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.
53 #define DEBUGME /* Debug mode, turns assertions on as well */
54 #define DASSERT /* Assertion mode */
58 * Pre PerlIO time when none of USE_PERLIO and PERLIO_IS_STDIO is defined
59 * Provide them with the necessary defines so they can build with pre-5.004.
62 #ifndef PERLIO_IS_STDIO
64 #define PerlIO_getc(x) getc(x)
65 #define PerlIO_putc(f,x) putc(x,f)
66 #define PerlIO_read(x,y,z) fread(y,1,z,x)
67 #define PerlIO_write(x,y,z) fwrite(y,1,z,x)
68 #define PerlIO_stdoutf printf
69 #endif /* PERLIO_IS_STDIO */
70 #endif /* USE_PERLIO */
73 * Earlier versions of perl might be used, we can't assume they have the latest!
76 #ifndef PERL_VERSION /* For perls < 5.6 */
77 #include <patchlevel.h>
78 #define PERL_REVISION 5
79 #define PERL_VERSION PATCHLEVEL
80 #define PERL_SUBVERSION SUBVERSION
82 #define newRV_noinc(sv) ((Sv = newRV(sv)), --SvREFCNT(SvRV(Sv)), Sv)
84 #if (PERL_VERSION <= 4) /* Older perls (<= 5.004) lack PL_ namespace */
85 #define PL_sv_yes sv_yes
86 #define PL_sv_no sv_no
87 #define PL_sv_undef sv_undef
88 #if (PERL_SUBVERSION <= 4) /* 5.004_04 has been reported to lack newSVpvn */
89 #define newSVpvn newSVpv
91 #endif /* PERL_VERSION <= 4 */
92 #ifndef HvSHAREKEYS_off
93 #define HvSHAREKEYS_off(hv) /* Ignore */
95 #ifndef AvFILLp /* Older perls (<=5.003) lack AvFILLp */
96 #define AvFILLp AvFILL
98 typedef double NV; /* Older perls lack the NV type */
99 #define IVdf "ld" /* Various printf formats for Perl types */
103 #define INT2PTR(t,v) (t)(IV)(v)
104 #define PTR2UV(v) (unsigned long)(v)
105 #endif /* PERL_VERSION -- perls < 5.6 */
107 #ifndef NVef /* The following were not part of perl 5.6 */
108 #if defined(USE_LONG_DOUBLE) && \
109 defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
110 #define NVef PERL_PRIeldbl
111 #define NVff PERL_PRIfldbl
112 #define NVgf PERL_PRIgldbl
127 * TRACEME() will only output things when the $Storable::DEBUGME is true.
130 #define TRACEME(x) do { \
131 if (SvTRUE(perl_get_sv("Storable::DEBUGME", TRUE))) \
132 { PerlIO_stdoutf x; PerlIO_stdoutf("\n"); } \
139 #define ASSERT(x,y) do { \
141 PerlIO_stdoutf("ASSERT FAILED (\"%s\", line %d): ", \
142 __FILE__, __LINE__); \
143 PerlIO_stdoutf y; PerlIO_stdoutf("\n"); \
154 #define C(x) ((char) (x)) /* For markers with dynamic retrieval handling */
156 #define SX_OBJECT C(0) /* Already stored object */
157 #define SX_LSCALAR C(1) /* Scalar (large binary) follows (length, data) */
158 #define SX_ARRAY C(2) /* Array forthcominng (size, item list) */
159 #define SX_HASH C(3) /* Hash forthcoming (size, key/value pair list) */
160 #define SX_REF C(4) /* Reference to object forthcoming */
161 #define SX_UNDEF C(5) /* Undefined scalar */
162 #define SX_INTEGER C(6) /* Integer forthcoming */
163 #define SX_DOUBLE C(7) /* Double forthcoming */
164 #define SX_BYTE C(8) /* (signed) byte forthcoming */
165 #define SX_NETINT C(9) /* Integer in network order forthcoming */
166 #define SX_SCALAR C(10) /* Scalar (binary, small) follows (length, data) */
167 #define SX_TIED_ARRAY C(11) /* Tied array forthcoming */
168 #define SX_TIED_HASH C(12) /* Tied hash forthcoming */
169 #define SX_TIED_SCALAR C(13) /* Tied scalar forthcoming */
170 #define SX_SV_UNDEF C(14) /* Perl's immortal PL_sv_undef */
171 #define SX_SV_YES C(15) /* Perl's immortal PL_sv_yes */
172 #define SX_SV_NO C(16) /* Perl's immortal PL_sv_no */
173 #define SX_BLESS C(17) /* Object is blessed */
174 #define SX_IX_BLESS C(18) /* Object is blessed, classname given by index */
175 #define SX_HOOK C(19) /* Stored via hook, user-defined */
176 #define SX_OVERLOAD C(20) /* Overloaded reference */
177 #define SX_TIED_KEY C(21) /* Tied magic key forthcoming */
178 #define SX_TIED_IDX C(22) /* Tied magic index forthcoming */
179 #define SX_UTF8STR C(23) /* UTF-8 string forthcoming (small) */
180 #define SX_LUTF8STR C(24) /* UTF-8 string forthcoming (large) */
181 #define SX_ERROR C(25) /* Error */
184 * Those are only used to retrieve "old" pre-0.6 binary images.
186 #define SX_ITEM 'i' /* An array item introducer */
187 #define SX_IT_UNDEF 'I' /* Undefined array item */
188 #define SX_KEY 'k' /* An hash key introducer */
189 #define SX_VALUE 'v' /* An hash value introducer */
190 #define SX_VL_UNDEF 'V' /* Undefined hash value */
193 * Those are only used to retrieve "old" pre-0.7 binary images
196 #define SX_CLASS 'b' /* Object is blessed, class name length <255 */
197 #define SX_LG_CLASS 'B' /* Object is blessed, class name length >255 */
198 #define SX_STORED 'X' /* End of object */
201 * Limits between short/long length representation.
204 #define LG_SCALAR 255 /* Large scalar length limit */
205 #define LG_BLESS 127 /* Large classname bless limit */
211 #define ST_STORE 0x1 /* Store operation */
212 #define ST_RETRIEVE 0x2 /* Retrieval operation */
213 #define ST_CLONE 0x4 /* Deep cloning operation */
216 * The following structure is used for hash table key retrieval. Since, when
217 * retrieving objects, we'll be facing blessed hash references, it's best
218 * to pre-allocate that buffer once and resize it as the need arises, never
219 * freeing it (keys will be saved away someplace else anyway, so even large
220 * keys are not enough a motivation to reclaim that space).
222 * This structure is also used for memory store/retrieve operations which
223 * happen in a fixed place before being malloc'ed elsewhere if persistency
224 * is required. Hence the aptr pointer.
227 char *arena; /* Will hold hash key strings, resized as needed */
228 STRLEN asiz; /* Size of aforementionned buffer */
229 char *aptr; /* Arena pointer, for in-place read/write ops */
230 char *aend; /* First invalid address */
235 * An hash table records the objects which have already been stored.
236 * Those are referred to as SX_OBJECT in the file, and their "tag" (i.e.
237 * an arbitrary sequence number) is used to identify them.
240 * An array table records the objects which have already been retrieved,
241 * as seen by the tag determind by counting the objects themselves. The
242 * reference to that retrieved object is kept in the table, and is returned
243 * when an SX_OBJECT is found bearing that same tag.
245 * The same processing is used to record "classname" for blessed objects:
246 * indexing by a hash at store time, and via an array at retrieve time.
249 typedef unsigned long stag_t; /* Used by pre-0.6 binary format */
252 * The following "thread-safe" related defines were contributed by
253 * Murray Nesbitt <murray@activestate.com> and integrated by RAM, who
254 * only renamed things a little bit to ensure consistency with surrounding
255 * code. -- RAM, 14/09/1999
257 * The original patch suffered from the fact that the stcxt_t structure
258 * was global. Murray tried to minimize the impact on the code as much as
261 * Starting with 0.7, Storable can be re-entrant, via the STORABLE_xxx hooks
262 * on objects. Therefore, the notion of context needs to be generalized,
266 #define MY_VERSION "Storable(" XS_VERSION ")"
269 * Fields s_tainted and s_dirty are prefixed with s_ because Perl's include
270 * files remap tainted and dirty when threading is enabled. That's bad for
271 * perl to remap such common words. -- RAM, 29/09/00
274 typedef struct stcxt {
275 int entry; /* flags recursion */
276 int optype; /* type of traversal operation */
277 HV *hseen; /* which objects have been seen, store time */
278 AV *hook_seen; /* which SVs were returned by STORABLE_freeze() */
279 AV *aseen; /* which objects have been seen, retrieve time */
280 HV *hclass; /* which classnames have been seen, store time */
281 AV *aclass; /* which classnames have been seen, retrieve time */
282 HV *hook; /* cache for hook methods per class name */
283 IV tagnum; /* incremented at store time for each seen object */
284 IV classnum; /* incremented at store time for each seen classname */
285 int netorder; /* true if network order used */
286 int s_tainted; /* true if input source is tainted, at retrieve time */
287 int forgive_me; /* whether to be forgiving... */
288 int canonical; /* whether to store hashes sorted by key */
289 int s_dirty; /* context is dirty due to CROAK() -- can be cleaned */
290 struct extendable keybuf; /* for hash key retrieval */
291 struct extendable membuf; /* for memory store/retrieve operations */
292 PerlIO *fio; /* where I/O are performed, NULL for memory */
293 int ver_major; /* major of version for retrieved object */
294 int ver_minor; /* minor of version for retrieved object */
295 SV *(**retrieve_vtbl)(); /* retrieve dispatch table */
296 struct stcxt *prev; /* contexts chained backwards in real recursion */
299 #if defined(MULTIPLICITY) || defined(PERL_OBJECT) || defined(PERL_CAPI)
301 #if (PERL_VERSION <= 4) && (PERL_SUBVERSION < 68)
303 SV *perinterp_sv = perl_get_sv(MY_VERSION, FALSE)
304 #else /* >= perl5.004_68 */
306 SV *perinterp_sv = *hv_fetch(PL_modglobal, \
307 MY_VERSION, sizeof(MY_VERSION)-1, TRUE)
308 #endif /* < perl5.004_68 */
310 #define dSTCXT_PTR(T,name) \
311 T name = (perinterp_sv && SvIOK(perinterp_sv) \
312 ? INT2PTR(T, SvIVX(perinterp_sv)) : (T) 0)
315 dSTCXT_PTR(stcxt_t *, cxt)
319 Newz(0, cxt, 1, stcxt_t); \
320 sv_setiv(perinterp_sv, PTR2IV(cxt))
322 #define SET_STCXT(x) do { \
324 sv_setiv(perinterp_sv, PTR2IV(x)); \
327 #else /* !MULTIPLICITY && !PERL_OBJECT && !PERL_CAPI */
329 static stcxt_t Context;
330 static stcxt_t *Context_ptr = &Context;
331 #define dSTCXT stcxt_t *cxt = Context_ptr
332 #define INIT_STCXT dSTCXT
333 #define SET_STCXT(x) Context_ptr = x
335 #endif /* MULTIPLICITY || PERL_OBJECT || PERL_CAPI */
339 * Croaking implies a memory leak, since we don't use setjmp/longjmp
340 * to catch the exit and free memory used during store or retrieve
341 * operations. This is not too difficult to fix, but I need to understand
342 * how Perl does it, and croaking is exceptional anyway, so I lack the
343 * motivation to do it.
345 * The current workaround is to mark the context as dirty when croaking,
346 * so that data structures can be freed whenever we renter Storable code
347 * (but only *then*: it's a workaround, not a fix).
349 * This is also imperfect, because we don't really know how far they trapped
350 * the croak(), and when we were recursing, we won't be able to clean anything
351 * but the topmost context stacked.
354 #define CROAK(x) do { cxt->s_dirty = 1; croak x; } while (0)
357 * End of "thread-safe" related definitions.
363 * Keep only the low 32 bits of a pointer (used for tags, which are not
368 #define LOW_32BITS(x) ((I32) (x))
370 #define LOW_32BITS(x) ((I32) ((unsigned long) (x) & 0xffffffffUL))
376 * Hack for Crays, where sizeof(I32) == 8, and which are big-endians.
377 * Used in the WLEN and RLEN macros.
381 #define oI(x) ((I32 *) ((char *) (x) + 4))
382 #define oS(x) ((x) - 4)
383 #define oC(x) (x = 0)
392 * key buffer handling
394 #define kbuf (cxt->keybuf).arena
395 #define ksiz (cxt->keybuf).asiz
396 #define KBUFINIT() do { \
398 TRACEME(("** allocating kbuf of 128 bytes")); \
399 New(10003, kbuf, 128, char); \
403 #define KBUFCHK(x) do { \
405 TRACEME(("** extending kbuf to %d bytes", x+1)); \
406 Renew(kbuf, x+1, char); \
412 * memory buffer handling
414 #define mbase (cxt->membuf).arena
415 #define msiz (cxt->membuf).asiz
416 #define mptr (cxt->membuf).aptr
417 #define mend (cxt->membuf).aend
419 #define MGROW (1 << 13)
420 #define MMASK (MGROW - 1)
422 #define round_mgrow(x) \
423 ((unsigned long) (((unsigned long) (x) + MMASK) & ~MMASK))
424 #define trunc_int(x) \
425 ((unsigned long) ((unsigned long) (x) & ~(sizeof(int)-1)))
426 #define int_aligned(x) \
427 ((unsigned long) (x) == trunc_int(x))
429 #define MBUF_INIT(x) do { \
431 TRACEME(("** allocating mbase of %d bytes", MGROW)); \
432 New(10003, mbase, MGROW, char); \
439 mend = mbase + msiz; \
442 #define MBUF_TRUNC(x) mptr = mbase + x
443 #define MBUF_SIZE() (mptr - mbase)
446 * Use SvPOKp(), because SvPOK() fails on tainted scalars.
447 * See store_scalar() for other usage of this workaround.
449 #define MBUF_LOAD(v) do { \
451 CROAK(("Not a scalar string")); \
452 mptr = mbase = SvPV(v, msiz); \
453 mend = mbase + msiz; \
456 #define MBUF_XTEND(x) do { \
457 int nsz = (int) round_mgrow((x)+msiz); \
458 int offset = mptr - mbase; \
459 TRACEME(("** extending mbase to %d bytes", nsz)); \
460 Renew(mbase, nsz, char); \
462 mptr = mbase + offset; \
463 mend = mbase + nsz; \
466 #define MBUF_CHK(x) do { \
467 if ((mptr + (x)) > mend) \
471 #define MBUF_GETC(x) do { \
473 x = (int) (unsigned char) *mptr++; \
479 #define MBUF_GETINT(x) do { \
481 if ((mptr + 4) <= mend) { \
482 memcpy(oI(&x), mptr, 4); \
488 #define MBUF_GETINT(x) do { \
489 if ((mptr + sizeof(int)) <= mend) { \
490 if (int_aligned(mptr)) \
493 memcpy(&x, mptr, sizeof(int)); \
494 mptr += sizeof(int); \
500 #define MBUF_READ(x,s) do { \
501 if ((mptr + (s)) <= mend) { \
502 memcpy(x, mptr, s); \
508 #define MBUF_SAFEREAD(x,s,z) do { \
509 if ((mptr + (s)) <= mend) { \
510 memcpy(x, mptr, s); \
518 #define MBUF_PUTC(c) do { \
520 *mptr++ = (char) c; \
523 *mptr++ = (char) c; \
528 #define MBUF_PUTINT(i) do { \
530 memcpy(mptr, oI(&i), 4); \
534 #define MBUF_PUTINT(i) do { \
535 MBUF_CHK(sizeof(int)); \
536 if (int_aligned(mptr)) \
539 memcpy(mptr, &i, sizeof(int)); \
540 mptr += sizeof(int); \
544 #define MBUF_WRITE(x,s) do { \
546 memcpy(mptr, x, s); \
551 * Possible return values for sv_type().
555 #define svis_SCALAR 1
559 #define svis_TIED_ITEM 5
566 #define SHF_TYPE_MASK 0x03
567 #define SHF_LARGE_CLASSLEN 0x04
568 #define SHF_LARGE_STRLEN 0x08
569 #define SHF_LARGE_LISTLEN 0x10
570 #define SHF_IDX_CLASSNAME 0x20
571 #define SHF_NEED_RECURSE 0x40
572 #define SHF_HAS_LIST 0x80
575 * Types for SX_HOOK (last 2 bits in flags).
581 #define SHT_EXTRA 3 /* Read extra byte for type */
584 * The following are held in the "extra byte"...
587 #define SHT_TSCALAR 4 /* 4 + 0 -- tied scalar */
588 #define SHT_TARRAY 5 /* 4 + 1 -- tied array */
589 #define SHT_THASH 6 /* 4 + 2 -- tied hash */
592 * Before 0.6, the magic string was "perl-store" (binary version number 0).
594 * Since 0.6 introduced many binary incompatibilities, the magic string has
595 * been changed to "pst0" to allow an old image to be properly retrieved by
596 * a newer Storable, but ensure a newer image cannot be retrieved with an
599 * At 0.7, objects are given the ability to serialize themselves, and the
600 * set of markers is extended, backward compatibility is not jeopardized,
601 * so the binary version number could have remained unchanged. To correctly
602 * spot errors if a file making use of 0.7-specific extensions is given to
603 * 0.6 for retrieval, the binary version was moved to "2". And I'm introducing
604 * a "minor" version, to better track this kind of evolution from now on.
607 static char old_magicstr[] = "perl-store"; /* Magic number before 0.6 */
608 static char magicstr[] = "pst0"; /* Used as a magic number */
610 #define STORABLE_BIN_MAJOR 2 /* Binary major "version" */
611 #define STORABLE_BIN_MINOR 4 /* Binary minor "version" */
614 * Useful store shortcuts...
617 #define PUTMARK(x) do { \
620 else if (PerlIO_putc(cxt->fio, x) == EOF) \
624 #define WRITE_I32(x) do { \
625 ASSERT(sizeof(x) == sizeof(I32), ("writing an I32")); \
628 else if (PerlIO_write(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
633 #define WLEN(x) do { \
634 if (cxt->netorder) { \
635 int y = (int) htonl(x); \
638 else if (PerlIO_write(cxt->fio,oI(&y),oS(sizeof(y))) != oS(sizeof(y))) \
643 else if (PerlIO_write(cxt->fio,oI(&x),oS(sizeof(x))) != oS(sizeof(x))) \
648 #define WLEN(x) WRITE_I32(x)
651 #define WRITE(x,y) do { \
654 else if (PerlIO_write(cxt->fio, x, y) != y) \
658 #define STORE_PV_LEN(pv, len, small, large) do { \
659 if (len <= LG_SCALAR) { \
660 unsigned char clen = (unsigned char) len; \
672 #define STORE_SCALAR(pv, len) STORE_PV_LEN(pv, len, SX_SCALAR, SX_LSCALAR)
675 * Conditional UTF8 support.
676 * On non-UTF8 perls, UTF8 strings are returned as normal strings.
680 #define STORE_UTF8STR(pv, len) STORE_PV_LEN(pv, len, SX_UTF8STR, SX_LUTF8STR)
683 #define STORE_UTF8STR(pv, len) CROAK(("panic: storing UTF8 in non-UTF8 perl"))
684 #define SvUTF8_on(sv) CROAK(("Cannot retrieve UTF8 data in non-UTF8 perl"))
688 * Store undef in arrays and hashes without recursing through store().
690 #define STORE_UNDEF() do { \
696 * Useful retrieve shortcuts...
700 (cxt->fio ? PerlIO_getc(cxt->fio) : (mptr >= mend ? EOF : (int) *mptr++))
702 #define GETMARK(x) do { \
705 else if ((int) (x = PerlIO_getc(cxt->fio)) == EOF) \
709 #define READ_I32(x) do { \
710 ASSERT(sizeof(x) == sizeof(I32), ("reading an I32")); \
714 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
719 #define RLEN(x) do { \
723 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
726 x = (int) ntohl(x); \
729 #define RLEN(x) READ_I32(x)
732 #define READ(x,y) do { \
735 else if (PerlIO_read(cxt->fio, x, y) != y) \
739 #define SAFEREAD(x,y,z) do { \
741 MBUF_SAFEREAD(x,y,z); \
742 else if (PerlIO_read(cxt->fio, x, y) != y) { \
749 * This macro is used at retrieve time, to remember where object 'y', bearing a
750 * given tag 'tagnum', has been retrieved. Next time we see an SX_OBJECT marker,
751 * we'll therefore know where it has been retrieved and will be able to
752 * share the same reference, as in the original stored memory image.
754 * We also need to bless objects ASAP for hooks (which may compute "ref $x"
755 * on the objects given to STORABLE_thaw and expect that to be defined), and
756 * also for overloaded objects (for which we might not find the stash if the
757 * object is not blessed yet--this might occur for overloaded objects that
758 * refer to themselves indirectly: if we blessed upon return from a sub
759 * retrieve(), the SX_OBJECT marker we'd found could not have overloading
760 * restored on it because the underlying object would not be blessed yet!).
762 * To achieve that, the class name of the last retrieved object is passed down
763 * recursively, and the first SEEN() call for which the class name is not NULL
764 * will bless the object.
766 #define SEEN(y,c) do { \
769 if (av_store(cxt->aseen, cxt->tagnum++, SvREFCNT_inc(y)) == 0) \
771 TRACEME(("aseen(#%d) = 0x%"UVxf" (refcnt=%d)", cxt->tagnum-1, \
772 PTR2UV(y), SvREFCNT(y)-1)); \
774 BLESS((SV *) (y), c); \
778 * Bless `s' in `p', via a temporary reference, required by sv_bless().
780 #define BLESS(s,p) do { \
783 TRACEME(("blessing 0x%"UVxf" in %s", PTR2UV(s), (p))); \
784 stash = gv_stashpv((p), TRUE); \
785 ref = newRV_noinc(s); \
786 (void) sv_bless(ref, stash); \
792 static SV *retrieve(stcxt_t *cxt, char *cname);
795 * Dynamic dispatching table for SV store.
798 static int store_ref(stcxt_t *cxt, SV *sv);
799 static int store_scalar(stcxt_t *cxt, SV *sv);
800 static int store_array(stcxt_t *cxt, AV *av);
801 static int store_hash(stcxt_t *cxt, HV *hv);
802 static int store_tied(stcxt_t *cxt, SV *sv);
803 static int store_tied_item(stcxt_t *cxt, SV *sv);
804 static int store_other(stcxt_t *cxt, SV *sv);
805 static int store_blessed(stcxt_t *cxt, SV *sv, int type, HV *pkg);
807 static int (*sv_store[])(stcxt_t *cxt, SV *sv) = {
808 store_ref, /* svis_REF */
809 store_scalar, /* svis_SCALAR */
810 (int (*)(stcxt_t *cxt, SV *sv)) store_array, /* svis_ARRAY */
811 (int (*)(stcxt_t *cxt, SV *sv)) store_hash, /* svis_HASH */
812 store_tied, /* svis_TIED */
813 store_tied_item, /* svis_TIED_ITEM */
814 store_other, /* svis_OTHER */
817 #define SV_STORE(x) (*sv_store[x])
820 * Dynamic dispatching tables for SV retrieval.
823 static SV *retrieve_lscalar(stcxt_t *cxt, char *cname);
824 static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname);
825 static SV *old_retrieve_array(stcxt_t *cxt, char *cname);
826 static SV *old_retrieve_hash(stcxt_t *cxt, char *cname);
827 static SV *retrieve_ref(stcxt_t *cxt, char *cname);
828 static SV *retrieve_undef(stcxt_t *cxt, char *cname);
829 static SV *retrieve_integer(stcxt_t *cxt, char *cname);
830 static SV *retrieve_double(stcxt_t *cxt, char *cname);
831 static SV *retrieve_byte(stcxt_t *cxt, char *cname);
832 static SV *retrieve_netint(stcxt_t *cxt, char *cname);
833 static SV *retrieve_scalar(stcxt_t *cxt, char *cname);
834 static SV *retrieve_utf8str(stcxt_t *cxt, char *cname);
835 static SV *retrieve_tied_array(stcxt_t *cxt, char *cname);
836 static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname);
837 static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname);
838 static SV *retrieve_other(stcxt_t *cxt, char *cname);
840 static SV *(*sv_old_retrieve[])(stcxt_t *cxt, char *cname) = {
841 0, /* SX_OBJECT -- entry unused dynamically */
842 retrieve_lscalar, /* SX_LSCALAR */
843 old_retrieve_array, /* SX_ARRAY -- for pre-0.6 binaries */
844 old_retrieve_hash, /* SX_HASH -- for pre-0.6 binaries */
845 retrieve_ref, /* SX_REF */
846 retrieve_undef, /* SX_UNDEF */
847 retrieve_integer, /* SX_INTEGER */
848 retrieve_double, /* SX_DOUBLE */
849 retrieve_byte, /* SX_BYTE */
850 retrieve_netint, /* SX_NETINT */
851 retrieve_scalar, /* SX_SCALAR */
852 retrieve_tied_array, /* SX_ARRAY */
853 retrieve_tied_hash, /* SX_HASH */
854 retrieve_tied_scalar, /* SX_SCALAR */
855 retrieve_other, /* SX_SV_UNDEF not supported */
856 retrieve_other, /* SX_SV_YES not supported */
857 retrieve_other, /* SX_SV_NO not supported */
858 retrieve_other, /* SX_BLESS not supported */
859 retrieve_other, /* SX_IX_BLESS not supported */
860 retrieve_other, /* SX_HOOK not supported */
861 retrieve_other, /* SX_OVERLOADED not supported */
862 retrieve_other, /* SX_TIED_KEY not supported */
863 retrieve_other, /* SX_TIED_IDX not supported */
864 retrieve_other, /* SX_UTF8STR not supported */
865 retrieve_other, /* SX_LUTF8STR not supported */
866 retrieve_other, /* SX_ERROR */
869 static SV *retrieve_array(stcxt_t *cxt, char *cname);
870 static SV *retrieve_hash(stcxt_t *cxt, char *cname);
871 static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname);
872 static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname);
873 static SV *retrieve_sv_no(stcxt_t *cxt, char *cname);
874 static SV *retrieve_blessed(stcxt_t *cxt, char *cname);
875 static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname);
876 static SV *retrieve_hook(stcxt_t *cxt, char *cname);
877 static SV *retrieve_overloaded(stcxt_t *cxt, char *cname);
878 static SV *retrieve_tied_key(stcxt_t *cxt, char *cname);
879 static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname);
881 static SV *(*sv_retrieve[])(stcxt_t *cxt, char *cname) = {
882 0, /* SX_OBJECT -- entry unused dynamically */
883 retrieve_lscalar, /* SX_LSCALAR */
884 retrieve_array, /* SX_ARRAY */
885 retrieve_hash, /* SX_HASH */
886 retrieve_ref, /* SX_REF */
887 retrieve_undef, /* SX_UNDEF */
888 retrieve_integer, /* SX_INTEGER */
889 retrieve_double, /* SX_DOUBLE */
890 retrieve_byte, /* SX_BYTE */
891 retrieve_netint, /* SX_NETINT */
892 retrieve_scalar, /* SX_SCALAR */
893 retrieve_tied_array, /* SX_ARRAY */
894 retrieve_tied_hash, /* SX_HASH */
895 retrieve_tied_scalar, /* SX_SCALAR */
896 retrieve_sv_undef, /* SX_SV_UNDEF */
897 retrieve_sv_yes, /* SX_SV_YES */
898 retrieve_sv_no, /* SX_SV_NO */
899 retrieve_blessed, /* SX_BLESS */
900 retrieve_idx_blessed, /* SX_IX_BLESS */
901 retrieve_hook, /* SX_HOOK */
902 retrieve_overloaded, /* SX_OVERLOAD */
903 retrieve_tied_key, /* SX_TIED_KEY */
904 retrieve_tied_idx, /* SX_TIED_IDX */
905 retrieve_utf8str, /* SX_UTF8STR */
906 retrieve_lutf8str, /* SX_LUTF8STR */
907 retrieve_other, /* SX_ERROR */
910 #define RETRIEVE(c,x) (*(c)->retrieve_vtbl[(x) >= SX_ERROR ? SX_ERROR : (x)])
912 static SV *mbuf2sv(void);
915 *** Context management.
921 * Called once per "thread" (interpreter) to initialize some global context.
923 static void init_perinterp(void)
927 cxt->netorder = 0; /* true if network order used */
928 cxt->forgive_me = -1; /* whether to be forgiving... */
934 * Initialize a new store context for real recursion.
936 static void init_store_context(
942 TRACEME(("init_store_context"));
944 cxt->netorder = network_order;
945 cxt->forgive_me = -1; /* Fetched from perl if needed */
946 cxt->canonical = -1; /* Idem */
947 cxt->tagnum = -1; /* Reset tag numbers */
948 cxt->classnum = -1; /* Reset class numbers */
949 cxt->fio = f; /* Where I/O are performed */
950 cxt->optype = optype; /* A store, or a deep clone */
951 cxt->entry = 1; /* No recursion yet */
954 * The `hseen' table is used to keep track of each SV stored and their
955 * associated tag numbers is special. It is "abused" because the
956 * values stored are not real SV, just integers cast to (SV *),
957 * which explains the freeing below.
959 * It is also one possible bottlneck to achieve good storing speed,
960 * so the "shared keys" optimization is turned off (unlikely to be
961 * of any use here), and the hash table is "pre-extended". Together,
962 * those optimizations increase the throughput by 12%.
965 cxt->hseen = newHV(); /* Table where seen objects are stored */
966 HvSHAREKEYS_off(cxt->hseen);
969 * The following does not work well with perl5.004_04, and causes
970 * a core dump later on, in a completely unrelated spot, which
971 * makes me think there is a memory corruption going on.
973 * Calling hv_ksplit(hseen, HBUCKETS) instead of manually hacking
974 * it below does not make any difference. It seems to work fine
975 * with perl5.004_68 but given the probable nature of the bug,
976 * that does not prove anything.
978 * It's a shame because increasing the amount of buckets raises
979 * store() throughput by 5%, but until I figure this out, I can't
980 * allow for this to go into production.
982 * It is reported fixed in 5.005, hence the #if.
984 #if PERL_VERSION >= 5
985 #define HBUCKETS 4096 /* Buckets for %hseen */
986 HvMAX(cxt->hseen) = HBUCKETS - 1; /* keys %hseen = $HBUCKETS; */
990 * The `hclass' hash uses the same settings as `hseen' above, but it is
991 * used to assign sequential tags (numbers) to class names for blessed
994 * We turn the shared key optimization on.
997 cxt->hclass = newHV(); /* Where seen classnames are stored */
999 #if PERL_VERSION >= 5
1000 HvMAX(cxt->hclass) = HBUCKETS - 1; /* keys %hclass = $HBUCKETS; */
1004 * The `hook' hash table is used to keep track of the references on
1005 * the STORABLE_freeze hook routines, when found in some class name.
1007 * It is assumed that the inheritance tree will not be changed during
1008 * storing, and that no new method will be dynamically created by the
1012 cxt->hook = newHV(); /* Table where hooks are cached */
1015 * The `hook_seen' array keeps track of all the SVs returned by
1016 * STORABLE_freeze hooks for us to serialize, so that they are not
1017 * reclaimed until the end of the serialization process. Each SV is
1018 * only stored once, the first time it is seen.
1021 cxt->hook_seen = newAV(); /* Lists SVs returned by STORABLE_freeze */
1025 * clean_store_context
1027 * Clean store context by
1029 static void clean_store_context(stcxt_t *cxt)
1033 TRACEME(("clean_store_context"));
1035 ASSERT(cxt->optype & ST_STORE, ("was performing a store()"));
1038 * Insert real values into hashes where we stored faked pointers.
1041 hv_iterinit(cxt->hseen);
1042 while ((he = hv_iternext(cxt->hseen)))
1043 HeVAL(he) = &PL_sv_undef;
1045 hv_iterinit(cxt->hclass);
1046 while ((he = hv_iternext(cxt->hclass)))
1047 HeVAL(he) = &PL_sv_undef;
1050 * And now dispose of them...
1052 * The surrounding if() protection has been added because there might be
1053 * some cases where this routine is called more than once, during
1054 * exceptionnal events. This was reported by Marc Lehmann when Storable
1055 * is executed from mod_perl, and the fix was suggested by him.
1056 * -- RAM, 20/12/2000
1060 HV *hseen = cxt->hseen;
1063 sv_free((SV *) hseen);
1067 HV *hclass = cxt->hclass;
1070 sv_free((SV *) hclass);
1074 HV *hook = cxt->hook;
1077 sv_free((SV *) hook);
1080 if (cxt->hook_seen) {
1081 AV *hook_seen = cxt->hook_seen;
1083 av_undef(hook_seen);
1084 sv_free((SV *) hook_seen);
1092 * init_retrieve_context
1094 * Initialize a new retrieve context for real recursion.
1096 static void init_retrieve_context(stcxt_t *cxt, int optype, int is_tainted)
1098 TRACEME(("init_retrieve_context"));
1101 * The hook hash table is used to keep track of the references on
1102 * the STORABLE_thaw hook routines, when found in some class name.
1104 * It is assumed that the inheritance tree will not be changed during
1105 * storing, and that no new method will be dynamically created by the
1109 cxt->hook = newHV(); /* Caches STORABLE_thaw */
1112 * If retrieving an old binary version, the cxt->retrieve_vtbl variable
1113 * was set to sv_old_retrieve. We'll need a hash table to keep track of
1114 * the correspondance between the tags and the tag number used by the
1115 * new retrieve routines.
1118 cxt->hseen = (cxt->retrieve_vtbl == sv_old_retrieve) ? newHV() : 0;
1120 cxt->aseen = newAV(); /* Where retrieved objects are kept */
1121 cxt->aclass = newAV(); /* Where seen classnames are kept */
1122 cxt->tagnum = 0; /* Have to count objects... */
1123 cxt->classnum = 0; /* ...and class names as well */
1124 cxt->optype = optype;
1125 cxt->s_tainted = is_tainted;
1126 cxt->entry = 1; /* No recursion yet */
1130 * clean_retrieve_context
1132 * Clean retrieve context by
1134 static void clean_retrieve_context(stcxt_t *cxt)
1136 TRACEME(("clean_retrieve_context"));
1138 ASSERT(cxt->optype & ST_RETRIEVE, ("was performing a retrieve()"));
1141 AV *aseen = cxt->aseen;
1144 sv_free((SV *) aseen);
1148 AV *aclass = cxt->aclass;
1151 sv_free((SV *) aclass);
1155 HV *hook = cxt->hook;
1158 sv_free((SV *) hook);
1162 HV *hseen = cxt->hseen;
1165 sv_free((SV *) hseen); /* optional HV, for backward compat. */
1175 * A workaround for the CROAK bug: cleanup the last context.
1177 static void clean_context(cxt)
1180 TRACEME(("clean_context"));
1182 ASSERT(cxt->s_dirty, ("dirty context"));
1184 if (cxt->optype & ST_RETRIEVE)
1185 clean_retrieve_context(cxt);
1187 clean_store_context(cxt);
1189 ASSERT(!cxt->s_dirty, ("context is clean"));
1195 * Allocate a new context and push it on top of the parent one.
1196 * This new context is made globally visible via SET_STCXT().
1198 static stcxt_t *allocate_context(parent_cxt)
1199 stcxt_t *parent_cxt;
1203 TRACEME(("allocate_context"));
1205 ASSERT(!parent_cxt->s_dirty, ("parent context clean"));
1207 Newz(0, cxt, 1, stcxt_t);
1208 cxt->prev = parent_cxt;
1217 * Free current context, which cannot be the "root" one.
1218 * Make the context underneath globally visible via SET_STCXT().
1220 static void free_context(cxt)
1223 stcxt_t *prev = cxt->prev;
1225 TRACEME(("free_context"));
1227 ASSERT(!cxt->s_dirty, ("clean context"));
1228 ASSERT(prev, ("not freeing root context"));
1246 * Tells whether we're in the middle of a store operation.
1248 int is_storing(void)
1252 return cxt->entry && (cxt->optype & ST_STORE);
1258 * Tells whether we're in the middle of a retrieve operation.
1260 int is_retrieving(void)
1264 return cxt->entry && (cxt->optype & ST_RETRIEVE);
1268 * last_op_in_netorder
1270 * Returns whether last operation was made using network order.
1272 * This is typically out-of-band information that might prove useful
1273 * to people wishing to convert native to network order data when used.
1275 int last_op_in_netorder(void)
1279 return cxt->netorder;
1283 *** Hook lookup and calling routines.
1289 * A wrapper on gv_fetchmethod_autoload() which caches results.
1291 * Returns the routine reference as an SV*, or null if neither the package
1292 * nor its ancestors know about the method.
1294 static SV *pkg_fetchmeth(
1303 * The following code is the same as the one performed by UNIVERSAL::can
1307 gv = gv_fetchmethod_autoload(pkg, method, FALSE);
1308 if (gv && isGV(gv)) {
1309 sv = newRV((SV*) GvCV(gv));
1310 TRACEME(("%s->%s: 0x%"UVxf, HvNAME(pkg), method, PTR2UV(sv)));
1312 sv = newSVsv(&PL_sv_undef);
1313 TRACEME(("%s->%s: not found", HvNAME(pkg), method));
1317 * Cache the result, ignoring failure: if we can't store the value,
1318 * it just won't be cached.
1321 (void) hv_store(cache, HvNAME(pkg), strlen(HvNAME(pkg)), sv, 0);
1323 return SvOK(sv) ? sv : (SV *) 0;
1329 * Force cached value to be undef: hook ignored even if present.
1331 static void pkg_hide(
1336 (void) hv_store(cache,
1337 HvNAME(pkg), strlen(HvNAME(pkg)), newSVsv(&PL_sv_undef), 0);
1343 * Discard cached value: a whole fetch loop will be retried at next lookup.
1345 static void pkg_uncache(
1350 (void) hv_delete(cache, HvNAME(pkg), strlen(HvNAME(pkg)), G_DISCARD);
1356 * Our own "UNIVERSAL::can", which caches results.
1358 * Returns the routine reference as an SV*, or null if the object does not
1359 * know about the method.
1369 TRACEME(("pkg_can for %s->%s", HvNAME(pkg), method));
1372 * Look into the cache to see whether we already have determined
1373 * where the routine was, if any.
1375 * NOTA BENE: we don't use `method' at all in our lookup, since we know
1376 * that only one hook (i.e. always the same) is cached in a given cache.
1379 svh = hv_fetch(cache, HvNAME(pkg), strlen(HvNAME(pkg)), FALSE);
1383 TRACEME(("cached %s->%s: not found", HvNAME(pkg), method));
1386 TRACEME(("cached %s->%s: 0x%"UVxf,
1387 HvNAME(pkg), method, PTR2UV(sv)));
1392 TRACEME(("not cached yet"));
1393 return pkg_fetchmeth(cache, pkg, method); /* Fetch and cache */
1399 * Call routine as obj->hook(av) in scalar context.
1400 * Propagates the single returned value if not called in void context.
1402 static SV *scalar_call(
1413 TRACEME(("scalar_call (cloning=%d)", cloning));
1420 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1422 SV **ary = AvARRAY(av);
1423 int cnt = AvFILLp(av) + 1;
1425 XPUSHs(ary[0]); /* Frozen string */
1426 for (i = 1; i < cnt; i++) {
1427 TRACEME(("pushing arg #%d (0x%"UVxf")...",
1428 i, PTR2UV(ary[i])));
1429 XPUSHs(sv_2mortal(newRV(ary[i])));
1434 TRACEME(("calling..."));
1435 count = perl_call_sv(hook, flags); /* Go back to Perl code */
1436 TRACEME(("count = %d", count));
1442 SvREFCNT_inc(sv); /* We're returning it, must stay alive! */
1455 * Call routine obj->hook(cloning) in list context.
1456 * Returns the list of returned values in an array.
1458 static AV *array_call(
1468 TRACEME(("array_call (cloning=%d)", cloning));
1474 XPUSHs(obj); /* Target object */
1475 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1478 count = perl_call_sv(hook, G_ARRAY); /* Go back to Perl code */
1483 for (i = count - 1; i >= 0; i--) {
1485 av_store(av, i, SvREFCNT_inc(sv));
1498 * Lookup the class name in the `hclass' table and either assign it a new ID
1499 * or return the existing one, by filling in `classnum'.
1501 * Return true if the class was known, false if the ID was just generated.
1503 static int known_class(
1505 char *name, /* Class name */
1506 int len, /* Name length */
1510 HV *hclass = cxt->hclass;
1512 TRACEME(("known_class (%s)", name));
1515 * Recall that we don't store pointers in this hash table, but tags.
1516 * Therefore, we need LOW_32BITS() to extract the relevant parts.
1519 svh = hv_fetch(hclass, name, len, FALSE);
1521 *classnum = LOW_32BITS(*svh);
1526 * Unknown classname, we need to record it.
1530 if (!hv_store(hclass, name, len, INT2PTR(SV*, cxt->classnum), 0))
1531 CROAK(("Unable to record new classname"));
1533 *classnum = cxt->classnum;
1538 *** Sepcific store routines.
1544 * Store a reference.
1545 * Layout is SX_REF <object> or SX_OVERLOAD <object>.
1547 static int store_ref(stcxt_t *cxt, SV *sv)
1549 TRACEME(("store_ref (0x%"UVxf")", PTR2UV(sv)));
1552 * Follow reference, and check if target is overloaded.
1558 HV *stash = (HV *) SvSTASH(sv);
1559 if (stash && Gv_AMG(stash)) {
1560 TRACEME(("ref (0x%"UVxf") is overloaded", PTR2UV(sv)));
1561 PUTMARK(SX_OVERLOAD);
1567 return store(cxt, sv);
1575 * Layout is SX_LSCALAR <length> <data>, SX_SCALAR <lenght> <data> or SX_UNDEF.
1576 * The <data> section is omitted if <length> is 0.
1578 * If integer or double, the layout is SX_INTEGER <data> or SX_DOUBLE <data>.
1579 * Small integers (within [-127, +127]) are stored as SX_BYTE <byte>.
1581 static int store_scalar(stcxt_t *cxt, SV *sv)
1586 U32 flags = SvFLAGS(sv); /* "cc -O" may put it in register */
1588 TRACEME(("store_scalar (0x%"UVxf")", PTR2UV(sv)));
1591 * For efficiency, break the SV encapsulation by peaking at the flags
1592 * directly without using the Perl macros to avoid dereferencing
1593 * sv->sv_flags each time we wish to check the flags.
1596 if (!(flags & SVf_OK)) { /* !SvOK(sv) */
1597 if (sv == &PL_sv_undef) {
1598 TRACEME(("immortal undef"));
1599 PUTMARK(SX_SV_UNDEF);
1601 TRACEME(("undef at 0x%"UVxf, PTR2UV(sv)));
1608 * Always store the string representation of a scalar if it exists.
1609 * Gisle Aas provided me with this test case, better than a long speach:
1611 * perl -MDevel::Peek -le '$a="abc"; $a+0; Dump($a)'
1612 * SV = PVNV(0x80c8520)
1614 * FLAGS = (NOK,POK,pNOK,pPOK)
1617 * PV = 0x80c83d0 "abc"\0
1621 * Write SX_SCALAR, length, followed by the actual data.
1623 * Otherwise, write an SX_BYTE, SX_INTEGER or an SX_DOUBLE as
1624 * appropriate, followed by the actual (binary) data. A double
1625 * is written as a string if network order, for portability.
1627 * NOTE: instead of using SvNOK(sv), we test for SvNOKp(sv).
1628 * The reason is that when the scalar value is tainted, the SvNOK(sv)
1631 * The test for a read-only scalar with both POK and NOK set is meant
1632 * to quickly detect &PL_sv_yes and &PL_sv_no without having to pay the
1633 * address comparison for each scalar we store.
1636 #define SV_MAYBE_IMMORTAL (SVf_READONLY|SVf_POK|SVf_NOK)
1638 if ((flags & SV_MAYBE_IMMORTAL) == SV_MAYBE_IMMORTAL) {
1639 if (sv == &PL_sv_yes) {
1640 TRACEME(("immortal yes"));
1642 } else if (sv == &PL_sv_no) {
1643 TRACEME(("immortal no"));
1646 pv = SvPV(sv, len); /* We know it's SvPOK */
1647 goto string; /* Share code below */
1649 } else if (flags & SVp_POK) { /* SvPOKp(sv) => string */
1650 I32 wlen; /* For 64-bit machines */
1654 * Will come here from below with pv and len set if double & netorder,
1655 * or from above if it was readonly, POK and NOK but neither &PL_sv_yes
1660 wlen = (I32) len; /* WLEN via STORE_SCALAR expects I32 */
1662 STORE_UTF8STR(pv, wlen);
1664 STORE_SCALAR(pv, wlen);
1665 TRACEME(("ok (scalar 0x%"UVxf" '%s', length = %"IVdf")",
1666 PTR2UV(sv), SvPVX(sv), (IV)len));
1668 } else if (flags & SVp_NOK) { /* SvNOKp(sv) => double */
1672 * Watch for number being an integer in disguise.
1674 if (nv == (NV) (iv = I_V(nv))) {
1675 TRACEME(("double %"NVff" is actually integer %"IVdf, nv, iv));
1676 goto integer; /* Share code below */
1679 if (cxt->netorder) {
1680 TRACEME(("double %"NVff" stored as string", nv));
1682 goto string; /* Share code above */
1686 WRITE(&nv, sizeof(nv));
1688 TRACEME(("ok (double 0x%"UVxf", value = %"NVff")", PTR2UV(sv), nv));
1690 } else if (flags & SVp_IOK) { /* SvIOKp(sv) => integer */
1694 * Will come here from above with iv set if double is an integer.
1699 * Optimize small integers into a single byte, otherwise store as
1700 * a real integer (converted into network order if they asked).
1703 if (iv >= -128 && iv <= 127) {
1704 unsigned char siv = (unsigned char) (iv + 128); /* [0,255] */
1707 TRACEME(("small integer stored as %d", siv));
1708 } else if (cxt->netorder) {
1711 niv = (I32) htonl(iv);
1712 TRACEME(("using network order"));
1715 TRACEME(("as-is for network order"));
1720 PUTMARK(SX_INTEGER);
1721 WRITE(&iv, sizeof(iv));
1724 TRACEME(("ok (integer 0x%"UVxf", value = %"IVdf")", PTR2UV(sv), iv));
1727 CROAK(("Can't determine type of %s(0x%"UVxf")",
1728 sv_reftype(sv, FALSE),
1731 return 0; /* Ok, no recursion on scalars */
1739 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
1740 * Each item is stored as <object>.
1742 static int store_array(stcxt_t *cxt, AV *av)
1745 I32 len = av_len(av) + 1;
1749 TRACEME(("store_array (0x%"UVxf")", PTR2UV(av)));
1752 * Signal array by emitting SX_ARRAY, followed by the array length.
1757 TRACEME(("size = %d", len));
1760 * Now store each item recursively.
1763 for (i = 0; i < len; i++) {
1764 sav = av_fetch(av, i, 0);
1766 TRACEME(("(#%d) undef item", i));
1770 TRACEME(("(#%d) item", i));
1771 if ((ret = store(cxt, *sav)))
1775 TRACEME(("ok (array)"));
1784 * Borrowed from perl source file pp_ctl.c, where it is used by pp_sort.
1787 sortcmp(const void *a, const void *b)
1789 return sv_cmp(*(SV * const *) a, *(SV * const *) b);
1796 * Store an hash table.
1798 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
1799 * Values are stored as <object>.
1800 * Keys are stored as <length> <data>, the <data> section being omitted
1803 static int store_hash(stcxt_t *cxt, HV *hv)
1805 I32 len = HvKEYS(hv);
1811 TRACEME(("store_hash (0x%"UVxf")", PTR2UV(hv)));
1814 * Signal hash by emitting SX_HASH, followed by the table length.
1819 TRACEME(("size = %d", len));
1822 * Save possible iteration state via each() on that table.
1825 riter = HvRITER(hv);
1826 eiter = HvEITER(hv);
1830 * Now store each item recursively.
1832 * If canonical is defined to some true value then store each
1833 * key/value pair in sorted order otherwise the order is random.
1834 * Canonical order is irrelevant when a deep clone operation is performed.
1836 * Fetch the value from perl only once per store() operation, and only
1841 !(cxt->optype & ST_CLONE) && (cxt->canonical == 1 ||
1842 (cxt->canonical < 0 && (cxt->canonical =
1843 SvTRUE(perl_get_sv("Storable::canonical", TRUE)) ? 1 : 0)))
1846 * Storing in order, sorted by key.
1847 * Run through the hash, building up an array of keys in a
1848 * mortal array, sort the array and then run through the
1854 TRACEME(("using canonical order"));
1856 for (i = 0; i < len; i++) {
1857 HE *he = hv_iternext(hv);
1858 SV *key = hv_iterkeysv(he);
1859 av_store(av, AvFILLp(av)+1, key); /* av_push(), really */
1862 qsort((char *) AvARRAY(av), len, sizeof(SV *), sortcmp);
1864 for (i = 0; i < len; i++) {
1867 SV *key = av_shift(av);
1868 HE *he = hv_fetch_ent(hv, key, 0, 0);
1869 SV *val = HeVAL(he);
1871 return 1; /* Internal error, not I/O error */
1874 * Store value first.
1877 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
1879 if ((ret = store(cxt, val)))
1884 * Keys are written after values to make sure retrieval
1885 * can be optimal in terms of memory usage, where keys are
1886 * read into a fixed unique buffer called kbuf.
1887 * See retrieve_hash() for details.
1890 keyval = hv_iterkey(he, &keylen);
1891 TRACEME(("(#%d) key '%s'", i, keyval));
1894 WRITE(keyval, keylen);
1898 * Free up the temporary array
1907 * Storing in "random" order (in the order the keys are stored
1908 * within the the hash). This is the default and will be faster!
1911 for (i = 0; i < len; i++) {
1914 SV *val = hv_iternextsv(hv, &key, &len);
1917 return 1; /* Internal error, not I/O error */
1920 * Store value first.
1923 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
1925 if ((ret = store(cxt, val)))
1930 * Keys are written after values to make sure retrieval
1931 * can be optimal in terms of memory usage, where keys are
1932 * read into a fixed unique buffer called kbuf.
1933 * See retrieve_hash() for details.
1936 TRACEME(("(#%d) key '%s'", i, key));
1943 TRACEME(("ok (hash 0x%"UVxf")", PTR2UV(hv)));
1946 HvRITER(hv) = riter; /* Restore hash iterator state */
1947 HvEITER(hv) = eiter;
1955 * When storing a tied object (be it a tied scalar, array or hash), we lay out
1956 * a special mark, followed by the underlying tied object. For instance, when
1957 * dealing with a tied hash, we store SX_TIED_HASH <hash object>, where
1958 * <hash object> stands for the serialization of the tied hash.
1960 static int store_tied(stcxt_t *cxt, SV *sv)
1964 int svt = SvTYPE(sv);
1967 TRACEME(("store_tied (0x%"UVxf")", PTR2UV(sv)));
1970 * We have a small run-time penalty here because we chose to factorise
1971 * all tieds objects into the same routine, and not have a store_tied_hash,
1972 * a store_tied_array, etc...
1974 * Don't use a switch() statement, as most compilers don't optimize that
1975 * well for 2/3 values. An if() else if() cascade is just fine. We put
1976 * tied hashes first, as they are the most likely beasts.
1979 if (svt == SVt_PVHV) {
1980 TRACEME(("tied hash"));
1981 PUTMARK(SX_TIED_HASH); /* Introduces tied hash */
1982 } else if (svt == SVt_PVAV) {
1983 TRACEME(("tied array"));
1984 PUTMARK(SX_TIED_ARRAY); /* Introduces tied array */
1986 TRACEME(("tied scalar"));
1987 PUTMARK(SX_TIED_SCALAR); /* Introduces tied scalar */
1991 if (!(mg = mg_find(sv, mtype)))
1992 CROAK(("No magic '%c' found while storing tied %s", mtype,
1993 (svt == SVt_PVHV) ? "hash" :
1994 (svt == SVt_PVAV) ? "array" : "scalar"));
1997 * The mg->mg_obj found by mg_find() above actually points to the
1998 * underlying tied Perl object implementation. For instance, if the
1999 * original SV was that of a tied array, then mg->mg_obj is an AV.
2001 * Note that we store the Perl object as-is. We don't call its FETCH
2002 * method along the way. At retrieval time, we won't call its STORE
2003 * method either, but the tieing magic will be re-installed. In itself,
2004 * that ensures that the tieing semantics are preserved since futher
2005 * accesses on the retrieved object will indeed call the magic methods...
2008 if ((ret = store(cxt, mg->mg_obj)))
2011 TRACEME(("ok (tied)"));
2019 * Stores a reference to an item within a tied structure:
2021 * . \$h{key}, stores both the (tied %h) object and 'key'.
2022 * . \$a[idx], stores both the (tied @a) object and 'idx'.
2024 * Layout is therefore either:
2025 * SX_TIED_KEY <object> <key>
2026 * SX_TIED_IDX <object> <index>
2028 static int store_tied_item(stcxt_t *cxt, SV *sv)
2033 TRACEME(("store_tied_item (0x%"UVxf")", PTR2UV(sv)));
2035 if (!(mg = mg_find(sv, 'p')))
2036 CROAK(("No magic 'p' found while storing reference to tied item"));
2039 * We discriminate between \$h{key} and \$a[idx] via mg_ptr.
2043 TRACEME(("store_tied_item: storing a ref to a tied hash item"));
2044 PUTMARK(SX_TIED_KEY);
2045 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2047 if ((ret = store(cxt, mg->mg_obj)))
2050 TRACEME(("store_tied_item: storing PTR 0x%"UVxf, PTR2UV(mg->mg_ptr)));
2052 if ((ret = store(cxt, (SV *) mg->mg_ptr)))
2055 I32 idx = mg->mg_len;
2057 TRACEME(("store_tied_item: storing a ref to a tied array item "));
2058 PUTMARK(SX_TIED_IDX);
2059 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2061 if ((ret = store(cxt, mg->mg_obj)))
2064 TRACEME(("store_tied_item: storing IDX %d", idx));
2069 TRACEME(("ok (tied item)"));
2075 * store_hook -- dispatched manually, not via sv_store[]
2077 * The blessed SV is serialized by a hook.
2081 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2083 * where <flags> indicates how long <len>, <len2> and <len3> are, whether
2084 * the trailing part [] is present, the type of object (scalar, array or hash).
2085 * There is also a bit which says how the classname is stored between:
2090 * and when the <index> form is used (classname already seen), the "large
2091 * classname" bit in <flags> indicates how large the <index> is.
2093 * The serialized string returned by the hook is of length <len2> and comes
2094 * next. It is an opaque string for us.
2096 * Those <len3> object IDs which are listed last represent the extra references
2097 * not directly serialized by the hook, but which are linked to the object.
2099 * When recursion is mandated to resolve object-IDs not yet seen, we have
2100 * instead, with <header> being flags with bits set to indicate the object type
2101 * and that recursion was indeed needed:
2103 * SX_HOOK <header> <object> <header> <object> <flags>
2105 * that same header being repeated between serialized objects obtained through
2106 * recursion, until we reach flags indicating no recursion, at which point
2107 * we know we've resynchronized with a single layout, after <flags>.
2109 * When storing a blessed ref to a tied variable, the following format is
2112 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
2114 * The first <flags> indication carries an object of type SHT_EXTRA, and the
2115 * real object type is held in the <extra> flag. At the very end of the
2116 * serialization stream, the underlying magic object is serialized, just like
2117 * any other tied variable.
2119 static int store_hook(
2132 int count; /* really len3 + 1 */
2133 unsigned char flags;
2136 int recursed = 0; /* counts recursion */
2137 int obj_type; /* object type, on 2 bits */
2140 int clone = cxt->optype & ST_CLONE;
2141 char mtype = 0; /* for blessed ref to tied structures */
2142 unsigned char eflags = 0; /* used when object type is SHT_EXTRA */
2144 TRACEME(("store_hook, class \"%s\", tagged #%d", HvNAME(pkg), cxt->tagnum));
2147 * Determine object type on 2 bits.
2152 obj_type = SHT_SCALAR;
2155 obj_type = SHT_ARRAY;
2158 obj_type = SHT_HASH;
2162 * Produced by a blessed ref to a tied data structure, $o in the
2163 * following Perl code.
2167 * my $o = bless \%h, 'BAR';
2169 * Signal the tie-ing magic by setting the object type as SHT_EXTRA
2170 * (since we have only 2 bits in <flags> to store the type), and an
2171 * <extra> byte flag will be emitted after the FIRST <flags> in the
2172 * stream, carrying what we put in `eflags'.
2174 obj_type = SHT_EXTRA;
2175 switch (SvTYPE(sv)) {
2177 eflags = (unsigned char) SHT_THASH;
2181 eflags = (unsigned char) SHT_TARRAY;
2185 eflags = (unsigned char) SHT_TSCALAR;
2191 CROAK(("Unexpected object type (%d) in store_hook()", type));
2193 flags = SHF_NEED_RECURSE | obj_type;
2195 class = HvNAME(pkg);
2196 len = strlen(class);
2199 * To call the hook, we need to fake a call like:
2201 * $object->STORABLE_freeze($cloning);
2203 * but we don't have the $object here. For instance, if $object is
2204 * a blessed array, what we have in `sv' is the array, and we can't
2205 * call a method on those.
2207 * Therefore, we need to create a temporary reference to the object and
2208 * make the call on that reference.
2211 TRACEME(("about to call STORABLE_freeze on class %s", class));
2213 ref = newRV_noinc(sv); /* Temporary reference */
2214 av = array_call(ref, hook, clone); /* @a = $object->STORABLE_freeze($c) */
2216 SvREFCNT_dec(ref); /* Reclaim temporary reference */
2218 count = AvFILLp(av) + 1;
2219 TRACEME(("store_hook, array holds %d items", count));
2222 * If they return an empty list, it means they wish to ignore the
2223 * hook for this class (and not just this instance -- that's for them
2224 * to handle if they so wish).
2226 * Simply disable the cached entry for the hook (it won't be recomputed
2227 * since it's present in the cache) and recurse to store_blessed().
2232 * They must not change their mind in the middle of a serialization.
2235 if (hv_fetch(cxt->hclass, class, len, FALSE))
2236 CROAK(("Too late to ignore hooks for %s class \"%s\"",
2237 (cxt->optype & ST_CLONE) ? "cloning" : "storing", class));
2239 pkg_hide(cxt->hook, pkg, "STORABLE_freeze");
2241 ASSERT(!pkg_can(cxt->hook, pkg, "STORABLE_freeze"), ("hook invisible"));
2242 TRACEME(("ignoring STORABLE_freeze in class \"%s\"", class));
2244 return store_blessed(cxt, sv, type, pkg);
2248 * Get frozen string.
2252 pv = SvPV(ary[0], len2);
2255 * If they returned more than one item, we need to serialize some
2256 * extra references if not already done.
2258 * Loop over the array, starting at postion #1, and for each item,
2259 * ensure it is a reference, serialize it if not already done, and
2260 * replace the entry with the tag ID of the corresponding serialized
2263 * We CHEAT by not calling av_fetch() and read directly within the
2267 for (i = 1; i < count; i++) {
2271 AV *av_hook = cxt->hook_seen;
2274 CROAK(("Item #%d returned by STORABLE_freeze "
2275 "for %s is not a reference", i, class));
2276 xsv = SvRV(rsv); /* Follow ref to know what to look for */
2279 * Look in hseen and see if we have a tag already.
2280 * Serialize entry if not done already, and get its tag.
2283 if ((svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE)))
2284 goto sv_seen; /* Avoid moving code too far to the right */
2286 TRACEME(("listed object %d at 0x%"UVxf" is unknown", i-1, PTR2UV(xsv)));
2289 * We need to recurse to store that object and get it to be known
2290 * so that we can resolve the list of object-IDs at retrieve time.
2292 * The first time we do this, we need to emit the proper header
2293 * indicating that we recursed, and what the type of object is (the
2294 * object we're storing via a user-hook). Indeed, during retrieval,
2295 * we'll have to create the object before recursing to retrieve the
2296 * others, in case those would point back at that object.
2299 /* [SX_HOOK] <flags> [<extra>] <object>*/
2303 if (obj_type == SHT_EXTRA)
2308 if ((ret = store(cxt, xsv))) /* Given by hook for us to store */
2311 svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE);
2313 CROAK(("Could not serialize item #%d from hook in %s", i, class));
2316 * It was the first time we serialized `xsv'.
2318 * Keep this SV alive until the end of the serialization: if we
2319 * disposed of it right now by decrementing its refcount, and it was
2320 * a temporary value, some next temporary value allocated during
2321 * another STORABLE_freeze might take its place, and we'd wrongly
2322 * assume that new SV was already serialized, based on its presence
2325 * Therefore, push it away in cxt->hook_seen.
2328 av_store(av_hook, AvFILLp(av_hook)+1, SvREFCNT_inc(xsv));
2332 * Dispose of the REF they returned. If we saved the `xsv' away
2333 * in the array of returned SVs, that will not cause the underlying
2334 * referenced SV to be reclaimed.
2337 ASSERT(SvREFCNT(xsv) > 1, ("SV will survive disposal of its REF"));
2338 SvREFCNT_dec(rsv); /* Dispose of reference */
2341 * Replace entry with its tag (not a real SV, so no refcnt increment)
2345 TRACEME(("listed object %d at 0x%"UVxf" is tag #%"UVuf,
2346 i-1, PTR2UV(xsv), PTR2UV(*svh)));
2350 * Allocate a class ID if not already done.
2352 * This needs to be done after the recursion above, since at retrieval
2353 * time, we'll see the inner objects first. Many thanks to
2354 * Salvador Ortiz Garcia <sog@msg.com.mx> who spot that bug and
2355 * proposed the right fix. -- RAM, 15/09/2000
2358 if (!known_class(cxt, class, len, &classnum)) {
2359 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2360 classnum = -1; /* Mark: we must store classname */
2362 TRACEME(("already seen class %s, ID = %d", class, classnum));
2366 * Compute leading flags.
2370 if (((classnum == -1) ? len : classnum) > LG_SCALAR)
2371 flags |= SHF_LARGE_CLASSLEN;
2373 flags |= SHF_IDX_CLASSNAME;
2374 if (len2 > LG_SCALAR)
2375 flags |= SHF_LARGE_STRLEN;
2377 flags |= SHF_HAS_LIST;
2378 if (count > (LG_SCALAR + 1))
2379 flags |= SHF_LARGE_LISTLEN;
2382 * We're ready to emit either serialized form:
2384 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2385 * SX_HOOK <flags> <index> <len2> <str> [<len3> <object-IDs>]
2387 * If we recursed, the SX_HOOK has already been emitted.
2390 TRACEME(("SX_HOOK (recursed=%d) flags=0x%x "
2391 "class=%"IVdf" len=%"IVdf" len2=%"IVdf" len3=%d",
2392 recursed, flags, (IV)classnum, (IV)len, (IV)len2, count-1));
2394 /* SX_HOOK <flags> [<extra>] */
2398 if (obj_type == SHT_EXTRA)
2403 /* <len> <classname> or <index> */
2404 if (flags & SHF_IDX_CLASSNAME) {
2405 if (flags & SHF_LARGE_CLASSLEN)
2408 unsigned char cnum = (unsigned char) classnum;
2412 if (flags & SHF_LARGE_CLASSLEN)
2415 unsigned char clen = (unsigned char) len;
2418 WRITE(class, len); /* Final \0 is omitted */
2421 /* <len2> <frozen-str> */
2422 if (flags & SHF_LARGE_STRLEN) {
2423 I32 wlen2 = len2; /* STRLEN might be 8 bytes */
2424 WLEN(wlen2); /* Must write an I32 for 64-bit machines */
2426 unsigned char clen = (unsigned char) len2;
2430 WRITE(pv, len2); /* Final \0 is omitted */
2432 /* [<len3> <object-IDs>] */
2433 if (flags & SHF_HAS_LIST) {
2434 int len3 = count - 1;
2435 if (flags & SHF_LARGE_LISTLEN)
2438 unsigned char clen = (unsigned char) len3;
2443 * NOTA BENE, for 64-bit machines: the ary[i] below does not yield a
2444 * real pointer, rather a tag number, well under the 32-bit limit.
2447 for (i = 1; i < count; i++) {
2448 I32 tagval = htonl(LOW_32BITS(ary[i]));
2450 TRACEME(("object %d, tag #%d", i-1, ntohl(tagval)));
2455 * Free the array. We need extra care for indices after 0, since they
2456 * don't hold real SVs but integers cast.
2460 AvFILLp(av) = 0; /* Cheat, nothing after 0 interests us */
2465 * If object was tied, need to insert serialization of the magic object.
2468 if (obj_type == SHT_EXTRA) {
2471 if (!(mg = mg_find(sv, mtype))) {
2472 int svt = SvTYPE(sv);
2473 CROAK(("No magic '%c' found while storing ref to tied %s with hook",
2474 mtype, (svt == SVt_PVHV) ? "hash" :
2475 (svt == SVt_PVAV) ? "array" : "scalar"));
2478 TRACEME(("handling the magic object 0x%"UVxf" part of 0x%"UVxf,
2479 PTR2UV(mg->mg_obj), PTR2UV(sv)));
2485 if ((ret = store(cxt, mg->mg_obj)))
2493 * store_blessed -- dispatched manually, not via sv_store[]
2495 * Check whether there is a STORABLE_xxx hook defined in the class or in one
2496 * of its ancestors. If there is, then redispatch to store_hook();
2498 * Otherwise, the blessed SV is stored using the following layout:
2500 * SX_BLESS <flag> <len> <classname> <object>
2502 * where <flag> indicates whether <len> is stored on 0 or 4 bytes, depending
2503 * on the high-order bit in flag: if 1, then length follows on 4 bytes.
2504 * Otherwise, the low order bits give the length, thereby giving a compact
2505 * representation for class names less than 127 chars long.
2507 * Each <classname> seen is remembered and indexed, so that the next time
2508 * an object in the blessed in the same <classname> is stored, the following
2511 * SX_IX_BLESS <flag> <index> <object>
2513 * where <index> is the classname index, stored on 0 or 4 bytes depending
2514 * on the high-order bit in flag (same encoding as above for <len>).
2516 static int store_blessed(
2527 TRACEME(("store_blessed, type %d, class \"%s\"", type, HvNAME(pkg)));
2530 * Look for a hook for this blessed SV and redirect to store_hook()
2534 hook = pkg_can(cxt->hook, pkg, "STORABLE_freeze");
2536 return store_hook(cxt, sv, type, pkg, hook);
2539 * This is a blessed SV without any serialization hook.
2542 class = HvNAME(pkg);
2543 len = strlen(class);
2545 TRACEME(("blessed 0x%"UVxf" in %s, no hook: tagged #%d",
2546 PTR2UV(sv), class, cxt->tagnum));
2549 * Determine whether it is the first time we see that class name (in which
2550 * case it will be stored in the SX_BLESS form), or whether we already
2551 * saw that class name before (in which case the SX_IX_BLESS form will be
2555 if (known_class(cxt, class, len, &classnum)) {
2556 TRACEME(("already seen class %s, ID = %d", class, classnum));
2557 PUTMARK(SX_IX_BLESS);
2558 if (classnum <= LG_BLESS) {
2559 unsigned char cnum = (unsigned char) classnum;
2562 unsigned char flag = (unsigned char) 0x80;
2567 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2569 if (len <= LG_BLESS) {
2570 unsigned char clen = (unsigned char) len;
2573 unsigned char flag = (unsigned char) 0x80;
2575 WLEN(len); /* Don't BER-encode, this should be rare */
2577 WRITE(class, len); /* Final \0 is omitted */
2581 * Now emit the <object> part.
2584 return SV_STORE(type)(cxt, sv);
2590 * We don't know how to store the item we reached, so return an error condition.
2591 * (it's probably a GLOB, some CODE reference, etc...)
2593 * If they defined the `forgive_me' variable at the Perl level to some
2594 * true value, then don't croak, just warn, and store a placeholder string
2597 static int store_other(stcxt_t *cxt, SV *sv)
2600 static char buf[80];
2602 TRACEME(("store_other"));
2605 * Fetch the value from perl only once per store() operation.
2609 cxt->forgive_me == 0 ||
2610 (cxt->forgive_me < 0 && !(cxt->forgive_me =
2611 SvTRUE(perl_get_sv("Storable::forgive_me", TRUE)) ? 1 : 0))
2613 CROAK(("Can't store %s items", sv_reftype(sv, FALSE)));
2615 warn("Can't store item %s(0x%"UVxf")",
2616 sv_reftype(sv, FALSE), PTR2UV(sv));
2619 * Store placeholder string as a scalar instead...
2622 (void) sprintf(buf, "You lost %s(0x%"UVxf")%c", sv_reftype(sv, FALSE),
2623 PTR2UV(sv), (char)0);
2626 STORE_SCALAR(buf, len);
2627 TRACEME(("ok (dummy \"%s\", length = %"IVdf")", buf, len));
2633 *** Store driving routines
2639 * WARNING: partially duplicates Perl's sv_reftype for speed.
2641 * Returns the type of the SV, identified by an integer. That integer
2642 * may then be used to index the dynamic routine dispatch table.
2644 static int sv_type(SV *sv)
2646 switch (SvTYPE(sv)) {
2651 * No need to check for ROK, that can't be set here since there
2652 * is no field capable of hodling the xrv_rv reference.
2660 * Starting from SVt_PV, it is possible to have the ROK flag
2661 * set, the pointer to the other SV being either stored in
2662 * the xrv_rv (in the case of a pure SVt_RV), or as the
2663 * xpv_pv field of an SVt_PV and its heirs.
2665 * However, those SV cannot be magical or they would be an
2666 * SVt_PVMG at least.
2668 return SvROK(sv) ? svis_REF : svis_SCALAR;
2670 case SVt_PVLV: /* Workaround for perl5.004_04 "LVALUE" bug */
2671 if (SvRMAGICAL(sv) && (mg_find(sv, 'p')))
2672 return svis_TIED_ITEM;
2675 if (SvRMAGICAL(sv) && (mg_find(sv, 'q')))
2677 return SvROK(sv) ? svis_REF : svis_SCALAR;
2679 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
2683 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
2696 * Recursively store objects pointed to by the sv to the specified file.
2698 * Layout is <content> or SX_OBJECT <tagnum> if we reach an already stored
2699 * object (one for which storage has started -- it may not be over if we have
2700 * a self-referenced structure). This data set forms a stored <object>.
2702 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 */
3191 int clone = cxt->optype & ST_CLONE;
3193 unsigned int extra_type = 0;
3195 TRACEME(("retrieve_hook (#%d)", cxt->tagnum));
3196 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3199 * Read flags, which tell us about the type, and whether we need to recurse.
3205 * Create the (empty) object, and mark it as seen.
3207 * This must be done now, because tags are incremented, and during
3208 * serialization, the object tag was affected before recursion could
3212 obj_type = flags & SHF_TYPE_MASK;
3218 sv = (SV *) newAV();
3221 sv = (SV *) newHV();
3225 * Read <extra> flag to know the type of the object.
3226 * Record associated magic type for later.
3228 GETMARK(extra_type);
3229 switch (extra_type) {
3235 sv = (SV *) newAV();
3239 sv = (SV *) newHV();
3243 return retrieve_other(cxt, 0); /* Let it croak */
3247 return retrieve_other(cxt, 0); /* Let it croak */
3249 SEEN(sv, 0); /* Don't bless yet */
3252 * Whilst flags tell us to recurse, do so.
3254 * We don't need to remember the addresses returned by retrieval, because
3255 * all the references will be obtained through indirection via the object
3256 * tags in the object-ID list.
3259 while (flags & SHF_NEED_RECURSE) {
3260 TRACEME(("retrieve_hook recursing..."));
3261 rv = retrieve(cxt, 0);
3264 TRACEME(("retrieve_hook back with rv=0x%"UVxf,
3269 if (flags & SHF_IDX_CLASSNAME) {
3274 * Fetch index from `aclass'
3277 if (flags & SHF_LARGE_CLASSLEN)
3282 sva = av_fetch(cxt->aclass, idx, FALSE);
3284 CROAK(("Class name #%"IVdf" should have been seen already",
3287 class = SvPVX(*sva); /* We know it's a PV, by construction */
3288 TRACEME(("class ID %d => %s", idx, class));
3292 * Decode class name length and read that name.
3294 * NOTA BENE: even if the length is stored on one byte, we don't read
3295 * on the stack. Just like retrieve_blessed(), we limit the name to
3296 * LG_BLESS bytes. This is an arbitrary decision.
3299 if (flags & SHF_LARGE_CLASSLEN)
3304 if (len > LG_BLESS) {
3305 TRACEME(("** allocating %d bytes for class name", len+1));
3306 New(10003, class, len+1, char);
3310 class[len] = '\0'; /* Mark string end */
3313 * Record new classname.
3316 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3320 TRACEME(("class name: %s", class));
3323 * Decode user-frozen string length and read it in a SV.
3325 * For efficiency reasons, we read data directly into the SV buffer.
3326 * To understand that code, read retrieve_scalar()
3329 if (flags & SHF_LARGE_STRLEN)
3334 frozen = NEWSV(10002, len2);
3336 SAFEREAD(SvPVX(frozen), len2, frozen);
3337 SvCUR_set(frozen, len2);
3338 *SvEND(frozen) = '\0';
3340 (void) SvPOK_only(frozen); /* Validates string pointer */
3341 if (cxt->s_tainted) /* Is input source tainted? */
3344 TRACEME(("frozen string: %d bytes", len2));
3347 * Decode object-ID list length, if present.
3350 if (flags & SHF_HAS_LIST) {
3351 if (flags & SHF_LARGE_LISTLEN)
3357 av_extend(av, len3 + 1); /* Leave room for [0] */
3358 AvFILLp(av) = len3; /* About to be filled anyway */
3362 TRACEME(("has %d object IDs to link", len3));
3365 * Read object-ID list into array.
3366 * Because we pre-extended it, we can cheat and fill it manually.
3368 * We read object tags and we can convert them into SV* on the fly
3369 * because we know all the references listed in there (as tags)
3370 * have been already serialized, hence we have a valid correspondance
3371 * between each of those tags and the recreated SV.
3375 SV **ary = AvARRAY(av);
3377 for (i = 1; i <= len3; i++) { /* We leave [0] alone */
3384 svh = av_fetch(cxt->aseen, tag, FALSE);
3386 CROAK(("Object #%"IVdf" should have been retrieved already", (IV)tag));
3388 ary[i] = SvREFCNT_inc(xsv);
3393 * Bless the object and look up the STORABLE_thaw hook.
3397 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3400 * Hook not found. Maybe they did not require the module where this
3401 * hook is defined yet?
3403 * If the require below succeeds, we'll be able to find the hook.
3404 * Still, it only works reliably when each class is defined in a
3408 SV *psv = newSVpvn("require ", 8);
3409 sv_catpv(psv, class);
3411 TRACEME(("No STORABLE_thaw defined for objects of class %s", class));
3412 TRACEME(("Going to require module '%s' with '%s'", class, SvPVX(psv)));
3414 perl_eval_sv(psv, G_DISCARD);
3418 * We cache results of pkg_can, so we need to uncache before attempting
3422 pkg_uncache(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3423 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3426 CROAK(("No STORABLE_thaw defined for objects of class %s "
3427 "(even after a \"require %s;\")", class, class));
3431 * If we don't have an `av' yet, prepare one.
3432 * Then insert the frozen string as item [0].
3440 AvARRAY(av)[0] = SvREFCNT_inc(frozen);
3445 * $object->STORABLE_thaw($cloning, $frozen, @refs);
3447 * where $object is our blessed (empty) object, $cloning is a boolean
3448 * telling whether we're running a deep clone, $frozen is the frozen
3449 * string the user gave us in his serializing hook, and @refs, which may
3450 * be empty, is the list of extra references he returned along for us
3453 * In effect, the hook is an alternate creation routine for the class,
3454 * the object itself being already created by the runtime.
3457 TRACEME(("calling STORABLE_thaw on %s at 0x%"UVxf" (%"IVdf" args)",
3458 class, PTR2UV(sv), AvFILLp(av) + 1));
3461 (void) scalar_call(rv, hook, clone, av, G_SCALAR|G_DISCARD);
3468 SvREFCNT_dec(frozen);
3471 if (!(flags & SHF_IDX_CLASSNAME) && class != buf)
3475 * If we had an <extra> type, then the object was not as simple, and
3476 * we need to restore extra magic now.
3482 TRACEME(("retrieving magic object for 0x%"UVxf"...", PTR2UV(sv)));
3484 rv = retrieve(cxt, 0); /* Retrieve <magic object> */
3486 TRACEME(("restoring the magic object 0x%"UVxf" part of 0x%"UVxf,
3487 PTR2UV(rv), PTR2UV(sv)));
3489 switch (extra_type) {
3491 sv_upgrade(sv, SVt_PVMG);
3494 sv_upgrade(sv, SVt_PVAV);
3495 AvREAL_off((AV *)sv);
3498 sv_upgrade(sv, SVt_PVHV);
3501 CROAK(("Forgot to deal with extra type %d", extra_type));
3506 * Adding the magic only now, well after the STORABLE_thaw hook was called
3507 * means the hook cannot know it deals with an object whose variable is
3508 * tied. But this is happening when retrieving $o in the following case:
3512 * my $o = bless \%h, 'BAR';
3514 * The 'BAR' class is NOT the one where %h is tied into. Therefore, as
3515 * far as the 'BAR' class is concerned, the fact that %h is not a REAL
3516 * hash but a tied one should not matter at all, and remain transparent.
3517 * This means the magic must be restored by Storable AFTER the hook is
3520 * That looks very reasonable to me, but then I've come up with this
3521 * after a bug report from David Nesting, who was trying to store such
3522 * an object and caused Storable to fail. And unfortunately, it was
3523 * also the easiest way to retrofit support for blessed ref to tied objects
3524 * into the existing design. -- RAM, 17/02/2001
3527 sv_magic(sv, rv, mtype, Nullch, 0);
3528 SvREFCNT_dec(rv); /* Undo refcnt inc from sv_magic() */
3536 * Retrieve reference to some other scalar.
3537 * Layout is SX_REF <object>, with SX_REF already read.
3539 static SV *retrieve_ref(stcxt_t *cxt, char *cname)
3544 TRACEME(("retrieve_ref (#%d)", cxt->tagnum));
3547 * We need to create the SV that holds the reference to the yet-to-retrieve
3548 * object now, so that we may record the address in the seen table.
3549 * Otherwise, if the object to retrieve references us, we won't be able
3550 * to resolve the SX_OBJECT we'll see at that point! Hence we cannot
3551 * do the retrieve first and use rv = newRV(sv) since it will be too late
3552 * for SEEN() recording.
3555 rv = NEWSV(10002, 0);
3556 SEEN(rv, cname); /* Will return if rv is null */
3557 sv = retrieve(cxt, 0); /* Retrieve <object> */
3559 return (SV *) 0; /* Failed */
3562 * WARNING: breaks RV encapsulation.
3564 * Now for the tricky part. We have to upgrade our existing SV, so that
3565 * it is now an RV on sv... Again, we cheat by duplicating the code
3566 * held in newSVrv(), since we already got our SV from retrieve().
3570 * SvRV(rv) = SvREFCNT_inc(sv);
3572 * here because the reference count we got from retrieve() above is
3573 * already correct: if the object was retrieved from the file, then
3574 * its reference count is one. Otherwise, if it was retrieved via
3575 * an SX_OBJECT indication, a ref count increment was done.
3578 sv_upgrade(rv, SVt_RV);
3579 SvRV(rv) = sv; /* $rv = \$sv */
3582 TRACEME(("ok (retrieve_ref at 0x%"UVxf")", PTR2UV(rv)));
3588 * retrieve_overloaded
3590 * Retrieve reference to some other scalar with overloading.
3591 * Layout is SX_OVERLOAD <object>, with SX_OVERLOAD already read.
3593 static SV *retrieve_overloaded(stcxt_t *cxt, char *cname)
3599 TRACEME(("retrieve_overloaded (#%d)", cxt->tagnum));
3602 * Same code as retrieve_ref(), duplicated to avoid extra call.
3605 rv = NEWSV(10002, 0);
3606 SEEN(rv, cname); /* Will return if rv is null */
3607 sv = retrieve(cxt, 0); /* Retrieve <object> */
3609 return (SV *) 0; /* Failed */
3612 * WARNING: breaks RV encapsulation.
3615 sv_upgrade(rv, SVt_RV);
3616 SvRV(rv) = sv; /* $rv = \$sv */
3620 * Restore overloading magic.
3623 stash = (HV *) SvSTASH (sv);
3624 if (!stash || !Gv_AMG(stash))
3625 CROAK(("Cannot restore overloading on %s(0x%"UVxf") (package %s)",
3626 sv_reftype(sv, FALSE),
3628 stash ? HvNAME(stash) : "<unknown>"));
3632 TRACEME(("ok (retrieve_overloaded at 0x%"UVxf")", PTR2UV(rv)));
3638 * retrieve_tied_array
3640 * Retrieve tied array
3641 * Layout is SX_TIED_ARRAY <object>, with SX_TIED_ARRAY already read.
3643 static SV *retrieve_tied_array(stcxt_t *cxt, char *cname)
3648 TRACEME(("retrieve_tied_array (#%d)", cxt->tagnum));
3650 tv = NEWSV(10002, 0);
3651 SEEN(tv, cname); /* Will return if tv is null */
3652 sv = retrieve(cxt, 0); /* Retrieve <object> */
3654 return (SV *) 0; /* Failed */
3656 sv_upgrade(tv, SVt_PVAV);
3657 AvREAL_off((AV *)tv);
3658 sv_magic(tv, sv, 'P', Nullch, 0);
3659 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3661 TRACEME(("ok (retrieve_tied_array at 0x%"UVxf")", PTR2UV(tv)));
3667 * retrieve_tied_hash
3669 * Retrieve tied hash
3670 * Layout is SX_TIED_HASH <object>, with SX_TIED_HASH already read.
3672 static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname)
3677 TRACEME(("retrieve_tied_hash (#%d)", cxt->tagnum));
3679 tv = NEWSV(10002, 0);
3680 SEEN(tv, cname); /* Will return if tv is null */
3681 sv = retrieve(cxt, 0); /* Retrieve <object> */
3683 return (SV *) 0; /* Failed */
3685 sv_upgrade(tv, SVt_PVHV);
3686 sv_magic(tv, sv, 'P', Nullch, 0);
3687 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3689 TRACEME(("ok (retrieve_tied_hash at 0x%"UVxf")", PTR2UV(tv)));
3695 * retrieve_tied_scalar
3697 * Retrieve tied scalar
3698 * Layout is SX_TIED_SCALAR <object>, with SX_TIED_SCALAR already read.
3700 static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname)
3705 TRACEME(("retrieve_tied_scalar (#%d)", cxt->tagnum));
3707 tv = NEWSV(10002, 0);
3708 SEEN(tv, cname); /* Will return if rv is null */
3709 sv = retrieve(cxt, 0); /* Retrieve <object> */
3711 return (SV *) 0; /* Failed */
3713 sv_upgrade(tv, SVt_PVMG);
3714 sv_magic(tv, sv, 'q', Nullch, 0);
3715 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3717 TRACEME(("ok (retrieve_tied_scalar at 0x%"UVxf")", PTR2UV(tv)));
3725 * Retrieve reference to value in a tied hash.
3726 * Layout is SX_TIED_KEY <object> <key>, with SX_TIED_KEY already read.
3728 static SV *retrieve_tied_key(stcxt_t *cxt, char *cname)
3734 TRACEME(("retrieve_tied_key (#%d)", cxt->tagnum));
3736 tv = NEWSV(10002, 0);
3737 SEEN(tv, cname); /* Will return if tv is null */
3738 sv = retrieve(cxt, 0); /* Retrieve <object> */
3740 return (SV *) 0; /* Failed */
3742 key = retrieve(cxt, 0); /* Retrieve <key> */
3744 return (SV *) 0; /* Failed */
3746 sv_upgrade(tv, SVt_PVMG);
3747 sv_magic(tv, sv, 'p', (char *)key, HEf_SVKEY);
3748 SvREFCNT_dec(key); /* Undo refcnt inc from sv_magic() */
3749 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3757 * Retrieve reference to value in a tied array.
3758 * Layout is SX_TIED_IDX <object> <idx>, with SX_TIED_IDX already read.
3760 static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname)
3766 TRACEME(("retrieve_tied_idx (#%d)", cxt->tagnum));
3768 tv = NEWSV(10002, 0);
3769 SEEN(tv, cname); /* Will return if tv is null */
3770 sv = retrieve(cxt, 0); /* Retrieve <object> */
3772 return (SV *) 0; /* Failed */
3774 RLEN(idx); /* Retrieve <idx> */
3776 sv_upgrade(tv, SVt_PVMG);
3777 sv_magic(tv, sv, 'p', Nullch, idx);
3778 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3787 * Retrieve defined long (string) scalar.
3789 * Layout is SX_LSCALAR <length> <data>, with SX_LSCALAR already read.
3790 * The scalar is "long" in that <length> is larger than LG_SCALAR so it
3791 * was not stored on a single byte.
3793 static SV *retrieve_lscalar(stcxt_t *cxt, char *cname)
3799 TRACEME(("retrieve_lscalar (#%d), len = %"IVdf, cxt->tagnum, len));
3802 * Allocate an empty scalar of the suitable length.
3805 sv = NEWSV(10002, len);
3806 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
3809 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
3811 * Now, for efficiency reasons, read data directly inside the SV buffer,
3812 * and perform the SV final settings directly by duplicating the final
3813 * work done by sv_setpv. Since we're going to allocate lots of scalars
3814 * this way, it's worth the hassle and risk.
3817 SAFEREAD(SvPVX(sv), len, sv);
3818 SvCUR_set(sv, len); /* Record C string length */
3819 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
3820 (void) SvPOK_only(sv); /* Validate string pointer */
3821 if (cxt->s_tainted) /* Is input source tainted? */
3822 SvTAINT(sv); /* External data cannot be trusted */
3824 TRACEME(("large scalar len %"IVdf" '%s'", len, SvPVX(sv)));
3825 TRACEME(("ok (retrieve_lscalar at 0x%"UVxf")", PTR2UV(sv)));
3833 * Retrieve defined short (string) scalar.
3835 * Layout is SX_SCALAR <length> <data>, with SX_SCALAR already read.
3836 * The scalar is "short" so <length> is single byte. If it is 0, there
3837 * is no <data> section.
3839 static SV *retrieve_scalar(stcxt_t *cxt, char *cname)
3845 TRACEME(("retrieve_scalar (#%d), len = %d", cxt->tagnum, len));
3848 * Allocate an empty scalar of the suitable length.
3851 sv = NEWSV(10002, len);
3852 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
3855 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
3860 * newSV did not upgrade to SVt_PV so the scalar is undefined.
3861 * To make it defined with an empty length, upgrade it now...
3863 sv_upgrade(sv, SVt_PV);
3865 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
3866 TRACEME(("ok (retrieve_scalar empty at 0x%"UVxf")", PTR2UV(sv)));
3869 * Now, for efficiency reasons, read data directly inside the SV buffer,
3870 * and perform the SV final settings directly by duplicating the final
3871 * work done by sv_setpv. Since we're going to allocate lots of scalars
3872 * this way, it's worth the hassle and risk.
3874 SAFEREAD(SvPVX(sv), len, sv);
3875 SvCUR_set(sv, len); /* Record C string length */
3876 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
3877 TRACEME(("small scalar len %d '%s'", len, SvPVX(sv)));
3880 (void) SvPOK_only(sv); /* Validate string pointer */
3881 if (cxt->s_tainted) /* Is input source tainted? */
3882 SvTAINT(sv); /* External data cannot be trusted */
3884 TRACEME(("ok (retrieve_scalar at 0x%"UVxf")", PTR2UV(sv)));
3891 * Like retrieve_scalar(), but tag result as utf8.
3892 * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
3894 static SV *retrieve_utf8str(stcxt_t *cxt, char *cname)
3898 TRACEME(("retrieve_utf8str"));
3900 sv = retrieve_scalar(cxt, cname);
3910 * Like retrieve_lscalar(), but tag result as utf8.
3911 * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
3913 static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname)
3917 TRACEME(("retrieve_lutf8str"));
3919 sv = retrieve_lscalar(cxt, cname);
3929 * Retrieve defined integer.
3930 * Layout is SX_INTEGER <data>, whith SX_INTEGER already read.
3932 static SV *retrieve_integer(stcxt_t *cxt, char *cname)
3937 TRACEME(("retrieve_integer (#%d)", cxt->tagnum));
3939 READ(&iv, sizeof(iv));
3941 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
3943 TRACEME(("integer %"IVdf, iv));
3944 TRACEME(("ok (retrieve_integer at 0x%"UVxf")", PTR2UV(sv)));
3952 * Retrieve defined integer in network order.
3953 * Layout is SX_NETINT <data>, whith SX_NETINT already read.
3955 static SV *retrieve_netint(stcxt_t *cxt, char *cname)
3960 TRACEME(("retrieve_netint (#%d)", cxt->tagnum));
3964 sv = newSViv((int) ntohl(iv));
3965 TRACEME(("network integer %d", (int) ntohl(iv)));
3968 TRACEME(("network integer (as-is) %d", iv));
3970 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
3972 TRACEME(("ok (retrieve_netint at 0x%"UVxf")", PTR2UV(sv)));
3980 * Retrieve defined double.
3981 * Layout is SX_DOUBLE <data>, whith SX_DOUBLE already read.
3983 static SV *retrieve_double(stcxt_t *cxt, char *cname)
3988 TRACEME(("retrieve_double (#%d)", cxt->tagnum));
3990 READ(&nv, sizeof(nv));
3992 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
3994 TRACEME(("double %"NVff, nv));
3995 TRACEME(("ok (retrieve_double at 0x%"UVxf")", PTR2UV(sv)));
4003 * Retrieve defined byte (small integer within the [-128, +127] range).
4004 * Layout is SX_BYTE <data>, whith SX_BYTE already read.
4006 static SV *retrieve_byte(stcxt_t *cxt, char *cname)
4010 signed char tmp; /* must use temp var to work around
4011 an AIX compiler bug --H.Merijn Brand */
4013 TRACEME(("retrieve_byte (#%d)", cxt->tagnum));
4016 TRACEME(("small integer read as %d", (unsigned char) siv));
4017 tmp = ((unsigned char)siv) - 128;
4020 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4022 TRACEME(("byte %d", tmp));
4023 TRACEME(("ok (retrieve_byte at 0x%"UVxf")", PTR2UV(sv)));
4031 * Return the undefined value.
4033 static SV *retrieve_undef(stcxt_t *cxt, char *cname)
4037 TRACEME(("retrieve_undef"));
4048 * Return the immortal undefined value.
4050 static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname)
4052 SV *sv = &PL_sv_undef;
4054 TRACEME(("retrieve_sv_undef"));
4063 * Return the immortal yes value.
4065 static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname)
4067 SV *sv = &PL_sv_yes;
4069 TRACEME(("retrieve_sv_yes"));
4078 * Return the immortal no value.
4080 static SV *retrieve_sv_no(stcxt_t *cxt, char *cname)
4084 TRACEME(("retrieve_sv_no"));
4093 * Retrieve a whole array.
4094 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
4095 * Each item is stored as <object>.
4097 * When we come here, SX_ARRAY has been read already.
4099 static SV *retrieve_array(stcxt_t *cxt, char *cname)
4106 TRACEME(("retrieve_array (#%d)", cxt->tagnum));
4109 * Read length, and allocate array, then pre-extend it.
4113 TRACEME(("size = %d", len));
4115 SEEN(av, cname); /* Will return if array not allocated nicely */
4119 return (SV *) av; /* No data follow if array is empty */
4122 * Now get each item in turn...
4125 for (i = 0; i < len; i++) {
4126 TRACEME(("(#%d) item", i));
4127 sv = retrieve(cxt, 0); /* Retrieve item */
4130 if (av_store(av, i, sv) == 0)
4134 TRACEME(("ok (retrieve_array at 0x%"UVxf")", PTR2UV(av)));
4142 * Retrieve a whole hash table.
4143 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4144 * Keys are stored as <length> <data>, the <data> section being omitted
4146 * Values are stored as <object>.
4148 * When we come here, SX_HASH has been read already.
4150 static SV *retrieve_hash(stcxt_t *cxt, char *cname)
4158 TRACEME(("retrieve_hash (#%d)", cxt->tagnum));
4161 * Read length, allocate table.
4165 TRACEME(("size = %d", len));
4167 SEEN(hv, cname); /* Will return if table not allocated properly */
4169 return (SV *) hv; /* No data follow if table empty */
4172 * Now get each key/value pair in turn...
4175 for (i = 0; i < len; i++) {
4180 TRACEME(("(#%d) value", i));
4181 sv = retrieve(cxt, 0);
4187 * Since we're reading into kbuf, we must ensure we're not
4188 * recursing between the read and the hv_store() where it's used.
4189 * Hence the key comes after the value.
4192 RLEN(size); /* Get key size */
4193 KBUFCHK(size); /* Grow hash key read pool if needed */
4196 kbuf[size] = '\0'; /* Mark string end, just in case */
4197 TRACEME(("(#%d) key '%s'", i, kbuf));
4200 * Enter key/value pair into hash table.
4203 if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
4207 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4213 * old_retrieve_array
4215 * Retrieve a whole array in pre-0.6 binary format.
4217 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
4218 * Each item is stored as SX_ITEM <object> or SX_IT_UNDEF for "holes".
4220 * When we come here, SX_ARRAY has been read already.
4222 static SV *old_retrieve_array(stcxt_t *cxt, char *cname)
4230 TRACEME(("old_retrieve_array (#%d)", cxt->tagnum));
4233 * Read length, and allocate array, then pre-extend it.
4237 TRACEME(("size = %d", len));
4239 SEEN(av, 0); /* Will return if array not allocated nicely */
4243 return (SV *) av; /* No data follow if array is empty */
4246 * Now get each item in turn...
4249 for (i = 0; i < len; i++) {
4251 if (c == SX_IT_UNDEF) {
4252 TRACEME(("(#%d) undef item", i));
4253 continue; /* av_extend() already filled us with undef */
4256 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
4257 TRACEME(("(#%d) item", i));
4258 sv = retrieve(cxt, 0); /* Retrieve item */
4261 if (av_store(av, i, sv) == 0)
4265 TRACEME(("ok (old_retrieve_array at 0x%"UVxf")", PTR2UV(av)));
4273 * Retrieve a whole hash table in pre-0.6 binary format.
4275 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4276 * Keys are stored as SX_KEY <length> <data>, the <data> section being omitted
4278 * Values are stored as SX_VALUE <object> or SX_VL_UNDEF for "holes".
4280 * When we come here, SX_HASH has been read already.
4282 static SV *old_retrieve_hash(stcxt_t *cxt, char *cname)
4290 static SV *sv_h_undef = (SV *) 0; /* hv_store() bug */
4292 TRACEME(("old_retrieve_hash (#%d)", cxt->tagnum));
4295 * Read length, allocate table.
4299 TRACEME(("size = %d", len));
4301 SEEN(hv, 0); /* Will return if table not allocated properly */
4303 return (SV *) hv; /* No data follow if table empty */
4306 * Now get each key/value pair in turn...
4309 for (i = 0; i < len; i++) {
4315 if (c == SX_VL_UNDEF) {
4316 TRACEME(("(#%d) undef value", i));
4318 * Due to a bug in hv_store(), it's not possible to pass
4319 * &PL_sv_undef to hv_store() as a value, otherwise the
4320 * associated key will not be creatable any more. -- RAM, 14/01/97
4323 sv_h_undef = newSVsv(&PL_sv_undef);
4324 sv = SvREFCNT_inc(sv_h_undef);
4325 } else if (c == SX_VALUE) {
4326 TRACEME(("(#%d) value", i));
4327 sv = retrieve(cxt, 0);
4331 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
4335 * Since we're reading into kbuf, we must ensure we're not
4336 * recursing between the read and the hv_store() where it's used.
4337 * Hence the key comes after the value.
4342 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
4343 RLEN(size); /* Get key size */
4344 KBUFCHK(size); /* Grow hash key read pool if needed */
4347 kbuf[size] = '\0'; /* Mark string end, just in case */
4348 TRACEME(("(#%d) key '%s'", i, kbuf));
4351 * Enter key/value pair into hash table.
4354 if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
4358 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4364 *** Retrieval engine.
4370 * Make sure the stored data we're trying to retrieve has been produced
4371 * on an ILP compatible system with the same byteorder. It croaks out in
4372 * case an error is detected. [ILP = integer-long-pointer sizes]
4373 * Returns null if error is detected, &PL_sv_undef otherwise.
4375 * Note that there's no byte ordering info emitted when network order was
4376 * used at store time.
4378 static SV *magic_check(stcxt_t *cxt)
4381 char byteorder[256];
4383 int use_network_order;
4385 int version_minor = 0;
4387 TRACEME(("magic_check"));
4390 * The "magic number" is only for files, not when freezing in memory.
4394 STRLEN len = sizeof(magicstr) - 1;
4397 READ(buf, len); /* Not null-terminated */
4398 buf[len] = '\0'; /* Is now */
4400 if (0 == strcmp(buf, magicstr))
4404 * Try to read more bytes to check for the old magic number, which
4408 old_len = sizeof(old_magicstr) - 1;
4409 READ(&buf[len], old_len - len);
4410 buf[old_len] = '\0'; /* Is now null-terminated */
4412 if (strcmp(buf, old_magicstr))
4413 CROAK(("File is not a perl storable"));
4418 * Starting with 0.6, the "use_network_order" byte flag is also used to
4419 * indicate the version number of the binary, and therefore governs the
4420 * setting of sv_retrieve_vtbl. See magic_write().
4423 GETMARK(use_network_order);
4424 version_major = use_network_order >> 1;
4425 cxt->retrieve_vtbl = version_major ? sv_retrieve : sv_old_retrieve;
4427 TRACEME(("magic_check: netorder = 0x%x", use_network_order));
4431 * Starting with 0.7 (binary major 2), a full byte is dedicated to the
4432 * minor version of the protocol. See magic_write().
4435 if (version_major > 1)
4436 GETMARK(version_minor);
4438 cxt->ver_major = version_major;
4439 cxt->ver_minor = version_minor;
4441 TRACEME(("binary image version is %d.%d", version_major, version_minor));
4444 * Inter-operability sanity check: we can't retrieve something stored
4445 * using a format more recent than ours, because we have no way to
4446 * know what has changed, and letting retrieval go would mean a probable
4447 * failure reporting a "corrupted" storable file.
4451 version_major > STORABLE_BIN_MAJOR ||
4452 (version_major == STORABLE_BIN_MAJOR &&
4453 version_minor > STORABLE_BIN_MINOR)
4455 CROAK(("Storable binary image v%d.%d more recent than I am (v%d.%d)",
4456 version_major, version_minor,
4457 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
4460 * If they stored using network order, there's no byte ordering
4461 * information to check.
4464 if ((cxt->netorder = (use_network_order & 0x1)))
4465 return &PL_sv_undef; /* No byte ordering info */
4467 sprintf(byteorder, "%lx", (unsigned long) BYTEORDER);
4469 READ(buf, c); /* Not null-terminated */
4470 buf[c] = '\0'; /* Is now */
4472 if (strcmp(buf, byteorder))
4473 CROAK(("Byte order is not compatible"));
4475 GETMARK(c); /* sizeof(int) */
4476 if ((int) c != sizeof(int))
4477 CROAK(("Integer size is not compatible"));
4479 GETMARK(c); /* sizeof(long) */
4480 if ((int) c != sizeof(long))
4481 CROAK(("Long integer size is not compatible"));
4483 GETMARK(c); /* sizeof(char *) */
4484 if ((int) c != sizeof(char *))
4485 CROAK(("Pointer integer size is not compatible"));
4487 if (version_major >= 2 && version_minor >= 2) {
4488 GETMARK(c); /* sizeof(NV) */
4489 if ((int) c != sizeof(NV))
4490 CROAK(("Double size is not compatible"));
4493 return &PL_sv_undef; /* OK */
4499 * Recursively retrieve objects from the specified file and return their
4500 * root SV (which may be an AV or an HV for what we care).
4501 * Returns null if there is a problem.
4503 static SV *retrieve(stcxt_t *cxt, char *cname)
4509 TRACEME(("retrieve"));
4512 * Grab address tag which identifies the object if we are retrieving
4513 * an older format. Since the new binary format counts objects and no
4514 * longer explicitely tags them, we must keep track of the correspondance
4517 * The following section will disappear one day when the old format is
4518 * no longer supported, hence the final "goto" in the "if" block.
4521 if (cxt->hseen) { /* Retrieving old binary */
4523 if (cxt->netorder) {
4525 READ(&nettag, sizeof(I32)); /* Ordered sequence of I32 */
4526 tag = (stag_t) nettag;
4528 READ(&tag, sizeof(stag_t)); /* Original address of the SV */
4531 if (type == SX_OBJECT) {
4533 svh = hv_fetch(cxt->hseen, (char *) &tag, sizeof(tag), FALSE);
4535 CROAK(("Old tag 0x%"UVxf" should have been mapped already", (UV)tag));
4536 tagn = SvIV(*svh); /* Mapped tag number computed earlier below */
4539 * The following code is common with the SX_OBJECT case below.
4542 svh = av_fetch(cxt->aseen, tagn, FALSE);
4544 CROAK(("Object #%"IVdf" should have been retrieved already", (IV)tagn));
4546 TRACEME(("has retrieved #%d at 0x%"UVxf, tagn, PTR2UV(sv)));
4547 SvREFCNT_inc(sv); /* One more reference to this same sv */
4548 return sv; /* The SV pointer where object was retrieved */
4552 * Map new object, but don't increase tagnum. This will be done
4553 * by each of the retrieve_* functions when they call SEEN().
4555 * The mapping associates the "tag" initially present with a unique
4556 * tag number. See test for SX_OBJECT above to see how this is perused.
4559 if (!hv_store(cxt->hseen, (char *) &tag, sizeof(tag),
4560 newSViv(cxt->tagnum), 0))
4567 * Regular post-0.6 binary format.
4573 TRACEME(("retrieve type = %d", type));
4576 * Are we dealing with an object we should have already retrieved?
4579 if (type == SX_OBJECT) {
4583 svh = av_fetch(cxt->aseen, tag, FALSE);
4585 CROAK(("Object #%"IVdf" should have been retrieved already",
4588 TRACEME(("had retrieved #%d at 0x%"UVxf, tag, PTR2UV(sv)));
4589 SvREFCNT_inc(sv); /* One more reference to this same sv */
4590 return sv; /* The SV pointer where object was retrieved */
4593 first_time: /* Will disappear when support for old format is dropped */
4596 * Okay, first time through for this one.
4599 sv = RETRIEVE(cxt, type)(cxt, cname);
4601 return (SV *) 0; /* Failed */
4604 * Old binary formats (pre-0.7).
4606 * Final notifications, ended by SX_STORED may now follow.
4607 * Currently, the only pertinent notification to apply on the
4608 * freshly retrieved object is either:
4609 * SX_CLASS <char-len> <classname> for short classnames.
4610 * SX_LG_CLASS <int-len> <classname> for larger one (rare!).
4611 * Class name is then read into the key buffer pool used by
4612 * hash table key retrieval.
4615 if (cxt->ver_major < 2) {
4616 while ((type = GETCHAR()) != SX_STORED) {
4620 GETMARK(len); /* Length coded on a single char */
4622 case SX_LG_CLASS: /* Length coded on a regular integer */
4627 return (SV *) 0; /* Failed */
4629 KBUFCHK(len); /* Grow buffer as necessary */
4632 kbuf[len] = '\0'; /* Mark string end */
4637 TRACEME(("ok (retrieved 0x%"UVxf", refcnt=%d, %s)", PTR2UV(sv),
4638 SvREFCNT(sv) - 1, sv_reftype(sv, FALSE)));
4646 * Retrieve data held in file and return the root object.
4647 * Common routine for pretrieve and mretrieve.
4649 static SV *do_retrieve(
4656 int is_tainted; /* Is input source tainted? */
4657 struct extendable msave; /* Where potentially valid mbuf is saved */
4659 TRACEME(("do_retrieve (optype = 0x%x)", optype));
4661 optype |= ST_RETRIEVE;
4664 * Sanity assertions for retrieve dispatch tables.
4667 ASSERT(sizeof(sv_old_retrieve) == sizeof(sv_retrieve),
4668 ("old and new retrieve dispatch table have same size"));
4669 ASSERT(sv_old_retrieve[SX_ERROR] == retrieve_other,
4670 ("SX_ERROR entry correctly initialized in old dispatch table"));
4671 ASSERT(sv_retrieve[SX_ERROR] == retrieve_other,
4672 ("SX_ERROR entry correctly initialized in new dispatch table"));
4675 * Workaround for CROAK leak: if they enter with a "dirty" context,
4676 * free up memory for them now.
4683 * Now that STORABLE_xxx hooks exist, it is possible that they try to
4684 * re-enter retrieve() via the hooks.
4688 cxt = allocate_context(cxt);
4692 ASSERT(cxt->entry == 1, ("starting new recursion"));
4693 ASSERT(!cxt->s_dirty, ("clean context"));
4698 * Data is loaded into the memory buffer when f is NULL, unless `in' is
4699 * also NULL, in which case we're expecting the data to already lie
4700 * in the buffer (dclone case).
4703 KBUFINIT(); /* Allocate hash key reading pool once */
4706 StructCopy(&cxt->membuf, &msave, struct extendable);
4712 * Magic number verifications.
4714 * This needs to be done before calling init_retrieve_context()
4715 * since the format indication in the file are necessary to conduct
4716 * some of the initializations.
4719 cxt->fio = f; /* Where I/O are performed */
4721 if (!magic_check(cxt))
4722 CROAK(("Magic number checking on storable %s failed",
4723 cxt->fio ? "file" : "string"));
4725 TRACEME(("data stored in %s format",
4726 cxt->netorder ? "net order" : "native"));
4729 * Check whether input source is tainted, so that we don't wrongly
4730 * taint perfectly good values...
4732 * We assume file input is always tainted. If both `f' and `in' are
4733 * NULL, then we come from dclone, and tainted is already filled in
4734 * the context. That's a kludge, but the whole dclone() thing is
4735 * already quite a kludge anyway! -- RAM, 15/09/2000.
4738 is_tainted = f ? 1 : (in ? SvTAINTED(in) : cxt->s_tainted);
4739 TRACEME(("input source is %s", is_tainted ? "tainted" : "trusted"));
4740 init_retrieve_context(cxt, optype, is_tainted);
4742 ASSERT(is_retrieving(), ("within retrieve operation"));
4744 sv = retrieve(cxt, 0); /* Recursively retrieve object, get root SV */
4751 StructCopy(&msave, &cxt->membuf, struct extendable);
4754 * The "root" context is never freed.
4757 clean_retrieve_context(cxt);
4758 if (cxt->prev) /* This context was stacked */
4759 free_context(cxt); /* It was not the "root" context */
4762 * Prepare returned value.
4766 TRACEME(("retrieve ERROR"));
4767 return &PL_sv_undef; /* Something went wrong, return undef */
4770 TRACEME(("retrieve got %s(0x%"UVxf")",
4771 sv_reftype(sv, FALSE), PTR2UV(sv)));
4774 * Backward compatibility with Storable-0.5@9 (which we know we
4775 * are retrieving if hseen is non-null): don't create an extra RV
4776 * for objects since we special-cased it at store time.
4778 * Build a reference to the SV returned by pretrieve even if it is
4779 * already one and not a scalar, for consistency reasons.
4781 * NB: although context might have been cleaned, the value of `cxt->hseen'
4782 * remains intact, and can be used as a flag.
4785 if (cxt->hseen) { /* Was not handling overloading by then */
4787 if (sv_type(sv) == svis_REF && (rv = SvRV(sv)) && SvOBJECT(rv))
4792 * If reference is overloaded, restore behaviour.
4794 * NB: minor glitch here: normally, overloaded refs are stored specially
4795 * so that we can croak when behaviour cannot be re-installed, and also
4796 * avoid testing for overloading magic at each reference retrieval.
4798 * Unfortunately, the root reference is implicitely stored, so we must
4799 * check for possible overloading now. Furthermore, if we don't restore
4800 * overloading, we cannot croak as if the original ref was, because we
4801 * have no way to determine whether it was an overloaded ref or not in
4804 * It's a pity that overloading magic is attached to the rv, and not to
4805 * the underlying sv as blessing is.
4809 HV *stash = (HV *) SvSTASH (sv);
4810 SV *rv = newRV_noinc(sv);
4811 if (stash && Gv_AMG(stash)) {
4813 TRACEME(("restored overloading on root reference"));
4818 return newRV_noinc(sv);
4824 * Retrieve data held in file and return the root object, undef on error.
4826 SV *pretrieve(PerlIO *f)
4828 TRACEME(("pretrieve"));
4829 return do_retrieve(f, Nullsv, 0);
4835 * Retrieve data held in scalar and return the root object, undef on error.
4837 SV *mretrieve(SV *sv)
4839 TRACEME(("mretrieve"));
4840 return do_retrieve((PerlIO*) 0, sv, 0);
4850 * Deep clone: returns a fresh copy of the original referenced SV tree.
4852 * This is achieved by storing the object in memory and restoring from
4853 * there. Not that efficient, but it should be faster than doing it from
4860 stcxt_t *real_context;
4863 TRACEME(("dclone"));
4866 * Workaround for CROAK leak: if they enter with a "dirty" context,
4867 * free up memory for them now.
4874 * do_store() optimizes for dclone by not freeing its context, should
4875 * we need to allocate one because we're deep cloning from a hook.
4878 if (!do_store((PerlIO*) 0, sv, ST_CLONE, FALSE, (SV**) 0))
4879 return &PL_sv_undef; /* Error during store */
4882 * Because of the above optimization, we have to refresh the context,
4883 * since a new one could have been allocated and stacked by do_store().
4886 { dSTCXT; real_context = cxt; } /* Sub-block needed for macro */
4887 cxt = real_context; /* And we need this temporary... */
4890 * Now, `cxt' may refer to a new context.
4893 ASSERT(!cxt->s_dirty, ("clean context"));
4894 ASSERT(!cxt->entry, ("entry will not cause new context allocation"));
4897 TRACEME(("dclone stored %d bytes", size));
4901 * Since we're passing do_retrieve() both a NULL file and sv, we need
4902 * to pre-compute the taintedness of the input by setting cxt->tainted
4903 * to whatever state our own input string was. -- RAM, 15/09/2000
4905 * do_retrieve() will free non-root context.
4908 cxt->s_tainted = SvTAINTED(sv);
4909 out = do_retrieve((PerlIO*) 0, Nullsv, ST_CLONE);
4911 TRACEME(("dclone returns 0x%"UVxf, PTR2UV(out)));
4921 * The Perl IO GV object distinguishes between input and output for sockets
4922 * but not for plain files. To allow Storable to transparently work on
4923 * plain files and sockets transparently, we have to ask xsubpp to fetch the
4924 * right object for us. Hence the OutputStream and InputStream declarations.
4926 * Before perl 5.004_05, those entries in the standard typemap are not
4927 * defined in perl include files, so we do that here.
4930 #ifndef OutputStream
4931 #define OutputStream PerlIO *
4932 #define InputStream PerlIO *
4933 #endif /* !OutputStream */
4935 MODULE = Storable PACKAGE = Storable
4973 last_op_in_netorder()