Commit | Line | Data |
---|---|---|
7a6a85bf RG |
1 | /* |
2 | * Store and retrieve mechanism. | |
3 | */ | |
4 | ||
5 | /* | |
6e0ac6f5 | 6 | * $Id: Storable.xs,v 1.0.1.10 2001/08/28 21:52:14 ram Exp $ |
7a6a85bf RG |
7 | * |
8 | * Copyright (c) 1995-2000, Raphael Manfredi | |
9 | * | |
9e21b3d0 JH |
10 | * You may redistribute only under the same terms as Perl 5, as specified |
11 | * in the README file that comes with the distribution. | |
7a6a85bf | 12 | * |
7a6a85bf RG |
13 | */ |
14 | ||
15 | #include <EXTERN.h> | |
16 | #include <perl.h> | |
e993d95c | 17 | #include <patchlevel.h> /* Perl's one, needed since 5.6 */ |
7a6a85bf RG |
18 | #include <XSUB.h> |
19 | ||
5dff01bc | 20 | #ifndef NETWARE |
e8189732 | 21 | #if 0 |
9e21b3d0 JH |
22 | #define DEBUGME /* Debug mode, turns assertions on as well */ |
23 | #define DASSERT /* Assertion mode */ | |
24 | #endif | |
5dff01bc JH |
25 | #else /* NETWARE */ |
26 | #if 0 /* On NetWare USE_PERLIO is not used */ | |
27 | #define DEBUGME /* Debug mode, turns assertions on as well */ | |
28 | #define DASSERT /* Assertion mode */ | |
29 | #endif | |
30 | #endif | |
7a6a85bf RG |
31 | |
32 | /* | |
33 | * Pre PerlIO time when none of USE_PERLIO and PERLIO_IS_STDIO is defined | |
34 | * Provide them with the necessary defines so they can build with pre-5.004. | |
35 | */ | |
36 | #ifndef USE_PERLIO | |
37 | #ifndef PERLIO_IS_STDIO | |
38 | #define PerlIO FILE | |
39 | #define PerlIO_getc(x) getc(x) | |
40 | #define PerlIO_putc(f,x) putc(x,f) | |
41 | #define PerlIO_read(x,y,z) fread(y,1,z,x) | |
42 | #define PerlIO_write(x,y,z) fwrite(y,1,z,x) | |
43 | #define PerlIO_stdoutf printf | |
44 | #endif /* PERLIO_IS_STDIO */ | |
45 | #endif /* USE_PERLIO */ | |
46 | ||
47 | /* | |
48 | * Earlier versions of perl might be used, we can't assume they have the latest! | |
49 | */ | |
f0ffaed8 JH |
50 | |
51 | #ifndef PERL_VERSION /* For perls < 5.6 */ | |
e993d95c | 52 | #define PERL_VERSION PATCHLEVEL |
7a6a85bf RG |
53 | #ifndef newRV_noinc |
54 | #define newRV_noinc(sv) ((Sv = newRV(sv)), --SvREFCNT(SvRV(Sv)), Sv) | |
55 | #endif | |
e993d95c | 56 | #if (PATCHLEVEL <= 4) /* Older perls (<= 5.004) lack PL_ namespace */ |
7a6a85bf RG |
57 | #define PL_sv_yes sv_yes |
58 | #define PL_sv_no sv_no | |
59 | #define PL_sv_undef sv_undef | |
e993d95c | 60 | #if (SUBVERSION <= 4) /* 5.004_04 has been reported to lack newSVpvn */ |
dd19458b | 61 | #define newSVpvn newSVpv |
7a6a85bf | 62 | #endif |
e993d95c | 63 | #endif /* PATCHLEVEL <= 4 */ |
7a6a85bf RG |
64 | #ifndef HvSHAREKEYS_off |
65 | #define HvSHAREKEYS_off(hv) /* Ignore */ | |
66 | #endif | |
f0ffaed8 JH |
67 | #ifndef AvFILLp /* Older perls (<=5.003) lack AvFILLp */ |
68 | #define AvFILLp AvFILL | |
69 | #endif | |
70 | typedef double NV; /* Older perls lack the NV type */ | |
cc964657 JH |
71 | #define IVdf "ld" /* Various printf formats for Perl types */ |
72 | #define UVuf "lu" | |
73 | #define UVof "lo" | |
74 | #define UVxf "lx" | |
75 | #define INT2PTR(t,v) (t)(IV)(v) | |
76 | #define PTR2UV(v) (unsigned long)(v) | |
f0ffaed8 | 77 | #endif /* PERL_VERSION -- perls < 5.6 */ |
7a6a85bf | 78 | |
cc964657 | 79 | #ifndef NVef /* The following were not part of perl 5.6 */ |
9e21b3d0 JH |
80 | #if defined(USE_LONG_DOUBLE) && \ |
81 | defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl) | |
82 | #define NVef PERL_PRIeldbl | |
83 | #define NVff PERL_PRIfldbl | |
84 | #define NVgf PERL_PRIgldbl | |
85 | #else | |
cc964657 JH |
86 | #define NVef "e" |
87 | #define NVff "f" | |
88 | #define NVgf "g" | |
89 | #endif | |
90 | #endif | |
91 | ||
7a6a85bf | 92 | #ifdef DEBUGME |
8be2b38b JH |
93 | |
94 | #ifndef DASSERT | |
95 | #define DASSERT | |
96 | #endif | |
97 | ||
90826881 JH |
98 | /* |
99 | * TRACEME() will only output things when the $Storable::DEBUGME is true. | |
100 | */ | |
101 | ||
111e03c1 RG |
102 | #define TRACEME(x) \ |
103 | STMT_START { \ | |
90826881 | 104 | if (SvTRUE(perl_get_sv("Storable::DEBUGME", TRUE))) \ |
111e03c1 RG |
105 | { PerlIO_stdoutf x; PerlIO_stdoutf("\n"); } \ |
106 | } STMT_END | |
7a6a85bf RG |
107 | #else |
108 | #define TRACEME(x) | |
8be2b38b | 109 | #endif /* DEBUGME */ |
7a6a85bf RG |
110 | |
111 | #ifdef DASSERT | |
111e03c1 RG |
112 | #define ASSERT(x,y) \ |
113 | STMT_START { \ | |
7a6a85bf RG |
114 | if (!(x)) { \ |
115 | PerlIO_stdoutf("ASSERT FAILED (\"%s\", line %d): ", \ | |
116 | __FILE__, __LINE__); \ | |
117 | PerlIO_stdoutf y; PerlIO_stdoutf("\n"); \ | |
118 | } \ | |
111e03c1 | 119 | } STMT_END |
7a6a85bf RG |
120 | #else |
121 | #define ASSERT(x,y) | |
122 | #endif | |
123 | ||
124 | /* | |
125 | * Type markers. | |
126 | */ | |
127 | ||
128 | #define C(x) ((char) (x)) /* For markers with dynamic retrieval handling */ | |
129 | ||
130 | #define SX_OBJECT C(0) /* Already stored object */ | |
dd19458b | 131 | #define SX_LSCALAR C(1) /* Scalar (large binary) follows (length, data) */ |
7a6a85bf RG |
132 | #define SX_ARRAY C(2) /* Array forthcominng (size, item list) */ |
133 | #define SX_HASH C(3) /* Hash forthcoming (size, key/value pair list) */ | |
134 | #define SX_REF C(4) /* Reference to object forthcoming */ | |
135 | #define SX_UNDEF C(5) /* Undefined scalar */ | |
136 | #define SX_INTEGER C(6) /* Integer forthcoming */ | |
137 | #define SX_DOUBLE C(7) /* Double forthcoming */ | |
138 | #define SX_BYTE C(8) /* (signed) byte forthcoming */ | |
139 | #define SX_NETINT C(9) /* Integer in network order forthcoming */ | |
dd19458b | 140 | #define SX_SCALAR C(10) /* Scalar (binary, small) follows (length, data) */ |
f062ea6c PN |
141 | #define SX_TIED_ARRAY C(11) /* Tied array forthcoming */ |
142 | #define SX_TIED_HASH C(12) /* Tied hash forthcoming */ | |
143 | #define SX_TIED_SCALAR C(13) /* Tied scalar forthcoming */ | |
7a6a85bf RG |
144 | #define SX_SV_UNDEF C(14) /* Perl's immortal PL_sv_undef */ |
145 | #define SX_SV_YES C(15) /* Perl's immortal PL_sv_yes */ | |
146 | #define SX_SV_NO C(16) /* Perl's immortal PL_sv_no */ | |
147 | #define SX_BLESS C(17) /* Object is blessed */ | |
148 | #define SX_IX_BLESS C(18) /* Object is blessed, classname given by index */ | |
149 | #define SX_HOOK C(19) /* Stored via hook, user-defined */ | |
150 | #define SX_OVERLOAD C(20) /* Overloaded reference */ | |
f062ea6c PN |
151 | #define SX_TIED_KEY C(21) /* Tied magic key forthcoming */ |
152 | #define SX_TIED_IDX C(22) /* Tied magic index forthcoming */ | |
153 | #define SX_UTF8STR C(23) /* UTF-8 string forthcoming (small) */ | |
154 | #define SX_LUTF8STR C(24) /* UTF-8 string forthcoming (large) */ | |
155 | #define SX_FLAG_HASH C(25) /* Hash with flags forthcoming (size, flags, key/flags/value triplet list) */ | |
e16e2ff8 | 156 | #define SX_ERROR C(26) /* Error */ |
7a6a85bf RG |
157 | |
158 | /* | |
159 | * Those are only used to retrieve "old" pre-0.6 binary images. | |
160 | */ | |
161 | #define SX_ITEM 'i' /* An array item introducer */ | |
162 | #define SX_IT_UNDEF 'I' /* Undefined array item */ | |
d1be9408 JF |
163 | #define SX_KEY 'k' /* A hash key introducer */ |
164 | #define SX_VALUE 'v' /* A hash value introducer */ | |
7a6a85bf RG |
165 | #define SX_VL_UNDEF 'V' /* Undefined hash value */ |
166 | ||
167 | /* | |
168 | * Those are only used to retrieve "old" pre-0.7 binary images | |
169 | */ | |
170 | ||
171 | #define SX_CLASS 'b' /* Object is blessed, class name length <255 */ | |
f062ea6c | 172 | #define SX_LG_CLASS 'B' /* Object is blessed, class name length >255 */ |
7a6a85bf RG |
173 | #define SX_STORED 'X' /* End of object */ |
174 | ||
175 | /* | |
176 | * Limits between short/long length representation. | |
177 | */ | |
178 | ||
179 | #define LG_SCALAR 255 /* Large scalar length limit */ | |
180 | #define LG_BLESS 127 /* Large classname bless limit */ | |
181 | ||
182 | /* | |
183 | * Operation types | |
184 | */ | |
185 | ||
186 | #define ST_STORE 0x1 /* Store operation */ | |
187 | #define ST_RETRIEVE 0x2 /* Retrieval operation */ | |
188 | #define ST_CLONE 0x4 /* Deep cloning operation */ | |
189 | ||
190 | /* | |
191 | * The following structure is used for hash table key retrieval. Since, when | |
192 | * retrieving objects, we'll be facing blessed hash references, it's best | |
193 | * to pre-allocate that buffer once and resize it as the need arises, never | |
194 | * freeing it (keys will be saved away someplace else anyway, so even large | |
195 | * keys are not enough a motivation to reclaim that space). | |
196 | * | |
197 | * This structure is also used for memory store/retrieve operations which | |
198 | * happen in a fixed place before being malloc'ed elsewhere if persistency | |
199 | * is required. Hence the aptr pointer. | |
200 | */ | |
201 | struct extendable { | |
202 | char *arena; /* Will hold hash key strings, resized as needed */ | |
203 | STRLEN asiz; /* Size of aforementionned buffer */ | |
204 | char *aptr; /* Arena pointer, for in-place read/write ops */ | |
205 | char *aend; /* First invalid address */ | |
206 | }; | |
207 | ||
208 | /* | |
209 | * At store time: | |
d1be9408 | 210 | * A hash table records the objects which have already been stored. |
7a6a85bf RG |
211 | * Those are referred to as SX_OBJECT in the file, and their "tag" (i.e. |
212 | * an arbitrary sequence number) is used to identify them. | |
213 | * | |
214 | * At retrieve time: | |
215 | * An array table records the objects which have already been retrieved, | |
216 | * as seen by the tag determind by counting the objects themselves. The | |
217 | * reference to that retrieved object is kept in the table, and is returned | |
218 | * when an SX_OBJECT is found bearing that same tag. | |
219 | * | |
220 | * The same processing is used to record "classname" for blessed objects: | |
221 | * indexing by a hash at store time, and via an array at retrieve time. | |
222 | */ | |
223 | ||
224 | typedef unsigned long stag_t; /* Used by pre-0.6 binary format */ | |
225 | ||
226 | /* | |
227 | * The following "thread-safe" related defines were contributed by | |
228 | * Murray Nesbitt <murray@activestate.com> and integrated by RAM, who | |
229 | * only renamed things a little bit to ensure consistency with surrounding | |
230 | * code. -- RAM, 14/09/1999 | |
231 | * | |
232 | * The original patch suffered from the fact that the stcxt_t structure | |
233 | * was global. Murray tried to minimize the impact on the code as much as | |
234 | * possible. | |
235 | * | |
236 | * Starting with 0.7, Storable can be re-entrant, via the STORABLE_xxx hooks | |
237 | * on objects. Therefore, the notion of context needs to be generalized, | |
238 | * threading or not. | |
239 | */ | |
240 | ||
241 | #define MY_VERSION "Storable(" XS_VERSION ")" | |
242 | ||
530b72ba NC |
243 | |
244 | /* | |
245 | * Conditional UTF8 support. | |
246 | * | |
247 | */ | |
248 | #ifdef SvUTF8_on | |
249 | #define STORE_UTF8STR(pv, len) STORE_PV_LEN(pv, len, SX_UTF8STR, SX_LUTF8STR) | |
250 | #define HAS_UTF8_SCALARS | |
251 | #ifdef HeKUTF8 | |
252 | #define HAS_UTF8_HASHES | |
253 | #define HAS_UTF8_ALL | |
254 | #else | |
255 | /* 5.6 perl has utf8 scalars but not hashes */ | |
256 | #endif | |
257 | #else | |
258 | #define SvUTF8(sv) 0 | |
259 | #define STORE_UTF8STR(pv, len) CROAK(("panic: storing UTF8 in non-UTF8 perl")) | |
260 | #endif | |
261 | #ifndef HAS_UTF8_ALL | |
262 | #define UTF8_CROAK() CROAK(("Cannot retrieve UTF8 data in non-UTF8 perl")) | |
263 | #endif | |
264 | ||
265 | #ifdef HvPLACEHOLDERS | |
266 | #define HAS_RESTRICTED_HASHES | |
267 | #else | |
268 | #define HVhek_PLACEHOLD 0x200 | |
269 | #define RESTRICTED_HASH_CROAK() CROAK(("Cannot retrieve restricted hash")) | |
270 | #endif | |
271 | ||
272 | #ifdef HvHASKFLAGS | |
273 | #define HAS_HASH_KEY_FLAGS | |
274 | #endif | |
275 | ||
dd19458b JH |
276 | /* |
277 | * Fields s_tainted and s_dirty are prefixed with s_ because Perl's include | |
278 | * files remap tainted and dirty when threading is enabled. That's bad for | |
279 | * perl to remap such common words. -- RAM, 29/09/00 | |
280 | */ | |
281 | ||
7a6a85bf RG |
282 | typedef struct stcxt { |
283 | int entry; /* flags recursion */ | |
284 | int optype; /* type of traversal operation */ | |
e993d95c JH |
285 | HV *hseen; /* which objects have been seen, store time */ |
286 | AV *hook_seen; /* which SVs were returned by STORABLE_freeze() */ | |
287 | AV *aseen; /* which objects have been seen, retrieve time */ | |
288 | HV *hclass; /* which classnames have been seen, store time */ | |
289 | AV *aclass; /* which classnames have been seen, retrieve time */ | |
290 | HV *hook; /* cache for hook methods per class name */ | |
291 | IV tagnum; /* incremented at store time for each seen object */ | |
292 | IV classnum; /* incremented at store time for each seen classname */ | |
293 | int netorder; /* true if network order used */ | |
294 | int s_tainted; /* true if input source is tainted, at retrieve time */ | |
295 | int forgive_me; /* whether to be forgiving... */ | |
296 | int canonical; /* whether to store hashes sorted by key */ | |
530b72ba NC |
297 | #ifndef HAS_RESTRICTED_HASHES |
298 | int derestrict; /* whether to downgrade restrcted hashes */ | |
299 | #endif | |
300 | #ifndef HAS_UTF8_ALL | |
301 | int use_bytes; /* whether to bytes-ify utf8 */ | |
302 | #endif | |
e8189732 | 303 | int accept_future_minor; /* croak immediately on future minor versions? */ |
dd19458b | 304 | int s_dirty; /* context is dirty due to CROAK() -- can be cleaned */ |
e993d95c JH |
305 | int membuf_ro; /* true means membuf is read-only and msaved is rw */ |
306 | struct extendable keybuf; /* for hash key retrieval */ | |
307 | struct extendable membuf; /* for memory store/retrieve operations */ | |
308 | struct extendable msaved; /* where potentially valid mbuf is saved */ | |
7a6a85bf RG |
309 | PerlIO *fio; /* where I/O are performed, NULL for memory */ |
310 | int ver_major; /* major of version for retrieved object */ | |
311 | int ver_minor; /* minor of version for retrieved object */ | |
312 | SV *(**retrieve_vtbl)(); /* retrieve dispatch table */ | |
111e03c1 RG |
313 | SV *prev; /* contexts chained backwards in real recursion */ |
314 | SV *my_sv; /* the blessed scalar who's SvPVX() I am */ | |
7a6a85bf RG |
315 | } stcxt_t; |
316 | ||
111e03c1 RG |
317 | #define NEW_STORABLE_CXT_OBJ(cxt) \ |
318 | STMT_START { \ | |
319 | SV *self = newSV(sizeof(stcxt_t) - 1); \ | |
320 | SV *my_sv = newRV_noinc(self); \ | |
321 | sv_bless(my_sv, gv_stashpv("Storable::Cxt", TRUE)); \ | |
322 | cxt = (stcxt_t *)SvPVX(self); \ | |
323 | Zero(cxt, 1, stcxt_t); \ | |
324 | cxt->my_sv = my_sv; \ | |
325 | } STMT_END | |
326 | ||
7a6a85bf RG |
327 | #if defined(MULTIPLICITY) || defined(PERL_OBJECT) || defined(PERL_CAPI) |
328 | ||
e993d95c | 329 | #if (PATCHLEVEL <= 4) && (SUBVERSION < 68) |
7a6a85bf RG |
330 | #define dSTCXT_SV \ |
331 | SV *perinterp_sv = perl_get_sv(MY_VERSION, FALSE) | |
332 | #else /* >= perl5.004_68 */ | |
333 | #define dSTCXT_SV \ | |
334 | SV *perinterp_sv = *hv_fetch(PL_modglobal, \ | |
335 | MY_VERSION, sizeof(MY_VERSION)-1, TRUE) | |
336 | #endif /* < perl5.004_68 */ | |
337 | ||
338 | #define dSTCXT_PTR(T,name) \ | |
111e03c1 | 339 | T name = ((perinterp_sv && SvIOK(perinterp_sv) && SvIVX(perinterp_sv) \ |
436c6dd3 | 340 | ? (T)SvPVX(SvRV(INT2PTR(SV*,SvIVX(perinterp_sv)))) : (T) 0)) |
7a6a85bf RG |
341 | #define dSTCXT \ |
342 | dSTCXT_SV; \ | |
343 | dSTCXT_PTR(stcxt_t *, cxt) | |
344 | ||
111e03c1 RG |
345 | #define INIT_STCXT \ |
346 | dSTCXT; \ | |
347 | NEW_STORABLE_CXT_OBJ(cxt); \ | |
348 | sv_setiv(perinterp_sv, PTR2IV(cxt->my_sv)) | |
7a6a85bf | 349 | |
111e03c1 RG |
350 | #define SET_STCXT(x) \ |
351 | STMT_START { \ | |
7a6a85bf | 352 | dSTCXT_SV; \ |
111e03c1 RG |
353 | sv_setiv(perinterp_sv, PTR2IV(x->my_sv)); \ |
354 | } STMT_END | |
7a6a85bf RG |
355 | |
356 | #else /* !MULTIPLICITY && !PERL_OBJECT && !PERL_CAPI */ | |
357 | ||
358 | static stcxt_t Context; | |
359 | static stcxt_t *Context_ptr = &Context; | |
360 | #define dSTCXT stcxt_t *cxt = Context_ptr | |
111e03c1 RG |
361 | #define INIT_STCXT \ |
362 | dSTCXT; \ | |
363 | NEW_STORABLE_CXT_OBJ(cxt) | |
364 | ||
365 | #define SET_STCXT(x) Context_ptr = x | |
7a6a85bf RG |
366 | |
367 | #endif /* MULTIPLICITY || PERL_OBJECT || PERL_CAPI */ | |
368 | ||
369 | /* | |
370 | * KNOWN BUG: | |
371 | * Croaking implies a memory leak, since we don't use setjmp/longjmp | |
372 | * to catch the exit and free memory used during store or retrieve | |
373 | * operations. This is not too difficult to fix, but I need to understand | |
374 | * how Perl does it, and croaking is exceptional anyway, so I lack the | |
375 | * motivation to do it. | |
376 | * | |
377 | * The current workaround is to mark the context as dirty when croaking, | |
378 | * so that data structures can be freed whenever we renter Storable code | |
379 | * (but only *then*: it's a workaround, not a fix). | |
380 | * | |
381 | * This is also imperfect, because we don't really know how far they trapped | |
382 | * the croak(), and when we were recursing, we won't be able to clean anything | |
383 | * but the topmost context stacked. | |
384 | */ | |
385 | ||
111e03c1 | 386 | #define CROAK(x) STMT_START { cxt->s_dirty = 1; croak x; } STMT_END |
7a6a85bf RG |
387 | |
388 | /* | |
389 | * End of "thread-safe" related definitions. | |
390 | */ | |
391 | ||
392 | /* | |
9e21b3d0 JH |
393 | * LOW_32BITS |
394 | * | |
395 | * Keep only the low 32 bits of a pointer (used for tags, which are not | |
396 | * really pointers). | |
397 | */ | |
398 | ||
399 | #if PTRSIZE <= 4 | |
400 | #define LOW_32BITS(x) ((I32) (x)) | |
401 | #else | |
402 | #define LOW_32BITS(x) ((I32) ((unsigned long) (x) & 0xffffffffUL)) | |
403 | #endif | |
404 | ||
405 | /* | |
406 | * oI, oS, oC | |
407 | * | |
408 | * Hack for Crays, where sizeof(I32) == 8, and which are big-endians. | |
409 | * Used in the WLEN and RLEN macros. | |
410 | */ | |
411 | ||
412 | #if INTSIZE > 4 | |
413 | #define oI(x) ((I32 *) ((char *) (x) + 4)) | |
414 | #define oS(x) ((x) - 4) | |
415 | #define oC(x) (x = 0) | |
416 | #define CRAY_HACK | |
417 | #else | |
418 | #define oI(x) (x) | |
419 | #define oS(x) (x) | |
420 | #define oC(x) | |
421 | #endif | |
422 | ||
423 | /* | |
7a6a85bf RG |
424 | * key buffer handling |
425 | */ | |
426 | #define kbuf (cxt->keybuf).arena | |
427 | #define ksiz (cxt->keybuf).asiz | |
111e03c1 RG |
428 | #define KBUFINIT() \ |
429 | STMT_START { \ | |
7a6a85bf RG |
430 | if (!kbuf) { \ |
431 | TRACEME(("** allocating kbuf of 128 bytes")); \ | |
432 | New(10003, kbuf, 128, char); \ | |
433 | ksiz = 128; \ | |
434 | } \ | |
111e03c1 RG |
435 | } STMT_END |
436 | #define KBUFCHK(x) \ | |
437 | STMT_START { \ | |
7a6a85bf | 438 | if (x >= ksiz) { \ |
e993d95c | 439 | TRACEME(("** extending kbuf to %d bytes (had %d)", x+1, ksiz)); \ |
7a6a85bf RG |
440 | Renew(kbuf, x+1, char); \ |
441 | ksiz = x+1; \ | |
442 | } \ | |
111e03c1 | 443 | } STMT_END |
7a6a85bf RG |
444 | |
445 | /* | |
446 | * memory buffer handling | |
447 | */ | |
448 | #define mbase (cxt->membuf).arena | |
449 | #define msiz (cxt->membuf).asiz | |
450 | #define mptr (cxt->membuf).aptr | |
451 | #define mend (cxt->membuf).aend | |
452 | ||
453 | #define MGROW (1 << 13) | |
454 | #define MMASK (MGROW - 1) | |
455 | ||
456 | #define round_mgrow(x) \ | |
457 | ((unsigned long) (((unsigned long) (x) + MMASK) & ~MMASK)) | |
458 | #define trunc_int(x) \ | |
459 | ((unsigned long) ((unsigned long) (x) & ~(sizeof(int)-1))) | |
460 | #define int_aligned(x) \ | |
461 | ((unsigned long) (x) == trunc_int(x)) | |
462 | ||
111e03c1 RG |
463 | #define MBUF_INIT(x) \ |
464 | STMT_START { \ | |
7a6a85bf RG |
465 | if (!mbase) { \ |
466 | TRACEME(("** allocating mbase of %d bytes", MGROW)); \ | |
467 | New(10003, mbase, MGROW, char); \ | |
468 | msiz = MGROW; \ | |
469 | } \ | |
470 | mptr = mbase; \ | |
471 | if (x) \ | |
472 | mend = mbase + x; \ | |
473 | else \ | |
474 | mend = mbase + msiz; \ | |
111e03c1 | 475 | } STMT_END |
7a6a85bf RG |
476 | |
477 | #define MBUF_TRUNC(x) mptr = mbase + x | |
478 | #define MBUF_SIZE() (mptr - mbase) | |
479 | ||
480 | /* | |
e993d95c JH |
481 | * MBUF_SAVE_AND_LOAD |
482 | * MBUF_RESTORE | |
483 | * | |
484 | * Those macros are used in do_retrieve() to save the current memory | |
485 | * buffer into cxt->msaved, before MBUF_LOAD() can be used to retrieve | |
486 | * data from a string. | |
487 | */ | |
111e03c1 RG |
488 | #define MBUF_SAVE_AND_LOAD(in) \ |
489 | STMT_START { \ | |
e993d95c JH |
490 | ASSERT(!cxt->membuf_ro, ("mbase not already saved")); \ |
491 | cxt->membuf_ro = 1; \ | |
492 | TRACEME(("saving mbuf")); \ | |
493 | StructCopy(&cxt->membuf, &cxt->msaved, struct extendable); \ | |
494 | MBUF_LOAD(in); \ | |
111e03c1 | 495 | } STMT_END |
e993d95c | 496 | |
111e03c1 RG |
497 | #define MBUF_RESTORE() \ |
498 | STMT_START { \ | |
e993d95c JH |
499 | ASSERT(cxt->membuf_ro, ("mbase is read-only")); \ |
500 | cxt->membuf_ro = 0; \ | |
501 | TRACEME(("restoring mbuf")); \ | |
502 | StructCopy(&cxt->msaved, &cxt->membuf, struct extendable); \ | |
111e03c1 | 503 | } STMT_END |
e993d95c JH |
504 | |
505 | /* | |
7a6a85bf RG |
506 | * Use SvPOKp(), because SvPOK() fails on tainted scalars. |
507 | * See store_scalar() for other usage of this workaround. | |
508 | */ | |
111e03c1 RG |
509 | #define MBUF_LOAD(v) \ |
510 | STMT_START { \ | |
e993d95c | 511 | ASSERT(cxt->membuf_ro, ("mbase is read-only")); \ |
7a6a85bf RG |
512 | if (!SvPOKp(v)) \ |
513 | CROAK(("Not a scalar string")); \ | |
514 | mptr = mbase = SvPV(v, msiz); \ | |
515 | mend = mbase + msiz; \ | |
111e03c1 | 516 | } STMT_END |
7a6a85bf | 517 | |
111e03c1 RG |
518 | #define MBUF_XTEND(x) \ |
519 | STMT_START { \ | |
7a6a85bf RG |
520 | int nsz = (int) round_mgrow((x)+msiz); \ |
521 | int offset = mptr - mbase; \ | |
e993d95c JH |
522 | ASSERT(!cxt->membuf_ro, ("mbase is not read-only")); \ |
523 | TRACEME(("** extending mbase from %d to %d bytes (wants %d new)", \ | |
524 | msiz, nsz, (x))); \ | |
7a6a85bf RG |
525 | Renew(mbase, nsz, char); \ |
526 | msiz = nsz; \ | |
527 | mptr = mbase + offset; \ | |
528 | mend = mbase + nsz; \ | |
111e03c1 | 529 | } STMT_END |
7a6a85bf | 530 | |
111e03c1 RG |
531 | #define MBUF_CHK(x) \ |
532 | STMT_START { \ | |
7a6a85bf RG |
533 | if ((mptr + (x)) > mend) \ |
534 | MBUF_XTEND(x); \ | |
111e03c1 | 535 | } STMT_END |
7a6a85bf | 536 | |
111e03c1 RG |
537 | #define MBUF_GETC(x) \ |
538 | STMT_START { \ | |
7a6a85bf RG |
539 | if (mptr < mend) \ |
540 | x = (int) (unsigned char) *mptr++; \ | |
541 | else \ | |
542 | return (SV *) 0; \ | |
111e03c1 | 543 | } STMT_END |
7a6a85bf | 544 | |
9e21b3d0 | 545 | #ifdef CRAY_HACK |
111e03c1 RG |
546 | #define MBUF_GETINT(x) \ |
547 | STMT_START { \ | |
9e21b3d0 JH |
548 | oC(x); \ |
549 | if ((mptr + 4) <= mend) { \ | |
550 | memcpy(oI(&x), mptr, 4); \ | |
551 | mptr += 4; \ | |
552 | } else \ | |
553 | return (SV *) 0; \ | |
111e03c1 | 554 | } STMT_END |
9e21b3d0 | 555 | #else |
111e03c1 RG |
556 | #define MBUF_GETINT(x) \ |
557 | STMT_START { \ | |
7a6a85bf RG |
558 | if ((mptr + sizeof(int)) <= mend) { \ |
559 | if (int_aligned(mptr)) \ | |
560 | x = *(int *) mptr; \ | |
561 | else \ | |
562 | memcpy(&x, mptr, sizeof(int)); \ | |
563 | mptr += sizeof(int); \ | |
564 | } else \ | |
565 | return (SV *) 0; \ | |
111e03c1 | 566 | } STMT_END |
9e21b3d0 | 567 | #endif |
7a6a85bf | 568 | |
111e03c1 RG |
569 | #define MBUF_READ(x,s) \ |
570 | STMT_START { \ | |
7a6a85bf RG |
571 | if ((mptr + (s)) <= mend) { \ |
572 | memcpy(x, mptr, s); \ | |
573 | mptr += s; \ | |
574 | } else \ | |
575 | return (SV *) 0; \ | |
111e03c1 | 576 | } STMT_END |
7a6a85bf | 577 | |
111e03c1 RG |
578 | #define MBUF_SAFEREAD(x,s,z) \ |
579 | STMT_START { \ | |
7a6a85bf RG |
580 | if ((mptr + (s)) <= mend) { \ |
581 | memcpy(x, mptr, s); \ | |
582 | mptr += s; \ | |
583 | } else { \ | |
584 | sv_free(z); \ | |
585 | return (SV *) 0; \ | |
586 | } \ | |
111e03c1 | 587 | } STMT_END |
7a6a85bf | 588 | |
111e03c1 RG |
589 | #define MBUF_PUTC(c) \ |
590 | STMT_START { \ | |
7a6a85bf RG |
591 | if (mptr < mend) \ |
592 | *mptr++ = (char) c; \ | |
593 | else { \ | |
594 | MBUF_XTEND(1); \ | |
595 | *mptr++ = (char) c; \ | |
596 | } \ | |
111e03c1 | 597 | } STMT_END |
7a6a85bf | 598 | |
9e21b3d0 | 599 | #ifdef CRAY_HACK |
111e03c1 RG |
600 | #define MBUF_PUTINT(i) \ |
601 | STMT_START { \ | |
9e21b3d0 JH |
602 | MBUF_CHK(4); \ |
603 | memcpy(mptr, oI(&i), 4); \ | |
604 | mptr += 4; \ | |
111e03c1 | 605 | } STMT_END |
9e21b3d0 | 606 | #else |
111e03c1 RG |
607 | #define MBUF_PUTINT(i) \ |
608 | STMT_START { \ | |
7a6a85bf RG |
609 | MBUF_CHK(sizeof(int)); \ |
610 | if (int_aligned(mptr)) \ | |
611 | *(int *) mptr = i; \ | |
612 | else \ | |
613 | memcpy(mptr, &i, sizeof(int)); \ | |
614 | mptr += sizeof(int); \ | |
111e03c1 | 615 | } STMT_END |
9e21b3d0 | 616 | #endif |
7a6a85bf | 617 | |
111e03c1 RG |
618 | #define MBUF_WRITE(x,s) \ |
619 | STMT_START { \ | |
7a6a85bf RG |
620 | MBUF_CHK(s); \ |
621 | memcpy(mptr, x, s); \ | |
622 | mptr += s; \ | |
111e03c1 | 623 | } STMT_END |
7a6a85bf RG |
624 | |
625 | /* | |
7a6a85bf RG |
626 | * Possible return values for sv_type(). |
627 | */ | |
628 | ||
629 | #define svis_REF 0 | |
630 | #define svis_SCALAR 1 | |
631 | #define svis_ARRAY 2 | |
632 | #define svis_HASH 3 | |
633 | #define svis_TIED 4 | |
634 | #define svis_TIED_ITEM 5 | |
635 | #define svis_OTHER 6 | |
636 | ||
637 | /* | |
638 | * Flags for SX_HOOK. | |
639 | */ | |
640 | ||
641 | #define SHF_TYPE_MASK 0x03 | |
642 | #define SHF_LARGE_CLASSLEN 0x04 | |
643 | #define SHF_LARGE_STRLEN 0x08 | |
644 | #define SHF_LARGE_LISTLEN 0x10 | |
645 | #define SHF_IDX_CLASSNAME 0x20 | |
646 | #define SHF_NEED_RECURSE 0x40 | |
647 | #define SHF_HAS_LIST 0x80 | |
648 | ||
649 | /* | |
b12202d0 | 650 | * Types for SX_HOOK (last 2 bits in flags). |
7a6a85bf RG |
651 | */ |
652 | ||
653 | #define SHT_SCALAR 0 | |
654 | #define SHT_ARRAY 1 | |
655 | #define SHT_HASH 2 | |
b12202d0 JH |
656 | #define SHT_EXTRA 3 /* Read extra byte for type */ |
657 | ||
658 | /* | |
659 | * The following are held in the "extra byte"... | |
660 | */ | |
661 | ||
662 | #define SHT_TSCALAR 4 /* 4 + 0 -- tied scalar */ | |
663 | #define SHT_TARRAY 5 /* 4 + 1 -- tied array */ | |
664 | #define SHT_THASH 6 /* 4 + 2 -- tied hash */ | |
7a6a85bf RG |
665 | |
666 | /* | |
e16e2ff8 NC |
667 | * per hash flags for flagged hashes |
668 | */ | |
669 | ||
670 | #define SHV_RESTRICTED 0x01 | |
671 | ||
672 | /* | |
673 | * per key flags for flagged hashes | |
674 | */ | |
675 | ||
676 | #define SHV_K_UTF8 0x01 | |
677 | #define SHV_K_WASUTF8 0x02 | |
678 | #define SHV_K_LOCKED 0x04 | |
679 | #define SHV_K_ISSV 0x08 | |
680 | #define SHV_K_PLACEHOLDER 0x10 | |
681 | ||
682 | /* | |
7a6a85bf RG |
683 | * Before 0.6, the magic string was "perl-store" (binary version number 0). |
684 | * | |
685 | * Since 0.6 introduced many binary incompatibilities, the magic string has | |
686 | * been changed to "pst0" to allow an old image to be properly retrieved by | |
687 | * a newer Storable, but ensure a newer image cannot be retrieved with an | |
688 | * older version. | |
689 | * | |
690 | * At 0.7, objects are given the ability to serialize themselves, and the | |
691 | * set of markers is extended, backward compatibility is not jeopardized, | |
692 | * so the binary version number could have remained unchanged. To correctly | |
693 | * spot errors if a file making use of 0.7-specific extensions is given to | |
694 | * 0.6 for retrieval, the binary version was moved to "2". And I'm introducing | |
695 | * a "minor" version, to better track this kind of evolution from now on. | |
696 | * | |
697 | */ | |
2aeb6432 NC |
698 | static const char old_magicstr[] = "perl-store"; /* Magic number before 0.6 */ |
699 | static const char magicstr[] = "pst0"; /* Used as a magic number */ | |
7a6a85bf | 700 | |
2aeb6432 NC |
701 | #define MAGICSTR_BYTES 'p','s','t','0' |
702 | #define OLDMAGICSTR_BYTES 'p','e','r','l','-','s','t','o','r','e' | |
703 | ||
704 | #if BYTEORDER == 0x1234 | |
705 | #define BYTEORDER_BYTES '1','2','3','4' | |
706 | #else | |
707 | #if BYTEORDER == 0x12345678 | |
708 | #define BYTEORDER_BYTES '1','2','3','4','5','6','7','8' | |
709 | #else | |
710 | #if BYTEORDER == 0x87654321 | |
711 | #define BYTEORDER_BYTES '8','7','6','5','4','3','2','1' | |
712 | #else | |
713 | #if BYTEORDER == 0x4321 | |
714 | #define BYTEORDER_BYTES '4','3','2','1' | |
715 | #else | |
716 | #error Unknown byteoder. Please append your byteorder to Storable.xs | |
717 | #endif | |
718 | #endif | |
719 | #endif | |
720 | #endif | |
721 | ||
722 | static const char byteorderstr[] = {BYTEORDER_BYTES, 0}; | |
530b72ba | 723 | |
e16e2ff8 | 724 | #define STORABLE_BIN_MAJOR 2 /* Binary major "version" */ |
530b72ba NC |
725 | #define STORABLE_BIN_MINOR 5 /* Binary minor "version" */ |
726 | ||
727 | /* If we aren't 5.7.3 or later, we won't be writing out files that use the | |
728 | * new flagged hash introdued in 2.5, so put 2.4 in the binary header to | |
729 | * maximise ease of interoperation with older Storables. | |
730 | * Could we write 2.3s if we're on 5.005_03? NWC | |
731 | */ | |
e16e2ff8 | 732 | #if (PATCHLEVEL <= 6) |
530b72ba | 733 | #define STORABLE_BIN_WRITE_MINOR 4 |
e16e2ff8 NC |
734 | #else |
735 | /* | |
736 | * As of perl 5.7.3, utf8 hash key is introduced. | |
737 | * So this must change -- dankogai | |
738 | */ | |
530b72ba | 739 | #define STORABLE_BIN_WRITE_MINOR 5 |
e16e2ff8 | 740 | #endif /* (PATCHLEVEL <= 6) */ |
7a6a85bf RG |
741 | |
742 | /* | |
743 | * Useful store shortcuts... | |
744 | */ | |
745 | ||
111e03c1 RG |
746 | #define PUTMARK(x) \ |
747 | STMT_START { \ | |
7a6a85bf RG |
748 | if (!cxt->fio) \ |
749 | MBUF_PUTC(x); \ | |
750 | else if (PerlIO_putc(cxt->fio, x) == EOF) \ | |
751 | return -1; \ | |
111e03c1 | 752 | } STMT_END |
7a6a85bf | 753 | |
111e03c1 RG |
754 | #define WRITE_I32(x) \ |
755 | STMT_START { \ | |
9e21b3d0 JH |
756 | ASSERT(sizeof(x) == sizeof(I32), ("writing an I32")); \ |
757 | if (!cxt->fio) \ | |
758 | MBUF_PUTINT(x); \ | |
759 | else if (PerlIO_write(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \ | |
760 | return -1; \ | |
111e03c1 | 761 | } STMT_END |
9e21b3d0 | 762 | |
7a6a85bf | 763 | #ifdef HAS_HTONL |
111e03c1 RG |
764 | #define WLEN(x) \ |
765 | STMT_START { \ | |
7a6a85bf RG |
766 | if (cxt->netorder) { \ |
767 | int y = (int) htonl(x); \ | |
768 | if (!cxt->fio) \ | |
769 | MBUF_PUTINT(y); \ | |
9e21b3d0 | 770 | else if (PerlIO_write(cxt->fio,oI(&y),oS(sizeof(y))) != oS(sizeof(y))) \ |
7a6a85bf RG |
771 | return -1; \ |
772 | } else { \ | |
773 | if (!cxt->fio) \ | |
774 | MBUF_PUTINT(x); \ | |
9e21b3d0 | 775 | else if (PerlIO_write(cxt->fio,oI(&x),oS(sizeof(x))) != oS(sizeof(x))) \ |
7a6a85bf RG |
776 | return -1; \ |
777 | } \ | |
111e03c1 | 778 | } STMT_END |
7a6a85bf | 779 | #else |
9e21b3d0 | 780 | #define WLEN(x) WRITE_I32(x) |
7a6a85bf RG |
781 | #endif |
782 | ||
111e03c1 RG |
783 | #define WRITE(x,y) \ |
784 | STMT_START { \ | |
7a6a85bf RG |
785 | if (!cxt->fio) \ |
786 | MBUF_WRITE(x,y); \ | |
787 | else if (PerlIO_write(cxt->fio, x, y) != y) \ | |
788 | return -1; \ | |
111e03c1 | 789 | } STMT_END |
7a6a85bf | 790 | |
111e03c1 RG |
791 | #define STORE_PV_LEN(pv, len, small, large) \ |
792 | STMT_START { \ | |
7a6a85bf RG |
793 | if (len <= LG_SCALAR) { \ |
794 | unsigned char clen = (unsigned char) len; \ | |
dd19458b | 795 | PUTMARK(small); \ |
7a6a85bf RG |
796 | PUTMARK(clen); \ |
797 | if (len) \ | |
798 | WRITE(pv, len); \ | |
799 | } else { \ | |
dd19458b | 800 | PUTMARK(large); \ |
7a6a85bf RG |
801 | WLEN(len); \ |
802 | WRITE(pv, len); \ | |
803 | } \ | |
111e03c1 | 804 | } STMT_END |
7a6a85bf | 805 | |
dd19458b JH |
806 | #define STORE_SCALAR(pv, len) STORE_PV_LEN(pv, len, SX_SCALAR, SX_LSCALAR) |
807 | ||
808 | /* | |
7a6a85bf RG |
809 | * Store undef in arrays and hashes without recursing through store(). |
810 | */ | |
111e03c1 RG |
811 | #define STORE_UNDEF() \ |
812 | STMT_START { \ | |
7a6a85bf RG |
813 | cxt->tagnum++; \ |
814 | PUTMARK(SX_UNDEF); \ | |
111e03c1 | 815 | } STMT_END |
7a6a85bf RG |
816 | |
817 | /* | |
818 | * Useful retrieve shortcuts... | |
819 | */ | |
820 | ||
821 | #define GETCHAR() \ | |
822 | (cxt->fio ? PerlIO_getc(cxt->fio) : (mptr >= mend ? EOF : (int) *mptr++)) | |
823 | ||
111e03c1 RG |
824 | #define GETMARK(x) \ |
825 | STMT_START { \ | |
7a6a85bf RG |
826 | if (!cxt->fio) \ |
827 | MBUF_GETC(x); \ | |
76df4757 | 828 | else if ((int) (x = PerlIO_getc(cxt->fio)) == EOF) \ |
7a6a85bf | 829 | return (SV *) 0; \ |
111e03c1 | 830 | } STMT_END |
7a6a85bf | 831 | |
111e03c1 RG |
832 | #define READ_I32(x) \ |
833 | STMT_START { \ | |
9e21b3d0 JH |
834 | ASSERT(sizeof(x) == sizeof(I32), ("reading an I32")); \ |
835 | oC(x); \ | |
7a6a85bf RG |
836 | if (!cxt->fio) \ |
837 | MBUF_GETINT(x); \ | |
9e21b3d0 | 838 | else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \ |
7a6a85bf | 839 | return (SV *) 0; \ |
111e03c1 | 840 | } STMT_END |
9e21b3d0 JH |
841 | |
842 | #ifdef HAS_NTOHL | |
111e03c1 RG |
843 | #define RLEN(x) \ |
844 | STMT_START { \ | |
9e21b3d0 | 845 | oC(x); \ |
7a6a85bf RG |
846 | if (!cxt->fio) \ |
847 | MBUF_GETINT(x); \ | |
9e21b3d0 | 848 | else if (PerlIO_read(cxt->fio, oI(&x), oS(sizeof(x))) != oS(sizeof(x))) \ |
7a6a85bf | 849 | return (SV *) 0; \ |
9e21b3d0 JH |
850 | if (cxt->netorder) \ |
851 | x = (int) ntohl(x); \ | |
111e03c1 | 852 | } STMT_END |
9e21b3d0 JH |
853 | #else |
854 | #define RLEN(x) READ_I32(x) | |
7a6a85bf RG |
855 | #endif |
856 | ||
111e03c1 RG |
857 | #define READ(x,y) \ |
858 | STMT_START { \ | |
7a6a85bf RG |
859 | if (!cxt->fio) \ |
860 | MBUF_READ(x, y); \ | |
861 | else if (PerlIO_read(cxt->fio, x, y) != y) \ | |
862 | return (SV *) 0; \ | |
111e03c1 | 863 | } STMT_END |
7a6a85bf | 864 | |
111e03c1 RG |
865 | #define SAFEREAD(x,y,z) \ |
866 | STMT_START { \ | |
7a6a85bf RG |
867 | if (!cxt->fio) \ |
868 | MBUF_SAFEREAD(x,y,z); \ | |
869 | else if (PerlIO_read(cxt->fio, x, y) != y) { \ | |
870 | sv_free(z); \ | |
871 | return (SV *) 0; \ | |
872 | } \ | |
111e03c1 | 873 | } STMT_END |
7a6a85bf RG |
874 | |
875 | /* | |
876 | * This macro is used at retrieve time, to remember where object 'y', bearing a | |
877 | * given tag 'tagnum', has been retrieved. Next time we see an SX_OBJECT marker, | |
878 | * we'll therefore know where it has been retrieved and will be able to | |
879 | * share the same reference, as in the original stored memory image. | |
b12202d0 JH |
880 | * |
881 | * We also need to bless objects ASAP for hooks (which may compute "ref $x" | |
882 | * on the objects given to STORABLE_thaw and expect that to be defined), and | |
883 | * also for overloaded objects (for which we might not find the stash if the | |
884 | * object is not blessed yet--this might occur for overloaded objects that | |
885 | * refer to themselves indirectly: if we blessed upon return from a sub | |
886 | * retrieve(), the SX_OBJECT marker we'd found could not have overloading | |
887 | * restored on it because the underlying object would not be blessed yet!). | |
888 | * | |
889 | * To achieve that, the class name of the last retrieved object is passed down | |
890 | * recursively, and the first SEEN() call for which the class name is not NULL | |
891 | * will bless the object. | |
7a6a85bf | 892 | */ |
111e03c1 RG |
893 | #define SEEN(y,c) \ |
894 | STMT_START { \ | |
7a6a85bf RG |
895 | if (!y) \ |
896 | return (SV *) 0; \ | |
897 | if (av_store(cxt->aseen, cxt->tagnum++, SvREFCNT_inc(y)) == 0) \ | |
898 | return (SV *) 0; \ | |
43d061fe | 899 | TRACEME(("aseen(#%d) = 0x%"UVxf" (refcnt=%d)", cxt->tagnum-1, \ |
b12202d0 JH |
900 | PTR2UV(y), SvREFCNT(y)-1)); \ |
901 | if (c) \ | |
902 | BLESS((SV *) (y), c); \ | |
111e03c1 | 903 | } STMT_END |
7a6a85bf RG |
904 | |
905 | /* | |
906 | * Bless `s' in `p', via a temporary reference, required by sv_bless(). | |
907 | */ | |
111e03c1 RG |
908 | #define BLESS(s,p) \ |
909 | STMT_START { \ | |
7a6a85bf RG |
910 | SV *ref; \ |
911 | HV *stash; \ | |
43d061fe | 912 | TRACEME(("blessing 0x%"UVxf" in %s", PTR2UV(s), (p))); \ |
7a6a85bf RG |
913 | stash = gv_stashpv((p), TRUE); \ |
914 | ref = newRV_noinc(s); \ | |
915 | (void) sv_bless(ref, stash); \ | |
916 | SvRV(ref) = 0; \ | |
917 | SvREFCNT_dec(ref); \ | |
111e03c1 | 918 | } STMT_END |
7a6a85bf RG |
919 | |
920 | static int store(); | |
b12202d0 | 921 | static SV *retrieve(stcxt_t *cxt, char *cname); |
7a6a85bf RG |
922 | |
923 | /* | |
924 | * Dynamic dispatching table for SV store. | |
925 | */ | |
926 | ||
927 | static int store_ref(stcxt_t *cxt, SV *sv); | |
928 | static int store_scalar(stcxt_t *cxt, SV *sv); | |
929 | static int store_array(stcxt_t *cxt, AV *av); | |
930 | static int store_hash(stcxt_t *cxt, HV *hv); | |
931 | static int store_tied(stcxt_t *cxt, SV *sv); | |
932 | static int store_tied_item(stcxt_t *cxt, SV *sv); | |
933 | static int store_other(stcxt_t *cxt, SV *sv); | |
f0ffaed8 | 934 | static int store_blessed(stcxt_t *cxt, SV *sv, int type, HV *pkg); |
7a6a85bf | 935 | |
862382c7 JH |
936 | static int (*sv_store[])(stcxt_t *cxt, SV *sv) = { |
937 | store_ref, /* svis_REF */ | |
938 | store_scalar, /* svis_SCALAR */ | |
939 | (int (*)(stcxt_t *cxt, SV *sv)) store_array, /* svis_ARRAY */ | |
940 | (int (*)(stcxt_t *cxt, SV *sv)) store_hash, /* svis_HASH */ | |
941 | store_tied, /* svis_TIED */ | |
942 | store_tied_item, /* svis_TIED_ITEM */ | |
943 | store_other, /* svis_OTHER */ | |
7a6a85bf RG |
944 | }; |
945 | ||
946 | #define SV_STORE(x) (*sv_store[x]) | |
947 | ||
948 | /* | |
949 | * Dynamic dispatching tables for SV retrieval. | |
950 | */ | |
951 | ||
b12202d0 JH |
952 | static SV *retrieve_lscalar(stcxt_t *cxt, char *cname); |
953 | static SV *retrieve_lutf8str(stcxt_t *cxt, char *cname); | |
954 | static SV *old_retrieve_array(stcxt_t *cxt, char *cname); | |
955 | static SV *old_retrieve_hash(stcxt_t *cxt, char *cname); | |
956 | static SV *retrieve_ref(stcxt_t *cxt, char *cname); | |
957 | static SV *retrieve_undef(stcxt_t *cxt, char *cname); | |
958 | static SV *retrieve_integer(stcxt_t *cxt, char *cname); | |
959 | static SV *retrieve_double(stcxt_t *cxt, char *cname); | |
960 | static SV *retrieve_byte(stcxt_t *cxt, char *cname); | |
961 | static SV *retrieve_netint(stcxt_t *cxt, char *cname); | |
962 | static SV *retrieve_scalar(stcxt_t *cxt, char *cname); | |
963 | static SV *retrieve_utf8str(stcxt_t *cxt, char *cname); | |
964 | static SV *retrieve_tied_array(stcxt_t *cxt, char *cname); | |
965 | static SV *retrieve_tied_hash(stcxt_t *cxt, char *cname); | |
966 | static SV *retrieve_tied_scalar(stcxt_t *cxt, char *cname); | |
967 | static SV *retrieve_other(stcxt_t *cxt, char *cname); | |
968 | ||
969 | static SV *(*sv_old_retrieve[])(stcxt_t *cxt, char *cname) = { | |
7a6a85bf RG |
970 | 0, /* SX_OBJECT -- entry unused dynamically */ |
971 | retrieve_lscalar, /* SX_LSCALAR */ | |
972 | old_retrieve_array, /* SX_ARRAY -- for pre-0.6 binaries */ | |
973 | old_retrieve_hash, /* SX_HASH -- for pre-0.6 binaries */ | |
974 | retrieve_ref, /* SX_REF */ | |
975 | retrieve_undef, /* SX_UNDEF */ | |
976 | retrieve_integer, /* SX_INTEGER */ | |
977 | retrieve_double, /* SX_DOUBLE */ | |
978 | retrieve_byte, /* SX_BYTE */ | |
979 | retrieve_netint, /* SX_NETINT */ | |
980 | retrieve_scalar, /* SX_SCALAR */ | |
981 | retrieve_tied_array, /* SX_ARRAY */ | |
982 | retrieve_tied_hash, /* SX_HASH */ | |
983 | retrieve_tied_scalar, /* SX_SCALAR */ | |
984 | retrieve_other, /* SX_SV_UNDEF not supported */ | |
985 | retrieve_other, /* SX_SV_YES not supported */ | |
986 | retrieve_other, /* SX_SV_NO not supported */ | |
987 | retrieve_other, /* SX_BLESS not supported */ | |
988 | retrieve_other, /* SX_IX_BLESS not supported */ | |
989 | retrieve_other, /* SX_HOOK not supported */ | |
990 | retrieve_other, /* SX_OVERLOADED not supported */ | |
991 | retrieve_other, /* SX_TIED_KEY not supported */ | |
992 | retrieve_other, /* SX_TIED_IDX not supported */ | |
dd19458b JH |
993 | retrieve_other, /* SX_UTF8STR not supported */ |
994 | retrieve_other, /* SX_LUTF8STR not supported */ | |
e16e2ff8 | 995 | retrieve_other, /* SX_FLAG_HASH not supported */ |
7a6a85bf RG |
996 | retrieve_other, /* SX_ERROR */ |
997 | }; | |
998 | ||
b12202d0 JH |
999 | static SV *retrieve_array(stcxt_t *cxt, char *cname); |
1000 | static SV *retrieve_hash(stcxt_t *cxt, char *cname); | |
1001 | static SV *retrieve_sv_undef(stcxt_t *cxt, char *cname); | |
1002 | static SV *retrieve_sv_yes(stcxt_t *cxt, char *cname); | |
1003 | static SV *retrieve_sv_no(stcxt_t *cxt, char *cname); | |
1004 | static SV *retrieve_blessed(stcxt_t *cxt, char *cname); | |
1005 | static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname); | |
1006 | static SV *retrieve_hook(stcxt_t *cxt, char *cname); | |
1007 | static SV *retrieve_overloaded(stcxt_t *cxt, char *cname); | |
1008 | static SV *retrieve_tied_key(stcxt_t *cxt, char *cname); | |
1009 | static SV *retrieve_tied_idx(stcxt_t *cxt, char *cname); | |
e16e2ff8 | 1010 | static SV *retrieve_flag_hash(stcxt_t *cxt, char *cname); |
b12202d0 JH |
1011 | |
1012 | static SV *(*sv_retrieve[])(stcxt_t *cxt, char *cname) = { | |
7a6a85bf RG |
1013 | 0, /* SX_OBJECT -- entry unused dynamically */ |
1014 | retrieve_lscalar, /* SX_LSCALAR */ | |
1015 | retrieve_array, /* SX_ARRAY */ | |
1016 | retrieve_hash, /* SX_HASH */ | |
1017 | retrieve_ref, /* SX_REF */ | |
1018 | retrieve_undef, /* SX_UNDEF */ | |
1019 | retrieve_integer, /* SX_INTEGER */ | |
1020 | retrieve_double, /* SX_DOUBLE */ | |
1021 | retrieve_byte, /* SX_BYTE */ | |
1022 | retrieve_netint, /* SX_NETINT */ | |
1023 | retrieve_scalar, /* SX_SCALAR */ | |
1024 | retrieve_tied_array, /* SX_ARRAY */ | |
1025 | retrieve_tied_hash, /* SX_HASH */ | |
1026 | retrieve_tied_scalar, /* SX_SCALAR */ | |
1027 | retrieve_sv_undef, /* SX_SV_UNDEF */ | |
1028 | retrieve_sv_yes, /* SX_SV_YES */ | |
1029 | retrieve_sv_no, /* SX_SV_NO */ | |
1030 | retrieve_blessed, /* SX_BLESS */ | |
1031 | retrieve_idx_blessed, /* SX_IX_BLESS */ | |
1032 | retrieve_hook, /* SX_HOOK */ | |
1033 | retrieve_overloaded, /* SX_OVERLOAD */ | |
1034 | retrieve_tied_key, /* SX_TIED_KEY */ | |
1035 | retrieve_tied_idx, /* SX_TIED_IDX */ | |
dd19458b JH |
1036 | retrieve_utf8str, /* SX_UTF8STR */ |
1037 | retrieve_lutf8str, /* SX_LUTF8STR */ | |
e16e2ff8 | 1038 | retrieve_flag_hash, /* SX_HASH */ |
7a6a85bf RG |
1039 | retrieve_other, /* SX_ERROR */ |
1040 | }; | |
1041 | ||
1042 | #define RETRIEVE(c,x) (*(c)->retrieve_vtbl[(x) >= SX_ERROR ? SX_ERROR : (x)]) | |
1043 | ||
f0ffaed8 | 1044 | static SV *mbuf2sv(void); |
7a6a85bf RG |
1045 | |
1046 | /*** | |
1047 | *** Context management. | |
1048 | ***/ | |
1049 | ||
1050 | /* | |
1051 | * init_perinterp | |
1052 | * | |
1053 | * Called once per "thread" (interpreter) to initialize some global context. | |
1054 | */ | |
f0ffaed8 JH |
1055 | static void init_perinterp(void) |
1056 | { | |
7a6a85bf RG |
1057 | INIT_STCXT; |
1058 | ||
1059 | cxt->netorder = 0; /* true if network order used */ | |
1060 | cxt->forgive_me = -1; /* whether to be forgiving... */ | |
1061 | } | |
1062 | ||
1063 | /* | |
e993d95c JH |
1064 | * reset_context |
1065 | * | |
1066 | * Called at the end of every context cleaning, to perform common reset | |
1067 | * operations. | |
1068 | */ | |
1069 | static void reset_context(stcxt_t *cxt) | |
1070 | { | |
1071 | cxt->entry = 0; | |
1072 | cxt->s_dirty = 0; | |
1073 | cxt->optype &= ~(ST_STORE|ST_RETRIEVE); /* Leave ST_CLONE alone */ | |
1074 | } | |
1075 | ||
1076 | /* | |
7a6a85bf RG |
1077 | * init_store_context |
1078 | * | |
1079 | * Initialize a new store context for real recursion. | |
1080 | */ | |
f0ffaed8 JH |
1081 | static void init_store_context( |
1082 | stcxt_t *cxt, | |
1083 | PerlIO *f, | |
1084 | int optype, | |
1085 | int network_order) | |
7a6a85bf RG |
1086 | { |
1087 | TRACEME(("init_store_context")); | |
1088 | ||
1089 | cxt->netorder = network_order; | |
1090 | cxt->forgive_me = -1; /* Fetched from perl if needed */ | |
1091 | cxt->canonical = -1; /* Idem */ | |
1092 | cxt->tagnum = -1; /* Reset tag numbers */ | |
1093 | cxt->classnum = -1; /* Reset class numbers */ | |
1094 | cxt->fio = f; /* Where I/O are performed */ | |
1095 | cxt->optype = optype; /* A store, or a deep clone */ | |
1096 | cxt->entry = 1; /* No recursion yet */ | |
1097 | ||
1098 | /* | |
1099 | * The `hseen' table is used to keep track of each SV stored and their | |
1100 | * associated tag numbers is special. It is "abused" because the | |
1101 | * values stored are not real SV, just integers cast to (SV *), | |
1102 | * which explains the freeing below. | |
1103 | * | |
1104 | * It is also one possible bottlneck to achieve good storing speed, | |
1105 | * so the "shared keys" optimization is turned off (unlikely to be | |
1106 | * of any use here), and the hash table is "pre-extended". Together, | |
1107 | * those optimizations increase the throughput by 12%. | |
1108 | */ | |
1109 | ||
1110 | cxt->hseen = newHV(); /* Table where seen objects are stored */ | |
1111 | HvSHAREKEYS_off(cxt->hseen); | |
1112 | ||
1113 | /* | |
1114 | * The following does not work well with perl5.004_04, and causes | |
1115 | * a core dump later on, in a completely unrelated spot, which | |
1116 | * makes me think there is a memory corruption going on. | |
1117 | * | |
1118 | * Calling hv_ksplit(hseen, HBUCKETS) instead of manually hacking | |
1119 | * it below does not make any difference. It seems to work fine | |
1120 | * with perl5.004_68 but given the probable nature of the bug, | |
1121 | * that does not prove anything. | |
1122 | * | |
1123 | * It's a shame because increasing the amount of buckets raises | |
1124 | * store() throughput by 5%, but until I figure this out, I can't | |
1125 | * allow for this to go into production. | |
1126 | * | |
1127 | * It is reported fixed in 5.005, hence the #if. | |
1128 | */ | |
f0ffaed8 | 1129 | #if PERL_VERSION >= 5 |
7a6a85bf RG |
1130 | #define HBUCKETS 4096 /* Buckets for %hseen */ |
1131 | HvMAX(cxt->hseen) = HBUCKETS - 1; /* keys %hseen = $HBUCKETS; */ | |
1132 | #endif | |
1133 | ||
1134 | /* | |
1135 | * The `hclass' hash uses the same settings as `hseen' above, but it is | |
1136 | * used to assign sequential tags (numbers) to class names for blessed | |
1137 | * objects. | |
1138 | * | |
1139 | * We turn the shared key optimization on. | |
1140 | */ | |
1141 | ||
1142 | cxt->hclass = newHV(); /* Where seen classnames are stored */ | |
1143 | ||
f0ffaed8 | 1144 | #if PERL_VERSION >= 5 |
7a6a85bf RG |
1145 | HvMAX(cxt->hclass) = HBUCKETS - 1; /* keys %hclass = $HBUCKETS; */ |
1146 | #endif | |
1147 | ||
1148 | /* | |
1149 | * The `hook' hash table is used to keep track of the references on | |
1150 | * the STORABLE_freeze hook routines, when found in some class name. | |
1151 | * | |
1152 | * It is assumed that the inheritance tree will not be changed during | |
1153 | * storing, and that no new method will be dynamically created by the | |
1154 | * hooks. | |
1155 | */ | |
1156 | ||
1157 | cxt->hook = newHV(); /* Table where hooks are cached */ | |
90826881 JH |
1158 | |
1159 | /* | |
1160 | * The `hook_seen' array keeps track of all the SVs returned by | |
1161 | * STORABLE_freeze hooks for us to serialize, so that they are not | |
1162 | * reclaimed until the end of the serialization process. Each SV is | |
1163 | * only stored once, the first time it is seen. | |
1164 | */ | |
1165 | ||
1166 | cxt->hook_seen = newAV(); /* Lists SVs returned by STORABLE_freeze */ | |
7a6a85bf RG |
1167 | } |
1168 | ||
1169 | /* | |
1170 | * clean_store_context | |
1171 | * | |
1172 | * Clean store context by | |
1173 | */ | |
f0ffaed8 | 1174 | static void clean_store_context(stcxt_t *cxt) |
7a6a85bf RG |
1175 | { |
1176 | HE *he; | |
1177 | ||
1178 | TRACEME(("clean_store_context")); | |
1179 | ||
1180 | ASSERT(cxt->optype & ST_STORE, ("was performing a store()")); | |
1181 | ||
1182 | /* | |
1183 | * Insert real values into hashes where we stored faked pointers. | |
1184 | */ | |
1185 | ||
e993d95c JH |
1186 | if (cxt->hseen) { |
1187 | hv_iterinit(cxt->hseen); | |
1188 | while ((he = hv_iternext(cxt->hseen))) /* Extra () for -Wall, grr.. */ | |
1189 | HeVAL(he) = &PL_sv_undef; | |
1190 | } | |
7a6a85bf | 1191 | |
e993d95c JH |
1192 | if (cxt->hclass) { |
1193 | hv_iterinit(cxt->hclass); | |
1194 | while ((he = hv_iternext(cxt->hclass))) /* Extra () for -Wall, grr.. */ | |
1195 | HeVAL(he) = &PL_sv_undef; | |
1196 | } | |
7a6a85bf RG |
1197 | |
1198 | /* | |
1199 | * And now dispose of them... | |
862382c7 JH |
1200 | * |
1201 | * The surrounding if() protection has been added because there might be | |
1202 | * some cases where this routine is called more than once, during | |
1203 | * exceptionnal events. This was reported by Marc Lehmann when Storable | |
1204 | * is executed from mod_perl, and the fix was suggested by him. | |
1205 | * -- RAM, 20/12/2000 | |
1206 | */ | |
1207 | ||
1208 | if (cxt->hseen) { | |
1209 | HV *hseen = cxt->hseen; | |
1210 | cxt->hseen = 0; | |
1211 | hv_undef(hseen); | |
1212 | sv_free((SV *) hseen); | |
1213 | } | |
7a6a85bf | 1214 | |
862382c7 JH |
1215 | if (cxt->hclass) { |
1216 | HV *hclass = cxt->hclass; | |
1217 | cxt->hclass = 0; | |
1218 | hv_undef(hclass); | |
1219 | sv_free((SV *) hclass); | |
1220 | } | |
7a6a85bf | 1221 | |
862382c7 JH |
1222 | if (cxt->hook) { |
1223 | HV *hook = cxt->hook; | |
1224 | cxt->hook = 0; | |
1225 | hv_undef(hook); | |
1226 | sv_free((SV *) hook); | |
1227 | } | |
7a6a85bf | 1228 | |
862382c7 JH |
1229 | if (cxt->hook_seen) { |
1230 | AV *hook_seen = cxt->hook_seen; | |
1231 | cxt->hook_seen = 0; | |
1232 | av_undef(hook_seen); | |
1233 | sv_free((SV *) hook_seen); | |
1234 | } | |
90826881 | 1235 | |
e8189732 NC |
1236 | cxt->forgive_me = -1; /* Fetched from perl if needed */ |
1237 | cxt->canonical = -1; /* Idem */ | |
1238 | ||
e993d95c | 1239 | reset_context(cxt); |
7a6a85bf RG |
1240 | } |
1241 | ||
1242 | /* | |
1243 | * init_retrieve_context | |
1244 | * | |
1245 | * Initialize a new retrieve context for real recursion. | |
1246 | */ | |
dd19458b | 1247 | static void init_retrieve_context(stcxt_t *cxt, int optype, int is_tainted) |
7a6a85bf RG |
1248 | { |
1249 | TRACEME(("init_retrieve_context")); | |
1250 | ||
1251 | /* | |
1252 | * The hook hash table is used to keep track of the references on | |
1253 | * the STORABLE_thaw hook routines, when found in some class name. | |
1254 | * | |
1255 | * It is assumed that the inheritance tree will not be changed during | |
1256 | * storing, and that no new method will be dynamically created by the | |
1257 | * hooks. | |
1258 | */ | |
1259 | ||
1260 | cxt->hook = newHV(); /* Caches STORABLE_thaw */ | |
1261 | ||
1262 | /* | |
1263 | * If retrieving an old binary version, the cxt->retrieve_vtbl variable | |
1264 | * was set to sv_old_retrieve. We'll need a hash table to keep track of | |
1265 | * the correspondance between the tags and the tag number used by the | |
1266 | * new retrieve routines. | |
1267 | */ | |
1268 | ||
e16e2ff8 | 1269 | cxt->hseen = ((cxt->retrieve_vtbl == sv_old_retrieve) ? newHV() : 0); |
7a6a85bf RG |
1270 | |
1271 | cxt->aseen = newAV(); /* Where retrieved objects are kept */ | |
1272 | cxt->aclass = newAV(); /* Where seen classnames are kept */ | |
1273 | cxt->tagnum = 0; /* Have to count objects... */ | |
1274 | cxt->classnum = 0; /* ...and class names as well */ | |
1275 | cxt->optype = optype; | |
dd19458b | 1276 | cxt->s_tainted = is_tainted; |
7a6a85bf | 1277 | cxt->entry = 1; /* No recursion yet */ |
530b72ba NC |
1278 | #ifndef HAS_RESTRICTED_HASHES |
1279 | cxt->derestrict = -1; /* Fetched from perl if needed */ | |
1280 | #endif | |
1281 | #ifndef HAS_UTF8_ALL | |
1282 | cxt->use_bytes = -1; /* Fetched from perl if needed */ | |
1283 | #endif | |
e8189732 | 1284 | cxt->accept_future_minor = -1; /* Fetched from perl if needed */ |
7a6a85bf RG |
1285 | } |
1286 | ||
1287 | /* | |
1288 | * clean_retrieve_context | |
1289 | * | |
1290 | * Clean retrieve context by | |
1291 | */ | |
dd19458b | 1292 | static void clean_retrieve_context(stcxt_t *cxt) |
7a6a85bf RG |
1293 | { |
1294 | TRACEME(("clean_retrieve_context")); | |
1295 | ||
1296 | ASSERT(cxt->optype & ST_RETRIEVE, ("was performing a retrieve()")); | |
1297 | ||
862382c7 JH |
1298 | if (cxt->aseen) { |
1299 | AV *aseen = cxt->aseen; | |
1300 | cxt->aseen = 0; | |
1301 | av_undef(aseen); | |
1302 | sv_free((SV *) aseen); | |
1303 | } | |
7a6a85bf | 1304 | |
862382c7 JH |
1305 | if (cxt->aclass) { |
1306 | AV *aclass = cxt->aclass; | |
1307 | cxt->aclass = 0; | |
1308 | av_undef(aclass); | |
1309 | sv_free((SV *) aclass); | |
1310 | } | |
7a6a85bf | 1311 | |
862382c7 JH |
1312 | if (cxt->hook) { |
1313 | HV *hook = cxt->hook; | |
1314 | cxt->hook = 0; | |
1315 | hv_undef(hook); | |
1316 | sv_free((SV *) hook); | |
1317 | } | |
7a6a85bf | 1318 | |
862382c7 JH |
1319 | if (cxt->hseen) { |
1320 | HV *hseen = cxt->hseen; | |
1321 | cxt->hseen = 0; | |
1322 | hv_undef(hseen); | |
1323 | sv_free((SV *) hseen); /* optional HV, for backward compat. */ | |
1324 | } | |
7a6a85bf | 1325 | |
e8189732 NC |
1326 | #ifndef HAS_RESTRICTED_HASHES |
1327 | cxt->derestrict = -1; /* Fetched from perl if needed */ | |
1328 | #endif | |
1329 | #ifndef HAS_UTF8_ALL | |
1330 | cxt->use_bytes = -1; /* Fetched from perl if needed */ | |
1331 | #endif | |
1332 | cxt->accept_future_minor = -1; /* Fetched from perl if needed */ | |
1333 | ||
e993d95c | 1334 | reset_context(cxt); |
7a6a85bf RG |
1335 | } |
1336 | ||
1337 | /* | |
1338 | * clean_context | |
1339 | * | |
1340 | * A workaround for the CROAK bug: cleanup the last context. | |
1341 | */ | |
e993d95c | 1342 | static void clean_context(stcxt_t *cxt) |
7a6a85bf RG |
1343 | { |
1344 | TRACEME(("clean_context")); | |
1345 | ||
dd19458b | 1346 | ASSERT(cxt->s_dirty, ("dirty context")); |
7a6a85bf | 1347 | |
e993d95c JH |
1348 | if (cxt->membuf_ro) |
1349 | MBUF_RESTORE(); | |
1350 | ||
1351 | ASSERT(!cxt->membuf_ro, ("mbase is not read-only")); | |
1352 | ||
7a6a85bf RG |
1353 | if (cxt->optype & ST_RETRIEVE) |
1354 | clean_retrieve_context(cxt); | |
e993d95c | 1355 | else if (cxt->optype & ST_STORE) |
7a6a85bf | 1356 | clean_store_context(cxt); |
e993d95c JH |
1357 | else |
1358 | reset_context(cxt); | |
862382c7 JH |
1359 | |
1360 | ASSERT(!cxt->s_dirty, ("context is clean")); | |
e993d95c | 1361 | ASSERT(cxt->entry == 0, ("context is reset")); |
7a6a85bf RG |
1362 | } |
1363 | ||
1364 | /* | |
1365 | * allocate_context | |
1366 | * | |
1367 | * Allocate a new context and push it on top of the parent one. | |
1368 | * This new context is made globally visible via SET_STCXT(). | |
1369 | */ | |
1370 | static stcxt_t *allocate_context(parent_cxt) | |
1371 | stcxt_t *parent_cxt; | |
1372 | { | |
1373 | stcxt_t *cxt; | |
1374 | ||
1375 | TRACEME(("allocate_context")); | |
1376 | ||
dd19458b | 1377 | ASSERT(!parent_cxt->s_dirty, ("parent context clean")); |
7a6a85bf | 1378 | |
111e03c1 RG |
1379 | NEW_STORABLE_CXT_OBJ(cxt); |
1380 | cxt->prev = parent_cxt->my_sv; | |
7a6a85bf RG |
1381 | SET_STCXT(cxt); |
1382 | ||
e993d95c JH |
1383 | ASSERT(!cxt->s_dirty, ("clean context")); |
1384 | ||
7a6a85bf RG |
1385 | return cxt; |
1386 | } | |
1387 | ||
1388 | /* | |
1389 | * free_context | |
1390 | * | |
1391 | * Free current context, which cannot be the "root" one. | |
1392 | * Make the context underneath globally visible via SET_STCXT(). | |
1393 | */ | |
1394 | static void free_context(cxt) | |
1395 | stcxt_t *cxt; | |
1396 | { | |
111e03c1 | 1397 | stcxt_t *prev = (stcxt_t *)(cxt->prev ? SvPVX(SvRV(cxt->prev)) : 0); |
7a6a85bf RG |
1398 | |
1399 | TRACEME(("free_context")); | |
1400 | ||
dd19458b | 1401 | ASSERT(!cxt->s_dirty, ("clean context")); |
7a6a85bf RG |
1402 | ASSERT(prev, ("not freeing root context")); |
1403 | ||
111e03c1 | 1404 | SvREFCNT_dec(cxt->my_sv); |
7a6a85bf | 1405 | SET_STCXT(prev); |
e993d95c JH |
1406 | |
1407 | ASSERT(cxt, ("context not void")); | |
7a6a85bf RG |
1408 | } |
1409 | ||
1410 | /*** | |
1411 | *** Predicates. | |
1412 | ***/ | |
1413 | ||
1414 | /* | |
1415 | * is_storing | |
1416 | * | |
1417 | * Tells whether we're in the middle of a store operation. | |
1418 | */ | |
f0ffaed8 | 1419 | int is_storing(void) |
7a6a85bf RG |
1420 | { |
1421 | dSTCXT; | |
1422 | ||
1423 | return cxt->entry && (cxt->optype & ST_STORE); | |
1424 | } | |
1425 | ||
1426 | /* | |
1427 | * is_retrieving | |
1428 | * | |
1429 | * Tells whether we're in the middle of a retrieve operation. | |
1430 | */ | |
f0ffaed8 | 1431 | int is_retrieving(void) |
7a6a85bf RG |
1432 | { |
1433 | dSTCXT; | |
1434 | ||
1435 | return cxt->entry && (cxt->optype & ST_RETRIEVE); | |
1436 | } | |
1437 | ||
1438 | /* | |
1439 | * last_op_in_netorder | |
1440 | * | |
1441 | * Returns whether last operation was made using network order. | |
1442 | * | |
1443 | * This is typically out-of-band information that might prove useful | |
1444 | * to people wishing to convert native to network order data when used. | |
1445 | */ | |
f0ffaed8 | 1446 | int last_op_in_netorder(void) |
7a6a85bf RG |
1447 | { |
1448 | dSTCXT; | |
1449 | ||
1450 | return cxt->netorder; | |
1451 | } | |
1452 | ||
1453 | /*** | |
1454 | *** Hook lookup and calling routines. | |
1455 | ***/ | |
1456 | ||
1457 | /* | |
1458 | * pkg_fetchmeth | |
1459 | * | |
1460 | * A wrapper on gv_fetchmethod_autoload() which caches results. | |
1461 | * | |
1462 | * Returns the routine reference as an SV*, or null if neither the package | |
1463 | * nor its ancestors know about the method. | |
1464 | */ | |
f0ffaed8 JH |
1465 | static SV *pkg_fetchmeth( |
1466 | HV *cache, | |
1467 | HV *pkg, | |
1468 | char *method) | |
7a6a85bf RG |
1469 | { |
1470 | GV *gv; | |
1471 | SV *sv; | |
7a6a85bf RG |
1472 | |
1473 | /* | |
1474 | * The following code is the same as the one performed by UNIVERSAL::can | |
1475 | * in the Perl core. | |
1476 | */ | |
1477 | ||
1478 | gv = gv_fetchmethod_autoload(pkg, method, FALSE); | |
1479 | if (gv && isGV(gv)) { | |
1480 | sv = newRV((SV*) GvCV(gv)); | |
9e21b3d0 | 1481 | TRACEME(("%s->%s: 0x%"UVxf, HvNAME(pkg), method, PTR2UV(sv))); |
7a6a85bf RG |
1482 | } else { |
1483 | sv = newSVsv(&PL_sv_undef); | |
1484 | TRACEME(("%s->%s: not found", HvNAME(pkg), method)); | |
1485 | } | |
1486 | ||
1487 | /* | |
1488 | * Cache the result, ignoring failure: if we can't store the value, | |
1489 | * it just won't be cached. | |
1490 | */ | |
1491 | ||
1492 | (void) hv_store(cache, HvNAME(pkg), strlen(HvNAME(pkg)), sv, 0); | |
1493 | ||
1494 | return SvOK(sv) ? sv : (SV *) 0; | |
1495 | } | |
1496 | ||
1497 | /* | |
1498 | * pkg_hide | |
1499 | * | |
1500 | * Force cached value to be undef: hook ignored even if present. | |
1501 | */ | |
f0ffaed8 JH |
1502 | static void pkg_hide( |
1503 | HV *cache, | |
1504 | HV *pkg, | |
1505 | char *method) | |
7a6a85bf RG |
1506 | { |
1507 | (void) hv_store(cache, | |
1508 | HvNAME(pkg), strlen(HvNAME(pkg)), newSVsv(&PL_sv_undef), 0); | |
1509 | } | |
1510 | ||
1511 | /* | |
212e9bde JH |
1512 | * pkg_uncache |
1513 | * | |
1514 | * Discard cached value: a whole fetch loop will be retried at next lookup. | |
1515 | */ | |
1516 | static void pkg_uncache( | |
1517 | HV *cache, | |
1518 | HV *pkg, | |
1519 | char *method) | |
1520 | { | |
1521 | (void) hv_delete(cache, HvNAME(pkg), strlen(HvNAME(pkg)), G_DISCARD); | |
1522 | } | |
1523 | ||
1524 | /* | |
7a6a85bf RG |
1525 | * pkg_can |
1526 | * | |
1527 | * Our own "UNIVERSAL::can", which caches results. | |
1528 | * | |
1529 | * Returns the routine reference as an SV*, or null if the object does not | |
1530 | * know about the method. | |
1531 | */ | |
f0ffaed8 JH |
1532 | static SV *pkg_can( |
1533 | HV *cache, | |
1534 | HV *pkg, | |
1535 | char *method) | |
7a6a85bf RG |
1536 | { |
1537 | SV **svh; | |
1538 | SV *sv; | |
1539 | ||
1540 | TRACEME(("pkg_can for %s->%s", HvNAME(pkg), method)); | |
1541 | ||
1542 | /* | |
1543 | * Look into the cache to see whether we already have determined | |
1544 | * where the routine was, if any. | |
1545 | * | |
1546 | * NOTA BENE: we don't use `method' at all in our lookup, since we know | |
1547 | * that only one hook (i.e. always the same) is cached in a given cache. | |
1548 | */ | |
1549 | ||
1550 | svh = hv_fetch(cache, HvNAME(pkg), strlen(HvNAME(pkg)), FALSE); | |
1551 | if (svh) { | |
1552 | sv = *svh; | |
1553 | if (!SvOK(sv)) { | |
1554 | TRACEME(("cached %s->%s: not found", HvNAME(pkg), method)); | |
1555 | return (SV *) 0; | |
1556 | } else { | |
43d061fe | 1557 | TRACEME(("cached %s->%s: 0x%"UVxf, |
9e21b3d0 | 1558 | HvNAME(pkg), method, PTR2UV(sv))); |
7a6a85bf RG |
1559 | return sv; |
1560 | } | |
1561 | } | |
1562 | ||
1563 | TRACEME(("not cached yet")); | |
1564 | return pkg_fetchmeth(cache, pkg, method); /* Fetch and cache */ | |
1565 | } | |
1566 | ||
1567 | /* | |
1568 | * scalar_call | |
1569 | * | |
1570 | * Call routine as obj->hook(av) in scalar context. | |
1571 | * Propagates the single returned value if not called in void context. | |
1572 | */ | |
f0ffaed8 JH |
1573 | static SV *scalar_call( |
1574 | SV *obj, | |
1575 | SV *hook, | |
1576 | int cloning, | |
1577 | AV *av, | |
1578 | I32 flags) | |
7a6a85bf RG |
1579 | { |
1580 | dSP; | |
1581 | int count; | |
1582 | SV *sv = 0; | |
1583 | ||
1584 | TRACEME(("scalar_call (cloning=%d)", cloning)); | |
1585 | ||
1586 | ENTER; | |
1587 | SAVETMPS; | |
1588 | ||
1589 | PUSHMARK(sp); | |
1590 | XPUSHs(obj); | |
1591 | XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */ | |
1592 | if (av) { | |
1593 | SV **ary = AvARRAY(av); | |
1594 | int cnt = AvFILLp(av) + 1; | |
1595 | int i; | |
1596 | XPUSHs(ary[0]); /* Frozen string */ | |
1597 | for (i = 1; i < cnt; i++) { | |
43d061fe JH |
1598 | TRACEME(("pushing arg #%d (0x%"UVxf")...", |
1599 | i, PTR2UV(ary[i]))); | |
7a6a85bf RG |
1600 | XPUSHs(sv_2mortal(newRV(ary[i]))); |
1601 | } | |
1602 | } | |
1603 | PUTBACK; | |
1604 | ||
1605 | TRACEME(("calling...")); | |
1606 | count = perl_call_sv(hook, flags); /* Go back to Perl code */ | |
1607 | TRACEME(("count = %d", count)); | |
1608 | ||
1609 | SPAGAIN; | |
1610 | ||
1611 | if (count) { | |
1612 | sv = POPs; | |
1613 | SvREFCNT_inc(sv); /* We're returning it, must stay alive! */ | |
1614 | } | |
1615 | ||
1616 | PUTBACK; | |
1617 | FREETMPS; | |
1618 | LEAVE; | |
1619 | ||
1620 | return sv; | |
1621 | } | |
1622 | ||
1623 | /* | |
1624 | * array_call | |
1625 | * | |
f9a1036d | 1626 | * Call routine obj->hook(cloning) in list context. |
7a6a85bf RG |
1627 | * Returns the list of returned values in an array. |
1628 | */ | |
f0ffaed8 JH |
1629 | static AV *array_call( |
1630 | SV *obj, | |
1631 | SV *hook, | |
1632 | int cloning) | |
7a6a85bf RG |
1633 | { |
1634 | dSP; | |
1635 | int count; | |
1636 | AV *av; | |
1637 | int i; | |
1638 | ||
f0ffaed8 | 1639 | TRACEME(("array_call (cloning=%d)", cloning)); |
7a6a85bf RG |
1640 | |
1641 | ENTER; | |
1642 | SAVETMPS; | |
1643 | ||
1644 | PUSHMARK(sp); | |
1645 | XPUSHs(obj); /* Target object */ | |
1646 | XPUSHs(sv_2mortal(newSViv(cloning))); /* Cloning flag */ | |
1647 | PUTBACK; | |
1648 | ||
1649 | count = perl_call_sv(hook, G_ARRAY); /* Go back to Perl code */ | |
1650 | ||
1651 | SPAGAIN; | |
1652 | ||
1653 | av = newAV(); | |
1654 | for (i = count - 1; i >= 0; i--) { | |
1655 | SV *sv = POPs; | |
1656 | av_store(av, i, SvREFCNT_inc(sv)); | |
1657 | } | |
1658 | ||
1659 | PUTBACK; | |
1660 | FREETMPS; | |
1661 | LEAVE; | |
1662 | ||
1663 | return av; | |
1664 | } | |
1665 | ||
1666 | /* | |
1667 | * known_class | |
1668 | * | |
1669 | * Lookup the class name in the `hclass' table and either assign it a new ID | |
1670 | * or return the existing one, by filling in `classnum'. | |
1671 | * | |
1672 | * Return true if the class was known, false if the ID was just generated. | |
1673 | */ | |
f0ffaed8 JH |
1674 | static int known_class( |
1675 | stcxt_t *cxt, | |
1676 | char *name, /* Class name */ | |
1677 | int len, /* Name length */ | |
1678 | I32 *classnum) | |
7a6a85bf RG |
1679 | { |
1680 | SV **svh; | |
1681 | HV *hclass = cxt->hclass; | |
1682 | ||
1683 | TRACEME(("known_class (%s)", name)); | |
1684 | ||
1685 | /* | |
1686 | * Recall that we don't store pointers in this hash table, but tags. | |
1687 | * Therefore, we need LOW_32BITS() to extract the relevant parts. | |
1688 | */ | |
1689 | ||
1690 | svh = hv_fetch(hclass, name, len, FALSE); | |
1691 | if (svh) { | |
1692 | *classnum = LOW_32BITS(*svh); | |
1693 | return TRUE; | |
1694 | } | |
1695 | ||
1696 | /* | |
1697 | * Unknown classname, we need to record it. | |
7a6a85bf RG |
1698 | */ |
1699 | ||
1700 | cxt->classnum++; | |
3341c981 | 1701 | if (!hv_store(hclass, name, len, INT2PTR(SV*, cxt->classnum), 0)) |
7a6a85bf RG |
1702 | CROAK(("Unable to record new classname")); |
1703 | ||
1704 | *classnum = cxt->classnum; | |
1705 | return FALSE; | |
1706 | } | |
1707 | ||
1708 | /*** | |
1709 | *** Sepcific store routines. | |
1710 | ***/ | |
1711 | ||
1712 | /* | |
1713 | * store_ref | |
1714 | * | |
1715 | * Store a reference. | |
1716 | * Layout is SX_REF <object> or SX_OVERLOAD <object>. | |
1717 | */ | |
f0ffaed8 | 1718 | static int store_ref(stcxt_t *cxt, SV *sv) |
7a6a85bf | 1719 | { |
43d061fe | 1720 | TRACEME(("store_ref (0x%"UVxf")", PTR2UV(sv))); |
7a6a85bf RG |
1721 | |
1722 | /* | |
1723 | * Follow reference, and check if target is overloaded. | |
1724 | */ | |
1725 | ||
1726 | sv = SvRV(sv); | |
1727 | ||
1728 | if (SvOBJECT(sv)) { | |
1729 | HV *stash = (HV *) SvSTASH(sv); | |
1730 | if (stash && Gv_AMG(stash)) { | |
9e21b3d0 | 1731 | TRACEME(("ref (0x%"UVxf") is overloaded", PTR2UV(sv))); |
7a6a85bf RG |
1732 | PUTMARK(SX_OVERLOAD); |
1733 | } else | |
1734 | PUTMARK(SX_REF); | |
1735 | } else | |
1736 | PUTMARK(SX_REF); | |
1737 | ||
1738 | return store(cxt, sv); | |
1739 | } | |
1740 | ||
1741 | /* | |
1742 | * store_scalar | |
1743 | * | |
1744 | * Store a scalar. | |
1745 | * | |
e16e2ff8 | 1746 | * Layout is SX_LSCALAR <length> <data>, SX_SCALAR <length> <data> or SX_UNDEF. |
7a6a85bf RG |
1747 | * The <data> section is omitted if <length> is 0. |
1748 | * | |
1749 | * If integer or double, the layout is SX_INTEGER <data> or SX_DOUBLE <data>. | |
1750 | * Small integers (within [-127, +127]) are stored as SX_BYTE <byte>. | |
1751 | */ | |
f0ffaed8 | 1752 | static int store_scalar(stcxt_t *cxt, SV *sv) |
7a6a85bf RG |
1753 | { |
1754 | IV iv; | |
1755 | char *pv; | |
1756 | STRLEN len; | |
1757 | U32 flags = SvFLAGS(sv); /* "cc -O" may put it in register */ | |
1758 | ||
43d061fe | 1759 | TRACEME(("store_scalar (0x%"UVxf")", PTR2UV(sv))); |
7a6a85bf RG |
1760 | |
1761 | /* | |
1762 | * For efficiency, break the SV encapsulation by peaking at the flags | |
1763 | * directly without using the Perl macros to avoid dereferencing | |
1764 | * sv->sv_flags each time we wish to check the flags. | |
1765 | */ | |
1766 | ||
1767 | if (!(flags & SVf_OK)) { /* !SvOK(sv) */ | |
1768 | if (sv == &PL_sv_undef) { | |
1769 | TRACEME(("immortal undef")); | |
1770 | PUTMARK(SX_SV_UNDEF); | |
1771 | } else { | |
86bbd6dc | 1772 | TRACEME(("undef at 0x%"UVxf, PTR2UV(sv))); |
7a6a85bf RG |
1773 | PUTMARK(SX_UNDEF); |
1774 | } | |
1775 | return 0; | |
1776 | } | |
1777 | ||
1778 | /* | |
1779 | * Always store the string representation of a scalar if it exists. | |
1780 | * Gisle Aas provided me with this test case, better than a long speach: | |
1781 | * | |
1782 | * perl -MDevel::Peek -le '$a="abc"; $a+0; Dump($a)' | |
1783 | * SV = PVNV(0x80c8520) | |
1784 | * REFCNT = 1 | |
1785 | * FLAGS = (NOK,POK,pNOK,pPOK) | |
1786 | * IV = 0 | |
1787 | * NV = 0 | |
1788 | * PV = 0x80c83d0 "abc"\0 | |
1789 | * CUR = 3 | |
1790 | * LEN = 4 | |
1791 | * | |
1792 | * Write SX_SCALAR, length, followed by the actual data. | |
1793 | * | |
1794 | * Otherwise, write an SX_BYTE, SX_INTEGER or an SX_DOUBLE as | |
1795 | * appropriate, followed by the actual (binary) data. A double | |
1796 | * is written as a string if network order, for portability. | |
1797 | * | |
1798 | * NOTE: instead of using SvNOK(sv), we test for SvNOKp(sv). | |
1799 | * The reason is that when the scalar value is tainted, the SvNOK(sv) | |
1800 | * value is false. | |
1801 | * | |
1802 | * The test for a read-only scalar with both POK and NOK set is meant | |
1803 | * to quickly detect &PL_sv_yes and &PL_sv_no without having to pay the | |
1804 | * address comparison for each scalar we store. | |
1805 | */ | |
1806 | ||
1807 | #define SV_MAYBE_IMMORTAL (SVf_READONLY|SVf_POK|SVf_NOK) | |
1808 | ||
1809 | if ((flags & SV_MAYBE_IMMORTAL) == SV_MAYBE_IMMORTAL) { | |
1810 | if (sv == &PL_sv_yes) { | |
1811 | TRACEME(("immortal yes")); | |
1812 | PUTMARK(SX_SV_YES); | |
1813 | } else if (sv == &PL_sv_no) { | |
1814 | TRACEME(("immortal no")); | |
1815 | PUTMARK(SX_SV_NO); | |
1816 | } else { | |
1817 | pv = SvPV(sv, len); /* We know it's SvPOK */ | |
1818 | goto string; /* Share code below */ | |
1819 | } | |
db670f21 NC |
1820 | } else if (flags & SVf_POK) { |
1821 | /* public string - go direct to string read. */ | |
1822 | goto string_readlen; | |
1823 | } else if ( | |
1824 | #if (PATCHLEVEL <= 6) | |
1825 | /* For 5.6 and earlier NV flag trumps IV flag, so only use integer | |
1826 | direct if NV flag is off. */ | |
1827 | (flags & (SVf_NOK | SVf_IOK)) == SVf_IOK | |
1828 | #else | |
1829 | /* 5.7 rules are that if IV public flag is set, IV value is as | |
1830 | good, if not better, than NV value. */ | |
1831 | flags & SVf_IOK | |
1832 | #endif | |
1833 | ) { | |
1834 | iv = SvIV(sv); | |
1835 | /* | |
1836 | * Will come here from below with iv set if double is an integer. | |
1837 | */ | |
1838 | integer: | |
7a6a85bf | 1839 | |
db670f21 NC |
1840 | /* Sorry. This isn't in 5.005_56 (IIRC) or earlier. */ |
1841 | #ifdef SVf_IVisUV | |
1842 | /* Need to do this out here, else 0xFFFFFFFF becomes iv of -1 | |
1843 | * (for example) and that ends up in the optimised small integer | |
1844 | * case. | |
1845 | */ | |
1846 | if ((flags & SVf_IVisUV) && SvUV(sv) > IV_MAX) { | |
1847 | TRACEME(("large unsigned integer as string, value = %"UVuf, SvUV(sv))); | |
1848 | goto string_readlen; | |
1849 | } | |
1850 | #endif | |
1851 | /* | |
1852 | * Optimize small integers into a single byte, otherwise store as | |
1853 | * a real integer (converted into network order if they asked). | |
1854 | */ | |
7a6a85bf | 1855 | |
db670f21 NC |
1856 | if (iv >= -128 && iv <= 127) { |
1857 | unsigned char siv = (unsigned char) (iv + 128); /* [0,255] */ | |
1858 | PUTMARK(SX_BYTE); | |
1859 | PUTMARK(siv); | |
1860 | TRACEME(("small integer stored as %d", siv)); | |
1861 | } else if (cxt->netorder) { | |
1862 | #ifndef HAS_HTONL | |
1863 | TRACEME(("no htonl, fall back to string for integer")); | |
1864 | goto string_readlen; | |
1865 | #else | |
1866 | I32 niv; | |
7a6a85bf | 1867 | |
7a6a85bf | 1868 | |
db670f21 NC |
1869 | #if IVSIZE > 4 |
1870 | if ( | |
1871 | #ifdef SVf_IVisUV | |
1872 | /* Sorry. This isn't in 5.005_56 (IIRC) or earlier. */ | |
1873 | ((flags & SVf_IVisUV) && SvUV(sv) > 0x7FFFFFFF) || | |
1874 | #endif | |
1875 | (iv > 0x7FFFFFFF) || (iv < -0x80000000)) { | |
1876 | /* Bigger than 32 bits. */ | |
1877 | TRACEME(("large network order integer as string, value = %"IVdf, iv)); | |
1878 | goto string_readlen; | |
1879 | } | |
1880 | #endif | |
7a6a85bf | 1881 | |
db670f21 NC |
1882 | niv = (I32) htonl((I32) iv); |
1883 | TRACEME(("using network order")); | |
1884 | PUTMARK(SX_NETINT); | |
1885 | WRITE_I32(niv); | |
1886 | #endif | |
1887 | } else { | |
1888 | PUTMARK(SX_INTEGER); | |
1889 | WRITE(&iv, sizeof(iv)); | |
1890 | } | |
1891 | ||
1892 | TRACEME(("ok (integer 0x%"UVxf", value = %"IVdf")", PTR2UV(sv), iv)); | |
1893 | } else if (flags & SVf_NOK) { | |
1894 | NV nv; | |
1895 | #if (PATCHLEVEL <= 6) | |
1896 | nv = SvNV(sv); | |
1897 | /* | |
1898 | * Watch for number being an integer in disguise. | |
1899 | */ | |
1900 | if (nv == (NV) (iv = I_V(nv))) { | |
1901 | TRACEME(("double %"NVff" is actually integer %"IVdf, nv, iv)); | |
1902 | goto integer; /* Share code above */ | |
1903 | } | |
1904 | #else | |
7a6a85bf | 1905 | |
db670f21 NC |
1906 | SvIV_please(sv); |
1907 | if (SvIOK(sv)) { | |
1908 | iv = SvIV(sv); | |
1909 | goto integer; /* Share code above */ | |
1910 | } | |
1911 | nv = SvNV(sv); | |
1912 | #endif | |
7a6a85bf | 1913 | |
db670f21 NC |
1914 | if (cxt->netorder) { |
1915 | TRACEME(("double %"NVff" stored as string", nv)); | |
1916 | goto string_readlen; /* Share code below */ | |
1917 | } | |
7a6a85bf | 1918 | |
db670f21 NC |
1919 | PUTMARK(SX_DOUBLE); |
1920 | WRITE(&nv, sizeof(nv)); | |
7a6a85bf | 1921 | |
db670f21 | 1922 | TRACEME(("ok (double 0x%"UVxf", value = %"NVff")", PTR2UV(sv), nv)); |
7a6a85bf | 1923 | |
db670f21 NC |
1924 | } else if (flags & (SVp_POK | SVp_NOK | SVp_IOK)) { |
1925 | I32 wlen; /* For 64-bit machines */ | |
7a6a85bf | 1926 | |
db670f21 NC |
1927 | string_readlen: |
1928 | pv = SvPV(sv, len); | |
7a6a85bf | 1929 | |
db670f21 NC |
1930 | /* |
1931 | * Will come here from above if it was readonly, POK and NOK but | |
1932 | * neither &PL_sv_yes nor &PL_sv_no. | |
1933 | */ | |
1934 | string: | |
1935 | ||
1936 | wlen = (I32) len; /* WLEN via STORE_SCALAR expects I32 */ | |
1937 | if (SvUTF8 (sv)) | |
1938 | STORE_UTF8STR(pv, wlen); | |
1939 | else | |
1940 | STORE_SCALAR(pv, wlen); | |
1941 | TRACEME(("ok (scalar 0x%"UVxf" '%s', length = %"IVdf")", | |
1942 | PTR2UV(sv), SvPVX(sv), (IV)len)); | |
7a6a85bf | 1943 | } else |
db670f21 NC |
1944 | CROAK(("Can't determine type of %s(0x%"UVxf")", |
1945 | sv_reftype(sv, FALSE), | |
1946 | PTR2UV(sv))); | |
1947 | return 0; /* Ok, no recursion on scalars */ | |
7a6a85bf RG |
1948 | } |
1949 | ||
1950 | /* | |
1951 | * store_array | |
1952 | * | |
1953 | * Store an array. | |
1954 | * | |
1955 | * Layout is SX_ARRAY <size> followed by each item, in increading index order. | |
1956 | * Each item is stored as <object>. | |
1957 | */ | |
f0ffaed8 | 1958 | static int store_array(stcxt_t *cxt, AV *av) |
7a6a85bf RG |
1959 | { |
1960 | SV **sav; | |
1961 | I32 len = av_len(av) + 1; | |
1962 | I32 i; | |
1963 | int ret; | |
1964 | ||
43d061fe | 1965 | TRACEME(("store_array (0x%"UVxf")", PTR2UV(av))); |
7a6a85bf RG |
1966 | |
1967 | /* | |
1968 | * Signal array by emitting SX_ARRAY, followed by the array length. | |
1969 | */ | |
1970 | ||
1971 | PUTMARK(SX_ARRAY); | |
1972 | WLEN(len); | |
1973 | TRACEME(("size = %d", len)); | |
1974 | ||
1975 | /* | |
1976 | * Now store each item recursively. | |
1977 | */ | |
1978 | ||
1979 | for (i = 0; i < len; i++) { | |
1980 | sav = av_fetch(av, i, 0); | |
1981 | if (!sav) { | |
1982 | TRACEME(("(#%d) undef item", i)); | |
1983 | STORE_UNDEF(); | |
1984 | continue; | |
1985 | } | |
1986 | TRACEME(("(#%d) item", i)); | |
e993d95c | 1987 | if ((ret = store(cxt, *sav))) /* Extra () for -Wall, grr... */ |
7a6a85bf RG |
1988 | return ret; |
1989 | } | |
1990 | ||
1991 | TRACEME(("ok (array)")); | |
1992 | ||
1993 | return 0; | |
1994 | } | |
1995 | ||
1996 | /* | |
1997 | * sortcmp | |
1998 | * | |
1999 | * Sort two SVs | |
2000 | * Borrowed from perl source file pp_ctl.c, where it is used by pp_sort. | |
2001 | */ | |
2002 | static int | |
f0ffaed8 | 2003 | sortcmp(const void *a, const void *b) |
7a6a85bf RG |
2004 | { |
2005 | return sv_cmp(*(SV * const *) a, *(SV * const *) b); | |
2006 | } | |
2007 | ||
2008 | ||
2009 | /* | |
2010 | * store_hash | |
2011 | * | |
d1be9408 | 2012 | * Store a hash table. |
7a6a85bf | 2013 | * |
e16e2ff8 NC |
2014 | * For a "normal" hash (not restricted, no utf8 keys): |
2015 | * | |
7a6a85bf RG |
2016 | * Layout is SX_HASH <size> followed by each key/value pair, in random order. |
2017 | * Values are stored as <object>. | |
2018 | * Keys are stored as <length> <data>, the <data> section being omitted | |
2019 | * if length is 0. | |
c194a0a3 TB |
2020 | * |
2021 | * For a "fancy" hash (restricted or utf8 keys): | |
2022 | * | |
2023 | * Layout is SX_FLAG_HASH <size> <hash flags> followed by each key/value pair, | |
e16e2ff8 NC |
2024 | * in random order. |
2025 | * Values are stored as <object>. | |
2026 | * Keys are stored as <flags> <length> <data>, the <data> section being omitted | |
2027 | * if length is 0. | |
2028 | * Currently the only hash flag is "restriced" | |
2029 | * Key flags are as for hv.h | |
7a6a85bf | 2030 | */ |
f0ffaed8 | 2031 | static int store_hash(stcxt_t *cxt, HV *hv) |
7a6a85bf | 2032 | { |
530b72ba NC |
2033 | I32 len = |
2034 | #ifdef HAS_RESTRICTED_HASHES | |
2035 | HvTOTALKEYS(hv); | |
2036 | #else | |
2037 | HvKEYS(hv); | |
2038 | #endif | |
7a6a85bf RG |
2039 | I32 i; |
2040 | int ret = 0; | |
2041 | I32 riter; | |
2042 | HE *eiter; | |
530b72ba NC |
2043 | int flagged_hash = ((SvREADONLY(hv) |
2044 | #ifdef HAS_HASH_KEY_FLAGS | |
2045 | || HvHASKFLAGS(hv) | |
2046 | #endif | |
2047 | ) ? 1 : 0); | |
e16e2ff8 | 2048 | unsigned char hash_flags = (SvREADONLY(hv) ? SHV_RESTRICTED : 0); |
7a6a85bf | 2049 | |
e16e2ff8 NC |
2050 | if (flagged_hash) { |
2051 | /* needs int cast for C++ compilers, doesn't it? */ | |
2052 | TRACEME(("store_hash (0x%"UVxf") (flags %x)", PTR2UV(hv), | |
2053 | (int) hash_flags)); | |
2054 | } else { | |
2055 | TRACEME(("store_hash (0x%"UVxf")", PTR2UV(hv))); | |
2056 | } | |
7a6a85bf RG |
2057 | |
2058 | /* | |
2059 | * Signal hash by emitting SX_HASH, followed by the table length. | |
2060 | */ | |
2061 | ||
e16e2ff8 NC |
2062 | if (flagged_hash) { |
2063 | PUTMARK(SX_FLAG_HASH); | |
2064 | PUTMARK(hash_flags); | |
2065 | } else { | |
2066 | PUTMARK(SX_HASH); | |
2067 | } | |
7a6a85bf RG |
2068 | WLEN(len); |
2069 | TRACEME(("size = %d", len)); | |
2070 | ||
2071 | /* | |
2072 | * Save possible iteration state via each() on that table. | |
2073 | */ | |
2074 | ||
2075 | riter = HvRITER(hv); | |
2076 | eiter = HvEITER(hv); | |
2077 | hv_iterinit(hv); | |
2078 | ||
2079 | /* | |
2080 | * Now store each item recursively. | |
2081 | * | |
2082 | * If canonical is defined to some true value then store each | |
2083 | * key/value pair in sorted order otherwise the order is random. | |
2084 | * Canonical order is irrelevant when a deep clone operation is performed. | |
2085 | * | |
2086 | * Fetch the value from perl only once per store() operation, and only | |
2087 | * when needed. | |
2088 | */ | |
2089 | ||
2090 | if ( | |
2091 | !(cxt->optype & ST_CLONE) && (cxt->canonical == 1 || | |
2092 | (cxt->canonical < 0 && (cxt->canonical = | |
e16e2ff8 | 2093 | (SvTRUE(perl_get_sv("Storable::canonical", TRUE)) ? 1 : 0)))) |
7a6a85bf RG |
2094 | ) { |
2095 | /* | |
2096 | * Storing in order, sorted by key. | |
2097 | * Run through the hash, building up an array of keys in a | |
2098 | * mortal array, sort the array and then run through the | |
2099 | * array. | |
2100 | */ | |
2101 | ||
2102 | AV *av = newAV(); | |
2103 | ||
e16e2ff8 NC |
2104 | /*av_extend (av, len);*/ |
2105 | ||
7a6a85bf RG |
2106 | TRACEME(("using canonical order")); |
2107 | ||
2108 | for (i = 0; i < len; i++) { | |
530b72ba | 2109 | #ifdef HAS_RESTRICTED_HASHES |
e16e2ff8 | 2110 | HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS); |
530b72ba NC |
2111 | #else |
2112 | HE *he = hv_iternext(hv); | |
2113 | #endif | |
7a6a85bf RG |
2114 | SV *key = hv_iterkeysv(he); |
2115 | av_store(av, AvFILLp(av)+1, key); /* av_push(), really */ | |
2116 | } | |
2117 | ||
2118 | qsort((char *) AvARRAY(av), len, sizeof(SV *), sortcmp); | |
2119 | ||
2120 | for (i = 0; i < len; i++) { | |
e16e2ff8 | 2121 | unsigned char flags; |
7a6a85bf | 2122 | char *keyval; |
e16e2ff8 NC |
2123 | STRLEN keylen_tmp; |
2124 | I32 keylen; | |
7a6a85bf RG |
2125 | SV *key = av_shift(av); |
2126 | HE *he = hv_fetch_ent(hv, key, 0, 0); | |
2127 | SV *val = HeVAL(he); | |
2128 | if (val == 0) | |
2129 | return 1; /* Internal error, not I/O error */ | |
2130 | ||
2131 | /* | |
2132 | * Store value first. | |
2133 | */ | |
2134 | ||
9e21b3d0 | 2135 | TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val))); |
7a6a85bf | 2136 | |
e993d95c | 2137 | if ((ret = store(cxt, val))) /* Extra () for -Wall, grr... */ |
7a6a85bf RG |
2138 | goto out; |
2139 | ||
2140 | /* | |
2141 | * Write key string. | |
2142 | * Keys are written after values to make sure retrieval | |
2143 | * can be optimal in terms of memory usage, where keys are | |
2144 | * read into a fixed unique buffer called kbuf. | |
2145 | * See retrieve_hash() for details. | |
2146 | */ | |
2147 | ||
e16e2ff8 NC |
2148 | /* Implementation of restricted hashes isn't nicely |
2149 | abstracted: */ | |
2150 | flags | |
2151 | = (((hash_flags & SHV_RESTRICTED) | |
2152 | && SvREADONLY(val)) | |
2153 | ? SHV_K_LOCKED : 0); | |
2154 | if (val == &PL_sv_undef) | |
2155 | flags |= SHV_K_PLACEHOLDER; | |
2156 | ||
2157 | keyval = SvPV(key, keylen_tmp); | |
2158 | keylen = keylen_tmp; | |
530b72ba NC |
2159 | #ifdef HAS_UTF8_HASHES |
2160 | /* If you build without optimisation on pre 5.6 | |
2161 | then nothing spots that SvUTF8(key) is always 0, | |
2162 | so the block isn't optimised away, at which point | |
2163 | the linker dislikes the reference to | |
2164 | bytes_from_utf8. */ | |
e16e2ff8 NC |
2165 | if (SvUTF8(key)) { |
2166 | const char *keysave = keyval; | |
2167 | bool is_utf8 = TRUE; | |
2168 | ||
2169 | /* Just casting the &klen to (STRLEN) won't work | |
2170 | well if STRLEN and I32 are of different widths. | |
2171 | --jhi */ | |
2172 | keyval = (char*)bytes_from_utf8((U8*)keyval, | |
2173 | &keylen_tmp, | |
2174 | &is_utf8); | |
2175 | ||
2176 | /* If we were able to downgrade here, then than | |
2177 | means that we have a key which only had chars | |
2178 | 0-255, but was utf8 encoded. */ | |
2179 | ||
2180 | if (keyval != keysave) { | |
2181 | keylen = keylen_tmp; | |
2182 | flags |= SHV_K_WASUTF8; | |
2183 | } else { | |
2184 | /* keylen_tmp can't have changed, so no need | |
2185 | to assign back to keylen. */ | |
2186 | flags |= SHV_K_UTF8; | |
2187 | } | |
2188 | } | |
530b72ba | 2189 | #endif |
e16e2ff8 NC |
2190 | |
2191 | if (flagged_hash) { | |
2192 | PUTMARK(flags); | |
2193 | TRACEME(("(#%d) key '%s' flags %x %u", i, keyval, flags, *keyval)); | |
2194 | } else { | |
2195 | assert (flags == 0); | |
2196 | TRACEME(("(#%d) key '%s'", i, keyval)); | |
2197 | } | |
7a6a85bf RG |
2198 | WLEN(keylen); |
2199 | if (keylen) | |
2200 | WRITE(keyval, keylen); | |
e16e2ff8 NC |
2201 | if (flags & SHV_K_WASUTF8) |
2202 | Safefree (keyval); | |
7a6a85bf RG |
2203 | } |
2204 | ||
2205 | /* | |
2206 | * Free up the temporary array | |
2207 | */ | |
2208 | ||
2209 | av_undef(av); | |
2210 | sv_free((SV *) av); | |
2211 | ||
2212 | } else { | |
2213 | ||
2214 | /* | |
2215 | * Storing in "random" order (in the order the keys are stored | |
2216 | * within the the hash). This is the default and will be faster! | |
2217 | */ | |
2218 | ||
2219 | for (i = 0; i < len; i++) { | |
2220 | char *key; | |
2221 | I32 len; | |
e16e2ff8 | 2222 | unsigned char flags; |
530b72ba | 2223 | #ifdef HV_ITERNEXT_WANTPLACEHOLDERS |
e16e2ff8 | 2224 | HE *he = hv_iternext_flags(hv, HV_ITERNEXT_WANTPLACEHOLDERS); |
530b72ba NC |
2225 | #else |
2226 | HE *he = hv_iternext(hv); | |
2227 | #endif | |
e16e2ff8 NC |
2228 | SV *val = (he ? hv_iterval(hv, he) : 0); |
2229 | SV *key_sv = NULL; | |
2230 | HEK *hek; | |
7a6a85bf RG |
2231 | |
2232 | if (val == 0) | |
2233 | return 1; /* Internal error, not I/O error */ | |
2234 | ||
2235 | /* | |
2236 | * Store value first. | |
2237 | */ | |
2238 | ||
9e21b3d0 | 2239 | TRACEME(("(#%d) value 0x%"UVxf, i, PTR2UV(val))); |
7a6a85bf | 2240 | |
e993d95c | 2241 | if ((ret = store(cxt, val))) /* Extra () for -Wall, grr... */ |
7a6a85bf RG |
2242 | goto out; |
2243 | ||
e16e2ff8 NC |
2244 | /* Implementation of restricted hashes isn't nicely |
2245 | abstracted: */ | |
2246 | flags | |
2247 | = (((hash_flags & SHV_RESTRICTED) | |
2248 | && SvREADONLY(val)) | |
2249 | ? SHV_K_LOCKED : 0); | |
2250 | if (val == &PL_sv_undef) | |
2251 | flags |= SHV_K_PLACEHOLDER; | |
2252 | ||
2253 | hek = HeKEY_hek(he); | |
2254 | len = HEK_LEN(hek); | |
2255 | if (len == HEf_SVKEY) { | |
2256 | /* This is somewhat sick, but the internal APIs are | |
2257 | * such that XS code could put one of these in in | |
2258 | * a regular hash. | |
2259 | * Maybe we should be capable of storing one if | |
2260 | * found. | |
2261 | */ | |
2262 | key_sv = HeKEY_sv(he); | |
2263 | flags |= SHV_K_ISSV; | |
2264 | } else { | |
2265 | /* Regular string key. */ | |
530b72ba | 2266 | #ifdef HAS_HASH_KEY_FLAGS |
e16e2ff8 NC |
2267 | if (HEK_UTF8(hek)) |
2268 | flags |= SHV_K_UTF8; | |
2269 | if (HEK_WASUTF8(hek)) | |
2270 | flags |= SHV_K_WASUTF8; | |
530b72ba | 2271 | #endif |
e16e2ff8 NC |
2272 | key = HEK_KEY(hek); |
2273 | } | |
7a6a85bf RG |
2274 | /* |
2275 | * Write key string. | |
2276 | * Keys are written after values to make sure retrieval | |
2277 | * can be optimal in terms of memory usage, where keys are | |
2278 | * read into a fixed unique buffer called kbuf. | |
2279 | * See retrieve_hash() for details. | |
2280 | */ | |
2281 | ||
e16e2ff8 NC |
2282 | if (flagged_hash) { |
2283 | PUTMARK(flags); | |
2284 | TRACEME(("(#%d) key '%s' flags %x", i, key, flags)); | |
2285 | } else { | |
2286 | assert (flags == 0); | |
2287 | TRACEME(("(#%d) key '%s'", i, key)); | |
2288 | } | |
2289 | if (flags & SHV_K_ISSV) { | |
2290 | store(cxt, key_sv); | |
2291 | } else { | |
2292 | WLEN(len); | |
2293 | if (len) | |
7a6a85bf | 2294 | WRITE(key, len); |
e16e2ff8 | 2295 | } |
7a6a85bf RG |
2296 | } |
2297 | } | |
2298 | ||
43d061fe | 2299 | TRACEME(("ok (hash 0x%"UVxf")", PTR2UV(hv))); |
7a6a85bf RG |
2300 | |
2301 | out: | |
2302 | HvRITER(hv) = riter; /* Restore hash iterator state */ | |
2303 | HvEITER(hv) = eiter; | |
2304 | ||
2305 | return ret; | |
2306 | } | |
2307 | ||
2308 | /* | |
2309 | * store_tied | |
2310 | * | |
2311 | * When storing a tied object (be it a tied scalar, array or hash), we lay out | |
2312 | * a special mark, followed by the underlying tied object. For instance, when | |
2313 | * dealing with a tied hash, we store SX_TIED_HASH <hash object>, where | |
2314 | * <hash object> stands for the serialization of the tied hash. | |
2315 | */ | |
f0ffaed8 | 2316 | static int store_tied(stcxt_t *cxt, SV *sv) |
7a6a85bf RG |
2317 | { |
2318 | MAGIC *mg; | |
2319 | int ret = 0; | |
2320 | int svt = SvTYPE(sv); | |
2321 | char mtype = 'P'; | |
2322 | ||
43d061fe | 2323 | TRACEME(("store_tied (0x%"UVxf")", PTR2UV(sv))); |
7a6a85bf RG |
2324 | |
2325 | /* | |
2326 | * We have a small run-time penalty here because we chose to factorise | |
2327 | * all tieds objects into the same routine, and not have a store_tied_hash, | |
2328 | * a store_tied_array, etc... | |
2329 | * | |
2330 | * Don't use a switch() statement, as most compilers don't optimize that | |
2331 | * well for 2/3 values. An if() else if() cascade is just fine. We put | |
2332 | * tied hashes first, as they are the most likely beasts. | |
2333 | */ | |
2334 | ||
2335 | if (svt == SVt_PVHV) { | |
2336 | TRACEME(("tied hash")); | |
2337 | PUTMARK(SX_TIED_HASH); /* Introduces tied hash */ | |
2338 | } else if (svt == SVt_PVAV) { | |
2339 | TRACEME(("tied array")); | |
2340 | PUTMARK(SX_TIED_ARRAY); /* Introduces tied array */ | |
2341 | } else { | |
2342 | TRACEME(("tied scalar")); | |
2343 | PUTMARK(SX_TIED_SCALAR); /* Introduces tied scalar */ | |
2344 | mtype = 'q'; | |
2345 | } | |
2346 | ||
2347 | if (!(mg = mg_find(sv, mtype))) | |
2348 | CROAK(("No magic '%c' found while storing tied %s", mtype, | |
2349 | (svt == SVt_PVHV) ? "hash" : | |
2350 | (svt == SVt_PVAV) ? "array" : "scalar")); | |
2351 | ||
2352 | /* | |
2353 | * The mg->mg_obj found by mg_find() above actually points to the | |
2354 | * underlying tied Perl object implementation. For instance, if the | |
2355 | * original SV was that of a tied array, then mg->mg_obj is an AV. | |
2356 | * | |
2357 | * Note that we store the Perl object as-is. We don't call its FETCH | |
2358 | * method along the way. At retrieval time, we won't call its STORE | |
2359 | * method either, but the tieing magic will be re-installed. In itself, | |
2360 | * that ensures that the tieing semantics are preserved since futher | |
2361 | * accesses on the retrieved object will indeed call the magic methods... | |
2362 | */ | |
2363 | ||
e993d95c | 2364 | if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */ |
7a6a85bf RG |
2365 | return ret; |
2366 | ||
2367 | TRACEME(("ok (tied)")); | |
2368 | ||
2369 | return 0; | |
2370 | } | |
2371 | ||
2372 | /* | |
2373 | * store_tied_item | |
2374 | * | |
2375 | * Stores a reference to an item within a tied structure: | |
2376 | * | |
2377 | * . \$h{key}, stores both the (tied %h) object and 'key'. | |
2378 | * . \$a[idx], stores both the (tied @a) object and 'idx'. | |
2379 | * | |
2380 | * Layout is therefore either: | |
2381 | * SX_TIED_KEY <object> <key> | |
2382 | * SX_TIED_IDX <object> <index> | |
2383 | */ | |
f0ffaed8 | 2384 | static int store_tied_item(stcxt_t *cxt, SV *sv) |
7a6a85bf RG |
2385 | { |
2386 | MAGIC *mg; | |
2387 | int ret; | |
2388 | ||
43d061fe | 2389 | TRACEME(("store_tied_item (0x%"UVxf")", PTR2UV(sv))); |
7a6a85bf RG |
2390 | |
2391 | if (!(mg = mg_find(sv, 'p'))) | |
2392 | CROAK(("No magic 'p' found while storing reference to tied item")); | |
2393 | ||
2394 | /* | |
2395 | * We discriminate between \$h{key} and \$a[idx] via mg_ptr. | |
2396 | */ | |
2397 | ||
2398 | if (mg->mg_ptr) { | |
2399 | TRACEME(("store_tied_item: storing a ref to a tied hash item")); | |
2400 | PUTMARK(SX_TIED_KEY); | |
9e21b3d0 | 2401 | TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj))); |
7a6a85bf | 2402 | |
e993d95c | 2403 | if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */ |
7a6a85bf RG |
2404 | return ret; |
2405 | ||
9e21b3d0 | 2406 | TRACEME(("store_tied_item: storing PTR 0x%"UVxf, PTR2UV(mg->mg_ptr))); |
7a6a85bf | 2407 | |
e993d95c | 2408 | if ((ret = store(cxt, (SV *) mg->mg_ptr))) /* Idem, for -Wall */ |
7a6a85bf RG |
2409 | return ret; |
2410 | } else { | |
2411 | I32 idx = mg->mg_len; | |
2412 | ||
2413 | TRACEME(("store_tied_item: storing a ref to a tied array item ")); | |
2414 | PUTMARK(SX_TIED_IDX); | |
9e21b3d0 | 2415 | TRACEME(("store_tied_item: storing OBJ 0x%"UVxf, PTR2UV(mg->mg_obj))); |
7a6a85bf | 2416 | |
e993d95c | 2417 | if ((ret = store(cxt, mg->mg_obj))) /* Idem, for -Wall */ |
7a6a85bf RG |
2418 | return ret; |
2419 | ||
2420 | TRACEME(("store_tied_item: storing IDX %d", idx)); | |
2421 | ||
2422 | WLEN(idx); | |
2423 | } | |
2424 | ||
2425 | TRACEME(("ok (tied item)")); | |
2426 | ||
2427 | return 0; | |
2428 | } | |
2429 | ||
2430 | /* | |
2431 | * store_hook -- dispatched manually, not via sv_store[] | |
2432 | * | |
2433 | * The blessed SV is serialized by a hook. | |
2434 | * | |
2435 | * Simple Layout is: | |
2436 | * | |
2437 | * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>] | |
2438 | * | |
2439 | * where <flags> indicates how long <len>, <len2> and <len3> are, whether | |
2440 | * the trailing part [] is present, the type of object (scalar, array or hash). | |
2441 | * There is also a bit which says how the classname is stored between: | |
2442 | * | |
2443 | * <len> <classname> | |
2444 | * <index> | |
2445 | * | |
2446 | * and when the <index> form is used (classname already seen), the "large | |
2447 | * classname" bit in <flags> indicates how large the <index> is. | |
2448 | * | |
2449 | * The serialized string returned by the hook is of length <len2> and comes | |
2450 | * next. It is an opaque string for us. | |
2451 | * | |
2452 | * Those <len3> object IDs which are listed last represent the extra references | |
2453 | * not directly serialized by the hook, but which are linked to the object. | |
2454 | * | |
2455 | * When recursion is mandated to resolve object-IDs not yet seen, we have | |
2456 | * instead, with <header> being flags with bits set to indicate the object type | |
2457 | * and that recursion was indeed needed: | |
2458 | * | |
2459 | * SX_HOOK <header> <object> <header> <object> <flags> | |
2460 | * | |
2461 | * that same header being repeated between serialized objects obtained through | |
2462 | * recursion, until we reach flags indicating no recursion, at which point | |
2463 | * we know we've resynchronized with a single layout, after <flags>. | |
b12202d0 JH |
2464 | * |
2465 | * When storing a blessed ref to a tied variable, the following format is | |
2466 | * used: | |
2467 | * | |
2468 | * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object> | |
2469 | * | |
2470 | * The first <flags> indication carries an object of type SHT_EXTRA, and the | |
2471 | * real object type is held in the <extra> flag. At the very end of the | |
2472 | * serialization stream, the underlying magic object is serialized, just like | |
2473 | * any other tied variable. | |
7a6a85bf | 2474 | */ |
f0ffaed8 JH |
2475 | static int store_hook( |
2476 | stcxt_t *cxt, | |
2477 | SV *sv, | |
2478 | int type, | |
2479 | HV *pkg, | |
2480 | SV *hook) | |
7a6a85bf RG |
2481 | { |
2482 | I32 len; | |
2483 | char *class; | |
2484 | STRLEN len2; | |
2485 | SV *ref; | |
2486 | AV *av; | |
2487 | SV **ary; | |
2488 | int count; /* really len3 + 1 */ | |
2489 | unsigned char flags; | |
2490 | char *pv; | |
2491 | int i; | |
2492 | int recursed = 0; /* counts recursion */ | |
2493 | int obj_type; /* object type, on 2 bits */ | |
2494 | I32 classnum; | |
2495 | int ret; | |
2496 | int clone = cxt->optype & ST_CLONE; | |
e993d95c JH |
2497 | char mtype = '\0'; /* for blessed ref to tied structures */ |
2498 | unsigned char eflags = '\0'; /* used when object type is SHT_EXTRA */ | |
7a6a85bf RG |
2499 | |
2500 | TRACEME(("store_hook, class \"%s\", tagged #%d", HvNAME(pkg), cxt->tagnum)); | |
2501 | ||
2502 | /* | |
2503 | * Determine object type on 2 bits. | |
2504 | */ | |
2505 | ||
2506 | switch (type) { | |
2507 | case svis_SCALAR: | |
2508 | obj_type = SHT_SCALAR; | |
2509 | break; | |
2510 | case svis_ARRAY: | |
2511 | obj_type = SHT_ARRAY; | |
2512 | break; | |
2513 | case svis_HASH: | |
2514 | obj_type = SHT_HASH; | |
2515 | break; | |
b12202d0 JH |
2516 | case svis_TIED: |
2517 | /* | |
2518 | * Produced by a blessed ref to a tied data structure, $o in the | |
2519 | * following Perl code. | |
2520 | * | |
2521 | * my %h; | |
2522 | * tie %h, 'FOO'; | |
2523 | * my $o = bless \%h, 'BAR'; | |
2524 | * | |
2525 | * Signal the tie-ing magic by setting the object type as SHT_EXTRA | |
2526 | * (since we have only 2 bits in <flags> to store the type), and an | |
2527 | * <extra> byte flag will be emitted after the FIRST <flags> in the | |
2528 | * stream, carrying what we put in `eflags'. | |
2529 | */ | |
2530 | obj_type = SHT_EXTRA; | |
2531 | switch (SvTYPE(sv)) { | |
2532 | case SVt_PVHV: | |
2533 | eflags = (unsigned char) SHT_THASH; | |
2534 | mtype = 'P'; | |
2535 | break; | |
2536 | case SVt_PVAV: | |
2537 | eflags = (unsigned char) SHT_TARRAY; | |
2538 | mtype = 'P'; | |
2539 | break; | |
2540 | default: | |
2541 | eflags = (unsigned char) SHT_TSCALAR; | |
2542 | mtype = 'q'; | |
2543 | break; | |
2544 | } | |
2545 | break; | |
7a6a85bf RG |
2546 | default: |
2547 | CROAK(("Unexpected object type (%d) in store_hook()", type)); | |
2548 | } | |
2549 | flags = SHF_NEED_RECURSE | obj_type; | |
2550 | ||
2551 | class = HvNAME(pkg); | |
2552 | len = strlen(class); | |
2553 | ||
2554 | /* | |
2555 | * To call the hook, we need to fake a call like: | |
2556 | * | |
2557 | * $object->STORABLE_freeze($cloning); | |
2558 | * | |
2559 | * but we don't have the $object here. For instance, if $object is | |
2560 | * a blessed array, what we have in `sv' is the array, and we can't | |
2561 | * call a method on those. | |
2562 | * | |
2563 | * Therefore, we need to create a temporary reference to the object and | |
2564 | * make the call on that reference. | |
2565 | */ | |
2566 | ||
2567 | TRACEME(("about to call STORABLE_freeze on class %s", class)); | |
2568 | ||
2569 | ref = newRV_noinc(sv); /* Temporary reference */ | |
2570 | av = array_call(ref, hook, clone); /* @a = $object->STORABLE_freeze($c) */ | |
2571 | SvRV(ref) = 0; | |
2572 | SvREFCNT_dec(ref); /* Reclaim temporary reference */ | |
2573 | ||
2574 | count = AvFILLp(av) + 1; | |
2575 | TRACEME(("store_hook, array holds %d items", count)); | |
2576 | ||
2577 | /* | |
2578 | * If they return an empty list, it means they wish to ignore the | |
2579 | * hook for this class (and not just this instance -- that's for them | |
2580 | * to handle if they so wish). | |
2581 | * | |
2582 | * Simply disable the cached entry for the hook (it won't be recomputed | |
2583 | * since it's present in the cache) and recurse to store_blessed(). | |
2584 | */ | |
2585 | ||
2586 | if (!count) { | |
2587 | /* | |
2588 | * They must not change their mind in the middle of a serialization. | |
2589 | */ | |
2590 | ||
2591 | if (hv_fetch(cxt->hclass, class, len, FALSE)) | |
2592 | CROAK(("Too late to ignore hooks for %s class \"%s\"", | |
2593 | (cxt->optype & ST_CLONE) ? "cloning" : "storing", class)); | |
2594 | ||
2595 | pkg_hide(cxt->hook, pkg, "STORABLE_freeze"); | |
2596 | ||
2597 | ASSERT(!pkg_can(cxt->hook, pkg, "STORABLE_freeze"), ("hook invisible")); | |
cc964657 | 2598 | TRACEME(("ignoring STORABLE_freeze in class \"%s\"", class)); |
7a6a85bf RG |
2599 | |
2600 | return store_blessed(cxt, sv, type, pkg); | |
2601 | } | |
2602 | ||
2603 | /* | |
2604 | * Get frozen string. | |
2605 | */ | |
2606 | ||
2607 | ary = AvARRAY(av); | |
2608 | pv = SvPV(ary[0], len2); | |
2609 | ||
2610 | /* | |
7a6a85bf RG |
2611 | * If they returned more than one item, we need to serialize some |
2612 | * extra references if not already done. | |
2613 | * | |
2614 | * Loop over the array, starting at postion #1, and for each item, | |
2615 | * ensure it is a reference, serialize it if not already done, and | |
2616 | * replace the entry with the tag ID of the corresponding serialized | |
2617 | * object. | |
2618 | * | |
2619 | * We CHEAT by not calling av_fetch() and read directly within the | |
2620 | * array, for speed. | |
2621 | */ | |
2622 | ||
2623 | for (i = 1; i < count; i++) { | |
2624 | SV **svh; | |
90826881 JH |
2625 | SV *rsv = ary[i]; |
2626 | SV *xsv; | |
2627 | AV *av_hook = cxt->hook_seen; | |
7a6a85bf | 2628 | |
90826881 JH |
2629 | if (!SvROK(rsv)) |
2630 | CROAK(("Item #%d returned by STORABLE_freeze " | |
2631 | "for %s is not a reference", i, class)); | |
2632 | xsv = SvRV(rsv); /* Follow ref to know what to look for */ | |
7a6a85bf RG |
2633 | |
2634 | /* | |
2635 | * Look in hseen and see if we have a tag already. | |
2636 | * Serialize entry if not done already, and get its tag. | |
2637 | */ | |
2638 | ||
13689cfe | 2639 | if ((svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE))) |
7a6a85bf RG |
2640 | goto sv_seen; /* Avoid moving code too far to the right */ |
2641 | ||
9e21b3d0 | 2642 | TRACEME(("listed object %d at 0x%"UVxf" is unknown", i-1, PTR2UV(xsv))); |
7a6a85bf RG |
2643 | |
2644 | /* | |
2645 | * We need to recurse to store that object and get it to be known | |
2646 | * so that we can resolve the list of object-IDs at retrieve time. | |
2647 | * | |
2648 | * The first time we do this, we need to emit the proper header | |
2649 | * indicating that we recursed, and what the type of object is (the | |
2650 | * object we're storing via a user-hook). Indeed, during retrieval, | |
2651 | * we'll have to create the object before recursing to retrieve the | |
2652 | * others, in case those would point back at that object. | |
2653 | */ | |
2654 | ||
b12202d0 JH |
2655 | /* [SX_HOOK] <flags> [<extra>] <object>*/ |
2656 | if (!recursed++) { | |
7a6a85bf | 2657 | PUTMARK(SX_HOOK); |
b12202d0 JH |
2658 | PUTMARK(flags); |
2659 | if (obj_type == SHT_EXTRA) | |
2660 | PUTMARK(eflags); | |
2661 | } else | |
2662 | PUTMARK(flags); | |
7a6a85bf | 2663 | |
e993d95c | 2664 | if ((ret = store(cxt, xsv))) /* Given by hook for us to store */ |
7a6a85bf RG |
2665 | return ret; |
2666 | ||
2667 | svh = hv_fetch(cxt->hseen, (char *) &xsv, sizeof(xsv), FALSE); | |
2668 | if (!svh) | |
2669 | CROAK(("Could not serialize item #%d from hook in %s", i, class)); | |
2670 | ||
2671 | /* | |
90826881 JH |
2672 | * It was the first time we serialized `xsv'. |
2673 | * | |
2674 | * Keep this SV alive until the end of the serialization: if we | |
2675 | * disposed of it right now by decrementing its refcount, and it was | |
2676 | * a temporary value, some next temporary value allocated during | |
2677 | * another STORABLE_freeze might take its place, and we'd wrongly | |
2678 | * assume that new SV was already serialized, based on its presence | |
2679 | * in cxt->hseen. | |
2680 | * | |
2681 | * Therefore, push it away in cxt->hook_seen. | |
7a6a85bf RG |
2682 | */ |
2683 | ||
90826881 JH |
2684 | av_store(av_hook, AvFILLp(av_hook)+1, SvREFCNT_inc(xsv)); |
2685 | ||
7a6a85bf | 2686 | sv_seen: |
90826881 JH |
2687 | /* |
2688 | * Dispose of the REF they returned. If we saved the `xsv' away | |
2689 | * in the array of returned SVs, that will not cause the underlying | |
2690 | * referenced SV to be reclaimed. | |
2691 | */ | |
2692 | ||
2693 | ASSERT(SvREFCNT(xsv) > 1, ("SV will survive disposal of its REF")); | |
2694 | SvREFCNT_dec(rsv); /* Dispose of reference */ | |
2695 | ||
2696 | /* | |
2697 | * Replace entry with its tag (not a real SV, so no refcnt increment) | |
2698 | */ | |
2699 | ||
7a6a85bf | 2700 | ary[i] = *svh; |
76edffbb | 2701 | TRACEME(("listed object %d at 0x%"UVxf" is tag #%"UVuf, |
d67b2c17 | 2702 | i-1, PTR2UV(xsv), PTR2UV(*svh))); |
7a6a85bf RG |
2703 | } |
2704 | ||
2705 | /* | |
dd19458b JH |
2706 | * Allocate a class ID if not already done. |
2707 | * | |
2708 | * This needs to be done after the recursion above, since at retrieval | |
2709 | * time, we'll see the inner objects first. Many thanks to | |
2710 | * Salvador Ortiz Garcia <sog@msg.com.mx> who spot that bug and | |
2711 | * proposed the right fix. -- RAM, 15/09/2000 | |
2712 | */ | |
2713 | ||
2714 | if (!known_class(cxt, class, len, &classnum)) { | |
2715 | TRACEME(("first time we see class %s, ID = %d", class, classnum)); | |
2716 | classnum = -1; /* Mark: we must store classname */ | |
2717 | } else { | |
2718 | TRACEME(("already seen class %s, ID = %d", class, classnum)); | |
2719 | } | |
2720 | ||
2721 | /* | |
7a6a85bf RG |
2722 | * Compute leading flags. |
2723 | */ | |
2724 | ||
2725 | flags = obj_type; | |
2726 | if (((classnum == -1) ? len : classnum) > LG_SCALAR) | |
2727 | flags |= SHF_LARGE_CLASSLEN; | |
2728 | if (classnum != -1) | |
2729 | flags |= SHF_IDX_CLASSNAME; | |
2730 | if (len2 > LG_SCALAR) | |
2731 | flags |= SHF_LARGE_STRLEN; | |
2732 | if (count > 1) | |
2733 | flags |= SHF_HAS_LIST; | |
2734 | if (count > (LG_SCALAR + 1)) | |
2735 | flags |= SHF_LARGE_LISTLEN; | |
2736 | ||
2737 | /* | |
2738 | * We're ready to emit either serialized form: | |
2739 | * | |
2740 | * SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>] | |
2741 | * SX_HOOK <flags> <index> <len2> <str> [<len3> <object-IDs>] | |
2742 | * | |
2743 | * If we recursed, the SX_HOOK has already been emitted. | |
2744 | */ | |
2745 | ||
9e21b3d0 JH |
2746 | TRACEME(("SX_HOOK (recursed=%d) flags=0x%x " |
2747 | "class=%"IVdf" len=%"IVdf" len2=%"IVdf" len3=%d", | |
d67b2c17 | 2748 | recursed, flags, (IV)classnum, (IV)len, (IV)len2, count-1)); |
7a6a85bf | 2749 | |
b12202d0 JH |
2750 | /* SX_HOOK <flags> [<extra>] */ |
2751 | if (!recursed) { | |
7a6a85bf | 2752 | PUTMARK(SX_HOOK); |
b12202d0 JH |
2753 | PUTMARK(flags); |
2754 | if (obj_type == SHT_EXTRA) | |
2755 | PUTMARK(eflags); | |
2756 | } else | |
2757 | PUTMARK(flags); | |
7a6a85bf RG |
2758 | |
2759 | /* <len> <classname> or <index> */ | |
2760 | if (flags & SHF_IDX_CLASSNAME) { | |
2761 | if (flags & SHF_LARGE_CLASSLEN) | |
2762 | WLEN(classnum); | |
2763 | else { | |
2764 | unsigned char cnum = (unsigned char) classnum; | |
2765 | PUTMARK(cnum); | |
2766 | } | |
2767 | } else { | |
2768 | if (flags & SHF_LARGE_CLASSLEN) | |
2769 | WLEN(len); | |
2770 | else { | |
2771 | unsigned char clen = (unsigned char) len; | |
2772 | PUTMARK(clen); | |
2773 | } | |
2774 | WRITE(class, len); /* Final \0 is omitted */ | |
2775 | } | |
2776 | ||
2777 | /* <len2> <frozen-str> */ | |
cc964657 JH |
2778 | if (flags & SHF_LARGE_STRLEN) { |
2779 | I32 wlen2 = len2; /* STRLEN might be 8 bytes */ | |
2780 | WLEN(wlen2); /* Must write an I32 for 64-bit machines */ | |
2781 | } else { | |
7a6a85bf RG |
2782 | unsigned char clen = (unsigned char) len2; |
2783 | PUTMARK(clen); | |
2784 | } | |
2785 | if (len2) | |
7c436af3 | 2786 | WRITE(pv, (SSize_t)len2); /* Final \0 is omitted */ |
7a6a85bf RG |
2787 | |
2788 | /* [<len3> <object-IDs>] */ | |
2789 | if (flags & SHF_HAS_LIST) { | |
2790 | int len3 = count - 1; | |
2791 | if (flags & SHF_LARGE_LISTLEN) | |
2792 | WLEN(len3); | |
2793 | else { | |
2794 | unsigned char clen = (unsigned char) len3; | |
2795 | PUTMARK(clen); | |
2796 | } | |
2797 | ||
2798 | /* | |
2799 | * NOTA BENE, for 64-bit machines: the ary[i] below does not yield a | |
2800 | * real pointer, rather a tag number, well under the 32-bit limit. | |
2801 | */ | |
2802 | ||
2803 | for (i = 1; i < count; i++) { | |
2804 | I32 tagval = htonl(LOW_32BITS(ary[i])); | |
9e21b3d0 | 2805 | WRITE_I32(tagval); |
7a6a85bf RG |
2806 | TRACEME(("object %d, tag #%d", i-1, ntohl(tagval))); |
2807 | } | |
2808 | } | |
2809 | ||
2810 | /* | |
2811 | * Free the array. We need extra care for indices after 0, since they | |
2812 | * don't hold real SVs but integers cast. | |
2813 | */ | |
2814 | ||
2815 | if (count > 1) | |
2816 | AvFILLp(av) = 0; /* Cheat, nothing after 0 interests us */ | |
2817 | av_undef(av); | |
2818 | sv_free((SV *) av); | |
2819 | ||
b12202d0 JH |
2820 | /* |
2821 | * If object was tied, need to insert serialization of the magic object. | |
2822 | */ | |
2823 | ||
2824 | if (obj_type == SHT_EXTRA) { | |
2825 | MAGIC *mg; | |
2826 | ||
2827 | if (!(mg = mg_find(sv, mtype))) { | |
2828 | int svt = SvTYPE(sv); | |
2829 | CROAK(("No magic '%c' found while storing ref to tied %s with hook", | |
2830 | mtype, (svt == SVt_PVHV) ? "hash" : | |
2831 | (svt == SVt_PVAV) ? "array" : "scalar")); | |
2832 | } | |
2833 | ||
2834 | TRACEME(("handling the magic object 0x%"UVxf" part of 0x%"UVxf, | |
2835 | PTR2UV(mg->mg_obj), PTR2UV(sv))); | |
2836 | ||
2837 | /* | |
2838 | * [<magic object>] | |
2839 | */ | |
2840 | ||
e993d95c | 2841 | if ((ret = store(cxt, mg->mg_obj))) /* Extra () for -Wall, grr... */ |
b12202d0 JH |
2842 | return ret; |
2843 | } | |
2844 | ||
7a6a85bf RG |
2845 | return 0; |
2846 | } | |
2847 | ||
2848 | /* | |
2849 | * store_blessed -- dispatched manually, not via sv_store[] | |
2850 | * | |
2851 | * Check whether there is a STORABLE_xxx hook defined in the class or in one | |
2852 | * of its ancestors. If there is, then redispatch to store_hook(); | |
2853 | * | |
2854 | * Otherwise, the blessed SV is stored using the following layout: | |
2855 | * | |
2856 | * SX_BLESS <flag> <len> <classname> <object> | |
2857 | * | |
2858 | * where <flag> indicates whether <len> is stored on 0 or 4 bytes, depending | |
2859 | * on the high-order bit in flag: if 1, then length follows on 4 bytes. | |
2860 | * Otherwise, the low order bits give the length, thereby giving a compact | |
2861 | * representation for class names less than 127 chars long. | |
2862 | * | |
2863 | * Each <classname> seen is remembered and indexed, so that the next time | |
2864 | * an object in the blessed in the same <classname> is stored, the following | |
2865 | * will be emitted: | |
2866 | * | |
2867 | * SX_IX_BLESS <flag> <index> <object> | |
2868 | * | |
2869 | * where <index> is the classname index, stored on 0 or 4 bytes depending | |
2870 | * on the high-order bit in flag (same encoding as above for <len>). | |
2871 | */ | |
f0ffaed8 JH |
2872 | static int store_blessed( |
2873 | stcxt_t *cxt, | |
2874 | SV *sv, | |
2875 | int type, | |
2876 | HV *pkg) | |
7a6a85bf RG |
2877 | { |
2878 | SV *hook; | |
2879 | I32 len; | |
2880 | char *class; | |
2881 | I32 classnum; | |
2882 | ||
2883 | TRACEME(("store_blessed, type %d, class \"%s\"", type, HvNAME(pkg))); | |
2884 | ||
2885 | /* | |
2886 | * Look for a hook for this blessed SV and redirect to store_hook() | |
2887 | * if needed. | |
2888 | */ | |
2889 | ||
2890 | hook = pkg_can(cxt->hook, pkg, "STORABLE_freeze"); | |
2891 | if (hook) | |
2892 | return store_hook(cxt, sv, type, pkg, hook); | |
2893 | ||
2894 | /* | |
2895 | * This is a blessed SV without any serialization hook. | |
2896 | */ | |
2897 | ||
2898 | class = HvNAME(pkg); | |
2899 | len = strlen(class); | |
2900 | ||
43d061fe JH |
2901 | TRACEME(("blessed 0x%"UVxf" in %s, no hook: tagged #%d", |
2902 | PTR2UV(sv), class, cxt->tagnum)); | |
7a6a85bf RG |
2903 | |
2904 | /* | |
2905 | * Determine whether it is the first time we see that class name (in which | |
2906 | * case it will be stored in the SX_BLESS form), or whether we already | |
2907 | * saw that class name before (in which case the SX_IX_BLESS form will be | |
2908 | * used). | |
2909 | */ | |
2910 | ||
2911 | if (known_class(cxt, class, len, &classnum)) { | |
2912 | TRACEME(("already seen class %s, ID = %d", class, classnum)); | |
2913 | PUTMARK(SX_IX_BLESS); | |
2914 | if (classnum <= LG_BLESS) { | |
2915 | unsigned char cnum = (unsigned char) classnum; | |
2916 | PUTMARK(cnum); | |
2917 | } else { | |
2918 | unsigned char flag = (unsigned char) 0x80; | |
2919 | PUTMARK(flag); | |
2920 | WLEN(classnum); | |
2921 | } | |
2922 | } else { | |
2923 | TRACEME(("first time we see class %s, ID = %d", class, classnum)); | |
2924 | PUTMARK(SX_BLESS); | |
2925 | if (len <= LG_BLESS) { | |
2926 | unsigned char clen = (unsigned char) len; | |
2927 | PUTMARK(clen); | |
2928 | } else { | |
2929 | unsigned char flag = (unsigned char) 0x80; | |
2930 | PUTMARK(flag); | |
2931 | WLEN(len); /* Don't BER-encode, this should be rare */ | |
2932 | } | |
2933 | WRITE(class, len); /* Final \0 is omitted */ | |
2934 | } | |
2935 | ||
2936 | /* | |
2937 | * Now emit the <object> part. | |
2938 | */ | |
2939 | ||
2940 | return SV_STORE(type)(cxt, sv); | |
2941 | } | |
2942 | ||
2943 | /* | |
2944 | * store_other | |
2945 | * | |
2946 | * We don't know how to store the item we reached, so return an error condition. | |
2947 | * (it's probably a GLOB, some CODE reference, etc...) | |
2948 | * | |
2949 | * If they defined the `forgive_me' variable at the Perl level to some | |
2950 | * true value, then don't croak, just warn, and store a placeholder string | |
2951 | * instead. | |
2952 | */ | |
f0ffaed8 | 2953 | static int store_other(stcxt_t *cxt, SV *sv) |
7a6a85bf | 2954 | { |
cc964657 | 2955 | I32 len; |
7a6a85bf RG |
2956 | static char buf[80]; |
2957 | ||
2958 | TRACEME(("store_other")); | |
2959 | ||
2960 | /* | |
2961 | * Fetch the value from perl only once per store() operation. | |
2962 | */ | |
2963 | ||
2964 | if ( | |
2965 | cxt->forgive_me == 0 || | |
2966 | (cxt->forgive_me < 0 && !(cxt->forgive_me = | |
2967 | SvTRUE(perl_get_sv("Storable::forgive_me", TRUE)) ? 1 : 0)) | |
2968 | ) | |
2969 | CROAK(("Can't store %s items", sv_reftype(sv, FALSE))); | |
2970 | ||
43d061fe JH |
2971 | warn("Can't store item %s(0x%"UVxf")", |
2972 | sv_reftype(sv, FALSE), PTR2UV(sv)); | |
7a6a85bf RG |
2973 | |
2974 | /* | |
2975 | * Store placeholder string as a scalar instead... | |
2976 | */ | |
2977 | ||
13689cfe | 2978 | (void) sprintf(buf, "You lost %s(0x%"UVxf")%c", sv_reftype(sv, FALSE), |
e993d95c | 2979 | PTR2UV(sv), (char) 0); |
7a6a85bf RG |
2980 | |
2981 | len = strlen(buf); | |
2982 | STORE_SCALAR(buf, len); | |
1cf92b12 | 2983 | TRACEME(("ok (dummy \"%s\", length = %"IVdf")", buf, (IV) len)); |
7a6a85bf RG |
2984 | |
2985 | return 0; | |
2986 | } | |
2987 | ||
2988 | /*** | |
2989 | *** Store driving routines | |
2990 | ***/ | |
2991 | ||
2992 | /* | |
2993 | * sv_type | |
2994 | * | |
2995 | * WARNING: partially duplicates Perl's sv_reftype for speed. | |
2996 | * | |
2997 | * Returns the type of the SV, identified by an integer. That integer | |
2998 | * may then be used to index the dynamic routine dispatch table. | |
2999 | */ | |
f0ffaed8 | 3000 | static int sv_type(SV *sv) |
7a6a85bf RG |
3001 | { |
3002 | switch (SvTYPE(sv)) { | |
3003 | case SVt_NULL: | |
3004 | case SVt_IV: | |
3005 | case SVt_NV: | |
3006 | /* | |
3007 | * No need to check for ROK, that can't be set here since there | |
3008 | * is no field capable of hodling the xrv_rv reference. | |
3009 | */ | |
3010 | return svis_SCALAR; | |
3011 | case SVt_PV: | |
3012 | case SVt_RV: | |
3013 | case SVt_PVIV: | |
3014 | case SVt_PVNV: | |
3015 | /* | |
3016 | * Starting from SVt_PV, it is possible to have the ROK flag | |
3017 | * set, the pointer to the other SV being either stored in | |
3018 | * the xrv_rv (in the case of a pure SVt_RV), or as the | |
3019 | * xpv_pv field of an SVt_PV and its heirs. | |
3020 | * | |
3021 | * However, those SV cannot be magical or they would be an | |
3022 | * SVt_PVMG at least. | |
3023 | */ | |
3024 | return SvROK(sv) ? svis_REF : svis_SCALAR; | |
3025 | case SVt_PVMG: | |
3026 | case SVt_PVLV: /* Workaround for perl5.004_04 "LVALUE" bug */ | |
3027 | if (SvRMAGICAL(sv) && (mg_find(sv, 'p'))) | |
3028 | return svis_TIED_ITEM; | |
3029 | /* FALL THROUGH */ | |
3030 | case SVt_PVBM: | |
3031 | if (SvRMAGICAL(sv) && (mg_find(sv, 'q'))) | |
3032 | return svis_TIED; | |
3033 | return SvROK(sv) ? svis_REF : svis_SCALAR; | |
3034 | case SVt_PVAV: | |
3035 | if (SvRMAGICAL(sv) && (mg_find(sv, 'P'))) | |
3036 | return svis_TIED; | |
3037 | return svis_ARRAY; | |
3038 | case SVt_PVHV: | |
3039 | if (SvRMAGICAL(sv) && (mg_find(sv, 'P'))) | |
3040 | return svis_TIED; | |
3041 | return svis_HASH; | |
3042 | default: | |
3043 | break; | |
3044 | } | |
3045 | ||
3046 | return svis_OTHER; | |
3047 | } | |
3048 | ||
3049 | /* | |
3050 | * store | |
3051 | * | |
3052 | * Recursively store objects pointed to by the sv to the specified file. | |
3053 | * | |
3054 | * Layout is <content> or SX_OBJECT <tagnum> if we reach an already stored | |
3055 | * object (one for which storage has started -- it may not be over if we have | |
3056 | * a self-referenced structure). This data set forms a stored <object>. | |
3057 | */ | |
f0ffaed8 | 3058 | static int store(stcxt_t *cxt, SV *sv) |
7a6a85bf RG |
3059 | { |
3060 | SV **svh; | |
3061 | int ret; | |
7a6a85bf | 3062 | int type; |
43d061fe | 3063 | HV *hseen = cxt->hseen; |
7a6a85bf | 3064 | |
43d061fe | 3065 | TRACEME(("store (0x%"UVxf")", PTR2UV(sv))); |
7a6a85bf RG |
3066 | |
3067 | /* | |
3068 | * If object has already been stored, do not duplicate data. | |
3069 | * Simply emit the SX_OBJECT marker followed by its tag data. | |
3070 | * The tag is always written in network order. | |
3071 | * | |
3072 | * NOTA BENE, for 64-bit machines: the "*svh" below does not yield a | |
3073 | * real pointer, rather a tag number (watch the insertion code below). | |
3074 | * That means it pobably safe to assume it is well under the 32-bit limit, | |
3075 | * and makes the truncation safe. | |
3076 | * -- RAM, 14/09/1999 | |
3077 | */ | |
3078 | ||
3079 | svh = hv_fetch(hseen, (char *) &sv, sizeof(sv), FALSE); | |
3080 | if (svh) { | |
3081 | I32 tagval = htonl(LOW_32BITS(*svh)); | |
3082 | ||
9e21b3d0 | 3083 | TRACEME(("object 0x%"UVxf" seen as #%d", PTR2UV(sv), ntohl(tagval))); |
7a6a85bf RG |
3084 | |
3085 | PUTMARK(SX_OBJECT); | |
9e21b3d0 | 3086 | WRITE_I32(tagval); |
7a6a85bf RG |
3087 | return 0; |
3088 | } | |
3089 | ||
3090 | /* | |
3091 | * Allocate a new tag and associate it with the address of the sv being | |
3092 | * stored, before recursing... | |
3093 | * | |
3094 | * In order to avoid creating new SvIVs to hold the tagnum we just | |
d1be9408 | 3095 | * cast the tagnum to an SV pointer and store that in the hash. This |
7a6a85bf RG |
3096 | * means that we must clean up the hash manually afterwards, but gives |
3097 | * us a 15% throughput increase. | |
3098 | * | |
7a6a85bf RG |
3099 | */ |
3100 | ||
3101 | cxt->tagnum++; | |
3102 | if (!hv_store(hseen, | |
3341c981 | 3103 | (char *) &sv, sizeof(sv), INT2PTR(SV*, cxt->tagnum), 0)) |
7a6a85bf RG |
3104 | return -1; |
3105 | ||
3106 | /* | |
3107 | * Store `sv' and everything beneath it, using appropriate routine. | |
3108 | * Abort immediately if we get a non-zero status back. | |
3109 | */ | |
3110 | ||
3111 | type = sv_type(sv); | |
3112 | ||
43d061fe JH |
3113 | TRACEME(("storing 0x%"UVxf" tag #%d, type %d...", |
3114 | PTR2UV(sv), cxt->tagnum, type)); | |
7a6a85bf RG |
3115 | |
3116 | if (SvOBJECT(sv)) { | |
3117 | HV *pkg = SvSTASH(sv); | |
3118 | ret = store_blessed(cxt, sv, type, pkg); | |
3119 | } else | |
3120 | ret = SV_STORE(type)(cxt, sv); | |
3121 | ||
43d061fe JH |
3122 | TRACEME(("%s (stored 0x%"UVxf", refcnt=%d, %s)", |
3123 | ret ? "FAILED" : "ok", PTR2UV(sv), | |
7a6a85bf RG |
3124 | SvREFCNT(sv), sv_reftype(sv, FALSE))); |
3125 | ||
3126 | return ret; | |
3127 | } | |
3128 | ||
3129 | /* | |
3130 | * magic_write | |
3131 | * | |
3132 | * Write magic number and system information into the file. | |
3133 | * Layout is <magic> <network> [<len> <byteorder> <sizeof int> <sizeof long> | |
3134 | * <sizeof ptr>] where <len> is the length of the byteorder hexa string. | |
3135 | * All size and lenghts are written as single characters here. | |
3136 | * | |
3137 | * Note that no byte ordering info is emitted when <network> is true, since | |
3138 | * integers will be emitted in network order in that case. | |
3139 | */ | |
f0ffaed8 | 3140 | static int magic_write(stcxt_t *cxt) |
7a6a85bf | 3141 | { |
2aeb6432 NC |
3142 | /* |
3143 | * Starting with 0.6, the "use_network_order" byte flag is also used to | |
3144 | * indicate the version number of the binary image, encoded in the upper | |
3145 | * bits. The bit 0 is always used to indicate network order. | |
3146 | */ | |
3147 | /* | |
3148 | * Starting with 0.7, a full byte is dedicated to the minor version of | |
3149 | * the binary format, which is incremented only when new markers are | |
3150 | * introduced, for instance, but when backward compatibility is preserved. | |
3151 | */ | |
7a6a85bf | 3152 | |
2aeb6432 NC |
3153 | /* Make these at compile time. The WRITE() macro is sufficiently complex |
3154 | that it saves about 200 bytes doing it this way and only using it | |
3155 | once. */ | |
3156 | static const unsigned char network_file_header[] = { | |
3157 | MAGICSTR_BYTES, | |
3158 | (STORABLE_BIN_MAJOR << 1) | 1, | |
3159 | STORABLE_BIN_WRITE_MINOR | |
3160 | }; | |
3161 | static const unsigned char file_header[] = { | |
3162 | MAGICSTR_BYTES, | |
3163 | (STORABLE_BIN_MAJOR << 1) | 0, | |
3164 | STORABLE_BIN_WRITE_MINOR, | |
3165 | /* sizeof the array includes the 0 byte at the end: */ | |
3166 | (char) sizeof (byteorderstr) - 1, | |
3167 | BYTEORDER_BYTES, | |
3168 | (unsigned char) sizeof(int), | |
3169 | (unsigned char) sizeof(long), | |
3170 | (unsigned char) sizeof(char *), | |
3171 | (unsigned char) sizeof(NV) | |
3172 | }; | |
3173 | const unsigned char *header; | |
3174 | SSize_t length; | |
3175 | ||
3176 | TRACEME(("magic_write on fd=%d", cxt->fio ? PerlIO_fileno(cxt->fio) : -1)); | |
3177 | ||
3178 | if (cxt->netorder) { | |
3179 | header = network_file_header; | |
3180 | length = sizeof (network_file_header); | |
3181 | } else { | |
3182 | header = file_header; | |
3183 | length = sizeof (file_header); | |
3184 | } | |
3185 | ||
3186 | if (!cxt->fio) { | |
3187 | /* sizeof the array includes the 0 byte at the end. */ | |
3188 | header += sizeof (magicstr) - 1; | |
3189 | length -= sizeof (magicstr) - 1; | |
3190 | } | |
3191 | ||
3192 | WRITE(header, length); | |
3193 | ||
3194 | if (!cxt->netorder) { | |
9e21b3d0 | 3195 | TRACEME(("ok (magic_write byteorder = 0x%lx [%d], I%d L%d P%d D%d)", |
2aeb6432 | 3196 | (unsigned long) BYTEORDER, (int) sizeof (byteorderstr) - 1, |
9e21b3d0 JH |
3197 | (int) sizeof(int), (int) sizeof(long), |
3198 | (int) sizeof(char *), (int) sizeof(NV))); | |
2aeb6432 NC |
3199 | } |
3200 | return 0; | |
7a6a85bf RG |
3201 | } |
3202 | ||
3203 | /* | |
3204 | * do_store | |
3205 | * | |
3206 | * Common code for store operations. | |
3207 | * | |
3208 | * When memory store is requested (f = NULL) and a non null SV* is given in | |
3209 | * `res', it is filled with a new SV created out of the memory buffer. | |
3210 | * | |
3211 | * It is required to provide a non-null `res' when the operation type is not | |
3212 | * dclone() and store() is performed to memory. | |
3213 | */ | |
f0ffaed8 JH |
3214 | static int do_store( |
3215 | PerlIO *f, | |
3216 | SV *sv, | |
3217 | int optype, | |
3218 | int network_order, | |
3219 | SV **res) | |
7a6a85bf RG |
3220 | { |
3221 | dSTCXT; | |
3222 | int status; | |
3223 | ||
3224 | ASSERT(!(f == 0 && !(optype & ST_CLONE)) || res, | |
3225 | ("must supply result SV pointer for real recursion to memory")); | |
3226 | ||
3227 | TRACEME(("do_store (optype=%d, netorder=%d)", | |
3228 | optype, network_order)); | |
3229 | ||
3230 | optype |= ST_STORE; | |
3231 | ||
3232 | /* | |
3233 | * Workaround for CROAK leak: if they enter with a "dirty" context, | |
3234 | * free up memory for them now. | |
3235 | */ | |
3236 | ||
dd19458b | 3237 | if (cxt->s_dirty) |
7a6a85bf RG |
3238 | clean_context(cxt); |
3239 | ||
3240 | /* | |
3241 | * Now that STORABLE_xxx hooks exist, it is possible that they try to | |
3242 | * re-enter store() via the hooks. We need to stack contexts. | |
3243 | */ | |
3244 | ||
3245 | if (cxt->entry) | |
3246 | cxt = allocate_context(cxt); | |
3247 | ||
3248 | cxt->entry++; | |
3249 | ||
3250 | ASSERT(cxt->entry == 1, ("starting new recursion")); | |
dd19458b | 3251 | ASSERT(!cxt->s_dirty, ("clean context")); |
7a6a85bf RG |
3252 | |
3253 | /* | |
3254 | * Ensure sv is actually a reference. From perl, we called something | |
3255 | * like: | |
3256 | * pstore(FILE, \@array); | |
3257 | * so we must get the scalar value behing that reference. | |
3258 | */ | |
3259 | ||
3260 | if (!SvROK(sv)) | |
3261 | CROAK(("Not a reference")); | |
3262 | sv = SvRV(sv); /* So follow it to know what to store */ | |
3263 | ||
3264 | /* | |
3265 | * If we're going to store to memory, reset the buffer. | |
3266 | */ | |
3267 | ||
3268 | if (!f) | |
3269 | MBUF_INIT(0); | |
3270 | ||
3271 | /* | |
3272 | * Prepare context and emit headers. | |
3273 | */ | |
3274 | ||
3275 | init_store_context(cxt, f, optype, network_order); | |
3276 | ||
3277 | if (-1 == magic_write(cxt)) /* Emit magic and ILP info */ | |
3278 | return 0; /* Error */ | |
3279 | ||
3280 | /* | |
3281 | * Recursively store object... | |
3282 | */ | |
3283 | ||
3284 | ASSERT(is_storing(), ("within store operation")); | |
3285 | ||
3286 | status = store(cxt, sv); /* Just do it! */ | |
3287 | ||
3288 | /* | |
3289 | * If they asked for a memory store and they provided an SV pointer, | |
3290 | * make an SV string out of the buffer and fill their pointer. | |
3291 | * | |
3292 | * When asking for ST_REAL, it's MANDATORY for the caller to provide | |
3293 | * an SV, since context cleanup might free the buffer if we did recurse. | |
3294 | * (unless caller is dclone(), which is aware of that). | |
3295 | */ | |
3296 | ||
3297 | if (!cxt->fio && res) | |
3298 | *res = mbuf2sv(); | |
3299 | ||
3300 | /* | |
3301 | * Final cleanup. | |
3302 | * | |
3303 | * The "root" context is never freed, since it is meant to be always | |
3304 | * handy for the common case where no recursion occurs at all (i.e. | |
3305 | * we enter store() outside of any Storable code and leave it, period). | |
3306 | * We know it's the "root" context because there's nothing stacked | |
3307 | * underneath it. | |
3308 | * | |
3309 | * OPTIMIZATION: | |
3310 | * | |
3311 | * When deep cloning, we don't free the context: doing so would force | |
3312 | * us to copy the data in the memory buffer. Sicne we know we're | |
3313 | * about to enter do_retrieve... | |
3314 | */ | |
3315 | ||
3316 | clean_store_context(cxt); | |
3317 | if (cxt->prev && !(cxt->optype & ST_CLONE)) | |
3318 | free_context(cxt); | |
3319 | ||
3320 | TRACEME(("do_store returns %d", status)); | |
3321 | ||
3322 | return status == 0; | |
3323 | } | |
3324 | ||
3325 | /* | |
3326 | * pstore | |
3327 | * | |
3328 | * Store the transitive data closure of given object to disk. | |
3329 | * Returns 0 on error, a true value otherwise. | |
3330 | */ | |
f0ffaed8 | 3331 | int pstore(PerlIO *f, SV *sv) |
7a6a85bf RG |
3332 | { |
3333 | TRACEME(("pstore")); | |
f0ffaed8 | 3334 | return do_store(f, sv, 0, FALSE, (SV**) 0); |
7a6a85bf RG |
3335 | |
3336 | } | |
3337 | ||
3338 | /* | |
3339 | * net_pstore | |
3340 | * | |
3341 | * Same as pstore(), but network order is used for integers and doubles are | |
3342 | * emitted as strings. | |
3343 | */ | |
f0ffaed8 | 3344 | int net_pstore(PerlIO *f, SV *sv) |
7a6a85bf RG |
3345 | { |
3346 | TRACEME(("net_pstore")); | |
f0ffaed8 | 3347 | return do_store(f, sv, 0, TRUE, (SV**) 0); |
7a6a85bf RG |
3348 | } |
3349 | ||
3350 | /*** | |
3351 | *** Memory stores. | |
3352 | ***/ | |
3353 | ||
3354 | /* | |
3355 | * mbuf2sv | |
3356 | * | |
3357 | * Build a new SV out of the content of the internal memory buffer. | |
3358 | */ | |
f0ffaed8 | 3359 | static SV *mbuf2sv(void) |
7a6a85bf RG |
3360 | { |
3361 | dSTCXT; | |
3362 | ||
3363 | return newSVpv(mbase, MBUF_SIZE()); | |
3364 | } | |
3365 | ||
3366 | /* | |
3367 | * mstore | |
3368 | * | |
3369 | * Store the transitive data closure of given object to memory. | |
3370 | * Returns undef on error, a scalar value containing the data otherwise. | |
3371 | */ | |
f0ffaed8 | 3372 | SV *mstore(SV *sv) |
7a6a85bf | 3373 | { |
7a6a85bf RG |
3374 | SV *out; |
3375 | ||
3376 | TRACEME(("mstore")); | |
3377 | ||
f0ffaed8 | 3378 | if (!do_store((PerlIO*) 0, sv, 0, FALSE, &out)) |
7a6a85bf RG |
3379 | return &PL_sv_undef; |
3380 | ||
3381 | return out; | |
3382 | } | |
3383 | ||
3384 | /* | |
3385 | * net_mstore | |
3386 | * | |
3387 | * Same as mstore(), but network order is used for integers and doubles are | |
3388 | * emitted as strings. | |
3389 | */ | |
f0ffaed8 | 3390 | SV *net_mstore(SV *sv) |
7a6a85bf | 3391 | { |
7a6a85bf RG |
3392 | SV *out; |
3393 | ||
3394 | TRACEME(("net_mstore")); | |
3395 | ||
f0ffaed8 | 3396 | if (!do_store((PerlIO*) 0, sv, 0, TRUE, &out)) |
7a6a85bf RG |
3397 | return &PL_sv_undef; |
3398 | ||
3399 | return out; | |
3400 | } | |
3401 | ||
3402 | /*** | |
3403 | *** Specific retrieve callbacks. | |
3404 | ***/ | |
3405 | ||
3406 | /* | |
3407 | * retrieve_other | |
3408 | * | |
3409 | * Return an error via croak, since it is not possible that we get here | |
3410 | * under normal conditions, when facing a file produced via pstore(). | |
3411 | */ | |
b12202d0 | 3412 | static SV *retrieve_other(stcxt_t *cxt, char *cname) |
7a6a85bf RG |
3413 | { |
3414 | if ( | |
3415 | cxt->ver_major != STORABLE_BIN_MAJOR && | |
3416 | cxt->ver_minor != STORABLE_BIN_MINOR | |
3417 | ) { | |
3418 | CROAK(("Corrupted storable %s (binary v%d.%d), current is v%d.%d", | |
3419 | cxt->fio ? "file" : "string", | |
3420 | cxt->ver_major, cxt->ver_minor, | |
3421 | STORABLE_BIN_MAJOR, STORABLE_BIN_MINOR)); | |
3422 | } else { | |
3423 | CROAK(("Corrupted storable %s (binary v%d.%d)", | |
3424 | cxt->fio ? "file" : "string", | |
3425 | cxt->ver_major, cxt->ver_minor)); | |
3426 | } | |
3427 | ||
3428 | return (SV *) 0; /* Just in case */ | |
3429 | } | |
3430 | ||
3431 | /* | |
3432 | * retrieve_idx_blessed | |
3433 | * | |
3434 | * Layout is SX_IX_BLESS <index> <object> with SX_IX_BLESS already read. | |
3435 | * <index> can be coded on either 1 or 5 bytes. | |
3436 | */ | |
b12202d0 | 3437 | static SV *retrieve_idx_blessed(stcxt_t *cxt, char *cname) |
7a6a85bf RG |
3438 | { |
3439 | I32 idx; | |
3440 | char *class; | |
3441 | SV **sva; | |
3442 | SV *sv; | |
3443 | ||
3444 | TRACEME(("retrieve_idx_blessed (#%d)", cxt->tagnum)); | |
b12202d0 | 3445 | ASSERT(!cname, ("no bless-into class given here, got %s", cname)); |
7a6a85bf RG |
3446 | |
3447 | GETMARK(idx); /* Index coded on a single char? */ | |
3448 | if (idx & 0x80) | |
3449 | RLEN(idx); | |
3450 | ||
3451 | /* | |
3452 | * Fetch classname in `aclass' | |
3453 | */ | |
3454 | ||
3455 | sva = av_fetch(cxt->aclass, idx, FALSE); | |
3456 | if (!sva) | |
e993d95c | 3457 | CROAK(("Class name #%"IVdf" should have been seen already", (IV) idx)); |
7a6a85bf RG |
3458 | |
3459 | class = SvPVX(*sva); /* We know it's a PV, by construction */ | |
3460 | ||
3461 | TRACEME(("class ID %d => %s", idx, class)); | |
3462 | ||
3463 | /* | |
3464 | * Retrieve object and bless it. | |
3465 | */ | |
3466 | ||
b12202d0 | 3467 | sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */ |
7a6a85bf RG |
3468 | |
3469 | return sv; | |
3470 | } | |
3471 | ||
3472 | /* | |
3473 | * retrieve_blessed | |
3474 | * | |
3475 | * Layout is SX_BLESS <len> <classname> <object> with SX_BLESS already read. | |
3476 | * <len> can be coded on either 1 or 5 bytes. | |
3477 | */ | |
b12202d0 | 3478 | static SV *retrieve_blessed(stcxt_t *cxt, char *cname) |
7a6a85bf RG |
3479 | { |
3480 | I32 len; | |
3481 | SV *sv; | |
3482 | char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */ | |
3483 | char *class = buf; | |
3484 | ||
3485 | TRACEME(("retrieve_blessed (#%d)", cxt->tagnum)); | |
b12202d0 | 3486 | ASSERT(!cname, ("no bless-into class given here, got %s", cname)); |
7a6a85bf RG |
3487 | |
3488 | /* | |
3489 | * Decode class name length and read that name. | |
3490 | * | |
3491 | * Short classnames have two advantages: their length is stored on one | |
3492 | * single byte, and the string can be read on the stack. | |
3493 | */ | |
3494 | ||
3495 | GETMARK(len); /* Length coded on a single char? */ | |
3496 | if (len & 0x80) { | |
3497 | RLEN(len); | |
3498 | TRACEME(("** allocating %d bytes for class name", len+1)); | |
3499 | New(10003, class, len+1, char); | |
3500 | } | |
3501 | READ(class, len); | |
3502 | class[len] = '\0'; /* Mark string end */ | |
3503 | ||
3504 | /* | |
3505 | * It's a new classname, otherwise it would have been an SX_IX_BLESS. | |
3506 | */ | |
3507 | ||
b12202d0 JH |
3508 | TRACEME(("new class name \"%s\" will bear ID = %d", class, cxt->classnum)); |
3509 | ||
7a6a85bf RG |
3510 | if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len))) |
3511 | return (SV *) 0; | |
3512 | ||
3513 | /* | |
3514 | * Retrieve object and bless it. | |
3515 | */ | |
3516 | ||
b12202d0 JH |
3517 | sv = retrieve(cxt, class); /* First SV which is SEEN will be blessed */ |
3518 | if (class != buf) | |
3519 | Safefree(class); | |
7a6a85bf RG |
3520 | |
3521 | return sv; | |
3522 | } | |
3523 | ||
3524 | /* | |
3525 | * retrieve_hook | |
3526 | * | |
3527 | * Layout: SX_HOOK <flags> <len> <classname> <len2> <str> [<len3> <object-IDs>] | |
3528 | * with leading mark already read, as usual. | |
3529 | * | |
3530 | * When recursion was involved during serialization of the object, there | |
3531 | * is an unknown amount of serialized objects after the SX_HOOK mark. Until | |
3532 | * we reach a <flags> marker with the recursion bit cleared. | |
b12202d0 JH |
3533 | * |
3534 | * If the first <flags> byte contains a type of SHT_EXTRA, then the real type | |
3535 | * is held in the <extra> byte, and if the object is tied, the serialized | |
3536 | * magic object comes at the very end: | |
3537 | * | |
3538 | * SX_HOOK <flags> <extra> ... [<len3> <object-IDs>] <magic object> | |
3539 | * | |
3540 | * This means the STORABLE_thaw hook will NOT get a tied variable during its | |
3541 | * processing (since we won't have seen the magic object by the time the hook | |
3542 | * is called). See comments below for why it was done that way. | |
7a6a85bf | 3543 | */ |
b12202d0 | 3544 | static SV *retrieve_hook(stcxt_t *cxt, char *cname) |
7a6a85bf RG |
3545 | { |
3546 | I32 len; | |
3547 | char buf[LG_BLESS + 1]; /* Avoid malloc() if possible */ | |
3548 | char *class = buf; | |
3549 | unsigned int flags; | |
3550 | I32 len2; | |
3551 | SV *frozen; | |
3552 | I32 len3 = 0; | |
3553 | AV *av = 0; | |
3554 | SV *hook; | |
3555 | SV *sv; | |
3556 | SV *rv; | |
3557 | int obj_type; | |
7a6a85bf | 3558 | int clone = cxt->optype & ST_CLONE; |
b12202d0 JH |
3559 | char mtype = '\0'; |
3560 | unsigned int extra_type = 0; | |
7a6a85bf RG |
3561 | |
3562 | TRACEME(("retrieve_hook (#%d)", cxt->tagnum)); | |
b12202d0 | 3563 | ASSERT(!cname, ("no bless-into class given here, got %s", cname)); |
7a6a85bf RG |
3564 | |
3565 | /* | |
3566 | * Read flags, which tell us about the type, and whether we need to recurse. | |
3567 | */ | |
3568 | ||
3569 | GETMARK(flags); | |
3570 | ||
3571 | /* | |
3572 | * Create the (empty) object, and mark it as seen. | |
3573 | * | |
3574 | * This must be done now, because tags are incremented, and during | |
3575 | * serialization, the object tag was affected before recursion could | |
3576 | * take place. | |
3577 | */ | |
3578 | ||
3579 | obj_type = flags & SHF_TYPE_MASK; | |
3580 | switch (obj_type) { | |
3581 | case SHT_SCALAR: | |
3582 | sv = newSV(0); | |
3583 | break; | |
3584 | case SHT_ARRAY: | |
3585 | sv = (SV *) newAV(); | |
3586 | break; | |
3587 | case SHT_HASH: | |
3588 | sv = (SV *) newHV(); | |
3589 | break; | |
b12202d0 JH |
3590 | case SHT_EXTRA: |
3591 | /* | |
3592 | * Read <extra> flag to know the type of the object. | |
3593 | * Record associated magic type for later. | |
3594 | */ | |
3595 | GETMARK(extra_type); | |
3596 | switch (extra_type) { | |
3597 | case SHT_TSCALAR: | |
3598 | sv = newSV(0); | |
3599 | mtype = 'q'; | |
3600 | break; | |
3601 | case SHT_TARRAY: | |
3602 | sv = (SV *) newAV(); | |
3603 | mtype = 'P'; | |
3604 | break; | |
3605 | case SHT_THASH: | |
3606 | sv = (SV *) newHV(); | |
3607 | mtype = 'P'; | |
3608 | break; | |
3609 | default: | |
3610 | return retrieve_other(cxt, 0); /* Let it croak */ | |
3611 | } | |
3612 | break; | |
7a6a85bf | 3613 | default: |
b12202d0 | 3614 | return retrieve_other(cxt, 0); /* Let it croak */ |
7a6a85bf | 3615 | } |
b12202d0 | 3616 | SEEN(sv, 0); /* Don't bless yet */ |
7a6a85bf RG |
3617 | |
3618 | /* | |
3619 | * Whilst flags tell us to recurse, do so. | |
3620 | * | |
3621 | * We don't need to remember the addresses returned by retrieval, because | |
3622 | * all the references will be obtained through indirection via the object | |
3623 | * tags in the object-ID list. | |
3624 | */ | |
3625 | ||
3626 | while (flags & SHF_NEED_RECURSE) { | |
3627 | TRACEME(("retrieve_hook recursing...")); | |
b12202d0 | 3628 | rv = retrieve(cxt, 0); |
7a6a85bf RG |
3629 | if (!rv) |
3630 | return (SV *) 0; | |
43d061fe JH |
3631 | TRACEME(("retrieve_hook back with rv=0x%"UVxf, |
3632 | PTR2UV(rv))); | |
7a6a85bf RG |
3633 | GETMARK(flags); |
3634 | } | |
3635 | ||
3636 | if (flags & SHF_IDX_CLASSNAME) { | |
3637 | SV **sva; | |
3638 | I32 idx; | |
3639 | ||
3640 | /* | |
3641 | * Fetch index from `aclass' | |
3642 | */ | |
3643 | ||
3644 | if (flags & SHF_LARGE_CLASSLEN) | |
3645 | RLEN(idx); | |
3646 | else | |
3647 | GETMARK(idx); | |
3648 | ||
3649 | sva = av_fetch(cxt->aclass, idx, FALSE); | |
3650 | if (!sva) | |
e993d95c JH |
3651 | CROAK(("Class name #%"IVdf" should have been seen already", |
3652 | (IV) idx)); | |
7a6a85bf RG |
3653 | |
3654 | class = SvPVX(*sva); /* We know it's a PV, by construction */ | |
3655 | TRACEME(("class ID %d => %s", idx, class)); | |
3656 | ||
3657 | } else { | |
3658 | /* | |
3659 | * Decode class name length and read that name. | |
3660 | * | |
3661 | * NOTA BENE: even if the length is stored on one byte, we don't read | |
3662 | * on the stack. Just like retrieve_blessed(), we limit the name to | |
3663 | * LG_BLESS bytes. This is an arbitrary decision. | |
3664 | */ | |
3665 | ||
3666 | if (flags & SHF_LARGE_CLASSLEN) | |
3667 | RLEN(len); | |
3668 | else | |
3669 | GETMARK(len); | |
3670 | ||
3671 | if (len > LG_BLESS) { | |
3672 | TRACEME(("** allocating %d bytes for class name", len+1)); | |
3673 | New(10003, class, len+1, char); | |
3674 | } | |
3675 | ||
3676 | READ(class, len); | |
3677 | class[len] = '\0'; /* Mark string end */ | |
3678 | ||
3679 | /* | |
3680 | * Record new classname. | |
3681 | */ | |
3682 | ||
3683 | if (!av_store(cxt->aclass, cxt->classnum++, newSVpvn(class, len))) | |
3684 | return (SV *) 0; | |
3685 | } | |
3686 | ||
3687 | TRACEME(("class name: %s", class)); | |
3688 | ||
3689 | /* | |
d1be9408 | 3690 | * Decode user-frozen string length and read it in an SV. |
7a6a85bf RG |
3691 | * |
3692 | * For efficiency reasons, we read data directly into the SV buffer. | |
3693 | * To understand that code, read retrieve_scalar() | |
3694 | */ | |
3695 | ||
3696 | if (flags & SHF_LARGE_STRLEN) | |
3697 | RLEN(len2); | |
3698 | else | |
3699 | GETMARK(len2); | |
3700 | ||
3701 | frozen = NEWSV(10002, len2); | |
3702 | if (len2) { | |
3703 | SAFEREAD(SvPVX(frozen), len2, frozen); | |
3704 | SvCUR_set(frozen, len2); | |
3705 | *SvEND(frozen) = '\0'; | |
3706 | } | |
3707 | (void) SvPOK_only(frozen); /* Validates string pointer */ | |
dd19458b JH |
3708 | if (cxt->s_tainted) /* Is input source tainted? */ |
3709 | SvTAINT(frozen); | |
7a6a85bf RG |
3710 | |
3711 | TRACEME(("frozen string: %d bytes", len2)); | |
3712 | ||
3713 | /* | |
3714 | * Decode object-ID list length, if present. | |
3715 | */ | |
3716 | ||
3717 | if (flags & SHF_HAS_LIST) { | |
3718 | if (flags & SHF_LARGE_LISTLEN) | |
3719 | RLEN(len3); | |
3720 | else | |
3721 | GETMARK(len3); | |
3722 | if (len3) { | |
3723 | av = newAV(); | |
3724 | av_extend(av, len3 + 1); /* Leave room for [0] */ | |
3725 | AvFILLp(av) = len3; /* About to be filled anyway */ | |
3726 | } | |
3727 | } | |
3728 | ||
3729 | TRACEME(("has %d object IDs to link", len3)); | |
3730 | ||
3731 | /* | |
3732 | * Read object-ID list into array. | |
3733 | * Because we pre-extended it, we can cheat and fill it manually. | |
3734 | * | |
3735 | * We read object tags and we can convert them into SV* on the fly | |
3736 | * because we know all the references listed in there (as tags) | |
3737 | * have been already serialized, hence we have a valid correspondance | |
3738 | * between each of those tags and the recreated SV. | |
3739 | */ | |
3740 | ||
3741 | if (av) { | |
3742 | SV **ary = AvARRAY(av); | |
3743 | int i; | |
3744 | for (i = 1; i <= len3; i++) { /* We leave [0] alone */ | |
3745 | I32 tag; | |
3746 | SV **svh; | |
3747 | SV *xsv; | |
3748 | ||
9e21b3d0 | 3749 | READ_I32(tag); |
7a6a85bf RG |
3750 | tag = ntohl(tag); |
3751 | svh = av_fetch(cxt->aseen, tag, FALSE); | |
3752 | if (!svh) | |
e993d95c JH |
3753 | CROAK(("Object #%"IVdf" should have been retrieved already", |
3754 | (IV) tag)); | |
7a6a85bf RG |
3755 | xsv = *svh; |
3756 | ary[i] = SvREFCNT_inc(xsv); | |
3757 | } | |
3758 | } | |
3759 | ||
3760 | /* | |
3761 | * Bless the object and look up the STORABLE_thaw hook. | |
3762 | */ | |
3763 | ||
3764 | BLESS(sv, class); | |
3765 | hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw"); | |
212e9bde JH |
3766 | if (!hook) { |
3767 | /* | |
3768 | * Hook not found. Maybe they did not require the module where this | |
3769 | * hook is defined yet? | |
3770 | * | |
3771 | * If the require below succeeds, we'll be able to find the hook. | |
3772 | * Still, it only works reliably when each class is defined in a | |
3773 | * file of its own. | |
3774 | */ | |
3775 | ||
3776 | SV *psv = newSVpvn("require ", 8); | |
3777 | sv_catpv(psv, class); | |
3778 | ||
3779 | TRACEME(("No STORABLE_thaw defined for objects of class %s", class)); | |
3780 | TRACEME(("Going to require module '%s' with '%s'", class, SvPVX(psv))); | |
3781 | ||
3782 | perl_eval_sv(psv, G_DISCARD); | |
3783 | sv_free(psv); | |
3784 | ||
3785 | /* | |
3786 | * We cache results of pkg_can, so we need to uncache before attempting | |
3787 | * the lookup again. | |
3788 | */ | |
3789 | ||
3790 | pkg_uncache(cxt->hook, SvSTASH(sv), "STORABLE_thaw"); | |
3791 | hook = pkg_can(cxt->hook, SvSTASH(sv), "STORABLE_thaw"); | |
3792 | ||
3793 | if (!hook) | |
3794 | CROAK(("No STORABLE_thaw defined for objects of class %s " | |
3795 | "(even after a \"require %s;\")", class, class)); | |
3796 | } | |
7a6a85bf RG |
3797 | |
3798 | /* | |
3799 | * If we don't have an `av' yet, prepare one. | |
3800 | * Then insert the frozen string as item [0]. | |
3801 | */ | |
3802 | ||
3803 | if (!av) { | |
3804 | av = newAV(); | |
3805 | av_extend(av, 1); | |