Commit | Line | Data |
265a0548 |
1 | #define PERL_NO_GET_CONTEXT |
2 | |
e98cedbf |
3 | #include "EXTERN.h" |
4 | #include "perl.h" |
5 | #include "XSUB.h" |
2eb93d08 |
6 | #include "ppport.h" |
e98cedbf |
7 | |
87372f42 |
8 | /* Not yet in ppport.h */ |
9 | #ifndef CvISXSUB |
10 | # define CvISXSUB(cv) (CvXSUB(cv) ? TRUE : FALSE) |
11 | #endif |
0e1f978f |
12 | #ifndef SvRV_const |
13 | # define SvRV_const(rv) SvRV(rv) |
14 | #endif |
87372f42 |
15 | |
9fc9ab86 |
16 | #ifdef _MSC_VER |
1a36ac09 |
17 | /* "structured exception" handling is a Microsoft extension to C and C++. |
18 | It's *not* C++ exception handling - C++ exception handling can't capture |
19 | SEGVs and suchlike, whereas this can. There's no known analagous |
20 | functionality on other platforms. */ |
21 | # include <excpt.h> |
22 | # define TRY_TO_CATCH_SEGV __try |
23 | # define CAUGHT_EXCEPTION __except(EXCEPTION EXCEPTION_EXECUTE_HANDLER) |
9fc9ab86 |
24 | #else |
1a36ac09 |
25 | # define TRY_TO_CATCH_SEGV if(1) |
26 | # define CAUGHT_EXCEPTION else |
9fc9ab86 |
27 | #endif |
28 | |
29 | #ifdef __GNUC__ |
30 | # define __attribute__(x) |
31 | #endif |
32 | |
b7621729 |
33 | #if 0 && defined(DEBUGGING) |
34 | #define dbg_printf(x) printf x |
35 | #else |
36 | #define dbg_printf(x) |
37 | #endif |
98ecbbc6 |
38 | |
0964064b |
39 | #define TAG /* printf( "# %s(%d)\n", __FILE__, __LINE__ ) */ |
6a9ad7ec |
40 | #define carp puts |
9fc9ab86 |
41 | |
30fe4f47 |
42 | /* The idea is to have a tree structure to store 1 bit per possible pointer |
43 | address. The lowest 16 bits are stored in a block of 8092 bytes. |
44 | The blocks are in a 256-way tree, indexed by the reset of the pointer. |
45 | This can cope with 32 and 64 bit pointers, and any address space layout, |
46 | without excessive memory needs. The assumption is that your CPU cache |
47 | works :-) (And that we're not going to bust it) */ |
48 | |
30fe4f47 |
49 | #define BYTE_BITS 3 |
50 | #define LEAF_BITS (16 - BYTE_BITS) |
51 | #define LEAF_MASK 0x1FFF |
9fc9ab86 |
52 | |
65db36c0 |
53 | struct state { |
eee00145 |
54 | UV total_size; |
65db36c0 |
55 | bool regex_whine; |
56 | bool fm_whine; |
57 | bool dangle_whine; |
58 | bool go_yell; |
59 | /* My hunch (not measured) is that for most architectures pointers will |
60 | start with 0 bits, hence the start of this array will be hot, and the |
61 | end unused. So put the flags next to the hot end. */ |
62 | void *tracking[256]; |
63 | }; |
64 | |
9fc9ab86 |
65 | /* |
66 | Checks to see if thing is in the bitstring. |
67 | Returns true or false, and |
68 | notes thing in the segmented bitstring. |
69 | */ |
2eb93d08 |
70 | static bool |
a4efdff3 |
71 | check_new(struct state *st, const void *const p) { |
30fe4f47 |
72 | unsigned int bits = 8 * sizeof(void*); |
73 | const size_t raw_p = PTR2nat(p); |
74 | /* This effectively rotates the value right by the number of low always-0 |
75 | bits in an aligned pointer. The assmption is that most (if not all) |
76 | pointers are aligned, and these will be in the same chain of nodes |
77 | (and hence hot in the cache) but we can still deal with any unaligned |
78 | pointers. */ |
79 | const size_t cooked_p |
f404ed48 |
80 | = (raw_p >> ALIGN_BITS) | (raw_p << (bits - ALIGN_BITS)); |
30fe4f47 |
81 | const U8 this_bit = 1 << (cooked_p & 0x7); |
82 | U8 **leaf_p; |
83 | U8 *leaf; |
84 | unsigned int i; |
302077b6 |
85 | void **tv_p = (void **) (st->tracking); |
30fe4f47 |
86 | |
302077b6 |
87 | if (NULL == p) return FALSE; |
1a36ac09 |
88 | TRY_TO_CATCH_SEGV { |
2eb93d08 |
89 | const char c = *(const char *)p; |
9fc9ab86 |
90 | } |
1a36ac09 |
91 | CAUGHT_EXCEPTION { |
a4efdff3 |
92 | if (st->dangle_whine) |
9fc9ab86 |
93 | warn( "Devel::Size: Encountered invalid pointer: %p\n", p ); |
94 | return FALSE; |
95 | } |
9fc9ab86 |
96 | TAG; |
30fe4f47 |
97 | |
98 | bits -= 8; |
99 | /* bits now 24 (32 bit pointers) or 56 (64 bit pointers) */ |
100 | |
101 | /* First level is always present. */ |
102 | do { |
103 | i = (unsigned int)((cooked_p >> bits) & 0xFF); |
104 | if (!tv_p[i]) |
105 | Newxz(tv_p[i], 256, void *); |
106 | tv_p = (void **)(tv_p[i]); |
107 | bits -= 8; |
108 | } while (bits > LEAF_BITS + BYTE_BITS); |
109 | /* bits now 16 always */ |
5f04c81d |
110 | #if !defined(MULTIPLICITY) || PERL_VERSION > 8 || (PERL_VERSION == 8 && PERL_SUBVERSION > 8) |
111 | /* 5.8.8 and early have an assert() macro that uses Perl_croak, hence needs |
112 | a my_perl under multiplicity */ |
30fe4f47 |
113 | assert(bits == 16); |
5f04c81d |
114 | #endif |
30fe4f47 |
115 | leaf_p = (U8 **)tv_p; |
116 | i = (unsigned int)((cooked_p >> bits) & 0xFF); |
117 | if (!leaf_p[i]) |
118 | Newxz(leaf_p[i], 1 << LEAF_BITS, U8); |
119 | leaf = leaf_p[i]; |
120 | |
9fc9ab86 |
121 | TAG; |
30fe4f47 |
122 | |
123 | i = (unsigned int)((cooked_p >> BYTE_BITS) & LEAF_MASK); |
124 | |
125 | if(leaf[i] & this_bit) |
126 | return FALSE; |
127 | |
128 | leaf[i] |= this_bit; |
9fc9ab86 |
129 | return TRUE; |
130 | } |
131 | |
e9716740 |
132 | static void |
30fe4f47 |
133 | free_tracking_at(void **tv, int level) |
134 | { |
135 | int i = 255; |
136 | |
137 | if (--level) { |
138 | /* Nodes */ |
139 | do { |
140 | if (tv[i]) { |
141 | free_tracking_at(tv[i], level); |
142 | Safefree(tv[i]); |
143 | } |
144 | } while (i--); |
145 | } else { |
146 | /* Leaves */ |
147 | do { |
148 | if (tv[i]) |
149 | Safefree(tv[i]); |
150 | } while (i--); |
151 | } |
152 | } |
153 | |
154 | static void |
a4efdff3 |
155 | free_state(struct state *st) |
e9716740 |
156 | { |
30fe4f47 |
157 | const int top_level = (sizeof(void *) * 8 - LEAF_BITS - BYTE_BITS) / 8; |
a4efdff3 |
158 | free_tracking_at((void **)st->tracking, top_level); |
159 | Safefree(st); |
e9716740 |
160 | } |
161 | |
f3cf7e20 |
162 | /* For now, this is somewhat a compatibility bodge until the plan comes |
163 | together for fine grained recursion control. total_size() would recurse into |
164 | hash and array members, whereas sv_size() would not. However, sv_size() is |
165 | called with CvSTASH() of a CV, which means that if it (also) starts to |
166 | recurse fully, then the size of any CV now becomes the size of the entire |
167 | symbol table reachable from it, and potentially the entire symbol table, if |
168 | any subroutine makes a reference to a global (such as %SIG). The historical |
169 | implementation of total_size() didn't report "everything", and changing the |
170 | only available size to "everything" doesn't feel at all useful. */ |
171 | |
172 | #define NO_RECURSION 0 |
173 | #define SOME_RECURSION 1 |
174 | #define TOTAL_SIZE_RECURSION 2 |
175 | |
176 | static bool sv_size(pTHX_ struct state *, const SV *const, const int recurse); |
db519f11 |
177 | |
7ccc7d88 |
178 | typedef enum { |
9fc9ab86 |
179 | OPc_NULL, /* 0 */ |
180 | OPc_BASEOP, /* 1 */ |
181 | OPc_UNOP, /* 2 */ |
182 | OPc_BINOP, /* 3 */ |
183 | OPc_LOGOP, /* 4 */ |
184 | OPc_LISTOP, /* 5 */ |
185 | OPc_PMOP, /* 6 */ |
186 | OPc_SVOP, /* 7 */ |
187 | OPc_PADOP, /* 8 */ |
188 | OPc_PVOP, /* 9 */ |
189 | OPc_LOOP, /* 10 */ |
190 | OPc_COP /* 11 */ |
7ccc7d88 |
191 | } opclass; |
192 | |
193 | static opclass |
9fc9ab86 |
194 | cc_opclass(const OP * const o) |
7ccc7d88 |
195 | { |
196 | if (!o) |
9fc9ab86 |
197 | return OPc_NULL; |
1a36ac09 |
198 | TRY_TO_CATCH_SEGV { |
9fc9ab86 |
199 | if (o->op_type == 0) |
200 | return (o->op_flags & OPf_KIDS) ? OPc_UNOP : OPc_BASEOP; |
7ccc7d88 |
201 | |
9fc9ab86 |
202 | if (o->op_type == OP_SASSIGN) |
203 | return ((o->op_private & OPpASSIGN_BACKWARDS) ? OPc_UNOP : OPc_BINOP); |
7ccc7d88 |
204 | |
9fc9ab86 |
205 | #ifdef USE_ITHREADS |
206 | if (o->op_type == OP_GV || o->op_type == OP_GVSV || o->op_type == OP_AELEMFAST) |
207 | return OPc_PADOP; |
208 | #endif |
7ccc7d88 |
209 | |
9fc9ab86 |
210 | if ((o->op_type == OP_TRANS)) { |
211 | return OPc_BASEOP; |
212 | } |
7ccc7d88 |
213 | |
9fc9ab86 |
214 | switch (PL_opargs[o->op_type] & OA_CLASS_MASK) { |
215 | case OA_BASEOP: TAG; |
216 | return OPc_BASEOP; |
217 | |
218 | case OA_UNOP: TAG; |
219 | return OPc_UNOP; |
220 | |
221 | case OA_BINOP: TAG; |
222 | return OPc_BINOP; |
62691e7c |
223 | |
9fc9ab86 |
224 | case OA_LOGOP: TAG; |
225 | return OPc_LOGOP; |
7ccc7d88 |
226 | |
9fc9ab86 |
227 | case OA_LISTOP: TAG; |
228 | return OPc_LISTOP; |
7ccc7d88 |
229 | |
9fc9ab86 |
230 | case OA_PMOP: TAG; |
231 | return OPc_PMOP; |
7ccc7d88 |
232 | |
9fc9ab86 |
233 | case OA_SVOP: TAG; |
234 | return OPc_SVOP; |
7ccc7d88 |
235 | |
9fc9ab86 |
236 | case OA_PADOP: TAG; |
237 | return OPc_PADOP; |
7ccc7d88 |
238 | |
9fc9ab86 |
239 | case OA_PVOP_OR_SVOP: TAG; |
240 | /* |
241 | * Character translations (tr///) are usually a PVOP, keeping a |
242 | * pointer to a table of shorts used to look up translations. |
243 | * Under utf8, however, a simple table isn't practical; instead, |
244 | * the OP is an SVOP, and the SV is a reference to a swash |
245 | * (i.e., an RV pointing to an HV). |
246 | */ |
247 | return (o->op_private & (OPpTRANS_TO_UTF|OPpTRANS_FROM_UTF)) |
248 | ? OPc_SVOP : OPc_PVOP; |
7ccc7d88 |
249 | |
9fc9ab86 |
250 | case OA_LOOP: TAG; |
251 | return OPc_LOOP; |
7ccc7d88 |
252 | |
9fc9ab86 |
253 | case OA_COP: TAG; |
254 | return OPc_COP; |
7ccc7d88 |
255 | |
9fc9ab86 |
256 | case OA_BASEOP_OR_UNOP: TAG; |
7ccc7d88 |
257 | /* |
9fc9ab86 |
258 | * UNI(OP_foo) in toke.c returns token UNI or FUNC1 depending on |
259 | * whether parens were seen. perly.y uses OPf_SPECIAL to |
260 | * signal whether a BASEOP had empty parens or none. |
261 | * Some other UNOPs are created later, though, so the best |
262 | * test is OPf_KIDS, which is set in newUNOP. |
7ccc7d88 |
263 | */ |
9fc9ab86 |
264 | return (o->op_flags & OPf_KIDS) ? OPc_UNOP : OPc_BASEOP; |
265 | |
266 | case OA_FILESTATOP: TAG; |
267 | /* |
268 | * The file stat OPs are created via UNI(OP_foo) in toke.c but use |
269 | * the OPf_REF flag to distinguish between OP types instead of the |
270 | * usual OPf_SPECIAL flag. As usual, if OPf_KIDS is set, then we |
271 | * return OPc_UNOP so that walkoptree can find our children. If |
272 | * OPf_KIDS is not set then we check OPf_REF. Without OPf_REF set |
273 | * (no argument to the operator) it's an OP; with OPf_REF set it's |
274 | * an SVOP (and op_sv is the GV for the filehandle argument). |
275 | */ |
276 | return ((o->op_flags & OPf_KIDS) ? OPc_UNOP : |
277 | #ifdef USE_ITHREADS |
278 | (o->op_flags & OPf_REF) ? OPc_PADOP : OPc_BASEOP); |
279 | #else |
280 | (o->op_flags & OPf_REF) ? OPc_SVOP : OPc_BASEOP); |
281 | #endif |
282 | case OA_LOOPEXOP: TAG; |
283 | /* |
284 | * next, last, redo, dump and goto use OPf_SPECIAL to indicate that a |
285 | * label was omitted (in which case it's a BASEOP) or else a term was |
286 | * seen. In this last case, all except goto are definitely PVOP but |
287 | * goto is either a PVOP (with an ordinary constant label), an UNOP |
288 | * with OPf_STACKED (with a non-constant non-sub) or an UNOP for |
289 | * OP_REFGEN (with goto &sub) in which case OPf_STACKED also seems to |
290 | * get set. |
291 | */ |
292 | if (o->op_flags & OPf_STACKED) |
293 | return OPc_UNOP; |
294 | else if (o->op_flags & OPf_SPECIAL) |
295 | return OPc_BASEOP; |
296 | else |
297 | return OPc_PVOP; |
298 | } |
299 | warn("Devel::Size: Can't determine class of operator %s, assuming BASEOP\n", |
300 | PL_op_name[o->op_type]); |
301 | } |
1a36ac09 |
302 | CAUGHT_EXCEPTION { } |
7ccc7d88 |
303 | return OPc_BASEOP; |
304 | } |
305 | |
6a9ad7ec |
306 | |
a6ea0805 |
307 | #if !defined(NV) |
308 | #define NV double |
309 | #endif |
310 | |
6a9ad7ec |
311 | /* Figure out how much magic is attached to the SV and return the |
312 | size */ |
eee00145 |
313 | static void |
b7130948 |
314 | magic_size(pTHX_ const SV * const thing, struct state *st) { |
980c6576 |
315 | MAGIC *magic_pointer = SvMAGIC(thing); |
6a9ad7ec |
316 | |
980c6576 |
317 | /* Have we seen the magic pointer? (NULL has always been seen before) */ |
e5c69bdd |
318 | while (check_new(st, magic_pointer)) { |
eee00145 |
319 | st->total_size += sizeof(MAGIC); |
6a9ad7ec |
320 | |
1a36ac09 |
321 | TRY_TO_CATCH_SEGV { |
9fc9ab86 |
322 | /* Have we seen the magic vtable? */ |
e5c69bdd |
323 | if (check_new(st, magic_pointer->mg_virtual)) { |
eee00145 |
324 | st->total_size += sizeof(MGVTBL); |
9fc9ab86 |
325 | } |
b7130948 |
326 | sv_size(aTHX_ st, magic_pointer->mg_obj, TOTAL_SIZE_RECURSION); |
d1888d0b |
327 | if (magic_pointer->mg_len == HEf_SVKEY) { |
328 | sv_size(aTHX_ st, (SV *)magic_pointer->mg_ptr, TOTAL_SIZE_RECURSION); |
329 | } |
330 | #if defined(PERL_MAGIC_utf8) && defined (PERL_MAGIC_UTF8_CACHESIZE) |
331 | else if (magic_pointer->mg_type == PERL_MAGIC_utf8) { |
332 | if (check_new(st, magic_pointer->mg_ptr)) { |
333 | st->total_size += PERL_MAGIC_UTF8_CACHESIZE * 2 * sizeof(STRLEN); |
334 | } |
335 | } |
336 | #endif |
337 | else if (magic_pointer->mg_len > 0) { |
338 | if (check_new(st, magic_pointer->mg_ptr)) { |
339 | st->total_size += magic_pointer->mg_len; |
340 | } |
341 | } |
6a9ad7ec |
342 | |
0964064b |
343 | /* Get the next in the chain */ |
9fc9ab86 |
344 | magic_pointer = magic_pointer->mg_moremagic; |
345 | } |
1a36ac09 |
346 | CAUGHT_EXCEPTION { |
a4efdff3 |
347 | if (st->dangle_whine) |
9fc9ab86 |
348 | warn( "Devel::Size: Encountered bad magic at: %p\n", magic_pointer ); |
349 | } |
6a9ad7ec |
350 | } |
6a9ad7ec |
351 | } |
352 | |
eee00145 |
353 | static void |
99684fd4 |
354 | check_new_and_strlen(struct state *st, const char *const p) { |
355 | if(check_new(st, p)) |
6ec51ae0 |
356 | st->total_size += 1 + strlen(p); |
99684fd4 |
357 | } |
358 | |
359 | static void |
eee00145 |
360 | regex_size(const REGEXP * const baseregex, struct state *st) { |
c1bfd7da |
361 | if(!check_new(st, baseregex)) |
362 | return; |
eee00145 |
363 | st->total_size += sizeof(REGEXP); |
9fc9ab86 |
364 | #if (PERL_VERSION < 11) |
6ea94d90 |
365 | /* Note the size of the paren offset thing */ |
eee00145 |
366 | st->total_size += sizeof(I32) * baseregex->nparens * 2; |
367 | st->total_size += strlen(baseregex->precomp); |
6ea94d90 |
368 | #else |
eee00145 |
369 | st->total_size += sizeof(struct regexp); |
370 | st->total_size += sizeof(I32) * SvANY(baseregex)->nparens * 2; |
371 | /*st->total_size += strlen(SvANY(baseregex)->subbeg);*/ |
6ea94d90 |
372 | #endif |
a4efdff3 |
373 | if (st->go_yell && !st->regex_whine) { |
6ea94d90 |
374 | carp("Devel::Size: Calculated sizes for compiled regexes are incompatible, and probably always will be"); |
a4efdff3 |
375 | st->regex_whine = 1; |
98ecbbc6 |
376 | } |
7ccc7d88 |
377 | } |
378 | |
eee00145 |
379 | static void |
1e5a8ad2 |
380 | op_size(pTHX_ const OP * const baseop, struct state *st) |
381 | { |
382 | TRY_TO_CATCH_SEGV { |
383 | TAG; |
384 | if(!check_new(st, baseop)) |
385 | return; |
386 | TAG; |
387 | op_size(aTHX_ baseop->op_next, st); |
388 | TAG; |
389 | switch (cc_opclass(baseop)) { |
390 | case OPc_BASEOP: TAG; |
391 | st->total_size += sizeof(struct op); |
392 | TAG;break; |
393 | case OPc_UNOP: TAG; |
394 | st->total_size += sizeof(struct unop); |
395 | op_size(aTHX_ cUNOPx(baseop)->op_first, st); |
396 | TAG;break; |
397 | case OPc_BINOP: TAG; |
398 | st->total_size += sizeof(struct binop); |
399 | op_size(aTHX_ cBINOPx(baseop)->op_first, st); |
400 | op_size(aTHX_ cBINOPx(baseop)->op_last, st); |
401 | TAG;break; |
402 | case OPc_LOGOP: TAG; |
403 | st->total_size += sizeof(struct logop); |
404 | op_size(aTHX_ cBINOPx(baseop)->op_first, st); |
405 | op_size(aTHX_ cLOGOPx(baseop)->op_other, st); |
406 | TAG;break; |
407 | case OPc_LISTOP: TAG; |
408 | st->total_size += sizeof(struct listop); |
409 | op_size(aTHX_ cLISTOPx(baseop)->op_first, st); |
410 | op_size(aTHX_ cLISTOPx(baseop)->op_last, st); |
411 | TAG;break; |
412 | case OPc_PMOP: TAG; |
413 | st->total_size += sizeof(struct pmop); |
414 | op_size(aTHX_ cPMOPx(baseop)->op_first, st); |
415 | op_size(aTHX_ cPMOPx(baseop)->op_last, st); |
5a83b7cf |
416 | #if PERL_VERSION < 9 || (PERL_VERSION == 9 && PERL_SUBVERSION < 5) |
1e5a8ad2 |
417 | op_size(aTHX_ cPMOPx(baseop)->op_pmreplroot, st); |
418 | op_size(aTHX_ cPMOPx(baseop)->op_pmreplstart, st); |
419 | op_size(aTHX_ (OP *)cPMOPx(baseop)->op_pmnext, st); |
5a83b7cf |
420 | #endif |
c1bfd7da |
421 | /* This is defined away in perl 5.8.x, but it is in there for |
422 | 5.6.x */ |
98ecbbc6 |
423 | #ifdef PM_GETRE |
c1bfd7da |
424 | regex_size(PM_GETRE(cPMOPx(baseop)), st); |
98ecbbc6 |
425 | #else |
c1bfd7da |
426 | regex_size(cPMOPx(baseop)->op_pmregexp, st); |
98ecbbc6 |
427 | #endif |
c1bfd7da |
428 | TAG;break; |
81f1c018 |
429 | case OPc_SVOP: TAG; |
430 | st->total_size += sizeof(struct pmop); |
574d9fd9 |
431 | if (!(baseop->op_type == OP_AELEMFAST |
432 | && baseop->op_flags & OPf_SPECIAL)) { |
433 | /* not an OP_PADAV replacement */ |
f3cf7e20 |
434 | sv_size(aTHX_ st, cSVOPx(baseop)->op_sv, SOME_RECURSION); |
574d9fd9 |
435 | } |
81f1c018 |
436 | TAG;break; |
9fc9ab86 |
437 | case OPc_PADOP: TAG; |
eee00145 |
438 | st->total_size += sizeof(struct padop); |
99684fd4 |
439 | TAG;break; |
440 | case OPc_PVOP: TAG; |
441 | check_new_and_strlen(st, cPVOPx(baseop)->op_pv); |
219b7d34 |
442 | TAG;break; |
1e5a8ad2 |
443 | case OPc_LOOP: TAG; |
444 | st->total_size += sizeof(struct loop); |
445 | op_size(aTHX_ cLOOPx(baseop)->op_first, st); |
446 | op_size(aTHX_ cLOOPx(baseop)->op_last, st); |
447 | op_size(aTHX_ cLOOPx(baseop)->op_redoop, st); |
448 | op_size(aTHX_ cLOOPx(baseop)->op_nextop, st); |
449 | op_size(aTHX_ cLOOPx(baseop)->op_lastop, st); |
450 | TAG;break; |
451 | case OPc_COP: TAG; |
9fc9ab86 |
452 | { |
453 | COP *basecop; |
454 | basecop = (COP *)baseop; |
eee00145 |
455 | st->total_size += sizeof(struct cop); |
9fc9ab86 |
456 | |
457 | /* Change 33656 by nicholas@mouse-mill on 2008/04/07 11:29:51 |
458 | Eliminate cop_label from struct cop by storing a label as the first |
459 | entry in the hints hash. Most statements don't have labels, so this |
460 | will save memory. Not sure how much. |
461 | The check below will be incorrect fail on bleadperls |
462 | before 5.11 @33656, but later than 5.10, producing slightly too |
463 | small memory sizes on these Perls. */ |
b7621729 |
464 | #if (PERL_VERSION < 11) |
99684fd4 |
465 | check_new_and_strlen(st, basecop->cop_label); |
b7621729 |
466 | #endif |
7ccc7d88 |
467 | #ifdef USE_ITHREADS |
99684fd4 |
468 | check_new_and_strlen(st, basecop->cop_file); |
469 | check_new_and_strlen(st, basecop->cop_stashpv); |
7ccc7d88 |
470 | #else |
f3cf7e20 |
471 | sv_size(aTHX_ st, (SV *)basecop->cop_stash, SOME_RECURSION); |
472 | sv_size(aTHX_ st, (SV *)basecop->cop_filegv, SOME_RECURSION); |
7ccc7d88 |
473 | #endif |
474 | |
9fc9ab86 |
475 | } |
476 | TAG;break; |
477 | default: |
478 | TAG;break; |
479 | } |
480 | } |
1a36ac09 |
481 | CAUGHT_EXCEPTION { |
a4efdff3 |
482 | if (st->dangle_whine) |
9fc9ab86 |
483 | warn( "Devel::Size: Encountered dangling pointer in opcode at: %p\n", baseop ); |
7ccc7d88 |
484 | } |
7ccc7d88 |
485 | } |
6a9ad7ec |
486 | |
24d37977 |
487 | #if PERL_VERSION > 9 || (PERL_VERSION == 9 && PERL_SUBVERSION > 2) |
488 | # define NEW_HEAD_LAYOUT |
489 | #endif |
490 | |
81f1c018 |
491 | static bool |
db519f11 |
492 | sv_size(pTHX_ struct state *const st, const SV * const orig_thing, |
f3cf7e20 |
493 | const int recurse) { |
9fc9ab86 |
494 | const SV *thing = orig_thing; |
eee00145 |
495 | |
81f1c018 |
496 | if(!check_new(st, thing)) |
497 | return FALSE; |
498 | |
eee00145 |
499 | st->total_size += sizeof(SV); |
b1e5ad85 |
500 | |
e98cedbf |
501 | switch (SvTYPE(thing)) { |
502 | /* Is it undef? */ |
9fc9ab86 |
503 | case SVt_NULL: TAG; |
504 | TAG;break; |
e98cedbf |
505 | /* Just a plain integer. This will be differently sized depending |
506 | on whether purify's been compiled in */ |
9fc9ab86 |
507 | case SVt_IV: TAG; |
24d37977 |
508 | #ifndef NEW_HEAD_LAYOUT |
509 | # ifdef PURIFY |
eee00145 |
510 | st->total_size += sizeof(sizeof(XPVIV)); |
24d37977 |
511 | # else |
eee00145 |
512 | st->total_size += sizeof(IV); |
24d37977 |
513 | # endif |
e98cedbf |
514 | #endif |
81f1c018 |
515 | if(recurse && SvROK(thing)) |
f3cf7e20 |
516 | sv_size(aTHX_ st, SvRV_const(thing), recurse); |
9fc9ab86 |
517 | TAG;break; |
e98cedbf |
518 | /* Is it a float? Like the int, it depends on purify */ |
9fc9ab86 |
519 | case SVt_NV: TAG; |
e98cedbf |
520 | #ifdef PURIFY |
eee00145 |
521 | st->total_size += sizeof(sizeof(XPVNV)); |
e98cedbf |
522 | #else |
eee00145 |
523 | st->total_size += sizeof(NV); |
e98cedbf |
524 | #endif |
9fc9ab86 |
525 | TAG;break; |
526 | #if (PERL_VERSION < 11) |
e98cedbf |
527 | /* Is it a reference? */ |
9fc9ab86 |
528 | case SVt_RV: TAG; |
24d37977 |
529 | #ifndef NEW_HEAD_LAYOUT |
eee00145 |
530 | st->total_size += sizeof(XRV); |
24d37977 |
531 | #endif |
81f1c018 |
532 | if(recurse && SvROK(thing)) |
f3cf7e20 |
533 | sv_size(aTHX_ st, SvRV_const(thing), recurse); |
9fc9ab86 |
534 | TAG;break; |
6ea94d90 |
535 | #endif |
e98cedbf |
536 | /* How about a plain string? In which case we need to add in how |
537 | much has been allocated */ |
9fc9ab86 |
538 | case SVt_PV: TAG; |
eee00145 |
539 | st->total_size += sizeof(XPV); |
db519f11 |
540 | if(recurse && SvROK(thing)) |
f3cf7e20 |
541 | sv_size(aTHX_ st, SvRV_const(thing), recurse); |
eee00145 |
542 | else |
543 | st->total_size += SvLEN(thing); |
9fc9ab86 |
544 | TAG;break; |
e98cedbf |
545 | /* A string with an integer part? */ |
9fc9ab86 |
546 | case SVt_PVIV: TAG; |
eee00145 |
547 | st->total_size += sizeof(XPVIV); |
db519f11 |
548 | if(recurse && SvROK(thing)) |
f3cf7e20 |
549 | sv_size(aTHX_ st, SvRV_const(thing), recurse); |
eee00145 |
550 | else |
551 | st->total_size += SvLEN(thing); |
0430b7f7 |
552 | if(SvOOK(thing)) { |
eee00145 |
553 | st->total_size += SvIVX(thing); |
9fc9ab86 |
554 | } |
555 | TAG;break; |
c8db37d3 |
556 | /* A scalar/string/reference with a float part? */ |
9fc9ab86 |
557 | case SVt_PVNV: TAG; |
eee00145 |
558 | st->total_size += sizeof(XPVNV); |
db519f11 |
559 | if(recurse && SvROK(thing)) |
f3cf7e20 |
560 | sv_size(aTHX_ st, SvRV_const(thing), recurse); |
eee00145 |
561 | else |
562 | st->total_size += SvLEN(thing); |
9fc9ab86 |
563 | TAG;break; |
564 | case SVt_PVMG: TAG; |
eee00145 |
565 | st->total_size += sizeof(XPVMG); |
db519f11 |
566 | if(recurse && SvROK(thing)) |
f3cf7e20 |
567 | sv_size(aTHX_ st, SvRV_const(thing), recurse); |
eee00145 |
568 | else |
569 | st->total_size += SvLEN(thing); |
b7130948 |
570 | magic_size(aTHX_ thing, st); |
9fc9ab86 |
571 | TAG;break; |
0430b7f7 |
572 | #if PERL_VERSION <= 8 |
9fc9ab86 |
573 | case SVt_PVBM: TAG; |
eee00145 |
574 | st->total_size += sizeof(XPVBM); |
db519f11 |
575 | if(recurse && SvROK(thing)) |
f3cf7e20 |
576 | sv_size(aTHX_ st, SvRV_const(thing), recurse); |
eee00145 |
577 | else |
578 | st->total_size += SvLEN(thing); |
b7130948 |
579 | magic_size(aTHX_ thing, st); |
9fc9ab86 |
580 | TAG;break; |
0430b7f7 |
581 | #endif |
9fc9ab86 |
582 | case SVt_PVLV: TAG; |
eee00145 |
583 | st->total_size += sizeof(XPVLV); |
db519f11 |
584 | if(recurse && SvROK(thing)) |
f3cf7e20 |
585 | sv_size(aTHX_ st, SvRV_const(thing), recurse); |
eee00145 |
586 | else |
587 | st->total_size += SvLEN(thing); |
b7130948 |
588 | magic_size(aTHX_ thing, st); |
9fc9ab86 |
589 | TAG;break; |
e98cedbf |
590 | /* How much space is dedicated to the array? Not counting the |
591 | elements in the array, mind, just the array itself */ |
9fc9ab86 |
592 | case SVt_PVAV: TAG; |
eee00145 |
593 | st->total_size += sizeof(XPVAV); |
e98cedbf |
594 | /* Is there anything in the array? */ |
595 | if (AvMAX(thing) != -1) { |
c8db37d3 |
596 | /* an array with 10 slots has AvMax() set to 9 - te 2007-04-22 */ |
eee00145 |
597 | st->total_size += sizeof(SV *) * (AvMAX(thing) + 1); |
598 | dbg_printf(("total_size: %li AvMAX: %li av_len: $i\n", st->total_size, AvMAX(thing), av_len((AV*)thing))); |
6c5ddc0d |
599 | |
600 | if (recurse >= TOTAL_SIZE_RECURSION) { |
601 | SSize_t i = AvFILLp(thing) + 1; |
602 | |
603 | while (i--) |
604 | sv_size(aTHX_ st, AvARRAY(thing)[i], recurse); |
605 | } |
e98cedbf |
606 | } |
607 | /* Add in the bits on the other side of the beginning */ |
0430b7f7 |
608 | |
b7621729 |
609 | dbg_printf(("total_size %li, sizeof(SV *) %li, AvARRAY(thing) %li, AvALLOC(thing)%li , sizeof(ptr) %li \n", |
eee00145 |
610 | st->total_size, sizeof(SV*), AvARRAY(thing), AvALLOC(thing), sizeof( thing ))); |
0430b7f7 |
611 | |
612 | /* under Perl 5.8.8 64bit threading, AvARRAY(thing) was a pointer while AvALLOC was 0, |
b1e5ad85 |
613 | resulting in grossly overstated sized for arrays. Technically, this shouldn't happen... */ |
0430b7f7 |
614 | if (AvALLOC(thing) != 0) { |
eee00145 |
615 | st->total_size += (sizeof(SV *) * (AvARRAY(thing) - AvALLOC(thing))); |
0430b7f7 |
616 | } |
795fc84c |
617 | #if (PERL_VERSION < 9) |
618 | /* Is there something hanging off the arylen element? |
619 | Post 5.9.something this is stored in magic, so will be found there, |
620 | and Perl_av_arylen_p() takes a non-const AV*, hence compilers rightly |
621 | complain about AvARYLEN() passing thing to it. */ |
f3cf7e20 |
622 | sv_size(aTHX_ st, AvARYLEN(thing), recurse); |
795fc84c |
623 | #endif |
b7130948 |
624 | magic_size(aTHX_ thing, st); |
9fc9ab86 |
625 | TAG;break; |
626 | case SVt_PVHV: TAG; |
a6ea0805 |
627 | /* First the base struct */ |
eee00145 |
628 | st->total_size += sizeof(XPVHV); |
a6ea0805 |
629 | /* Now the array of buckets */ |
eee00145 |
630 | st->total_size += (sizeof(HE *) * (HvMAX(thing) + 1)); |
a6ea0805 |
631 | /* Now walk the bucket chain */ |
6a9ad7ec |
632 | if (HvARRAY(thing)) { |
a6ea0805 |
633 | HE *cur_entry; |
9fc9ab86 |
634 | UV cur_bucket = 0; |
a6ea0805 |
635 | for (cur_bucket = 0; cur_bucket <= HvMAX(thing); cur_bucket++) { |
9fc9ab86 |
636 | cur_entry = *(HvARRAY(thing) + cur_bucket); |
637 | while (cur_entry) { |
eee00145 |
638 | st->total_size += sizeof(HE); |
9fc9ab86 |
639 | if (cur_entry->hent_hek) { |
640 | /* Hash keys can be shared. Have we seen this before? */ |
a4efdff3 |
641 | if (check_new(st, cur_entry->hent_hek)) { |
eee00145 |
642 | st->total_size += HEK_BASESIZE + cur_entry->hent_hek->hek_len + 2; |
9fc9ab86 |
643 | } |
644 | } |
f3cf7e20 |
645 | if (recurse >= TOTAL_SIZE_RECURSION) |
646 | sv_size(aTHX_ st, HeVAL(cur_entry), recurse); |
9fc9ab86 |
647 | cur_entry = cur_entry->hent_next; |
648 | } |
a6ea0805 |
649 | } |
650 | } |
b7130948 |
651 | magic_size(aTHX_ thing, st); |
9fc9ab86 |
652 | TAG;break; |
653 | case SVt_PVCV: TAG; |
eee00145 |
654 | st->total_size += sizeof(XPVCV); |
b7130948 |
655 | magic_size(aTHX_ thing, st); |
7ccc7d88 |
656 | |
eee00145 |
657 | st->total_size += ((XPVIO *) SvANY(thing))->xpv_len; |
f3cf7e20 |
658 | sv_size(aTHX_ st, (SV *)CvSTASH(thing), SOME_RECURSION); |
659 | sv_size(aTHX_ st, (SV *)SvSTASH(thing), SOME_RECURSION); |
660 | sv_size(aTHX_ st, (SV *)CvGV(thing), SOME_RECURSION); |
6c5ddc0d |
661 | sv_size(aTHX_ st, (SV *)CvPADLIST(thing), SOME_RECURSION); |
f3cf7e20 |
662 | sv_size(aTHX_ st, (SV *)CvOUTSIDE(thing), recurse); |
66f50dda |
663 | if (CvISXSUB(thing)) { |
f3cf7e20 |
664 | sv_size(aTHX_ st, cv_const_sv((CV *)thing), recurse); |
66f50dda |
665 | } else { |
1e5a8ad2 |
666 | op_size(aTHX_ CvSTART(thing), st); |
667 | op_size(aTHX_ CvROOT(thing), st); |
7ccc7d88 |
668 | } |
669 | |
9fc9ab86 |
670 | TAG;break; |
671 | case SVt_PVGV: TAG; |
b7130948 |
672 | magic_size(aTHX_ thing, st); |
eee00145 |
673 | st->total_size += sizeof(XPVGV); |
4a3d023d |
674 | if(isGV_with_GP(thing)) { |
675 | st->total_size += GvNAMELEN(thing); |
78dfb4e7 |
676 | #ifdef GvFILE |
2b217e71 |
677 | # if !defined(USE_ITHREADS) || (PERL_VERSION > 8 || (PERL_VERSION == 8 && PERL_SUBVERSION > 8)) |
678 | /* With itreads, before 5.8.9, this can end up pointing to freed memory |
679 | if the GV was created in an eval, as GvFILE() points to CopFILE(), |
680 | and the relevant COP has been freed on scope cleanup after the eval. |
681 | 5.8.9 adds a binary compatible fudge that catches the vast majority |
682 | of cases. 5.9.something added a proper fix, by converting the GP to |
683 | use a shared hash key (porperly reference counted), instead of a |
684 | char * (owned by who knows? possibly no-one now) */ |
4a3d023d |
685 | check_new_and_strlen(st, GvFILE(thing)); |
2b217e71 |
686 | # endif |
78dfb4e7 |
687 | #endif |
4a3d023d |
688 | /* Is there something hanging off the glob? */ |
689 | if (check_new(st, GvGP(thing))) { |
690 | st->total_size += sizeof(GP); |
f3cf7e20 |
691 | sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_sv), recurse); |
692 | sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_form), recurse); |
693 | sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_av), recurse); |
694 | sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_hv), recurse); |
695 | sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_egv), recurse); |
696 | sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_cv), recurse); |
4a3d023d |
697 | } |
5c2e1b12 |
698 | } |
9fc9ab86 |
699 | TAG;break; |
700 | case SVt_PVFM: TAG; |
eee00145 |
701 | st->total_size += sizeof(XPVFM); |
b7130948 |
702 | magic_size(aTHX_ thing, st); |
eee00145 |
703 | st->total_size += ((XPVIO *) SvANY(thing))->xpv_len; |
6c5ddc0d |
704 | sv_size(aTHX_ st, (SV *)CvPADLIST(thing), SOME_RECURSION); |
f3cf7e20 |
705 | sv_size(aTHX_ st, (SV *)CvOUTSIDE(thing), recurse); |
7ccc7d88 |
706 | |
a4efdff3 |
707 | if (st->go_yell && !st->fm_whine) { |
5073b933 |
708 | carp("Devel::Size: Calculated sizes for FMs are incomplete"); |
a4efdff3 |
709 | st->fm_whine = 1; |
ebb2c5b9 |
710 | } |
9fc9ab86 |
711 | TAG;break; |
712 | case SVt_PVIO: TAG; |
eee00145 |
713 | st->total_size += sizeof(XPVIO); |
b7130948 |
714 | magic_size(aTHX_ thing, st); |
a4efdff3 |
715 | if (check_new(st, (SvPVX_const(thing)))) { |
eee00145 |
716 | st->total_size += ((XPVIO *) SvANY(thing))->xpv_cur; |
ebb2c5b9 |
717 | } |
5073b933 |
718 | /* Some embedded char pointers */ |
99684fd4 |
719 | check_new_and_strlen(st, ((XPVIO *) SvANY(thing))->xio_top_name); |
720 | check_new_and_strlen(st, ((XPVIO *) SvANY(thing))->xio_fmt_name); |
721 | check_new_and_strlen(st, ((XPVIO *) SvANY(thing))->xio_bottom_name); |
5073b933 |
722 | /* Throw the GVs on the list to be walked if they're not-null */ |
f3cf7e20 |
723 | sv_size(aTHX_ st, (SV *)((XPVIO *) SvANY(thing))->xio_top_gv, recurse); |
724 | sv_size(aTHX_ st, (SV *)((XPVIO *) SvANY(thing))->xio_bottom_gv, recurse); |
725 | sv_size(aTHX_ st, (SV *)((XPVIO *) SvANY(thing))->xio_fmt_gv, recurse); |
5073b933 |
726 | |
727 | /* Only go trotting through the IO structures if they're really |
728 | trottable. If USE_PERLIO is defined we can do this. If |
729 | not... we can't, so we don't even try */ |
730 | #ifdef USE_PERLIO |
731 | /* Dig into xio_ifp and xio_ofp here */ |
9fc9ab86 |
732 | warn("Devel::Size: Can't size up perlio layers yet\n"); |
5073b933 |
733 | #endif |
9fc9ab86 |
734 | TAG;break; |
e98cedbf |
735 | default: |
9fc9ab86 |
736 | warn("Devel::Size: Unknown variable type: %d encountered\n", SvTYPE(thing) ); |
e98cedbf |
737 | } |
81f1c018 |
738 | return TRUE; |
e98cedbf |
739 | } |
740 | |
ec404c23 |
741 | /* Frustratingly, the vtables aren't const in perl.h |
742 | gcc is happy enough to have non-const initialisers in a static array. |
743 | VC seems not to be. (Is it actually treating the file as C++?) |
744 | So do the maximally portable thing, unless we know it's gcc, in which case |
745 | we can do the more space efficient version. */ |
746 | |
747 | #if __GNUC__ |
d9b022a1 |
748 | void *vtables[] = { |
749 | #include "vtables.inc" |
750 | NULL |
751 | }; |
ec404c23 |
752 | #endif |
d9b022a1 |
753 | |
a4efdff3 |
754 | static struct state * |
755 | new_state(pTHX) |
65db36c0 |
756 | { |
757 | SV *warn_flag; |
a4efdff3 |
758 | struct state *st; |
ec404c23 |
759 | #if __GNUC__ |
d9b022a1 |
760 | void **vt_p = vtables; |
ec404c23 |
761 | #endif |
d9b022a1 |
762 | |
a4efdff3 |
763 | Newxz(st, 1, struct state); |
764 | st->go_yell = TRUE; |
65db36c0 |
765 | if (NULL != (warn_flag = perl_get_sv("Devel::Size::warn", FALSE))) { |
a4efdff3 |
766 | st->dangle_whine = st->go_yell = SvIV(warn_flag) ? TRUE : FALSE; |
65db36c0 |
767 | } |
768 | if (NULL != (warn_flag = perl_get_sv("Devel::Size::dangle", FALSE))) { |
a4efdff3 |
769 | st->dangle_whine = SvIV(warn_flag) ? TRUE : FALSE; |
65db36c0 |
770 | } |
a52ceccd |
771 | check_new(st, &PL_sv_undef); |
772 | check_new(st, &PL_sv_no); |
773 | check_new(st, &PL_sv_yes); |
ec404c23 |
774 | #if __GNUC__ |
d9b022a1 |
775 | while(*vt_p) |
776 | check_new(st, *vt_p++); |
ec404c23 |
777 | #else |
778 | #include "vtables.inc" |
779 | #endif |
a4efdff3 |
780 | return st; |
65db36c0 |
781 | } |
782 | |
9fc9ab86 |
783 | MODULE = Devel::Size PACKAGE = Devel::Size |
e98cedbf |
784 | |
fea63ffa |
785 | PROTOTYPES: DISABLE |
786 | |
eee00145 |
787 | UV |
a6ea0805 |
788 | size(orig_thing) |
789 | SV *orig_thing |
13683e3a |
790 | ALIAS: |
791 | total_size = TOTAL_SIZE_RECURSION |
e98cedbf |
792 | CODE: |
793 | { |
6a9ad7ec |
794 | SV *thing = orig_thing; |
a4efdff3 |
795 | struct state *st = new_state(aTHX); |
ebb2c5b9 |
796 | |
6a9ad7ec |
797 | /* If they passed us a reference then dereference it. This is the |
798 | only way we can check the sizes of arrays and hashes */ |
b7621729 |
799 | if (SvROK(thing)) { |
800 | thing = SvRV(thing); |
801 | } |
b7621729 |
802 | |
13683e3a |
803 | sv_size(aTHX_ st, thing, ix); |
eee00145 |
804 | RETVAL = st->total_size; |
a4efdff3 |
805 | free_state(st); |
6a9ad7ec |
806 | } |
807 | OUTPUT: |
808 | RETVAL |