This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
use lengths in sv_setpv() calls
[perl5.git] / ext / Storable / Storable.xs
CommitLineData
7a6a85bf 1/*
6ad89a2f 2 * Store and retrieve mechanism.
7a6a85bf
RG
3 *
4 * Copyright (c) 1995-2000, Raphael Manfredi
5 *
9e21b3d0
JH
6 * You may redistribute only under the same terms as Perl 5, as specified
7 * in the README file that comes with the distribution.
7a6a85bf 8 *
7a6a85bf
RG
9 */
10
138ec36d 11#define PERL_NO_GET_CONTEXT /* we want efficiency */
7a6a85bf
RG
12#include <EXTERN.h>
13#include <perl.h>
a3bf621f
JH
14#include <XSUB.h>
15
eadddfac 16#ifndef PATCHLEVEL
fac63a07 17#include <patchlevel.h> /* Perl's one, needed since 5.6 */
069d7f71 18#endif
7a6a85bf 19
fac63a07 20#if !defined(PERL_VERSION) || PERL_VERSION < 8
3f575d8d
NC
21#include "ppport.h" /* handle old perls */
22#endif
23
e8189732 24#if 0
9e21b3d0
JH
25#define DEBUGME /* Debug mode, turns assertions on as well */
26#define DASSERT /* Assertion mode */
27#endif
7a6a85bf
RG
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 */
f0ffaed8
JH
47
48#ifndef PERL_VERSION /* For perls < 5.6 */
e993d95c 49#define PERL_VERSION PATCHLEVEL
7a6a85bf
RG
50#ifndef newRV_noinc
51#define newRV_noinc(sv) ((Sv = newRV(sv)), --SvREFCNT(SvRV(Sv)), Sv)
52#endif
e993d95c 53#if (PATCHLEVEL <= 4) /* Older perls (<= 5.004) lack PL_ namespace */
7a6a85bf
RG
54#define PL_sv_yes sv_yes
55#define PL_sv_no sv_no
56#define PL_sv_undef sv_undef
e993d95c 57#if (SUBVERSION <= 4) /* 5.004_04 has been reported to lack newSVpvn */
dd19458b 58#define newSVpvn newSVpv
7a6a85bf 59#endif
e993d95c 60#endif /* PATCHLEVEL <= 4 */
7a6a85bf
RG
61#ifndef HvSHAREKEYS_off
62#define HvSHAREKEYS_off(hv) /* Ignore */
63#endif
f0ffaed8
JH
64#ifndef AvFILLp /* Older perls (<=5.003) lack AvFILLp */
65#define AvFILLp AvFILL
66#endif
67typedef double NV; /* Older perls lack the NV type */
cc964657
JH
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)
f0ffaed8 74#endif /* PERL_VERSION -- perls < 5.6 */
7a6a85bf 75
cc964657 76#ifndef NVef /* The following were not part of perl 5.6 */
9e21b3d0
JH
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
cc964657
JH
83#define NVef "e"
84#define NVff "f"
85#define NVgf "g"
86#endif
87#endif
88
0bb78401
AMS
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
aec614a5
NC
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
27da23d5 104# else
aec614a5 105# define PERL_UNUSED_DECL
27da23d5 106# endif
27da23d5
JH
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
bfcb3514 117#ifndef HvRITER_set
0bb78401 118# define HvRITER_set(hv,r) (HvRITER(hv) = r)
bfcb3514
NC
119#endif
120#ifndef HvEITER_set
0bb78401 121# define HvEITER_set(hv,r) (HvEITER(hv) = r)
bfcb3514
NC
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
ca732855
NC
135#ifndef HvPLACEHOLDERS_get
136# define HvPLACEHOLDERS_get HvPLACEHOLDERS
137#endif
138
7a6a85bf 139#ifdef DEBUGME
8be2b38b
JH
140
141#ifndef DASSERT
142#define DASSERT
143#endif
144
90826881
JH
145/*
146 * TRACEME() will only output things when the $Storable::DEBUGME is true.
147 */
148
111e03c1
RG
149#define TRACEME(x) \
150 STMT_START { \
90826881 151 if (SvTRUE(perl_get_sv("Storable::DEBUGME", TRUE))) \
111e03c1
RG
152 { PerlIO_stdoutf x; PerlIO_stdoutf("\n"); } \
153 } STMT_END
7a6a85bf
RG
154#else
155#define TRACEME(x)
8be2b38b 156#endif /* DEBUGME */
7a6a85bf
RG
157
158#ifdef DASSERT
111e03c1
RG
159#define ASSERT(x,y) \
160 STMT_START { \
7a6a85bf
RG
161 if (!(x)) { \
162 PerlIO_stdoutf("ASSERT FAILED (\"%s\", line %d): ", \
163 __FILE__, __LINE__); \
164 PerlIO_stdoutf y; PerlIO_stdoutf("\n"); \
165 } \
111e03c1 166 } STMT_END
7a6a85bf
RG
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 */
dd19458b 178#define SX_LSCALAR C(1) /* Scalar (large binary) follows (length, data) */
7a6a85bf
RG
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 */
dd19458b 187#define SX_SCALAR C(10) /* Scalar (binary, small) follows (length, data) */
f062ea6c
PN
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 */
7a6a85bf
RG
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 */
f062ea6c
PN
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) */
464b080a 203#define SX_CODE C(26) /* Code references as perl source code */
c3c53033
NC
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 */
7a6a85bf
RG
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 */
d1be9408
JF
213#define SX_KEY 'k' /* A hash key introducer */
214#define SX_VALUE 'v' /* A hash value introducer */
7a6a85bf
RG
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 */
f062ea6c 222#define SX_LG_CLASS 'B' /* Object is blessed, class name length >255 */
7a6a85bf
RG
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 */
251struct 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:
d1be9408 260 * A hash table records the objects which have already been stored.
7a6a85bf
RG
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
274typedef 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
530b72ba
NC
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
c3c53033
NC
314#ifndef SvWEAKREF
315#define WEAKREF_CROAK() CROAK(("Cannot retrieve weak references in this perl"))
316#endif
530b72ba
NC
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
ab923da1
NC
329#ifdef ptr_table_new
330#define USE_PTR_TABLE
331#endif
332
dd19458b
JH
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
0723351e 339struct stcxt;
7a6a85bf
RG
340typedef struct stcxt {
341 int entry; /* flags recursion */
342 int optype; /* type of traversal operation */
ab923da1
NC
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;
e993d95c
JH
353 AV *hook_seen; /* which SVs were returned by STORABLE_freeze() */
354 AV *aseen; /* which objects have been seen, retrieve time */
dfd91409 355 IV where_is_undef; /* index in aseen of PL_sv_undef */
e993d95c
JH
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... */
464b080a
SR
364 int deparse; /* whether to deparse code refs */
365 SV *eval; /* whether to eval source code */
e993d95c 366 int canonical; /* whether to store hashes sorted by key */
530b72ba
NC
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
e8189732 373 int accept_future_minor; /* croak immediately on future minor versions? */
dd19458b 374 int s_dirty; /* context is dirty due to CROAK() -- can be cleaned */
e993d95c
JH
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 */
7a6a85bf
RG
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 */
0723351e 382 SV *(**retrieve_vtbl)(pTHX_ struct stcxt *, char *); /* retrieve dispatch table */
111e03c1
RG
383 SV *prev; /* contexts chained backwards in real recursion */
384 SV *my_sv; /* the blessed scalar who's SvPVX() I am */
7a6a85bf
RG
385} stcxt_t;
386
111e03c1
RG
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
7a6a85bf
RG
397#if defined(MULTIPLICITY) || defined(PERL_OBJECT) || defined(PERL_CAPI)
398
e993d95c 399#if (PATCHLEVEL <= 4) && (SUBVERSION < 68)
7a6a85bf
RG
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) \
111e03c1 409 T name = ((perinterp_sv && SvIOK(perinterp_sv) && SvIVX(perinterp_sv) \
436c6dd3 410 ? (T)SvPVX(SvRV(INT2PTR(SV*,SvIVX(perinterp_sv)))) : (T) 0))
7a6a85bf
RG
411#define dSTCXT \
412 dSTCXT_SV; \
413 dSTCXT_PTR(stcxt_t *, cxt)
414
111e03c1
RG
415#define INIT_STCXT \
416 dSTCXT; \
417 NEW_STORABLE_CXT_OBJ(cxt); \
418 sv_setiv(perinterp_sv, PTR2IV(cxt->my_sv))
7a6a85bf 419
111e03c1
RG
420#define SET_STCXT(x) \
421 STMT_START { \
7a6a85bf 422 dSTCXT_SV; \
111e03c1
RG
423 sv_setiv(perinterp_sv, PTR2IV(x->my_sv)); \
424 } STMT_END
7a6a85bf
RG
425
426#else /* !MULTIPLICITY && !PERL_OBJECT && !PERL_CAPI */
427
85535365 428static stcxt_t *Context_ptr = NULL;
7a6a85bf 429#define dSTCXT stcxt_t *cxt = Context_ptr
85535365 430#define SET_STCXT(x) Context_ptr = x
111e03c1
RG
431#define INIT_STCXT \
432 dSTCXT; \
85535365
RG
433 NEW_STORABLE_CXT_OBJ(cxt); \
434 SET_STCXT(cxt)
111e03c1 435
7a6a85bf
RG
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
111e03c1 456#define CROAK(x) STMT_START { cxt->s_dirty = 1; croak x; } STMT_END
7a6a85bf
RG
457
458/*
459 * End of "thread-safe" related definitions.
460 */
461
462/*
9e21b3d0
JH
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/*
7a6a85bf
RG
494 * key buffer handling
495 */
496#define kbuf (cxt->keybuf).arena
497#define ksiz (cxt->keybuf).asiz
111e03c1
RG
498#define KBUFINIT() \
499 STMT_START { \
7a6a85bf
RG
500 if (!kbuf) { \
501 TRACEME(("** allocating kbuf of 128 bytes")); \
502 New(10003, kbuf, 128, char); \
503 ksiz = 128; \
504 } \
111e03c1
RG
505 } STMT_END
506#define KBUFCHK(x) \
507 STMT_START { \
7a6a85bf 508 if (x >= ksiz) { \
e993d95c 509 TRACEME(("** extending kbuf to %d bytes (had %d)", x+1, ksiz)); \
7a6a85bf
RG
510 Renew(kbuf, x+1, char); \
511 ksiz = x+1; \
512 } \
111e03c1 513 } STMT_END
7a6a85bf
RG
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
111e03c1
RG
533#define MBUF_INIT(x) \
534 STMT_START { \
7a6a85bf
RG
535 if (!mbase) { \
536 TRACEME(("** allocating mbase of %d bytes", MGROW)); \
537 New(10003, mbase, MGROW, char); \
2cc1b180 538 msiz = (STRLEN)MGROW; \
7a6a85bf
RG
539 } \
540 mptr = mbase; \
541 if (x) \
542 mend = mbase + x; \
543 else \
544 mend = mbase + msiz; \
111e03c1 545 } STMT_END
7a6a85bf
RG
546
547#define MBUF_TRUNC(x) mptr = mbase + x
548#define MBUF_SIZE() (mptr - mbase)
549
550/*
e993d95c
JH
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 */
111e03c1
RG
558#define MBUF_SAVE_AND_LOAD(in) \
559 STMT_START { \
e993d95c
JH
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); \
111e03c1 565 } STMT_END
e993d95c 566
111e03c1
RG
567#define MBUF_RESTORE() \
568 STMT_START { \
e993d95c
JH
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); \
111e03c1 573 } STMT_END
e993d95c
JH
574
575/*
7a6a85bf
RG
576 * Use SvPOKp(), because SvPOK() fails on tainted scalars.
577 * See store_scalar() for other usage of this workaround.
578 */
111e03c1
RG
579#define MBUF_LOAD(v) \
580 STMT_START { \
e993d95c 581 ASSERT(cxt->membuf_ro, ("mbase is read-only")); \
7a6a85bf
RG
582 if (!SvPOKp(v)) \
583 CROAK(("Not a scalar string")); \
584 mptr = mbase = SvPV(v, msiz); \
585 mend = mbase + msiz; \
111e03c1 586 } STMT_END
7a6a85bf 587
111e03c1
RG
588#define MBUF_XTEND(x) \
589 STMT_START { \
7a6a85bf
RG
590 int nsz = (int) round_mgrow((x)+msiz); \
591 int offset = mptr - mbase; \
e993d95c
JH
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))); \
7a6a85bf
RG
595 Renew(mbase, nsz, char); \
596 msiz = nsz; \
597 mptr = mbase + offset; \
598 mend = mbase + nsz; \
111e03c1 599 } STMT_END
7a6a85bf 600
111e03c1
RG
601#define MBUF_CHK(x) \
602 STMT_START { \
7a6a85bf
RG
603 if ((mptr + (x)) > mend) \
604 MBUF_XTEND(x); \
111e03c1 605 } STMT_END
7a6a85bf 606
111e03c1
RG
607#define MBUF_GETC(x) \
608 STMT_START { \
7a6a85bf
RG
609 if (mptr < mend) \
610 x = (int) (unsigned char) *mptr++; \
611 else \
612 return (SV *) 0; \
111e03c1 613 } STMT_END
7a6a85bf 614
9e21b3d0 615#ifdef CRAY_HACK
111e03c1
RG
616#define MBUF_GETINT(x) \
617 STMT_START { \
9e21b3d0
JH
618 oC(x); \
619 if ((mptr + 4) <= mend) { \
620 memcpy(oI(&x), mptr, 4); \
621 mptr += 4; \
622 } else \
623 return (SV *) 0; \
111e03c1 624 } STMT_END
9e21b3d0 625#else
111e03c1
RG
626#define MBUF_GETINT(x) \
627 STMT_START { \
7a6a85bf
RG
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; \
111e03c1 636 } STMT_END
9e21b3d0 637#endif
7a6a85bf 638
111e03c1
RG
639#define MBUF_READ(x,s) \
640 STMT_START { \
7a6a85bf
RG
641 if ((mptr + (s)) <= mend) { \
642 memcpy(x, mptr, s); \
643 mptr += s; \
644 } else \
645 return (SV *) 0; \
111e03c1 646 } STMT_END
7a6a85bf 647
111e03c1
RG
648#define MBUF_SAFEREAD(x,s,z) \
649 STMT_START { \
7a6a85bf
RG
650 if ((mptr + (s)) <= mend) { \
651 memcpy(x, mptr, s); \
652 mptr += s; \
653 } else { \
654 sv_free(z); \
655 return (SV *) 0; \
656 } \
111e03c1 657 } STMT_END
7a6a85bf 658
111e03c1
RG
659#define MBUF_PUTC(c) \
660 STMT_START { \
7a6a85bf
RG
661 if (mptr < mend) \
662 *mptr++ = (char) c; \
663 else { \
664 MBUF_XTEND(1); \
665 *mptr++ = (char) c; \
666 } \
111e03c1 667 } STMT_END
7a6a85bf 668
9e21b3d0 669#ifdef CRAY_HACK
111e03c1
RG
670#define MBUF_PUTINT(i) \
671 STMT_START { \
9e21b3d0
JH
672 MBUF_CHK(4); \
673 memcpy(mptr, oI(&i), 4); \
674 mptr += 4; \
111e03c1 675 } STMT_END
9e21b3d0 676#else
111e03c1
RG
677#define MBUF_PUTINT(i) \
678 STMT_START { \
7a6a85bf
RG
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); \
111e03c1 685 } STMT_END
9e21b3d0 686#endif
7a6a85bf 687
111e03c1
RG
688#define MBUF_WRITE(x,s) \
689 STMT_START { \
7a6a85bf
RG
690 MBUF_CHK(s); \
691 memcpy(mptr, x, s); \
692 mptr += s; \
111e03c1 693 } STMT_END
7a6a85bf
RG
694
695/*
7a6a85bf
RG
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
464b080a
SR
705#define svis_CODE 6
706#define svis_OTHER 7
7a6a85bf
RG
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/*
b12202d0 721 * Types for SX_HOOK (last 2 bits in flags).
7a6a85bf
RG
722 */
723
724#define SHT_SCALAR 0
725#define SHT_ARRAY 1
726#define SHT_HASH 2
b12202d0
JH
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 */
7a6a85bf
RG
736
737/*
e16e2ff8
NC
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/*
7a6a85bf
RG
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 */
2aeb6432
NC
769static const char old_magicstr[] = "perl-store"; /* Magic number before 0.6 */
770static const char magicstr[] = "pst0"; /* Used as a magic number */
7a6a85bf 771
2aeb6432
NC
772#define MAGICSTR_BYTES 'p','s','t','0'
773#define OLDMAGICSTR_BYTES 'p','e','r','l','-','s','t','o','r','e'
774
ee0f7aac
NC
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
2aeb6432
NC
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'
ee0f7aac
NC
809#ifdef USE_56_INTERWORK_KLUDGE
810#define BYTEORDER_BYTES_56 '1','2','3','4'
811#endif
2aeb6432
NC
812#else
813#if BYTEORDER == 0x87654321
814#define BYTEORDER_BYTES '8','7','6','5','4','3','2','1'
ee0f7aac
NC
815#ifdef USE_56_INTERWORK_KLUDGE
816#define BYTEORDER_BYTES_56 '4','3','2','1'
817#endif
2aeb6432
NC
818#else
819#if BYTEORDER == 0x4321
820#define BYTEORDER_BYTES '4','3','2','1'
821#else
c597ea9d 822#error Unknown byteorder. Please append your byteorder to Storable.xs
2aeb6432
NC
823#endif
824#endif
825#endif
826#endif
827
828static const char byteorderstr[] = {BYTEORDER_BYTES, 0};
ee0f7aac
NC
829#ifdef USE_56_INTERWORK_KLUDGE
830static const char byteorderstr_56[] = {BYTEORDER_BYTES_56, 0};
831#endif
530b72ba 832
e16e2ff8 833#define STORABLE_BIN_MAJOR 2 /* Binary major "version" */
c3c53033 834#define STORABLE_BIN_MINOR 7 /* Binary minor "version" */
530b72ba 835
c3c53033 836#if (PATCHLEVEL <= 5)
530b72ba 837#define STORABLE_BIN_WRITE_MINOR 4
e16e2ff8 838#else
c3c53033
NC
839/*
840 * Perl 5.6.0 onwards can do weak references.
e16e2ff8 841*/
c3c53033
NC
842#define STORABLE_BIN_WRITE_MINOR 7
843#endif /* (PATCHLEVEL <= 5) */
7a6a85bf 844
e9822705 845#if (PATCHLEVEL < 8 || (PATCHLEVEL == 8 && SUBVERSION < 1))
fcaa57e7
AMS
846#define PL_sv_placeholder PL_sv_undef
847#endif
848
7a6a85bf
RG
849/*
850 * Useful store shortcuts...
851 */
852
a8b7ef86
AMS
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 */
111e03c1
RG
860#define PUTMARK(x) \
861 STMT_START { \
7a6a85bf
RG
862 if (!cxt->fio) \
863 MBUF_PUTC(x); \
864 else if (PerlIO_putc(cxt->fio, x) == EOF) \
865 return -1; \
111e03c1 866 } STMT_END
7a6a85bf 867
111e03c1
RG
868#define WRITE_I32(x) \
869 STMT_START { \
9e21b3d0
JH
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; \
111e03c1 875 } STMT_END
9e21b3d0 876
7a6a85bf 877#ifdef HAS_HTONL
111e03c1
RG
878#define WLEN(x) \
879 STMT_START { \
7a6a85bf
RG
880 if (cxt->netorder) { \
881 int y = (int) htonl(x); \
882 if (!cxt->fio) \
883 MBUF_PUTINT(y); \
9e21b3d0 884 else if (PerlIO_write(cxt->fio,oI(&y),oS(sizeof(y))) != oS(sizeof(y))) \
7a6a85bf
RG
885 return -1; \
886 } else { \
887 if (!cxt->fio) \
888 MBUF_PUTINT(x); \
9e21b3d0 889 else if (PerlIO_write(cxt->fio,oI(&x),oS(sizeof(x))) != oS(sizeof(x))) \
7a6a85bf
RG
890 return -1; \
891 } \
111e03c1 892 } STMT_END
7a6a85bf 893#else
9e21b3d0 894#define WLEN(x) WRITE_I32(x)
7a6a85bf
RG
895#endif
896
111e03c1
RG
897#define WRITE(x,y) \
898 STMT_START { \
7a6a85bf
RG
899 if (!cxt->fio) \
900 MBUF_WRITE(x,y); \
901 else if (PerlIO_write(cxt->fio, x, y) != y) \
902 return -1; \
111e03c1 903 } STMT_END
7a6a85bf 904
111e03c1
RG
905#define STORE_PV_LEN(pv, len, small, large) \
906 STMT_START { \
7a6a85bf
RG
907 if (len <= LG_SCALAR) { \
908 unsigned char clen = (unsigned char) len; \
dd19458b 909 PUTMARK(small); \
7a6a85bf
RG
910 PUTMARK(clen); \
911 if (len) \
912 WRITE(pv, len); \
913 } else { \
dd19458b 914 PUTMARK(large); \
7a6a85bf
RG
915 WLEN(len); \
916 WRITE(pv, len); \
917 } \
111e03c1 918 } STMT_END
7a6a85bf 919
dd19458b
JH
920#define STORE_SCALAR(pv, len) STORE_PV_LEN(pv, len, SX_SCALAR, SX_LSCALAR)
921
922/*
20bb3f55 923 * Store &PL_sv_undef in arrays without recursing through store().
7a6a85bf 924 */
20bb3f55 925#define STORE_SV_UNDEF() \
111e03c1 926 STMT_START { \
7a6a85bf 927 cxt->tagnum++; \
20bb3f55 928 PUTMARK(SX_SV_UNDEF); \
111e03c1 929 } STMT_END
7a6a85bf
RG
930
931/*
932 * Useful retrieve shortcuts...
933 */
934
935#define GETCHAR() \
936 (cxt->fio ? PerlIO_getc(cxt->fio) : (mptr >= mend ? EOF : (int) *mptr++))
937
111e03c1
RG
938#define GETMARK(x) \
939 STMT_START { \
7a6a85bf
RG
940 if (!cxt->fio) \
941 MBUF_GETC(x); \
76df4757 942 else if ((int) (x = PerlIO_getc(cxt->fio)) == EOF) \
7a6a85bf 943 return (SV *) 0; \
111e03c1 944 } STMT_END
7a6a85bf 945
111e03c1
RG
946#define READ_I32(x) \
947 STMT_START { \
9e21b3d0
JH
948 ASSERT(sizeof(x) == sizeof(I32), ("reading an I32")); \
949 oC(x); \
7a6a85bf
RG
950 if (!cxt->fio) \
951 MBUF_GETINT(x); \
9e21b3d0 952 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
7a6a85bf 953 return (SV *) 0; \
111e03c1 954 } STMT_END
9e21b3d0
JH
955
956#ifdef HAS_NTOHL
111e03c1
RG
957#define RLEN(x) \
958 STMT_START { \
9e21b3d0 959 oC(x); \
7a6a85bf
RG
960 if (!cxt->fio) \
961 MBUF_GETINT(x); \
9e21b3d0 962 else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \
7a6a85bf 963 return (SV *) 0; \
9e21b3d0
JH
964 if (cxt->netorder) \
965 x = (int) ntohl(x); \
111e03c1 966 } STMT_END
9e21b3d0
JH
967#else
968#define RLEN(x) READ_I32(x)
7a6a85bf
RG
969#endif
970
111e03c1
RG
971#define READ(x,y) \
972 STMT_START { \
7a6a85bf
RG
973 if (!cxt->fio) \
974 MBUF_READ(x, y); \
975 else if (PerlIO_read(cxt->fio, x, y) != y) \
976 return (SV *) 0; \
111e03c1 977 } STMT_END
7a6a85bf 978
111e03c1
RG
979#define SAFEREAD(x,y,z) \
980 STMT_START { \
7a6a85bf
RG
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 } \
111e03c1 987 } STMT_END
7a6a85bf
RG
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.
b12202d0
JH
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.
dfd91409
NC
1006 *
1007 * i should be true iff sv is immortal (ie PL_sv_yes, PL_sv_no or PL_sv_undef)
7a6a85bf 1008 */
dfd91409 1009#define SEEN(y,c,i) \
111e03c1 1010 STMT_START { \
7a6a85bf
RG
1011 if (!y) \
1012 return (SV *) 0; \
dfd91409 1013 if (av_store(cxt->aseen, cxt->tagnum++, i ? (SV*)(y) : SvREFCNT_inc(y)) == 0) \
7a6a85bf 1014 return (SV *) 0; \
43d061fe 1015 TRACEME(("aseen(#%d) = 0x%"UVxf" (refcnt=%d)", cxt->tagnum-1, \
b12202d0
JH
1016 PTR2UV(y), SvREFCNT(y)-1)); \
1017 if (c) \
1018 BLESS((SV *) (y), c); \
111e03c1 1019 } STMT_END
7a6a85bf
RG
1020
1021/*
1022 * Bless `s' in `p', via a temporary reference, required by sv_bless().
1023 */
111e03c1
RG
1024#define BLESS(s,p) \
1025 STMT_START { \
7a6a85bf
RG
1026 SV *ref; \
1027 HV *stash; \
43d061fe 1028 TRACEME(("blessing 0x%"UVxf" in %s", PTR2UV(s), (p))); \
7a6a85bf
RG
1029 stash = gv_stashpv((p), TRUE); \
1030 ref = newRV_noinc(s); \
1031 (void) sv_bless(ref, stash); \
b162af07 1032 SvRV_set(ref, NULL); \
7a6a85bf 1033 SvREFCNT_dec(ref); \
111e03c1 1034 } STMT_END
138ec36d 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 */
7a6a85bf 1053
138ec36d 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
1066static int store(pTHX_ stcxt_t *cxt, SV *sv);
1067static SV *retrieve(pTHX_ stcxt_t *cxt, char *cname);
7a6a85bf
RG
1068
1069/*
1070 * Dynamic dispatching table for SV store.
1071 */
1072
138ec36d 1073static int store_ref(pTHX_ stcxt_t *cxt, SV *sv);
1074static int store_scalar(pTHX_ stcxt_t *cxt, SV *sv);
1075static int store_array(pTHX_ stcxt_t *cxt, AV *av);
1076static int store_hash(pTHX_ stcxt_t *cxt, HV *hv);
1077static int store_tied(pTHX_ stcxt_t *cxt, SV *sv);
1078static int store_tied_item(pTHX_ stcxt_t *cxt, SV *sv);
1079static int store_code(pTHX_ stcxt_t *cxt, CV *cv);
1080static int store_other(pTHX_ stcxt_t *cxt, SV *sv);
1081static int store_blessed(pTHX_ stcxt_t *cxt, SV *sv, int type, HV *pkg);
1082
93ad979b
MB
1083typedef int (*sv_store_t)(pTHX_ stcxt_t *cxt, SV *sv);
1084
d5748a18 1085static sv_store_t sv_store[] = {
93ad979b
MB
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 */
7a6a85bf
RG
1094};
1095
1096#define SV_STORE(x) (*sv_store[x])
1097
1098/*
1099 * Dynamic dispatching tables for SV retrieval.
1100 */
1101
138ec36d 1102static SV *retrieve_lscalar(pTHX_ stcxt_t *cxt, char *cname);
1103static SV *retrieve_lutf8str(pTHX_ stcxt_t *cxt, char *cname);
1104static SV *old_retrieve_array(pTHX_ stcxt_t *cxt, char *cname);
1105static SV *old_retrieve_hash(pTHX_ stcxt_t *cxt, char *cname);
1106static SV *retrieve_ref(pTHX_ stcxt_t *cxt, char *cname);
1107static SV *retrieve_undef(pTHX_ stcxt_t *cxt, char *cname);
1108static SV *retrieve_integer(pTHX_ stcxt_t *cxt, char *cname);
1109static SV *retrieve_double(pTHX_ stcxt_t *cxt, char *cname);
1110static SV *retrieve_byte(pTHX_ stcxt_t *cxt, char *cname);
1111static SV *retrieve_netint(pTHX_ stcxt_t *cxt, char *cname);
1112static SV *retrieve_scalar(pTHX_ stcxt_t *cxt, char *cname);
1113static SV *retrieve_utf8str(pTHX_ stcxt_t *cxt, char *cname);
1114static SV *retrieve_tied_array(pTHX_ stcxt_t *cxt, char *cname);
1115static SV *retrieve_tied_hash(pTHX_ stcxt_t *cxt, char *cname);
1116static SV *retrieve_tied_scalar(pTHX_ stcxt_t *cxt, char *cname);
1117static SV *retrieve_other(pTHX_ stcxt_t *cxt, char *cname);
1118
93ad979b
MB
1119typedef SV* (*sv_retrieve_t)(pTHX_ stcxt_t *cxt, char *name);
1120
1121static 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 */
7a6a85bf
RG
1152};
1153
138ec36d 1154static SV *retrieve_array(pTHX_ stcxt_t *cxt, char *cname);
1155static SV *retrieve_hash(pTHX_ stcxt_t *cxt, char *cname);
1156static SV *retrieve_sv_undef(pTHX_ stcxt_t *cxt, char *cname);
1157static SV *retrieve_sv_yes(pTHX_ stcxt_t *cxt, char *cname);
1158static SV *retrieve_sv_no(pTHX_ stcxt_t *cxt, char *cname);
1159static SV *retrieve_blessed(pTHX_ stcxt_t *cxt, char *cname);
1160static SV *retrieve_idx_blessed(pTHX_ stcxt_t *cxt, char *cname);
1161static SV *retrieve_hook(pTHX_ stcxt_t *cxt, char *cname);
1162static SV *retrieve_overloaded(pTHX_ stcxt_t *cxt, char *cname);
1163static SV *retrieve_tied_key(pTHX_ stcxt_t *cxt, char *cname);
1164static SV *retrieve_tied_idx(pTHX_ stcxt_t *cxt, char *cname);
1165static SV *retrieve_flag_hash(pTHX_ stcxt_t *cxt, char *cname);
1166static SV *retrieve_code(pTHX_ stcxt_t *cxt, char *cname);
c3c53033
NC
1167static SV *retrieve_weakref(pTHX_ stcxt_t *cxt, char *cname);
1168static SV *retrieve_weakoverloaded(pTHX_ stcxt_t *cxt, char *cname);
138ec36d 1169
93ad979b 1170static const sv_retrieve_t sv_retrieve[] = {
7a6a85bf 1171 0, /* SX_OBJECT -- entry unused dynamically */
93ad979b
MB
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 */
7a6a85bf
RG
1201};
1202
1203#define RETRIEVE(c,x) (*(c)->retrieve_vtbl[(x) >= SX_ERROR ? SX_ERROR : (x)])
1204
138ec36d 1205static SV *mbuf2sv(pTHX);
7a6a85bf
RG
1206
1207/***
1208 *** Context management.
1209 ***/
1210
1211/*
1212 * init_perinterp
1213 *
1214 * Called once per "thread" (interpreter) to initialize some global context.
1215 */
138ec36d 1216static void init_perinterp(pTHX)
f0ffaed8 1217{
7a6a85bf
RG
1218 INIT_STCXT;
1219
1220 cxt->netorder = 0; /* true if network order used */
1221 cxt->forgive_me = -1; /* whether to be forgiving... */
0b6a08b2 1222 cxt->accept_future_minor = -1; /* would otherwise occur too late */
7a6a85bf
RG
1223}
1224
1225/*
e993d95c
JH
1226 * reset_context
1227 *
1228 * Called at the end of every context cleaning, to perform common reset
1229 * operations.
1230 */
1231static 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/*
7a6a85bf
RG
1239 * init_store_context
1240 *
1241 * Initialize a new store context for real recursion.
1242 */
f0ffaed8 1243static void init_store_context(
138ec36d 1244 pTHX_
f0ffaed8
JH
1245 stcxt_t *cxt,
1246 PerlIO *f,
1247 int optype,
1248 int network_order)
7a6a85bf
RG
1249{
1250 TRACEME(("init_store_context"));
1251
1252 cxt->netorder = network_order;
1253 cxt->forgive_me = -1; /* Fetched from perl if needed */
464b080a
SR
1254 cxt->deparse = -1; /* Idem */
1255 cxt->eval = NULL; /* Idem */
7a6a85bf
RG
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
ab923da1
NC
1275#ifdef USE_PTR_TABLE
1276 cxt->pseen = ptr_table_new();
1277 cxt->hseen = 0;
1278#else
7a6a85bf
RG
1279 cxt->hseen = newHV(); /* Table where seen objects are stored */
1280 HvSHAREKEYS_off(cxt->hseen);
ab923da1 1281#endif
7a6a85bf
RG
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 */
f0ffaed8 1298#if PERL_VERSION >= 5
7a6a85bf 1299#define HBUCKETS 4096 /* Buckets for %hseen */
ab923da1 1300#ifndef USE_PTR_TABLE
7a6a85bf
RG
1301 HvMAX(cxt->hseen) = HBUCKETS - 1; /* keys %hseen = $HBUCKETS; */
1302#endif
ab923da1 1303#endif
7a6a85bf
RG
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
f0ffaed8 1315#if PERL_VERSION >= 5
7a6a85bf
RG
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 */
90826881
JH
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 */
7a6a85bf
RG
1338}
1339
1340/*
1341 * clean_store_context
1342 *
1343 * Clean store context by
1344 */
138ec36d 1345static void clean_store_context(pTHX_ stcxt_t *cxt)
7a6a85bf
RG
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
ab923da1 1357#ifndef USE_PTR_TABLE
e993d95c
JH
1358 if (cxt->hseen) {
1359 hv_iterinit(cxt->hseen);
1360 while ((he = hv_iternext(cxt->hseen))) /* Extra () for -Wall, grr.. */
da5add9b 1361 HeVAL(he) = &PL_sv_undef;
e993d95c 1362 }
ab923da1 1363#endif
7a6a85bf 1364
e993d95c
JH
1365 if (cxt->hclass) {
1366 hv_iterinit(cxt->hclass);
1367 while ((he = hv_iternext(cxt->hclass))) /* Extra () for -Wall, grr.. */
da5add9b 1368 HeVAL(he) = &PL_sv_undef;
e993d95c 1369 }
7a6a85bf
RG
1370
1371 /*
1372 * And now dispose of them...
862382c7
JH
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
ab923da1
NC
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
862382c7
JH
1389 if (cxt->hseen) {
1390 HV *hseen = cxt->hseen;
1391 cxt->hseen = 0;
1392 hv_undef(hseen);
1393 sv_free((SV *) hseen);
1394 }
ab923da1 1395#endif
7a6a85bf 1396
862382c7
JH
1397 if (cxt->hclass) {
1398 HV *hclass = cxt->hclass;
1399 cxt->hclass = 0;
1400 hv_undef(hclass);
1401 sv_free((SV *) hclass);
1402 }
7a6a85bf 1403
862382c7
JH
1404 if (cxt->hook) {
1405 HV *hook = cxt->hook;
1406 cxt->hook = 0;
1407 hv_undef(hook);
1408 sv_free((SV *) hook);
1409 }
7a6a85bf 1410
862382c7
JH
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 }
90826881 1417
e8189732 1418 cxt->forgive_me = -1; /* Fetched from perl if needed */
464b080a
SR
1419 cxt->deparse = -1; /* Idem */
1420 if (cxt->eval) {
1421 SvREFCNT_dec(cxt->eval);
1422 }
1423 cxt->eval = NULL; /* Idem */
e8189732
NC
1424 cxt->canonical = -1; /* Idem */
1425
e993d95c 1426 reset_context(cxt);
7a6a85bf
RG
1427}
1428
1429/*
1430 * init_retrieve_context
1431 *
1432 * Initialize a new retrieve context for real recursion.
1433 */
138ec36d 1434static void init_retrieve_context(pTHX_ stcxt_t *cxt, int optype, int is_tainted)
7a6a85bf
RG
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
ab923da1
NC
1449#ifdef USE_PTR_TABLE
1450 cxt->pseen = 0;
1451#endif
1452
7a6a85bf
RG
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
2cc1b180
JH
1460 cxt->hseen = (((void*)cxt->retrieve_vtbl == (void*)sv_old_retrieve)
1461 ? newHV() : 0);
7a6a85bf
RG
1462
1463 cxt->aseen = newAV(); /* Where retrieved objects are kept */
dfd91409 1464 cxt->where_is_undef = -1; /* Special case for PL_sv_undef */
7a6a85bf
RG
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;
dd19458b 1469 cxt->s_tainted = is_tainted;
7a6a85bf 1470 cxt->entry = 1; /* No recursion yet */
530b72ba
NC
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
e8189732 1477 cxt->accept_future_minor = -1; /* Fetched from perl if needed */
7a6a85bf
RG
1478}
1479
1480/*
1481 * clean_retrieve_context
1482 *
1483 * Clean retrieve context by
1484 */
138ec36d 1485static void clean_retrieve_context(pTHX_ stcxt_t *cxt)
7a6a85bf
RG
1486{
1487 TRACEME(("clean_retrieve_context"));
1488
1489 ASSERT(cxt->optype & ST_RETRIEVE, ("was performing a retrieve()"));
1490
862382c7
JH
1491 if (cxt->aseen) {
1492 AV *aseen = cxt->aseen;
1493 cxt->aseen = 0;
1494 av_undef(aseen);
1495 sv_free((SV *) aseen);
1496 }
dfd91409 1497 cxt->where_is_undef = -1;
7a6a85bf 1498
862382c7
JH
1499 if (cxt->aclass) {
1500 AV *aclass = cxt->aclass;
1501 cxt->aclass = 0;
1502 av_undef(aclass);
1503 sv_free((SV *) aclass);
1504 }
7a6a85bf 1505
862382c7
JH
1506 if (cxt->hook) {
1507 HV *hook = cxt->hook;
1508 cxt->hook = 0;
1509 hv_undef(hook);
1510 sv_free((SV *) hook);
1511 }
7a6a85bf 1512
862382c7
JH
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 }
7a6a85bf 1519
e8189732
NC
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
e993d95c 1528 reset_context(cxt);
7a6a85bf
RG
1529}
1530
1531/*
1532 * clean_context
1533 *
1534 * A workaround for the CROAK bug: cleanup the last context.
1535 */
138ec36d 1536static void clean_context(pTHX_ stcxt_t *cxt)
7a6a85bf
RG
1537{
1538 TRACEME(("clean_context"));
1539
dd19458b 1540 ASSERT(cxt->s_dirty, ("dirty context"));
7a6a85bf 1541
e993d95c
JH
1542 if (cxt->membuf_ro)
1543 MBUF_RESTORE();
1544
1545 ASSERT(!cxt->membuf_ro, ("mbase is not read-only"));
1546
7a6a85bf 1547 if (cxt->optype & ST_RETRIEVE)
138ec36d 1548 clean_retrieve_context(aTHX_ cxt);
e993d95c 1549 else if (cxt->optype & ST_STORE)
138ec36d 1550 clean_store_context(aTHX_ cxt);
e993d95c
JH
1551 else
1552 reset_context(cxt);
862382c7
JH
1553
1554 ASSERT(!cxt->s_dirty, ("context is clean"));
e993d95c 1555 ASSERT(cxt->entry == 0, ("context is reset"));
7a6a85bf
RG
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 */
138ec36d 1564static stcxt_t *allocate_context(pTHX_ stcxt_t *parent_cxt)
7a6a85bf
RG
1565{
1566 stcxt_t *cxt;
1567
1568 TRACEME(("allocate_context"));
1569
dd19458b 1570 ASSERT(!parent_cxt->s_dirty, ("parent context clean"));
7a6a85bf 1571
111e03c1
RG
1572 NEW_STORABLE_CXT_OBJ(cxt);
1573 cxt->prev = parent_cxt->my_sv;
7a6a85bf
RG
1574 SET_STCXT(cxt);
1575
e993d95c
JH
1576 ASSERT(!cxt->s_dirty, ("clean context"));
1577
7a6a85bf
RG
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 */
138ec36d 1587static void free_context(pTHX_ stcxt_t *cxt)
7a6a85bf 1588{
111e03c1 1589 stcxt_t *prev = (stcxt_t *)(cxt->prev ? SvPVX(SvRV(cxt->prev)) : 0);
7a6a85bf
RG
1590
1591 TRACEME(("free_context"));
1592
dd19458b 1593 ASSERT(!cxt->s_dirty, ("clean context"));
7a6a85bf
RG
1594 ASSERT(prev, ("not freeing root context"));
1595
111e03c1 1596 SvREFCNT_dec(cxt->my_sv);
7a6a85bf 1597 SET_STCXT(prev);
e993d95c
JH
1598
1599 ASSERT(cxt, ("context not void"));
7a6a85bf
RG
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 */
138ec36d 1611int is_storing(pTHX)
7a6a85bf
RG
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 */
138ec36d 1623int is_retrieving(pTHX)
7a6a85bf
RG
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 */
138ec36d 1638int last_op_in_netorder(pTHX)
7a6a85bf
RG
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 */
f0ffaed8 1657static SV *pkg_fetchmeth(
138ec36d 1658 pTHX_
f0ffaed8
JH
1659 HV *cache,
1660 HV *pkg,
1661 char *method)
7a6a85bf
RG
1662{
1663 GV *gv;
1664 SV *sv;
bfcb3514
NC
1665 const char *hvname = HvNAME_get(pkg);
1666
7a6a85bf
RG
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));
bfcb3514 1676 TRACEME(("%s->%s: 0x%"UVxf, hvname, method, PTR2UV(sv)));
7a6a85bf
RG
1677 } else {
1678 sv = newSVsv(&PL_sv_undef);
bfcb3514 1679 TRACEME(("%s->%s: not found", hvname, method));
7a6a85bf
RG
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
bfcb3514 1687 (void) hv_store(cache, hvname, strlen(hvname), sv, 0);
7a6a85bf
RG
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 */
f0ffaed8 1697static void pkg_hide(
138ec36d 1698 pTHX_
f0ffaed8
JH
1699 HV *cache,
1700 HV *pkg,
1701 char *method)
7a6a85bf 1702{
bfcb3514 1703 const char *hvname = HvNAME_get(pkg);
7a6a85bf 1704 (void) hv_store(cache,
bfcb3514 1705 hvname, strlen(hvname), newSVsv(&PL_sv_undef), 0);
7a6a85bf
RG
1706}
1707
1708/*
212e9bde
JH
1709 * pkg_uncache
1710 *
1711 * Discard cached value: a whole fetch loop will be retried at next lookup.
1712 */
1713static void pkg_uncache(
138ec36d 1714 pTHX_
212e9bde
JH
1715 HV *cache,
1716 HV *pkg,
1717 char *method)
1718{
bfcb3514
NC
1719 const char *hvname = HvNAME_get(pkg);
1720 (void) hv_delete(cache, hvname, strlen(hvname), G_DISCARD);
212e9bde
JH
1721}
1722
1723/*
7a6a85bf
RG
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 */
f0ffaed8 1731static SV *pkg_can(
138ec36d 1732 pTHX_
f0ffaed8
JH
1733 HV *cache,
1734 HV *pkg,
1735 char *method)
7a6a85bf
RG
1736{
1737 SV **svh;
1738 SV *sv;
bfcb3514 1739 const char *hvname = HvNAME_get(pkg);
7a6a85bf 1740
bfcb3514 1741 TRACEME(("pkg_can for %s->%s", hvname, method));
7a6a85bf
RG
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
bfcb3514 1751 svh = hv_fetch(cache, hvname, strlen(hvname), FALSE);
7a6a85bf
RG
1752 if (svh) {
1753 sv = *svh;
1754 if (!SvOK(sv)) {
bfcb3514 1755 TRACEME(("cached %s->%s: not found", hvname, method));
7a6a85bf
RG
1756 return (SV *) 0;
1757 } else {
43d061fe 1758 TRACEME(("cached %s->%s: 0x%"UVxf,
bfcb3514 1759 hvname, method, PTR2UV(sv)));
7a6a85bf
RG
1760 return sv;
1761 }
1762 }
1763
1764 TRACEME(("not cached yet"));
138ec36d 1765 return pkg_fetchmeth(aTHX_ cache, pkg, method); /* Fetch and cache */
7a6a85bf
RG
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 */
f0ffaed8 1774static SV *scalar_call(
138ec36d 1775 pTHX_
f0ffaed8
JH
1776 SV *obj,
1777 SV *hook,
1778 int cloning,
1779 AV *av,
1780 I32 flags)
7a6a85bf
RG
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++) {
43d061fe
JH
1800 TRACEME(("pushing arg #%d (0x%"UVxf")...",
1801 i, PTR2UV(ary[i])));
7a6a85bf
RG
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 *
f9a1036d 1828 * Call routine obj->hook(cloning) in list context.
7a6a85bf
RG
1829 * Returns the list of returned values in an array.
1830 */
f0ffaed8 1831static AV *array_call(
138ec36d 1832 pTHX_
f0ffaed8
JH
1833 SV *obj,
1834 SV *hook,
1835 int cloning)
7a6a85bf
RG
1836{
1837 dSP;
1838 int count;
1839 AV *av;
1840 int i;
1841
f0ffaed8 1842 TRACEME(("array_call (cloning=%d)", cloning));
7a6a85bf
RG
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 */
f0ffaed8 1877static int known_class(
138ec36d 1878 pTHX_
f0ffaed8
JH
1879 stcxt_t *cxt,
1880 char *name, /* Class name */
1881 int len, /* Name length */
1882 I32 *classnum)
7a6a85bf
RG
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.
7a6a85bf
RG
1902 */
1903
1904 cxt->classnum++;
3341c981 1905 if (!hv_store(hclass, name, len, INT2PTR(SV*, cxt->classnum), 0))
7a6a85bf
RG
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 */
138ec36d 1922static int store_ref(pTHX_ stcxt_t *cxt, SV *sv)
7a6a85bf 1923{
c3c53033 1924 int is_weak = 0;
43d061fe 1925 TRACEME(("store_ref (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
1926
1927 /*
1928 * Follow reference, and check if target is overloaded.
1929 */
1930
96466a21 1931#ifdef SvWEAKREF
c3c53033
NC
1932 if (SvWEAKREF(sv))
1933 is_weak = 1;
1934 TRACEME(("ref (0x%"UVxf") is%s weak", PTR2UV(sv), is_weak ? "" : "n't"));
1935#endif
7a6a85bf
RG
1936 sv = SvRV(sv);
1937
1938 if (SvOBJECT(sv)) {
1939 HV *stash = (HV *) SvSTASH(sv);
1940 if (stash && Gv_AMG(stash)) {
9e21b3d0 1941 TRACEME(("ref (0x%"UVxf") is overloaded", PTR2UV(sv)));
c3c53033 1942 PUTMARK(is_weak ? SX_WEAKOVERLOAD : SX_OVERLOAD);
7a6a85bf 1943 } else
c3c53033 1944 PUTMARK(is_weak ? SX_WEAKREF : SX_REF);
7a6a85bf 1945 } else
c3c53033 1946 PUTMARK(is_weak ? SX_WEAKREF : SX_REF);
7a6a85bf 1947
138ec36d 1948 return store(aTHX_ cxt, sv);
7a6a85bf
RG
1949}
1950
1951/*
1952 * store_scalar
1953 *
1954 * Store a scalar.
1955 *
e16e2ff8 1956 * Layout is SX_LSCALAR <length> <data>, SX_SCALAR <length> <data> or SX_UNDEF.
7a6a85bf
RG
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 */
138ec36d 1962static int store_scalar(pTHX_ stcxt_t *cxt, SV *sv)
7a6a85bf
RG
1963{
1964 IV iv;
1965 char *pv;
1966 STRLEN len;
1967 U32 flags = SvFLAGS(sv); /* "cc -O" may put it in register */
1968
43d061fe 1969 TRACEME(("store_scalar (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
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 {
86bbd6dc 1982 TRACEME(("undef at 0x%"UVxf, PTR2UV(sv)));
7a6a85bf
RG
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 }
db670f21
NC
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:
7a6a85bf 2049
db670f21
NC
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 */
7a6a85bf 2065
db670f21
NC
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;
7a6a85bf 2077
7a6a85bf 2078
db670f21
NC
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
7a6a85bf 2091
db670f21
NC
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
7a6a85bf 2115
db670f21 2116 SvIV_please(sv);
3ddd445a 2117 if (SvIOK_notUV(sv)) {
db670f21
NC
2118 iv = SvIV(sv);
2119 goto integer; /* Share code above */
2120 }
2121 nv = SvNV(sv);
2122#endif
7a6a85bf 2123
db670f21
NC
2124 if (cxt->netorder) {
2125 TRACEME(("double %"NVff" stored as string", nv));
2126 goto string_readlen; /* Share code below */
2127 }
7a6a85bf 2128
db670f21
NC
2129 PUTMARK(SX_DOUBLE);
2130 WRITE(&nv, sizeof(nv));
7a6a85bf 2131
db670f21 2132 TRACEME(("ok (double 0x%"UVxf", value = %"NVff")", PTR2UV(sv), nv));
7a6a85bf 2133
db670f21
NC
2134 } else if (flags & (SVp_POK | SVp_NOK | SVp_IOK)) {
2135 I32 wlen; /* For 64-bit machines */
7a6a85bf 2136
db670f21
NC
2137 string_readlen:
2138 pv = SvPV(sv, len);
7a6a85bf 2139
db670f21
NC
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));
7a6a85bf 2153 } else
db670f21
NC
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 */
7a6a85bf
RG
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 */
138ec36d 2168static int store_array(pTHX_ stcxt_t *cxt, AV *av)
7a6a85bf
RG
2169{
2170 SV **sav;
2171 I32 len = av_len(av) + 1;
2172 I32 i;
2173 int ret;
2174
43d061fe 2175 TRACEME(("store_array (0x%"UVxf")", PTR2UV(av)));
7a6a85bf
RG
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));
20bb3f55 2193 STORE_SV_UNDEF();
7a6a85bf
RG
2194 continue;
2195 }
2196 TRACEME(("(#%d) item", i));
138ec36d 2197 if ((ret = store(aTHX_ cxt, *sav))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2198 return ret;
2199 }
2200
2201 TRACEME(("ok (array)"));
2202
2203 return 0;
2204}
2205
138ec36d 2206
2207#if (PATCHLEVEL <= 6)
2208
7a6a85bf
RG
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 */
2215static int
f0ffaed8 2216sortcmp(const void *a, const void *b)
7a6a85bf 2217{
138ec36d 2218#if defined(USE_ITHREADS)
2219 dTHX;
2220#endif /* USE_ITHREADS */
2221 return sv_cmp(*(SV * const *) a, *(SV * const *) b);
7a6a85bf
RG
2222}
2223
138ec36d 2224#endif /* PATCHLEVEL <= 6 */
7a6a85bf
RG
2225
2226/*
2227 * store_hash
2228 *
d1be9408 2229 * Store a hash table.
7a6a85bf 2230 *
e16e2ff8
NC
2231 * For a "normal" hash (not restricted, no utf8 keys):
2232 *
7a6a85bf
RG
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.
c194a0a3
TB
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,
e16e2ff8
NC
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
7a6a85bf 2247 */
138ec36d 2248static int store_hash(pTHX_ stcxt_t *cxt, HV *hv)
7a6a85bf 2249{
27da23d5 2250 dVAR;
530b72ba
NC
2251 I32 len =
2252#ifdef HAS_RESTRICTED_HASHES
2253 HvTOTALKEYS(hv);
2254#else
2255 HvKEYS(hv);
2256#endif
7a6a85bf
RG
2257 I32 i;
2258 int ret = 0;
2259 I32 riter;
2260 HE *eiter;
530b72ba
NC
2261 int flagged_hash = ((SvREADONLY(hv)
2262#ifdef HAS_HASH_KEY_FLAGS
2263 || HvHASKFLAGS(hv)
2264#endif
2265 ) ? 1 : 0);
e16e2ff8 2266 unsigned char hash_flags = (SvREADONLY(hv) ? SHV_RESTRICTED : 0);
7a6a85bf 2267
e16e2ff8
NC
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 }
7a6a85bf
RG
2275
2276 /*
2277 * Signal hash by emitting SX_HASH, followed by the table length.
2278 */
2279
e16e2ff8
NC
2280 if (flagged_hash) {
2281 PUTMARK(SX_FLAG_HASH);
2282 PUTMARK(hash_flags);
2283 } else {
2284 PUTMARK(SX_HASH);
2285 }
7a6a85bf
RG
2286 WLEN(len);
2287 TRACEME(("size = %d", len));
2288
2289 /*
2290 * Save possible iteration state via each() on that table.
2291 */
2292
bfcb3514
NC
2293 riter = HvRITER_get(hv);
2294 eiter = HvEITER_get(hv);
7a6a85bf
RG
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 =
e16e2ff8 2311 (SvTRUE(perl_get_sv("Storable::canonical", TRUE)) ? 1 : 0))))
7a6a85bf
RG
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
e16e2ff8
NC
2322 /*av_extend (av, len);*/
2323
7a6a85bf
RG
2324 TRACEME(("using canonical order"));
2325
2326 for (i = 0; i < len; i++) {
530b72ba 2327#ifdef HAS_RESTRICTED_HASHES
e16e2ff8 2328 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
530b72ba
NC
2329#else
2330 HE *he = hv_iternext(hv);
2331#endif
7a6a85bf
RG
2332 SV *key = hv_iterkeysv(he);
2333 av_store(av, AvFILLp(av)+1, key); /* av_push(), really */
2334 }
2335
138ec36d 2336 STORE_HASH_SORT;
7a6a85bf
RG
2337
2338 for (i = 0; i < len; i++) {
dfd91409 2339#ifdef HAS_RESTRICTED_HASHES
ca732855 2340 int placeholders = (int)HvPLACEHOLDERS_get(hv);
dfd91409
NC
2341#endif
2342 unsigned char flags = 0;
7a6a85bf 2343 char *keyval;
e16e2ff8
NC
2344 STRLEN keylen_tmp;
2345 I32 keylen;
7a6a85bf 2346 SV *key = av_shift(av);
dfd91409
NC
2347 /* This will fail if key is a placeholder.
2348 Track how many placeholders we have, and error if we
2349 "see" too many. */
7a6a85bf 2350 HE *he = hv_fetch_ent(hv, key, 0, 0);
dfd91409
NC
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 }
7a6a85bf
RG
2377
2378 /*
2379 * Store value first.
2380 */
2381
9e21b3d0 2382 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
7a6a85bf 2383
138ec36d 2384 if ((ret = store(aTHX_ cxt, val))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
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
e16e2ff8
NC
2395 /* Implementation of restricted hashes isn't nicely
2396 abstracted: */
dfd91409
NC
2397 if ((hash_flags & SHV_RESTRICTED) && SvREADONLY(val)) {
2398 flags |= SHV_K_LOCKED;
2399 }
e16e2ff8
NC
2400
2401 keyval = SvPV(key, keylen_tmp);
2402 keylen = keylen_tmp;
530b72ba
NC
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. */
e16e2ff8
NC
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 }
530b72ba 2433#endif
e16e2ff8
NC
2434
2435 if (flagged_hash) {
2436 PUTMARK(flags);
2437 TRACEME(("(#%d) key '%s' flags %x %u", i, keyval, flags, *keyval));
2438 } else {
fcaa57e7
AMS
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);
e16e2ff8
NC
2446 TRACEME(("(#%d) key '%s'", i, keyval));
2447 }
7a6a85bf
RG
2448 WLEN(keylen);
2449 if (keylen)
2450 WRITE(keyval, keylen);
e16e2ff8
NC
2451 if (flags & SHV_K_WASUTF8)
2452 Safefree (keyval);
7a6a85bf
RG
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
a6d05634 2466 * within the hash). This is the default and will be faster!
7a6a85bf
RG
2467 */
2468
2469 for (i = 0; i < len; i++) {
0bb78401 2470 char *key = 0;
7a6a85bf 2471 I32 len;
e16e2ff8 2472 unsigned char flags;
530b72ba 2473#ifdef HV_ITERNEXT_WANTPLACEHOLDERS
e16e2ff8 2474 HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS);
530b72ba
NC
2475#else
2476 HE *he = hv_iternext(hv);
2477#endif
e16e2ff8
NC
2478 SV *val = (he ? hv_iterval(hv, he) : 0);
2479 SV *key_sv = NULL;
2480 HEK *hek;
7a6a85bf
RG
2481
2482 if (val == 0)
2483 return 1; /* Internal error, not I/O error */
2484
dfd91409
NC
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
7a6a85bf
RG
2497 /*
2498 * Store value first.
2499 */
2500
9e21b3d0 2501 TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val)));
7a6a85bf 2502
138ec36d 2503 if ((ret = store(aTHX_ cxt, val))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2504 goto out;
2505
e16e2ff8
NC
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. */
530b72ba 2520#ifdef HAS_HASH_KEY_FLAGS
e16e2ff8
NC
2521 if (HEK_UTF8(hek))
2522 flags |= SHV_K_UTF8;
2523 if (HEK_WASUTF8(hek))
2524 flags |= SHV_K_WASUTF8;
530b72ba 2525#endif
e16e2ff8
NC
2526 key = HEK_KEY(hek);
2527 }
7a6a85bf
RG
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
e16e2ff8
NC
2536 if (flagged_hash) {
2537 PUTMARK(flags);
2538 TRACEME(("(#%d) key '%s' flags %x", i, key, flags));
2539 } else {
fcaa57e7
AMS
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);
e16e2ff8
NC
2547 TRACEME(("(#%d) key '%s'", i, key));
2548 }
2549 if (flags & SHV_K_ISSV) {
138ec36d 2550 store(aTHX_ cxt, key_sv);
e16e2ff8
NC
2551 } else {
2552 WLEN(len);
2553 if (len)
7a6a85bf 2554 WRITE(key, len);
e16e2ff8 2555 }
7a6a85bf
RG
2556 }
2557 }
2558
43d061fe 2559 TRACEME(("ok (hash 0x%"UVxf")", PTR2UV(hv)));
7a6a85bf
RG
2560
2561out:
bfcb3514
NC
2562 HvRITER_set(hv, riter); /* Restore hash iterator state */
2563 HvEITER_set(hv, eiter);
7a6a85bf
RG
2564
2565 return ret;
2566}
2567
2568/*
464b080a
SR
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 */
138ec36d 2576static int store_code(pTHX_ stcxt_t *cxt, CV *cv)
464b080a
SR
2577{
2578#if PERL_VERSION < 6
2579 /*
2580 * retrieve_code does not work with perl 5.005 or less
2581 */
138ec36d 2582 return store_other(aTHX_ cxt, (SV*)cv);
464b080a
SR
2583#else
2584 dSP;
2585 I32 len;
c5661c80 2586 int count, reallen;
464b080a
SR
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 ) {
138ec36d 2596 return store_other(aTHX_ cxt, (SV*)cv);
464b080a
SR
2597 }
2598
2599 /*
2600 * Require B::Deparse. At least B::Deparse 0.61 is needed for
2601 * blessed code references.
2602 */
17625bd2 2603 /* Ownership of both SVs is passed to load_module, which frees them. */
464b080a
SR
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;
dfe4365a 2636 len = SvCUR(text);
e3feee4e 2637 reallen = strlen(SvPV_nolen(text));
464b080a
SR
2638
2639 /*
2640 * Empty code references or XS functions are deparsed as
2641 * "(prototype) ;" or ";".
2642 */
2643
e3feee4e 2644 if (len == 0 || *(SvPV_nolen(text)+reallen-1) == ';') {
464b080a
SR
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);
a8b7ef86 2653 cxt->tagnum++; /* necessary, as SX_CODE is a SEEN() candidate */
464b080a 2654 TRACEME(("size = %d", len));
e3feee4e 2655 TRACEME(("code = %s", SvPV_nolen(text)));
464b080a
SR
2656
2657 /*
2658 * Now store the source code.
2659 */
2660
e3feee4e 2661 STORE_SCALAR(SvPV_nolen(text), len);
464b080a
SR
2662
2663 FREETMPS;
2664 LEAVE;
2665
2666 TRACEME(("ok (code)"));
2667
2668 return 0;
2669#endif
2670}
2671
2672/*
7a6a85bf
RG
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 */
138ec36d 2680static int store_tied(pTHX_ stcxt_t *cxt, SV *sv)
7a6a85bf
RG
2681{
2682 MAGIC *mg;
72edffd8 2683 SV *obj = NULL;
7a6a85bf
RG
2684 int ret = 0;
2685 int svt = SvTYPE(sv);
2686 char mtype = 'P';
2687
43d061fe 2688 TRACEME(("store_tied (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
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
72edffd8
AMS
2729 /* [#17040] mg_obj is NULL for scalar self-ties. AMS 20030416 */
2730 obj = mg->mg_obj ? mg->mg_obj : newSV(0);
138ec36d 2731 if ((ret = store(aTHX_ cxt, obj)))
7a6a85bf
RG
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 */
138ec36d 2751static int store_tied_item(pTHX_ stcxt_t *cxt, SV *sv)
7a6a85bf
RG
2752{
2753 MAGIC *mg;
2754 int ret;
2755
43d061fe 2756 TRACEME(("store_tied_item (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
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);
9e21b3d0 2768 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
7a6a85bf 2769
138ec36d 2770 if ((ret = store(aTHX_ cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
7a6a85bf
RG
2771 return ret;
2772
9e21b3d0 2773 TRACEME(("store_tied_item: storing PTR 0x%"UVxf, PTR2UV(mg->mg_ptr)));
7a6a85bf 2774
138ec36d 2775 if ((ret = store(aTHX_ cxt, (SV *) mg->mg_ptr))) /* Idem, for -Wall */
7a6a85bf
RG
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);
9e21b3d0 2782 TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj)));
7a6a85bf 2783
138ec36d 2784 if ((ret = store(aTHX_ cxt, mg->mg_obj))) /* Idem, for -Wall */
7a6a85bf
RG
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>.
b12202d0
JH
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.
7a6a85bf 2841 */
f0ffaed8 2842static int store_hook(
138ec36d 2843 pTHX_
f0ffaed8
JH
2844 stcxt_t *cxt,
2845 SV *sv,
2846 int type,
2847 HV *pkg,
2848 SV *hook)
7a6a85bf
RG
2849{
2850 I32 len;
0723351e 2851 char *classname;
7a6a85bf
RG
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;
e993d95c
JH
2865 char mtype = '\0'; /* for blessed ref to tied structures */
2866 unsigned char eflags = '\0'; /* used when object type is SHT_EXTRA */
7a6a85bf 2867
bfcb3514 2868 TRACEME(("store_hook, classname \"%s\", tagged #%d", HvNAME_get(pkg), cxt->tagnum));
7a6a85bf
RG
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;
b12202d0
JH
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;
7a6a85bf
RG
2914 default:
2915 CROAK(("Unexpected object type (%d) in store_hook()", type));
2916 }
2917 flags = SHF_NEED_RECURSE | obj_type;
2918
bfcb3514 2919 classname = HvNAME_get(pkg);
0723351e 2920 len = strlen(classname);
7a6a85bf
RG
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
0723351e 2935 TRACEME(("about to call STORABLE_freeze on class %s", classname));
7a6a85bf
RG
2936
2937 ref = newRV_noinc(sv); /* Temporary reference */
138ec36d 2938 av = array_call(aTHX_ ref, hook, clone); /* @a = $object->STORABLE_freeze($c) */
b162af07 2939 SvRV_set(ref, NULL);
7a6a85bf
RG
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
0723351e 2959 if (hv_fetch(cxt->hclass, classname, len, FALSE))
7a6a85bf 2960 CROAK(("Too late to ignore hooks for %s class \"%s\"",
0723351e 2961 (cxt->optype & ST_CLONE) ? "cloning" : "storing", classname));
7a6a85bf 2962
138ec36d 2963 pkg_hide(aTHX_ cxt->hook, pkg, "STORABLE_freeze");
7a6a85bf 2964
138ec36d 2965 ASSERT(!pkg_can(aTHX_ cxt->hook, pkg, "STORABLE_freeze"), ("hook invisible"));
0723351e 2966 TRACEME(("ignoring STORABLE_freeze in class \"%s\"", classname));
7a6a85bf 2967
138ec36d 2968 return store_blessed(aTHX_ cxt, sv, type, pkg);
7a6a85bf
RG
2969 }
2970
2971 /*
2972 * Get frozen string.
2973 */
2974
2975 ary = AvARRAY(av);
2976 pv = SvPV(ary[0], len2);
2f796f32
AMS
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 }
7a6a85bf
RG
2987
2988 /*
7a6a85bf
RG
2989 * If they returned more than one item, we need to serialize some
2990 * extra references if not already done.
2991 *
10ffa93f 2992 * Loop over the array, starting at position #1, and for each item,
7a6a85bf
RG
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++) {
ab923da1
NC
3002#ifdef USE_PTR_TABLE
3003 char *fake_tag;
3004#else
7a6a85bf 3005 SV **svh;
ab923da1 3006#endif
90826881
JH
3007 SV *rsv = ary[i];
3008 SV *xsv;
ab923da1 3009 SV *tag;
90826881 3010 AV *av_hook = cxt->hook_seen;
7a6a85bf 3011
90826881
JH
3012 if (!SvROK(rsv))
3013 CROAK(("Item #%d returned by STORABLE_freeze "
0723351e 3014 "for %s is not a reference", i, classname));
90826881 3015 xsv = SvRV(rsv); /* Follow ref to know what to look for */
7a6a85bf
RG
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 */
ab923da1
NC
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 */
0bb78401 3027 if ((fake_tag = ptr_table_fetch(cxt->pseen, xsv)))
ab923da1
NC
3028 goto sv_seen; /* Avoid moving code too far to the right */
3029#else
13689cfe 3030 if ((svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE)))
7a6a85bf 3031 goto sv_seen; /* Avoid moving code too far to the right */
ab923da1 3032#endif
7a6a85bf 3033
9e21b3d0 3034 TRACEME(("listed object %d at 0x%"UVxf" is unknown", i-1, PTR2UV(xsv)));
7a6a85bf
RG
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
b12202d0
JH
3047 /* [SX_HOOK] <flags> [<extra>] <object>*/
3048 if (!recursed++) {
7a6a85bf 3049 PUTMARK(SX_HOOK);
b12202d0
JH
3050 PUTMARK(flags);
3051 if (obj_type == SHT_EXTRA)
3052 PUTMARK(eflags);
3053 } else
3054 PUTMARK(flags);
7a6a85bf 3055
138ec36d 3056 if ((ret = store(aTHX_ cxt, xsv))) /* Given by hook for us to store */
7a6a85bf
RG
3057 return ret;
3058
ab923da1
NC
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
7a6a85bf
RG
3064 svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE);
3065 if (!svh)
0723351e 3066 CROAK(("Could not serialize item #%d from hook in %s", i, classname));
ab923da1 3067#endif
7a6a85bf 3068 /*
90826881
JH
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.
7a6a85bf
RG
3079 */
3080
90826881
JH
3081 av_store(av_hook, AvFILLp(av_hook)+1, SvREFCNT_inc(xsv));
3082
7a6a85bf 3083 sv_seen:
90826881
JH
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
ab923da1
NC
3097#ifdef USE_PTR_TABLE
3098 tag = (SV *)--fake_tag;
3099#else
3100 tag = *svh;
3101#endif
3102 ary[i] = tag
76edffbb 3103 TRACEME(("listed object %d at 0x%"UVxf" is tag #%"UVuf,
ab923da1 3104 i-1, PTR2UV(xsv), PTR2UV(tag)));
7a6a85bf
RG
3105 }
3106
3107 /*
dd19458b
JH
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
2f796f32 3116check_done:
0723351e
NC
3117 if (!known_class(aTHX_ cxt, classname, len, &classnum)) {
3118 TRACEME(("first time we see class %s, ID = %d", classname, classnum));
dd19458b
JH
3119 classnum = -1; /* Mark: we must store classname */
3120 } else {
0723351e 3121 TRACEME(("already seen class %s, ID = %d", classname, classnum));
dd19458b
JH
3122 }
3123
3124 /*
7a6a85bf
RG
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
9e21b3d0
JH
3149 TRACEME(("SX_HOOK (recursed=%d) flags=0x%x "
3150 "class=%"IVdf" len=%"IVdf" len2=%"IVdf" len3=%d",
d67b2c17 3151 recursed, flags, (IV)classnum, (IV)len, (IV)len2, count-1));
7a6a85bf 3152
b12202d0
JH
3153 /* SX_HOOK <flags> [<extra>] */
3154 if (!recursed) {
7a6a85bf 3155 PUTMARK(SX_HOOK);
b12202d0
JH
3156 PUTMARK(flags);
3157 if (obj_type == SHT_EXTRA)
3158 PUTMARK(eflags);
3159 } else
3160 PUTMARK(flags);
7a6a85bf
RG
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 }
0723351e 3177 WRITE(classname, len); /* Final \0 is omitted */
7a6a85bf
RG
3178 }
3179
3180 /* <len2> <frozen-str> */
cc964657
JH
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 {
7a6a85bf
RG
3185 unsigned char clen = (unsigned char) len2;
3186 PUTMARK(clen);
3187 }
3188 if (len2)
7c436af3 3189 WRITE(pv, (SSize_t)len2); /* Final \0 is omitted */
7a6a85bf
RG
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]));
9e21b3d0 3208 WRITE_I32(tagval);
7a6a85bf
RG
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
b12202d0
JH
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
138ec36d 3244 if ((ret = store(aTHX_ cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */
b12202d0
JH
3245 return ret;
3246 }
3247
7a6a85bf
RG
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 */
f0ffaed8 3275static int store_blessed(
138ec36d 3276 pTHX_
f0ffaed8
JH
3277 stcxt_t *cxt,
3278 SV *sv,
3279 int type,
3280 HV *pkg)
7a6a85bf
RG
3281{
3282 SV *hook;
3283 I32 len;
0723351e 3284 char *classname;
7a6a85bf
RG
3285 I32 classnum;
3286
bfcb3514 3287 TRACEME(("store_blessed, type %d, class \"%s\"", type, HvNAME_get(pkg)));
7a6a85bf
RG
3288
3289 /*
3290 * Look for a hook for this blessed SV and redirect to store_hook()
3291 * if needed.
3292 */
3293
138ec36d 3294 hook = pkg_can(aTHX_ cxt->hook, pkg, "STORABLE_freeze");
7a6a85bf 3295 if (hook)
138ec36d 3296 return store_hook(aTHX_ cxt, sv, type, pkg, hook);
7a6a85bf
RG
3297
3298 /*
3299 * This is a blessed SV without any serialization hook.
3300 */
3301
bfcb3514 3302 classname = HvNAME_get(pkg);
0723351e 3303 len = strlen(classname);
7a6a85bf 3304
43d061fe 3305 TRACEME(("blessed 0x%"UVxf" in %s, no hook: tagged #%d",
5e081687 3306 PTR2UV(sv), classname, cxt->tagnum));
7a6a85bf
RG
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
0723351e
NC
3315 if (known_class(aTHX_ cxt, classname, len, &classnum)) {
3316 TRACEME(("already seen class %s, ID = %d", classname, classnum));
7a6a85bf
RG
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 {
0723351e 3327 TRACEME(("first time we see class %s, ID = %d", classname, classnum));
7a6a85bf
RG
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 }
0723351e 3337 WRITE(classname, len); /* Final \0 is omitted */
7a6a85bf
RG
3338 }
3339
3340 /*
3341 * Now emit the <object> part.
3342 */
3343
138ec36d 3344 return SV_STORE(type)(aTHX_ cxt, sv);
7a6a85bf
RG
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 */
138ec36d 3357static int store_other(pTHX_ stcxt_t *cxt, SV *sv)
7a6a85bf 3358{
cc964657 3359 I32 len;
27da23d5 3360 char buf[80];
7a6a85bf
RG
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
43d061fe
JH
3375 warn("Can't store item %s(0x%"UVxf")",
3376 sv_reftype(sv, FALSE), PTR2UV(sv));
7a6a85bf
RG
3377
3378 /*
3379 * Store placeholder string as a scalar instead...
3380 */
3381
13689cfe 3382 (void) sprintf(buf, "You lost %s(0x%"UVxf")%c", sv_reftype(sv, FALSE),
e993d95c 3383 PTR2UV(sv), (char) 0);
7a6a85bf
RG
3384
3385 len = strlen(buf);
3386 STORE_SCALAR(buf, len);
1cf92b12 3387 TRACEME(("ok (dummy \"%s\", length = %"IVdf")", buf, (IV) len));
7a6a85bf
RG
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 */
138ec36d 3404static int sv_type(pTHX_ SV *sv)
7a6a85bf
RG
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;
464b080a
SR
3446 case SVt_PVCV:
3447 return svis_CODE;
7a6a85bf
RG
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 */
138ec36d 3464static int store(pTHX_ stcxt_t *cxt, SV *sv)
7a6a85bf
RG
3465{
3466 SV **svh;
3467 int ret;
7a6a85bf 3468 int type;
ab923da1
NC
3469#ifdef USE_PTR_TABLE
3470 struct ptr_tbl *pseen = cxt->pseen;
3471#else
43d061fe 3472 HV *hseen = cxt->hseen;
ab923da1 3473#endif
7a6a85bf 3474
43d061fe 3475 TRACEME(("store (0x%"UVxf")", PTR2UV(sv)));
7a6a85bf
RG
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).
464b080a 3484 * That means it probably safe to assume it is well under the 32-bit limit,
7a6a85bf
RG
3485 * and makes the truncation safe.
3486 * -- RAM, 14/09/1999
3487 */
3488
ab923da1
NC
3489#ifdef USE_PTR_TABLE
3490 svh = ptr_table_fetch(pseen, sv);
3491#else
7a6a85bf 3492 svh = hv_fetch(hseen, (char *) &sv, sizeof(sv), FALSE);
ab923da1 3493#endif
7a6a85bf 3494 if (svh) {
dfd91409
NC
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
17625bd2 3508 SvREFCNT_dec()ed, and as Storable cheats horribly
dfd91409
NC
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
ab923da1
NC
3527#ifdef USE_PTR_TABLE
3528 tagval = htonl(LOW_32BITS(((char *)svh)-1));
3529#else
dfd91409 3530 tagval = htonl(LOW_32BITS(*svh));
ab923da1 3531#endif
7a6a85bf 3532
9e21b3d0 3533 TRACEME(("object 0x%"UVxf" seen as #%d", PTR2UV(sv), ntohl(tagval)));
7a6a85bf
RG
3534
3535 PUTMARK(SX_OBJECT);
9e21b3d0 3536 WRITE_I32(tagval);
7a6a85bf
RG
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
d1be9408 3545 * cast the tagnum to an SV pointer and store that in the hash. This
7a6a85bf
RG
3546 * means that we must clean up the hash manually afterwards, but gives
3547 * us a 15% throughput increase.
3548 *
7a6a85bf
RG
3549 */
3550
3551 cxt->tagnum++;
ab923da1
NC
3552#ifdef USE_PTR_TABLE
3553 ptr_table_store(pseen, sv, INT2PTR(SV*, 1 + cxt->tagnum));
3554#else
7a6a85bf 3555 if (!hv_store(hseen,
3341c981 3556 (char *) &sv, sizeof(sv), INT2PTR(SV*, cxt->tagnum), 0))
7a6a85bf 3557 return -1;
ab923da1 3558#endif
7a6a85bf
RG
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
138ec36d 3565 type = sv_type(aTHX_ sv);
7a6a85bf 3566
dfd91409 3567undef_special_case:
43d061fe
JH
3568 TRACEME(("storing 0x%"UVxf" tag #%d, type %d...",
3569 PTR2UV(sv), cxt->tagnum, type));
7a6a85bf
RG
3570
3571 if (SvOBJECT(sv)) {
3572 HV *pkg = SvSTASH(sv);
138ec36d 3573 ret = store_blessed(aTHX_ cxt, sv, type, pkg);
7a6a85bf 3574 } else
138ec36d 3575 ret = SV_STORE(type)(aTHX_ cxt, sv);
7a6a85bf 3576
43d061fe
JH
3577 TRACEME(("%s (stored 0x%"UVxf", refcnt=%d, %s)",
3578 ret ? "FAILED" : "ok", PTR2UV(sv),
7a6a85bf
RG
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 */
138ec36d 3595static int magic_write(pTHX_ stcxt_t *cxt)
7a6a85bf 3596{
2aeb6432
NC
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 */
7a6a85bf 3607
2aeb6432
NC
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 };
ee0f7aac
NC
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(ch