merged intset code into the split files

src/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 release.o networking.o util.o object.o db.o replication.o rdb.o t_string.o t_list.o t_set.o t_zset.o t_hash.o config.o aof.o vm.o pubsub.o multi.o debug.o sort.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 release.o networking.o util.o object.o db.o replication.o rdb.o t_string.o t_list.o t_set.o t_zset.o t_hash.o config.o aof.o vm.o pubsub.o multi.o debug.o sort.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 @@ sds.o: sds.c sds.h zmalloc.h
 sha1.o: sha1.c sha1.h
 ziplist.o: ziplist.c zmalloc.h ziplist.h
 zipmap.o: zipmap.c zmalloc.h
+intset.o: intset.c zmalloc.h
 zmalloc.o: zmalloc.c config.h
 
 redis-server: $(OBJ)

src/config.c

@@ -199,6 +199,8 @@ 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;
         }

src/intset.c

+#include <stdio.h>
+#include <stdlib.h>
+#include <string.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 = -1;
+    int64_t cur = -1;
+
+    /* The value can never be found when the set is empty */
+    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) {
+        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 {
+        /* 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);
+        if (pos < is->length) intsetMoveTail(is,pos,pos+1);
+    }
+
+    INTSET_SET(is,pos,value);
+    is->length++;
+    return is;
+}
+
+/* Delete integer from intset */
+intset *intsetRemove(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);
+}
+
+/* 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>
+
+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 = intsetRemove(is,v2,NULL);
+            assert(!intsetFind(is,v2));
+        }
+        checkConsistency(is);
+        ok();
+    }
+}
+#endif

src/intset.h

+#ifndef __INTSET_H
+#define __INTSET_H
+#include <stdint.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 *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

src/object.c

@@ -73,7 +73,16 @@ robj *createZiplistObject(void) {
 
 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;
+}
+
+robj *createIntsetObject(void) {
+    intset *is = intsetNew();
+    robj *o = createObject(REDIS_SET,is);
+    o->encoding = REDIS_ENCODING_INTSET;
+    return o;
 }
 
 robj *createHashObject(void) {
@@ -114,7 +123,16 @@ void freeListObject(robj *o) {
 }
 
 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");
+    }
 }
 
 void freeZsetObject(robj *o) {
@@ -356,7 +374,7 @@ int getLongLongFromObject(robj *o, long long *target) {
         }
     }
 
-    *target = value;
+    if (target) *target = value;
     return REDIS_OK;
 }
 
@@ -400,6 +418,7 @@ char *strEncoding(int encoding) {
     case REDIS_ENCODING_ZIPMAP: return "zipmap";
     case REDIS_ENCODING_LINKEDLIST: return "linkedlist";
     case REDIS_ENCODING_ZIPLIST: return "ziplist";
+    case REDIS_ENCODING_INTSET: return "intset";
     default: return "unknown";
     }
 }

src/rdb.c

@@ -260,17 +260,29 @@ 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 (rdbSaveLen(fp,dictSize(set)) == -1) return -1;
-        while((de = dictNext(di)) != NULL) {
-            robj *eleobj = dictGetEntryKey(de);
+        if (o->encoding == REDIS_ENCODING_HT) {
+            dict *set = o->ptr;
+            dictIterator *di = dictGetIterator(set);
+            dictEntry *de;
 
-            if (rdbSaveStringObject(fp,eleobj) == -1) return -1;
+            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 (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;
@@ -628,6 +640,7 @@ int rdbLoadDoubleValue(FILE *fp, double *val) {
 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) {
@@ -669,16 +682,39 @@ 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--) {
+        for (i = 0; i < len; i++) {
+            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);
+                    dictExpand(o->ptr,len);
+                }
+            }
+
+            /* 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 */

src/redis.c

@@ -731,6 +731,7 @@ 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();

src/redis.h

@@ -25,6 +25,7 @@
 #include "anet.h"   /* Networking the easy way */
 #include "zipmap.h" /* Compact string -> string data structure */
 #include "ziplist.h" /* Compact list data structure */
+#include "intset.h" /* Compact integer set structure */
 #include "version.h"
 
 /* Error codes */
@@ -81,6 +82,7 @@
 #define REDIS_ENCODING_ZIPMAP 3  /* Encoded as zipmap */
 #define REDIS_ENCODING_LINKEDLIST 4 /* Encoded as regular linked list */
 #define REDIS_ENCODING_ZIPLIST 5 /* Encoded as ziplist */
+#define REDIS_ENCODING_INTSET 6  /* Encoded as intset */
 
 /* Object types only used for dumping to disk */
 #define REDIS_EXPIRETIME 253
@@ -187,6 +189,7 @@
 #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
 
 /* Sets operations codes */
 #define REDIS_OP_UNION 0
@@ -396,6 +399,7 @@ 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;
@@ -533,6 +537,14 @@ typedef struct {
     listNode *ln;       /* Entry in linked list */
 } listTypeEntry;
 
+/* Structure to hold set iteration abstraction. */
+typedef struct {
+    robj *subject;
+    int encoding;
+    int ii; /* intset iterator */
+    dictIterator *di;
+} setIterator;
+
 /* Structure to hold hash iteration abstration. Note that iteration over
  * hashes involves both fields and values. Because it is possible that
  * not both are required, store pointers in the iterator to avoid
@@ -631,6 +643,7 @@ robj *createStringObjectFromLongLong(long long value);
 robj *createListObject(void);
 robj *createZiplistObject(void);
 robj *createSetObject(void);
+robj *createIntsetObject(void);
 robj *createHashObject(void);
 robj *createZsetObject(void);
 int getLongFromObjectOrReply(redisClient *c, robj *o, long *target, const char *msg);
@@ -714,6 +727,18 @@ int dontWaitForSwappedKey(redisClient *c, robj *key);
 void handleClientsBlockedOnSwappedKey(redisDb *db, robj *key);
 vmpointer *vmSwapObjectBlocking(robj *val);
 
+/* Set data type */
+robj *setTypeCreate(robj *value);
+int setTypeAdd(robj *subject, robj *value);
+int setTypeRemove(robj *subject, robj *value);
+int setTypeIsMember(robj *subject, robj *value);
+setIterator *setTypeInitIterator(robj *subject);
+void setTypeReleaseIterator(setIterator *si);
+robj *setTypeNext(setIterator *si);
+robj *setTypeRandomElement(robj *subject);
+unsigned long setTypeSize(robj *subject);
+void setTypeConvert(robj *subject, int enc);
+
 /* Hash data type */
 void convertToRealHash(robj *o);
 void hashTypeTryConversion(robj *subject, robj **argv, int start, int end);

tests/unit/type/set.tcl

-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}}
-
-    test {SADD adding the same element multiple times} {
-        r sadd myset foo
-        r sadd myset foo
-        r sadd myset foo
-        r scard myset
-    } {2}
+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 }
+    }
+
+    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 $err
-    } {ERR*kind*}
-
-    test {SREM basics} {
-        r sadd myset ciao
-        r srem myset foo
-        lsort [r smembers myset]
-    } {bar ciao}
-
-    test {Mass SADD and SINTER with two sets} {
-        for {set i 0} {$i < 1000} {incr i} {
+        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 $i }
+        assert_encoding intset myset
+        assert_equal 1 [r sadd myset 512]
+        assert_encoding hashtable myset
+    }
+
+    test "Set encoding after DEBUG RELOAD" {
+        r del myintset myhashset mylargeintset
+        for {set i 0} {$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" $i] }
+        assert_encoding intset myintset
+        assert_encoding hashtable mylargeintset
+        assert_encoding hashtable myhashset
+
+        r debug reload
+        assert_encoding intset myintset
+        assert_encoding hashtable mylargeintset
+        assert_encoding hashtable myhashset
+    }
+
+    test {SREM basics - regular set} {
+        create_set myset {foo bar ciao}
+        assert_encoding hashtable myset
+        assert_equal 0 [r srem myset qux]
+        assert_equal 1 [r srem myset foo]
+        assert_equal {bar ciao} [lsort [r smembers myset]]
+    }
+
+    test {SREM basics - intset} {
+        create_set myset {3 4 5}
+        assert_encoding intset myset
+        assert_equal 0 [r srem myset 6]
+        assert_equal 1 [r srem myset 4]
+        assert_equal {3 5} [lsort [r smembers myset]]
+    }
+
+    foreach {type} {hashtable intset} {
+        for {set i 1} {$i <= 5} {incr i} {
+            r del [format "set%d" $i]
+        }
+        for {set i 0} {$i < 200} {incr i} {
             r sadd set1 $i
-            r sadd set2 [expr $i+995]
+            r sadd set2 [expr $i+195]
+        }
+        foreach i {199 195 1000 2000} {
+            r sadd set3 $i
+        }
+        for {set i 5} {$i < 200} {incr i} {
+            r sadd set4 $i
         }
-        lsort [r sinter set1 set2]
-    } {995 996 997 998 999}
+        r sadd set5 0
 
-    test {SUNION with two sets} {
-        lsort [r sunion set1 set2]
-    } [lsort -uniq "[r smembers set1] [r smembers set2]"]
+        # 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
+            }
+        }
 
-    test {SINTERSTORE with two sets} {
-        r sinterstore setres set1 set2
-        lsort [r smembers setres]
-    } {995 996 997 998 999}
+        test "Generated sets must be encoded as $type" {
+            for {set i 1} {$i <= 5} {incr i} {
+                assert_encoding $type [format "set%d" $i]
+            }
+        }
 
-    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 "SINTER with two sets - $type" {
+            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 {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 {195 196 197 198 199} [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 {SUNIONSTORE with two sets} {
-        r sunionstore setres set1 set2
-        lsort [r smembers setres]
-    } [lsort -uniq "[r smembers set1] [r smembers set2]"]
+        test "SINTER against three sets - $type" {
+            assert_equal {195 199} [lsort [r sinter set1 set2 set3]]
+        }
+
+        test "SINTERSTORE with three sets - $type" {
+            r sinterstore setres set1 set2 set3
+            assert_equal {195 199} [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 {SUNIONSTORE against non existing keys} {
+        test "SDIFF with three sets - $type" {
+            assert_equal {1 2 3 4} [lsort [r sdiff set1 set4 set5]]
+        }
+
+        test "SDIFFSTORE with three sets - $type" {
+            r sdiffstore setres set1 set4 set5
+            assert_encoding intset setres
+            assert_equal {1 2 3 4} [lsort [r smembers setres]]
+        }
+    }
+
+    test "SINTER against non-set should throw error" {
+        r set key1 x
+        assert_error "ERR*wrong kind*" {r sinter key1 noset}
+    }
+
+    test "SUNION against non-set should throw error" {
+        r set key1 x
+        assert_error "ERR*wrong kind*" {r sunion key1 noset}
+    }
+
+    test "SINTERSTORE against non existing keys should delete dstkey" {
         r set setres xxx
-        list [r sunionstore setres foo111 bar222] [r exists xxx]
-    } {0 0}
-
-    test {SINTER against three sets} {
-        r sadd set3 999
-        r sadd set3 995
-        r sadd set3 1000
-        r sadd set3 2000
-        lsort [r sinter set1 set2 set3]
-    } {995 999}
-
-    test {SINTERSTORE with three sets} {
-        r sinterstore setres set1 set2 set3
-        lsort [r smembers setres]
-    } {995 999}
-
-    test {SUNION with non existing keys} {
-        lsort [r sunion nokey1 set1 set2 nokey2]
-    } [lsort -uniq "[r smembers set1] [r smembers set2]"]
-
-    test {SDIFF with two sets} {
-        for {set i 5} {$i < 1000} {incr i} {
-            r sadd set4 $i
+        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]
         }
-        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 "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 {SDIFFSTORE with three sets} {
-        r sdiffstore sres set1 set4 set5
-        lsort [r smembers sres]
-    } {1 2 3 4}
+    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 {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
-        }
-        lsort [array names myset]
-    } {a b c}
-    
-    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}}
-
-    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 non existing src set} {
-        list [r smove noset myset2 foo] [lsort [r smembers myset2]]
-    } {0 {a x y z}}
-
-    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 wrong src key type} {
+    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
+
+        # 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 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 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}
+    }
 }