X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=Size.xs;h=2d4f39b574bbf4ede4c38c4a28a0a45392fcda7c;hb=eb73dc893b6bf09bce7b81b00dd3818882ba2b4c;hp=14be0be10aebb2a363a0dc8538d4dc9e49b8a41a;hpb=ec404c2370ba877301e093207215b59f5de80a7a;p=p5sagit%2FDevel-Size.git diff --git a/Size.xs b/Size.xs index 14be0be..2d4f39b 100644 --- a/Size.xs +++ b/Size.xs @@ -1,3 +1,8 @@ +/* -*- mode: C -*- */ + +#undef NDEBUG /* XXX */ +#include + #define PERL_NO_GET_CONTEXT #include "EXTERN.h" @@ -12,6 +17,25 @@ #ifndef SvRV_const # define SvRV_const(rv) SvRV(rv) #endif +#ifndef SvOOK_offset +# define SvOOK_offset(sv, len) STMT_START { len = SvIVX(sv); } STMT_END +#endif +#ifndef SvIsCOW +# define SvIsCOW(sv) ((SvFLAGS(sv) & (SVf_FAKE | SVf_READONLY)) == \ + (SVf_FAKE | SVf_READONLY)) +#endif +#ifndef SvIsCOW_shared_hash +# define SvIsCOW_shared_hash(sv) (SvIsCOW(sv) && SvLEN(sv) == 0) +#endif +#ifndef SvSHARED_HEK_FROM_PV +# define SvSHARED_HEK_FROM_PV(pvx) \ + ((struct hek*)(pvx - STRUCT_OFFSET(struct hek, hek_key))) +#endif + +#if PERL_VERSION < 6 +# define PL_opargs opargs +# define PL_op_name op_name +#endif #ifdef _MSC_VER /* "structured exception" handling is a Microsoft extension to C and C++. @@ -20,7 +44,7 @@ functionality on other platforms. */ # include # define TRY_TO_CATCH_SEGV __try -# define CAUGHT_EXCEPTION __except(EXCEPTION EXCEPTION_EXECUTE_HANDLER) +# define CAUGHT_EXCEPTION __except(EXCEPTION_EXECUTE_HANDLER) #else # define TRY_TO_CATCH_SEGV if(1) # define CAUGHT_EXCEPTION else @@ -50,6 +74,16 @@ #define LEAF_BITS (16 - BYTE_BITS) #define LEAF_MASK 0x1FFF +typedef struct npath_node_st npath_node_t; +struct npath_node_st { + npath_node_t *prev; + const void *id; + U8 type; + U8 flags; + UV seqn; + U16 depth; +}; + struct state { UV total_size; bool regex_whine; @@ -60,8 +94,260 @@ struct state { start with 0 bits, hence the start of this array will be hot, and the end unused. So put the flags next to the hot end. */ void *tracking[256]; + int min_recurse_threshold; + /* callback hooks and data */ + int (*add_attr_cb)(struct state *st, npath_node_t *npath_node, UV attr_type, const char *name, UV value); + void (*free_state_cb)(pTHX_ struct state *st); + UV seqn; + void *state_cb_data; /* free'd by free_state() after free_state_cb() call */ + /* this stuff wil be moved to state_cb_data later */ + FILE *node_stream_fh; + char *node_stream_name; +}; + +#define ADD_SIZE(st, leafname, bytes) (NPathAddSizeCb(st, leafname, bytes) (st)->total_size += (bytes)) + +#define PATH_TRACKING +#ifdef PATH_TRACKING + +#define pPATH npath_node_t *NPathArg + +/* A subtle point here is that dNPathNodes and NPathPushNode leave NP pointing + * to the next unused slot (though with prev already filled in) + * whereas NPathLink leaves NP unchanged, it just fills in the slot NP points + * to and passes that NP value to the function being called. + * seqn==0 indicates the node is new (hasn't been output yet) + */ +#define dNPathNodes(nodes, prev_np) \ + npath_node_t name_path_nodes[nodes+1]; /* +1 for NPathLink */ \ + npath_node_t *NP = &name_path_nodes[0]; \ + NP->seqn = NP->type = 0; NP->id = Nullch; /* safety/debug */ \ + NP->prev = prev_np +#define NPathPushNode(nodeid, nodetype) \ + NP->id = nodeid; \ + NP->type = nodetype; \ + NP->seqn = 0; \ + if(0)fprintf(stderr,"NPathPushNode (%p <-) %p <- [%d %s]\n", NP->prev, NP, nodetype,(char*)nodeid);\ + NP++; \ + NP->id = Nullch; /* safety/debug */ \ + NP->seqn = 0; \ + NP->prev = (NP-1) +#define NPathSetNode(nodeid, nodetype) \ + (NP-1)->id = nodeid; \ + (NP-1)->type = nodetype; \ + if(0)fprintf(stderr,"NPathSetNode (%p <-) %p <- [%d %s]\n", (NP-1)->prev, (NP-1), nodetype,(char*)nodeid);\ + (NP-1)->seqn = 0; +#define NPathPopNode \ + --NP + +/* dNPathUseParent points NP directly the the parents' name_path_nodes array + * So the function can only safely call ADD_*() but not NPathLink, unless the + * caller has spare nodes in its name_path_nodes. + */ +#define dNPathUseParent(prev_np) npath_node_t *NP = (((prev_np+1)->prev = prev_np), prev_np+1) + +#define NPtype_NAME 0x01 +#define NPtype_LINK 0x02 +#define NPtype_SV 0x03 +#define NPtype_MAGIC 0x04 +#define NPtype_OP 0x05 + +#define NPattr_LEAFSIZE 0x00 +#define NPattr_NAME 0x01 +#define NPattr_PADFAKE 0x02 +#define NPattr_PADNAME 0x03 +#define NPattr_PADTMP 0x04 +#define NPattr_NOTE 0x05 + +#define _NPathLink(np, nid, ntype) (((np)->id=nid), ((np)->type=ntype), ((np)->seqn=0)) +#define NPathLink(nid) (_NPathLink(NP, nid, NPtype_LINK), NP) +/* add a link and a name node to the path - a special case for op_size */ +#define NPathLinkAndNode(nid, nid2) (_NPathLink(NP, nid, NPtype_LINK), _NPathLink(NP+1, nid2, NPtype_NAME), ((NP+1)->prev=NP), (NP+1)) +#define NPathOpLink (NPathArg) +#define NPathAddSizeCb(st, name, bytes) (st->add_attr_cb && st->add_attr_cb(st, NP-1, NPattr_LEAFSIZE, (name), (bytes))), +#define ADD_ATTR(st, attr_type, attr_name, attr_value) (st->add_attr_cb && st->add_attr_cb(st, NP-1, attr_type, attr_name, attr_value)) + +#else + +#define NPathAddSizeCb(st, name, bytes) +#define pPATH void *npath_dummy /* XXX ideally remove */ +#define dNPathNodes(nodes, prev_np) dNOOP +#define NPathLink(nodeid, nodetype) NULL +#define NPathOpLink NULL +#define ADD_ATTR(st, attr_type, attr_name, attr_value) NOOP + +#endif /* PATH_TRACKING */ + + + + +#ifdef PATH_TRACKING + +static const char *svtypenames[SVt_LAST] = { +#if PERL_VERSION < 9 + "NULL", "IV", "NV", "RV", "PV", "PVIV", "PVNV", "PVMG", "PVBM", "PVLV", "PVAV", "PVHV", "PVCV", "PVGV", "PVFM", "PVIO", +#elif PERL_VERSION == 10 && PERL_SUBVERSION == 0 + "NULL", "BIND", "IV", "NV", "RV", "PV", "PVIV", "PVNV", "PVMG", "PVGV", "PVLV", "PVAV", "PVHV", "PVCV", "PVFM", "PVIO", +#elif PERL_VERSION == 10 && PERL_SUBVERSION == 1 + "NULL", "BIND", "IV", "NV", "RV", "PV", "PVIV", "PVNV", "PVMG", "PVGV", "PVLV", "PVAV", "PVHV", "PVCV", "PVFM", "PVIO", +#elif PERL_VERSION < 13 + "NULL", "BIND", "IV", "NV", "PV", "PVIV", "PVNV", "PVMG", "REGEXP", "PVGV", "PVLV", "PVAV", "PVHV", "PVCV", "PVFM", "PVIO", +#else + "NULL", "BIND", "IV", "NV", "PV", "PVIV", "PVNV", "PVMG", "REGEXP", "PVGV", "PVLV", "PVAV", "PVHV", "PVCV", "PVFM", "PVIO", +#endif }; +int +np_print_node_name(FILE *fp, npath_node_t *npath_node) +{ + char buf[1024]; /* XXX */ + + switch (npath_node->type) { + case NPtype_SV: { /* id is pointer to the SV sv_size was called on */ + const SV *sv = (SV*)npath_node->id; + int type = SvTYPE(sv); + char *typename = (type == SVt_IV && SvROK(sv)) ? "RV" : svtypenames[type]; + fprintf(fp, "SV(%s)", typename); + switch(type) { /* add some useful details */ + case SVt_PVAV: fprintf(fp, " fill=%d/%ld", av_len((AV*)sv), AvMAX((AV*)sv)); break; + case SVt_PVHV: fprintf(fp, " fill=%ld/%ld", HvFILL((HV*)sv), HvMAX((HV*)sv)); break; + } + break; + } + case NPtype_OP: { /* id is pointer to the OP op_size was called on */ + const OP *op = (OP*)npath_node->id; + fprintf(fp, "OP(%s)", OP_NAME(op)); + break; + } + case NPtype_MAGIC: { /* id is pointer to the MAGIC struct */ + MAGIC *magic_pointer = (MAGIC*)npath_node->id; + /* XXX it would be nice if we could reuse mg_names.c [sigh] */ + fprintf(fp, "MAGIC(%c)", magic_pointer->mg_type ? magic_pointer->mg_type : '0'); + break; + } + case NPtype_LINK: + fprintf(fp, "%s", npath_node->id); + break; + case NPtype_NAME: + fprintf(fp, "%s", npath_node->id); + break; + default: /* assume id is a string pointer */ + fprintf(fp, "UNKNOWN(%d,%p)", npath_node->type, npath_node->id); + break; + } + return 0; +} + +void +np_dump_indent(int depth) { + while (depth-- > 0) + fprintf(stderr, ": "); +} + +int +np_walk_new_nodes(struct state *st, + npath_node_t *npath_node, + npath_node_t *npath_node_deeper, + int (*cb)(struct state *st, npath_node_t *npath_node, npath_node_t *npath_node_deeper)) +{ + if (npath_node->seqn) /* node already output */ + return 0; + + if (npath_node->prev) { + np_walk_new_nodes(st, npath_node->prev, npath_node, cb); /* recurse */ + npath_node->depth = npath_node->prev->depth + 1; + } + else npath_node->depth = 0; + npath_node->seqn = ++st->seqn; + + if (cb) { + if (cb(st, npath_node, npath_node_deeper)) { + /* ignore this node */ + assert(npath_node->prev); + assert(npath_node->depth); + assert(npath_node_deeper); + npath_node->depth--; + npath_node->seqn = --st->seqn; + npath_node_deeper->prev = npath_node->prev; + } + } + + return 0; +} + +int +np_dump_formatted_node(struct state *st, npath_node_t *npath_node, npath_node_t *npath_node_deeper) { + if (0 && npath_node->type == NPtype_LINK) + return 1; + np_dump_indent(npath_node->depth); + np_print_node_name(stderr, npath_node); + if (npath_node->type == NPtype_LINK) + fprintf(stderr, "->"); /* cosmetic */ + fprintf(stderr, "\t\t[#%ld @%u] ", npath_node->seqn, npath_node->depth); + fprintf(stderr, "\n"); + return 0; +} + +int +np_dump_node_path_info(struct state *st, npath_node_t *npath_node, UV attr_type, const char *attr_name, UV attr_value) +{ + if (attr_type == NPattr_LEAFSIZE && !attr_value) + return 0; /* ignore zero sized leaf items */ + np_walk_new_nodes(st, npath_node, NULL, np_dump_formatted_node); + np_dump_indent(npath_node->depth+1); + switch (attr_type) { + case NPattr_LEAFSIZE: + fprintf(stderr, "+%ld %s =%ld", attr_value, attr_name, attr_value+st->total_size); + break; + case NPattr_NAME: + fprintf(stderr, "~NAMED('%s') %lu", attr_name, attr_value); + break; + case NPattr_NOTE: + fprintf(stderr, "~note %s %lu", attr_name, attr_value); + break; + case NPattr_PADTMP: + case NPattr_PADNAME: + case NPattr_PADFAKE: + fprintf(stderr, "~pad%lu %s %lu", attr_type, attr_name, attr_value); + break; + default: + fprintf(stderr, "~??? %s %lu", attr_name, attr_value); + break; + } + fprintf(stderr, "\n"); + return 0; +} + +int +np_stream_formatted_node(struct state *st, npath_node_t *npath_node, npath_node_t *npath_node_deeper) { + fprintf(st->node_stream_fh, "-%u %lu %u ", + npath_node->type, npath_node->seqn, (unsigned)npath_node->depth + ); + np_print_node_name(st->node_stream_fh, npath_node); + fprintf(st->node_stream_fh, "\n"); + return 0; +} + +int +np_stream_node_path_info(struct state *st, npath_node_t *npath_node, UV attr_type, const char *attr_name, UV attr_value) +{ + if (!attr_type && !attr_value) + return 0; /* ignore zero sized leaf items */ + np_walk_new_nodes(st, npath_node, NULL, np_stream_formatted_node); + if (attr_type) { /* Attribute type, name and value */ + fprintf(st->node_stream_fh, "%lu %lu ", attr_type, npath_node->seqn); + } + else { /* Leaf name and memory size */ + fprintf(st->node_stream_fh, "L %lu ", npath_node->seqn); + } + fprintf(st->node_stream_fh, "%lu %s\n", attr_value, attr_name); + return 0; +} + + +#endif /* PATH_TRACKING */ + + /* Checks to see if thing is in the bitstring. Returns true or false, and @@ -138,7 +424,7 @@ free_tracking_at(void **tv, int level) /* Nodes */ do { if (tv[i]) { - free_tracking_at(tv[i], level); + free_tracking_at((void **) tv[i], level); Safefree(tv[i]); } } while (i--); @@ -155,6 +441,10 @@ static void free_state(struct state *st) { const int top_level = (sizeof(void *) * 8 - LEAF_BITS - BYTE_BITS) / 8; + if (st->free_state_cb) + st->free_state_cb(st); + if (st->state_cb_data) + Safefree(st->state_cb_data); free_tracking_at((void **)st->tracking, top_level); Safefree(st); } @@ -173,7 +463,7 @@ free_state(struct state *st) #define SOME_RECURSION 1 #define TOTAL_SIZE_RECURSION 2 -static bool sv_size(pTHX_ struct state *, const SV *const, const int recurse); +static void sv_size(pTHX_ struct state *, pPATH, const SV *const, const int recurse); typedef enum { OPc_NULL, /* 0 */ @@ -188,6 +478,13 @@ typedef enum { OPc_PVOP, /* 9 */ OPc_LOOP, /* 10 */ OPc_COP /* 11 */ +#ifdef OA_CONDOP + , OPc_CONDOP /* 12 */ +#endif +#ifdef OA_GVOP + , OPc_GVOP /* 13 */ +#endif + } opclass; static opclass @@ -233,9 +530,17 @@ cc_opclass(const OP * const o) case OA_SVOP: TAG; return OPc_SVOP; +#ifdef OA_PADOP case OA_PADOP: TAG; return OPc_PADOP; +#endif + +#ifdef OA_GVOP + case OA_GVOP: TAG; + return OPc_GVOP; +#endif +#ifdef OA_PVOP_OR_SVOP case OA_PVOP_OR_SVOP: TAG; /* * Character translations (tr///) are usually a PVOP, keeping a @@ -246,6 +551,7 @@ cc_opclass(const OP * const o) */ return (o->op_private & (OPpTRANS_TO_UTF|OPpTRANS_FROM_UTF)) ? OPc_SVOP : OPc_PVOP; +#endif case OA_LOOP: TAG; return OPc_LOOP; @@ -295,6 +601,11 @@ cc_opclass(const OP * const o) return OPc_BASEOP; else return OPc_PVOP; + +#ifdef OA_CONDOP + case OA_CONDOP: TAG; + return OPc_CONDOP; +#endif } warn("Devel::Size: Can't determine class of operator %s, assuming BASEOP\n", PL_op_name[o->op_type]); @@ -303,40 +614,56 @@ cc_opclass(const OP * const o) return OPc_BASEOP; } - -#if !defined(NV) -#define NV double -#endif - /* Figure out how much magic is attached to the SV and return the size */ static void -magic_size(pTHX_ const SV * const thing, struct state *st) { +magic_size(pTHX_ const SV * const thing, struct state *st, pPATH) { + dNPathNodes(1, NPathArg); MAGIC *magic_pointer = SvMAGIC(thing); + if (!magic_pointer) + return; + + if (!SvMAGICAL(thing)) { + if (0) { + warn("Ignoring suspect magic on this SV\n"); + sv_dump((SV*)thing); + } + return; + } + + /* push a dummy node for NPathSetNode to update inside the while loop */ + NPathPushNode("dummy", NPtype_NAME); + /* Have we seen the magic pointer? (NULL has always been seen before) */ while (check_new(st, magic_pointer)) { - st->total_size += sizeof(MAGIC); + + NPathSetNode(magic_pointer, NPtype_MAGIC); + + ADD_SIZE(st, "mg", sizeof(MAGIC)); + /* magic vtables aren't freed when magic is freed, so don't count them. + (They are static structures. Anything that assumes otherwise is buggy.) + */ + TRY_TO_CATCH_SEGV { - /* Have we seen the magic vtable? */ - if (check_new(st, magic_pointer->mg_virtual)) { - st->total_size += sizeof(MGVTBL); - } - sv_size(aTHX_ st, magic_pointer->mg_obj, TOTAL_SIZE_RECURSION); + /* XXX only chase mg_obj if mg->mg_flags & MGf_REFCOUNTED ? */ + sv_size(aTHX_ st, NPathLink("mg_obj"), 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); + sv_size(aTHX_ st, NPathLink("mg_ptr"), (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); + ADD_SIZE(st, "PERL_MAGIC_utf8", PERL_MAGIC_UTF8_CACHESIZE * 2 * sizeof(STRLEN)); } } #endif + /* XXX also handle mg->mg_type == PERL_MAGIC_utf8 ? */ else if (magic_pointer->mg_len > 0) { + if(0)do_magic_dump(0, Perl_debug_log, magic_pointer, 0, 0, FALSE, 0); if (check_new(st, magic_pointer->mg_ptr)) { - st->total_size += magic_pointer->mg_len; + ADD_SIZE(st, "mg_len", magic_pointer->mg_len); } } @@ -351,24 +678,29 @@ magic_size(pTHX_ const SV * const thing, struct state *st) { } static void -check_new_and_strlen(struct state *st, const char *const p) { - if(check_new(st, p)) - st->total_size += 1 + strlen(p); +check_new_and_strlen(struct state *st, const char *const p, pPATH) { + dNPathNodes(1, NPathArg->prev); + if(check_new(st, p)) { + NPathPushNode(NPathArg->id, NPtype_NAME); + ADD_SIZE(st, NPathArg->id, 1 + strlen(p)); + } } static void -regex_size(const REGEXP * const baseregex, struct state *st) { +regex_size(const REGEXP * const baseregex, struct state *st, pPATH) { + dNPathNodes(1, NPathArg); if(!check_new(st, baseregex)) return; - st->total_size += sizeof(REGEXP); + NPathPushNode("regex_size", NPtype_NAME); + ADD_SIZE(st, "REGEXP", sizeof(REGEXP)); #if (PERL_VERSION < 11) /* Note the size of the paren offset thing */ - st->total_size += sizeof(I32) * baseregex->nparens * 2; - st->total_size += strlen(baseregex->precomp); + ADD_SIZE(st, "nparens", sizeof(I32) * baseregex->nparens * 2); + ADD_SIZE(st, "precomp", strlen(baseregex->precomp)); #else - st->total_size += sizeof(struct regexp); - st->total_size += sizeof(I32) * SvANY(baseregex)->nparens * 2; - /*st->total_size += strlen(SvANY(baseregex)->subbeg);*/ + ADD_SIZE(st, "regexp", sizeof(struct regexp)); + ADD_SIZE(st, "nparens", sizeof(I32) * SvANY(baseregex)->nparens * 2); + /*ADD_SIZE(st, strlen(SvANY(baseregex)->subbeg));*/ #endif if (st->go_yell && !st->regex_whine) { carp("Devel::Size: Calculated sizes for compiled regexes are incompatible, and probably always will be"); @@ -377,82 +709,104 @@ regex_size(const REGEXP * const baseregex, struct state *st) { } static void -op_size(pTHX_ const OP * const baseop, struct state *st) +op_size(pTHX_ const OP * const baseop, struct state *st, pPATH) { + /* op_size recurses to follow the chain of opcodes. For the node path we + * don't want the chain to be 'nested' in the path so we use dNPathUseParent(). + * Also, to avoid a link-to-a-link the caller should use NPathLinkAndNode() + * instead of NPathLink(). + */ + dNPathUseParent(NPathArg); + TRY_TO_CATCH_SEGV { TAG; if(!check_new(st, baseop)) return; TAG; - op_size(aTHX_ baseop->op_next, st); + op_size(aTHX_ baseop->op_next, st, NPathOpLink); TAG; switch (cc_opclass(baseop)) { case OPc_BASEOP: TAG; - st->total_size += sizeof(struct op); + ADD_SIZE(st, "op", sizeof(struct op)); TAG;break; case OPc_UNOP: TAG; - st->total_size += sizeof(struct unop); - op_size(aTHX_ cUNOPx(baseop)->op_first, st); + ADD_SIZE(st, "unop", sizeof(struct unop)); + op_size(aTHX_ ((UNOP *)baseop)->op_first, st, NPathOpLink); TAG;break; case OPc_BINOP: TAG; - st->total_size += sizeof(struct binop); - op_size(aTHX_ cBINOPx(baseop)->op_first, st); - op_size(aTHX_ cBINOPx(baseop)->op_last, st); + ADD_SIZE(st, "binop", sizeof(struct binop)); + op_size(aTHX_ ((BINOP *)baseop)->op_first, st, NPathOpLink); + op_size(aTHX_ ((BINOP *)baseop)->op_last, st, NPathOpLink); TAG;break; case OPc_LOGOP: TAG; - st->total_size += sizeof(struct logop); - op_size(aTHX_ cBINOPx(baseop)->op_first, st); - op_size(aTHX_ cLOGOPx(baseop)->op_other, st); + ADD_SIZE(st, "logop", sizeof(struct logop)); + op_size(aTHX_ ((BINOP *)baseop)->op_first, st, NPathOpLink); + op_size(aTHX_ ((LOGOP *)baseop)->op_other, st, NPathOpLink); TAG;break; +#ifdef OA_CONDOP + case OPc_CONDOP: TAG; + ADD_SIZE(st, "condop", sizeof(struct condop)); + op_size(aTHX_ ((BINOP *)baseop)->op_first, st, NPathOpLink); + op_size(aTHX_ ((CONDOP *)baseop)->op_true, st, NPathOpLink); + op_size(aTHX_ ((CONDOP *)baseop)->op_false, st, NPathOpLink); + TAG;break; +#endif case OPc_LISTOP: TAG; - st->total_size += sizeof(struct listop); - op_size(aTHX_ cLISTOPx(baseop)->op_first, st); - op_size(aTHX_ cLISTOPx(baseop)->op_last, st); + ADD_SIZE(st, "listop", sizeof(struct listop)); + op_size(aTHX_ ((LISTOP *)baseop)->op_first, st, NPathOpLink); + op_size(aTHX_ ((LISTOP *)baseop)->op_last, st, NPathOpLink); TAG;break; case OPc_PMOP: TAG; - st->total_size += sizeof(struct pmop); - op_size(aTHX_ cPMOPx(baseop)->op_first, st); - op_size(aTHX_ cPMOPx(baseop)->op_last, st); + ADD_SIZE(st, "pmop", sizeof(struct pmop)); + op_size(aTHX_ ((PMOP *)baseop)->op_first, st, NPathOpLink); + op_size(aTHX_ ((PMOP *)baseop)->op_last, st, NPathOpLink); #if PERL_VERSION < 9 || (PERL_VERSION == 9 && PERL_SUBVERSION < 5) - op_size(aTHX_ cPMOPx(baseop)->op_pmreplroot, st); - op_size(aTHX_ cPMOPx(baseop)->op_pmreplstart, st); - op_size(aTHX_ (OP *)cPMOPx(baseop)->op_pmnext, st); + op_size(aTHX_ ((PMOP *)baseop)->op_pmreplroot, st, NPathOpLink); + op_size(aTHX_ ((PMOP *)baseop)->op_pmreplstart, st, NPathOpLink); #endif /* This is defined away in perl 5.8.x, but it is in there for 5.6.x */ #ifdef PM_GETRE - regex_size(PM_GETRE(cPMOPx(baseop)), st); + regex_size(PM_GETRE((PMOP *)baseop), st, NPathLink("PM_GETRE")); #else - regex_size(cPMOPx(baseop)->op_pmregexp, st); + regex_size(((PMOP *)baseop)->op_pmregexp, st, NPathLink("op_pmregexp")); #endif TAG;break; case OPc_SVOP: TAG; - st->total_size += sizeof(struct pmop); + ADD_SIZE(st, "svop", sizeof(struct svop)); if (!(baseop->op_type == OP_AELEMFAST && baseop->op_flags & OPf_SPECIAL)) { /* not an OP_PADAV replacement */ - sv_size(aTHX_ st, cSVOPx(baseop)->op_sv, SOME_RECURSION); + sv_size(aTHX_ st, NPathLink("SVOP"), ((SVOP *)baseop)->op_sv, SOME_RECURSION); } TAG;break; +#ifdef OA_PADOP case OPc_PADOP: TAG; - st->total_size += sizeof(struct padop); + ADD_SIZE(st, "padop", sizeof(struct padop)); + TAG;break; +#endif +#ifdef OA_GVOP + case OPc_GVOP: TAG; + ADD_SIZE(st, "gvop", sizeof(struct gvop)); + sv_size(aTHX_ st, NPathLink("GVOP"), ((GVOP *)baseop)->op_gv, SOME_RECURSION); TAG;break; +#endif case OPc_PVOP: TAG; - check_new_and_strlen(st, cPVOPx(baseop)->op_pv); + check_new_and_strlen(st, ((PVOP *)baseop)->op_pv, NPathLink("op_pv")); TAG;break; case OPc_LOOP: TAG; - st->total_size += sizeof(struct loop); - op_size(aTHX_ cLOOPx(baseop)->op_first, st); - op_size(aTHX_ cLOOPx(baseop)->op_last, st); - op_size(aTHX_ cLOOPx(baseop)->op_redoop, st); - op_size(aTHX_ cLOOPx(baseop)->op_nextop, st); - op_size(aTHX_ cLOOPx(baseop)->op_lastop, st); + ADD_SIZE(st, "loop", sizeof(struct loop)); + op_size(aTHX_ ((LOOP *)baseop)->op_first, st, NPathOpLink); + op_size(aTHX_ ((LOOP *)baseop)->op_last, st, NPathOpLink); + op_size(aTHX_ ((LOOP *)baseop)->op_redoop, st, NPathOpLink); + op_size(aTHX_ ((LOOP *)baseop)->op_nextop, st, NPathOpLink); + op_size(aTHX_ ((LOOP *)baseop)->op_lastop, st, NPathOpLink); TAG;break; case OPc_COP: TAG; { COP *basecop; basecop = (COP *)baseop; - st->total_size += sizeof(struct cop); + ADD_SIZE(st, "cop", sizeof(struct cop)); /* Change 33656 by nicholas@mouse-mill on 2008/04/07 11:29:51 Eliminate cop_label from struct cop by storing a label as the first @@ -462,14 +816,15 @@ op_size(pTHX_ const OP * const baseop, struct state *st) before 5.11 @33656, but later than 5.10, producing slightly too small memory sizes on these Perls. */ #if (PERL_VERSION < 11) - check_new_and_strlen(st, basecop->cop_label); + check_new_and_strlen(st, basecop->cop_label, NPathLink("cop_label")); #endif #ifdef USE_ITHREADS - check_new_and_strlen(st, basecop->cop_file); - check_new_and_strlen(st, basecop->cop_stashpv); + check_new_and_strlen(st, basecop->cop_file, NPathLink("cop_file")); + check_new_and_strlen(st, basecop->cop_stashpv, NPathLink("cop_stashpv")); #else - sv_size(aTHX_ st, (SV *)basecop->cop_stash, SOME_RECURSION); - sv_size(aTHX_ st, (SV *)basecop->cop_filegv, SOME_RECURSION); + if (SvREFCNT(basecop->cop_stash) == 1) /* XXX hack? */ + sv_size(aTHX_ st, NPathLink("cop_stash"), (SV *)basecop->cop_stash, SOME_RECURSION); + sv_size(aTHX_ st, NPathLink("cop_filegv"), (SV *)basecop->cop_filegv, SOME_RECURSION); #endif } @@ -484,150 +839,238 @@ op_size(pTHX_ const OP * const baseop, struct state *st) } } -#if PERL_VERSION > 9 || (PERL_VERSION == 9 && PERL_SUBVERSION > 2) -# define NEW_HEAD_LAYOUT -#endif - -static bool -sv_size(pTHX_ struct state *const st, const SV * const orig_thing, - const int recurse) { - const SV *thing = orig_thing; +static void +hek_size(pTHX_ struct state *st, HEK *hek, U32 shared, pPATH) +{ + dNPathNodes(1, NPathArg); - if(!check_new(st, thing)) - return FALSE; + /* Hash keys can be shared. Have we seen this before? */ + if (!check_new(st, hek)) + return; + NPathPushNode("hek", NPtype_NAME); + ADD_SIZE(st, "hek_len", HEK_BASESIZE + hek->hek_len +#if PERL_VERSION < 8 + + 1 /* No hash key flags prior to 5.8.0 */ +#else + + 2 +#endif + ); + if (shared) { +#if PERL_VERSION < 10 + ADD_SIZE(st, "he", sizeof(struct he)); +#else + ADD_SIZE(st, "shared_he", STRUCT_OFFSET(struct shared_he, shared_he_hek)); +#endif + } +} - st->total_size += sizeof(SV); - switch (SvTYPE(thing)) { - /* Is it undef? */ - case SVt_NULL: TAG; - TAG;break; - /* Just a plain integer. This will be differently sized depending - on whether purify's been compiled in */ - case SVt_IV: TAG; -#ifndef NEW_HEAD_LAYOUT -# ifdef PURIFY - st->total_size += sizeof(sizeof(XPVIV)); -# else - st->total_size += sizeof(IV); -# endif +#if PERL_VERSION < 8 || PERL_SUBVERSION < 9 +# define SVt_LAST 16 #endif - if(recurse && SvROK(thing)) - 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; + #ifdef PURIFY - st->total_size += sizeof(sizeof(XPVNV)); +# define MAYBE_PURIFY(normal, pure) (pure) +# define MAYBE_OFFSET(struct_name, member) 0 #else - st->total_size += sizeof(NV); -#endif - TAG;break; -#if (PERL_VERSION < 11) - /* Is it a reference? */ - case SVt_RV: TAG; -#ifndef NEW_HEAD_LAYOUT - st->total_size += sizeof(XRV); +# define MAYBE_PURIFY(normal, pure) (normal) +# define MAYBE_OFFSET(struct_name, member) STRUCT_OFFSET(struct_name, member) #endif - if(recurse && SvROK(thing)) - sv_size(aTHX_ st, SvRV_const(thing), recurse); - TAG;break; + +const U8 body_sizes[SVt_LAST] = { +#if PERL_VERSION < 9 + 0, /* SVt_NULL */ + MAYBE_PURIFY(sizeof(IV), sizeof(XPVIV)), /* SVt_IV */ + MAYBE_PURIFY(sizeof(NV), sizeof(XPVNV)), /* SVt_NV */ + sizeof(XRV), /* SVt_RV */ + sizeof(XPV), /* SVt_PV */ + sizeof(XPVIV), /* SVt_PVIV */ + sizeof(XPVNV), /* SVt_PVNV */ + sizeof(XPVMG), /* SVt_PVMG */ + sizeof(XPVBM), /* SVt_PVBM */ + sizeof(XPVLV), /* SVt_PVLV */ + sizeof(XPVAV), /* SVt_PVAV */ + sizeof(XPVHV), /* SVt_PVHV */ + sizeof(XPVCV), /* SVt_PVCV */ + sizeof(XPVGV), /* SVt_PVGV */ + sizeof(XPVFM), /* SVt_PVFM */ + sizeof(XPVIO) /* SVt_PVIO */ +#elif PERL_VERSION == 10 && PERL_SUBVERSION == 0 + 0, /* SVt_NULL */ + 0, /* SVt_BIND */ + 0, /* SVt_IV */ + MAYBE_PURIFY(sizeof(NV), sizeof(XPVNV)), /* SVt_NV */ + 0, /* SVt_RV */ + MAYBE_PURIFY(sizeof(xpv_allocated), sizeof(XPV)), /* SVt_PV */ + MAYBE_PURIFY(sizeof(xpviv_allocated), sizeof(XPVIV)),/* SVt_PVIV */ + sizeof(XPVNV), /* SVt_PVNV */ + sizeof(XPVMG), /* SVt_PVMG */ + sizeof(XPVGV), /* SVt_PVGV */ + sizeof(XPVLV), /* SVt_PVLV */ + MAYBE_PURIFY(sizeof(xpvav_allocated), sizeof(XPVAV)),/* SVt_PVAV */ + MAYBE_PURIFY(sizeof(xpvhv_allocated), sizeof(XPVHV)),/* SVt_PVHV */ + MAYBE_PURIFY(sizeof(xpvcv_allocated), sizeof(XPVCV)),/* SVt_PVCV */ + MAYBE_PURIFY(sizeof(xpvfm_allocated), sizeof(XPVFM)),/* SVt_PVFM */ + sizeof(XPVIO), /* SVt_PVIO */ +#elif PERL_VERSION == 10 && PERL_SUBVERSION == 1 + 0, /* SVt_NULL */ + 0, /* SVt_BIND */ + 0, /* SVt_IV */ + MAYBE_PURIFY(sizeof(NV), sizeof(XPVNV)), /* SVt_NV */ + 0, /* SVt_RV */ + sizeof(XPV) - MAYBE_OFFSET(XPV, xpv_cur), /* SVt_PV */ + sizeof(XPVIV) - MAYBE_OFFSET(XPV, xpv_cur), /* SVt_PVIV */ + sizeof(XPVNV), /* SVt_PVNV */ + sizeof(XPVMG), /* SVt_PVMG */ + sizeof(XPVGV), /* SVt_PVGV */ + sizeof(XPVLV), /* SVt_PVLV */ + sizeof(XPVAV) - MAYBE_OFFSET(XPVAV, xav_fill), /* SVt_PVAV */ + sizeof(XPVHV) - MAYBE_OFFSET(XPVHV, xhv_fill), /* SVt_PVHV */ + sizeof(XPVCV) - MAYBE_OFFSET(XPVCV, xpv_cur), /* SVt_PVCV */ + sizeof(XPVFM) - MAYBE_OFFSET(XPVFM, xpv_cur), /* SVt_PVFM */ + sizeof(XPVIO) /* SVt_PVIO */ +#elif PERL_VERSION < 13 + 0, /* SVt_NULL */ + 0, /* SVt_BIND */ + 0, /* SVt_IV */ + MAYBE_PURIFY(sizeof(NV), sizeof(XPVNV)), /* SVt_NV */ + sizeof(XPV) - MAYBE_OFFSET(XPV, xpv_cur), /* SVt_PV */ + sizeof(XPVIV) - MAYBE_OFFSET(XPV, xpv_cur), /* SVt_PVIV */ + sizeof(XPVNV), /* SVt_PVNV */ + sizeof(XPVMG), /* SVt_PVMG */ + sizeof(regexp) - MAYBE_OFFSET(regexp, xpv_cur), /* SVt_REGEXP */ + sizeof(XPVGV), /* SVt_PVGV */ + sizeof(XPVLV), /* SVt_PVLV */ + sizeof(XPVAV) - MAYBE_OFFSET(XPVAV, xav_fill), /* SVt_PVAV */ + sizeof(XPVHV) - MAYBE_OFFSET(XPVHV, xhv_fill), /* SVt_PVHV */ + sizeof(XPVCV) - MAYBE_OFFSET(XPVCV, xpv_cur), /* SVt_PVCV */ + sizeof(XPVFM) - MAYBE_OFFSET(XPVFM, xpv_cur), /* SVt_PVFM */ + sizeof(XPVIO) /* SVt_PVIO */ +#else + 0, /* SVt_NULL */ + 0, /* SVt_BIND */ + 0, /* SVt_IV */ + MAYBE_PURIFY(sizeof(NV), sizeof(XPVNV)), /* SVt_NV */ + sizeof(XPV) - MAYBE_OFFSET(XPV, xpv_cur), /* SVt_PV */ + sizeof(XPVIV) - MAYBE_OFFSET(XPV, xpv_cur), /* SVt_PVIV */ + sizeof(XPVNV) - MAYBE_OFFSET(XPV, xpv_cur), /* SVt_PVNV */ + sizeof(XPVMG), /* SVt_PVMG */ + sizeof(regexp), /* SVt_REGEXP */ + sizeof(XPVGV), /* SVt_PVGV */ + sizeof(XPVLV), /* SVt_PVLV */ + sizeof(XPVAV), /* SVt_PVAV */ + sizeof(XPVHV), /* SVt_PVHV */ + sizeof(XPVCV), /* SVt_PVCV */ + sizeof(XPVFM), /* SVt_PVFM */ + sizeof(XPVIO) /* SVt_PVIO */ #endif - /* How about a plain string? In which case we need to add in how - much has been allocated */ - case SVt_PV: TAG; - st->total_size += sizeof(XPV); - if(recurse && SvROK(thing)) - sv_size(aTHX_ st, SvRV_const(thing), recurse); - else - st->total_size += SvLEN(thing); - TAG;break; - /* A string with an integer part? */ - case SVt_PVIV: TAG; - st->total_size += sizeof(XPVIV); - if(recurse && SvROK(thing)) - sv_size(aTHX_ st, SvRV_const(thing), recurse); - else - st->total_size += SvLEN(thing); - if(SvOOK(thing)) { - st->total_size += SvIVX(thing); +}; + + +/* based on Perl_do_dump_pad() - wraps sv_size and adds ADD_ATTR calls for the pad names */ +static void +padlist_size(pTHX_ struct state *const st, pPATH, PADLIST *padlist, + const int recurse) +{ + dNPathUseParent(NPathArg); + const AV *pad_name; + SV **pname; + I32 ix; + + if (!padlist) { + return; } - TAG;break; - /* A scalar/string/reference with a float part? */ - case SVt_PVNV: TAG; - st->total_size += sizeof(XPVNV); - if(recurse && SvROK(thing)) - 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), recurse); - else - st->total_size += SvLEN(thing); - 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), recurse); - else - st->total_size += SvLEN(thing); - magic_size(aTHX_ thing, st); - TAG;break; + pad_name = MUTABLE_AV(*av_fetch(MUTABLE_AV(padlist), 0, FALSE)); + pname = AvARRAY(pad_name); + + for (ix = 1; ix <= AvFILLp(pad_name); ix++) { + const SV *namesv = pname[ix]; + if (namesv && namesv == &PL_sv_undef) { + namesv = NULL; + } + if (namesv) { + if (SvFAKE(namesv)) + ADD_ATTR(st, NPattr_PADFAKE, SvPVX_const(namesv), ix); + else + ADD_ATTR(st, NPattr_PADNAME, SvPVX_const(namesv), ix); + } + else { + ADD_ATTR(st, NPattr_PADTMP, "SVs_PADTMP", ix); + } + + } + sv_size(aTHX_ st, NPathArg, (SV*)padlist, recurse); +} + + +static void +sv_size(pTHX_ struct state *const st, pPATH, const SV * const orig_thing, + const int recurse) { + const SV *thing = orig_thing; + dNPathNodes(3, NPathArg); + U32 type; + + if(!check_new(st, orig_thing)) + return; + + type = SvTYPE(thing); + if (type > SVt_LAST) { + warn("Devel::Size: Unknown variable type: %d encountered\n", type); + return; + } + NPathPushNode(thing, NPtype_SV); + ADD_SIZE(st, "sv", sizeof(SV) + body_sizes[type]); + + switch (type) { +#if (PERL_VERSION < 11) + /* Is it a reference? */ + case SVt_RV: TAG; +#else + case SVt_IV: TAG; #endif - case SVt_PVLV: TAG; - st->total_size += sizeof(XPVLV); if(recurse && SvROK(thing)) - sv_size(aTHX_ st, SvRV_const(thing), recurse); - else - st->total_size += SvLEN(thing); - magic_size(aTHX_ thing, st); + sv_size(aTHX_ st, NPathLink("RV"), SvRV_const(thing), recurse); TAG;break; - /* How much space is dedicated to the array? Not counting the - elements in the array, mind, just the array itself */ + case SVt_PVAV: TAG; - st->total_size += sizeof(XPVAV); /* Is there anything in the array? */ if (AvMAX(thing) != -1) { /* an array with 10 slots has AvMax() set to 9 - te 2007-04-22 */ - st->total_size += sizeof(SV *) * (AvMAX(thing) + 1); + ADD_SIZE(st, "av_max", 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) { + if (recurse >= st->min_recurse_threshold) { SSize_t i = AvFILLp(thing) + 1; while (i--) - sv_size(aTHX_ st, AvARRAY(thing)[i], recurse); + sv_size(aTHX_ st, NPathLink("AVelem"), AvARRAY(thing)[i], recurse); } } /* Add in the bits on the other side of the beginning */ dbg_printf(("total_size %li, sizeof(SV *) %li, AvARRAY(thing) %li, AvALLOC(thing)%li , sizeof(ptr) %li \n", - st->total_size, sizeof(SV*), AvARRAY(thing), AvALLOC(thing), sizeof( thing ))); + st->total_size, sizeof(SV*), AvARRAY(thing), AvALLOC(thing), sizeof( thing ))); /* under Perl 5.8.8 64bit threading, AvARRAY(thing) was a pointer while AvALLOC was 0, resulting in grossly overstated sized for arrays. Technically, this shouldn't happen... */ if (AvALLOC(thing) != 0) { - st->total_size += (sizeof(SV *) * (AvARRAY(thing) - AvALLOC(thing))); + ADD_SIZE(st, "AvALLOC", (sizeof(SV *) * (AvARRAY(thing) - AvALLOC(thing)))); } #if (PERL_VERSION < 9) /* Is there something hanging off the arylen element? 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), recurse); + sv_size(aTHX_ st, NPathLink("ARYLEN"), AvARYLEN(thing), recurse); #endif - magic_size(aTHX_ thing, st); TAG;break; + case SVt_PVHV: TAG; - /* First the base struct */ - st->total_size += sizeof(XPVHV); /* Now the array of buckets */ - st->total_size += (sizeof(HE *) * (HvMAX(thing) + 1)); + ADD_SIZE(st, "hv_max", (sizeof(HE *) * (HvMAX(thing) + 1))); + if (HvENAME(thing)) { + ADD_ATTR(st, NPattr_NAME, HvENAME(thing), 0); + } /* Now walk the bucket chain */ if (HvARRAY(thing)) { HE *cur_entry; @@ -635,45 +1078,125 @@ sv_size(pTHX_ struct state *const st, const SV * const orig_thing, for (cur_bucket = 0; cur_bucket <= HvMAX(thing); cur_bucket++) { cur_entry = *(HvARRAY(thing) + cur_bucket); while (cur_entry) { - st->total_size += sizeof(HE); - if (cur_entry->hent_hek) { - /* Hash keys can be shared. Have we seen this before? */ - if (check_new(st, cur_entry->hent_hek)) { - 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); +/* XXX a HE should probably be a node so the keys and values are seen as pairs */ + ADD_SIZE(st, "he", sizeof(HE)); + hek_size(aTHX_ st, cur_entry->hent_hek, HvSHAREKEYS(thing), NPathLink("hent_hek")); + if (recurse >= st->min_recurse_threshold) { +/* I've seen a PL_strtab HeVAL == 0xC and 0x40C etc + * just running perl -Mblib -Mstrict -MDevel::Size=:all -MCarp -e 'warn perl_size()' + * but it seemed like a corruption - it would change come and go with irrelevant code changes. + * so we protect against that here, but I'd like to know the cause. + */ +if (PTR2UV(HeVAL(cur_entry)) > 0xFFF) + sv_size(aTHX_ st, NPathLink("HeVAL"), HeVAL(cur_entry), recurse); +else warn("skipped suspect HeVAL %p", HeVAL(cur_entry)); + } cur_entry = cur_entry->hent_next; } } } - magic_size(aTHX_ thing, st); +#ifdef HvAUX + if (SvOOK(thing)) { + /* This direct access is arguably "naughty": */ + struct mro_meta *meta = HvAUX(thing)->xhv_mro_meta; +#if PERL_VERSION > 13 || PERL_SUBVERSION > 8 + /* As is this: */ + I32 count = HvAUX(thing)->xhv_name_count; + + if (count) { + HEK **names = HvAUX(thing)->xhv_name_u.xhvnameu_names; + if (count < 0) + count = -count; + while (--count) + hek_size(aTHX_ st, names[count], 1, NPathLink("HvAUXelem")); + } + else +#endif + { + hek_size(aTHX_ st, HvNAME_HEK(thing), 1, NPathLink("HvNAME_HEK")); + } + + ADD_SIZE(st, "xpvhv_aux", sizeof(struct xpvhv_aux)); + if (meta) { + ADD_SIZE(st, "mro_meta", sizeof(struct mro_meta)); + sv_size(aTHX_ st, NPathLink("mro_nextmethod"), (SV *)meta->mro_nextmethod, TOTAL_SIZE_RECURSION); +#if PERL_VERSION > 10 || (PERL_VERSION == 10 && PERL_SUBVERSION > 0) + sv_size(aTHX_ st, NPathLink("isa"), (SV *)meta->isa, TOTAL_SIZE_RECURSION); +#endif +#if PERL_VERSION > 10 + sv_size(aTHX_ st, NPathLink("mro_linear_all"), (SV *)meta->mro_linear_all, TOTAL_SIZE_RECURSION); + sv_size(aTHX_ st, NPathLink("mro_linear_current"), meta->mro_linear_current, TOTAL_SIZE_RECURSION); +#else + sv_size(aTHX_ st, NPathLink("mro_linear_dfs"), (SV *)meta->mro_linear_dfs, TOTAL_SIZE_RECURSION); + sv_size(aTHX_ st, NPathLink("mro_linear_c3"), (SV *)meta->mro_linear_c3, TOTAL_SIZE_RECURSION); +#endif + } + } +#else + check_new_and_strlen(st, HvNAME_get(thing), NPathLink("HvNAME")); +#endif TAG;break; + + + case SVt_PVFM: TAG; + padlist_size(aTHX_ st, NPathLink("CvPADLIST"), CvPADLIST(thing), recurse); + sv_size(aTHX_ st, NPathLink("CvOUTSIDE"), (SV *)CvOUTSIDE(thing), recurse); + + if (st->go_yell && !st->fm_whine) { + carp("Devel::Size: Calculated sizes for FMs are incomplete"); + st->fm_whine = 1; + } + goto freescalar; + case SVt_PVCV: TAG; - st->total_size += sizeof(XPVCV); - magic_size(aTHX_ thing, st); - - st->total_size += ((XPVIO *) SvANY(thing))->xpv_len; - 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); + /* not CvSTASH, per https://rt.cpan.org/Ticket/Display.html?id=79366 */ + sv_size(aTHX_ st, NPathLink("SvSTASH"), (SV *)SvSTASH(thing), SOME_RECURSION); + sv_size(aTHX_ st, NPathLink("CvGV"), (SV *)CvGV(thing), SOME_RECURSION); + padlist_size(aTHX_ st, NPathLink("CvPADLIST"), CvPADLIST(thing), recurse); + sv_size(aTHX_ st, NPathLink("CvOUTSIDE"), (SV *)CvOUTSIDE(thing), recurse); if (CvISXSUB(thing)) { - sv_size(aTHX_ st, cv_const_sv((CV *)thing), recurse); + sv_size(aTHX_ st, NPathLink("cv_const_sv"), cv_const_sv((CV *)thing), recurse); } else { - op_size(aTHX_ CvSTART(thing), st); - op_size(aTHX_ CvROOT(thing), st); + if(1)op_size(aTHX_ CvSTART(thing), st, NPathLinkAndNode("CvSTART", "OPs")); /* XXX ? */ + op_size(aTHX_ CvROOT(thing), st, NPathLinkAndNode("CvROOT", "OPs")); } + goto freescalar; + + case SVt_PVIO: TAG; + /* Some embedded char pointers */ + check_new_and_strlen(st, ((XPVIO *) SvANY(thing))->xio_top_name, NPathLink("xio_top_name")); + check_new_and_strlen(st, ((XPVIO *) SvANY(thing))->xio_fmt_name, NPathLink("xio_fmt_name")); + check_new_and_strlen(st, ((XPVIO *) SvANY(thing))->xio_bottom_name, NPathLink("xio_bottom_name")); + /* Throw the GVs on the list to be walked if they're not-null */ + sv_size(aTHX_ st, NPathLink("xio_top_gv"), (SV *)((XPVIO *) SvANY(thing))->xio_top_gv, recurse); + sv_size(aTHX_ st, NPathLink("xio_bottom_gv"), (SV *)((XPVIO *) SvANY(thing))->xio_bottom_gv, recurse); + sv_size(aTHX_ st, NPathLink("xio_fmt_gv"), (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 + not... we can't, so we don't even try */ +#ifdef USE_PERLIO + /* Dig into xio_ifp and xio_ofp here */ + warn("Devel::Size: Can't size up perlio layers yet\n"); +#endif + goto freescalar; + + case SVt_PVLV: TAG; +#if (PERL_VERSION < 9) + goto freescalar; +#endif - TAG;break; case SVt_PVGV: TAG; - magic_size(aTHX_ thing, st); - st->total_size += sizeof(XPVGV); if(isGV_with_GP(thing)) { - st->total_size += GvNAMELEN(thing); -#ifdef GvFILE +#ifdef GvNAME_HEK + hek_size(aTHX_ st, GvNAME_HEK(thing), 1, NPathLink("GvNAME_HEK")); +#else + ADD_SIZE(st, "GvNAMELEN", GvNAMELEN(thing)); +#endif + ADD_ATTR(st, NPattr_NAME, GvNAME_get(thing), 0); +#ifdef GvFILE_HEK + hek_size(aTHX_ st, GvFILE_HEK(thing), 1, NPathLink("GvFILE_HEK")); +#elif defined(GvFILE) # 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(), @@ -682,86 +1205,78 @@ sv_size(pTHX_ struct state *const st, const SV * const orig_thing, 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)); + check_new_and_strlen(st, GvFILE(thing), NPathLink("GvFILE")); # endif #endif /* 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); + ADD_SIZE(st, "GP", sizeof(GP)); + sv_size(aTHX_ st, NPathLink("gp_sv"), (SV *)(GvGP(thing)->gp_sv), recurse); + sv_size(aTHX_ st, NPathLink("gp_av"), (SV *)(GvGP(thing)->gp_av), recurse); + sv_size(aTHX_ st, NPathLink("gp_hv"), (SV *)(GvGP(thing)->gp_hv), recurse); + sv_size(aTHX_ st, NPathLink("gp_cv"), (SV *)(GvGP(thing)->gp_cv), recurse); + sv_size(aTHX_ st, NPathLink("gp_egv"), (SV *)(GvGP(thing)->gp_egv), recurse); + sv_size(aTHX_ st, NPathLink("gp_form"), (SV *)(GvGP(thing)->gp_form), recurse); } +#if (PERL_VERSION >= 9) + TAG; break; +#endif } - TAG;break; - case SVt_PVFM: TAG; - st->total_size += sizeof(XPVFM); - magic_size(aTHX_ thing, st); - st->total_size += ((XPVIO *) SvANY(thing))->xpv_len; - sv_size(aTHX_ st, (SV *)CvPADLIST(thing), SOME_RECURSION); - sv_size(aTHX_ st, (SV *)CvOUTSIDE(thing), recurse); +#if PERL_VERSION <= 8 + case SVt_PVBM: TAG; +#endif + case SVt_PVMG: TAG; + case SVt_PVNV: TAG; + case SVt_PVIV: TAG; + case SVt_PV: TAG; + freescalar: + if(recurse && SvROK(thing)) + sv_size(aTHX_ st, NPathLink("RV"), SvRV_const(thing), recurse); + else if (SvIsCOW_shared_hash(thing)) + hek_size(aTHX_ st, SvSHARED_HEK_FROM_PV(SvPVX(thing)), 1, NPathLink("SvSHARED_HEK_FROM_PV")); + else + ADD_SIZE(st, "SvLEN", SvLEN(thing)); - if (st->go_yell && !st->fm_whine) { - carp("Devel::Size: Calculated sizes for FMs are incomplete"); - st->fm_whine = 1; + if(SvOOK(thing)) { + STRLEN len; + SvOOK_offset(thing, len); + ADD_SIZE(st, "SvOOK", len); } TAG;break; - case SVt_PVIO: TAG; - st->total_size += sizeof(XPVIO); - magic_size(aTHX_ thing, st); - if (check_new(st, (SvPVX_const(thing)))) { - st->total_size += ((XPVIO *) SvANY(thing))->xpv_cur; - } - /* Some embedded char pointers */ - check_new_and_strlen(st, ((XPVIO *) SvANY(thing))->xio_top_name); - 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, 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 - not... we can't, so we don't even try */ -#ifdef USE_PERLIO - /* Dig into xio_ifp and xio_ofp here */ - warn("Devel::Size: Can't size up perlio layers yet\n"); -#endif - TAG;break; - default: - warn("Devel::Size: Unknown variable type: %d encountered\n", SvTYPE(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 (type >= SVt_PVMG) { + magic_size(aTHX_ thing, st, NPathLink("MG")); + } + + return; +} -#if __GNUC__ -void *vtables[] = { -#include "vtables.inc" - NULL -}; -#endif +static void +free_memnode_state(pTHX_ struct state *st) +{ + if (st->node_stream_fh && st->node_stream_name) { + if (*st->node_stream_name == '|') { + if (pclose(st->node_stream_fh)) + warn("%s exited with an error status\n", st->node_stream_name); + } + else { + if (fclose(st->node_stream_fh)) + warn("Error closing %s: %s\n", st->node_stream_name, strerror(errno)); + } + } +} 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; + st->min_recurse_threshold = TOTAL_SIZE_RECURSION; if (NULL != (warn_flag = perl_get_sv("Devel::Size::warn", FALSE))) { st->dangle_whine = st->go_yell = SvIV(warn_flag) ? TRUE : FALSE; } @@ -771,15 +1286,60 @@ 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" +#if PERL_VERSION > 8 || (PERL_VERSION == 8 && PERL_SUBVERSION > 0) + check_new(st, &PL_sv_placeholder); +#endif +#ifdef PATH_TRACKING + if (getenv("MEMNODES") && *getenv("MEMNODES")) { /* XXX quick hack */ + st->node_stream_name = getenv("MEMNODES"); + if (*st->node_stream_name == '|') + st->node_stream_fh = popen(st->node_stream_name+1, "w"); + else + st->node_stream_fh = fopen(st->node_stream_name, "wb"); + if (!st->node_stream_fh) + croak("Can't open '%s' for writing: %s", st->node_stream_name, strerror(errno)); + setlinebuf(st->node_stream_fh); /* XXX temporary for debugging */ + st->add_attr_cb = np_stream_node_path_info; + } + else + st->add_attr_cb = np_dump_node_path_info; + st->free_state_cb = free_memnode_state; #endif return st; } +/* XXX based on S_visit() in sv.c */ +static void +unseen_sv_size(pTHX_ struct state *st, pPATH) +{ + dVAR; + SV* sva; + I32 visited = 0; + dNPathNodes(1, NPathArg); + + NPathPushNode("unseen", NPtype_NAME); + + /* by this point we should have visited all the SVs + * so now we'll run through all the SVs via the arenas + * in order to find any thet we've missed for some reason. + * Once the rest of the code is finding all the SVs then any + * found here will be leaks. + */ + for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) { + const SV * const svend = &sva[SvREFCNT(sva)]; + SV* sv; + for (sv = sva + 1; sv < svend; ++sv) { + if (SvTYPE(sv) != (svtype)SVTYPEMASK && SvREFCNT(sv)) { + sv_size(aTHX_ st, NPathLink("arena"), sv, TOTAL_SIZE_RECURSION); + } + else if (check_new(st, sv)) { /* sanity check */ + warn("unseen_sv_size encountered freed SV unexpectedly"); + sv_dump(sv); + } + } + } +} + MODULE = Devel::Size PACKAGE = Devel::Size PROTOTYPES: DISABLE @@ -800,7 +1360,93 @@ CODE: thing = SvRV(thing); } - sv_size(aTHX_ st, thing, ix); + sv_size(aTHX_ st, NULL, thing, ix); + RETVAL = st->total_size; + free_state(st); +} +OUTPUT: + RETVAL + +UV +perl_size() +CODE: +{ + struct state *st = new_state(aTHX); + dNPathNodes(3, NULL); + + st->min_recurse_threshold = NO_RECURSION; /* so always recurse */ + + NPathPushNode("perl_size", NPtype_NAME); /* provide a root node */ + /* start with PL_defstash to get everything reachable from \%main:: */ + sv_size(aTHX_ st, NPathLink("PL_defstash"), (SV*)PL_defstash, TOTAL_SIZE_RECURSION); + + NPathPushNode("others", NPtype_NAME); /* group these (typically much smaller) items */ + sv_size(aTHX_ st, NPathLink("PL_defgv"), (SV*)PL_defgv, TOTAL_SIZE_RECURSION); + sv_size(aTHX_ st, NPathLink("PL_incgv"), (SV*)PL_incgv, TOTAL_SIZE_RECURSION); + sv_size(aTHX_ st, NPathLink("PL_rs"), (SV*)PL_rs, TOTAL_SIZE_RECURSION); + sv_size(aTHX_ st, NPathLink("PL_fdpid"), (SV*)PL_fdpid, TOTAL_SIZE_RECURSION); + sv_size(aTHX_ st, NPathLink("PL_modglobal"), (SV*)PL_modglobal, TOTAL_SIZE_RECURSION); + sv_size(aTHX_ st, NPathLink("PL_errors"), (SV*)PL_errors, TOTAL_SIZE_RECURSION); + sv_size(aTHX_ st, NPathLink("PL_stashcache"), (SV*)PL_stashcache, TOTAL_SIZE_RECURSION); + sv_size(aTHX_ st, NPathLink("PL_patchlevel"), (SV*)PL_patchlevel, TOTAL_SIZE_RECURSION); + sv_size(aTHX_ st, NPathLink("PL_apiversion"), (SV*)PL_apiversion, TOTAL_SIZE_RECURSION); + sv_size(aTHX_ st, NPathLink("PL_registered_mros"), (SV*)PL_registered_mros, TOTAL_SIZE_RECURSION); +#ifdef USE_ITHREADS + sv_size(aTHX_ st, NPathLink("PL_regex_padav"), (SV*)PL_regex_padav, TOTAL_SIZE_RECURSION); +#endif + sv_size(aTHX_ st, NPathLink("PL_warnhook"), (SV*)PL_warnhook, TOTAL_SIZE_RECURSION); + sv_size(aTHX_ st, NPathLink("PL_diehook"), (SV*)PL_diehook, TOTAL_SIZE_RECURSION); + sv_size(aTHX_ st, NPathLink("PL_endav"), (SV*)PL_endav, TOTAL_SIZE_RECURSION); + sv_size(aTHX_ st, NPathLink("PL_main_cv"), (SV*)PL_main_cv, TOTAL_SIZE_RECURSION); + sv_size(aTHX_ st, NPathLink("PL_main_root"), (SV*)PL_main_root, TOTAL_SIZE_RECURSION); + sv_size(aTHX_ st, NPathLink("PL_main_start"), (SV*)PL_main_start, TOTAL_SIZE_RECURSION); + /* TODO PL_pidstatus */ + /* TODO PL_stashpad */ + /* TODO PL_compiling? COP */ + + /* TODO stacks: cur, main, tmps, mark, scope, save */ + /* TODO PL_exitlist */ + /* TODO PL_reentrant_buffers etc */ + /* TODO environ */ + /* TODO PerlIO? PL_known_layers PL_def_layerlist PL_perlio_fd_refcnt etc */ + /* TODO threads? */ + /* TODO anything missed? */ + + /* --- by this point we should have seen all reachable SVs --- */ + + /* in theory we shouldn't have any elements in PL_strtab that haven't been seen yet */ + sv_size(aTHX_ st, NPathLink("PL_strtab"), (SV*)PL_strtab, TOTAL_SIZE_RECURSION); + + /* unused space in sv head arenas */ + if (PL_sv_root) { + SV *p = PL_sv_root; + UV free_heads = 1; +# define SvARENA_CHAIN(sv) SvANY(sv) /* XXX */ + while ((p = MUTABLE_SV(SvARENA_CHAIN(p)))) { + if (!check_new(st, p)) /* sanity check */ + warn("Free'd SV head unexpectedly already seen"); + ++free_heads; + } + NPathPushNode("unused_sv_heads", NPtype_NAME); + ADD_SIZE(st, "sv", free_heads * sizeof(SV)); + NPathPopNode; + } + /* XXX iterate over bodies_by_type and crawl the free chains for each */ + + /* iterate over all SVs to find any we've not accounted for yet */ + /* once the code above is visiting all SVs, any found here have been leaked */ + unseen_sv_size(aTHX_ st, NPathLink("unaccounted")); + + if (1) { + struct mstats ms = mstats(); + NPathSetNode("unused malloc space", NPtype_NAME); + ADD_SIZE(st, "bytes_free", ms.bytes_free); + ADD_ATTR(st, NPattr_NOTE, "bytes_total", ms.bytes_total); + ADD_ATTR(st, NPattr_NOTE, "bytes_used", ms.bytes_used); + ADD_ATTR(st, NPattr_NOTE, "chunks_used", ms.chunks_used); + ADD_ATTR(st, NPattr_NOTE, "chunks_free", ms.chunks_free); + } + RETVAL = st->total_size; free_state(st); }