1 /*
2  *  Store and retrieve mechanism.
3  *
4  *  Copyright (c) 1995-2000, Raphael Manfredi
5  *
6  *  You may redistribute only under the same terms as Perl 5, as specified
7  *  in the README file that comes with the distribution.
8  *
9  */
10 
11 #define PERL_NO_GET_CONTEXT     /* we want efficiency */
12 #include <EXTERN.h>
13 #include <perl.h>
14 #include <XSUB.h>
15 
16 #ifndef PATCHLEVEL
17 #include <patchlevel.h>		/* Perl's one, needed since 5.6 */
18 #endif
19 
20 #if !defined(PERL_VERSION) || PERL_VERSION < 8
21 #include "ppport.h"             /* handle old perls */
22 #endif
23 
24 #if 0
25 #define DEBUGME /* Debug mode, turns assertions on as well */
26 #define DASSERT /* Assertion mode */
27 #endif
28 
29 /*
30  * Pre PerlIO time when none of USE_PERLIO and PERLIO_IS_STDIO is defined
31  * Provide them with the necessary defines so they can build with pre-5.004.
32  */
33 #ifndef USE_PERLIO
34 #ifndef PERLIO_IS_STDIO
35 #define PerlIO FILE
36 #define PerlIO_getc(x) getc(x)
37 #define PerlIO_putc(f,x) putc(x,f)
38 #define PerlIO_read(x,y,z) fread(y,1,z,x)
39 #define PerlIO_write(x,y,z) fwrite(y,1,z,x)
40 #define PerlIO_stdoutf printf
41 #endif	/* PERLIO_IS_STDIO */
42 #endif	/* USE_PERLIO */
43 
44 /*
45  * Earlier versions of perl might be used, we can't assume they have the latest!
46  */
47 
48 #ifndef PERL_VERSION		/* For perls < 5.6 */
49 #define PERL_VERSION PATCHLEVEL
50 #ifndef newRV_noinc
51 #define newRV_noinc(sv)		((Sv = newRV(sv)), --SvREFCNT(SvRV(Sv)), Sv)
52 #endif
53 #if (PATCHLEVEL <= 4)		/* Older perls (<= 5.004) lack PL_ namespace */
54 #define PL_sv_yes	sv_yes
55 #define PL_sv_no	sv_no
56 #define PL_sv_undef	sv_undef
57 #if (SUBVERSION <= 4)		/* 5.004_04 has been reported to lack newSVpvn */
58 #define newSVpvn newSVpv
59 #endif
60 #endif						/* PATCHLEVEL <= 4 */
61 #ifndef HvSHAREKEYS_off
62 #define HvSHAREKEYS_off(hv)	/* Ignore */
63 #endif
64 #ifndef AvFILLp				/* Older perls (<=5.003) lack AvFILLp */
65 #define AvFILLp AvFILL
66 #endif
67 typedef double NV;			/* Older perls lack the NV type */
68 #define	IVdf		"ld"	/* Various printf formats for Perl types */
69 #define	UVuf		"lu"
70 #define	UVof		"lo"
71 #define	UVxf		"lx"
72 #define INT2PTR(t,v) (t)(IV)(v)
73 #define PTR2UV(v)    (unsigned long)(v)
74 #endif						/* PERL_VERSION -- perls < 5.6 */
75 
76 #ifndef NVef				/* The following were not part of perl 5.6 */
77 #if defined(USE_LONG_DOUBLE) && \
78 	defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
79 #define NVef		PERL_PRIeldbl
80 #define NVff		PERL_PRIfldbl
81 #define NVgf		PERL_PRIgldbl
82 #else
83 #define	NVef		"e"
84 #define	NVff		"f"
85 #define	NVgf		"g"
86 #endif
87 #endif
88 
89 #ifndef SvRV_set
90 #define SvRV_set(sv, val) \
91     STMT_START { \
92         assert(SvTYPE(sv) >=  SVt_RV); \
93         (((XRV*)SvANY(sv))->xrv_rv = (val)); \
94     } STMT_END
95 #endif
96 
97 #ifndef PERL_UNUSED_DECL
98 #  ifdef HASATTRIBUTE
99 #    if (defined(__GNUC__) && defined(__cplusplus)) || defined(__INTEL_COMPILER)
100 #      define PERL_UNUSED_DECL
101 #    else
102 #      define PERL_UNUSED_DECL __attribute__((unused))
103 #    endif
104 #  else
105 #    define PERL_UNUSED_DECL
106 #  endif
107 #endif
108 
109 #ifndef dNOOP
110 #define dNOOP extern int Perl___notused PERL_UNUSED_DECL
111 #endif
112 
113 #ifndef dVAR
114 #define dVAR dNOOP
115 #endif
116 
117 #ifndef HvRITER_set
118 #  define HvRITER_set(hv,r)	(HvRITER(hv) = r)
119 #endif
120 #ifndef HvEITER_set
121 #  define HvEITER_set(hv,r)	(HvEITER(hv) = r)
122 #endif
123 
124 #ifndef HvRITER_get
125 #  define HvRITER_get HvRITER
126 #endif
127 #ifndef HvEITER_get
128 #  define HvEITER_get HvEITER
129 #endif
130 
131 #ifndef HvNAME_get
132 #define HvNAME_get HvNAME
133 #endif
134 
135 #ifndef HvPLACEHOLDERS_get
136 #  define HvPLACEHOLDERS_get HvPLACEHOLDERS
137 #endif
138 
139 #ifdef DEBUGME
140 
141 #ifndef DASSERT
142 #define DASSERT
143 #endif
144 
145 /*
146  * TRACEME() will only output things when the $Storable::DEBUGME is true.
147  */
148 
149 #define TRACEME(x)										\
150   STMT_START {											\
151 	if (SvTRUE(perl_get_sv("Storable::DEBUGME", TRUE)))	\
152 		{ PerlIO_stdoutf x; PerlIO_stdoutf("\n"); }		\
153   } STMT_END
154 #else
155 #define TRACEME(x)
156 #endif	/* DEBUGME */
157 
158 #ifdef DASSERT
159 #define ASSERT(x,y)										\
160   STMT_START {											\
161 	if (!(x)) {												\
162 		PerlIO_stdoutf("ASSERT FAILED (\"%s\", line %d): ",	\
163 			__FILE__, __LINE__);							\
164 		PerlIO_stdoutf y; PerlIO_stdoutf("\n");				\
165 	}														\
166   } STMT_END
167 #else
168 #define ASSERT(x,y)
169 #endif
170 
171 /*
172  * Type markers.
173  */
174 
175 #define C(x) ((char) (x))	/* For markers with dynamic retrieval handling */
176 
177 #define SX_OBJECT	C(0)	/* Already stored object */
178 #define SX_LSCALAR	C(1)	/* Scalar (large binary) follows (length, data) */
179 #define SX_ARRAY	C(2)	/* Array forthcominng (size, item list) */
180 #define SX_HASH		C(3)	/* Hash forthcoming (size, key/value pair list) */
181 #define SX_REF		C(4)	/* Reference to object forthcoming */
182 #define SX_UNDEF	C(5)	/* Undefined scalar */
183 #define SX_INTEGER	C(6)	/* Integer forthcoming */
184 #define SX_DOUBLE	C(7)	/* Double forthcoming */
185 #define SX_BYTE		C(8)	/* (signed) byte forthcoming */
186 #define SX_NETINT	C(9)	/* Integer in network order forthcoming */
187 #define SX_SCALAR	C(10)	/* Scalar (binary, small) follows (length, data) */
188 #define SX_TIED_ARRAY	C(11)	/* Tied array forthcoming */
189 #define SX_TIED_HASH	C(12)	/* Tied hash forthcoming */
190 #define SX_TIED_SCALAR	C(13)	/* Tied scalar forthcoming */
191 #define SX_SV_UNDEF	C(14)	/* Perl's immortal PL_sv_undef */
192 #define SX_SV_YES	C(15)	/* Perl's immortal PL_sv_yes */
193 #define SX_SV_NO	C(16)	/* Perl's immortal PL_sv_no */
194 #define SX_BLESS	C(17)	/* Object is blessed */
195 #define SX_IX_BLESS	C(18)	/* Object is blessed, classname given by index */
196 #define SX_HOOK		C(19)	/* Stored via hook, user-defined */
197 #define SX_OVERLOAD	C(20)	/* Overloaded reference */
198 #define SX_TIED_KEY	C(21)	/* Tied magic key forthcoming */
199 #define SX_TIED_IDX	C(22)	/* Tied magic index forthcoming */
200 #define SX_UTF8STR	C(23)	/* UTF-8 string forthcoming (small) */
201 #define SX_LUTF8STR	C(24)	/* UTF-8 string forthcoming (large) */
202 #define SX_FLAG_HASH	C(25)	/* Hash with flags forthcoming (size, flags, key/flags/value triplet list) */
203 #define SX_CODE         C(26)   /* Code references as perl source code */
204 #define SX_WEAKREF	C(27)	/* Weak reference to object forthcoming */
205 #define SX_WEAKOVERLOAD	C(28)	/* Overloaded weak reference */
206 #define SX_ERROR	C(29)	/* Error */
207 
208 /*
209  * Those are only used to retrieve "old" pre-0.6 binary images.
210  */
211 #define SX_ITEM		'i'		/* An array item introducer */
212 #define SX_IT_UNDEF	'I'		/* Undefined array item */
213 #define SX_KEY		'k'		/* A hash key introducer */
214 #define SX_VALUE	'v'		/* A hash value introducer */
215 #define SX_VL_UNDEF	'V'		/* Undefined hash value */
216 
217 /*
218  * Those are only used to retrieve "old" pre-0.7 binary images
219  */
220 
221 #define SX_CLASS	'b'		/* Object is blessed, class name length <255 */
222 #define SX_LG_CLASS	'B'		/* Object is blessed, class name length >255 */
223 #define SX_STORED	'X'		/* End of object */
224 
225 /*
226  * Limits between short/long length representation.
227  */
228 
229 #define LG_SCALAR	255		/* Large scalar length limit */
230 #define LG_BLESS	127		/* Large classname bless limit */
231 
232 /*
233  * Operation types
234  */
235 
236 #define ST_STORE	0x1		/* Store operation */
237 #define ST_RETRIEVE	0x2		/* Retrieval operation */
238 #define ST_CLONE	0x4		/* Deep cloning operation */
239 
240 /*
241  * The following structure is used for hash table key retrieval. Since, when
242  * retrieving objects, we'll be facing blessed hash references, it's best
243  * to pre-allocate that buffer once and resize it as the need arises, never
244  * freeing it (keys will be saved away someplace else anyway, so even large
245  * keys are not enough a motivation to reclaim that space).
246  *
247  * This structure is also used for memory store/retrieve operations which
248  * happen in a fixed place before being malloc'ed elsewhere if persistency
249  * is required. Hence the aptr pointer.
250  */
251 struct extendable {
252 	char *arena;		/* Will hold hash key strings, resized as needed */
253 	STRLEN asiz;		/* Size of aforementionned buffer */
254 	char *aptr;			/* Arena pointer, for in-place read/write ops */
255 	char *aend;			/* First invalid address */
256 };
257 
258 /*
259  * At store time:
260  * A hash table records the objects which have already been stored.
261  * Those are referred to as SX_OBJECT in the file, and their "tag" (i.e.
262  * an arbitrary sequence number) is used to identify them.
263  *
264  * At retrieve time:
265  * An array table records the objects which have already been retrieved,
266  * as seen by the tag determind by counting the objects themselves. The
267  * reference to that retrieved object is kept in the table, and is returned
268  * when an SX_OBJECT is found bearing that same tag.
269  *
270  * The same processing is used to record "classname" for blessed objects:
271  * indexing by a hash at store time, and via an array at retrieve time.
272  */
273 
274 typedef unsigned long stag_t;	/* Used by pre-0.6 binary format */
275 
276 /*
277  * The following "thread-safe" related defines were contributed by
278  * Murray Nesbitt <murray@activestate.com> and integrated by RAM, who
279  * only renamed things a little bit to ensure consistency with surrounding
280  * code.	-- RAM, 14/09/1999
281  *
282  * The original patch suffered from the fact that the stcxt_t structure
283  * was global.  Murray tried to minimize the impact on the code as much as
284  * possible.
285  *
286  * Starting with 0.7, Storable can be re-entrant, via the STORABLE_xxx hooks
287  * on objects.  Therefore, the notion of context needs to be generalized,
288  * threading or not.
289  */
290 
291 #define MY_VERSION "Storable(" XS_VERSION ")"
292 
293 
294 /*
295  * Conditional UTF8 support.
296  *
297  */
298 #ifdef SvUTF8_on
299 #define STORE_UTF8STR(pv, len)	STORE_PV_LEN(pv, len, SX_UTF8STR, SX_LUTF8STR)
300 #define HAS_UTF8_SCALARS
301 #ifdef HeKUTF8
302 #define HAS_UTF8_HASHES
303 #define HAS_UTF8_ALL
304 #else
305 /* 5.6 perl has utf8 scalars but not hashes */
306 #endif
307 #else
308 #define SvUTF8(sv) 0
309 #define STORE_UTF8STR(pv, len) CROAK(("panic: storing UTF8 in non-UTF8 perl"))
310 #endif
311 #ifndef HAS_UTF8_ALL
312 #define UTF8_CROAK() CROAK(("Cannot retrieve UTF8 data in non-UTF8 perl"))
313 #endif
314 #ifndef SvWEAKREF
315 #define WEAKREF_CROAK() CROAK(("Cannot retrieve weak references in this perl"))
316 #endif
317 
318 #ifdef HvPLACEHOLDERS
319 #define HAS_RESTRICTED_HASHES
320 #else
321 #define HVhek_PLACEHOLD	0x200
322 #define RESTRICTED_HASH_CROAK() CROAK(("Cannot retrieve restricted hash"))
323 #endif
324 
325 #ifdef HvHASKFLAGS
326 #define HAS_HASH_KEY_FLAGS
327 #endif
328 
329 #ifdef ptr_table_new
330 #define USE_PTR_TABLE
331 #endif
332 
333 /*
334  * Fields s_tainted and s_dirty are prefixed with s_ because Perl's include
335  * files remap tainted and dirty when threading is enabled.  That's bad for
336  * perl to remap such common words.	-- RAM, 29/09/00
337  */
338 
339 struct stcxt;
340 typedef struct stcxt {
341 	int entry;			/* flags recursion */
342 	int optype;			/* type of traversal operation */
343 	/* which objects have been seen, store time.
344 	   tags are numbers, which are cast to (SV *) and stored directly */
345 #ifdef USE_PTR_TABLE
346 	/* use pseen if we have ptr_tables. We have to store tag+1, because
347 	   tag numbers start at 0, and we can't store (SV *) 0 in a ptr_table
348 	   without it being confused for a fetch lookup failure.  */
349 	struct ptr_tbl *pseen;
350 	/* Still need hseen for the 0.6 file format code. */
351 #endif
352 	HV *hseen;
353 	AV *hook_seen;		/* which SVs were returned by STORABLE_freeze() */
354 	AV *aseen;			/* which objects have been seen, retrieve time */
355 	IV where_is_undef;		/* index in aseen of PL_sv_undef */
356 	HV *hclass;			/* which classnames have been seen, store time */
357 	AV *aclass;			/* which classnames have been seen, retrieve time */
358 	HV *hook;			/* cache for hook methods per class name */
359 	IV tagnum;			/* incremented at store time for each seen object */
360 	IV classnum;		/* incremented at store time for each seen classname */
361 	int netorder;		/* true if network order used */
362 	int s_tainted;		/* true if input source is tainted, at retrieve time */
363 	int forgive_me;		/* whether to be forgiving... */
364 	int deparse;        /* whether to deparse code refs */
365 	SV *eval;           /* whether to eval source code */
366 	int canonical;		/* whether to store hashes sorted by key */
367 #ifndef HAS_RESTRICTED_HASHES
368         int derestrict;         /* whether to downgrade restrcted hashes */
369 #endif
370 #ifndef HAS_UTF8_ALL
371         int use_bytes;         /* whether to bytes-ify utf8 */
372 #endif
373         int accept_future_minor; /* croak immediately on future minor versions?  */
374 	int s_dirty;		/* context is dirty due to CROAK() -- can be cleaned */
375 	int membuf_ro;		/* true means membuf is read-only and msaved is rw */
376 	struct extendable keybuf;	/* for hash key retrieval */
377 	struct extendable membuf;	/* for memory store/retrieve operations */
378 	struct extendable msaved;	/* where potentially valid mbuf is saved */
379 	PerlIO *fio;		/* where I/O are performed, NULL for memory */
380 	int ver_major;		/* major of version for retrieved object */
381 	int ver_minor;		/* minor of version for retrieved object */
382 	SV *(**retrieve_vtbl)(pTHX_ struct stcxt *, char *);	/* retrieve dispatch table */
383 	SV *prev;		/* contexts chained backwards in real recursion */
384 	SV *my_sv;		/* the blessed scalar who's SvPVX() I am */
385 } stcxt_t;
386 
387 #define NEW_STORABLE_CXT_OBJ(cxt)					\
388   STMT_START {										\
389 	SV *self = newSV(sizeof(stcxt_t) - 1);			\
390 	SV *my_sv = newRV_noinc(self);					\
391 	sv_bless(my_sv, gv_stashpv("Storable::Cxt", TRUE));	\
392 	cxt = (stcxt_t *)SvPVX(self);					\
393 	Zero(cxt, 1, stcxt_t);							\
394 	cxt->my_sv = my_sv;								\
395   } STMT_END
396 
397 #if defined(MULTIPLICITY) || defined(PERL_OBJECT) || defined(PERL_CAPI)
398 
399 #if (PATCHLEVEL <= 4) && (SUBVERSION < 68)
400 #define dSTCXT_SV 									\
401 	SV *perinterp_sv = perl_get_sv(MY_VERSION, FALSE)
402 #else	/* >= perl5.004_68 */
403 #define dSTCXT_SV									\
404 	SV *perinterp_sv = *hv_fetch(PL_modglobal,		\
405 		MY_VERSION, sizeof(MY_VERSION)-1, TRUE)
406 #endif	/* < perl5.004_68 */
407 
408 #define dSTCXT_PTR(T,name)							\
409 	T name = ((perinterp_sv && SvIOK(perinterp_sv) && SvIVX(perinterp_sv)	\
410 				? (T)SvPVX(SvRV(INT2PTR(SV*,SvIVX(perinterp_sv)))) : (T) 0))
411 #define dSTCXT										\
412 	dSTCXT_SV;										\
413 	dSTCXT_PTR(stcxt_t *, cxt)
414 
415 #define INIT_STCXT							\
416 	dSTCXT;									\
417 	NEW_STORABLE_CXT_OBJ(cxt);				\
418 	sv_setiv(perinterp_sv, PTR2IV(cxt->my_sv))
419 
420 #define SET_STCXT(x)								\
421   STMT_START {										\
422 	dSTCXT_SV;										\
423 	sv_setiv(perinterp_sv, PTR2IV(x->my_sv));		\
424   } STMT_END
425 
426 #else /* !MULTIPLICITY && !PERL_OBJECT && !PERL_CAPI */
427 
428 static stcxt_t *Context_ptr = NULL;
429 #define dSTCXT			stcxt_t *cxt = Context_ptr
430 #define SET_STCXT(x)		Context_ptr = x
431 #define INIT_STCXT						\
432 	dSTCXT;								\
433 	NEW_STORABLE_CXT_OBJ(cxt);			\
434 	SET_STCXT(cxt)
435 
436 
437 #endif /* MULTIPLICITY || PERL_OBJECT || PERL_CAPI */
438 
439 /*
440  * KNOWN BUG:
441  *   Croaking implies a memory leak, since we don't use setjmp/longjmp
442  *   to catch the exit and free memory used during store or retrieve
443  *   operations.  This is not too difficult to fix, but I need to understand
444  *   how Perl does it, and croaking is exceptional anyway, so I lack the
445  *   motivation to do it.
446  *
447  * The current workaround is to mark the context as dirty when croaking,
448  * so that data structures can be freed whenever we renter Storable code
449  * (but only *then*: it's a workaround, not a fix).
450  *
451  * This is also imperfect, because we don't really know how far they trapped
452  * the croak(), and when we were recursing, we won't be able to clean anything
453  * but the topmost context stacked.
454  */
455 
456 #define CROAK(x)	STMT_START { cxt->s_dirty = 1; croak x; } STMT_END
457 
458 /*
459  * End of "thread-safe" related definitions.
460  */
461 
462 /*
463  * LOW_32BITS
464  *
465  * Keep only the low 32 bits of a pointer (used for tags, which are not
466  * really pointers).
467  */
468 
469 #if PTRSIZE <= 4
470 #define LOW_32BITS(x)	((I32) (x))
471 #else
472 #define LOW_32BITS(x)	((I32) ((unsigned long) (x) & 0xffffffffUL))
473 #endif
474 
475 /*
476  * oI, oS, oC
477  *
478  * Hack for Crays, where sizeof(I32) == 8, and which are big-endians.
479  * Used in the WLEN and RLEN macros.
480  */
481 
482 #if INTSIZE > 4
483 #define oI(x)	((I32 *) ((char *) (x) + 4))
484 #define oS(x)	((x) - 4)
485 #define oC(x)	(x = 0)
486 #define CRAY_HACK
487 #else
488 #define oI(x)	(x)
489 #define oS(x)	(x)
490 #define oC(x)
491 #endif
492 
493 /*
494  * key buffer handling
495  */
496 #define kbuf	(cxt->keybuf).arena
497 #define ksiz	(cxt->keybuf).asiz
498 #define KBUFINIT()						\
499   STMT_START {							\
500 	if (!kbuf) {						\
501 		TRACEME(("** allocating kbuf of 128 bytes")); \
502 		New(10003, kbuf, 128, char);	\
503 		ksiz = 128;						\
504 	}									\
505   } STMT_END
506 #define KBUFCHK(x)				\
507   STMT_START {					\
508 	if (x >= ksiz) {			\
509 		TRACEME(("** extending kbuf to %d bytes (had %d)", x+1, ksiz)); \
510 		Renew(kbuf, x+1, char);	\
511 		ksiz = x+1;				\
512 	}							\
513   } STMT_END
514 
515 /*
516  * memory buffer handling
517  */
518 #define mbase	(cxt->membuf).arena
519 #define msiz	(cxt->membuf).asiz
520 #define mptr	(cxt->membuf).aptr
521 #define mend	(cxt->membuf).aend
522 
523 #define MGROW	(1 << 13)
524 #define MMASK	(MGROW - 1)
525 
526 #define round_mgrow(x)	\
527 	((unsigned long) (((unsigned long) (x) + MMASK) & ~MMASK))
528 #define trunc_int(x)	\
529 	((unsigned long) ((unsigned long) (x) & ~(sizeof(int)-1)))
530 #define int_aligned(x)	\
531 	((unsigned long) (x) == trunc_int(x))
532 
533 #define MBUF_INIT(x)					\
534   STMT_START {							\
535 	if (!mbase) {						\
536 		TRACEME(("** allocating mbase of %d bytes", MGROW)); \
537 		New(10003, mbase, MGROW, char);	\
538 		msiz = (STRLEN)MGROW;					\
539 	}									\
540 	mptr = mbase;						\
541 	if (x)								\
542 		mend = mbase + x;				\
543 	else								\
544 		mend = mbase + msiz;			\
545   } STMT_END
546 
547 #define MBUF_TRUNC(x)	mptr = mbase + x
548 #define MBUF_SIZE()		(mptr - mbase)
549 
550 /*
551  * MBUF_SAVE_AND_LOAD
552  * MBUF_RESTORE
553  *
554  * Those macros are used in do_retrieve() to save the current memory
555  * buffer into cxt->msaved, before MBUF_LOAD() can be used to retrieve
556  * data from a string.
557  */
558 #define MBUF_SAVE_AND_LOAD(in)			\
559   STMT_START {							\
560 	ASSERT(!cxt->membuf_ro, ("mbase not already saved")); \
561 	cxt->membuf_ro = 1;					\
562 	TRACEME(("saving mbuf"));			\
563 	StructCopy(&cxt->membuf, &cxt->msaved, struct extendable); \
564 	MBUF_LOAD(in);						\
565   } STMT_END
566 
567 #define MBUF_RESTORE() 					\
568   STMT_START {							\
569 	ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
570 	cxt->membuf_ro = 0;					\
571 	TRACEME(("restoring mbuf"));		\
572 	StructCopy(&cxt->msaved, &cxt->membuf, struct extendable); \
573   } STMT_END
574 
575 /*
576  * Use SvPOKp(), because SvPOK() fails on tainted scalars.
577  * See store_scalar() for other usage of this workaround.
578  */
579 #define MBUF_LOAD(v) 					\
580   STMT_START {							\
581 	ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
582 	if (!SvPOKp(v))						\
583 		CROAK(("Not a scalar string"));	\
584 	mptr = mbase = SvPV(v, msiz);		\
585 	mend = mbase + msiz;				\
586   } STMT_END
587 
588 #define MBUF_XTEND(x) 				\
589   STMT_START {						\
590 	int nsz = (int) round_mgrow((x)+msiz);	\
591 	int offset = mptr - mbase;		\
592 	ASSERT(!cxt->membuf_ro, ("mbase is not read-only")); \
593 	TRACEME(("** extending mbase from %d to %d bytes (wants %d new)", \
594 		msiz, nsz, (x)));			\
595 	Renew(mbase, nsz, char);		\
596 	msiz = nsz;						\
597 	mptr = mbase + offset;			\
598 	mend = mbase + nsz;				\
599   } STMT_END
600 
601 #define MBUF_CHK(x) 				\
602   STMT_START {						\
603 	if ((mptr + (x)) > mend)		\
604 		MBUF_XTEND(x);				\
605   } STMT_END
606 
607 #define MBUF_GETC(x) 				\
608   STMT_START {						\
609 	if (mptr < mend)				\
610 		x = (int) (unsigned char) *mptr++;	\
611 	else							\
612 		return (SV *) 0;			\
613   } STMT_END
614 
615 #ifdef CRAY_HACK
616 #define MBUF_GETINT(x) 					\
617   STMT_START {							\
618 	oC(x);								\
619 	if ((mptr + 4) <= mend) {			\
620 		memcpy(oI(&x), mptr, 4);		\
621 		mptr += 4;						\
622 	} else								\
623 		return (SV *) 0;				\
624   } STMT_END
625 #else
626 #define MBUF_GETINT(x) 					\
627   STMT_START {							\
628 	if ((mptr + sizeof(int)) <= mend) {	\
629 		if (int_aligned(mptr))			\
630 			x = *(int *) mptr;			\
631 		else							\
632 			memcpy(&x, mptr, sizeof(int));	\
633 		mptr += sizeof(int);			\
634 	} else								\
635 		return (SV *) 0;				\
636   } STMT_END
637 #endif
638 
639 #define MBUF_READ(x,s) 				\
640   STMT_START {						\
641 	if ((mptr + (s)) <= mend) {		\
642 		memcpy(x, mptr, s);			\
643 		mptr += s;					\
644 	} else							\
645 		return (SV *) 0;			\
646   } STMT_END
647 
648 #define MBUF_SAFEREAD(x,s,z) 		\
649   STMT_START {						\
650 	if ((mptr + (s)) <= mend) {		\
651 		memcpy(x, mptr, s);			\
652 		mptr += s;					\
653 	} else {						\
654 		sv_free(z);					\
655 		return (SV *) 0;			\
656 	}								\
657   } STMT_END
658 
659 #define MBUF_PUTC(c) 				\
660   STMT_START {						\
661 	if (mptr < mend)				\
662 		*mptr++ = (char) c;			\
663 	else {							\
664 		MBUF_XTEND(1);				\
665 		*mptr++ = (char) c;			\
666 	}								\
667   } STMT_END
668 
669 #ifdef CRAY_HACK
670 #define MBUF_PUTINT(i) 				\
671   STMT_START {						\
672 	MBUF_CHK(4);					\
673 	memcpy(mptr, oI(&i), 4);		\
674 	mptr += 4;						\
675   } STMT_END
676 #else
677 #define MBUF_PUTINT(i) 				\
678   STMT_START {						\
679 	MBUF_CHK(sizeof(int));			\
680 	if (int_aligned(mptr))			\
681 		*(int *) mptr = i;			\
682 	else							\
683 		memcpy(mptr, &i, sizeof(int));	\
684 	mptr += sizeof(int);			\
685   } STMT_END
686 #endif
687 
688 #define MBUF_WRITE(x,s) 			\
689   STMT_START {						\
690 	MBUF_CHK(s);					\
691 	memcpy(mptr, x, s);				\
692 	mptr += s;						\
693   } STMT_END
694 
695 /*
696  * Possible return values for sv_type().
697  */
698 
699 #define svis_REF		0
700 #define svis_SCALAR		1
701 #define svis_ARRAY		2
702 #define svis_HASH		3
703 #define svis_TIED		4
704 #define svis_TIED_ITEM	5
705 #define svis_CODE		6
706 #define svis_OTHER		7
707 
708 /*
709  * Flags for SX_HOOK.
710  */
711 
712 #define SHF_TYPE_MASK		0x03
713 #define SHF_LARGE_CLASSLEN	0x04
714 #define SHF_LARGE_STRLEN	0x08
715 #define SHF_LARGE_LISTLEN	0x10
716 #define SHF_IDX_CLASSNAME	0x20
717 #define SHF_NEED_RECURSE	0x40
718 #define SHF_HAS_LIST		0x80
719 
720 /*
721  * Types for SX_HOOK (last 2 bits in flags).
722  */
723 
724 #define SHT_SCALAR			0
725 #define SHT_ARRAY			1
726 #define SHT_HASH			2
727 #define SHT_EXTRA			3		/* Read extra byte for type */
728 
729 /*
730  * The following are held in the "extra byte"...
731  */
732 
733 #define SHT_TSCALAR			4		/* 4 + 0 -- tied scalar */
734 #define SHT_TARRAY			5		/* 4 + 1 -- tied array */
735 #define SHT_THASH			6		/* 4 + 2 -- tied hash */
736 
737 /*
738  * per hash flags for flagged hashes
739  */
740 
741 #define SHV_RESTRICTED		0x01
742 
743 /*
744  * per key flags for flagged hashes
745  */
746 
747 #define SHV_K_UTF8		0x01
748 #define SHV_K_WASUTF8		0x02
749 #define SHV_K_LOCKED		0x04
750 #define SHV_K_ISSV		0x08
751 #define SHV_K_PLACEHOLDER	0x10
752 
753 /*
754  * Before 0.6, the magic string was "perl-store" (binary version number 0).
755  *
756  * Since 0.6 introduced many binary incompatibilities, the magic string has
757  * been changed to "pst0" to allow an old image to be properly retrieved by
758  * a newer Storable, but ensure a newer image cannot be retrieved with an
759  * older version.
760  *
761  * At 0.7, objects are given the ability to serialize themselves, and the
762  * set of markers is extended, backward compatibility is not jeopardized,
763  * so the binary version number could have remained unchanged.  To correctly
764  * spot errors if a file making use of 0.7-specific extensions is given to
765  * 0.6 for retrieval, the binary version was moved to "2".  And I'm introducing
766  * a "minor" version, to better track this kind of evolution from now on.
767  *
768  */
769 static const char old_magicstr[] = "perl-store"; /* Magic number before 0.6 */
770 static const char magicstr[] = "pst0";		 /* Used as a magic number */
771 
772 #define MAGICSTR_BYTES  'p','s','t','0'
773 #define OLDMAGICSTR_BYTES  'p','e','r','l','-','s','t','o','r','e'
774 
775 /* 5.6.x introduced the ability to have IVs as long long.
776    However, Configure still defined BYTEORDER based on the size of a long.
777    Storable uses the BYTEORDER value as part of the header, but doesn't
778    explicity store sizeof(IV) anywhere in the header.  Hence on 5.6.x built
779    with IV as long long on a platform that uses Configure (ie most things
780    except VMS and Windows) headers are identical for the different IV sizes,
781    despite the files containing some fields based on sizeof(IV)
782    Erk. Broken-ness.
783    5.8 is consistent - the following redifinition kludge is only needed on
784    5.6.x, but the interwork is needed on 5.8 while data survives in files
785    with the 5.6 header.
786 
787 */
788 
789 #if defined (IVSIZE) && (IVSIZE == 8) && (LONGSIZE == 4)
790 #ifndef NO_56_INTERWORK_KLUDGE
791 #define USE_56_INTERWORK_KLUDGE
792 #endif
793 #if BYTEORDER == 0x1234
794 #undef BYTEORDER
795 #define BYTEORDER 0x12345678
796 #else
797 #if BYTEORDER == 0x4321
798 #undef BYTEORDER
799 #define BYTEORDER 0x87654321
800 #endif
801 #endif
802 #endif
803 
804 #if BYTEORDER == 0x1234
805 #define BYTEORDER_BYTES  '1','2','3','4'
806 #else
807 #if BYTEORDER == 0x12345678
808 #define BYTEORDER_BYTES  '1','2','3','4','5','6','7','8'
809 #ifdef USE_56_INTERWORK_KLUDGE
810 #define BYTEORDER_BYTES_56  '1','2','3','4'
811 #endif
812 #else
813 #if BYTEORDER == 0x87654321
814 #define BYTEORDER_BYTES  '8','7','6','5','4','3','2','1'
815 #ifdef USE_56_INTERWORK_KLUDGE
816 #define BYTEORDER_BYTES_56  '4','3','2','1'
817 #endif
818 #else
819 #if BYTEORDER == 0x4321
820 #define BYTEORDER_BYTES  '4','3','2','1'
821 #else
822 #error Unknown byteorder. Please append your byteorder to Storable.xs
823 #endif
824 #endif
825 #endif
826 #endif
827 
828 static const char byteorderstr[] = {BYTEORDER_BYTES, 0};
829 #ifdef USE_56_INTERWORK_KLUDGE
830 static const char byteorderstr_56[] = {BYTEORDER_BYTES_56, 0};
831 #endif
832 
833 #define STORABLE_BIN_MAJOR	2		/* Binary major "version" */
834 #define STORABLE_BIN_MINOR	7		/* Binary minor "version" */
835 
836 #if (PATCHLEVEL <= 5)
837 #define STORABLE_BIN_WRITE_MINOR	4
838 #else
839 /*
840  * Perl 5.6.0 onwards can do weak references.
841 */
842 #define STORABLE_BIN_WRITE_MINOR	7
843 #endif /* (PATCHLEVEL <= 5) */
844 
845 #if (PATCHLEVEL < 8 || (PATCHLEVEL == 8 && SUBVERSION < 1))
846 #define PL_sv_placeholder PL_sv_undef
847 #endif
848 
849 /*
850  * Useful store shortcuts...
851  */
852 
853 /*
854  * Note that if you put more than one mark for storing a particular
855  * type of thing, *and* in the retrieve_foo() function you mark both
856  * the thingy's you get off with SEEN(), you *must* increase the
857  * tagnum with cxt->tagnum++ along with this macro!
858  *     - samv 20Jan04
859  */
860 #define PUTMARK(x) 							\
861   STMT_START {								\
862 	if (!cxt->fio)							\
863 		MBUF_PUTC(x);						\
864 	else if (PerlIO_putc(cxt->fio, x) == EOF)	\
865 		return -1;							\
866   } STMT_END
867 
868 #define WRITE_I32(x)					\
869   STMT_START {							\
870 	ASSERT(sizeof(x) == sizeof(I32), ("writing an I32"));	\
871 	if (!cxt->fio)						\
872 		MBUF_PUTINT(x);					\
873 	else if (PerlIO_write(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
874 		return -1;					\
875   } STMT_END
876 
877 #ifdef HAS_HTONL
878 #define WLEN(x)						\
879   STMT_START {						\
880 	if (cxt->netorder) {			\
881 		int y = (int) htonl(x);		\
882 		if (!cxt->fio)				\
883 			MBUF_PUTINT(y);			\
884 		else if (PerlIO_write(cxt->fio,oI(&y),oS(sizeof(y))) != oS(sizeof(y))) \
885 			return -1;				\
886 	} else {						\
887 		if (!cxt->fio)				\
888 			MBUF_PUTINT(x);			\
889 		else if (PerlIO_write(cxt->fio,oI(&x),oS(sizeof(x))) != oS(sizeof(x))) \
890 			return -1;				\
891 	}								\
892   } STMT_END
893 #else
894 #define WLEN(x)	WRITE_I32(x)
895 #endif
896 
897 #define WRITE(x,y) 							\
898   STMT_START {								\
899 	if (!cxt->fio)							\
900 		MBUF_WRITE(x,y);					\
901 	else if (PerlIO_write(cxt->fio, x, y) != y)	\
902 		return -1;							\
903   } STMT_END
904 
905 #define STORE_PV_LEN(pv, len, small, large)			\
906   STMT_START {							\
907 	if (len <= LG_SCALAR) {				\
908 		unsigned char clen = (unsigned char) len;	\
909 		PUTMARK(small);					\
910 		PUTMARK(clen);					\
911 		if (len)						\
912 			WRITE(pv, len);				\
913 	} else {							\
914 		PUTMARK(large);					\
915 		WLEN(len);						\
916 		WRITE(pv, len);					\
917 	}									\
918   } STMT_END
919 
920 #define STORE_SCALAR(pv, len)	STORE_PV_LEN(pv, len, SX_SCALAR, SX_LSCALAR)
921 
922 /*
923  * Store &PL_sv_undef in arrays without recursing through store().
924  */
925 #define STORE_SV_UNDEF() 					\
926   STMT_START {							\
927 	cxt->tagnum++;						\
928 	PUTMARK(SX_SV_UNDEF);					\
929   } STMT_END
930 
931 /*
932  * Useful retrieve shortcuts...
933  */
934 
935 #define GETCHAR() \
936 	(cxt->fio ? PerlIO_getc(cxt->fio) : (mptr >= mend ? EOF : (int) *mptr++))
937 
938 #define GETMARK(x) 								\
939   STMT_START {									\
940 	if (!cxt->fio)								\
941 		MBUF_GETC(x);							\
942 	else if ((int) (x = PerlIO_getc(cxt->fio)) == EOF)	\
943 		return (SV *) 0;						\
944   } STMT_END
945 
946 #define READ_I32(x)						\
947   STMT_START {							\
948 	ASSERT(sizeof(x) == sizeof(I32), ("reading an I32"));	\
949 	oC(x);								\
950 	if (!cxt->fio)						\
951 		MBUF_GETINT(x);					\
952 	else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x)))	\
953 		return (SV *) 0;				\
954   } STMT_END
955 
956 #ifdef HAS_NTOHL
957 #define RLEN(x)							\
958   STMT_START {							\
959 	oC(x);								\
960 	if (!cxt->fio)						\
961 		MBUF_GETINT(x);					\
962 	else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x)))	\
963 		return (SV *) 0;				\
964 	if (cxt->netorder)					\
965 		x = (int) ntohl(x);				\
966   } STMT_END
967 #else
968 #define RLEN(x) READ_I32(x)
969 #endif
970 
971 #define READ(x,y) 							\
972   STMT_START {								\
973 	if (!cxt->fio)							\
974 		MBUF_READ(x, y);					\
975 	else if (PerlIO_read(cxt->fio, x, y) != y)	\
976 		return (SV *) 0;					\
977   } STMT_END
978 
979 #define SAFEREAD(x,y,z)		 					\
980   STMT_START {									\
981 	if (!cxt->fio)								\
982 		MBUF_SAFEREAD(x,y,z);					\
983 	else if (PerlIO_read(cxt->fio, x, y) != y)	 {	\
984 		sv_free(z);								\
985 		return (SV *) 0;						\
986 	}											\
987   } STMT_END
988 
989 /*
990  * This macro is used at retrieve time, to remember where object 'y', bearing a
991  * given tag 'tagnum', has been retrieved. Next time we see an SX_OBJECT marker,
992  * we'll therefore know where it has been retrieved and will be able to
993  * share the same reference, as in the original stored memory image.
994  *
995  * We also need to bless objects ASAP for hooks (which may compute "ref $x"
996  * on the objects given to STORABLE_thaw and expect that to be defined), and
997  * also for overloaded objects (for which we might not find the stash if the
998  * object is not blessed yet--this might occur for overloaded objects that
999  * refer to themselves indirectly: if we blessed upon return from a sub
1000  * retrieve(), the SX_OBJECT marker we'd found could not have overloading
1001  * restored on it because the underlying object would not be blessed yet!).
1002  *
1003  * To achieve that, the class name of the last retrieved object is passed down
1004  * recursively, and the first SEEN() call for which the class name is not NULL
1005  * will bless the object.
1006  *
1007  * i should be true iff sv is immortal (ie PL_sv_yes, PL_sv_no or PL_sv_undef)
1008  */
1009 #define SEEN(y,c,i) 							\
1010   STMT_START {								\
1011 	if (!y)									\
1012 		return (SV *) 0;					\
1013 	if (av_store(cxt->aseen, cxt->tagnum++, i ? (SV*)(y) : SvREFCNT_inc(y)) == 0) \
1014 		return (SV *) 0;					\
1015 	TRACEME(("aseen(#%d) = 0x%"UVxf" (refcnt=%d)", cxt->tagnum-1, \
1016 		 PTR2UV(y), SvREFCNT(y)-1));		\
1017 	if (c)									\
1018 		BLESS((SV *) (y), c);				\
1019   } STMT_END
1020 
1021 /*
1022  * Bless `s' in `p', via a temporary reference, required by sv_bless().
1023  */
1024 #define BLESS(s,p) 							\
1025   STMT_START {								\
1026 	SV *ref;								\
1027 	HV *stash;								\
1028 	TRACEME(("blessing 0x%"UVxf" in %s", PTR2UV(s), (p))); \
1029 	stash = gv_stashpv((p), TRUE);			\
1030 	ref = newRV_noinc(s);					\
1031 	(void) sv_bless(ref, stash);			\
1032 	SvRV_set(ref, NULL);						\
1033 	SvREFCNT_dec(ref);						\
1034   } STMT_END
1035 /*
1036  * sort (used in store_hash) - conditionally use qsort when
1037  * sortsv is not available ( <= 5.6.1 ).
1038  */
1039 
1040 #if (PATCHLEVEL <= 6)
1041 
1042 #if defined(USE_ITHREADS)
1043 
1044 #define STORE_HASH_SORT \
1045         ENTER; { \
1046         PerlInterpreter *orig_perl = PERL_GET_CONTEXT; \
1047         SAVESPTR(orig_perl); \
1048         PERL_SET_CONTEXT(aTHX); \
1049         qsort((char *) AvARRAY(av), len, sizeof(SV *), sortcmp); \
1050         } LEAVE;
1051 
1052 #else /* ! USE_ITHREADS */
1053 
1054 #define STORE_HASH_SORT \
1055         qsort((char *) AvARRAY(av), len, sizeof(SV *), sortcmp);
1056 
1057 #endif  /* USE_ITHREADS */
1058 
1059 #else /* PATCHLEVEL > 6 */
1060 
1061 #define STORE_HASH_SORT \
1062         sortsv(AvARRAY(av), len, Perl_sv_cmp);
1063 
1064 #endif /* PATCHLEVEL <= 6 */
1065 
1066 static int store(pTHX_ stcxt_t *cxt, SV *sv);
1067 static SV *retrieve(pTHX_ stcxt_t *cxt, char *cname);
1068 
1069 /*
1070  * Dynamic dispatching table for SV store.
1071  */
1072 
1073 static int store_ref(pTHX_ stcxt_t *cxt, SV *sv);
1074 static int store_scalar(pTHX_ stcxt_t *cxt, SV *sv);
1075 static int store_array(pTHX_ stcxt_t *cxt, AV *av);
1076 static int store_hash(pTHX_ stcxt_t *cxt, HV *hv);
1077 static int store_tied(pTHX_ stcxt_t *cxt, SV *sv);
1078 static int store_tied_item(pTHX_ stcxt_t *cxt, SV *sv);
1079 static int store_code(pTHX_ stcxt_t *cxt, CV *cv);
1080 static int store_other(pTHX_ stcxt_t *cxt, SV *sv);
1081 static int store_blessed(pTHX_ stcxt_t *cxt, SV *sv, int type, HV *pkg);
1082 
1083 typedef int (*sv_store_t)(pTHX_ stcxt_t *cxt, SV *sv);
1084 
1085 static sv_store_t sv_store[] = {
1086 	(sv_store_t)store_ref,		/* svis_REF */
1087 	(sv_store_t)store_scalar,	/* svis_SCALAR */
1088 	(sv_store_t)store_array,	/* svis_ARRAY */
1089 	(sv_store_t)store_hash,		/* svis_HASH */
1090 	(sv_store_t)store_tied,		/* svis_TIED */
1091 	(sv_store_t)store_tied_item,	/* svis_TIED_ITEM */
1092 	(sv_store_t)store_code,		/* svis_CODE */
1093 	(sv_store_t)store_other,	/* svis_OTHER */
1094 };
1095 
1096 #define SV_STORE(x)	(*sv_store[x])
1097 
1098 /*
1099  * Dynamic dispatching tables for SV retrieval.
1100  */
1101 
1102 static SV *retrieve_lscalar(pTHX_ stcxt_t *cxt, char *cname);
1103 static SV *retrieve_lutf8str(pTHX_ stcxt_t *cxt, char *cname);
1104 static SV *old_retrieve_array(pTHX_ stcxt_t *cxt, char *cname);
1105 static SV *old_retrieve_hash(pTHX_ stcxt_t *cxt, char *cname);
1106 static SV *retrieve_ref(pTHX_ stcxt_t *cxt, char *cname);
1107 static SV *retrieve_undef(pTHX_ stcxt_t *cxt, char *cname);
1108 static SV *retrieve_integer(pTHX_ stcxt_t *cxt, char *cname);
1109 static SV *retrieve_double(pTHX_ stcxt_t *cxt, char *cname);
1110 static SV *retrieve_byte(pTHX_ stcxt_t *cxt, char *cname);
1111 static SV *retrieve_netint(pTHX_ stcxt_t *cxt, char *cname);
1112 static SV *retrieve_scalar(pTHX_ stcxt_t *cxt, char *cname);
1113 static SV *retrieve_utf8str(pTHX_ stcxt_t *cxt, char *cname);
1114 static SV *retrieve_tied_array(pTHX_ stcxt_t *cxt, char *cname);
1115 static SV *retrieve_tied_hash(pTHX_ stcxt_t *cxt, char *cname);
1116 static SV *retrieve_tied_scalar(pTHX_ stcxt_t *cxt, char *cname);
1117 static SV *retrieve_other(pTHX_ stcxt_t *cxt, char *cname);
1118 
1119 typedef SV* (*sv_retrieve_t)(pTHX_ stcxt_t *cxt, char *name);
1120 
1121 static const sv_retrieve_t sv_old_retrieve[] = {
1122 	0,			/* SX_OBJECT -- entry unused dynamically */
1123 	(sv_retrieve_t)retrieve_lscalar,	/* SX_LSCALAR */
1124 	(sv_retrieve_t)old_retrieve_array,	/* SX_ARRAY -- for pre-0.6 binaries */
1125 	(sv_retrieve_t)old_retrieve_hash,	/* SX_HASH -- for pre-0.6 binaries */
1126 	(sv_retrieve_t)retrieve_ref,		/* SX_REF */
1127 	(sv_retrieve_t)retrieve_undef,		/* SX_UNDEF */
1128 	(sv_retrieve_t)retrieve_integer,	/* SX_INTEGER */
1129 	(sv_retrieve_t)retrieve_double,		/* SX_DOUBLE */
1130 	(sv_retrieve_t)retrieve_byte,		/* SX_BYTE */
1131 	(sv_retrieve_t)retrieve_netint,		/* SX_NETINT */
1132 	(sv_retrieve_t)retrieve_scalar,		/* SX_SCALAR */
1133 	(sv_retrieve_t)retrieve_tied_array,	/* SX_ARRAY */
1134 	(sv_retrieve_t)retrieve_tied_hash,	/* SX_HASH */
1135 	(sv_retrieve_t)retrieve_tied_scalar,	/* SX_SCALAR */
1136 	(sv_retrieve_t)retrieve_other,	/* SX_SV_UNDEF not supported */
1137 	(sv_retrieve_t)retrieve_other,	/* SX_SV_YES not supported */
1138 	(sv_retrieve_t)retrieve_other,	/* SX_SV_NO not supported */
1139 	(sv_retrieve_t)retrieve_other,	/* SX_BLESS not supported */
1140 	(sv_retrieve_t)retrieve_other,	/* SX_IX_BLESS not supported */
1141 	(sv_retrieve_t)retrieve_other,	/* SX_HOOK not supported */
1142 	(sv_retrieve_t)retrieve_other,	/* SX_OVERLOADED not supported */
1143 	(sv_retrieve_t)retrieve_other,	/* SX_TIED_KEY not supported */
1144 	(sv_retrieve_t)retrieve_other,	/* SX_TIED_IDX not supported */
1145 	(sv_retrieve_t)retrieve_other,	/* SX_UTF8STR not supported */
1146 	(sv_retrieve_t)retrieve_other,	/* SX_LUTF8STR not supported */
1147 	(sv_retrieve_t)retrieve_other,	/* SX_FLAG_HASH not supported */
1148 	(sv_retrieve_t)retrieve_other,	/* SX_CODE not supported */
1149 	(sv_retrieve_t)retrieve_other,	/* SX_WEAKREF not supported */
1150 	(sv_retrieve_t)retrieve_other,	/* SX_WEAKOVERLOAD not supported */
1151 	(sv_retrieve_t)retrieve_other,	/* SX_ERROR */
1152 };
1153 
1154 static SV *retrieve_array(pTHX_ stcxt_t *cxt, char *cname);
1155 static SV *retrieve_hash(pTHX_ stcxt_t *cxt, char *cname);
1156 static SV *retrieve_sv_undef(pTHX_ stcxt_t *cxt, char *cname);
1157 static SV *retrieve_sv_yes(pTHX_ stcxt_t *cxt, char *cname);
1158 static SV *retrieve_sv_no(pTHX_ stcxt_t *cxt, char *cname);
1159 static SV *retrieve_blessed(pTHX_ stcxt_t *cxt, char *cname);
1160 static SV *retrieve_idx_blessed(pTHX_ stcxt_t *cxt, char *cname);
1161 static SV *retrieve_hook(pTHX_ stcxt_t *cxt, char *cname);
1162 static SV *retrieve_overloaded(pTHX_ stcxt_t *cxt, char *cname);
1163 static SV *retrieve_tied_key(pTHX_ stcxt_t *cxt, char *cname);
1164 static SV *retrieve_tied_idx(pTHX_ stcxt_t *cxt, char *cname);
1165 static SV *retrieve_flag_hash(pTHX_ stcxt_t *cxt, char *cname);
1166 static SV *retrieve_code(pTHX_ stcxt_t *cxt, char *cname);
1167 static SV *retrieve_weakref(pTHX_ stcxt_t *cxt, char *cname);
1168 static SV *retrieve_weakoverloaded(pTHX_ stcxt_t *cxt, char *cname);
1169 
1170 static const sv_retrieve_t sv_retrieve[] = {
1171 	0,			/* SX_OBJECT -- entry unused dynamically */
1172 	(sv_retrieve_t)retrieve_lscalar,	/* SX_LSCALAR */
1173 	(sv_retrieve_t)retrieve_array,		/* SX_ARRAY */
1174 	(sv_retrieve_t)retrieve_hash,		/* SX_HASH */
1175 	(sv_retrieve_t)retrieve_ref,		/* SX_REF */
1176 	(sv_retrieve_t)retrieve_undef,		/* SX_UNDEF */
1177 	(sv_retrieve_t)retrieve_integer,	/* SX_INTEGER */
1178 	(sv_retrieve_t)retrieve_double,		/* SX_DOUBLE */
1179 	(sv_retrieve_t)retrieve_byte,		/* SX_BYTE */
1180 	(sv_retrieve_t)retrieve_netint,		/* SX_NETINT */
1181 	(sv_retrieve_t)retrieve_scalar,		/* SX_SCALAR */
1182 	(sv_retrieve_t)retrieve_tied_array,	/* SX_ARRAY */
1183 	(sv_retrieve_t)retrieve_tied_hash,	/* SX_HASH */
1184 	(sv_retrieve_t)retrieve_tied_scalar,	/* SX_SCALAR */
1185 	(sv_retrieve_t)retrieve_sv_undef,	/* SX_SV_UNDEF */
1186 	(sv_retrieve_t)retrieve_sv_yes,		/* SX_SV_YES */
1187 	(sv_retrieve_t)retrieve_sv_no,		/* SX_SV_NO */
1188 	(sv_retrieve_t)retrieve_blessed,	/* SX_BLESS */
1189 	(sv_retrieve_t)retrieve_idx_blessed,	/* SX_IX_BLESS */
1190 	(sv_retrieve_t)retrieve_hook,		/* SX_HOOK */
1191 	(sv_retrieve_t)retrieve_overloaded,	/* SX_OVERLOAD */
1192 	(sv_retrieve_t)retrieve_tied_key,	/* SX_TIED_KEY */
1193 	(sv_retrieve_t)retrieve_tied_idx,	/* SX_TIED_IDX */
1194 	(sv_retrieve_t)retrieve_utf8str,	/* SX_UTF8STR  */
1195 	(sv_retrieve_t)retrieve_lutf8str,	/* SX_LUTF8STR */
1196 	(sv_retrieve_t)retrieve_flag_hash,	/* SX_HASH */
1197 	(sv_retrieve_t)retrieve_code,		/* SX_CODE */
1198 	(sv_retrieve_t)retrieve_weakref,	/* SX_WEAKREF */
1199 	(sv_retrieve_t)retrieve_weakoverloaded,	/* SX_WEAKOVERLOAD */
1200 	(sv_retrieve_t)retrieve_other,		/* SX_ERROR */
1201 };
1202 
1203 #define RETRIEVE(c,x) (*(c)->retrieve_vtbl[(x) >= SX_ERROR ? SX_ERROR : (x)])
1204 
1205 static SV *mbuf2sv(pTHX);
1206 
1207 /***
1208  *** Context management.
1209  ***/
1210 
1211 /*
1212  * init_perinterp
1213  *
1214  * Called once per "thread" (interpreter) to initialize some global context.
1215  */
init_perinterp(pTHX)1216 static void init_perinterp(pTHX)
1217 {
1218     INIT_STCXT;
1219 
1220     cxt->netorder = 0;		/* true if network order used */
1221     cxt->forgive_me = -1;	/* whether to be forgiving... */
1222     cxt->accept_future_minor = -1; /* would otherwise occur too late */
1223 }
1224 
1225 /*
1226  * reset_context
1227  *
1228  * Called at the end of every context cleaning, to perform common reset
1229  * operations.
1230  */
reset_context(stcxt_t * cxt)1231 static void reset_context(stcxt_t *cxt)
1232 {
1233 	cxt->entry = 0;
1234 	cxt->s_dirty = 0;
1235 	cxt->optype &= ~(ST_STORE|ST_RETRIEVE);		/* Leave ST_CLONE alone */
1236 }
1237 
1238 /*
1239  * init_store_context
1240  *
1241  * Initialize a new store context for real recursion.
1242  */
init_store_context(pTHX_ stcxt_t * cxt,PerlIO * f,int optype,int network_order)1243 static void init_store_context(
1244         pTHX_
1245 	stcxt_t *cxt,
1246 	PerlIO *f,
1247 	int optype,
1248 	int network_order)
1249 {
1250 	TRACEME(("init_store_context"));
1251 
1252 	cxt->netorder = network_order;
1253 	cxt->forgive_me = -1;			/* Fetched from perl if needed */
1254 	cxt->deparse = -1;				/* Idem */
1255 	cxt->eval = NULL;				/* Idem */
1256 	cxt->canonical = -1;			/* Idem */
1257 	cxt->tagnum = -1;				/* Reset tag numbers */
1258 	cxt->classnum = -1;				/* Reset class numbers */
1259 	cxt->fio = f;					/* Where I/O are performed */
1260 	cxt->optype = optype;			/* A store, or a deep clone */
1261 	cxt->entry = 1;					/* No recursion yet */
1262 
1263 	/*
1264 	 * The `hseen' table is used to keep track of each SV stored and their
1265 	 * associated tag numbers is special. It is "abused" because the
1266 	 * values stored are not real SV, just integers cast to (SV *),
1267 	 * which explains the freeing below.
1268 	 *
1269 	 * It is also one possible bottlneck to achieve good storing speed,
1270 	 * so the "shared keys" optimization is turned off (unlikely to be
1271 	 * of any use here), and the hash table is "pre-extended". Together,
1272 	 * those optimizations increase the throughput by 12%.
1273 	 */
1274 
1275 #ifdef USE_PTR_TABLE
1276 	cxt->pseen = ptr_table_new();
1277 	cxt->hseen = 0;
1278 #else
1279 	cxt->hseen = newHV();			/* Table where seen objects are stored */
1280 	HvSHAREKEYS_off(cxt->hseen);
1281 #endif
1282 	/*
1283 	 * The following does not work well with perl5.004_04, and causes
1284 	 * a core dump later on, in a completely unrelated spot, which
1285 	 * makes me think there is a memory corruption going on.
1286 	 *
1287 	 * Calling hv_ksplit(hseen, HBUCKETS) instead of manually hacking
1288 	 * it below does not make any difference. It seems to work fine
1289 	 * with perl5.004_68 but given the probable nature of the bug,
1290 	 * that does not prove anything.
1291 	 *
1292 	 * It's a shame because increasing the amount of buckets raises
1293 	 * store() throughput by 5%, but until I figure this out, I can't
1294 	 * allow for this to go into production.
1295 	 *
1296 	 * It is reported fixed in 5.005, hence the #if.
1297 	 */
1298 #if PERL_VERSION >= 5
1299 #define HBUCKETS	4096				/* Buckets for %hseen */
1300 #ifndef USE_PTR_TABLE
1301 	HvMAX(cxt->hseen) = HBUCKETS - 1;	/* keys %hseen = $HBUCKETS; */
1302 #endif
1303 #endif
1304 
1305 	/*
1306 	 * The `hclass' hash uses the same settings as `hseen' above, but it is
1307 	 * used to assign sequential tags (numbers) to class names for blessed
1308 	 * objects.
1309 	 *
1310 	 * We turn the shared key optimization on.
1311 	 */
1312 
1313 	cxt->hclass = newHV();			/* Where seen classnames are stored */
1314 
1315 #if PERL_VERSION >= 5
1316 	HvMAX(cxt->hclass) = HBUCKETS - 1;	/* keys %hclass = $HBUCKETS; */
1317 #endif
1318 
1319 	/*
1320 	 * The `hook' hash table is used to keep track of the references on
1321 	 * the STORABLE_freeze hook routines, when found in some class name.
1322 	 *
1323 	 * It is assumed that the inheritance tree will not be changed during
1324 	 * storing, and that no new method will be dynamically created by the
1325 	 * hooks.
1326 	 */
1327 
1328 	cxt->hook = newHV();			/* Table where hooks are cached */
1329 
1330 	/*
1331 	 * The `hook_seen' array keeps track of all the SVs returned by
1332 	 * STORABLE_freeze hooks for us to serialize, so that they are not
1333 	 * reclaimed until the end of the serialization process.  Each SV is
1334 	 * only stored once, the first time it is seen.
1335 	 */
1336 
1337 	cxt->hook_seen = newAV();		/* Lists SVs returned by STORABLE_freeze */
1338 }
1339 
1340 /*
1341  * clean_store_context
1342  *
1343  * Clean store context by
1344  */
clean_store_context(pTHX_ stcxt_t * cxt)1345 static void clean_store_context(pTHX_ stcxt_t *cxt)
1346 {
1347 	HE *he;
1348 
1349 	TRACEME(("clean_store_context"));
1350 
1351 	ASSERT(cxt->optype & ST_STORE, ("was performing a store()"));
1352 
1353 	/*
1354 	 * Insert real values into hashes where we stored faked pointers.
1355 	 */
1356 
1357 #ifndef USE_PTR_TABLE
1358 	if (cxt->hseen) {
1359 		hv_iterinit(cxt->hseen);
1360 		while ((he = hv_iternext(cxt->hseen)))	/* Extra () for -Wall, grr.. */
1361 			HeVAL(he) = &PL_sv_undef;
1362 	}
1363 #endif
1364 
1365 	if (cxt->hclass) {
1366 		hv_iterinit(cxt->hclass);
1367 		while ((he = hv_iternext(cxt->hclass)))	/* Extra () for -Wall, grr.. */
1368 			HeVAL(he) = &PL_sv_undef;
1369 	}
1370 
1371 	/*
1372 	 * And now dispose of them...
1373 	 *
1374 	 * The surrounding if() protection has been added because there might be
1375 	 * some cases where this routine is called more than once, during
1376 	 * exceptionnal events.  This was reported by Marc Lehmann when Storable
1377 	 * is executed from mod_perl, and the fix was suggested by him.
1378 	 * 		-- RAM, 20/12/2000
1379 	 */
1380 
1381 #ifdef USE_PTR_TABLE
1382 	if (cxt->pseen) {
1383 		struct ptr_tbl *pseen = cxt->pseen;
1384 		cxt->pseen = 0;
1385 		ptr_table_free(pseen);
1386 	}
1387 	assert(!cxt->hseen);
1388 #else
1389 	if (cxt->hseen) {
1390 		HV *hseen = cxt->hseen;
1391 		cxt->hseen = 0;
1392 		hv_undef(hseen);
1393 		sv_free((SV *) hseen);
1394 	}
1395 #endif
1396 
1397 	if (cxt->hclass) {
1398 		HV *hclass = cxt->hclass;
1399 		cxt->hclass = 0;
1400 		hv_undef(hclass);
1401 		sv_free((SV *) hclass);
1402 	}
1403 
1404 	if (cxt->hook) {
1405 		HV *hook = cxt->hook;
1406 		cxt->hook = 0;
1407 		hv_undef(hook);
1408 		sv_free((SV *) hook);
1409 	}
1410 
1411 	if (cxt->hook_seen) {
1412 		AV *hook_seen = cxt->hook_seen;
1413 		cxt->hook_seen = 0;
1414 		av_undef(hook_seen);
1415 		sv_free((SV *) hook_seen);
1416 	}
1417 
1418 	cxt->forgive_me = -1;			/* Fetched from perl if needed */
1419 	cxt->deparse = -1;				/* Idem */
1420 	if (cxt->eval) {
1421 	    SvREFCNT_dec(cxt->eval);
1422 	}
1423 	cxt->eval = NULL;				/* Idem */
1424 	cxt->canonical = -1;			/* Idem */
1425 
1426 	reset_context(cxt);
1427 }
1428 
1429 /*
1430  * init_retrieve_context
1431  *
1432  * Initialize a new retrieve context for real recursion.
1433  */
init_retrieve_context(pTHX_ stcxt_t * cxt,int optype,int is_tainted)1434 static void init_retrieve_context(pTHX_ stcxt_t *cxt, int optype, int is_tainted)
1435 {
1436 	TRACEME(("init_retrieve_context"));
1437 
1438 	/*
1439 	 * The hook hash table is used to keep track of the references on
1440 	 * the STORABLE_thaw hook routines, when found in some class name.
1441 	 *
1442 	 * It is assumed that the inheritance tree will not be changed during
1443 	 * storing, and that no new method will be dynamically created by the
1444 	 * hooks.
1445 	 */
1446 
1447 	cxt->hook  = newHV();			/* Caches STORABLE_thaw */
1448 
1449 #ifdef USE_PTR_TABLE
1450 	cxt->pseen = 0;
1451 #endif
1452 
1453 	/*
1454 	 * If retrieving an old binary version, the cxt->retrieve_vtbl variable
1455 	 * was set to sv_old_retrieve. We'll need a hash table to keep track of
1456 	 * the correspondance between the tags and the tag number used by the
1457 	 * new retrieve routines.
1458 	 */
1459 
1460 	cxt->hseen = (((void*)cxt->retrieve_vtbl == (void*)sv_old_retrieve)
1461 		      ? newHV() : 0);
1462 
1463 	cxt->aseen = newAV();			/* Where retrieved objects are kept */
1464 	cxt->where_is_undef = -1;		/* Special case for PL_sv_undef */
1465 	cxt->aclass = newAV();			/* Where seen classnames are kept */
1466 	cxt->tagnum = 0;				/* Have to count objects... */
1467 	cxt->classnum = 0;				/* ...and class names as well */
1468 	cxt->optype = optype;
1469 	cxt->s_tainted = is_tainted;
1470 	cxt->entry = 1;					/* No recursion yet */
1471 #ifndef HAS_RESTRICTED_HASHES
1472         cxt->derestrict = -1;		/* Fetched from perl if needed */
1473 #endif
1474 #ifndef HAS_UTF8_ALL
1475         cxt->use_bytes = -1;		/* Fetched from perl if needed */
1476 #endif
1477         cxt->accept_future_minor = -1;	/* Fetched from perl if needed */
1478 }
1479 
1480 /*
1481  * clean_retrieve_context
1482  *
1483  * Clean retrieve context by
1484  */
clean_retrieve_context(pTHX_ stcxt_t * cxt)1485 static void clean_retrieve_context(pTHX_ stcxt_t *cxt)
1486 {
1487 	TRACEME(("clean_retrieve_context"));
1488 
1489 	ASSERT(cxt->optype & ST_RETRIEVE, ("was performing a retrieve()"));
1490 
1491 	if (cxt->aseen) {
1492 		AV *aseen = cxt->aseen;
1493 		cxt->aseen = 0;
1494 		av_undef(aseen);
1495 		sv_free((SV *) aseen);
1496 	}
1497 	cxt->where_is_undef = -1;
1498 
1499 	if (cxt->aclass) {
1500 		AV *aclass = cxt->aclass;
1501 		cxt->aclass = 0;
1502 		av_undef(aclass);
1503 		sv_free((SV *) aclass);
1504 	}
1505 
1506 	if (cxt->hook) {
1507 		HV *hook = cxt->hook;
1508 		cxt->hook = 0;
1509 		hv_undef(hook);
1510 		sv_free((SV *) hook);
1511 	}
1512 
1513 	if (cxt->hseen) {
1514 		HV *hseen = cxt->hseen;
1515 		cxt->hseen = 0;
1516 		hv_undef(hseen);
1517 		sv_free((SV *) hseen);		/* optional HV, for backward compat. */
1518 	}
1519 
1520 #ifndef HAS_RESTRICTED_HASHES
1521         cxt->derestrict = -1;		/* Fetched from perl if needed */
1522 #endif
1523 #ifndef HAS_UTF8_ALL
1524         cxt->use_bytes = -1;		/* Fetched from perl if needed */
1525 #endif
1526         cxt->accept_future_minor = -1;	/* Fetched from perl if needed */
1527 
1528 	reset_context(cxt);
1529 }
1530 
1531 /*
1532  * clean_context
1533  *
1534  * A workaround for the CROAK bug: cleanup the last context.
1535  */
clean_context(pTHX_ stcxt_t * cxt)1536 static void clean_context(pTHX_ stcxt_t *cxt)
1537 {
1538 	TRACEME(("clean_context"));
1539 
1540 	ASSERT(cxt->s_dirty, ("dirty context"));
1541 
1542 	if (cxt->membuf_ro)
1543 		MBUF_RESTORE();
1544 
1545 	ASSERT(!cxt->membuf_ro, ("mbase is not read-only"));
1546 
1547 	if (cxt->optype & ST_RETRIEVE)
1548 		clean_retrieve_context(aTHX_ cxt);
1549 	else if (cxt->optype & ST_STORE)
1550 		clean_store_context(aTHX_ cxt);
1551 	else
1552 		reset_context(cxt);
1553 
1554 	ASSERT(!cxt->s_dirty, ("context is clean"));
1555 	ASSERT(cxt->entry == 0, ("context is reset"));
1556 }
1557 
1558 /*
1559  * allocate_context
1560  *
1561  * Allocate a new context and push it on top of the parent one.
1562  * This new context is made globally visible via SET_STCXT().
1563  */
allocate_context(pTHX_ stcxt_t * parent_cxt)1564 static stcxt_t *allocate_context(pTHX_ stcxt_t *parent_cxt)
1565 {
1566 	stcxt_t *cxt;
1567 
1568 	TRACEME(("allocate_context"));
1569 
1570 	ASSERT(!parent_cxt->s_dirty, ("parent context clean"));
1571 
1572 	NEW_STORABLE_CXT_OBJ(cxt);
1573 	cxt->prev = parent_cxt->my_sv;
1574 	SET_STCXT(cxt);
1575 
1576 	ASSERT(!cxt->s_dirty, ("clean context"));
1577 
1578 	return cxt;
1579 }
1580 
1581 /*
1582  * free_context
1583  *
1584  * Free current context, which cannot be the "root" one.
1585  * Make the context underneath globally visible via SET_STCXT().
1586  */
free_context(pTHX_ stcxt_t * cxt)1587 static void free_context(pTHX_ stcxt_t *cxt)
1588 {
1589 	stcxt_t *prev = (stcxt_t *)(cxt->prev ? SvPVX(SvRV(cxt->prev)) : 0);
1590 
1591 	TRACEME(("free_context"));
1592 
1593 	ASSERT(!cxt->s_dirty, ("clean context"));
1594 	ASSERT(prev, ("not freeing root context"));
1595 
1596 	SvREFCNT_dec(cxt->my_sv);
1597 	SET_STCXT(prev);
1598 
1599 	ASSERT(cxt, ("context not void"));
1600 }
1601 
1602 /***
1603  *** Predicates.
1604  ***/
1605 
1606 /*
1607  * is_storing
1608  *
1609  * Tells whether we're in the middle of a store operation.
1610  */
is_storing(pTHX)1611 int is_storing(pTHX)
1612 {
1613 	dSTCXT;
1614 
1615 	return cxt->entry && (cxt->optype & ST_STORE);
1616 }
1617 
1618 /*
1619  * is_retrieving
1620  *
1621  * Tells whether we're in the middle of a retrieve operation.
1622  */
is_retrieving(pTHX)1623 int is_retrieving(pTHX)
1624 {
1625 	dSTCXT;
1626 
1627 	return cxt->entry && (cxt->optype & ST_RETRIEVE);
1628 }
1629 
1630 /*
1631  * last_op_in_netorder
1632  *
1633  * Returns whether last operation was made using network order.
1634  *
1635  * This is typically out-of-band information that might prove useful
1636  * to people wishing to convert native to network order data when used.
1637  */
last_op_in_netorder(pTHX)1638 int last_op_in_netorder(pTHX)
1639 {
1640 	dSTCXT;
1641 
1642 	return cxt->netorder;
1643 }
1644 
1645 /***
1646  *** Hook lookup and calling routines.
1647  ***/
1648 
1649 /*
1650  * pkg_fetchmeth
1651  *
1652  * A wrapper on gv_fetchmethod_autoload() which caches results.
1653  *
1654  * Returns the routine reference as an SV*, or null if neither the package
1655  * nor its ancestors know about the method.
1656  */
pkg_fetchmeth(pTHX_ HV * cache,HV * pkg,char * method)1657 static SV *pkg_fetchmeth(
1658         pTHX_
1659 	HV *cache,
1660 	HV *pkg,
1661 	char *method)
1662 {
1663 	GV *gv;
1664 	SV *sv;
1665 	const char *hvname = HvNAME_get(pkg);
1666 
1667 
1668 	/*
1669 	 * The following code is the same as the one performed by UNIVERSAL::can
1670 	 * in the Perl core.
1671 	 */
1672 
1673 	gv = gv_fetchmethod_autoload(pkg, method, FALSE);
1674 	if (gv && isGV(gv)) {
1675 		sv = newRV((SV*) GvCV(gv));
1676 		TRACEME(("%s->%s: 0x%"UVxf, hvname, method, PTR2UV(sv)));
1677 	} else {
1678 		sv = newSVsv(&PL_sv_undef);
1679 		TRACEME(("%s->%s: not found", hvname, method));
1680 	}
1681 
1682 	/*
1683 	 * Cache the result, ignoring failure: if we can't store the value,
1684 	 * it just won't be cached.
1685 	 */
1686 
1687 	(void) hv_store(cache, hvname, strlen(hvname), sv, 0);
1688 
1689 	return SvOK(sv) ? sv : (SV *) 0;
1690 }
1691 
1692 /*
1693  * pkg_hide
1694  *
1695  * Force cached value to be undef: hook ignored even if present.
1696  */
pkg_hide(pTHX_ HV * cache,HV * pkg,char * method)1697 static void pkg_hide(
1698         pTHX_
1699 	HV *cache,
1700 	HV *pkg,
1701 	char *method)
1702 {
1703 	const char *hvname = HvNAME_get(pkg);
1704 	(void) hv_store(cache,
1705 		hvname, strlen(hvname), newSVsv(&PL_sv_undef), 0);
1706 }
1707 
1708 /*
1709  * pkg_uncache
1710  *
1711  * Discard cached value: a whole fetch loop will be retried at next lookup.
1712  */
pkg_uncache(pTHX_ HV * cache,HV * pkg,char * method)1713 static void pkg_uncache(
1714         pTHX_
1715 	HV *cache,
1716 	HV *pkg,
1717 	char *method)
1718 {
1719 	const char *hvname = HvNAME_get(pkg);
1720 	(void) hv_delete(cache, hvname, strlen(hvname), G_DISCARD);
1721 }
1722 
1723 /*
1724  * pkg_can
1725  *
1726  * Our own "UNIVERSAL::can", which caches results.
1727  *
1728  * Returns the routine reference as an SV*, or null if the object does not
1729  * know about the method.
1730  */
pkg_can(pTHX_ HV * cache,HV * pkg,char * method)1731 static SV *pkg_can(
1732         pTHX_
1733 	HV *cache,
1734 	HV *pkg,
1735 	char *method)
1736 {
1737 	SV **svh;
1738 	SV *sv;
1739 	const char *hvname = HvNAME_get(pkg);
1740 
1741 	TRACEME(("pkg_can for %s->%s", hvname, method));
1742 
1743 	/*
1744 	 * Look into the cache to see whether we already have determined
1745 	 * where the routine was, if any.
1746 	 *
1747 	 * NOTA BENE: we don't use `method' at all in our lookup, since we know
1748 	 * that only one hook (i.e. always the same) is cached in a given cache.
1749 	 */
1750 
1751 	svh = hv_fetch(cache, hvname, strlen(hvname), FALSE);
1752 	if (svh) {
1753 		sv = *svh;
1754 		if (!SvOK(sv)) {
1755 			TRACEME(("cached %s->%s: not found", hvname, method));
1756 			return (SV *) 0;
1757 		} else {
1758 			TRACEME(("cached %s->%s: 0x%"UVxf,
1759 				hvname, method, PTR2UV(sv)));
1760 			return sv;
1761 		}
1762 	}
1763 
1764 	TRACEME(("not cached yet"));
1765 	return pkg_fetchmeth(aTHX_ cache, pkg, method);		/* Fetch and cache */
1766 }
1767 
1768 /*
1769  * scalar_call
1770  *
1771  * Call routine as obj->hook(av) in scalar context.
1772  * Propagates the single returned value if not called in void context.
1773  */
scalar_call(pTHX_ SV * obj,SV * hook,int cloning,AV * av,I32 flags)1774 static SV *scalar_call(
1775         pTHX_
1776 	SV *obj,
1777 	SV *hook,
1778 	int cloning,
1779 	AV *av,
1780 	I32 flags)
1781 {
1782 	dSP;
1783 	int count;
1784 	SV *sv = 0;
1785 
1786 	TRACEME(("scalar_call (cloning=%d)", cloning));
1787 
1788 	ENTER;
1789 	SAVETMPS;
1790 
1791 	PUSHMARK(sp);
1792 	XPUSHs(obj);
1793 	XPUSHs(sv_2mortal(newSViv(cloning)));		/* Cloning flag */
1794 	if (av) {
1795 		SV **ary = AvARRAY(av);
1796 		int cnt = AvFILLp(av) + 1;
1797 		int i;
1798 		XPUSHs(ary[0]);							/* Frozen string */
1799 		for (i = 1; i < cnt; i++) {
1800 			TRACEME(("pushing arg #%d (0x%"UVxf")...",
1801 				 i, PTR2UV(ary[i])));
1802 			XPUSHs(sv_2mortal(newRV(ary[i])));
1803 		}
1804 	}
1805 	PUTBACK;
1806 
1807 	TRACEME(("calling..."));
1808 	count = perl_call_sv(hook, flags);		/* Go back to Perl code */
1809 	TRACEME(("count = %d", count));
1810 
1811 	SPAGAIN;
1812 
1813 	if (count) {
1814 		sv = POPs;
1815 		SvREFCNT_inc(sv);		/* We're returning it, must stay alive! */
1816 	}
1817 
1818 	PUTBACK;
1819 	FREETMPS;
1820 	LEAVE;
1821 
1822 	return sv;
1823 }
1824 
1825 /*
1826  * array_call
1827  *
1828  * Call routine obj->hook(cloning) in list context.
1829  * Returns the list of returned values in an array.
1830  */
array_call(pTHX_ SV * obj,SV * hook,int cloning)1831 static AV *array_call(
1832         pTHX_
1833 	SV *obj,
1834 	SV *hook,
1835 	int cloning)
1836 {
1837 	dSP;
1838 	int count;
1839 	AV *av;
1840 	int i;
1841 
1842 	TRACEME(("array_call (cloning=%d)", cloning));
1843 
1844 	ENTER;
1845 	SAVETMPS;
1846 
1847 	PUSHMARK(sp);
1848 	XPUSHs(obj);								/* Target object */
1849 	XPUSHs(sv_2mortal(newSViv(cloning)));		/* Cloning flag */
1850 	PUTBACK;
1851 
1852 	count = perl_call_sv(hook, G_ARRAY);		/* Go back to Perl code */
1853 
1854 	SPAGAIN;
1855 
1856 	av = newAV();
1857 	for (i = count - 1; i >= 0; i--) {
1858 		SV *sv = POPs;
1859 		av_store(av, i, SvREFCNT_inc(sv));
1860 	}
1861 
1862 	PUTBACK;
1863 	FREETMPS;
1864 	LEAVE;
1865 
1866 	return av;
1867 }
1868 
1869 /*
1870  * known_class
1871  *
1872  * Lookup the class name in the `hclass' table and either assign it a new ID
1873  * or return the existing one, by filling in `classnum'.
1874  *
1875  * Return true if the class was known, false if the ID was just generated.
1876  */
known_class(pTHX_ stcxt_t * cxt,char * name,int len,I32 * classnum)1877 static int known_class(
1878         pTHX_
1879 	stcxt_t *cxt,
1880 	char *name,		/* Class name */
1881 	int len,		/* Name length */
1882 	I32 *classnum)
1883 {
1884 	SV **svh;
1885 	HV *hclass = cxt->hclass;
1886 
1887 	TRACEME(("known_class (%s)", name));
1888 
1889 	/*
1890 	 * Recall that we don't store pointers in this hash table, but tags.
1891 	 * Therefore, we need LOW_32BITS() to extract the relevant parts.
1892 	 */
1893 
1894 	svh = hv_fetch(hclass, name, len, FALSE);
1895 	if (svh) {
1896 		*classnum = LOW_32BITS(*svh);
1897 		return TRUE;
1898 	}
1899 
1900 	/*
1901 	 * Unknown classname, we need to record it.
1902 	 */
1903 
1904 	cxt->classnum++;
1905 	if (!hv_store(hclass, name, len, INT2PTR(SV*, cxt->classnum), 0))
1906 		CROAK(("Unable to record new classname"));
1907 
1908 	*classnum = cxt->classnum;
1909 	return FALSE;
1910 }
1911 
1912 /***
1913  *** Sepcific store routines.
1914  ***/
1915 
1916 /*
1917  * store_ref
1918  *
1919  * Store a reference.
1920  * Layout is SX_REF <object> or SX_OVERLOAD <object>.
1921  */
store_ref(pTHX_ stcxt_t * cxt,SV * sv)1922 static int store_ref(pTHX_ stcxt_t *cxt, SV *sv)
1923 {
1924 	int is_weak = 0;
1925 	TRACEME(("store_ref (0x%"UVxf")", PTR2UV(sv)));
1926 
1927 	/*
1928 	 * Follow reference, and check if target is overloaded.
1929 	 */
1930 
1931 #ifdef SvWEAKREF
1932 	if (SvWEAKREF(sv))
1933 		is_weak = 1;
1934 	TRACEME(("ref (0x%"UVxf") is%s weak", PTR2UV(sv), is_weak ? "" : "n't"));
1935 #endif
1936 	sv = SvRV(sv);
1937 
1938 	if (SvOBJECT(sv)) {
1939 		HV *stash = (HV *) SvSTASH(sv);
1940 		if (stash && Gv_AMG(stash)) {
1941 			TRACEME(("ref (0x%"UVxf") is overloaded", PTR2UV(sv)));
1942 			PUTMARK(is_weak ? SX_WEAKOVERLOAD : SX_OVERLOAD);
1943 		} else
1944 			PUTMARK(is_weak ? SX_WEAKREF : SX_REF);
1945 	} else
1946 		PUTMARK(is_weak ? SX_WEAKREF : SX_REF);
1947 
1948 	return store(aTHX_ cxt, sv);
1949 }
1950 
1951 /*
1952  * store_scalar
1953  *
1954  * Store a scalar.
1955  *
1956  * Layout is SX_LSCALAR <length> <data>, SX_SCALAR <length> <data> or SX_UNDEF.
1957  * The <data> section is omitted if <length> is 0.
1958  *
1959  * If integer or double, the layout is SX_INTEGER <data> or SX_DOUBLE <data>.
1960  * Small integers (within [-127, +127]) are stored as SX_BYTE <byte>.
1961  */
store_scalar(pTHX_ stcxt_t * cxt,SV * sv)1962 static int store_scalar(pTHX_ stcxt_t *cxt, SV *sv)
1963 {
1964 	IV iv;
1965 	char *pv;
1966 	STRLEN len;
1967 	U32 flags = SvFLAGS(sv);			/* "cc -O" may put it in register */
1968 
1969 	TRACEME(("store_scalar (0x%"UVxf")", PTR2UV(sv)));
1970 
1971 	/*
1972 	 * For efficiency, break the SV encapsulation by peaking at the flags
1973 	 * directly without using the Perl macros to avoid dereferencing
1974 	 * sv->sv_flags each time we wish to check the flags.
1975 	 */
1976 
1977 	if (!(flags & SVf_OK)) {			/* !SvOK(sv) */
1978 		if (sv == &PL_sv_undef) {
1979 			TRACEME(("immortal undef"));
1980 			PUTMARK(SX_SV_UNDEF);
1981 		} else {
1982 			TRACEME(("undef at 0x%"UVxf, PTR2UV(sv)));
1983 			PUTMARK(SX_UNDEF);
1984 		}
1985 		return 0;
1986 	}
1987 
1988 	/*
1989 	 * Always store the string representation of a scalar if it exists.
1990 	 * Gisle Aas provided me with this test case, better than a long speach:
1991 	 *
1992 	 *  perl -MDevel::Peek -le '$a="abc"; $a+0; Dump($a)'
1993 	 *  SV = PVNV(0x80c8520)
1994 	 *       REFCNT = 1
1995 	 *       FLAGS = (NOK,POK,pNOK,pPOK)
1996 	 *       IV = 0
1997 	 *       NV = 0
1998 	 *       PV = 0x80c83d0 "abc"\0
1999 	 *       CUR = 3
2000 	 *       LEN = 4
2001 	 *
2002 	 * Write SX_SCALAR, length, followed by the actual data.
2003 	 *
2004 	 * Otherwise, write an SX_BYTE, SX_INTEGER or an SX_DOUBLE as
2005 	 * appropriate, followed by the actual (binary) data. A double
2006 	 * is written as a string if network order, for portability.
2007 	 *
2008 	 * NOTE: instead of using SvNOK(sv), we test for SvNOKp(sv).
2009 	 * The reason is that when the scalar value is tainted, the SvNOK(sv)
2010 	 * value is false.
2011 	 *
2012 	 * The test for a read-only scalar with both POK and NOK set is meant
2013 	 * to quickly detect &PL_sv_yes and &PL_sv_no without having to pay the
2014 	 * address comparison for each scalar we store.
2015 	 */
2016 
2017 #define SV_MAYBE_IMMORTAL (SVf_READONLY|SVf_POK|SVf_NOK)
2018 
2019 	if ((flags & SV_MAYBE_IMMORTAL) == SV_MAYBE_IMMORTAL) {
2020 		if (sv == &PL_sv_yes) {
2021 			TRACEME(("immortal yes"));
2022 			PUTMARK(SX_SV_YES);
2023 		} else if (sv == &PL_sv_no) {
2024 			TRACEME(("immortal no"));
2025 			PUTMARK(SX_SV_NO);
2026 		} else {
2027 			pv = SvPV(sv, len);			/* We know it's SvPOK */
2028 			goto string;				/* Share code below */
2029 		}
2030 	} else if (flags & SVf_POK) {
2031             /* public string - go direct to string read.  */
2032             goto string_readlen;
2033         } else if (
2034 #if (PATCHLEVEL <= 6)
2035             /* For 5.6 and earlier NV flag trumps IV flag, so only use integer
2036                direct if NV flag is off.  */
2037             (flags & (SVf_NOK | SVf_IOK)) == SVf_IOK
2038 #else
2039             /* 5.7 rules are that if IV public flag is set, IV value is as
2040                good, if not better, than NV value.  */
2041             flags & SVf_IOK
2042 #endif
2043             ) {
2044             iv = SvIV(sv);
2045             /*
2046              * Will come here from below with iv set if double is an integer.
2047              */
2048           integer:
2049 
2050             /* Sorry. This isn't in 5.005_56 (IIRC) or earlier.  */
2051 #ifdef SVf_IVisUV
2052             /* Need to do this out here, else 0xFFFFFFFF becomes iv of -1
2053              * (for example) and that ends up in the optimised small integer
2054              * case.
2055              */
2056             if ((flags & SVf_IVisUV) && SvUV(sv) > IV_MAX) {
2057                 TRACEME(("large unsigned integer as string, value = %"UVuf, SvUV(sv)));
2058                 goto string_readlen;
2059             }
2060 #endif
2061             /*
2062              * Optimize small integers into a single byte, otherwise store as
2063              * a real integer (converted into network order if they asked).
2064              */
2065 
2066             if (iv >= -128 && iv <= 127) {
2067                 unsigned char siv = (unsigned char) (iv + 128);	/* [0,255] */
2068                 PUTMARK(SX_BYTE);
2069                 PUTMARK(siv);
2070                 TRACEME(("small integer stored as %d", siv));
2071             } else if (cxt->netorder) {
2072 #ifndef HAS_HTONL
2073                 TRACEME(("no htonl, fall back to string for integer"));
2074                 goto string_readlen;
2075 #else
2076                 I32 niv;
2077 
2078 
2079 #if IVSIZE > 4
2080                 if (
2081 #ifdef SVf_IVisUV
2082                     /* Sorry. This isn't in 5.005_56 (IIRC) or earlier.  */
2083                     ((flags & SVf_IVisUV) && SvUV(sv) > 0x7FFFFFFF) ||
2084 #endif
2085                     (iv > 0x7FFFFFFF) || (iv < -0x80000000)) {
2086                     /* Bigger than 32 bits.  */
2087                     TRACEME(("large network order integer as string, value = %"IVdf, iv));
2088                     goto string_readlen;
2089                 }
2090 #endif
2091 
2092                 niv = (I32) htonl((I32) iv);
2093                 TRACEME(("using network order"));
2094                 PUTMARK(SX_NETINT);
2095                 WRITE_I32(niv);
2096 #endif
2097             } else {
2098                 PUTMARK(SX_INTEGER);
2099                 WRITE(&iv, sizeof(iv));
2100             }
2101 
2102             TRACEME(("ok (integer 0x%"UVxf", value = %"IVdf")", PTR2UV(sv), iv));
2103 	} else if (flags & SVf_NOK) {
2104             NV nv;
2105 #if (PATCHLEVEL <= 6)
2106             nv = SvNV(sv);
2107             /*
2108              * Watch for number being an integer in disguise.
2109              */
2110             if (nv == (NV) (iv = I_V(nv))) {
2111                 TRACEME(("double %"NVff" is actually integer %"IVdf, nv, iv));
2112                 goto integer;		/* Share code above */
2113             }
2114 #else
2115 
2116             SvIV_please(sv);
2117 	    if (SvIOK_notUV(sv)) {
2118                 iv = SvIV(sv);
2119                 goto integer;		/* Share code above */
2120             }
2121             nv = SvNV(sv);
2122 #endif
2123 
2124             if (cxt->netorder) {
2125                 TRACEME(("double %"NVff" stored as string", nv));
2126                 goto string_readlen;		/* Share code below */
2127             }
2128 
2129             PUTMARK(SX_DOUBLE);
2130             WRITE(&nv, sizeof(nv));
2131 
2132             TRACEME(("ok (double 0x%"UVxf", value = %"NVff")", PTR2UV(sv), nv));
2133 
2134 	} else if (flags & (SVp_POK | SVp_NOK | SVp_IOK)) {
2135             I32 wlen; /* For 64-bit machines */
2136 
2137           string_readlen:
2138             pv = SvPV(sv, len);
2139 
2140             /*
2141              * Will come here from above  if it was readonly, POK and NOK but
2142              * neither &PL_sv_yes nor &PL_sv_no.
2143              */
2144           string:
2145 
2146             wlen = (I32) len; /* WLEN via STORE_SCALAR expects I32 */
2147             if (SvUTF8 (sv))
2148                 STORE_UTF8STR(pv, wlen);
2149             else
2150                 STORE_SCALAR(pv, wlen);
2151             TRACEME(("ok (scalar 0x%"UVxf" '%s', length = %"IVdf")",
2152                      PTR2UV(sv), SvPVX(sv), (IV)len));
2153 	} else
2154             CROAK(("Can't determine type of %s(0x%"UVxf")",
2155                    sv_reftype(sv, FALSE),
2156                    PTR2UV(sv)));
2157         return 0;		/* Ok, no recursion on scalars */
2158 }
2159 
2160 /*
2161  * store_array
2162  *
2163  * Store an array.
2164  *
2165  * Layout is SX_ARRAY <size> followed by each item, in increading index order.
2166  * Each item is stored as <object>.
2167  */
store_array(pTHX_ stcxt_t * cxt,AV * av)2168 static int store_array(pTHX_ stcxt_t *cxt, AV *av)
2169 {
2170 	SV **sav;
2171 	I32 len = av_len(av) + 1;
2172 	I32 i;
2173 	int ret;
2174 
2175 	TRACEME(("store_array (0x%"UVxf")", PTR2UV(av)));
2176 
2177 	/*
2178 	 * Signal array by emitting SX_ARRAY, followed by the array length.
2179 	 */
2180 
2181 	PUTMARK(SX_ARRAY);
2182 	WLEN(len);
2183 	TRACEME(("size = %d", len));
2184 
2185 	/*
2186 	 * Now store each item recursively.
2187 	 */
2188 
2189 	for (i = 0; i < len; i++) {
2190 		sav = av_fetch(av, i, 0);
2191 		if (!sav) {
2192 			TRACEME(("(#%d) undef item", i));
2193 			STORE_SV_UNDEF();
2194 			continue;
2195 		}
2196 		TRACEME(("(#%d) item", i));
2197 		if ((ret = store(aTHX_ cxt, *sav)))	/* Extra () for -Wall, grr... */
2198 			return ret;
2199 	}
2200 
2201 	TRACEME(("ok (array)"));
2202 
2203 	return 0;
2204 }
2205 
2206 
2207 #if (PATCHLEVEL <= 6)
2208 
2209 /*
2210  * sortcmp
2211  *
2212  * Sort two SVs
2213  * Borrowed from perl source file pp_ctl.c, where it is used by pp_sort.
2214  */
2215 static int
sortcmp(const void * a,const void * b)2216 sortcmp(const void *a, const void *b)
2217 {
2218 #if defined(USE_ITHREADS)
2219         dTHX;
2220 #endif /* USE_ITHREADS */
2221         return sv_cmp(*(SV * const *) a, *(SV * const *) b);
2222 }
2223 
2224 #endif /* PATCHLEVEL <= 6 */
2225 
2226 /*
2227  * store_hash
2228  *
2229  * Store a hash table.
2230  *
2231  * For a "normal" hash (not restricted, no utf8 keys):
2232  *
2233  * Layout is SX_HASH <size> followed by each key/value pair, in random order.
2234  * Values are stored as <object>.
2235  * Keys are stored as <length> <data>, the <data> section being omitted
2236  * if length is 0.
2237  *
2238  * For a "fancy" hash (restricted or utf8 keys):
2239  *
2240  * Layout is SX_FLAG_HASH <size> <hash flags> followed by each key/value pair,
2241  * in random order.
2242  * Values are stored as <object>.
2243  * Keys are stored as <flags> <length> <data>, the <data> section being omitted
2244  * if length is 0.
2245  * Currently the only hash flag is "restriced"
2246  * Key flags are as for hv.h
2247  */
store_hash(pTHX_ stcxt_t * cxt,HV * hv)2248 static int store_hash(pTHX_ stcxt_t *cxt, HV *hv)
2249 {
2250 	dVAR;
2251 	I32 len =
2252 #ifdef HAS_RESTRICTED_HASHES
2253             HvTOTALKEYS(hv);
2254 #else
2255             HvKEYS(hv);
2256 #endif
2257 	I32 i;
2258 	int ret = 0;
2259 	I32 riter;
2260 	HE *eiter;
2261         int flagged_hash = ((SvREADONLY(hv)
2262 #ifdef HAS_HASH_KEY_FLAGS
2263                              || HvHASKFLAGS(hv)
2264 #endif
2265                                 ) ? 1 : 0);
2266         unsigned char hash_flags = (SvREADONLY(hv) ? SHV_RESTRICTED : 0);
2267 
2268         if (flagged_hash) {
2269             /* needs int cast for C++ compilers, doesn't it?  */
2270             TRACEME(("store_hash (0x%"UVxf") (flags %x)", PTR2UV(hv),
2271                      (int) hash_flags));
2272         } else {
2273             TRACEME(("store_hash (0x%"UVxf")", PTR2UV(hv)));
2274         }
2275 
2276 	/*
2277 	 * Signal hash by emitting SX_HASH, followed by the table length.
2278 	 */
2279 
2280         if (flagged_hash) {
2281             PUTMARK(SX_FLAG_HASH);
2282             PUTMARK(hash_flags);
2283         } else {
2284             PUTMARK(SX_HASH);
2285         }
2286 	WLEN(len);
2287 	TRACEME(("size = %d", len));
2288 
2289 	/*
2290 	 * Save possible iteration state via each() on that table.
2291 	 */
2292 
2293 	riter = HvRITER_get(hv);
2294 	eiter = HvEITER_get(hv);
2295 	hv_iterinit(hv);
2296 
2297 	/*
2298 	 * Now store each item recursively.
2299 	 *
2300      * If canonical is defined to some true value then store each
2301      * key/value pair in sorted order otherwise the order is random.
2302 	 * Canonical order is irrelevant when a deep clone operation is performed.
2303 	 *
2304 	 * Fetch the value from perl only once per store() operation, and only
2305 	 * when needed.
2306 	 */
2307 
2308 	if (
2309 		!(cxt->optype & ST_CLONE) && (cxt->canonical == 1 ||
2310 		(cxt->canonical < 0 && (cxt->canonical =
2311 			(SvTRUE(perl_get_sv("Storable::canonical", TRUE)) ? 1 : 0))))
2312 	) {
2313 		/*
2314 		 * Storing in order, sorted by key.
2315 		 * Run through the hash, building up an array of keys in a
2316 		 * mortal array, sort the array and then run through the
2317 		 * array.
2318 		 */
2319 
2320 		AV *av = newAV();
2321 
2322                 /*av_extend (av, len);*/
2323 
2324 		TRACEME(("using canonical order"));
2325 
2326 		for (i = 0; i < len; i++) {
2327 #ifdef HAS_RESTRICTED_HASHES
2328 			HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
2329 #else
2330 			HE *he = hv_iternext(hv);
2331 #endif
2332 			SV *key = hv_iterkeysv(he);
2333 			av_store(av, AvFILLp(av)+1, key);	/* av_push(), really */
2334 		}
2335 
2336 		STORE_HASH_SORT;
2337 
2338 		for (i = 0; i < len; i++) {
2339 #ifdef HAS_RESTRICTED_HASHES
2340 			int placeholders = (int)HvPLACEHOLDERS_get(hv);
2341 #endif
2342                         unsigned char flags = 0;
2343 			char *keyval;
2344 			STRLEN keylen_tmp;
2345                         I32 keylen;
2346 			SV *key = av_shift(av);
2347 			/* This will fail if key is a placeholder.
2348 			   Track how many placeholders we have, and error if we
2349 			   "see" too many.  */
2350 			HE *he  = hv_fetch_ent(hv, key, 0, 0);
2351 			SV *val;
2352 
2353 			if (he) {
2354 				if (!(val =  HeVAL(he))) {
2355 					/* Internal error, not I/O error */
2356 					return 1;
2357 				}
2358 			} else {
2359 #ifdef HAS_RESTRICTED_HASHES
2360 				/* Should be a placeholder.  */
2361 				if (placeholders-- < 0) {
2362 					/* This should not happen - number of
2363 					   retrieves should be identical to
2364 					   number of placeholders.  */
2365 			  		return 1;
2366 				}
2367 				/* Value is never needed, and PL_sv_undef is
2368 				   more space efficient to store.  */
2369 				val = &PL_sv_undef;
2370 				ASSERT (flags == 0,
2371 					("Flags not 0 but %d", flags));
2372 				flags = SHV_K_PLACEHOLDER;
2373 #else
2374 				return 1;
2375 #endif
2376 			}
2377 
2378 			/*
2379 			 * Store value first.
2380 			 */
2381 
2382 			TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
2383 
2384 			if ((ret = store(aTHX_ cxt, val)))	/* Extra () for -Wall, grr... */
2385 				goto out;
2386 
2387 			/*
2388 			 * Write key string.
2389 			 * Keys are written after values to make sure retrieval
2390 			 * can be optimal in terms of memory usage, where keys are
2391 			 * read into a fixed unique buffer called kbuf.
2392 			 * See retrieve_hash() for details.
2393 			 */
2394 
2395                         /* Implementation of restricted hashes isn't nicely
2396                            abstracted:  */
2397 			if ((hash_flags & SHV_RESTRICTED) && SvREADONLY(val)) {
2398 				flags |= SHV_K_LOCKED;
2399 			}
2400 
2401 			keyval = SvPV(key, keylen_tmp);
2402                         keylen = keylen_tmp;
2403 #ifdef HAS_UTF8_HASHES
2404                         /* If you build without optimisation on pre 5.6
2405                            then nothing spots that SvUTF8(key) is always 0,
2406                            so the block isn't optimised away, at which point
2407                            the linker dislikes the reference to
2408                            bytes_from_utf8.  */
2409 			if (SvUTF8(key)) {
2410                             const char *keysave = keyval;
2411                             bool is_utf8 = TRUE;
2412 
2413                             /* Just casting the &klen to (STRLEN) won't work
2414                                well if STRLEN and I32 are of different widths.
2415                                --jhi */
2416                             keyval = (char*)bytes_from_utf8((U8*)keyval,
2417                                                             &keylen_tmp,
2418                                                             &is_utf8);
2419 
2420                             /* If we were able to downgrade here, then than
2421                                means that we have  a key which only had chars
2422                                0-255, but was utf8 encoded.  */
2423 
2424                             if (keyval != keysave) {
2425                                 keylen = keylen_tmp;
2426                                 flags |= SHV_K_WASUTF8;
2427                             } else {
2428                                 /* keylen_tmp can't have changed, so no need
2429                                    to assign back to keylen.  */
2430                                 flags |= SHV_K_UTF8;
2431                             }
2432                         }
2433 #endif
2434 
2435                         if (flagged_hash) {
2436                             PUTMARK(flags);
2437                             TRACEME(("(#%d) key '%s' flags %x %u", i, keyval, flags, *keyval));
2438                         } else {
2439                             /* This is a workaround for a bug in 5.8.0
2440                                that causes the HEK_WASUTF8 flag to be
2441                                set on an HEK without the hash being
2442                                marked as having key flags. We just
2443                                cross our fingers and drop the flag.
2444                                AMS 20030901 */
2445                             assert (flags == 0 || flags == SHV_K_WASUTF8);
2446                             TRACEME(("(#%d) key '%s'", i, keyval));
2447                         }
2448 			WLEN(keylen);
2449 			if (keylen)
2450 				WRITE(keyval, keylen);
2451                         if (flags & SHV_K_WASUTF8)
2452                             Safefree (keyval);
2453 		}
2454 
2455 		/*
2456 		 * Free up the temporary array
2457 		 */
2458 
2459 		av_undef(av);
2460 		sv_free((SV *) av);
2461 
2462 	} else {
2463 
2464 		/*
2465 		 * Storing in "random" order (in the order the keys are stored
2466 		 * within the hash).  This is the default and will be faster!
2467 		 */
2468 
2469 		for (i = 0; i < len; i++) {
2470 			char *key = 0;
2471 			I32 len;
2472                         unsigned char flags;
2473 #ifdef HV_ITERNEXT_WANTPLACEHOLDERS
2474                         HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
2475 #else
2476                         HE *he = hv_iternext(hv);
2477 #endif
2478 			SV *val = (he ? hv_iterval(hv, he) : 0);
2479                         SV *key_sv = NULL;
2480                         HEK *hek;
2481 
2482 			if (val == 0)
2483 				return 1;		/* Internal error, not I/O error */
2484 
2485                         /* Implementation of restricted hashes isn't nicely
2486                            abstracted:  */
2487                         flags
2488                             = (((hash_flags & SHV_RESTRICTED)
2489                                 && SvREADONLY(val))
2490                                              ? SHV_K_LOCKED : 0);
2491 
2492                         if (val == &PL_sv_placeholder) {
2493                             flags |= SHV_K_PLACEHOLDER;
2494 			    val = &PL_sv_undef;
2495 			}
2496 
2497 			/*
2498 			 * Store value first.
2499 			 */
2500 
2501 			TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
2502 
2503 			if ((ret = store(aTHX_ cxt, val)))	/* Extra () for -Wall, grr... */
2504 				goto out;
2505 
2506 
2507                         hek = HeKEY_hek(he);
2508                         len = HEK_LEN(hek);
2509                         if (len == HEf_SVKEY) {
2510                             /* This is somewhat sick, but the internal APIs are
2511                              * such that XS code could put one of these in in
2512                              * a regular hash.
2513                              * Maybe we should be capable of storing one if
2514                              * found.
2515                              */
2516                             key_sv = HeKEY_sv(he);
2517                             flags |= SHV_K_ISSV;
2518                         } else {
2519                             /* Regular string key. */
2520 #ifdef HAS_HASH_KEY_FLAGS
2521                             if (HEK_UTF8(hek))
2522                                 flags |= SHV_K_UTF8;
2523                             if (HEK_WASUTF8(hek))
2524                                 flags |= SHV_K_WASUTF8;
2525 #endif
2526                             key = HEK_KEY(hek);
2527                         }
2528 			/*
2529 			 * Write key string.
2530 			 * Keys are written after values to make sure retrieval
2531 			 * can be optimal in terms of memory usage, where keys are
2532 			 * read into a fixed unique buffer called kbuf.
2533 			 * See retrieve_hash() for details.
2534 			 */
2535 
2536                         if (flagged_hash) {
2537                             PUTMARK(flags);
2538                             TRACEME(("(#%d) key '%s' flags %x", i, key, flags));
2539                         } else {
2540                             /* This is a workaround for a bug in 5.8.0
2541                                that causes the HEK_WASUTF8 flag to be
2542                                set on an HEK without the hash being
2543                                marked as having key flags. We just
2544                                cross our fingers and drop the flag.
2545                                AMS 20030901 */
2546                             assert (flags == 0 || flags == SHV_K_WASUTF8);
2547                             TRACEME(("(#%d) key '%s'", i, key));
2548                         }
2549                         if (flags & SHV_K_ISSV) {
2550                             store(aTHX_ cxt, key_sv);
2551                         } else {
2552                             WLEN(len);
2553                             if (len)
2554 				WRITE(key, len);
2555                         }
2556 		}
2557     }
2558 
2559 	TRACEME(("ok (hash 0x%"UVxf")", PTR2UV(hv)));
2560 
2561 out:
2562 	HvRITER_set(hv, riter);		/* Restore hash iterator state */
2563 	HvEITER_set(hv, eiter);
2564 
2565 	return ret;
2566 }
2567 
2568 /*
2569  * store_code
2570  *
2571  * Store a code reference.
2572  *
2573  * Layout is SX_CODE <length> followed by a scalar containing the perl
2574  * source code of the code reference.
2575  */
store_code(pTHX_ stcxt_t * cxt,CV * cv)2576 static int store_code(pTHX_ stcxt_t *cxt, CV *cv)
2577 {
2578 #if PERL_VERSION < 6
2579     /*
2580 	 * retrieve_code does not work with perl 5.005 or less
2581 	 */
2582 	return store_other(aTHX_ cxt, (SV*)cv);
2583 #else
2584 	dSP;
2585 	I32 len;
2586 	int count, reallen;
2587 	SV *text, *bdeparse;
2588 
2589 	TRACEME(("store_code (0x%"UVxf")", PTR2UV(cv)));
2590 
2591 	if (
2592 		cxt->deparse == 0 ||
2593 		(cxt->deparse < 0 && !(cxt->deparse =
2594 			SvTRUE(perl_get_sv("Storable::Deparse", TRUE)) ? 1 : 0))
2595 	) {
2596 		return store_other(aTHX_ cxt, (SV*)cv);
2597 	}
2598 
2599 	/*
2600 	 * Require B::Deparse. At least B::Deparse 0.61 is needed for
2601 	 * blessed code references.
2602 	 */
2603 	/* Ownership of both SVs is passed to load_module, which frees them. */
2604 	load_module(PERL_LOADMOD_NOIMPORT, newSVpvn("B::Deparse",10), newSVnv(0.61));
2605 
2606 	ENTER;
2607 	SAVETMPS;
2608 
2609 	/*
2610 	 * create the B::Deparse object
2611 	 */
2612 
2613 	PUSHMARK(sp);
2614 	XPUSHs(sv_2mortal(newSVpvn("B::Deparse",10)));
2615 	PUTBACK;
2616 	count = call_method("new", G_SCALAR);
2617 	SPAGAIN;
2618 	if (count != 1)
2619 		CROAK(("Unexpected return value from B::Deparse::new\n"));
2620 	bdeparse = POPs;
2621 
2622 	/*
2623 	 * call the coderef2text method
2624 	 */
2625 
2626 	PUSHMARK(sp);
2627 	XPUSHs(bdeparse); /* XXX is this already mortal? */
2628 	XPUSHs(sv_2mortal(newRV_inc((SV*)cv)));
2629 	PUTBACK;
2630 	count = call_method("coderef2text", G_SCALAR);
2631 	SPAGAIN;
2632 	if (count != 1)
2633 		CROAK(("Unexpected return value from B::Deparse::coderef2text\n"));
2634 
2635 	text = POPs;
2636 	len = SvLEN(text);
2637 	reallen = strlen(SvPV_nolen(text));
2638 
2639 	/*
2640 	 * Empty code references or XS functions are deparsed as
2641 	 * "(prototype) ;" or ";".
2642 	 */
2643 
2644 	if (len == 0 || *(SvPV_nolen(text)+reallen-1) == ';') {
2645 	    CROAK(("The result of B::Deparse::coderef2text was empty - maybe you're trying to serialize an XS function?\n"));
2646 	}
2647 
2648 	/*
2649 	 * Signal code by emitting SX_CODE.
2650 	 */
2651 
2652 	PUTMARK(SX_CODE);
2653 	cxt->tagnum++;   /* necessary, as SX_CODE is a SEEN() candidate */
2654 	TRACEME(("size = %d", len));
2655 	TRACEME(("code = %s", SvPV_nolen(text)));
2656 
2657 	/*
2658 	 * Now store the source code.
2659 	 */
2660 
2661 	STORE_SCALAR(SvPV_nolen(text), len);
2662 
2663 	FREETMPS;
2664 	LEAVE;
2665 
2666 	TRACEME(("ok (code)"));
2667 
2668 	return 0;
2669 #endif
2670 }
2671 
2672 /*
2673  * store_tied
2674  *
2675  * When storing a tied object (be it a tied scalar, array or hash), we lay out
2676  * a special mark, followed by the underlying tied object. For instance, when
2677  * dealing with a tied hash, we store SX_TIED_HASH <hash object>, where
2678  * <hash object> stands for the serialization of the tied hash.
2679  */
store_tied(pTHX_ stcxt_t * cxt,SV * sv)2680 static int store_tied(pTHX_ stcxt_t *cxt, SV *sv)
2681 {
2682 	MAGIC *mg;
2683 	SV *obj = NULL;
2684 	int ret = 0;
2685 	int svt = SvTYPE(sv);
2686 	char mtype = 'P';
2687 
2688 	TRACEME(("store_tied (0x%"UVxf")", PTR2UV(sv)));
2689 
2690 	/*
2691 	 * We have a small run-time penalty here because we chose to factorise
2692 	 * all tieds objects into the same routine, and not have a store_tied_hash,
2693 	 * a store_tied_array, etc...
2694 	 *
2695 	 * Don't use a switch() statement, as most compilers don't optimize that
2696 	 * well for 2/3 values. An if() else if() cascade is just fine. We put
2697 	 * tied hashes first, as they are the most likely beasts.
2698 	 */
2699 
2700 	if (svt == SVt_PVHV) {
2701 		TRACEME(("tied hash"));
2702 		PUTMARK(SX_TIED_HASH);			/* Introduces tied hash */
2703 	} else if (svt == SVt_PVAV) {
2704 		TRACEME(("tied array"));
2705 		PUTMARK(SX_TIED_ARRAY);			/* Introduces tied array */
2706 	} else {
2707 		TRACEME(("tied scalar"));
2708 		PUTMARK(SX_TIED_SCALAR);		/* Introduces tied scalar */
2709 		mtype = 'q';
2710 	}
2711 
2712 	if (!(mg = mg_find(sv, mtype)))
2713 		CROAK(("No magic '%c' found while storing tied %s", mtype,
2714 			(svt == SVt_PVHV) ? "hash" :
2715 				(svt == SVt_PVAV) ? "array" : "scalar"));
2716 
2717 	/*
2718 	 * The mg->mg_obj found by mg_find() above actually points to the
2719 	 * underlying tied Perl object implementation. For instance, if the
2720 	 * original SV was that of a tied array, then mg->mg_obj is an AV.
2721 	 *
2722 	 * Note that we store the Perl object as-is. We don't call its FETCH
2723 	 * method along the way. At retrieval time, we won't call its STORE
2724 	 * method either, but the tieing magic will be re-installed. In itself,
2725 	 * that ensures that the tieing semantics are preserved since futher
2726 	 * accesses on the retrieved object will indeed call the magic methods...
2727 	 */
2728 
2729 	/* [#17040] mg_obj is NULL for scalar self-ties. AMS 20030416 */
2730 	obj = mg->mg_obj ? mg->mg_obj : newSV(0);
2731 	if ((ret = store(aTHX_ cxt, obj)))
2732 		return ret;
2733 
2734 	TRACEME(("ok (tied)"));
2735 
2736 	return 0;
2737 }
2738 
2739 /*
2740  * store_tied_item
2741  *
2742  * Stores a reference to an item within a tied structure:
2743  *
2744  *  . \$h{key}, stores both the (tied %h) object and 'key'.
2745  *  . \$a[idx], stores both the (tied @a) object and 'idx'.
2746  *
2747  * Layout is therefore either:
2748  *     SX_TIED_KEY <object> <key>
2749  *     SX_TIED_IDX <object> <index>
2750  */
store_tied_item(pTHX_ stcxt_t * cxt,SV * sv)2751 static int store_tied_item(pTHX_ stcxt_t *cxt, SV *sv)
2752 {
2753 	MAGIC *mg;
2754 	int ret;
2755 
2756 	TRACEME(("store_tied_item (0x%"UVxf")", PTR2UV(sv)));
2757 
2758 	if (!(mg = mg_find(sv, 'p')))
2759 		CROAK(("No magic 'p' found while storing reference to tied item"));
2760 
2761 	/*
2762 	 * We discriminate between \$h{key} and \$a[idx] via mg_ptr.
2763 	 */
2764 
2765 	if (mg->mg_ptr) {
2766 		TRACEME(("store_tied_item: storing a ref to a tied hash item"));
2767 		PUTMARK(SX_TIED_KEY);
2768 		TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2769 
2770 		if ((ret = store(aTHX_ cxt, mg->mg_obj)))		/* Extra () for -Wall, grr... */
2771 			return ret;
2772 
2773 		TRACEME(("store_tied_item: storing PTR 0x%"UVxf, PTR2UV(mg->mg_ptr)));
2774 
2775 		if ((ret = store(aTHX_ cxt, (SV *) mg->mg_ptr)))	/* Idem, for -Wall */
2776 			return ret;
2777 	} else {
2778 		I32 idx = mg->mg_len;
2779 
2780 		TRACEME(("store_tied_item: storing a ref to a tied array item "));
2781 		PUTMARK(SX_TIED_IDX);
2782 		TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
2783 
2784 		if ((ret = store(aTHX_ cxt, mg->mg_obj)))		/* Idem, for -Wall */
2785 			return ret;
2786 
2787 		TRACEME(("store_tied_item: storing IDX %d", idx));
2788 
2789 		WLEN(idx);
2790 	}
2791 
2792 	TRACEME(("ok (tied item)"));
2793 
2794 	return 0;
2795 }
2796 
2797 /*
2798  * store_hook		-- dispatched manually, not via sv_store[]
2799  *
2800  * The blessed SV is serialized by a hook.
2801  *
2802  * Simple Layout is:
2803  *
2804  *     SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
2805  *
2806  * where <flags> indicates how long <len>, <len2> and <len3> are, whether
2807  * the trailing part [] is present, the type of object (scalar, array or hash).
2808  * There is also a bit which says how the classname is stored between:
2809  *
2810  *     <len> <classname>
2811  *     <index>
2812  *
2813  * and when the <index> form is used (classname already seen), the "large
2814  * classname" bit in <flags> indicates how large the <index> is.
2815  *
2816  * The serialized string returned by the hook is of length <len2> and comes
2817  * next.  It is an opaque string for us.
2818  *
2819  * Those <len3> object IDs which are listed last represent the extra references
2820  * not directly serialized by the hook, but which are linked to the object.
2821  *
2822  * When recursion is mandated to resolve object-IDs not yet seen, we have
2823  * instead, with <header> being flags with bits set to indicate the object type
2824  * and that recursion was indeed needed:
2825  *
2826  *     SX_HOOK <header> <object> <header> <object> <flags>
2827  *
2828  * that same header being repeated between serialized objects obtained through
2829  * recursion, until we reach flags indicating no recursion, at which point
2830  * we know we've resynchronized with a single layout, after <flags>.
2831  *
2832  * When storing a blessed ref to a tied variable, the following format is
2833  * used:
2834  *
2835  *     SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
2836  *
2837  * The first <flags> indication carries an object of type SHT_EXTRA, and the
2838  * real object type is held in the <extra> flag.  At the very end of the
2839  * serialization stream, the underlying magic object is serialized, just like
2840  * any other tied variable.
2841  */
store_hook(pTHX_ stcxt_t * cxt,SV * sv,int type,HV * pkg,SV * hook)2842 static int store_hook(
2843         pTHX_
2844 	stcxt_t *cxt,
2845 	SV *sv,
2846 	int type,
2847 	HV *pkg,
2848 	SV *hook)
2849 {
2850 	I32 len;
2851 	char *classname;
2852 	STRLEN len2;
2853 	SV *ref;
2854 	AV *av;
2855 	SV **ary;
2856 	int count;				/* really len3 + 1 */
2857 	unsigned char flags;
2858 	char *pv;
2859 	int i;
2860 	int recursed = 0;		/* counts recursion */
2861 	int obj_type;			/* object type, on 2 bits */
2862 	I32 classnum;
2863 	int ret;
2864 	int clone = cxt->optype & ST_CLONE;
2865 	char mtype = '\0';				/* for blessed ref to tied structures */
2866 	unsigned char eflags = '\0';	/* used when object type is SHT_EXTRA */
2867 
2868 	TRACEME(("store_hook, classname \"%s\", tagged #%d", HvNAME_get(pkg), cxt->tagnum));
2869 
2870 	/*
2871 	 * Determine object type on 2 bits.
2872 	 */
2873 
2874 	switch (type) {
2875 	case svis_SCALAR:
2876 		obj_type = SHT_SCALAR;
2877 		break;
2878 	case svis_ARRAY:
2879 		obj_type = SHT_ARRAY;
2880 		break;
2881 	case svis_HASH:
2882 		obj_type = SHT_HASH;
2883 		break;
2884 	case svis_TIED:
2885 		/*
2886 		 * Produced by a blessed ref to a tied data structure, $o in the
2887 		 * following Perl code.
2888 		 *
2889 		 * 	my %h;
2890 		 *  tie %h, 'FOO';
2891 		 *	my $o = bless \%h, 'BAR';
2892 		 *
2893 		 * Signal the tie-ing magic by setting the object type as SHT_EXTRA
2894 		 * (since we have only 2 bits in <flags> to store the type), and an
2895 		 * <extra> byte flag will be emitted after the FIRST <flags> in the
2896 		 * stream, carrying what we put in `eflags'.
2897 		 */
2898 		obj_type = SHT_EXTRA;
2899 		switch (SvTYPE(sv)) {
2900 		case SVt_PVHV:
2901 			eflags = (unsigned char) SHT_THASH;
2902 			mtype = 'P';
2903 			break;
2904 		case SVt_PVAV:
2905 			eflags = (unsigned char) SHT_TARRAY;
2906 			mtype = 'P';
2907 			break;
2908 		default:
2909 			eflags = (unsigned char) SHT_TSCALAR;
2910 			mtype = 'q';
2911 			break;
2912 		}
2913 		break;
2914 	default:
2915 		CROAK(("Unexpected object type (%d) in store_hook()", type));
2916 	}
2917 	flags = SHF_NEED_RECURSE | obj_type;
2918 
2919 	classname = HvNAME_get(pkg);
2920 	len = strlen(classname);
2921 
2922 	/*
2923 	 * To call the hook, we need to fake a call like:
2924 	 *
2925 	 *    $object->STORABLE_freeze($cloning);
2926 	 *
2927 	 * but we don't have the $object here.  For instance, if $object is
2928 	 * a blessed array, what we have in `sv' is the array, and we can't
2929 	 * call a method on those.
2930 	 *
2931 	 * Therefore, we need to create a temporary reference to the object and
2932 	 * make the call on that reference.
2933 	 */
2934 
2935 	TRACEME(("about to call STORABLE_freeze on class %s", classname));
2936 
2937 	ref = newRV_noinc(sv);				/* Temporary reference */
2938 	av = array_call(aTHX_ ref, hook, clone);	/* @a = $object->STORABLE_freeze($c) */
2939 	SvRV_set(ref, NULL);
2940 	SvREFCNT_dec(ref);					/* Reclaim temporary reference */
2941 
2942 	count = AvFILLp(av) + 1;
2943 	TRACEME(("store_hook, array holds %d items", count));
2944 
2945 	/*
2946 	 * If they return an empty list, it means they wish to ignore the
2947 	 * hook for this class (and not just this instance -- that's for them
2948 	 * to handle if they so wish).
2949 	 *
2950 	 * Simply disable the cached entry for the hook (it won't be recomputed
2951 	 * since it's present in the cache) and recurse to store_blessed().
2952 	 */
2953 
2954 	if (!count) {
2955 		/*
2956 		 * They must not change their mind in the middle of a serialization.
2957 		 */
2958 
2959 		if (hv_fetch(cxt->hclass, classname, len, FALSE))
2960 			CROAK(("Too late to ignore hooks for %s class \"%s\"",
2961 				(cxt->optype & ST_CLONE) ? "cloning" : "storing", classname));
2962 
2963 		pkg_hide(aTHX_ cxt->hook, pkg, "STORABLE_freeze");
2964 
2965 		ASSERT(!pkg_can(aTHX_ cxt->hook, pkg, "STORABLE_freeze"), ("hook invisible"));
2966 		TRACEME(("ignoring STORABLE_freeze in class \"%s\"", classname));
2967 
2968 		return store_blessed(aTHX_ cxt, sv, type, pkg);
2969 	}
2970 
2971 	/*
2972 	 * Get frozen string.
2973 	 */
2974 
2975 	ary = AvARRAY(av);
2976 	pv = SvPV(ary[0], len2);
2977 	/* We can't use pkg_can here because it only caches one method per
2978 	 * package */
2979 	{
2980 	    GV* gv = gv_fetchmethod_autoload(pkg, "STORABLE_attach", FALSE);
2981 	    if (gv && isGV(gv)) {
2982 	        if (count > 1)
2983 	            CROAK(("Freeze cannot return references if %s class is using STORABLE_attach", classname));
2984 	        goto check_done;
2985 	    }
2986 	}
2987 
2988 	/*
2989 	 * If they returned more than one item, we need to serialize some
2990 	 * extra references if not already done.
2991 	 *
2992 	 * Loop over the array, starting at position #1, and for each item,
2993 	 * ensure it is a reference, serialize it if not already done, and
2994 	 * replace the entry with the tag ID of the corresponding serialized
2995 	 * object.
2996 	 *
2997 	 * We CHEAT by not calling av_fetch() and read directly within the
2998 	 * array, for speed.
2999 	 */
3000 
3001 	for (i = 1; i < count; i++) {
3002 #ifdef USE_PTR_TABLE
3003 		char *fake_tag;
3004 #else
3005 		SV **svh;
3006 #endif
3007 		SV *rsv = ary[i];
3008 		SV *xsv;
3009 		SV *tag;
3010 		AV *av_hook = cxt->hook_seen;
3011 
3012 		if (!SvROK(rsv))
3013 			CROAK(("Item #%d returned by STORABLE_freeze "
3014 				"for %s is not a reference", i, classname));
3015 		xsv = SvRV(rsv);		/* Follow ref to know what to look for */
3016 
3017 		/*
3018 		 * Look in hseen and see if we have a tag already.
3019 		 * Serialize entry if not done already, and get its tag.
3020 		 */
3021 
3022 #ifdef USE_PTR_TABLE
3023 		/* Fakery needed because ptr_table_fetch returns zero for a
3024 		   failure, whereas the existing code assumes that it can
3025 		   safely store a tag zero. So for ptr_tables we store tag+1
3026 		*/
3027 		if ((fake_tag = ptr_table_fetch(cxt->pseen, xsv)))
3028 			goto sv_seen;		/* Avoid moving code too far to the right */
3029 #else
3030 		if ((svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE)))
3031 			goto sv_seen;		/* Avoid moving code too far to the right */
3032 #endif
3033 
3034 		TRACEME(("listed object %d at 0x%"UVxf" is unknown", i-1, PTR2UV(xsv)));
3035 
3036 		/*
3037 		 * We need to recurse to store that object and get it to be known
3038 		 * so that we can resolve the list of object-IDs at retrieve time.
3039 		 *
3040 		 * The first time we do this, we need to emit the proper header
3041 		 * indicating that we recursed, and what the type of object is (the
3042 		 * object we're storing via a user-hook).  Indeed, during retrieval,
3043 		 * we'll have to create the object before recursing to retrieve the
3044 		 * others, in case those would point back at that object.
3045 		 */
3046 
3047 		/* [SX_HOOK] <flags> [<extra>] <object>*/
3048 		if (!recursed++) {
3049 			PUTMARK(SX_HOOK);
3050 			PUTMARK(flags);
3051 			if (obj_type == SHT_EXTRA)
3052 				PUTMARK(eflags);
3053 		} else
3054 			PUTMARK(flags);
3055 
3056 		if ((ret = store(aTHX_ cxt, xsv)))	/* Given by hook for us to store */
3057 			return ret;
3058 
3059 #ifdef USE_PTR_TABLE
3060 		fake_tag = ptr_table_fetch(cxt->pseen, xsv);
3061 		if (!sv)
3062 			CROAK(("Could not serialize item #%d from hook in %s", i, classname));
3063 #else
3064 		svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE);
3065 		if (!svh)
3066 			CROAK(("Could not serialize item #%d from hook in %s", i, classname));
3067 #endif
3068 		/*
3069 		 * It was the first time we serialized `xsv'.
3070 		 *
3071 		 * Keep this SV alive until the end of the serialization: if we
3072 		 * disposed of it right now by decrementing its refcount, and it was
3073 		 * a temporary value, some next temporary value allocated during
3074 		 * another STORABLE_freeze might take its place, and we'd wrongly
3075 		 * assume that new SV was already serialized, based on its presence
3076 		 * in cxt->hseen.
3077 		 *
3078 		 * Therefore, push it away in cxt->hook_seen.
3079 		 */
3080 
3081 		av_store(av_hook, AvFILLp(av_hook)+1, SvREFCNT_inc(xsv));
3082 
3083 	sv_seen:
3084 		/*
3085 		 * Dispose of the REF they returned.  If we saved the `xsv' away
3086 		 * in the array of returned SVs, that will not cause the underlying
3087 		 * referenced SV to be reclaimed.
3088 		 */
3089 
3090 		ASSERT(SvREFCNT(xsv) > 1, ("SV will survive disposal of its REF"));
3091 		SvREFCNT_dec(rsv);			/* Dispose of reference */
3092 
3093 		/*
3094 		 * Replace entry with its tag (not a real SV, so no refcnt increment)
3095 		 */
3096 
3097 #ifdef USE_PTR_TABLE
3098 		tag = (SV *)--fake_tag;
3099 #else
3100 		tag = *svh;
3101 #endif
3102 		ary[i] = tag
3103 		TRACEME(("listed object %d at 0x%"UVxf" is tag #%"UVuf,
3104 			 i-1, PTR2UV(xsv), PTR2UV(tag)));
3105 	}
3106 
3107 	/*
3108 	 * Allocate a class ID if not already done.
3109 	 *
3110 	 * This needs to be done after the recursion above, since at retrieval
3111 	 * time, we'll see the inner objects first.  Many thanks to
3112 	 * Salvador Ortiz Garcia <sog@msg.com.mx> who spot that bug and
3113 	 * proposed the right fix.  -- RAM, 15/09/2000
3114 	 */
3115 
3116 check_done:
3117 	if (!known_class(aTHX_ cxt, classname, len, &classnum)) {
3118 		TRACEME(("first time we see class %s, ID = %d", classname, classnum));
3119 		classnum = -1;				/* Mark: we must store classname */
3120 	} else {
3121 		TRACEME(("already seen class %s, ID = %d", classname, classnum));
3122 	}
3123 
3124 	/*
3125 	 * Compute leading flags.
3126 	 */
3127 
3128 	flags = obj_type;
3129 	if (((classnum == -1) ? len : classnum) > LG_SCALAR)
3130 		flags |= SHF_LARGE_CLASSLEN;
3131 	if (classnum != -1)
3132 		flags |= SHF_IDX_CLASSNAME;
3133 	if (len2 > LG_SCALAR)
3134 		flags |= SHF_LARGE_STRLEN;
3135 	if (count > 1)
3136 		flags |= SHF_HAS_LIST;
3137 	if (count > (LG_SCALAR + 1))
3138 		flags |= SHF_LARGE_LISTLEN;
3139 
3140 	/*
3141 	 * We're ready to emit either serialized form:
3142 	 *
3143 	 *   SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
3144 	 *   SX_HOOK <flags> <index>           <len2> <str> [<len3> <object-IDs>]
3145 	 *
3146 	 * If we recursed, the SX_HOOK has already been emitted.
3147 	 */
3148 
3149 	TRACEME(("SX_HOOK (recursed=%d) flags=0x%x "
3150 			"class=%"IVdf" len=%"IVdf" len2=%"IVdf" len3=%d",
3151 		 recursed, flags, (IV)classnum, (IV)len, (IV)len2, count-1));
3152 
3153 	/* SX_HOOK <flags> [<extra>] */
3154 	if (!recursed) {
3155 		PUTMARK(SX_HOOK);
3156 		PUTMARK(flags);
3157 		if (obj_type == SHT_EXTRA)
3158 			PUTMARK(eflags);
3159 	} else
3160 		PUTMARK(flags);
3161 
3162 	/* <len> <classname> or <index> */
3163 	if (flags & SHF_IDX_CLASSNAME) {
3164 		if (flags & SHF_LARGE_CLASSLEN)
3165 			WLEN(classnum);
3166 		else {
3167 			unsigned char cnum = (unsigned char) classnum;
3168 			PUTMARK(cnum);
3169 		}
3170 	} else {
3171 		if (flags & SHF_LARGE_CLASSLEN)
3172 			WLEN(len);
3173 		else {
3174 			unsigned char clen = (unsigned char) len;
3175 			PUTMARK(clen);
3176 		}
3177 		WRITE(classname, len);		/* Final \0 is omitted */
3178 	}
3179 
3180 	/* <len2> <frozen-str> */
3181 	if (flags & SHF_LARGE_STRLEN) {
3182 		I32 wlen2 = len2;		/* STRLEN might be 8 bytes */
3183 		WLEN(wlen2);			/* Must write an I32 for 64-bit machines */
3184 	} else {
3185 		unsigned char clen = (unsigned char) len2;
3186 		PUTMARK(clen);
3187 	}
3188 	if (len2)
3189 		WRITE(pv, (SSize_t)len2);	/* Final \0 is omitted */
3190 
3191 	/* [<len3> <object-IDs>] */
3192 	if (flags & SHF_HAS_LIST) {
3193 		int len3 = count - 1;
3194 		if (flags & SHF_LARGE_LISTLEN)
3195 			WLEN(len3);
3196 		else {
3197 			unsigned char clen = (unsigned char) len3;
3198 			PUTMARK(clen);
3199 		}
3200 
3201 		/*
3202 		 * NOTA BENE, for 64-bit machines: the ary[i] below does not yield a
3203 		 * real pointer, rather a tag number, well under the 32-bit limit.
3204 		 */
3205 
3206 		for (i = 1; i < count; i++) {
3207 			I32 tagval = htonl(LOW_32BITS(ary[i]));
3208 			WRITE_I32(tagval);
3209 			TRACEME(("object %d, tag #%d", i-1, ntohl(tagval)));
3210 		}
3211 	}
3212 
3213 	/*
3214 	 * Free the array.  We need extra care for indices after 0, since they
3215 	 * don't hold real SVs but integers cast.
3216 	 */
3217 
3218 	if (count > 1)
3219 		AvFILLp(av) = 0;	/* Cheat, nothing after 0 interests us */
3220 	av_undef(av);
3221 	sv_free((SV *) av);
3222 
3223 	/*
3224 	 * If object was tied, need to insert serialization of the magic object.
3225 	 */
3226 
3227 	if (obj_type == SHT_EXTRA) {
3228 		MAGIC *mg;
3229 
3230 		if (!(mg = mg_find(sv, mtype))) {
3231 			int svt = SvTYPE(sv);
3232 			CROAK(("No magic '%c' found while storing ref to tied %s with hook",
3233 				mtype, (svt == SVt_PVHV) ? "hash" :
3234 					(svt == SVt_PVAV) ? "array" : "scalar"));
3235 		}
3236 
3237 		TRACEME(("handling the magic object 0x%"UVxf" part of 0x%"UVxf,
3238 			PTR2UV(mg->mg_obj), PTR2UV(sv)));
3239 
3240 		/*
3241 		 * [<magic object>]
3242 		 */
3243 
3244 		if ((ret = store(aTHX_ cxt, mg->mg_obj)))	/* Extra () for -Wall, grr... */
3245 			return ret;
3246 	}
3247 
3248 	return 0;
3249 }
3250 
3251 /*
3252  * store_blessed	-- dispatched manually, not via sv_store[]
3253  *
3254  * Check whether there is a STORABLE_xxx hook defined in the class or in one
3255  * of its ancestors.  If there is, then redispatch to store_hook();
3256  *
3257  * Otherwise, the blessed SV is stored using the following layout:
3258  *
3259  *    SX_BLESS <flag> <len> <classname> <object>
3260  *
3261  * where <flag> indicates whether <len> is stored on 0 or 4 bytes, depending
3262  * on the high-order bit in flag: if 1, then length follows on 4 bytes.
3263  * Otherwise, the low order bits give the length, thereby giving a compact
3264  * representation for class names less than 127 chars long.
3265  *
3266  * Each <classname> seen is remembered and indexed, so that the next time
3267  * an object in the blessed in the same <classname> is stored, the following
3268  * will be emitted:
3269  *
3270  *    SX_IX_BLESS <flag> <index> <object>
3271  *
3272  * where <index> is the classname index, stored on 0 or 4 bytes depending
3273  * on the high-order bit in flag (same encoding as above for <len>).
3274  */
store_blessed(pTHX_ stcxt_t * cxt,SV * sv,int type,HV * pkg)3275 static int store_blessed(
3276         pTHX_
3277 	stcxt_t *cxt,
3278 	SV *sv,
3279 	int type,
3280 	HV *pkg)
3281 {
3282 	SV *hook;
3283 	I32 len;
3284 	char *classname;
3285 	I32 classnum;
3286 
3287 	TRACEME(("store_blessed, type %d, class \"%s\"", type, HvNAME_get(pkg)));
3288 
3289 	/*
3290 	 * Look for a hook for this blessed SV and redirect to store_hook()
3291 	 * if needed.
3292 	 */
3293 
3294 	hook = pkg_can(aTHX_ cxt->hook, pkg, "STORABLE_freeze");
3295 	if (hook)
3296 		return store_hook(aTHX_ cxt, sv, type, pkg, hook);
3297 
3298 	/*
3299 	 * This is a blessed SV without any serialization hook.
3300 	 */
3301 
3302 	classname = HvNAME_get(pkg);
3303 	len = strlen(classname);
3304 
3305 	TRACEME(("blessed 0x%"UVxf" in %s, no hook: tagged #%d",
3306 		 PTR2UV(sv), classname, cxt->tagnum));
3307 
3308 	/*
3309 	 * Determine whether it is the first time we see that class name (in which
3310 	 * case it will be stored in the SX_BLESS form), or whether we already
3311 	 * saw that class name before (in which case the SX_IX_BLESS form will be
3312 	 * used).
3313 	 */
3314 
3315 	if (known_class(aTHX_ cxt, classname, len, &classnum)) {
3316 		TRACEME(("already seen class %s, ID = %d", classname, classnum));
3317 		PUTMARK(SX_IX_BLESS);
3318 		if (classnum <= LG_BLESS) {
3319 			unsigned char cnum = (unsigned char) classnum;
3320 			PUTMARK(cnum);
3321 		} else {
3322 			unsigned char flag = (unsigned char) 0x80;
3323 			PUTMARK(flag);
3324 			WLEN(classnum);
3325 		}
3326 	} else {
3327 		TRACEME(("first time we see class %s, ID = %d", classname, classnum));
3328 		PUTMARK(SX_BLESS);
3329 		if (len <= LG_BLESS) {
3330 			unsigned char clen = (unsigned char) len;
3331 			PUTMARK(clen);
3332 		} else {
3333 			unsigned char flag = (unsigned char) 0x80;
3334 			PUTMARK(flag);
3335 			WLEN(len);					/* Don't BER-encode, this should be rare */
3336 		}
3337 		WRITE(classname, len);				/* Final \0 is omitted */
3338 	}
3339 
3340 	/*
3341 	 * Now emit the <object> part.
3342 	 */
3343 
3344 	return SV_STORE(type)(aTHX_ cxt, sv);
3345 }
3346 
3347 /*
3348  * store_other
3349  *
3350  * We don't know how to store the item we reached, so return an error condition.
3351  * (it's probably a GLOB, some CODE reference, etc...)
3352  *
3353  * If they defined the `forgive_me' variable at the Perl level to some
3354  * true value, then don't croak, just warn, and store a placeholder string
3355  * instead.
3356  */
store_other(pTHX_ stcxt_t * cxt,SV * sv)3357 static int store_other(pTHX_ stcxt_t *cxt, SV *sv)
3358 {
3359 	I32 len;
3360 	char buf[80];
3361 
3362 	TRACEME(("store_other"));
3363 
3364 	/*
3365 	 * Fetch the value from perl only once per store() operation.
3366 	 */
3367 
3368 	if (
3369 		cxt->forgive_me == 0 ||
3370 		(cxt->forgive_me < 0 && !(cxt->forgive_me =
3371 			SvTRUE(perl_get_sv("Storable::forgive_me", TRUE)) ? 1 : 0))
3372 	)
3373 		CROAK(("Can't store %s items", sv_reftype(sv, FALSE)));
3374 
3375 	warn("Can't store item %s(0x%"UVxf")",
3376 		sv_reftype(sv, FALSE), PTR2UV(sv));
3377 
3378 	/*
3379 	 * Store placeholder string as a scalar instead...
3380 	 */
3381 
3382 	(void) sprintf(buf, "You lost %s(0x%"UVxf")%c", sv_reftype(sv, FALSE),
3383 		       PTR2UV(sv), (char) 0);
3384 
3385 	len = strlen(buf);
3386 	STORE_SCALAR(buf, len);
3387 	TRACEME(("ok (dummy \"%s\", length = %"IVdf")", buf, (IV) len));
3388 
3389 	return 0;
3390 }
3391 
3392 /***
3393  *** Store driving routines
3394  ***/
3395 
3396 /*
3397  * sv_type
3398  *
3399  * WARNING: partially duplicates Perl's sv_reftype for speed.
3400  *
3401  * Returns the type of the SV, identified by an integer. That integer
3402  * may then be used to index the dynamic routine dispatch table.
3403  */
sv_type(pTHX_ SV * sv)3404 static int sv_type(pTHX_ SV *sv)
3405 {
3406 	switch (SvTYPE(sv)) {
3407 	case SVt_NULL:
3408 	case SVt_IV:
3409 	case SVt_NV:
3410 		/*
3411 		 * No need to check for ROK, that can't be set here since there
3412 		 * is no field capable of hodling the xrv_rv reference.
3413 		 */
3414 		return svis_SCALAR;
3415 	case SVt_PV:
3416 	case SVt_RV:
3417 	case SVt_PVIV:
3418 	case SVt_PVNV:
3419 		/*
3420 		 * Starting from SVt_PV, it is possible to have the ROK flag
3421 		 * set, the pointer to the other SV being either stored in
3422 		 * the xrv_rv (in the case of a pure SVt_RV), or as the
3423 		 * xpv_pv field of an SVt_PV and its heirs.
3424 		 *
3425 		 * However, those SV cannot be magical or they would be an
3426 		 * SVt_PVMG at least.
3427 		 */
3428 		return SvROK(sv) ? svis_REF : svis_SCALAR;
3429 	case SVt_PVMG:
3430 	case SVt_PVLV:		/* Workaround for perl5.004_04 "LVALUE" bug */
3431 		if (SvRMAGICAL(sv) && (mg_find(sv, 'p')))
3432 			return svis_TIED_ITEM;
3433 		/* FALL THROUGH */
3434 	case SVt_PVBM:
3435 		if (SvRMAGICAL(sv) && (mg_find(sv, 'q')))
3436 			return svis_TIED;
3437 		return SvROK(sv) ? svis_REF : svis_SCALAR;
3438 	case SVt_PVAV:
3439 		if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
3440 			return svis_TIED;
3441 		return svis_ARRAY;
3442 	case SVt_PVHV:
3443 		if (SvRMAGICAL(sv) && (mg_find(sv, 'P')))
3444 			return svis_TIED;
3445 		return svis_HASH;
3446 	case SVt_PVCV:
3447 		return svis_CODE;
3448 	default:
3449 		break;
3450 	}
3451 
3452 	return svis_OTHER;
3453 }
3454 
3455 /*
3456  * store
3457  *
3458  * Recursively store objects pointed to by the sv to the specified file.
3459  *
3460  * Layout is <content> or SX_OBJECT <tagnum> if we reach an already stored
3461  * object (one for which storage has started -- it may not be over if we have
3462  * a self-referenced structure). This data set forms a stored <object>.
3463  */
store(pTHX_ stcxt_t * cxt,SV * sv)3464 static int store(pTHX_ stcxt_t *cxt, SV *sv)
3465 {
3466 	SV **svh;
3467 	int ret;
3468 	int type;
3469 #ifdef USE_PTR_TABLE
3470 	struct ptr_tbl *pseen = cxt->pseen;
3471 #else
3472 	HV *hseen = cxt->hseen;
3473 #endif
3474 
3475 	TRACEME(("store (0x%"UVxf")", PTR2UV(sv)));
3476 
3477 	/*
3478 	 * If object has already been stored, do not duplicate data.
3479 	 * Simply emit the SX_OBJECT marker followed by its tag data.
3480 	 * The tag is always written in network order.
3481 	 *
3482 	 * NOTA BENE, for 64-bit machines: the "*svh" below does not yield a
3483 	 * real pointer, rather a tag number (watch the insertion code below).
3484 	 * That means it probably safe to assume it is well under the 32-bit limit,
3485 	 * and makes the truncation safe.
3486 	 *		-- RAM, 14/09/1999
3487 	 */
3488 
3489 #ifdef USE_PTR_TABLE
3490 	svh = ptr_table_fetch(pseen, sv);
3491 #else
3492 	svh = hv_fetch(hseen, (char *) &sv, sizeof(sv), FALSE);
3493 #endif
3494 	if (svh) {
3495 		I32 tagval;
3496 
3497 		if (sv == &PL_sv_undef) {
3498 			/* We have seen PL_sv_undef before, but fake it as
3499 			   if we have not.
3500 
3501 			   Not the simplest solution to making restricted
3502 			   hashes work on 5.8.0, but it does mean that
3503 			   repeated references to the one true undef will
3504 			   take up less space in the output file.
3505 			*/
3506 			/* Need to jump past the next hv_store, because on the
3507 			   second store of undef the old hash value will be
3508 			   SvREFCNT_dec()ed, and as Storable cheats horribly
3509 			   by storing non-SVs in the hash a SEGV will ensure.
3510 			   Need to increase the tag number so that the
3511 			   receiver has no idea what games we're up to.  This
3512 			   special casing doesn't affect hooks that store
3513 			   undef, as the hook routine does its own lookup into
3514 			   hseen.  Also this means that any references back
3515 			   to PL_sv_undef (from the pathological case of hooks
3516 			   storing references to it) will find the seen hash
3517 			   entry for the first time, as if we didn't have this
3518 			   hackery here. (That hseen lookup works even on 5.8.0
3519 			   because it's a key of &PL_sv_undef and a value
3520 			   which is a tag number, not a value which is
3521 			   PL_sv_undef.)  */
3522 			cxt->tagnum++;
3523 			type = svis_SCALAR;
3524 			goto undef_special_case;
3525 		}
3526 
3527 #ifdef USE_PTR_TABLE
3528 		tagval = htonl(LOW_32BITS(((char *)svh)-1));
3529 #else
3530 		tagval = htonl(LOW_32BITS(*svh));
3531 #endif
3532 
3533 		TRACEME(("object 0x%"UVxf" seen as #%d", PTR2UV(sv), ntohl(tagval)));
3534 
3535 		PUTMARK(SX_OBJECT);
3536 		WRITE_I32(tagval);
3537 		return 0;
3538 	}
3539 
3540 	/*
3541 	 * Allocate a new tag and associate it with the address of the sv being
3542 	 * stored, before recursing...
3543 	 *
3544 	 * In order to avoid creating new SvIVs to hold the tagnum we just
3545 	 * cast the tagnum to an SV pointer and store that in the hash.  This
3546 	 * means that we must clean up the hash manually afterwards, but gives
3547 	 * us a 15% throughput increase.
3548 	 *
3549 	 */
3550 
3551 	cxt->tagnum++;
3552 #ifdef USE_PTR_TABLE
3553 	ptr_table_store(pseen, sv, INT2PTR(SV*, 1 + cxt->tagnum));
3554 #else
3555 	if (!hv_store(hseen,
3556 			(char *) &sv, sizeof(sv), INT2PTR(SV*, cxt->tagnum), 0))
3557 		return -1;
3558 #endif
3559 
3560 	/*
3561 	 * Store `sv' and everything beneath it, using appropriate routine.
3562 	 * Abort immediately if we get a non-zero status back.
3563 	 */
3564 
3565 	type = sv_type(aTHX_ sv);
3566 
3567 undef_special_case:
3568 	TRACEME(("storing 0x%"UVxf" tag #%d, type %d...",
3569 		 PTR2UV(sv), cxt->tagnum, type));
3570 
3571 	if (SvOBJECT(sv)) {
3572 		HV *pkg = SvSTASH(sv);
3573 		ret = store_blessed(aTHX_ cxt, sv, type, pkg);
3574 	} else
3575 		ret = SV_STORE(type)(aTHX_ cxt, sv);
3576 
3577 	TRACEME(("%s (stored 0x%"UVxf", refcnt=%d, %s)",
3578 		ret ? "FAILED" : "ok", PTR2UV(sv),
3579 		SvREFCNT(sv), sv_reftype(sv, FALSE)));
3580 
3581 	return ret;
3582 }
3583 
3584 /*
3585  * magic_write
3586  *
3587  * Write magic number and system information into the file.
3588  * Layout is <magic> <network> [<len> <byteorder> <sizeof int> <sizeof long>
3589  * <sizeof ptr>] where <len> is the length of the byteorder hexa string.
3590  * All size and lenghts are written as single characters here.
3591  *
3592  * Note that no byte ordering info is emitted when <network> is true, since
3593  * integers will be emitted in network order in that case.
3594  */
magic_write(pTHX_ stcxt_t * cxt)3595 static int magic_write(pTHX_ stcxt_t *cxt)
3596 {
3597     /*
3598      * Starting with 0.6, the "use_network_order" byte flag is also used to
3599      * indicate the version number of the binary image, encoded in the upper
3600      * bits. The bit 0 is always used to indicate network order.
3601      */
3602     /*
3603      * Starting with 0.7, a full byte is dedicated to the minor version of
3604      * the binary format, which is incremented only when new markers are
3605      * introduced, for instance, but when backward compatibility is preserved.
3606      */
3607 
3608     /* Make these at compile time.  The WRITE() macro is sufficiently complex
3609        that it saves about 200 bytes doing it this way and only using it
3610        once.  */
3611     static const unsigned char network_file_header[] = {
3612         MAGICSTR_BYTES,
3613         (STORABLE_BIN_MAJOR << 1) | 1,
3614         STORABLE_BIN_WRITE_MINOR
3615     };
3616     static const unsigned char file_header[] = {
3617         MAGICSTR_BYTES,
3618         (STORABLE_BIN_MAJOR << 1) | 0,
3619         STORABLE_BIN_WRITE_MINOR,
3620         /* sizeof the array includes the 0 byte at the end:  */
3621         (char) sizeof (byteorderstr) - 1,
3622         BYTEORDER_BYTES,
3623         (unsigned char) sizeof(int),
3624 	(unsigned char) sizeof(long),
3625         (unsigned char) sizeof(char *),
3626 	(unsigned char) sizeof(NV)
3627     };
3628 #ifdef USE_56_INTERWORK_KLUDGE
3629     static const unsigned char file_header_56[] = {
3630         MAGICSTR_BYTES,
3631         (STORABLE_BIN_MAJOR << 1) | 0,
3632         STORABLE_BIN_WRITE_MINOR,
3633         /* sizeof the array includes the 0 byte at the end:  */
3634         (char) sizeof (byteorderstr_56) - 1,
3635         BYTEORDER_BYTES_56,
3636         (unsigned char) sizeof(int),
3637 	(unsigned char) sizeof(long),
3638         (unsigned char) sizeof(char *),
3639 	(unsigned char) sizeof(NV)
3640     };
3641 #endif
3642     const unsigned char *header;
3643     SSize_t length;
3644 
3645     TRACEME(("magic_write on fd=%d", cxt->fio ? PerlIO_fileno(cxt->fio) : -1));
3646 
3647     if (cxt->netorder) {
3648         header = network_file_header;
3649         length = sizeof (network_file_header);
3650     } else {
3651 #ifdef USE_56_INTERWORK_KLUDGE
3652         if (SvTRUE(perl_get_sv("Storable::interwork_56_64bit", TRUE))) {
3653             header = file_header_56;
3654             length = sizeof (file_header_56);
3655         } else
3656 #endif
3657         {
3658             header = file_header;
3659             length = sizeof (file_header);
3660         }
3661     }
3662 
3663     if (!cxt->fio) {
3664         /* sizeof the array includes the 0 byte at the end.  */
3665         header += sizeof (magicstr) - 1;
3666         length -= sizeof (magicstr) - 1;
3667     }
3668 
3669     WRITE( (unsigned char*) header, length);
3670 
3671     if (!cxt->netorder) {
3672 	TRACEME(("ok (magic_write byteorder = 0x%lx [%d], I%d L%d P%d D%d)",
3673 		 (unsigned long) BYTEORDER, (int) sizeof (byteorderstr) - 1,
3674 		 (int) sizeof(int), (int) sizeof(long),
3675 		 (int) sizeof(char *), (int) sizeof(NV)));
3676     }
3677     return 0;
3678 }
3679 
3680 /*
3681  * do_store
3682  *
3683  * Common code for store operations.
3684  *
3685  * When memory store is requested (f = NULL) and a non null SV* is given in
3686  * `res', it is filled with a new SV created out of the memory buffer.
3687  *
3688  * It is required to provide a non-null `res' when the operation type is not
3689  * dclone() and store() is performed to memory.
3690  */
do_store(pTHX_ PerlIO * f,SV * sv,int optype,int network_order,SV ** res)3691 static int do_store(
3692         pTHX_
3693 	PerlIO *f,
3694 	SV *sv,
3695 	int optype,
3696 	int network_order,
3697 	SV **res)
3698 {
3699 	dSTCXT;
3700 	int status;
3701 
3702 	ASSERT(!(f == 0 && !(optype & ST_CLONE)) || res,
3703 		("must supply result SV pointer for real recursion to memory"));
3704 
3705 	TRACEME(("do_store (optype=%d, netorder=%d)",
3706 		optype, network_order));
3707 
3708 	optype |= ST_STORE;
3709 
3710 	/*
3711 	 * Workaround for CROAK leak: if they enter with a "dirty" context,
3712 	 * free up memory for them now.
3713 	 */
3714 
3715 	if (cxt->s_dirty)
3716 		clean_context(aTHX_ cxt);
3717 
3718 	/*
3719 	 * Now that STORABLE_xxx hooks exist, it is possible that they try to
3720 	 * re-enter store() via the hooks.  We need to stack contexts.
3721 	 */
3722 
3723 	if (cxt->entry)
3724 		cxt = allocate_context(aTHX_ cxt);
3725 
3726 	cxt->entry++;
3727 
3728 	ASSERT(cxt->entry == 1, ("starting new recursion"));
3729 	ASSERT(!cxt->s_dirty, ("clean context"));
3730 
3731 	/*
3732 	 * Ensure sv is actually a reference. From perl, we called something
3733 	 * like:
3734 	 *       pstore(aTHX_ FILE, \@array);
3735 	 * so we must get the scalar value behing that reference.
3736 	 */
3737 
3738 	if (!SvROK(sv))
3739 		CROAK(("Not a reference"));
3740 	sv = SvRV(sv);			/* So follow it to know what to store */
3741 
3742 	/*
3743 	 * If we're going to store to memory, reset the buffer.
3744 	 */
3745 
3746 	if (!f)
3747 		MBUF_INIT(0);
3748 
3749 	/*
3750 	 * Prepare context and emit headers.
3751 	 */
3752 
3753 	init_store_context(aTHX_ cxt, f, optype, network_order);
3754 
3755 	if (-1 == magic_write(aTHX_ cxt))		/* Emit magic and ILP info */
3756 		return 0;					/* Error */
3757 
3758 	/*
3759 	 * Recursively store object...
3760 	 */
3761 
3762 	ASSERT(is_storing(aTHX), ("within store operation"));
3763 
3764 	status = store(aTHX_ cxt, sv);		/* Just do it! */
3765 
3766 	/*
3767 	 * If they asked for a memory store and they provided an SV pointer,
3768 	 * make an SV string out of the buffer and fill their pointer.
3769 	 *
3770 	 * When asking for ST_REAL, it's MANDATORY for the caller to provide
3771 	 * an SV, since context cleanup might free the buffer if we did recurse.
3772 	 * (unless caller is dclone(), which is aware of that).
3773 	 */
3774 
3775 	if (!cxt->fio && res)
3776 		*res = mbuf2sv(aTHX);
3777 
3778 	/*
3779 	 * Final cleanup.
3780 	 *
3781 	 * The "root" context is never freed, since it is meant to be always
3782 	 * handy for the common case where no recursion occurs at all (i.e.
3783 	 * we enter store() outside of any Storable code and leave it, period).
3784 	 * We know it's the "root" context because there's nothing stacked
3785 	 * underneath it.
3786 	 *
3787 	 * OPTIMIZATION:
3788 	 *
3789 	 * When deep cloning, we don't free the context: doing so would force
3790 	 * us to copy the data in the memory buffer.  Sicne we know we're
3791 	 * about to enter do_retrieve...
3792 	 */
3793 
3794 	clean_store_context(aTHX_ cxt);
3795 	if (cxt->prev && !(cxt->optype & ST_CLONE))
3796 		free_context(aTHX_ cxt);
3797 
3798 	TRACEME(("do_store returns %d", status));
3799 
3800 	return status == 0;
3801 }
3802 
3803 /*
3804  * pstore
3805  *
3806  * Store the transitive data closure of given object to disk.
3807  * Returns 0 on error, a true value otherwise.
3808  */
pstore(pTHX_ PerlIO * f,SV * sv)3809 int pstore(pTHX_ PerlIO *f, SV *sv)
3810 {
3811 	TRACEME(("pstore"));
3812 	return do_store(aTHX_ f, sv, 0, FALSE, (SV**) 0);
3813 
3814 }
3815 
3816 /*
3817  * net_pstore
3818  *
3819  * Same as pstore(), but network order is used for integers and doubles are
3820  * emitted as strings.
3821  */
net_pstore(pTHX_ PerlIO * f,SV * sv)3822 int net_pstore(pTHX_ PerlIO *f, SV *sv)
3823 {
3824 	TRACEME(("net_pstore"));
3825 	return do_store(aTHX_ f, sv, 0, TRUE, (SV**) 0);
3826 }
3827 
3828 /***
3829  *** Memory stores.
3830  ***/
3831 
3832 /*
3833  * mbuf2sv
3834  *
3835  * Build a new SV out of the content of the internal memory buffer.
3836  */
mbuf2sv(pTHX)3837 static SV *mbuf2sv(pTHX)
3838 {
3839 	dSTCXT;
3840 
3841 	return newSVpv(mbase, MBUF_SIZE());
3842 }
3843 
3844 /*
3845  * mstore
3846  *
3847  * Store the transitive data closure of given object to memory.
3848  * Returns undef on error, a scalar value containing the data otherwise.
3849  */
mstore(pTHX_ SV * sv)3850 SV *mstore(pTHX_ SV *sv)
3851 {
3852 	SV *out;
3853 
3854 	TRACEME(("mstore"));
3855 
3856 	if (!do_store(aTHX_ (PerlIO*) 0, sv, 0, FALSE, &out))
3857 		return &PL_sv_undef;
3858 
3859 	return out;
3860 }
3861 
3862 /*
3863  * net_mstore
3864  *
3865  * Same as mstore(), but network order is used for integers and doubles are
3866  * emitted as strings.
3867  */
net_mstore(pTHX_ SV * sv)3868 SV *net_mstore(pTHX_ SV *sv)
3869 {
3870 	SV *out;
3871 
3872 	TRACEME(("net_mstore"));
3873 
3874 	if (!do_store(aTHX_ (PerlIO*) 0, sv, 0, TRUE, &out))
3875 		return &PL_sv_undef;
3876 
3877 	return out;
3878 }
3879 
3880 /***
3881  *** Specific retrieve callbacks.
3882  ***/
3883 
3884 /*
3885  * retrieve_other
3886  *
3887  * Return an error via croak, since it is not possible that we get here
3888  * under normal conditions, when facing a file produced via pstore().
3889  */
retrieve_other(pTHX_ stcxt_t * cxt,char * cname)3890 static SV *retrieve_other(pTHX_ stcxt_t *cxt, char *cname)
3891 {
3892 	if (
3893 		cxt->ver_major != STORABLE_BIN_MAJOR &&
3894 		cxt->ver_minor != STORABLE_BIN_MINOR
3895 	) {
3896 		CROAK(("Corrupted storable %s (binary v%d.%d), current is v%d.%d",
3897 			cxt->fio ? "file" : "string",
3898 			cxt->ver_major, cxt->ver_minor,
3899 			STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
3900 	} else {
3901 		CROAK(("Corrupted storable %s (binary v%d.%d)",
3902 			cxt->fio ? "file" : "string",
3903 			cxt->ver_major, cxt->ver_minor));
3904 	}
3905 
3906 	return (SV *) 0;		/* Just in case */
3907 }
3908 
3909 /*
3910  * retrieve_idx_blessed
3911  *
3912  * Layout is SX_IX_BLESS <index> <object> with SX_IX_BLESS already read.
3913  * <index> can be coded on either 1 or 5 bytes.
3914  */
retrieve_idx_blessed(pTHX_ stcxt_t * cxt,char * cname)3915 static SV *retrieve_idx_blessed(pTHX_ stcxt_t *cxt, char *cname)
3916 {
3917 	I32 idx;
3918 	char *classname;
3919 	SV **sva;
3920 	SV *sv;
3921 
3922 	TRACEME(("retrieve_idx_blessed (#%d)", cxt->tagnum));
3923 	ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3924 
3925 	GETMARK(idx);			/* Index coded on a single char? */
3926 	if (idx & 0x80)
3927 		RLEN(idx);
3928 
3929 	/*
3930 	 * Fetch classname in `aclass'
3931 	 */
3932 
3933 	sva = av_fetch(cxt->aclass, idx, FALSE);
3934 	if (!sva)
3935 		CROAK(("Class name #%"IVdf" should have been seen already", (IV) idx));
3936 
3937 	classname = SvPVX(*sva);	/* We know it's a PV, by construction */
3938 
3939 	TRACEME(("class ID %d => %s", idx, classname));
3940 
3941 	/*
3942 	 * Retrieve object and bless it.
3943 	 */
3944 
3945 	sv = retrieve(aTHX_ cxt, classname);	/* First SV which is SEEN will be blessed */
3946 
3947 	return sv;
3948 }
3949 
3950 /*
3951  * retrieve_blessed
3952  *
3953  * Layout is SX_BLESS <len> <classname> <object> with SX_BLESS already read.
3954  * <len> can be coded on either 1 or 5 bytes.
3955  */
retrieve_blessed(pTHX_ stcxt_t * cxt,char * cname)3956 static SV *retrieve_blessed(pTHX_ stcxt_t *cxt, char *cname)
3957 {
3958 	I32 len;
3959 	SV *sv;
3960 	char buf[LG_BLESS + 1];		/* Avoid malloc() if possible */
3961 	char *classname = buf;
3962 
3963 	TRACEME(("retrieve_blessed (#%d)", cxt->tagnum));
3964 	ASSERT(!cname, ("no bless-into class given here, got %s", cname));
3965 
3966 	/*
3967 	 * Decode class name length and read that name.
3968 	 *
3969 	 * Short classnames have two advantages: their length is stored on one
3970 	 * single byte, and the string can be read on the stack.
3971 	 */
3972 
3973 	GETMARK(len);			/* Length coded on a single char? */
3974 	if (len & 0x80) {
3975 		RLEN(len);
3976 		TRACEME(("** allocating %d bytes for class name", len+1));
3977 		New(10003, classname, len+1, char);
3978 	}
3979 	READ(classname, len);
3980 	classname[len] = '\0';		/* Mark string end */
3981 
3982 	/*
3983 	 * It's a new classname, otherwise it would have been an SX_IX_BLESS.
3984 	 */
3985 
3986 	TRACEME(("new class name \"%s\" will bear ID = %d", classname, cxt->classnum));
3987 
3988 	if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(classname, len)))
3989 		return (SV *) 0;
3990 
3991 	/*
3992 	 * Retrieve object and bless it.
3993 	 */
3994 
3995 	sv = retrieve(aTHX_ cxt, classname);	/* First SV which is SEEN will be blessed */
3996 	if (classname != buf)
3997 		Safefree(classname);
3998 
3999 	return sv;
4000 }
4001 
4002 /*
4003  * retrieve_hook
4004  *
4005  * Layout: SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>]
4006  * with leading mark already read, as usual.
4007  *
4008  * When recursion was involved during serialization of the object, there
4009  * is an unknown amount of serialized objects after the SX_HOOK mark.  Until
4010  * we reach a <flags> marker with the recursion bit cleared.
4011  *
4012  * If the first <flags> byte contains a type of SHT_EXTRA, then the real type
4013  * is held in the <extra> byte, and if the object is tied, the serialized
4014  * magic object comes at the very end:
4015  *
4016  *     SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object>
4017  *
4018  * This means the STORABLE_thaw hook will NOT get a tied variable during its
4019  * processing (since we won't have seen the magic object by the time the hook
4020  * is called).  See comments below for why it was done that way.
4021  */
retrieve_hook(pTHX_ stcxt_t * cxt,char * cname)4022 static SV *retrieve_hook(pTHX_ stcxt_t *cxt, char *cname)
4023 {
4024 	I32 len;
4025 	char buf[LG_BLESS + 1];		/* Avoid malloc() if possible */
4026 	char *classname = buf;
4027 	unsigned int flags;
4028 	I32 len2;
4029 	SV *frozen;
4030 	I32 len3 = 0;
4031 	AV *av = 0;
4032 	SV *hook;
4033 	SV *sv;
4034 	SV *rv;
4035 	GV *attach;
4036 	int obj_type;
4037 	int clone = cxt->optype & ST_CLONE;
4038 	char mtype = '\0';
4039 	unsigned int extra_type = 0;
4040 
4041 	TRACEME(("retrieve_hook (#%d)", cxt->tagnum));
4042 	ASSERT(!cname, ("no bless-into class given here, got %s", cname));
4043 
4044 	/*
4045 	 * Read flags, which tell us about the type, and whether we need to recurse.
4046 	 */
4047 
4048 	GETMARK(flags);
4049 
4050 	/*
4051 	 * Create the (empty) object, and mark it as seen.
4052 	 *
4053 	 * This must be done now, because tags are incremented, and during
4054 	 * serialization, the object tag was affected before recursion could
4055 	 * take place.
4056 	 */
4057 
4058 	obj_type = flags & SHF_TYPE_MASK;
4059 	switch (obj_type) {
4060 	case SHT_SCALAR:
4061 		sv = newSV(0);
4062 		break;
4063 	case SHT_ARRAY:
4064 		sv = (SV *) newAV();
4065 		break;
4066 	case SHT_HASH:
4067 		sv = (SV *) newHV();
4068 		break;
4069 	case SHT_EXTRA:
4070 		/*
4071 		 * Read <extra> flag to know the type of the object.
4072 		 * Record associated magic type for later.
4073 		 */
4074 		GETMARK(extra_type);
4075 		switch (extra_type) {
4076 		case SHT_TSCALAR:
4077 			sv = newSV(0);
4078 			mtype = 'q';
4079 			break;
4080 		case SHT_TARRAY:
4081 			sv = (SV *) newAV();
4082 			mtype = 'P';
4083 			break;
4084 		case SHT_THASH:
4085 			sv = (SV *) newHV();
4086 			mtype = 'P';
4087 			break;
4088 		default:
4089 			return retrieve_other(aTHX_ cxt, 0);	/* Let it croak */
4090 		}
4091 		break;
4092 	default:
4093 		return retrieve_other(aTHX_ cxt, 0);		/* Let it croak */
4094 	}
4095 	SEEN(sv, 0, 0);							/* Don't bless yet */
4096 
4097 	/*
4098 	 * Whilst flags tell us to recurse, do so.
4099 	 *
4100 	 * We don't need to remember the addresses returned by retrieval, because
4101 	 * all the references will be obtained through indirection via the object
4102 	 * tags in the object-ID list.
4103 	 *
4104 	 * We need to decrement the reference count for these objects
4105 	 * because, if the user doesn't save a reference to them in the hook,
4106 	 * they must be freed when this context is cleaned.
4107 	 */
4108 
4109 	while (flags & SHF_NEED_RECURSE) {
4110 		TRACEME(("retrieve_hook recursing..."));
4111 		rv = retrieve(aTHX_ cxt, 0);
4112 		if (!rv)
4113 			return (SV *) 0;
4114 		SvREFCNT_dec(rv);
4115 		TRACEME(("retrieve_hook back with rv=0x%"UVxf,
4116 			 PTR2UV(rv)));
4117 		GETMARK(flags);
4118 	}
4119 
4120 	if (flags & SHF_IDX_CLASSNAME) {
4121 		SV **sva;
4122 		I32 idx;
4123 
4124 		/*
4125 		 * Fetch index from `aclass'
4126 		 */
4127 
4128 		if (flags & SHF_LARGE_CLASSLEN)
4129 			RLEN(idx);
4130 		else
4131 			GETMARK(idx);
4132 
4133 		sva = av_fetch(cxt->aclass, idx, FALSE);
4134 		if (!sva)
4135 			CROAK(("Class name #%"IVdf" should have been seen already",
4136 				(IV) idx));
4137 
4138 		classname = SvPVX(*sva);	/* We know it's a PV, by construction */
4139 		TRACEME(("class ID %d => %s", idx, classname));
4140 
4141 	} else {
4142 		/*
4143 		 * Decode class name length and read that name.
4144 		 *
4145 		 * NOTA BENE: even if the length is stored on one byte, we don't read
4146 		 * on the stack.  Just like retrieve_blessed(), we limit the name to
4147 		 * LG_BLESS bytes.  This is an arbitrary decision.
4148 		 */
4149 
4150 		if (flags & SHF_LARGE_CLASSLEN)
4151 			RLEN(len);
4152 		else
4153 			GETMARK(len);
4154 
4155 		if (len > LG_BLESS) {
4156 			TRACEME(("** allocating %d bytes for class name", len+1));
4157 			New(10003, classname, len+1, char);
4158 		}
4159 
4160 		READ(classname, len);
4161 		classname[len] = '\0';		/* Mark string end */
4162 
4163 		/*
4164 		 * Record new classname.
4165 		 */
4166 
4167 		if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(classname, len)))
4168 			return (SV *) 0;
4169 	}
4170 
4171 	TRACEME(("class name: %s", classname));
4172 
4173 	/*
4174 	 * Decode user-frozen string length and read it in an SV.
4175 	 *
4176 	 * For efficiency reasons, we read data directly into the SV buffer.
4177 	 * To understand that code, read retrieve_scalar()
4178 	 */
4179 
4180 	if (flags & SHF_LARGE_STRLEN)
4181 		RLEN(len2);
4182 	else
4183 		GETMARK(len2);
4184 
4185 	frozen = NEWSV(10002, len2);
4186 	if (len2) {
4187 		SAFEREAD(SvPVX(frozen), len2, frozen);
4188 		SvCUR_set(frozen, len2);
4189 		*SvEND(frozen) = '\0';
4190 	}
4191 	(void) SvPOK_only(frozen);		/* Validates string pointer */
4192 	if (cxt->s_tainted)				/* Is input source tainted? */
4193 		SvTAINT(frozen);
4194 
4195 	TRACEME(("frozen string: %d bytes", len2));
4196 
4197 	/*
4198 	 * Decode object-ID list length, if present.
4199 	 */
4200 
4201 	if (flags & SHF_HAS_LIST) {
4202 		if (flags & SHF_LARGE_LISTLEN)
4203 			RLEN(len3);
4204 		else
4205 			GETMARK(len3);
4206 		if (len3) {
4207 			av = newAV();
4208 			av_extend(av, len3 + 1);	/* Leave room for [0] */
4209 			AvFILLp(av) = len3;			/* About to be filled anyway */
4210 		}
4211 	}
4212 
4213 	TRACEME(("has %d object IDs to link", len3));
4214 
4215 	/*
4216 	 * Read object-ID list into array.
4217 	 * Because we pre-extended it, we can cheat and fill it manually.
4218 	 *
4219 	 * We read object tags and we can convert them into SV* on the fly
4220 	 * because we know all the references listed in there (as tags)
4221 	 * have been already serialized, hence we have a valid correspondance
4222 	 * between each of those tags and the recreated SV.
4223 	 */
4224 
4225 	if (av) {
4226 		SV **ary = AvARRAY(av);
4227 		int i;
4228 		for (i = 1; i <= len3; i++) {	/* We leave [0] alone */
4229 			I32 tag;
4230 			SV **svh;
4231 			SV *xsv;
4232 
4233 			READ_I32(tag);
4234 			tag = ntohl(tag);
4235 			svh = av_fetch(cxt->aseen, tag, FALSE);
4236 			if (!svh) {
4237 				if (tag == cxt->where_is_undef) {
4238 					/* av_fetch uses PL_sv_undef internally, hence this
4239 					   somewhat gruesome hack. */
4240 					xsv = &PL_sv_undef;
4241 					svh = &xsv;
4242 				} else {
4243 					CROAK(("Object #%"IVdf" should have been retrieved already",
4244 					       (IV) tag));
4245 				}
4246 			}
4247 			xsv = *svh;
4248 			ary[i] = SvREFCNT_inc(xsv);
4249 		}
4250 	}
4251 
4252 	/*
4253 	 * Bless the object and look up the STORABLE_thaw hook.
4254 	 */
4255 
4256 	BLESS(sv, classname);
4257 
4258 	/* Handle attach case; again can't use pkg_can because it only
4259 	 * caches one method */
4260 	attach = gv_fetchmethod_autoload(SvSTASH(sv), "STORABLE_attach", FALSE);
4261 	if (attach && isGV(attach)) {
4262 	    SV* attached;
4263 	    SV* attach_hook = newRV((SV*) GvCV(attach));
4264 
4265 	    if (av)
4266 	        CROAK(("STORABLE_attach called with unexpected references"));
4267 	    av = newAV();
4268 	    av_extend(av, 1);
4269 	    AvFILLp(av) = 0;
4270 	    AvARRAY(av)[0] = SvREFCNT_inc(frozen);
4271 	    rv = newSVpv(classname, 0);
4272 	    attached = scalar_call(aTHX_ rv, attach_hook, clone, av, G_SCALAR);
4273 	    if (attached &&
4274 	        SvROK(attached) &&
4275 	        sv_derived_from(attached, classname))
4276 	        return SvRV(attached);
4277 	    CROAK(("STORABLE_attach did not return a %s object", classname));
4278 	}
4279 
4280 	hook = pkg_can(aTHX_ cxt->hook, SvSTASH(sv), "STORABLE_thaw");
4281 	if (!hook) {
4282 		/*
4283 		 * Hook not found.  Maybe they did not require the module where this
4284 		 * hook is defined yet?
4285 		 *
4286 		 * If the require below succeeds, we'll be able to find the hook.
4287 		 * Still, it only works reliably when each class is defined in a
4288 		 * file of its own.
4289 		 */
4290 
4291 		SV *psv = newSVpvn("require ", 8);
4292 		sv_catpv(psv, classname);
4293 
4294 		TRACEME(("No STORABLE_thaw defined for objects of class %s", classname));
4295 		TRACEME(("Going to require module '%s' with '%s'", classname, SvPVX(psv)));
4296 
4297 		perl_eval_sv(psv, G_DISCARD);
4298 		sv_free(psv);
4299 
4300 		/*
4301 		 * We cache results of pkg_can, so we need to uncache before attempting
4302 		 * the lookup again.
4303 		 */
4304 
4305 		pkg_uncache(aTHX_ cxt->hook, SvSTASH(sv), "STORABLE_thaw");
4306 		hook = pkg_can(aTHX_ cxt->hook, SvSTASH(sv), "STORABLE_thaw");
4307 
4308 		if (!hook)
4309 			CROAK(("No STORABLE_thaw defined for objects of class %s "
4310 					"(even after a \"require %s;\")", classname, classname));
4311 	}
4312 
4313 	/*
4314 	 * If we don't have an `av' yet, prepare one.
4315 	 * Then insert the frozen string as item [0].
4316 	 */
4317 
4318 	if (!av) {
4319 		av = newAV();
4320 		av_extend(av, 1);
4321 		AvFILLp(av) = 0;
4322 	}
4323 	AvARRAY(av)[0] = SvREFCNT_inc(frozen);
4324 
4325 	/*
4326 	 * Call the hook as:
4327 	 *
4328 	 *   $object->STORABLE_thaw($cloning, $frozen, @refs);
4329 	 *
4330 	 * where $object is our blessed (empty) object, $cloning is a boolean
4331 	 * telling whether we're running a deep clone, $frozen is the frozen
4332 	 * string the user gave us in his serializing hook, and @refs, which may
4333 	 * be empty, is the list of extra references he returned along for us
4334 	 * to serialize.
4335 	 *
4336 	 * In effect, the hook is an alternate creation routine for the class,
4337 	 * the object itself being already created by the runtime.
4338 	 */
4339 
4340 	TRACEME(("calling STORABLE_thaw on %s at 0x%"UVxf" (%"IVdf" args)",
4341 		 classname, PTR2UV(sv), (IV) AvFILLp(av) + 1));
4342 
4343 	rv = newRV(sv);
4344 	(void) scalar_call(aTHX_ rv, hook, clone, av, G_SCALAR|G_DISCARD);
4345 	SvREFCNT_dec(rv);
4346 
4347 	/*
4348 	 * Final cleanup.
4349 	 */
4350 
4351 	SvREFCNT_dec(frozen);
4352 	av_undef(av);
4353 	sv_free((SV *) av);
4354 	if (!(flags & SHF_IDX_CLASSNAME) && classname != buf)
4355 		Safefree(classname);
4356 
4357 	/*
4358 	 * If we had an <extra> type, then the object was not as simple, and
4359 	 * we need to restore extra magic now.
4360 	 */
4361 
4362 	if (!extra_type)
4363 		return sv;
4364 
4365 	TRACEME(("retrieving magic object for 0x%"UVxf"...", PTR2UV(sv)));
4366 
4367 	rv = retrieve(aTHX_ cxt, 0);		/* Retrieve <magic object> */
4368 
4369 	TRACEME(("restoring the magic object 0x%"UVxf" part of 0x%"UVxf,
4370 		PTR2UV(rv), PTR2UV(sv)));
4371 
4372 	switch (extra_type) {
4373 	case SHT_TSCALAR:
4374 		sv_upgrade(sv, SVt_PVMG);
4375 		break;
4376 	case SHT_TARRAY:
4377 		sv_upgrade(sv, SVt_PVAV);
4378 		AvREAL_off((AV *)sv);
4379 		break;
4380 	case SHT_THASH:
4381 		sv_upgrade(sv, SVt_PVHV);
4382 		break;
4383 	default:
4384 		CROAK(("Forgot to deal with extra type %d", extra_type));
4385 		break;
4386 	}
4387 
4388 	/*
4389 	 * Adding the magic only now, well after the STORABLE_thaw hook was called
4390 	 * means the hook cannot know it deals with an object whose variable is
4391 	 * tied.  But this is happening when retrieving $o in the following case:
4392 	 *
4393 	 *	my %h;
4394 	 *  tie %h, 'FOO';
4395 	 *	my $o = bless \%h, 'BAR';
4396 	 *
4397 	 * The 'BAR' class is NOT the one where %h is tied into.  Therefore, as
4398 	 * far as the 'BAR' class is concerned, the fact that %h is not a REAL
4399 	 * hash but a tied one should not matter at all, and remain transparent.
4400 	 * This means the magic must be restored by Storable AFTER the hook is
4401 	 * called.
4402 	 *
4403 	 * That looks very reasonable to me, but then I've come up with this
4404 	 * after a bug report from David Nesting, who was trying to store such
4405 	 * an object and caused Storable to fail.  And unfortunately, it was
4406 	 * also the easiest way to retrofit support for blessed ref to tied objects
4407 	 * into the existing design.  -- RAM, 17/02/2001
4408 	 */
4409 
4410 	sv_magic(sv, rv, mtype, Nullch, 0);
4411 	SvREFCNT_dec(rv);			/* Undo refcnt inc from sv_magic() */
4412 
4413 	return sv;
4414 }
4415 
4416 /*
4417  * retrieve_ref
4418  *
4419  * Retrieve reference to some other scalar.
4420  * Layout is SX_REF <object>, with SX_REF already read.
4421  */
retrieve_ref(pTHX_ stcxt_t * cxt,char * cname)4422 static SV *retrieve_ref(pTHX_ stcxt_t *cxt, char *cname)
4423 {
4424 	SV *rv;
4425 	SV *sv;
4426 
4427 	TRACEME(("retrieve_ref (#%d)", cxt->tagnum));
4428 
4429 	/*
4430 	 * We need to create the SV that holds the reference to the yet-to-retrieve
4431 	 * object now, so that we may record the address in the seen table.
4432 	 * Otherwise, if the object to retrieve references us, we won't be able
4433 	 * to resolve the SX_OBJECT we'll see at that point! Hence we cannot
4434 	 * do the retrieve first and use rv = newRV(sv) since it will be too late
4435 	 * for SEEN() recording.
4436 	 */
4437 
4438 	rv = NEWSV(10002, 0);
4439 	SEEN(rv, cname, 0);		/* Will return if rv is null */
4440 	sv = retrieve(aTHX_ cxt, 0);	/* Retrieve <object> */
4441 	if (!sv)
4442 		return (SV *) 0;	/* Failed */
4443 
4444 	/*
4445 	 * WARNING: breaks RV encapsulation.
4446 	 *
4447 	 * Now for the tricky part. We have to upgrade our existing SV, so that
4448 	 * it is now an RV on sv... Again, we cheat by duplicating the code
4449 	 * held in newSVrv(), since we already got our SV from retrieve().
4450 	 *
4451 	 * We don't say:
4452 	 *
4453 	 *		SvRV(rv) = SvREFCNT_inc(sv);
4454 	 *
4455 	 * here because the reference count we got from retrieve() above is
4456 	 * already correct: if the object was retrieved from the file, then
4457 	 * its reference count is one. Otherwise, if it was retrieved via
4458 	 * an SX_OBJECT indication, a ref count increment was done.
4459 	 */
4460 
4461 	if (cname) {
4462 		/* No need to do anything, as rv will already be PVMG.  */
4463 		assert (SvTYPE(rv) >= SVt_RV);
4464 	} else {
4465 		sv_upgrade(rv, SVt_RV);
4466 	}
4467 
4468 	SvRV_set(rv, sv);				/* $rv = \$sv */
4469 	SvROK_on(rv);
4470 
4471 	TRACEME(("ok (retrieve_ref at 0x%"UVxf")", PTR2UV(rv)));
4472 
4473 	return rv;
4474 }
4475 
4476 /*
4477  * retrieve_weakref
4478  *
4479  * Retrieve weak reference to some other scalar.
4480  * Layout is SX_WEAKREF <object>, with SX_WEAKREF already read.
4481  */
retrieve_weakref(pTHX_ stcxt_t * cxt,char * cname)4482 static SV *retrieve_weakref(pTHX_ stcxt_t *cxt, char *cname)
4483 {
4484 	SV *sv;
4485 
4486 	TRACEME(("retrieve_weakref (#%d)", cxt->tagnum));
4487 
4488 	sv = retrieve_ref(aTHX_ cxt, cname);
4489 	if (sv) {
4490 #ifdef SvWEAKREF
4491 		sv_rvweaken(sv);
4492 #else
4493 		WEAKREF_CROAK();
4494 #endif
4495 	}
4496 	return sv;
4497 }
4498 
4499 /*
4500  * retrieve_overloaded
4501  *
4502  * Retrieve reference to some other scalar with overloading.
4503  * Layout is SX_OVERLOAD <object>, with SX_OVERLOAD already read.
4504  */
retrieve_overloaded(pTHX_ stcxt_t * cxt,char * cname)4505 static SV *retrieve_overloaded(pTHX_ stcxt_t *cxt, char *cname)
4506 {
4507 	SV *rv;
4508 	SV *sv;
4509 	HV *stash;
4510 
4511 	TRACEME(("retrieve_overloaded (#%d)", cxt->tagnum));
4512 
4513 	/*
4514 	 * Same code as retrieve_ref(), duplicated to avoid extra call.
4515 	 */
4516 
4517 	rv = NEWSV(10002, 0);
4518 	SEEN(rv, cname, 0);		/* Will return if rv is null */
4519 	sv = retrieve(aTHX_ cxt, 0);	/* Retrieve <object> */
4520 	if (!sv)
4521 		return (SV *) 0;	/* Failed */
4522 
4523 	/*
4524 	 * WARNING: breaks RV encapsulation.
4525 	 */
4526 
4527 	sv_upgrade(rv, SVt_RV);
4528 	SvRV_set(rv, sv);				/* $rv = \$sv */
4529 	SvROK_on(rv);
4530 
4531 	/*
4532 	 * Restore overloading magic.
4533 	 */
4534 
4535 	stash = SvTYPE(sv) ? (HV *) SvSTASH (sv) : 0;
4536 	if (!stash) {
4537 		CROAK(("Cannot restore overloading on %s(0x%"UVxf
4538 		       ") (package <unknown>)",
4539 		       sv_reftype(sv, FALSE),
4540 		       PTR2UV(sv)));
4541 	}
4542 	if (!Gv_AMG(stash)) {
4543 		SV *psv = newSVpvn("require ", 8);
4544 		const char *package = HvNAME_get(stash);
4545 		sv_catpv(psv, package);
4546 
4547 		TRACEME(("No overloading defined for package %s", package));
4548 		TRACEME(("Going to require module '%s' with '%s'", package, SvPVX(psv)));
4549 
4550 		perl_eval_sv(psv, G_DISCARD);
4551 		sv_free(psv);
4552 		if (!Gv_AMG(stash)) {
4553 			CROAK(("Cannot restore overloading on %s(0x%"UVxf
4554 			       ") (package %s) (even after a \"require %s;\")",
4555 			       sv_reftype(sv, FALSE),
4556 			       PTR2UV(sv),
4557 			       package, package));
4558 		}
4559 	}
4560 
4561 	SvAMAGIC_on(rv);
4562 
4563 	TRACEME(("ok (retrieve_overloaded at 0x%"UVxf")", PTR2UV(rv)));
4564 
4565 	return rv;
4566 }
4567 
4568 /*
4569  * retrieve_weakoverloaded
4570  *
4571  * Retrieve weak overloaded reference to some other scalar.
4572  * Layout is SX_WEAKOVERLOADED <object>, with SX_WEAKOVERLOADED already read.
4573  */
retrieve_weakoverloaded(pTHX_ stcxt_t * cxt,char * cname)4574 static SV *retrieve_weakoverloaded(pTHX_ stcxt_t *cxt, char *cname)
4575 {
4576 	SV *sv;
4577 
4578 	TRACEME(("retrieve_weakoverloaded (#%d)", cxt->tagnum));
4579 
4580 	sv = retrieve_overloaded(aTHX_ cxt, cname);
4581 	if (sv) {
4582 #ifdef SvWEAKREF
4583 		sv_rvweaken(sv);
4584 #else
4585 		WEAKREF_CROAK();
4586 #endif
4587 	}
4588 	return sv;
4589 }
4590 
4591 /*
4592  * retrieve_tied_array
4593  *
4594  * Retrieve tied array
4595  * Layout is SX_TIED_ARRAY <object>, with SX_TIED_ARRAY already read.
4596  */
retrieve_tied_array(pTHX_ stcxt_t * cxt,char * cname)4597 static SV *retrieve_tied_array(pTHX_ stcxt_t *cxt, char *cname)
4598 {
4599 	SV *tv;
4600 	SV *sv;
4601 
4602 	TRACEME(("retrieve_tied_array (#%d)", cxt->tagnum));
4603 
4604 	tv = NEWSV(10002, 0);
4605 	SEEN(tv, cname, 0);			/* Will return if tv is null */
4606 	sv = retrieve(aTHX_ cxt, 0);		/* Retrieve <object> */
4607 	if (!sv)
4608 		return (SV *) 0;		/* Failed */
4609 
4610 	sv_upgrade(tv, SVt_PVAV);
4611 	AvREAL_off((AV *)tv);
4612 	sv_magic(tv, sv, 'P', Nullch, 0);
4613 	SvREFCNT_dec(sv);			/* Undo refcnt inc from sv_magic() */
4614 
4615 	TRACEME(("ok (retrieve_tied_array at 0x%"UVxf")", PTR2UV(tv)));
4616 
4617 	return tv;
4618 }
4619 
4620 /*
4621  * retrieve_tied_hash
4622  *
4623  * Retrieve tied hash
4624  * Layout is SX_TIED_HASH <object>, with SX_TIED_HASH already read.
4625  */
retrieve_tied_hash(pTHX_ stcxt_t * cxt,char * cname)4626 static SV *retrieve_tied_hash(pTHX_ stcxt_t *cxt, char *cname)
4627 {
4628 	SV *tv;
4629 	SV *sv;
4630 
4631 	TRACEME(("retrieve_tied_hash (#%d)", cxt->tagnum));
4632 
4633 	tv = NEWSV(10002, 0);
4634 	SEEN(tv, cname, 0);			/* Will return if tv is null */
4635 	sv = retrieve(aTHX_ cxt, 0);		/* Retrieve <object> */
4636 	if (!sv)
4637 		return (SV *) 0;		/* Failed */
4638 
4639 	sv_upgrade(tv, SVt_PVHV);
4640 	sv_magic(tv, sv, 'P', Nullch, 0);
4641 	SvREFCNT_dec(sv);			/* Undo refcnt inc from sv_magic() */
4642 
4643 	TRACEME(("ok (retrieve_tied_hash at 0x%"UVxf")", PTR2UV(tv)));
4644 
4645 	return tv;
4646 }
4647 
4648 /*
4649  * retrieve_tied_scalar
4650  *
4651  * Retrieve tied scalar
4652  * Layout is SX_TIED_SCALAR <object>, with SX_TIED_SCALAR already read.
4653  */
retrieve_tied_scalar(pTHX_ stcxt_t * cxt,char * cname)4654 static SV *retrieve_tied_scalar(pTHX_ stcxt_t *cxt, char *cname)
4655 {
4656 	SV *tv;
4657 	SV *sv, *obj = NULL;
4658 
4659 	TRACEME(("retrieve_tied_scalar (#%d)", cxt->tagnum));
4660 
4661 	tv = NEWSV(10002, 0);
4662 	SEEN(tv, cname, 0);			/* Will return if rv is null */
4663 	sv = retrieve(aTHX_ cxt, 0);		/* Retrieve <object> */
4664 	if (!sv) {
4665 		return (SV *) 0;		/* Failed */
4666 	}
4667 	else if (SvTYPE(sv) != SVt_NULL) {
4668 		obj = sv;
4669 	}
4670 
4671 	sv_upgrade(tv, SVt_PVMG);
4672 	sv_magic(tv, obj, 'q', Nullch, 0);
4673 
4674 	if (obj) {
4675 		/* Undo refcnt inc from sv_magic() */
4676 		SvREFCNT_dec(obj);
4677 	}
4678 
4679 	TRACEME(("ok (retrieve_tied_scalar at 0x%"UVxf")", PTR2UV(tv)));
4680 
4681 	return tv;
4682 }
4683 
4684 /*
4685  * retrieve_tied_key
4686  *
4687  * Retrieve reference to value in a tied hash.
4688  * Layout is SX_TIED_KEY <object> <key>, with SX_TIED_KEY already read.
4689  */
retrieve_tied_key(pTHX_ stcxt_t * cxt,char * cname)4690 static SV *retrieve_tied_key(pTHX_ stcxt_t *cxt, char *cname)
4691 {
4692 	SV *tv;
4693 	SV *sv;
4694 	SV *key;
4695 
4696 	TRACEME(("retrieve_tied_key (#%d)", cxt->tagnum));
4697 
4698 	tv = NEWSV(10002, 0);
4699 	SEEN(tv, cname, 0);			/* Will return if tv is null */
4700 	sv = retrieve(aTHX_ cxt, 0);		/* Retrieve <object> */
4701 	if (!sv)
4702 		return (SV *) 0;		/* Failed */
4703 
4704 	key = retrieve(aTHX_ cxt, 0);		/* Retrieve <key> */
4705 	if (!key)
4706 		return (SV *) 0;		/* Failed */
4707 
4708 	sv_upgrade(tv, SVt_PVMG);
4709 	sv_magic(tv, sv, 'p', (char *)key, HEf_SVKEY);
4710 	SvREFCNT_dec(key);			/* Undo refcnt inc from sv_magic() */
4711 	SvREFCNT_dec(sv);			/* Undo refcnt inc from sv_magic() */
4712 
4713 	return tv;
4714 }
4715 
4716 /*
4717  * retrieve_tied_idx
4718  *
4719  * Retrieve reference to value in a tied array.
4720  * Layout is SX_TIED_IDX <object> <idx>, with SX_TIED_IDX already read.
4721  */
retrieve_tied_idx(pTHX_ stcxt_t * cxt,char * cname)4722 static SV *retrieve_tied_idx(pTHX_ stcxt_t *cxt, char *cname)
4723 {
4724 	SV *tv;
4725 	SV *sv;
4726 	I32 idx;
4727 
4728 	TRACEME(("retrieve_tied_idx (#%d)", cxt->tagnum));
4729 
4730 	tv = NEWSV(10002, 0);
4731 	SEEN(tv, cname, 0);			/* Will return if tv is null */
4732 	sv = retrieve(aTHX_ cxt, 0);		/* Retrieve <object> */
4733 	if (!sv)
4734 		return (SV *) 0;		/* Failed */
4735 
4736 	RLEN(idx);					/* Retrieve <idx> */
4737 
4738 	sv_upgrade(tv, SVt_PVMG);
4739 	sv_magic(tv, sv, 'p', Nullch, idx);
4740 	SvREFCNT_dec(sv);			/* Undo refcnt inc from sv_magic() */
4741 
4742 	return tv;
4743 }
4744 
4745 
4746 /*
4747  * retrieve_lscalar
4748  *
4749  * Retrieve defined long (string) scalar.
4750  *
4751  * Layout is SX_LSCALAR <length> <data>, with SX_LSCALAR already read.
4752  * The scalar is "long" in that <length> is larger than LG_SCALAR so it
4753  * was not stored on a single byte.
4754  */
retrieve_lscalar(pTHX_ stcxt_t * cxt,char * cname)4755 static SV *retrieve_lscalar(pTHX_ stcxt_t *cxt, char *cname)
4756 {
4757 	I32 len;
4758 	SV *sv;
4759 
4760 	RLEN(len);
4761 	TRACEME(("retrieve_lscalar (#%d), len = %"IVdf, cxt->tagnum, (IV) len));
4762 
4763 	/*
4764 	 * Allocate an empty scalar of the suitable length.
4765 	 */
4766 
4767 	sv = NEWSV(10002, len);
4768 	SEEN(sv, cname, 0);	/* Associate this new scalar with tag "tagnum" */
4769 
4770 	/*
4771 	 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
4772 	 *
4773 	 * Now, for efficiency reasons, read data directly inside the SV buffer,
4774 	 * and perform the SV final settings directly by duplicating the final
4775 	 * work done by sv_setpv. Since we're going to allocate lots of scalars
4776 	 * this way, it's worth the hassle and risk.
4777 	 */
4778 
4779 	SAFEREAD(SvPVX(sv), len, sv);
4780 	SvCUR_set(sv, len);				/* Record C string length */
4781 	*SvEND(sv) = '\0';				/* Ensure it's null terminated anyway */
4782 	(void) SvPOK_only(sv);			/* Validate string pointer */
4783 	if (cxt->s_tainted)				/* Is input source tainted? */
4784 		SvTAINT(sv);				/* External data cannot be trusted */
4785 
4786 	TRACEME(("large scalar len %"IVdf" '%s'", (IV) len, SvPVX(sv)));
4787 	TRACEME(("ok (retrieve_lscalar at 0x%"UVxf")", PTR2UV(sv)));
4788 
4789 	return sv;
4790 }
4791 
4792 /*
4793  * retrieve_scalar
4794  *
4795  * Retrieve defined short (string) scalar.
4796  *
4797  * Layout is SX_SCALAR <length> <data>, with SX_SCALAR already read.
4798  * The scalar is "short" so <length> is single byte. If it is 0, there
4799  * is no <data> section.
4800  */
retrieve_scalar(pTHX_ stcxt_t * cxt,char * cname)4801 static SV *retrieve_scalar(pTHX_ stcxt_t *cxt, char *cname)
4802 {
4803 	int len;
4804 	SV *sv;
4805 
4806 	GETMARK(len);
4807 	TRACEME(("retrieve_scalar (#%d), len = %d", cxt->tagnum, len));
4808 
4809 	/*
4810 	 * Allocate an empty scalar of the suitable length.
4811 	 */
4812 
4813 	sv = NEWSV(10002, len);
4814 	SEEN(sv, cname, 0);	/* Associate this new scalar with tag "tagnum" */
4815 
4816 	/*
4817 	 * WARNING: duplicates parts of sv_setpv and breaks SV data encapsulation.
4818 	 */
4819 
4820 	if (len == 0) {
4821 		/*
4822 		 * newSV did not upgrade to SVt_PV so the scalar is undefined.
4823 		 * To make it defined with an empty length, upgrade it now...
4824 		 * Don't upgrade to a PV if the original type contains more
4825 		 * information than a scalar.
4826 		 */
4827 		if (SvTYPE(sv) <= SVt_PV) {
4828 			sv_upgrade(sv, SVt_PV);
4829 		}
4830 		SvGROW(sv, 1);
4831 		*SvEND(sv) = '\0';			/* Ensure it's null terminated anyway */
4832 		TRACEME(("ok (retrieve_scalar empty at 0x%"UVxf")", PTR2UV(sv)));
4833 	} else {
4834 		/*
4835 		 * Now, for efficiency reasons, read data directly inside the SV buffer,
4836 		 * and perform the SV final settings directly by duplicating the final
4837 		 * work done by sv_setpv. Since we're going to allocate lots of scalars
4838 		 * this way, it's worth the hassle and risk.
4839 		 */
4840 		SAFEREAD(SvPVX(sv), len, sv);
4841 		SvCUR_set(sv, len);			/* Record C string length */
4842 		*SvEND(sv) = '\0';			/* Ensure it's null terminated anyway */
4843 		TRACEME(("small scalar len %d '%s'", len, SvPVX(sv)));
4844 	}
4845 
4846 	(void) SvPOK_only(sv);			/* Validate string pointer */
4847 	if (cxt->s_tainted)				/* Is input source tainted? */
4848 		SvTAINT(sv);				/* External data cannot be trusted */
4849 
4850 	TRACEME(("ok (retrieve_scalar at 0x%"UVxf")", PTR2UV(sv)));
4851 	return sv;
4852 }
4853 
4854 /*
4855  * retrieve_utf8str
4856  *
4857  * Like retrieve_scalar(), but tag result as utf8.
4858  * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
4859  */
retrieve_utf8str(pTHX_ stcxt_t * cxt,char * cname)4860 static SV *retrieve_utf8str(pTHX_ stcxt_t *cxt, char *cname)
4861 {
4862     SV *sv;
4863 
4864     TRACEME(("retrieve_utf8str"));
4865 
4866     sv = retrieve_scalar(aTHX_ cxt, cname);
4867     if (sv) {
4868 #ifdef HAS_UTF8_SCALARS
4869         SvUTF8_on(sv);
4870 #else
4871         if (cxt->use_bytes < 0)
4872             cxt->use_bytes
4873                 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4874                    ? 1 : 0);
4875         if (cxt->use_bytes == 0)
4876             UTF8_CROAK();
4877 #endif
4878     }
4879 
4880     return sv;
4881 }
4882 
4883 /*
4884  * retrieve_lutf8str
4885  *
4886  * Like retrieve_lscalar(), but tag result as utf8.
4887  * If we're retrieving UTF8 data in a non-UTF8 perl, croaks.
4888  */
retrieve_lutf8str(pTHX_ stcxt_t * cxt,char * cname)4889 static SV *retrieve_lutf8str(pTHX_ stcxt_t *cxt, char *cname)
4890 {
4891     SV *sv;
4892 
4893     TRACEME(("retrieve_lutf8str"));
4894 
4895     sv = retrieve_lscalar(aTHX_ cxt, cname);
4896     if (sv) {
4897 #ifdef HAS_UTF8_SCALARS
4898         SvUTF8_on(sv);
4899 #else
4900         if (cxt->use_bytes < 0)
4901             cxt->use_bytes
4902                 = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
4903                    ? 1 : 0);
4904         if (cxt->use_bytes == 0)
4905             UTF8_CROAK();
4906 #endif
4907     }
4908     return sv;
4909 }
4910 
4911 /*
4912  * retrieve_integer
4913  *
4914  * Retrieve defined integer.
4915  * Layout is SX_INTEGER <data>, whith SX_INTEGER already read.
4916  */
retrieve_integer(pTHX_ stcxt_t * cxt,char * cname)4917 static SV *retrieve_integer(pTHX_ stcxt_t *cxt, char *cname)
4918 {
4919 	SV *sv;
4920 	IV iv;
4921 
4922 	TRACEME(("retrieve_integer (#%d)", cxt->tagnum));
4923 
4924 	READ(&iv, sizeof(iv));
4925 	sv = newSViv(iv);
4926 	SEEN(sv, cname, 0);	/* Associate this new scalar with tag "tagnum" */
4927 
4928 	TRACEME(("integer %"IVdf, iv));
4929 	TRACEME(("ok (retrieve_integer at 0x%"UVxf")", PTR2UV(sv)));
4930 
4931 	return sv;
4932 }
4933 
4934 /*
4935  * retrieve_netint
4936  *
4937  * Retrieve defined integer in network order.
4938  * Layout is SX_NETINT <data>, whith SX_NETINT already read.
4939  */
retrieve_netint(pTHX_ stcxt_t * cxt,char * cname)4940 static SV *retrieve_netint(pTHX_ stcxt_t *cxt, char *cname)
4941 {
4942 	SV *sv;
4943 	I32 iv;
4944 
4945 	TRACEME(("retrieve_netint (#%d)", cxt->tagnum));
4946 
4947 	READ_I32(iv);
4948 #ifdef HAS_NTOHL
4949 	sv = newSViv((int) ntohl(iv));
4950 	TRACEME(("network integer %d", (int) ntohl(iv)));
4951 #else
4952 	sv = newSViv(iv);
4953 	TRACEME(("network integer (as-is) %d", iv));
4954 #endif
4955 	SEEN(sv, cname, 0);	/* Associate this new scalar with tag "tagnum" */
4956 
4957 	TRACEME(("ok (retrieve_netint at 0x%"UVxf")", PTR2UV(sv)));
4958 
4959 	return sv;
4960 }
4961 
4962 /*
4963  * retrieve_double
4964  *
4965  * Retrieve defined double.
4966  * Layout is SX_DOUBLE <data>, whith SX_DOUBLE already read.
4967  */
retrieve_double(pTHX_ stcxt_t * cxt,char * cname)4968 static SV *retrieve_double(pTHX_ stcxt_t *cxt, char *cname)
4969 {
4970 	SV *sv;
4971 	NV nv;
4972 
4973 	TRACEME(("retrieve_double (#%d)", cxt->tagnum));
4974 
4975 	READ(&nv, sizeof(nv));
4976 	sv = newSVnv(nv);
4977 	SEEN(sv, cname, 0);	/* Associate this new scalar with tag "tagnum" */
4978 
4979 	TRACEME(("double %"NVff, nv));
4980 	TRACEME(("ok (retrieve_double at 0x%"UVxf")", PTR2UV(sv)));
4981 
4982 	return sv;
4983 }
4984 
4985 /*
4986  * retrieve_byte
4987  *
4988  * Retrieve defined byte (small integer within the [-128, +127] range).
4989  * Layout is SX_BYTE <data>, whith SX_BYTE already read.
4990  */
retrieve_byte(pTHX_ stcxt_t * cxt,char * cname)4991 static SV *retrieve_byte(pTHX_ stcxt_t *cxt, char *cname)
4992 {
4993 	SV *sv;
4994 	int siv;
4995 	signed char tmp;	/* Workaround for AIX cc bug --H.Merijn Brand */
4996 
4997 	TRACEME(("retrieve_byte (#%d)", cxt->tagnum));
4998 
4999 	GETMARK(siv);
5000 	TRACEME(("small integer read as %d", (unsigned char) siv));
5001 	tmp = (unsigned char) siv - 128;
5002 	sv = newSViv(tmp);
5003 	SEEN(sv, cname, 0);	/* Associate this new scalar with tag "tagnum" */
5004 
5005 	TRACEME(("byte %d", tmp));
5006 	TRACEME(("ok (retrieve_byte at 0x%"UVxf")", PTR2UV(sv)));
5007 
5008 	return sv;
5009 }
5010 
5011 /*
5012  * retrieve_undef
5013  *
5014  * Return the undefined value.
5015  */
retrieve_undef(pTHX_ stcxt_t * cxt,char * cname)5016 static SV *retrieve_undef(pTHX_ stcxt_t *cxt, char *cname)
5017 {
5018 	SV* sv;
5019 
5020 	TRACEME(("retrieve_undef"));
5021 
5022 	sv = newSV(0);
5023 	SEEN(sv, cname, 0);
5024 
5025 	return sv;
5026 }
5027 
5028 /*
5029  * retrieve_sv_undef
5030  *
5031  * Return the immortal undefined value.
5032  */
retrieve_sv_undef(pTHX_ stcxt_t * cxt,char * cname)5033 static SV *retrieve_sv_undef(pTHX_ stcxt_t *cxt, char *cname)
5034 {
5035 	SV *sv = &PL_sv_undef;
5036 
5037 	TRACEME(("retrieve_sv_undef"));
5038 
5039 	/* Special case PL_sv_undef, as av_fetch uses it internally to mark
5040 	   deleted elements, and will return NULL (fetch failed) whenever it
5041 	   is fetched.  */
5042 	if (cxt->where_is_undef == -1) {
5043 		cxt->where_is_undef = cxt->tagnum;
5044 	}
5045 	SEEN(sv, cname, 1);
5046 	return sv;
5047 }
5048 
5049 /*
5050  * retrieve_sv_yes
5051  *
5052  * Return the immortal yes value.
5053  */
retrieve_sv_yes(pTHX_ stcxt_t * cxt,char * cname)5054 static SV *retrieve_sv_yes(pTHX_ stcxt_t *cxt, char *cname)
5055 {
5056 	SV *sv = &PL_sv_yes;
5057 
5058 	TRACEME(("retrieve_sv_yes"));
5059 
5060 	SEEN(sv, cname, 1);
5061 	return sv;
5062 }
5063 
5064 /*
5065  * retrieve_sv_no
5066  *
5067  * Return the immortal no value.
5068  */
retrieve_sv_no(pTHX_ stcxt_t * cxt,char * cname)5069 static SV *retrieve_sv_no(pTHX_ stcxt_t *cxt, char *cname)
5070 {
5071 	SV *sv = &PL_sv_no;
5072 
5073 	TRACEME(("retrieve_sv_no"));
5074 
5075 	SEEN(sv, cname, 1);
5076 	return sv;
5077 }
5078 
5079 /*
5080  * retrieve_array
5081  *
5082  * Retrieve a whole array.
5083  * Layout is SX_ARRAY <size> followed by each item, in increading index order.
5084  * Each item is stored as <object>.
5085  *
5086  * When we come here, SX_ARRAY has been read already.
5087  */
retrieve_array(pTHX_ stcxt_t * cxt,char * cname)5088 static SV *retrieve_array(pTHX_ stcxt_t *cxt, char *cname)
5089 {
5090 	I32 len;
5091 	I32 i;
5092 	AV *av;
5093 	SV *sv;
5094 
5095 	TRACEME(("retrieve_array (#%d)", cxt->tagnum));
5096 
5097 	/*
5098 	 * Read length, and allocate array, then pre-extend it.
5099 	 */
5100 
5101 	RLEN(len);
5102 	TRACEME(("size = %d", len));
5103 	av = newAV();
5104 	SEEN(av, cname, 0);			/* Will return if array not allocated nicely */
5105 	if (len)
5106 		av_extend(av, len);
5107 	else
5108 		return (SV *) av;		/* No data follow if array is empty */
5109 
5110 	/*
5111 	 * Now get each item in turn...
5112 	 */
5113 
5114 	for (i = 0; i < len; i++) {
5115 		TRACEME(("(#%d) item", i));
5116 		sv = retrieve(aTHX_ cxt, 0);			/* Retrieve item */
5117 		if (!sv)
5118 			return (SV *) 0;
5119 		if (av_store(av, i, sv) == 0)
5120 			return (SV *) 0;
5121 	}
5122 
5123 	TRACEME(("ok (retrieve_array at 0x%"UVxf")", PTR2UV(av)));
5124 
5125 	return (SV *) av;
5126 }
5127 
5128 /*
5129  * retrieve_hash
5130  *
5131  * Retrieve a whole hash table.
5132  * Layout is SX_HASH <size> followed by each key/value pair, in random order.
5133  * Keys are stored as <length> <data>, the <data> section being omitted
5134  * if length is 0.
5135  * Values are stored as <object>.
5136  *
5137  * When we come here, SX_HASH has been read already.
5138  */
retrieve_hash(pTHX_ stcxt_t * cxt,char * cname)5139 static SV *retrieve_hash(pTHX_ stcxt_t *cxt, char *cname)
5140 {
5141 	I32 len;
5142 	I32 size;
5143 	I32 i;
5144 	HV *hv;
5145 	SV *sv;
5146 
5147 	TRACEME(("retrieve_hash (#%d)", cxt->tagnum));
5148 
5149 	/*
5150 	 * Read length, allocate table.
5151 	 */
5152 
5153 	RLEN(len);
5154 	TRACEME(("size = %d", len));
5155 	hv = newHV();
5156 	SEEN(hv, cname, 0);		/* Will return if table not allocated properly */
5157 	if (len == 0)
5158 		return (SV *) hv;	/* No data follow if table empty */
5159 	hv_ksplit(hv, len);		/* pre-extend hash to save multiple splits */
5160 
5161 	/*
5162 	 * Now get each key/value pair in turn...
5163 	 */
5164 
5165 	for (i = 0; i < len; i++) {
5166 		/*
5167 		 * Get value first.
5168 		 */
5169 
5170 		TRACEME(("(#%d) value", i));
5171 		sv = retrieve(aTHX_ cxt, 0);
5172 		if (!sv)
5173 			return (SV *) 0;
5174 
5175 		/*
5176 		 * Get key.
5177 		 * Since we're reading into kbuf, we must ensure we're not
5178 		 * recursing between the read and the hv_store() where it's used.
5179 		 * Hence the key comes after the value.
5180 		 */
5181 
5182 		RLEN(size);						/* Get key size */
5183 		KBUFCHK((STRLEN)size);					/* Grow hash key read pool if needed */
5184 		if (size)
5185 			READ(kbuf, size);
5186 		kbuf[size] = '\0';				/* Mark string end, just in case */
5187 		TRACEME(("(#%d) key '%s'", i, kbuf));
5188 
5189 		/*
5190 		 * Enter key/value pair into hash table.
5191 		 */
5192 
5193 		if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
5194 			return (SV *) 0;
5195 	}
5196 
5197 	TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
5198 
5199 	return (SV *) hv;
5200 }
5201 
5202 /*
5203  * retrieve_hash
5204  *
5205  * Retrieve a whole hash table.
5206  * Layout is SX_HASH <size> followed by each key/value pair, in random order.
5207  * Keys are stored as <length> <data>, the <data> section being omitted
5208  * if length is 0.
5209  * Values are stored as <object>.
5210  *
5211  * When we come here, SX_HASH has been read already.
5212  */
retrieve_flag_hash(pTHX_ stcxt_t * cxt,char * cname)5213 static SV *retrieve_flag_hash(pTHX_ stcxt_t *cxt, char *cname)
5214 {
5215     dVAR;
5216     I32 len;
5217     I32 size;
5218     I32 i;
5219     HV *hv;
5220     SV *sv;
5221     int hash_flags;
5222 
5223     GETMARK(hash_flags);
5224     TRACEME(("retrieve_flag_hash (#%d)", cxt->tagnum));
5225     /*
5226      * Read length, allocate table.
5227      */
5228 
5229 #ifndef HAS_RESTRICTED_HASHES
5230     if (hash_flags & SHV_RESTRICTED) {
5231         if (cxt->derestrict < 0)
5232             cxt->derestrict
5233                 = (SvTRUE(perl_get_sv("Storable::downgrade_restricted", TRUE))
5234                    ? 1 : 0);
5235         if (cxt->derestrict == 0)
5236             RESTRICTED_HASH_CROAK();
5237     }
5238 #endif
5239 
5240     RLEN(len);
5241     TRACEME(("size = %d, flags = %d", len, hash_flags));
5242     hv = newHV();
5243     SEEN(hv, cname, 0);		/* Will return if table not allocated properly */
5244     if (len == 0)
5245         return (SV *) hv;	/* No data follow if table empty */
5246     hv_ksplit(hv, len);		/* pre-extend hash to save multiple splits */
5247 
5248     /*
5249      * Now get each key/value pair in turn...
5250      */
5251 
5252     for (i = 0; i < len; i++) {
5253         int flags;
5254         int store_flags = 0;
5255         /*
5256          * Get value first.
5257          */
5258 
5259         TRACEME(("(#%d) value", i));
5260         sv = retrieve(aTHX_ cxt, 0);
5261         if (!sv)
5262             return (SV *) 0;
5263 
5264         GETMARK(flags);
5265 #ifdef HAS_RESTRICTED_HASHES
5266         if ((hash_flags & SHV_RESTRICTED) && (flags & SHV_K_LOCKED))
5267             SvREADONLY_on(sv);
5268 #endif
5269 
5270         if (flags & SHV_K_ISSV) {
5271             /* XXX you can't set a placeholder with an SV key.
5272                Then again, you can't get an SV key.
5273                Without messing around beyond what the API is supposed to do.
5274             */
5275             SV *keysv;
5276             TRACEME(("(#%d) keysv, flags=%d", i, flags));
5277             keysv = retrieve(aTHX_ cxt, 0);
5278             if (!keysv)
5279                 return (SV *) 0;
5280 
5281             if (!hv_store_ent(hv, keysv, sv, 0))
5282                 return (SV *) 0;
5283         } else {
5284             /*
5285              * Get key.
5286              * Since we're reading into kbuf, we must ensure we're not
5287              * recursing between the read and the hv_store() where it's used.
5288              * Hence the key comes after the value.
5289              */
5290 
5291             if (flags & SHV_K_PLACEHOLDER) {
5292                 SvREFCNT_dec (sv);
5293                 sv = &PL_sv_placeholder;
5294 		store_flags |= HVhek_PLACEHOLD;
5295 	    }
5296             if (flags & SHV_K_UTF8) {
5297 #ifdef HAS_UTF8_HASHES
5298                 store_flags |= HVhek_UTF8;
5299 #else
5300                 if (cxt->use_bytes < 0)
5301                     cxt->use_bytes
5302                         = (SvTRUE(perl_get_sv("Storable::drop_utf8", TRUE))
5303                            ? 1 : 0);
5304                 if (cxt->use_bytes == 0)
5305                     UTF8_CROAK();
5306 #endif
5307             }
5308 #ifdef HAS_UTF8_HASHES
5309             if (flags & SHV_K_WASUTF8)
5310 		store_flags |= HVhek_WASUTF8;
5311 #endif
5312 
5313             RLEN(size);						/* Get key size */
5314             KBUFCHK((STRLEN)size);				/* Grow hash key read pool if needed */
5315             if (size)
5316                 READ(kbuf, size);
5317             kbuf[size] = '\0';				/* Mark string end, just in case */
5318             TRACEME(("(#%d) key '%s' flags %X store_flags %X", i, kbuf,
5319 		     flags, store_flags));
5320 
5321             /*
5322              * Enter key/value pair into hash table.
5323              */
5324 
5325 #ifdef HAS_RESTRICTED_HASHES
5326             if (hv_store_flags(hv, kbuf, size, sv, 0, store_flags) == 0)
5327                 return (SV *) 0;
5328 #else
5329             if (!(store_flags & HVhek_PLACEHOLD))
5330                 if (hv_store(hv, kbuf, size, sv, 0) == 0)
5331                     return (SV *) 0;
5332 #endif
5333 	}
5334     }
5335 #ifdef HAS_RESTRICTED_HASHES
5336     if (hash_flags & SHV_RESTRICTED)
5337         SvREADONLY_on(hv);
5338 #endif
5339 
5340     TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
5341 
5342     return (SV *) hv;
5343 }
5344 
5345 /*
5346  * retrieve_code
5347  *
5348  * Return a code reference.
5349  */
retrieve_code(pTHX_ stcxt_t * cxt,char * cname)5350 static SV *retrieve_code(pTHX_ stcxt_t *cxt, char *cname)
5351 {
5352 #if PERL_VERSION < 6
5353     CROAK(("retrieve_code does not work with perl 5.005 or less\n"));
5354 #else
5355 	dSP;
5356 	int type, count, tagnum;
5357 	SV *cv;
5358 	SV *sv, *text, *sub;
5359 
5360 	TRACEME(("retrieve_code (#%d)", cxt->tagnum));
5361 
5362 	/*
5363 	 *  Insert dummy SV in the aseen array so that we don't screw
5364 	 *  up the tag numbers.  We would just make the internal
5365 	 *  scalar an untagged item in the stream, but
5366 	 *  retrieve_scalar() calls SEEN().  So we just increase the
5367 	 *  tag number.
5368 	 */
5369 	tagnum = cxt->tagnum;
5370 	sv = newSViv(0);
5371 	SEEN(sv, cname, 0);
5372 
5373 	/*
5374 	 * Retrieve the source of the code reference
5375 	 * as a small or large scalar
5376 	 */
5377 
5378 	GETMARK(type);
5379 	switch (type) {
5380 	case SX_SCALAR:
5381 		text = retrieve_scalar(aTHX_ cxt, cname);
5382 		break;
5383 	case SX_LSCALAR:
5384 		text = retrieve_lscalar(aTHX_ cxt, cname);
5385 		break;
5386 	default:
5387 		CROAK(("Unexpected type %d in retrieve_code\n", type));
5388 	}
5389 
5390 	/*
5391 	 * prepend "sub " to the source
5392 	 */
5393 
5394 	sub = newSVpvn("sub ", 4);
5395 	sv_catpv(sub, SvPV_nolen(text)); /* XXX no sv_catsv! */
5396 	SvREFCNT_dec(text);
5397 
5398 	/*
5399 	 * evaluate the source to a code reference and use the CV value
5400 	 */
5401 
5402 	if (cxt->eval == NULL) {
5403 		cxt->eval = perl_get_sv("Storable::Eval", TRUE);
5404 		SvREFCNT_inc(cxt->eval);
5405 	}
5406 	if (!SvTRUE(cxt->eval)) {
5407 		if (
5408 			cxt->forgive_me == 0 ||
5409 			(cxt->forgive_me < 0 && !(cxt->forgive_me =
5410 				SvTRUE(perl_get_sv("Storable::forgive_me", TRUE)) ? 1 : 0))
5411 		) {
5412 			CROAK(("Can't eval, please set $Storable::Eval to a true value"));
5413 		} else {
5414 			sv = newSVsv(sub);
5415 			/* fix up the dummy entry... */
5416 			av_store(cxt->aseen, tagnum, SvREFCNT_inc(sv));
5417 			return sv;
5418 		}
5419 	}
5420 
5421 	ENTER;
5422 	SAVETMPS;
5423 
5424 	if (SvROK(cxt->eval) && SvTYPE(SvRV(cxt->eval)) == SVt_PVCV) {
5425 		SV* errsv = get_sv("@", TRUE);
5426 		sv_setpvn(errsv, "", 0);	/* clear $@ */
5427 		PUSHMARK(sp);
5428 		XPUSHs(sv_2mortal(newSVsv(sub)));
5429 		PUTBACK;
5430 		count = call_sv(cxt->eval, G_SCALAR);
5431 		SPAGAIN;
5432 		if (count != 1)
5433 			CROAK(("Unexpected return value from $Storable::Eval callback\n"));
5434 		cv = POPs;
5435 		if (SvTRUE(errsv)) {
5436 			CROAK(("code %s caused an error: %s",
5437 				SvPV_nolen(sub), SvPV_nolen(errsv)));
5438 		}
5439 		PUTBACK;
5440 	} else {
5441 		cv = eval_pv(SvPV_nolen(sub), TRUE);
5442 	}
5443 	if (cv && SvROK(cv) && SvTYPE(SvRV(cv)) == SVt_PVCV) {
5444 	    sv = SvRV(cv);
5445 	} else {
5446 	    CROAK(("code %s did not evaluate to a subroutine reference\n", SvPV_nolen(sub)));
5447 	}
5448 
5449 	SvREFCNT_inc(sv); /* XXX seems to be necessary */
5450 	SvREFCNT_dec(sub);
5451 
5452 	FREETMPS;
5453 	LEAVE;
5454 	/* fix up the dummy entry... */
5455 	av_store(cxt->aseen, tagnum, SvREFCNT_inc(sv));
5456 
5457 	return sv;
5458 #endif
5459 }
5460 
5461 /*
5462  * old_retrieve_array
5463  *
5464  * Retrieve a whole array in pre-0.6 binary format.
5465  *
5466  * Layout is SX_ARRAY <size> followed by each item, in increading index order.
5467  * Each item is stored as SX_ITEM <object> or SX_IT_UNDEF for "holes".
5468  *
5469  * When we come here, SX_ARRAY has been read already.
5470  */
old_retrieve_array(pTHX_ stcxt_t * cxt,char * cname)5471 static SV *old_retrieve_array(pTHX_ stcxt_t *cxt, char *cname)
5472 {
5473 	I32 len;
5474 	I32 i;
5475 	AV *av;
5476 	SV *sv;
5477 	int c;
5478 
5479 	TRACEME(("old_retrieve_array (#%d)", cxt->tagnum));
5480 
5481 	/*
5482 	 * Read length, and allocate array, then pre-extend it.
5483 	 */
5484 
5485 	RLEN(len);
5486 	TRACEME(("size = %d", len));
5487 	av = newAV();
5488 	SEEN(av, 0, 0);				/* Will return if array not allocated nicely */
5489 	if (len)
5490 		av_extend(av, len);
5491 	else
5492 		return (SV *) av;		/* No data follow if array is empty */
5493 
5494 	/*
5495 	 * Now get each item in turn...
5496 	 */
5497 
5498 	for (i = 0; i < len; i++) {
5499 		GETMARK(c);
5500 		if (c == SX_IT_UNDEF) {
5501 			TRACEME(("(#%d) undef item", i));
5502 			continue;			/* av_extend() already filled us with undef */
5503 		}
5504 		if (c != SX_ITEM)
5505 			(void) retrieve_other(aTHX_ (stcxt_t *) 0, 0);	/* Will croak out */
5506 		TRACEME(("(#%d) item", i));
5507 		sv = retrieve(aTHX_ cxt, 0);						/* Retrieve item */
5508 		if (!sv)
5509 			return (SV *) 0;
5510 		if (av_store(av, i, sv) == 0)
5511 			return (SV *) 0;
5512 	}
5513 
5514 	TRACEME(("ok (old_retrieve_array at 0x%"UVxf")", PTR2UV(av)));
5515 
5516 	return (SV *) av;
5517 }
5518 
5519 /*
5520  * old_retrieve_hash
5521  *
5522  * Retrieve a whole hash table in pre-0.6 binary format.
5523  *
5524  * Layout is SX_HASH <size> followed by each key/value pair, in random order.
5525  * Keys are stored as SX_KEY <length> <data>, the <data> section being omitted
5526  * if length is 0.
5527  * Values are stored as SX_VALUE <object> or SX_VL_UNDEF for "holes".
5528  *
5529  * When we come here, SX_HASH has been read already.
5530  */
old_retrieve_hash(pTHX_ stcxt_t * cxt,char * cname)5531 static SV *old_retrieve_hash(pTHX_ stcxt_t *cxt, char *cname)
5532 {
5533 	I32 len;
5534 	I32 size;
5535 	I32 i;
5536 	HV *hv;
5537 	SV *sv = (SV *) 0;
5538 	int c;
5539 	SV *sv_h_undef = (SV *) 0;		/* hv_store() bug */
5540 
5541 	TRACEME(("old_retrieve_hash (#%d)", cxt->tagnum));
5542 
5543 	/*
5544 	 * Read length, allocate table.
5545 	 */
5546 
5547 	RLEN(len);
5548 	TRACEME(("size = %d", len));
5549 	hv = newHV();
5550 	SEEN(hv, 0, 0);			/* Will return if table not allocated properly */
5551 	if (len == 0)
5552 		return (SV *) hv;	/* No data follow if table empty */
5553 	hv_ksplit(hv, len);		/* pre-extend hash to save multiple splits */
5554 
5555 	/*
5556 	 * Now get each key/value pair in turn...
5557 	 */
5558 
5559 	for (i = 0; i < len; i++) {
5560 		/*
5561 		 * Get value first.
5562 		 */
5563 
5564 		GETMARK(c);
5565 		if (c == SX_VL_UNDEF) {
5566 			TRACEME(("(#%d) undef value", i));
5567 			/*
5568 			 * Due to a bug in hv_store(), it's not possible to pass
5569 			 * &PL_sv_undef to hv_store() as a value, otherwise the
5570 			 * associated key will not be creatable any more. -- RAM, 14/01/97
5571 			 */
5572 			if (!sv_h_undef)
5573 				sv_h_undef = newSVsv(&PL_sv_undef);
5574 			sv = SvREFCNT_inc(sv_h_undef);
5575 		} else if (c == SX_VALUE) {
5576 			TRACEME(("(#%d) value", i));
5577 			sv = retrieve(aTHX_ cxt, 0);
5578 			if (!sv)
5579 				return (SV *) 0;
5580 		} else
5581 			(void) retrieve_other(aTHX_ (stcxt_t *) 0, 0);	/* Will croak out */
5582 
5583 		/*
5584 		 * Get key.
5585 		 * Since we're reading into kbuf, we must ensure we're not
5586 		 * recursing between the read and the hv_store() where it's used.
5587 		 * Hence the key comes after the value.
5588 		 */
5589 
5590 		GETMARK(c);
5591 		if (c != SX_KEY)
5592 			(void) retrieve_other(aTHX_ (stcxt_t *) 0, 0);	/* Will croak out */
5593 		RLEN(size);						/* Get key size */
5594 		KBUFCHK((STRLEN)size);					/* Grow hash key read pool if needed */
5595 		if (size)
5596 			READ(kbuf, size);
5597 		kbuf[size] = '\0';				/* Mark string end, just in case */
5598 		TRACEME(("(#%d) key '%s'", i, kbuf));
5599 
5600 		/*
5601 		 * Enter key/value pair into hash table.
5602 		 */
5603 
5604 		if (hv_store(hv, kbuf, (U32) size, sv, 0) == 0)
5605 			return (SV *) 0;
5606 	}
5607 
5608 	TRACEME(("ok (retrieve_hash at 0x%"UVxf")", PTR2UV(hv)));
5609 
5610 	return (SV *) hv;
5611 }
5612 
5613 /***
5614  *** Retrieval engine.
5615  ***/
5616 
5617 /*
5618  * magic_check
5619  *
5620  * Make sure the stored data we're trying to retrieve has been produced
5621  * on an ILP compatible system with the same byteorder. It croaks out in
5622  * case an error is detected. [ILP = integer-long-pointer sizes]
5623  * Returns null if error is detected, &PL_sv_undef otherwise.
5624  *
5625  * Note that there's no byte ordering info emitted when network order was
5626  * used at store time.
5627  */
magic_check(pTHX_ stcxt_t * cxt)5628 static SV *magic_check(pTHX_ stcxt_t *cxt)
5629 {
5630     /* The worst case for a malicious header would be old magic (which is
5631        longer), major, minor, byteorder length byte of 255, 255 bytes of
5632        garbage, sizeof int, long, pointer, NV.
5633        So the worse of that we can read is 255 bytes of garbage plus 4.
5634        Err, I am assuming 8 bit bytes here. Please file a bug report if you're
5635        compiling perl on a system with chars that are larger than 8 bits.
5636        (Even Crays aren't *that* perverse).
5637     */
5638     unsigned char buf[4 + 255];
5639     unsigned char *current;
5640     int c;
5641     int length;
5642     int use_network_order;
5643     int use_NV_size;
5644     int version_major;
5645     int version_minor = 0;
5646 
5647     TRACEME(("magic_check"));
5648 
5649     /*
5650      * The "magic number" is only for files, not when freezing in memory.
5651      */
5652 
5653     if (cxt->fio) {
5654         /* This includes the '\0' at the end.  I want to read the extra byte,
5655            which is usually going to be the major version number.  */
5656         STRLEN len = sizeof(magicstr);
5657         STRLEN old_len;
5658 
5659         READ(buf, (SSize_t)(len));	/* Not null-terminated */
5660 
5661         /* Point at the byte after the byte we read.  */
5662         current = buf + --len;	/* Do the -- outside of macros.  */
5663 
5664         if (memNE(buf, magicstr, len)) {
5665             /*
5666              * Try to read more bytes to check for the old magic number, which
5667              * was longer.
5668              */
5669 
5670             TRACEME(("trying for old magic number"));
5671 
5672             old_len = sizeof(old_magicstr) - 1;
5673             READ(current + 1, (SSize_t)(old_len - len));
5674 
5675             if (memNE(buf, old_magicstr, old_len))
5676                 CROAK(("File is not a perl storable"));
5677             current = buf + old_len;
5678         }
5679         use_network_order = *current;
5680     } else
5681 	GETMARK(use_network_order);
5682 
5683     /*
5684      * Starting with 0.6, the "use_network_order" byte flag is also used to
5685      * indicate the version number of the binary, and therefore governs the
5686      * setting of sv_retrieve_vtbl. See magic_write().
5687      */
5688 
5689     version_major = use_network_order >> 1;
5690     cxt->retrieve_vtbl = (SV*(**)(pTHX_ stcxt_t *cxt, char *cname)) (version_major ? sv_retrieve : sv_old_retrieve);
5691 
5692     TRACEME(("magic_check: netorder = 0x%x", use_network_order));
5693 
5694 
5695     /*
5696      * Starting with 0.7 (binary major 2), a full byte is dedicated to the
5697      * minor version of the protocol.  See magic_write().
5698      */
5699 
5700     if (version_major > 1)
5701         GETMARK(version_minor);
5702 
5703     cxt->ver_major = version_major;
5704     cxt->ver_minor = version_minor;
5705 
5706     TRACEME(("binary image version is %d.%d", version_major, version_minor));
5707 
5708     /*
5709      * Inter-operability sanity check: we can't retrieve something stored
5710      * using a format more recent than ours, because we have no way to
5711      * know what has changed, and letting retrieval go would mean a probable
5712      * failure reporting a "corrupted" storable file.
5713      */
5714 
5715     if (
5716         version_major > STORABLE_BIN_MAJOR ||
5717         (version_major == STORABLE_BIN_MAJOR &&
5718          version_minor > STORABLE_BIN_MINOR)
5719         ) {
5720         int croak_now = 1;
5721         TRACEME(("but I am version is %d.%d", STORABLE_BIN_MAJOR,
5722                  STORABLE_BIN_MINOR));
5723 
5724         if (version_major == STORABLE_BIN_MAJOR) {
5725             TRACEME(("cxt->accept_future_minor is %d",
5726                      cxt->accept_future_minor));
5727             if (cxt->accept_future_minor < 0)
5728                 cxt->accept_future_minor
5729                     = (SvTRUE(perl_get_sv("Storable::accept_future_minor",
5730                                           TRUE))
5731                        ? 1 : 0);
5732             if (cxt->accept_future_minor == 1)
5733                 croak_now = 0;  /* Don't croak yet.  */
5734         }
5735         if (croak_now) {
5736             CROAK(("Storable binary image v%d.%d more recent than I am (v%d.%d)",
5737                    version_major, version_minor,
5738                    STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR));
5739         }
5740     }
5741 
5742     /*
5743      * If they stored using network order, there's no byte ordering
5744      * information to check.
5745      */
5746 
5747     if ((cxt->netorder = (use_network_order & 0x1)))	/* Extra () for -Wall */
5748         return &PL_sv_undef;			/* No byte ordering info */
5749 
5750     /* In C truth is 1, falsehood is 0. Very convienient.  */
5751     use_NV_size = version_major >= 2 && version_minor >= 2;
5752 
5753     GETMARK(c);
5754     length = c + 3 + use_NV_size;
5755     READ(buf, length);	/* Not null-terminated */
5756 
5757     TRACEME(("byte order '%.*s' %d", c, buf, c));
5758 
5759 #ifdef USE_56_INTERWORK_KLUDGE
5760     /* No point in caching this in the context as we only need it once per
5761        retrieve, and we need to recheck it each read.  */
5762     if (SvTRUE(perl_get_sv("Storable::interwork_56_64bit", TRUE))) {
5763         if ((c != (sizeof (byteorderstr_56) - 1))
5764             || memNE(buf, byteorderstr_56, c))
5765             CROAK(("Byte order is not compatible"));
5766     } else
5767 #endif
5768     {
5769         if ((c != (sizeof (byteorderstr) - 1)) || memNE(buf, byteorderstr, c))
5770             CROAK(("Byte order is not compatible"));
5771     }
5772 
5773     current = buf + c;
5774 
5775     /* sizeof(int) */
5776     if ((int) *current++ != sizeof(int))
5777         CROAK(("Integer size is not compatible"));
5778 
5779     /* sizeof(long) */
5780     if ((int) *current++ != sizeof(long))
5781         CROAK(("Long integer size is not compatible"));
5782 
5783     /* sizeof(char *) */
5784     if ((int) *current != sizeof(char *))
5785         CROAK(("Pointer size is not compatible"));
5786 
5787     if (use_NV_size) {
5788         /* sizeof(NV) */
5789         if ((int) *++current != sizeof(NV))
5790             CROAK(("Double size is not compatible"));
5791     }
5792 
5793     return &PL_sv_undef;	/* OK */
5794 }
5795 
5796 /*
5797  * retrieve
5798  *
5799  * Recursively retrieve objects from the specified file and return their
5800  * root SV (which may be an AV or an HV for what we care).
5801  * Returns null if there is a problem.
5802  */
retrieve(pTHX_ stcxt_t * cxt,char * cname)5803 static SV *retrieve(pTHX_ stcxt_t *cxt, char *cname)
5804 {
5805 	int type;
5806 	SV **svh;
5807 	SV *sv;
5808 
5809 	TRACEME(("retrieve"));
5810 
5811 	/*
5812 	 * Grab address tag which identifies the object if we are retrieving
5813 	 * an older format. Since the new binary format counts objects and no
5814 	 * longer explicitely tags them, we must keep track of the correspondance
5815 	 * ourselves.
5816 	 *
5817 	 * The following section will disappear one day when the old format is
5818 	 * no longer supported, hence the final "goto" in the "if" block.
5819 	 */
5820 
5821 	if (cxt->hseen) {						/* Retrieving old binary */
5822 		stag_t tag;
5823 		if (cxt->netorder) {
5824 			I32 nettag;
5825 			READ(&nettag, sizeof(I32));		/* Ordered sequence of I32 */
5826 			tag = (stag_t) nettag;
5827 		} else
5828 			READ(&tag, sizeof(stag_t));		/* Original address of the SV */
5829 
5830 		GETMARK(type);
5831 		if (type == SX_OBJECT) {
5832 			I32 tagn;
5833 			svh = hv_fetch(cxt->hseen, (char *) &tag, sizeof(tag), FALSE);
5834 			if (!svh)
5835 				CROAK(("Old tag 0x%"UVxf" should have been mapped already",
5836 					(UV) tag));
5837 			tagn = SvIV(*svh);	/* Mapped tag number computed earlier below */
5838 
5839 			/*
5840 			 * The following code is common with the SX_OBJECT case below.
5841 			 */
5842 
5843 			svh = av_fetch(cxt->aseen, tagn, FALSE);
5844 			if (!svh)
5845 				CROAK(("Object #%"IVdf" should have been retrieved already",
5846 					(IV) tagn));
5847 			sv = *svh;
5848 			TRACEME(("has retrieved #%d at 0x%"UVxf, tagn, PTR2UV(sv)));
5849 			SvREFCNT_inc(sv);	/* One more reference to this same sv */
5850 			return sv;			/* The SV pointer where object was retrieved */
5851 		}
5852 
5853 		/*
5854 		 * Map new object, but don't increase tagnum. This will be done
5855 		 * by each of the retrieve_* functions when they call SEEN().
5856 		 *
5857 		 * The mapping associates the "tag" initially present with a unique
5858 		 * tag number. See test for SX_OBJECT above to see how this is perused.
5859 		 */
5860 
5861 		if (!hv_store(cxt->hseen, (char *) &tag, sizeof(tag),
5862 				newSViv(cxt->tagnum), 0))
5863 			return (SV *) 0;
5864 
5865 		goto first_time;
5866 	}
5867 
5868 	/*
5869 	 * Regular post-0.6 binary format.
5870 	 */
5871 
5872 	GETMARK(type);
5873 
5874 	TRACEME(("retrieve type = %d", type));
5875 
5876 	/*
5877 	 * Are we dealing with an object we should have already retrieved?
5878 	 */
5879 
5880 	if (type == SX_OBJECT) {
5881 		I32 tag;
5882 		READ_I32(tag);
5883 		tag = ntohl(tag);
5884 		svh = av_fetch(cxt->aseen, tag, FALSE);
5885 		if (!svh)
5886 			CROAK(("Object #%"IVdf" should have been retrieved already",
5887 				(IV) tag));
5888 		sv = *svh;
5889 		TRACEME(("had retrieved #%d at 0x%"UVxf, tag, PTR2UV(sv)));
5890 		SvREFCNT_inc(sv);	/* One more reference to this same sv */
5891 		return sv;			/* The SV pointer where object was retrieved */
5892 	} else if (type >= SX_ERROR && cxt->ver_minor > STORABLE_BIN_MINOR) {
5893             if (cxt->accept_future_minor < 0)
5894                 cxt->accept_future_minor
5895                     = (SvTRUE(perl_get_sv("Storable::accept_future_minor",
5896                                           TRUE))
5897                        ? 1 : 0);
5898             if (cxt->accept_future_minor == 1) {
5899                 CROAK(("Storable binary image v%d.%d contains data of type %d. "
5900                        "This Storable is v%d.%d and can only handle data types up to %d",
5901                        cxt->ver_major, cxt->ver_minor, type,
5902                        STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR, SX_ERROR - 1));
5903             }
5904         }
5905 
5906 first_time:		/* Will disappear when support for old format is dropped */
5907 
5908 	/*
5909 	 * Okay, first time through for this one.
5910 	 */
5911 
5912 	sv = RETRIEVE(cxt, type)(aTHX_ cxt, cname);
5913 	if (!sv)
5914 		return (SV *) 0;			/* Failed */
5915 
5916 	/*
5917 	 * Old binary formats (pre-0.7).
5918 	 *
5919 	 * Final notifications, ended by SX_STORED may now follow.
5920 	 * Currently, the only pertinent notification to apply on the
5921 	 * freshly retrieved object is either:
5922 	 *    SX_CLASS <char-len> <classname> for short classnames.
5923 	 *    SX_LG_CLASS <int-len> <classname> for larger one (rare!).
5924 	 * Class name is then read into the key buffer pool used by
5925 	 * hash table key retrieval.
5926 	 */
5927 
5928 	if (cxt->ver_major < 2) {
5929 		while ((type = GETCHAR()) != SX_STORED) {
5930 			I32 len;
5931 			switch (type) {
5932 			case SX_CLASS:
5933 				GETMARK(len);			/* Length coded on a single char */
5934 				break;
5935 			case SX_LG_CLASS:			/* Length coded on a regular integer */
5936 				RLEN(len);
5937 				break;
5938 			case EOF:
5939 			default:
5940 				return (SV *) 0;		/* Failed */
5941 			}
5942 			KBUFCHK((STRLEN)len);			/* Grow buffer as necessary */
5943 			if (len)
5944 				READ(kbuf, len);
5945 			kbuf[len] = '\0';			/* Mark string end */
5946 			BLESS(sv, kbuf);
5947 		}
5948 	}
5949 
5950 	TRACEME(("ok (retrieved 0x%"UVxf", refcnt=%d, %s)", PTR2UV(sv),
5951 		SvREFCNT(sv) - 1, sv_reftype(sv, FALSE)));
5952 
5953 	return sv;	/* Ok */
5954 }
5955 
5956 /*
5957  * do_retrieve
5958  *
5959  * Retrieve data held in file and return the root object.
5960  * Common routine for pretrieve and mretrieve.
5961  */
do_retrieve(pTHX_ PerlIO * f,SV * in,int optype)5962 static SV *do_retrieve(
5963         pTHX_
5964 	PerlIO *f,
5965 	SV *in,
5966 	int optype)
5967 {
5968 	dSTCXT;
5969 	SV *sv;
5970 	int is_tainted;				/* Is input source tainted? */
5971 	int pre_06_fmt = 0;			/* True with pre Storable 0.6 formats */
5972 
5973 	TRACEME(("do_retrieve (optype = 0x%x)", optype));
5974 
5975 	optype |= ST_RETRIEVE;
5976 
5977 	/*
5978 	 * Sanity assertions for retrieve dispatch tables.
5979 	 */
5980 
5981 	ASSERT(sizeof(sv_old_retrieve) == sizeof(sv_retrieve),
5982 		("old and new retrieve dispatch table have same size"));
5983 	ASSERT(sv_old_retrieve[SX_ERROR] == retrieve_other,
5984 		("SX_ERROR entry correctly initialized in old dispatch table"));
5985 	ASSERT(sv_retrieve[SX_ERROR] == retrieve_other,
5986 		("SX_ERROR entry correctly initialized in new dispatch table"));
5987 
5988 	/*
5989 	 * Workaround for CROAK leak: if they enter with a "dirty" context,
5990 	 * free up memory for them now.
5991 	 */
5992 
5993 	if (cxt->s_dirty)
5994 		clean_context(aTHX_ cxt);
5995 
5996 	/*
5997 	 * Now that STORABLE_xxx hooks exist, it is possible that they try to
5998 	 * re-enter retrieve() via the hooks.
5999 	 */
6000 
6001 	if (cxt->entry)
6002 		cxt = allocate_context(aTHX_ cxt);
6003 
6004 	cxt->entry++;
6005 
6006 	ASSERT(cxt->entry == 1, ("starting new recursion"));
6007 	ASSERT(!cxt->s_dirty, ("clean context"));
6008 
6009 	/*
6010 	 * Prepare context.
6011 	 *
6012 	 * Data is loaded into the memory buffer when f is NULL, unless `in' is
6013 	 * also NULL, in which case we're expecting the data to already lie
6014 	 * in the buffer (dclone case).
6015 	 */
6016 
6017 	KBUFINIT();			 		/* Allocate hash key reading pool once */
6018 
6019 	if (!f && in) {
6020 #ifdef SvUTF8_on
6021 		if (SvUTF8(in)) {
6022 			STRLEN length;
6023 			const char *orig = SvPV(in, length);
6024 			char *asbytes;
6025 			/* This is quite deliberate. I want the UTF8 routines
6026 			   to encounter the '\0' which perl adds at the end
6027 			   of all scalars, so that any new string also has
6028 			   this.
6029 			*/
6030 			STRLEN klen_tmp = length + 1;
6031 			bool is_utf8 = TRUE;
6032 
6033 			/* Just casting the &klen to (STRLEN) won't work
6034 			   well if STRLEN and I32 are of different widths.
6035 			   --jhi */
6036 			asbytes = (char*)bytes_from_utf8((U8*)orig,
6037 							 &klen_tmp,
6038 							 &is_utf8);
6039 			if (is_utf8) {
6040 				CROAK(("Frozen string corrupt - contains characters outside 0-255"));
6041 			}
6042 			if (asbytes != orig) {
6043 				/* String has been converted.
6044 				   There is no need to keep any reference to
6045 				   the old string.  */
6046 				in = sv_newmortal();
6047 				/* We donate the SV the malloc()ed string
6048 				   bytes_from_utf8 returned us.  */
6049 				SvUPGRADE(in, SVt_PV);
6050 				SvPOK_on(in);
6051 				SvPV_set(in, asbytes);
6052 				SvLEN_set(in, klen_tmp);
6053 				SvCUR_set(in, klen_tmp - 1);
6054 			}
6055 		}
6056 #endif
6057 		MBUF_SAVE_AND_LOAD(in);
6058 	}
6059 
6060 	/*
6061 	 * Magic number verifications.
6062 	 *
6063 	 * This needs to be done before calling init_retrieve_context()
6064 	 * since the format indication in the file are necessary to conduct
6065 	 * some of the initializations.
6066 	 */
6067 
6068 	cxt->fio = f;				/* Where I/O are performed */
6069 
6070 	if (!magic_check(aTHX_ cxt))
6071 		CROAK(("Magic number checking on storable %s failed",
6072 			cxt->fio ? "file" : "string"));
6073 
6074 	TRACEME(("data stored in %s format",
6075 		cxt->netorder ? "net order" : "native"));
6076 
6077 	/*
6078 	 * Check whether input source is tainted, so that we don't wrongly
6079 	 * taint perfectly good values...
6080 	 *
6081 	 * We assume file input is always tainted.  If both `f' and `in' are
6082 	 * NULL, then we come from dclone, and tainted is already filled in
6083 	 * the context.  That's a kludge, but the whole dclone() thing is
6084 	 * already quite a kludge anyway! -- RAM, 15/09/2000.
6085 	 */
6086 
6087 	is_tainted = f ? 1 : (in ? SvTAINTED(in) : cxt->s_tainted);
6088 	TRACEME(("input source is %s", is_tainted ? "tainted" : "trusted"));
6089 	init_retrieve_context(aTHX_ cxt, optype, is_tainted);
6090 
6091 	ASSERT(is_retrieving(aTHX), ("within retrieve operation"));
6092 
6093 	sv = retrieve(aTHX_ cxt, 0);		/* Recursively retrieve object, get root SV */
6094 
6095 	/*
6096 	 * Final cleanup.
6097 	 */
6098 
6099 	if (!f && in)
6100 		MBUF_RESTORE();
6101 
6102 	pre_06_fmt = cxt->hseen != NULL;	/* Before we clean context */
6103 
6104 	/*
6105 	 * The "root" context is never freed.
6106 	 */
6107 
6108 	clean_retrieve_context(aTHX_ cxt);
6109 	if (cxt->prev)				/* This context was stacked */
6110 		free_context(aTHX_ cxt);		/* It was not the "root" context */
6111 
6112 	/*
6113 	 * Prepare returned value.
6114 	 */
6115 
6116 	if (!sv) {
6117 		TRACEME(("retrieve ERROR"));
6118 #if (PATCHLEVEL <= 4)
6119 		/* perl 5.00405 seems to screw up at this point with an
6120 		   'attempt to modify a read only value' error reported in the
6121 		   eval { $self = pretrieve(*FILE) } in _retrieve.
6122 		   I can't see what the cause of this error is, but I suspect a
6123 		   bug in 5.004, as it seems to be capable of issuing spurious
6124 		   errors or core dumping with matches on $@. I'm not going to
6125 		   spend time on what could be a fruitless search for the cause,
6126 		   so here's a bodge. If you're running 5.004 and don't like
6127 		   this inefficiency, either upgrade to a newer perl, or you are
6128 		   welcome to find the problem and send in a patch.
6129 		 */
6130 		return newSV(0);
6131 #else
6132 		return &PL_sv_undef;		/* Something went wrong, return undef */
6133 #endif
6134 	}
6135 
6136 	TRACEME(("retrieve got %s(0x%"UVxf")",
6137 		sv_reftype(sv, FALSE), PTR2UV(sv)));
6138 
6139 	/*
6140 	 * Backward compatibility with Storable-0.5@9 (which we know we
6141 	 * are retrieving if hseen is non-null): don't create an extra RV
6142 	 * for objects since we special-cased it at store time.
6143 	 *
6144 	 * Build a reference to the SV returned by pretrieve even if it is
6145 	 * already one and not a scalar, for consistency reasons.
6146 	 */
6147 
6148 	if (pre_06_fmt) {			/* Was not handling overloading by then */
6149 		SV *rv;
6150 		TRACEME(("fixing for old formats -- pre 0.6"));
6151 		if (sv_type(aTHX_ sv) == svis_REF && (rv = SvRV(sv)) && SvOBJECT(rv)) {
6152 			TRACEME(("ended do_retrieve() with an object -- pre 0.6"));
6153 			return sv;
6154 		}
6155 	}
6156 
6157 	/*
6158 	 * If reference is overloaded, restore behaviour.
6159 	 *
6160 	 * NB: minor glitch here: normally, overloaded refs are stored specially
6161 	 * so that we can croak when behaviour cannot be re-installed, and also
6162 	 * avoid testing for overloading magic at each reference retrieval.
6163 	 *
6164 	 * Unfortunately, the root reference is implicitely stored, so we must
6165 	 * check for possible overloading now.  Furthermore, if we don't restore
6166 	 * overloading, we cannot croak as if the original ref was, because we
6167 	 * have no way to determine whether it was an overloaded ref or not in
6168 	 * the first place.
6169 	 *
6170 	 * It's a pity that overloading magic is attached to the rv, and not to
6171 	 * the underlying sv as blessing is.
6172 	 */
6173 
6174 	if (SvOBJECT(sv)) {
6175 		HV *stash = (HV *) SvSTASH(sv);
6176 		SV *rv = newRV_noinc(sv);
6177 		if (stash && Gv_AMG(stash)) {
6178 			SvAMAGIC_on(rv);
6179 			TRACEME(("restored overloading on root reference"));
6180 		}
6181 		TRACEME(("ended do_retrieve() with an object"));
6182 		return rv;
6183 	}
6184 
6185 	TRACEME(("regular do_retrieve() end"));
6186 
6187 	return newRV_noinc(sv);
6188 }
6189 
6190 /*
6191  * pretrieve
6192  *
6193  * Retrieve data held in file and return the root object, undef on error.
6194  */
pretrieve(pTHX_ PerlIO * f)6195 SV *pretrieve(pTHX_ PerlIO *f)
6196 {
6197 	TRACEME(("pretrieve"));
6198 	return do_retrieve(aTHX_ f, Nullsv, 0);
6199 }
6200 
6201 /*
6202  * mretrieve
6203  *
6204  * Retrieve data held in scalar and return the root object, undef on error.
6205  */
mretrieve(pTHX_ SV * sv)6206 SV *mretrieve(pTHX_ SV *sv)
6207 {
6208 	TRACEME(("mretrieve"));
6209 	return do_retrieve(aTHX_ (PerlIO*) 0, sv, 0);
6210 }
6211 
6212 /***
6213  *** Deep cloning
6214  ***/
6215 
6216 /*
6217  * dclone
6218  *
6219  * Deep clone: returns a fresh copy of the original referenced SV tree.
6220  *
6221  * This is achieved by storing the object in memory and restoring from
6222  * there. Not that efficient, but it should be faster than doing it from
6223  * pure perl anyway.
6224  */
dclone(pTHX_ SV * sv)6225 SV *dclone(pTHX_ SV *sv)
6226 {
6227 	dSTCXT;
6228 	int size;
6229 	stcxt_t *real_context;
6230 	SV *out;
6231 
6232 	TRACEME(("dclone"));
6233 
6234 	/*
6235 	 * Workaround for CROAK leak: if they enter with a "dirty" context,
6236 	 * free up memory for them now.
6237 	 */
6238 
6239 	if (cxt->s_dirty)
6240 		clean_context(aTHX_ cxt);
6241 
6242 	/*
6243 	 * do_store() optimizes for dclone by not freeing its context, should
6244 	 * we need to allocate one because we're deep cloning from a hook.
6245 	 */
6246 
6247 	if (!do_store(aTHX_ (PerlIO*) 0, sv, ST_CLONE, FALSE, (SV**) 0))
6248 		return &PL_sv_undef;				/* Error during store */
6249 
6250 	/*
6251 	 * Because of the above optimization, we have to refresh the context,
6252 	 * since a new one could have been allocated and stacked by do_store().
6253 	 */
6254 
6255 	{ dSTCXT; real_context = cxt; }		/* Sub-block needed for macro */
6256 	cxt = real_context;					/* And we need this temporary... */
6257 
6258 	/*
6259 	 * Now, `cxt' may refer to a new context.
6260 	 */
6261 
6262 	ASSERT(!cxt->s_dirty, ("clean context"));
6263 	ASSERT(!cxt->entry, ("entry will not cause new context allocation"));
6264 
6265 	size = MBUF_SIZE();
6266 	TRACEME(("dclone stored %d bytes", size));
6267 	MBUF_INIT(size);
6268 
6269 	/*
6270 	 * Since we're passing do_retrieve() both a NULL file and sv, we need
6271 	 * to pre-compute the taintedness of the input by setting cxt->tainted
6272 	 * to whatever state our own input string was.	-- RAM, 15/09/2000
6273 	 *
6274 	 * do_retrieve() will free non-root context.
6275 	 */
6276 
6277 	cxt->s_tainted = SvTAINTED(sv);
6278 	out = do_retrieve(aTHX_ (PerlIO*) 0, Nullsv, ST_CLONE);
6279 
6280 	TRACEME(("dclone returns 0x%"UVxf, PTR2UV(out)));
6281 
6282 	return out;
6283 }
6284 
6285 /***
6286  *** Glue with perl.
6287  ***/
6288 
6289 /*
6290  * The Perl IO GV object distinguishes between input and output for sockets
6291  * but not for plain files. To allow Storable to transparently work on
6292  * plain files and sockets transparently, we have to ask xsubpp to fetch the
6293  * right object for us. Hence the OutputStream and InputStream declarations.
6294  *
6295  * Before perl 5.004_05, those entries in the standard typemap are not
6296  * defined in perl include files, so we do that here.
6297  */
6298 
6299 #ifndef OutputStream
6300 #define OutputStream	PerlIO *
6301 #define InputStream		PerlIO *
6302 #endif	/* !OutputStream */
6303 
6304 MODULE = Storable	PACKAGE = Storable::Cxt
6305 
6306 void
6307 DESTROY(self)
6308     SV *self
6309 PREINIT:
6310 	stcxt_t *cxt = (stcxt_t *)SvPVX(SvRV(self));
6311 PPCODE:
6312 	if (kbuf)
6313 		Safefree(kbuf);
6314 	if (!cxt->membuf_ro && mbase)
6315 		Safefree(mbase);
6316 	if (cxt->membuf_ro && (cxt->msaved).arena)
6317 		Safefree((cxt->msaved).arena);
6318 
6319 
6320 MODULE = Storable	PACKAGE = Storable
6321 
6322 PROTOTYPES: ENABLE
6323 
6324 BOOT:
6325     init_perinterp(aTHX);
6326     gv_fetchpv("Storable::drop_utf8",   GV_ADDMULTI, SVt_PV);
6327 #ifdef DEBUGME
6328     /* Only disable the used only once warning if we are in debugging mode.  */
6329     gv_fetchpv("Storable::DEBUGME",   GV_ADDMULTI, SVt_PV);
6330 #endif
6331 #ifdef USE_56_INTERWORK_KLUDGE
6332     gv_fetchpv("Storable::interwork_56_64bit",   GV_ADDMULTI, SVt_PV);
6333 #endif
6334 
6335 void
6336 init_perinterp()
6337  CODE:
6338   init_perinterp(aTHX);
6339 
6340 int
6341 pstore(f,obj)
6342 OutputStream	f
6343 SV *	obj
6344  CODE:
6345   RETVAL = pstore(aTHX_ f, obj);
6346  OUTPUT:
6347   RETVAL
6348 
6349 int
6350 net_pstore(f,obj)
6351 OutputStream	f
6352 SV *	obj
6353  CODE:
6354   RETVAL = net_pstore(aTHX_ f, obj);
6355  OUTPUT:
6356   RETVAL
6357 
6358 SV *
6359 mstore(obj)
6360 SV *	obj
6361  CODE:
6362   RETVAL = mstore(aTHX_ obj);
6363  OUTPUT:
6364   RETVAL
6365 
6366 SV *
6367 net_mstore(obj)
6368 SV *	obj
6369  CODE:
6370   RETVAL = net_mstore(aTHX_ obj);
6371  OUTPUT:
6372   RETVAL
6373 
6374 SV *
6375 pretrieve(f)
6376 InputStream	f
6377  CODE:
6378   RETVAL = pretrieve(aTHX_ f);
6379  OUTPUT:
6380   RETVAL
6381 
6382 SV *
6383 mretrieve(sv)
6384 SV *	sv
6385  CODE:
6386   RETVAL = mretrieve(aTHX_ sv);
6387  OUTPUT:
6388   RETVAL
6389 
6390 SV *
6391 dclone(sv)
6392 SV *	sv
6393  CODE:
6394   RETVAL = dclone(aTHX_ sv);
6395  OUTPUT:
6396   RETVAL
6397 
6398 int
6399 last_op_in_netorder()
6400  CODE:
6401   RETVAL = last_op_in_netorder(aTHX);
6402  OUTPUT:
6403   RETVAL
6404 
6405 int
6406 is_storing()
6407  CODE:
6408   RETVAL = is_storing(aTHX);
6409  OUTPUT:
6410   RETVAL
6411 
6412 int
6413 is_retrieving()
6414  CODE:
6415   RETVAL = is_retrieving(aTHX);
6416  OUTPUT:
6417   RETVAL
6418