X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=hv.c;h=69bbdcff68f9fb5f879545fbc965799a320cf997;hb=6bc89f92370990b836308b02c6551b4e758e7401;hp=5443771fea2145da6d917959b2b83e84ef7aec26;hpb=16580ff596d3844ac4e62fa769c913cc47285a8f;p=p5sagit%2Fp5-mst-13.2.git diff --git a/hv.c b/hv.c index 5443771..69bbdcf 100644 --- a/hv.c +++ b/hv.c @@ -1,7 +1,7 @@ /* hv.c * * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, - * 2000, 2001, 2002, 2003, 2004, 2005, by Larry Wall and others + * 2000, 2001, 2002, 2003, 2004, 2005, 2006, by Larry Wall and others * * You may distribute under the terms of either the GNU General Public * License or the Artistic License, as specified in the README file. @@ -33,17 +33,21 @@ holds the key and hash value. #define HV_MAX_LENGTH_BEFORE_SPLIT 14 +static const char S_strtab_error[] + = "Cannot modify shared string table in hv_%s"; + STATIC void S_more_he(pTHX) { + dVAR; HE* he; HE* heend; - New(54, he, PERL_ARENA_SIZE/sizeof(HE), HE); - HeNEXT(he) = PL_he_arenaroot; - PL_he_arenaroot = he; + Newx(he, PERL_ARENA_SIZE/sizeof(HE), HE); + HeNEXT(he) = (HE*) PL_body_arenaroots[HE_SVSLOT]; + PL_body_arenaroots[HE_SVSLOT] = he; heend = &he[PERL_ARENA_SIZE / sizeof(HE) - 1]; - PL_he_root = ++he; + PL_body_roots[HE_SVSLOT] = ++he; while (he < heend) { HeNEXT(he) = (HE*)(he + 1); he++; @@ -51,37 +55,39 @@ S_more_he(pTHX) HeNEXT(he) = 0; } +#ifdef PURIFY + +#define new_HE() (HE*)safemalloc(sizeof(HE)) +#define del_HE(p) safefree((char*)p) + +#else + STATIC HE* S_new_he(pTHX) { + dVAR; HE* he; + void ** const root = &PL_body_roots[HE_SVSLOT]; + LOCK_SV_MUTEX; - if (!PL_he_root) + if (!*root) S_more_he(aTHX); - he = PL_he_root; - PL_he_root = HeNEXT(he); + he = *root; + *root = HeNEXT(he); UNLOCK_SV_MUTEX; return he; } -STATIC void -S_del_he(pTHX_ HE *p) -{ - LOCK_SV_MUTEX; - HeNEXT(p) = (HE*)PL_he_root; - PL_he_root = p; - UNLOCK_SV_MUTEX; -} - -#ifdef PURIFY - -#define new_HE() (HE*)safemalloc(sizeof(HE)) -#define del_HE(p) safefree((char*)p) +#define new_HE() new_he() +#define del_HE(p) \ + STMT_START { \ + LOCK_SV_MUTEX; \ + HeNEXT(p) = (HE*)(PL_body_roots[HE_SVSLOT]); \ + PL_body_roots[HE_SVSLOT] = p; \ + UNLOCK_SV_MUTEX; \ + } STMT_END -#else -#define new_HE() new_he() -#define del_HE(p) del_he(p) #endif @@ -92,7 +98,7 @@ S_save_hek_flags(pTHX_ const char *str, I32 len, U32 hash, int flags) char *k; register HEK *hek; - New(54, k, HEK_BASESIZE + len + 2, char); + Newx(k, HEK_BASESIZE + len + 2, char); hek = (HEK*)k; Copy(str, HEK_KEY(hek), len, char); HEK_KEY(hek)[len] = 0; @@ -105,17 +111,17 @@ S_save_hek_flags(pTHX_ const char *str, I32 len, U32 hash, int flags) return hek; } -/* free the pool of temporary HE/HEK pairs retunrned by hv_fetch_ent +/* free the pool of temporary HE/HEK pairs returned by hv_fetch_ent * for tied hashes */ void Perl_free_tied_hv_pool(pTHX) { - HE *ohe; + dVAR; HE *he = PL_hv_fetch_ent_mh; while (he) { + HE * const ohe = he; Safefree(HeKEY_hek(he)); - ohe = he; he = HeNEXT(he); del_HE(ohe); } @@ -123,8 +129,28 @@ Perl_free_tied_hv_pool(pTHX) } #if defined(USE_ITHREADS) +HEK * +Perl_hek_dup(pTHX_ HEK *source, CLONE_PARAMS* param) +{ + HEK *shared = (HEK*)ptr_table_fetch(PL_ptr_table, source); + + PERL_UNUSED_ARG(param); + + if (shared) { + /* We already shared this hash key. */ + (void)share_hek_hek(shared); + } + else { + shared + = share_hek_flags(HEK_KEY(source), HEK_LEN(source), + HEK_HASH(source), HEK_FLAGS(source)); + ptr_table_store(PL_ptr_table, source, shared); + } + return shared; +} + HE * -Perl_he_dup(pTHX_ HE *e, bool shared, CLONE_PARAMS* param) +Perl_he_dup(pTHX_ const HE *e, bool shared, CLONE_PARAMS* param) { HE *ret; @@ -142,13 +168,28 @@ Perl_he_dup(pTHX_ HE *e, bool shared, CLONE_PARAMS* param) HeNEXT(ret) = he_dup(HeNEXT(e),shared, param); if (HeKLEN(e) == HEf_SVKEY) { char *k; - New(54, k, HEK_BASESIZE + sizeof(SV*), char); + Newx(k, HEK_BASESIZE + sizeof(SV*), char); HeKEY_hek(ret) = (HEK*)k; HeKEY_sv(ret) = SvREFCNT_inc(sv_dup(HeKEY_sv(e), param)); } - else if (shared) - HeKEY_hek(ret) = share_hek_flags(HeKEY(e), HeKLEN(e), HeHASH(e), - HeKFLAGS(e)); + else if (shared) { + /* This is hek_dup inlined, which seems to be important for speed + reasons. */ + HEK * const source = HeKEY_hek(e); + HEK *shared = (HEK*)ptr_table_fetch(PL_ptr_table, source); + + if (shared) { + /* We already shared this hash key. */ + (void)share_hek_hek(shared); + } + else { + shared + = share_hek_flags(HEK_KEY(source), HEK_LEN(source), + HEK_HASH(source), HEK_FLAGS(source)); + ptr_table_store(PL_ptr_table, source, shared); + } + HeKEY_hek(ret) = shared; + } else HeKEY_hek(ret) = save_hek_flags(HeKEY(e), HeKLEN(e), HeHASH(e), HeKFLAGS(e)); @@ -161,7 +202,7 @@ static void S_hv_notallowed(pTHX_ int flags, const char *key, I32 klen, const char *msg) { - SV *sv = sv_newmortal(); + SV * const sv = sv_newmortal(); if (!(flags & HVhek_FREEKEY)) { sv_setpvn(sv, key, klen); } @@ -229,11 +270,12 @@ Perl_hv_store(pTHX_ HV *hv, const char *key, I32 klen_i32, SV *val, U32 hash) return hek ? &HeVAL(hek) : NULL; } +/* XXX This looks like an ideal candidate to inline */ SV** Perl_hv_store_flags(pTHX_ HV *hv, const char *key, I32 klen, SV *val, register U32 hash, int flags) { - HE *hek = hv_fetch_common (hv, NULL, key, klen, flags, + HE * const hek = hv_fetch_common (hv, NULL, key, klen, flags, (HV_FETCH_ISSTORE|HV_FETCH_JUST_SV), val, hash); return hek ? &HeVAL(hek) : NULL; } @@ -267,6 +309,7 @@ information on how to use this function on tied hashes. =cut */ +/* XXX This looks like an ideal candidate to inline */ HE * Perl_hv_store_ent(pTHX_ HV *hv, SV *keysv, SV *val, U32 hash) { @@ -328,7 +371,7 @@ Perl_hv_fetch(pTHX_ HV *hv, const char *key, I32 klen_i32, I32 lval) flags = 0; } hek = hv_fetch_common (hv, NULL, key, klen, flags, - HV_FETCH_JUST_SV | (lval ? HV_FETCH_LVALUE : 0), + lval ? (HV_FETCH_JUST_SV | HV_FETCH_LVALUE) : HV_FETCH_JUST_SV, Nullsv, 0); return hek ? &HeVAL(hek) : NULL; } @@ -343,6 +386,7 @@ computed. =cut */ +/* XXX This looks like an ideal candidate to inline */ bool Perl_hv_exists_ent(pTHX_ HV *hv, SV *keysv, U32 hash) { @@ -382,7 +426,6 @@ S_hv_fetch_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, { dVAR; XPVHV* xhv; - U32 n_links; HE *entry; HE **oentry; SV *sv; @@ -390,12 +433,12 @@ S_hv_fetch_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, int masked_flags; if (!hv) - return 0; + return NULL; if (keysv) { if (flags & HVhek_FREEKEY) Safefree(key); - key = SvPV(keysv, klen); + key = SvPV_const(keysv, klen); flags = 0; is_utf8 = (SvUTF8(keysv) != 0); } else { @@ -429,7 +472,7 @@ S_hv_fetch_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, else { char *k; entry = new_HE(); - New(54, k, HEK_BASESIZE + sizeof(SV*), char); + Newx(k, HEK_BASESIZE + sizeof(SV*), char); HeKEY_hek(entry) = (HEK*)k; } HeNEXT(entry) = Nullhe; @@ -453,7 +496,7 @@ S_hv_fetch_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, if (isLOWER(key[i])) { /* Would be nice if we had a routine to do the copy and upercase in a single pass through. */ - const char *nkey = strupr(savepvn(key,klen)); + const char * const nkey = strupr(savepvn(key,klen)); /* Note that this fetch is for nkey (the uppercased key) whereas the store is for key (the original) */ entry = hv_fetch_common(hv, Nullsv, nkey, klen, @@ -479,10 +522,9 @@ S_hv_fetch_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, } /* ISFETCH */ else if (SvRMAGICAL(hv) && (action & HV_FETCH_ISEXISTS)) { if (mg_find((SV*)hv, PERL_MAGIC_tied) || SvGMAGICAL((SV*)hv)) { - SV* svret; /* I don't understand why hv_exists_ent has svret and sv, whereas hv_exists only had one. */ - svret = sv_newmortal(); + SV * const svret = sv_newmortal(); sv = sv_newmortal(); if (keysv || is_utf8) { @@ -507,7 +549,7 @@ S_hv_fetch_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, #ifdef ENV_IS_CASELESS else if (mg_find((SV*)hv, PERL_MAGIC_env)) { /* XXX This code isn't UTF8 clean. */ - const char *keysave = key; + char * const keysave = (char * const)key; /* Will need to free this, so set FREEKEY flag. */ key = savepvn(key,klen); key = (const char*)strupr((char*)key); @@ -573,10 +615,13 @@ S_hv_fetch_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, #ifdef DYNAMIC_ENV_FETCH /* if it's an %ENV lookup, we may get it on the fly */ || (SvRMAGICAL((SV*)hv) && mg_find((SV*)hv, PERL_MAGIC_env)) #endif - ) - Newz(503, HvARRAY(hv), + ) { + char *array; + Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */), - HE*); + char); + HvARRAY(hv) = (HE**)array; + } #ifdef DYNAMIC_ENV_FETCH else if (action & HV_FETCH_ISEXISTS) { /* for an %ENV exists, if we do an insert it's by a recursive @@ -593,7 +638,7 @@ S_hv_fetch_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, } if (is_utf8) { - const char *keysave = key; + char * const keysave = (char * const)key; key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8); if (is_utf8) flags |= HVhek_UTF8; @@ -616,14 +661,13 @@ S_hv_fetch_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, flags |= HVhek_REHASH; } else if (!hash) { if (keysv && (SvIsCOW_shared_hash(keysv))) { - hash = SvUVX(keysv); + hash = SvSHARED_HASH(keysv); } else { PERL_HASH(hash, key, klen); } } masked_flags = (flags & HVhek_MASK); - n_links = 0; #ifdef DYNAMIC_ENV_FETCH if (!HvARRAY(hv)) entry = Null(HE*); @@ -632,7 +676,7 @@ S_hv_fetch_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, { entry = (HvARRAY(hv))[hash & (I32) HvMAX(hv)]; } - for (; entry; ++n_links, entry = HeNEXT(entry)) { + for (; entry; entry = HeNEXT(entry)) { if (HeHASH(entry) != hash) /* strings can't be equal */ continue; if (HeKLEN(entry) != (I32)klen) @@ -657,6 +701,14 @@ S_hv_fetch_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, unshare_hek (HeKEY_hek(entry)); HeKEY_hek(entry) = new_hek; } + else if (hv == PL_strtab) { + /* PL_strtab is usually the only hash without HvSHAREKEYS, + so putting this test here is cheap */ + if (flags & HVhek_FREEKEY) + Safefree(key); + Perl_croak(aTHX_ S_strtab_error, + action & HV_FETCH_LVALUE ? "fetch" : "store"); + } else HeKFLAGS(entry) = masked_flags; if (masked_flags & HVhek_ENABLEHVKFLAGS) @@ -703,7 +755,7 @@ S_hv_fetch_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, if (!(action & HV_FETCH_ISSTORE) && SvRMAGICAL((SV*)hv) && mg_find((SV*)hv, PERL_MAGIC_env)) { unsigned long len; - char *env = PerlEnv_ENVgetenv_len(key,&len); + const char * const env = PerlEnv_ENVgetenv_len(key,&len); if (env) { sv = newSVpvn(env,len); SvTAINTED_on(sv); @@ -714,7 +766,7 @@ S_hv_fetch_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, #endif if (!entry && SvREADONLY(hv) && !(action & HV_FETCH_ISEXISTS)) { - S_hv_notallowed(aTHX_ flags, key, klen, + hv_notallowed(flags, key, klen, "Attempt to access disallowed key '%"SVf"' in" " a restricted hash"); } @@ -744,9 +796,11 @@ S_hv_fetch_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, /* Not sure if we can get here. I think the only case of oentry being NULL is for %ENV with dynamic env fetch. But that should disappear with magic in the previous code. */ - Newz(503, HvARRAY(hv), + char *array; + Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */), - HE*); + char); + HvARRAY(hv) = (HE**)array; } oentry = &(HvARRAY(hv))[hash & (I32) xhv->xhv_max]; @@ -756,6 +810,14 @@ S_hv_fetch_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, bad API design. */ if (HvSHAREKEYS(hv)) HeKEY_hek(entry) = share_hek_flags(key, klen, hash, flags); + else if (hv == PL_strtab) { + /* PL_strtab is usually the only hash without HvSHAREKEYS, so putting + this test here is cheap */ + if (flags & HVhek_FREEKEY) + Safefree(key); + Perl_croak(aTHX_ S_strtab_error, + action & HV_FETCH_LVALUE ? "fetch" : "store"); + } else /* gotta do the real thing */ HeKEY_hek(entry) = save_hek_flags(key, klen, hash, flags); HeVAL(entry) = val; @@ -767,18 +829,30 @@ S_hv_fetch_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, if (masked_flags & HVhek_ENABLEHVKFLAGS) HvHASKFLAGS_on(hv); - xhv->xhv_keys++; /* HvKEYS(hv)++ */ - if (!n_links) { /* initial entry? */ - xhv->xhv_fill++; /* HvFILL(hv)++ */ - } else if ((xhv->xhv_keys > (IV)xhv->xhv_max) - || ((n_links > HV_MAX_LENGTH_BEFORE_SPLIT) && !HvREHASH(hv))) { - /* Use only the old HvKEYS(hv) > HvMAX(hv) condition to limit bucket - splits on a rehashed hash, as we're not going to split it again, - and if someone is lucky (evil) enough to get all the keys in one - list they could exhaust our memory as we repeatedly double the - number of buckets on every entry. Linear search feels a less worse - thing to do. */ - hsplit(hv); + { + const HE *counter = HeNEXT(entry); + + xhv->xhv_keys++; /* HvKEYS(hv)++ */ + if (!counter) { /* initial entry? */ + xhv->xhv_fill++; /* HvFILL(hv)++ */ + } else if (xhv->xhv_keys > (IV)xhv->xhv_max) { + hsplit(hv); + } else if(!HvREHASH(hv)) { + U32 n_links = 1; + + while ((counter = HeNEXT(counter))) + n_links++; + + if (n_links > HV_MAX_LENGTH_BEFORE_SPLIT) { + /* Use only the old HvKEYS(hv) > HvMAX(hv) condition to limit + bucket splits on a rehashed hash, as we're not going to + split it again, and if someone is lucky (evil) enough to + get all the keys in one list they could exhaust our memory + as we repeatedly double the number of buckets on every + entry. Linear search feels a less worse thing to do. */ + hsplit(hv); + } + } } return entry; @@ -793,10 +867,9 @@ S_hv_magic_check(pTHX_ HV *hv, bool *needs_copy, bool *needs_store) while (mg) { if (isUPPER(mg->mg_type)) { *needs_copy = TRUE; - switch (mg->mg_type) { - case PERL_MAGIC_tied: - case PERL_MAGIC_sig: + if (mg->mg_type == PERL_MAGIC_tied) { *needs_store = FALSE; + return; /* We've set all there is to set. */ } } mg = mg->mg_moremagic; @@ -814,13 +887,13 @@ Evaluates the hash in scalar context and returns the result. Handles magic when SV * Perl_hv_scalar(pTHX_ HV *hv) { - MAGIC *mg; SV *sv; - - if ((SvRMAGICAL(hv) && (mg = mg_find((SV*)hv, PERL_MAGIC_tied)))) { - sv = magic_scalarpack(hv, mg); - return sv; - } + + if (SvRMAGICAL(hv)) { + MAGIC * const mg = mg_find((SV*)hv, PERL_MAGIC_tied); + if (mg) + return magic_scalarpack(hv, mg); + } sv = sv_newmortal(); if (HvFILL((HV*)hv)) @@ -869,6 +942,7 @@ precomputed hash value, or 0 to ask for it to be computed. =cut */ +/* XXX This looks like an ideal candidate to inline */ SV * Perl_hv_delete_ent(pTHX_ HV *hv, SV *keysv, I32 flags, U32 hash) { @@ -881,9 +955,9 @@ S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, { dVAR; register XPVHV* xhv; - register I32 i; register HE *entry; register HE **oentry; + HE *const *first_entry; SV *sv; bool is_utf8; int masked_flags; @@ -894,7 +968,7 @@ S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, if (keysv) { if (k_flags & HVhek_FREEKEY) Safefree(key); - key = SvPV(keysv, klen); + key = SvPV_const(keysv, klen); k_flags = 0; is_utf8 = (SvUTF8(keysv) != 0); } else { @@ -944,8 +1018,8 @@ S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, return Nullsv; if (is_utf8) { - const char *keysave = key; - key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8); + const char * const keysave = key; + key = (char*)bytes_from_utf8((U8*)key, &klen, &is_utf8); if (is_utf8) k_flags |= HVhek_UTF8; @@ -966,7 +1040,7 @@ S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, PERL_HASH_INTERNAL(hash, key, klen); } else if (!hash) { if (keysv && (SvIsCOW_shared_hash(keysv))) { - hash = SvUVX(keysv); + hash = SvSHARED_HASH(keysv); } else { PERL_HASH(hash, key, klen); } @@ -974,10 +1048,9 @@ S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, masked_flags = (k_flags & HVhek_MASK); - oentry = &(HvARRAY(hv))[hash & (I32) HvMAX(hv)]; + first_entry = oentry = &(HvARRAY(hv))[hash & (I32) HvMAX(hv)]; entry = *oentry; - i = 1; - for (; entry; i=0, oentry = &HeNEXT(entry), entry = *oentry) { + for (; entry; oentry = &HeNEXT(entry), entry = *oentry) { if (HeHASH(entry) != hash) /* strings can't be equal */ continue; if (HeKLEN(entry) != (I32)klen) @@ -987,6 +1060,12 @@ S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, if ((HeKFLAGS(entry) ^ masked_flags) & HVhek_UTF8) continue; + if (hv == PL_strtab) { + if (k_flags & HVhek_FREEKEY) + Safefree(key); + Perl_croak(aTHX_ S_strtab_error, "delete"); + } + /* if placeholder is here, it's already been deleted.... */ if (HeVAL(entry) == &PL_sv_placeholder) { @@ -1023,10 +1102,10 @@ S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, HvPLACEHOLDERS(hv)++; } else { *oentry = HeNEXT(entry); - if (i && !*oentry) + if(!*first_entry) { xhv->xhv_fill--; /* HvFILL(hv)-- */ - if (xhv->xhv_aux && entry - == ((struct xpvhv_aux *)xhv->xhv_aux)->xhv_eiter /* HvEITER(hv) */) + } + if (SvOOK(hv) && entry == HvAUX(hv)->xhv_eiter /* HvEITER(hv) */) HvLAZYDEL_on(hv); else hv_free_ent(hv, entry); @@ -1050,6 +1129,7 @@ S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, STATIC void S_hsplit(pTHX_ HV *hv) { + dVAR; register XPVHV* xhv = (XPVHV*)SvANY(hv); const I32 oldsize = (I32) xhv->xhv_max+1; /* HvMAX(hv)+1 (sick) */ register I32 newsize = oldsize * 2; @@ -1063,7 +1143,7 @@ S_hsplit(pTHX_ HV *hv) /*PerlIO_printf(PerlIO_stderr(), "hsplit called for %p which had %d\n", hv, (int) oldsize);*/ - if (HvPLACEHOLDERS(hv) && !SvREADONLY(hv)) { + if (HvPLACEHOLDERS_get(hv) && !SvREADONLY(hv)) { /* Can make this clear any placeholders first for non-restricted hashes, even though Storable rebuilds restricted hashes by putting in all the placeholders (first) before turning on the readonly flag, because @@ -1073,21 +1153,30 @@ S_hsplit(pTHX_ HV *hv) PL_nomemok = TRUE; #if defined(STRANGE_MALLOC) || defined(MYMALLOC) - Renew(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char); + Renew(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize) + + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char); if (!a) { PL_nomemok = FALSE; return; } + if (SvOOK(hv)) { + Copy(&a[oldsize * sizeof(HE*)], &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux); + } #else - New(2, a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char); + Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize) + + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char); if (!a) { PL_nomemok = FALSE; return; } Copy(HvARRAY(hv), a, oldsize * sizeof(HE*), char); + if (SvOOK(hv)) { + Copy(HvAUX(hv), &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux); + } if (oldsize >= 64) { offer_nice_chunk(HvARRAY(hv), - PERL_HV_ARRAY_ALLOC_BYTES(oldsize)); + PERL_HV_ARRAY_ALLOC_BYTES(oldsize) + + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0)); } else Safefree(HvARRAY(hv)); @@ -1152,7 +1241,12 @@ S_hsplit(pTHX_ HV *hv) longest_chain, HvTOTALKEYS(hv), HvFILL(hv), 1+HvMAX(hv));*/ ++newsize; - Newz(2, a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char); + Newxz(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize) + + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char); + if (SvOOK(hv)) { + Copy(HvAUX(hv), &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux); + } + was_shared = HvSHAREKEYS(hv); xhv->xhv_fill = 0; @@ -1166,7 +1260,7 @@ S_hsplit(pTHX_ HV *hv) while (entry) { /* We're going to trash this HE's next pointer when we chain it into the new hash below, so store where we go next. */ - HE *next = HeNEXT(entry); + HE * const next = HeNEXT(entry); UV hash; HE **bep; @@ -1175,7 +1269,7 @@ S_hsplit(pTHX_ HV *hv) if (was_shared) { /* Unshare it. */ - HEK *new_hek + HEK * const new_hek = save_hek_flags(HeKEY(entry), HeKLEN(entry), hash, HeKFLAGS(entry)); unshare_hek (HeKEY_hek(entry)); @@ -1205,6 +1299,7 @@ S_hsplit(pTHX_ HV *hv) void Perl_hv_ksplit(pTHX_ HV *hv, IV newmax) { + dVAR; register XPVHV* xhv = (XPVHV*)SvANY(hv); const I32 oldsize = (I32) xhv->xhv_max+1; /* HvMAX(hv)+1 (sick) */ register I32 newsize; @@ -1229,21 +1324,30 @@ Perl_hv_ksplit(pTHX_ HV *hv, IV newmax) if (a) { PL_nomemok = TRUE; #if defined(STRANGE_MALLOC) || defined(MYMALLOC) - Renew(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char); + Renew(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize) + + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char); if (!a) { PL_nomemok = FALSE; return; } + if (SvOOK(hv)) { + Copy(&a[oldsize * sizeof(HE*)], &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux); + } #else - New(2, a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char); + Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize) + + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0), char); if (!a) { PL_nomemok = FALSE; return; } Copy(HvARRAY(hv), a, oldsize * sizeof(HE*), char); + if (SvOOK(hv)) { + Copy(HvAUX(hv), &a[newsize * sizeof(HE*)], 1, struct xpvhv_aux); + } if (oldsize >= 64) { offer_nice_chunk(HvARRAY(hv), - PERL_HV_ARRAY_ALLOC_BYTES(oldsize)); + PERL_HV_ARRAY_ALLOC_BYTES(oldsize) + + (SvOOK(hv) ? sizeof(struct xpvhv_aux) : 0)); } else Safefree(HvARRAY(hv)); @@ -1252,7 +1356,7 @@ Perl_hv_ksplit(pTHX_ HV *hv, IV newmax) Zero(&a[oldsize * sizeof(HE*)], (newsize-oldsize) * sizeof(HE*), char); /* zero 2nd half*/ } else { - Newz(0, a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char); + Newxz(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char); } xhv->xhv_max = --newsize; /* HvMAX(hv) = --newsize */ HvARRAY(hv) = (HE **) a; @@ -1292,10 +1396,9 @@ Creates a new HV. The reference count is set to 1. HV * Perl_newHV(pTHX) { - register HV *hv; register XPVHV* xhv; + HV * const hv = (HV*)NEWSV(502,0); - hv = (HV*)NEWSV(502,0); sv_upgrade((SV *)hv, SVt_PVHV); xhv = (XPVHV*)SvANY(hv); SvPOK_off(hv); @@ -1306,14 +1409,13 @@ Perl_newHV(pTHX) xhv->xhv_max = 7; /* HvMAX(hv) = 7 (start with 8 buckets) */ xhv->xhv_fill = 0; /* HvFILL(hv) = 0 */ - xhv->xhv_aux = 0; return hv; } HV * Perl_newHVhv(pTHX_ HV *ohv) { - HV *hv = newHV(); + HV * const hv = newHV(); STRLEN hv_max, hv_fill; if (!ohv || (hv_fill = HvFILL(ohv)) == 0) @@ -1324,14 +1426,15 @@ Perl_newHVhv(pTHX_ HV *ohv) /* It's an ordinary hash, so copy it fast. AMS 20010804 */ STRLEN i; const bool shared = !!HvSHAREKEYS(ohv); - HE **ents, **oents = (HE **)HvARRAY(ohv); + HE **ents, ** const oents = (HE **)HvARRAY(ohv); char *a; - New(0, a, PERL_HV_ARRAY_ALLOC_BYTES(hv_max+1), char); + Newx(a, PERL_HV_ARRAY_ALLOC_BYTES(hv_max+1), char); ents = (HE**)a; /* In each bucket... */ for (i = 0; i <= hv_max; i++) { - HE *prev = NULL, *ent = NULL, *oent = oents[i]; + HE *prev = NULL, *ent = NULL; + HE *oent = oents[i]; if (!oent) { ents[i] = NULL; @@ -1339,7 +1442,7 @@ Perl_newHVhv(pTHX_ HV *ohv) } /* Copy the linked list of entries. */ - for (oent = oents[i]; oent; oent = HeNEXT(oent)) { + for (; oent; oent = HeNEXT(oent)) { const U32 hash = HeHASH(oent); const char * const key = HeKEY(oent); const STRLEN len = HeKLEN(oent); @@ -1363,7 +1466,7 @@ Perl_newHVhv(pTHX_ HV *ohv) HvFILL(hv) = hv_fill; HvTOTALKEYS(hv) = HvTOTALKEYS(ohv); HvARRAY(hv) = ents; - } + } /* not magical */ else { /* Iterate over ohv, copying keys and values one at a time. */ HE *entry; @@ -1391,6 +1494,7 @@ Perl_newHVhv(pTHX_ HV *ohv) void Perl_hv_free_ent(pTHX_ HV *hv, register HE *entry) { + dVAR; SV *val; if (!entry) @@ -1413,20 +1517,15 @@ Perl_hv_free_ent(pTHX_ HV *hv, register HE *entry) void Perl_hv_delayfree_ent(pTHX_ HV *hv, register HE *entry) { + dVAR; if (!entry) return; - if (isGV(HeVAL(entry)) && GvCVu(HeVAL(entry)) && HvNAME_get(hv)) - PL_sub_generation++; /* may be deletion of method from stash */ - sv_2mortal(HeVAL(entry)); /* free between statements */ + /* SvREFCNT_inc to counter the SvREFCNT_dec in hv_free_ent */ + sv_2mortal(SvREFCNT_inc(HeVAL(entry))); /* free between statements */ if (HeKLEN(entry) == HEf_SVKEY) { - sv_2mortal(HeKEY_sv(entry)); - Safefree(HeKEY_hek(entry)); + sv_2mortal(SvREFCNT_inc(HeKEY_sv(entry))); } - else if (HvSHAREKEYS(hv)) - unshare_hek(HeKEY_hek(entry)); - else - Safefree(HeKEY_hek(entry)); - del_HE(entry); + hv_free_ent(hv, entry); } /* @@ -1451,8 +1550,8 @@ Perl_hv_clear(pTHX_ HV *hv) if (SvREADONLY(hv) && HvARRAY(hv) != NULL) { /* restricted hash: convert all keys to placeholders */ - I32 i; - for (i = 0; i <= (I32) xhv->xhv_max; i++) { + STRLEN i; + for (i = 0; i <= xhv->xhv_max; i++) { HE *entry = (HvARRAY(hv))[i]; for (; entry; entry = HeNEXT(entry)) { /* not already placeholder */ @@ -1484,7 +1583,7 @@ Perl_hv_clear(pTHX_ HV *hv) HvHASKFLAGS_off(hv); HvREHASH_off(hv); reset: - if (xhv->xhv_aux) { + if (SvOOK(hv)) { HvEITER_set(hv, NULL); } } @@ -1507,37 +1606,35 @@ void Perl_hv_clear_placeholders(pTHX_ HV *hv) { dVAR; - I32 items = (I32)HvPLACEHOLDERS(hv); - I32 i = HvMAX(hv); + I32 items = (I32)HvPLACEHOLDERS_get(hv); + I32 i; if (items == 0) return; + i = HvMAX(hv); do { /* Loop down the linked list heads */ bool first = 1; HE **oentry = &(HvARRAY(hv))[i]; - HE *entry = *oentry; - - if (!entry) - continue; + HE *entry; - for (; entry; entry = *oentry) { + while ((entry = *oentry)) { if (HeVAL(entry) == &PL_sv_placeholder) { *oentry = HeNEXT(entry); if (first && !*oentry) HvFILL(hv)--; /* This linked list is now empty. */ - if (HvEITER_get(hv)) + if (entry == HvEITER_get(hv)) HvLAZYDEL_on(hv); else hv_free_ent(hv, entry); if (--items == 0) { /* Finished. */ - HvTOTALKEYS(hv) -= (IV)HvPLACEHOLDERS(hv); + HvTOTALKEYS(hv) -= (IV)HvPLACEHOLDERS_get(hv); if (HvKEYS(hv) == 0) HvHASKFLAGS_off(hv); - HvPLACEHOLDERS(hv) = 0; + HvPLACEHOLDERS_set(hv, 0); return; } } else { @@ -1554,51 +1651,136 @@ Perl_hv_clear_placeholders(pTHX_ HV *hv) STATIC void S_hfreeentries(pTHX_ HV *hv) { - register HE **array; - register HE *entry; - I32 riter; - I32 max; - struct xpvhv_aux *iter; + /* This is the array that we're going to restore */ + HE **orig_array; + HEK *name; + int attempts = 100; - if (!hv) - return; if (!HvARRAY(hv)) return; - riter = 0; - max = HvMAX(hv); - array = HvARRAY(hv); - /* make everyone else think the array is empty, so that the destructors - * called for freed entries can't recusively mess with us */ - HvARRAY(hv) = Null(HE**); - HvFILL(hv) = 0; - ((XPVHV*) SvANY(hv))->xhv_keys = 0; - - entry = array[0]; - for (;;) { - if (entry) { - register HE *oentry = entry; - entry = HeNEXT(entry); - hv_free_ent(hv, oentry); + if (SvOOK(hv)) { + /* If the hash is actually a symbol table with a name, look after the + name. */ + struct xpvhv_aux *iter = HvAUX(hv); + + name = iter->xhv_name; + iter->xhv_name = NULL; + } else { + name = NULL; + } + + orig_array = HvARRAY(hv); + /* orig_array remains unchanged throughout the loop. If after freeing all + the entries it turns out that one of the little blighters has triggered + an action that has caused HvARRAY to be re-allocated, then we set + array to the new HvARRAY, and try again. */ + + while (1) { + /* This is the one we're going to try to empty. First time round + it's the original array. (Hopefully there will only be 1 time + round) */ + HE **array = HvARRAY(hv); + I32 i = HvMAX(hv); + + /* Because we have taken xhv_name out, the only allocated pointer + in the aux structure that might exist is the backreference array. + */ + + if (SvOOK(hv)) { + HE *entry; + struct xpvhv_aux *iter = HvAUX(hv); + /* If there are weak references to this HV, we need to avoid + freeing them up here. In particular we need to keep the AV + visible as what we're deleting might well have weak references + back to this HV, so the for loop below may well trigger + the removal of backreferences from this array. */ + + if (iter->xhv_backreferences) { + /* So donate them to regular backref magic to keep them safe. + The sv_magic will increase the reference count of the AV, + so we need to drop it first. */ + SvREFCNT_dec(iter->xhv_backreferences); + if (AvFILLp(iter->xhv_backreferences) == -1) { + /* Turns out that the array is empty. Just free it. */ + SvREFCNT_dec(iter->xhv_backreferences); + + } else { + sv_magic((SV*)hv, (SV*)iter->xhv_backreferences, + PERL_MAGIC_backref, NULL, 0); + } + iter->xhv_backreferences = NULL; + } + + entry = iter->xhv_eiter; /* HvEITER(hv) */ + if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */ + HvLAZYDEL_off(hv); + hv_free_ent(hv, entry); + } + iter->xhv_riter = -1; /* HvRITER(hv) = -1 */ + iter->xhv_eiter = Null(HE*); /* HvEITER(hv) = Null(HE*) */ + + /* There are now no allocated pointers in the aux structure. */ + + SvFLAGS(hv) &= ~SVf_OOK; /* Goodbye, aux structure. */ + /* What aux structure? */ } - if (!entry) { - if (++riter > max) - break; - entry = array[riter]; + + /* make everyone else think the array is empty, so that the destructors + * called for freed entries can't recusively mess with us */ + HvARRAY(hv) = NULL; + HvFILL(hv) = 0; + ((XPVHV*) SvANY(hv))->xhv_keys = 0; + + + do { + /* Loop down the linked list heads */ + HE *entry = array[i]; + + while (entry) { + register HE * const oentry = entry; + entry = HeNEXT(entry); + hv_free_ent(hv, oentry); + } + } while (--i >= 0); + + /* As there are no allocated pointers in the aux structure, it's now + safe to free the array we just cleaned up, if it's not the one we're + going to put back. */ + if (array != orig_array) { + Safefree(array); } - } - HvARRAY(hv) = array; - iter = ((XPVHV*) SvANY(hv))->xhv_aux; - if (iter) { - entry = iter->xhv_eiter; /* HvEITER(hv) */ - if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */ - HvLAZYDEL_off(hv); - hv_free_ent(hv, entry); + if (!HvARRAY(hv)) { + /* Good. No-one added anything this time round. */ + break; + } + + if (SvOOK(hv)) { + /* Someone attempted to iterate or set the hash name while we had + the array set to 0. We'll catch backferences on the next time + round the while loop. */ + assert(HvARRAY(hv)); + + if (HvAUX(hv)->xhv_name) { + unshare_hek_or_pvn(HvAUX(hv)->xhv_name, 0, 0, 0); + } } - Safefree(iter->xhv_name); - Safefree(iter); - ((XPVHV*) SvANY(hv))->xhv_aux = 0; + + if (--attempts == 0) { + Perl_die(aTHX_ "panic: hfreeentries failed to free hash - something is repeatedly re-creating entries"); + } + }; + + HvARRAY(hv) = orig_array; + + /* If the hash was actually a symbol table, put the name back. */ + if (name) { + /* We have restored the original array. If name is non-NULL, then + the original array had an aux structure at the end. So this is + valid: */ + SvFLAGS(hv) |= SVf_OOK; + HvAUX(hv)->xhv_name = name; } } @@ -1613,20 +1795,22 @@ Undefines the hash. void Perl_hv_undef(pTHX_ HV *hv) { + dVAR; register XPVHV* xhv; const char *name; + if (!hv) return; DEBUG_A(Perl_hv_assert(aTHX_ hv)); xhv = (XPVHV*)SvANY(hv); hfreeentries(hv); - Safefree(HvARRAY(hv)); if ((name = HvNAME_get(hv))) { - /* FIXME - strlen HvNAME */ if(PL_stashcache) - hv_delete(PL_stashcache, name, strlen(name), G_DISCARD); - Perl_hv_name_set(aTHX_ hv, 0, 0, 0); + hv_delete(PL_stashcache, name, HvNAMELEN_get(hv), G_DISCARD); + hv_name_set(hv, Nullch, 0, 0); } + SvFLAGS(hv) &= ~SVf_OOK; + Safefree(HvARRAY(hv)); xhv->xhv_max = 7; /* HvMAX(hv) = 7 (it's a normal hash) */ HvARRAY(hv) = 0; HvPLACEHOLDERS_set(hv, 0); @@ -1635,16 +1819,28 @@ Perl_hv_undef(pTHX_ HV *hv) mg_clear((SV*)hv); } -struct xpvhv_aux* -S_hv_auxinit(pTHX) { +static struct xpvhv_aux* +S_hv_auxinit(pTHX_ HV *hv) { struct xpvhv_aux *iter; + char *array; - New(0, iter, 1, struct xpvhv_aux); + if (!HvARRAY(hv)) { + Newxz(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1) + + sizeof(struct xpvhv_aux), char); + } else { + array = (char *) HvARRAY(hv); + Renew(array, PERL_HV_ARRAY_ALLOC_BYTES(HvMAX(hv) + 1) + + sizeof(struct xpvhv_aux), char); + } + HvARRAY(hv) = (HE**) array; + /* SvOOK_on(hv) attacks the IV flags. */ + SvFLAGS(hv) |= SVf_OOK; + iter = HvAUX(hv); iter->xhv_riter = -1; /* HvRITER(hv) = -1 */ iter->xhv_eiter = Null(HE*); /* HvEITER(hv) = Null(HE*) */ iter->xhv_name = 0; - + iter->xhv_backreferences = 0; return iter; } @@ -1666,17 +1862,12 @@ value, you can get it through the macro C. I32 Perl_hv_iterinit(pTHX_ HV *hv) { - register XPVHV* xhv; - HE *entry; - struct xpvhv_aux *iter; - if (!hv) Perl_croak(aTHX_ "Bad hash"); - xhv = (XPVHV*)SvANY(hv); - iter = xhv->xhv_aux; - if (iter) { - entry = iter->xhv_eiter; /* HvEITER(hv) */ + if (SvOOK(hv)) { + struct xpvhv_aux *iter = HvAUX(hv); + HE * const entry = iter->xhv_eiter; /* HvEITER(hv) */ if (entry && HvLAZYDEL(hv)) { /* was deleted earlier? */ HvLAZYDEL_off(hv); hv_free_ent(hv, entry); @@ -1684,11 +1875,11 @@ Perl_hv_iterinit(pTHX_ HV *hv) iter->xhv_riter = -1; /* HvRITER(hv) = -1 */ iter->xhv_eiter = Null(HE*); /* HvEITER(hv) = Null(HE*) */ } else { - xhv->xhv_aux = S_hv_auxinit(aTHX); + S_hv_auxinit(aTHX_ hv); } /* used to be xhv->xhv_fill before 5.004_65 */ - return XHvTOTALKEYS(xhv); + return HvTOTALKEYS(hv); } I32 * @@ -1698,10 +1889,7 @@ Perl_hv_riter_p(pTHX_ HV *hv) { if (!hv) Perl_croak(aTHX_ "Bad hash"); - iter = ((XPVHV *)SvANY(hv))->xhv_aux; - if (!iter) { - ((XPVHV *)SvANY(hv))->xhv_aux = iter = S_hv_auxinit(aTHX); - } + iter = SvOOK(hv) ? HvAUX(hv) : S_hv_auxinit(aTHX_ hv); return &(iter->xhv_riter); } @@ -1712,10 +1900,7 @@ Perl_hv_eiter_p(pTHX_ HV *hv) { if (!hv) Perl_croak(aTHX_ "Bad hash"); - iter = ((XPVHV *)SvANY(hv))->xhv_aux; - if (!iter) { - ((XPVHV *)SvANY(hv))->xhv_aux = iter = S_hv_auxinit(aTHX); - } + iter = SvOOK(hv) ? HvAUX(hv) : S_hv_auxinit(aTHX_ hv); return &(iter->xhv_eiter); } @@ -1726,13 +1911,13 @@ Perl_hv_riter_set(pTHX_ HV *hv, I32 riter) { if (!hv) Perl_croak(aTHX_ "Bad hash"); - - iter = ((XPVHV *)SvANY(hv))->xhv_aux; - if (!iter) { + if (SvOOK(hv)) { + iter = HvAUX(hv); + } else { if (riter == -1) return; - ((XPVHV *)SvANY(hv))->xhv_aux = iter = S_hv_auxinit(aTHX); + iter = S_hv_auxinit(aTHX_ hv); } iter->xhv_riter = riter; } @@ -1744,47 +1929,69 @@ Perl_hv_eiter_set(pTHX_ HV *hv, HE *eiter) { if (!hv) Perl_croak(aTHX_ "Bad hash"); - iter = ((XPVHV *)SvANY(hv))->xhv_aux; - if (!iter) { + if (SvOOK(hv)) { + iter = HvAUX(hv); + } else { /* 0 is the default so don't go malloc()ing a new structure just to hold 0. */ if (!eiter) return; - ((XPVHV *)SvANY(hv))->xhv_aux = iter = S_hv_auxinit(aTHX); + iter = S_hv_auxinit(aTHX_ hv); } iter->xhv_eiter = eiter; } - -char ** -Perl_hv_name_p(pTHX_ HV *hv) +void +Perl_hv_name_set(pTHX_ HV *hv, const char *name, I32 len, int flags) { - struct xpvhv_aux *iter = ((XPVHV *)SvANY(hv))->xhv_aux; + dVAR; + struct xpvhv_aux *iter; + U32 hash; + + PERL_UNUSED_ARG(flags); + + if (SvOOK(hv)) { + iter = HvAUX(hv); + if (iter->xhv_name) { + unshare_hek_or_pvn(iter->xhv_name, 0, 0, 0); + } + } else { + if (name == 0) + return; - if (!iter) { - ((XPVHV *)SvANY(hv))->xhv_aux = iter = S_hv_auxinit(aTHX); + iter = S_hv_auxinit(aTHX_ hv); } - return &(iter->xhv_name); + PERL_HASH(hash, name, len); + iter->xhv_name = name ? share_hek(name, len, hash) : 0; +} + +AV ** +Perl_hv_backreferences_p(pTHX_ HV *hv) { + struct xpvhv_aux *iter; + + iter = SvOOK(hv) ? HvAUX(hv) : S_hv_auxinit(aTHX_ hv); + return &(iter->xhv_backreferences); } void -Perl_hv_name_set(pTHX_ HV *hv, const char *name, STRLEN len, int flags) -{ - struct xpvhv_aux *iter = ((XPVHV *)SvANY(hv))->xhv_aux; +Perl_hv_kill_backrefs(pTHX_ HV *hv) { + AV *av; - if (iter) { - Safefree(iter->xhv_name); - } else { - if (name == 0) - return; + if (!SvOOK(hv)) + return; - ((XPVHV *)SvANY(hv))->xhv_aux = iter = S_hv_auxinit(aTHX); + av = HvAUX(hv)->xhv_backreferences; + + if (av) { + HvAUX(hv)->xhv_backreferences = 0; + Perl_sv_kill_backrefs(aTHX_ (SV*) hv, av); } - iter->xhv_name = savepvn(name, len); } /* +hv_iternext is implemented as a macro in hv.h + =for apidoc hv_iternext Returns entries from a hash iterator. See C. @@ -1797,16 +2004,6 @@ to free the entry on the next call to C, so you must not discard your iterator immediately else the entry will leak - call C to trigger the resource deallocation. -=cut -*/ - -HE * -Perl_hv_iternext(pTHX_ HV *hv) -{ - return hv_iternext_flags(hv, 0); -} - -/* =for apidoc hv_iternext_flags Returns entries from a hash iterator. See C and C. @@ -1834,20 +2031,19 @@ Perl_hv_iternext_flags(pTHX_ HV *hv, I32 flags) if (!hv) Perl_croak(aTHX_ "Bad hash"); xhv = (XPVHV*)SvANY(hv); - iter = xhv->xhv_aux; - if (!iter) { + if (!SvOOK(hv)) { /* Too many things (well, pp_each at least) merrily assume that you can call iv_iternext without calling hv_iterinit, so we'll have to deal with it. */ hv_iterinit(hv); - iter = ((XPVHV *)SvANY(hv))->xhv_aux; } + iter = HvAUX(hv); oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */ if ((mg = SvTIED_mg((SV*)hv, PERL_MAGIC_tied))) { - SV *key = sv_newmortal(); + SV * const key = sv_newmortal(); if (entry) { sv_setsv(key, HeSVKEY_force(entry)); SvREFCNT_dec(HeSVKEY(entry)); /* get rid of previous key */ @@ -1859,7 +2055,7 @@ Perl_hv_iternext_flags(pTHX_ HV *hv, I32 flags) /* one HE per MAGICAL hash */ iter->xhv_eiter = entry = new_HE(); /* HvEITER(hv) = new_HE() */ Zero(entry, 1, HE); - Newz(54, k, HEK_BASESIZE + sizeof(SV*), char); + Newxz(k, HEK_BASESIZE + sizeof(SV*), char); hek = (HEK*)k; HeKEY_hek(entry) = hek; HeKLEN(entry) = HEf_SVKEY; @@ -1878,17 +2074,22 @@ Perl_hv_iternext_flags(pTHX_ HV *hv, I32 flags) return Null(HE*); } #ifdef DYNAMIC_ENV_FETCH /* set up %ENV for iteration */ - if (!entry && SvRMAGICAL((SV*)hv) && mg_find((SV*)hv, PERL_MAGIC_env)) + if (!entry && SvRMAGICAL((SV*)hv) && mg_find((SV*)hv, PERL_MAGIC_env)) { prime_env_iter(); +#ifdef VMS + /* The prime_env_iter() on VMS just loaded up new hash values + * so the iteration count needs to be reset back to the beginning + */ + hv_iterinit(hv); + iter = HvAUX(hv); + oldentry = entry = iter->xhv_eiter; /* HvEITER(hv) */ #endif - - if (!HvARRAY(hv)) { - char *darray; - Newz(506, darray, - PERL_HV_ARRAY_ALLOC_BYTES(xhv->xhv_max+1 /* HvMAX(hv)+1 */), - char); - HvARRAY(hv) = (HE**) darray; } +#endif + + /* hv_iterint now ensures this. */ + assert (HvARRAY(hv)); + /* At start of hash, entry is NULL. */ if (entry) { @@ -1951,7 +2152,7 @@ Perl_hv_iterkey(pTHX_ register HE *entry, I32 *retlen) { if (HeKLEN(entry) == HEf_SVKEY) { STRLEN len; - char *p = SvPV(HeKEY_sv(entry), len); + char * const p = SvPV(HeKEY_sv(entry), len); *retlen = len; return p; } @@ -1975,39 +2176,7 @@ see C. SV * Perl_hv_iterkeysv(pTHX_ register HE *entry) { - if (HeKLEN(entry) != HEf_SVKEY) { - HEK *hek = HeKEY_hek(entry); - const int flags = HEK_FLAGS(hek); - SV *sv; - - if (flags & HVhek_WASUTF8) { - /* Trouble :-) - Andreas would like keys he put in as utf8 to come back as utf8 - */ - STRLEN utf8_len = HEK_LEN(hek); - U8 *as_utf8 = bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len); - - sv = newSVpvn ((char*)as_utf8, utf8_len); - SvUTF8_on (sv); - Safefree (as_utf8); /* bytes_to_utf8() allocates a new string */ - } else if (flags & HVhek_REHASH) { - /* We don't have a pointer to the hv, so we have to replicate the - flag into every HEK. This hv is using custom a hasing - algorithm. Hence we can't return a shared string scalar, as - that would contain the (wrong) hash value, and might get passed - into an hv routine with a regular hash */ - - sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek)); - if (HEK_UTF8(hek)) - SvUTF8_on (sv); - } else { - sv = newSVpvn_share(HEK_KEY(hek), - (HEK_UTF8(hek) ? -HEK_LEN(hek) : HEK_LEN(hek)), - HEK_HASH(hek)); - } - return sv_2mortal(sv); - } - return sv_mortalcopy(HeKEY_sv(entry)); + return sv_2mortal(newSVhek(HeKEY_hek(entry))); } /* @@ -2024,7 +2193,7 @@ Perl_hv_iterval(pTHX_ HV *hv, register HE *entry) { if (SvRMAGICAL(hv)) { if (mg_find((SV*)hv, PERL_MAGIC_tied)) { - SV* sv = sv_newmortal(); + SV* const sv = sv_newmortal(); if (HeKLEN(entry) == HEf_SVKEY) mg_copy((SV*)hv, sv, (char*)HeKEY_sv(entry), HEf_SVKEY); else @@ -2047,14 +2216,18 @@ operation. SV * Perl_hv_iternextsv(pTHX_ HV *hv, char **key, I32 *retlen) { - HE *he; - if ( (he = hv_iternext_flags(hv, 0)) == NULL) + HE * const he = hv_iternext_flags(hv, 0); + + if (!he) return NULL; *key = hv_iterkey(he, retlen); return hv_iterval(hv, he); } /* + +Now a macro in hv.h + =for apidoc hv_magic Adds magic to a hash. See C. @@ -2062,22 +2235,6 @@ Adds magic to a hash. See C. =cut */ -void -Perl_hv_magic(pTHX_ HV *hv, GV *gv, int how) -{ - sv_magic((SV*)hv, (SV*)gv, how, Nullch, 0); -} - -#if 0 /* use the macro from hv.h instead */ - -char* -Perl_sharepvn(pTHX_ const char *sv, I32 len, U32 hash) -{ - return HEK_KEY(share_hek(sv, len, hash)); -} - -#endif - /* possibly free a shared string if no one has access to it * len and hash must both be valid for str. */ @@ -2099,18 +2256,37 @@ Perl_unshare_hek(pTHX_ HEK *hek) are used. If so, len and hash must both be valid for str. */ STATIC void -S_unshare_hek_or_pvn(pTHX_ HEK *hek, const char *str, I32 len, U32 hash) +S_unshare_hek_or_pvn(pTHX_ const HEK *hek, const char *str, I32 len, U32 hash) { + dVAR; register XPVHV* xhv; - register HE *entry; + HE *entry; register HE **oentry; - register I32 i = 1; + HE **first; bool found = 0; bool is_utf8 = FALSE; int k_flags = 0; - const char *save = str; + const char * const save = str; + struct shared_he *he = NULL; if (hek) { + /* Find the shared he which is just before us in memory. */ + he = (struct shared_he *)(((char *)hek) + - STRUCT_OFFSET(struct shared_he, + shared_he_hek)); + + /* Assert that the caller passed us a genuine (or at least consistent) + shared hek */ + assert (he->shared_he_he.hent_hek == hek); + + LOCK_STRTAB_MUTEX; + if (he->shared_he_he.hent_val - 1) { + --he->shared_he_he.hent_val; + UNLOCK_STRTAB_MUTEX; + return; + } + UNLOCK_STRTAB_MUTEX; + hash = HEK_HASH(hek); } else if (len < 0) { STRLEN tmplen = -len; @@ -2132,17 +2308,18 @@ S_unshare_hek_or_pvn(pTHX_ HEK *hek, const char *str, I32 len, U32 hash) xhv = (XPVHV*)SvANY(PL_strtab); /* assert(xhv_array != 0) */ LOCK_STRTAB_MUTEX; - oentry = &(HvARRAY(PL_strtab))[hash & (I32) HvMAX(PL_strtab)]; - if (hek) { - for (entry = *oentry; entry; i=0, oentry = &HeNEXT(entry), entry = *oentry) { - if (HeKEY_hek(entry) != hek) + first = oentry = &(HvARRAY(PL_strtab))[hash & (I32) HvMAX(PL_strtab)]; + if (he) { + const HE *const he_he = &(he->shared_he_he); + for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) { + if (entry != he_he) continue; found = 1; break; } } else { const int flags_masked = k_flags & HVhek_MASK; - for (entry = *oentry; entry; i=0, oentry = &HeNEXT(entry), entry = *oentry) { + for (entry = *oentry; entry; oentry = &HeNEXT(entry), entry = *oentry) { if (HeHASH(entry) != hash) /* strings can't be equal */ continue; if (HeKLEN(entry) != len) @@ -2159,10 +2336,11 @@ S_unshare_hek_or_pvn(pTHX_ HEK *hek, const char *str, I32 len, U32 hash) if (found) { if (--HeVAL(entry) == Nullsv) { *oentry = HeNEXT(entry); - if (i && !*oentry) + if (!*first) { + /* There are now no entries in our slot. */ xhv->xhv_fill--; /* HvFILL(hv)-- */ - Safefree(HeKEY_hek(entry)); - del_HE(entry); + } + Safefree(entry); xhv->xhv_keys--; /* HvKEYS(hv)-- */ } } @@ -2187,7 +2365,7 @@ Perl_share_hek(pTHX_ const char *str, I32 len, register U32 hash) { bool is_utf8 = FALSE; int flags = 0; - const char *save = str; + const char * const save = str; if (len < 0) { STRLEN tmplen = -len; @@ -2212,12 +2390,10 @@ Perl_share_hek(pTHX_ const char *str, I32 len, register U32 hash) STATIC HEK * S_share_hek_flags(pTHX_ const char *str, I32 len, register U32 hash, int flags) { - register XPVHV* xhv; + dVAR; register HE *entry; - register HE **oentry; - register I32 i = 1; - I32 found = 0; const int flags_masked = flags & HVhek_MASK; + const U32 hindex = hash & (I32) HvMAX(PL_strtab); /* what follows is the moral equivalent of: @@ -2227,11 +2403,11 @@ S_share_hek_flags(pTHX_ const char *str, I32 len, register U32 hash, int flags) Can't rehash the shared string table, so not sure if it's worth counting the number of entries in the linked list */ - xhv = (XPVHV*)SvANY(PL_strtab); + register XPVHV * const xhv = (XPVHV*)SvANY(PL_strtab); /* assert(xhv_array != 0) */ LOCK_STRTAB_MUTEX; - oentry = &(HvARRAY(PL_strtab))[hash & (I32) HvMAX(PL_strtab)]; - for (entry = *oentry; entry; i=0, entry = HeNEXT(entry)) { + entry = (HvARRAY(PL_strtab))[hindex]; + for (;entry; entry = HeNEXT(entry)) { if (HeHASH(entry) != hash) /* strings can't be equal */ continue; if (HeKLEN(entry) != len) @@ -2240,17 +2416,46 @@ S_share_hek_flags(pTHX_ const char *str, I32 len, register U32 hash, int flags) continue; if (HeKFLAGS(entry) != flags_masked) continue; - found = 1; break; } - if (!found) { - entry = new_HE(); - HeKEY_hek(entry) = save_hek_flags(str, len, hash, flags_masked); + + if (!entry) { + /* What used to be head of the list. + If this is NULL, then we're the first entry for this slot, which + means we need to increate fill. */ + struct shared_he *new_entry; + HEK *hek; + char *k; + HE **const head = &HvARRAY(PL_strtab)[hindex]; + HE *const next = *head; + + /* We don't actually store a HE from the arena and a regular HEK. + Instead we allocate one chunk of memory big enough for both, + and put the HEK straight after the HE. This way we can find the + HEK directly from the HE. + */ + + Newx(k, STRUCT_OFFSET(struct shared_he, + shared_he_hek.hek_key[0]) + len + 2, char); + new_entry = (struct shared_he *)k; + entry = &(new_entry->shared_he_he); + hek = &(new_entry->shared_he_hek); + + Copy(str, HEK_KEY(hek), len, char); + HEK_KEY(hek)[len] = 0; + HEK_LEN(hek) = len; + HEK_HASH(hek) = hash; + HEK_FLAGS(hek) = (unsigned char)flags_masked; + + /* Still "point" to the HEK, so that other code need not know what + we're up to. */ + HeKEY_hek(entry) = hek; HeVAL(entry) = Nullsv; - HeNEXT(entry) = *oentry; - *oentry = entry; + HeNEXT(entry) = next; + *head = entry; + xhv->xhv_keys++; /* HvKEYS(hv)++ */ - if (i) { /* initial entry? */ + if (!next) { /* initial entry? */ xhv->xhv_fill++; /* HvFILL(hv)++ */ } else if (xhv->xhv_keys > (IV)xhv->xhv_max /* HvKEYS(hv) > HvMAX(hv) */) { hsplit(PL_strtab); @@ -2287,7 +2492,7 @@ I32 Perl_hv_placeholders_get(pTHX_ HV *hv) { dVAR; - MAGIC *mg = mg_find((SV*)hv, PERL_MAGIC_rhash); + MAGIC * const mg = mg_find((SV*)hv, PERL_MAGIC_rhash); return mg ? mg->mg_len : 0; } @@ -2296,7 +2501,7 @@ void Perl_hv_placeholders_set(pTHX_ HV *hv, I32 ph) { dVAR; - MAGIC *mg = mg_find((SV*)hv, PERL_MAGIC_rhash); + MAGIC * const mg = mg_find((SV*)hv, PERL_MAGIC_rhash); if (mg) { mg->mg_len = ph; @@ -2357,10 +2562,10 @@ Perl_hv_assert(pTHX_ HV *hv) (int) real, (int) HvUSEDKEYS(hv)); bad = 1; } - if (HvPLACEHOLDERS(hv) != placeholders) { + if (HvPLACEHOLDERS_get(hv) != placeholders) { PerlIO_printf(Perl_debug_log, "Count %d placeholder(s), but hash reports %d\n", - (int) placeholders, (int) HvPLACEHOLDERS(hv)); + (int) placeholders, (int) HvPLACEHOLDERS_get(hv)); bad = 1; } }