2 * Store and retrieve mechanism.
6 * $Id: Storable.xs,v 1.0.1.7 2001/02/17 12:25:26 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 1.0.1.7 2001/02/17 12:25:26 ram
15 * patch8: now bless objects ASAP at retrieve time
16 * patch8: added support for blessed ref to tied structures
18 * Revision 1.0.1.6 2001/01/03 09:40:40 ram
19 * patch7: prototype and casting cleanup
20 * patch7: trace offending package when overloading cannot be restored
21 * patch7: made context cleanup safer to avoid dup freeing
23 * Revision 1.0.1.5 2000/11/05 17:21:24 ram
24 * patch6: fixed severe "object lost" bug for STORABLE_freeze returns
26 * Revision 1.0.1.4 2000/10/26 17:11:04 ram
27 * patch5: auto requires module of blessed ref when STORABLE_thaw misses
29 * Revision 1.0.1.3 2000/09/29 19:49:57 ram
30 * patch3: avoid using "tainted" and "dirty" since Perl remaps them via cpp
32 * $Log: Storable.xs,v $
33 * Revision 1.0 2000/09/01 19:40:41 ram
34 * Baseline for first official release.
40 #include <patchlevel.h> /* Perl's one, needed since 5.6 */
44 #define DEBUGME /* Debug mode, turns assertions on as well */
45 #define DASSERT /* Assertion mode */
49 * Pre PerlIO time when none of USE_PERLIO and PERLIO_IS_STDIO is defined
50 * Provide them with the necessary defines so they can build with pre-5.004.
53 #ifndef PERLIO_IS_STDIO
55 #define PerlIO_getc(x) getc(x)
56 #define PerlIO_putc(f,x) putc(x,f)
57 #define PerlIO_read(x,y,z) fread(y,1,z,x)
58 #define PerlIO_write(x,y,z) fwrite(y,1,z,x)
59 #define PerlIO_stdoutf printf
60 #endif /* PERLIO_IS_STDIO */
61 #endif /* USE_PERLIO */
64 * Earlier versions of perl might be used, we can't assume they have the latest!
67 #ifndef PERL_VERSION /* For perls < 5.6 */
68 #define PERL_VERSION PATCHLEVEL
70 #define newRV_noinc(sv) ((Sv = newRV(sv)), --SvREFCNT(SvRV(Sv)), Sv)
72 #if (PATCHLEVEL <= 4) /* Older perls (<= 5.004) lack PL_ namespace */
73 #define PL_sv_yes sv_yes
74 #define PL_sv_no sv_no
75 #define PL_sv_undef sv_undef
76 #if (SUBVERSION <= 4) /* 5.004_04 has been reported to lack newSVpvn */
77 #define newSVpvn newSVpv
79 #endif /* PATCHLEVEL <= 4 */
80 #ifndef HvSHAREKEYS_off
81 #define HvSHAREKEYS_off(hv) /* Ignore */
83 #ifndef AvFILLp /* Older perls (<=5.003) lack AvFILLp */
84 #define AvFILLp AvFILL
86 typedef double NV; /* Older perls lack the NV type */
87 #define IVdf "ld" /* Various printf formats for Perl types */
91 #define INT2PTR(t,v) (t)(IV)(v)
92 #define PTR2UV(v) (unsigned long)(v)
93 #endif /* PERL_VERSION -- perls < 5.6 */
95 #ifndef NVef /* The following were not part of perl 5.6 */
96 #if defined(USE_LONG_DOUBLE) && \
97 defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
98 #define NVef PERL_PRIeldbl
99 #define NVff PERL_PRIfldbl
100 #define NVgf PERL_PRIgldbl
110 * TRACEME() will only output things when the $Storable::DEBUGME is true.
113 #define TRACEME(x) do { \
114 if (SvTRUE(perl_get_sv("Storable::DEBUGME", TRUE))) \
115 { PerlIO_stdoutf x; PerlIO_stdoutf("\n"); } \
125 #define ASSERT(x,y) do { \
127 PerlIO_stdoutf("ASSERT FAILED (\"%s\", line %d): ", \
128 __FILE__, __LINE__); \
129 PerlIO_stdoutf y; PerlIO_stdoutf("\n"); \
140 #define C(x) ((char) (x)) /* For markers with dynamic retrieval handling */
142 #define SX_OBJECT C(0) /* Already stored object */
143 #define SX_LSCALAR C(1) /* Scalar (large binary) follows (length, data) */
144 #define SX_ARRAY C(2) /* Array forthcominng (size, item list) */
145 #define SX_HASH C(3) /* Hash forthcoming (size, key/value pair list) */
146 #define SX_REF C(4) /* Reference to object forthcoming */
147 #define SX_UNDEF C(5) /* Undefined scalar */
148 #define SX_INTEGER C(6) /* Integer forthcoming */
149 #define SX_DOUBLE C(7) /* Double forthcoming */
150 #define SX_BYTE C(8) /* (signed) byte forthcoming */
151 #define SX_NETINT C(9) /* Integer in network order forthcoming */
152 #define SX_SCALAR C(10) /* Scalar (binary, small) follows (length, data) */
153 #define SX_TIED_ARRAY C(11) /* Tied array forthcoming */
154 #define SX_TIED_HASH C(12) /* Tied hash forthcoming */
155 #define SX_TIED_SCALAR C(13) /* Tied scalar forthcoming */
156 #define SX_SV_UNDEF C(14) /* Perl's immortal PL_sv_undef */
157 #define SX_SV_YES C(15) /* Perl's immortal PL_sv_yes */
158 #define SX_SV_NO C(16) /* Perl's immortal PL_sv_no */
159 #define SX_BLESS C(17) /* Object is blessed */
160 #define SX_IX_BLESS C(18) /* Object is blessed, classname given by index */
161 #define SX_HOOK C(19) /* Stored via hook, user-defined */
162 #define SX_OVERLOAD C(20) /* Overloaded reference */
163 #define SX_TIED_KEY C(21) /* Tied magic key forthcoming */
164 #define SX_TIED_IDX C(22) /* Tied magic index forthcoming */
165 #define SX_UTF8STR C(23) /* UTF-8 string forthcoming (small) */
166 #define SX_LUTF8STR C(24) /* UTF-8 string forthcoming (large) */
167 #define SX_ERROR C(25) /* Error */
170 * Those are only used to retrieve "old" pre-0.6 binary images.
172 #define SX_ITEM 'i' /* An array item introducer */
173 #define SX_IT_UNDEF 'I' /* Undefined array item */
174 #define SX_KEY 'k' /* An hash key introducer */
175 #define SX_VALUE 'v' /* An hash value introducer */
176 #define SX_VL_UNDEF 'V' /* Undefined hash value */
179 * Those are only used to retrieve "old" pre-0.7 binary images
182 #define SX_CLASS 'b' /* Object is blessed, class name length <255 */
183 #define SX_LG_CLASS 'B' /* Object is blessed, class name length >255 */
184 #define SX_STORED 'X' /* End of object */
187 * Limits between short/long length representation.
190 #define LG_SCALAR 255 /* Large scalar length limit */
191 #define LG_BLESS 127 /* Large classname bless limit */
197 #define ST_STORE 0x1 /* Store operation */
198 #define ST_RETRIEVE 0x2 /* Retrieval operation */
199 #define ST_CLONE 0x4 /* Deep cloning operation */
202 * The following structure is used for hash table key retrieval. Since, when
203 * retrieving objects, we'll be facing blessed hash references, it's best
204 * to pre-allocate that buffer once and resize it as the need arises, never
205 * freeing it (keys will be saved away someplace else anyway, so even large
206 * keys are not enough a motivation to reclaim that space).
208 * This structure is also used for memory store/retrieve operations which
209 * happen in a fixed place before being malloc'ed elsewhere if persistency
210 * is required. Hence the aptr pointer.
213 char *arena; /* Will hold hash key strings, resized as needed */
214 STRLEN asiz; /* Size of aforementionned buffer */
215 char *aptr; /* Arena pointer, for in-place read/write ops */
216 char *aend; /* First invalid address */
221 * An hash table records the objects which have already been stored.
222 * Those are referred to as SX_OBJECT in the file, and their "tag" (i.e.
223 * an arbitrary sequence number) is used to identify them.
226 * An array table records the objects which have already been retrieved,
227 * as seen by the tag determind by counting the objects themselves. The
228 * reference to that retrieved object is kept in the table, and is returned
229 * when an SX_OBJECT is found bearing that same tag.
231 * The same processing is used to record "classname" for blessed objects:
232 * indexing by a hash at store time, and via an array at retrieve time.
235 typedef unsigned long stag_t; /* Used by pre-0.6 binary format */
238 * The following "thread-safe" related defines were contributed by
239 * Murray Nesbitt <murray@activestate.com> and integrated by RAM, who
240 * only renamed things a little bit to ensure consistency with surrounding
241 * code. -- RAM, 14/09/1999
243 * The original patch suffered from the fact that the stcxt_t structure
244 * was global. Murray tried to minimize the impact on the code as much as
247 * Starting with 0.7, Storable can be re-entrant, via the STORABLE_xxx hooks
248 * on objects. Therefore, the notion of context needs to be generalized,
252 #define MY_VERSION "Storable(" XS_VERSION ")"
255 * Fields s_tainted and s_dirty are prefixed with s_ because Perl's include
256 * files remap tainted and dirty when threading is enabled. That's bad for
257 * perl to remap such common words. -- RAM, 29/09/00
260 typedef struct stcxt {
261 int entry; /* flags recursion */
262 int optype; /* type of traversal operation */
263 HV *hseen; /* which objects have been seen, store time */
264 AV *hook_seen; /* which SVs were returned by STORABLE_freeze() */
265 AV *aseen; /* which objects have been seen, retrieve time */
266 HV *hclass; /* which classnames have been seen, store time */
267 AV *aclass; /* which classnames have been seen, retrieve time */
268 HV *hook; /* cache for hook methods per class name */
269 I32 tagnum; /* incremented at store time for each seen object */
270 I32 classnum; /* incremented at store time for each seen classname */
271 int netorder; /* true if network order used */
272 int s_tainted; /* true if input source is tainted, at retrieve time */
273 int forgive_me; /* whether to be forgiving... */
274 int canonical; /* whether to store hashes sorted by key */
275 int s_dirty; /* context is dirty due to CROAK() -- can be cleaned */
276 struct extendable keybuf; /* for hash key retrieval */
277 struct extendable membuf; /* for memory store/retrieve operations */
278 PerlIO *fio; /* where I/O are performed, NULL for memory */
279 int ver_major; /* major of version for retrieved object */
280 int ver_minor; /* minor of version for retrieved object */
281 SV *(**retrieve_vtbl)(); /* retrieve dispatch table */
282 struct stcxt *prev; /* contexts chained backwards in real recursion */
285 #if defined(MULTIPLICITY) || defined(PERL_OBJECT) || defined(PERL_CAPI)
287 #if (PATCHLEVEL <= 4) && (SUBVERSION < 68)
289 SV *perinterp_sv = perl_get_sv(MY_VERSION, FALSE)
290 #else /* >= perl5.004_68 */
292 SV *perinterp_sv = *hv_fetch(PL_modglobal, \
293 MY_VERSION, sizeof(MY_VERSION)-1, TRUE)
294 #endif /* < perl5.004_68 */
296 #define dSTCXT_PTR(T,name) \
297 T name = (perinterp_sv && SvIOK(perinterp_sv) \
298 ? INT2PTR(T, SvIVX(perinterp_sv)) : (T) 0)
301 dSTCXT_PTR(stcxt_t *, cxt)
305 Newz(0, cxt, 1, stcxt_t); \
306 sv_setiv(perinterp_sv, PTR2IV(cxt))
308 #define SET_STCXT(x) do { \
310 sv_setiv(perinterp_sv, PTR2IV(x)); \
313 #else /* !MULTIPLICITY && !PERL_OBJECT && !PERL_CAPI */
315 static stcxt_t Context;
316 static stcxt_t *Context_ptr = &Context;
317 #define dSTCXT stcxt_t *cxt = Context_ptr
318 #define INIT_STCXT dSTCXT
319 #define SET_STCXT(x) Context_ptr = x
321 #endif /* MULTIPLICITY || PERL_OBJECT || PERL_CAPI */
325 * Croaking implies a memory leak, since we don't use setjmp/longjmp
326 * to catch the exit and free memory used during store or retrieve
327 * operations. This is not too difficult to fix, but I need to understand
328 * how Perl does it, and croaking is exceptional anyway, so I lack the
329 * motivation to do it.
331 * The current workaround is to mark the context as dirty when croaking,
332 * so that data structures can be freed whenever we renter Storable code
333 * (but only *then*: it's a workaround, not a fix).
335 * This is also imperfect, because we don't really know how far they trapped
336 * the croak(), and when we were recursing, we won't be able to clean anything
337 * but the topmost context stacked.
340 #define CROAK(x) do { cxt->s_dirty = 1; croak x; } while (0)
343 * End of "thread-safe" related definitions.
349 * Keep only the low 32 bits of a pointer (used for tags, which are not
354 #define LOW_32BITS(x) ((I32) (x))
356 #define LOW_32BITS(x) ((I32) ((unsigned long) (x) & 0xffffffffUL))
362 * Hack for Crays, where sizeof(I32) == 8, and which are big-endians.
363 * Used in the WLEN and RLEN macros.
367 #define oI(x) ((I32 *) ((char *) (x) + 4))
368 #define oS(x) ((x) - 4)
369 #define oC(x) (x = 0)
378 * key buffer handling
380 #define kbuf (cxt->keybuf).arena
381 #define ksiz (cxt->keybuf).asiz
382 #define KBUFINIT() do { \
384 TRACEME(("** allocating kbuf of 128 bytes")); \
385 New(10003, kbuf, 128, char); \
389 #define KBUFCHK(x) do { \
391 TRACEME(("** extending kbuf to %d bytes", x+1)); \
392 Renew(kbuf, x+1, char); \
398 * memory buffer handling
400 #define mbase (cxt->membuf).arena
401 #define msiz (cxt->membuf).asiz
402 #define mptr (cxt->membuf).aptr
403 #define mend (cxt->membuf).aend
405 #define MGROW (1 << 13)
406 #define MMASK (MGROW - 1)
408 #define round_mgrow(x) \
409 ((unsigned long) (((unsigned long) (x) + MMASK) & ~MMASK))
410 #define trunc_int(x) \
411 ((unsigned long) ((unsigned long) (x) & ~(sizeof(int)-1)))
412 #define int_aligned(x) \
413 ((unsigned long) (x) == trunc_int(x))
415 #define MBUF_INIT(x) do { \
417 TRACEME(("** allocating mbase of %d bytes", MGROW)); \
418 New(10003, mbase, MGROW, char); \
425 mend = mbase + msiz; \
428 #define MBUF_TRUNC(x) mptr = mbase + x
429 #define MBUF_SIZE() (mptr - mbase)
432 * Use SvPOKp(), because SvPOK() fails on tainted scalars.
433 * See store_scalar() for other usage of this workaround.
435 #define MBUF_LOAD(v) do { \
437 CROAK(("Not a scalar string")); \
438 mptr = mbase = SvPV(v, msiz); \
439 mend = mbase + msiz; \
442 #define MBUF_XTEND(x) do { \
443 int nsz = (int) round_mgrow((x)+msiz); \
444 int offset = mptr - mbase; \
445 TRACEME(("** extending mbase to %d bytes", nsz)); \
446 Renew(mbase, nsz, char); \
448 mptr = mbase + offset; \
449 mend = mbase + nsz; \
452 #define MBUF_CHK(x) do { \
453 if ((mptr + (x)) > mend) \
457 #define MBUF_GETC(x) do { \
459 x = (int) (unsigned char) *mptr++; \
465 #define MBUF_GETINT(x) do { \
467 if ((mptr + 4) <= mend) { \
468 memcpy(oI(&x), mptr, 4); \
474 #define MBUF_GETINT(x) do { \
475 if ((mptr + sizeof(int)) <= mend) { \
476 if (int_aligned(mptr)) \
479 memcpy(&x, mptr, sizeof(int)); \
480 mptr += sizeof(int); \
486 #define MBUF_READ(x,s) do { \
487 if ((mptr + (s)) <= mend) { \
488 memcpy(x, mptr, s); \
494 #define MBUF_SAFEREAD(x,s,z) do { \
495 if ((mptr + (s)) <= mend) { \
496 memcpy(x, mptr, s); \
504 #define MBUF_PUTC(c) do { \
506 *mptr++ = (char) c; \
509 *mptr++ = (char) c; \
514 #define MBUF_PUTINT(i) do { \
516 memcpy(mptr, oI(&i), 4); \
520 #define MBUF_PUTINT(i) do { \
521 MBUF_CHK(sizeof(int)); \
522 if (int_aligned(mptr)) \
525 memcpy(mptr, &i, sizeof(int)); \
526 mptr += sizeof(int); \
530 #define MBUF_WRITE(x,s) do { \
532 memcpy(mptr, x, s); \
537 * Possible return values for sv_type().
541 #define svis_SCALAR 1
545 #define svis_TIED_ITEM 5
552 #define SHF_TYPE_MASK 0x03
553 #define SHF_LARGE_CLASSLEN 0x04
554 #define SHF_LARGE_STRLEN 0x08
555 #define SHF_LARGE_LISTLEN 0x10
556 #define SHF_IDX_CLASSNAME 0x20
557 #define SHF_NEED_RECURSE 0x40
558 #define SHF_HAS_LIST 0x80
561 * Types for SX_HOOK (last 2 bits in flags).
567 #define SHT_EXTRA 3 /* Read extra byte for type */
570 * The following are held in the "extra byte"...
573 #define SHT_TSCALAR 4 /* 4 + 0 -- tied scalar */
574 #define SHT_TARRAY 5 /* 4 + 1 -- tied array */
575 #define SHT_THASH 6 /* 4 + 2 -- tied hash */
578 * Before 0.6, the magic string was "perl-store" (binary version number 0).
580 * Since 0.6 introduced many binary incompatibilities, the magic string has
581 * been changed to "pst0" to allow an old image to be properly retrieved by
582 * a newer Storable, but ensure a newer image cannot be retrieved with an
585 * At 0.7, objects are given the ability to serialize themselves, and the
586 * set of markers is extended, backward compatibility is not jeopardized,
587 * so the binary version number could have remained unchanged. To correctly
588 * spot errors if a file making use of 0.7-specific extensions is given to
589 * 0.6 for retrieval, the binary version was moved to "2". And I'm introducing
590 * a "minor" version, to better track this kind of evolution from now on.
593 static char old_magicstr[] = "perl-store"; /* Magic number before 0.6 */
594 static char magicstr[] = "pst0"; /* Used as a magic number */
596 #define STORABLE_BIN_MAJOR 2 /* Binary major "version" */
597 #define STORABLE_BIN_MINOR 4 /* Binary minor "version" */
600 * Useful store shortcuts...
603 #define PUTMARK(x) do { \
606 else if (PerlIO_putc(cxt->fio, x) == EOF) \
610 #define WRITE_I32(x) do { \
611 ASSERT(sizeof(x) == sizeof(I32), ("writing an I32")); \
614 else if (PerlIO_write(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
619 #define WLEN(x) do { \
620 if (cxt->netorder) { \
621 int y = (int) htonl(x); \
624 else if (PerlIO_write(cxt->fio,oI(&y),oS(sizeof(y))) != oS(sizeof(y))) \
629 else if (PerlIO_write(cxt->fio,oI(&x),oS(sizeof(x))) != oS(sizeof(x))) \
634 #define WLEN(x) WRITE_I32(x)
637 #define WRITE(x,y) do { \
640 else if (PerlIO_write(cxt->fio, x, y) != y) \
644 #define STORE_PV_LEN(pv, len, small, large) do { \
645 if (len <= LG_SCALAR) { \
646 unsigned char clen = (unsigned char) len; \
658 #define STORE_SCALAR(pv, len) STORE_PV_LEN(pv, len, SX_SCALAR, SX_LSCALAR)
661 * Conditional UTF8 support.
662 * On non-UTF8 perls, UTF8 strings are returned as normal strings.
666 #define STORE_UTF8STR(pv, len) STORE_PV_LEN(pv, len, SX_UTF8STR, SX_LUTF8STR)
669 #define STORE_UTF8STR(pv, len) CROAK(("panic: storing UTF8 in non-UTF8 perl"))
670 #define SvUTF8_on(sv) CROAK(("Cannot retrieve UTF8 data in non-UTF8 perl"))
674 * Store undef in arrays and hashes without recursing through store().
676 #define STORE_UNDEF() do { \
682 * Useful retrieve shortcuts...
686 (cxt->fio ? PerlIO_getc(cxt->fio) : (mptr >= mend ? EOF : (int) *mptr++))
688 #define GETMARK(x) do { \
691 else if ((int) (x = PerlIO_getc(cxt->fio)) == EOF) \
695 #define READ_I32(x) do { \
696 ASSERT(sizeof(x) == sizeof(I32), ("reading an I32")); \
700 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
705 #define RLEN(x) do { \
709 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
712 x = (int) ntohl(x); \
715 #define RLEN(x) READ_I32(x)
718 #define READ(x,y) do { \
721 else if (PerlIO_read(cxt->fio, x, y) != y) \
725 #define SAFEREAD(x,y,z) do { \
727 MBUF_SAFEREAD(x,y,z); \
728 else if (PerlIO_read(cxt->fio, x, y) != y) { \
735 * This macro is used at retrieve time, to remember where object 'y', bearing a
736 * given tag 'tagnum', has been retrieved. Next time we see an SX_OBJECT marker,
737 * we'll therefore know where it has been retrieved and will be able to
738 * share the same reference, as in the original stored memory image.
740 * We also need to bless objects ASAP for hooks (which may compute "ref $x"
741 * on the objects given to STORABLE_thaw and expect that to be defined), and
742 * also for overloaded objects (for which we might not find the stash if the
743 * object is not blessed yet--this might occur for overloaded objects that
744 * refer to themselves indirectly: if we blessed upon return from a sub
745 * retrieve(), the SX_OBJECT marker we'd found could not have overloading
746 * restored on it because the underlying object would not be blessed yet!).
748 * To achieve that, the class name of the last retrieved object is passed down
749 * recursively, and the first SEEN() call for which the class name is not NULL
750 * will bless the object.
752 #define SEEN(y,c) do { \
755 if (av_store(cxt->aseen, cxt->tagnum++, SvREFCNT_inc(y)) == 0) \
757 TRACEME(("aseen(#%d) = 0x%"UVxf" (refcnt=%d)", cxt->tagnum-1, \
758 PTR2UV(y), SvREFCNT(y)-1)); \
760 BLESS((SV *) (y), c); \
764 * Bless `s' in `p', via a temporary reference, required by sv_bless().
766 #define BLESS(s,p) do { \
769 TRACEME(("blessing 0x%"UVxf" in %s", PTR2UV(s), (p))); \
770 stash = gv_stashpv((p), TRUE); \
771 ref = newRV_noinc(s); \
772 (void) sv_bless(ref, stash); \
778 static SV *retrieve(stcxt_t *cxt, char *cname);
781 * Dynamic dispatching table for SV store.
784 static int store_ref(stcxt_t *cxt, SV *sv);
785 static int store_scalar(stcxt_t *cxt, SV *sv);
786 static int store_array(stcxt_t *cxt, AV *av);
787 static int store_hash(stcxt_t *cxt, HV *hv);
788 static int store_tied(stcxt_t *cxt, SV *sv);
789 static int store_tied_item(stcxt_t *cxt, SV *sv);
790 static int store_other(stcxt_t *cxt, SV *sv);
791 static int store_blessed(stcxt_t *cxt, SV *sv, int type, HV *pkg);
793 static int (*sv_store[])(stcxt_t *cxt, SV *sv) = {
794 store_ref, /* svis_REF */
795 store_scalar, /* svis_SCALAR */
796 (int (*)(stcxt_t *cxt, SV *sv)) store_array, /* svis_ARRAY */
797 (int (*)(stcxt_t *cxt, SV *sv)) store_hash, /* svis_HASH */
798 store_tied, /* svis_TIED */
799 store_tied_item, /* svis_TIED_ITEM */
800 store_other, /* svis_OTHER */
803 #define SV_STORE(x) (*sv_store[x])
806 * Dynamic dispatching tables for SV retrieval.
809 static SV *retrieve_lscalar(stcxt_t *cxt, char *cname);
810 static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname);
811 static SV *old_retrieve_array(stcxt_t *cxt, char *cname);
812 static SV *old_retrieve_hash(stcxt_t *cxt, char *cname);
813 static SV *retrieve_ref(stcxt_t *cxt, char *cname);
814 static SV *retrieve_undef(stcxt_t *cxt, char *cname);
815 static SV *retrieve_integer(stcxt_t *cxt, char *cname);
816 static SV *retrieve_double(stcxt_t *cxt, char *cname);
817 static SV *retrieve_byte(stcxt_t *cxt, char *cname);
818 static SV *retrieve_netint(stcxt_t *cxt, char *cname);
819 static SV *retrieve_scalar(stcxt_t *cxt, char *cname);
820 static SV *retrieve_utf8str(stcxt_t *cxt, char *cname);
821 static SV *retrieve_tied_array(stcxt_t *cxt, char *cname);
822 static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname);
823 static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname);
824 static SV *retrieve_other(stcxt_t *cxt, char *cname);
826 static SV *(*sv_old_retrieve[])(stcxt_t *cxt, char *cname) = {
827 0, /* SX_OBJECT -- entry unused dynamically */
828 retrieve_lscalar, /* SX_LSCALAR */
829 old_retrieve_array, /* SX_ARRAY -- for pre-0.6 binaries */
830 old_retrieve_hash, /* SX_HASH -- for pre-0.6 binaries */
831 retrieve_ref, /* SX_REF */
832 retrieve_undef, /* SX_UNDEF */
833 retrieve_integer, /* SX_INTEGER */
834 retrieve_double, /* SX_DOUBLE */
835 retrieve_byte, /* SX_BYTE */
836 retrieve_netint, /* SX_NETINT */
837 retrieve_scalar, /* SX_SCALAR */
838 retrieve_tied_array, /* SX_ARRAY */
839 retrieve_tied_hash, /* SX_HASH */
840 retrieve_tied_scalar, /* SX_SCALAR */
841 retrieve_other, /* SX_SV_UNDEF not supported */
842 retrieve_other, /* SX_SV_YES not supported */
843 retrieve_other, /* SX_SV_NO not supported */
844 retrieve_other, /* SX_BLESS not supported */
845 retrieve_other, /* SX_IX_BLESS not supported */
846 retrieve_other, /* SX_HOOK not supported */
847 retrieve_other, /* SX_OVERLOADED not supported */
848 retrieve_other, /* SX_TIED_KEY not supported */
849 retrieve_other, /* SX_TIED_IDX not supported */
850 retrieve_other, /* SX_UTF8STR not supported */
851 retrieve_other, /* SX_LUTF8STR not supported */
852 retrieve_other, /* SX_ERROR */
855 static SV *retrieve_array(stcxt_t *cxt, char *cname);
856 static SV *retrieve_hash(stcxt_t *cxt, char *cname);
857 static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname);
858 static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname);
859 static SV *retrieve_sv_no(stcxt_t *cxt, char *cname);
860 static SV *retrieve_blessed(stcxt_t *cxt, char *cname);
861 static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname);
862 static SV *retrieve_hook(stcxt_t *cxt, char *cname);
863 static SV *retrieve_overloaded(stcxt_t *cxt, char *cname);
864 static SV *retrieve_tied_key(stcxt_t *cxt, char *cname);
865 static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname);
867 static SV *(*sv_retrieve[])(stcxt_t *cxt, char *cname) = {
868 0, /* SX_OBJECT -- entry unused dynamically */
869 retrieve_lscalar, /* SX_LSCALAR */
870 retrieve_array, /* SX_ARRAY */
871 retrieve_hash, /* SX_HASH */
872 retrieve_ref, /* SX_REF */
873 retrieve_undef, /* SX_UNDEF */
874 retrieve_integer, /* SX_INTEGER */
875 retrieve_double, /* SX_DOUBLE */
876 retrieve_byte, /* SX_BYTE */
877 retrieve_netint, /* SX_NETINT */
878 retrieve_scalar, /* SX_SCALAR */
879 retrieve_tied_array, /* SX_ARRAY */
880 retrieve_tied_hash, /* SX_HASH */
881 retrieve_tied_scalar, /* SX_SCALAR */
882 retrieve_sv_undef, /* SX_SV_UNDEF */
883 retrieve_sv_yes, /* SX_SV_YES */
884 retrieve_sv_no, /* SX_SV_NO */
885 retrieve_blessed, /* SX_BLESS */
886 retrieve_idx_blessed, /* SX_IX_BLESS */
887 retrieve_hook, /* SX_HOOK */
888 retrieve_overloaded, /* SX_OVERLOAD */
889 retrieve_tied_key, /* SX_TIED_KEY */
890 retrieve_tied_idx, /* SX_TIED_IDX */
891 retrieve_utf8str, /* SX_UTF8STR */
892 retrieve_lutf8str, /* SX_LUTF8STR */
893 retrieve_other, /* SX_ERROR */
896 #define RETRIEVE(c,x) (*(c)->retrieve_vtbl[(x) >= SX_ERROR ? SX_ERROR : (x)])
898 static SV *mbuf2sv(void);
901 *** Context management.
907 * Called once per "thread" (interpreter) to initialize some global context.
909 static void init_perinterp(void)
913 cxt->netorder = 0; /* true if network order used */
914 cxt->forgive_me = -1; /* whether to be forgiving... */
920 * Initialize a new store context for real recursion.
922 static void init_store_context(
928 TRACEME(("init_store_context"));
930 cxt->netorder = network_order;
931 cxt->forgive_me = -1; /* Fetched from perl if needed */
932 cxt->canonical = -1; /* Idem */
933 cxt->tagnum = -1; /* Reset tag numbers */
934 cxt->classnum = -1; /* Reset class numbers */
935 cxt->fio = f; /* Where I/O are performed */
936 cxt->optype = optype; /* A store, or a deep clone */
937 cxt->entry = 1; /* No recursion yet */
940 * The `hseen' table is used to keep track of each SV stored and their
941 * associated tag numbers is special. It is "abused" because the
942 * values stored are not real SV, just integers cast to (SV *),
943 * which explains the freeing below.
945 * It is also one possible bottlneck to achieve good storing speed,
946 * so the "shared keys" optimization is turned off (unlikely to be
947 * of any use here), and the hash table is "pre-extended". Together,
948 * those optimizations increase the throughput by 12%.
951 cxt->hseen = newHV(); /* Table where seen objects are stored */
952 HvSHAREKEYS_off(cxt->hseen);
955 * The following does not work well with perl5.004_04, and causes
956 * a core dump later on, in a completely unrelated spot, which
957 * makes me think there is a memory corruption going on.
959 * Calling hv_ksplit(hseen, HBUCKETS) instead of manually hacking
960 * it below does not make any difference. It seems to work fine
961 * with perl5.004_68 but given the probable nature of the bug,
962 * that does not prove anything.
964 * It's a shame because increasing the amount of buckets raises
965 * store() throughput by 5%, but until I figure this out, I can't
966 * allow for this to go into production.
968 * It is reported fixed in 5.005, hence the #if.
970 #if PERL_VERSION >= 5
971 #define HBUCKETS 4096 /* Buckets for %hseen */
972 HvMAX(cxt->hseen) = HBUCKETS - 1; /* keys %hseen = $HBUCKETS; */
976 * The `hclass' hash uses the same settings as `hseen' above, but it is
977 * used to assign sequential tags (numbers) to class names for blessed
980 * We turn the shared key optimization on.
983 cxt->hclass = newHV(); /* Where seen classnames are stored */
985 #if PERL_VERSION >= 5
986 HvMAX(cxt->hclass) = HBUCKETS - 1; /* keys %hclass = $HBUCKETS; */
990 * The `hook' hash table is used to keep track of the references on
991 * the STORABLE_freeze hook routines, when found in some class name.
993 * It is assumed that the inheritance tree will not be changed during
994 * storing, and that no new method will be dynamically created by the
998 cxt->hook = newHV(); /* Table where hooks are cached */
1001 * The `hook_seen' array keeps track of all the SVs returned by
1002 * STORABLE_freeze hooks for us to serialize, so that they are not
1003 * reclaimed until the end of the serialization process. Each SV is
1004 * only stored once, the first time it is seen.
1007 cxt->hook_seen = newAV(); /* Lists SVs returned by STORABLE_freeze */
1011 * clean_store_context
1013 * Clean store context by
1015 static void clean_store_context(stcxt_t *cxt)
1019 TRACEME(("clean_store_context"));
1021 ASSERT(cxt->optype & ST_STORE, ("was performing a store()"));
1024 * Insert real values into hashes where we stored faked pointers.
1027 hv_iterinit(cxt->hseen);
1028 while (he = hv_iternext(cxt->hseen))
1029 HeVAL(he) = &PL_sv_undef;
1031 hv_iterinit(cxt->hclass);
1032 while (he = hv_iternext(cxt->hclass))
1033 HeVAL(he) = &PL_sv_undef;
1036 * And now dispose of them...
1038 * The surrounding if() protection has been added because there might be
1039 * some cases where this routine is called more than once, during
1040 * exceptionnal events. This was reported by Marc Lehmann when Storable
1041 * is executed from mod_perl, and the fix was suggested by him.
1042 * -- RAM, 20/12/2000
1046 HV *hseen = cxt->hseen;
1049 sv_free((SV *) hseen);
1053 HV *hclass = cxt->hclass;
1056 sv_free((SV *) hclass);
1060 HV *hook = cxt->hook;
1063 sv_free((SV *) hook);
1066 if (cxt->hook_seen) {
1067 AV *hook_seen = cxt->hook_seen;
1069 av_undef(hook_seen);
1070 sv_free((SV *) hook_seen);
1078 * init_retrieve_context
1080 * Initialize a new retrieve context for real recursion.
1082 static void init_retrieve_context(stcxt_t *cxt, int optype, int is_tainted)
1084 TRACEME(("init_retrieve_context"));
1087 * The hook hash table is used to keep track of the references on
1088 * the STORABLE_thaw hook routines, when found in some class name.
1090 * It is assumed that the inheritance tree will not be changed during
1091 * storing, and that no new method will be dynamically created by the
1095 cxt->hook = newHV(); /* Caches STORABLE_thaw */
1098 * If retrieving an old binary version, the cxt->retrieve_vtbl variable
1099 * was set to sv_old_retrieve. We'll need a hash table to keep track of
1100 * the correspondance between the tags and the tag number used by the
1101 * new retrieve routines.
1104 cxt->hseen = (cxt->retrieve_vtbl == sv_old_retrieve) ? newHV() : 0;
1106 cxt->aseen = newAV(); /* Where retrieved objects are kept */
1107 cxt->aclass = newAV(); /* Where seen classnames are kept */
1108 cxt->tagnum = 0; /* Have to count objects... */
1109 cxt->classnum = 0; /* ...and class names as well */
1110 cxt->optype = optype;
1111 cxt->s_tainted = is_tainted;
1112 cxt->entry = 1; /* No recursion yet */
1116 * clean_retrieve_context
1118 * Clean retrieve context by
1120 static void clean_retrieve_context(stcxt_t *cxt)
1122 TRACEME(("clean_retrieve_context"));
1124 ASSERT(cxt->optype & ST_RETRIEVE, ("was performing a retrieve()"));
1127 AV *aseen = cxt->aseen;
1130 sv_free((SV *) aseen);
1134 AV *aclass = cxt->aclass;
1137 sv_free((SV *) aclass);
1141 HV *hook = cxt->hook;
1144 sv_free((SV *) hook);
1148 HV *hseen = cxt->hseen;
1151 sv_free((SV *) hseen); /* optional HV, for backward compat. */
1161 * A workaround for the CROAK bug: cleanup the last context.
1163 static void clean_context(cxt)
1166 TRACEME(("clean_context"));
1168 ASSERT(cxt->s_dirty, ("dirty context"));
1170 if (cxt->optype & ST_RETRIEVE)
1171 clean_retrieve_context(cxt);
1173 clean_store_context(cxt);
1175 ASSERT(!cxt->s_dirty, ("context is clean"));
1181 * Allocate a new context and push it on top of the parent one.
1182 * This new context is made globally visible via SET_STCXT().
1184 static stcxt_t *allocate_context(parent_cxt)
1185 stcxt_t *parent_cxt;
1189 TRACEME(("allocate_context"));
1191 ASSERT(!parent_cxt->s_dirty, ("parent context clean"));
1193 Newz(0, cxt, 1, stcxt_t);
1194 cxt->prev = parent_cxt;
1203 * Free current context, which cannot be the "root" one.
1204 * Make the context underneath globally visible via SET_STCXT().
1206 static void free_context(cxt)
1209 stcxt_t *prev = cxt->prev;
1211 TRACEME(("free_context"));
1213 ASSERT(!cxt->s_dirty, ("clean context"));
1214 ASSERT(prev, ("not freeing root context"));
1232 * Tells whether we're in the middle of a store operation.
1234 int is_storing(void)
1238 return cxt->entry && (cxt->optype & ST_STORE);
1244 * Tells whether we're in the middle of a retrieve operation.
1246 int is_retrieving(void)
1250 return cxt->entry && (cxt->optype & ST_RETRIEVE);
1254 * last_op_in_netorder
1256 * Returns whether last operation was made using network order.
1258 * This is typically out-of-band information that might prove useful
1259 * to people wishing to convert native to network order data when used.
1261 int last_op_in_netorder(void)
1265 return cxt->netorder;
1269 *** Hook lookup and calling routines.
1275 * A wrapper on gv_fetchmethod_autoload() which caches results.
1277 * Returns the routine reference as an SV*, or null if neither the package
1278 * nor its ancestors know about the method.
1280 static SV *pkg_fetchmeth(
1290 * The following code is the same as the one performed by UNIVERSAL::can
1294 gv = gv_fetchmethod_autoload(pkg, method, FALSE);
1295 if (gv && isGV(gv)) {
1296 sv = newRV((SV*) GvCV(gv));
1297 TRACEME(("%s->%s: 0x%"UVxf, HvNAME(pkg), method, PTR2UV(sv)));
1299 sv = newSVsv(&PL_sv_undef);
1300 TRACEME(("%s->%s: not found", HvNAME(pkg), method));
1304 * Cache the result, ignoring failure: if we can't store the value,
1305 * it just won't be cached.
1308 (void) hv_store(cache, HvNAME(pkg), strlen(HvNAME(pkg)), sv, 0);
1310 return SvOK(sv) ? sv : (SV *) 0;
1316 * Force cached value to be undef: hook ignored even if present.
1318 static void pkg_hide(
1323 (void) hv_store(cache,
1324 HvNAME(pkg), strlen(HvNAME(pkg)), newSVsv(&PL_sv_undef), 0);
1330 * Discard cached value: a whole fetch loop will be retried at next lookup.
1332 static void pkg_uncache(
1337 (void) hv_delete(cache, HvNAME(pkg), strlen(HvNAME(pkg)), G_DISCARD);
1343 * Our own "UNIVERSAL::can", which caches results.
1345 * Returns the routine reference as an SV*, or null if the object does not
1346 * know about the method.
1356 TRACEME(("pkg_can for %s->%s", HvNAME(pkg), method));
1359 * Look into the cache to see whether we already have determined
1360 * where the routine was, if any.
1362 * NOTA BENE: we don't use `method' at all in our lookup, since we know
1363 * that only one hook (i.e. always the same) is cached in a given cache.
1366 svh = hv_fetch(cache, HvNAME(pkg), strlen(HvNAME(pkg)), FALSE);
1370 TRACEME(("cached %s->%s: not found", HvNAME(pkg), method));
1373 TRACEME(("cached %s->%s: 0x%"UVxf,
1374 HvNAME(pkg), method, PTR2UV(sv)));
1379 TRACEME(("not cached yet"));
1380 return pkg_fetchmeth(cache, pkg, method); /* Fetch and cache */
1386 * Call routine as obj->hook(av) in scalar context.
1387 * Propagates the single returned value if not called in void context.
1389 static SV *scalar_call(
1400 TRACEME(("scalar_call (cloning=%d)", cloning));
1407 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1409 SV **ary = AvARRAY(av);
1410 int cnt = AvFILLp(av) + 1;
1412 XPUSHs(ary[0]); /* Frozen string */
1413 for (i = 1; i < cnt; i++) {
1414 TRACEME(("pushing arg #%d (0x%"UVxf")...",
1415 i, PTR2UV(ary[i])));
1416 XPUSHs(sv_2mortal(newRV(ary[i])));
1421 TRACEME(("calling..."));
1422 count = perl_call_sv(hook, flags); /* Go back to Perl code */
1423 TRACEME(("count = %d", count));
1429 SvREFCNT_inc(sv); /* We're returning it, must stay alive! */
1442 * Call routine obj->hook(cloning) in list context.
1443 * Returns the list of returned values in an array.
1445 static AV *array_call(
1455 TRACEME(("array_call (cloning=%d)", cloning));
1461 XPUSHs(obj); /* Target object */
1462 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1465 count = perl_call_sv(hook, G_ARRAY); /* Go back to Perl code */
1470 for (i = count - 1; i >= 0; i--) {
1472 av_store(av, i, SvREFCNT_inc(sv));
1485 * Lookup the class name in the `hclass' table and either assign it a new ID
1486 * or return the existing one, by filling in `classnum'.
1488 * Return true if the class was known, false if the ID was just generated.
1490 static int known_class(
1492 char *name, /* Class name */
1493 int len, /* Name length */
1497 HV *hclass = cxt->hclass;
1499 TRACEME(("known_class (%s)", name));
1502 * Recall that we don't store pointers in this hash table, but tags.
1503 * Therefore, we need LOW_32BITS() to extract the relevant parts.
1506 svh = hv_fetch(hclass, name, len, FALSE);
1508 *classnum = LOW_32BITS(*svh);
1513 * Unknown classname, we need to record it.
1517 if (!hv_store(hclass, name, len, INT2PTR(SV*, cxt->classnum), 0))
1518 CROAK(("Unable to record new classname"));
1520 *classnum = cxt->classnum;
1525 *** Sepcific store routines.
1531 * Store a reference.
1532 * Layout is SX_REF <object> or SX_OVERLOAD <object>.
1534 static int store_ref(stcxt_t *cxt, SV *sv)
1536 TRACEME(("store_ref (0x%"UVxf")", PTR2UV(sv)));
1539 * Follow reference, and check if target is overloaded.
1545 HV *stash = (HV *) SvSTASH(sv);
1546 if (stash && Gv_AMG(stash)) {
1547 TRACEME(("ref (0x%"UVxf") is overloaded", PTR2UV(sv)));
1548 PUTMARK(SX_OVERLOAD);
1554 return store(cxt, sv);
1562 * Layout is SX_LSCALAR <length> <data>, SX_SCALAR <lenght> <data> or SX_UNDEF.
1563 * The <data> section is omitted if <length> is 0.
1565 * If integer or double, the layout is SX_INTEGER <data> or SX_DOUBLE <data>.
1566 * Small integers (within [-127, +127]) are stored as SX_BYTE <byte>.
1568 static int store_scalar(stcxt_t *cxt, SV *sv)
1573 U32 flags = SvFLAGS(sv); /* "cc -O" may put it in register */
1575 TRACEME(("store_scalar (0x%"UVxf")", PTR2UV(sv)));
1578 * For efficiency, break the SV encapsulation by peaking at the flags
1579 * directly without using the Perl macros to avoid dereferencing
1580 * sv->sv_flags each time we wish to check the flags.
1583 if (!(flags & SVf_OK)) { /* !SvOK(sv) */
1584 if (sv == &PL_sv_undef) {
1585 TRACEME(("immortal undef"));
1586 PUTMARK(SX_SV_UNDEF);
1588 TRACEME(("undef at 0x%"UVxf, PTR2UV(sv)));
1595 * Always store the string representation of a scalar if it exists.
1596 * Gisle Aas provided me with this test case, better than a long speach:
1598 * perl -MDevel::Peek -le '$a="abc"; $a+0; Dump($a)'
1599 * SV = PVNV(0x80c8520)
1601 * FLAGS = (NOK,POK,pNOK,pPOK)
1604 * PV = 0x80c83d0 "abc"\0
1608 * Write SX_SCALAR, length, followed by the actual data.
1610 * Otherwise, write an SX_BYTE, SX_INTEGER or an SX_DOUBLE as
1611 * appropriate, followed by the actual (binary) data. A double
1612 * is written as a string if network order, for portability.
1614 * NOTE: instead of using SvNOK(sv), we test for SvNOKp(sv).
1615 * The reason is that when the scalar value is tainted, the SvNOK(sv)
1618 * The test for a read-only scalar with both POK and NOK set is meant
1619 * to quickly detect &PL_sv_yes and &PL_sv_no without having to pay the
1620 * address comparison for each scalar we store.
1623 #define SV_MAYBE_IMMORTAL (SVf_READONLY|SVf_POK|SVf_NOK)
1625 if ((flags & SV_MAYBE_IMMORTAL) == SV_MAYBE_IMMORTAL) {
1626 if (sv == &PL_sv_yes) {
1627 TRACEME(("immortal yes"));
1629 } else if (sv == &PL_sv_no) {
1630 TRACEME(("immortal no"));
1633 pv = SvPV(sv, len); /* We know it's SvPOK */
1634 goto string; /* Share code below */
1636 } else if (flags & SVp_POK) { /* SvPOKp(sv) => string */
1637 I32 wlen; /* For 64-bit machines */
1641 * Will come here from below with pv and len set if double & netorder,
1642 * or from above if it was readonly, POK and NOK but neither &PL_sv_yes
1647 wlen = (I32) len; /* WLEN via STORE_SCALAR expects I32 */
1649 STORE_UTF8STR(pv, wlen);
1651 STORE_SCALAR(pv, wlen);
1652 TRACEME(("ok (scalar 0x%"UVxf" '%s', length = %"IVdf")",
1653 PTR2UV(sv), SvPVX(sv), (IV)len));
1655 } else if (flags & SVp_NOK) { /* SvNOKp(sv) => double */
1659 * Watch for number being an integer in disguise.
1661 if (nv == (NV) (iv = I_V(nv))) {
1662 TRACEME(("double %"NVff" is actually integer %"IVdf, nv, iv));
1663 goto integer; /* Share code below */
1666 if (cxt->netorder) {
1667 TRACEME(("double %"NVff" stored as string", nv));
1669 goto string; /* Share code above */
1673 WRITE(&nv, sizeof(nv));
1675 TRACEME(("ok (double 0x%"UVxf", value = %"NVff")", PTR2UV(sv), nv));
1677 } else if (flags & SVp_IOK) { /* SvIOKp(sv) => integer */
1681 * Will come here from above with iv set if double is an integer.
1686 * Optimize small integers into a single byte, otherwise store as
1687 * a real integer (converted into network order if they asked).
1690 if (iv >= -128 && iv <= 127) {
1691 unsigned char siv = (unsigned char) (iv + 128); /* [0,255] */
1694 TRACEME(("small integer stored as %d", siv));
1695 } else if (cxt->netorder) {
1698 niv = (I32) htonl(iv);
1699 TRACEME(("using network order"));
1702 TRACEME(("as-is for network order"));
1707 PUTMARK(SX_INTEGER);
1708 WRITE(&iv, sizeof(iv));
1711 TRACEME(("ok (integer 0x%"UVxf", value = %"IVdf")", PTR2UV(sv), iv));
1714 CROAK(("Can't determine type of %s(0x%"UVxf")",
1715 sv_reftype(sv, FALSE),
1718 return 0; /* Ok, no recursion on scalars */
1726 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
1727 * Each item is stored as <object>.
1729 static int store_array(stcxt_t *cxt, AV *av)
1732 I32 len = av_len(av) + 1;
1736 TRACEME(("store_array (0x%"UVxf")", PTR2UV(av)));
1739 * Signal array by emitting SX_ARRAY, followed by the array length.
1744 TRACEME(("size = %d", len));
1747 * Now store each item recursively.
1750 for (i = 0; i < len; i++) {
1751 sav = av_fetch(av, i, 0);
1753 TRACEME(("(#%d) undef item", i));
1757 TRACEME(("(#%d) item", i));
1758 if (ret = store(cxt, *sav))
1762 TRACEME(("ok (array)"));
1771 * Borrowed from perl source file pp_ctl.c, where it is used by pp_sort.
1774 sortcmp(const void *a, const void *b)
1776 return sv_cmp(*(SV * const *) a, *(SV * const *) b);
1783 * Store an hash table.
1785 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
1786 * Values are stored as <object>.
1787 * Keys are stored as <length> <data>, the <data> section being omitted
1790 static int store_hash(stcxt_t *cxt, HV *hv)
1792 I32 len = HvKEYS(hv);
1798 TRACEME(("store_hash (0x%"UVxf")", PTR2UV(hv)));
1801 * Signal hash by emitting SX_HASH, followed by the table length.
1806 TRACEME(("size = %d", len));
1809 * Save possible iteration state via each() on that table.
1812 riter = HvRITER(hv);
1813 eiter = HvEITER(hv);
1817 * Now store each item recursively.
1819 * If canonical is defined to some true value then store each
1820 * key/value pair in sorted order otherwise the order is random.
1821 * Canonical order is irrelevant when a deep clone operation is performed.
1823 * Fetch the value from perl only once per store() operation, and only
1828 !(cxt->optype & ST_CLONE) && (cxt->canonical == 1 ||
1829 (cxt->canonical < 0 && (cxt->canonical =
1830 SvTRUE(perl_get_sv("Storable::canonical", TRUE)) ? 1 : 0)))
1833 * Storing in order, sorted by key.
1834 * Run through the hash, building up an array of keys in a
1835 * mortal array, sort the array and then run through the
1841 TRACEME(("using canonical order"));
1843 for (i = 0; i < len; i++) {
1844 HE *he = hv_iternext(hv);
1845 SV *key = hv_iterkeysv(he);
1846 av_store(av, AvFILLp(av)+1, key); /* av_push(), really */
1849 qsort((char *) AvARRAY(av), len, sizeof(SV *), sortcmp);
1851 for (i = 0; i < len; i++) {
1854 SV *key = av_shift(av);
1855 HE *he = hv_fetch_ent(hv, key, 0, 0);
1856 SV *val = HeVAL(he);
1858 return 1; /* Internal error, not I/O error */
1861 * Store value first.
1864 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
1866 if (ret = store(cxt, val))
1871 * Keys are written after values to make sure retrieval
1872 * can be optimal in terms of memory usage, where keys are
1873 * read into a fixed unique buffer called kbuf.
1874 * See retrieve_hash() for details.
1877 keyval = hv_iterkey(he, &keylen);
1878 TRACEME(("(#%d) key '%s'", i, keyval));
1881 WRITE(keyval, keylen);
1885 * Free up the temporary array
1894 * Storing in "random" order (in the order the keys are stored
1895 * within the the hash). This is the default and will be faster!
1898 for (i = 0; i < len; i++) {
1901 SV *val = hv_iternextsv(hv, &key, &len);
1904 return 1; /* Internal error, not I/O error */
1907 * Store value first.
1910 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
1912 if (ret = store(cxt, val))
1917 * Keys are written after values to make sure retrieval
1918 * can be optimal in terms of memory usage, where keys are
1919 * read into a fixed unique buffer called kbuf.
1920 * See retrieve_hash() for details.
1923 TRACEME(("(#%d) key '%s'", i, key));
1930 TRACEME(("ok (hash 0x%"UVxf")", PTR2UV(hv)));
1933 HvRITER(hv) = riter; /* Restore hash iterator state */
1934 HvEITER(hv) = eiter;
1942 * When storing a tied object (be it a tied scalar, array or hash), we lay out
1943 * a special mark, followed by the underlying tied object. For instance, when
1944 * dealing with a tied hash, we store SX_TIED_HASH <hash object>, where
1945 * <hash object> stands for the serialization of the tied hash.
1947 static int store_tied(stcxt_t *cxt, SV *sv)
1951 int svt = SvTYPE(sv);
1954 TRACEME(("store_tied (0x%"UVxf")", PTR2UV(sv)));
1957 * We have a small run-time penalty here because we chose to factorise
1958 * all tieds objects into the same routine, and not have a store_tied_hash,
1959 * a store_tied_array, etc...
1961 * Don't use a switch() statement, as most compilers don't optimize that
1962 * well for 2/3 values. An if() else if() cascade is just fine. We put
1963 * tied hashes first, as they are the most likely beasts.
1966 if (svt == SVt_PVHV) {
1967 TRACEME(("tied hash"));
1968 PUTMARK(SX_TIED_HASH); /* Introduces tied hash */
1969 } else if (svt == SVt_PVAV) {
1970 TRACEME(("tied array"));
1971 PUTMARK(SX_TIED_ARRAY); /* Introduces tied array */
1973 TRACEME(("tied scalar"));
1974 PUTMARK(SX_TIED_SCALAR); /* Introduces tied scalar */
1978 if (!(mg = mg_find(sv, mtype)))
1979 CROAK(("No magic '%c' found while storing tied %s", mtype,
1980 (svt == SVt_PVHV) ? "hash" :
1981 (svt == SVt_PVAV) ? "array" : "scalar"));
1984 * The mg->mg_obj found by mg_find() above actually points to the
1985 * underlying tied Perl object implementation. For instance, if the
1986 * original SV was that of a tied array, then mg->mg_obj is an AV.
1988 * Note that we store the Perl object as-is. We don't call its FETCH
1989 * method along the way. At retrieval time, we won't call its STORE
1990 * method either, but the tieing magic will be re-installed. In itself,
1991 * that ensures that the tieing semantics are preserved since futher
1992 * accesses on the retrieved object will indeed call the magic methods...
1995 if (ret = store(cxt, mg->mg_obj))
1998 TRACEME(("ok (tied)"));
2006 * Stores a reference to an item within a tied structure:
2008 * . \$h{key}, stores both the (tied %h) object and 'key'.
2009 * . \$a[idx], stores both the (tied @a) object and 'idx'.
2011 * Layout is therefore either:
2012 * SX_TIED_KEY <object> <key>
2013 * SX_TIED_IDX <object> <index>
2015 static int store_tied_item(stcxt_t *cxt, SV *sv)
2020 TRACEME(("store_tied_item (0x%"UVxf")", PTR2UV(sv)));
2022 if (!(mg = mg_find(sv, 'p')))
2023 CROAK(("No magic 'p' found while storing reference to tied item"));
2026 * We discriminate between \$h{key} and \$a[idx] via mg_ptr.
2030 TRACEME(("store_tied_item: storing a ref to a tied hash item"));
2031 PUTMARK(SX_TIED_KEY);
2032 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2034 if (ret = store(cxt, mg->mg_obj))
2037 TRACEME(("store_tied_item: storing PTR 0x%"UVxf, PTR2UV(mg->mg_ptr)));
2039 if (ret = store(cxt, (SV *) mg->mg_ptr))
2042 I32 idx = mg->mg_len;
2044 TRACEME(("store_tied_item: storing a ref to a tied array item "));
2045 PUTMARK(SX_TIED_IDX);
2046 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2048 if (ret = store(cxt, mg->mg_obj))
2051 TRACEME(("store_tied_item: storing IDX %d", idx));
2056 TRACEME(("ok (tied item)"));
2062 * store_hook -- dispatched manually, not via sv_store[]
2064 * The blessed SV is serialized by a hook.
2068 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2070 * where <flags> indicates how long <len>, <len2> and <len3> are, whether
2071 * the trailing part [] is present, the type of object (scalar, array or hash).
2072 * There is also a bit which says how the classname is stored between:
2077 * and when the <index> form is used (classname already seen), the "large
2078 * classname" bit in <flags> indicates how large the <index> is.
2080 * The serialized string returned by the hook is of length <len2> and comes
2081 * next. It is an opaque string for us.
2083 * Those <len3> object IDs which are listed last represent the extra references
2084 * not directly serialized by the hook, but which are linked to the object.
2086 * When recursion is mandated to resolve object-IDs not yet seen, we have
2087 * instead, with <header> being flags with bits set to indicate the object type
2088 * and that recursion was indeed needed:
2090 * SX_HOOK <header> <object> <header> <object> <flags>
2092 * that same header being repeated between serialized objects obtained through
2093 * recursion, until we reach flags indicating no recursion, at which point
2094 * we know we've resynchronized with a single layout, after <flags>.
2096 * When storing a blessed ref to a tied variable, the following format is
2099 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
2101 * The first <flags> indication carries an object of type SHT_EXTRA, and the
2102 * real object type is held in the <extra> flag. At the very end of the
2103 * serialization stream, the underlying magic object is serialized, just like
2104 * any other tied variable.
2106 static int store_hook(
2119 int count; /* really len3 + 1 */
2120 unsigned char flags;
2123 int recursed = 0; /* counts recursion */
2124 int obj_type; /* object type, on 2 bits */
2127 int clone = cxt->optype & ST_CLONE;
2128 char mtype; /* for blessed ref to tied structures */
2129 unsigned char eflags; /* used when object type is SHT_EXTRA */
2131 TRACEME(("store_hook, class \"%s\", tagged #%d", HvNAME(pkg), cxt->tagnum));
2134 * Determine object type on 2 bits.
2139 obj_type = SHT_SCALAR;
2142 obj_type = SHT_ARRAY;
2145 obj_type = SHT_HASH;
2149 * Produced by a blessed ref to a tied data structure, $o in the
2150 * following Perl code.
2154 * my $o = bless \%h, 'BAR';
2156 * Signal the tie-ing magic by setting the object type as SHT_EXTRA
2157 * (since we have only 2 bits in <flags> to store the type), and an
2158 * <extra> byte flag will be emitted after the FIRST <flags> in the
2159 * stream, carrying what we put in `eflags'.
2161 obj_type = SHT_EXTRA;
2162 switch (SvTYPE(sv)) {
2164 eflags = (unsigned char) SHT_THASH;
2168 eflags = (unsigned char) SHT_TARRAY;
2172 eflags = (unsigned char) SHT_TSCALAR;
2178 CROAK(("Unexpected object type (%d) in store_hook()", type));
2180 flags = SHF_NEED_RECURSE | obj_type;
2182 class = HvNAME(pkg);
2183 len = strlen(class);
2186 * To call the hook, we need to fake a call like:
2188 * $object->STORABLE_freeze($cloning);
2190 * but we don't have the $object here. For instance, if $object is
2191 * a blessed array, what we have in `sv' is the array, and we can't
2192 * call a method on those.
2194 * Therefore, we need to create a temporary reference to the object and
2195 * make the call on that reference.
2198 TRACEME(("about to call STORABLE_freeze on class %s", class));
2200 ref = newRV_noinc(sv); /* Temporary reference */
2201 av = array_call(ref, hook, clone); /* @a = $object->STORABLE_freeze($c) */
2203 SvREFCNT_dec(ref); /* Reclaim temporary reference */
2205 count = AvFILLp(av) + 1;
2206 TRACEME(("store_hook, array holds %d items", count));
2209 * If they return an empty list, it means they wish to ignore the
2210 * hook for this class (and not just this instance -- that's for them
2211 * to handle if they so wish).
2213 * Simply disable the cached entry for the hook (it won't be recomputed
2214 * since it's present in the cache) and recurse to store_blessed().
2219 * They must not change their mind in the middle of a serialization.
2222 if (hv_fetch(cxt->hclass, class, len, FALSE))
2223 CROAK(("Too late to ignore hooks for %s class \"%s\"",
2224 (cxt->optype & ST_CLONE) ? "cloning" : "storing", class));
2226 pkg_hide(cxt->hook, pkg, "STORABLE_freeze");
2228 ASSERT(!pkg_can(cxt->hook, pkg, "STORABLE_freeze"), ("hook invisible"));
2229 TRACEME(("ignoring STORABLE_freeze in class \"%s\"", class));
2231 return store_blessed(cxt, sv, type, pkg);
2235 * Get frozen string.
2239 pv = SvPV(ary[0], len2);
2242 * If they returned more than one item, we need to serialize some
2243 * extra references if not already done.
2245 * Loop over the array, starting at postion #1, and for each item,
2246 * ensure it is a reference, serialize it if not already done, and
2247 * replace the entry with the tag ID of the corresponding serialized
2250 * We CHEAT by not calling av_fetch() and read directly within the
2254 for (i = 1; i < count; i++) {
2258 AV *av_hook = cxt->hook_seen;
2261 CROAK(("Item #%d returned by STORABLE_freeze "
2262 "for %s is not a reference", i, class));
2263 xsv = SvRV(rsv); /* Follow ref to know what to look for */
2266 * Look in hseen and see if we have a tag already.
2267 * Serialize entry if not done already, and get its tag.
2270 if (svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE))
2271 goto sv_seen; /* Avoid moving code too far to the right */
2273 TRACEME(("listed object %d at 0x%"UVxf" is unknown", i-1, PTR2UV(xsv)));
2276 * We need to recurse to store that object and get it to be known
2277 * so that we can resolve the list of object-IDs at retrieve time.
2279 * The first time we do this, we need to emit the proper header
2280 * indicating that we recursed, and what the type of object is (the
2281 * object we're storing via a user-hook). Indeed, during retrieval,
2282 * we'll have to create the object before recursing to retrieve the
2283 * others, in case those would point back at that object.
2286 /* [SX_HOOK] <flags> [<extra>] <object>*/
2290 if (obj_type == SHT_EXTRA)
2295 if (ret = store(cxt, xsv)) /* Given by hook for us to store */
2298 svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE);
2300 CROAK(("Could not serialize item #%d from hook in %s", i, class));
2303 * It was the first time we serialized `xsv'.
2305 * Keep this SV alive until the end of the serialization: if we
2306 * disposed of it right now by decrementing its refcount, and it was
2307 * a temporary value, some next temporary value allocated during
2308 * another STORABLE_freeze might take its place, and we'd wrongly
2309 * assume that new SV was already serialized, based on its presence
2312 * Therefore, push it away in cxt->hook_seen.
2315 av_store(av_hook, AvFILLp(av_hook)+1, SvREFCNT_inc(xsv));
2319 * Dispose of the REF they returned. If we saved the `xsv' away
2320 * in the array of returned SVs, that will not cause the underlying
2321 * referenced SV to be reclaimed.
2324 ASSERT(SvREFCNT(xsv) > 1, ("SV will survive disposal of its REF"));
2325 SvREFCNT_dec(rsv); /* Dispose of reference */
2328 * Replace entry with its tag (not a real SV, so no refcnt increment)
2332 TRACEME(("listed object %d at 0x%"UVxf" is tag #%"UVuf,
2333 i-1, PTR2UV(xsv), PTR2UV(*svh)));
2337 * Allocate a class ID if not already done.
2339 * This needs to be done after the recursion above, since at retrieval
2340 * time, we'll see the inner objects first. Many thanks to
2341 * Salvador Ortiz Garcia <sog@msg.com.mx> who spot that bug and
2342 * proposed the right fix. -- RAM, 15/09/2000
2345 if (!known_class(cxt, class, len, &classnum)) {
2346 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2347 classnum = -1; /* Mark: we must store classname */
2349 TRACEME(("already seen class %s, ID = %d", class, classnum));
2353 * Compute leading flags.
2357 if (((classnum == -1) ? len : classnum) > LG_SCALAR)
2358 flags |= SHF_LARGE_CLASSLEN;
2360 flags |= SHF_IDX_CLASSNAME;
2361 if (len2 > LG_SCALAR)
2362 flags |= SHF_LARGE_STRLEN;
2364 flags |= SHF_HAS_LIST;
2365 if (count > (LG_SCALAR + 1))
2366 flags |= SHF_LARGE_LISTLEN;
2369 * We're ready to emit either serialized form:
2371 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2372 * SX_HOOK <flags> <index> <len2> <str> [<len3> <object-IDs>]
2374 * If we recursed, the SX_HOOK has already been emitted.
2377 TRACEME(("SX_HOOK (recursed=%d) flags=0x%x "
2378 "class=%"IVdf" len=%"IVdf" len2=%"IVdf" len3=%d",
2379 recursed, flags, (IV)classnum, (IV)len, (IV)len2, count-1));
2381 /* SX_HOOK <flags> [<extra>] */
2385 if (obj_type == SHT_EXTRA)
2390 /* <len> <classname> or <index> */
2391 if (flags & SHF_IDX_CLASSNAME) {
2392 if (flags & SHF_LARGE_CLASSLEN)
2395 unsigned char cnum = (unsigned char) classnum;
2399 if (flags & SHF_LARGE_CLASSLEN)
2402 unsigned char clen = (unsigned char) len;
2405 WRITE(class, len); /* Final \0 is omitted */
2408 /* <len2> <frozen-str> */
2409 if (flags & SHF_LARGE_STRLEN) {
2410 I32 wlen2 = len2; /* STRLEN might be 8 bytes */
2411 WLEN(wlen2); /* Must write an I32 for 64-bit machines */
2413 unsigned char clen = (unsigned char) len2;
2417 WRITE(pv, len2); /* Final \0 is omitted */
2419 /* [<len3> <object-IDs>] */
2420 if (flags & SHF_HAS_LIST) {
2421 int len3 = count - 1;
2422 if (flags & SHF_LARGE_LISTLEN)
2425 unsigned char clen = (unsigned char) len3;
2430 * NOTA BENE, for 64-bit machines: the ary[i] below does not yield a
2431 * real pointer, rather a tag number, well under the 32-bit limit.
2434 for (i = 1; i < count; i++) {
2435 I32 tagval = htonl(LOW_32BITS(ary[i]));
2437 TRACEME(("object %d, tag #%d", i-1, ntohl(tagval)));
2442 * Free the array. We need extra care for indices after 0, since they
2443 * don't hold real SVs but integers cast.
2447 AvFILLp(av) = 0; /* Cheat, nothing after 0 interests us */
2452 * If object was tied, need to insert serialization of the magic object.
2455 if (obj_type == SHT_EXTRA) {
2458 if (!(mg = mg_find(sv, mtype))) {
2459 int svt = SvTYPE(sv);
2460 CROAK(("No magic '%c' found while storing ref to tied %s with hook",
2461 mtype, (svt == SVt_PVHV) ? "hash" :
2462 (svt == SVt_PVAV) ? "array" : "scalar"));
2465 TRACEME(("handling the magic object 0x%"UVxf" part of 0x%"UVxf,
2466 PTR2UV(mg->mg_obj), PTR2UV(sv)));
2472 if (ret = store(cxt, mg->mg_obj))
2480 * store_blessed -- dispatched manually, not via sv_store[]
2482 * Check whether there is a STORABLE_xxx hook defined in the class or in one
2483 * of its ancestors. If there is, then redispatch to store_hook();
2485 * Otherwise, the blessed SV is stored using the following layout:
2487 * SX_BLESS <flag> <len> <classname> <object>
2489 * where <flag> indicates whether <len> is stored on 0 or 4 bytes, depending
2490 * on the high-order bit in flag: if 1, then length follows on 4 bytes.
2491 * Otherwise, the low order bits give the length, thereby giving a compact
2492 * representation for class names less than 127 chars long.
2494 * Each <classname> seen is remembered and indexed, so that the next time
2495 * an object in the blessed in the same <classname> is stored, the following
2498 * SX_IX_BLESS <flag> <index> <object>
2500 * where <index> is the classname index, stored on 0 or 4 bytes depending
2501 * on the high-order bit in flag (same encoding as above for <len>).
2503 static int store_blessed(
2514 TRACEME(("store_blessed, type %d, class \"%s\"", type, HvNAME(pkg)));
2517 * Look for a hook for this blessed SV and redirect to store_hook()
2521 hook = pkg_can(cxt->hook, pkg, "STORABLE_freeze");
2523 return store_hook(cxt, sv, type, pkg, hook);
2526 * This is a blessed SV without any serialization hook.
2529 class = HvNAME(pkg);
2530 len = strlen(class);
2532 TRACEME(("blessed 0x%"UVxf" in %s, no hook: tagged #%d",
2533 PTR2UV(sv), class, cxt->tagnum));
2536 * Determine whether it is the first time we see that class name (in which
2537 * case it will be stored in the SX_BLESS form), or whether we already
2538 * saw that class name before (in which case the SX_IX_BLESS form will be
2542 if (known_class(cxt, class, len, &classnum)) {
2543 TRACEME(("already seen class %s, ID = %d", class, classnum));
2544 PUTMARK(SX_IX_BLESS);
2545 if (classnum <= LG_BLESS) {
2546 unsigned char cnum = (unsigned char) classnum;
2549 unsigned char flag = (unsigned char) 0x80;
2554 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2556 if (len <= LG_BLESS) {
2557 unsigned char clen = (unsigned char) len;
2560 unsigned char flag = (unsigned char) 0x80;
2562 WLEN(len); /* Don't BER-encode, this should be rare */
2564 WRITE(class, len); /* Final \0 is omitted */
2568 * Now emit the <object> part.
2571 return SV_STORE(type)(cxt, sv);
2577 * We don't know how to store the item we reached, so return an error condition.
2578 * (it's probably a GLOB, some CODE reference, etc...)
2580 * If they defined the `forgive_me' variable at the Perl level to some
2581 * true value, then don't croak, just warn, and store a placeholder string
2584 static int store_other(stcxt_t *cxt, SV *sv)
2587 static char buf[80];
2589 TRACEME(("store_other"));
2592 * Fetch the value from perl only once per store() operation.
2596 cxt->forgive_me == 0 ||
2597 (cxt->forgive_me < 0 && !(cxt->forgive_me =
2598 SvTRUE(perl_get_sv("Storable::forgive_me", TRUE)) ? 1 : 0))
2600 CROAK(("Can't store %s items", sv_reftype(sv, FALSE)));
2602 warn("Can't store item %s(0x%"UVxf")",
2603 sv_reftype(sv, FALSE), PTR2UV(sv));
2606 * Store placeholder string as a scalar instead...
2609 (void) sprintf(buf, "You lost %s(0x%"UVxf")\0", sv_reftype(sv, FALSE),
2613 STORE_SCALAR(buf, len);
2614 TRACEME(("ok (dummy \"%s\", length = %"IVdf")", buf, len));
2620 *** Store driving routines
2626 * WARNING: partially duplicates Perl's sv_reftype for speed.
2628 * Returns the type of the SV, identified by an integer. That integer
2629 * may then be used to index the dynamic routine dispatch table.
2631 static int sv_type(SV *sv)
2633 switch (SvTYPE(sv)) {
2638 * No need to check for ROK, that can't be set here since there
2639 * is no field capable of hodling the xrv_rv reference.
2647 * Starting from SVt_PV, it is possible to have the ROK flag
2648 * set, the pointer to the other SV being either stored in
2649 * the xrv_rv (in the case of a pure SVt_RV), or as the
2650 * xpv_pv field of an SVt_PV and its heirs.
2652 * However, those SV cannot be magical or they would be an
2653 * SVt_PVMG at least.
2655 return SvROK(sv) ? svis_REF : svis_SCALAR;
2657 case SVt_PVLV: /* Workaround for perl5.004_04 "LVALUE" bug */
2658 if (SvRMAGICAL(sv) && (mg_find(sv, 'p')))
2659 return svis_TIED_ITEM;
2662 if (SvRMAGICAL(sv) && (mg_find(sv, 'q')))
2664 return SvROK(sv) ? svis_REF : svis_SCALAR;
2666 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
2670 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
2683 * Recursively store objects pointed to by the sv to the specified file.
2685 * Layout is <content> or SX_OBJECT <tagnum> if we reach an already stored
2686 * object (one for which storage has started -- it may not be over if we have
2687 * a self-referenced structure). This data set forms a stored <object>.
2689 static int store(stcxt_t *cxt, SV *sv)
2695 HV *hseen = cxt->hseen;
2697 TRACEME(("store (0x%"UVxf")", PTR2UV(sv)));
2700 * If object has already been stored, do not duplicate data.
2701 * Simply emit the SX_OBJECT marker followed by its tag data.
2702 * The tag is always written in network order.
2704 * NOTA BENE, for 64-bit machines: the "*svh" below does not yield a
2705 * real pointer, rather a tag number (watch the insertion code below).
2706 * That means it pobably safe to assume it is well under the 32-bit limit,
2707 * and makes the truncation safe.
2708 * -- RAM, 14/09/1999
2711 svh = hv_fetch(hseen, (char *) &sv, sizeof(sv), FALSE);
2713 I32 tagval = htonl(LOW_32BITS(*svh));
2715 TRACEME(("object 0x%"UVxf" seen as #%d", PTR2UV(sv), ntohl(tagval)));
2723 * Allocate a new tag and associate it with the address of the sv being
2724 * stored, before recursing...
2726 * In order to avoid creating new SvIVs to hold the tagnum we just
2727 * cast the tagnum to a SV pointer and store that in the hash. This
2728 * means that we must clean up the hash manually afterwards, but gives
2729 * us a 15% throughput increase.
2734 if (!hv_store(hseen,
2735 (char *) &sv, sizeof(sv), INT2PTR(SV*, cxt->tagnum), 0))
2739 * Store `sv' and everything beneath it, using appropriate routine.
2740 * Abort immediately if we get a non-zero status back.
2745 TRACEME(("storing 0x%"UVxf" tag #%d, type %d...",
2746 PTR2UV(sv), cxt->tagnum, type));
2749 HV *pkg = SvSTASH(sv);
2750 ret = store_blessed(cxt, sv, type, pkg);
2752 ret = SV_STORE(type)(cxt, sv);
2754 TRACEME(("%s (stored 0x%"UVxf", refcnt=%d, %s)",
2755 ret ? "FAILED" : "ok", PTR2UV(sv),
2756 SvREFCNT(sv), sv_reftype(sv, FALSE)));
2764 * Write magic number and system information into the file.
2765 * Layout is <magic> <network> [<len> <byteorder> <sizeof int> <sizeof long>
2766 * <sizeof ptr>] where <len> is the length of the byteorder hexa string.
2767 * All size and lenghts are written as single characters here.
2769 * Note that no byte ordering info is emitted when <network> is true, since
2770 * integers will be emitted in network order in that case.
2772 static int magic_write(stcxt_t *cxt)
2774 char buf[256]; /* Enough room for 256 hexa digits */
2776 int use_network_order = cxt->netorder;
2778 TRACEME(("magic_write on fd=%d", cxt->fio ? fileno(cxt->fio) : -1));
2781 WRITE(magicstr, strlen(magicstr)); /* Don't write final \0 */
2784 * Starting with 0.6, the "use_network_order" byte flag is also used to
2785 * indicate the version number of the binary image, encoded in the upper
2786 * bits. The bit 0 is always used to indicate network order.
2790 ((use_network_order ? 0x1 : 0x0) | (STORABLE_BIN_MAJOR << 1));
2794 * Starting with 0.7, a full byte is dedicated to the minor version of
2795 * the binary format, which is incremented only when new markers are
2796 * introduced, for instance, but when backward compatibility is preserved.
2799 PUTMARK((unsigned char) STORABLE_BIN_MINOR);
2801 if (use_network_order)
2802 return 0; /* Don't bother with byte ordering */
2804 sprintf(buf, "%lx", (unsigned long) BYTEORDER);
2805 c = (unsigned char) strlen(buf);
2807 WRITE(buf, (unsigned int) c); /* Don't write final \0 */
2808 PUTMARK((unsigned char) sizeof(int));
2809 PUTMARK((unsigned char) sizeof(long));
2810 PUTMARK((unsigned char) sizeof(char *));
2811 PUTMARK((unsigned char) sizeof(NV));
2813 TRACEME(("ok (magic_write byteorder = 0x%lx [%d], I%d L%d P%d D%d)",
2814 (unsigned long) BYTEORDER, (int) c,
2815 (int) sizeof(int), (int) sizeof(long),
2816 (int) sizeof(char *), (int) sizeof(NV)));
2824 * Common code for store operations.
2826 * When memory store is requested (f = NULL) and a non null SV* is given in
2827 * `res', it is filled with a new SV created out of the memory buffer.
2829 * It is required to provide a non-null `res' when the operation type is not
2830 * dclone() and store() is performed to memory.
2832 static int do_store(
2842 ASSERT(!(f == 0 && !(optype & ST_CLONE)) || res,
2843 ("must supply result SV pointer for real recursion to memory"));
2845 TRACEME(("do_store (optype=%d, netorder=%d)",
2846 optype, network_order));
2851 * Workaround for CROAK leak: if they enter with a "dirty" context,
2852 * free up memory for them now.
2859 * Now that STORABLE_xxx hooks exist, it is possible that they try to
2860 * re-enter store() via the hooks. We need to stack contexts.
2864 cxt = allocate_context(cxt);
2868 ASSERT(cxt->entry == 1, ("starting new recursion"));
2869 ASSERT(!cxt->s_dirty, ("clean context"));
2872 * Ensure sv is actually a reference. From perl, we called something
2874 * pstore(FILE, \@array);
2875 * so we must get the scalar value behing that reference.
2879 CROAK(("Not a reference"));
2880 sv = SvRV(sv); /* So follow it to know what to store */
2883 * If we're going to store to memory, reset the buffer.
2890 * Prepare context and emit headers.
2893 init_store_context(cxt, f, optype, network_order);
2895 if (-1 == magic_write(cxt)) /* Emit magic and ILP info */
2896 return 0; /* Error */
2899 * Recursively store object...
2902 ASSERT(is_storing(), ("within store operation"));
2904 status = store(cxt, sv); /* Just do it! */
2907 * If they asked for a memory store and they provided an SV pointer,
2908 * make an SV string out of the buffer and fill their pointer.
2910 * When asking for ST_REAL, it's MANDATORY for the caller to provide
2911 * an SV, since context cleanup might free the buffer if we did recurse.
2912 * (unless caller is dclone(), which is aware of that).
2915 if (!cxt->fio && res)
2921 * The "root" context is never freed, since it is meant to be always
2922 * handy for the common case where no recursion occurs at all (i.e.
2923 * we enter store() outside of any Storable code and leave it, period).
2924 * We know it's the "root" context because there's nothing stacked
2929 * When deep cloning, we don't free the context: doing so would force
2930 * us to copy the data in the memory buffer. Sicne we know we're
2931 * about to enter do_retrieve...
2934 clean_store_context(cxt);
2935 if (cxt->prev && !(cxt->optype & ST_CLONE))
2938 TRACEME(("do_store returns %d", status));
2946 * Store the transitive data closure of given object to disk.
2947 * Returns 0 on error, a true value otherwise.
2949 int pstore(PerlIO *f, SV *sv)
2951 TRACEME(("pstore"));
2952 return do_store(f, sv, 0, FALSE, (SV**) 0);
2959 * Same as pstore(), but network order is used for integers and doubles are
2960 * emitted as strings.
2962 int net_pstore(PerlIO *f, SV *sv)
2964 TRACEME(("net_pstore"));
2965 return do_store(f, sv, 0, TRUE, (SV**) 0);
2975 * Build a new SV out of the content of the internal memory buffer.
2977 static SV *mbuf2sv(void)
2981 return newSVpv(mbase, MBUF_SIZE());
2987 * Store the transitive data closure of given object to memory.
2988 * Returns undef on error, a scalar value containing the data otherwise.
2995 TRACEME(("mstore"));
2997 if (!do_store((PerlIO*) 0, sv, 0, FALSE, &out))
2998 return &PL_sv_undef;
3006 * Same as mstore(), but network order is used for integers and doubles are
3007 * emitted as strings.
3009 SV *net_mstore(SV *sv)
3014 TRACEME(("net_mstore"));
3016 if (!do_store((PerlIO*) 0, sv, 0, TRUE, &out))
3017 return &PL_sv_undef;
3023 *** Specific retrieve callbacks.
3029 * Return an error via croak, since it is not possible that we get here
3030 * under normal conditions, when facing a file produced via pstore().
3032 static SV *retrieve_other(stcxt_t *cxt, char *cname)
3035 cxt->ver_major != STORABLE_BIN_MAJOR &&
3036 cxt->ver_minor != STORABLE_BIN_MINOR
3038 CROAK(("Corrupted storable %s (binary v%d.%d), current is v%d.%d",
3039 cxt->fio ? "file" : "string",
3040 cxt->ver_major, cxt->ver_minor,
3041 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
3043 CROAK(("Corrupted storable %s (binary v%d.%d)",
3044 cxt->fio ? "file" : "string",
3045 cxt->ver_major, cxt->ver_minor));
3048 return (SV *) 0; /* Just in case */
3052 * retrieve_idx_blessed
3054 * Layout is SX_IX_BLESS <index> <object> with SX_IX_BLESS already read.
3055 * <index> can be coded on either 1 or 5 bytes.
3057 static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname)
3064 TRACEME(("retrieve_idx_blessed (#%d)", cxt->tagnum));
3065 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3067 GETMARK(idx); /* Index coded on a single char? */
3072 * Fetch classname in `aclass'
3075 sva = av_fetch(cxt->aclass, idx, FALSE);
3077 CROAK(("Class name #%d should have been seen already", (int)idx));
3079 class = SvPVX(*sva); /* We know it's a PV, by construction */
3081 TRACEME(("class ID %d => %s", idx, class));
3084 * Retrieve object and bless it.
3087 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
3095 * Layout is SX_BLESS <len> <classname> <object> with SX_BLESS already read.
3096 * <len> can be coded on either 1 or 5 bytes.
3098 static SV *retrieve_blessed(stcxt_t *cxt, char *cname)
3102 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3105 TRACEME(("retrieve_blessed (#%d)", cxt->tagnum));
3106 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3109 * Decode class name length and read that name.
3111 * Short classnames have two advantages: their length is stored on one
3112 * single byte, and the string can be read on the stack.
3115 GETMARK(len); /* Length coded on a single char? */
3118 TRACEME(("** allocating %d bytes for class name", len+1));
3119 New(10003, class, len+1, char);
3122 class[len] = '\0'; /* Mark string end */
3125 * It's a new classname, otherwise it would have been an SX_IX_BLESS.
3128 TRACEME(("new class name \"%s\" will bear ID = %d", class, cxt->classnum));
3130 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3134 * Retrieve object and bless it.
3137 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
3147 * Layout: SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
3148 * with leading mark already read, as usual.
3150 * When recursion was involved during serialization of the object, there
3151 * is an unknown amount of serialized objects after the SX_HOOK mark. Until
3152 * we reach a <flags> marker with the recursion bit cleared.
3154 * If the first <flags> byte contains a type of SHT_EXTRA, then the real type
3155 * is held in the <extra> byte, and if the object is tied, the serialized
3156 * magic object comes at the very end:
3158 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
3160 * This means the STORABLE_thaw hook will NOT get a tied variable during its
3161 * processing (since we won't have seen the magic object by the time the hook
3162 * is called). See comments below for why it was done that way.
3164 static SV *retrieve_hook(stcxt_t *cxt, char *cname)
3167 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3179 int clone = cxt->optype & ST_CLONE;
3181 unsigned int extra_type = 0;
3183 TRACEME(("retrieve_hook (#%d)", cxt->tagnum));
3184 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3187 * Read flags, which tell us about the type, and whether we need to recurse.
3193 * Create the (empty) object, and mark it as seen.
3195 * This must be done now, because tags are incremented, and during
3196 * serialization, the object tag was affected before recursion could
3200 obj_type = flags & SHF_TYPE_MASK;
3206 sv = (SV *) newAV();
3209 sv = (SV *) newHV();
3213 * Read <extra> flag to know the type of the object.
3214 * Record associated magic type for later.
3216 GETMARK(extra_type);
3217 switch (extra_type) {
3223 sv = (SV *) newAV();
3227 sv = (SV *) newHV();
3231 return retrieve_other(cxt, 0); /* Let it croak */
3235 return retrieve_other(cxt, 0); /* Let it croak */
3237 SEEN(sv, 0); /* Don't bless yet */
3240 * Whilst flags tell us to recurse, do so.
3242 * We don't need to remember the addresses returned by retrieval, because
3243 * all the references will be obtained through indirection via the object
3244 * tags in the object-ID list.
3247 while (flags & SHF_NEED_RECURSE) {
3248 TRACEME(("retrieve_hook recursing..."));
3249 rv = retrieve(cxt, 0);
3252 TRACEME(("retrieve_hook back with rv=0x%"UVxf,
3257 if (flags & SHF_IDX_CLASSNAME) {
3262 * Fetch index from `aclass'
3265 if (flags & SHF_LARGE_CLASSLEN)
3270 sva = av_fetch(cxt->aclass, idx, FALSE);
3272 CROAK(("Class name #%d should have been seen already", (int)idx));
3274 class = SvPVX(*sva); /* We know it's a PV, by construction */
3275 TRACEME(("class ID %d => %s", idx, class));
3279 * Decode class name length and read that name.
3281 * NOTA BENE: even if the length is stored on one byte, we don't read
3282 * on the stack. Just like retrieve_blessed(), we limit the name to
3283 * LG_BLESS bytes. This is an arbitrary decision.
3286 if (flags & SHF_LARGE_CLASSLEN)
3291 if (len > LG_BLESS) {
3292 TRACEME(("** allocating %d bytes for class name", len+1));
3293 New(10003, class, len+1, char);
3297 class[len] = '\0'; /* Mark string end */
3300 * Record new classname.
3303 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3307 TRACEME(("class name: %s", class));
3310 * Decode user-frozen string length and read it in a SV.
3312 * For efficiency reasons, we read data directly into the SV buffer.
3313 * To understand that code, read retrieve_scalar()
3316 if (flags & SHF_LARGE_STRLEN)
3321 frozen = NEWSV(10002, len2);
3323 SAFEREAD(SvPVX(frozen), len2, frozen);
3324 SvCUR_set(frozen, len2);
3325 *SvEND(frozen) = '\0';
3327 (void) SvPOK_only(frozen); /* Validates string pointer */
3328 if (cxt->s_tainted) /* Is input source tainted? */
3331 TRACEME(("frozen string: %d bytes", len2));
3334 * Decode object-ID list length, if present.
3337 if (flags & SHF_HAS_LIST) {
3338 if (flags & SHF_LARGE_LISTLEN)
3344 av_extend(av, len3 + 1); /* Leave room for [0] */
3345 AvFILLp(av) = len3; /* About to be filled anyway */
3349 TRACEME(("has %d object IDs to link", len3));
3352 * Read object-ID list into array.
3353 * Because we pre-extended it, we can cheat and fill it manually.
3355 * We read object tags and we can convert them into SV* on the fly
3356 * because we know all the references listed in there (as tags)
3357 * have been already serialized, hence we have a valid correspondance
3358 * between each of those tags and the recreated SV.
3362 SV **ary = AvARRAY(av);
3364 for (i = 1; i <= len3; i++) { /* We leave [0] alone */
3371 svh = av_fetch(cxt->aseen, tag, FALSE);
3373 CROAK(("Object #%d should have been retrieved already", (int)tag));
3375 ary[i] = SvREFCNT_inc(xsv);
3380 * Bless the object and look up the STORABLE_thaw hook.
3384 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3387 * Hook not found. Maybe they did not require the module where this
3388 * hook is defined yet?
3390 * If the require below succeeds, we'll be able to find the hook.
3391 * Still, it only works reliably when each class is defined in a
3395 SV *psv = newSVpvn("require ", 8);
3396 sv_catpv(psv, class);
3398 TRACEME(("No STORABLE_thaw defined for objects of class %s", class));
3399 TRACEME(("Going to require module '%s' with '%s'", class, SvPVX(psv)));
3401 perl_eval_sv(psv, G_DISCARD);
3405 * We cache results of pkg_can, so we need to uncache before attempting
3409 pkg_uncache(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3410 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
3413 CROAK(("No STORABLE_thaw defined for objects of class %s "
3414 "(even after a \"require %s;\")", class, class));
3418 * If we don't have an `av' yet, prepare one.
3419 * Then insert the frozen string as item [0].
3427 AvARRAY(av)[0] = SvREFCNT_inc(frozen);
3432 * $object->STORABLE_thaw($cloning, $frozen, @refs);
3434 * where $object is our blessed (empty) object, $cloning is a boolean
3435 * telling whether we're running a deep clone, $frozen is the frozen
3436 * string the user gave us in his serializing hook, and @refs, which may
3437 * be empty, is the list of extra references he returned along for us
3440 * In effect, the hook is an alternate creation routine for the class,
3441 * the object itself being already created by the runtime.
3444 TRACEME(("calling STORABLE_thaw on %s at 0x%"UVxf" (%"IVdf" args)",
3445 class, PTR2UV(sv), AvFILLp(av) + 1));
3448 (void) scalar_call(rv, hook, clone, av, G_SCALAR|G_DISCARD);
3455 SvREFCNT_dec(frozen);
3458 if (!(flags & SHF_IDX_CLASSNAME) && class != buf)
3462 * If we had an <extra> type, then the object was not as simple, and
3463 * we need to restore extra magic now.
3469 TRACEME(("retrieving magic object for 0x%"UVxf"...", PTR2UV(sv)));
3471 rv = retrieve(cxt, 0); /* Retrieve <magic object> */
3473 TRACEME(("restoring the magic object 0x%"UVxf" part of 0x%"UVxf,
3474 PTR2UV(rv), PTR2UV(sv)));
3476 switch (extra_type) {
3478 sv_upgrade(sv, SVt_PVMG);
3481 sv_upgrade(sv, SVt_PVAV);
3482 AvREAL_off((AV *)sv);
3485 sv_upgrade(sv, SVt_PVHV);
3488 CROAK(("Forgot to deal with extra type %d", extra_type));
3493 * Adding the magic only now, well after the STORABLE_thaw hook was called
3494 * means the hook cannot know it deals with an object whose variable is
3495 * tied. But this is happening when retrieving $o in the following case:
3499 * my $o = bless \%h, 'BAR';
3501 * The 'BAR' class is NOT the one where %h is tied into. Therefore, as
3502 * far as the 'BAR' class is concerned, the fact that %h is not a REAL
3503 * hash but a tied one should not matter at all, and remain transparent.
3504 * This means the magic must be restored by Storable AFTER the hook is
3507 * That looks very reasonable to me, but then I've come up with this
3508 * after a bug report from David Nesting, who was trying to store such
3509 * an object and caused Storable to fail. And unfortunately, it was
3510 * also the easiest way to retrofit support for blessed ref to tied objects
3511 * into the existing design. -- RAM, 17/02/2001
3514 sv_magic(sv, rv, mtype, Nullch, 0);
3515 SvREFCNT_dec(rv); /* Undo refcnt inc from sv_magic() */
3523 * Retrieve reference to some other scalar.
3524 * Layout is SX_REF <object>, with SX_REF already read.
3526 static SV *retrieve_ref(stcxt_t *cxt, char *cname)
3531 TRACEME(("retrieve_ref (#%d)", cxt->tagnum));
3534 * We need to create the SV that holds the reference to the yet-to-retrieve
3535 * object now, so that we may record the address in the seen table.
3536 * Otherwise, if the object to retrieve references us, we won't be able
3537 * to resolve the SX_OBJECT we'll see at that point! Hence we cannot
3538 * do the retrieve first and use rv = newRV(sv) since it will be too late
3539 * for SEEN() recording.
3542 rv = NEWSV(10002, 0);
3543 SEEN(rv, cname); /* Will return if rv is null */
3544 sv = retrieve(cxt, 0); /* Retrieve <object> */
3546 return (SV *) 0; /* Failed */
3549 * WARNING: breaks RV encapsulation.
3551 * Now for the tricky part. We have to upgrade our existing SV, so that
3552 * it is now an RV on sv... Again, we cheat by duplicating the code
3553 * held in newSVrv(), since we already got our SV from retrieve().
3557 * SvRV(rv) = SvREFCNT_inc(sv);
3559 * here because the reference count we got from retrieve() above is
3560 * already correct: if the object was retrieved from the file, then
3561 * its reference count is one. Otherwise, if it was retrieved via
3562 * an SX_OBJECT indication, a ref count increment was done.
3565 sv_upgrade(rv, SVt_RV);
3566 SvRV(rv) = sv; /* $rv = \$sv */
3569 TRACEME(("ok (retrieve_ref at 0x%"UVxf")", PTR2UV(rv)));
3575 * retrieve_overloaded
3577 * Retrieve reference to some other scalar with overloading.
3578 * Layout is SX_OVERLOAD <object>, with SX_OVERLOAD already read.
3580 static SV *retrieve_overloaded(stcxt_t *cxt, char *cname)
3586 TRACEME(("retrieve_overloaded (#%d)", cxt->tagnum));
3589 * Same code as retrieve_ref(), duplicated to avoid extra call.
3592 rv = NEWSV(10002, 0);
3593 SEEN(rv, cname); /* Will return if rv is null */
3594 sv = retrieve(cxt, 0); /* Retrieve <object> */
3596 return (SV *) 0; /* Failed */
3599 * WARNING: breaks RV encapsulation.
3602 sv_upgrade(rv, SVt_RV);
3603 SvRV(rv) = sv; /* $rv = \$sv */
3607 * Restore overloading magic.
3610 stash = (HV *) SvSTASH (sv);
3611 if (!stash || !Gv_AMG(stash))
3612 CROAK(("Cannot restore overloading on %s(0x%"UVxf") (package %s)",
3613 sv_reftype(sv, FALSE),
3615 stash ? HvNAME(stash) : "<unknown>"));
3619 TRACEME(("ok (retrieve_overloaded at 0x%"UVxf")", PTR2UV(rv)));
3625 * retrieve_tied_array
3627 * Retrieve tied array
3628 * Layout is SX_TIED_ARRAY <object>, with SX_TIED_ARRAY already read.
3630 static SV *retrieve_tied_array(stcxt_t *cxt, char *cname)
3635 TRACEME(("retrieve_tied_array (#%d)", cxt->tagnum));
3637 tv = NEWSV(10002, 0);
3638 SEEN(tv, cname); /* Will return if tv is null */
3639 sv = retrieve(cxt, 0); /* Retrieve <object> */
3641 return (SV *) 0; /* Failed */
3643 sv_upgrade(tv, SVt_PVAV);
3644 AvREAL_off((AV *)tv);
3645 sv_magic(tv, sv, 'P', Nullch, 0);
3646 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3648 TRACEME(("ok (retrieve_tied_array at 0x%"UVxf")", PTR2UV(tv)));
3654 * retrieve_tied_hash
3656 * Retrieve tied hash
3657 * Layout is SX_TIED_HASH <object>, with SX_TIED_HASH already read.
3659 static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname)
3664 TRACEME(("retrieve_tied_hash (#%d)", cxt->tagnum));
3666 tv = NEWSV(10002, 0);
3667 SEEN(tv, cname); /* Will return if tv is null */
3668 sv = retrieve(cxt, 0); /* Retrieve <object> */
3670 return (SV *) 0; /* Failed */
3672 sv_upgrade(tv, SVt_PVHV);
3673 sv_magic(tv, sv, 'P', Nullch, 0);
3674 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3676 TRACEME(("ok (retrieve_tied_hash at 0x%"UVxf")", PTR2UV(tv)));
3682 * retrieve_tied_scalar
3684 * Retrieve tied scalar
3685 * Layout is SX_TIED_SCALAR <object>, with SX_TIED_SCALAR already read.
3687 static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname)
3692 TRACEME(("retrieve_tied_scalar (#%d)", cxt->tagnum));
3694 tv = NEWSV(10002, 0);
3695 SEEN(tv, cname); /* Will return if rv is null */
3696 sv = retrieve(cxt, 0); /* Retrieve <object> */
3698 return (SV *) 0; /* Failed */
3700 sv_upgrade(tv, SVt_PVMG);
3701 sv_magic(tv, sv, 'q', Nullch, 0);
3702 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3704 TRACEME(("ok (retrieve_tied_scalar at 0x%"UVxf")", PTR2UV(tv)));
3712 * Retrieve reference to value in a tied hash.
3713 * Layout is SX_TIED_KEY <object> <key>, with SX_TIED_KEY already read.
3715 static SV *retrieve_tied_key(stcxt_t *cxt, char *cname)
3721 TRACEME(("retrieve_tied_key (#%d)", cxt->tagnum));
3723 tv = NEWSV(10002, 0);
3724 SEEN(tv, cname); /* Will return if tv is null */
3725 sv = retrieve(cxt, 0); /* Retrieve <object> */
3727 return (SV *) 0; /* Failed */
3729 key = retrieve(cxt, 0); /* Retrieve <key> */
3731 return (SV *) 0; /* Failed */
3733 sv_upgrade(tv, SVt_PVMG);
3734 sv_magic(tv, sv, 'p', (char *)key, HEf_SVKEY);
3735 SvREFCNT_dec(key); /* Undo refcnt inc from sv_magic() */
3736 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3744 * Retrieve reference to value in a tied array.
3745 * Layout is SX_TIED_IDX <object> <idx>, with SX_TIED_IDX already read.
3747 static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname)
3753 TRACEME(("retrieve_tied_idx (#%d)", cxt->tagnum));
3755 tv = NEWSV(10002, 0);
3756 SEEN(tv, cname); /* Will return if tv is null */
3757 sv = retrieve(cxt, 0); /* Retrieve <object> */
3759 return (SV *) 0; /* Failed */
3761 RLEN(idx); /* Retrieve <idx> */
3763 sv_upgrade(tv, SVt_PVMG);
3764 sv_magic(tv, sv, 'p', Nullch, idx);
3765 SvREFCNT_dec(sv); /* Undo refcnt inc from sv_magic() */
3774 * Retrieve defined long (string) scalar.
3776 * Layout is SX_LSCALAR <length> <data>, with SX_LSCALAR already read.
3777 * The scalar is "long" in that <length> is larger than LG_SCALAR so it
3778 * was not stored on a single byte.
3780 static SV *retrieve_lscalar(stcxt_t *cxt, char *cname)
3786 TRACEME(("retrieve_lscalar (#%d), len = %"IVdf, cxt->tagnum, len));
3789 * Allocate an empty scalar of the suitable length.
3792 sv = NEWSV(10002, len);
3793 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
3796 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
3798 * Now, for efficiency reasons, read data directly inside the SV buffer,
3799 * and perform the SV final settings directly by duplicating the final
3800 * work done by sv_setpv. Since we're going to allocate lots of scalars
3801 * this way, it's worth the hassle and risk.
3804 SAFEREAD(SvPVX(sv), len, sv);
3805 SvCUR_set(sv, len); /* Record C string length */
3806 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
3807 (void) SvPOK_only(sv); /* Validate string pointer */
3808 if (cxt->s_tainted) /* Is input source tainted? */
3809 SvTAINT(sv); /* External data cannot be trusted */
3811 TRACEME(("large scalar len %"IVdf" '%s'", len, SvPVX(sv)));
3812 TRACEME(("ok (retrieve_lscalar at 0x%"UVxf")", PTR2UV(sv)));
3820 * Retrieve defined short (string) scalar.
3822 * Layout is SX_SCALAR <length> <data>, with SX_SCALAR already read.
3823 * The scalar is "short" so <length> is single byte. If it is 0, there
3824 * is no <data> section.
3826 static SV *retrieve_scalar(stcxt_t *cxt, char *cname)
3832 TRACEME(("retrieve_scalar (#%d), len = %d", cxt->tagnum, len));
3835 * Allocate an empty scalar of the suitable length.
3838 sv = NEWSV(10002, len);
3839 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
3842 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
3847 * newSV did not upgrade to SVt_PV so the scalar is undefined.
3848 * To make it defined with an empty length, upgrade it now...
3850 sv_upgrade(sv, SVt_PV);
3852 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
3853 TRACEME(("ok (retrieve_scalar empty at 0x%"UVxf")", PTR2UV(sv)));
3856 * Now, for efficiency reasons, read data directly inside the SV buffer,
3857 * and perform the SV final settings directly by duplicating the final
3858 * work done by sv_setpv. Since we're going to allocate lots of scalars
3859 * this way, it's worth the hassle and risk.
3861 SAFEREAD(SvPVX(sv), len, sv);
3862 SvCUR_set(sv, len); /* Record C string length */
3863 *SvEND(sv) = '\0'; /* Ensure it's null terminated anyway */
3864 TRACEME(("small scalar len %d '%s'", len, SvPVX(sv)));
3867 (void) SvPOK_only(sv); /* Validate string pointer */
3868 if (cxt->s_tainted) /* Is input source tainted? */
3869 SvTAINT(sv); /* External data cannot be trusted */
3871 TRACEME(("ok (retrieve_scalar at 0x%"UVxf")", PTR2UV(sv)));
3878 * Like retrieve_scalar(), but tag result as utf8.
3879 * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
3881 static SV *retrieve_utf8str(stcxt_t *cxt, char *cname)
3885 TRACEME(("retrieve_utf8str"));
3887 sv = retrieve_scalar(cxt, cname);
3897 * Like retrieve_lscalar(), but tag result as utf8.
3898 * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
3900 static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname)
3904 TRACEME(("retrieve_lutf8str"));
3906 sv = retrieve_lscalar(cxt, cname);
3916 * Retrieve defined integer.
3917 * Layout is SX_INTEGER <data>, whith SX_INTEGER already read.
3919 static SV *retrieve_integer(stcxt_t *cxt, char *cname)
3924 TRACEME(("retrieve_integer (#%d)", cxt->tagnum));
3926 READ(&iv, sizeof(iv));
3928 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
3930 TRACEME(("integer %"IVdf, iv));
3931 TRACEME(("ok (retrieve_integer at 0x%"UVxf")", PTR2UV(sv)));
3939 * Retrieve defined integer in network order.
3940 * Layout is SX_NETINT <data>, whith SX_NETINT already read.
3942 static SV *retrieve_netint(stcxt_t *cxt, char *cname)
3947 TRACEME(("retrieve_netint (#%d)", cxt->tagnum));
3951 sv = newSViv((int) ntohl(iv));
3952 TRACEME(("network integer %d", (int) ntohl(iv)));
3955 TRACEME(("network integer (as-is) %d", iv));
3957 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
3959 TRACEME(("ok (retrieve_netint at 0x%"UVxf")", PTR2UV(sv)));
3967 * Retrieve defined double.
3968 * Layout is SX_DOUBLE <data>, whith SX_DOUBLE already read.
3970 static SV *retrieve_double(stcxt_t *cxt, char *cname)
3975 TRACEME(("retrieve_double (#%d)", cxt->tagnum));
3977 READ(&nv, sizeof(nv));
3979 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
3981 TRACEME(("double %"NVff, nv));
3982 TRACEME(("ok (retrieve_double at 0x%"UVxf")", PTR2UV(sv)));
3990 * Retrieve defined byte (small integer within the [-128, +127] range).
3991 * Layout is SX_BYTE <data>, whith SX_BYTE already read.
3993 static SV *retrieve_byte(stcxt_t *cxt, char *cname)
3999 TRACEME(("retrieve_byte (#%d)", cxt->tagnum));
4002 TRACEME(("small integer read as %d", (unsigned char) siv));
4003 tmp = ((unsigned char)siv) - 128;
4006 SEEN(sv, cname); /* Associate this new scalar with tag "tagnum" */
4008 TRACEME(("byte %d", tmp));
4009 TRACEME(("ok (retrieve_byte at 0x%"UVxf")", PTR2UV(sv)));
4017 * Return the undefined value.
4019 static SV *retrieve_undef(stcxt_t *cxt, char *cname)
4023 TRACEME(("retrieve_undef"));
4034 * Return the immortal undefined value.
4036 static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname)
4038 SV *sv = &PL_sv_undef;
4040 TRACEME(("retrieve_sv_undef"));
4049 * Return the immortal yes value.
4051 static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname)
4053 SV *sv = &PL_sv_yes;
4055 TRACEME(("retrieve_sv_yes"));
4064 * Return the immortal no value.
4066 static SV *retrieve_sv_no(stcxt_t *cxt, char *cname)
4070 TRACEME(("retrieve_sv_no"));
4079 * Retrieve a whole array.
4080 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
4081 * Each item is stored as <object>.
4083 * When we come here, SX_ARRAY has been read already.
4085 static SV *retrieve_array(stcxt_t *cxt, char *cname)
4092 TRACEME(("retrieve_array (#%d)", cxt->tagnum));
4095 * Read length, and allocate array, then pre-extend it.
4099 TRACEME(("size = %d", len));
4101 SEEN(av, cname); /* Will return if array not allocated nicely */
4105 return (SV *) av; /* No data follow if array is empty */
4108 * Now get each item in turn...
4111 for (i = 0; i < len; i++) {
4112 TRACEME(("(#%d) item", i));
4113 sv = retrieve(cxt, 0); /* Retrieve item */
4116 if (av_store(av, i, sv) == 0)
4120 TRACEME(("ok (retrieve_array at 0x%"UVxf")", PTR2UV(av)));
4128 * Retrieve a whole hash table.
4129 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4130 * Keys are stored as <length> <data>, the <data> section being omitted
4132 * Values are stored as <object>.
4134 * When we come here, SX_HASH has been read already.
4136 static SV *retrieve_hash(stcxt_t *cxt, char *cname)
4143 static SV *sv_h_undef = (SV *) 0; /* hv_store() bug */
4145 TRACEME(("retrieve_hash (#%d)", cxt->tagnum));
4148 * Read length, allocate table.
4152 TRACEME(("size = %d", len));
4154 SEEN(hv, cname); /* Will return if table not allocated properly */
4156 return (SV *) hv; /* No data follow if table empty */
4159 * Now get each key/value pair in turn...
4162 for (i = 0; i < len; i++) {
4167 TRACEME(("(#%d) value", i));
4168 sv = retrieve(cxt, 0);
4174 * Since we're reading into kbuf, we must ensure we're not
4175 * recursing between the read and the hv_store() where it's used.
4176 * Hence the key comes after the value.
4179 RLEN(size); /* Get key size */
4180 KBUFCHK(size); /* Grow hash key read pool if needed */
4183 kbuf[size] = '\0'; /* Mark string end, just in case */
4184 TRACEME(("(#%d) key '%s'", i, kbuf));
4187 * Enter key/value pair into hash table.
4190 if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
4194 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4200 * old_retrieve_array
4202 * Retrieve a whole array in pre-0.6 binary format.
4204 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
4205 * Each item is stored as SX_ITEM <object> or SX_IT_UNDEF for "holes".
4207 * When we come here, SX_ARRAY has been read already.
4209 static SV *old_retrieve_array(stcxt_t *cxt, char *cname)
4217 TRACEME(("old_retrieve_array (#%d)", cxt->tagnum));
4220 * Read length, and allocate array, then pre-extend it.
4224 TRACEME(("size = %d", len));
4226 SEEN(av, 0); /* Will return if array not allocated nicely */
4230 return (SV *) av; /* No data follow if array is empty */
4233 * Now get each item in turn...
4236 for (i = 0; i < len; i++) {
4238 if (c == SX_IT_UNDEF) {
4239 TRACEME(("(#%d) undef item", i));
4240 continue; /* av_extend() already filled us with undef */
4243 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
4244 TRACEME(("(#%d) item", i));
4245 sv = retrieve(cxt, 0); /* Retrieve item */
4248 if (av_store(av, i, sv) == 0)
4252 TRACEME(("ok (old_retrieve_array at 0x%"UVxf")", PTR2UV(av)));
4260 * Retrieve a whole hash table in pre-0.6 binary format.
4262 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
4263 * Keys are stored as SX_KEY <length> <data>, the <data> section being omitted
4265 * Values are stored as SX_VALUE <object> or SX_VL_UNDEF for "holes".
4267 * When we come here, SX_HASH has been read already.
4269 static SV *old_retrieve_hash(stcxt_t *cxt, char *cname)
4277 static SV *sv_h_undef = (SV *) 0; /* hv_store() bug */
4279 TRACEME(("old_retrieve_hash (#%d)", cxt->tagnum));
4282 * Read length, allocate table.
4286 TRACEME(("size = %d", len));
4288 SEEN(hv, 0); /* Will return if table not allocated properly */
4290 return (SV *) hv; /* No data follow if table empty */
4293 * Now get each key/value pair in turn...
4296 for (i = 0; i < len; i++) {
4302 if (c == SX_VL_UNDEF) {
4303 TRACEME(("(#%d) undef value", i));
4305 * Due to a bug in hv_store(), it's not possible to pass
4306 * &PL_sv_undef to hv_store() as a value, otherwise the
4307 * associated key will not be creatable any more. -- RAM, 14/01/97
4310 sv_h_undef = newSVsv(&PL_sv_undef);
4311 sv = SvREFCNT_inc(sv_h_undef);
4312 } else if (c == SX_VALUE) {
4313 TRACEME(("(#%d) value", i));
4314 sv = retrieve(cxt, 0);
4318 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
4322 * Since we're reading into kbuf, we must ensure we're not
4323 * recursing between the read and the hv_store() where it's used.
4324 * Hence the key comes after the value.
4329 (void) retrieve_other((stcxt_t *) 0, 0); /* Will croak out */
4330 RLEN(size); /* Get key size */
4331 KBUFCHK(size); /* Grow hash key read pool if needed */
4334 kbuf[size] = '\0'; /* Mark string end, just in case */
4335 TRACEME(("(#%d) key '%s'", i, kbuf));
4338 * Enter key/value pair into hash table.
4341 if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
4345 TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
4351 *** Retrieval engine.
4357 * Make sure the stored data we're trying to retrieve has been produced
4358 * on an ILP compatible system with the same byteorder. It croaks out in
4359 * case an error is detected. [ILP = integer-long-pointer sizes]
4360 * Returns null if error is detected, &PL_sv_undef otherwise.
4362 * Note that there's no byte ordering info emitted when network order was
4363 * used at store time.
4365 static SV *magic_check(stcxt_t *cxt)
4368 char byteorder[256];
4370 int use_network_order;
4372 int version_minor = 0;
4374 TRACEME(("magic_check"));
4377 * The "magic number" is only for files, not when freezing in memory.
4381 STRLEN len = sizeof(magicstr) - 1;
4384 READ(buf, len); /* Not null-terminated */
4385 buf[len] = '\0'; /* Is now */
4387 if (0 == strcmp(buf, magicstr))
4391 * Try to read more bytes to check for the old magic number, which
4395 old_len = sizeof(old_magicstr) - 1;
4396 READ(&buf[len], old_len - len);
4397 buf[old_len] = '\0'; /* Is now null-terminated */
4399 if (strcmp(buf, old_magicstr))
4400 CROAK(("File is not a perl storable"));
4405 * Starting with 0.6, the "use_network_order" byte flag is also used to
4406 * indicate the version number of the binary, and therefore governs the
4407 * setting of sv_retrieve_vtbl. See magic_write().
4410 GETMARK(use_network_order);
4411 version_major = use_network_order >> 1;
4412 cxt->retrieve_vtbl = version_major ? sv_retrieve : sv_old_retrieve;
4414 TRACEME(("magic_check: netorder = 0x%x", use_network_order));
4418 * Starting with 0.7 (binary major 2), a full byte is dedicated to the
4419 * minor version of the protocol. See magic_write().
4422 if (version_major > 1)
4423 GETMARK(version_minor);
4425 cxt->ver_major = version_major;
4426 cxt->ver_minor = version_minor;
4428 TRACEME(("binary image version is %d.%d", version_major, version_minor));
4431 * Inter-operability sanity check: we can't retrieve something stored
4432 * using a format more recent than ours, because we have no way to
4433 * know what has changed, and letting retrieval go would mean a probable
4434 * failure reporting a "corrupted" storable file.
4438 version_major > STORABLE_BIN_MAJOR ||
4439 (version_major == STORABLE_BIN_MAJOR &&
4440 version_minor > STORABLE_BIN_MINOR)
4442 CROAK(("Storable binary image v%d.%d more recent than I am (v%d.%d)",
4443 version_major, version_minor,
4444 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
4447 * If they stored using network order, there's no byte ordering
4448 * information to check.
4451 if (cxt->netorder = (use_network_order & 0x1))
4452 return &PL_sv_undef; /* No byte ordering info */
4454 sprintf(byteorder, "%lx", (unsigned long) BYTEORDER);
4456 READ(buf, c); /* Not null-terminated */
4457 buf[c] = '\0'; /* Is now */
4459 if (strcmp(buf, byteorder))
4460 CROAK(("Byte order is not compatible"));
4462 GETMARK(c); /* sizeof(int) */
4463 if ((int) c != sizeof(int))
4464 CROAK(("Integer size is not compatible"));
4466 GETMARK(c); /* sizeof(long) */
4467 if ((int) c != sizeof(long))
4468 CROAK(("Long integer size is not compatible"));
4470 GETMARK(c); /* sizeof(char *) */
4471 if ((int) c != sizeof(char *))
4472 CROAK(("Pointer integer size is not compatible"));
4474 if (version_major >= 2 && version_minor >= 2) {
4475 GETMARK(c); /* sizeof(NV) */
4476 if ((int) c != sizeof(NV))
4477 CROAK(("Double size is not compatible"));
4480 return &PL_sv_undef; /* OK */
4486 * Recursively retrieve objects from the specified file and return their
4487 * root SV (which may be an AV or an HV for what we care).
4488 * Returns null if there is a problem.
4490 static SV *retrieve(stcxt_t *cxt, char *cname)
4496 TRACEME(("retrieve"));
4499 * Grab address tag which identifies the object if we are retrieving
4500 * an older format. Since the new binary format counts objects and no
4501 * longer explicitely tags them, we must keep track of the correspondance
4504 * The following section will disappear one day when the old format is
4505 * no longer supported, hence the final "goto" in the "if" block.
4508 if (cxt->hseen) { /* Retrieving old binary */
4510 if (cxt->netorder) {
4512 READ(&nettag, sizeof(I32)); /* Ordered sequence of I32 */
4513 tag = (stag_t) nettag;
4515 READ(&tag, sizeof(stag_t)); /* Original address of the SV */
4518 if (type == SX_OBJECT) {
4520 svh = hv_fetch(cxt->hseen, (char *) &tag, sizeof(tag), FALSE);
4522 CROAK(("Old tag 0x%x should have been mapped already", (unsigned)tag));
4523 tagn = SvIV(*svh); /* Mapped tag number computed earlier below */
4526 * The following code is common with the SX_OBJECT case below.
4529 svh = av_fetch(cxt->aseen, tagn, FALSE);
4531 CROAK(("Object #%d should have been retrieved already", (int)tagn));
4533 TRACEME(("has retrieved #%d at 0x%"UVxf, tagn, PTR2UV(sv)));
4534 SvREFCNT_inc(sv); /* One more reference to this same sv */
4535 return sv; /* The SV pointer where object was retrieved */
4539 * Map new object, but don't increase tagnum. This will be done
4540 * by each of the retrieve_* functions when they call SEEN().
4542 * The mapping associates the "tag" initially present with a unique
4543 * tag number. See test for SX_OBJECT above to see how this is perused.
4546 if (!hv_store(cxt->hseen, (char *) &tag, sizeof(tag),
4547 newSViv(cxt->tagnum), 0))
4554 * Regular post-0.6 binary format.
4560 TRACEME(("retrieve type = %d", type));
4563 * Are we dealing with an object we should have already retrieved?
4566 if (type == SX_OBJECT) {
4570 svh = av_fetch(cxt->aseen, tag, FALSE);
4572 CROAK(("Object #%d should have been retrieved already", (int)tag));
4574 TRACEME(("had retrieved #%d at 0x%"UVxf, tag, PTR2UV(sv)));
4575 SvREFCNT_inc(sv); /* One more reference to this same sv */
4576 return sv; /* The SV pointer where object was retrieved */
4579 first_time: /* Will disappear when support for old format is dropped */
4582 * Okay, first time through for this one.
4585 sv = RETRIEVE(cxt, type)(cxt, cname);
4587 return (SV *) 0; /* Failed */
4590 * Old binary formats (pre-0.7).
4592 * Final notifications, ended by SX_STORED may now follow.
4593 * Currently, the only pertinent notification to apply on the
4594 * freshly retrieved object is either:
4595 * SX_CLASS <char-len> <classname> for short classnames.
4596 * SX_LG_CLASS <int-len> <classname> for larger one (rare!).
4597 * Class name is then read into the key buffer pool used by
4598 * hash table key retrieval.
4601 if (cxt->ver_major < 2) {
4602 while ((type = GETCHAR()) != SX_STORED) {
4606 GETMARK(len); /* Length coded on a single char */
4608 case SX_LG_CLASS: /* Length coded on a regular integer */
4613 return (SV *) 0; /* Failed */
4615 KBUFCHK(len); /* Grow buffer as necessary */
4618 kbuf[len] = '\0'; /* Mark string end */
4623 TRACEME(("ok (retrieved 0x%"UVxf", refcnt=%d, %s)", PTR2UV(sv),
4624 SvREFCNT(sv) - 1, sv_reftype(sv, FALSE)));
4632 * Retrieve data held in file and return the root object.
4633 * Common routine for pretrieve and mretrieve.
4635 static SV *do_retrieve(
4642 int is_tainted; /* Is input source tainted? */
4643 struct extendable msave; /* Where potentially valid mbuf is saved */
4645 TRACEME(("do_retrieve (optype = 0x%x)", optype));
4647 optype |= ST_RETRIEVE;
4650 * Sanity assertions for retrieve dispatch tables.
4653 ASSERT(sizeof(sv_old_retrieve) == sizeof(sv_retrieve),
4654 ("old and new retrieve dispatch table have same size"));
4655 ASSERT(sv_old_retrieve[SX_ERROR] == retrieve_other,
4656 ("SX_ERROR entry correctly initialized in old dispatch table"));
4657 ASSERT(sv_retrieve[SX_ERROR] == retrieve_other,
4658 ("SX_ERROR entry correctly initialized in new dispatch table"));
4661 * Workaround for CROAK leak: if they enter with a "dirty" context,
4662 * free up memory for them now.
4669 * Now that STORABLE_xxx hooks exist, it is possible that they try to
4670 * re-enter retrieve() via the hooks.
4674 cxt = allocate_context(cxt);
4678 ASSERT(cxt->entry == 1, ("starting new recursion"));
4679 ASSERT(!cxt->s_dirty, ("clean context"));
4684 * Data is loaded into the memory buffer when f is NULL, unless `in' is
4685 * also NULL, in which case we're expecting the data to already lie
4686 * in the buffer (dclone case).
4689 KBUFINIT(); /* Allocate hash key reading pool once */
4692 StructCopy(&cxt->membuf, &msave, struct extendable);
4698 * Magic number verifications.
4700 * This needs to be done before calling init_retrieve_context()
4701 * since the format indication in the file are necessary to conduct
4702 * some of the initializations.
4705 cxt->fio = f; /* Where I/O are performed */
4707 if (!magic_check(cxt))
4708 CROAK(("Magic number checking on storable %s failed",
4709 cxt->fio ? "file" : "string"));
4711 TRACEME(("data stored in %s format",
4712 cxt->netorder ? "net order" : "native"));
4715 * Check whether input source is tainted, so that we don't wrongly
4716 * taint perfectly good values...
4718 * We assume file input is always tainted. If both `f' and `in' are
4719 * NULL, then we come from dclone, and tainted is already filled in
4720 * the context. That's a kludge, but the whole dclone() thing is
4721 * already quite a kludge anyway! -- RAM, 15/09/2000.
4724 is_tainted = f ? 1 : (in ? SvTAINTED(in) : cxt->s_tainted);
4725 TRACEME(("input source is %s", is_tainted ? "tainted" : "trusted"));
4726 init_retrieve_context(cxt, optype, is_tainted);
4728 ASSERT(is_retrieving(), ("within retrieve operation"));
4730 sv = retrieve(cxt, 0); /* Recursively retrieve object, get root SV */
4737 StructCopy(&msave, &cxt->membuf, struct extendable);
4740 * The "root" context is never freed.
4743 clean_retrieve_context(cxt);
4744 if (cxt->prev) /* This context was stacked */
4745 free_context(cxt); /* It was not the "root" context */
4748 * Prepare returned value.
4752 TRACEME(("retrieve ERROR"));
4753 return &PL_sv_undef; /* Something went wrong, return undef */
4756 TRACEME(("retrieve got %s(0x%"UVxf")",
4757 sv_reftype(sv, FALSE), PTR2UV(sv)));
4760 * Backward compatibility with Storable-0.5@9 (which we know we
4761 * are retrieving if hseen is non-null): don't create an extra RV
4762 * for objects since we special-cased it at store time.
4764 * Build a reference to the SV returned by pretrieve even if it is
4765 * already one and not a scalar, for consistency reasons.
4767 * NB: although context might have been cleaned, the value of `cxt->hseen'
4768 * remains intact, and can be used as a flag.
4771 if (cxt->hseen) { /* Was not handling overloading by then */
4773 if (sv_type(sv) == svis_REF && (rv = SvRV(sv)) && SvOBJECT(rv))
4778 * If reference is overloaded, restore behaviour.
4780 * NB: minor glitch here: normally, overloaded refs are stored specially
4781 * so that we can croak when behaviour cannot be re-installed, and also
4782 * avoid testing for overloading magic at each reference retrieval.
4784 * Unfortunately, the root reference is implicitely stored, so we must
4785 * check for possible overloading now. Furthermore, if we don't restore
4786 * overloading, we cannot croak as if the original ref was, because we
4787 * have no way to determine whether it was an overloaded ref or not in
4790 * It's a pity that overloading magic is attached to the rv, and not to
4791 * the underlying sv as blessing is.
4795 HV *stash = (HV *) SvSTASH (sv);
4796 SV *rv = newRV_noinc(sv);
4797 if (stash && Gv_AMG(stash)) {
4799 TRACEME(("restored overloading on root reference"));
4804 return newRV_noinc(sv);
4810 * Retrieve data held in file and return the root object, undef on error.
4812 SV *pretrieve(PerlIO *f)
4814 TRACEME(("pretrieve"));
4815 return do_retrieve(f, Nullsv, 0);
4821 * Retrieve data held in scalar and return the root object, undef on error.
4823 SV *mretrieve(SV *sv)
4825 TRACEME(("mretrieve"));
4826 return do_retrieve((PerlIO*) 0, sv, 0);
4836 * Deep clone: returns a fresh copy of the original referenced SV tree.
4838 * This is achieved by storing the object in memory and restoring from
4839 * there. Not that efficient, but it should be faster than doing it from
4846 stcxt_t *real_context;
4849 TRACEME(("dclone"));
4852 * Workaround for CROAK leak: if they enter with a "dirty" context,
4853 * free up memory for them now.
4860 * do_store() optimizes for dclone by not freeing its context, should
4861 * we need to allocate one because we're deep cloning from a hook.
4864 if (!do_store((PerlIO*) 0, sv, ST_CLONE, FALSE, (SV**) 0))
4865 return &PL_sv_undef; /* Error during store */
4868 * Because of the above optimization, we have to refresh the context,
4869 * since a new one could have been allocated and stacked by do_store().
4872 { dSTCXT; real_context = cxt; } /* Sub-block needed for macro */
4873 cxt = real_context; /* And we need this temporary... */
4876 * Now, `cxt' may refer to a new context.
4879 ASSERT(!cxt->s_dirty, ("clean context"));
4880 ASSERT(!cxt->entry, ("entry will not cause new context allocation"));
4883 TRACEME(("dclone stored %d bytes", size));
4887 * Since we're passing do_retrieve() both a NULL file and sv, we need
4888 * to pre-compute the taintedness of the input by setting cxt->tainted
4889 * to whatever state our own input string was. -- RAM, 15/09/2000
4891 * do_retrieve() will free non-root context.
4894 cxt->s_tainted = SvTAINTED(sv);
4895 out = do_retrieve((PerlIO*) 0, Nullsv, ST_CLONE);
4897 TRACEME(("dclone returns 0x%"UVxf, PTR2UV(out)));
4907 * The Perl IO GV object distinguishes between input and output for sockets
4908 * but not for plain files. To allow Storable to transparently work on
4909 * plain files and sockets transparently, we have to ask xsubpp to fetch the
4910 * right object for us. Hence the OutputStream and InputStream declarations.
4912 * Before perl 5.004_05, those entries in the standard typemap are not
4913 * defined in perl include files, so we do that here.
4916 #ifndef OutputStream
4917 #define OutputStream PerlIO *
4918 #define InputStream PerlIO *
4919 #endif /* !OutputStream */
4921 MODULE = Storable PACKAGE = Storable
4959 last_op_in_netorder()