This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Clean up copy-on-write macros and debug facilities (new flag 'C').
[perl5.git] / ext / Storable / Storable.xs
CommitLineData
7a6a85bf 1/*
6ad89a2f 2 * Store and retrieve mechanism.
7a6a85bf
RG
3 *
4 * Copyright (c) 1995-2000, Raphael Manfredi
5 *
9e21b3d0
JH
6 * You may redistribute only under the same terms as Perl 5, as specified
7 * in the README file that comes with the distribution.
7a6a85bf 8 *
7a6a85bf
RG
9 */
10
11#include <EXTERN.h>
12#include <perl.h>
e993d95c 13#include <patchlevel.h> /* Perl's one, needed since 5.6 */
7a6a85bf
RG
14#include <XSUB.h>
15
5dff01bc 16#ifndef NETWARE
e8189732 17#if 0
9e21b3d0
JH
18#define DEBUGME /* Debug mode, turns assertions on as well */
19#define DASSERT /* Assertion mode */
20#endif
5dff01bc
JH
21#else /* NETWARE */
22#if 0 /* On NetWare USE_PERLIO is not used */
23#define DEBUGME /* Debug mode, turns assertions on as well */
24#define DASSERT /* Assertion mode */
25#endif
26#endif
7a6a85bf
RG
27
28/*
29 * Pre PerlIO time when none of USE_PERLIO and PERLIO_IS_STDIO is defined
30 * Provide them with the necessary defines so they can build with pre-5.004.
31 */
32#ifndef USE_PERLIO
33#ifndef PERLIO_IS_STDIO
34#define PerlIO FILE
35#define PerlIO_getc(x) getc(x)
36#define PerlIO_putc(f,x) putc(x,f)
37#define PerlIO_read(x,y,z) fread(y,1,z,x)
38#define PerlIO_write(x,y,z) fwrite(y,1,z,x)
39#define PerlIO_stdoutf printf
40#endif /* PERLIO_IS_STDIO */
41#endif /* USE_PERLIO */
42
43/*
44 * Earlier versions of perl might be used, we can't assume they have the latest!
45 */
f0ffaed8
JH
46
47#ifndef PERL_VERSION /* For perls < 5.6 */
e993d95c 48#define PERL_VERSION PATCHLEVEL
7a6a85bf
RG
49#ifndef newRV_noinc
50#define newRV_noinc(sv) ((Sv = newRV(sv)), --SvREFCNT(SvRV(Sv)), Sv)
51#endif
e993d95c 52#if (PATCHLEVEL <= 4) /* Older perls (<= 5.004) lack PL_ namespace */
7a6a85bf
RG
53#define PL_sv_yes sv_yes
54#define PL_sv_no sv_no
55#define PL_sv_undef sv_undef
e993d95c 56#if (SUBVERSION <= 4) /* 5.004_04 has been reported to lack newSVpvn */
dd19458b 57#define newSVpvn newSVpv
7a6a85bf 58#endif
e993d95c 59#endif /* PATCHLEVEL <= 4 */
7a6a85bf
RG
60#ifndef HvSHAREKEYS_off
61#define HvSHAREKEYS_off(hv) /* Ignore */
62#endif
f0ffaed8
JH
63#ifndef AvFILLp /* Older perls (<=5.003) lack AvFILLp */
64#define AvFILLp AvFILL
65#endif
66typedef double NV; /* Older perls lack the NV type */
cc964657
JH
67#define IVdf "ld" /* Various printf formats for Perl types */
68#define UVuf "lu"
69#define UVof "lo"
70#define UVxf "lx"
71#define INT2PTR(t,v) (t)(IV)(v)
72#define PTR2UV(v) (unsigned long)(v)
f0ffaed8 73#endif /* PERL_VERSION -- perls < 5.6 */
7a6a85bf 74
cc964657 75#ifndef NVef /* The following were not part of perl 5.6 */
9e21b3d0
JH
76#if defined(USE_LONG_DOUBLE) && \
77 defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
78#define NVef PERL_PRIeldbl
79#define NVff PERL_PRIfldbl
80#define NVgf PERL_PRIgldbl
81#else
cc964657
JH
82#define NVef "e"
83#define NVff "f"
84#define NVgf "g"
85#endif
86#endif
87
7a6a85bf 88#ifdef DEBUGME
8be2b38b
JH
89
90#ifndef DASSERT
91#define DASSERT
92#endif
93
90826881
JH
94/*
95 * TRACEME() will only output things when the $Storable::DEBUGME is true.
96 */
97
111e03c1
RG
98#define TRACEME(x) \
99 STMT_START { \
90826881 100 if (SvTRUE(perl_get_sv("Storable::DEBUGME", TRUE))) \
111e03c1
RG
101 { PerlIO_stdoutf x; PerlIO_stdoutf("\n"); } \
102 } STMT_END
7a6a85bf
RG
103#else
104#define TRACEME(x)
8be2b38b 105#endif /* DEBUGME */
7a6a85bf
RG
106
107#ifdef DASSERT
111e03c1
RG
108#define ASSERT(x,y) \
109 STMT_START { \
7a6a85bf
RG
110 if (!(x)) { \
111 PerlIO_stdoutf("ASSERT FAILED (\"%s\", line %d): ", \
112 __FILE__, __LINE__); \
113 PerlIO_stdoutf y; PerlIO_stdoutf("\n"); \
114 } \
111e03c1 115 } STMT_END
7a6a85bf
RG
116#else
117#define ASSERT(x,y)
118#endif
119
120/*
121 * Type markers.
122 */
123
124#define C(x) ((char) (x)) /* For markers with dynamic retrieval handling */
125
126#define SX_OBJECT C(0) /* Already stored object */
dd19458b 127#define SX_LSCALAR C(1) /* Scalar (large binary) follows (length, data) */
7a6a85bf
RG
128#define SX_ARRAY C(2) /* Array forthcominng (size, item list) */
129#define SX_HASH C(3) /* Hash forthcoming (size, key/value pair list) */
130#define SX_REF C(4) /* Reference to object forthcoming */
131#define SX_UNDEF C(5) /* Undefined scalar */
132#define SX_INTEGER C(6) /* Integer forthcoming */
133#define SX_DOUBLE C(7) /* Double forthcoming */
134#define SX_BYTE C(8) /* (signed) byte forthcoming */
135#define SX_NETINT C(9) /* Integer in network order forthcoming */
dd19458b 136#define SX_SCALAR C(10) /* Scalar (binary, small) follows (length, data) */
f062ea6c
PN
137#define SX_TIED_ARRAY C(11) /* Tied array forthcoming */
138#define SX_TIED_HASH C(12) /* Tied hash forthcoming */
139#define SX_TIED_SCALAR C(13) /* Tied scalar forthcoming */
7a6a85bf
RG
140#define SX_SV_UNDEF C(14) /* Perl's immortal PL_sv_undef */
141#define SX_SV_YES C(15) /* Perl's immortal PL_sv_yes */
142#define SX_SV_NO C(16) /* Perl's immortal PL_sv_no */
143#define SX_BLESS C(17) /* Object is blessed */
144#define SX_IX_BLESS C(18) /* Object is blessed, classname given by index */
145#define SX_HOOK C(19) /* Stored via hook, user-defined */
146#define SX_OVERLOAD C(20) /* Overloaded reference */
f062ea6c
PN
147#define SX_TIED_KEY C(21) /* Tied magic key forthcoming */
148#define SX_TIED_IDX C(22) /* Tied magic index forthcoming */
149#define SX_UTF8STR C(23) /* UTF-8 string forthcoming (small) */
150#define SX_LUTF8STR C(24) /* UTF-8 string forthcoming (large) */
151#define SX_FLAG_HASH C(25) /* Hash with flags forthcoming (size, flags, key/flags/value triplet list) */
e16e2ff8 152#define SX_ERROR C(26) /* Error */
7a6a85bf
RG
153
154/*
155 * Those are only used to retrieve "old" pre-0.6 binary images.
156 */
157#define SX_ITEM 'i' /* An array item introducer */
158#define SX_IT_UNDEF 'I' /* Undefined array item */
d1be9408
JF
159#define SX_KEY 'k' /* A hash key introducer */
160#define SX_VALUE 'v' /* A hash value introducer */
7a6a85bf
RG
161#define SX_VL_UNDEF 'V' /* Undefined hash value */
162
163/*
164 * Those are only used to retrieve "old" pre-0.7 binary images
165 */
166
167#define SX_CLASS 'b' /* Object is blessed, class name length <255 */
f062ea6c 168#define SX_LG_CLASS 'B' /* Object is blessed, class name length >255 */
7a6a85bf
RG
169#define SX_STORED 'X' /* End of object */
170
171/*
172 * Limits between short/long length representation.
173 */
174
175#define LG_SCALAR 255 /* Large scalar length limit */
176#define LG_BLESS 127 /* Large classname bless limit */
177
178/*
179 * Operation types
180 */
181
182#define ST_STORE 0x1 /* Store operation */
183#define ST_RETRIEVE 0x2 /* Retrieval operation */
184#define ST_CLONE 0x4 /* Deep cloning operation */
185
186/*
187 * The following structure is used for hash table key retrieval. Since, when
188 * retrieving objects, we'll be facing blessed hash references, it's best
189 * to pre-allocate that buffer once and resize it as the need arises, never
190 * freeing it (keys will be saved away someplace else anyway, so even large
191 * keys are not enough a motivation to reclaim that space).
192 *
193 * This structure is also used for memory store/retrieve operations which
194 * happen in a fixed place before being malloc'ed elsewhere if persistency
195 * is required. Hence the aptr pointer.
196 */
197struct extendable {
198 char *arena; /* Will hold hash key strings, resized as needed */
199 STRLEN asiz; /* Size of aforementionned buffer */
200 char *aptr; /* Arena pointer, for in-place read/write ops */
201 char *aend; /* First invalid address */
202};
203
204/*
205 * At store time:
d1be9408 206 * A hash table records the objects which have already been stored.
7a6a85bf
RG
207 * Those are referred to as SX_OBJECT in the file, and their "tag" (i.e.
208 * an arbitrary sequence number) is used to identify them.
209 *
210 * At retrieve time:
211 * An array table records the objects which have already been retrieved,
212 * as seen by the tag determind by counting the objects themselves. The
213 * reference to that retrieved object is kept in the table, and is returned
214 * when an SX_OBJECT is found bearing that same tag.
215 *
216 * The same processing is used to record "classname" for blessed objects:
217 * indexing by a hash at store time, and via an array at retrieve time.
218 */
219
220typedef unsigned long stag_t; /* Used by pre-0.6 binary format */
221
222/*
223 * The following "thread-safe" related defines were contributed by
224 * Murray Nesbitt <murray@activestate.com> and integrated by RAM, who
225 * only renamed things a little bit to ensure consistency with surrounding
226 * code. -- RAM, 14/09/1999
227 *
228 * The original patch suffered from the fact that the stcxt_t structure
229 * was global. Murray tried to minimize the impact on the code as much as
230 * possible.
231 *
232 * Starting with 0.7, Storable can be re-entrant, via the STORABLE_xxx hooks
233 * on objects. Therefore, the notion of context needs to be generalized,
234 * threading or not.
235 */
236
237#define MY_VERSION "Storable(" XS_VERSION ")"
238
530b72ba
NC
239
240/*
241 * Conditional UTF8 support.
242 *
243 */
244#ifdef SvUTF8_on
245#define STORE_UTF8STR(pv, len) STORE_PV_LEN(pv, len, SX_UTF8STR, SX_LUTF8STR)
246#define HAS_UTF8_SCALARS
247#ifdef HeKUTF8
248#define HAS_UTF8_HASHES
249#define HAS_UTF8_ALL
250#else
251/* 5.6 perl has utf8 scalars but not hashes */
252#endif
253#else
254#define SvUTF8(sv) 0
255#define STORE_UTF8STR(pv, len) CROAK(("panic: storing UTF8 in non-UTF8 perl"))
256#endif
257#ifndef HAS_UTF8_ALL
258#define UTF8_CROAK() CROAK(("Cannot retrieve UTF8 data in non-UTF8 perl"))
259#endif
260
261#ifdef HvPLACEHOLDERS
262#define HAS_RESTRICTED_HASHES
263#else
264#define HVhek_PLACEHOLD 0x200
265#define RESTRICTED_HASH_CROAK() CROAK(("Cannot retrieve restricted hash"))
266#endif
267
268#ifdef HvHASKFLAGS
269#define HAS_HASH_KEY_FLAGS
270#endif
271
dd19458b
JH
272/*
273 * Fields s_tainted and s_dirty are prefixed with s_ because Perl's include
274 * files remap tainted and dirty when threading is enabled. That's bad for
275 * perl to remap such common words. -- RAM, 29/09/00
276 */
277
7a6a85bf
RG
278typedef struct stcxt {
279 int entry; /* flags recursion */
280 int optype; /* type of traversal operation */
e993d95c
JH
281 HV *hseen; /* which objects have been seen, store time */
282 AV *hook_seen; /* which SVs were returned by STORABLE_freeze() */
283 AV *aseen; /* which objects have been seen, retrieve time */
284 HV *hclass; /* which classnames have been seen, store time */
285 AV *aclass; /* which classnames have been seen, retrieve time */
286 HV *hook; /* cache for hook methods per class name */
287 IV tagnum; /* incremented at store time for each seen object */
288 IV classnum; /* incremented at store time for each seen classname */
289 int netorder; /* true if network order used */
290 int s_tainted; /* true if input source is tainted, at retrieve time */
291 int forgive_me; /* whether to be forgiving... */
292 int canonical; /* whether to store hashes sorted by key */
530b72ba
NC
293#ifndef HAS_RESTRICTED_HASHES
294 int derestrict; /* whether to downgrade restrcted hashes */
295#endif
296#ifndef HAS_UTF8_ALL
297 int use_bytes; /* whether to bytes-ify utf8 */
298#endif
e8189732 299 int accept_future_minor; /* croak immediately on future minor versions? */
dd19458b 300 int s_dirty; /* context is dirty due to CROAK() -- can be cleaned */
e993d95c
JH
301 int membuf_ro; /* true means membuf is read-only and msaved is rw */
302 struct extendable keybuf; /* for hash key retrieval */
303 struct extendable membuf; /* for memory store/retrieve operations */
304 struct extendable msaved; /* where potentially valid mbuf is saved */
7a6a85bf
RG
305 PerlIO *fio; /* where I/O are performed, NULL for memory */
306 int ver_major; /* major of version for retrieved object */
307 int ver_minor; /* minor of version for retrieved object */
308 SV *(**retrieve_vtbl)(); /* retrieve dispatch table */
111e03c1
RG
309 SV *prev; /* contexts chained backwards in real recursion */
310 SV *my_sv; /* the blessed scalar who's SvPVX() I am */
7a6a85bf
RG
311} stcxt_t;
312
111e03c1
RG
313#define NEW_STORABLE_CXT_OBJ(cxt) \
314 STMT_START { \
315 SV *self = newSV(sizeof(stcxt_t) - 1); \
316 SV *my_sv = newRV_noinc(self); \
317 sv_bless(my_sv, gv_stashpv("Storable::Cxt", TRUE)); \
318 cxt = (stcxt_t *)SvPVX(self); \
319 Zero(cxt, 1, stcxt_t); \
320 cxt->my_sv = my_sv; \
321 } STMT_END
322
7a6a85bf
RG
323#if defined(MULTIPLICITY) || defined(PERL_OBJECT) || defined(PERL_CAPI)
324
e993d95c 325#if (PATCHLEVEL <= 4) && (SUBVERSION < 68)
7a6a85bf
RG
326#define dSTCXT_SV \
327 SV *perinterp_sv = perl_get_sv(MY_VERSION, FALSE)
328#else /* >= perl5.004_68 */
329#define dSTCXT_SV \
330 SV *perinterp_sv = *hv_fetch(PL_modglobal, \
331 MY_VERSION, sizeof(MY_VERSION)-1, TRUE)
332#endif /* < perl5.004_68 */
333
334#define dSTCXT_PTR(T,name) \
111e03c1 335 T name = ((perinterp_sv && SvIOK(perinterp_sv) && SvIVX(perinterp_sv) \
436c6dd3 336 ? (T)SvPVX(SvRV(INT2PTR(SV*,SvIVX(perinterp_sv)))) : (T) 0))
7a6a85bf
RG
337#define dSTCXT \
338 dSTCXT_SV; \
339 dSTCXT_PTR(stcxt_t *, cxt)
340
111e03c1
RG
341#define INIT_STCXT \
342 dSTCXT; \
343 NEW_STORABLE_CXT_OBJ(cxt); \
344 sv_setiv(perinterp_sv, PTR2IV(cxt->my_sv))
7a6a85bf 345
111e03c1
RG
346#define SET_STCXT(x) \
347 STMT_START { \
7a6a85bf 348 dSTCXT_SV; \
111e03c1
RG
349 sv_setiv(perinterp_sv, PTR2IV(x->my_sv)); \
350 } STMT_END
7a6a85bf
RG
351
352#else /* !MULTIPLICITY && !PERL_OBJECT && !PERL_CAPI */
353
85535365 354static stcxt_t *Context_ptr = NULL;
7a6a85bf 355#define dSTCXT stcxt_t *cxt = Context_ptr
85535365 356#define SET_STCXT(x) Context_ptr = x
111e03c1
RG
357#define INIT_STCXT \
358 dSTCXT; \
85535365
RG
359 NEW_STORABLE_CXT_OBJ(cxt); \
360 SET_STCXT(cxt)
111e03c1 361
7a6a85bf
RG
362
363#endif /* MULTIPLICITY || PERL_OBJECT || PERL_CAPI */
364
365/*
366 * KNOWN BUG:
367 * Croaking implies a memory leak, since we don't use setjmp/longjmp
368 * to catch the exit and free memory used during store or retrieve
369 * operations. This is not too difficult to fix, but I need to understand
370 * how Perl does it, and croaking is exceptional anyway, so I lack the
371 * motivation to do it.
372 *
373 * The current workaround is to mark the context as dirty when croaking,
374 * so that data structures can be freed whenever we renter Storable code
375 * (but only *then*: it's a workaround, not a fix).
376 *
377 * This is also imperfect, because we don't really know how far they trapped
378 * the croak(), and when we were recursing, we won't be able to clean anything
379 * but the topmost context stacked.
380 */
381
111e03c1 382#define CROAK(x) STMT_START { cxt->s_dirty = 1; croak x; } STMT_END
7a6a85bf
RG
383
384/*
385 * End of "thread-safe" related definitions.
386 */
387
388/*
9e21b3d0
JH
389 * LOW_32BITS
390 *
391 * Keep only the low 32 bits of a pointer (used for tags, which are not
392 * really pointers).
393 */
394
395#if PTRSIZE <= 4
396#define LOW_32BITS(x) ((I32) (x))
397#else
398#define LOW_32BITS(x) ((I32) ((unsigned long) (x) & 0xffffffffUL))
399#endif
400
401/*
402 * oI, oS, oC
403 *
404 * Hack for Crays, where sizeof(I32) == 8, and which are big-endians.
405 * Used in the WLEN and RLEN macros.
406 */
407
408#if INTSIZE > 4
409#define oI(x) ((I32 *) ((char *) (x) + 4))
410#define oS(x) ((x) - 4)
411#define oC(x) (x = 0)
412#define CRAY_HACK
413#else
414#define oI(x) (x)
415#define oS(x) (x)
416#define oC(x)
417#endif
418
419/*
7a6a85bf
RG
420 * key buffer handling
421 */
422#define kbuf (cxt->keybuf).arena
423#define ksiz (cxt->keybuf).asiz
111e03c1
RG
424#define KBUFINIT() \
425 STMT_START { \
7a6a85bf
RG
426 if (!kbuf) { \
427 TRACEME(("** allocating kbuf of 128 bytes")); \
428 New(10003, kbuf, 128, char); \
429 ksiz = 128; \
430 } \
111e03c1
RG
431 } STMT_END
432#define KBUFCHK(x) \
433 STMT_START { \
7a6a85bf 434 if (x >= ksiz) { \
e993d95c 435 TRACEME(("** extending kbuf to %d bytes (had %d)", x+1, ksiz)); \
7a6a85bf
RG
436 Renew(kbuf, x+1, char); \
437 ksiz = x+1; \
438 } \
111e03c1 439 } STMT_END
7a6a85bf
RG
440
441/*
442 * memory buffer handling
443 */
444#define mbase (cxt->membuf).arena
445#define msiz (cxt->membuf).asiz
446#define mptr (cxt->membuf).aptr
447#define mend (cxt->membuf).aend
448
449#define MGROW (1 << 13)
450#define MMASK (MGROW - 1)
451
452#define round_mgrow(x) \
453 ((unsigned long) (((unsigned long) (x) + MMASK) & ~MMASK))
454#define trunc_int(x) \
455 ((unsigned long) ((unsigned long) (x) & ~(sizeof(int)-1)))
456#define int_aligned(x) \
457 ((unsigned long) (x) == trunc_int(x))
458
111e03c1
RG
459#define MBUF_INIT(x) \
460 STMT_START { \
7a6a85bf
RG
461 if (!mbase) { \
462 TRACEME(("** allocating mbase of %d bytes", MGROW)); \
463 New(10003, mbase, MGROW, char); \
464 msiz = MGROW; \
465 } \
466 mptr = mbase; \
467 if (x) \
468 mend = mbase + x; \
469 else \
470 mend = mbase + msiz; \
111e03c1 471 } STMT_END
7a6a85bf
RG
472
473#define MBUF_TRUNC(x) mptr = mbase + x
474#define MBUF_SIZE() (mptr - mbase)
475
476/*
e993d95c
JH
477 * MBUF_SAVE_AND_LOAD
478 * MBUF_RESTORE
479 *
480 * Those macros are used in do_retrieve() to save the current memory
481 * buffer into cxt->msaved, before MBUF_LOAD() can be used to retrieve
482 * data from a string.
483 */
111e03c1
RG
484#define MBUF_SAVE_AND_LOAD(in) \
485 STMT_START { \
e993d95c
JH
486 ASSERT(!cxt->membuf_ro, ("mbase not already saved")); \
487 cxt->membuf_ro = 1; \
488 TRACEME(("saving mbuf")); \
489 StructCopy(&cxt->membuf, &cxt->msaved, struct extendable); \
490 MBUF_LOAD(in); \
111e03c1 491 } STMT_END
e993d95c 492
111e03c1
RG
493#define MBUF_RESTORE() \
494 STMT_START { \
e993d95c
JH
495 ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
496 cxt->membuf_ro = 0; \
497 TRACEME(("restoring mbuf")); \
498 StructCopy(&cxt->msaved, &cxt->membuf, struct extendable); \
111e03c1 499 } STMT_END
e993d95c
JH
500
501/*
7a6a85bf
RG
502 * Use SvPOKp(), because SvPOK() fails on tainted scalars.
503 * See store_scalar() for other usage of this workaround.
504 */
111e03c1
RG
505#define MBUF_LOAD(v) \
506 STMT_START { \
e993d95c 507 ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
7a6a85bf
RG
508 if (!SvPOKp(v)) \
509 CROAK(("Not a scalar string")); \
510 mptr = mbase = SvPV(v, msiz); \
511 mend = mbase + msiz; \
111e03c1 512 } STMT_END
7a6a85bf 513
111e03c1
RG
514#define MBUF_XTEND(x) \
515 STMT_START { \
7a6a85bf
RG
516 int nsz = (int) round_mgrow((x)+msiz); \
517 int offset = mptr - mbase; \
e993d95c
JH
518 ASSERT(!cxt->membuf_ro, ("mbase is not read-only")); \
519 TRACEME(("** extending mbase from %d to %d bytes (wants %d new)", \
520 msiz, nsz, (x))); \
7a6a85bf
RG
521 Renew(mbase, nsz, char); \
522 msiz = nsz; \
523 mptr = mbase + offset; \
524 mend = mbase + nsz; \
111e03c1 525 } STMT_END
7a6a85bf 526
111e03c1
RG
527#define MBUF_CHK(x) \
528 STMT_START { \
7a6a85bf
RG
529 if ((mptr + (x)) > mend) \
530 MBUF_XTEND(x); \
111e03c1 531 } STMT_END
7a6a85bf 532
111e03c1
RG
533#define MBUF_GETC(x) \
534 STMT_START { \
7a6a85bf
RG
535 if (mptr < mend) \
536 x = (int) (unsigned char) *mptr++; \
537 else \
538 return (SV *) 0; \
111e03c1 539 } STMT_END
7a6a85bf 540
9e21b3d0 541#ifdef CRAY_HACK
111e03c1
RG
542#define MBUF_GETINT(x) \
543 STMT_START { \
9e21b3d0
JH
544 oC(x); \
545 if ((mptr + 4) <= mend) { \
546 memcpy(oI(&x), mptr, 4); \
547 mptr += 4; \
548 } else \
549 return (SV *) 0; \
111e03c1 550 } STMT_END
9e21b3d0 551#else
111e03c1
RG
552#define MBUF_GETINT(x) \
553 STMT_START { \
7a6a85bf
RG
554 if ((mptr + sizeof(int)) <= mend) { \
555 if (int_aligned(mptr)) \
556 x = *(int *) mptr; \
557 else \
558 memcpy(&x, mptr, sizeof(int)); \
559 mptr += sizeof(int); \
560 } else \
561 return (SV *) 0; \
111e03c1 562 } STMT_END
9e21b3d0 563#endif
7a6a85bf 564
111e03c1
RG
565#define MBUF_READ(x,s) \
566 STMT_START { \
7a6a85bf
RG
567 if ((mptr + (s)) <= mend) { \
568 memcpy(x, mptr, s); \
569 mptr += s; \
570 } else \
571 return (SV *) 0; \
111e03c1 572 } STMT_END
7a6a85bf 573
111e03c1
RG
574#define MBUF_SAFEREAD(x,s,z) \
575 STMT_START { \
7a6a85bf
RG
576 if ((mptr + (s)) <= mend) { \
577 memcpy(x, mptr, s); \
578 mptr += s; \
579 } else { \
580 sv_free(z); \
581 return (SV *) 0; \
582 } \
111e03c1 583 } STMT_END
7a6a85bf 584
111e03c1
RG
585#define MBUF_PUTC(c) \
586 STMT_START { \
7a6a85bf
RG
587 if (mptr < mend) \
588 *mptr++ = (char) c; \
589 else { \
590 MBUF_XTEND(1); \
591 *mptr++ = (char) c; \
592 } \
111e03c1 593 } STMT_END
7a6a85bf 594
9e21b3d0 595#ifdef CRAY_HACK
111e03c1
RG
596#define MBUF_PUTINT(i) \
597 STMT_START { \
9e21b3d0
JH
598 MBUF_CHK(4); \
599 memcpy(mptr, oI(&i), 4); \
600 mptr += 4; \
111e03c1 601 } STMT_END
9e21b3d0 602#else
111e03c1
RG
603#define MBUF_PUTINT(i) \
604 STMT_START { \
7a6a85bf
RG
605 MBUF_CHK(sizeof(int)); \
606 if (int_aligned(mptr)) \
607 *(int *) mptr = i; \
608 else \
609 memcpy(mptr, &i, sizeof(int)); \
610 mptr += sizeof(int); \
111e03c1 611 } STMT_END
9e21b3d0 612#endif
7a6a85bf 613
111e03c1
RG
614#define MBUF_WRITE(x,s) \
615 STMT_START { \
7a6a85bf
RG
616 MBUF_CHK(s); \
617 memcpy(mptr, x, s); \
618 mptr += s; \
111e03c1 619 } STMT_END
7a6a85bf
RG
620
621/*
7a6a85bf
RG
622 * Possible return values for sv_type().
623 */
624
625#define svis_REF 0
626#define svis_SCALAR 1
627#define svis_ARRAY 2
628#define svis_HASH 3
629#define svis_TIED 4
630#define svis_TIED_ITEM 5
631#define svis_OTHER 6
632
633/*
634 * Flags for SX_HOOK.
635 */
636
637#define SHF_TYPE_MASK 0x03
638#define SHF_LARGE_CLASSLEN 0x04
639#define SHF_LARGE_STRLEN 0x08
640#define SHF_LARGE_LISTLEN 0x10
641#define SHF_IDX_CLASSNAME 0x20
642#define SHF_NEED_RECURSE 0x40
643#define SHF_HAS_LIST 0x80
644
645/*
b12202d0 646 * Types for SX_HOOK (last 2 bits in flags).
7a6a85bf
RG
647 */
648
649#define SHT_SCALAR 0
650#define SHT_ARRAY 1
651#define SHT_HASH 2
b12202d0
JH
652#define SHT_EXTRA 3 /* Read extra byte for type */
653
654/*
655 * The following are held in the "extra byte"...
656 */
657
658#define SHT_TSCALAR 4 /* 4 + 0 -- tied scalar */
659#define SHT_TARRAY 5 /* 4 + 1 -- tied array */
660#define SHT_THASH 6 /* 4 + 2 -- tied hash */
7a6a85bf
RG
661
662/*
e16e2ff8
NC
663 * per hash flags for flagged hashes
664 */
665
666#define SHV_RESTRICTED 0x01
667
668/*
669 * per key flags for flagged hashes
670 */
671
672#define SHV_K_UTF8 0x01
673#define SHV_K_WASUTF8 0x02
674#define SHV_K_LOCKED 0x04
675#define SHV_K_ISSV 0x08
676#define SHV_K_PLACEHOLDER 0x10
677
678/*
7a6a85bf
RG
679 * Before 0.6, the magic string was "perl-store" (binary version number 0).
680 *
681 * Since 0.6 introduced many binary incompatibilities, the magic string has
682 * been changed to "pst0" to allow an old image to be properly retrieved by
683 * a newer Storable, but ensure a newer image cannot be retrieved with an
684 * older version.
685 *
686 * At 0.7, objects are given the ability to serialize themselves, and the
687 * set of markers is extended, backward compatibility is not jeopardized,
688 * so the binary version number could have remained unchanged. To correctly
689 * spot errors if a file making use of 0.7-specific extensions is given to
690 * 0.6 for retrieval, the binary version was moved to "2". And I'm introducing
691 * a "minor" version, to better track this kind of evolution from now on.
692 *
693 */
2aeb6432
NC
694static const char old_magicstr[] = "perl-store"; /* Magic number before 0.6 */
695static const char magicstr[] = "pst0"; /* Used as a magic number */
7a6a85bf 696
2aeb6432
NC
697#define MAGICSTR_BYTES 'p','s','t','0'
698#define OLDMAGICSTR_BYTES 'p','e','r','l','-','s','t','o','r','e'
699
ee0f7aac
NC
700/* 5.6.x introduced the ability to have IVs as long long.
701 However, Configure still defined BYTEORDER based on the size of a long.
702 Storable uses the BYTEORDER value as part of the header, but doesn't
703 explicity store sizeof(IV) anywhere in the header. Hence on 5.6.x built
704 with IV as long long on a platform that uses Configure (ie most things
705 except VMS and Windows) headers are identical for the different IV sizes,
706 despite the files containing some fields based on sizeof(IV)
707 Erk. Broken-ness.
708 5.8 is consistent - the following redifinition kludge is only needed on
709 5.6.x, but the interwork is needed on 5.8 while data survives in files
710 with the 5.6 header.
711
712*/
713
714#if defined (IVSIZE) && (IVSIZE == 8) && (LONGSIZE == 4)
715#ifndef NO_56_INTERWORK_KLUDGE
716#define USE_56_INTERWORK_KLUDGE
717#endif
718#if BYTEORDER == 0x1234
719#undef BYTEORDER
720#define BYTEORDER 0x12345678
721#else
722#if BYTEORDER == 0x4321
723#undef BYTEORDER
724#define BYTEORDER 0x87654321
725#endif
726#endif
727#endif
728
2aeb6432
NC
729#if BYTEORDER == 0x1234
730#define BYTEORDER_BYTES '1','2','3','4'
731#else
732#if BYTEORDER == 0x12345678
733#define BYTEORDER_BYTES '1','2','3','4','5','6','7','8'
ee0f7aac
NC
734#ifdef USE_56_INTERWORK_KLUDGE
735#define BYTEORDER_BYTES_56 '1','2','3','4'
736#endif
2aeb6432
NC
737#else
738#if BYTEORDER == 0x87654321
739#define BYTEORDER_BYTES '8','7','6','5','4','3','2','1'
ee0f7aac
NC
740#ifdef USE_56_INTERWORK_KLUDGE
741#define BYTEORDER_BYTES_56 '4','3','2','1'
742#endif
2aeb6432
NC
743#else
744#if BYTEORDER == 0x4321
745#define BYTEORDER_BYTES '4','3','2','1'
746#else
747#error Unknown byteoder. Please append your byteorder to Storable.xs
748#endif
749#endif
750#endif
751#endif
752
753static const char byteorderstr[] = {BYTEORDER_BYTES, 0};
ee0f7aac
NC
754#ifdef USE_56_INTERWORK_KLUDGE
755static const char byteorderstr_56[] = {BYTEORDER_BYTES_56, 0};
756#endif
530b72ba 757
e16e2ff8 758#define STORABLE_BIN_MAJOR 2 /* Binary major "version" */
530b72ba
NC
759#define STORABLE_BIN_MINOR 5 /* Binary minor "version" */
760
761/* If we aren't 5.7.3 or later, we won't be writing out files that use the
762 * new flagged hash introdued in 2.5, so put 2.4 in the binary header to
763 * maximise ease of interoperation with older Storables.
764 * Could we write 2.3s if we're on 5.005_03? NWC
765 */
e16e2ff8 766#if (PATCHLEVEL <= 6)
530b72ba 767#define STORABLE_BIN_WRITE_MINOR 4
e16e2ff8
NC
768#else
769/*
770 * As of perl 5.7.3, utf8 hash key is introduced.
771 * So this must change -- dankogai
772*/
530b72ba 773#define STORABLE_BIN_WRITE_MINOR 5
e16e2ff8 774#endif /* (PATCHLEVEL <= 6) */
7a6a85bf
RG
775
776/*
777 * Useful store shortcuts...
778 */
779
111e03c1
RG
780#define PUTMARK(x) \
781 STMT_START { \
7a6a85bf
RG
782 if (!cxt->fio) \
783 MBUF_PUTC(x); \
784 else if (PerlIO_putc(cxt->fio, x) == EOF) \
785 return -1; \
111e03c1 786 } STMT_END
7a6a85bf 787
111e03c1
RG
788#define WRITE_I32(x) \
789 STMT_START { \
9e21b3d0
JH
790 ASSERT(sizeof(x) == sizeof(I32), ("writing an I32")); \
791 if (!cxt->fio) \
792 MBUF_PUTINT(x); \
793 else if (PerlIO_write(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
794 return -1; \
111e03c1 795 } STMT_END
9e21b3d0 796
7a6a85bf 797#ifdef HAS_HTONL
111e03c1
RG
798#define WLEN(x) \
799 STMT_START { \
7a6a85bf
RG
800 if (cxt->netorder) { \
801 int y = (int) htonl(x); \
802 if (!cxt->fio) \
803 MBUF_PUTINT(y); \
9e21b3d0 804 else if (PerlIO_write(cxt->fio,oI(&y),oS(sizeof(y))) != oS(sizeof(y))) \
7a6a85bf
RG
805 return -1; \
806 } else { \
807 if (!cxt->fio) \
808 MBUF_PUTINT(x); \
9e21b3d0 809 else if (PerlIO_write(cxt->fio,oI(&x),oS(sizeof(x))) != oS(sizeof(x))) \
7a6a85bf
RG
810 return -1; \
811 } \
111e03c1 812 } STMT_END
7a6a85bf 813#else
9e21b3d0 814#define WLEN(x) WRITE_I32(x)
7a6a85bf
RG
815#endif
816
111e03c1
RG
817#define WRITE(x,y) \
818 STMT_START { \
7a6a85bf
RG
819 if (!cxt->fio) \
820 MBUF_WRITE(x,y); \
821 else if (PerlIO_write(cxt->fio, x, y) != y) \
822 return -1; \
111e03c1 823 } STMT_END
7a6a85bf 824
111e03c1
RG
825#define STORE_PV_LEN(pv, len, small, large) \
826 STMT_START { \
7a6a85bf
RG
827 if (len <= LG_SCALAR) { \
828 unsigned char clen = (unsigned char) len; \
dd19458b 829 PUTMARK(small); \
7a6a85bf
RG
830 PUTMARK(clen); \
831 if (len) \
832 WRITE(pv, len); \
833 } else { \
dd19458b 834 PUTMARK(large); \
7a6a85bf
RG
835 WLEN(len); \
836 WRITE(pv, len); \
837 } \
111e03c1 838 } STMT_END
7a6a85bf 839
dd19458b
JH
840#define STORE_SCALAR(pv, len) STORE_PV_LEN(pv, len, SX_SCALAR, SX_LSCALAR)
841
842/*
7a6a85bf
RG
843 * Store undef in arrays and hashes without recursing through store().
844 */
111e03c1
RG
845#define STORE_UNDEF() \
846 STMT_START { \
7a6a85bf
RG
847 cxt->tagnum++; \
848 PUTMARK(SX_UNDEF); \
111e03c1 849 } STMT_END
7a6a85bf
RG
850
851/*
852 * Useful retrieve shortcuts...
853 */
854
855#define GETCHAR() \
856 (cxt->fio ? PerlIO_getc(cxt->fio) : (mptr >= mend ? EOF : (int) *mptr++))
857
111e03c1
RG
858#define GETMARK(x) \
859 STMT_START { \
7a6a85bf
RG
860 if (!cxt->fio) \
861 MBUF_GETC(x); \
76df4757 862 else if ((int) (x = PerlIO_getc(cxt->fio)) == EOF) \
7a6a85bf 863 return (SV *) 0; \
111e03c1 864 } STMT_END
7a6a85bf 865
111e03c1
RG
866#define READ_I32(x) \
867 STMT_START { \
9e21b3d0
JH
868 ASSERT(sizeof(x) == sizeof(I32), ("reading an I32")); \
869 oC(x); \
7a6a85bf
RG
870 if (!cxt->fio) \
871 MBUF_GETINT(x); \
9e21b3d0 872 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
7a6a85bf 873 return (SV *) 0; \
111e03c1 874 } STMT_END
9e21b3d0
JH
875
876#ifdef HAS_NTOHL
111e03c1
RG
877#define RLEN(x) \
878 STMT_START { \
9e21b3d0 879 oC(x); \
7a6a85bf
RG
880 if (!cxt->fio) \
881 MBUF_GETINT(x); \
9e21b3d0 882 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
7a6a85bf 883 return (SV *) 0; \
9e21b3d0
JH
884 if (cxt->netorder) \
885 x = (int) ntohl(x); \
111e03c1 886 } STMT_END
9e21b3d0
JH
887#else
888#define RLEN(x) READ_I32(x)
7a6a85bf
RG
889#endif
890
111e03c1
RG
891#define READ(x,y) \
892 STMT_START { \
7a6a85bf
RG
893 if (!cxt->fio) \
894 MBUF_READ(x, y); \
895 else if (PerlIO_read(cxt->fio, x, y) != y) \
896 return (SV *) 0; \
111e03c1 897 } STMT_END
7a6a85bf 898
111e03c1
RG
899#define SAFEREAD(x,y,z) \
900 STMT_START { \
7a6a85bf
RG
901 if (!cxt->fio) \
902 MBUF_SAFEREAD(x,y,z); \
903 else if (PerlIO_read(cxt->fio, x, y) != y) { \
904 sv_free(z); \
905 return (SV *) 0; \
906 } \
111e03c1 907 } STMT_END
7a6a85bf
RG
908
909/*
910 * This macro is used at retrieve time, to remember where object 'y', bearing a
911 * given tag 'tagnum', has been retrieved. Next time we see an SX_OBJECT marker,
912 * we'll therefore know where it has been retrieved and will be able to
913 * share the same reference, as in the original stored memory image.
b12202d0
JH
914 *
915 * We also need to bless objects ASAP for hooks (which may compute "ref $x"
916 * on the objects given to STORABLE_thaw and expect that to be defined), and
917 * also for overloaded objects (for which we might not find the stash if the
918 * object is not blessed yet--this might occur for overloaded objects that
919 * refer to themselves indirectly: if we blessed upon return from a sub
920 * retrieve(), the SX_OBJECT marker we'd found could not have overloading
921 * restored on it because the underlying object would not be blessed yet!).
922 *
923 * To achieve that, the class name of the last retrieved object is passed down
924 * recursively, and the first SEEN() call for which the class name is not NULL
925 * will bless the object.
7a6a85bf 926 */
111e03c1
RG
927#define SEEN(y,c) \
928 STMT_START { \
7a6a85bf
RG
929 if (!y) \
930 return (SV *) 0; \
931 if (av_store(cxt->aseen, cxt->tagnum++, SvREFCNT_inc(y)) == 0) \
932 return (SV *) 0; \
43d061fe 933 TRACEME(("aseen(#%d) = 0x%"UVxf" (refcnt=%d)", cxt->tagnum-1, \
b12202d0
JH
934 PTR2UV(y), SvREFCNT(y)-1)); \
935 if (c) \
936 BLESS((SV *) (y), c); \
111e03c1 937 } STMT_END
7a6a85bf
RG
938
939/*
940 * Bless `s' in `p', via a temporary reference, required by sv_bless().
941 */
111e03c1
RG
942#define BLESS(s,p) \
943 STMT_START { \
7a6a85bf
RG
944 SV *ref; \
945 HV *stash; \
43d061fe 946 TRACEME(("blessing 0x%"UVxf" in %s", PTR2UV(s), (p))); \
7a6a85bf
RG
947 stash = gv_stashpv((p), TRUE); \
948 ref = newRV_noinc(s); \
949 (void) sv_bless(ref, stash); \
950 SvRV(ref) = 0; \
951 SvREFCNT_dec(ref); \
111e03c1 952 } STMT_END
7a6a85bf
RG
953
954static int store();
b12202d0 955static SV *retrieve(stcxt_t *cxt, char *cname);
7a6a85bf
RG
956
957/*
958 * Dynamic dispatching table for SV store.
959 */
960
961static int store_ref(stcxt_t *cxt, SV *sv);
962static int store_scalar(stcxt_t *cxt, SV *sv);
963static int store_array(stcxt_t *cxt, AV *av);
964static int store_hash(stcxt_t *cxt, HV *hv);
965static int store_tied(stcxt_t *cxt, SV *sv);
966static int store_tied_item(stcxt_t *cxt, SV *sv);
967static int store_other(stcxt_t *cxt, SV *sv);
f0ffaed8 968static int store_blessed(stcxt_t *cxt, SV *sv, int type, HV *pkg);
7a6a85bf 969
862382c7
JH
970static int (*sv_store[])(stcxt_t *cxt, SV *sv) = {
971 store_ref, /* svis_REF */
972 store_scalar, /* svis_SCALAR */
973 (int (*)(stcxt_t *cxt, SV *sv)) store_array, /* svis_ARRAY */
974 (int (*)(stcxt_t *cxt, SV *sv)) store_hash, /* svis_HASH */
975 store_tied, /* svis_TIED */
976 store_tied_item, /* svis_TIED_ITEM */
977 store_other, /* svis_OTHER */
7a6a85bf
RG
978};
979
980#define SV_STORE(x) (*sv_store[x])
981
982/*
983 * Dynamic dispatching tables for SV retrieval.
984 */
985
b12202d0
JH
986static SV *retrieve_lscalar(stcxt_t *cxt, char *cname);
987static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname);
988static SV *old_retrieve_array(stcxt_t *cxt, char *cname);
989static SV *old_retrieve_hash(stcxt_t *cxt, char *cname);
990static SV *retrieve_ref(stcxt_t *cxt, char *cname);
991static SV *retrieve_undef(stcxt_t *cxt, char *cname);
992static SV *retrieve_integer(stcxt_t *cxt, char *cname);
993static SV *retrieve_double(stcxt_t *cxt, char *cname);
994static SV *retrieve_byte(stcxt_t *cxt, char *cname);
995static SV *retrieve_netint(stcxt_t *cxt, char *cname);
996static SV *retrieve_scalar(stcxt_t *cxt, char *cname);
997static SV *retrieve_utf8str(stcxt_t *cxt, char *cname);
998static SV *retrieve_tied_array(stcxt_t *cxt, char *cname);
999static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname);
1000static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname);
1001static SV *retrieve_other(stcxt_t *cxt, char *cname);
1002
1003static SV *(*sv_old_retrieve[])(stcxt_t *cxt, char *cname) = {
7a6a85bf
RG
1004 0, /* SX_OBJECT -- entry unused dynamically */
1005 retrieve_lscalar, /* SX_LSCALAR */
1006 old_retrieve_array, /* SX_ARRAY -- for pre-0.6 binaries */
1007 old_retrieve_hash, /* SX_HASH -- for pre-0.6 binaries */
1008 retrieve_ref, /* SX_REF */
1009 retrieve_undef, /* SX_UNDEF */
1010 retrieve_integer, /* SX_INTEGER */
1011 retrieve_double, /* SX_DOUBLE */
1012 retrieve_byte, /* SX_BYTE */
1013 retrieve_netint, /* SX_NETINT */
1014 retrieve_scalar, /* SX_SCALAR */
1015 retrieve_tied_array, /* SX_ARRAY */
1016 retrieve_tied_hash, /* SX_HASH */
1017 retrieve_tied_scalar, /* SX_SCALAR */
1018 retrieve_other, /* SX_SV_UNDEF not supported */
1019 retrieve_other, /* SX_SV_YES not supported */
1020 retrieve_other, /* SX_SV_NO not supported */
1021 retrieve_other, /* SX_BLESS not supported */
1022 retrieve_other, /* SX_IX_BLESS not supported */
1023 retrieve_other, /* SX_HOOK not supported */
1024 retrieve_other, /* SX_OVERLOADED not supported */
1025 retrieve_other, /* SX_TIED_KEY not supported */
1026 retrieve_other, /* SX_TIED_IDX not supported */
dd19458b
JH
1027 retrieve_other, /* SX_UTF8STR not supported */
1028 retrieve_other, /* SX_LUTF8STR not supported */
e16e2ff8 1029 retrieve_other, /* SX_FLAG_HASH not supported */
7a6a85bf
RG
1030 retrieve_other, /* SX_ERROR */
1031};
1032
b12202d0
JH
1033static SV *retrieve_array(stcxt_t *cxt, char *cname);
1034static SV *retrieve_hash(stcxt_t *cxt, char *cname);
1035static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname);
1036static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname);
1037static SV *retrieve_sv_no(stcxt_t *cxt, char *cname);
1038static SV *retrieve_blessed(stcxt_t *cxt, char *cname);
1039static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname);
1040static SV *retrieve_hook(stcxt_t *cxt, char *cname);
1041static SV *retrieve_overloaded(stcxt_t *cxt, char *cname);
1042static SV *retrieve_tied_key(stcxt_t *cxt, char *cname);
1043static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname);
e16e2ff8 1044static SV *retrieve_flag_hash(stcxt_t *cxt, char *cname);
b12202d0
JH
1045
1046static SV *(*sv_retrieve[])(stcxt_t *cxt, char *cname) = {
7a6a85bf
RG
1047 0, /* SX_OBJECT -- entry unused dynamically */
1048 retrieve_lscalar, /* SX_LSCALAR */
1049 retrieve_array, /* SX_ARRAY */
1050 retrieve_hash, /* SX_HASH */
1051 retrieve_ref, /* SX_REF */
1052 retrieve_undef, /* SX_UNDEF */
1053 retrieve_integer, /* SX_INTEGER */
1054 retrieve_double, /* SX_DOUBLE */
1055 retrieve_byte, /* SX_BYTE */
1056 retrieve_netint, /* SX_NETINT */
1057 retrieve_scalar, /* SX_SCALAR */
1058 retrieve_tied_array, /* SX_ARRAY */
1059 retrieve_tied_hash, /* SX_HASH */
1060 retrieve_tied_scalar, /* SX_SCALAR */
1061 retrieve_sv_undef, /* SX_SV_UNDEF */
1062 retrieve_sv_yes, /* SX_SV_YES */
1063 retrieve_sv_no, /* SX_SV_NO */
1064 retrieve_blessed, /* SX_BLESS */
1065 retrieve_idx_blessed, /* SX_IX_BLESS */
1066 retrieve_hook, /* SX_HOOK */
1067 retrieve_overloaded, /* SX_OVERLOAD */
1068 retrieve_tied_key, /* SX_TIED_KEY */
1069 retrieve_tied_idx, /* SX_TIED_IDX */
dd19458b
JH
1070 retrieve_utf8str, /* SX_UTF8STR */
1071 retrieve_lutf8str, /* SX_LUTF8STR */
e16e2ff8 1072 retrieve_flag_hash, /* SX_HASH */
7a6a85bf
RG
1073 retrieve_other, /* SX_ERROR */
1074};
1075
1076#define RETRIEVE(c,x) (*(c)->retrieve_vtbl[(x) >= SX_ERROR ? SX_ERROR : (x)])
1077
f0ffaed8 1078static SV *mbuf2sv(void);
7a6a85bf
RG
1079
1080/***
1081 *** Context management.
1082 ***/
1083
1084/*
1085 * init_perinterp
1086 *
1087 * Called once per "thread" (interpreter) to initialize some global context.
1088 */
f0ffaed8
JH
1089static void init_perinterp(void)
1090{
7a6a85bf
RG
1091 INIT_STCXT;
1092
1093 cxt->netorder = 0; /* true if network order used */
1094 cxt->forgive_me = -1; /* whether to be forgiving... */
1095}
1096
1097/*
e993d95c
JH
1098 * reset_context
1099 *
1100 * Called at the end of every context cleaning, to perform common reset
1101 * operations.
1102 */
1103static void reset_context(stcxt_t *cxt)
1104{
1105 cxt->entry = 0;
1106 cxt->s_dirty = 0;
1107 cxt->optype &= ~(ST_STORE|ST_RETRIEVE); /* Leave ST_CLONE alone */
1108}
1109
1110/*
7a6a85bf
RG
1111 * init_store_context
1112 *
1113 * Initialize a new store context for real recursion.
1114 */
f0ffaed8
JH
1115static void init_store_context(
1116 stcxt_t *cxt,
1117 PerlIO *f,
1118 int optype,
1119 int network_order)
7a6a85bf
RG
1120{
1121 TRACEME(("init_store_context"));
1122
1123 cxt->netorder = network_order;
1124 cxt->forgive_me = -1; /* Fetched from perl if needed */
1125 cxt->canonical = -1; /* Idem */
1126 cxt->tagnum = -1; /* Reset tag numbers */
1127 cxt->classnum = -1; /* Reset class numbers */
1128 cxt->fio = f; /* Where I/O are performed */
1129 cxt->optype = optype; /* A store, or a deep clone */
1130 cxt->entry = 1; /* No recursion yet */
1131
1132 /*
1133 * The `hseen' table is used to keep track of each SV stored and their
1134 * associated tag numbers is special. It is "abused" because the
1135 * values stored are not real SV, just integers cast to (SV *),
1136 * which explains the freeing below.
1137 *
1138 * It is also one possible bottlneck to achieve good storing speed,
1139 * so the "shared keys" optimization is turned off (unlikely to be
1140 * of any use here), and the hash table is "pre-extended". Together,
1141 * those optimizations increase the throughput by 12%.
1142 */
1143
1144 cxt->hseen = newHV(); /* Table where seen objects are stored */
1145 HvSHAREKEYS_off(cxt->hseen);
1146
1147 /*
1148 * The following does not work well with perl5.004_04, and causes
1149 * a core dump later on, in a completely unrelated spot, which
1150 * makes me think there is a memory corruption going on.
1151 *
1152 * Calling hv_ksplit(hseen, HBUCKETS) instead of manually hacking
1153 * it below does not make any difference. It seems to work fine
1154 * with perl5.004_68 but given the probable nature of the bug,
1155 * that does not prove anything.
1156 *
1157 * It's a shame because increasing the amount of buckets raises
1158 * store() throughput by 5%, but until I figure this out, I can't
1159 * allow for this to go into production.
1160 *
1161 * It is reported fixed in 5.005, hence the #if.
1162 */
f0ffaed8 1163#if PERL_VERSION >= 5
7a6a85bf
RG
1164#define HBUCKETS 4096 /* Buckets for %hseen */
1165 HvMAX(cxt->hseen) = HBUCKETS - 1; /* keys %hseen = $HBUCKETS; */
1166#endif
1167
1168 /*
1169 * The `hclass' hash uses the same settings as `hseen' above, but it is
1170 * used to assign sequential tags (numbers) to class names for blessed
1171 * objects.
1172 *
1173 * We turn the shared key optimization on.
1174 */
1175
1176 cxt->hclass = newHV(); /* Where seen classnames are stored */
1177
f0ffaed8 1178#if PERL_VERSION >= 5
7a6a85bf
RG
1179 HvMAX(cxt->hclass) = HBUCKETS - 1; /* keys %hclass = $HBUCKETS; */
1180#endif
1181
1182 /*
1183 * The `hook' hash table is used to keep track of the references on
1184 * the STORABLE_freeze hook routines, when found in some class name.
1185 *
1186 * It is assumed that the inheritance tree will not be changed during
1187 * storing, and that no new method will be dynamically created by the
1188 * hooks.
1189 */
1190
1191 cxt->hook = newHV(); /* Table where hooks are cached */
90826881
JH
1192
1193 /*
1194 * The `hook_seen' array keeps track of all the SVs returned by
1195 * STORABLE_freeze hooks for us to serialize, so that they are not
1196 * reclaimed until the end of the serialization process. Each SV is
1197 * only stored once, the first time it is seen.
1198 */
1199
1200 cxt->hook_seen = newAV(); /* Lists SVs returned by STORABLE_freeze */
7a6a85bf
RG
1201}
1202
1203/*
1204 * clean_store_context
1205 *
1206 * Clean store context by
1207 */
f0ffaed8 1208static void clean_store_context(stcxt_t *cxt)
7a6a85bf
RG
1209{
1210 HE *he;
1211
1212 TRACEME(("clean_store_context"));
1213
1214 ASSERT(cxt->optype & ST_STORE, ("was performing a store()"));
1215
1216 /*
1217 * Insert real values into hashes where we stored faked pointers.
1218 */
1219
e993d95c
JH
1220 if (cxt->hseen) {
1221 hv_iterinit(cxt->hseen);
1222 while ((he = hv_iternext(cxt->hseen))) /* Extra () for -Wall, grr.. */
1223 HeVAL(he) = &PL_sv_undef;
1224 }
7a6a85bf 1225
e993d95c
JH
1226 if (cxt->hclass) {
1227 hv_iterinit(cxt->hclass);
1228 while ((he = hv_iternext(cxt->hclass))) /* Extra () for -Wall, grr.. */
1229 HeVAL(he) = &PL_sv_undef;
1230 }
7a6a85bf
RG
1231
1232 /*
1233 * And now dispose of them...
862382c7
JH
1234 *
1235 * The surrounding if() protection has been added because there might be
1236 * some cases where this routine is called more than once, during
1237 * exceptionnal events. This was reported by Marc Lehmann when Storable
1238 * is executed from mod_perl, and the fix was suggested by him.
1239 * -- RAM, 20/12/2000
1240 */
1241
1242 if (cxt->hseen) {
1243 HV *hseen = cxt->hseen;
1244 cxt->hseen = 0;
1245 hv_undef(hseen);
1246 sv_free((SV *) hseen);
1247 }
7a6a85bf 1248
862382c7
JH
1249 if (cxt->hclass) {
1250 HV *hclass = cxt->hclass;
1251 cxt->hclass = 0;
1252 hv_undef(hclass);
1253 sv_free((SV *) hclass);
1254 }
7a6a85bf 1255
862382c7
JH
1256 if (cxt->hook) {
1257 HV *hook = cxt->hook;
1258 cxt->hook = 0;
1259 hv_undef(hook);
1260 sv_free((SV *) hook);
1261 }
7a6a85bf 1262
862382c7
JH
1263 if (cxt->hook_seen) {
1264 AV *hook_seen = cxt->hook_seen;
1265 cxt->hook_seen = 0;
1266 av_undef(hook_seen);
1267 sv_free((SV *) hook_seen);
1268 }
90826881 1269
e8189732
NC
1270 cxt->forgive_me = -1; /* Fetched from perl if needed */
1271 cxt->canonical = -1; /* Idem */
1272
e993d95c 1273 reset_context(cxt);
7a6a85bf
RG
1274}
1275
1276/*
1277 * init_retrieve_context
1278 *
1279 * Initialize a new retrieve context for real recursion.
1280 */
dd19458b 1281static void init_retrieve_context(stcxt_t *cxt, int optype, int is_tainted)
7a6a85bf
RG
1282{
1283 TRACEME(("init_retrieve_context"));
1284
1285 /*
1286 * The hook hash table is used to keep track of the references on
1287 * the STORABLE_thaw hook routines, when found in some class name.
1288 *
1289 * It is assumed that the inheritance tree will not be changed during
1290 * storing, and that no new method will be dynamically created by the
1291 * hooks.
1292 */
1293
1294 cxt->hook = newHV(); /* Caches STORABLE_thaw */
1295
1296 /*
1297 * If retrieving an old binary version, the cxt->retrieve_vtbl variable
1298 * was set to sv_old_retrieve. We'll need a hash table to keep track of
1299 * the correspondance between the tags and the tag number used by the
1300 * new retrieve routines.
1301 */
1302
e16e2ff8 1303 cxt->hseen = ((cxt->retrieve_vtbl == sv_old_retrieve) ? newHV() : 0);
7a6a85bf
RG
1304
1305 cxt->aseen = newAV(); /* Where retrieved objects are kept */
1306 cxt->aclass = newAV(); /* Where seen classnames are kept */
1307 cxt->tagnum = 0; /* Have to count objects... */
1308 cxt->classnum = 0; /* ...and class names as well */
1309 cxt->optype = optype;
dd19458b 1310 cxt->s_tainted = is_tainted;
7a6a85bf 1311 cxt->entry = 1; /* No recursion yet */
530b72ba
NC
1312#ifndef HAS_RESTRICTED_HASHES
1313 cxt->derestrict = -1; /* Fetched from perl if needed */
1314#endif
1315#ifndef HAS_UTF8_ALL
1316 cxt->use_bytes = -1; /* Fetched from perl if needed */
1317#endif
e8189732 1318 cxt->accept_future_minor = -1; /* Fetched from perl if needed */
7a6a85bf
RG
1319}
1320
1321/*
1322 * clean_retrieve_context
1323 *
1324 * Clean retrieve context by
1325 */
dd19458b 1326static void clean_retrieve_context(stcxt_t *cxt)
7a6a85bf
RG
1327{
1328 TRACEME(("clean_retrieve_context"));
1329
1330 ASSERT(cxt->optype & ST_RETRIEVE, ("was performing a retrieve()"));
1331
862382c7
JH
1332 if (cxt->aseen) {
1333 AV *aseen = cxt->aseen;
1334 cxt->aseen = 0;
1335 av_undef(aseen);
1336 sv_free((SV *) aseen);
1337 }
7a6a85bf 1338
862382c7
JH
1339 if (cxt->aclass) {
1340 AV *aclass = cxt->aclass;
1341 cxt->aclass = 0;
1342 av_undef(aclass);
1343 sv_free((SV *) aclass);
1344 }
7a6a85bf 1345
862382c7
JH
1346 if (cxt->hook) {
1347 HV *hook = cxt->hook;
1348 cxt->hook = 0;
1349 hv_undef(hook);
1350 sv_free((SV *) hook);
1351 }
7a6a85bf 1352
862382c7
JH
1353 if (cxt->hseen) {
1354 HV *hseen = cxt->hseen;
1355 cxt->hseen = 0;
1356 hv_undef(hseen);
1357 sv_free((SV *) hseen); /* optional HV, for backward compat. */
1358 }
7a6a85bf 1359
e8189732
NC
1360#ifndef HAS_RESTRICTED_HASHES
1361 cxt->derestrict = -1; /* Fetched from perl if needed */
1362#endif
1363#ifndef HAS_UTF8_ALL
1364 cxt->use_bytes = -1; /* Fetched from perl if needed */
1365#endif
1366 cxt->accept_future_minor = -1; /* Fetched from perl if needed */
1367
e993d95c 1368 reset_context(cxt);
7a6a85bf
RG
1369}
1370
1371/*
1372 * clean_context
1373 *
1374 * A workaround for the CROAK bug: cleanup the last context.
1375 */
e993d95c 1376static void clean_context(stcxt_t *cxt)
7a6a85bf
RG
1377{
1378 TRACEME(("clean_context"));
1379
dd19458b 1380 ASSERT(cxt->s_dirty, ("dirty context"));
7a6a85bf 1381
e993d95c
JH
1382 if (cxt->membuf_ro)
1383 MBUF_RESTORE();
1384
1385 ASSERT(!cxt->membuf_ro, ("mbase is not read-only"));
1386
7a6a85bf
RG
1387 if (cxt->optype & ST_RETRIEVE)
1388 clean_retrieve_context(cxt);
e993d95c 1389 else if (cxt->optype & ST_STORE)
7a6a85bf 1390 clean_store_context(cxt);
e993d95c
JH
1391 else
1392 reset_context(cxt);
862382c7
JH
1393
1394 ASSERT(!cxt->s_dirty, ("context is clean"));
e993d95c 1395 ASSERT(cxt->entry == 0, ("context is reset"));
7a6a85bf
RG
1396}
1397
1398/*
1399 * allocate_context
1400 *
1401 * Allocate a new context and push it on top of the parent one.
1402 * This new context is made globally visible via SET_STCXT().
1403 */
1404static stcxt_t *allocate_context(parent_cxt)
1405stcxt_t *parent_cxt;
1406{
1407 stcxt_t *cxt;
1408
1409 TRACEME(("allocate_context"));
1410
dd19458b 1411 ASSERT(!parent_cxt->s_dirty, ("parent context clean"));
7a6a85bf 1412
111e03c1
RG
1413 NEW_STORABLE_CXT_OBJ(cxt);
1414 cxt->prev = parent_cxt->my_sv;
7a6a85bf
RG
1415 SET_STCXT(cxt);
1416
e993d95c
JH
1417 ASSERT(!cxt->s_dirty, ("clean context"));
1418
7a6a85bf
RG
1419 return cxt;
1420}
1421
1422/*
1423 * free_context
1424 *
1425 * Free current context, which cannot be the "root" one.
1426 * Make the context underneath globally visible via SET_STCXT().
1427 */
1428static void free_context(cxt)
1429stcxt_t *cxt;
1430{
111e03c1 1431 stcxt_t *prev = (stcxt_t *)(cxt->prev ? SvPVX(SvRV(cxt->prev)) : 0);
7a6a85bf
RG
1432
1433 TRACEME(("free_context"));
1434
dd19458b 1435 ASSERT(!cxt->s_dirty, ("clean context"));
7a6a85bf
RG
1436 ASSERT(prev, ("not freeing root context"));
1437
111e03c1 1438 SvREFCNT_dec(cxt->my_sv);
7a6a85bf 1439 SET_STCXT(prev);
e993d95c
JH
1440
1441 ASSERT(cxt, ("context not void"));
7a6a85bf
RG
1442}
1443
1444/***
1445 *** Predicates.
1446 ***/
1447
1448/*
1449 * is_storing
1450 *
1451 * Tells whether we're in the middle of a store operation.
1452 */
f0ffaed8 1453int is_storing(void)
7a6a85bf
RG
1454{
1455 dSTCXT;
1456
1457 return cxt->entry && (cxt->optype & ST_STORE);
1458}
1459
1460/*
1461 * is_retrieving
1462 *
1463 * Tells whether we're in the middle of a retrieve operation.
1464 */
f0ffaed8 1465int is_retrieving(void)
7a6a85bf
RG
1466{
1467 dSTCXT;
1468
1469 return cxt->entry && (cxt->optype & ST_RETRIEVE);
1470}
1471
1472/*
1473 * last_op_in_netorder
1474 *
1475 * Returns whether last operation was made using network order.
1476 *
1477 * This is typically out-of-band information that might prove useful
1478 * to people wishing to convert native to network order data when used.
1479 */
f0ffaed8 1480int last_op_in_netorder(void)
7a6a85bf
RG
1481{
1482 dSTCXT;
1483
1484 return cxt->netorder;
1485}
1486
1487/***
1488 *** Hook lookup and calling routines.
1489 ***/
1490
1491/*
1492 * pkg_fetchmeth
1493 *
1494 * A wrapper on gv_fetchmethod_autoload() which caches results.
1495 *
1496 * Returns the routine reference as an SV*, or null if neither the package
1497 * nor its ancestors know about the method.
1498 */
f0ffaed8
JH
1499static SV *pkg_fetchmeth(
1500 HV *cache,
1501 HV *pkg,
1502 char *method)
7a6a85bf
RG
1503{
1504 GV *gv;
1505 SV *sv;
7a6a85bf
RG
1506
1507 /*
1508 * The following code is the same as the one performed by UNIVERSAL::can
1509 * in the Perl core.
1510 */
1511
1512 gv = gv_fetchmethod_autoload(pkg, method, FALSE);
1513 if (gv && isGV(gv)) {
1514 sv = newRV((SV*) GvCV(gv));
9e21b3d0 1515 TRACEME(("%s->%s: 0x%"UVxf, HvNAME(pkg), method, PTR2UV(sv)));
7a6a85bf
RG
1516 } else {
1517 sv = newSVsv(&PL_sv_undef);
1518 TRACEME(("%s->%s: not found", HvNAME(pkg), method));
1519 }
1520
1521 /*
1522 * Cache the result, ignoring failure: if we can't store the value,
1523 * it just won't be cached.
1524 */
1525
1526 (void) hv_store(cache, HvNAME(pkg), strlen(HvNAME(pkg)), sv, 0);
1527
1528 return SvOK(sv) ? sv : (SV *) 0;
1529}
1530
1531/*
1532 * pkg_hide
1533 *
1534 * Force cached value to be undef: hook ignored even if present.
1535 */
f0ffaed8
JH
1536static void pkg_hide(
1537 HV *cache,
1538 HV *pkg,
1539 char *method)
7a6a85bf
RG
1540{
1541 (void) hv_store(cache,
1542 HvNAME(pkg), strlen(HvNAME(pkg)), newSVsv(&PL_sv_undef), 0);
1543}
1544
1545/*
212e9bde
JH
1546 * pkg_uncache
1547 *
1548 * Discard cached value: a whole fetch loop will be retried at next lookup.
1549 */
1550static void pkg_uncache(
1551 HV *cache,
1552 HV *pkg,
1553 char *method)
1554{
1555 (void) hv_delete(cache, HvNAME(pkg), strlen(HvNAME(pkg)), G_DISCARD);
1556}
1557
1558/*
7a6a85bf
RG
1559 * pkg_can
1560 *
1561 * Our own "UNIVERSAL::can", which caches results.
1562 *
1563 * Returns the routine reference as an SV*, or null if the object does not
1564 * know about the method.
1565 */
f0ffaed8
JH
1566static SV *pkg_can(
1567 HV *cache,
1568 HV *pkg,
1569 char *method)
7a6a85bf
RG
1570{
1571 SV **svh;
1572 SV *sv;
1573
1574 TRACEME(("pkg_can for %s->%s", HvNAME(pkg), method));
1575
1576 /*
1577 * Look into the cache to see whether we already have determined
1578 * where the routine was, if any.
1579 *
1580 * NOTA BENE: we don't use `method' at all in our lookup, since we know
1581 * that only one hook (i.e. always the same) is cached in a given cache.
1582 */
1583
1584 svh = hv_fetch(cache, HvNAME(pkg), strlen(HvNAME(pkg)), FALSE);
1585 if (svh) {
1586 sv = *svh;
1587 if (!SvOK(sv)) {
1588 TRACEME(("cached %s->%s: not found", HvNAME(pkg), method));
1589 return (SV *) 0;
1590 } else {
43d061fe 1591 TRACEME(("cached %s->%s: 0x%"UVxf,
9e21b3d0 1592 HvNAME(pkg), method, PTR2UV(sv)));
7a6a85bf
RG
1593 return sv;
1594 }
1595 }
1596
1597 TRACEME(("not cached yet"));
1598 return pkg_fetchmeth(cache, pkg, method); /* Fetch and cache */
1599}
1600
1601/*
1602 * scalar_call
1603 *
1604 * Call routine as obj->hook(av) in scalar context.
1605 * Propagates the single returned value if not called in void context.
1606 */
f0ffaed8
JH
1607static SV *scalar_call(
1608 SV *obj,
1609 SV *hook,
1610 int cloning,
1611 AV *av,
1612 I32 flags)
7a6a85bf
RG
1613{
1614 dSP;
1615 int count;
1616 SV *sv = 0;
1617
1618 TRACEME(("scalar_call (cloning=%d)", cloning));
1619
1620 ENTER;
1621 SAVETMPS;
1622
1623 PUSHMARK(sp);
1624 XPUSHs(obj);
1625 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1626 if (av) {
1627 SV **ary = AvARRAY(av);
1628 int cnt = AvFILLp(av) + 1;
1629 int i;
1630 XPUSHs(ary[0]); /* Frozen string */
1631 for (i = 1; i < cnt; i++) {
43d061fe
JH
1632 TRACEME(("pushing arg #%d (0x%"UVxf")...",
1633 i, PTR2UV(ary[i])));
7a6a85bf
RG
1634 XPUSHs(sv_2mortal(newRV(ary[i])));
1635 }
1636 }
1637 PUTBACK;
1638
1639 TRACEME(("calling..."));
1640 count = perl_call_sv(hook, flags); /* Go back to Perl code */
1641 TRACEME(("count = %d", count));
1642
1643 SPAGAIN;
1644
1645 if (count) {
1646 sv = POPs;
1647 SvREFCNT_inc(sv); /* We're returning it, must stay alive! */
1648 }
1649
1650 PUTBACK;
1651 FREETMPS;
1652 LEAVE;
1653
1654 return sv;
1655}
1656
1657/*
1658 * array_call
1659 *
f9a1036d 1660 * Call routine obj->hook(cloning) in list context.
7a6a85bf
RG
1661 * Returns the list of returned values in an array.
1662 */
f0ffaed8
JH
1663static AV *array_call(
1664 SV *obj,
1665 SV *hook,
1666 int cloning)
7a6a85bf
RG
1667{
1668 dSP;
1669 int count;
1670 AV *av;
1671 int i;
1672
f0ffaed8 1673 TRACEME(("array_call (cloning=%d)", cloning));
7a6a85bf
RG
1674
1675 ENTER;
1676 SAVETMPS;
1677
1678 PUSHMARK(sp);
1679 XPUSHs(obj); /* Target object */
1680 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1681 PUTBACK;
1682
1683 count = perl_call_sv(hook, G_ARRAY); /* Go back to Perl code */
1684
1685 SPAGAIN;
1686
1687 av = newAV();
1688 for (i = count - 1; i >= 0; i--) {
1689 SV *sv = POPs;
1690 av_store(av, i, SvREFCNT_inc(sv));
1691 }
1692
1693 PUTBACK;
1694 FREETMPS;
1695 LEAVE;
1696
1697 return av;
1698}
1699
1700/*
1701 * known_class
1702 *
1703 * Lookup the class name in the `hclass' table and either assign it a new ID
1704 * or return the existing one, by filling in `classnum'.
1705 *
1706 * Return true if the class was known, false if the ID was just generated.
1707 */
f0ffaed8
JH
1708static int known_class(
1709 stcxt_t *cxt,
1710 char *name, /* Class name */
1711 int len, /* Name length */
1712 I32 *classnum)
7a6a85bf
RG
1713{
1714 SV **svh;
1715 HV *hclass = cxt->hclass;
1716
1717 TRACEME(("known_class (%s)", name));
1718
1719 /*
1720 * Recall that we don't store pointers in this hash table, but tags.
1721 * Therefore, we need LOW_32BITS() to extract the relevant parts.
1722 */
1723
1724 svh = hv_fetch(hclass, name, len, FALSE);
1725 if (svh) {
1726 *classnum = LOW_32BITS(*svh);
1727 return TRUE;
1728 }
1729
1730 /*
1731 * Unknown classname, we need to record it.
7a6a85bf
RG
1732 */
1733
1734 cxt->classnum++;
3341c981 1735 if (!hv_store(hclass, name, len, INT2PTR(SV*, cxt->classnum), 0))
7a6a85bf
RG
1736 CROAK(("Unable to record new classname"));
1737
1738 *classnum = cxt->classnum;
1739 return FALSE;
1740}
1741
1742/***
1743 *** Sepcific store routines.
1744 ***/
1745
1746/*
1747 * store_ref
1748 *
1749 * Store a reference.
1750 * Layout is SX_REF <object> or SX_OVERLOAD <object>.
1751 */
f0ffaed8 1752static int store_ref(stcxt_t *cxt, SV *sv)
7a6a85bf 1753{
43d061fe 1754 TRACEME(("store_ref (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
1755
1756 /*
1757 * Follow reference, and check if target is overloaded.
1758 */
1759
1760 sv = SvRV(sv);
1761
1762 if (SvOBJECT(sv)) {
1763 HV *stash = (HV *) SvSTASH(sv);
1764 if (stash && Gv_AMG(stash)) {
9e21b3d0 1765 TRACEME(("ref (0x%"UVxf") is overloaded", PTR2UV(sv)));
7a6a85bf
RG
1766 PUTMARK(SX_OVERLOAD);
1767 } else
1768 PUTMARK(SX_REF);
1769 } else
1770 PUTMARK(SX_REF);
1771
1772 return store(cxt, sv);
1773}
1774
1775/*
1776 * store_scalar
1777 *
1778 * Store a scalar.
1779 *
e16e2ff8 1780 * Layout is SX_LSCALAR <length> <data>, SX_SCALAR <length> <data> or SX_UNDEF.
7a6a85bf
RG
1781 * The <data> section is omitted if <length> is 0.
1782 *
1783 * If integer or double, the layout is SX_INTEGER <data> or SX_DOUBLE <data>.
1784 * Small integers (within [-127, +127]) are stored as SX_BYTE <byte>.
1785 */
f0ffaed8 1786static int store_scalar(stcxt_t *cxt, SV *sv)
7a6a85bf
RG
1787{
1788 IV iv;
1789 char *pv;
1790 STRLEN len;
1791 U32 flags = SvFLAGS(sv); /* "cc -O" may put it in register */
1792
43d061fe 1793 TRACEME(("store_scalar (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
1794
1795 /*
1796 * For efficiency, break the SV encapsulation by peaking at the flags
1797 * directly without using the Perl macros to avoid dereferencing
1798 * sv->sv_flags each time we wish to check the flags.
1799 */
1800
1801 if (!(flags & SVf_OK)) { /* !SvOK(sv) */
1802 if (sv == &PL_sv_undef) {
1803 TRACEME(("immortal undef"));
1804 PUTMARK(SX_SV_UNDEF);
1805 } else {
86bbd6dc 1806 TRACEME(("undef at 0x%"UVxf, PTR2UV(sv)));
7a6a85bf
RG
1807 PUTMARK(SX_UNDEF);
1808 }
1809 return 0;
1810 }
1811
1812 /*
1813 * Always store the string representation of a scalar if it exists.
1814 * Gisle Aas provided me with this test case, better than a long speach:
1815 *
1816 * perl -MDevel::Peek -le '$a="abc"; $a+0; Dump($a)'
1817 * SV = PVNV(0x80c8520)
1818 * REFCNT = 1
1819 * FLAGS = (NOK,POK,pNOK,pPOK)
1820 * IV = 0
1821 * NV = 0
1822 * PV = 0x80c83d0 "abc"\0
1823 * CUR = 3
1824 * LEN = 4
1825 *
1826 * Write SX_SCALAR, length, followed by the actual data.
1827 *
1828 * Otherwise, write an SX_BYTE, SX_INTEGER or an SX_DOUBLE as
1829 * appropriate, followed by the actual (binary) data. A double
1830 * is written as a string if network order, for portability.
1831 *
1832 * NOTE: instead of using SvNOK(sv), we test for SvNOKp(sv).
1833 * The reason is that when the scalar value is tainted, the SvNOK(sv)
1834 * value is false.
1835 *
1836 * The test for a read-only scalar with both POK and NOK set is meant
1837 * to quickly detect &PL_sv_yes and &PL_sv_no without having to pay the
1838 * address comparison for each scalar we store.
1839 */
1840
1841#define SV_MAYBE_IMMORTAL (SVf_READONLY|SVf_POK|SVf_NOK)
1842
1843 if ((flags & SV_MAYBE_IMMORTAL) == SV_MAYBE_IMMORTAL) {
1844 if (sv == &PL_sv_yes) {
1845 TRACEME(("immortal yes"));
1846 PUTMARK(SX_SV_YES);
1847 } else if (sv == &PL_sv_no) {
1848 TRACEME(("immortal no"));
1849 PUTMARK(SX_SV_NO);
1850 } else {
1851 pv = SvPV(sv, len); /* We know it's SvPOK */
1852 goto string; /* Share code below */
1853 }
db670f21
NC
1854 } else if (flags & SVf_POK) {
1855 /* public string - go direct to string read. */
1856 goto string_readlen;
1857 } else if (
1858#if (PATCHLEVEL <= 6)
1859 /* For 5.6 and earlier NV flag trumps IV flag, so only use integer
1860 direct if NV flag is off. */
1861 (flags & (SVf_NOK | SVf_IOK)) == SVf_IOK
1862#else
1863 /* 5.7 rules are that if IV public flag is set, IV value is as
1864 good, if not better, than NV value. */
1865 flags & SVf_IOK
1866#endif
1867 ) {
1868 iv = SvIV(sv);
1869 /*
1870 * Will come here from below with iv set if double is an integer.
1871 */
1872 integer:
7a6a85bf 1873
db670f21
NC
1874 /* Sorry. This isn't in 5.005_56 (IIRC) or earlier. */
1875#ifdef SVf_IVisUV
1876 /* Need to do this out here, else 0xFFFFFFFF becomes iv of -1
1877 * (for example) and that ends up in the optimised small integer
1878 * case.
1879 */
1880 if ((flags & SVf_IVisUV) && SvUV(sv) > IV_MAX) {
1881 TRACEME(("large unsigned integer as string, value = %"UVuf, SvUV(sv)));
1882 goto string_readlen;
1883 }
1884#endif
1885 /*
1886 * Optimize small integers into a single byte, otherwise store as
1887 * a real integer (converted into network order if they asked).
1888 */
7a6a85bf 1889
db670f21
NC
1890 if (iv >= -128 && iv <= 127) {
1891 unsigned char siv = (unsigned char) (iv + 128); /* [0,255] */
1892 PUTMARK(SX_BYTE);
1893 PUTMARK(siv);
1894 TRACEME(("small integer stored as %d", siv));
1895 } else if (cxt->netorder) {
1896#ifndef HAS_HTONL
1897 TRACEME(("no htonl, fall back to string for integer"));
1898 goto string_readlen;
1899#else
1900 I32 niv;
7a6a85bf 1901
7a6a85bf 1902
db670f21
NC
1903#if IVSIZE > 4
1904 if (
1905#ifdef SVf_IVisUV
1906 /* Sorry. This isn't in 5.005_56 (IIRC) or earlier. */
1907 ((flags & SVf_IVisUV) && SvUV(sv) > 0x7FFFFFFF) ||
1908#endif
1909 (iv > 0x7FFFFFFF) || (iv < -0x80000000)) {
1910 /* Bigger than 32 bits. */
1911 TRACEME(("large network order integer as string, value = %"IVdf, iv));
1912 goto string_readlen;
1913 }
1914#endif
7a6a85bf 1915
db670f21
NC
1916 niv = (I32) htonl((I32) iv);
1917 TRACEME(("using network order"));
1918 PUTMARK(SX_NETINT);
1919 WRITE_I32(niv);
1920#endif
1921 } else {
1922 PUTMARK(SX_INTEGER);
1923 WRITE(&iv, sizeof(iv));
1924 }
1925
1926 TRACEME(("ok (integer 0x%"UVxf", value = %"IVdf")", PTR2UV(sv), iv));
1927 } else if (flags & SVf_NOK) {
1928 NV nv;
1929#if (PATCHLEVEL <= 6)
1930 nv = SvNV(sv);
1931 /*
1932 * Watch for number being an integer in disguise.
1933 */
1934 if (nv == (NV) (iv = I_V(nv))) {
1935 TRACEME(("double %"NVff" is actually integer %"IVdf, nv, iv));
1936 goto integer; /* Share code above */
1937 }
1938#else
7a6a85bf 1939
db670f21
NC
1940 SvIV_please(sv);
1941 if (SvIOK(sv)) {
1942 iv = SvIV(sv);
1943 goto integer; /* Share code above */
1944 }
1945 nv = SvNV(sv);
1946#endif
7a6a85bf 1947
db670f21
NC
1948 if (cxt->netorder) {
1949 TRACEME(("double %"NVff" stored as string", nv));
1950 goto string_readlen; /* Share code below */
1951 }
7a6a85bf 1952
db670f21
NC
1953 PUTMARK(SX_DOUBLE);
1954 WRITE(&nv, sizeof(nv));
7a6a85bf 1955
db670f21 1956 TRACEME(("ok (double 0x%"UVxf", value = %"NVff")", PTR2UV(sv), nv));
7a6a85bf 1957
db670f21
NC
1958 } else if (flags & (SVp_POK | SVp_NOK | SVp_IOK)) {
1959 I32 wlen; /* For 64-bit machines */
7a6a85bf 1960
db670f21
NC
1961 string_readlen:
1962 pv = SvPV(sv, len);
7a6a85bf 1963
db670f21
NC
1964 /*
1965 * Will come here from above if it was readonly, POK and NOK but
1966 * neither &PL_sv_yes nor &PL_sv_no.
1967 */
1968 string:
1969
1970 wlen = (I32) len; /* WLEN via STORE_SCALAR expects I32 */
1971 if (SvUTF8 (sv))
1972 STORE_UTF8STR(pv, wlen);
1973 else
1974 STORE_SCALAR(pv, wlen);
1975 TRACEME(("ok (scalar 0x%"UVxf" '%s', length = %"IVdf")",
1976 PTR2UV(sv), SvPVX(sv), (IV)len));
7a6a85bf 1977 } else
db670f21
NC
1978 CROAK(("Can't determine type of %s(0x%"UVxf")",
1979 sv_reftype(sv, FALSE),
1980 PTR2UV(sv)));
1981 return 0; /* Ok, no recursion on scalars */
7a6a85bf
RG
1982}
1983
1984/*
1985 * store_array
1986 *
1987 * Store an array.
1988 *
1989 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
1990 * Each item is stored as <object>.
1991 */
f0ffaed8 1992static int store_array(stcxt_t *cxt, AV *av)
7a6a85bf
RG
1993{
1994 SV **sav;
1995 I32 len = av_len(av) + 1;
1996 I32 i;
1997 int ret;
1998
43d061fe 1999 TRACEME(("store_array (0x%"UVxf")", PTR2UV(av)));
7a6a85bf
RG
2000
2001 /*
2002 * Signal array by emitting SX_ARRAY, followed by the array length.
2003 */
2004
2005 PUTMARK(SX_ARRAY);
2006 WLEN(len);
2007 TRACEME(("size = %d", len));
2008
2009 /*
2010 * Now store each item recursively.
2011 */
2012
2013 for (i = 0; i < len; i++) {
2014 sav = av_fetch(av, i, 0);
2015 if (!sav) {
2016 TRACEME(("(#%d) undef item", i));
2017 STORE_UNDEF();
2018 continue;
2019 }
2020 TRACEME(("(#%d) item", i));
e993d95c 2021 if ((ret = store(cxt, *sav))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2022 return ret;
2023 }
2024
2025 TRACEME(("ok (array)"));
2026
2027 return 0;
2028}
2029
2030/*
2031 * sortcmp
2032 *
2033 * Sort two SVs
2034 * Borrowed from perl source file pp_ctl.c, where it is used by pp_sort.
2035 */
2036static int
f0ffaed8 2037sortcmp(const void *a, const void *b)
7a6a85bf
RG
2038{
2039 return sv_cmp(*(SV * const *) a, *(SV * const *) b);
2040}
2041
2042
2043/*
2044 * store_hash
2045 *
d1be9408 2046 * Store a hash table.
7a6a85bf 2047 *
e16e2ff8
NC
2048 * For a "normal" hash (not restricted, no utf8 keys):
2049 *
7a6a85bf
RG
2050 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
2051 * Values are stored as <object>.
2052 * Keys are stored as <length> <data>, the <data> section being omitted
2053 * if length is 0.
c194a0a3
TB
2054 *
2055 * For a "fancy" hash (restricted or utf8 keys):
2056 *
2057 * Layout is SX_FLAG_HASH <size> <hash flags> followed by each key/value pair,
e16e2ff8
NC
2058 * in random order.
2059 * Values are stored as <object>.
2060 * Keys are stored as <flags> <length> <data>, the <data> section being omitted
2061 * if length is 0.
2062 * Currently the only hash flag is "restriced"
2063 * Key flags are as for hv.h
7a6a85bf 2064 */
f0ffaed8 2065static int store_hash(stcxt_t *cxt, HV *hv)
7a6a85bf 2066{
530b72ba
NC
2067 I32 len =
2068#ifdef HAS_RESTRICTED_HASHES
2069 HvTOTALKEYS(hv);
2070#else
2071 HvKEYS(hv);
2072#endif
7a6a85bf
RG
2073 I32 i;
2074 int ret = 0;
2075 I32 riter;
2076 HE *eiter;
530b72ba
NC
2077 int flagged_hash = ((SvREADONLY(hv)
2078#ifdef HAS_HASH_KEY_FLAGS
2079 || HvHASKFLAGS(hv)
2080#endif
2081 ) ? 1 : 0);
e16e2ff8 2082 unsigned char hash_flags = (SvREADONLY(hv) ? SHV_RESTRICTED : 0);
7a6a85bf 2083
e16e2ff8
NC
2084 if (flagged_hash) {
2085 /* needs int cast for C++ compilers, doesn't it? */
2086 TRACEME(("store_hash (0x%"UVxf") (flags %x)", PTR2UV(hv),
2087 (int) hash_flags));
2088 } else {
2089 TRACEME(("store_hash (0x%"UVxf")", PTR2UV(hv)));
2090 }
7a6a85bf
RG
2091
2092 /*
2093 * Signal hash by emitting SX_HASH, followed by the table length.
2094 */
2095
e16e2ff8
NC
2096 if (flagged_hash) {
2097 PUTMARK(SX_FLAG_HASH);
2098 PUTMARK(hash_flags);
2099 } else {
2100 PUTMARK(SX_HASH);
2101 }
7a6a85bf
RG
2102 WLEN(len);
2103 TRACEME(("size = %d", len));
2104
2105 /*
2106 * Save possible iteration state via each() on that table.
2107 */
2108
2109 riter = HvRITER(hv);
2110 eiter = HvEITER(hv);
2111 hv_iterinit(hv);
2112
2113 /*
2114 * Now store each item recursively.
2115 *
2116 * If canonical is defined to some true value then store each
2117 * key/value pair in sorted order otherwise the order is random.
2118 * Canonical order is irrelevant when a deep clone operation is performed.
2119 *
2120 * Fetch the value from perl only once per store() operation, and only
2121 * when needed.
2122 */
2123
2124 if (
2125 !(cxt->optype & ST_CLONE) && (cxt->canonical == 1 ||
2126 (cxt->canonical < 0 && (cxt->canonical =
e16e2ff8 2127 (SvTRUE(perl_get_sv("Storable::canonical", TRUE)) ? 1 : 0))))
7a6a85bf
RG
2128 ) {
2129 /*
2130 * Storing in order, sorted by key.
2131 * Run through the hash, building up an array of keys in a
2132 * mortal array, sort the array and then run through the
2133 * array.
2134 */
2135
2136 AV *av = newAV();
2137
e16e2ff8
NC
2138 /*av_extend (av, len);*/
2139
7a6a85bf
RG
2140 TRACEME(("using canonical order"));
2141
2142 for (i = 0; i < len; i++) {
530b72ba 2143#ifdef HAS_RESTRICTED_HASHES
e16e2ff8 2144 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
530b72ba
NC
2145#else
2146 HE *he = hv_iternext(hv);
2147#endif
7a6a85bf
RG
2148 SV *key = hv_iterkeysv(he);
2149 av_store(av, AvFILLp(av)+1, key); /* av_push(), really */
2150 }
2151
2152 qsort((char *) AvARRAY(av), len, sizeof(SV *), sortcmp);
2153
2154 for (i = 0; i < len; i++) {
e16e2ff8 2155 unsigned char flags;
7a6a85bf 2156 char *keyval;
e16e2ff8
NC
2157 STRLEN keylen_tmp;
2158 I32 keylen;
7a6a85bf
RG
2159 SV *key = av_shift(av);
2160 HE *he = hv_fetch_ent(hv, key, 0, 0);
2161 SV *val = HeVAL(he);
2162 if (val == 0)
2163 return 1; /* Internal error, not I/O error */
2164
2165 /*
2166 * Store value first.
2167 */
2168
9e21b3d0 2169 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
7a6a85bf 2170
e993d95c 2171 if ((ret = store(cxt, val))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2172 goto out;
2173
2174 /*
2175 * Write key string.
2176 * Keys are written after values to make sure retrieval
2177 * can be optimal in terms of memory usage, where keys are
2178 * read into a fixed unique buffer called kbuf.
2179 * See retrieve_hash() for details.
2180 */
2181
e16e2ff8
NC
2182 /* Implementation of restricted hashes isn't nicely
2183 abstracted: */
2184 flags
2185 = (((hash_flags & SHV_RESTRICTED)
2186 && SvREADONLY(val))
2187 ? SHV_K_LOCKED : 0);
2188 if (val == &PL_sv_undef)
2189 flags |= SHV_K_PLACEHOLDER;
2190
2191 keyval = SvPV(key, keylen_tmp);
2192 keylen = keylen_tmp;
530b72ba
NC
2193#ifdef HAS_UTF8_HASHES
2194 /* If you build without optimisation on pre 5.6
2195 then nothing spots that SvUTF8(key) is always 0,
2196 so the block isn't optimised away, at which point
2197 the linker dislikes the reference to
2198 bytes_from_utf8. */
e16e2ff8
NC
2199 if (SvUTF8(key)) {
2200 const char *keysave = keyval;
2201 bool is_utf8 = TRUE;
2202
2203 /* Just casting the &klen to (STRLEN) won't work
2204 well if STRLEN and I32 are of different widths.
2205 --jhi */
2206 keyval = (char*)bytes_from_utf8((U8*)keyval,
2207 &keylen_tmp,
2208 &is_utf8);
2209
2210 /* If we were able to downgrade here, then than
2211 means that we have a key which only had chars
2212 0-255, but was utf8 encoded. */
2213
2214 if (keyval != keysave) {
2215 keylen = keylen_tmp;
2216 flags |= SHV_K_WASUTF8;
2217 } else {
2218 /* keylen_tmp can't have changed, so no need
2219 to assign back to keylen. */
2220 flags |= SHV_K_UTF8;
2221 }
2222 }
530b72ba 2223#endif
e16e2ff8
NC
2224
2225 if (flagged_hash) {
2226 PUTMARK(flags);
2227 TRACEME(("(#%d) key '%s' flags %x %u", i, keyval, flags, *keyval));
2228 } else {
2229 assert (flags == 0);
2230 TRACEME(("(#%d) key '%s'", i, keyval));
2231 }
7a6a85bf
RG
2232 WLEN(keylen);
2233 if (keylen)
2234 WRITE(keyval, keylen);
e16e2ff8
NC
2235 if (flags & SHV_K_WASUTF8)
2236 Safefree (keyval);
7a6a85bf
RG
2237 }
2238
2239 /*
2240 * Free up the temporary array
2241 */
2242
2243 av_undef(av);
2244 sv_free((SV *) av);
2245
2246 } else {
2247
2248 /*
2249 * Storing in "random" order (in the order the keys are stored
2250 * within the the hash). This is the default and will be faster!
2251 */
2252
2253 for (i = 0; i < len; i++) {
2254 char *key;
2255 I32 len;
e16e2ff8 2256 unsigned char flags;
530b72ba 2257#ifdef HV_ITERNEXT_WANTPLACEHOLDERS
e16e2ff8 2258 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
530b72ba
NC
2259#else
2260 HE *he = hv_iternext(hv);
2261#endif
e16e2ff8
NC
2262 SV *val = (he ? hv_iterval(hv, he) : 0);
2263 SV *key_sv = NULL;
2264 HEK *hek;
7a6a85bf
RG
2265
2266 if (val == 0)
2267 return 1; /* Internal error, not I/O error */
2268
2269 /*
2270 * Store value first.
2271 */
2272
9e21b3d0 2273 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
7a6a85bf 2274
e993d95c 2275 if ((ret = store(cxt, val))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2276 goto out;
2277
e16e2ff8
NC
2278 /* Implementation of restricted hashes isn't nicely
2279 abstracted: */
2280 flags
2281 = (((hash_flags & SHV_RESTRICTED)
2282 && SvREADONLY(val))
2283 ? SHV_K_LOCKED : 0);
2284 if (val == &PL_sv_undef)
2285 flags |= SHV_K_PLACEHOLDER;
2286
2287 hek = HeKEY_hek(he);
2288 len = HEK_LEN(hek);
2289 if (len == HEf_SVKEY) {
2290 /* This is somewhat sick, but the internal APIs are
2291 * such that XS code could put one of these in in
2292 * a regular hash.
2293 * Maybe we should be capable of storing one if
2294 * found.
2295 */
2296 key_sv = HeKEY_sv(he);
2297 flags |= SHV_K_ISSV;
2298 } else {
2299 /* Regular string key. */
530b72ba 2300#ifdef HAS_HASH_KEY_FLAGS
e16e2ff8
NC
2301 if (HEK_UTF8(hek))
2302 flags |= SHV_K_UTF8;
2303 if (HEK_WASUTF8(hek))
2304 flags |= SHV_K_WASUTF8;
530b72ba 2305#endif
e16e2ff8
NC
2306 key = HEK_KEY(hek);
2307 }
7a6a85bf
RG
2308 /*
2309 * Write key string.
2310 * Keys are written after values to make sure retrieval
2311 * can be optimal in terms of memory usage, where keys are
2312 * read into a fixed unique buffer called kbuf.
2313 * See retrieve_hash() for details.
2314 */
2315
e16e2ff8
NC
2316 if (flagged_hash) {
2317 PUTMARK(flags);
2318 TRACEME(("(#%d) key '%s' flags %x", i, key, flags));
2319 } else {
2320 assert (flags == 0);
2321 TRACEME(("(#%d) key '%s'", i, key));
2322 }
2323 if (flags & SHV_K_ISSV) {
2324 store(cxt, key_sv);
2325 } else {
2326 WLEN(len);
2327 if (len)
7a6a85bf 2328 WRITE(key, len);
e16e2ff8 2329 }
7a6a85bf
RG
2330 }
2331 }
2332
43d061fe 2333 TRACEME(("ok (hash 0x%"UVxf")", PTR2UV(hv)));
7a6a85bf
RG
2334
2335out:
2336 HvRITER(hv) = riter; /* Restore hash iterator state */
2337 HvEITER(hv) = eiter;
2338
2339 return ret;
2340}
2341
2342/*
2343 * store_tied
2344 *
2345 * When storing a tied object (be it a tied scalar, array or hash), we lay out
2346 * a special mark, followed by the underlying tied object. For instance, when
2347 * dealing with a tied hash, we store SX_TIED_HASH <hash object>, where
2348 * <hash object> stands for the serialization of the tied hash.
2349 */
f0ffaed8 2350static int store_tied(stcxt_t *cxt, SV *sv)
7a6a85bf
RG
2351{
2352 MAGIC *mg;
2353 int ret = 0;
2354 int svt = SvTYPE(sv);
2355 char mtype = 'P';
2356
43d061fe 2357 TRACEME(("store_tied (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
2358
2359 /*
2360 * We have a small run-time penalty here because we chose to factorise
2361 * all tieds objects into the same routine, and not have a store_tied_hash,
2362 * a store_tied_array, etc...
2363 *
2364 * Don't use a switch() statement, as most compilers don't optimize that
2365 * well for 2/3 values. An if() else if() cascade is just fine. We put
2366 * tied hashes first, as they are the most likely beasts.
2367 */
2368
2369 if (svt == SVt_PVHV) {
2370 TRACEME(("tied hash"));
2371 PUTMARK(SX_TIED_HASH); /* Introduces tied hash */
2372 } else if (svt == SVt_PVAV) {
2373 TRACEME(("tied array"));
2374 PUTMARK(SX_TIED_ARRAY); /* Introduces tied array */
2375 } else {
2376 TRACEME(("tied scalar"));
2377 PUTMARK(SX_TIED_SCALAR); /* Introduces tied scalar */
2378 mtype = 'q';
2379 }
2380
2381 if (!(mg = mg_find(sv, mtype)))
2382 CROAK(("No magic '%c' found while storing tied %s", mtype,
2383 (svt == SVt_PVHV) ? "hash" :
2384 (svt == SVt_PVAV) ? "array" : "scalar"));
2385
2386 /*
2387 * The mg->mg_obj found by mg_find() above actually points to the
2388 * underlying tied Perl object implementation. For instance, if the
2389 * original SV was that of a tied array, then mg->mg_obj is an AV.
2390 *
2391 * Note that we store the Perl object as-is. We don't call its FETCH
2392 * method along the way. At retrieval time, we won't call its STORE
2393 * method either, but the tieing magic will be re-installed. In itself,
2394 * that ensures that the tieing semantics are preserved since futher
2395 * accesses on the retrieved object will indeed call the magic methods...
2396 */
2397
e993d95c 2398 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2399 return ret;
2400
2401 TRACEME(("ok (tied)"));
2402
2403 return 0;
2404}
2405
2406/*
2407 * store_tied_item
2408 *
2409 * Stores a reference to an item within a tied structure:
2410 *
2411 * . \$h{key}, stores both the (tied %h) object and 'key'.
2412 * . \$a[idx], stores both the (tied @a) object and 'idx'.
2413 *
2414 * Layout is therefore either:
2415 * SX_TIED_KEY <object> <key>
2416 * SX_TIED_IDX <object> <index>
2417 */
f0ffaed8 2418static int store_tied_item(stcxt_t *cxt, SV *sv)
7a6a85bf
RG
2419{
2420 MAGIC *mg;
2421 int ret;
2422
43d061fe 2423 TRACEME(("store_tied_item (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
2424
2425 if (!(mg = mg_find(sv, 'p')))
2426 CROAK(("No magic 'p' found while storing reference to tied item"));
2427
2428 /*
2429 * We discriminate between \$h{key} and \$a[idx] via mg_ptr.
2430 */
2431
2432 if (mg->mg_ptr) {
2433 TRACEME(("store_tied_item: storing a ref to a tied hash item"));
2434 PUTMARK(SX_TIED_KEY);
9e21b3d0 2435 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
7a6a85bf 2436
e993d95c 2437 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2438 return ret;
2439
9e21b3d0 2440 TRACEME(("store_tied_item: storing PTR 0x%"UVxf, PTR2UV(mg->mg_ptr)));
7a6a85bf 2441
e993d95c 2442 if ((ret = store(cxt, (SV *) mg->mg_ptr))) /* Idem, for -Wall */
7a6a85bf
RG
2443 return ret;
2444 } else {
2445 I32 idx = mg->mg_len;
2446
2447 TRACEME(("store_tied_item: storing a ref to a tied array item "));
2448 PUTMARK(SX_TIED_IDX);
9e21b3d0 2449 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
7a6a85bf 2450
e993d95c 2451 if ((ret = store(cxt, mg->mg_obj))) /* Idem, for -Wall */
7a6a85bf
RG
2452 return ret;
2453
2454 TRACEME(("store_tied_item: storing IDX %d", idx));
2455
2456 WLEN(idx);
2457 }
2458
2459 TRACEME(("ok (tied item)"));
2460
2461 return 0;
2462}
2463
2464/*
2465 * store_hook -- dispatched manually, not via sv_store[]
2466 *
2467 * The blessed SV is serialized by a hook.
2468 *
2469 * Simple Layout is:
2470 *
2471 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2472 *
2473 * where <flags> indicates how long <len>, <len2> and <len3> are, whether
2474 * the trailing part [] is present, the type of object (scalar, array or hash).
2475 * There is also a bit which says how the classname is stored between:
2476 *
2477 * <len> <classname>
2478 * <index>
2479 *
2480 * and when the <index> form is used (classname already seen), the "large
2481 * classname" bit in <flags> indicates how large the <index> is.
2482 *
2483 * The serialized string returned by the hook is of length <len2> and comes
2484 * next. It is an opaque string for us.
2485 *
2486 * Those <len3> object IDs which are listed last represent the extra references
2487 * not directly serialized by the hook, but which are linked to the object.
2488 *
2489 * When recursion is mandated to resolve object-IDs not yet seen, we have
2490 * instead, with <header> being flags with bits set to indicate the object type
2491 * and that recursion was indeed needed:
2492 *
2493 * SX_HOOK <header> <object> <header> <object> <flags>
2494 *
2495 * that same header being repeated between serialized objects obtained through
2496 * recursion, until we reach flags indicating no recursion, at which point
2497 * we know we've resynchronized with a single layout, after <flags>.
b12202d0
JH
2498 *
2499 * When storing a blessed ref to a tied variable, the following format is
2500 * used:
2501 *
2502 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
2503 *
2504 * The first <flags> indication carries an object of type SHT_EXTRA, and the
2505 * real object type is held in the <extra> flag. At the very end of the
2506 * serialization stream, the underlying magic object is serialized, just like
2507 * any other tied variable.
7a6a85bf 2508 */
f0ffaed8
JH
2509static int store_hook(
2510 stcxt_t *cxt,
2511 SV *sv,
2512 int type,
2513 HV *pkg,
2514 SV *hook)
7a6a85bf
RG
2515{
2516 I32 len;
2517 char *class;
2518 STRLEN len2;
2519 SV *ref;
2520 AV *av;
2521 SV **ary;
2522 int count; /* really len3 + 1 */
2523 unsigned char flags;
2524 char *pv;
2525 int i;
2526 int recursed = 0; /* counts recursion */
2527 int obj_type; /* object type, on 2 bits */
2528 I32 classnum;
2529 int ret;
2530 int clone = cxt->optype & ST_CLONE;
e993d95c
JH
2531 char mtype = '\0'; /* for blessed ref to tied structures */
2532 unsigned char eflags = '\0'; /* used when object type is SHT_EXTRA */
7a6a85bf
RG
2533
2534 TRACEME(("store_hook, class \"%s\", tagged #%d", HvNAME(pkg), cxt->tagnum));
2535
2536 /*
2537 * Determine object type on 2 bits.
2538 */
2539
2540 switch (type) {
2541 case svis_SCALAR:
2542 obj_type = SHT_SCALAR;
2543 break;
2544 case svis_ARRAY:
2545 obj_type = SHT_ARRAY;
2546 break;
2547 case svis_HASH:
2548 obj_type = SHT_HASH;
2549 break;
b12202d0
JH
2550 case svis_TIED:
2551 /*
2552 * Produced by a blessed ref to a tied data structure, $o in the
2553 * following Perl code.
2554 *
2555 * my %h;
2556 * tie %h, 'FOO';
2557 * my $o = bless \%h, 'BAR';
2558 *
2559 * Signal the tie-ing magic by setting the object type as SHT_EXTRA
2560 * (since we have only 2 bits in <flags> to store the type), and an
2561 * <extra> byte flag will be emitted after the FIRST <flags> in the
2562 * stream, carrying what we put in `eflags'.
2563 */
2564 obj_type = SHT_EXTRA;
2565 switch (SvTYPE(sv)) {
2566 case SVt_PVHV:
2567 eflags = (unsigned char) SHT_THASH;
2568 mtype = 'P';
2569 break;
2570 case SVt_PVAV:
2571 eflags = (unsigned char) SHT_TARRAY;
2572 mtype = 'P';
2573 break;
2574 default:
2575 eflags = (unsigned char) SHT_TSCALAR;
2576 mtype = 'q';
2577 break;
2578 }
2579 break;
7a6a85bf
RG
2580 default:
2581 CROAK(("Unexpected object type (%d) in store_hook()", type));
2582 }
2583 flags = SHF_NEED_RECURSE | obj_type;
2584
2585 class = HvNAME(pkg);
2586 len = strlen(class);
2587
2588 /*
2589 * To call the hook, we need to fake a call like:
2590 *
2591 * $object->STORABLE_freeze($cloning);
2592 *
2593 * but we don't have the $object here. For instance, if $object is
2594 * a blessed array, what we have in `sv' is the array, and we can't
2595 * call a method on those.
2596 *
2597 * Therefore, we need to create a temporary reference to the object and
2598 * make the call on that reference.
2599 */
2600
2601 TRACEME(("about to call STORABLE_freeze on class %s", class));
2602
2603 ref = newRV_noinc(sv); /* Temporary reference */
2604 av = array_call(ref, hook, clone); /* @a = $object->STORABLE_freeze($c) */
2605 SvRV(ref) = 0;
2606 SvREFCNT_dec(ref); /* Reclaim temporary reference */
2607
2608 count = AvFILLp(av) + 1;
2609 TRACEME(("store_hook, array holds %d items", count));
2610
2611 /*
2612 * If they return an empty list, it means they wish to ignore the
2613 * hook for this class (and not just this instance -- that's for them
2614 * to handle if they so wish).
2615 *
2616 * Simply disable the cached entry for the hook (it won't be recomputed
2617 * since it's present in the cache) and recurse to store_blessed().
2618 */
2619
2620 if (!count) {
2621 /*
2622 * They must not change their mind in the middle of a serialization.
2623 */
2624
2625 if (hv_fetch(cxt->hclass, class, len, FALSE))
2626 CROAK(("Too late to ignore hooks for %s class \"%s\"",
2627 (cxt->optype & ST_CLONE) ? "cloning" : "storing", class));
2628
2629 pkg_hide(cxt->hook, pkg, "STORABLE_freeze");
2630
2631 ASSERT(!pkg_can(cxt->hook, pkg, "STORABLE_freeze"), ("hook invisible"));
cc964657 2632 TRACEME(("ignoring STORABLE_freeze in class \"%s\"", class));
7a6a85bf
RG
2633
2634 return store_blessed(cxt, sv, type, pkg);
2635 }
2636
2637 /*
2638 * Get frozen string.
2639 */
2640
2641 ary = AvARRAY(av);
2642 pv = SvPV(ary[0], len2);
2643
2644 /*
7a6a85bf
RG
2645 * If they returned more than one item, we need to serialize some
2646 * extra references if not already done.
2647 *
10ffa93f 2648 * Loop over the array, starting at position #1, and for each item,
7a6a85bf
RG
2649 * ensure it is a reference, serialize it if not already done, and
2650 * replace the entry with the tag ID of the corresponding serialized
2651 * object.
2652 *
2653 * We CHEAT by not calling av_fetch() and read directly within the
2654 * array, for speed.
2655 */
2656
2657 for (i = 1; i < count; i++) {
2658 SV **svh;
90826881
JH
2659 SV *rsv = ary[i];
2660 SV *xsv;
2661 AV *av_hook = cxt->hook_seen;
7a6a85bf 2662
90826881
JH
2663 if (!SvROK(rsv))
2664 CROAK(("Item #%d returned by STORABLE_freeze "
2665 "for %s is not a reference", i, class));
2666 xsv = SvRV(rsv); /* Follow ref to know what to look for */
7a6a85bf
RG
2667
2668 /*
2669 * Look in hseen and see if we have a tag already.
2670 * Serialize entry if not done already, and get its tag.
2671 */
2672
13689cfe 2673 if ((svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE)))
7a6a85bf
RG
2674 goto sv_seen; /* Avoid moving code too far to the right */
2675
9e21b3d0 2676 TRACEME(("listed object %d at 0x%"UVxf" is unknown", i-1, PTR2UV(xsv)));
7a6a85bf
RG
2677
2678 /*
2679 * We need to recurse to store that object and get it to be known
2680 * so that we can resolve the list of object-IDs at retrieve time.
2681 *
2682 * The first time we do this, we need to emit the proper header
2683 * indicating that we recursed, and what the type of object is (the
2684 * object we're storing via a user-hook). Indeed, during retrieval,
2685 * we'll have to create the object before recursing to retrieve the
2686 * others, in case those would point back at that object.
2687 */
2688
b12202d0
JH
2689 /* [SX_HOOK] <flags> [<extra>] <object>*/
2690 if (!recursed++) {
7a6a85bf 2691 PUTMARK(SX_HOOK);
b12202d0
JH
2692 PUTMARK(flags);
2693 if (obj_type == SHT_EXTRA)
2694 PUTMARK(eflags);
2695 } else
2696 PUTMARK(flags);
7a6a85bf 2697
e993d95c 2698 if ((ret = store(cxt, xsv))) /* Given by hook for us to store */
7a6a85bf
RG
2699 return ret;
2700
2701 svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE);
2702 if (!svh)
2703 CROAK(("Could not serialize item #%d from hook in %s", i, class));
2704
2705 /*
90826881
JH
2706 * It was the first time we serialized `xsv'.
2707 *
2708 * Keep this SV alive until the end of the serialization: if we
2709 * disposed of it right now by decrementing its refcount, and it was
2710 * a temporary value, some next temporary value allocated during
2711 * another STORABLE_freeze might take its place, and we'd wrongly
2712 * assume that new SV was already serialized, based on its presence
2713 * in cxt->hseen.
2714 *
2715 * Therefore, push it away in cxt->hook_seen.
7a6a85bf
RG
2716 */
2717
90826881
JH
2718 av_store(av_hook, AvFILLp(av_hook)+1, SvREFCNT_inc(xsv));
2719
7a6a85bf 2720 sv_seen:
90826881
JH
2721 /*
2722 * Dispose of the REF they returned. If we saved the `xsv' away
2723 * in the array of returned SVs, that will not cause the underlying
2724 * referenced SV to be reclaimed.
2725 */
2726
2727 ASSERT(SvREFCNT(xsv) > 1, ("SV will survive disposal of its REF"));
2728 SvREFCNT_dec(rsv); /* Dispose of reference */
2729
2730 /*
2731 * Replace entry with its tag (not a real SV, so no refcnt increment)
2732 */
2733
7a6a85bf 2734 ary[i] = *svh;
76edffbb 2735 TRACEME(("listed object %d at 0x%"UVxf" is tag #%"UVuf,
d67b2c17 2736 i-1, PTR2UV(xsv), PTR2UV(*svh)));
7a6a85bf
RG
2737 }
2738
2739 /*
dd19458b
JH
2740 * Allocate a class ID if not already done.
2741 *
2742 * This needs to be done after the recursion above, since at retrieval
2743 * time, we'll see the inner objects first. Many thanks to
2744 * Salvador Ortiz Garcia <sog@msg.com.mx> who spot that bug and
2745 * proposed the right fix. -- RAM, 15/09/2000
2746 */
2747
2748 if (!known_class(cxt, class, len, &classnum)) {
2749 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2750 classnum = -1; /* Mark: we must store classname */
2751 } else {
2752 TRACEME(("already seen class %s, ID = %d", class, classnum));
2753 }
2754
2755 /*
7a6a85bf
RG
2756 * Compute leading flags.
2757 */
2758
2759 flags = obj_type;
2760 if (((classnum == -1) ? len : classnum) > LG_SCALAR)
2761 flags |= SHF_LARGE_CLASSLEN;
2762 if (classnum != -1)
2763 flags |= SHF_IDX_CLASSNAME;
2764 if (len2 > LG_SCALAR)
2765 flags |= SHF_LARGE_STRLEN;
2766 if (count > 1)
2767 flags |= SHF_HAS_LIST;
2768 if (count > (LG_SCALAR + 1))
2769 flags |= SHF_LARGE_LISTLEN;
2770
2771 /*
2772 * We're ready to emit either serialized form:
2773 *
2774 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2775 * SX_HOOK <flags> <index> <len2> <str> [<len3> <object-IDs>]
2776 *
2777 * If we recursed, the SX_HOOK has already been emitted.
2778 */
2779
9e21b3d0
JH
2780 TRACEME(("SX_HOOK (recursed=%d) flags=0x%x "
2781 "class=%"IVdf" len=%"IVdf" len2=%"IVdf" len3=%d",
d67b2c17 2782 recursed, flags, (IV)classnum, (IV)len, (IV)len2, count-1));
7a6a85bf 2783
b12202d0
JH
2784 /* SX_HOOK <flags> [<extra>] */
2785 if (!recursed) {
7a6a85bf 2786 PUTMARK(SX_HOOK);
b12202d0
JH
2787 PUTMARK(flags);
2788 if (obj_type == SHT_EXTRA)
2789 PUTMARK(eflags);
2790 } else
2791 PUTMARK(flags);
7a6a85bf
RG
2792
2793 /* <len> <classname> or <index> */
2794 if (flags & SHF_IDX_CLASSNAME) {
2795 if (flags & SHF_LARGE_CLASSLEN)
2796 WLEN(classnum);
2797 else {
2798 unsigned char cnum = (unsigned char) classnum;
2799 PUTMARK(cnum);
2800 }
2801 } else {
2802 if (flags & SHF_LARGE_CLASSLEN)
2803 WLEN(len);
2804 else {
2805 unsigned char clen = (unsigned char) len;
2806 PUTMARK(clen);
2807 }
2808 WRITE(class, len); /* Final \0 is omitted */
2809 }
2810
2811 /* <len2> <frozen-str> */
cc964657
JH
2812 if (flags & SHF_LARGE_STRLEN) {
2813 I32 wlen2 = len2; /* STRLEN might be 8 bytes */
2814 WLEN(wlen2); /* Must write an I32 for 64-bit machines */
2815 } else {
7a6a85bf
RG
2816 unsigned char clen = (unsigned char) len2;
2817 PUTMARK(clen);
2818 }
2819 if (len2)
7c436af3 2820 WRITE(pv, (SSize_t)len2); /* Final \0 is omitted */
7a6a85bf
RG
2821
2822 /* [<len3> <object-IDs>] */
2823 if (flags & SHF_HAS_LIST) {
2824 int len3 = count - 1;
2825 if (flags & SHF_LARGE_LISTLEN)
2826 WLEN(len3);
2827 else {
2828 unsigned char clen = (unsigned char) len3;
2829 PUTMARK(clen);
2830 }
2831
2832 /*
2833 * NOTA BENE, for 64-bit machines: the ary[i] below does not yield a
2834 * real pointer, rather a tag number, well under the 32-bit limit.
2835 */
2836
2837 for (i = 1; i < count; i++) {
2838 I32 tagval = htonl(LOW_32BITS(ary[i]));
9e21b3d0 2839 WRITE_I32(tagval);
7a6a85bf
RG
2840 TRACEME(("object %d, tag #%d", i-1, ntohl(tagval)));
2841 }
2842 }
2843
2844 /*
2845 * Free the array. We need extra care for indices after 0, since they
2846 * don't hold real SVs but integers cast.
2847 */
2848
2849 if (count > 1)
2850 AvFILLp(av) = 0; /* Cheat, nothing after 0 interests us */
2851 av_undef(av);
2852 sv_free((SV *) av);
2853
b12202d0
JH
2854 /*
2855 * If object was tied, need to insert serialization of the magic object.
2856 */
2857
2858 if (obj_type == SHT_EXTRA) {
2859 MAGIC *mg;
2860
2861 if (!(mg = mg_find(sv, mtype))) {
2862 int svt = SvTYPE(sv);
2863 CROAK(("No magic '%c' found while storing ref to tied %s with hook",
2864 mtype, (svt == SVt_PVHV) ? "hash" :
2865 (svt == SVt_PVAV) ? "array" : "scalar"));
2866 }
2867
2868 TRACEME(("handling the magic object 0x%"UVxf" part of 0x%"UVxf,
2869 PTR2UV(mg->mg_obj), PTR2UV(sv)));
2870
2871 /*
2872 * [<magic object>]
2873 */
2874
e993d95c 2875 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
b12202d0
JH
2876 return ret;
2877 }
2878
7a6a85bf
RG
2879 return 0;
2880}
2881
2882/*
2883 * store_blessed -- dispatched manually, not via sv_store[]
2884 *
2885 * Check whether there is a STORABLE_xxx hook defined in the class or in one
2886 * of its ancestors. If there is, then redispatch to store_hook();
2887 *
2888 * Otherwise, the blessed SV is stored using the following layout:
2889 *
2890 * SX_BLESS <flag> <len> <classname> <object>
2891 *
2892 * where <flag> indicates whether <len> is stored on 0 or 4 bytes, depending
2893 * on the high-order bit in flag: if 1, then length follows on 4 bytes.
2894 * Otherwise, the low order bits give the length, thereby giving a compact
2895 * representation for class names less than 127 chars long.
2896 *
2897 * Each <classname> seen is remembered and indexed, so that the next time
2898 * an object in the blessed in the same <classname> is stored, the following
2899 * will be emitted:
2900 *
2901 * SX_IX_BLESS <flag> <index> <object>
2902 *
2903 * where <index> is the classname index, stored on 0 or 4 bytes depending
2904 * on the high-order bit in flag (same encoding as above for <len>).
2905 */
f0ffaed8
JH
2906static int store_blessed(
2907 stcxt_t *cxt,
2908 SV *sv,
2909 int type,
2910 HV *pkg)
7a6a85bf
RG
2911{
2912 SV *hook;
2913 I32 len;
2914 char *class;
2915 I32 classnum;
2916
2917 TRACEME(("store_blessed, type %d, class \"%s\"", type, HvNAME(pkg)));
2918
2919 /*
2920 * Look for a hook for this blessed SV and redirect to store_hook()
2921 * if needed.
2922 */
2923
2924 hook = pkg_can(cxt->hook, pkg, "STORABLE_freeze");
2925 if (hook)
2926 return store_hook(cxt, sv, type, pkg, hook);
2927
2928 /*
2929 * This is a blessed SV without any serialization hook.
2930 */
2931
2932 class = HvNAME(pkg);
2933 len = strlen(class);
2934
43d061fe
JH
2935 TRACEME(("blessed 0x%"UVxf" in %s, no hook: tagged #%d",
2936 PTR2UV(sv), class, cxt->tagnum));
7a6a85bf
RG
2937
2938 /*
2939 * Determine whether it is the first time we see that class name (in which
2940 * case it will be stored in the SX_BLESS form), or whether we already
2941 * saw that class name before (in which case the SX_IX_BLESS form will be
2942 * used).
2943 */
2944
2945 if (known_class(cxt, class, len, &classnum)) {
2946 TRACEME(("already seen class %s, ID = %d", class, classnum));
2947 PUTMARK(SX_IX_BLESS);
2948 if (classnum <= LG_BLESS) {
2949 unsigned char cnum = (unsigned char) classnum;
2950 PUTMARK(cnum);
2951 } else {
2952 unsigned char flag = (unsigned char) 0x80;
2953 PUTMARK(flag);
2954 WLEN(classnum);
2955 }
2956 } else {
2957 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2958 PUTMARK(SX_BLESS);
2959 if (len <= LG_BLESS) {
2960 unsigned char clen = (unsigned char) len;
2961 PUTMARK(clen);
2962 } else {
2963 unsigned char flag = (unsigned char) 0x80;
2964 PUTMARK(flag);
2965 WLEN(len); /* Don't BER-encode, this should be rare */
2966 }
2967 WRITE(class, len); /* Final \0 is omitted */
2968 }
2969
2970 /*
2971 * Now emit the <object> part.
2972 */
2973
2974 return SV_STORE(type)(cxt, sv);
2975}
2976
2977/*
2978 * store_other
2979 *
2980 * We don't know how to store the item we reached, so return an error condition.
2981 * (it's probably a GLOB, some CODE reference, etc...)
2982 *
2983 * If they defined the `forgive_me' variable at the Perl level to some
2984 * true value, then don't croak, just warn, and store a placeholder string
2985 * instead.
2986 */
f0ffaed8 2987static int store_other(stcxt_t *cxt, SV *sv)
7a6a85bf 2988{
cc964657 2989 I32 len;
7a6a85bf
RG
2990 static char buf[80];
2991
2992 TRACEME(("store_other"));
2993
2994 /*
2995 * Fetch the value from perl only once per store() operation.
2996 */
2997
2998 if (
2999 cxt->forgive_me == 0 ||
3000 (cxt->forgive_me < 0 && !(cxt->forgive_me =
3001 SvTRUE(perl_get_sv("Storable::forgive_me", TRUE)) ? 1 : 0))
3002 )
3003 CROAK(("Can't store %s items", sv_reftype(sv, FALSE)));
3004
43d061fe
JH
3005 warn("Can't store item %s(0x%"UVxf")",
3006 sv_reftype(sv, FALSE), PTR2UV(sv));
7a6a85bf
RG
3007
3008 /*
3009 * Store placeholder string as a scalar instead...
3010 */
3011
13689cfe 3012 (void) sprintf(buf, "You lost %s(0x%"UVxf")%c", sv_reftype(sv, FALSE),
e993d95c 3013 PTR2UV(sv), (char) 0);
7a6a85bf
RG
3014
3015 len = strlen(buf);
3016 STORE_SCALAR(buf, len);
1cf92b12 3017 TRACEME(("ok (dummy \"%s\", length = %"IVdf")", buf, (IV) len));
7a6a85bf
RG
3018
3019 return 0;
3020}
3021
3022/***
3023 *** Store driving routines
3024 ***/
3025
3026/*
3027 * sv_type
3028 *
3029 * WARNING: partially duplicates Perl's sv_reftype for speed.
3030 *
3031 * Returns the type of the SV, identified by an integer. That integer
3032 * may then be used to index the dynamic routine dispatch table.
3033 */
f0ffaed8 3034static int sv_type(SV *sv)
7a6a85bf
RG
3035{
3036 switch (SvTYPE(sv)) {
3037 case SVt_NULL:
3038 case SVt_IV:
3039 case SVt_NV:
3040 /*
3041 * No need to check for ROK, that can't be set here since there
3042 * is no field capable of hodling the xrv_rv reference.
3043 */
3044 return svis_SCALAR;
3045 case SVt_PV:
3046 case SVt_RV:
3047 case SVt_PVIV:
3048 case SVt_PVNV:
3049 /*
3050 * Starting from SVt_PV, it is possible to have the ROK flag
3051 * set, the pointer to the other SV being either stored in
3052 * the xrv_rv (in the case of a pure SVt_RV), or as the
3053 * xpv_pv field of an SVt_PV and its heirs.
3054 *
3055 * However, those SV cannot be magical or they would be an
3056 * SVt_PVMG at least.
3057 */
3058 return SvROK(sv) ? svis_REF : svis_SCALAR;
3059 case SVt_PVMG:
3060 case SVt_PVLV: /* Workaround for perl5.004_04 "LVALUE" bug */
3061 if (SvRMAGICAL(sv) && (mg_find(sv, 'p')))
3062 return svis_TIED_ITEM;
3063 /* FALL THROUGH */
3064 case SVt_PVBM:
3065 if (SvRMAGICAL(sv) && (mg_find(sv, 'q')))
3066 return svis_TIED;
3067 return SvROK(sv) ? svis_REF : svis_SCALAR;
3068 case SVt_PVAV:
3069 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
3070 return svis_TIED;
3071 return svis_ARRAY;
3072 case SVt_PVHV:
3073 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
3074 return svis_TIED;
3075 return svis_HASH;
3076 default:
3077 break;
3078 }
3079
3080 return svis_OTHER;
3081}
3082
3083/*
3084 * store
3085 *
3086 * Recursively store objects pointed to by the sv to the specified file.
3087 *
3088 * Layout is <content> or SX_OBJECT <tagnum> if we reach an already stored
3089 * object (one for which storage has started -- it may not be over if we have
3090 * a self-referenced structure). This data set forms a stored <object>.
3091 */
f0ffaed8 3092static int store(stcxt_t *cxt, SV *sv)
7a6a85bf
RG
3093{
3094 SV **svh;
3095 int ret;
7a6a85bf 3096 int type;
43d061fe 3097 HV *hseen = cxt->hseen;
7a6a85bf 3098
43d061fe 3099 TRACEME(("store (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
3100
3101 /*
3102 * If object has already been stored, do not duplicate data.
3103 * Simply emit the SX_OBJECT marker followed by its tag data.
3104 * The tag is always written in network order.
3105 *
3106 * NOTA BENE, for 64-bit machines: the "*svh" below does not yield a
3107 * real pointer, rather a tag number (watch the insertion code below).
3108 * That means it pobably safe to assume it is well under the 32-bit limit,
3109 * and makes the truncation safe.
3110 * -- RAM, 14/09/1999
3111 */
3112
3113 svh = hv_fetch(hseen, (char *) &sv, sizeof(sv), FALSE);
3114 if (svh) {
3115 I32 tagval = htonl(LOW_32BITS(*svh));
3116
9e21b3d0 3117 TRACEME(("object 0x%"UVxf" seen as #%d", PTR2UV(sv), ntohl(tagval)));
7a6a85bf
RG
3118
3119 PUTMARK(SX_OBJECT);
9e21b3d0 3120 WRITE_I32(tagval);
7a6a85bf
RG
3121 return 0;
3122 }
3123
3124 /*
3125 * Allocate a new tag and associate it with the address of the sv being
3126 * stored, before recursing...
3127 *
3128 * In order to avoid creating new SvIVs to hold the tagnum we just
d1be9408 3129 * cast the tagnum to an SV pointer and store that in the hash. This
7a6a85bf
RG
3130 * means that we must clean up the hash manually afterwards, but gives
3131 * us a 15% throughput increase.
3132 *
7a6a85bf
RG
3133 */
3134
3135 cxt->tagnum++;
3136 if (!hv_store(hseen,
3341c981 3137 (char *) &sv, sizeof(sv), INT2PTR(SV*, cxt->tagnum), 0))
7a6a85bf
RG
3138 return -1;
3139
3140 /*
3141 * Store `sv' and everything beneath it, using appropriate routine.
3142 * Abort immediately if we get a non-zero status back.
3143 */
3144
3145 type = sv_type(sv);
3146
43d061fe
JH
3147 TRACEME(("storing 0x%"UVxf" tag #%d, type %d...",
3148 PTR2UV(sv), cxt->tagnum, type));
7a6a85bf
RG
3149
3150 if (SvOBJECT(sv)) {
3151 HV *pkg = SvSTASH(sv);
3152 ret = store_blessed(cxt, sv, type, pkg);
3153 } else
3154 ret = SV_STORE(type)(cxt, sv);
3155
43d061fe
JH
3156 TRACEME(("%s (stored 0x%"UVxf", refcnt=%d, %s)",
3157 ret ? "FAILED" : "ok", PTR2UV(sv),
7a6a85bf
RG
3158 SvREFCNT(sv), sv_reftype(sv, FALSE)));
3159
3160 return ret;
3161}
3162
3163/*
3164 * magic_write
3165 *
3166 * Write magic number and system information into the file.
3167 * Layout is <magic> <network> [<len> <byteorder> <sizeof int> <sizeof long>
3168 * <sizeof ptr>] where <len> is the length of the byteorder hexa string.
3169 * All size and lenghts are written as single characters here.
3170 *
3171 * Note that no byte ordering info is emitted when <network> is true, since
3172 * integers will be emitted in network order in that case.
3173 */
f0ffaed8 3174static int magic_write(stcxt_t *cxt)
7a6a85bf 3175{
2aeb6432
NC
3176 /*
3177 * Starting with 0.6, the "use_network_order" byte flag is also used to
3178 * indicate the version number of the binary image, encoded in the upper
3179 * bits. The bit 0 is always used to indicate network order.
3180 */
3181 /*
3182 * Starting with 0.7, a full byte is dedicated to the minor version of
3183 * the binary format, which is incremented only when new markers are
3184 * introduced, for instance, but when backward compatibility is preserved.
3185 */
7a6a85bf 3186
2aeb6432
NC
3187 /* Make these at compile time. The WRITE() macro is sufficiently complex
3188 that it saves about 200 bytes doing it this way and only using it
3189 once. */
3190 static const unsigned char network_file_header[] = {
3191 MAGICSTR_BYTES,
3192 (STORABLE_BIN_MAJOR << 1) | 1,
3193 STORABLE_BIN_WRITE_MINOR
3194 };
3195 static const unsigned char file_header[] = {
3196 MAGICSTR_BYTES,
3197 (STORABLE_BIN_MAJOR << 1) | 0,
3198 STORABLE_BIN_WRITE_MINOR,
3199 /* sizeof the array includes the 0 byte at the end: */
3200 (char) sizeof (byteorderstr) - 1,
3201 BYTEORDER_BYTES,
3202 (unsigned char) sizeof(int),
3203 (unsigned char) sizeof(long),
3204 (unsigned char) sizeof(char *),
3205 (unsigned char) sizeof(NV)
3206 };
ee0f7aac
NC
3207#ifdef USE_56_INTERWORK_KLUDGE
3208 static const unsigned char file_header_56[] = {
3209 MAGICSTR_BYTES,
3210 (STORABLE_BIN_MAJOR << 1) | 0,
3211 STORABLE_BIN_WRITE_MINOR,
3212 /* sizeof the array includes the 0 byte at the end: */
3213 (char) sizeof (byteorderstr_56) - 1,
3214 BYTEORDER_BYTES_56,
3215 (unsigned char) sizeof(int),
3216 (unsigned char) sizeof(long),
3217 (unsigned char) sizeof(char *),
3218 (unsigned char) sizeof(NV)
3219 };
3220#endif
2aeb6432
NC
3221 const unsigned char *header;
3222 SSize_t length;
3223
3224 TRACEME(("magic_write on fd=%d", cxt->fio ? PerlIO_fileno(cxt->fio) : -1));
3225
3226 if (cxt->netorder) {
3227 header = network_file_header;
3228 length = sizeof (network_file_header);
3229 } else {
ee0f7aac
NC
3230#ifdef USE_56_INTERWORK_KLUDGE
3231 if (SvTRUE(perl_get_sv("Storable::interwork_56_64bit", TRUE))) {
3232 header = file_header_56;
3233 length = sizeof (file_header_56);
3234 } else
3235#endif
3236 {
3237 header = file_header;
3238 length = sizeof (file_header);
3239 }
2aeb6432
NC
3240 }
3241
3242 if (!cxt->fio) {
3243 /* sizeof the array includes the 0 byte at the end. */
3244 header += sizeof (magicstr) - 1;
3245 length -= sizeof (magicstr) - 1;
3246 }
3247
3248 WRITE(header, length);
3249
3250 if (!cxt->netorder) {
9e21b3d0 3251 TRACEME(("ok (magic_write byteorder = 0x%lx [%d], I%d L%d P%d D%d)",
2aeb6432 3252 (unsigned long) BYTEORDER, (int) sizeof (byteorderstr) - 1,
9e21b3d0
JH
3253 (int) sizeof(int), (int) sizeof(long),
3254 (int) sizeof(char *), (int) sizeof(NV)));
2aeb6432
NC
3255 }
3256 return 0;
7a6a85bf
RG
3257}
3258
3259/*
3260 * do_store
3261 *
3262 * Common code for store operations.
3263 *
3264 * When memory store is requested (f = NULL) and a non null SV* is given in
3265 * `res', it is filled with a new SV created out of the memory buffer.
3266 *
3267 * It is required to provide a non-null `res' when the operation type is not
3268 * dclone() and store() is performed to memory.
3269 */
f0ffaed8
JH
3270static int do_store(
3271 PerlIO *f,
3272 SV *sv,
3273 int optype,
3274 int network_order,
3275 SV **res)
7a6a85bf
RG
3276{
3277 dSTCXT;
3278 int status;
3279
3280 ASSERT(!(f == 0 && !(optype & ST_CLONE)) || res,
3281 ("must supply result SV pointer for real recursion to memory"));
3282
3283 TRACEME(("do_store (optype=%d, netorder=%d)",
3284 optype, network_order));
3285
3286 optype |= ST_STORE;
3287
3288 /*
3289 * Workaround for CROAK leak: if they enter with a "dirty" context,
3290 * free up memory for them now.
3291 */
3292
dd19458b 3293 if (cxt->s_dirty)
7a6a85bf
RG
3294 clean_context(cxt);
3295
3296 /*
3297 * Now that STORABLE_xxx hooks exist, it is possible that they try to
3298 * re-enter store() via the hooks. We need to stack contexts.
3299 */
3300
3301 if (cxt->entry)
3302 cxt = allocate_context(cxt);
3303
3304 cxt->entry++;
3305
3306 ASSERT(cxt->entry == 1, ("starting new recursion"));
dd19458b 3307 ASSERT(!cxt->s_dirty, ("clean context"));
7a6a85bf
RG
3308
3309 /*
3310 * Ensure sv is actually a reference. From perl, we called something
3311 * like:
3312 * pstore(FILE, \@array);
3313 * so we must get the scalar value behing that reference.
3314 */
3315
3316 if (!SvROK(sv))
3317 CROAK(("Not a reference"));
3318 sv = SvRV(sv); /* So follow it to know what to store */
3319
3320 /*
3321 * If we're going to store to memory, reset the buffer.
3322 */
3323
3324 if (!f)
3325 MBUF_INIT(0);
3326
3327 /*
3328 * Prepare context and emit headers.
3329 */
3330
3331 init_store_context(cxt, f, optype, network_order);
3332
3333 if (-1 == magic_write(cxt)) /* Emit magic and ILP info */
3334 return 0; /* Error */
3335
3336 /*
3337 * Recursively store object...
3338 */
3339
3340 ASSERT(is_storing(), ("within store operation"));
3341
3342 status = store(cxt, sv); /* Just do it! */
3343
3344 /*
3345 * If they asked for a memory store and they provided an SV pointer,
3346 * make an SV string out of the buffer and fill their pointer.
3347 *
3348 * When asking for ST_REAL, it's MANDATORY for the caller to provide
3349 * an SV, since context cleanup might free the buffer if we did recurse.
3350 * (unless caller is dclone(), which is aware of that).
3351 */
3352
3353 if (!cxt->fio && res)
3354 *res = mbuf2sv();
3355
3356 /*
3357 * Final cleanup.
3358 *
3359 * The "root" context is never freed, since it is meant to be always
3360 * handy for the common case where no recursion occurs at all (i.e.
3361 * we enter store() outside of any Storable code and leave it, period).
3362 * We know it's the "root" context because there's nothing stacked
3363 * underneath it.
3364 *
3365 * OPTIMIZATION:
3366 *
3367 * When deep cloning, we don't free the context: doing so would force
3368 * us to copy the data in the memory buffer. Sicne we know we're
3369 * about to enter do_retrieve...
3370 */
3371
3372 clean_store_context(cxt);
3373 if (cxt->prev && !(cxt->optype & ST_CLONE))
3374 free_context(cxt);
3375
3376 TRACEME(("do_store returns %d", status));
3377
3378 return status == 0;
3379}
3380
3381/*
3382 * pstore
3383 *
3384 * Store the transitive data closure of given object to disk.
3385 * Returns 0 on error, a true value otherwise.
3386 */
f0ffaed8 3387int pstore(PerlIO *f, SV *sv)
7a6a85bf
RG
3388{
3389 TRACEME(("pstore"));
f0ffaed8 3390 return do_store(f, sv, 0, FALSE, (SV**) 0);
7a6a85bf
RG
3391
3392}
3393
3394/*
3395 * net_pstore
3396 *
3397 * Same as pstore(), but network order is used for integers and doubles are
3398 * emitted as strings.
3399 */
f0ffaed8 3400int net_pstore(PerlIO *f, SV *sv)
7a6a85bf
RG
3401{
3402 TRACEME(("net_pstore"));
f0ffaed8 3403 return do_store(f, sv, 0, TRUE, (SV**) 0);
7a6a85bf
RG
3404}
3405
3406/***
3407 *** Memory stores.
3408 ***/
3409
3410/*
3411 * mbuf2sv
3412 *
3413 * Build a new SV out of the content of the internal memory buffer.
3414 */
f0ffaed8 3415static SV *mbuf2sv(void)
7a6a85bf
RG
3416{
3417 dSTCXT;
3418
3419 return newSVpv(mbase, MBUF_SIZE());
3420}
3421
3422/*
3423 * mstore
3424 *
3425 * Store the transitive data closure of given object to memory.
3426 * Returns undef on error, a scalar value containing the data otherwise.
3427 */
f0ffaed8 3428SV *mstore(SV *sv)
7a6a85bf 3429{
7a6a85bf
RG
3430 SV *out;
3431
3432 TRACEME(("mstore"));
3433
f0ffaed8 3434 if (!do_store((PerlIO*) 0, sv, 0, FALSE, &out))
7a6a85bf
RG
3435 return &PL_sv_undef;
3436
3437 return out;
3438}
3439
3440/*
3441 * net_mstore
3442 *
3443 * Same as mstore(), but network order is used for integers and doubles are
3444 * emitted as strings.
3445 */
f0ffaed8 3446SV *net_mstore(SV *sv)
7a6a85bf 3447{
7a6a85bf
RG
3448 SV *out;
3449
3450 TRACEME(("net_mstore"));
3451
f0ffaed8 3452 if (!do_store((PerlIO*) 0, sv, 0, TRUE, &out))
7a6a85bf
RG
3453 return &PL_sv_undef;
3454
3455 return out;
3456}
3457
3458/***
3459 *** Specific retrieve callbacks.
3460 ***/
3461
3462/*
3463 * retrieve_other
3464 *
3465 * Return an error via croak, since it is not possible that we get here
3466 * under normal conditions, when facing a file produced via pstore().
3467 */
b12202d0 3468static SV *retrieve_other(stcxt_t *cxt, char *cname)
7a6a85bf
RG
3469{
3470 if (
3471 cxt->ver_major != STORABLE_BIN_MAJOR &&
3472 cxt->ver_minor != STORABLE_BIN_MINOR
3473 ) {
3474 CROAK(("Corrupted storable %s (binary v%d.%d), current is v%d.%d",
3475 cxt->fio ? "file" : "string",
3476 cxt->ver_major, cxt->ver_minor,
3477 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
3478 } else {
3479 CROAK(("Corrupted storable %s (binary v%d.%d)",
3480 cxt->fio ? "file" : "string",
3481 cxt->ver_major, cxt->ver_minor));
3482 }
3483
3484 return (SV *) 0; /* Just in case */
3485}
3486
3487/*
3488 * retrieve_idx_blessed
3489 *
3490 * Layout is SX_IX_BLESS <index> <object> with SX_IX_BLESS already read.
3491 * <index> can be coded on either 1 or 5 bytes.
3492 */
b12202d0 3493static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname)
7a6a85bf
RG
3494{
3495 I32 idx;
3496 char *class;
3497 SV **sva;
3498 SV *sv;
3499
3500 TRACEME(("retrieve_idx_blessed (#%d)", cxt->tagnum));
b12202d0 3501 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
7a6a85bf
RG
3502
3503 GETMARK(idx); /* Index coded on a single char? */
3504 if (idx & 0x80)
3505 RLEN(idx);
3506
3507 /*
3508 * Fetch classname in `aclass'
3509 */
3510
3511 sva = av_fetch(cxt->aclass, idx, FALSE);
3512 if (!sva)
e993d95c 3513 CROAK(("Class name #%"IVdf" should have been seen already", (IV) idx));
7a6a85bf
RG
3514
3515 class = SvPVX(*sva); /* We know it's a PV, by construction */
3516
3517 TRACEME(("class ID %d => %s", idx, class));
3518
3519 /*
3520 * Retrieve object and bless it.
3521 */
3522
b12202d0 3523 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
7a6a85bf
RG
3524
3525 return sv;
3526}
3527
3528/*
3529 * retrieve_blessed
3530 *
3531 * Layout is SX_BLESS <len> <classname> <object> with SX_BLESS already read.
3532 * <len> can be coded on either 1 or 5 bytes.
3533 */
b12202d0 3534static SV *retrieve_blessed(stcxt_t *cxt, char *cname)
7a6a85bf
RG
3535{
3536 I32 len;
3537 SV *sv;
3538 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3539 char *class = buf;
3540
3541 TRACEME(("retrieve_blessed (#%d)", cxt->tagnum));
b12202d0 3542 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
7a6a85bf
RG
3543
3544 /*
3545 * Decode class name length and read that name.
3546 *
3547 * Short classnames have two advantages: their length is stored on one
3548 * single byte, and the string can be read on the stack.
3549 */
3550
3551 GETMARK(len); /* Length coded on a single char? */
3552 if (len & 0x80) {
3553 RLEN(len);
3554 TRACEME(("** allocating %d bytes for class name", len+1));
3555 New(10003, class, len+1, char);
3556 }
3557 READ(class, len);
3558 class[len] = '\0'; /* Mark string end */
3559
3560 /*
3561 * It's a new classname, otherwise it would have been an SX_IX_BLESS.
3562 */
3563
b12202d0
JH
3564 TRACEME(("new class name \"%s\" will bear ID = %d", class, cxt->classnum));
3565
7a6a85bf
RG
3566 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3567 return (SV *) 0;
3568
3569 /*
3570 * Retrieve object and bless it.
3571 */
3572
b12202d0
JH
3573 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
3574 if (class != buf)
3575 Safefree(class);
7a6a85bf
RG
3576
3577 return sv;
3578}
3579
3580/*
3581 * retrieve_hook
3582 *
3583 * Layout: SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
3584 * with leading mark already read, as usual.
3585 *
3586 * When recursion was involved during serialization of the object, there
3587 * is an unknown amount of serialized objects after the SX_HOOK mark. Until
3588 * we reach a <flags> marker with the recursion bit cleared.
b12202d0
JH
3589 *
3590 * If the first <flags> byte contains a type of SHT_EXTRA, then the real type
3591 * is held in the <extra> byte, and if the object is tied, the serialized
3592 * magic object comes at the very end:
3593 *
3594 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
3595 *
3596 * This means the STORABLE_thaw hook will NOT get a tied variable during its
3597 * processing (since we won't have seen the magic object by the time the hook
3598 * is called). See comments below for why it was done that way.
7a6a85bf 3599 */
b12202d0 3600static SV *retrieve_hook(stcxt_t *cxt, char *cname)
7a6a85bf
RG
3601{
3602 I32 len;
3603 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3604 char *class = buf;
3605 unsigned int flags;
3606 I32 len2;
3607 SV *frozen;
3608 I32 len3 = 0;
3609 AV *av = 0;
3610 SV *hook;
3611 SV *sv;
3612 SV *rv;
3613 int obj_type;
7a6a85bf 3614 int clone = cxt->optype & ST_CLONE;
b12202d0
JH
3615 char mtype = '\0';
3616 unsigned int extra_type = 0;
7a6a85bf
RG
3617
3618 TRACEME(("retrieve_hook (#%d)", cxt->tagnum));
b12202d0 3619 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
7a6a85bf
RG
3620
3621 /*
3622 * Read flags, which tell us about the type, and whether we need to recurse.
3623 */
3624
3625 GETMARK(flags);
3626
3627 /*
3628 * Create the (empty) object, and mark it as seen.
3629 *
3630 * This must be done now, because tags are incremented, and during
3631 * serialization, the object tag was affected before recursion could
3632 * take place.
3633 */
3634
3635 obj_type = flags & SHF_TYPE_MASK;
3636 switch (obj_type) {
3637 case SHT_SCALAR:
3638 sv = newSV(0);
3639 break;
3640 case SHT_ARRAY:
3641 sv = (SV *) newAV();
3642 break;
3643 case SHT_HASH:
3644 sv = (SV *) newHV();
3645 break;
b12202d0
JH
3646 case SHT_EXTRA:
3647 /*
3648 * Read <extra> flag to know the type of the object.
3649 * Record associated magic type for later.
3650 */
3651 GETMARK(extra_type);
3652 switch (extra_type) {
3653 case SHT_TSCALAR:
3654 sv = newSV(0);
3655 mtype = 'q';
3656 break;
3657 case SHT_TARRAY:
3658 sv = (SV *) newAV();
3659 mtype = 'P';
3660 break;
3661 case SHT_THASH:
3662 sv = (SV *) newHV();
3663 mtype = 'P';
3664 break;
3665 default:
3666 return retrieve_other(cxt, 0); /* Let it croak */
3667 }
3668 break;
7a6a85bf 3669 default:
b12202d0 3670 return retrieve_other(cxt, 0); /* Let it croak */
7a6a85bf 3671 }
b12202d0 3672 SEEN(sv, 0); /* Don't bless yet */
7a6a85bf
RG
3673
3674 /*
3675 * Whilst flags tell us to recurse, do so.
3676 *
3677 * We don't need to remember the addresses returned by retrieval, because
3678 * all the references will be obtained through indirection via the object
3679 * tags in the object-ID list.
10ffa93f
RG
3680 *
3681 * We need to decrement the reference count for these objects
3682 * because, if the user doesn't save a reference to them in the hook,
3683 * they must be freed when this context is cleaned.
7a6a85bf
RG
3684 */
3685
3686 while (flags & SHF_NEED_RECURSE) {
3687 TRACEME(("retrieve_hook recursing..."));
b12202d0 3688 rv = retrieve(cxt, 0);
7a6a85bf
RG
3689 if (!rv)
3690 return (SV *) 0;
10ffa93f 3691 SvREFCNT_dec(rv);
43d061fe
JH
3692 TRACEME(("retrieve_hook back with rv=0x%"UVxf,
3693 PTR2UV(rv)));
7a6a85bf
RG
3694 GETMARK(flags);
3695 }
3696
3697 if (flags & SHF_IDX_CLASSNAME) {
3698 SV **sva;
3699 I32 idx;
3700
3701 /*
3702 * Fetch index from `aclass'
3703 */
3704
3705 if (flags & SHF_LARGE_CLASSLEN)
3706 RLEN(idx);
3707 else
3708 GETMARK(idx);
3709
3710 sva = av_fetch(cxt->aclass, idx, FALSE);
3711 if (!sva)
e993d95c
JH
3712 CROAK(("Class name #%"IVdf" should have been seen already",
3713 (IV) idx));
7a6a85bf
RG
3714
3715 class = SvPVX(*sva); /* We know it's a PV, by construction */
3716 TRACEME(("class ID %d => %s", idx, class));
3717
3718 } else {
3719 /*
3720 * Decode class name length and read that name.
3721 *
3722 * NOTA BENE: even if the length is stored on one byte, we don't read
3723 * on the stack. Just like retrieve_blessed(), we limit the name to
3724 * LG_BLESS bytes. This is an arbitrary decision.
3725 */
3726
3727 if (flags & SHF_LARGE_CLASSLEN)
3728 RLEN(len);
3729 else
3730 GETMARK(len);
3731
3732 if (len > LG_BLESS) {
3733 TRACEME(("** allocating %d bytes for class name", len+1));
3734 New(10003, class, len+1, char);
3735 }
3736
3737 READ(class, len);
3738 class[len] = '\0'; /* Mark string end */
3739
3740 /*
3741 * Record new classname.
3742 */
3743
3744 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3745 return (SV *) 0;
3746 }
3747
3748 TRACEME(("class name: %s", class));
3749
3750 /*
d1be9408 3751 * Decode user-frozen string length and read it in an SV.
7a6a85bf
RG
3752 *
3753 * For efficiency reasons, we read data directly into the SV buffer.
3754 * To understand that code, read retrieve_scalar()
3755 */
3756
3757 if (flags & SHF_LARGE_STRLEN)
3758 RLEN(len2);
3759 else
3760 GETMARK(len2);
3761
3762 frozen = NEWSV(10002, len2);
3763 if (len2) {
3764 SAFEREAD(SvPVX(frozen), len2, frozen);
3765 SvCUR_set(frozen, len2);
3766 *SvEND(frozen) = '\0';
3767 }
3768 (void) SvPOK_only(frozen); /* Validates string pointer */
dd19458b
JH
3769 if (cxt->s_tainted) /* Is input source tainted? */
3770 SvTAINT(frozen);
7a6a85bf
RG
3771
3772 TRACEME(("frozen string: %d bytes", len2));
3773
3774 /*
3775 * Decode object-ID list length, if present.
3776 */
3777
3778 if (flags & SHF_HAS_LIST) {
3779 if (flags & SHF_LARGE_LISTLEN)
3780 RLEN(len3);
3781 else
3782 GETMARK(len3);
3783 if (len3) {
3784 av = newAV();
3785 av_extend(av, len3 + 1); /* Leave room for [0] */
3786 AvFILLp(av) = len3; /* About to be filled anyway */
3787 }
3788 }
3789
3790 TRACEME(("has %d object IDs to link", len3));
3791
3792 /*
3793 * Read object-ID list into array.