diff options
Diffstat (limited to 'hv.c')
| -rw-r--r-- | hv.c | 35 |
1 files changed, 35 insertions, 0 deletions
@@ -1560,6 +1560,34 @@ S_hv_delete_common(pTHX_ HV *hv, SV *keysv, const char *key, STRLEN klen, return NULL; } +/* HVs are used for (at least) three things + 1) objects + 2) symbol tables + 3) associative arrays + + shared hash keys benefit the first two greatly, because keys are likely + to be re-used between objects, or for constants in the optree + + However, for large associative arrays (lookup tables, "seen" hashes) keys are + unlikely to be re-used. Hence having those keys in the shared string table as + well as the hash is a memory hit, if they are never actually shared with a + second hash. Hence we turn off shared hash keys if a (regular) hash gets + large. + + This is a heuristic. There might be a better answer than 42, but for now + we'll use it. +*/ +static bool +S_large_hash_heuristic(pTHX_ HV *hv, STRLEN size) { + if (size > 42 + && !SvOBJECT(hv) + && !(SvOOK(hv) && HvENAME_get(hv))) { + /* This hash appears to be growing quite large. + We gamble that it is not sharing keys with other hashes. */ + return TRUE; + } + return FALSE; +} STATIC void S_hsplit(pTHX_ HV *hv, STRLEN const oldsize, STRLEN newsize) @@ -1595,6 +1623,11 @@ S_hsplit(pTHX_ HV *hv, STRLEN const oldsize, STRLEN newsize) if (!HvTOTALKEYS(hv)) /* skip rest if no entries */ return; + /* don't share keys in large simple hashes */ + if (S_large_hash_heuristic(aTHX_ hv, HvTOTALKEYS(hv))) + HvSHAREKEYS_off(hv); + + newsize--; aep = (HE**)a; do { @@ -1687,6 +1720,8 @@ Perl_hv_ksplit(pTHX_ HV *hv, IV newmax) } #endif } else { + if (S_large_hash_heuristic(aTHX_ hv, newmax)) + HvSHAREKEYS_off(hv); Newxz(a, PERL_HV_ARRAY_ALLOC_BYTES(newsize), char); xhv->xhv_max = newsize - 1; HvARRAY(hv) = (HE **) a; |