, the layout is SX_INTEGER <data> or SX_DOUBLE <data>.
1578 * Small integers (within [-127, +127]) are stored as SX_BYTE <byte>.
1580 static int store_scalar(stcxt_t *cxt, SV *sv)
1585 U32 flags = SvFLAGS(sv); /* "cc -O" may put it in register */
1587 TRACEME(("store_scalar (0x%"UVxf")", PTR2UV(sv)));
1590 * For efficiency, break the SV encapsulation by peaking at the flags
1591 * directly without using the Perl macros to avoid dereferencing
1592 * sv->sv_flags each time we wish to check the flags.
1595 if (!(flags & SVf_OK)) { /* !SvOK(sv) */
1596 if (sv == &PL_sv_undef) {
1597 TRACEME(("immortal undef"));
1598 PUTMARK(SX_SV_UNDEF);
1600 TRACEME(("undef at 0x%"UVxf, PTR2UV(sv)));
1607 * Always store the string representation of a scalar if it exists.
1608 * Gisle Aas provided me with this test case, better than a long speach:
1610 * perl -MDevel::Peek -le '$a="abc"; $a+0; Dump($a)'
1611 * SV = PVNV(0x80c8520)
1613 * FLAGS = (NOK,POK,pNOK,pPOK)
1616 * PV = 0x80c83d0 "abc"\0
1620 * Write SX_SCALAR, length, followed by the actual data.
1622 * Otherwise, write an SX_BYTE, SX_INTEGER or an SX_DOUBLE as
1623 * appropriate, followed by the actual (binary) data. A double
1624 * is written as a string if network order, for portability.
1626 * NOTE: instead of using SvNOK(sv), we test for SvNOKp(sv).
1627 * The reason is that when the scalar value is tainted, the SvNOK(sv)
1630 * The test for a read-only scalar with both POK and NOK set is meant
1631 * to quickly detect &PL_sv_yes and &PL_sv_no without having to pay the
1632 * address comparison for each scalar we store.
1635 #define SV_MAYBE_IMMORTAL (SVf_READONLY|SVf_POK|SVf_NOK)
1637 if ((flags & SV_MAYBE_IMMORTAL) == SV_MAYBE_IMMORTAL) {
1638 if (sv == &PL_sv_yes) {
1639 TRACEME(("immortal yes"));
1641 } else if (sv == &PL_sv_no) {
1642 TRACEME(("immortal no"));
1645 pv = SvPV(sv, len); /* We know it's SvPOK */
1646 goto string; /* Share code below */
1648 } else if (flags & SVp_POK) { /* SvPOKp(sv) => string */
1649 I32 wlen; /* For 64-bit machines */
1653 * Will come here from below with pv and len set if double & netorder,
1654 * or from above if it was readonly, POK and NOK but neither &PL_sv_yes
1659 wlen = (I32) len; /* WLEN via STORE_SCALAR expects I32 */
1661 STORE_UTF8STR(pv, wlen);
1663 STORE_SCALAR(pv, wlen);
1664 TRACEME(("ok (scalar 0x%"UVxf" '%s', length = %"IVdf")",
1665 PTR2UV(sv), SvPVX(sv), (IV)len));
1667 } else if (flags & SVp_NOK) { /* SvNOKp(sv) => double */
1671 * Watch for number being an integer in disguise.
1673 if (nv == (NV) (iv = I_V(nv))) {
1674 TRACEME(("double %"NVff" is actually integer %"IVdf, nv, iv));
1675 goto integer; /* Share code below */
1678 if (cxt->netorder) {
1679 TRACEME(("double %"NVff" stored as string", nv));
1681 goto string; /* Share code above */
1685 WRITE(&nv, sizeof(nv));
1687 TRACEME(("ok (double 0x%"UVxf", value = %"NVff")", PTR2UV(sv), nv));
1689 } else if (flags & SVp_IOK) { /* SvIOKp(sv) => integer */
1693 * Will come here from above with iv set if double is an integer.
1698 * Optimize small integers into a single byte, otherwise store as
1699 * a real integer (converted into network order if they asked).
1702 if (iv >= -128 && iv <= 127) {
1703 unsigned char siv = (unsigned char) (iv + 128); /* [0,255] */
1706 TRACEME(("small integer stored as %d", siv));
1707 } else if (cxt->netorder) {
1710 niv = (I32) htonl(iv);
1711 TRACEME(("using network order"));
1714 TRACEME(("as-is for network order"));
1719 PUTMARK(SX_INTEGER);
1720 WRITE(&iv, sizeof(iv));
1723 TRACEME(("ok (integer 0x%"UVxf", value = %"IVdf")", PTR2UV(sv), iv));
1726 CROAK(("Can't determine type of %s(0x%"UVxf")",
1727 sv_reftype(sv, FALSE),
1730 return 0; /* Ok, no recursion on scalars */
1738 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
1739 * Each item is stored as <object>.
1741 static int store_array(stcxt_t *cxt, AV *av)
1744 I32 len = av_len(av) + 1;
1748 TRACEME(("store_array (0x%"UVxf")", PTR2UV(av)));
1751 * Signal array by emitting SX_ARRAY, followed by the array length.
1756 TRACEME(("size = %d", len));
1759 * Now store each item recursively.
1762 for (i = 0; i < len; i++) {
1763 sav = av_fetch(av, i, 0);
1765 TRACEME(("(#%d) undef item", i));
1769 TRACEME(("(#%d) item", i));
1770 if (ret = store(cxt, *sav))
1774 TRACEME(("ok (array)"));
1783 * Borrowed from perl source file pp_ctl.c, where it is used by pp_sort.
1786 sortcmp(const void *a, const void *b)
1788 return sv_cmp(*(SV * const *) a, *(SV * const *) b);
1795 * Store an hash table.
1797 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
1798 * Values are stored as <object>.
1799 * Keys are stored as <length> <data>, the <data> section being omitted
1802 static int store_hash(stcxt_t *cxt, HV *hv)
1804 I32 len = HvKEYS(hv);
1810 TRACEME(("store_hash (0x%"UVxf")", PTR2UV(hv)));
1813 * Signal hash by emitting SX_HASH, followed by the table length.
1818 TRACEME(("size = %d", len));
1821 * Save possible iteration state via each() on that table.
1824 riter = HvRITER(hv);
1825 eiter = HvEITER(hv);
1829 * Now store each item recursively.
1831 * If canonical is defined to some true value then store each
1832 * key/value pair in sorted order otherwise the order is random.
1833 * Canonical order is irrelevant when a deep clone operation is performed.
1835 * Fetch the value from perl only once per store() operation, and only
1840 !(cxt->optype & ST_CLONE) && (cxt->canonical == 1 ||
1841 (cxt->canonical < 0 && (cxt->canonical =
1842 SvTRUE(perl_get_sv("Storable::canonical", TRUE)) ? 1 : 0)))
1845 * Storing in order, sorted by key.
1846 * Run through the hash, building up an array of keys in a
1847 * mortal array, sort the array and then run through the
1853 TRACEME(("using canonical order"));
1855 for (i = 0; i < len; i++) {
1856 HE *he = hv_iternext(hv);
1857 SV *key = hv_iterkeysv(he);
1858 av_store(av, AvFILLp(av)+1, key); /* av_push(), really */
1861 qsort((char *) AvARRAY(av), len, sizeof(SV *), sortcmp);
1863 for (i = 0; i < len; i++) {
1866 SV *key = av_shift(av);
1867 HE *he = hv_fetch_ent(hv, key, 0, 0);
1868 SV *val = HeVAL(he);
1870 return 1; /* Internal error, not I/O error */
1873 * Store value first.
1876 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
1878 if (ret = store(cxt, val))
1883 * Keys are written after values to make sure retrieval
1884 * can be optimal in terms of memory usage, where keys are
1885 * read into a fixed unique buffer called kbuf.
1886 * See retrieve_hash() for details.
1889 keyval = hv_iterkey(he, &keylen);
1890 TRACEME(("(#%d) key '%s'", i, keyval));
1893 WRITE(keyval, keylen);
1897 * Free up the temporary array
1906 * Storing in "random" order (in the order the keys are stored
1907 * within the the hash). This is the default and will be faster!
1910 for (i = 0; i < len; i++) {
1913 SV *val = hv_iternextsv(hv, &key, &len);
1916 return 1; /* Internal error, not I/O error */
1919 * Store value first.
1922 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
1924 if (ret = store(cxt, val))
1929 * Keys are written after values to make sure retrieval
1930 * can be optimal in terms of memory usage, where keys are
1931 * read into a fixed unique buffer called kbuf.
1932 * See retrieve_hash() for details.
1935 TRACEME(("(#%d) key '%s'", i, key));
1942 TRACEME(("ok (hash 0x%"UVxf")", PTR2UV(hv)));
1945 HvRITER(hv) = riter; /* Restore hash iterator state */
1946 HvEITER(hv) = eiter;
1954 * When storing a tied object (be it a tied scalar, array or hash), we lay out
1955 * a special mark, followed by the underlying tied object. For instance, when
1956 * dealing with a tied hash, we store SX_TIED_HASH <hash object>, where
1957 * <hash object> stands for the serialization of the tied hash.
1959 static int store_tied(stcxt_t *cxt, SV *sv)
1963 int svt = SvTYPE(sv);
1966 TRACEME(("store_tied (0x%"UVxf")", PTR2UV(sv)));
1969 * We have a small run-time penalty here because we chose to factorise
1970 * all tieds objects into the same routine, and not have a store_tied_hash,
1971 * a store_tied_array, etc...
1973 * Don't use a switch() statement, as most compilers don't optimize that
1974 * well for 2/3 values. An if() else if() cascade is just fine. We put
1975 * tied hashes first, as they are the most likely beasts.
1978 if (svt == SVt_PVHV) {
1979 TRACEME(("tied hash"));
1980 PUTMARK(SX_TIED_HASH); /* Introduces tied hash */
1981 } else if (svt == SVt_PVAV) {
1982 TRACEME(("tied array"));
1983 PUTMARK(SX_TIED_ARRAY); /* Introduces tied array */
1985 TRACEME(("tied scalar"));
1986 PUTMARK(SX_TIED_SCALAR); /* Introduces tied scalar */
1990 if (!(mg = mg_find(sv, mtype)))
1991 CROAK(("No magic '%c' found while storing tied %s", mtype,
1992 (svt == SVt_PVHV) ? "hash" :
1993 (svt == SVt_PVAV) ? "array" : "scalar"));
1996 * The mg->mg_obj found by mg_find() above actually points to the
1997 * underlying tied Perl object implementation. For instance, if the
1998 * original SV was that of a tied array, then mg->mg_obj is an AV.
2000 * Note that we store the Perl object as-is. We don't call its FETCH
2001 * method along the way. At retrieval time, we won't call its STORE
2002 * method either, but the tieing magic will be re-installed. In itself,
2003 * that ensures that the tieing semantics are preserved since futher
2004 * accesses on the retrieved object will indeed call the magic methods...
2007 if (ret = store(cxt, mg->mg_obj))
2010 TRACEME(("ok (tied)"));
2018 * Stores a reference to an item within a tied structure:
2020 * . \$h{key}, stores both the (tied %h) object and 'key'.
2021 * . \$a[idx], stores both the (tied @a) object and 'idx'.
2023 * Layout is therefore either:
2024 * SX_TIED_KEY <object> <key>
2025 * SX_TIED_IDX <object> <index>
2027 static int store_tied_item(stcxt_t *cxt, SV *sv)
2032 TRACEME(("store_tied_item (0x%"UVxf")", PTR2UV(sv)));
2034 if (!(mg = mg_find(sv, 'p')))
2035 CROAK(("No magic 'p' found while storing reference to tied item"));
2038 * We discriminate between \$h{key} and \$a[idx] via mg_ptr.
2042 TRACEME(("store_tied_item: storing a ref to a tied hash item"));
2043 PUTMARK(SX_TIED_KEY);
2044 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2046 if (ret = store(cxt, mg->mg_obj))
2049 TRACEME(("store_tied_item: storing PTR 0x%"UVxf, PTR2UV(mg->mg_ptr)));
2051 if (ret = store(cxt, (SV *) mg->mg_ptr))
2054 I32 idx = mg->mg_len;
2056 TRACEME(("store_tied_item: storing a ref to a tied array item "));
2057 PUTMARK(SX_TIED_IDX);
2058 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2060 if (ret = store(cxt, mg->mg_obj))
2063 TRACEME(("store_tied_item: storing IDX %d", idx));
2068 TRACEME(("ok (tied item)"));
2074 * store_hook -- dispatched manually, not via sv_store[]
2076 * The blessed SV is serialized by a hook.
2080 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2082 * where <flags> indicates how long <len>, <len2> and <len3> are, whether
2083 * the trailing part [] is present, the type of object (scalar, array or hash).
2084 * There is also a bit which says how the classname is stored between:
2089 * and when the <index> form is used (classname already seen), the "large
2090 * classname" bit in <flags> indicates how large the <index> is.
2092 * The serialized string returned by the hook is of length <len2> and comes
2093 * next. It is an opaque string for us.
2095 * Those <len3> object IDs which are listed last represent the extra references
2096 * not directly serialized by the hook, but which are linked to the object.
2098 * When recursion is mandated to resolve object-IDs not yet seen, we have
2099 * instead, with <header> being flags with bits set to indicate the object type
2100 * and that recursion was indeed needed:
2102 * SX_HOOK <header> <object> <header> <object> <flags>
2104 * that same header being repeated between serialized objects obtained through
2105 * recursion, until we reach flags indicating no recursion, at which point
2106 * we know we've resynchronized with a single layout, after <flags>.
2108 * When storing a blessed ref to a tied variable, the following format is
2111 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
2113 * The first <flags> indication carries an object of type SHT_EXTRA, and the
2114 * real object type is held in the <extra> flag. At the very end of the
2115 * serialization stream, the underlying magic object is serialized, just like
2116 * any other tied variable.
2118 static int store_hook(
2131 int count; /* really len3 + 1 */
2132 unsigned char flags;
2135 int recursed = 0; /* counts recursion */
2136 int obj_type; /* object type, on 2 bits */
2139 int clone = cxt->optype & ST_CLONE;
2140 char mtype; /* for blessed ref to tied structures */
2141 unsigned char eflags; /* used when object type is SHT_EXTRA */
2143 TRACEME(("store_hook, class \"%s\", tagged #%d", HvNAME(pkg), cxt->tagnum));
2146 * Determine object type on 2 bits.
2151 obj_type = SHT_SCALAR;
2154 obj_type = SHT_ARRAY;
2157 obj_type = SHT_HASH;
2161 * Produced by a blessed ref to a tied data structure, $o in the
2162 * following Perl code.
2166 * my $o = bless \%h, 'BAR';
2168 * Signal the tie-ing magic by setting the object type as SHT_EXTRA
2169 * (since we have only 2 bits in <flags> to store the type), and an
2170 * <extra> byte flag will be emitted after the FIRST <flags> in the
2171 * stream, carrying what we put in `eflags'.
2173 obj_type = SHT_EXTRA;
2174 switch (SvTYPE(sv)) {
2176 eflags = (unsigned char) SHT_THASH;
2180 eflags = (unsigned char) SHT_TARRAY;
2184 eflags = (unsigned char) SHT_TSCALAR;
2190 CROAK(("Unexpected object type (%d) in store_hook()", type));
2192 flags = SHF_NEED_RECURSE | obj_type;
2194 class = HvNAME(pkg);
2195 len = strlen(class);
2198 * To call the hook, we need to fake a call like:
2200 * $object->STORABLE_freeze($cloning);
2202 * but we don't have the $object here. For instance, if $object is
2203 * a blessed array, what we have in `sv' is the array, and we can't
2204 * call a method on those.
2206 * Therefore, we need to create a temporary reference to the object and
2207 * make the call on that reference.
2210 TRACEME(("about to call STORABLE_freeze on class %s", class));
2212 ref = newRV_noinc(sv); /* Temporary reference */
2213 av = array_call(ref, hook, clone); /* @a = $object->STORABLE_freeze($c) */
2215 SvREFCNT_dec(ref); /* Reclaim temporary reference */
2217 count = AvFILLp(av) + 1;
2218 TRACEME(("store_hook, array holds %d items", count));
2221 * If they return an empty list, it means they wish to ignore the
2222 * hook for this class (and not just this instance -- that's for them
2223 * to handle if they so wish).
2225 * Simply disable the cached entry for the hook (it won't be recomputed
2226 * since it's present in the cache) and recurse to store_blessed().
2231 * They must not change their mind in the middle of a serialization.
2234 if (hv_fetch(cxt->hclass, class, len, FALSE))
2235 CROAK(("Too late to ignore hooks for %s class \"%s\"",
2236 (cxt->optype & ST_CLONE) ? "cloning" : "storing", class));
2238 pkg_hide(cxt->hook, pkg, "STORABLE_freeze");
2240 ASSERT(!pkg_can(cxt->hook, pkg, "STORABLE_freeze"), ("hook invisible"));
2241 TRACEME(("ignoring STORABLE_freeze in class \"%s\"", class));
2243 return store_blessed(cxt, sv, type, pkg);
2247 * Get frozen string.
2251 pv = SvPV(ary[0], len2);
2254 * If they returned more than one item, we need to serialize some
2255 * extra references if not already done.
2257 * Loop over the array, starting at postion #1, and for each item,
2258 * ensure it is a reference, serialize it if not already done, and
2259 * replace the entry with the tag ID of the corresponding serialized
2262 * We CHEAT by not calling av_fetch() and read directly within the
2266 for (i = 1; i < count; i++) {
2270 AV *av_hook = cxt->hook_seen;
2273 CROAK(("Item #%d returned by STORABLE_freeze "
2274 "for %s is not a reference", i, class));
2275 xsv = SvRV(rsv); /* Follow ref to know what to look for */
2278 * Look in hseen and see if we have a tag already.
2279 * Serialize entry if not done already, and get its tag.
2282 if (svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE))
2283 goto sv_seen; /* Avoid moving code too far to the right */
2285 TRACEME(("listed object %d at 0x%"UVxf" is unknown", i-1, PTR2UV(xsv)));
2288 * We need to recurse to store that object and get it to be known
2289 * so that we can resolve the list of object-IDs at retrieve time.
2291 * The first time we do this, we need to emit the proper header
2292 * indicating that we recursed, and what the type of object is (the
2293 * object we're storing via a user-hook). Indeed, during retrieval,
2294 * we'll have to create the object before recursing to retrieve the
2295 * others, in case those would point back at that object.
2298 /* [SX_HOOK] <flags> [<extra>] <object>*/
2302 if (obj_type == SHT_EXTRA)
2307 if (ret = store(cxt, xsv)) /* Given by hook for us to store */
2310 svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE);
2312 CROAK(("Could not serialize item #%d from hook in %s", i, class));
2315 * It was the first time we serialized `xsv'.
2317 * Keep this SV alive until the end of the serialization: if we
2318 * disposed of it right now by decrementing its refcount, and it was
2319 * a temporary value, some next temporary value allocated during
2320 * another STORABLE_freeze might take its place, and we'd wrongly
2321 * assume that new SV was already serialized, based on its presence
2324 * Therefore, push it away in cxt->hook_seen.
2327 av_store(av_hook, AvFILLp(av_hook)+1, SvREFCNT_inc(xsv));
2331 * Dispose of the REF they returned. If we saved the `xsv' away
2332 * in the array of returned SVs, that will not cause the underlying
2333 * referenced SV to be reclaimed.
2336 ASSERT(SvREFCNT(xsv) > 1, ("SV will survive disposal of its REF"));
2337 SvREFCNT_dec(rsv); /* Dispose of reference */
2340 * Replace entry with its tag (not a real SV, so no refcnt increment)
2344 TRACEME(("listed object %d at 0x%"UVxf" is tag #%"UVuf,
2345 i-1, PTR2UV(xsv), PTR2UV(*svh)));
2349 * Allocate a class ID if not already done.
2351 * This needs to be done after the recursion above, since at retrieval
2352 * time, we'll see the inner objects first. Many thanks to
2353 * Salvador Ortiz Garcia <sog@msg.com.mx> who spot that bug and
2354 * proposed the right fix. -- RAM, 15/09/2000
2357 if (!known_class(cxt, class, len, &classnum)) {
2358 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2359 classnum = -1; /* Mark: we must store classname */
2361 TRACEME(("already seen class %s, ID = %d", class, classnum));
2365 * Compute leading flags.
2369 if (((classnum == -1) ? len : classnum) > LG_SCALAR)
2370 flags |= SHF_LARGE_CLASSLEN;
2372 flags |= SHF_IDX_CLASSNAME;
2373 if (len2 > LG_SCALAR)
2374 flags |= SHF_LARGE_STRLEN;
2376 flags |= SHF_HAS_LIST;
2377 if (count > (LG_SCALAR + 1))
2378 flags |= SHF_LARGE_LISTLEN;
2381 * We're ready to emit either serialized form:
2383 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2384 * SX_HOOK <flags> <index> <len2> <str> [<len3> <object-IDs>]
2386 * If we recursed, the SX_HOOK has already been emitted.
2389 TRACEME(("SX_HOOK (recursed=%d) flags=0x%x "
2390 "class=%"IVdf" len=%"IVdf" len2=%"IVdf" len3=%d",
2391 recursed, flags, (IV)classnum, (IV)len, (IV)len2, count-1));
2393 /* SX_HOOK <flags> [<extra>] */
2397 if (obj_type == SHT_EXTRA)
2402 /* <len> <classname> or <index> */
2403 if (flags & SHF_IDX_CLASSNAME) {
2404 if (flags & SHF_LARGE_CLASSLEN)
2407 unsigned char cnum = (unsigned char) classnum;
2411 if (flags & SHF_LARGE_CLASSLEN)
2414 unsigned char clen = (unsigned char) len;
2417 WRITE(class, len); /* Final \0 is omitted */
2420 /* <len2> <frozen-str> */
2421 if (flags & SHF_LARGE_STRLEN) {
2422 I32 wlen2 = len2; /* STRLEN might be 8 bytes */
2423 WLEN(wlen2); /* Must write an I32 for 64-bit machines */
2425 unsigned char clen = (unsigned char) len2;
2429 WRITE(pv, len2); /* Final \0 is omitted */
2431 /* [<len3> <object-IDs>] */
2432 if (flags & SHF_HAS_LIST) {
2433 int len3 = count - 1;
2434 if (flags & SHF_LARGE_LISTLEN)
2437 unsigned char clen = (unsigned char) len3;
2442 * NOTA BENE, for 64-bit machines: the ary[i] below does not yield a
2443 * real pointer, rather a tag number, well under the 32-bit limit.
2446 for (i = 1; i < count; i++) {
2447 I32 tagval = htonl(LOW_32BITS(ary[i]));
2449 TRACEME(("object %d, tag #%d", i-1, ntohl(tagval)));
2454 * Free the array. We need extra care for indices after 0, since they
2455 * don't hold real SVs but integers cast.
2459 AvFILLp(av) = 0; /* Cheat, nothing after 0 interests us */
2464 * If object was tied, need to insert serialization of the magic object.
2467 if (obj_type == SHT_EXTRA) {
2470 if (!(mg = mg_find(sv, mtype))) {
2471 int svt = SvTYPE(sv);
2472 CROAK(("No magic '%c' found while storing ref to tied %s with hook",
2473 mtype, (svt == SVt_PVHV) ? "hash" :
2474 (svt == SVt_PVAV) ? "array" : "scalar"));
2477 TRACEME(("handling the magic object 0x%"UVxf" part of 0x%"UVxf,
2478 PTR2UV(mg->mg_obj), PTR2UV(sv)));
2484 if (ret = store(cxt, mg->mg_obj))
2492 * store_blessed -- dispatched manually, not via sv_store[]
2494 * Check whether there is a STORABLE_xxx hook defined in the class or in one
2495 * of its ancestors. If there is, then redispatch to store_hook();
2497 * Otherwise, the blessed SV is stored using the following layout:
2499 * SX_BLESS <flag> <len> <classname> <object>
2501 * where <flag> indicates whether <len> is stored on 0 or 4 bytes, depending
2502 * on the high-order bit in flag: if 1, then length follows on 4 bytes.
2503 * Otherwise, the low order bits give the length, thereby giving a compact
2504 * representation for class names less than 127 chars long.
2506 * Each <classname> seen is remembered and indexed, so that the next time
2507 * an object in the blessed in the same <classname> is stored, the following
2510 * SX_IX_BLESS <flag> <index> <object>
2512 * where <index> is the classname index, stored on 0 or 4 bytes depending
2513 * on the high-order bit in flag (same encoding as above for <len>).
2515 static int store_blessed(
2526 TRACEME(("store_blessed, type %d, class \"%s\"", type, HvNAME(pkg)));
2529 * Look for a hook for this blessed SV and redirect to store_hook()
2533 hook = pkg_can(cxt->hook, pkg, "STORABLE_freeze");
2535 return store_hook(cxt, sv, type, pkg, hook);
2538 * This is a blessed SV without any serialization hook.
2541 class = HvNAME(pkg);
2542 len = strlen(class);
2544 TRACEME(("blessed 0x%"UVxf" in %s, no hook: tagged #%d",
2545 PTR2UV(sv), class, cxt->tagnum));
2548 * Determine whether it is the first time we see that class name (in which
2549 * case it will be stored in the SX_BLESS form), or whether we already
2550 * saw that class name before (in which case the SX_IX_BLESS form will be
2554 if (known_class(cxt, class, len, &classnum)) {
2555 TRACEME(("already seen class %s, ID = %d", class, classnum));
2556 PUTMARK(SX_IX_BLESS);
2557 if (classnum <= LG_BLESS) {
2558 unsigned char cnum = (unsigned char) classnum;
2561 unsigned char flag = (unsigned char) 0x80;
2566 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2568 if (len <= LG_BLESS) {
2569 unsigned char clen = (unsigned char) len;
2572 unsigned char flag = (unsigned char) 0x80;
2574 WLEN(len); /* Don't BER-encode, this should be rare */
2576 WRITE(class, len); /* Final \0 is omitted */
2580 * Now emit the <object> part.
2583 return SV_STORE(type)(cxt, sv);
2589 * We don't know how to store the item we reached, so return an error condition.
2590 * (it's probably a GLOB, some CODE reference, etc...)
2592 * If they defined the `forgive_me' variable at the Perl level to some
2593 * true value, then don't croak, just warn, and store a placeholder string
2596 static int store_other(stcxt_t *cxt, SV *sv)
2599 static char buf[80];
2601 TRACEME(("store_other"));
2604 * Fetch the value from perl only once per store() operation.
2608 cxt->forgive_me == 0 ||
2609 (cxt->forgive_me < 0 && !(cxt->forgive_me =
2610 SvTRUE(perl_get_sv("Storable::forgive_me", TRUE)) ? 1 : 0))
2612 CROAK(("Can't store %s items", sv_reftype(sv, FALSE)));
2614 warn("Can't store item %s(0x%"UVxf")",
2615 sv_reftype(sv, FALSE), PTR2UV(sv));
2618 * Store placeholder string as a scalar instead...
2621 (void) sprintf(buf, "You lost %s(0x%"UVxf")\0", sv_reftype(sv, FALSE),
2625 STORE_SCALAR(buf, len);
2626 TRACEME(("ok (dummy \"%s\", length = %"IVdf")", buf, len));
2632 *** Store driving routines
2638 * WARNING: partially duplicates Perl's sv_reftype for speed.
2640 * Returns the type of the SV, identified by an integer. That integer
2641 * may then be used to index the dynamic routine dispatch table.
2643 static int sv_type(SV *sv)
2645 switch (SvTYPE(sv)) {
2650 * No need to check for ROK, that can't be set here since there
2651 * is no field capable of hodling the xrv_rv reference.
2659 * Starting from SVt_PV, it is possible to have the ROK flag
2660 * set, the pointer to the other SV being either stored in
2661 * the xrv_rv (in the case of a pure SVt_RV), or as the
2662 * xpv_pv field of an SVt_PV and its heirs.
2664 * However, those SV cannot be magical or they would be an
2665 * SVt_PVMG at least.
2667 return SvROK(sv) ? svis_REF : svis_SCALAR;
2669 case SVt_PVLV: /* Workaround for perl5.004_04 "LVALUE" bug */
2670 if (SvRMAGICAL(sv) && (mg_find(sv, 'p')))
2671 return svis_TIED_ITEM;
2674 if (SvRMAGICAL(sv) && (mg_find(sv, 'q')))
2676 return SvROK(sv) ? svis_REF : svis_SCALAR;
2678 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
2682 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
2695 * Recursively store objects pointed to by the sv to the specified file.
2697 * Layout is <content> or SX_OBJECT <tagnum> if we reach an already stored
2698 * object (one for which storage has started -- it may not be over if we have
2699 * a self-referenced structure). This data set forms a stored <object>.
2701 static int store(stcxt_t *cxt, SV *sv)
2707 HV *hseen = cxt->hseen;
2709 TRACEME(("store (0x%"UVxf")", PTR2UV(sv)));
2712 * If object has already been stored, do not duplicate data.
2713 * Simply emit the SX_OBJECT marker followed by its tag data.
2714 * The tag is always written in network order.
2716 * NOTA BENE, for 64-bit machines: the "*svh" below does not yield a
2717 * real pointer, rather a tag number (watch the insertion code below).
2718 * That means it pobably safe to assume it is well under the 32-bit limit,
2719 * and makes the truncation safe.
2720 * -- RAM, 14/09/1999
2723 svh = hv_fetch(hseen, (char *) &sv, sizeof(sv), FALSE);
2725 I32 tagval = htonl(LOW_32BITS(*svh));
2727 TRACEME(("object 0x%"UVxf" seen as #%d", PTR2UV(sv), ntohl(tagval)));
2735 * Allocate a new tag and associate it with the address of the sv being
2736 * stored, before recursing...
2738 * In order to avoid creating new SvIVs to hold the tagnum we just
2739 * cast the tagnum to a SV pointer and store that in the hash. This
2740 * means that we must clean up the hash manually afterwards, but gives
2741 * us a 15% throughput increase.
2746 if (!hv_store(hseen,
2747 (char *) &sv, sizeof(sv), INT2PTR(SV*, cxt->tagnum), 0))
2751 * Store `sv' and everything beneath it, using appropriate routine.
2752 * Abort immediately if we get a non-zero status back.
2757 TRACEME(("storing 0x%"UVxf" tag #%d, type %d...",
2758 PTR2UV(sv), cxt->tagnum, type));
2761 HV *pkg = SvSTASH(sv);
2762 ret = store_blessed(cxt, sv, type, pkg);
2764 ret = SV_STORE(type)(cxt, sv);
2766 TRACEME(("%s (stored 0x%"UVxf", refcnt=%d, %s)",
2767 ret ? "FAILED" : "ok", PTR2UV(sv),
2768 SvREFCNT(sv), sv_reftype(sv, FALSE)));
2776 * Write magic number and system information into the file.
2777 * Layout is <magic> <network> [<len> <byteorder> <sizeof int> <sizeof long>
2778 * <sizeof ptr>] where <len> is the length of the byteorder hexa string.
2779 * All size and lenghts are written as single characters here.
2781 * Note that no byte ordering info is emitted when <network> is true, since
2782 * integers will be emitted in network order in that case.
2784 static int magic_write(stcxt_t *cxt)
2786 char buf[256]; /* Enough room for 256 hexa digits */
2788 int use_network_order = cxt->netorder;
2790 TRACEME(("magic_write on fd=%d", cxt->fio ? fileno(cxt->fio) : -1));
2793 WRITE(magicstr, strlen(magicstr)); /* Don't write final \0 */
2796 * Starting with 0.6, the "use_network_order" byte flag is also used to
2797 * indicate the version number of the binary image, encoded in the upper
2798 * bits. The bit 0 is always used to indicate network order.
2802 ((use_network_order ? 0x1 : 0x0) | (STORABLE_BIN_MAJOR << 1));
2806 * Starting with 0.7, a full byte is dedicated to the minor version of
2807 * the binary format, which is incremented only when new markers are
2808 * introduced, for instance, but when backward compatibility is preserved.
2811 PUTMARK((unsigned char) STORABLE_BIN_MINOR);
2813 if (use_network_order)
2814 return 0; /* Don't bother with byte ordering */
2816 sprintf(buf, "%lx", (unsigned long) BYTEORDER);
2817 c = (unsigned char) strlen(buf);
2819 WRITE(buf, (unsigned int) c); /* Don't write final \0 */
2820 PUTMARK((unsigned char) sizeof(int));
2821 PUTMARK((unsigned char) sizeof(long));
2822 PUTMARK((unsigned char) sizeof(char *));
2823 PUTMARK((unsigned char) sizeof(NV));
2825 TRACEME(("ok (magic_write byteorder = 0x%lx [%d], I%d L%d P%d D%d)",
2826 (unsigned long) BYTEORDER, (int) c,
2827 (int) sizeof(int), (int) sizeof(long),
2828 (int) sizeof(char *), (int) sizeof(NV)));
2836 * Common code for store operations.
2838 * When memory store is requested (f = NULL) and a non null SV* is given in
2839 * `res', it is filled with a new SV created out of the memory buffer.
2841 * It is required to provide a non-null `res' when the operation type is not
2842 * dclone() and store() is performed to memory.
2844 static int do_store(
2854 ASSERT(!(f == 0 && !(optype & ST_CLONE)) || res,
2855 ("must supply result SV pointer for real recursion to memory"));
2857 TRACEME(("do_store (optype=%d, netorder=%d)",
2858 optype, network_order));
2863 * Workaround for CROAK leak: if they enter with a "dirty" context,
2864 * free up memory for them now.
2871 * Now that STORABLE_xxx hooks exist, it is possible that they try to
2872 * re-enter store() via the hooks. We need to stack contexts.
2876 cxt = allocate_context(cxt);
2880 ASSERT(cxt->entry == 1, ("starting new recursion"));
2881 ASSERT(!cxt->s_dirty, ("clean context"));
2884 * Ensure sv is actually a reference. From perl, we called something
2886 * pstore(FILE, \@array);
2887 * so we must get the scalar value behing that reference.
2891 CROAK(("Not a reference"));
2892 sv = SvRV(sv); /* So follow it to know what to store */
2895 * If we're going to store to memory, reset the buffer.
2902 * Prepare context and emit headers.
2905 init_store_context(cxt, f, optype, network_order);
2907 if (-1 == magic_write(cxt)) /* Emit magic and ILP info */
2908 return 0; /* Error */
2911 * Recursively store object...
2914 ASSERT(is_storing(), ("within store operation"));
2916 status = store(cxt, sv); /* Just do it! */
2919 * If they asked for a memory store and they provided an SV pointer,
2920 * make an SV string out of the buffer and fill their pointer.
2922 * When asking for ST_REAL, it's MANDATORY for the caller to provide
2923 * an SV, since context cleanup might free the buffer if we did recurse.
2924 * (unless caller is dclone(), which is aware of that).
2927 if (!cxt->fio && res)
2933 * The "root" context is never freed, since it is meant to be always
2934 * handy for the common case where no recursion occurs at all (i.e.
2935 * we enter store() outside of any Storable code and leave it, period).
2936 * We know it's the "root" context because there's nothing stacked
2941 * When deep cloning, we don't free the context: doing so would force
2942 * us to copy the data in the memory buffer. Sicne we know we're
2943 * about to enter do_retrieve...
2946 clean_store_context(cxt);
2947 if (cxt->prev && !(cxt->optype & ST_CLONE))
2950 TRACEME(("do_store returns %d", status));
2958 * Store the transitive data closure of given object to disk.
2959 * Returns 0 on error, a true value otherwise.
2961 int pstore(PerlIO *f, SV *sv)
2963 TRACEME(("pstore"));
2964 return do_store(f, sv, 0, FALSE, (SV**) 0);
2971 * Same as pstore(), but network order is used for integers and doubles are
2972 * emitted as strings.
2974 int net_pstore(PerlIO *f, SV *sv)
2976 TRACEME(("net_pstore"));
2977 return do_store(f, sv, 0, TRUE, (SV**) 0);
2987 * Build a new SV out of the content of the internal memory buffer.
2989 static SV *mbuf2sv(void)
2993 return newSVpv(mbase, MBUF_SIZE());
2999 * Store the transitive data closure of given object to memory.
3000 * Returns undef on error, a scalar value containing the data otherwise.
3007 TRACEME(("mstore"));
3009 if (!do_store((PerlIO*) 0, sv, 0, FALSE, &out))
3010 return &PL_sv_undef;
3018 * Same as mstore(), but network order is used for integers and doubles are
3019 * emitted as strings.
3021 SV *net_mstore(SV *sv)
3026 TRACEME(("net_mstore"));
3028 if (!do_store((PerlIO*) 0, sv, 0, TRUE, &out))
3029 return &PL_sv_undef;
3035 *** Specific retrieve callbacks.
3041 * Return an error via croak, since it is not possible that we get here
3042 * under normal conditions, when facing a file produced via pstore().
3044 static SV *retrieve_other(stcxt_t *cxt, char *cname)
3047 cxt->ver_major != STORABLE_BIN_MAJOR &&
3048 cxt->ver_minor != STORABLE_BIN_MINOR
3050 CROAK(("Corrupted storable %s (binary v%d.%d), current is v%d.%d",
3051 cxt->fio ? "file" : "string",
3052 cxt->ver_major, cxt->ver_minor,
3053 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
3055 CROAK(("Corrupted storable %s (binary v%d.%d)",
3056 cxt->fio ? "file" : "string",
3057 cxt->ver_major, cxt->ver_minor));
3060 return (SV *) 0; /* Just in case */
3064 * retrieve_idx_blessed
3066 * Layout is SX_IX_BLESS <index> <object> with SX_IX_BLESS already read.
3067 * <index> can be coded on either 1 or 5 bytes.
3069 static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname)
3076 TRACEME(("retrieve_idx_blessed (#%d)", cxt->tagnum));
3077 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3079 GETMARK(idx); /* Index coded on a single char? */
3084 * Fetch classname in `aclass'
3087 sva = av_fetch(cxt->aclass, idx, FALSE);
3089 CROAK(("Class name #%"IVdf" should have been seen already",
3092 class = SvPVX(*sva); /* We know it's a PV, by construction */
3094 TRACEME(("class ID %d => %s", idx, class));
3097 * Retrieve object and bless it.
3100 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
3108 * Layout is SX_BLESS <len> <classname> <object> with SX_BLESS already read.
3109 * <len> can be coded on either 1 or 5 bytes.
3111 static SV *retrieve_blessed(stcxt_t *cxt, char *cname)
3115 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3118 TRACEME(("retrieve_blessed (#%d)", cxt->tagnum));
3119 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3122 * Decode class name length and read that name.
3124 * Short classnames have two advantages: their length is stored on one
3125 * single byte, and the string can be read on the stack.
3128 GETMARK(len); /* Length coded on a single char? */
3131 TRACEME(("** allocating %d bytes for class name", len+1));
3132 New(10003, class, len+1, char);
3135 class[len] = '\0'; /* Mark string end */
3138 * It's a new classname, otherwise it would have been an SX_IX_BLESS.
3141 TRACEME(("new class name \"%s\" will bear ID = %d", class, cxt->classnum));
3143 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3147 * Retrieve object and bless it.
3150 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
3160 * Layout: SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
3161 * with leading mark already read, as usual.
3163 * When recursion was involved during serialization of the object, there
3164 * is an unknown amount of serialized objects after the SX_HOOK mark. Until
3165 * we reach a <flags> marker with the recursion bit cleared.
3167 * If the first <flags> byte contains a type of SHT_EXTRA, then the real type
3168 * is held in the <extra> byte, and if the object is tied, the serialized
3169 * magic object comes at the very end:
3171 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
3173 * This means the STORABLE_thaw hook will NOT get a tied variable during its
3174 * processing (since we won't have seen the magic object by the time the hook
3175 * is called). See comments below for why it was done that way.
3177 static SV *retrieve_hook(stcxt_t *cxt, char *cname)
3180 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3192 int clone = cxt->optype & ST_CLONE;
3194 unsigned int extra_type = 0;
3196 TRACEME(("retrieve_hook (#%d)", cxt->tagnum));
3197 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3200 * Read flags, which tell us about the type, and whether we need to recurse.
3206 * Create the (empty) object, and mark it as seen.
3208 * This must be done now, because tags are incremented, and during
3209 * serialization, the object tag was affected before recursion could
3213 obj_type = flags & SHF_TYPE_MASK;
3219 sv = (SV *) newAV();
3222 sv = (SV *) newHV();
3226 * Read <extra> flag to know the type of the object.
3227 * Record associated magic type for later.
3229 GETMARK(extra_type);
3230 switch (extra_type) {
3236 sv = (SV *) newAV();
3240 sv = (SV *) newHV();
3244 return retrieve_other(cxt, 0); /* Let it croak */
3248 return retrieve_other(cxt, 0); /* Let it croak */
3250 SEEN(sv, 0); /* Don't bless yet */
3253 * Whilst flags tell us to recurse, do so.
3255 * We don't need to remember the addresses returned by retrieval, because
3256 * all the references will be obtained through indirection via the object
3257 * tags in the object-ID list.
3260 while (flags & SHF_NEED_RECURSE) {
3261 TRACEME(("retrieve_hook recursing..."));
3262 rv = retrieve(cxt, 0);
3265 TRACEME(("retrieve_hook back with rv=0x%"UVxf,
3270 if (flags & SHF_IDX_CLASSNAME) {
3275 * Fetch index from `aclass'
3278 if (flags & SHF_LARGE_CLASSLEN)
3283 sva = av_fetch(cxt->aclass, idx, FALSE);
3285 CROAK(("Class name #%"IVdf" should have been seen already",
3288 class = SvPVX(*sva); /* We know it's a PV, by construction */
3289 TRACEME(("class ID %d => %s", idx, class));
3293 * Decode class name length and read that name.
3295 * NOTA BENE: even if the length is stored on one byte, we don't read
3296 * on the stack. Just like retrieve_blessed(), we limit the name to
3297 * LG_BLESS bytes. This is an arbitrary decision.
3300 if (flags & SHF_LARGE_CLASSLEN)
3305 if (len > LG_BLESS) {
3306 TRACEME(("** allocating %d bytes for class name", len+1));
3307 New(10003, class, len+1, char);
3311 class[len] = '\0'; /* Mark string end */
3314 * Record new classname.
3317 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3321 TRACEME(("class name: %s", class));
3324 * Decode user-frozen string length and read it in a SV.
3326 * For efficiency reasons, we read data directly into the SV buffer.
3327 * To understand that code, read retrieve_scalar()
3330 if (flags & SHF_LARGE_STRLEN)
3335 frozen = NEWSV(10002, len2);
3337 SAFEREAD(SvPVX(frozen), len2, frozen);
3338 SvCUR_set(frozen, len2);
3339 *SvEND(frozen) = '\0';
3341 (void) SvPOK_only(frozen); /* Validates string pointer */
3342 if (cxt->s_tainted) /* Is input source tainted? */
3345 TRACEME(("frozen string: %d bytes", len2));
3348 * Decode object-ID list length, if present.
3351 if (flags & SHF_HAS_LIST) {
3352 if (flags & SHF_LARGE_LISTLEN)
3358 av_extend(av, len3 + 1); /* Leave room for [0] */
3359 AvFILLp(av) = len3; /* About to be filled anyway */
3363 TRACEME(("has %d object IDs to link", len3));
3366 * Read object-ID list into array.
3367 * Because we pre-extended it, we can cheat and fill it manually.
3369 * We read object tags and we can convert them into SV* on the fly
3370 * because we know all the references listed in there (as tags)
3371 * have been already serialized, hence we have a valid correspondance
3372 * between each of those tags and the recreated SV.
3376 SV **ary = AvARRAY(av);
3378 for (i = 1; i <= len3; i++) { /* We leave [0] alone */
3385 svh = av_fetch(cxt->aseen, tag, FALSE);
3387 CROAK(("Object #%"IVdf" should have been retrieved already", (IV)tag));
3389 ary[i] = SvREFCNT_inc(xsv);
3394 * Bless the object and look up the STORABLE_thaw hook.
3398 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3401 * Hook not found. Maybe they did not require the module where this
3402 * hook is defined yet?
3404 * If the require below succeeds, we'll be able to find the hook.
3405 * Still, it only works reliably when each class is defined in a
3409 SV *psv = newSVpvn("require ", 8);
3410 sv_catpv(psv, class);
3412 TRACEME(("No STORABLE_thaw defined for objects of class %s", class));
3413 TRACEME(("Going to require module '%s' with '%s'", class, SvPVX(psv)));
3415 perl_eval_sv(psv, G_DISCARD);
3419 * We cache results of pkg_can, so we need to uncache before attempting
3423 pkg_uncache(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3424 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3427 CROAK(("No STORABLE_thaw defined for objects of class %s "
3428 "(even after a \"require %s;\")", class, class));
3432 * If we don't have an `av' yet, prepare one.
3433 * Then insert the frozen string as item [0].
3441 AvARRAY(av)[0] = SvREFCNT_inc(frozen);
3446 * $object->STORABLE_thaw($cloning, $frozen, @refs);
3448 * where $object is our blessed (empty) object, $cloning is a boolean
3449 * telling whether we're running a deep clone, $frozen is the frozen
3450 * string the user gave us in his serializing hook, and @refs, which may
3451 * be empty, is the list of extra references he returned along for us
3454 * In effect, the hook is an alternate creation routine for the class,
3455 * the object itself being already created by the runtime.
3458 TRACEME(("calling STORABLE_thaw on %s at 0x%"UVxf" (%"IVdf" args)",
3459 class, PTR2UV(sv), AvFILLp(av) + 1));
3462 (void) scalar_call(rv, hook, clone, av, G_SCALAR|G_DISCARD);
3469 SvREFCNT_dec(frozen);
3472 if (!(flags & SHF_IDX_CLASSNAME) && class != buf)
3476 * If we had an <extra> type, then the object was not as simple, and
3477 * we need to restore extra magic now.
3483 TRACEME(("retrieving magic object for 0x%"UVxf"...", PTR2UV(sv)));
3485 rv = retrieve(cxt, 0); /* Retrieve <magic object> */
3487 TRACEME(("restoring the magic object 0x%"UVxf" part of 0x%"UVxf,
3488 PTR2UV(rv), PTR2UV(sv)));
3490 switch (extra_type) {
3492 sv_upgrade(sv, SVt_PVMG);
3495 sv_upgrade(sv, SVt_PVAV);
3496 AvREAL_off((AV *)sv);
3499 sv_upgrade(sv, SVt_PVHV);
3502 CROAK(("Forgot to deal with extra type %d", extra_type));
3507 * Adding the magic only now, well after the STORABLE_thaw hook was called
3508 * means the hook cannot know it deals with an object whose variable is
3509 * tied. But this is happening when retrieving $o in the following case:
3513 * my $o = bless \%h, 'BAR';
3515 * The 'BAR' class is NOT the one where %h is tied into. Therefore, as
3516 * far as the 'BAR' class is concerned, the fact that %h is not a REAL
3517 * hash but a tied one should not matter at all, and remain transparent.
3518 * This means the magic must be restored by Storable AFTER the hook is
3521 * That looks very reasonable to me, but then I've come up with this
3522 * after a bug report from David Nesting, who was trying to store such
3523 * an object and caused Storable to fail. And unfortunately, it was
3524 * also the easiest way to retrofit support for blessed ref to tied objects
3525 * into the existing design. -- RAM, 17/02/2001
3528 sv_magic(sv, rv, mtype, Nullch, 0);
3529 SvREFCNT_dec(rv); /* Undo refcnt inc from sv_magic() */
3537 * Retrieve reference to some other scalar.
3538 * Layout is SX_REF <object>, with SX_REF already read.
3540 static SV *retrieve_ref(stcxt_t *cxt, char *cname)
3545 TRACEME(("retrieve_ref (#%d)", cxt->tagnum));
3548 * We need to create the SV that holds the reference to the yet-to-retrieve
3549 * object now, so that we may record the address in the seen table.
3550 * Otherwise, if the object to retrieve references us, we won't be able
3551 * to resolve the SX_OBJECT we'll see at that point! Hence we cannot
3552 * do the retrieve first and use rv = newRV(sv) since it will be too late
3553 * for SEEN() recording.
3556 rv = NEWSV(10002, 0);
3557 SEEN(rv, cname); /* Will return if rv is null */
3558 sv = retrieve(cxt, 0); /* Retrieve <object> */
3560 return (SV *) 0; /* Failed */
3563 * WARNING: breaks RV encapsulation.
3565 * Now for the tricky part. We have to upgrade our existing SV, so that
3566 * it is now an RV on sv... Again, we cheat by duplicating the code
3567 * held in newSVrv(), since we already got our SV from retrieve().
3571 * SvRV(rv) = SvREFCNT_inc(sv);
3573 * here because the reference count we got from retrieve() above is
3574 * already correct: if the object was retrieved from the file, then
3575 * its reference count is one. Otherwise, if it was retrieved via
3576 * an SX_OBJECT indication, a ref count increment was done.
3579 sv_upgrade(rv, SVt_RV);
3580 SvRV(rv) = sv; /* $rv = \$sv */
3583 TRACEME(("ok (retrieve_ref at 0x%"UVxf")", PTR2UV(rv)));
3589 * retrieve_overloaded
3591 * Retrieve reference to some other scalar with overloading.
3592 * Layout is SX_OVERLOAD <object>, with SX_OVERLOAD already read.
3594 static SV *retrieve_overloaded(stcxt_t *cxt, char *cname)
3600 TRACEME(("retrieve_overloaded (#%d)", cxt->tagnum));
3603 * Same code as retrieve_ref(), duplicated to avoid extra call.
3606 rv = NEWSV(10002, 0);
3607 SEEN(rv, cname); /* Will return if rv is null */
3608 sv = retrieve(cxt, 0); /* Retrieve <object> */
3610 return (SV *) 0; /* Failed */
3613 * WARNING: breaks RV encapsulation.
3616 sv_upgrade(rv, SVt_RV);
3617 SvRV(rv) = sv; /* $rv = \$sv */
3621 * Restore overloading magic.
3624 stash = (HV *) SvSTASH (sv);
3625 if (!stash || !Gv_AMG(stash))
3626 CROAK(("Cannot restore overloading on %s(0x%"UVxf") (package %s)",
3627 sv_reftype(sv, FALSE),
3629 stash ? HvNAME(stash) : "<unknown>"));
3633 TRACEME(("ok (retrieve_overloaded at 0x%"UVxf")", PTR2UV(rv)));
3639 * retrieve_tied_array
3641 * Retrieve tied array
3642 * Layout is SX_TIED_ARRAY <object>, with SX_TIED_ARRAY already read.
3644 static SV *retrieve_tied_array(stcxt_t *cxt, char *cname)
3649 TRACEME(("retrieve_tied_array (#%d)", cxt->tagnum));
3651 tv = NEWSV(10002, 0);
3652 SEEN(tv, cname); /* Will return if tv is null */
3653 sv = retrieve(cxt, 0); /* Retrieve <object> */
3655 return (SV *) 0; /* Failed */
3657 sv_upgrade(tv, SVt_PVAV);
3658 AvREAL_off((AV *)tv);
3659 sv_magic(tv, sv, 'P', Nullch, 0);
3660 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3662 TRACEME(("ok (retrieve_tied_array at 0x%"UVxf")", PTR2UV(tv)));
3668 * retrieve_tied_hash
3670 * Retrieve tied hash
3671 * Layout is SX_TIED_HASH <object>, with SX_TIED_HASH already read.
3673 static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname)
3678 TRACEME(("retrieve_tied_hash (#%d)", cxt->tagnum));
3680 tv = NEWSV(10002, 0);
3681 SEEN(tv, cname); /* Will return if tv is null */
3682 sv = retrieve(cxt, 0); /* Retrieve <object> */
3684 return (SV *) 0; /* Failed */
3686 sv_upgrade(tv, SVt_PVHV);
3687 sv_magic(tv, sv, 'P', Nullch, 0);
3688 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3690 TRACEME(("ok (retrieve_tied_hash at 0x%"UVxf")", PTR2UV(tv)));
3696 * retrieve_tied_scalar
3698 * Retrieve tied scalar
3699 * Layout is SX_TIED_SCALAR <object>, with SX_TIED_SCALAR already read.
3701 static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname)
3706 TRACEME(("retrieve_tied_scalar (#%d)", cxt->tagnum));
3708 tv = NEWSV(10002, 0);
3709 SEEN(tv, cname); /* Will return if rv is null */
3710 sv = retrieve(cxt, 0); /* Retrieve <object> */
3712 return (SV *) 0; /* Failed */
3714 sv_upgrade(tv, SVt_PVMG);
3715 sv_magic(tv, sv, 'q', Nullch, 0);
3716 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3718 TRACEME(("ok (retrieve_tied_scalar at 0x%"UVxf")", PTR2UV(tv)));
3726 * Retrieve reference to value in a tied hash.
3727 * Layout is SX_TIED_KEY <object> <key>, with SX_TIED_KEY already read.
3729 static SV *retrieve_tied_key(stcxt_t *cxt, char *cname)
3735 TRACEME(("retrieve_tied_key (#%d)", cxt->tagnum));
3737 tv = NEWSV(10002, 0);
3738 SEEN(tv, cname); /* Will return if tv is null */
3739 sv = retrieve(cxt, 0); /* Retrieve <object> */
3741 return (SV *) 0; /* Failed */
3743 key = retrieve(cxt, 0); /* Retrieve <key> */
3745 return (SV *) 0; /* Failed */
3747 sv_upgrade(tv, SVt_PVMG);
3748 sv_magic(tv, sv, 'p', (char *)key, HEf_SVKEY);
3749 SvREFCNT_dec(key); /* Undo refcnt inc from sv_magic() */
3750 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3758 * Retrieve reference to value in a tied array.
3759 * Layout is SX_TIED_IDX <object> <idx>, with SX_TIED_IDX already read.
3761 static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname)
3767 TRACEME(("retrieve_tied_idx (#%d)", cxt->tagnum));
3769 tv = NEWSV(10002, 0);
3770 SEEN(tv, cname); /* Will return if tv is null */
3771 sv = retrieve(cxt, 0); /* Retrieve <object> */
3773 return (SV *) 0; /* Failed */
3775 RLEN(idx); /* Retrieve <idx> */
3777 sv_upgrade(tv, SVt_PVMG);
3778 sv_magic(tv, sv, 'p', Nullch, idx);
3779 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3788 * Retrieve defined long (string) scalar.
3790 * Layout is SX_LSCALAR <length> <data>, with SX_LSCALAR already read.
3791 * The scalar is "long" in that <length> is larger than LG_SCALAR so it
3792 * was not stored on a single byte.
3794 static SV *retrieve_lscalar(stcxt_t *cxt, char *cname)
3800 TRACEME(("retrieve_lscalar (#%d), len = %"IVdf, cxt->tagnum, len));
3803 * Allocate an empty scalar of the suitable length.
3806 sv = NEWSV(10002, len);
3807 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
3810 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
3812 * Now, for efficiency reasons, read data directly inside the SV buffer,
3813 * and perform the SV final settings directly by duplicating the final
3814 * work done by sv_setpv. Since we're going to allocate lots of scalars
3815 * this way, it's worth the hassle and risk.
3818 SAFEREAD(SvPVX(sv), len, sv);
3819 SvCUR_set(sv, len); /* Record C string length */
3820 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
3821 (void) SvPOK_only(sv); /* Validate string pointer */
3822 if (cxt->s_tainted) /* Is input source tainted? */
3823 SvTAINT(sv); /* External data cannot be trusted */
3825 TRACEME(("large scalar len %"IVdf" '%s'", len, SvPVX(sv)));
3826 TRACEME(("ok (retrieve_lscalar at 0x%"UVxf")", PTR2UV(sv)));
3834 * Retrieve defined short (string) scalar.
3836 * Layout is SX_SCALAR <length> <data>, with SX_SCALAR already read.
3837 * The scalar is "short" so <length> is single byte. If it is 0, there
3838 * is no <data> section.
3840 static SV *retrieve_scalar(stcxt_t *cxt, char *cname)
3846 TRACEME(("retrieve_scalar (#%d), len = %d", cxt->tagnum, len));
3849 * Allocate an empty scalar of the suitable length.
3852 sv = NEWSV(10002, len);
3853 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
3856 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
3861 * newSV did not upgrade to SVt_PV so the scalar is undefined.
3862 * To make it defined with an empty length, upgrade it now...
3864 sv_upgrade(sv, SVt_PV);
3866 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
3867 TRACEME(("ok (retrieve_scalar empty at 0x%"UVxf")", PTR2UV(sv)));
3870 * Now, for efficiency reasons, read data directly inside the SV buffer,
3871 * and perform the SV final settings directly by duplicating the final
3872 * work done by sv_setpv. Since we're going to allocate lots of scalars
3873 * this way, it's worth the hassle and risk.
3875 SAFEREAD(SvPVX(sv), len, sv);
3876 SvCUR_set(sv, len); /* Record C string length */
3877 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
3878 TRACEME(("small scalar len %d '%s'", len, SvPVX(sv)));
3881 (void) SvPOK_only(sv); /* Validate string pointer */
3882 if (cxt->s_tainted) /* Is input source tainted? */
3883 SvTAINT(sv); /* External data cannot be trusted */
3885 TRACEME(("ok (retrieve_scalar at 0x%"UVxf")", PTR2UV(sv)));
3892 * Like retrieve_scalar(), but tag result as utf8.
3893 * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
3895 static SV *retrieve_utf8str(stcxt_t *cxt, char *cname)
3899 TRACEME(("retrieve_utf8str"));
3901 sv = retrieve_scalar(cxt, cname);
3911 * Like retrieve_lscalar(), but tag result as utf8.
3912 * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
3914 static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname)
3918 TRACEME(("retrieve_lutf8str"));
3920 sv = retrieve_lscalar(cxt, cname);
3930 * Retrieve defined integer.
3931 * Layout is SX_INTEGER <data>, whith SX_INTEGER already read.
3933 static SV *retrieve_integer(stcxt_t *cxt, char *cname)
3938 TRACEME(("retrieve_integer (#%d)", cxt->tagnum));
3940 READ(&iv, sizeof(iv));
3942 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
3944 TRACEME(("integer %"IVdf, iv));
3945 TRACEME(("ok (retrieve_integer at 0x%"UVxf")", PTR2UV(sv)));
3953 * Retrieve defined integer in network order.
3954 * Layout is SX_NETINT <data>, whith SX_NETINT already read.
3956 static SV *retrieve_netint(stcxt_t *cxt, char *cname)
3961 TRACEME(("retrieve_netint (#%d)", cxt->tagnum));
3965 sv = newSViv((int) ntohl(iv));
3966 TRACEME(("network integer %d", (int) ntohl(iv)));
3969 TRACEME(("network integer (as-is) %d", iv));
3971 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
3973 TRACEME(("ok (retrieve_netint at 0x%"UVxf")", PTR2UV(sv)));
3981 * Retrieve defined double.
3982 * Layout is SX_DOUBLE <data>, whith SX_DOUBLE already read.
3984 static SV *retrieve_double(stcxt_t *cxt, char *cname)
3989 TRACEME(("retrieve_double (#%d)", cxt->tagnum));
3991 READ(&nv, sizeof(nv));
3993 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
3995 TRACEME(("double %"NVff, nv));
3996 TRACEME(("ok (retrieve_double at 0x%"UVxf")", PTR2UV(sv)));
4004 * Retrieve defined byte (small integer within the [-128, +127] range).
4005 * Layout is SX_BYTE <data>, whith SX_BYTE already read.
4007 static SV *retrieve_byte(stcxt_t *cxt, char *cname)
4011 signed char tmp; /* must use temp var to work around
4012 an AIX compiler bug --H.Merijn Brand */
4014 TRACEME(("retrieve_byte (#%d)", cxt->tagnum));
4017 TRACEME(("small integer read as %d", (unsigned char) siv));
4018 tmp = ((unsigned char)siv) - 128;
4021 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4023 TRACEME(("byte %d", tmp));
4024 TRACEME(("ok (retrieve_byte at 0x%"UVxf")", PTR2UV(sv)));
4032 * Return the undefined value.
4034 static SV *retrieve_undef(stcxt_t *cxt, char *cname)
4038 TRACEME(("retrieve_undef"));
4049 * Return the immortal undefined value.
4051 static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname)
4053 SV *sv = &PL_sv_undef;
4055 TRACEME(("retrieve_sv_undef"));
4064 * Return the immortal yes value.
4066 static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname)
4068 SV *sv = &PL_sv_yes;
4070 TRACEME(("retrieve_sv_yes"));
4079 * Return the immortal no value.
4081 static SV *retrieve_sv_no(stcxt_t *cxt, char *cname)
4085 TRACEME(("retrieve_sv_no"));
4094 * Retrieve a whole array.
4095 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
4096 * Each item is stored as <object>.
4098 * When we come here, SX_ARRAY has been read already.
4100 static SV *retrieve_array(stcxt_t *cxt, char *cname)
4107 TRACEME(("retrieve_array (#%d)", cxt->tagnum));
4110 * Read length, and allocate array, then pre-extend it.
4114 TRACEME(("size = %d", len));
4116 SEEN(av, cname); /* Will return if array not allocated nicely */
4120 return (SV *) av; /* No data follow if array is empty */
4123 * Now get each item in turn...
4126 for (i = 0; i < len; i++) {
4127 TRACEME(("(#%d) item", i));
4128 sv = retrieve(cxt, 0); /* Retrieve item */
4131 if (av_store(av, i, sv) == 0)
4135 TRACEME(("ok (retrieve_array at 0x%"UVxf")", PTR2UV(av)));
4143 * Retrieve a whole hash table.
4144 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4145 * Keys are stored as <length> <data>, the <data> section being omitted
4147 * Values are stored as <object>.
4149 * When we come here, SX_HASH has been read already.
4151 static SV *retrieve_hash(stcxt_t *cxt, char *cname)
4158 static SV *sv_h_undef = (SV *) 0; /* hv_store() bug */
4160 TRACEME(("retrieve_hash (#%d)", cxt->tagnum));
4163 * Read length, allocate table.
4167 TRACEME(("size = %d", len));
4169 SEEN(hv, cname); /* Will return if table not allocated properly */
4171 return (SV *) hv; /* No data follow if table empty */
4174 * Now get each key/value pair in turn...
4177 for (i = 0; i < len; i++) {
4182 TRACEME(("(#%d) value", i));
4183 sv = retrieve(cxt, 0);
4189 * Since we're reading into kbuf, we must ensure we're not
4190 * recursing between the read and the hv_store() where it's used.
4191 * Hence the key comes after the value.
4194 RLEN(size); /* Get key size */
4195 KBUFCHK(size); /* Grow hash key read pool if needed */
4198 kbuf[size] = '\0'; /* Mark string end, just in case */
4199 TRACEME(("(#%d) key '%s'", i, kbuf));
4202 * Enter key/value pair into hash table.
4205 if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
4209 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4215 * old_retrieve_array
4217 * Retrieve a whole array in pre-0.6 binary format.
4219 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
4220 * Each item is stored as SX_ITEM <object> or SX_IT_UNDEF for "holes".
4222 * When we come here, SX_ARRAY has been read already.
4224 static SV *old_retrieve_array(stcxt_t *cxt, char *cname)
4232 TRACEME(("old_retrieve_array (#%d)", cxt->tagnum));
4235 * Read length, and allocate array, then pre-extend it.
4239 TRACEME(("size = %d", len));
4241 SEEN(av, 0); /* Will return if array not allocated nicely */
4245 return (SV *) av; /* No data follow if array is empty */
4248 * Now get each item in turn...
4251 for (i = 0; i < len; i++) {
4253 if (c == SX_IT_UNDEF) {
4254 TRACEME(("(#%d) undef item", i));
4255 continue; /* av_extend() already filled us with undef */
4258 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
4259 TRACEME(("(#%d) item", i));
4260 sv = retrieve(cxt, 0); /* Retrieve item */
4263 if (av_store(av, i, sv) == 0)
4267 TRACEME(("ok (old_retrieve_array at 0x%"UVxf")", PTR2UV(av)));
4275 * Retrieve a whole hash table in pre-0.6 binary format.
4277 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4278 * Keys are stored as SX_KEY <length> <data>, the <data> section being omitted
4280 * Values are stored as SX_VALUE <object> or SX_VL_UNDEF for "holes".
4282 * When we come here, SX_HASH has been read already.
4284 static SV *old_retrieve_hash(stcxt_t *cxt, char *cname)
4292 static SV *sv_h_undef = (SV *) 0; /* hv_store() bug */
4294 TRACEME(("old_retrieve_hash (#%d)", cxt->tagnum));
4297 * Read length, allocate table.
4301 TRACEME(("size = %d", len));
4303 SEEN(hv, 0); /* Will return if table not allocated properly */
4305 return (SV *) hv; /* No data follow if table empty */
4308 * Now get each key/value pair in turn...
4311 for (i = 0; i < len; i++) {
4317 if (c == SX_VL_UNDEF) {
4318 TRACEME(("(#%d) undef value", i));
4320 * Due to a bug in hv_store(), it's not possible to pass
4321 * &PL_sv_undef to hv_store() as a value, otherwise the
4322 * associated key will not be creatable any more. -- RAM, 14/01/97
4325 sv_h_undef = newSVsv(&PL_sv_undef);
4326 sv = SvREFCNT_inc(sv_h_undef);
4327 } else if (c == SX_VALUE) {
4328 TRACEME(("(#%d) value", i));
4329 sv = retrieve(cxt, 0);
4333 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
4337 * Since we're reading into kbuf, we must ensure we're not
4338 * recursing between the read and the hv_store() where it's used.
4339 * Hence the key comes after the value.
4344 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
4345 RLEN(size); /* Get key size */
4346 KBUFCHK(size); /* Grow hash key read pool if needed */
4349 kbuf[size] = '\0'; /* Mark string end, just in case */
4350 TRACEME(("(#%d) key '%s'", i, kbuf));
4353 * Enter key/value pair into hash table.
4356 if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
4360 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4366 *** Retrieval engine.
4372 * Make sure the stored data we're trying to retrieve has been produced
4373 * on an ILP compatible system with the same byteorder. It croaks out in
4374 * case an error is detected. [ILP = integer-long-pointer sizes]
4375 * Returns null if error is detected, &PL_sv_undef otherwise.
4377 * Note that there's no byte ordering info emitted when network order was
4378 * used at store time.
4380 static SV *magic_check(stcxt_t *cxt)
4383 char byteorder[256];
4385 int use_network_order;
4387 int version_minor = 0;
4389 TRACEME(("magic_check"));
4392 * The "magic number" is only for files, not when freezing in memory.
4396 STRLEN len = sizeof(magicstr) - 1;
4399 READ(buf, len); /* Not null-terminated */
4400 buf[len] = '\0'; /* Is now */
4402 if (0 == strcmp(buf, magicstr))
4406 * Try to read more bytes to check for the old magic number, which
4410 old_len = sizeof(old_magicstr) - 1;
4411 READ(&buf[len], old_len - len);
4412 buf[old_len] = '\0'; /* Is now null-terminated */
4414 if (strcmp(buf, old_magicstr))
4415 CROAK(("File is not a perl storable"));
4420 * Starting with 0.6, the "use_network_order" byte flag is also used to
4421 * indicate the version number of the binary, and therefore governs the
4422 * setting of sv_retrieve_vtbl. See magic_write().
4425 GETMARK(use_network_order);
4426 version_major = use_network_order >> 1;
4427 cxt->retrieve_vtbl = version_major ? sv_retrieve : sv_old_retrieve;
4429 TRACEME(("magic_check: netorder = 0x%x", use_network_order));
4433 * Starting with 0.7 (binary major 2), a full byte is dedicated to the
4434 * minor version of the protocol. See magic_write().
4437 if (version_major > 1)
4438 GETMARK(version_minor);
4440 cxt->ver_major = version_major;
4441 cxt->ver_minor = version_minor;
4443 TRACEME(("binary image version is %d.%d", version_major, version_minor));
4446 * Inter-operability sanity check: we can't retrieve something stored
4447 * using a format more recent than ours, because we have no way to
4448 * know what has changed, and letting retrieval go would mean a probable
4449 * failure reporting a "corrupted" storable file.
4453 version_major > STORABLE_BIN_MAJOR ||
4454 (version_major == STORABLE_BIN_MAJOR &&
4455 version_minor > STORABLE_BIN_MINOR)
4457 CROAK(("Storable binary image v%d.%d more recent than I am (v%d.%d)",
4458 version_major, version_minor,
4459 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
4462 * If they stored using network order, there's no byte ordering
4463 * information to check.
4466 if (cxt->netorder = (use_network_order & 0x1))
4467 return &PL_sv_undef; /* No byte ordering info */
4469 sprintf(byteorder, "%lx", (unsigned long) BYTEORDER);
4471 READ(buf, c); /* Not null-terminated */
4472 buf[c] = '\0'; /* Is now */
4474 if (strcmp(buf, byteorder))
4475 CROAK(("Byte order is not compatible"));
4477 GETMARK(c); /* sizeof(int) */
4478 if ((int) c != sizeof(int))
4479 CROAK(("Integer size is not compatible"));
4481 GETMARK(c); /* sizeof(long) */
4482 if ((int) c != sizeof(long))
4483 CROAK(("Long integer size is not compatible"));
4485 GETMARK(c); /* sizeof(char *) */
4486 if ((int) c != sizeof(char *))
4487 CROAK(("Pointer integer size is not compatible"));
4489 if (version_major >= 2 && version_minor >= 2) {
4490 GETMARK(c); /* sizeof(NV) */
4491 if ((int) c != sizeof(NV))
4492 CROAK(("Double size is not compatible"));
4495 return &PL_sv_undef; /* OK */
4501 * Recursively retrieve objects from the specified file and return their
4502 * root SV (which may be an AV or an HV for what we care).
4503 * Returns null if there is a problem.
4505 static SV *retrieve(stcxt_t *cxt, char *cname)
4511 TRACEME(("retrieve"));
4514 * Grab address tag which identifies the object if we are retrieving
4515 * an older format. Since the new binary format counts objects and no
4516 * longer explicitely tags them, we must keep track of the correspondance
4519 * The following section will disappear one day when the old format is
4520 * no longer supported, hence the final "goto" in the "if" block.
4523 if (cxt->hseen) { /* Retrieving old binary */
4525 if (cxt->netorder) {
4527 READ(&nettag, sizeof(I32)); /* Ordered sequence of I32 */
4528 tag = (stag_t) nettag;
4530 READ(&tag, sizeof(stag_t)); /* Original address of the SV */
4533 if (type == SX_OBJECT) {
4535 svh = hv_fetch(cxt->hseen, (char *) &tag, sizeof(tag), FALSE);
4537 CROAK(("Old tag 0x%"UVxf" should have been mapped already", (UV)tag));
4538 tagn = SvIV(*svh); /* Mapped tag number computed earlier below */
4541 * The following code is common with the SX_OBJECT case below.
4544 svh = av_fetch(cxt->aseen, tagn, FALSE);
4546 CROAK(("Object #%"IVdf" should have been retrieved already", (IV)tagn));
4548 TRACEME(("has retrieved #%d at 0x%"UVxf, tagn, PTR2UV(sv)));
4549 SvREFCNT_inc(sv); /* One more reference to this same sv */
4550 return sv; /* The SV pointer where object was retrieved */
4554 * Map new object, but don't increase tagnum. This will be done
4555 * by each of the retrieve_* functions when they call SEEN().
4557 * The mapping associates the "tag" initially present with a unique
4558 * tag number. See test for SX_OBJECT above to see how this is perused.
4561 if (!hv_store(cxt->hseen, (char *) &tag, sizeof(tag),
4562 newSViv(cxt->tagnum), 0))
4569 * Regular post-0.6 binary format.
4575 TRACEME(("retrieve type = %d", type));
4578 * Are we dealing with an object we should have already retrieved?
4581 if (type == SX_OBJECT) {
4585 svh = av_fetch(cxt->aseen, tag, FALSE);
4587 CROAK(("Object #%"IVdf" should have been retrieved already",
4590 TRACEME(("had retrieved #%d at 0x%"UVxf, tag, PTR2UV(sv)));
4591 SvREFCNT_inc(sv); /* One more reference to this same sv */
4592 return sv; /* The SV pointer where object was retrieved */
4595 first_time: /* Will disappear when support for old format is dropped */
4598 * Okay, first time through for this one.
4601 sv = RETRIEVE(cxt, type)(cxt, cname);
4603 return (SV *) 0; /* Failed */
4606 * Old binary formats (pre-0.7).
4608 * Final notifications, ended by SX_STORED may now follow.
4609 * Currently, the only pertinent notification to apply on the
4610 * freshly retrieved object is either:
4611 * SX_CLASS <char-len> <classname> for short classnames.
4612 * SX_LG_CLASS <int-len> <classname> for larger one (rare!).
4613 * Class name is then read into the key buffer pool used by
4614 * hash table key retrieval.
4617 if (cxt->ver_major < 2) {
4618 while ((type = GETCHAR()) != SX_STORED) {
4622 GETMARK(len); /* Length coded on a single char */
4624 case SX_LG_CLASS: /* Length coded on a regular integer */
4629 return (SV *) 0; /* Failed */
4631 KBUFCHK(len); /* Grow buffer as necessary */
4634 kbuf[len] = '\0'; /* Mark string end */
4639 TRACEME(("ok (retrieved 0x%"UVxf", refcnt=%d, %s)", PTR2UV(sv),
4640 SvREFCNT(sv) - 1, sv_reftype(sv, FALSE)));
4648 * Retrieve data held in file and return the root object.
4649 * Common routine for pretrieve and mretrieve.
4651 static SV *do_retrieve(
4658 int is_tainted; /* Is input source tainted? */
4659 struct extendable msave; /* Where potentially valid mbuf is saved */
4661 TRACEME(("do_retrieve (optype = 0x%x)", optype));
4663 optype |= ST_RETRIEVE;
4666 * Sanity assertions for retrieve dispatch tables.
4669 ASSERT(sizeof(sv_old_retrieve) == sizeof(sv_retrieve),
4670 ("old and new retrieve dispatch table have same size"));
4671 ASSERT(sv_old_retrieve[SX_ERROR] == retrieve_other,
4672 ("SX_ERROR entry correctly initialized in old dispatch table"));
4673 ASSERT(sv_retrieve[SX_ERROR] == retrieve_other,
4674 ("SX_ERROR entry correctly initialized in new dispatch table"));
4677 * Workaround for CROAK leak: if they enter with a "dirty" context,
4678 * free up memory for them now.
4685 * Now that STORABLE_xxx hooks exist, it is possible that they try to
4686 * re-enter retrieve() via the hooks.
4690 cxt = allocate_context(cxt);
4694 ASSERT(cxt->entry == 1, ("starting new recursion"));
4695 ASSERT(!cxt->s_dirty, ("clean context"));
4700 * Data is loaded into the memory buffer when f is NULL, unless `in' is
4701 * also NULL, in which case we're expecting the data to already lie
4702 * in the buffer (dclone case).
4705 KBUFINIT(); /* Allocate hash key reading pool once */
4708 StructCopy(&cxt->membuf, &msave, struct extendable);
4714 * Magic number verifications.
4716 * This needs to be done before calling init_retrieve_context()
4717 * since the format indication in the file are necessary to conduct
4718 * some of the initializations.
4721 cxt->fio = f; /* Where I/O are performed */
4723 if (!magic_check(cxt))
4724 CROAK(("Magic number checking on storable %s failed",
4725 cxt->fio ? "file" : "string"));
4727 TRACEME(("data stored in %s format",
4728 cxt->netorder ? "net order" : "native"));
4731 * Check whether input source is tainted, so that we don't wrongly
4732 * taint perfectly good values...
4734 * We assume file input is always tainted. If both `f' and `in' are
4735 * NULL, then we come from dclone, and tainted is already filled in
4736 * the context. That's a kludge, but the whole dclone() thing is
4737 * already quite a kludge anyway! -- RAM, 15/09/2000.
4740 is_tainted = f ? 1 : (in ? SvTAINTED(in) : cxt->s_tainted);
4741 TRACEME(("input source is %s", is_tainted ? "tainted" : "trusted"));
4742 init_retrieve_context(cxt, optype, is_tainted);
4744 ASSERT(is_retrieving(), ("within retrieve operation"));
4746 sv = retrieve(cxt, 0); /* Recursively retrieve object, get root SV */
4753 StructCopy(&msave, &cxt->membuf, struct extendable);
4756 * The "root" context is never freed.
4759 clean_retrieve_context(cxt);
4760 if (cxt->prev) /* This context was stacked */
4761 free_context(cxt); /* It was not the "root" context */
4764 * Prepare returned value.
4768 TRACEME(("retrieve ERROR"));
4769 return &PL_sv_undef; /* Something went wrong, return undef */
4772 TRACEME(("retrieve got %s(0x%"UVxf")",
4773 sv_reftype(sv, FALSE), PTR2UV(sv)));
4776 * Backward compatibility with Storable-0.5@9 (which we know we
4777 * are retrieving if hseen is non-null): don't create an extra RV
4778 * for objects since we special-cased it at store time.
4780 * Build a reference to the SV returned by pretrieve even if it is
4781 * already one and not a scalar, for consistency reasons.
4783 * NB: although context might have been cleaned, the value of `cxt->hseen'
4784 * remains intact, and can be used as a flag.
4787 if (cxt->hseen) { /* Was not handling overloading by then */
4789 if (sv_type(sv) == svis_REF && (rv = SvRV(sv)) && SvOBJECT(rv))
4794 * If reference is overloaded, restore behaviour.
4796 * NB: minor glitch here: normally, overloaded refs are stored specially
4797 * so that we can croak when behaviour cannot be re-installed, and also
4798 * avoid testing for overloading magic at each reference retrieval.
4800 * Unfortunately, the root reference is implicitely stored, so we must
4801 * check for possible overloading now. Furthermore, if we don't restore
4802 * overloading, we cannot croak as if the original ref was, because we
4803 * have no way to determine whether it was an overloaded ref or not in
4806 * It's a pity that overloading magic is attached to the rv, and not to
4807 * the underlying sv as blessing is.
4811 HV *stash = (HV *) SvSTASH (sv);
4812 SV *rv = newRV_noinc(sv);
4813 if (stash && Gv_AMG(stash)) {
4815 TRACEME(("restored overloading on root reference"));
4820 return newRV_noinc(sv);
4826 * Retrieve data held in file and return the root object, undef on error.
4828 SV *pretrieve(PerlIO *f)
4830 TRACEME(("pretrieve"));
4831 return do_retrieve(f, Nullsv, 0);
4837 * Retrieve data held in scalar and return the root object, undef on error.
4839 SV *mretrieve(SV *sv)
4841 TRACEME(("mretrieve"));
4842 return do_retrieve((PerlIO*) 0, sv, 0);
4852 * Deep clone: returns a fresh copy of the original referenced SV tree.
4854 * This is achieved by storing the object in memory and restoring from
4855 * there. Not that efficient, but it should be faster than doing it from
4862 stcxt_t *real_context;
4865 TRACEME(("dclone"));
4868 * Workaround for CROAK leak: if they enter with a "dirty" context,
4869 * free up memory for them now.
4876 * do_store() optimizes for dclone by not freeing its context, should
4877 * we need to allocate one because we're deep cloning from a hook.
4880 if (!do_store((PerlIO*) 0, sv, ST_CLONE, FALSE, (SV**) 0))
4881 return &PL_sv_undef; /* Error during store */
4884 * Because of the above optimization, we have to refresh the context,
4885 * since a new one could have been allocated and stacked by do_store().
4888 { dSTCXT; real_context = cxt; } /* Sub-block needed for macro */
4889 cxt = real_context; /* And we need this temporary... */
4892 * Now, `cxt' may refer to a new context.
4895 ASSERT(!cxt->s_dirty, ("clean context"));
4896 ASSERT(!cxt->entry, ("entry will not cause new context allocation"));
4899 TRACEME(("dclone stored %d bytes", size));
4903 * Since we're passing do_retrieve() both a NULL file and sv, we need
4904 * to pre-compute the taintedness of the input by setting cxt->tainted
4905 * to whatever state our own input string was. -- RAM, 15/09/2000
4907 * do_retrieve() will free non-root context.
4910 cxt->s_tainted = SvTAINTED(sv);
4911 out = do_retrieve((PerlIO*) 0, Nullsv, ST_CLONE);
4913 TRACEME(("dclone returns 0x%"UVxf, PTR2UV(out)));
4923 * The Perl IO GV object distinguishes between input and output for sockets
4924 * but not for plain files. To allow Storable to transparently work on
4925 * plain files and sockets transparently, we have to ask xsubpp to fetch the
4926 * right object for us. Hence the OutputStream and InputStream declarations.
4928 * Before perl 5.004_05, those entries in the standard typemap are not
4929 * defined in perl include files, so we do that here.
4932 #ifndef OutputStream
4933 #define OutputStream PerlIO *
4934 #define InputStream PerlIO *
4935 #endif /* !OutputStream */
4937 MODULE = Storable PACKAGE = Storable
4975 last_op_in_netorder()