X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=Size.xs;h=14be0be10aebb2a363a0dc8538d4dc9e49b8a41a;hb=ec404c2370ba877301e093207215b59f5de80a7a;hp=b849cc184e3de938e03b9d15ee06edc33e57546f;hpb=219b7d3497fb2c1bca6e63f521f61b725e5afc7d;p=p5sagit%2FDevel-Size.git diff --git a/Size.xs b/Size.xs index b849cc1..14be0be 100644 --- a/Size.xs +++ b/Size.xs @@ -46,7 +46,6 @@ without excessive memory needs. The assumption is that your CPU cache works :-) (And that we're not going to bust it) */ -#define ALIGN_BITS ( sizeof(void*) >> 1 ) #define BYTE_BITS 3 #define LEAF_BITS (16 - BYTE_BITS) #define LEAF_MASK 0x1FFF @@ -78,7 +77,7 @@ check_new(struct state *st, const void *const p) { (and hence hot in the cache) but we can still deal with any unaligned pointers. */ const size_t cooked_p - = (raw_p >> ALIGN_BITS) | (raw_p << (bits - BYTE_BITS)); + = (raw_p >> ALIGN_BITS) | (raw_p << (bits - ALIGN_BITS)); const U8 this_bit = 1 << (cooked_p & 0x7); U8 **leaf_p; U8 *leaf; @@ -160,7 +159,21 @@ free_state(struct state *st) Safefree(st); } -static bool sv_size(pTHX_ struct state *, const SV *const, bool recurse); +/* For now, this is somewhat a compatibility bodge until the plan comes + together for fine grained recursion control. total_size() would recurse into + hash and array members, whereas sv_size() would not. However, sv_size() is + called with CvSTASH() of a CV, which means that if it (also) starts to + recurse fully, then the size of any CV now becomes the size of the entire + symbol table reachable from it, and potentially the entire symbol table, if + any subroutine makes a reference to a global (such as %SIG). The historical + implementation of total_size() didn't report "everything", and changing the + only available size to "everything" doesn't feel at all useful. */ + +#define NO_RECURSION 0 +#define SOME_RECURSION 1 +#define TOTAL_SIZE_RECURSION 2 + +static bool sv_size(pTHX_ struct state *, const SV *const, const int recurse); typedef enum { OPc_NULL, /* 0 */ @@ -298,19 +311,10 @@ cc_opclass(const OP * const o) /* Figure out how much magic is attached to the SV and return the size */ static void -magic_size(const SV * const thing, struct state *st) { - MAGIC *magic_pointer; - - /* Is there any? */ - if (!SvMAGIC(thing)) { - /* No, bail */ - return; - } - - /* Get the base magic pointer */ - magic_pointer = SvMAGIC(thing); +magic_size(pTHX_ const SV * const thing, struct state *st) { + MAGIC *magic_pointer = SvMAGIC(thing); - /* Have we seen the magic pointer? */ + /* Have we seen the magic pointer? (NULL has always been seen before) */ while (check_new(st, magic_pointer)) { st->total_size += sizeof(MAGIC); @@ -319,6 +323,22 @@ magic_size(const SV * const thing, struct state *st) { if (check_new(st, magic_pointer->mg_virtual)) { st->total_size += sizeof(MGVTBL); } + sv_size(aTHX_ st, magic_pointer->mg_obj, TOTAL_SIZE_RECURSION); + if (magic_pointer->mg_len == HEf_SVKEY) { + sv_size(aTHX_ st, (SV *)magic_pointer->mg_ptr, TOTAL_SIZE_RECURSION); + } +#if defined(PERL_MAGIC_utf8) && defined (PERL_MAGIC_UTF8_CACHESIZE) + else if (magic_pointer->mg_type == PERL_MAGIC_utf8) { + if (check_new(st, magic_pointer->mg_ptr)) { + st->total_size += PERL_MAGIC_UTF8_CACHESIZE * 2 * sizeof(STRLEN); + } + } +#endif + else if (magic_pointer->mg_len > 0) { + if (check_new(st, magic_pointer->mg_ptr)) { + st->total_size += magic_pointer->mg_len; + } + } /* Get the next in the chain */ magic_pointer = magic_pointer->mg_moremagic; @@ -411,7 +431,7 @@ op_size(pTHX_ const OP * const baseop, struct state *st) if (!(baseop->op_type == OP_AELEMFAST && baseop->op_flags & OPf_SPECIAL)) { /* not an OP_PADAV replacement */ - sv_size(aTHX_ st, cSVOPx(baseop)->op_sv, TRUE); + sv_size(aTHX_ st, cSVOPx(baseop)->op_sv, SOME_RECURSION); } TAG;break; case OPc_PADOP: TAG; @@ -448,8 +468,8 @@ op_size(pTHX_ const OP * const baseop, struct state *st) check_new_and_strlen(st, basecop->cop_file); check_new_and_strlen(st, basecop->cop_stashpv); #else - sv_size(aTHX_ st, (SV *)basecop->cop_stash, TRUE); - sv_size(aTHX_ st, (SV *)basecop->cop_filegv, TRUE); + sv_size(aTHX_ st, (SV *)basecop->cop_stash, SOME_RECURSION); + sv_size(aTHX_ st, (SV *)basecop->cop_filegv, SOME_RECURSION); #endif } @@ -470,7 +490,7 @@ op_size(pTHX_ const OP * const baseop, struct state *st) static bool sv_size(pTHX_ struct state *const st, const SV * const orig_thing, - const bool recurse) { + const int recurse) { const SV *thing = orig_thing; if(!check_new(st, thing)) @@ -493,7 +513,7 @@ sv_size(pTHX_ struct state *const st, const SV * const orig_thing, # endif #endif if(recurse && SvROK(thing)) - sv_size(aTHX_ st, SvRV_const(thing), TRUE); + sv_size(aTHX_ st, SvRV_const(thing), recurse); TAG;break; /* Is it a float? Like the int, it depends on purify */ case SVt_NV: TAG; @@ -510,7 +530,7 @@ sv_size(pTHX_ struct state *const st, const SV * const orig_thing, st->total_size += sizeof(XRV); #endif if(recurse && SvROK(thing)) - sv_size(aTHX_ st, SvRV_const(thing), TRUE); + sv_size(aTHX_ st, SvRV_const(thing), recurse); TAG;break; #endif /* How about a plain string? In which case we need to add in how @@ -518,7 +538,7 @@ sv_size(pTHX_ struct state *const st, const SV * const orig_thing, case SVt_PV: TAG; st->total_size += sizeof(XPV); if(recurse && SvROK(thing)) - sv_size(aTHX_ st, SvRV_const(thing), TRUE); + sv_size(aTHX_ st, SvRV_const(thing), recurse); else st->total_size += SvLEN(thing); TAG;break; @@ -526,7 +546,7 @@ sv_size(pTHX_ struct state *const st, const SV * const orig_thing, case SVt_PVIV: TAG; st->total_size += sizeof(XPVIV); if(recurse && SvROK(thing)) - sv_size(aTHX_ st, SvRV_const(thing), TRUE); + sv_size(aTHX_ st, SvRV_const(thing), recurse); else st->total_size += SvLEN(thing); if(SvOOK(thing)) { @@ -537,35 +557,35 @@ sv_size(pTHX_ struct state *const st, const SV * const orig_thing, case SVt_PVNV: TAG; st->total_size += sizeof(XPVNV); if(recurse && SvROK(thing)) - sv_size(aTHX_ st, SvRV_const(thing), TRUE); + sv_size(aTHX_ st, SvRV_const(thing), recurse); else st->total_size += SvLEN(thing); TAG;break; case SVt_PVMG: TAG; st->total_size += sizeof(XPVMG); if(recurse && SvROK(thing)) - sv_size(aTHX_ st, SvRV_const(thing), TRUE); + sv_size(aTHX_ st, SvRV_const(thing), recurse); else st->total_size += SvLEN(thing); - magic_size(thing, st); + magic_size(aTHX_ thing, st); TAG;break; #if PERL_VERSION <= 8 case SVt_PVBM: TAG; st->total_size += sizeof(XPVBM); if(recurse && SvROK(thing)) - sv_size(aTHX_ st, SvRV_const(thing), TRUE); + sv_size(aTHX_ st, SvRV_const(thing), recurse); else st->total_size += SvLEN(thing); - magic_size(thing, st); + magic_size(aTHX_ thing, st); TAG;break; #endif case SVt_PVLV: TAG; st->total_size += sizeof(XPVLV); if(recurse && SvROK(thing)) - sv_size(aTHX_ st, SvRV_const(thing), TRUE); + sv_size(aTHX_ st, SvRV_const(thing), recurse); else st->total_size += SvLEN(thing); - magic_size(thing, st); + magic_size(aTHX_ thing, st); TAG;break; /* How much space is dedicated to the array? Not counting the elements in the array, mind, just the array itself */ @@ -576,6 +596,13 @@ sv_size(pTHX_ struct state *const st, const SV * const orig_thing, /* an array with 10 slots has AvMax() set to 9 - te 2007-04-22 */ st->total_size += sizeof(SV *) * (AvMAX(thing) + 1); dbg_printf(("total_size: %li AvMAX: %li av_len: $i\n", st->total_size, AvMAX(thing), av_len((AV*)thing))); + + if (recurse >= TOTAL_SIZE_RECURSION) { + SSize_t i = AvFILLp(thing) + 1; + + while (i--) + sv_size(aTHX_ st, AvARRAY(thing)[i], recurse); + } } /* Add in the bits on the other side of the beginning */ @@ -592,9 +619,9 @@ sv_size(pTHX_ struct state *const st, const SV * const orig_thing, Post 5.9.something this is stored in magic, so will be found there, and Perl_av_arylen_p() takes a non-const AV*, hence compilers rightly complain about AvARYLEN() passing thing to it. */ - sv_size(aTHX_ st, AvARYLEN(thing), TRUE); + sv_size(aTHX_ st, AvARYLEN(thing), recurse); #endif - magic_size(thing, st); + magic_size(aTHX_ thing, st); TAG;break; case SVt_PVHV: TAG; /* First the base struct */ @@ -615,24 +642,26 @@ sv_size(pTHX_ struct state *const st, const SV * const orig_thing, st->total_size += HEK_BASESIZE + cur_entry->hent_hek->hek_len + 2; } } + if (recurse >= TOTAL_SIZE_RECURSION) + sv_size(aTHX_ st, HeVAL(cur_entry), recurse); cur_entry = cur_entry->hent_next; } } } - magic_size(thing, st); + magic_size(aTHX_ thing, st); TAG;break; case SVt_PVCV: TAG; st->total_size += sizeof(XPVCV); - magic_size(thing, st); + magic_size(aTHX_ thing, st); st->total_size += ((XPVIO *) SvANY(thing))->xpv_len; - sv_size(aTHX_ st, (SV *)CvSTASH(thing), TRUE); - sv_size(aTHX_ st, (SV *)SvSTASH(thing), TRUE); - sv_size(aTHX_ st, (SV *)CvGV(thing), TRUE); - sv_size(aTHX_ st, (SV *)CvPADLIST(thing), TRUE); - sv_size(aTHX_ st, (SV *)CvOUTSIDE(thing), TRUE); + sv_size(aTHX_ st, (SV *)CvSTASH(thing), SOME_RECURSION); + sv_size(aTHX_ st, (SV *)SvSTASH(thing), SOME_RECURSION); + sv_size(aTHX_ st, (SV *)CvGV(thing), SOME_RECURSION); + sv_size(aTHX_ st, (SV *)CvPADLIST(thing), SOME_RECURSION); + sv_size(aTHX_ st, (SV *)CvOUTSIDE(thing), recurse); if (CvISXSUB(thing)) { - sv_size(aTHX_ st, cv_const_sv((CV *)thing), TRUE); + sv_size(aTHX_ st, cv_const_sv((CV *)thing), recurse); } else { op_size(aTHX_ CvSTART(thing), st); op_size(aTHX_ CvROOT(thing), st); @@ -640,32 +669,40 @@ sv_size(pTHX_ struct state *const st, const SV * const orig_thing, TAG;break; case SVt_PVGV: TAG; - magic_size(thing, st); + magic_size(aTHX_ thing, st); st->total_size += sizeof(XPVGV); - st->total_size += GvNAMELEN(thing); + if(isGV_with_GP(thing)) { + st->total_size += GvNAMELEN(thing); #ifdef GvFILE - /* Is there a file? */ - check_new_and_strlen(st, GvFILE(thing)); +# if !defined(USE_ITHREADS) || (PERL_VERSION > 8 || (PERL_VERSION == 8 && PERL_SUBVERSION > 8)) + /* With itreads, before 5.8.9, this can end up pointing to freed memory + if the GV was created in an eval, as GvFILE() points to CopFILE(), + and the relevant COP has been freed on scope cleanup after the eval. + 5.8.9 adds a binary compatible fudge that catches the vast majority + of cases. 5.9.something added a proper fix, by converting the GP to + use a shared hash key (porperly reference counted), instead of a + char * (owned by who knows? possibly no-one now) */ + check_new_and_strlen(st, GvFILE(thing)); +# endif #endif - /* Is there something hanging off the glob? */ - if (GvGP(thing)) { - if (check_new(st, GvGP(thing))) { - st->total_size += sizeof(GP); - sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_sv), TRUE); - sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_form), TRUE); - sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_av), TRUE); - sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_hv), TRUE); - sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_egv), TRUE); - sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_cv), TRUE); - } + /* Is there something hanging off the glob? */ + if (check_new(st, GvGP(thing))) { + st->total_size += sizeof(GP); + sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_sv), recurse); + sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_form), recurse); + sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_av), recurse); + sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_hv), recurse); + sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_egv), recurse); + sv_size(aTHX_ st, (SV *)(GvGP(thing)->gp_cv), recurse); + } } TAG;break; case SVt_PVFM: TAG; st->total_size += sizeof(XPVFM); - magic_size(thing, st); + magic_size(aTHX_ thing, st); st->total_size += ((XPVIO *) SvANY(thing))->xpv_len; - sv_size(aTHX_ st, (SV *)CvPADLIST(thing), TRUE); - sv_size(aTHX_ st, (SV *)CvOUTSIDE(thing), TRUE); + sv_size(aTHX_ st, (SV *)CvPADLIST(thing), SOME_RECURSION); + sv_size(aTHX_ st, (SV *)CvOUTSIDE(thing), recurse); if (st->go_yell && !st->fm_whine) { carp("Devel::Size: Calculated sizes for FMs are incomplete"); @@ -674,7 +711,7 @@ sv_size(pTHX_ struct state *const st, const SV * const orig_thing, TAG;break; case SVt_PVIO: TAG; st->total_size += sizeof(XPVIO); - magic_size(thing, st); + magic_size(aTHX_ thing, st); if (check_new(st, (SvPVX_const(thing)))) { st->total_size += ((XPVIO *) SvANY(thing))->xpv_cur; } @@ -683,9 +720,9 @@ sv_size(pTHX_ struct state *const st, const SV * const orig_thing, check_new_and_strlen(st, ((XPVIO *) SvANY(thing))->xio_fmt_name); check_new_and_strlen(st, ((XPVIO *) SvANY(thing))->xio_bottom_name); /* Throw the GVs on the list to be walked if they're not-null */ - sv_size(aTHX_ st, (SV *)((XPVIO *) SvANY(thing))->xio_top_gv, TRUE); - sv_size(aTHX_ st, (SV *)((XPVIO *) SvANY(thing))->xio_bottom_gv, TRUE); - sv_size(aTHX_ st, (SV *)((XPVIO *) SvANY(thing))->xio_fmt_gv, TRUE); + sv_size(aTHX_ st, (SV *)((XPVIO *) SvANY(thing))->xio_top_gv, recurse); + sv_size(aTHX_ st, (SV *)((XPVIO *) SvANY(thing))->xio_bottom_gv, recurse); + sv_size(aTHX_ st, (SV *)((XPVIO *) SvANY(thing))->xio_fmt_gv, recurse); /* Only go trotting through the IO structures if they're really trottable. If USE_PERLIO is defined we can do this. If @@ -701,11 +738,28 @@ sv_size(pTHX_ struct state *const st, const SV * const orig_thing, return TRUE; } +/* Frustratingly, the vtables aren't const in perl.h + gcc is happy enough to have non-const initialisers in a static array. + VC seems not to be. (Is it actually treating the file as C++?) + So do the maximally portable thing, unless we know it's gcc, in which case + we can do the more space efficient version. */ + +#if __GNUC__ +void *vtables[] = { +#include "vtables.inc" + NULL +}; +#endif + static struct state * new_state(pTHX) { SV *warn_flag; struct state *st; +#if __GNUC__ + void **vt_p = vtables; +#endif + Newxz(st, 1, struct state); st->go_yell = TRUE; if (NULL != (warn_flag = perl_get_sv("Devel::Size::warn", FALSE))) { @@ -717,6 +771,12 @@ new_state(pTHX) check_new(st, &PL_sv_undef); check_new(st, &PL_sv_no); check_new(st, &PL_sv_yes); +#if __GNUC__ + while(*vt_p) + check_new(st, *vt_p++); +#else +#include "vtables.inc" +#endif return st; } @@ -727,6 +787,8 @@ PROTOTYPES: DISABLE UV size(orig_thing) SV *orig_thing +ALIAS: + total_size = TOTAL_SIZE_RECURSION CODE: { SV *thing = orig_thing; @@ -734,147 +796,13 @@ CODE: /* If they passed us a reference then dereference it. This is the only way we can check the sizes of arrays and hashes */ -#if (PERL_VERSION < 11) - if (SvOK(thing) && SvROK(thing)) { - thing = SvRV(thing); - } -#else if (SvROK(thing)) { thing = SvRV(thing); } -#endif - sv_size(aTHX_ st, thing, FALSE); + sv_size(aTHX_ st, thing, ix); RETVAL = st->total_size; free_state(st); } OUTPUT: RETVAL - - -UV -total_size(orig_thing) - SV *orig_thing -CODE: -{ - SV *thing = orig_thing; - /* Array with things we still need to do */ - AV *pending_array; - IV size = 0; - struct state *st = new_state(aTHX); - - /* Size starts at zero */ - RETVAL = 0; - - pending_array = newAV(); - - /* If they passed us a reference then dereference it. - This is the only way we can check the sizes of arrays and hashes. */ - if (SvROK(thing)) { - thing = SvRV(thing); - } - - /* Put it on the pending array */ - av_push(pending_array, thing); - - /* Now just yank things off the end of the array until it's done */ - while (av_len(pending_array) >= 0) { - thing = av_pop(pending_array); - /* Process it if we've not seen it */ - if (sv_size(aTHX_ st, thing, TRUE)) { - dbg_printf(("# Found type %i at %p\n", SvTYPE(thing), thing)); - switch (SvTYPE(thing)) { - /* fix for bug #24846 (Does not correctly recurse into references in a PVNV-type scalar) */ - case SVt_PVNV: TAG; - if (SvROK(thing)) - { - av_push(pending_array, SvRV(thing)); - } - TAG;break; -#if (PERL_VERSION < 11) - case SVt_RV: TAG; -#else - case SVt_IV: TAG; -#endif - dbg_printf(("# Found RV\n")); - if (SvROK(thing)) { - dbg_printf(("# Found RV\n")); - av_push(pending_array, SvRV(thing)); - } - TAG;break; - - case SVt_PVAV: TAG; - { - AV *tempAV = (AV *)thing; - SV **tempSV; - - dbg_printf(("# Found type AV\n")); - /* Quick alias to cut down on casting */ - - /* Any elements? */ - if (av_len(tempAV) != -1) { - IV index; - /* Run through them all */ - for (index = 0; index <= av_len(tempAV); index++) { - /* Did we get something? */ - if ((tempSV = av_fetch(tempAV, index, 0))) { - /* Was it undef? */ - if (*tempSV != &PL_sv_undef) { - /* Apparently not. Save it for later */ - av_push(pending_array, *tempSV); - } - } - } - } - } - TAG;break; - - case SVt_PVHV: TAG; - dbg_printf(("# Found type HV\n")); - /* Is there anything in here? */ - if (hv_iterinit((HV *)thing)) { - HE *temp_he; - while ((temp_he = hv_iternext((HV *)thing))) { - av_push(pending_array, hv_iterval((HV *)thing, temp_he)); - } - } - TAG;break; - - case SVt_PVGV: TAG; - dbg_printf(("# Found type GV\n")); - /* Run through all the pieces and push the ones with bits */ - if (GvSV(thing)) { - av_push(pending_array, (SV *)GvSV(thing)); - } - if (GvFORM(thing)) { - av_push(pending_array, (SV *)GvFORM(thing)); - } - if (GvAV(thing)) { - av_push(pending_array, (SV *)GvAV(thing)); - } - if (GvHV(thing)) { - av_push(pending_array, (SV *)GvHV(thing)); - } - if (GvCV(thing)) { - av_push(pending_array, (SV *)GvCV(thing)); - } - TAG;break; - default: - TAG;break; - } - } else { - /* check_new() returned false: */ -#ifdef DEVEL_SIZE_DEBUGGING - if (SvOK(sv)) printf("# Ignore ref copy 0x%x\n", sv); - else printf("# Ignore non-sv 0x%x\n", sv); -#endif - } - } /* end while */ - - RETVAL = st->total_size; - free_state(st); - SvREFCNT_dec(pending_array); -} -OUTPUT: - RETVAL -