This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Update Changes.
[perl5.git] / ext / Storable / Storable.xs
CommitLineData
7a6a85bf
RG
1/*
2 * Store and retrieve mechanism.
3 */
4
5/*
6e0ac6f5 6 * $Id: Storable.xs,v 1.0.1.10 2001/08/28 21:52:14 ram Exp $
7a6a85bf
RG
7 *
8 * Copyright (c) 1995-2000, Raphael Manfredi
9 *
9e21b3d0
JH
10 * You may redistribute only under the same terms as Perl 5, as specified
11 * in the README file that comes with the distribution.
7a6a85bf
RG
12 *
13 * $Log: Storable.xs,v $
6e0ac6f5
JH
14 * Revision 1.0.1.10 2001/08/28 21:52:14 ram
15 * patch13: removed spurious debugging messages
16 *
e993d95c
JH
17 * Revision 1.0.1.9 2001/07/01 11:25:02 ram
18 * patch12: fixed memory corruption on croaks during thaw()
19 * patch12: made code compile cleanly with -Wall (Jarkko Hietaniemi)
20 * patch12: changed tagnum and classnum from I32 to IV in context
21 *
8be2b38b
JH
22 * Revision 1.0.1.8 2001/03/15 00:20:55 ram
23 * patch11: last version was wrongly compiling with assertions on
24 *
b12202d0
JH
25 * Revision 1.0.1.7 2001/02/17 12:25:26 ram
26 * patch8: now bless objects ASAP at retrieve time
27 * patch8: added support for blessed ref to tied structures
28 *
862382c7
JH
29 * Revision 1.0.1.6 2001/01/03 09:40:40 ram
30 * patch7: prototype and casting cleanup
31 * patch7: trace offending package when overloading cannot be restored
32 * patch7: made context cleanup safer to avoid dup freeing
33 *
90826881
JH
34 * Revision 1.0.1.5 2000/11/05 17:21:24 ram
35 * patch6: fixed severe "object lost" bug for STORABLE_freeze returns
36 *
212e9bde
JH
37 * Revision 1.0.1.4 2000/10/26 17:11:04 ram
38 * patch5: auto requires module of blessed ref when STORABLE_thaw misses
39 *
40 * Revision 1.0.1.3 2000/09/29 19:49:57 ram
41 * patch3: avoid using "tainted" and "dirty" since Perl remaps them via cpp
42 *
8be2b38b
JH
43 * Revision 1.0.1.2 2000/09/28 21:43:10 ram
44 * patch2: perls before 5.004_04 lack newSVpvn
45 *
46 * Revision 1.0.1.1 2000/09/17 16:47:49 ram
47 * patch1: now only taint retrieved data when source was tainted
48 * patch1: added support for UTF-8 strings
49 * patch1: fixed store hook bug: was allocating class id too soon
50 *
9e21b3d0
JH
51 * Revision 1.0 2000/09/01 19:40:41 ram
52 * Baseline for first official release.
7a6a85bf
RG
53 *
54 */
55
56#include <EXTERN.h>
57#include <perl.h>
e993d95c 58#include <patchlevel.h> /* Perl's one, needed since 5.6 */
7a6a85bf
RG
59#include <XSUB.h>
60
5dff01bc 61#ifndef NETWARE
e8189732 62#if 0
9e21b3d0
JH
63#define DEBUGME /* Debug mode, turns assertions on as well */
64#define DASSERT /* Assertion mode */
65#endif
5dff01bc
JH
66#else /* NETWARE */
67#if 0 /* On NetWare USE_PERLIO is not used */
68#define DEBUGME /* Debug mode, turns assertions on as well */
69#define DASSERT /* Assertion mode */
70#endif
71#endif
7a6a85bf
RG
72
73/*
74 * Pre PerlIO time when none of USE_PERLIO and PERLIO_IS_STDIO is defined
75 * Provide them with the necessary defines so they can build with pre-5.004.
76 */
77#ifndef USE_PERLIO
78#ifndef PERLIO_IS_STDIO
79#define PerlIO FILE
80#define PerlIO_getc(x) getc(x)
81#define PerlIO_putc(f,x) putc(x,f)
82#define PerlIO_read(x,y,z) fread(y,1,z,x)
83#define PerlIO_write(x,y,z) fwrite(y,1,z,x)
84#define PerlIO_stdoutf printf
85#endif /* PERLIO_IS_STDIO */
86#endif /* USE_PERLIO */
87
88/*
89 * Earlier versions of perl might be used, we can't assume they have the latest!
90 */
f0ffaed8
JH
91
92#ifndef PERL_VERSION /* For perls < 5.6 */
e993d95c 93#define PERL_VERSION PATCHLEVEL
7a6a85bf
RG
94#ifndef newRV_noinc
95#define newRV_noinc(sv) ((Sv = newRV(sv)), --SvREFCNT(SvRV(Sv)), Sv)
96#endif
e993d95c 97#if (PATCHLEVEL <= 4) /* Older perls (<= 5.004) lack PL_ namespace */
7a6a85bf
RG
98#define PL_sv_yes sv_yes
99#define PL_sv_no sv_no
100#define PL_sv_undef sv_undef
e993d95c 101#if (SUBVERSION <= 4) /* 5.004_04 has been reported to lack newSVpvn */
dd19458b 102#define newSVpvn newSVpv
7a6a85bf 103#endif
e993d95c 104#endif /* PATCHLEVEL <= 4 */
7a6a85bf
RG
105#ifndef HvSHAREKEYS_off
106#define HvSHAREKEYS_off(hv) /* Ignore */
107#endif
f0ffaed8
JH
108#ifndef AvFILLp /* Older perls (<=5.003) lack AvFILLp */
109#define AvFILLp AvFILL
110#endif
111typedef double NV; /* Older perls lack the NV type */
cc964657
JH
112#define IVdf "ld" /* Various printf formats for Perl types */
113#define UVuf "lu"
114#define UVof "lo"
115#define UVxf "lx"
116#define INT2PTR(t,v) (t)(IV)(v)
117#define PTR2UV(v) (unsigned long)(v)
f0ffaed8 118#endif /* PERL_VERSION -- perls < 5.6 */
7a6a85bf 119
cc964657 120#ifndef NVef /* The following were not part of perl 5.6 */
9e21b3d0
JH
121#if defined(USE_LONG_DOUBLE) && \
122 defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
123#define NVef PERL_PRIeldbl
124#define NVff PERL_PRIfldbl
125#define NVgf PERL_PRIgldbl
126#else
cc964657
JH
127#define NVef "e"
128#define NVff "f"
129#define NVgf "g"
130#endif
131#endif
132
7a6a85bf 133#ifdef DEBUGME
8be2b38b
JH
134
135#ifndef DASSERT
136#define DASSERT
137#endif
138
90826881
JH
139/*
140 * TRACEME() will only output things when the $Storable::DEBUGME is true.
141 */
142
111e03c1
RG
143#define TRACEME(x) \
144 STMT_START { \
90826881 145 if (SvTRUE(perl_get_sv("Storable::DEBUGME", TRUE))) \
111e03c1
RG
146 { PerlIO_stdoutf x; PerlIO_stdoutf("\n"); } \
147 } STMT_END
7a6a85bf
RG
148#else
149#define TRACEME(x)
8be2b38b 150#endif /* DEBUGME */
7a6a85bf
RG
151
152#ifdef DASSERT
111e03c1
RG
153#define ASSERT(x,y) \
154 STMT_START { \
7a6a85bf
RG
155 if (!(x)) { \
156 PerlIO_stdoutf("ASSERT FAILED (\"%s\", line %d): ", \
157 __FILE__, __LINE__); \
158 PerlIO_stdoutf y; PerlIO_stdoutf("\n"); \
159 } \
111e03c1 160 } STMT_END
7a6a85bf
RG
161#else
162#define ASSERT(x,y)
163#endif
164
165/*
166 * Type markers.
167 */
168
169#define C(x) ((char) (x)) /* For markers with dynamic retrieval handling */
170
171#define SX_OBJECT C(0) /* Already stored object */
dd19458b 172#define SX_LSCALAR C(1) /* Scalar (large binary) follows (length, data) */
7a6a85bf
RG
173#define SX_ARRAY C(2) /* Array forthcominng (size, item list) */
174#define SX_HASH C(3) /* Hash forthcoming (size, key/value pair list) */
175#define SX_REF C(4) /* Reference to object forthcoming */
176#define SX_UNDEF C(5) /* Undefined scalar */
177#define SX_INTEGER C(6) /* Integer forthcoming */
178#define SX_DOUBLE C(7) /* Double forthcoming */
179#define SX_BYTE C(8) /* (signed) byte forthcoming */
180#define SX_NETINT C(9) /* Integer in network order forthcoming */
dd19458b 181#define SX_SCALAR C(10) /* Scalar (binary, small) follows (length, data) */
f062ea6c
PN
182#define SX_TIED_ARRAY C(11) /* Tied array forthcoming */
183#define SX_TIED_HASH C(12) /* Tied hash forthcoming */
184#define SX_TIED_SCALAR C(13) /* Tied scalar forthcoming */
7a6a85bf
RG
185#define SX_SV_UNDEF C(14) /* Perl's immortal PL_sv_undef */
186#define SX_SV_YES C(15) /* Perl's immortal PL_sv_yes */
187#define SX_SV_NO C(16) /* Perl's immortal PL_sv_no */
188#define SX_BLESS C(17) /* Object is blessed */
189#define SX_IX_BLESS C(18) /* Object is blessed, classname given by index */
190#define SX_HOOK C(19) /* Stored via hook, user-defined */
191#define SX_OVERLOAD C(20) /* Overloaded reference */
f062ea6c
PN
192#define SX_TIED_KEY C(21) /* Tied magic key forthcoming */
193#define SX_TIED_IDX C(22) /* Tied magic index forthcoming */
194#define SX_UTF8STR C(23) /* UTF-8 string forthcoming (small) */
195#define SX_LUTF8STR C(24) /* UTF-8 string forthcoming (large) */
196#define SX_FLAG_HASH C(25) /* Hash with flags forthcoming (size, flags, key/flags/value triplet list) */
e16e2ff8 197#define SX_ERROR C(26) /* Error */
7a6a85bf
RG
198
199/*
200 * Those are only used to retrieve "old" pre-0.6 binary images.
201 */
202#define SX_ITEM 'i' /* An array item introducer */
203#define SX_IT_UNDEF 'I' /* Undefined array item */
d1be9408
JF
204#define SX_KEY 'k' /* A hash key introducer */
205#define SX_VALUE 'v' /* A hash value introducer */
7a6a85bf
RG
206#define SX_VL_UNDEF 'V' /* Undefined hash value */
207
208/*
209 * Those are only used to retrieve "old" pre-0.7 binary images
210 */
211
212#define SX_CLASS 'b' /* Object is blessed, class name length <255 */
f062ea6c 213#define SX_LG_CLASS 'B' /* Object is blessed, class name length >255 */
7a6a85bf
RG
214#define SX_STORED 'X' /* End of object */
215
216/*
217 * Limits between short/long length representation.
218 */
219
220#define LG_SCALAR 255 /* Large scalar length limit */
221#define LG_BLESS 127 /* Large classname bless limit */
222
223/*
224 * Operation types
225 */
226
227#define ST_STORE 0x1 /* Store operation */
228#define ST_RETRIEVE 0x2 /* Retrieval operation */
229#define ST_CLONE 0x4 /* Deep cloning operation */
230
231/*
232 * The following structure is used for hash table key retrieval. Since, when
233 * retrieving objects, we'll be facing blessed hash references, it's best
234 * to pre-allocate that buffer once and resize it as the need arises, never
235 * freeing it (keys will be saved away someplace else anyway, so even large
236 * keys are not enough a motivation to reclaim that space).
237 *
238 * This structure is also used for memory store/retrieve operations which
239 * happen in a fixed place before being malloc'ed elsewhere if persistency
240 * is required. Hence the aptr pointer.
241 */
242struct extendable {
243 char *arena; /* Will hold hash key strings, resized as needed */
244 STRLEN asiz; /* Size of aforementionned buffer */
245 char *aptr; /* Arena pointer, for in-place read/write ops */
246 char *aend; /* First invalid address */
247};
248
249/*
250 * At store time:
d1be9408 251 * A hash table records the objects which have already been stored.
7a6a85bf
RG
252 * Those are referred to as SX_OBJECT in the file, and their "tag" (i.e.
253 * an arbitrary sequence number) is used to identify them.
254 *
255 * At retrieve time:
256 * An array table records the objects which have already been retrieved,
257 * as seen by the tag determind by counting the objects themselves. The
258 * reference to that retrieved object is kept in the table, and is returned
259 * when an SX_OBJECT is found bearing that same tag.
260 *
261 * The same processing is used to record "classname" for blessed objects:
262 * indexing by a hash at store time, and via an array at retrieve time.
263 */
264
265typedef unsigned long stag_t; /* Used by pre-0.6 binary format */
266
267/*
268 * The following "thread-safe" related defines were contributed by
269 * Murray Nesbitt <murray@activestate.com> and integrated by RAM, who
270 * only renamed things a little bit to ensure consistency with surrounding
271 * code. -- RAM, 14/09/1999
272 *
273 * The original patch suffered from the fact that the stcxt_t structure
274 * was global. Murray tried to minimize the impact on the code as much as
275 * possible.
276 *
277 * Starting with 0.7, Storable can be re-entrant, via the STORABLE_xxx hooks
278 * on objects. Therefore, the notion of context needs to be generalized,
279 * threading or not.
280 */
281
282#define MY_VERSION "Storable(" XS_VERSION ")"
283
530b72ba
NC
284
285/*
286 * Conditional UTF8 support.
287 *
288 */
289#ifdef SvUTF8_on
290#define STORE_UTF8STR(pv, len) STORE_PV_LEN(pv, len, SX_UTF8STR, SX_LUTF8STR)
291#define HAS_UTF8_SCALARS
292#ifdef HeKUTF8
293#define HAS_UTF8_HASHES
294#define HAS_UTF8_ALL
295#else
296/* 5.6 perl has utf8 scalars but not hashes */
297#endif
298#else
299#define SvUTF8(sv) 0
300#define STORE_UTF8STR(pv, len) CROAK(("panic: storing UTF8 in non-UTF8 perl"))
301#endif
302#ifndef HAS_UTF8_ALL
303#define UTF8_CROAK() CROAK(("Cannot retrieve UTF8 data in non-UTF8 perl"))
304#endif
305
306#ifdef HvPLACEHOLDERS
307#define HAS_RESTRICTED_HASHES
308#else
309#define HVhek_PLACEHOLD 0x200
310#define RESTRICTED_HASH_CROAK() CROAK(("Cannot retrieve restricted hash"))
311#endif
312
313#ifdef HvHASKFLAGS
314#define HAS_HASH_KEY_FLAGS
315#endif
316
dd19458b
JH
317/*
318 * Fields s_tainted and s_dirty are prefixed with s_ because Perl's include
319 * files remap tainted and dirty when threading is enabled. That's bad for
320 * perl to remap such common words. -- RAM, 29/09/00
321 */
322
7a6a85bf
RG
323typedef struct stcxt {
324 int entry; /* flags recursion */
325 int optype; /* type of traversal operation */
e993d95c
JH
326 HV *hseen; /* which objects have been seen, store time */
327 AV *hook_seen; /* which SVs were returned by STORABLE_freeze() */
328 AV *aseen; /* which objects have been seen, retrieve time */
329 HV *hclass; /* which classnames have been seen, store time */
330 AV *aclass; /* which classnames have been seen, retrieve time */
331 HV *hook; /* cache for hook methods per class name */
332 IV tagnum; /* incremented at store time for each seen object */
333 IV classnum; /* incremented at store time for each seen classname */
334 int netorder; /* true if network order used */
335 int s_tainted; /* true if input source is tainted, at retrieve time */
336 int forgive_me; /* whether to be forgiving... */
337 int canonical; /* whether to store hashes sorted by key */
530b72ba
NC
338#ifndef HAS_RESTRICTED_HASHES
339 int derestrict; /* whether to downgrade restrcted hashes */
340#endif
341#ifndef HAS_UTF8_ALL
342 int use_bytes; /* whether to bytes-ify utf8 */
343#endif
e8189732 344 int accept_future_minor; /* croak immediately on future minor versions? */
dd19458b 345 int s_dirty; /* context is dirty due to CROAK() -- can be cleaned */
e993d95c
JH
346 int membuf_ro; /* true means membuf is read-only and msaved is rw */
347 struct extendable keybuf; /* for hash key retrieval */
348 struct extendable membuf; /* for memory store/retrieve operations */
349 struct extendable msaved; /* where potentially valid mbuf is saved */
7a6a85bf
RG
350 PerlIO *fio; /* where I/O are performed, NULL for memory */
351 int ver_major; /* major of version for retrieved object */
352 int ver_minor; /* minor of version for retrieved object */
353 SV *(**retrieve_vtbl)(); /* retrieve dispatch table */
111e03c1
RG
354 SV *prev; /* contexts chained backwards in real recursion */
355 SV *my_sv; /* the blessed scalar who's SvPVX() I am */
7a6a85bf
RG
356} stcxt_t;
357
111e03c1
RG
358#define NEW_STORABLE_CXT_OBJ(cxt) \
359 STMT_START { \
360 SV *self = newSV(sizeof(stcxt_t) - 1); \
361 SV *my_sv = newRV_noinc(self); \
362 sv_bless(my_sv, gv_stashpv("Storable::Cxt", TRUE)); \
363 cxt = (stcxt_t *)SvPVX(self); \
364 Zero(cxt, 1, stcxt_t); \
365 cxt->my_sv = my_sv; \
366 } STMT_END
367
7a6a85bf
RG
368#if defined(MULTIPLICITY) || defined(PERL_OBJECT) || defined(PERL_CAPI)
369
e993d95c 370#if (PATCHLEVEL <= 4) && (SUBVERSION < 68)
7a6a85bf
RG
371#define dSTCXT_SV \
372 SV *perinterp_sv = perl_get_sv(MY_VERSION, FALSE)
373#else /* >= perl5.004_68 */
374#define dSTCXT_SV \
375 SV *perinterp_sv = *hv_fetch(PL_modglobal, \
376 MY_VERSION, sizeof(MY_VERSION)-1, TRUE)
377#endif /* < perl5.004_68 */
378
379#define dSTCXT_PTR(T,name) \
111e03c1 380 T name = ((perinterp_sv && SvIOK(perinterp_sv) && SvIVX(perinterp_sv) \
436c6dd3 381 ? (T)SvPVX(SvRV(INT2PTR(SV*,SvIVX(perinterp_sv)))) : (T) 0))
7a6a85bf
RG
382#define dSTCXT \
383 dSTCXT_SV; \
384 dSTCXT_PTR(stcxt_t *, cxt)
385
111e03c1
RG
386#define INIT_STCXT \
387 dSTCXT; \
388 NEW_STORABLE_CXT_OBJ(cxt); \
389 sv_setiv(perinterp_sv, PTR2IV(cxt->my_sv))
7a6a85bf 390
111e03c1
RG
391#define SET_STCXT(x) \
392 STMT_START { \
7a6a85bf 393 dSTCXT_SV; \
111e03c1
RG
394 sv_setiv(perinterp_sv, PTR2IV(x->my_sv)); \
395 } STMT_END
7a6a85bf
RG
396
397#else /* !MULTIPLICITY && !PERL_OBJECT && !PERL_CAPI */
398
399static stcxt_t Context;
400static stcxt_t *Context_ptr = &Context;
401#define dSTCXT stcxt_t *cxt = Context_ptr
111e03c1
RG
402#define INIT_STCXT \
403 dSTCXT; \
404 NEW_STORABLE_CXT_OBJ(cxt)
405
406#define SET_STCXT(x) Context_ptr = x
7a6a85bf
RG
407
408#endif /* MULTIPLICITY || PERL_OBJECT || PERL_CAPI */
409
410/*
411 * KNOWN BUG:
412 * Croaking implies a memory leak, since we don't use setjmp/longjmp
413 * to catch the exit and free memory used during store or retrieve
414 * operations. This is not too difficult to fix, but I need to understand
415 * how Perl does it, and croaking is exceptional anyway, so I lack the
416 * motivation to do it.
417 *
418 * The current workaround is to mark the context as dirty when croaking,
419 * so that data structures can be freed whenever we renter Storable code
420 * (but only *then*: it's a workaround, not a fix).
421 *
422 * This is also imperfect, because we don't really know how far they trapped
423 * the croak(), and when we were recursing, we won't be able to clean anything
424 * but the topmost context stacked.
425 */
426
111e03c1 427#define CROAK(x) STMT_START { cxt->s_dirty = 1; croak x; } STMT_END
7a6a85bf
RG
428
429/*
430 * End of "thread-safe" related definitions.
431 */
432
433/*
9e21b3d0
JH
434 * LOW_32BITS
435 *
436 * Keep only the low 32 bits of a pointer (used for tags, which are not
437 * really pointers).
438 */
439
440#if PTRSIZE <= 4
441#define LOW_32BITS(x) ((I32) (x))
442#else
443#define LOW_32BITS(x) ((I32) ((unsigned long) (x) & 0xffffffffUL))
444#endif
445
446/*
447 * oI, oS, oC
448 *
449 * Hack for Crays, where sizeof(I32) == 8, and which are big-endians.
450 * Used in the WLEN and RLEN macros.
451 */
452
453#if INTSIZE > 4
454#define oI(x) ((I32 *) ((char *) (x) + 4))
455#define oS(x) ((x) - 4)
456#define oC(x) (x = 0)
457#define CRAY_HACK
458#else
459#define oI(x) (x)
460#define oS(x) (x)
461#define oC(x)
462#endif
463
464/*
7a6a85bf
RG
465 * key buffer handling
466 */
467#define kbuf (cxt->keybuf).arena
468#define ksiz (cxt->keybuf).asiz
111e03c1
RG
469#define KBUFINIT() \
470 STMT_START { \
7a6a85bf
RG
471 if (!kbuf) { \
472 TRACEME(("** allocating kbuf of 128 bytes")); \
473 New(10003, kbuf, 128, char); \
474 ksiz = 128; \
475 } \
111e03c1
RG
476 } STMT_END
477#define KBUFCHK(x) \
478 STMT_START { \
7a6a85bf 479 if (x >= ksiz) { \
e993d95c 480 TRACEME(("** extending kbuf to %d bytes (had %d)", x+1, ksiz)); \
7a6a85bf
RG
481 Renew(kbuf, x+1, char); \
482 ksiz = x+1; \
483 } \
111e03c1 484 } STMT_END
7a6a85bf
RG
485
486/*
487 * memory buffer handling
488 */
489#define mbase (cxt->membuf).arena
490#define msiz (cxt->membuf).asiz
491#define mptr (cxt->membuf).aptr
492#define mend (cxt->membuf).aend
493
494#define MGROW (1 << 13)
495#define MMASK (MGROW - 1)
496
497#define round_mgrow(x) \
498 ((unsigned long) (((unsigned long) (x) + MMASK) & ~MMASK))
499#define trunc_int(x) \
500 ((unsigned long) ((unsigned long) (x) & ~(sizeof(int)-1)))
501#define int_aligned(x) \
502 ((unsigned long) (x) == trunc_int(x))
503
111e03c1
RG
504#define MBUF_INIT(x) \
505 STMT_START { \
7a6a85bf
RG
506 if (!mbase) { \
507 TRACEME(("** allocating mbase of %d bytes", MGROW)); \
508 New(10003, mbase, MGROW, char); \
509 msiz = MGROW; \
510 } \
511 mptr = mbase; \
512 if (x) \
513 mend = mbase + x; \
514 else \
515 mend = mbase + msiz; \
111e03c1 516 } STMT_END
7a6a85bf
RG
517
518#define MBUF_TRUNC(x) mptr = mbase + x
519#define MBUF_SIZE() (mptr - mbase)
520
521/*
e993d95c
JH
522 * MBUF_SAVE_AND_LOAD
523 * MBUF_RESTORE
524 *
525 * Those macros are used in do_retrieve() to save the current memory
526 * buffer into cxt->msaved, before MBUF_LOAD() can be used to retrieve
527 * data from a string.
528 */
111e03c1
RG
529#define MBUF_SAVE_AND_LOAD(in) \
530 STMT_START { \
e993d95c
JH
531 ASSERT(!cxt->membuf_ro, ("mbase not already saved")); \
532 cxt->membuf_ro = 1; \
533 TRACEME(("saving mbuf")); \
534 StructCopy(&cxt->membuf, &cxt->msaved, struct extendable); \
535 MBUF_LOAD(in); \
111e03c1 536 } STMT_END
e993d95c 537
111e03c1
RG
538#define MBUF_RESTORE() \
539 STMT_START { \
e993d95c
JH
540 ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
541 cxt->membuf_ro = 0; \
542 TRACEME(("restoring mbuf")); \
543 StructCopy(&cxt->msaved, &cxt->membuf, struct extendable); \
111e03c1 544 } STMT_END
e993d95c
JH
545
546/*
7a6a85bf
RG
547 * Use SvPOKp(), because SvPOK() fails on tainted scalars.
548 * See store_scalar() for other usage of this workaround.
549 */
111e03c1
RG
550#define MBUF_LOAD(v) \
551 STMT_START { \
e993d95c 552 ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
7a6a85bf
RG
553 if (!SvPOKp(v)) \
554 CROAK(("Not a scalar string")); \
555 mptr = mbase = SvPV(v, msiz); \
556 mend = mbase + msiz; \
111e03c1 557 } STMT_END
7a6a85bf 558
111e03c1
RG
559#define MBUF_XTEND(x) \
560 STMT_START { \
7a6a85bf
RG
561 int nsz = (int) round_mgrow((x)+msiz); \
562 int offset = mptr - mbase; \
e993d95c
JH
563 ASSERT(!cxt->membuf_ro, ("mbase is not read-only")); \
564 TRACEME(("** extending mbase from %d to %d bytes (wants %d new)", \
565 msiz, nsz, (x))); \
7a6a85bf
RG
566 Renew(mbase, nsz, char); \
567 msiz = nsz; \
568 mptr = mbase + offset; \
569 mend = mbase + nsz; \
111e03c1 570 } STMT_END
7a6a85bf 571
111e03c1
RG
572#define MBUF_CHK(x) \
573 STMT_START { \
7a6a85bf
RG
574 if ((mptr + (x)) > mend) \
575 MBUF_XTEND(x); \
111e03c1 576 } STMT_END
7a6a85bf 577
111e03c1
RG
578#define MBUF_GETC(x) \
579 STMT_START { \
7a6a85bf
RG
580 if (mptr < mend) \
581 x = (int) (unsigned char) *mptr++; \
582 else \
583 return (SV *) 0; \
111e03c1 584 } STMT_END
7a6a85bf 585
9e21b3d0 586#ifdef CRAY_HACK
111e03c1
RG
587#define MBUF_GETINT(x) \
588 STMT_START { \
9e21b3d0
JH
589 oC(x); \
590 if ((mptr + 4) <= mend) { \
591 memcpy(oI(&x), mptr, 4); \
592 mptr += 4; \
593 } else \
594 return (SV *) 0; \
111e03c1 595 } STMT_END
9e21b3d0 596#else
111e03c1
RG
597#define MBUF_GETINT(x) \
598 STMT_START { \
7a6a85bf
RG
599 if ((mptr + sizeof(int)) <= mend) { \
600 if (int_aligned(mptr)) \
601 x = *(int *) mptr; \
602 else \
603 memcpy(&x, mptr, sizeof(int)); \
604 mptr += sizeof(int); \
605 } else \
606 return (SV *) 0; \
111e03c1 607 } STMT_END
9e21b3d0 608#endif
7a6a85bf 609
111e03c1
RG
610#define MBUF_READ(x,s) \
611 STMT_START { \
7a6a85bf
RG
612 if ((mptr + (s)) <= mend) { \
613 memcpy(x, mptr, s); \
614 mptr += s; \
615 } else \
616 return (SV *) 0; \
111e03c1 617 } STMT_END
7a6a85bf 618
111e03c1
RG
619#define MBUF_SAFEREAD(x,s,z) \
620 STMT_START { \
7a6a85bf
RG
621 if ((mptr + (s)) <= mend) { \
622 memcpy(x, mptr, s); \
623 mptr += s; \
624 } else { \
625 sv_free(z); \
626 return (SV *) 0; \
627 } \
111e03c1 628 } STMT_END
7a6a85bf 629
111e03c1
RG
630#define MBUF_PUTC(c) \
631 STMT_START { \
7a6a85bf
RG
632 if (mptr < mend) \
633 *mptr++ = (char) c; \
634 else { \
635 MBUF_XTEND(1); \
636 *mptr++ = (char) c; \
637 } \
111e03c1 638 } STMT_END
7a6a85bf 639
9e21b3d0 640#ifdef CRAY_HACK
111e03c1
RG
641#define MBUF_PUTINT(i) \
642 STMT_START { \
9e21b3d0
JH
643 MBUF_CHK(4); \
644 memcpy(mptr, oI(&i), 4); \
645 mptr += 4; \
111e03c1 646 } STMT_END
9e21b3d0 647#else
111e03c1
RG
648#define MBUF_PUTINT(i) \
649 STMT_START { \
7a6a85bf
RG
650 MBUF_CHK(sizeof(int)); \
651 if (int_aligned(mptr)) \
652 *(int *) mptr = i; \
653 else \
654 memcpy(mptr, &i, sizeof(int)); \
655 mptr += sizeof(int); \
111e03c1 656 } STMT_END
9e21b3d0 657#endif
7a6a85bf 658
111e03c1
RG
659#define MBUF_WRITE(x,s) \
660 STMT_START { \
7a6a85bf
RG
661 MBUF_CHK(s); \
662 memcpy(mptr, x, s); \
663 mptr += s; \
111e03c1 664 } STMT_END
7a6a85bf
RG
665
666/*
7a6a85bf
RG
667 * Possible return values for sv_type().
668 */
669
670#define svis_REF 0
671#define svis_SCALAR 1
672#define svis_ARRAY 2
673#define svis_HASH 3
674#define svis_TIED 4
675#define svis_TIED_ITEM 5
676#define svis_OTHER 6
677
678/*
679 * Flags for SX_HOOK.
680 */
681
682#define SHF_TYPE_MASK 0x03
683#define SHF_LARGE_CLASSLEN 0x04
684#define SHF_LARGE_STRLEN 0x08
685#define SHF_LARGE_LISTLEN 0x10
686#define SHF_IDX_CLASSNAME 0x20
687#define SHF_NEED_RECURSE 0x40
688#define SHF_HAS_LIST 0x80
689
690/*
b12202d0 691 * Types for SX_HOOK (last 2 bits in flags).
7a6a85bf
RG
692 */
693
694#define SHT_SCALAR 0
695#define SHT_ARRAY 1
696#define SHT_HASH 2
b12202d0
JH
697#define SHT_EXTRA 3 /* Read extra byte for type */
698
699/*
700 * The following are held in the "extra byte"...
701 */
702
703#define SHT_TSCALAR 4 /* 4 + 0 -- tied scalar */
704#define SHT_TARRAY 5 /* 4 + 1 -- tied array */
705#define SHT_THASH 6 /* 4 + 2 -- tied hash */
7a6a85bf
RG
706
707/*
e16e2ff8
NC
708 * per hash flags for flagged hashes
709 */
710
711#define SHV_RESTRICTED 0x01
712
713/*
714 * per key flags for flagged hashes
715 */
716
717#define SHV_K_UTF8 0x01
718#define SHV_K_WASUTF8 0x02
719#define SHV_K_LOCKED 0x04
720#define SHV_K_ISSV 0x08
721#define SHV_K_PLACEHOLDER 0x10
722
723/*
7a6a85bf
RG
724 * Before 0.6, the magic string was "perl-store" (binary version number 0).
725 *
726 * Since 0.6 introduced many binary incompatibilities, the magic string has
727 * been changed to "pst0" to allow an old image to be properly retrieved by
728 * a newer Storable, but ensure a newer image cannot be retrieved with an
729 * older version.
730 *
731 * At 0.7, objects are given the ability to serialize themselves, and the
732 * set of markers is extended, backward compatibility is not jeopardized,
733 * so the binary version number could have remained unchanged. To correctly
734 * spot errors if a file making use of 0.7-specific extensions is given to
735 * 0.6 for retrieval, the binary version was moved to "2". And I'm introducing
736 * a "minor" version, to better track this kind of evolution from now on.
737 *
738 */
739static char old_magicstr[] = "perl-store"; /* Magic number before 0.6 */
740static char magicstr[] = "pst0"; /* Used as a magic number */
741
530b72ba 742
e16e2ff8 743#define STORABLE_BIN_MAJOR 2 /* Binary major "version" */
530b72ba
NC
744#define STORABLE_BIN_MINOR 5 /* Binary minor "version" */
745
746/* If we aren't 5.7.3 or later, we won't be writing out files that use the
747 * new flagged hash introdued in 2.5, so put 2.4 in the binary header to
748 * maximise ease of interoperation with older Storables.
749 * Could we write 2.3s if we're on 5.005_03? NWC
750 */
e16e2ff8 751#if (PATCHLEVEL <= 6)
530b72ba 752#define STORABLE_BIN_WRITE_MINOR 4
e16e2ff8
NC
753#else
754/*
755 * As of perl 5.7.3, utf8 hash key is introduced.
756 * So this must change -- dankogai
757*/
530b72ba 758#define STORABLE_BIN_WRITE_MINOR 5
e16e2ff8 759#endif /* (PATCHLEVEL <= 6) */
7a6a85bf
RG
760
761/*
762 * Useful store shortcuts...
763 */
764
111e03c1
RG
765#define PUTMARK(x) \
766 STMT_START { \
7a6a85bf
RG
767 if (!cxt->fio) \
768 MBUF_PUTC(x); \
769 else if (PerlIO_putc(cxt->fio, x) == EOF) \
770 return -1; \
111e03c1 771 } STMT_END
7a6a85bf 772
111e03c1
RG
773#define WRITE_I32(x) \
774 STMT_START { \
9e21b3d0
JH
775 ASSERT(sizeof(x) == sizeof(I32), ("writing an I32")); \
776 if (!cxt->fio) \
777 MBUF_PUTINT(x); \
778 else if (PerlIO_write(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
779 return -1; \
111e03c1 780 } STMT_END
9e21b3d0 781
7a6a85bf 782#ifdef HAS_HTONL
111e03c1
RG
783#define WLEN(x) \
784 STMT_START { \
7a6a85bf
RG
785 if (cxt->netorder) { \
786 int y = (int) htonl(x); \
787 if (!cxt->fio) \
788 MBUF_PUTINT(y); \
9e21b3d0 789 else if (PerlIO_write(cxt->fio,oI(&y),oS(sizeof(y))) != oS(sizeof(y))) \
7a6a85bf
RG
790 return -1; \
791 } else { \
792 if (!cxt->fio) \
793 MBUF_PUTINT(x); \
9e21b3d0 794 else if (PerlIO_write(cxt->fio,oI(&x),oS(sizeof(x))) != oS(sizeof(x))) \
7a6a85bf
RG
795 return -1; \
796 } \
111e03c1 797 } STMT_END
7a6a85bf 798#else
9e21b3d0 799#define WLEN(x) WRITE_I32(x)
7a6a85bf
RG
800#endif
801
111e03c1
RG
802#define WRITE(x,y) \
803 STMT_START { \
7a6a85bf
RG
804 if (!cxt->fio) \
805 MBUF_WRITE(x,y); \
806 else if (PerlIO_write(cxt->fio, x, y) != y) \
807 return -1; \
111e03c1 808 } STMT_END
7a6a85bf 809
111e03c1
RG
810#define STORE_PV_LEN(pv, len, small, large) \
811 STMT_START { \
7a6a85bf
RG
812 if (len <= LG_SCALAR) { \
813 unsigned char clen = (unsigned char) len; \
dd19458b 814 PUTMARK(small); \
7a6a85bf
RG
815 PUTMARK(clen); \
816 if (len) \
817 WRITE(pv, len); \
818 } else { \
dd19458b 819 PUTMARK(large); \
7a6a85bf
RG
820 WLEN(len); \
821 WRITE(pv, len); \
822 } \
111e03c1 823 } STMT_END
7a6a85bf 824
dd19458b
JH
825#define STORE_SCALAR(pv, len) STORE_PV_LEN(pv, len, SX_SCALAR, SX_LSCALAR)
826
827/*
7a6a85bf
RG
828 * Store undef in arrays and hashes without recursing through store().
829 */
111e03c1
RG
830#define STORE_UNDEF() \
831 STMT_START { \
7a6a85bf
RG
832 cxt->tagnum++; \
833 PUTMARK(SX_UNDEF); \
111e03c1 834 } STMT_END
7a6a85bf
RG
835
836/*
837 * Useful retrieve shortcuts...
838 */
839
840#define GETCHAR() \
841 (cxt->fio ? PerlIO_getc(cxt->fio) : (mptr >= mend ? EOF : (int) *mptr++))
842
111e03c1
RG
843#define GETMARK(x) \
844 STMT_START { \
7a6a85bf
RG
845 if (!cxt->fio) \
846 MBUF_GETC(x); \
76df4757 847 else if ((int) (x = PerlIO_getc(cxt->fio)) == EOF) \
7a6a85bf 848 return (SV *) 0; \
111e03c1 849 } STMT_END
7a6a85bf 850
111e03c1
RG
851#define READ_I32(x) \
852 STMT_START { \
9e21b3d0
JH
853 ASSERT(sizeof(x) == sizeof(I32), ("reading an I32")); \
854 oC(x); \
7a6a85bf
RG
855 if (!cxt->fio) \
856 MBUF_GETINT(x); \
9e21b3d0 857 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
7a6a85bf 858 return (SV *) 0; \
111e03c1 859 } STMT_END
9e21b3d0
JH
860
861#ifdef HAS_NTOHL
111e03c1
RG
862#define RLEN(x) \
863 STMT_START { \
9e21b3d0 864 oC(x); \
7a6a85bf
RG
865 if (!cxt->fio) \
866 MBUF_GETINT(x); \
9e21b3d0 867 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
7a6a85bf 868 return (SV *) 0; \
9e21b3d0
JH
869 if (cxt->netorder) \
870 x = (int) ntohl(x); \
111e03c1 871 } STMT_END
9e21b3d0
JH
872#else
873#define RLEN(x) READ_I32(x)
7a6a85bf
RG
874#endif
875
111e03c1
RG
876#define READ(x,y) \
877 STMT_START { \
7a6a85bf
RG
878 if (!cxt->fio) \
879 MBUF_READ(x, y); \
880 else if (PerlIO_read(cxt->fio, x, y) != y) \
881 return (SV *) 0; \
111e03c1 882 } STMT_END
7a6a85bf 883
111e03c1
RG
884#define SAFEREAD(x,y,z) \
885 STMT_START { \
7a6a85bf
RG
886 if (!cxt->fio) \
887 MBUF_SAFEREAD(x,y,z); \
888 else if (PerlIO_read(cxt->fio, x, y) != y) { \
889 sv_free(z); \
890 return (SV *) 0; \
891 } \
111e03c1 892 } STMT_END
7a6a85bf
RG
893
894/*
895 * This macro is used at retrieve time, to remember where object 'y', bearing a
896 * given tag 'tagnum', has been retrieved. Next time we see an SX_OBJECT marker,
897 * we'll therefore know where it has been retrieved and will be able to
898 * share the same reference, as in the original stored memory image.
b12202d0
JH
899 *
900 * We also need to bless objects ASAP for hooks (which may compute "ref $x"
901 * on the objects given to STORABLE_thaw and expect that to be defined), and
902 * also for overloaded objects (for which we might not find the stash if the
903 * object is not blessed yet--this might occur for overloaded objects that
904 * refer to themselves indirectly: if we blessed upon return from a sub
905 * retrieve(), the SX_OBJECT marker we'd found could not have overloading
906 * restored on it because the underlying object would not be blessed yet!).
907 *
908 * To achieve that, the class name of the last retrieved object is passed down
909 * recursively, and the first SEEN() call for which the class name is not NULL
910 * will bless the object.
7a6a85bf 911 */
111e03c1
RG
912#define SEEN(y,c) \
913 STMT_START { \
7a6a85bf
RG
914 if (!y) \
915 return (SV *) 0; \
916 if (av_store(cxt->aseen, cxt->tagnum++, SvREFCNT_inc(y)) == 0) \
917 return (SV *) 0; \
43d061fe 918 TRACEME(("aseen(#%d) = 0x%"UVxf" (refcnt=%d)", cxt->tagnum-1, \
b12202d0
JH
919 PTR2UV(y), SvREFCNT(y)-1)); \
920 if (c) \
921 BLESS((SV *) (y), c); \
111e03c1 922 } STMT_END
7a6a85bf
RG
923
924/*
925 * Bless `s' in `p', via a temporary reference, required by sv_bless().
926 */
111e03c1
RG
927#define BLESS(s,p) \
928 STMT_START { \
7a6a85bf
RG
929 SV *ref; \
930 HV *stash; \
43d061fe 931 TRACEME(("blessing 0x%"UVxf" in %s", PTR2UV(s), (p))); \
7a6a85bf
RG
932 stash = gv_stashpv((p), TRUE); \
933 ref = newRV_noinc(s); \
934 (void) sv_bless(ref, stash); \
935 SvRV(ref) = 0; \
936 SvREFCNT_dec(ref); \
111e03c1 937 } STMT_END
7a6a85bf
RG
938
939static int store();
b12202d0 940static SV *retrieve(stcxt_t *cxt, char *cname);
7a6a85bf
RG
941
942/*
943 * Dynamic dispatching table for SV store.
944 */
945
946static int store_ref(stcxt_t *cxt, SV *sv);
947static int store_scalar(stcxt_t *cxt, SV *sv);
948static int store_array(stcxt_t *cxt, AV *av);
949static int store_hash(stcxt_t *cxt, HV *hv);
950static int store_tied(stcxt_t *cxt, SV *sv);
951static int store_tied_item(stcxt_t *cxt, SV *sv);
952static int store_other(stcxt_t *cxt, SV *sv);
f0ffaed8 953static int store_blessed(stcxt_t *cxt, SV *sv, int type, HV *pkg);
7a6a85bf 954
862382c7
JH
955static int (*sv_store[])(stcxt_t *cxt, SV *sv) = {
956 store_ref, /* svis_REF */
957 store_scalar, /* svis_SCALAR */
958 (int (*)(stcxt_t *cxt, SV *sv)) store_array, /* svis_ARRAY */
959 (int (*)(stcxt_t *cxt, SV *sv)) store_hash, /* svis_HASH */
960 store_tied, /* svis_TIED */
961 store_tied_item, /* svis_TIED_ITEM */
962 store_other, /* svis_OTHER */
7a6a85bf
RG
963};
964
965#define SV_STORE(x) (*sv_store[x])
966
967/*
968 * Dynamic dispatching tables for SV retrieval.
969 */
970
b12202d0
JH
971static SV *retrieve_lscalar(stcxt_t *cxt, char *cname);
972static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname);
973static SV *old_retrieve_array(stcxt_t *cxt, char *cname);
974static SV *old_retrieve_hash(stcxt_t *cxt, char *cname);
975static SV *retrieve_ref(stcxt_t *cxt, char *cname);
976static SV *retrieve_undef(stcxt_t *cxt, char *cname);
977static SV *retrieve_integer(stcxt_t *cxt, char *cname);
978static SV *retrieve_double(stcxt_t *cxt, char *cname);
979static SV *retrieve_byte(stcxt_t *cxt, char *cname);
980static SV *retrieve_netint(stcxt_t *cxt, char *cname);
981static SV *retrieve_scalar(stcxt_t *cxt, char *cname);
982static SV *retrieve_utf8str(stcxt_t *cxt, char *cname);
983static SV *retrieve_tied_array(stcxt_t *cxt, char *cname);
984static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname);
985static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname);
986static SV *retrieve_other(stcxt_t *cxt, char *cname);
987
988static SV *(*sv_old_retrieve[])(stcxt_t *cxt, char *cname) = {
7a6a85bf
RG
989 0, /* SX_OBJECT -- entry unused dynamically */
990 retrieve_lscalar, /* SX_LSCALAR */
991 old_retrieve_array, /* SX_ARRAY -- for pre-0.6 binaries */
992 old_retrieve_hash, /* SX_HASH -- for pre-0.6 binaries */
993 retrieve_ref, /* SX_REF */
994 retrieve_undef, /* SX_UNDEF */
995 retrieve_integer, /* SX_INTEGER */
996 retrieve_double, /* SX_DOUBLE */
997 retrieve_byte, /* SX_BYTE */
998 retrieve_netint, /* SX_NETINT */
999 retrieve_scalar, /* SX_SCALAR */
1000 retrieve_tied_array, /* SX_ARRAY */
1001 retrieve_tied_hash, /* SX_HASH */
1002 retrieve_tied_scalar, /* SX_SCALAR */
1003 retrieve_other, /* SX_SV_UNDEF not supported */
1004 retrieve_other, /* SX_SV_YES not supported */
1005 retrieve_other, /* SX_SV_NO not supported */
1006 retrieve_other, /* SX_BLESS not supported */
1007 retrieve_other, /* SX_IX_BLESS not supported */
1008 retrieve_other, /* SX_HOOK not supported */
1009 retrieve_other, /* SX_OVERLOADED not supported */
1010 retrieve_other, /* SX_TIED_KEY not supported */
1011 retrieve_other, /* SX_TIED_IDX not supported */
dd19458b
JH
1012 retrieve_other, /* SX_UTF8STR not supported */
1013 retrieve_other, /* SX_LUTF8STR not supported */
e16e2ff8 1014 retrieve_other, /* SX_FLAG_HASH not supported */
7a6a85bf
RG
1015 retrieve_other, /* SX_ERROR */
1016};
1017
b12202d0
JH
1018static SV *retrieve_array(stcxt_t *cxt, char *cname);
1019static SV *retrieve_hash(stcxt_t *cxt, char *cname);
1020static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname);
1021static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname);
1022static SV *retrieve_sv_no(stcxt_t *cxt, char *cname);
1023static SV *retrieve_blessed(stcxt_t *cxt, char *cname);
1024static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname);
1025static SV *retrieve_hook(stcxt_t *cxt, char *cname);
1026static SV *retrieve_overloaded(stcxt_t *cxt, char *cname);
1027static SV *retrieve_tied_key(stcxt_t *cxt, char *cname);
1028static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname);
e16e2ff8 1029static SV *retrieve_flag_hash(stcxt_t *cxt, char *cname);
b12202d0
JH
1030
1031static SV *(*sv_retrieve[])(stcxt_t *cxt, char *cname) = {
7a6a85bf
RG
1032 0, /* SX_OBJECT -- entry unused dynamically */
1033 retrieve_lscalar, /* SX_LSCALAR */
1034 retrieve_array, /* SX_ARRAY */
1035 retrieve_hash, /* SX_HASH */
1036 retrieve_ref, /* SX_REF */
1037 retrieve_undef, /* SX_UNDEF */
1038 retrieve_integer, /* SX_INTEGER */
1039 retrieve_double, /* SX_DOUBLE */
1040 retrieve_byte, /* SX_BYTE */
1041 retrieve_netint, /* SX_NETINT */
1042 retrieve_scalar, /* SX_SCALAR */
1043 retrieve_tied_array, /* SX_ARRAY */
1044 retrieve_tied_hash, /* SX_HASH */
1045 retrieve_tied_scalar, /* SX_SCALAR */
1046 retrieve_sv_undef, /* SX_SV_UNDEF */
1047 retrieve_sv_yes, /* SX_SV_YES */
1048 retrieve_sv_no, /* SX_SV_NO */
1049 retrieve_blessed, /* SX_BLESS */
1050 retrieve_idx_blessed, /* SX_IX_BLESS */
1051 retrieve_hook, /* SX_HOOK */
1052 retrieve_overloaded, /* SX_OVERLOAD */
1053 retrieve_tied_key, /* SX_TIED_KEY */
1054 retrieve_tied_idx, /* SX_TIED_IDX */
dd19458b
JH
1055 retrieve_utf8str, /* SX_UTF8STR */
1056 retrieve_lutf8str, /* SX_LUTF8STR */
e16e2ff8 1057 retrieve_flag_hash, /* SX_HASH */
7a6a85bf
RG
1058 retrieve_other, /* SX_ERROR */
1059};
1060
1061#define RETRIEVE(c,x) (*(c)->retrieve_vtbl[(x) >= SX_ERROR ? SX_ERROR : (x)])
1062
f0ffaed8 1063static SV *mbuf2sv(void);
7a6a85bf
RG
1064
1065/***
1066 *** Context management.
1067 ***/
1068
1069/*
1070 * init_perinterp
1071 *
1072 * Called once per "thread" (interpreter) to initialize some global context.
1073 */
f0ffaed8
JH
1074static void init_perinterp(void)
1075{
7a6a85bf
RG
1076 INIT_STCXT;
1077
1078 cxt->netorder = 0; /* true if network order used */
1079 cxt->forgive_me = -1; /* whether to be forgiving... */
1080}
1081
1082/*
e993d95c
JH
1083 * reset_context
1084 *
1085 * Called at the end of every context cleaning, to perform common reset
1086 * operations.
1087 */
1088static void reset_context(stcxt_t *cxt)
1089{
1090 cxt->entry = 0;
1091 cxt->s_dirty = 0;
1092 cxt->optype &= ~(ST_STORE|ST_RETRIEVE); /* Leave ST_CLONE alone */
1093}
1094
1095/*
7a6a85bf
RG
1096 * init_store_context
1097 *
1098 * Initialize a new store context for real recursion.
1099 */
f0ffaed8
JH
1100static void init_store_context(
1101 stcxt_t *cxt,
1102 PerlIO *f,
1103 int optype,
1104 int network_order)
7a6a85bf
RG
1105{
1106 TRACEME(("init_store_context"));
1107
1108 cxt->netorder = network_order;
1109 cxt->forgive_me = -1; /* Fetched from perl if needed */
1110 cxt->canonical = -1; /* Idem */
1111 cxt->tagnum = -1; /* Reset tag numbers */
1112 cxt->classnum = -1; /* Reset class numbers */
1113 cxt->fio = f; /* Where I/O are performed */
1114 cxt->optype = optype; /* A store, or a deep clone */
1115 cxt->entry = 1; /* No recursion yet */
1116
1117 /*
1118 * The `hseen' table is used to keep track of each SV stored and their
1119 * associated tag numbers is special. It is "abused" because the
1120 * values stored are not real SV, just integers cast to (SV *),
1121 * which explains the freeing below.
1122 *
1123 * It is also one possible bottlneck to achieve good storing speed,
1124 * so the "shared keys" optimization is turned off (unlikely to be
1125 * of any use here), and the hash table is "pre-extended". Together,
1126 * those optimizations increase the throughput by 12%.
1127 */
1128
1129 cxt->hseen = newHV(); /* Table where seen objects are stored */
1130 HvSHAREKEYS_off(cxt->hseen);
1131
1132 /*
1133 * The following does not work well with perl5.004_04, and causes
1134 * a core dump later on, in a completely unrelated spot, which
1135 * makes me think there is a memory corruption going on.
1136 *
1137 * Calling hv_ksplit(hseen, HBUCKETS) instead of manually hacking
1138 * it below does not make any difference. It seems to work fine
1139 * with perl5.004_68 but given the probable nature of the bug,
1140 * that does not prove anything.
1141 *
1142 * It's a shame because increasing the amount of buckets raises
1143 * store() throughput by 5%, but until I figure this out, I can't
1144 * allow for this to go into production.
1145 *
1146 * It is reported fixed in 5.005, hence the #if.
1147 */
f0ffaed8 1148#if PERL_VERSION >= 5
7a6a85bf
RG
1149#define HBUCKETS 4096 /* Buckets for %hseen */
1150 HvMAX(cxt->hseen) = HBUCKETS - 1; /* keys %hseen = $HBUCKETS; */
1151#endif
1152
1153 /*
1154 * The `hclass' hash uses the same settings as `hseen' above, but it is
1155 * used to assign sequential tags (numbers) to class names for blessed
1156 * objects.
1157 *
1158 * We turn the shared key optimization on.
1159 */
1160
1161 cxt->hclass = newHV(); /* Where seen classnames are stored */
1162
f0ffaed8 1163#if PERL_VERSION >= 5
7a6a85bf
RG
1164 HvMAX(cxt->hclass) = HBUCKETS - 1; /* keys %hclass = $HBUCKETS; */
1165#endif
1166
1167 /*
1168 * The `hook' hash table is used to keep track of the references on
1169 * the STORABLE_freeze hook routines, when found in some class name.
1170 *
1171 * It is assumed that the inheritance tree will not be changed during
1172 * storing, and that no new method will be dynamically created by the
1173 * hooks.
1174 */
1175
1176 cxt->hook = newHV(); /* Table where hooks are cached */
90826881
JH
1177
1178 /*
1179 * The `hook_seen' array keeps track of all the SVs returned by
1180 * STORABLE_freeze hooks for us to serialize, so that they are not
1181 * reclaimed until the end of the serialization process. Each SV is
1182 * only stored once, the first time it is seen.
1183 */
1184
1185 cxt->hook_seen = newAV(); /* Lists SVs returned by STORABLE_freeze */
7a6a85bf
RG
1186}
1187
1188/*
1189 * clean_store_context
1190 *
1191 * Clean store context by
1192 */
f0ffaed8 1193static void clean_store_context(stcxt_t *cxt)
7a6a85bf
RG
1194{
1195 HE *he;
1196
1197 TRACEME(("clean_store_context"));
1198
1199 ASSERT(cxt->optype & ST_STORE, ("was performing a store()"));
1200
1201 /*
1202 * Insert real values into hashes where we stored faked pointers.
1203 */
1204
e993d95c
JH
1205 if (cxt->hseen) {
1206 hv_iterinit(cxt->hseen);
1207 while ((he = hv_iternext(cxt->hseen))) /* Extra () for -Wall, grr.. */
1208 HeVAL(he) = &PL_sv_undef;
1209 }
7a6a85bf 1210
e993d95c
JH
1211 if (cxt->hclass) {
1212 hv_iterinit(cxt->hclass);
1213 while ((he = hv_iternext(cxt->hclass))) /* Extra () for -Wall, grr.. */
1214 HeVAL(he) = &PL_sv_undef;
1215 }
7a6a85bf
RG
1216
1217 /*
1218 * And now dispose of them...
862382c7
JH
1219 *
1220 * The surrounding if() protection has been added because there might be
1221 * some cases where this routine is called more than once, during
1222 * exceptionnal events. This was reported by Marc Lehmann when Storable
1223 * is executed from mod_perl, and the fix was suggested by him.
1224 * -- RAM, 20/12/2000
1225 */
1226
1227 if (cxt->hseen) {
1228 HV *hseen = cxt->hseen;
1229 cxt->hseen = 0;
1230 hv_undef(hseen);
1231 sv_free((SV *) hseen);
1232 }
7a6a85bf 1233
862382c7
JH
1234 if (cxt->hclass) {
1235 HV *hclass = cxt->hclass;
1236 cxt->hclass = 0;
1237 hv_undef(hclass);
1238 sv_free((SV *) hclass);
1239 }
7a6a85bf 1240
862382c7
JH
1241 if (cxt->hook) {
1242 HV *hook = cxt->hook;
1243 cxt->hook = 0;
1244 hv_undef(hook);
1245 sv_free((SV *) hook);
1246 }
7a6a85bf 1247
862382c7
JH
1248 if (cxt->hook_seen) {
1249 AV *hook_seen = cxt->hook_seen;
1250 cxt->hook_seen = 0;
1251 av_undef(hook_seen);
1252 sv_free((SV *) hook_seen);
1253 }
90826881 1254
e8189732
NC
1255 cxt->forgive_me = -1; /* Fetched from perl if needed */
1256 cxt->canonical = -1; /* Idem */
1257
e993d95c 1258 reset_context(cxt);
7a6a85bf
RG
1259}
1260
1261/*
1262 * init_retrieve_context
1263 *
1264 * Initialize a new retrieve context for real recursion.
1265 */
dd19458b 1266static void init_retrieve_context(stcxt_t *cxt, int optype, int is_tainted)
7a6a85bf
RG
1267{
1268 TRACEME(("init_retrieve_context"));
1269
1270 /*
1271 * The hook hash table is used to keep track of the references on
1272 * the STORABLE_thaw hook routines, when found in some class name.
1273 *
1274 * It is assumed that the inheritance tree will not be changed during
1275 * storing, and that no new method will be dynamically created by the
1276 * hooks.
1277 */
1278
1279 cxt->hook = newHV(); /* Caches STORABLE_thaw */
1280
1281 /*
1282 * If retrieving an old binary version, the cxt->retrieve_vtbl variable
1283 * was set to sv_old_retrieve. We'll need a hash table to keep track of
1284 * the correspondance between the tags and the tag number used by the
1285 * new retrieve routines.
1286 */
1287
e16e2ff8 1288 cxt->hseen = ((cxt->retrieve_vtbl == sv_old_retrieve) ? newHV() : 0);
7a6a85bf
RG
1289
1290 cxt->aseen = newAV(); /* Where retrieved objects are kept */
1291 cxt->aclass = newAV(); /* Where seen classnames are kept */
1292 cxt->tagnum = 0; /* Have to count objects... */
1293 cxt->classnum = 0; /* ...and class names as well */
1294 cxt->optype = optype;
dd19458b 1295 cxt->s_tainted = is_tainted;
7a6a85bf 1296 cxt->entry = 1; /* No recursion yet */
530b72ba
NC
1297#ifndef HAS_RESTRICTED_HASHES
1298 cxt->derestrict = -1; /* Fetched from perl if needed */
1299#endif
1300#ifndef HAS_UTF8_ALL
1301 cxt->use_bytes = -1; /* Fetched from perl if needed */
1302#endif
e8189732 1303 cxt->accept_future_minor = -1; /* Fetched from perl if needed */
7a6a85bf
RG
1304}
1305
1306/*
1307 * clean_retrieve_context
1308 *
1309 * Clean retrieve context by
1310 */
dd19458b 1311static void clean_retrieve_context(stcxt_t *cxt)
7a6a85bf
RG
1312{
1313 TRACEME(("clean_retrieve_context"));
1314
1315 ASSERT(cxt->optype & ST_RETRIEVE, ("was performing a retrieve()"));
1316
862382c7
JH
1317 if (cxt->aseen) {
1318 AV *aseen = cxt->aseen;
1319 cxt->aseen = 0;
1320 av_undef(aseen);
1321 sv_free((SV *) aseen);
1322 }
7a6a85bf 1323
862382c7
JH
1324 if (cxt->aclass) {
1325 AV *aclass = cxt->aclass;
1326 cxt->aclass = 0;
1327 av_undef(aclass);
1328 sv_free((SV *) aclass);
1329 }
7a6a85bf 1330
862382c7
JH
1331 if (cxt->hook) {
1332 HV *hook = cxt->hook;
1333 cxt->hook = 0;
1334 hv_undef(hook);
1335 sv_free((SV *) hook);
1336 }
7a6a85bf 1337
862382c7
JH
1338 if (cxt->hseen) {
1339 HV *hseen = cxt->hseen;
1340 cxt->hseen = 0;
1341 hv_undef(hseen);
1342 sv_free((SV *) hseen); /* optional HV, for backward compat. */
1343 }
7a6a85bf 1344
e8189732
NC
1345#ifndef HAS_RESTRICTED_HASHES
1346 cxt->derestrict = -1; /* Fetched from perl if needed */
1347#endif
1348#ifndef HAS_UTF8_ALL
1349 cxt->use_bytes = -1; /* Fetched from perl if needed */
1350#endif
1351 cxt->accept_future_minor = -1; /* Fetched from perl if needed */
1352
e993d95c 1353 reset_context(cxt);
7a6a85bf
RG
1354}
1355
1356/*
1357 * clean_context
1358 *
1359 * A workaround for the CROAK bug: cleanup the last context.
1360 */
e993d95c 1361static void clean_context(stcxt_t *cxt)
7a6a85bf
RG
1362{
1363 TRACEME(("clean_context"));
1364
dd19458b 1365 ASSERT(cxt->s_dirty, ("dirty context"));
7a6a85bf 1366
e993d95c
JH
1367 if (cxt->membuf_ro)
1368 MBUF_RESTORE();
1369
1370 ASSERT(!cxt->membuf_ro, ("mbase is not read-only"));
1371
7a6a85bf
RG
1372 if (cxt->optype & ST_RETRIEVE)
1373 clean_retrieve_context(cxt);
e993d95c 1374 else if (cxt->optype & ST_STORE)
7a6a85bf 1375 clean_store_context(cxt);
e993d95c
JH
1376 else
1377 reset_context(cxt);
862382c7
JH
1378
1379 ASSERT(!cxt->s_dirty, ("context is clean"));
e993d95c 1380 ASSERT(cxt->entry == 0, ("context is reset"));
7a6a85bf
RG
1381}
1382
1383/*
1384 * allocate_context
1385 *
1386 * Allocate a new context and push it on top of the parent one.
1387 * This new context is made globally visible via SET_STCXT().
1388 */
1389static stcxt_t *allocate_context(parent_cxt)
1390stcxt_t *parent_cxt;
1391{
1392 stcxt_t *cxt;
1393
1394 TRACEME(("allocate_context"));
1395
dd19458b 1396 ASSERT(!parent_cxt->s_dirty, ("parent context clean"));
7a6a85bf 1397
111e03c1
RG
1398 NEW_STORABLE_CXT_OBJ(cxt);
1399 cxt->prev = parent_cxt->my_sv;
7a6a85bf
RG
1400 SET_STCXT(cxt);
1401
e993d95c
JH
1402 ASSERT(!cxt->s_dirty, ("clean context"));
1403
7a6a85bf
RG
1404 return cxt;
1405}
1406
1407/*
1408 * free_context
1409 *
1410 * Free current context, which cannot be the "root" one.
1411 * Make the context underneath globally visible via SET_STCXT().
1412 */
1413static void free_context(cxt)
1414stcxt_t *cxt;
1415{
111e03c1 1416 stcxt_t *prev = (stcxt_t *)(cxt->prev ? SvPVX(SvRV(cxt->prev)) : 0);
7a6a85bf
RG
1417
1418 TRACEME(("free_context"));
1419
dd19458b 1420 ASSERT(!cxt->s_dirty, ("clean context"));
7a6a85bf
RG
1421 ASSERT(prev, ("not freeing root context"));
1422
111e03c1 1423 SvREFCNT_dec(cxt->my_sv);
7a6a85bf 1424 SET_STCXT(prev);
e993d95c
JH
1425
1426 ASSERT(cxt, ("context not void"));
7a6a85bf
RG
1427}
1428
1429/***
1430 *** Predicates.
1431 ***/
1432
1433/*
1434 * is_storing
1435 *
1436 * Tells whether we're in the middle of a store operation.
1437 */
f0ffaed8 1438int is_storing(void)
7a6a85bf
RG
1439{
1440 dSTCXT;
1441
1442 return cxt->entry && (cxt->optype & ST_STORE);
1443}
1444
1445/*
1446 * is_retrieving
1447 *
1448 * Tells whether we're in the middle of a retrieve operation.
1449 */
f0ffaed8 1450int is_retrieving(void)
7a6a85bf
RG
1451{
1452 dSTCXT;
1453
1454 return cxt->entry && (cxt->optype & ST_RETRIEVE);
1455}
1456
1457/*
1458 * last_op_in_netorder
1459 *
1460 * Returns whether last operation was made using network order.
1461 *
1462 * This is typically out-of-band information that might prove useful
1463 * to people wishing to convert native to network order data when used.
1464 */
f0ffaed8 1465int last_op_in_netorder(void)
7a6a85bf
RG
1466{
1467 dSTCXT;
1468
1469 return cxt->netorder;
1470}
1471
1472/***
1473 *** Hook lookup and calling routines.
1474 ***/
1475
1476/*
1477 * pkg_fetchmeth
1478 *
1479 * A wrapper on gv_fetchmethod_autoload() which caches results.
1480 *
1481 * Returns the routine reference as an SV*, or null if neither the package
1482 * nor its ancestors know about the method.
1483 */
f0ffaed8
JH
1484static SV *pkg_fetchmeth(
1485 HV *cache,
1486 HV *pkg,
1487 char *method)
7a6a85bf
RG
1488{
1489 GV *gv;
1490 SV *sv;
7a6a85bf
RG
1491
1492 /*
1493 * The following code is the same as the one performed by UNIVERSAL::can
1494 * in the Perl core.
1495 */
1496
1497 gv = gv_fetchmethod_autoload(pkg, method, FALSE);
1498 if (gv && isGV(gv)) {
1499 sv = newRV((SV*) GvCV(gv));
9e21b3d0 1500 TRACEME(("%s->%s: 0x%"UVxf, HvNAME(pkg), method, PTR2UV(sv)));
7a6a85bf
RG
1501 } else {
1502 sv = newSVsv(&PL_sv_undef);
1503 TRACEME(("%s->%s: not found", HvNAME(pkg), method));
1504 }
1505
1506 /*
1507 * Cache the result, ignoring failure: if we can't store the value,
1508 * it just won't be cached.
1509 */
1510
1511 (void) hv_store(cache, HvNAME(pkg), strlen(HvNAME(pkg)), sv, 0);
1512
1513 return SvOK(sv) ? sv : (SV *) 0;
1514}
1515
1516/*
1517 * pkg_hide
1518 *
1519 * Force cached value to be undef: hook ignored even if present.
1520 */
f0ffaed8
JH
1521static void pkg_hide(
1522 HV *cache,
1523 HV *pkg,
1524 char *method)
7a6a85bf
RG
1525{
1526 (void) hv_store(cache,
1527 HvNAME(pkg), strlen(HvNAME(pkg)), newSVsv(&PL_sv_undef), 0);
1528}
1529
1530/*
212e9bde
JH
1531 * pkg_uncache
1532 *
1533 * Discard cached value: a whole fetch loop will be retried at next lookup.
1534 */
1535static void pkg_uncache(
1536 HV *cache,
1537 HV *pkg,
1538 char *method)
1539{
1540 (void) hv_delete(cache, HvNAME(pkg), strlen(HvNAME(pkg)), G_DISCARD);
1541}
1542
1543/*
7a6a85bf
RG
1544 * pkg_can
1545 *
1546 * Our own "UNIVERSAL::can", which caches results.
1547 *
1548 * Returns the routine reference as an SV*, or null if the object does not
1549 * know about the method.
1550 */
f0ffaed8
JH
1551static SV *pkg_can(
1552 HV *cache,
1553 HV *pkg,
1554 char *method)
7a6a85bf
RG
1555{
1556 SV **svh;
1557 SV *sv;
1558
1559 TRACEME(("pkg_can for %s->%s", HvNAME(pkg), method));
1560
1561 /*
1562 * Look into the cache to see whether we already have determined
1563 * where the routine was, if any.
1564 *
1565 * NOTA BENE: we don't use `method' at all in our lookup, since we know
1566 * that only one hook (i.e. always the same) is cached in a given cache.
1567 */
1568
1569 svh = hv_fetch(cache, HvNAME(pkg), strlen(HvNAME(pkg)), FALSE);
1570 if (svh) {
1571 sv = *svh;
1572 if (!SvOK(sv)) {
1573 TRACEME(("cached %s->%s: not found", HvNAME(pkg), method));
1574 return (SV *) 0;
1575 } else {
43d061fe 1576 TRACEME(("cached %s->%s: 0x%"UVxf,
9e21b3d0 1577 HvNAME(pkg), method, PTR2UV(sv)));
7a6a85bf
RG
1578 return sv;
1579 }
1580 }
1581
1582 TRACEME(("not cached yet"));
1583 return pkg_fetchmeth(cache, pkg, method); /* Fetch and cache */
1584}
1585
1586/*
1587 * scalar_call
1588 *
1589 * Call routine as obj->hook(av) in scalar context.
1590 * Propagates the single returned value if not called in void context.
1591 */
f0ffaed8
JH
1592static SV *scalar_call(
1593 SV *obj,
1594 SV *hook,
1595 int cloning,
1596 AV *av,
1597 I32 flags)
7a6a85bf
RG
1598{
1599 dSP;
1600 int count;
1601 SV *sv = 0;
1602
1603 TRACEME(("scalar_call (cloning=%d)", cloning));
1604
1605 ENTER;
1606 SAVETMPS;
1607
1608 PUSHMARK(sp);
1609 XPUSHs(obj);
1610 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1611 if (av) {
1612 SV **ary = AvARRAY(av);
1613 int cnt = AvFILLp(av) + 1;
1614 int i;
1615 XPUSHs(ary[0]); /* Frozen string */
1616 for (i = 1; i < cnt; i++) {
43d061fe
JH
1617 TRACEME(("pushing arg #%d (0x%"UVxf")...",
1618 i, PTR2UV(ary[i])));
7a6a85bf
RG
1619 XPUSHs(sv_2mortal(newRV(ary[i])));
1620 }
1621 }
1622 PUTBACK;
1623
1624 TRACEME(("calling..."));
1625 count = perl_call_sv(hook, flags); /* Go back to Perl code */
1626 TRACEME(("count = %d", count));
1627
1628 SPAGAIN;
1629
1630 if (count) {
1631 sv = POPs;
1632 SvREFCNT_inc(sv); /* We're returning it, must stay alive! */
1633 }
1634
1635 PUTBACK;
1636 FREETMPS;
1637 LEAVE;
1638
1639 return sv;
1640}
1641
1642/*
1643 * array_call
1644 *
f9a1036d 1645 * Call routine obj->hook(cloning) in list context.
7a6a85bf
RG
1646 * Returns the list of returned values in an array.
1647 */
f0ffaed8
JH
1648static AV *array_call(
1649 SV *obj,
1650 SV *hook,
1651 int cloning)
7a6a85bf
RG
1652{
1653 dSP;
1654 int count;
1655 AV *av;
1656 int i;
1657
f0ffaed8 1658 TRACEME(("array_call (cloning=%d)", cloning));
7a6a85bf
RG
1659
1660 ENTER;
1661 SAVETMPS;
1662
1663 PUSHMARK(sp);
1664 XPUSHs(obj); /* Target object */
1665 XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */
1666 PUTBACK;
1667
1668 count = perl_call_sv(hook, G_ARRAY); /* Go back to Perl code */
1669
1670 SPAGAIN;
1671
1672 av = newAV();
1673 for (i = count - 1; i >= 0; i--) {
1674 SV *sv = POPs;
1675 av_store(av, i, SvREFCNT_inc(sv));
1676 }
1677
1678 PUTBACK;
1679 FREETMPS;
1680 LEAVE;
1681
1682 return av;
1683}
1684
1685/*
1686 * known_class
1687 *
1688 * Lookup the class name in the `hclass' table and either assign it a new ID
1689 * or return the existing one, by filling in `classnum'.
1690 *
1691 * Return true if the class was known, false if the ID was just generated.
1692 */
f0ffaed8
JH
1693static int known_class(
1694 stcxt_t *cxt,
1695 char *name, /* Class name */
1696 int len, /* Name length */
1697 I32 *classnum)
7a6a85bf
RG
1698{
1699 SV **svh;
1700 HV *hclass = cxt->hclass;
1701
1702 TRACEME(("known_class (%s)", name));
1703
1704 /*
1705 * Recall that we don't store pointers in this hash table, but tags.
1706 * Therefore, we need LOW_32BITS() to extract the relevant parts.
1707 */
1708
1709 svh = hv_fetch(hclass, name, len, FALSE);
1710 if (svh) {
1711 *classnum = LOW_32BITS(*svh);
1712 return TRUE;
1713 }
1714
1715 /*
1716 * Unknown classname, we need to record it.
7a6a85bf
RG
1717 */
1718
1719 cxt->classnum++;
3341c981 1720 if (!hv_store(hclass, name, len, INT2PTR(SV*, cxt->classnum), 0))
7a6a85bf
RG
1721 CROAK(("Unable to record new classname"));
1722
1723 *classnum = cxt->classnum;
1724 return FALSE;
1725}
1726
1727/***
1728 *** Sepcific store routines.
1729 ***/
1730
1731/*
1732 * store_ref
1733 *
1734 * Store a reference.
1735 * Layout is SX_REF <object> or SX_OVERLOAD <object>.
1736 */
f0ffaed8 1737static int store_ref(stcxt_t *cxt, SV *sv)
7a6a85bf 1738{
43d061fe 1739 TRACEME(("store_ref (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
1740
1741 /*
1742 * Follow reference, and check if target is overloaded.
1743 */
1744
1745 sv = SvRV(sv);
1746
1747 if (SvOBJECT(sv)) {
1748 HV *stash = (HV *) SvSTASH(sv);
1749 if (stash && Gv_AMG(stash)) {
9e21b3d0 1750 TRACEME(("ref (0x%"UVxf") is overloaded", PTR2UV(sv)));
7a6a85bf
RG
1751 PUTMARK(SX_OVERLOAD);
1752 } else
1753 PUTMARK(SX_REF);
1754 } else
1755 PUTMARK(SX_REF);
1756
1757 return store(cxt, sv);
1758}
1759
1760/*
1761 * store_scalar
1762 *
1763 * Store a scalar.
1764 *
e16e2ff8 1765 * Layout is SX_LSCALAR <length> <data>, SX_SCALAR <length> <data> or SX_UNDEF.
7a6a85bf
RG
1766 * The <data> section is omitted if <length> is 0.
1767 *
1768 * If integer or double, the layout is SX_INTEGER <data> or SX_DOUBLE <data>.
1769 * Small integers (within [-127, +127]) are stored as SX_BYTE <byte>.
1770 */
f0ffaed8 1771static int store_scalar(stcxt_t *cxt, SV *sv)
7a6a85bf
RG
1772{
1773 IV iv;
1774 char *pv;
1775 STRLEN len;
1776 U32 flags = SvFLAGS(sv); /* "cc -O" may put it in register */
1777
43d061fe 1778 TRACEME(("store_scalar (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
1779
1780 /*
1781 * For efficiency, break the SV encapsulation by peaking at the flags
1782 * directly without using the Perl macros to avoid dereferencing
1783 * sv->sv_flags each time we wish to check the flags.
1784 */
1785
1786 if (!(flags & SVf_OK)) { /* !SvOK(sv) */
1787 if (sv == &PL_sv_undef) {
1788 TRACEME(("immortal undef"));
1789 PUTMARK(SX_SV_UNDEF);
1790 } else {
86bbd6dc 1791 TRACEME(("undef at 0x%"UVxf, PTR2UV(sv)));
7a6a85bf
RG
1792 PUTMARK(SX_UNDEF);
1793 }
1794 return 0;
1795 }
1796
1797 /*
1798 * Always store the string representation of a scalar if it exists.
1799 * Gisle Aas provided me with this test case, better than a long speach:
1800 *
1801 * perl -MDevel::Peek -le '$a="abc"; $a+0; Dump($a)'
1802 * SV = PVNV(0x80c8520)
1803 * REFCNT = 1
1804 * FLAGS = (NOK,POK,pNOK,pPOK)
1805 * IV = 0
1806 * NV = 0
1807 * PV = 0x80c83d0 "abc"\0
1808 * CUR = 3
1809 * LEN = 4
1810 *
1811 * Write SX_SCALAR, length, followed by the actual data.
1812 *
1813 * Otherwise, write an SX_BYTE, SX_INTEGER or an SX_DOUBLE as
1814 * appropriate, followed by the actual (binary) data. A double
1815 * is written as a string if network order, for portability.
1816 *
1817 * NOTE: instead of using SvNOK(sv), we test for SvNOKp(sv).
1818 * The reason is that when the scalar value is tainted, the SvNOK(sv)
1819 * value is false.
1820 *
1821 * The test for a read-only scalar with both POK and NOK set is meant
1822 * to quickly detect &PL_sv_yes and &PL_sv_no without having to pay the
1823 * address comparison for each scalar we store.
1824 */
1825
1826#define SV_MAYBE_IMMORTAL (SVf_READONLY|SVf_POK|SVf_NOK)
1827
1828 if ((flags & SV_MAYBE_IMMORTAL) == SV_MAYBE_IMMORTAL) {
1829 if (sv == &PL_sv_yes) {
1830 TRACEME(("immortal yes"));
1831 PUTMARK(SX_SV_YES);
1832 } else if (sv == &PL_sv_no) {
1833 TRACEME(("immortal no"));
1834 PUTMARK(SX_SV_NO);
1835 } else {
1836 pv = SvPV(sv, len); /* We know it's SvPOK */
1837 goto string; /* Share code below */
1838 }
db670f21
NC
1839 } else if (flags & SVf_POK) {
1840 /* public string - go direct to string read. */
1841 goto string_readlen;
1842 } else if (
1843#if (PATCHLEVEL <= 6)
1844 /* For 5.6 and earlier NV flag trumps IV flag, so only use integer
1845 direct if NV flag is off. */
1846 (flags & (SVf_NOK | SVf_IOK)) == SVf_IOK
1847#else
1848 /* 5.7 rules are that if IV public flag is set, IV value is as
1849 good, if not better, than NV value. */
1850 flags & SVf_IOK
1851#endif
1852 ) {
1853 iv = SvIV(sv);
1854 /*
1855 * Will come here from below with iv set if double is an integer.
1856 */
1857 integer:
7a6a85bf 1858
db670f21
NC
1859 /* Sorry. This isn't in 5.005_56 (IIRC) or earlier. */
1860#ifdef SVf_IVisUV
1861 /* Need to do this out here, else 0xFFFFFFFF becomes iv of -1
1862 * (for example) and that ends up in the optimised small integer
1863 * case.
1864 */
1865 if ((flags & SVf_IVisUV) && SvUV(sv) > IV_MAX) {
1866 TRACEME(("large unsigned integer as string, value = %"UVuf, SvUV(sv)));
1867 goto string_readlen;
1868 }
1869#endif
1870 /*
1871 * Optimize small integers into a single byte, otherwise store as
1872 * a real integer (converted into network order if they asked).
1873 */
7a6a85bf 1874
db670f21
NC
1875 if (iv >= -128 && iv <= 127) {
1876 unsigned char siv = (unsigned char) (iv + 128); /* [0,255] */
1877 PUTMARK(SX_BYTE);
1878 PUTMARK(siv);
1879 TRACEME(("small integer stored as %d", siv));
1880 } else if (cxt->netorder) {
1881#ifndef HAS_HTONL
1882 TRACEME(("no htonl, fall back to string for integer"));
1883 goto string_readlen;
1884#else
1885 I32 niv;
7a6a85bf 1886
7a6a85bf 1887
db670f21
NC
1888#if IVSIZE > 4
1889 if (
1890#ifdef SVf_IVisUV
1891 /* Sorry. This isn't in 5.005_56 (IIRC) or earlier. */
1892 ((flags & SVf_IVisUV) && SvUV(sv) > 0x7FFFFFFF) ||
1893#endif
1894 (iv > 0x7FFFFFFF) || (iv < -0x80000000)) {
1895 /* Bigger than 32 bits. */
1896 TRACEME(("large network order integer as string, value = %"IVdf, iv));
1897 goto string_readlen;
1898 }
1899#endif
7a6a85bf 1900
db670f21
NC
1901 niv = (I32) htonl((I32) iv);
1902 TRACEME(("using network order"));
1903 PUTMARK(SX_NETINT);
1904 WRITE_I32(niv);
1905#endif
1906 } else {
1907 PUTMARK(SX_INTEGER);
1908 WRITE(&iv, sizeof(iv));
1909 }
1910
1911 TRACEME(("ok (integer 0x%"UVxf", value = %"IVdf")", PTR2UV(sv), iv));
1912 } else if (flags & SVf_NOK) {
1913 NV nv;
1914#if (PATCHLEVEL <= 6)
1915 nv = SvNV(sv);
1916 /*
1917 * Watch for number being an integer in disguise.
1918 */
1919 if (nv == (NV) (iv = I_V(nv))) {
1920 TRACEME(("double %"NVff" is actually integer %"IVdf, nv, iv));
1921 goto integer; /* Share code above */
1922 }
1923#else
7a6a85bf 1924
db670f21
NC
1925 SvIV_please(sv);
1926 if (SvIOK(sv)) {
1927 iv = SvIV(sv);
1928 goto integer; /* Share code above */
1929 }
1930 nv = SvNV(sv);
1931#endif
7a6a85bf 1932
db670f21
NC
1933 if (cxt->netorder) {
1934 TRACEME(("double %"NVff" stored as string", nv));
1935 goto string_readlen; /* Share code below */
1936 }
7a6a85bf 1937
db670f21
NC
1938 PUTMARK(SX_DOUBLE);
1939 WRITE(&nv, sizeof(nv));
7a6a85bf 1940
db670f21 1941 TRACEME(("ok (double 0x%"UVxf", value = %"NVff")", PTR2UV(sv), nv));
7a6a85bf 1942
db670f21
NC
1943 } else if (flags & (SVp_POK | SVp_NOK | SVp_IOK)) {
1944 I32 wlen; /* For 64-bit machines */
7a6a85bf 1945
db670f21
NC
1946 string_readlen:
1947 pv = SvPV(sv, len);
7a6a85bf 1948
db670f21
NC
1949 /*
1950 * Will come here from above if it was readonly, POK and NOK but
1951 * neither &PL_sv_yes nor &PL_sv_no.
1952 */
1953 string:
1954
1955 wlen = (I32) len; /* WLEN via STORE_SCALAR expects I32 */
1956 if (SvUTF8 (sv))
1957 STORE_UTF8STR(pv, wlen);
1958 else
1959 STORE_SCALAR(pv, wlen);
1960 TRACEME(("ok (scalar 0x%"UVxf" '%s', length = %"IVdf")",
1961 PTR2UV(sv), SvPVX(sv), (IV)len));
7a6a85bf 1962 } else
db670f21
NC
1963 CROAK(("Can't determine type of %s(0x%"UVxf")",
1964 sv_reftype(sv, FALSE),
1965 PTR2UV(sv)));
1966 return 0; /* Ok, no recursion on scalars */
7a6a85bf
RG
1967}
1968
1969/*
1970 * store_array
1971 *
1972 * Store an array.
1973 *
1974 * Layout is SX_ARRAY <size> followed by each item, in increading index order.
1975 * Each item is stored as <object>.
1976 */
f0ffaed8 1977static int store_array(stcxt_t *cxt, AV *av)
7a6a85bf
RG
1978{
1979 SV **sav;
1980 I32 len = av_len(av) + 1;
1981 I32 i;
1982 int ret;
1983
43d061fe 1984 TRACEME(("store_array (0x%"UVxf")", PTR2UV(av)));
7a6a85bf
RG
1985
1986 /*
1987 * Signal array by emitting SX_ARRAY, followed by the array length.
1988 */
1989
1990 PUTMARK(SX_ARRAY);
1991 WLEN(len);
1992 TRACEME(("size = %d", len));
1993
1994 /*
1995 * Now store each item recursively.
1996 */
1997
1998 for (i = 0; i < len; i++) {
1999 sav = av_fetch(av, i, 0);
2000 if (!sav) {
2001 TRACEME(("(#%d) undef item", i));
2002 STORE_UNDEF();
2003 continue;
2004 }
2005 TRACEME(("(#%d) item", i));
e993d95c 2006 if ((ret = store(cxt, *sav))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2007 return ret;
2008 }
2009
2010 TRACEME(("ok (array)"));
2011
2012 return 0;
2013}
2014
2015/*
2016 * sortcmp
2017 *
2018 * Sort two SVs
2019 * Borrowed from perl source file pp_ctl.c, where it is used by pp_sort.
2020 */
2021static int
f0ffaed8 2022sortcmp(const void *a, const void *b)
7a6a85bf
RG
2023{
2024 return sv_cmp(*(SV * const *) a, *(SV * const *) b);
2025}
2026
2027
2028/*
2029 * store_hash
2030 *
d1be9408 2031 * Store a hash table.
7a6a85bf 2032 *
e16e2ff8
NC
2033 * For a "normal" hash (not restricted, no utf8 keys):
2034 *
7a6a85bf
RG
2035 * Layout is SX_HASH <size> followed by each key/value pair, in random order.
2036 * Values are stored as <object>.
2037 * Keys are stored as <length> <data>, the <data> section being omitted
2038 * if length is 0.
c194a0a3
TB
2039 *
2040 * For a "fancy" hash (restricted or utf8 keys):
2041 *
2042 * Layout is SX_FLAG_HASH <size> <hash flags> followed by each key/value pair,
e16e2ff8
NC
2043 * in random order.
2044 * Values are stored as <object>.
2045 * Keys are stored as <flags> <length> <data>, the <data> section being omitted
2046 * if length is 0.
2047 * Currently the only hash flag is "restriced"
2048 * Key flags are as for hv.h
7a6a85bf 2049 */
f0ffaed8 2050static int store_hash(stcxt_t *cxt, HV *hv)
7a6a85bf 2051{
530b72ba
NC
2052 I32 len =
2053#ifdef HAS_RESTRICTED_HASHES
2054 HvTOTALKEYS(hv);
2055#else
2056 HvKEYS(hv);
2057#endif
7a6a85bf
RG
2058 I32 i;
2059 int ret = 0;
2060 I32 riter;
2061 HE *eiter;
530b72ba
NC
2062 int flagged_hash = ((SvREADONLY(hv)
2063#ifdef HAS_HASH_KEY_FLAGS
2064 || HvHASKFLAGS(hv)
2065#endif
2066 ) ? 1 : 0);
e16e2ff8 2067 unsigned char hash_flags = (SvREADONLY(hv) ? SHV_RESTRICTED : 0);
7a6a85bf 2068
e16e2ff8
NC
2069 if (flagged_hash) {
2070 /* needs int cast for C++ compilers, doesn't it? */
2071 TRACEME(("store_hash (0x%"UVxf") (flags %x)", PTR2UV(hv),
2072 (int) hash_flags));
2073 } else {
2074 TRACEME(("store_hash (0x%"UVxf")", PTR2UV(hv)));
2075 }
7a6a85bf
RG
2076
2077 /*
2078 * Signal hash by emitting SX_HASH, followed by the table length.
2079 */
2080
e16e2ff8
NC
2081 if (flagged_hash) {
2082 PUTMARK(SX_FLAG_HASH);
2083 PUTMARK(hash_flags);
2084 } else {
2085 PUTMARK(SX_HASH);
2086 }
7a6a85bf
RG
2087 WLEN(len);
2088 TRACEME(("size = %d", len));
2089
2090 /*
2091 * Save possible iteration state via each() on that table.
2092 */
2093
2094 riter = HvRITER(hv);
2095 eiter = HvEITER(hv);
2096 hv_iterinit(hv);
2097
2098 /*
2099 * Now store each item recursively.
2100 *
2101 * If canonical is defined to some true value then store each
2102 * key/value pair in sorted order otherwise the order is random.
2103 * Canonical order is irrelevant when a deep clone operation is performed.
2104 *
2105 * Fetch the value from perl only once per store() operation, and only
2106 * when needed.
2107 */
2108
2109 if (
2110 !(cxt->optype & ST_CLONE) && (cxt->canonical == 1 ||
2111 (cxt->canonical < 0 && (cxt->canonical =
e16e2ff8 2112 (SvTRUE(perl_get_sv("Storable::canonical", TRUE)) ? 1 : 0))))
7a6a85bf
RG
2113 ) {
2114 /*
2115 * Storing in order, sorted by key.
2116 * Run through the hash, building up an array of keys in a
2117 * mortal array, sort the array and then run through the
2118 * array.
2119 */
2120
2121 AV *av = newAV();
2122
e16e2ff8
NC
2123 /*av_extend (av, len);*/
2124
7a6a85bf
RG
2125 TRACEME(("using canonical order"));
2126
2127 for (i = 0; i < len; i++) {
530b72ba 2128#ifdef HAS_RESTRICTED_HASHES
e16e2ff8 2129 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
530b72ba
NC
2130#else
2131 HE *he = hv_iternext(hv);
2132#endif
7a6a85bf
RG
2133 SV *key = hv_iterkeysv(he);
2134 av_store(av, AvFILLp(av)+1, key); /* av_push(), really */
2135 }
2136
2137 qsort((char *) AvARRAY(av), len, sizeof(SV *), sortcmp);
2138
2139 for (i = 0; i < len; i++) {
e16e2ff8 2140 unsigned char flags;
7a6a85bf 2141 char *keyval;
e16e2ff8
NC
2142 STRLEN keylen_tmp;
2143 I32 keylen;
7a6a85bf
RG
2144 SV *key = av_shift(av);
2145 HE *he = hv_fetch_ent(hv, key, 0, 0);
2146 SV *val = HeVAL(he);
2147 if (val == 0)
2148 return 1; /* Internal error, not I/O error */
2149
2150 /*
2151 * Store value first.
2152 */
2153
9e21b3d0 2154 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
7a6a85bf 2155
e993d95c 2156 if ((ret = store(cxt, val))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2157 goto out;
2158
2159 /*
2160 * Write key string.
2161 * Keys are written after values to make sure retrieval
2162 * can be optimal in terms of memory usage, where keys are
2163 * read into a fixed unique buffer called kbuf.
2164 * See retrieve_hash() for details.
2165 */
2166
e16e2ff8
NC
2167 /* Implementation of restricted hashes isn't nicely
2168 abstracted: */
2169 flags
2170 = (((hash_flags & SHV_RESTRICTED)
2171 && SvREADONLY(val))
2172 ? SHV_K_LOCKED : 0);
2173 if (val == &PL_sv_undef)
2174 flags |= SHV_K_PLACEHOLDER;
2175
2176 keyval = SvPV(key, keylen_tmp);
2177 keylen = keylen_tmp;
530b72ba
NC
2178#ifdef HAS_UTF8_HASHES
2179 /* If you build without optimisation on pre 5.6
2180 then nothing spots that SvUTF8(key) is always 0,
2181 so the block isn't optimised away, at which point
2182 the linker dislikes the reference to
2183 bytes_from_utf8. */
e16e2ff8
NC
2184 if (SvUTF8(key)) {
2185 const char *keysave = keyval;
2186 bool is_utf8 = TRUE;
2187
2188 /* Just casting the &klen to (STRLEN) won't work
2189 well if STRLEN and I32 are of different widths.
2190 --jhi */
2191 keyval = (char*)bytes_from_utf8((U8*)keyval,
2192 &keylen_tmp,
2193 &is_utf8);
2194
2195 /* If we were able to downgrade here, then than
2196 means that we have a key which only had chars
2197 0-255, but was utf8 encoded. */
2198
2199 if (keyval != keysave) {
2200 keylen = keylen_tmp;
2201 flags |= SHV_K_WASUTF8;
2202 } else {
2203 /* keylen_tmp can't have changed, so no need
2204 to assign back to keylen. */
2205 flags |= SHV_K_UTF8;
2206 }
2207 }
530b72ba 2208#endif
e16e2ff8
NC
2209
2210 if (flagged_hash) {
2211 PUTMARK(flags);
2212 TRACEME(("(#%d) key '%s' flags %x %u", i, keyval, flags, *keyval));
2213 } else {
2214 assert (flags == 0);
2215 TRACEME(("(#%d) key '%s'", i, keyval));
2216 }
7a6a85bf
RG
2217 WLEN(keylen);
2218 if (keylen)
2219 WRITE(keyval, keylen);
e16e2ff8
NC
2220 if (flags & SHV_K_WASUTF8)
2221 Safefree (keyval);
7a6a85bf
RG
2222 }
2223
2224 /*
2225 * Free up the temporary array
2226 */
2227
2228 av_undef(av);
2229 sv_free((SV *) av);
2230
2231 } else {
2232
2233 /*
2234 * Storing in "random" order (in the order the keys are stored
2235 * within the the hash). This is the default and will be faster!
2236 */
2237
2238 for (i = 0; i < len; i++) {
2239 char *key;
2240 I32 len;
e16e2ff8 2241 unsigned char flags;
530b72ba 2242#ifdef HV_ITERNEXT_WANTPLACEHOLDERS
e16e2ff8 2243 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
530b72ba
NC
2244#else
2245 HE *he = hv_iternext(hv);
2246#endif
e16e2ff8
NC
2247 SV *val = (he ? hv_iterval(hv, he) : 0);
2248 SV *key_sv = NULL;
2249 HEK *hek;
7a6a85bf
RG
2250
2251 if (val == 0)
2252 return 1; /* Internal error, not I/O error */
2253
2254 /*
2255 * Store value first.
2256 */
2257
9e21b3d0 2258 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
7a6a85bf 2259
e993d95c 2260 if ((ret = store(cxt, val))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2261 goto out;
2262
e16e2ff8
NC
2263 /* Implementation of restricted hashes isn't nicely
2264 abstracted: */
2265 flags
2266 = (((hash_flags & SHV_RESTRICTED)
2267 && SvREADONLY(val))
2268 ? SHV_K_LOCKED : 0);
2269 if (val == &PL_sv_undef)
2270 flags |= SHV_K_PLACEHOLDER;
2271
2272 hek = HeKEY_hek(he);
2273 len = HEK_LEN(hek);
2274 if (len == HEf_SVKEY) {
2275 /* This is somewhat sick, but the internal APIs are
2276 * such that XS code could put one of these in in
2277 * a regular hash.
2278 * Maybe we should be capable of storing one if
2279 * found.
2280 */
2281 key_sv = HeKEY_sv(he);
2282 flags |= SHV_K_ISSV;
2283 } else {
2284 /* Regular string key. */
530b72ba 2285#ifdef HAS_HASH_KEY_FLAGS
e16e2ff8
NC
2286 if (HEK_UTF8(hek))
2287 flags |= SHV_K_UTF8;
2288 if (HEK_WASUTF8(hek))
2289 flags |= SHV_K_WASUTF8;
530b72ba 2290#endif
e16e2ff8
NC
2291 key = HEK_KEY(hek);
2292 }
7a6a85bf
RG
2293 /*
2294 * Write key string.
2295 * Keys are written after values to make sure retrieval
2296 * can be optimal in terms of memory usage, where keys are
2297 * read into a fixed unique buffer called kbuf.
2298 * See retrieve_hash() for details.
2299 */
2300
e16e2ff8
NC
2301 if (flagged_hash) {
2302 PUTMARK(flags);
2303 TRACEME(("(#%d) key '%s' flags %x", i, key, flags));
2304 } else {
2305 assert (flags == 0);
2306 TRACEME(("(#%d) key '%s'", i, key));
2307 }
2308 if (flags & SHV_K_ISSV) {
2309 store(cxt, key_sv);
2310 } else {
2311 WLEN(len);
2312 if (len)
7a6a85bf 2313 WRITE(key, len);
e16e2ff8 2314 }
7a6a85bf
RG
2315 }
2316 }
2317
43d061fe 2318 TRACEME(("ok (hash 0x%"UVxf")", PTR2UV(hv)));
7a6a85bf
RG
2319
2320out:
2321 HvRITER(hv) = riter; /* Restore hash iterator state */
2322 HvEITER(hv) = eiter;
2323
2324 return ret;
2325}
2326
2327/*
2328 * store_tied
2329 *
2330 * When storing a tied object (be it a tied scalar, array or hash), we lay out
2331 * a special mark, followed by the underlying tied object. For instance, when
2332 * dealing with a tied hash, we store SX_TIED_HASH <hash object>, where
2333 * <hash object> stands for the serialization of the tied hash.
2334 */
f0ffaed8 2335static int store_tied(stcxt_t *cxt, SV *sv)
7a6a85bf
RG
2336{
2337 MAGIC *mg;
2338 int ret = 0;
2339 int svt = SvTYPE(sv);
2340 char mtype = 'P';
2341
43d061fe 2342 TRACEME(("store_tied (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
2343
2344 /*
2345 * We have a small run-time penalty here because we chose to factorise
2346 * all tieds objects into the same routine, and not have a store_tied_hash,
2347 * a store_tied_array, etc...
2348 *
2349 * Don't use a switch() statement, as most compilers don't optimize that
2350 * well for 2/3 values. An if() else if() cascade is just fine. We put
2351 * tied hashes first, as they are the most likely beasts.
2352 */
2353
2354 if (svt == SVt_PVHV) {
2355 TRACEME(("tied hash"));
2356 PUTMARK(SX_TIED_HASH); /* Introduces tied hash */
2357 } else if (svt == SVt_PVAV) {
2358 TRACEME(("tied array"));
2359 PUTMARK(SX_TIED_ARRAY); /* Introduces tied array */
2360 } else {
2361 TRACEME(("tied scalar"));
2362 PUTMARK(SX_TIED_SCALAR); /* Introduces tied scalar */
2363 mtype = 'q';
2364 }
2365
2366 if (!(mg = mg_find(sv, mtype)))
2367 CROAK(("No magic '%c' found while storing tied %s", mtype,
2368 (svt == SVt_PVHV) ? "hash" :
2369 (svt == SVt_PVAV) ? "array" : "scalar"));
2370
2371 /*
2372 * The mg->mg_obj found by mg_find() above actually points to the
2373 * underlying tied Perl object implementation. For instance, if the
2374 * original SV was that of a tied array, then mg->mg_obj is an AV.
2375 *
2376 * Note that we store the Perl object as-is. We don't call its FETCH
2377 * method along the way. At retrieval time, we won't call its STORE
2378 * method either, but the tieing magic will be re-installed. In itself,
2379 * that ensures that the tieing semantics are preserved since futher
2380 * accesses on the retrieved object will indeed call the magic methods...
2381 */
2382
e993d95c 2383 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2384 return ret;
2385
2386 TRACEME(("ok (tied)"));
2387
2388 return 0;
2389}
2390
2391/*
2392 * store_tied_item
2393 *
2394 * Stores a reference to an item within a tied structure:
2395 *
2396 * . \$h{key}, stores both the (tied %h) object and 'key'.
2397 * . \$a[idx], stores both the (tied @a) object and 'idx'.
2398 *
2399 * Layout is therefore either:
2400 * SX_TIED_KEY <object> <key>
2401 * SX_TIED_IDX <object> <index>
2402 */
f0ffaed8 2403static int store_tied_item(stcxt_t *cxt, SV *sv)
7a6a85bf
RG
2404{
2405 MAGIC *mg;
2406 int ret;
2407
43d061fe 2408 TRACEME(("store_tied_item (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
2409
2410 if (!(mg = mg_find(sv, 'p')))
2411 CROAK(("No magic 'p' found while storing reference to tied item"));
2412
2413 /*
2414 * We discriminate between \$h{key} and \$a[idx] via mg_ptr.
2415 */
2416
2417 if (mg->mg_ptr) {
2418 TRACEME(("store_tied_item: storing a ref to a tied hash item"));
2419 PUTMARK(SX_TIED_KEY);
9e21b3d0 2420 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
7a6a85bf 2421
e993d95c 2422 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2423 return ret;
2424
9e21b3d0 2425 TRACEME(("store_tied_item: storing PTR 0x%"UVxf, PTR2UV(mg->mg_ptr)));
7a6a85bf 2426
e993d95c 2427 if ((ret = store(cxt, (SV *) mg->mg_ptr))) /* Idem, for -Wall */
7a6a85bf
RG
2428 return ret;
2429 } else {
2430 I32 idx = mg->mg_len;
2431
2432 TRACEME(("store_tied_item: storing a ref to a tied array item "));
2433 PUTMARK(SX_TIED_IDX);
9e21b3d0 2434 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
7a6a85bf 2435
e993d95c 2436 if ((ret = store(cxt, mg->mg_obj))) /* Idem, for -Wall */
7a6a85bf
RG
2437 return ret;
2438
2439 TRACEME(("store_tied_item: storing IDX %d", idx));
2440
2441 WLEN(idx);
2442 }
2443
2444 TRACEME(("ok (tied item)"));
2445
2446 return 0;
2447}
2448
2449/*
2450 * store_hook -- dispatched manually, not via sv_store[]
2451 *
2452 * The blessed SV is serialized by a hook.
2453 *
2454 * Simple Layout is:
2455 *
2456 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2457 *
2458 * where <flags> indicates how long <len>, <len2> and <len3> are, whether
2459 * the trailing part [] is present, the type of object (scalar, array or hash).
2460 * There is also a bit which says how the classname is stored between:
2461 *
2462 * <len> <classname>
2463 * <index>
2464 *
2465 * and when the <index> form is used (classname already seen), the "large
2466 * classname" bit in <flags> indicates how large the <index> is.
2467 *
2468 * The serialized string returned by the hook is of length <len2> and comes
2469 * next. It is an opaque string for us.
2470 *
2471 * Those <len3> object IDs which are listed last represent the extra references
2472 * not directly serialized by the hook, but which are linked to the object.
2473 *
2474 * When recursion is mandated to resolve object-IDs not yet seen, we have
2475 * instead, with <header> being flags with bits set to indicate the object type
2476 * and that recursion was indeed needed:
2477 *
2478 * SX_HOOK <header> <object> <header> <object> <flags>
2479 *
2480 * that same header being repeated between serialized objects obtained through
2481 * recursion, until we reach flags indicating no recursion, at which point
2482 * we know we've resynchronized with a single layout, after <flags>.
b12202d0
JH
2483 *
2484 * When storing a blessed ref to a tied variable, the following format is
2485 * used:
2486 *
2487 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
2488 *
2489 * The first <flags> indication carries an object of type SHT_EXTRA, and the
2490 * real object type is held in the <extra> flag. At the very end of the
2491 * serialization stream, the underlying magic object is serialized, just like
2492 * any other tied variable.
7a6a85bf 2493 */
f0ffaed8
JH
2494static int store_hook(
2495 stcxt_t *cxt,
2496 SV *sv,
2497 int type,
2498 HV *pkg,
2499 SV *hook)
7a6a85bf
RG
2500{
2501 I32 len;
2502 char *class;
2503 STRLEN len2;
2504 SV *ref;
2505 AV *av;
2506 SV **ary;
2507 int count; /* really len3 + 1 */
2508 unsigned char flags;
2509 char *pv;
2510 int i;
2511 int recursed = 0; /* counts recursion */
2512 int obj_type; /* object type, on 2 bits */
2513 I32 classnum;
2514 int ret;
2515 int clone = cxt->optype & ST_CLONE;
e993d95c
JH
2516 char mtype = '\0'; /* for blessed ref to tied structures */
2517 unsigned char eflags = '\0'; /* used when object type is SHT_EXTRA */
7a6a85bf
RG
2518
2519 TRACEME(("store_hook, class \"%s\", tagged #%d", HvNAME(pkg), cxt->tagnum));
2520
2521 /*
2522 * Determine object type on 2 bits.
2523 */
2524
2525 switch (type) {
2526 case svis_SCALAR:
2527 obj_type = SHT_SCALAR;
2528 break;
2529 case svis_ARRAY:
2530 obj_type = SHT_ARRAY;
2531 break;
2532 case svis_HASH:
2533 obj_type = SHT_HASH;
2534 break;
b12202d0
JH
2535 case svis_TIED:
2536 /*
2537 * Produced by a blessed ref to a tied data structure, $o in the
2538 * following Perl code.
2539 *
2540 * my %h;
2541 * tie %h, 'FOO';
2542 * my $o = bless \%h, 'BAR';
2543 *
2544 * Signal the tie-ing magic by setting the object type as SHT_EXTRA
2545 * (since we have only 2 bits in <flags> to store the type), and an
2546 * <extra> byte flag will be emitted after the FIRST <flags> in the
2547 * stream, carrying what we put in `eflags'.
2548 */
2549 obj_type = SHT_EXTRA;
2550 switch (SvTYPE(sv)) {
2551 case SVt_PVHV:
2552 eflags = (unsigned char) SHT_THASH;
2553 mtype = 'P';
2554 break;
2555 case SVt_PVAV:
2556 eflags = (unsigned char) SHT_TARRAY;
2557 mtype = 'P';
2558 break;
2559 default:
2560 eflags = (unsigned char) SHT_TSCALAR;
2561 mtype = 'q';
2562 break;
2563 }
2564 break;
7a6a85bf
RG
2565 default:
2566 CROAK(("Unexpected object type (%d) in store_hook()", type));
2567 }
2568 flags = SHF_NEED_RECURSE | obj_type;
2569
2570 class = HvNAME(pkg);
2571 len = strlen(class);
2572
2573 /*
2574 * To call the hook, we need to fake a call like:
2575 *
2576 * $object->STORABLE_freeze($cloning);
2577 *
2578 * but we don't have the $object here. For instance, if $object is
2579 * a blessed array, what we have in `sv' is the array, and we can't
2580 * call a method on those.
2581 *
2582 * Therefore, we need to create a temporary reference to the object and
2583 * make the call on that reference.
2584 */
2585
2586 TRACEME(("about to call STORABLE_freeze on class %s", class));
2587
2588 ref = newRV_noinc(sv); /* Temporary reference */
2589 av = array_call(ref, hook, clone); /* @a = $object->STORABLE_freeze($c) */
2590 SvRV(ref) = 0;
2591 SvREFCNT_dec(ref); /* Reclaim temporary reference */
2592
2593 count = AvFILLp(av) + 1;
2594 TRACEME(("store_hook, array holds %d items", count));
2595
2596 /*
2597 * If they return an empty list, it means they wish to ignore the
2598 * hook for this class (and not just this instance -- that's for them
2599 * to handle if they so wish).
2600 *
2601 * Simply disable the cached entry for the hook (it won't be recomputed
2602 * since it's present in the cache) and recurse to store_blessed().
2603 */
2604
2605 if (!count) {
2606 /*
2607 * They must not change their mind in the middle of a serialization.
2608 */
2609
2610 if (hv_fetch(cxt->hclass, class, len, FALSE))
2611 CROAK(("Too late to ignore hooks for %s class \"%s\"",
2612 (cxt->optype & ST_CLONE) ? "cloning" : "storing", class));
2613
2614 pkg_hide(cxt->hook, pkg, "STORABLE_freeze");
2615
2616 ASSERT(!pkg_can(cxt->hook, pkg, "STORABLE_freeze"), ("hook invisible"));
cc964657 2617 TRACEME(("ignoring STORABLE_freeze in class \"%s\"", class));
7a6a85bf
RG
2618
2619 return store_blessed(cxt, sv, type, pkg);
2620 }
2621
2622 /*
2623 * Get frozen string.
2624 */
2625
2626 ary = AvARRAY(av);
2627 pv = SvPV(ary[0], len2);
2628
2629 /*
7a6a85bf
RG
2630 * If they returned more than one item, we need to serialize some
2631 * extra references if not already done.
2632 *
2633 * Loop over the array, starting at postion #1, and for each item,
2634 * ensure it is a reference, serialize it if not already done, and
2635 * replace the entry with the tag ID of the corresponding serialized
2636 * object.
2637 *
2638 * We CHEAT by not calling av_fetch() and read directly within the
2639 * array, for speed.
2640 */
2641
2642 for (i = 1; i < count; i++) {
2643 SV **svh;
90826881
JH
2644 SV *rsv = ary[i];
2645 SV *xsv;
2646 AV *av_hook = cxt->hook_seen;
7a6a85bf 2647
90826881
JH
2648 if (!SvROK(rsv))
2649 CROAK(("Item #%d returned by STORABLE_freeze "
2650 "for %s is not a reference", i, class));
2651 xsv = SvRV(rsv); /* Follow ref to know what to look for */
7a6a85bf
RG
2652
2653 /*
2654 * Look in hseen and see if we have a tag already.
2655 * Serialize entry if not done already, and get its tag.
2656 */
2657
13689cfe 2658 if ((svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE)))
7a6a85bf
RG
2659 goto sv_seen; /* Avoid moving code too far to the right */
2660
9e21b3d0 2661 TRACEME(("listed object %d at 0x%"UVxf" is unknown", i-1, PTR2UV(xsv)));
7a6a85bf
RG
2662
2663 /*
2664 * We need to recurse to store that object and get it to be known
2665 * so that we can resolve the list of object-IDs at retrieve time.
2666 *
2667 * The first time we do this, we need to emit the proper header
2668 * indicating that we recursed, and what the type of object is (the
2669 * object we're storing via a user-hook). Indeed, during retrieval,
2670 * we'll have to create the object before recursing to retrieve the
2671 * others, in case those would point back at that object.
2672 */
2673
b12202d0
JH
2674 /* [SX_HOOK] <flags> [<extra>] <object>*/
2675 if (!recursed++) {
7a6a85bf 2676 PUTMARK(SX_HOOK);
b12202d0
JH
2677 PUTMARK(flags);
2678 if (obj_type == SHT_EXTRA)
2679 PUTMARK(eflags);
2680 } else
2681 PUTMARK(flags);
7a6a85bf 2682
e993d95c 2683 if ((ret = store(cxt, xsv))) /* Given by hook for us to store */
7a6a85bf
RG
2684 return ret;
2685
2686 svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE);
2687 if (!svh)
2688 CROAK(("Could not serialize item #%d from hook in %s", i, class));
2689
2690 /*
90826881
JH
2691 * It was the first time we serialized `xsv'.
2692 *
2693 * Keep this SV alive until the end of the serialization: if we
2694 * disposed of it right now by decrementing its refcount, and it was
2695 * a temporary value, some next temporary value allocated during
2696 * another STORABLE_freeze might take its place, and we'd wrongly
2697 * assume that new SV was already serialized, based on its presence
2698 * in cxt->hseen.
2699 *
2700 * Therefore, push it away in cxt->hook_seen.
7a6a85bf
RG
2701 */
2702
90826881
JH
2703 av_store(av_hook, AvFILLp(av_hook)+1, SvREFCNT_inc(xsv));
2704
7a6a85bf 2705 sv_seen:
90826881
JH
2706 /*
2707 * Dispose of the REF they returned. If we saved the `xsv' away
2708 * in the array of returned SVs, that will not cause the underlying
2709 * referenced SV to be reclaimed.
2710 */
2711
2712 ASSERT(SvREFCNT(xsv) > 1, ("SV will survive disposal of its REF"));
2713 SvREFCNT_dec(rsv); /* Dispose of reference */
2714
2715 /*
2716 * Replace entry with its tag (not a real SV, so no refcnt increment)
2717 */
2718
7a6a85bf 2719 ary[i] = *svh;
76edffbb 2720 TRACEME(("listed object %d at 0x%"UVxf" is tag #%"UVuf,
d67b2c17 2721 i-1, PTR2UV(xsv), PTR2UV(*svh)));
7a6a85bf
RG
2722 }
2723
2724 /*
dd19458b
JH
2725 * Allocate a class ID if not already done.
2726 *
2727 * This needs to be done after the recursion above, since at retrieval
2728 * time, we'll see the inner objects first. Many thanks to
2729 * Salvador Ortiz Garcia <sog@msg.com.mx> who spot that bug and
2730 * proposed the right fix. -- RAM, 15/09/2000
2731 */
2732
2733 if (!known_class(cxt, class, len, &classnum)) {
2734 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2735 classnum = -1; /* Mark: we must store classname */
2736 } else {
2737 TRACEME(("already seen class %s, ID = %d", class, classnum));
2738 }
2739
2740 /*
7a6a85bf
RG
2741 * Compute leading flags.
2742 */
2743
2744 flags = obj_type;
2745 if (((classnum == -1) ? len : classnum) > LG_SCALAR)
2746 flags |= SHF_LARGE_CLASSLEN;
2747 if (classnum != -1)
2748 flags |= SHF_IDX_CLASSNAME;
2749 if (len2 > LG_SCALAR)
2750 flags |= SHF_LARGE_STRLEN;
2751 if (count > 1)
2752 flags |= SHF_HAS_LIST;
2753 if (count > (LG_SCALAR + 1))
2754 flags |= SHF_LARGE_LISTLEN;
2755
2756 /*
2757 * We're ready to emit either serialized form:
2758 *
2759 * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2760 * SX_HOOK <flags> <index> <len2> <str> [<len3> <object-IDs>]
2761 *
2762 * If we recursed, the SX_HOOK has already been emitted.
2763 */
2764
9e21b3d0
JH
2765 TRACEME(("SX_HOOK (recursed=%d) flags=0x%x "
2766 "class=%"IVdf" len=%"IVdf" len2=%"IVdf" len3=%d",
d67b2c17 2767 recursed, flags, (IV)classnum, (IV)len, (IV)len2, count-1));
7a6a85bf 2768
b12202d0
JH
2769 /* SX_HOOK <flags> [<extra>] */
2770 if (!recursed) {
7a6a85bf 2771 PUTMARK(SX_HOOK);
b12202d0
JH
2772 PUTMARK(flags);
2773 if (obj_type == SHT_EXTRA)
2774 PUTMARK(eflags);
2775 } else
2776 PUTMARK(flags);
7a6a85bf
RG
2777
2778 /* <len> <classname> or <index> */
2779 if (flags & SHF_IDX_CLASSNAME) {
2780 if (flags & SHF_LARGE_CLASSLEN)
2781 WLEN(classnum);
2782 else {
2783 unsigned char cnum = (unsigned char) classnum;
2784 PUTMARK(cnum);
2785 }
2786 } else {
2787 if (flags & SHF_LARGE_CLASSLEN)
2788 WLEN(len);
2789 else {
2790 unsigned char clen = (unsigned char) len;
2791 PUTMARK(clen);
2792 }
2793 WRITE(class, len); /* Final \0 is omitted */
2794 }
2795
2796 /* <len2> <frozen-str> */
cc964657
JH
2797 if (flags & SHF_LARGE_STRLEN) {
2798 I32 wlen2 = len2; /* STRLEN might be 8 bytes */
2799 WLEN(wlen2); /* Must write an I32 for 64-bit machines */
2800 } else {
7a6a85bf
RG
2801 unsigned char clen = (unsigned char) len2;
2802 PUTMARK(clen);
2803 }
2804 if (len2)
7c436af3 2805 WRITE(pv, (SSize_t)len2); /* Final \0 is omitted */
7a6a85bf
RG
2806
2807 /* [<len3> <object-IDs>] */
2808 if (flags & SHF_HAS_LIST) {
2809 int len3 = count - 1;
2810 if (flags & SHF_LARGE_LISTLEN)
2811 WLEN(len3);
2812 else {
2813 unsigned char clen = (unsigned char) len3;
2814 PUTMARK(clen);
2815 }
2816
2817 /*
2818 * NOTA BENE, for 64-bit machines: the ary[i] below does not yield a
2819 * real pointer, rather a tag number, well under the 32-bit limit.
2820 */
2821
2822 for (i = 1; i < count; i++) {
2823 I32 tagval = htonl(LOW_32BITS(ary[i]));
9e21b3d0 2824 WRITE_I32(tagval);
7a6a85bf
RG
2825 TRACEME(("object %d, tag #%d", i-1, ntohl(tagval)));
2826 }
2827 }
2828
2829 /*
2830 * Free the array. We need extra care for indices after 0, since they
2831 * don't hold real SVs but integers cast.
2832 */
2833
2834 if (count > 1)
2835 AvFILLp(av) = 0; /* Cheat, nothing after 0 interests us */
2836 av_undef(av);
2837 sv_free((SV *) av);
2838
b12202d0
JH
2839 /*
2840 * If object was tied, need to insert serialization of the magic object.
2841 */
2842
2843 if (obj_type == SHT_EXTRA) {
2844 MAGIC *mg;
2845
2846 if (!(mg = mg_find(sv, mtype))) {
2847 int svt = SvTYPE(sv);
2848 CROAK(("No magic '%c' found while storing ref to tied %s with hook",
2849 mtype, (svt == SVt_PVHV) ? "hash" :
2850 (svt == SVt_PVAV) ? "array" : "scalar"));
2851 }
2852
2853 TRACEME(("handling the magic object 0x%"UVxf" part of 0x%"UVxf,
2854 PTR2UV(mg->mg_obj), PTR2UV(sv)));
2855
2856 /*
2857 * [<magic object>]
2858 */
2859
e993d95c 2860 if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
b12202d0
JH
2861 return ret;
2862 }
2863
7a6a85bf
RG
2864 return 0;
2865}
2866
2867/*
2868 * store_blessed -- dispatched manually, not via sv_store[]
2869 *
2870 * Check whether there is a STORABLE_xxx hook defined in the class or in one
2871 * of its ancestors. If there is, then redispatch to store_hook();
2872 *
2873 * Otherwise, the blessed SV is stored using the following layout:
2874 *
2875 * SX_BLESS <flag> <len> <classname> <object>
2876 *
2877 * where <flag> indicates whether <len> is stored on 0 or 4 bytes, depending
2878 * on the high-order bit in flag: if 1, then length follows on 4 bytes.
2879 * Otherwise, the low order bits give the length, thereby giving a compact
2880 * representation for class names less than 127 chars long.
2881 *
2882 * Each <classname> seen is remembered and indexed, so that the next time
2883 * an object in the blessed in the same <classname> is stored, the following
2884 * will be emitted:
2885 *
2886 * SX_IX_BLESS <flag> <index> <object>
2887 *
2888 * where <index> is the classname index, stored on 0 or 4 bytes depending
2889 * on the high-order bit in flag (same encoding as above for <len>).
2890 */
f0ffaed8
JH
2891static int store_blessed(
2892 stcxt_t *cxt,
2893 SV *sv,
2894 int type,
2895 HV *pkg)
7a6a85bf
RG
2896{
2897 SV *hook;
2898 I32 len;
2899 char *class;
2900 I32 classnum;
2901
2902 TRACEME(("store_blessed, type %d, class \"%s\"", type, HvNAME(pkg)));
2903
2904 /*
2905 * Look for a hook for this blessed SV and redirect to store_hook()
2906 * if needed.
2907 */
2908
2909 hook = pkg_can(cxt->hook, pkg, "STORABLE_freeze");
2910 if (hook)
2911 return store_hook(cxt, sv, type, pkg, hook);
2912
2913 /*
2914 * This is a blessed SV without any serialization hook.
2915 */
2916
2917 class = HvNAME(pkg);
2918 len = strlen(class);
2919
43d061fe
JH
2920 TRACEME(("blessed 0x%"UVxf" in %s, no hook: tagged #%d",
2921 PTR2UV(sv), class, cxt->tagnum));
7a6a85bf
RG
2922
2923 /*
2924 * Determine whether it is the first time we see that class name (in which
2925 * case it will be stored in the SX_BLESS form), or whether we already
2926 * saw that class name before (in which case the SX_IX_BLESS form will be
2927 * used).
2928 */
2929
2930 if (known_class(cxt, class, len, &classnum)) {
2931 TRACEME(("already seen class %s, ID = %d", class, classnum));
2932 PUTMARK(SX_IX_BLESS);
2933 if (classnum <= LG_BLESS) {
2934 unsigned char cnum = (unsigned char) classnum;
2935 PUTMARK(cnum);
2936 } else {
2937 unsigned char flag = (unsigned char) 0x80;
2938 PUTMARK(flag);
2939 WLEN(classnum);
2940 }
2941 } else {
2942 TRACEME(("first time we see class %s, ID = %d", class, classnum));
2943 PUTMARK(SX_BLESS);
2944 if (len <= LG_BLESS) {
2945 unsigned char clen = (unsigned char) len;
2946 PUTMARK(clen);
2947 } else {
2948 unsigned char flag = (unsigned char) 0x80;
2949 PUTMARK(flag);
2950 WLEN(len); /* Don't BER-encode, this should be rare */
2951 }
2952 WRITE(class, len); /* Final \0 is omitted */
2953 }
2954
2955 /*
2956 * Now emit the <object> part.
2957 */
2958
2959 return SV_STORE(type)(cxt, sv);
2960}
2961
2962/*
2963 * store_other
2964 *
2965 * We don't know how to store the item we reached, so return an error condition.
2966 * (it's probably a GLOB, some CODE reference, etc...)
2967 *
2968 * If they defined the `forgive_me' variable at the Perl level to some
2969 * true value, then don't croak, just warn, and store a placeholder string
2970 * instead.
2971 */
f0ffaed8 2972static int store_other(stcxt_t *cxt, SV *sv)
7a6a85bf 2973{
cc964657 2974 I32 len;
7a6a85bf
RG
2975 static char buf[80];
2976
2977 TRACEME(("store_other"));
2978
2979 /*
2980 * Fetch the value from perl only once per store() operation.
2981 */
2982
2983 if (
2984 cxt->forgive_me == 0 ||
2985 (cxt->forgive_me < 0 && !(cxt->forgive_me =
2986 SvTRUE(perl_get_sv("Storable::forgive_me", TRUE)) ? 1 : 0))
2987 )
2988 CROAK(("Can't store %s items", sv_reftype(sv, FALSE)));
2989
43d061fe
JH
2990 warn("Can't store item %s(0x%"UVxf")",
2991 sv_reftype(sv, FALSE), PTR2UV(sv));
7a6a85bf
RG
2992
2993 /*
2994 * Store placeholder string as a scalar instead...
2995 */
2996
13689cfe 2997 (void) sprintf(buf, "You lost %s(0x%"UVxf")%c", sv_reftype(sv, FALSE),
e993d95c 2998 PTR2UV(sv), (char) 0);
7a6a85bf
RG
2999
3000 len = strlen(buf);
3001 STORE_SCALAR(buf, len);
1cf92b12 3002 TRACEME(("ok (dummy \"%s\", length = %"IVdf")", buf, (IV) len));
7a6a85bf
RG
3003
3004 return 0;
3005}
3006
3007/***
3008 *** Store driving routines
3009 ***/
3010
3011/*
3012 * sv_type
3013 *
3014 * WARNING: partially duplicates Perl's sv_reftype for speed.
3015 *
3016 * Returns the type of the SV, identified by an integer. That integer
3017 * may then be used to index the dynamic routine dispatch table.
3018 */
f0ffaed8 3019static int sv_type(SV *sv)
7a6a85bf
RG
3020{
3021 switch (SvTYPE(sv)) {
3022 case SVt_NULL:
3023 case SVt_IV:
3024 case SVt_NV:
3025 /*
3026 * No need to check for ROK, that can't be set here since there
3027 * is no field capable of hodling the xrv_rv reference.
3028 */
3029 return svis_SCALAR;
3030 case SVt_PV:
3031 case SVt_RV:
3032 case SVt_PVIV:
3033 case SVt_PVNV:
3034 /*
3035 * Starting from SVt_PV, it is possible to have the ROK flag
3036 * set, the pointer to the other SV being either stored in
3037 * the xrv_rv (in the case of a pure SVt_RV), or as the
3038 * xpv_pv field of an SVt_PV and its heirs.
3039 *
3040 * However, those SV cannot be magical or they would be an
3041 * SVt_PVMG at least.
3042 */
3043 return SvROK(sv) ? svis_REF : svis_SCALAR;
3044 case SVt_PVMG:
3045 case SVt_PVLV: /* Workaround for perl5.004_04 "LVALUE" bug */
3046 if (SvRMAGICAL(sv) && (mg_find(sv, 'p')))
3047 return svis_TIED_ITEM;
3048 /* FALL THROUGH */
3049 case SVt_PVBM:
3050 if (SvRMAGICAL(sv) && (mg_find(sv, 'q')))
3051 return svis_TIED;
3052 return SvROK(sv) ? svis_REF : svis_SCALAR;
3053 case SVt_PVAV:
3054 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
3055 return svis_TIED;
3056 return svis_ARRAY;
3057 case SVt_PVHV:
3058 if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
3059 return svis_TIED;
3060 return svis_HASH;
3061 default:
3062 break;
3063 }
3064
3065 return svis_OTHER;
3066}
3067
3068/*
3069 * store
3070 *
3071 * Recursively store objects pointed to by the sv to the specified file.
3072 *
3073 * Layout is <content> or SX_OBJECT <tagnum> if we reach an already stored
3074 * object (one for which storage has started -- it may not be over if we have
3075 * a self-referenced structure). This data set forms a stored <object>.
3076 */
f0ffaed8 3077static int store(stcxt_t *cxt, SV *sv)
7a6a85bf
RG
3078{
3079 SV **svh;
3080 int ret;
7a6a85bf 3081 int type;
43d061fe 3082 HV *hseen = cxt->hseen;
7a6a85bf 3083
43d061fe 3084 TRACEME(("store (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
3085
3086 /*
3087 * If object has already been stored, do not duplicate data.
3088 * Simply emit the SX_OBJECT marker followed by its tag data.
3089 * The tag is always written in network order.
3090 *
3091 * NOTA BENE, for 64-bit machines: the "*svh" below does not yield a
3092 * real pointer, rather a tag number (watch the insertion code below).
3093 * That means it pobably safe to assume it is well under the 32-bit limit,
3094 * and makes the truncation safe.
3095 * -- RAM, 14/09/1999
3096 */
3097
3098 svh = hv_fetch(hseen, (char *) &sv, sizeof(sv), FALSE);
3099 if (svh) {
3100 I32 tagval = htonl(LOW_32BITS(*svh));
3101
9e21b3d0 3102 TRACEME(("object 0x%"UVxf" seen as #%d", PTR2UV(sv), ntohl(tagval)));
7a6a85bf
RG
3103
3104 PUTMARK(SX_OBJECT);
9e21b3d0 3105 WRITE_I32(tagval);
7a6a85bf
RG
3106 return 0;
3107 }
3108
3109 /*
3110 * Allocate a new tag and associate it with the address of the sv being
3111 * stored, before recursing...
3112 *
3113 * In order to avoid creating new SvIVs to hold the tagnum we just
d1be9408 3114 * cast the tagnum to an SV pointer and store that in the hash. This
7a6a85bf
RG
3115 * means that we must clean up the hash manually afterwards, but gives
3116 * us a 15% throughput increase.
3117 *
7a6a85bf
RG
3118 */
3119
3120 cxt->tagnum++;
3121 if (!hv_store(hseen,
3341c981 3122 (char *) &sv, sizeof(sv), INT2PTR(SV*, cxt->tagnum), 0))
7a6a85bf
RG
3123 return -1;
3124
3125 /*
3126 * Store `sv' and everything beneath it, using appropriate routine.
3127 * Abort immediately if we get a non-zero status back.
3128 */
3129
3130 type = sv_type(sv);
3131
43d061fe
JH
3132 TRACEME(("storing 0x%"UVxf" tag #%d, type %d...",
3133 PTR2UV(sv), cxt->tagnum, type));
7a6a85bf
RG
3134
3135 if (SvOBJECT(sv)) {
3136 HV *pkg = SvSTASH(sv);
3137 ret = store_blessed(cxt, sv, type, pkg);
3138 } else
3139 ret = SV_STORE(type)(cxt, sv);
3140
43d061fe
JH
3141 TRACEME(("%s (stored 0x%"UVxf", refcnt=%d, %s)",
3142 ret ? "FAILED" : "ok", PTR2UV(sv),
7a6a85bf
RG
3143 SvREFCNT(sv), sv_reftype(sv, FALSE)));
3144
3145 return ret;
3146}
3147
3148/*
3149 * magic_write
3150 *
3151 * Write magic number and system information into the file.
3152 * Layout is <magic> <network> [<len> <byteorder> <sizeof int> <sizeof long>
3153 * <sizeof ptr>] where <len> is the length of the byteorder hexa string.
3154 * All size and lenghts are written as single characters here.
3155 *
3156 * Note that no byte ordering info is emitted when <network> is true, since
3157 * integers will be emitted in network order in that case.
3158 */
f0ffaed8 3159static int magic_write(stcxt_t *cxt)
7a6a85bf
RG
3160{
3161 char buf[256]; /* Enough room for 256 hexa digits */
3162 unsigned char c;
3163 int use_network_order = cxt->netorder;
3164
e16e2ff8
NC
3165 TRACEME(("magic_write on fd=%d", cxt->fio ? PerlIO_fileno(cxt->fio)
3166 : -1));
7a6a85bf
RG
3167
3168 if (cxt->fio)
7c436af3 3169 WRITE(magicstr, (SSize_t)strlen(magicstr)); /* Don't write final \0 */
7a6a85bf
RG
3170
3171 /*
3172 * Starting with 0.6, the "use_network_order" byte flag is also used to
3173 * indicate the version number of the binary image, encoded in the upper
3174 * bits. The bit 0 is always used to indicate network order.
3175 */
3176
3177 c = (unsigned char)
3178 ((use_network_order ? 0x1 : 0x0) | (STORABLE_BIN_MAJOR << 1));
3179 PUTMARK(c);
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 */
3186
530b72ba 3187 PUTMARK((unsigned char) STORABLE_BIN_WRITE_MINOR);
7a6a85bf
RG
3188
3189 if (use_network_order)
3190 return 0; /* Don't bother with byte ordering */
3191
3192 sprintf(buf, "%lx", (unsigned long) BYTEORDER);
3193 c = (unsigned char) strlen(buf);
3194 PUTMARK(c);
7c436af3 3195 WRITE(buf, (SSize_t)c); /* Don't write final \0 */
7a6a85bf
RG
3196 PUTMARK((unsigned char) sizeof(int));
3197 PUTMARK((unsigned char) sizeof(long));
3198 PUTMARK((unsigned char) sizeof(char *));
9e21b3d0 3199 PUTMARK((unsigned char) sizeof(NV));
7a6a85bf 3200
9e21b3d0 3201 TRACEME(("ok (magic_write byteorder = 0x%lx [%d], I%d L%d P%d D%d)",
43d061fe 3202 (unsigned long) BYTEORDER, (int) c,
9e21b3d0
JH
3203 (int) sizeof(int), (int) sizeof(long),
3204 (int) sizeof(char *), (int) sizeof(NV)));
7a6a85bf
RG
3205
3206 return 0;
3207}
3208
3209/*
3210 * do_store
3211 *
3212 * Common code for store operations.
3213 *
3214 * When memory store is requested (f = NULL) and a non null SV* is given in
3215 * `res', it is filled with a new SV created out of the memory buffer.
3216 *
3217 * It is required to provide a non-null `res' when the operation type is not
3218 * dclone() and store() is performed to memory.
3219 */
f0ffaed8
JH
3220static int do_store(
3221 PerlIO *f,
3222 SV *sv,
3223 int optype,
3224 int network_order,
3225 SV **res)
7a6a85bf
RG
3226{
3227 dSTCXT;
3228 int status;
3229
3230 ASSERT(!(f == 0 && !(optype & ST_CLONE)) || res,
3231 ("must supply result SV pointer for real recursion to memory"));
3232
3233 TRACEME(("do_store (optype=%d, netorder=%d)",
3234 optype, network_order));
3235
3236 optype |= ST_STORE;
3237
3238 /*
3239 * Workaround for CROAK leak: if they enter with a "dirty" context,
3240 * free up memory for them now.
3241 */
3242
dd19458b 3243 if (cxt->s_dirty)
7a6a85bf
RG
3244 clean_context(cxt);
3245
3246 /*
3247 * Now that STORABLE_xxx hooks exist, it is possible that they try to
3248 * re-enter store() via the hooks. We need to stack contexts.
3249 */
3250
3251 if (cxt->entry)
3252 cxt = allocate_context(cxt);
3253
3254 cxt->entry++;
3255
3256 ASSERT(cxt->entry == 1, ("starting new recursion"));
dd19458b 3257 ASSERT(!cxt->s_dirty, ("clean context"));
7a6a85bf
RG
3258
3259 /*
3260 * Ensure sv is actually a reference. From perl, we called something
3261 * like:
3262 * pstore(FILE, \@array);
3263 * so we must get the scalar value behing that reference.
3264 */
3265
3266 if (!SvROK(sv))
3267 CROAK(("Not a reference"));
3268 sv = SvRV(sv); /* So follow it to know what to store */
3269
3270 /*
3271 * If we're going to store to memory, reset the buffer.
3272 */
3273
3274 if (!f)
3275 MBUF_INIT(0);
3276
3277 /*
3278 * Prepare context and emit headers.
3279 */
3280
3281 init_store_context(cxt, f, optype, network_order);
3282
3283 if (-1 == magic_write(cxt)) /* Emit magic and ILP info */
3284 return 0; /* Error */
3285
3286 /*
3287 * Recursively store object...
3288 */
3289
3290 ASSERT(is_storing(), ("within store operation"));
3291
3292 status = store(cxt, sv); /* Just do it! */
3293
3294 /*
3295 * If they asked for a memory store and they provided an SV pointer,
3296 * make an SV string out of the buffer and fill their pointer.
3297 *
3298 * When asking for ST_REAL, it's MANDATORY for the caller to provide
3299 * an SV, since context cleanup might free the buffer if we did recurse.
3300 * (unless caller is dclone(), which is aware of that).
3301 */
3302
3303 if (!cxt->fio && res)
3304 *res = mbuf2sv();
3305
3306 /*
3307 * Final cleanup.
3308 *
3309 * The "root" context is never freed, since it is meant to be always
3310 * handy for the common case where no recursion occurs at all (i.e.
3311 * we enter store() outside of any Storable code and leave it, period).
3312 * We know it's the "root" context because there's nothing stacked
3313 * underneath it.
3314 *
3315 * OPTIMIZATION:
3316 *
3317 * When deep cloning, we don't free the context: doing so would force
3318 * us to copy the data in the memory buffer. Sicne we know we're
3319 * about to enter do_retrieve...
3320 */
3321
3322 clean_store_context(cxt);
3323 if (cxt->prev && !(cxt->optype & ST_CLONE))
3324 free_context(cxt);
3325
3326 TRACEME(("do_store returns %d", status));
3327
3328 return status == 0;
3329}
3330
3331/*
3332 * pstore
3333 *
3334 * Store the transitive data closure of given object to disk.
3335 * Returns 0 on error, a true value otherwise.
3336 */
f0ffaed8 3337int pstore(PerlIO *f, SV *sv)
7a6a85bf
RG
3338{
3339 TRACEME(("pstore"));
f0ffaed8 3340 return do_store(f, sv, 0, FALSE, (SV**) 0);
7a6a85bf
RG
3341
3342}
3343
3344/*
3345 * net_pstore
3346 *
3347 * Same as pstore(), but network order is used for integers and doubles are
3348 * emitted as strings.
3349 */
f0ffaed8 3350int net_pstore(PerlIO *f, SV *sv)
7a6a85bf
RG
3351{
3352 TRACEME(("net_pstore"));
f0ffaed8 3353 return do_store(f, sv, 0, TRUE, (SV**) 0);
7a6a85bf
RG
3354}
3355
3356/***
3357 *** Memory stores.
3358 ***/
3359
3360/*
3361 * mbuf2sv
3362 *
3363 * Build a new SV out of the content of the internal memory buffer.
3364 */
f0ffaed8 3365static SV *mbuf2sv(void)
7a6a85bf
RG
3366{
3367 dSTCXT;
3368
3369 return newSVpv(mbase, MBUF_SIZE());
3370}
3371
3372/*
3373 * mstore
3374 *
3375 * Store the transitive data closure of given object to memory.
3376 * Returns undef on error, a scalar value containing the data otherwise.
3377 */
f0ffaed8 3378SV *mstore(SV *sv)
7a6a85bf 3379{
7a6a85bf
RG
3380 SV *out;
3381
3382 TRACEME(("mstore"));
3383
f0ffaed8 3384 if (!do_store((PerlIO*) 0, sv, 0, FALSE, &out))
7a6a85bf
RG
3385 return &PL_sv_undef;
3386
3387 return out;
3388}
3389
3390/*
3391 * net_mstore
3392 *
3393 * Same as mstore(), but network order is used for integers and doubles are
3394 * emitted as strings.
3395 */
f0ffaed8 3396SV *net_mstore(SV *sv)
7a6a85bf 3397{
7a6a85bf
RG
3398 SV *out;
3399
3400 TRACEME(("net_mstore"));
3401
f0ffaed8 3402 if (!do_store((PerlIO*) 0, sv, 0, TRUE, &out))
7a6a85bf
RG
3403 return &PL_sv_undef;
3404
3405 return out;
3406}
3407
3408/***
3409 *** Specific retrieve callbacks.
3410 ***/
3411
3412/*
3413 * retrieve_other
3414 *
3415 * Return an error via croak, since it is not possible that we get here
3416 * under normal conditions, when facing a file produced via pstore().
3417 */
b12202d0 3418static SV *retrieve_other(stcxt_t *cxt, char *cname)
7a6a85bf
RG
3419{
3420 if (
3421 cxt->ver_major != STORABLE_BIN_MAJOR &&
3422 cxt->ver_minor != STORABLE_BIN_MINOR
3423 ) {
3424 CROAK(("Corrupted storable %s (binary v%d.%d), current is v%d.%d",
3425 cxt->fio ? "file" : "string",
3426 cxt->ver_major, cxt->ver_minor,
3427 STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
3428 } else {
3429 CROAK(("Corrupted storable %s (binary v%d.%d)",
3430 cxt->fio ? "file" : "string",
3431 cxt->ver_major, cxt->ver_minor));
3432 }
3433
3434 return (SV *) 0; /* Just in case */
3435}
3436
3437/*
3438 * retrieve_idx_blessed
3439 *
3440 * Layout is SX_IX_BLESS <index> <object> with SX_IX_BLESS already read.
3441 * <index> can be coded on either 1 or 5 bytes.
3442 */
b12202d0 3443static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname)
7a6a85bf
RG
3444{
3445 I32 idx;
3446 char *class;
3447 SV **sva;
3448 SV *sv;
3449
3450 TRACEME(("retrieve_idx_blessed (#%d)", cxt->tagnum));
b12202d0 3451 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
7a6a85bf
RG
3452
3453 GETMARK(idx); /* Index coded on a single char? */
3454 if (idx & 0x80)
3455 RLEN(idx);
3456
3457 /*
3458 * Fetch classname in `aclass'
3459 */
3460
3461 sva = av_fetch(cxt->aclass, idx, FALSE);
3462 if (!sva)
e993d95c 3463 CROAK(("Class name #%"IVdf" should have been seen already", (IV) idx));
7a6a85bf
RG
3464
3465 class = SvPVX(*sva); /* We know it's a PV, by construction */
3466
3467 TRACEME(("class ID %d => %s", idx, class));
3468
3469 /*
3470 * Retrieve object and bless it.
3471 */
3472
b12202d0 3473 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
7a6a85bf
RG
3474
3475 return sv;
3476}
3477
3478/*
3479 * retrieve_blessed
3480 *
3481 * Layout is SX_BLESS <len> <classname> <object> with SX_BLESS already read.
3482 * <len> can be coded on either 1 or 5 bytes.
3483 */
b12202d0 3484static SV *retrieve_blessed(stcxt_t *cxt, char *cname)
7a6a85bf
RG
3485{
3486 I32 len;
3487 SV *sv;
3488 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3489 char *class = buf;
3490
3491 TRACEME(("retrieve_blessed (#%d)", cxt->tagnum));
b12202d0 3492 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
7a6a85bf
RG
3493
3494 /*
3495 * Decode class name length and read that name.
3496 *
3497 * Short classnames have two advantages: their length is stored on one
3498 * single byte, and the string can be read on the stack.
3499 */
3500
3501 GETMARK(len); /* Length coded on a single char? */
3502 if (len & 0x80) {
3503 RLEN(len);
3504 TRACEME(("** allocating %d bytes for class name", len+1));
3505 New(10003, class, len+1, char);
3506 }
3507 READ(class, len);
3508 class[len] = '\0'; /* Mark string end */
3509
3510 /*
3511 * It's a new classname, otherwise it would have been an SX_IX_BLESS.
3512 */
3513
b12202d0
JH
3514 TRACEME(("new class name \"%s\" will bear ID = %d", class, cxt->classnum));
3515
7a6a85bf
RG
3516 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3517 return (SV *) 0;
3518
3519 /*
3520 * Retrieve object and bless it.
3521 */
3522
b12202d0
JH
3523 sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */
3524 if (class != buf)
3525 Safefree(class);
7a6a85bf
RG
3526
3527 return sv;
3528}
3529
3530/*
3531 * retrieve_hook
3532 *
3533 * Layout: SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
3534 * with leading mark already read, as usual.
3535 *
3536 * When recursion was involved during serialization of the object, there
3537 * is an unknown amount of serialized objects after the SX_HOOK mark. Until
3538 * we reach a <flags> marker with the recursion bit cleared.
b12202d0
JH
3539 *
3540 * If the first <flags> byte contains a type of SHT_EXTRA, then the real type
3541 * is held in the <extra> byte, and if the object is tied, the serialized
3542 * magic object comes at the very end:
3543 *
3544 * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
3545 *
3546 * This means the STORABLE_thaw hook will NOT get a tied variable during its
3547 * processing (since we won't have seen the magic object by the time the hook
3548 * is called). See comments below for why it was done that way.
7a6a85bf 3549 */
b12202d0 3550static SV *retrieve_hook(stcxt_t *cxt, char *cname)
7a6a85bf
RG
3551{
3552 I32 len;
3553 char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */
3554 char *class = buf;
3555 unsigned int flags;
3556 I32 len2;
3557 SV *frozen;
3558 I32 len3 = 0;
3559 AV *av = 0;
3560 SV *hook;
3561 SV *sv;
3562 SV *rv;
3563 int obj_type;
7a6a85bf 3564 int clone = cxt->optype & ST_CLONE;
b12202d0
JH
3565 char mtype = '\0';
3566 unsigned int extra_type = 0;
7a6a85bf
RG
3567
3568 TRACEME(("retrieve_hook (#%d)", cxt->tagnum));
b12202d0 3569 ASSERT(!cname, ("no bless-into class given here, got %s", cname));
7a6a85bf
RG
3570
3571 /*
3572 * Read flags, which tell us about the type, and whether we need to recurse.
3573 */
3574
3575 GETMARK(flags);
3576
3577 /*
3578 * Create the (empty) object, and mark it as seen.
3579 *
3580 * This must be done now, because tags are incremented, and during
3581 * serialization, the object tag was affected before recursion could
3582 * take place.
3583 */
3584
3585 obj_type = flags & SHF_TYPE_MASK;
3586 switch (obj_type) {
3587 case SHT_SCALAR:
3588 sv = newSV(0);
3589 break;
3590 case SHT_ARRAY:
3591 sv = (SV *) newAV();
3592 break;
3593 case SHT_HASH:
3594 sv = (SV *) newHV();
3595 break;
b12202d0
JH
3596 case SHT_EXTRA:
3597 /*
3598 * Read <extra> flag to know the type of the object.
3599 * Record associated magic type for later.
3600 */
3601 GETMARK(extra_type);
3602 switch (extra_type) {
3603 case SHT_TSCALAR:
3604 sv = newSV(0);
3605 mtype = 'q';
3606 break;
3607 case SHT_TARRAY:
3608 sv = (SV *) newAV();
3609 mtype = 'P';
3610 break;
3611 case SHT_THASH:
3612 sv = (SV *) newHV();
3613 mtype = 'P';
3614 break;
3615 default:
3616 return retrieve_other(cxt, 0); /* Let it croak */
3617 }
3618 break;
7a6a85bf 3619 default:
b12202d0 3620 return retrieve_other(cxt, 0); /* Let it croak */
7a6a85bf 3621 }
b12202d0 3622 SEEN(sv, 0); /* Don't bless yet */
7a6a85bf
RG
3623
3624 /*
3625 * Whilst flags tell us to recurse, do so.
3626 *
3627 * We don't need to remember the addresses returned by retrieval, because
3628 * all the references will be obtained through indirection via the object
3629 * tags in the object-ID list.
3630 */
3631
3632 while (flags & SHF_NEED_RECURSE) {
3633 TRACEME(("retrieve_hook recursing..."));
b12202d0 3634 rv = retrieve(cxt, 0);
7a6a85bf
RG
3635 if (!rv)
3636 return (SV *) 0;
43d061fe
JH
3637 TRACEME(("retrieve_hook back with rv=0x%"UVxf,
3638 PTR2UV(rv)));
7a6a85bf
RG
3639 GETMARK(flags);
3640 }
3641
3642 if (flags & SHF_IDX_CLASSNAME) {
3643 SV **sva;
3644 I32 idx;
3645
3646 /*
3647 * Fetch index from `aclass'
3648 */
3649
3650 if (flags & SHF_LARGE_CLASSLEN)
3651 RLEN(idx);
3652 else
3653 GETMARK(idx);
3654
3655 sva = av_fetch(cxt->aclass, idx, FALSE);
3656 if (!sva)
e993d95c
JH
3657 CROAK(("Class name #%"IVdf" should have been seen already",
3658 (IV) idx));
7a6a85bf
RG
3659
3660 class = SvPVX(*sva); /* We know it's a PV, by construction */
3661 TRACEME(("class ID %d => %s", idx, class));
3662
3663 } else {
3664 /*
3665 * Decode class name length and read that name.
3666 *
3667 * NOTA BENE: even if the length is stored on one byte, we don't read
3668 * on the stack. Just like retrieve_blessed(), we limit the name to
3669 * LG_BLESS bytes. This is an arbitrary decision.
3670 */
3671
3672 if (flags & SHF_LARGE_CLASSLEN)
3673 RLEN(len);
3674 else
3675 GETMARK(len);
3676
3677 if (len > LG_BLESS) {
3678 TRACEME(("** allocating %d bytes for class name", len+1));
3679 New(10003, class, len+1, char);
3680 }
3681
3682 READ(class, len);
3683 class[len] = '\0'; /* Mark string end */
3684
3685 /*
3686 * Record new classname.
3687 */
3688
3689 if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len)))
3690 return (SV *) 0;
3691 }
3692
3693 TRACEME(("class name: %s", class));
3694
3695 /*
d1be9408 3696 * Decode user-frozen string length and read it in an SV.
7a6a85bf
RG
3697 *
3698 * For efficiency reasons, we read data directly into the SV buffer.
3699 * To understand that code, read retrieve_scalar()
3700 */
3701
3702 if (flags & SHF_LARGE_STRLEN)
3703 RLEN(len2);
3704 else
3705 GETMARK(len2);
3706
3707 frozen = NEWSV(10002, len2);
3708 if (len2) {
3709 SAFEREAD(SvPVX(frozen), len2, frozen);
3710 SvCUR_set(frozen, len2);
3711 *SvEND(frozen) = '\0';
3712 }
3713 (void) SvPOK_only(frozen); /* Validates string pointer */
dd19458b
JH
3714 if (cxt->s_tainted) /* Is input source tainted? */
3715 SvTAINT(frozen);
7a6a85bf
RG
3716
3717 TRACEME(("frozen string: %d bytes", len2));
3718
3719 /*
3720 * Decode object-ID list length, if present.
3721 */
3722
3723 if (flags & SHF_HAS_LIST) {
3724 if (flags & SHF_LARGE_LISTLEN)
3725 RLEN(len3);
3726 else
3727 GETMARK(len3);
3728 if (len3) {
3729 av = newAV();
3730 av_extend(av, len3 + 1); /* Leave room for [0] */
3731 AvFILLp(av) = len3; /* About to be filled anyway */
3732 }
3733 }
3734
3735 TRACEME(("has %d object IDs to link", len3));
3736
3737 /*
3738 * Read object-ID list into array.
3739 * Because we pre-extended it, we can cheat and fill it manually.
3740 *
3741 * We read object tags and we can convert them into SV* on the fly
3742 * because we know all the references listed in there (as tags)
3743 * have been already serialized, hence we have a valid correspondance
3744 * between each of those tags and the recreated SV.
3745 */
3746
3747 if (av) {
3748 SV **ary = AvARRAY(av);
3749 int i;
3750 for (i = 1; i <= len3; i++) { /* We leave [0] alone */
3751 I32 tag;
3752 SV **svh;
3753 SV *xsv;
3754
9e21b3d0 3755 READ_I32(tag);
7a6a85bf
RG
3756 tag = ntohl(tag);
3757 svh = av_fetch(cxt->aseen, tag, FALSE);
3758 if (!svh)
e993d95c
JH
3759 CROAK(("Object #%"IVdf" should have been retrieved already",
3760 (IV) tag));
7a6a85bf
RG
3761 xsv = *svh;
3762 ary[i] = SvREFCNT_inc(xsv);
3763 }
3764 }
3765
3766 /*
3767 * Bless the object and look up the STORABLE_thaw hook.
3768 */
3769
3770 BLESS(sv, class);
3771 hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw");
212e9bde
JH
3772 if (!hook) {
3773 /*
3774 * Hook not found. Maybe they did not require the module where this
3775 * hook is defined yet?
3776 *
3777 * If the require below succeeds, we'll be able to find the hook.
3778 * Still, it only works reliably when each class is defined in a
3779 * file of its own.
3780 */
3781
3782 SV *psv = newSVpvn("require ", 8);
3783 sv_catpv(psv, class);
3784
3785 TRACEME(("No STORABLE_thaw defined for objects of class %s", class));
3786 TRACEME(("Going to require module '%s' with '%s'", class, SvPVX(psv)));
3787
3788 perl_eval_sv(psv, G_DISCARD);
3789 sv_free(psv);
3790
3791 /*
3792 * We cache results of pkg_can, so we need to uncache before attempting