summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--hv.c50
-rw-r--r--hv.h11
2 files changed, 46 insertions, 15 deletions
diff --git a/hv.c b/hv.c
index 7a1d25bf9a..485e2206e4 100644
--- a/hv.c
+++ b/hv.c
@@ -20,6 +20,8 @@
#define PERL_IN_HV_C
#include "perl.h"
+#define HV_MAX_LENGTH_BEFORE_SPLIT 4
+
STATIC HE*
S_new_he(pTHX)
{
@@ -313,7 +315,7 @@ S_hv_fetch_flags(pTHX_ HV *hv, const char *key, I32 klen, I32 lval, int flags)
}
else
HeKFLAGS(entry) = flags;
- if (flags)
+ if (flags & HVhek_ENABLEHVKFLAGS)
HvHASKFLAGS_on(hv);
}
if (flags & HVhek_FREEKEY)
@@ -487,7 +489,7 @@ Perl_hv_fetch_ent(pTHX_ HV *hv, SV *keysv, I32 lval, register U32 hash)
}
else
HeKFLAGS(entry) = flags;
- if (flags)
+ if (flags & HVhek_ENABLEHVKFLAGS)
HvHASKFLAGS_on(hv);
}
if (key != keysave)
@@ -603,7 +605,7 @@ Perl_hv_store_flags(pTHX_ HV *hv, const char *key, I32 klen, SV *val,
register U32 hash, int flags)
{
register XPVHV* xhv;
- register I32 i;
+ register U32 n_links;
register HE *entry;
register HE **oentry;
@@ -650,9 +652,10 @@ Perl_hv_store_flags(pTHX_ HV *hv, const char *key, I32 klen, SV *val,
/* oentry = &(HvARRAY(hv))[hash & (I32) HvMAX(hv)]; */
oentry = &((HE**)xhv->xhv_array)[hash & (I32) xhv->xhv_max];
- i = 1;
- for (entry = *oentry; entry; i=0, entry = HeNEXT(entry)) {
+ n_links = 0;
+
+ for (entry = *oentry; entry; ++n_links, entry = HeNEXT(entry)) {
if (HeHASH(entry) != hash) /* strings can't be equal */
continue;
if (HeKLEN(entry) != (I32)klen)
@@ -719,9 +722,17 @@ Perl_hv_store_flags(pTHX_ HV *hv, const char *key, I32 klen, SV *val,
*oentry = entry;
xhv->xhv_keys++; /* HvKEYS(hv)++ */
- if (i) { /* initial entry? */
+ if (!n_links) { /* initial entry? */
xhv->xhv_fill++; /* HvFILL(hv)++ */
- } else if (xhv->xhv_keys > (IV)xhv->xhv_max /* HvKEYS(hv) > HvMAX(hv) */) {
+ } else if ((n_links > HV_MAX_LENGTH_BEFORE_SPLIT)
+ && (!HvREHASH(hv)
+ || (xhv->xhv_keys > (IV)xhv->xhv_max))) {
+ /* Use 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);
}
@@ -763,7 +774,7 @@ Perl_hv_store_ent(pTHX_ HV *hv, SV *keysv, SV *val, U32 hash)
XPVHV* xhv;
char *key;
STRLEN klen;
- I32 i;
+ U32 n_links;
HE *entry;
HE **oentry;
bool is_utf8;
@@ -830,9 +841,9 @@ Perl_hv_store_ent(pTHX_ HV *hv, SV *keysv, SV *val, U32 hash)
/* oentry = &(HvARRAY(hv))[hash & (I32) HvMAX(hv)]; */
oentry = &((HE**)xhv->xhv_array)[hash & (I32) xhv->xhv_max];
- i = 1;
+ n_links = 0;
entry = *oentry;
- for (; entry; i=0, entry = HeNEXT(entry)) {
+ for (; entry; ++n_links, entry = HeNEXT(entry)) {
if (HeHASH(entry) != hash) /* strings can't be equal */
continue;
if (HeKLEN(entry) != (I32)klen)
@@ -886,10 +897,17 @@ Perl_hv_store_ent(pTHX_ HV *hv, SV *keysv, SV *val, U32 hash)
*oentry = entry;
xhv->xhv_keys++; /* HvKEYS(hv)++ */
- if (i) { /* initial entry? */
+ if (!n_links) { /* initial entry? */
xhv->xhv_fill++; /* HvFILL(hv)++ */
- } else if (xhv->xhv_keys > (IV)xhv->xhv_max /* HvKEYS(hv) > HvMAX(hv) */) {
- hsplit(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);
}
return entry;
@@ -1511,7 +1529,7 @@ S_hsplit(pTHX_ HV *hv)
/* Pick your policy for "hashing isn't working" here: */
- if (longest_chain < 8 || longest_chain * 2 < HvTOTALKEYS(hv)
+ if (longest_chain <= HV_MAX_LENGTH_BEFORE_SPLIT /* split worked? */
|| HvREHASH(hv)) {
return;
}
@@ -1533,7 +1551,6 @@ S_hsplit(pTHX_ HV *hv)
xhv->xhv_fill = 0;
HvSHAREKEYS_off(hv);
HvREHASH_on(hv);
- HvHASKFLAGS_on(hv);
aep = (HE **) xhv->xhv_array;
@@ -2385,6 +2402,9 @@ S_share_hek_flags(pTHX_ const char *str, I32 len, register U32 hash, int flags)
if (!(Svp = hv_fetch(PL_strtab, str, len, FALSE)))
hv_store(PL_strtab, str, len, Nullsv, hash);
+
+ 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);
/* assert(xhv_array != 0) */
diff --git a/hv.h b/hv.h
index 1c7fcda241..5f8344004d 100644
--- a/hv.h
+++ b/hv.h
@@ -212,6 +212,7 @@ C<SV*>.
* is utf8 (including 8 bit keys that were entered as utf8, and need upgrading
* when retrieved during iteration. It may still be set when there are no longer
* any utf8 keys.
+ * See HVhek_ENABLEHVKFLAGS for the trigger.
*/
#define HvHASKFLAGS(hv) (SvFLAGS(hv) & SVphv_HASKFLAGS)
#define HvHASKFLAGS_on(hv) (SvFLAGS(hv) |= SVphv_HASKFLAGS)
@@ -283,6 +284,16 @@ C<SV*>.
* (may change, but Storable is a core module) */
#define HVhek_MASK 0xFF
+/* Which flags enable HvHASKFLAGS? Somewhat a hack on a hack, as
+ HVhek_REHASH is only needed because the rehash flag has to be duplicated
+ into all keys as hv_iternext has no access to the hash flags. At this
+ point Storable's tests get upset, because sometimes hashes are "keyed"
+ and sometimes not, depending on the order of data insertion, and whether
+ it triggered rehashing. So currently HVhek_REHAS is exempt.
+*/
+
+#define HVhek_ENABLEHVKFLAGS (HVhek_MASK - HVhek_REHASH)
+
#define HEK_UTF8(hek) (HEK_FLAGS(hek) & HVhek_UTF8)
#define HEK_UTF8_on(hek) (HEK_FLAGS(hek) |= HVhek_UTF8)
#define HEK_UTF8_off(hek) (HEK_FLAGS(hek) &= ~HVhek_UTF8)