From 144b0094c31b0a3604e5bd2686a7df15c0fd324d Mon Sep 17 00:00:00 2001
From: Pieter Noordhuis <pcnoordhuis@gmail.com>
Date: Thu, 3 Jun 2010 16:06:18 +0200
Subject: [PATCH 01/10] initial implementation for the intset

---
 intset.c | 379 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
 intset.h |  16 +++
 2 files changed, 395 insertions(+)
 create mode 100644 intset.c
 create mode 100644 intset.h

diff --git a/intset.c b/intset.c
new file mode 100644
index 00000000..8ffaef25
--- /dev/null
+++ b/intset.c
@@ -0,0 +1,379 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdint.h>
+#include "intset.h"
+#include "zmalloc.h"
+
+/* Note that these encodings are ordered, so:
+ * INTSET_ENC_INT16 < INTSET_ENC_INT32 < INTSET_ENC_INT64. */
+#define INTSET_ENC_INT16 (sizeof(int16_t))
+#define INTSET_ENC_INT32 (sizeof(int32_t))
+#define INTSET_ENC_INT64 (sizeof(int64_t))
+
+/* Accessors for each type of encoding */
+#define INTSET_VALUE_ENCODING(__val) (((__val) < INT32_MIN || (__val) > INT32_MAX) ? \
+    INTSET_ENC_INT64 : (((__val) < INT16_MIN || (__val) > INT16_MAX) ? \
+    INTSET_ENC_INT32 : INTSET_ENC_INT16))
+#define INTSET_GET_ENCODED(__is,__pos,__enc) ((__enc == INTSET_ENC_INT64) ? \
+    ((int64_t*)(__is)->contents)[__pos] : ((__enc == INTSET_ENC_INT32) ? \
+    ((int32_t*)(__is)->contents)[__pos] : ((int16_t*)(__is)->contents)[__pos]))
+#define INTSET_GET(__is,__pos) (INTSET_GET_ENCODED(__is,__pos,(__is)->encoding))
+#define INTSET_SET(__is,__pos,__val) { \
+    if ((__is)->encoding == INTSET_ENC_INT64) \
+        ((int64_t*)(__is)->contents)[__pos] = (__val); \
+    else if ((__is)->encoding == INTSET_ENC_INT32) \
+        ((int32_t*)(__is)->contents)[__pos] = (__val); \
+    else \
+        ((int16_t*)(__is)->contents)[__pos] = (__val); }
+
+/* Create an empty intset. */
+intset *intsetNew(void) {
+    intset *is = zmalloc(sizeof(intset));
+    is->encoding = INTSET_ENC_INT16;
+    is->length = 0;
+    return is;
+}
+
+/* Resize the intset */
+static intset *intsetResize(intset *is, uint32_t len) {
+    uint32_t size = len*is->encoding;
+    is = zrealloc(is,sizeof(intset)+size);
+    return is;
+}
+
+static intset *intsetUpgrade(intset *is, uint8_t newenc, uint8_t extra, uint8_t offset) {
+    uint8_t curenc = is->encoding;
+    int length = is->length;
+
+    /* First set new encoding and resize */
+    is->encoding = newenc;
+    is = intsetResize(is,is->length+extra);
+
+    /* Upgrade back-to-front so we don't overwrite values */
+    while(length--)
+        INTSET_SET(is,length+offset,INTSET_GET_ENCODED(is,length,curenc));
+    return is;
+}
+
+/* Search for the position of "value". Return 1 when the value was found and
+ * sets "pos" to the position of the value within the intset. Return 0 when
+ * the value is not present in the intset and sets "pos" to the position
+ * where "value" can be inserted. */
+static uint8_t intsetSearch(intset *is, int64_t value, uint32_t *pos) {
+    int min = 0, max = is->length-1, mid;
+    int64_t cur;
+
+    /* The value can never be found when the set is empty */
+    if (is->length == 0) {
+        if (pos) *pos = 0;
+        return 0;
+    }
+
+    while(max >= min) {
+        mid = (min+max)/2;
+        cur = INTSET_GET(is,mid);
+        if (value > cur) {
+            min = mid+1;
+        } else if (value < cur) {
+            max = mid-1;
+        } else {
+            break;
+        }
+    }
+
+    if (value == cur) {
+        if (pos) *pos = mid;
+        return 1;
+    } else {
+        if (pos) *pos = min;
+        return 0;
+    }
+}
+
+static void intsetMoveTail(intset *is, uint32_t from, uint32_t to) {
+    void *src, *dst;
+    uint32_t bytes = is->length-from;
+    if (is->encoding == INTSET_ENC_INT64) {
+        src = (int64_t*)is->contents+from;
+        dst = (int64_t*)is->contents+to;
+        bytes *= sizeof(int64_t);
+    } else if (is->encoding == INTSET_ENC_INT32) {
+        src = (int32_t*)is->contents+from;
+        dst = (int32_t*)is->contents+to;
+        bytes *= sizeof(int32_t);
+    } else {
+        src = (int16_t*)is->contents+from;
+        dst = (int16_t*)is->contents+to;
+        bytes *= sizeof(int16_t);
+    }
+    memmove(dst,src,bytes);
+}
+
+/* Insert an integer in the intset */
+intset *intsetAdd(intset *is, int64_t value, uint8_t *success) {
+    uint8_t valenc = INTSET_VALUE_ENCODING(value);
+    uint32_t pos, offset;
+    if (success) *success = 1;
+
+    /* Upgrade encoding if necessary. If we need to upgrade, we know that
+     * this value should be either appended (if > 0) or prepended (if < 0),
+     * because it lies outside the range of existing values. */
+    if (valenc > is->encoding) {
+        offset = value < 0 ? 1 : 0;
+        is = intsetUpgrade(is,valenc,1,offset);
+        pos = (value < 0) ? 0 : is->length;
+    } else {
+        if (is->length == 0) {
+            pos = 0;
+        } else {
+            /* Check for the case where we know the insert position */
+            if (value > INTSET_GET(is,is->length-1)) {
+                pos = is->length;
+            } else if (value < INTSET_GET(is,0)) {
+                pos = 0;
+            } else {
+                /* Abort if the value is already present in the set */
+                if (intsetSearch(is,value,&pos)) {
+                    if (success) *success = 0;
+                    return is;
+                }
+            }
+        }
+
+        is = intsetResize(is,is->length+1);
+        if (pos < is->length) intsetMoveTail(is,pos,pos+1);
+    }
+
+    INTSET_SET(is,pos,value);
+    is->length++;
+    return is;
+}
+
+/* Delete integer from intset */
+intset *intsetDelete(intset *is, int64_t value, uint8_t *success) {
+    uint8_t valenc = INTSET_VALUE_ENCODING(value);
+    uint32_t pos;
+    if (success) *success = 0;
+
+    if (valenc <= is->encoding && intsetSearch(is,value,&pos)) {
+        /* We know we can delete */
+        if (success) *success = 1;
+
+        /* Overwrite value with tail and update length */
+        if (pos < (is->length-1)) intsetMoveTail(is,pos+1,pos);
+        is = intsetResize(is,is->length-1);
+        is->length--;
+    }
+    return is;
+}
+
+/* Determine whether a value belongs to this set */
+uint8_t intsetFind(intset *is, int64_t value) {
+    uint8_t valenc = INTSET_VALUE_ENCODING(value);
+    return valenc <= is->encoding && intsetSearch(is,value,NULL);
+}
+
+/* Return random member */
+int64_t intsetRandom(intset *is) {
+    return INTSET_GET(is,rand()%is->length);
+}
+
+#ifdef INTSET_TEST_MAIN
+#include <sys/time.h>
+
+void intsetRepr(intset *is) {
+    int i;
+    for (i = 0; i < is->length; i++) {
+        printf("%lld\n", (uint64_t)INTSET_GET(is,i));
+    }
+    printf("\n");
+}
+
+void error(char *err) {
+    printf("%s\n", err);
+    exit(1);
+}
+
+void ok(void) {
+    printf("OK\n");
+}
+
+long long usec(void) {
+    struct timeval tv;
+    gettimeofday(&tv,NULL);
+    return (((long long)tv.tv_sec)*1000000)+tv.tv_usec;
+}
+
+#define assert(_e) ((_e)?(void)0:(_assert(#_e,__FILE__,__LINE__),exit(1)))
+void _assert(char *estr, char *file, int line) {
+    printf("\n\n=== ASSERTION FAILED ===\n");
+    printf("==> %s:%d '%s' is not true\n",file,line,estr);
+}
+
+intset *createSet(int bits, int size) {
+    uint64_t mask = (1<<bits)-1;
+    uint64_t i, value;
+    intset *is = intsetNew();
+
+    for (i = 0; i < size; i++) {
+        if (bits > 32) {
+            value = (rand()*rand()) & mask;
+        } else {
+            value = rand() & mask;
+        }
+        is = intsetAdd(is,value,NULL);
+    }
+    return is;
+}
+
+void checkConsistency(intset *is) {
+    int i;
+
+    for (i = 0; i < (is->length-1); i++) {
+        if (is->encoding == INTSET_ENC_INT16) {
+            int16_t *i16 = (int16_t*)is->contents;
+            assert(i16[i] < i16[i+1]);
+        } else if (is->encoding == INTSET_ENC_INT32) {
+            int32_t *i32 = (int32_t*)is->contents;
+            assert(i32[i] < i32[i+1]);
+        } else {
+            int64_t *i64 = (int64_t*)is->contents;
+            assert(i64[i] < i64[i+1]);
+        }
+    }
+}
+
+int main(int argc, char **argv) {
+    uint8_t success;
+    int i;
+    intset *is;
+    sranddev();
+
+    printf("Value encodings: "); {
+        assert(INTSET_VALUE_ENCODING(-32768) == INTSET_ENC_INT16);
+        assert(INTSET_VALUE_ENCODING(+32767) == INTSET_ENC_INT16);
+        assert(INTSET_VALUE_ENCODING(-32769) == INTSET_ENC_INT32);
+        assert(INTSET_VALUE_ENCODING(+32768) == INTSET_ENC_INT32);
+        assert(INTSET_VALUE_ENCODING(-2147483648) == INTSET_ENC_INT32);
+        assert(INTSET_VALUE_ENCODING(+2147483647) == INTSET_ENC_INT32);
+        assert(INTSET_VALUE_ENCODING(-2147483649) == INTSET_ENC_INT64);
+        assert(INTSET_VALUE_ENCODING(+2147483648) == INTSET_ENC_INT64);
+        assert(INTSET_VALUE_ENCODING(-9223372036854775808ull) == INTSET_ENC_INT64);
+        assert(INTSET_VALUE_ENCODING(+9223372036854775807ull) == INTSET_ENC_INT64);
+        ok();
+    }
+
+    printf("Basic adding: "); {
+        is = intsetNew();
+        is = intsetAdd(is,5,&success); assert(success);
+        is = intsetAdd(is,6,&success); assert(success);
+        is = intsetAdd(is,4,&success); assert(success);
+        is = intsetAdd(is,4,&success); assert(!success);
+        ok();
+    }
+
+    printf("Large number of random adds: "); {
+        int inserts = 0;
+        is = intsetNew();
+        for (i = 0; i < 1024; i++) {
+            is = intsetAdd(is,rand()%0x800,&success);
+            if (success) inserts++;
+        }
+        assert(is->length == inserts);
+        checkConsistency(is);
+        ok();
+    }
+
+    printf("Upgrade from int16 to int32: "); {
+        is = intsetNew();
+        is = intsetAdd(is,32,NULL);
+        assert(is->encoding == INTSET_ENC_INT16);
+        is = intsetAdd(is,65535,NULL);
+        assert(is->encoding == INTSET_ENC_INT32);
+        assert(intsetFind(is,32));
+        assert(intsetFind(is,65535));
+        checkConsistency(is);
+
+        is = intsetNew();
+        is = intsetAdd(is,32,NULL);
+        assert(is->encoding == INTSET_ENC_INT16);
+        is = intsetAdd(is,-65535,NULL);
+        assert(is->encoding == INTSET_ENC_INT32);
+        assert(intsetFind(is,32));
+        assert(intsetFind(is,-65535));
+        checkConsistency(is);
+        ok();
+    }
+
+    printf("Upgrade from int16 to int64: "); {
+        is = intsetNew();
+        is = intsetAdd(is,32,NULL);
+        assert(is->encoding == INTSET_ENC_INT16);
+        is = intsetAdd(is,4294967295,NULL);
+        assert(is->encoding == INTSET_ENC_INT64);
+        assert(intsetFind(is,32));
+        assert(intsetFind(is,4294967295));
+        checkConsistency(is);
+
+        is = intsetNew();
+        is = intsetAdd(is,32,NULL);
+        assert(is->encoding == INTSET_ENC_INT16);
+        is = intsetAdd(is,-4294967295,NULL);
+        assert(is->encoding == INTSET_ENC_INT64);
+        assert(intsetFind(is,32));
+        assert(intsetFind(is,-4294967295));
+        checkConsistency(is);
+        ok();
+    }
+
+    printf("Upgrade from int32 to int64: "); {
+        is = intsetNew();
+        is = intsetAdd(is,65535,NULL);
+        assert(is->encoding == INTSET_ENC_INT32);
+        is = intsetAdd(is,4294967295,NULL);
+        assert(is->encoding == INTSET_ENC_INT64);
+        assert(intsetFind(is,65535));
+        assert(intsetFind(is,4294967295));
+        checkConsistency(is);
+
+        is = intsetNew();
+        is = intsetAdd(is,65535,NULL);
+        assert(is->encoding == INTSET_ENC_INT32);
+        is = intsetAdd(is,-4294967295,NULL);
+        assert(is->encoding == INTSET_ENC_INT64);
+        assert(intsetFind(is,65535));
+        assert(intsetFind(is,-4294967295));
+        checkConsistency(is);
+        ok();
+    }
+
+    printf("Stress lookups: "); {
+        long num = 100000, size = 10000;
+        int i, bits = 20;
+        long long start;
+        is = createSet(bits,size);
+        checkConsistency(is);
+
+        start = usec();
+        for (i = 0; i < num; i++) intsetSearch(is,rand() % ((1<<bits)-1),NULL);
+        printf("%ld lookups, %ld element set, %lldusec\n",num,size,usec()-start);
+    }
+
+    printf("Stress add+delete: "); {
+        int i, v1, v2;
+        is = intsetNew();
+        for (i = 0; i < 0xffff; i++) {
+            v1 = rand() % 0xfff;
+            is = intsetAdd(is,v1,NULL);
+            assert(intsetFind(is,v1));
+
+            v2 = rand() % 0xfff;
+            is = intsetDelete(is,v2,NULL);
+            assert(!intsetFind(is,v2));
+        }
+        checkConsistency(is);
+        ok();
+    }
+}
+#endif
diff --git a/intset.h b/intset.h
new file mode 100644
index 00000000..9d91710a
--- /dev/null
+++ b/intset.h
@@ -0,0 +1,16 @@
+#ifndef __INTSET_H
+#define __INTSET_H
+
+typedef struct intset {
+    uint32_t encoding;
+    uint32_t length;
+    int8_t contents[];
+} intset;
+
+intset *intsetNew(void);
+intset *intsetAdd(intset *is, int64_t value, uint8_t *success);
+intset *intsetDelete(intset *is, int64_t value, uint8_t *success);
+uint8_t intsetFind(intset *is, int64_t value);
+int64_t intsetRandom(intset *is);
+
+#endif // __INTSET_H

From 3ab98cef4e9aaefbb1febbe51309f657c701b1df Mon Sep 17 00:00:00 2001
From: Pieter Noordhuis <pcnoordhuis@gmail.com>
Date: Thu, 3 Jun 2010 22:21:39 +0200
Subject: [PATCH 02/10] move logic for obvious corner cases to intsetSearch

---
 intset.c | 31 ++++++++++++++++---------------
 1 file changed, 16 insertions(+), 15 deletions(-)

diff --git a/intset.c b/intset.c
index 8ffaef25..f1bbcde3 100644
--- a/intset.c
+++ b/intset.c
@@ -68,6 +68,16 @@ static uint8_t intsetSearch(intset *is, int64_t value, uint32_t *pos) {
     if (is->length == 0) {
         if (pos) *pos = 0;
         return 0;
+    } else {
+        /* Check for the case where we know we cannot find the value,
+         * but do know the insert position. */
+        if (value > INTSET_GET(is,is->length-1)) {
+            if (pos) *pos = is->length;
+            return 0;
+        } else if (value < INTSET_GET(is,0)) {
+            if (pos) *pos = 0;
+            return 0;
+        }
     }
 
     while(max >= min) {
@@ -124,21 +134,12 @@ intset *intsetAdd(intset *is, int64_t value, uint8_t *success) {
         is = intsetUpgrade(is,valenc,1,offset);
         pos = (value < 0) ? 0 : is->length;
     } else {
-        if (is->length == 0) {
-            pos = 0;
-        } else {
-            /* Check for the case where we know the insert position */
-            if (value > INTSET_GET(is,is->length-1)) {
-                pos = is->length;
-            } else if (value < INTSET_GET(is,0)) {
-                pos = 0;
-            } else {
-                /* Abort if the value is already present in the set */
-                if (intsetSearch(is,value,&pos)) {
-                    if (success) *success = 0;
-                    return is;
-                }
-            }
+        /* Abort if the value is already present in the set.
+         * This call will populate "pos" with the right position to insert
+         * the value when it cannot be found. */
+        if (intsetSearch(is,value,&pos)) {
+            if (success) *success = 0;
+            return is;
         }
 
         is = intsetResize(is,is->length+1);

From 35cabcb50557eae1433a68c7730ad8aff478b68a Mon Sep 17 00:00:00 2001
From: Pieter Noordhuis <pcnoordhuis@gmail.com>
Date: Fri, 11 Jun 2010 18:35:57 +0200
Subject: [PATCH 03/10] wrapper functions for the set type to support multiple
 encodings

---
 redis.c | 282 ++++++++++++++++++++++++++++++++++----------------------
 1 file changed, 174 insertions(+), 108 deletions(-)

diff --git a/redis.c b/redis.c
index 3e7a17d8..cb7bca8e 100644
--- a/redis.c
+++ b/redis.c
@@ -3064,7 +3064,9 @@ static robj *createZiplistObject(void) {
 
 static robj *createSetObject(void) {
     dict *d = dictCreate(&setDictType,NULL);
-    return createObject(REDIS_SET,d);
+    robj *o = createObject(REDIS_SET,d);
+    o->encoding = REDIS_ENCODING_HT;
+    return o;
 }
 
 static robj *createHashObject(void) {
@@ -5457,6 +5459,99 @@ static void rpoplpushcommand(redisClient *c) {
 
 /* ==================================== Sets ================================ */
 
+static int setTypeAdd(robj *subject, robj *value) {
+    if (subject->encoding == REDIS_ENCODING_HT) {
+        if (dictAdd(subject->ptr,value,NULL) == DICT_OK) {
+            incrRefCount(value);
+            return 1;
+        }
+    } else {
+        redisPanic("Unknown set encoding");
+    }
+    return 0;
+}
+
+static int setTypeRemove(robj *subject, robj *value) {
+    if (subject->encoding == REDIS_ENCODING_HT) {
+        if (dictDelete(subject->ptr,value) == DICT_OK) {
+            if (htNeedsResize(subject->ptr)) dictResize(subject->ptr);
+            return 1;
+        }
+    } else {
+        redisPanic("Unknown set encoding");
+    }
+    return 0;
+}
+
+static int setTypeIsMember(robj *subject, robj *value) {
+    if (subject->encoding == REDIS_ENCODING_HT) {
+        return dictFind((dict*)subject->ptr,value) != NULL;
+    } else {
+        redisPanic("Unknown set encoding");
+    }
+}
+
+/* Structure to hold set iteration abstraction. */
+typedef struct {
+    int encoding;
+    dictIterator *di;
+} setIterator;
+
+static setIterator *setTypeInitIterator(robj *subject) {
+    setIterator *si = zmalloc(sizeof(setIterator));
+    si->encoding = subject->encoding;
+    if (si->encoding == REDIS_ENCODING_HT) {
+        si->di = dictGetIterator(subject->ptr);
+    } else {
+        redisPanic("Unknown set encoding");
+    }
+    return si;
+}
+
+static void setTypeReleaseIterator(setIterator *si) {
+    if (si->encoding == REDIS_ENCODING_HT)
+        dictReleaseIterator(si->di);
+    zfree(si);
+}
+
+/* Move to the next entry in the set. Returns the object at the current
+ * position, or NULL when the end is reached. This object will have its
+ * refcount incremented, so the caller needs to take care of this. */
+static robj *setTypeNext(setIterator *si) {
+    robj *ret = NULL;
+    if (si->encoding == REDIS_ENCODING_HT) {
+        dictEntry *de = dictNext(si->di);
+        if (de != NULL) {
+            ret = dictGetEntryKey(de);
+            incrRefCount(ret);
+        }
+    }
+    return ret;
+}
+
+
+/* Return random element from set. The returned object will always have
+ * an incremented refcount. */
+robj *setTypeRandomElement(robj *subject) {
+    robj *ret = NULL;
+    if (subject->encoding == REDIS_ENCODING_HT) {
+        dictEntry *de = dictGetRandomKey(subject->ptr);
+        ret = dictGetEntryKey(de);
+        incrRefCount(ret);
+    } else {
+        redisPanic("Unknown set encoding");
+    }
+    return ret;
+}
+
+static unsigned long setTypeSize(robj *subject) {
+    if (subject->encoding == REDIS_ENCODING_HT) {
+        return dictSize((dict*)subject->ptr);
+    } else {
+        redisPanic("Unknown set encoding");
+    }
+}
+
 static void saddCommand(redisClient *c) {
     robj *set;
 
@@ -5470,8 +5565,7 @@ static void saddCommand(redisClient *c) {
             return;
         }
     }
-    if (dictAdd(set->ptr,c->argv[2],NULL) == DICT_OK) {
-        incrRefCount(c->argv[2]);
+    if (setTypeAdd(set,c->argv[2])) {
         server.dirty++;
         addReply(c,shared.cone);
     } else {
@@ -5485,10 +5579,9 @@ static void sremCommand(redisClient *c) {
     if ((set = lookupKeyWriteOrReply(c,c->argv[1],shared.czero)) == NULL ||
         checkType(c,set,REDIS_SET)) return;
 
-    if (dictDelete(set->ptr,c->argv[2]) == DICT_OK) {
+    if (setTypeRemove(set,c->argv[2])) {
+        if (setTypeSize(set) == 0) dbDelete(c->db,c->argv[1]);
         server.dirty++;
-        if (htNeedsResize(set->ptr)) dictResize(set->ptr);
-        if (dictSize((dict*)set->ptr) == 0) dbDelete(c->db,c->argv[1]);
         addReply(c,shared.cone);
     } else {
         addReply(c,shared.czero);
@@ -5513,12 +5606,12 @@ static void smoveCommand(redisClient *c) {
         return;
     }
     /* Remove the element from the source set */
-    if (dictDelete(srcset->ptr,c->argv[3]) == DICT_ERR) {
+    if (!setTypeRemove(srcset,c->argv[3])) {
         /* Key not found in the src set! return zero */
         addReply(c,shared.czero);
         return;
     }
-    if (dictSize((dict*)srcset->ptr) == 0 && srcset != dstset)
+    if (setTypeSize(srcset) == 0 && srcset != dstset)
         dbDelete(c->db,c->argv[1]);
     server.dirty++;
     /* Add the element to the destination set */
@@ -5526,8 +5619,7 @@ static void smoveCommand(redisClient *c) {
         dstset = createSetObject();
         dbAdd(c->db,c->argv[2],dstset);
     }
-    if (dictAdd(dstset->ptr,c->argv[3],NULL) == DICT_OK)
-        incrRefCount(c->argv[3]);
+    setTypeAdd(dstset,c->argv[3]);
     addReply(c,shared.cone);
 }
 
@@ -5537,7 +5629,7 @@ static void sismemberCommand(redisClient *c) {
     if ((set = lookupKeyReadOrReply(c,c->argv[1],shared.czero)) == NULL ||
         checkType(c,set,REDIS_SET)) return;
 
-    if (dictFind(set->ptr,c->argv[2]))
+    if (setTypeIsMember(set,c->argv[2]))
         addReply(c,shared.cone);
     else
         addReply(c,shared.czero);
@@ -5545,74 +5637,62 @@ static void sismemberCommand(redisClient *c) {
 
 static void scardCommand(redisClient *c) {
     robj *o;
-    dict *s;
 
     if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.czero)) == NULL ||
         checkType(c,o,REDIS_SET)) return;
 
-    s = o->ptr;
-    addReplyUlong(c,dictSize(s));
+    addReplyUlong(c,setTypeSize(o));
 }
 
 static void spopCommand(redisClient *c) {
-    robj *set;
-    dictEntry *de;
+    robj *set, *ele;
 
     if ((set = lookupKeyWriteOrReply(c,c->argv[1],shared.nullbulk)) == NULL ||
         checkType(c,set,REDIS_SET)) return;
 
-    de = dictGetRandomKey(set->ptr);
-    if (de == NULL) {
+    ele = setTypeRandomElement(set);
+    if (ele == NULL) {
         addReply(c,shared.nullbulk);
     } else {
-        robj *ele = dictGetEntryKey(de);
-
+        setTypeRemove(set,ele);
         addReplyBulk(c,ele);
-        dictDelete(set->ptr,ele);
-        if (htNeedsResize(set->ptr)) dictResize(set->ptr);
-        if (dictSize((dict*)set->ptr) == 0) dbDelete(c->db,c->argv[1]);
+        decrRefCount(ele);
+        if (setTypeSize(set) == 0) dbDelete(c->db,c->argv[1]);
         server.dirty++;
     }
 }
 
 static void srandmemberCommand(redisClient *c) {
-    robj *set;
-    dictEntry *de;
+    robj *set, *ele;
 
     if ((set = lookupKeyReadOrReply(c,c->argv[1],shared.nullbulk)) == NULL ||
         checkType(c,set,REDIS_SET)) return;
 
-    de = dictGetRandomKey(set->ptr);
-    if (de == NULL) {
+    ele = setTypeRandomElement(set);
+    if (ele == NULL) {
         addReply(c,shared.nullbulk);
     } else {
-        robj *ele = dictGetEntryKey(de);
-
         addReplyBulk(c,ele);
+        decrRefCount(ele);
     }
 }
 
 static int qsortCompareSetsByCardinality(const void *s1, const void *s2) {
-    dict **d1 = (void*) s1, **d2 = (void*) s2;
-
-    return dictSize(*d1)-dictSize(*d2);
+    return setTypeSize(*(robj**)s1)-setTypeSize(*(robj**)s2);
 }
 
-static void sinterGenericCommand(redisClient *c, robj **setskeys, unsigned long setsnum, robj *dstkey) {
-    dict **dv = zmalloc(sizeof(dict*)*setsnum);
-    dictIterator *di;
-    dictEntry *de;
-    robj *lenobj = NULL, *dstset = NULL;
+static void sinterGenericCommand(redisClient *c, robj **setkeys, unsigned long setnum, robj *dstkey) {
+    robj **sets = zmalloc(sizeof(robj*)*setnum);
+    setIterator *si;
+    robj *ele, *lenobj = NULL, *dstset = NULL;
     unsigned long j, cardinality = 0;
 
-    for (j = 0; j < setsnum; j++) {
-        robj *setobj;
-
-        setobj = dstkey ?
-                    lookupKeyWrite(c->db,setskeys[j]) :
-                    lookupKeyRead(c->db,setskeys[j]);
+    for (j = 0; j < setnum; j++) {
+        robj *setobj = dstkey ?
+            lookupKeyWrite(c->db,setkeys[j]) :
+            lookupKeyRead(c->db,setkeys[j]);
         if (!setobj) {
-            zfree(dv);
+            zfree(sets);
             if (dstkey) {
                 if (dbDelete(c->db,dstkey))
                     server.dirty++;
@@ -5622,16 +5702,15 @@ static void sinterGenericCommand(redisClient *c, robj **setskeys, unsigned long
             }
             return;
         }
-        if (setobj->type != REDIS_SET) {
-            zfree(dv);
-            addReply(c,shared.wrongtypeerr);
+        if (checkType(c,setobj,REDIS_SET)) {
+            zfree(sets);
             return;
         }
-        dv[j] = setobj->ptr;
+        sets[j] = setobj;
     }
     /* Sort sets from the smallest to largest, this will improve our
      * algorithm's performace */
-    qsort(dv,setsnum,sizeof(dict*),qsortCompareSetsByCardinality);
+    qsort(sets,setnum,sizeof(robj*),qsortCompareSetsByCardinality);
 
     /* The first thing we should output is the total number of elements...
      * since this is a multi-bulk write, but at this stage we don't know
@@ -5651,33 +5730,31 @@ static void sinterGenericCommand(redisClient *c, robj **setskeys, unsigned long
     /* Iterate all the elements of the first (smallest) set, and test
      * the element against all the other sets, if at least one set does
      * not include the element it is discarded */
-    di = dictGetIterator(dv[0]);
+    si = setTypeInitIterator(sets[0]);
+    while((ele = setTypeNext(si)) != NULL) {
+        for (j = 1; j < setnum; j++)
+            if (!setTypeIsMember(sets[j],ele)) break;
 
-    while((de = dictNext(di)) != NULL) {
-        robj *ele;
-
-        for (j = 1; j < setsnum; j++)
-            if (dictFind(dv[j],dictGetEntryKey(de)) == NULL) break;
-        if (j != setsnum)
-            continue; /* at least one set does not contain the member */
-        ele = dictGetEntryKey(de);
-        if (!dstkey) {
-            addReplyBulk(c,ele);
-            cardinality++;
-        } else {
-            dictAdd(dstset->ptr,ele,NULL);
-            incrRefCount(ele);
+        /* Only take action when all sets contain the member */
+        if (j == setnum) {
+            if (!dstkey) {
+                addReplyBulk(c,ele);
+                cardinality++;
+            } else {
+                setTypeAdd(dstset,ele);
+            }
         }
+        decrRefCount(ele);
     }
-    dictReleaseIterator(di);
+    setTypeReleaseIterator(si);
 
     if (dstkey) {
         /* Store the resulting set into the target, if the intersection
          * is not an empty set. */
         dbDelete(c->db,dstkey);
-        if (dictSize((dict*)dstset->ptr) > 0) {
+        if (setTypeSize(dstset) > 0) {
             dbAdd(c->db,dstkey,dstset);
-            addReplyLongLong(c,dictSize((dict*)dstset->ptr));
+            addReplyLongLong(c,setTypeSize(dstset));
         } else {
             decrRefCount(dstset);
             addReply(c,shared.czero);
@@ -5686,7 +5763,7 @@ static void sinterGenericCommand(redisClient *c, robj **setskeys, unsigned long
     } else {
         lenobj->ptr = sdscatprintf(sdsempty(),"*%lu\r\n",cardinality);
     }
-    zfree(dv);
+    zfree(sets);
 }
 
 static void sinterCommand(redisClient *c) {
@@ -5701,29 +5778,25 @@ static void sinterstoreCommand(redisClient *c) {
 #define REDIS_OP_DIFF 1
 #define REDIS_OP_INTER 2
 
-static void sunionDiffGenericCommand(redisClient *c, robj **setskeys, int setsnum, robj *dstkey, int op) {
-    dict **dv = zmalloc(sizeof(dict*)*setsnum);
-    dictIterator *di;
-    dictEntry *de;
-    robj *dstset = NULL;
+static void sunionDiffGenericCommand(redisClient *c, robj **setkeys, int setnum, robj *dstkey, int op) {
+    robj **sets = zmalloc(sizeof(robj*)*setnum);
+    setIterator *si;
+    robj *ele, *dstset = NULL;
     int j, cardinality = 0;
 
-    for (j = 0; j < setsnum; j++) {
-        robj *setobj;
-
-        setobj = dstkey ?
-                    lookupKeyWrite(c->db,setskeys[j]) :
-                    lookupKeyRead(c->db,setskeys[j]);
+    for (j = 0; j < setnum; j++) {
+        robj *setobj = dstkey ?
+            lookupKeyWrite(c->db,setkeys[j]) :
+            lookupKeyRead(c->db,setkeys[j]);
         if (!setobj) {
-            dv[j] = NULL;
+            sets[j] = NULL;
             continue;
         }
-        if (setobj->type != REDIS_SET) {
-            zfree(dv);
-            addReply(c,shared.wrongtypeerr);
+        if (checkType(c,setobj,REDIS_SET)) {
+            zfree(sets);
             return;
         }
-        dv[j] = setobj->ptr;
+        sets[j] = setobj;
     }
 
     /* We need a temp set object to store our union. If the dstkey
@@ -5733,60 +5806,53 @@ static void sunionDiffGenericCommand(redisClient *c, robj **setskeys, int setsnu
 
     /* Iterate all the elements of all the sets, add every element a single
      * time to the result set */
-    for (j = 0; j < setsnum; j++) {
-        if (op == REDIS_OP_DIFF && j == 0 && !dv[j]) break; /* result set is empty */
-        if (!dv[j]) continue; /* non existing keys are like empty sets */
+    for (j = 0; j < setnum; j++) {
+        if (op == REDIS_OP_DIFF && j == 0 && !sets[j]) break; /* result set is empty */
+        if (!sets[j]) continue; /* non existing keys are like empty sets */
 
-        di = dictGetIterator(dv[j]);
-
-        while((de = dictNext(di)) != NULL) {
-            robj *ele;
-
-            /* dictAdd will not add the same element multiple times */
-            ele = dictGetEntryKey(de);
+        si = setTypeInitIterator(sets[j]);
+        while((ele = setTypeNext(si)) != NULL) {
             if (op == REDIS_OP_UNION || j == 0) {
-                if (dictAdd(dstset->ptr,ele,NULL) == DICT_OK) {
-                    incrRefCount(ele);
+                if (setTypeAdd(dstset,ele)) {
                     cardinality++;
                 }
             } else if (op == REDIS_OP_DIFF) {
-                if (dictDelete(dstset->ptr,ele) == DICT_OK) {
+                if (setTypeRemove(dstset,ele)) {
                     cardinality--;
                 }
             }
+            decrRefCount(ele);
         }
-        dictReleaseIterator(di);
+        setTypeReleaseIterator(si);
 
-        /* result set is empty? Exit asap. */
+        /* Exit when result set is empty. */
         if (op == REDIS_OP_DIFF && cardinality == 0) break;
     }
 
     /* Output the content of the resulting set, if not in STORE mode */
     if (!dstkey) {
         addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n",cardinality));
-        di = dictGetIterator(dstset->ptr);
-        while((de = dictNext(di)) != NULL) {
-            robj *ele;
-
-            ele = dictGetEntryKey(de);
+        si = setTypeInitIterator(dstset);
+        while((ele = setTypeNext(si)) != NULL) {
             addReplyBulk(c,ele);
+            decrRefCount(ele);
         }
-        dictReleaseIterator(di);
+        setTypeReleaseIterator(si);
         decrRefCount(dstset);
     } else {
         /* If we have a target key where to store the resulting set
          * create this key with the result set inside */
         dbDelete(c->db,dstkey);
-        if (dictSize((dict*)dstset->ptr) > 0) {
+        if (setTypeSize(dstset) > 0) {
             dbAdd(c->db,dstkey,dstset);
-            addReplyLongLong(c,dictSize((dict*)dstset->ptr));
+            addReplyLongLong(c,setTypeSize(dstset));
         } else {
             decrRefCount(dstset);
             addReply(c,shared.czero);
         }
         server.dirty++;
     }
-    zfree(dv);
+    zfree(sets);
 }
 
 static void sunionCommand(redisClient *c) {

From e24d93762f8945de0ee03f23b15d686cfa08bb58 Mon Sep 17 00:00:00 2001
From: Pieter Noordhuis <pcnoordhuis@gmail.com>
Date: Fri, 11 Jun 2010 19:22:27 +0200
Subject: [PATCH 04/10] intset housekeeping

---
 intset.c | 5 ++---
 intset.h | 3 ++-
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/intset.c b/intset.c
index f1bbcde3..1ee8263c 100644
--- a/intset.c
+++ b/intset.c
@@ -1,7 +1,6 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
-#include <stdint.h>
 #include "intset.h"
 #include "zmalloc.h"
 
@@ -152,7 +151,7 @@ intset *intsetAdd(intset *is, int64_t value, uint8_t *success) {
 }
 
 /* Delete integer from intset */
-intset *intsetDelete(intset *is, int64_t value, uint8_t *success) {
+intset *intsetRemove(intset *is, int64_t value, uint8_t *success) {
     uint8_t valenc = INTSET_VALUE_ENCODING(value);
     uint32_t pos;
     if (success) *success = 0;
@@ -370,7 +369,7 @@ int main(int argc, char **argv) {
             assert(intsetFind(is,v1));
 
             v2 = rand() % 0xfff;
-            is = intsetDelete(is,v2,NULL);
+            is = intsetRemove(is,v2,NULL);
             assert(!intsetFind(is,v2));
         }
         checkConsistency(is);
diff --git a/intset.h b/intset.h
index 9d91710a..b90c44e0 100644
--- a/intset.h
+++ b/intset.h
@@ -1,5 +1,6 @@
 #ifndef __INTSET_H
 #define __INTSET_H
+#include <stdint.h>
 
 typedef struct intset {
     uint32_t encoding;
@@ -9,7 +10,7 @@ typedef struct intset {
 
 intset *intsetNew(void);
 intset *intsetAdd(intset *is, int64_t value, uint8_t *success);
-intset *intsetDelete(intset *is, int64_t value, uint8_t *success);
+intset *intsetRemove(intset *is, int64_t value, uint8_t *success);
 uint8_t intsetFind(intset *is, int64_t value);
 int64_t intsetRandom(intset *is);
 

From d0b58d530027185a7fccd08bcda31efe06ef366b Mon Sep 17 00:00:00 2001
From: Pieter Noordhuis <pcnoordhuis@gmail.com>
Date: Sat, 12 Jun 2010 22:25:22 +0200
Subject: [PATCH 05/10] intset encoding for sets, refactored set tests to test
 both encodings

---
 Makefile                |   3 +-
 intset.c                |  19 ++-
 intset.h                |   2 +
 redis.c                 | 135 +++++++++++++++++---
 tests/unit/type/set.tcl | 264 +++++++++++++++++++++++++---------------
 5 files changed, 306 insertions(+), 117 deletions(-)

diff --git a/Makefile b/Makefile
index 46df88bb..72524f7c 100644
--- a/Makefile
+++ b/Makefile
@@ -15,7 +15,7 @@ endif
 CCOPT= $(CFLAGS) $(CCLINK) $(ARCH) $(PROF)
 DEBUG?= -g -rdynamic -ggdb 
 
-OBJ = adlist.o ae.o anet.o dict.o redis.o sds.o zmalloc.o lzf_c.o lzf_d.o pqsort.o zipmap.o sha1.o ziplist.o
+OBJ = adlist.o ae.o anet.o dict.o redis.o sds.o zmalloc.o lzf_c.o lzf_d.o pqsort.o zipmap.o sha1.o ziplist.o intset.o
 BENCHOBJ = ae.o anet.o redis-benchmark.o sds.o adlist.o zmalloc.o
 CLIOBJ = anet.o sds.o adlist.o redis-cli.o zmalloc.o linenoise.o
 CHECKDUMPOBJ = redis-check-dump.o lzf_c.o lzf_d.o
@@ -54,6 +54,7 @@ redis.o: redis.c fmacros.h config.h redis.h ae.h sds.h anet.h dict.h \
 sds.o: sds.c sds.h zmalloc.h
 zipmap.o: zipmap.c zmalloc.h
 ziplist.o: ziplist.c zmalloc.h
+intset.o: intset.c zmalloc.h
 zmalloc.o: zmalloc.c config.h
 
 redis-server: $(OBJ)
diff --git a/intset.c b/intset.c
index 1ee8263c..2532582e 100644
--- a/intset.c
+++ b/intset.c
@@ -60,8 +60,8 @@ static intset *intsetUpgrade(intset *is, uint8_t newenc, uint8_t extra, uint8_t
  * the value is not present in the intset and sets "pos" to the position
  * where "value" can be inserted. */
 static uint8_t intsetSearch(intset *is, int64_t value, uint32_t *pos) {
-    int min = 0, max = is->length-1, mid;
-    int64_t cur;
+    int min = 0, max = is->length-1, mid = -1;
+    int64_t cur = -1;
 
     /* The value can never be found when the set is empty */
     if (is->length == 0) {
@@ -179,6 +179,21 @@ int64_t intsetRandom(intset *is) {
     return INTSET_GET(is,rand()%is->length);
 }
 
+/* Sets the value to the value at the given position. When this position is
+ * out of range the function returns 0, when in range it returns 1. */
+uint8_t intsetGet(intset *is, uint32_t pos, int64_t *value) {
+    if (pos < is->length) {
+        *value = INTSET_GET(is,pos);
+        return 1;
+    }
+    return 0;
+}
+
+/* Return intset length */
+uint32_t intsetLen(intset *is) {
+    return is->length;
+}
+
 #ifdef INTSET_TEST_MAIN
 #include <sys/time.h>
 
diff --git a/intset.h b/intset.h
index b90c44e0..25afc18d 100644
--- a/intset.h
+++ b/intset.h
@@ -13,5 +13,7 @@ intset *intsetAdd(intset *is, int64_t value, uint8_t *success);
 intset *intsetRemove(intset *is, int64_t value, uint8_t *success);
 uint8_t intsetFind(intset *is, int64_t value);
 int64_t intsetRandom(intset *is);
+uint8_t intsetGet(intset *is, uint32_t pos, int64_t *value);
+uint32_t intsetLen(intset *is);
 
 #endif // __INTSET_H
diff --git a/redis.c b/redis.c
index cb7bca8e..c1df3293 100644
--- a/redis.c
+++ b/redis.c
@@ -76,6 +76,7 @@
 #include "pqsort.h" /* Partial qsort for SORT+LIMIT */
 #include "zipmap.h" /* Compact dictionary-alike data structure */
 #include "ziplist.h" /* Compact list data structure */
+#include "intset.h" /* Compact integer set structure */
 #include "sha1.h"   /* SHA1 is used for DEBUG DIGEST */
 #include "release.h" /* Release and/or git repository information */
 
@@ -132,9 +133,10 @@
 #define REDIS_ENCODING_ZIPMAP 3  /* Encoded as zipmap */
 #define REDIS_ENCODING_LIST 4    /* Encoded as zipmap */
 #define REDIS_ENCODING_ZIPLIST 5 /* Encoded as ziplist */
+#define REDIS_ENCODING_INTSET 6  /* Encoded as intset */
 
 static char* strencoding[] = {
-    "raw", "int", "hashtable", "zipmap", "list", "ziplist"
+    "raw", "int", "hashtable", "zipmap", "list", "ziplist", "intset"
 };
 
 /* Object types only used for dumping to disk */
@@ -651,6 +653,7 @@ static void call(redisClient *c, struct redisCommand *cmd);
 static void resetClient(redisClient *c);
 static void convertToRealHash(robj *o);
 static void listTypeConvert(robj *o, int enc);
+static void setTypeConvert(robj *o, int enc);
 static int pubsubUnsubscribeAllChannels(redisClient *c, int notify);
 static int pubsubUnsubscribeAllPatterns(redisClient *c, int notify);
 static void freePubsubPattern(void *p);
@@ -3069,6 +3072,13 @@ static robj *createSetObject(void) {
     return o;
 }
 
+static robj *createIntsetObject(void) {
+    intset *is = intsetNew();
+    robj *o = createObject(REDIS_SET,is);
+    o->encoding = REDIS_ENCODING_INTSET;
+    return o;
+}
+
 static robj *createHashObject(void) {
     /* All the Hashes start as zipmaps. Will be automatically converted
      * into hash tables if there are enough elements or big elements
@@ -3107,7 +3117,16 @@ static void freeListObject(robj *o) {
 }
 
 static void freeSetObject(robj *o) {
-    dictRelease((dict*) o->ptr);
+    switch (o->encoding) {
+    case REDIS_ENCODING_HT:
+        dictRelease((dict*) o->ptr);
+        break;
+    case REDIS_ENCODING_INTSET:
+        zfree(o->ptr);
+        break;
+    default:
+        redisPanic("Unknown set encoding type");
+    }
 }
 
 static void freeZsetObject(robj *o) {
@@ -3371,7 +3390,7 @@ static int getLongLongFromObject(robj *o, long long *target) {
         }
     }
 
-    *target = value;
+    if (target) *target = value;
     return REDIS_OK;
 }
 
@@ -3789,17 +3808,29 @@ static int rdbSaveObject(FILE *fp, robj *o) {
         }
     } else if (o->type == REDIS_SET) {
         /* Save a set value */
-        dict *set = o->ptr;
-        dictIterator *di = dictGetIterator(set);
-        dictEntry *de;
+        if (o->encoding == REDIS_ENCODING_HT) {
+            dict *set = o->ptr;
+            dictIterator *di = dictGetIterator(set);
+            dictEntry *de;
 
-        if (rdbSaveLen(fp,dictSize(set)) == -1) return -1;
-        while((de = dictNext(di)) != NULL) {
-            robj *eleobj = dictGetEntryKey(de);
+            if (rdbSaveLen(fp,dictSize(set)) == -1) return -1;
+            while((de = dictNext(di)) != NULL) {
+                robj *eleobj = dictGetEntryKey(de);
+                if (rdbSaveStringObject(fp,eleobj) == -1) return -1;
+            }
+            dictReleaseIterator(di);
+        } else if (o->encoding == REDIS_ENCODING_INTSET) {
+            intset *is = o->ptr;
+            long long llval;
+            int i = 0;
 
-            if (rdbSaveStringObject(fp,eleobj) == -1) return -1;
+            if (rdbSaveLen(fp,intsetLen(is)) == -1) return -1;
+            while(intsetGet(is,i++,&llval)) {
+                if (rdbSaveLongLongAsStringObject(fp,llval) == -1) return -1;
+            }
+        } else {
+            redisPanic("Unknown set encoding");
         }
-        dictReleaseIterator(di);
     } else if (o->type == REDIS_ZSET) {
         /* Save a set value */
         zset *zs = o->ptr;
@@ -5459,12 +5490,37 @@ static void rpoplpushcommand(redisClient *c) {
 
 /* ==================================== Sets ================================ */
 
+/* Factory method to return a set that *can* hold "value". When the object has
+ * an integer-encodable value, an intset will be returned. Otherwise a regular
+ * hash table. */
+static robj *setTypeCreate(robj *value) {
+    if (getLongLongFromObject(value,NULL) == REDIS_OK)
+        return createIntsetObject();
+    return createSetObject();
+}
+
 static int setTypeAdd(robj *subject, robj *value) {
+    long long llval;
     if (subject->encoding == REDIS_ENCODING_HT) {
         if (dictAdd(subject->ptr,value,NULL) == DICT_OK) {
             incrRefCount(value);
             return 1;
         }
+    } else if (subject->encoding == REDIS_ENCODING_INTSET) {
+        if (getLongLongFromObject(value,&llval) == REDIS_OK) {
+            uint8_t success;
+            subject->ptr = intsetAdd(subject->ptr,llval,&success);
+            if (success) return 1;
+        } else {
+            /* Failed to get integer from object, convert to regular set. */
+            setTypeConvert(subject,REDIS_ENCODING_HT);
+
+            /* The set *was* an intset and this value is not integer
+             * encodable, so dictAdd should always work. */
+            redisAssert(dictAdd(subject->ptr,value,NULL) == DICT_OK);
+            incrRefCount(value);
+            return 1;
+        }
     } else {
         redisPanic("Unknown set encoding");
     }
@@ -5472,11 +5528,18 @@ static int setTypeAdd(robj *subject, robj *value) {
 }
 
 static int setTypeRemove(robj *subject, robj *value) {
+    long long llval;
     if (subject->encoding == REDIS_ENCODING_HT) {
         if (dictDelete(subject->ptr,value) == DICT_OK) {
             if (htNeedsResize(subject->ptr)) dictResize(subject->ptr);
             return 1;
         }
+    } else if (subject->encoding == REDIS_ENCODING_INTSET) {
+        if (getLongLongFromObject(value,&llval) == REDIS_OK) {
+            uint8_t success;
+            subject->ptr = intsetRemove(subject->ptr,llval,&success);
+            if (success) return 1;
+        }
     } else {
         redisPanic("Unknown set encoding");
     }
@@ -5484,24 +5547,35 @@ static int setTypeRemove(robj *subject, robj *value) {
 }
 
 static int setTypeIsMember(robj *subject, robj *value) {
+    long long llval;
     if (subject->encoding == REDIS_ENCODING_HT) {
         return dictFind((dict*)subject->ptr,value) != NULL;
+    } else if (subject->encoding == REDIS_ENCODING_INTSET) {
+        if (getLongLongFromObject(value,&llval) == REDIS_OK) {
+            return intsetFind((intset*)subject->ptr,llval);
+        }
     } else {
         redisPanic("Unknown set encoding");
     }
+    return 0;
 }
 
 /* Structure to hold set iteration abstraction. */
 typedef struct {
+    robj *subject;
     int encoding;
+    int ii; /* intset iterator */
     dictIterator *di;
 } setIterator;
 
 static setIterator *setTypeInitIterator(robj *subject) {
     setIterator *si = zmalloc(sizeof(setIterator));
+    si->subject = subject;
     si->encoding = subject->encoding;
     if (si->encoding == REDIS_ENCODING_HT) {
         si->di = dictGetIterator(subject->ptr);
+    } else if (si->encoding == REDIS_ENCODING_INTSET) {
+        si->ii = 0;
     } else {
         redisPanic("Unknown set encoding");
     }
@@ -5525,6 +5599,10 @@ static robj *setTypeNext(setIterator *si) {
             ret = dictGetEntryKey(de);
             incrRefCount(ret);
         }
+    } else if (si->encoding == REDIS_ENCODING_INTSET) {
+        long long llval;
+        if (intsetGet(si->subject->ptr,si->ii++,&llval))
+            ret = createStringObjectFromLongLong(llval);
     }
     return ret;
 }
@@ -5538,6 +5616,9 @@ robj *setTypeRandomElement(robj *subject) {
         dictEntry *de = dictGetRandomKey(subject->ptr);
         ret = dictGetEntryKey(de);
         incrRefCount(ret);
+    } else if (subject->encoding == REDIS_ENCODING_INTSET) {
+        long long llval = intsetRandom(subject->ptr);
+        ret = createStringObjectFromLongLong(llval);
     } else {
         redisPanic("Unknown set encoding");
     }
@@ -5547,17 +5628,41 @@ robj *setTypeRandomElement(robj *subject) {
 static unsigned long setTypeSize(robj *subject) {
     if (subject->encoding == REDIS_ENCODING_HT) {
         return dictSize((dict*)subject->ptr);
+    } else if (subject->encoding == REDIS_ENCODING_INTSET) {
+        return intsetLen((intset*)subject->ptr);
     } else {
         redisPanic("Unknown set encoding");
     }
 }
 
+static void setTypeConvert(robj *subject, int enc) {
+    setIterator *si;
+    robj *element;
+    redisAssert(subject->type == REDIS_SET);
+
+    if (enc == REDIS_ENCODING_HT) {
+        dict *d = dictCreate(&setDictType,NULL);
+
+        /* setTypeGet returns a robj with incremented refcount */
+        si = setTypeInitIterator(subject);
+        while ((element = setTypeNext(si)) != NULL)
+            redisAssert(dictAdd(d,element,NULL) == DICT_OK);
+        setTypeReleaseIterator(si);
+
+        subject->encoding = REDIS_ENCODING_HT;
+        zfree(subject->ptr);
+        subject->ptr = d;
+    } else {
+        redisPanic("Unsupported set conversion");
+    }
+}
+
 static void saddCommand(redisClient *c) {
     robj *set;
 
     set = lookupKeyWrite(c->db,c->argv[1]);
     if (set == NULL) {
-        set = createSetObject();
+        set = setTypeCreate(c->argv[2]);
         dbAdd(c->db,c->argv[1],set);
     } else {
         if (set->type != REDIS_SET) {
@@ -5616,7 +5721,7 @@ static void smoveCommand(redisClient *c) {
     server.dirty++;
     /* Add the element to the destination set */
     if (!dstset) {
-        dstset = createSetObject();
+        dstset = setTypeCreate(c->argv[3]);
         dbAdd(c->db,c->argv[2],dstset);
     }
     setTypeAdd(dstset,c->argv[3]);
@@ -5724,7 +5829,7 @@ static void sinterGenericCommand(redisClient *c, robj **setkeys, unsigned long s
     } else {
         /* If we have a target key where to store the resulting set
          * create this key with an empty set inside */
-        dstset = createSetObject();
+        dstset = createIntsetObject();
     }
 
     /* Iterate all the elements of the first (smallest) set, and test
@@ -5802,7 +5907,7 @@ static void sunionDiffGenericCommand(redisClient *c, robj **setkeys, int setnum,
     /* We need a temp set object to store our union. If the dstkey
      * is not NULL (that is, we are inside an SUNIONSTORE operation) then
      * this set object will be the resulting object to set into the target key*/
-    dstset = createSetObject();
+    dstset = createIntsetObject();
 
     /* Iterate all the elements of all the sets, add every element a single
      * time to the result set */
diff --git a/tests/unit/type/set.tcl b/tests/unit/type/set.tcl
index 58ea2b5b..c4fd4d76 100644
--- a/tests/unit/type/set.tcl
+++ b/tests/unit/type/set.tcl
@@ -1,119 +1,185 @@
 start_server {tags {"set"}} {
-    test {SADD, SCARD, SISMEMBER, SMEMBERS basics} {
-        r sadd myset foo
-        r sadd myset bar
-        list [r scard myset] [r sismember myset foo] \
-            [r sismember myset bar] [r sismember myset bla] \
-            [lsort [r smembers myset]]
-    } {2 1 1 0 {bar foo}}
+    proc create_set {key entries} {
+        r del $key
+        foreach entry $entries { r sadd $key $entry }
+    }
 
-    test {SADD adding the same element multiple times} {
-        r sadd myset foo
-        r sadd myset foo
-        r sadd myset foo
-        r scard myset
-    } {2}
+    test {SADD, SCARD, SISMEMBER, SMEMBERS basics - regular set} {
+        create_set myset {foo}
+        assert_encoding hashtable myset
+        assert_equal 1 [r sadd myset bar]
+        assert_equal 0 [r sadd myset bar]
+        assert_equal 2 [r scard myset]
+        assert_equal 1 [r sismember myset foo]
+        assert_equal 1 [r sismember myset bar]
+        assert_equal 0 [r sismember myset bla]
+        assert_equal {bar foo} [lsort [r smembers myset]]
+    }
+
+    test {SADD, SCARD, SISMEMBER, SMEMBERS basics - intset} {
+        create_set myset {17}
+        assert_encoding intset myset
+        assert_equal 1 [r sadd myset 16]
+        assert_equal 0 [r sadd myset 16]
+        assert_equal 2 [r scard myset]
+        assert_equal 1 [r sismember myset 16]
+        assert_equal 1 [r sismember myset 17]
+        assert_equal 0 [r sismember myset 18]
+        assert_equal {16 17} [lsort [r smembers myset]]
+    }
 
     test {SADD against non set} {
         r lpush mylist foo
-        catch {r sadd mylist bar} err
-        format $err
-    } {ERR*kind*}
+        assert_error ERR*kind* {r sadd mylist bar}
+    }
 
-    test {SREM basics} {
-        r sadd myset ciao
-        r srem myset foo
-        lsort [r smembers myset]
-    } {bar ciao}
+    test {SREM basics - regular set} {
+        create_set myset {foo bar ciao}
+        assert_encoding hashtable myset
+        assert_equal 0 [r srem myset qux]
+        assert_equal 1 [r srem myset foo]
+        assert_equal {bar ciao} [lsort [r smembers myset]]
+    }
 
-    test {Mass SADD and SINTER with two sets} {
+    test {SREM basics - intset} {
+        create_set myset {3 4 5}
+        assert_encoding intset myset
+        assert_equal 0 [r srem myset 6]
+        assert_equal 1 [r srem myset 4]
+        assert_equal {3 5} [lsort [r smembers myset]]
+    }
+
+    foreach {type} {hashtable intset} {
+        for {set i 1} {$i <= 5} {incr i} {
+            r del [format "set%d" $i]
+        }
         for {set i 0} {$i < 1000} {incr i} {
             r sadd set1 $i
             r sadd set2 [expr $i+995]
         }
-        lsort [r sinter set1 set2]
-    } {995 996 997 998 999}
-
-    test {SUNION with two sets} {
-        lsort [r sunion set1 set2]
-    } [lsort -uniq "[r smembers set1] [r smembers set2]"]
-
-    test {SINTERSTORE with two sets} {
-        r sinterstore setres set1 set2
-        lsort [r smembers setres]
-    } {995 996 997 998 999}
-
-    test {SINTERSTORE with two sets, after a DEBUG RELOAD} {
-        r debug reload
-        r sinterstore setres set1 set2
-        lsort [r smembers setres]
-    } {995 996 997 998 999}
-
-    test {SUNIONSTORE with two sets} {
-        r sunionstore setres set1 set2
-        lsort [r smembers setres]
-    } [lsort -uniq "[r smembers set1] [r smembers set2]"]
-
-    test {SUNIONSTORE against non existing keys} {
-        r set setres xxx
-        list [r sunionstore setres foo111 bar222] [r exists xxx]
-    } {0 0}
-
-    test {SINTER against three sets} {
-        r sadd set3 999
-        r sadd set3 995
-        r sadd set3 1000
-        r sadd set3 2000
-        lsort [r sinter set1 set2 set3]
-    } {995 999}
-
-    test {SINTERSTORE with three sets} {
-        r sinterstore setres set1 set2 set3
-        lsort [r smembers setres]
-    } {995 999}
-
-    test {SUNION with non existing keys} {
-        lsort [r sunion nokey1 set1 set2 nokey2]
-    } [lsort -uniq "[r smembers set1] [r smembers set2]"]
-
-    test {SDIFF with two sets} {
+        foreach i {999 995 1000 2000} {
+            r sadd set3 $i
+        }
         for {set i 5} {$i < 1000} {incr i} {
             r sadd set4 $i
         }
-        lsort [r sdiff set1 set4]
-    } {0 1 2 3 4}
-
-    test {SDIFF with three sets} {
         r sadd set5 0
-        lsort [r sdiff set1 set4 set5]
-    } {1 2 3 4}
 
-    test {SDIFFSTORE with three sets} {
-        r sdiffstore sres set1 set4 set5
-        lsort [r smembers sres]
-    } {1 2 3 4}
-
-    test {SPOP basics} {
-        r del myset
-        r sadd myset 1
-        r sadd myset 2
-        r sadd myset 3
-        list [lsort [list [r spop myset] [r spop myset] [r spop myset]]] [r scard myset]
-    } {{1 2 3} 0}
-    
-    test {SRANDMEMBER} {
-        r del myset
-        r sadd myset a
-        r sadd myset b
-        r sadd myset c
-        unset -nocomplain myset
-        array set myset {}
-        for {set i 0} {$i < 100} {incr i} {
-            set myset([r srandmember myset]) 1
+        # it is possible that a hashtable encoded only contains integers,
+        # because it is converted from an intset to a hashtable when a
+        # non-integer element is added and then removed.
+        if {$type eq "hashtable"} {
+            for {set i 1} {$i <= 5} {incr i} {
+                r sadd [format "set%d" $i] foo
+                r srem [format "set%d" $i] foo
+            }
         }
-        lsort [array names myset]
-    } {a b c}
-    
+
+        test "Generated sets must be encoded as $type" {
+            for {set i 1} {$i <= 5} {incr i} {
+                assert_encoding $type [format "set%d" $i]
+            }
+        }
+
+        test "SINTER with two sets - $type" {
+            assert_equal {995 996 997 998 999} [lsort [r sinter set1 set2]]
+        }
+
+        test "SINTERSTORE with two sets - $type" {
+            r sinterstore setres set1 set2
+            assert_encoding intset setres
+            assert_equal {995 996 997 998 999} [lsort [r smembers setres]]
+        }
+
+        test "SINTERSTORE with two sets, after a DEBUG RELOAD - $type" {
+            r debug reload
+            r sinterstore setres set1 set2
+            assert_encoding intset setres
+            assert_equal {995 996 997 998 999} [lsort [r smembers setres]]
+        }
+
+        test "SUNION with two sets - $type" {
+            set expected [lsort -uniq "[r smembers set1] [r smembers set2]"]
+            assert_equal $expected [lsort [r sunion set1 set2]]
+        }
+
+        test "SUNIONSTORE with two sets - $type" {
+            r sunionstore setres set1 set2
+            assert_encoding intset setres
+            set expected [lsort -uniq "[r smembers set1] [r smembers set2]"]
+            assert_equal $expected [lsort [r smembers setres]]
+        }
+
+        test "SINTER against three sets - $type" {
+            assert_equal {995 999} [lsort [r sinter set1 set2 set3]]
+        }
+
+        test "SINTERSTORE with three sets - $type" {
+            r sinterstore setres set1 set2 set3
+            assert_equal {995 999} [r smembers setres]
+        }
+
+        test "SUNION with non existing keys - $type" {
+            set expected [lsort -uniq "[r smembers set1] [r smembers set2]"]
+            assert_equal $expected [lsort [r sunion nokey1 set1 set2 nokey2]]
+        }
+
+        test "SDIFF with two sets - $type" {
+            assert_equal {0 1 2 3 4} [lsort [r sdiff set1 set4]]
+        }
+
+        test "SDIFF with three sets - $type" {
+            assert_equal {1 2 3 4} [lsort [r sdiff set1 set4 set5]]
+        }
+
+        test "SDIFFSTORE with three sets - $type" {
+            r sdiffstore setres set1 set4 set5
+            assert_encoding intset setres
+            assert_equal {1 2 3 4} [lsort [r smembers setres]]
+        }
+    }
+
+    test "SINTER against non-set should throw error" {
+        r set key1 x
+        assert_error "ERR*wrong kind*" {r sinter key1 noset}
+    }
+
+    test "SUNION against non-set should throw error" {
+        r set key1 x
+        assert_error "ERR*wrong kind*" {r sunion key1 noset}
+    }
+
+    test "SINTERSTORE against non existing keys should delete dstkey" {
+        r set setres xxx
+        assert_equal 0 [r sinterstore setres foo111 bar222]
+        assert_equal 0 [r exists setres]
+    }
+
+    test "SUNIONSTORE against non existing keys should delete dstkey" {
+        r set setres xxx
+        assert_equal 0 [r sunionstore setres foo111 bar222]
+        assert_equal 0 [r exists setres]
+    }
+
+    foreach {type contents} {hashtable {a b c} intset {1 2 3}} {
+        test "SPOP basics - $type" {
+            create_set myset $contents
+            assert_encoding $type myset
+            assert_equal $contents [lsort [list [r spop myset] [r spop myset] [r spop myset]]]
+            assert_equal 0 [r scard myset]
+        }
+
+        test "SRANDMEMBER - $type" {
+            create_set myset $contents
+            unset -nocomplain myset
+            array set myset {}
+            for {set i 0} {$i < 100} {incr i} {
+                set myset([r srandmember myset]) 1
+            }
+            assert_equal $contents [lsort [array names myset]]
+        }
+    }
+
     test {SMOVE basics} {
         r sadd myset1 a
         r sadd myset1 b

From b978abbf022b1810031b5aaa171ef0899f7fe77d Mon Sep 17 00:00:00 2001
From: Pieter Noordhuis <pcnoordhuis@gmail.com>
Date: Sun, 13 Jun 2010 11:16:18 +0200
Subject: [PATCH 06/10] small refactor of SMOVE and tests for SMOVE on sets
 with different encoding

---
 redis.c                 | 51 +++++++++++++----------
 tests/unit/type/set.tcl | 89 ++++++++++++++++++++++++++++-------------
 2 files changed, 92 insertions(+), 48 deletions(-)

diff --git a/redis.c b/redis.c
index c1df3293..51ca6d2a 100644
--- a/redis.c
+++ b/redis.c
@@ -5694,37 +5694,46 @@ static void sremCommand(redisClient *c) {
 }
 
 static void smoveCommand(redisClient *c) {
-    robj *srcset, *dstset;
-
+    robj *srcset, *dstset, *ele;
     srcset = lookupKeyWrite(c->db,c->argv[1]);
     dstset = lookupKeyWrite(c->db,c->argv[2]);
+    ele = c->argv[3];
 
-    /* If the source key does not exist return 0, if it's of the wrong type
-     * raise an error */
-    if (srcset == NULL || srcset->type != REDIS_SET) {
-        addReply(c, srcset ? shared.wrongtypeerr : shared.czero);
-        return;
-    }
-    /* Error if the destination key is not a set as well */
-    if (dstset && dstset->type != REDIS_SET) {
-        addReply(c,shared.wrongtypeerr);
-        return;
-    }
-    /* Remove the element from the source set */
-    if (!setTypeRemove(srcset,c->argv[3])) {
-        /* Key not found in the src set! return zero */
+    /* If the source key does not exist return 0 */
+    if (srcset == NULL) {
         addReply(c,shared.czero);
         return;
     }
-    if (setTypeSize(srcset) == 0 && srcset != dstset)
-        dbDelete(c->db,c->argv[1]);
+
+    /* If the source key has the wrong type, or the destination key
+     * is set and has the wrong type, return with an error. */
+    if (checkType(c,srcset,REDIS_SET) ||
+        (dstset && checkType(c,dstset,REDIS_SET))) return;
+
+    /* If srcset and dstset are equal, SMOVE is a no-op */
+    if (srcset == dstset) {
+        addReply(c,shared.cone);
+        return;
+    }
+
+    /* If the element cannot be removed from the src set, return 0. */
+    if (!setTypeRemove(srcset,ele)) {
+        addReply(c,shared.czero);
+        return;
+    }
+
+    /* Remove the src set from the database when empty */
+    if (setTypeSize(srcset) == 0) dbDelete(c->db,c->argv[1]);
     server.dirty++;
-    /* Add the element to the destination set */
+
+    /* Create the destination set when it doesn't exist */
     if (!dstset) {
-        dstset = setTypeCreate(c->argv[3]);
+        dstset = setTypeCreate(ele);
         dbAdd(c->db,c->argv[2],dstset);
     }
-    setTypeAdd(dstset,c->argv[3]);
+
+    /* An extra key has changed when ele was successfully added to dstset */
+    if (setTypeAdd(dstset,ele)) server.dirty++;
     addReply(c,shared.cone);
 }
 
diff --git a/tests/unit/type/set.tcl b/tests/unit/type/set.tcl
index c4fd4d76..5b8d961e 100644
--- a/tests/unit/type/set.tcl
+++ b/tests/unit/type/set.tcl
@@ -180,38 +180,73 @@ start_server {tags {"set"}} {
         }
     }
 
-    test {SMOVE basics} {
-        r sadd myset1 a
-        r sadd myset1 b
-        r sadd myset1 c
-        r sadd myset2 x
-        r sadd myset2 y
-        r sadd myset2 z
-        r smove myset1 myset2 a
-        list [lsort [r smembers myset2]] [lsort [r smembers myset1]]
-    } {{a x y z} {b c}}
+    proc setup_move {} {
+        r del myset3 myset4
+        create_set myset1 {1 a b}
+        create_set myset2 {2 3 4}
+        assert_encoding hashtable myset1
+        assert_encoding intset myset2
+    }
 
-    test {SMOVE non existing key} {
-        list [r smove myset1 myset2 foo] [lsort [r smembers myset2]] [lsort [r smembers myset1]]
-    } {0 {a x y z} {b c}}
+    test "SMOVE basics - from regular set to intset" {
+        # move a non-integer element to an intset should convert encoding
+        setup_move
+        assert_equal 1 [r smove myset1 myset2 a]
+        assert_equal {1 b} [lsort [r smembers myset1]]
+        assert_equal {2 3 4 a} [lsort [r smembers myset2]]
+        assert_encoding hashtable myset2
 
-    test {SMOVE non existing src set} {
-        list [r smove noset myset2 foo] [lsort [r smembers myset2]]
-    } {0 {a x y z}}
+        # move an integer element should not convert the encoding
+        setup_move
+        assert_equal 1 [r smove myset1 myset2 1]
+        assert_equal {a b} [lsort [r smembers myset1]]
+        assert_equal {1 2 3 4} [lsort [r smembers myset2]]
+        assert_encoding intset myset2
+    }
 
-    test {SMOVE non existing dst set} {
-        list [r smove myset2 myset3 y] [lsort [r smembers myset2]] [lsort [r smembers myset3]]
-    } {1 {a x z} y}
+    test "SMOVE basics - from intset to regular set" {
+        setup_move
+        assert_equal 1 [r smove myset2 myset1 2]
+        assert_equal {1 2 a b} [lsort [r smembers myset1]]
+        assert_equal {3 4} [lsort [r smembers myset2]]
+    }
 
-    test {SMOVE wrong src key type} {
+    test "SMOVE non existing key" {
+        setup_move
+        assert_equal 0 [r smove myset1 myset2 foo]
+        assert_equal {1 a b} [lsort [r smembers myset1]]
+        assert_equal {2 3 4} [lsort [r smembers myset2]]
+    }
+
+    test "SMOVE non existing src set" {
+        setup_move
+        assert_equal 0 [r smove noset myset2 foo]
+        assert_equal {2 3 4} [lsort [r smembers myset2]]
+    }
+
+    test "SMOVE from regular set to non existing destination set" {
+        setup_move
+        assert_equal 1 [r smove myset1 myset3 a]
+        assert_equal {1 b} [lsort [r smembers myset1]]
+        assert_equal {a} [lsort [r smembers myset3]]
+        assert_encoding hashtable myset3
+    }
+
+    test "SMOVE from intset to non existing destination set" {
+        setup_move
+        assert_equal 1 [r smove myset2 myset3 2]
+        assert_equal {3 4} [lsort [r smembers myset2]]
+        assert_equal {2} [lsort [r smembers myset3]]
+        assert_encoding intset myset3
+    }
+
+    test "SMOVE wrong src key type" {
         r set x 10
-        catch {r smove x myset2 foo} err
-        format $err
-    } {ERR*}
+        assert_error "ERR*wrong kind*" {r smove x myset2 foo}
+    }
 
-    test {SMOVE wrong dst key type} {
+    test "SMOVE wrong dst key type" {
         r set x 10
-        catch {r smove myset2 x foo} err
-        format $err
-    } {ERR*}
+        assert_error "ERR*wrong kind*" {r smove myset2 x foo}
+    }
 }

From 70ff3511bc427fe05df22e1aba4071910f33e289 Mon Sep 17 00:00:00 2001
From: Pieter Noordhuis <pcnoordhuis@gmail.com>
Date: Sun, 13 Jun 2010 15:21:25 +0200
Subject: [PATCH 07/10] configure maximum number of entries in an intset

---
 redis.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)

diff --git a/redis.c b/redis.c
index 51ca6d2a..0cf768ac 100644
--- a/redis.c
+++ b/redis.c
@@ -244,6 +244,7 @@ static char* strencoding[] = {
 #define REDIS_HASH_MAX_ZIPMAP_VALUE 512
 #define REDIS_LIST_MAX_ZIPLIST_ENTRIES 1024
 #define REDIS_LIST_MAX_ZIPLIST_VALUE 32
+#define REDIS_SET_MAX_INTSET_ENTRIES 4096
 
 /* We can print the stacktrace, so our assert is defined this way: */
 #define redisAssert(_e) ((_e)?(void)0 : (_redisAssert(#_e,__FILE__,__LINE__),_exit(1)))
@@ -434,6 +435,7 @@ struct redisServer {
     size_t hash_max_zipmap_value;
     size_t list_max_ziplist_entries;
     size_t list_max_ziplist_value;
+    size_t set_max_intset_entries;
     /* Virtual memory state */
     FILE *vm_fp;
     int vm_fd;
@@ -1765,6 +1767,7 @@ static void initServerConfig() {
     server.hash_max_zipmap_value = REDIS_HASH_MAX_ZIPMAP_VALUE;
     server.list_max_ziplist_entries = REDIS_LIST_MAX_ZIPLIST_ENTRIES;
     server.list_max_ziplist_value = REDIS_LIST_MAX_ZIPLIST_VALUE;
+    server.set_max_intset_entries = REDIS_SET_MAX_INTSET_ENTRIES;
     server.shutdown_asap = 0;
 
     resetServerSaveParams();
@@ -2047,6 +2050,8 @@ static void loadServerConfig(char *filename) {
             server.list_max_ziplist_entries = memtoll(argv[1], NULL);
         } else if (!strcasecmp(argv[0],"list-max-ziplist-value") && argc == 2){
             server.list_max_ziplist_value = memtoll(argv[1], NULL);
+        } else if (!strcasecmp(argv[0],"set-max-intset-entries") && argc == 2){
+            server.set_max_intset_entries = memtoll(argv[1], NULL);
         } else {
             err = "Bad directive or wrong number of arguments"; goto loaderr;
         }
@@ -5508,9 +5513,15 @@ static int setTypeAdd(robj *subject, robj *value) {
         }
     } else if (subject->encoding == REDIS_ENCODING_INTSET) {
         if (getLongLongFromObject(value,&llval) == REDIS_OK) {
-            uint8_t success;
+            uint8_t success = 0;
             subject->ptr = intsetAdd(subject->ptr,llval,&success);
-            if (success) return 1;
+            if (success) {
+                /* Convert to regular set when the intset contains
+                 * too many entries. */
+                if (intsetLen(subject->ptr) > server.set_max_intset_entries)
+                    setTypeConvert(subject,REDIS_ENCODING_HT);
+                return 1;
+            }
         } else {
             /* Failed to get integer from object, convert to regular set. */
             setTypeConvert(subject,REDIS_ENCODING_HT);

From ab37269c38a1e60ec03f3d537b1c7ed33f761225 Mon Sep 17 00:00:00 2001
From: Pieter Noordhuis <pcnoordhuis@gmail.com>
Date: Sun, 13 Jun 2010 21:23:43 +0200
Subject: [PATCH 08/10] use max number of intset entries in tests and make
 SUNION/SINTER/SDIFF tests use less entries

---
 tests/unit/type/set.tcl | 40 ++++++++++++++++++++++++++++++----------
 1 file changed, 30 insertions(+), 10 deletions(-)

diff --git a/tests/unit/type/set.tcl b/tests/unit/type/set.tcl
index 5b8d961e..f9e7f4bd 100644
--- a/tests/unit/type/set.tcl
+++ b/tests/unit/type/set.tcl
@@ -1,4 +1,9 @@
-start_server {tags {"set"}} {
+start_server {
+    tags {"set"}
+    overrides {
+        "set-max-intset-entries" 512
+    }
+} {
     proc create_set {key entries} {
         r del $key
         foreach entry $entries { r sadd $key $entry }
@@ -33,6 +38,21 @@ start_server {tags {"set"}} {
         assert_error ERR*kind* {r sadd mylist bar}
     }
 
+    test "SADD a non-integer against an intset" {
+        create_set myset {1 2 3}
+        assert_encoding intset myset
+        assert_equal 1 [r sadd myset a]
+        assert_encoding hashtable myset
+    }
+
+    test "SADD overflows the maximum allowed integers in an intset" {
+        r del myset
+        for {set i 0} {$i < 512} {incr i} { r sadd myset $i }
+        assert_encoding intset myset
+        assert_equal 1 [r sadd myset 512]
+        assert_encoding hashtable myset
+    }
+
     test {SREM basics - regular set} {
         create_set myset {foo bar ciao}
         assert_encoding hashtable myset
@@ -53,14 +73,14 @@ start_server {tags {"set"}} {
         for {set i 1} {$i <= 5} {incr i} {
             r del [format "set%d" $i]
         }
-        for {set i 0} {$i < 1000} {incr i} {
+        for {set i 0} {$i < 200} {incr i} {
             r sadd set1 $i
-            r sadd set2 [expr $i+995]
+            r sadd set2 [expr $i+195]
         }
-        foreach i {999 995 1000 2000} {
+        foreach i {199 195 1000 2000} {
             r sadd set3 $i
         }
-        for {set i 5} {$i < 1000} {incr i} {
+        for {set i 5} {$i < 200} {incr i} {
             r sadd set4 $i
         }
         r sadd set5 0
@@ -82,20 +102,20 @@ start_server {tags {"set"}} {
         }
 
         test "SINTER with two sets - $type" {
-            assert_equal {995 996 997 998 999} [lsort [r sinter set1 set2]]
+            assert_equal {195 196 197 198 199} [lsort [r sinter set1 set2]]
         }
 
         test "SINTERSTORE with two sets - $type" {
             r sinterstore setres set1 set2
             assert_encoding intset setres
-            assert_equal {995 996 997 998 999} [lsort [r smembers setres]]
+            assert_equal {195 196 197 198 199} [lsort [r smembers setres]]
         }
 
         test "SINTERSTORE with two sets, after a DEBUG RELOAD - $type" {
             r debug reload
             r sinterstore setres set1 set2
             assert_encoding intset setres
-            assert_equal {995 996 997 998 999} [lsort [r smembers setres]]
+            assert_equal {195 196 197 198 199} [lsort [r smembers setres]]
         }
 
         test "SUNION with two sets - $type" {
@@ -111,12 +131,12 @@ start_server {tags {"set"}} {
         }
 
         test "SINTER against three sets - $type" {
-            assert_equal {995 999} [lsort [r sinter set1 set2 set3]]
+            assert_equal {195 199} [lsort [r sinter set1 set2 set3]]
         }
 
         test "SINTERSTORE with three sets - $type" {
             r sinterstore setres set1 set2 set3
-            assert_equal {995 999} [r smembers setres]
+            assert_equal {195 199} [r smembers setres]
         }
 
         test "SUNION with non existing keys - $type" {

From 273f6169301eba5461d90f07ec683ae06572e931 Mon Sep 17 00:00:00 2001
From: Pieter Noordhuis <pcnoordhuis@gmail.com>
Date: Sun, 13 Jun 2010 21:42:04 +0200
Subject: [PATCH 09/10] make sure sets have the right encoding when loaded from
 rdb

---
 redis.c                 | 34 ++++++++++++++++++++++++++++------
 tests/unit/type/set.tcl | 15 +++++++++++++++
 2 files changed, 43 insertions(+), 6 deletions(-)

diff --git a/redis.c b/redis.c
index 0cf768ac..0d6648ee 100644
--- a/redis.c
+++ b/redis.c
@@ -4234,16 +4234,38 @@ static robj *rdbLoadObject(int type, FILE *fp) {
     } else if (type == REDIS_SET) {
         /* Read list/set value */
         if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
-        o = createSetObject();
-        /* It's faster to expand the dict to the right size asap in order
-         * to avoid rehashing */
-        if (len > DICT_HT_INITIAL_SIZE)
-            dictExpand(o->ptr,len);
+
+        /* Use a regular set when there are too many entries. */
+        if (len > server.set_max_intset_entries) {
+            o = createSetObject();
+            /* It's faster to expand the dict to the right size asap in order
+             * to avoid rehashing */
+            if (len > DICT_HT_INITIAL_SIZE)
+                dictExpand(o->ptr,len);
+        } else {
+            o = createIntsetObject();
+        }
+
         /* Load every single element of the list/set */
         while(len--) {
+            long long llval;
             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
             ele = tryObjectEncoding(ele);
-            dictAdd((dict*)o->ptr,ele,NULL);
+
+            if (o->encoding == REDIS_ENCODING_INTSET) {
+                /* Fetch integer value from element */
+                if (getLongLongFromObject(ele,&llval) == REDIS_OK) {
+                    o->ptr = intsetAdd(o->ptr,llval,NULL);
+                } else {
+                    setTypeConvert(o,REDIS_ENCODING_HT);
+                }
+            }
+
+            /* This will also be called when the set was just converted
+             * to regular hashtable encoded set */
+            if (o->encoding == REDIS_ENCODING_HT) {
+                dictAdd((dict*)o->ptr,ele,NULL);
+            }
         }
     } else if (type == REDIS_ZSET) {
         /* Read list/set value */
diff --git a/tests/unit/type/set.tcl b/tests/unit/type/set.tcl
index f9e7f4bd..a1b655ef 100644
--- a/tests/unit/type/set.tcl
+++ b/tests/unit/type/set.tcl
@@ -53,6 +53,21 @@ start_server {
         assert_encoding hashtable myset
     }
 
+    test "Set encoding after DEBUG RELOAD" {
+        r del myintset myhashset mylargeintset
+        for {set i 0} {$i <  100} {incr i} { r sadd myintset $i }
+        for {set i 0} {$i < 1280} {incr i} { r sadd mylargeintset $i }
+        for {set i 0} {$i <  256} {incr i} { r sadd myhashset [format "i%03d" $i] }
+        assert_encoding intset myintset
+        assert_encoding hashtable mylargeintset
+        assert_encoding hashtable myhashset
+
+        r debug reload
+        assert_encoding intset myintset
+        assert_encoding hashtable mylargeintset
+        assert_encoding hashtable myhashset
+    }
+
     test {SREM basics - regular set} {
         create_set myset {foo bar ciao}
         assert_encoding hashtable myset

From 400aea2b1343232e45a0b7a9435a69d0fe8a0956 Mon Sep 17 00:00:00 2001
From: Pieter Noordhuis <pcnoordhuis@gmail.com>
Date: Sun, 13 Jun 2010 21:52:07 +0200
Subject: [PATCH 10/10] expand the dictionary of the target set to the right
 size when converting from intset

---
 redis.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/redis.c b/redis.c
index 0d6648ee..28b1466c 100644
--- a/redis.c
+++ b/redis.c
@@ -4193,6 +4193,7 @@ static int rdbLoadDoubleValue(FILE *fp, double *val) {
 static robj *rdbLoadObject(int type, FILE *fp) {
     robj *o, *ele, *dec;
     size_t len;
+    unsigned int i;
 
     redisLog(REDIS_DEBUG,"LOADING OBJECT %d (at %d)\n",type,ftell(fp));
     if (type == REDIS_STRING) {
@@ -4247,7 +4248,7 @@ static robj *rdbLoadObject(int type, FILE *fp) {
         }
 
         /* Load every single element of the list/set */
-        while(len--) {
+        for (i = 0; i < len; i++) {
             long long llval;
             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
             ele = tryObjectEncoding(ele);
@@ -4258,6 +4259,7 @@ static robj *rdbLoadObject(int type, FILE *fp) {
                     o->ptr = intsetAdd(o->ptr,llval,NULL);
                 } else {
                     setTypeConvert(o,REDIS_ENCODING_HT);
+                    dictExpand(o->ptr,len);
                 }
             }
 
@@ -5668,6 +5670,9 @@ static unsigned long setTypeSize(robj *subject) {
     }
 }
 
+/* Convert the set to specified encoding. The resulting dict (when converting
+ * to a hashtable) is presized to hold the number of elements in the original
+ * set. */
 static void setTypeConvert(robj *subject, int enc) {
     setIterator *si;
     robj *element;
@@ -5675,6 +5680,8 @@ static void setTypeConvert(robj *subject, int enc) {
 
     if (enc == REDIS_ENCODING_HT) {
         dict *d = dictCreate(&setDictType,NULL);
+        /* Presize the dict to avoid rehashing */
+        dictExpand(d,intsetLen(subject->ptr));
 
         /* setTypeGet returns a robj with incremented refcount */
         si = setTypeInitIterator(subject);