diff options
| -rw-r--r-- | caching-tests/test.ur | 17 | ||||
| -rw-r--r-- | include/urweb/types_cpp.h | 31 | ||||
| -rw-r--r-- | include/urweb/urweb_cpp.h | 8 | ||||
| -rw-r--r-- | include/urweb/uthash.h | 963 | ||||
| -rw-r--r-- | src/c/urweb.c | 147 | ||||
| -rw-r--r-- | src/lru_cache.sml | 171 | ||||
| -rw-r--r-- | src/sources | 1 | ||||
| -rw-r--r-- | src/sqlcache.sml | 115 |
8 files changed, 1397 insertions, 56 deletions
diff --git a/caching-tests/test.ur b/caching-tests/test.ur index 8035e336..842fd77d 100644 --- a/caching-tests/test.ur +++ b/caching-tests/test.ur @@ -16,13 +16,24 @@ fun flush id = FROM tab WHERE tab.Id = {[id]}); (case res of - None => dml (INSERT INTO tab (Id, Val) - VALUES ({[id]}, 0)) + None => return () (* dml (INSERT INTO tab (Id, Val) *) + (* VALUES ({[id]}, 0)) *) | Some row => dml (UPDATE tab SET Val = {[row.Tab.Val + 1]} - WHERE Id = {[id]})); + WHERE Id = {[id + 1]} OR Id = {[id]} (* OR Id = {[id - 1]} *))); return <xml><body> {case res of None => <xml>Initialized {[id]}!</xml> | Some row => <xml>Incremented {[id]}!</xml>} </body></xml> + +(* task periodic 5 = *) +(* fn () => *) +(* t <- now; *) +(* let *) +(* val n = toSeconds t % 2 *) +(* in *) +(* dml (UPDATE tab *) +(* SET Val = 9001 *) +(* WHERE Id = {[n]} OR Id = {[n+1]}) *) +(* end *) diff --git a/include/urweb/types_cpp.h b/include/urweb/types_cpp.h index 0c431ff8..2f154e1f 100644 --- a/include/urweb/types_cpp.h +++ b/include/urweb/types_cpp.h @@ -119,4 +119,35 @@ typedef struct { char *start, *front, *back; } uw_buffer; +// Caching + +#include "uthash.h" + +typedef struct CacheValue { + char *result; + char *output; +} CacheValue; + +typedef struct CacheEntry { + char *key; + void *value; + time_t timeValid; + struct CacheEntry *prev; + struct CacheEntry *next; + UT_hash_handle hh; +} CacheEntry; + +typedef struct CacheList { + CacheEntry *first; + CacheEntry *last; + int size; +} CacheList; + +typedef struct Cache { + CacheEntry *table; + time_t timeInvalid; + CacheList *lru; + int height; +} Cache; + #endif diff --git a/include/urweb/urweb_cpp.h b/include/urweb/urweb_cpp.h index a9d42554..3ae5b69e 100644 --- a/include/urweb/urweb_cpp.h +++ b/include/urweb/urweb_cpp.h @@ -402,4 +402,12 @@ void uw_set_remoteSock(struct uw_context *, int sock); void uw_Basis_writec(struct uw_context *, char); +// Sqlcache. + +#include "uthash.h" + +CacheValue *check(Cache *, char **); +CacheValue *store(Cache *, char **, CacheValue *); +CacheValue *flush(Cache *, char **); + #endif diff --git a/include/urweb/uthash.h b/include/urweb/uthash.h new file mode 100644 index 00000000..367d295a --- /dev/null +++ b/include/urweb/uthash.h @@ -0,0 +1,963 @@ +/* +Copyright (c) 2003-2014, Troy D. Hanson http://troydhanson.github.com/uthash/ +All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS +IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED +TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A +PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER +OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, +EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, +PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR +PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF +LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING +NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef UTHASH_H +#define UTHASH_H + +#include <string.h> /* memcmp,strlen */ +#include <stddef.h> /* ptrdiff_t */ +#include <stdlib.h> /* exit() */ + +/* These macros use decltype or the earlier __typeof GNU extension. + As decltype is only available in newer compilers (VS2010 or gcc 4.3+ + when compiling c++ source) this code uses whatever method is needed + or, for VS2008 where neither is available, uses casting workarounds. */ +#if defined(_MSC_VER) /* MS compiler */ +#if _MSC_VER >= 1600 && defined(__cplusplus) /* VS2010 or newer in C++ mode */ +#define DECLTYPE(x) (decltype(x)) +#else /* VS2008 or older (or VS2010 in C mode) */ +#define NO_DECLTYPE +#define DECLTYPE(x) +#endif +#elif defined(__BORLANDC__) || defined(__LCC__) || defined(__WATCOMC__) +#define NO_DECLTYPE +#define DECLTYPE(x) +#else /* GNU, Sun and other compilers */ +#define DECLTYPE(x) (__typeof(x)) +#endif + +#ifdef NO_DECLTYPE +#define DECLTYPE_ASSIGN(dst,src) \ +do { \ + char **_da_dst = (char**)(&(dst)); \ + *_da_dst = (char*)(src); \ +} while(0) +#else +#define DECLTYPE_ASSIGN(dst,src) \ +do { \ + (dst) = DECLTYPE(dst)(src); \ +} while(0) +#endif + +/* a number of the hash function use uint32_t which isn't defined on Pre VS2010 */ +#if defined (_WIN32) +#if defined(_MSC_VER) && _MSC_VER >= 1600 +#include <stdint.h> +#elif defined(__WATCOMC__) +#include <stdint.h> +#else +typedef unsigned int uint32_t; +typedef unsigned char uint8_t; +#endif +#else +#include <stdint.h> +#endif + +#define UTHASH_VERSION 1.9.9 + +#ifndef uthash_fatal +#define uthash_fatal(msg) exit(-1) /* fatal error (out of memory,etc) */ +#endif +#ifndef uthash_malloc +#define uthash_malloc(sz) malloc(sz) /* malloc fcn */ +#endif +#ifndef uthash_free +#define uthash_free(ptr,sz) free(ptr) /* free fcn */ +#endif + +#ifndef uthash_noexpand_fyi +#define uthash_noexpand_fyi(tbl) /* can be defined to log noexpand */ +#endif +#ifndef uthash_expand_fyi +#define uthash_expand_fyi(tbl) /* can be defined to log expands */ +#endif + +/* initial number of buckets */ +#define HASH_INITIAL_NUM_BUCKETS 32U /* initial number of buckets */ +#define HASH_INITIAL_NUM_BUCKETS_LOG2 5U /* lg2 of initial number of buckets */ +#define HASH_BKT_CAPACITY_THRESH 10U /* expand when bucket count reaches */ + +/* calculate the element whose hash handle address is hhe */ +#define ELMT_FROM_HH(tbl,hhp) ((void*)(((char*)(hhp)) - ((tbl)->hho))) + +#define HASH_FIND(hh,head,keyptr,keylen,out) \ +do { \ + out=NULL; \ + if (head != NULL) { \ + unsigned _hf_bkt,_hf_hashv; \ + HASH_FCN(keyptr,keylen, (head)->hh.tbl->num_buckets, _hf_hashv, _hf_bkt); \ + if (HASH_BLOOM_TEST((head)->hh.tbl, _hf_hashv) != 0) { \ + HASH_FIND_IN_BKT((head)->hh.tbl, hh, (head)->hh.tbl->buckets[ _hf_bkt ], \ + keyptr,keylen,out); \ + } \ + } \ +} while (0) + +#ifdef HASH_BLOOM +#define HASH_BLOOM_BITLEN (1UL << HASH_BLOOM) +#define HASH_BLOOM_BYTELEN (HASH_BLOOM_BITLEN/8UL) + (((HASH_BLOOM_BITLEN%8UL)!=0UL) ? 1UL : 0UL) +#define HASH_BLOOM_MAKE(tbl) \ +do { \ + (tbl)->bloom_nbits = HASH_BLOOM; \ + (tbl)->bloom_bv = (uint8_t*)uthash_malloc(HASH_BLOOM_BYTELEN); \ + if (!((tbl)->bloom_bv)) { uthash_fatal( "out of memory"); } \ + memset((tbl)->bloom_bv, 0, HASH_BLOOM_BYTELEN); \ + (tbl)->bloom_sig = HASH_BLOOM_SIGNATURE; \ +} while (0) + +#define HASH_BLOOM_FREE(tbl) \ +do { \ + uthash_free((tbl)->bloom_bv, HASH_BLOOM_BYTELEN); \ +} while (0) + +#define HASH_BLOOM_BITSET(bv,idx) (bv[(idx)/8U] |= (1U << ((idx)%8U))) +#define HASH_BLOOM_BITTEST(bv,idx) (bv[(idx)/8U] & (1U << ((idx)%8U))) + +#define HASH_BLOOM_ADD(tbl,hashv) \ + HASH_BLOOM_BITSET((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1U))) + +#define HASH_BLOOM_TEST(tbl,hashv) \ + HASH_BLOOM_BITTEST((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1U))) + +#else +#define HASH_BLOOM_MAKE(tbl) +#define HASH_BLOOM_FREE(tbl) +#define HASH_BLOOM_ADD(tbl,hashv) +#define HASH_BLOOM_TEST(tbl,hashv) (1) +#define HASH_BLOOM_BYTELEN 0U +#endif + +#define HASH_MAKE_TABLE(hh,head) \ +do { \ + (head)->hh.tbl = (UT_hash_table*)uthash_malloc( \ + sizeof(UT_hash_table)); \ + if (!((head)->hh.tbl)) { uthash_fatal( "out of memory"); } \ + memset((head)->hh.tbl, 0, sizeof(UT_hash_table)); \ + (head)->hh.tbl->tail = &((head)->hh); \ + (head)->hh.tbl->num_buckets = HASH_INITIAL_NUM_BUCKETS; \ + (head)->hh.tbl->log2_num_buckets = HASH_INITIAL_NUM_BUCKETS_LOG2; \ + (head)->hh.tbl->hho = (char*)(&(head)->hh) - (char*)(head); \ + (head)->hh.tbl->buckets = (UT_hash_bucket*)uthash_malloc( \ + HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \ + if (! (head)->hh.tbl->buckets) { uthash_fatal( "out of memory"); } \ + memset((head)->hh.tbl->buckets, 0, \ + HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \ + HASH_BLOOM_MAKE((head)->hh.tbl); \ + (head)->hh.tbl->signature = HASH_SIGNATURE; \ +} while(0) + +#define HASH_ADD(hh,head,fieldname,keylen_in,add) \ + HASH_ADD_KEYPTR(hh,head,&((add)->fieldname),keylen_in,add) + +#define HASH_REPLACE(hh,head,fieldname,keylen_in,add,replaced) \ +do { \ + replaced=NULL; \ + HASH_FIND(hh,head,&((add)->fieldname),keylen_in,replaced); \ + if (replaced!=NULL) { \ + HASH_DELETE(hh,head,replaced); \ + } \ + HASH_ADD(hh,head,fieldname,keylen_in,add); \ +} while(0) + +#define HASH_ADD_KEYPTR(hh,head,keyptr,keylen_in,add) \ +do { \ + unsigned _ha_bkt; \ + (add)->hh.next = NULL; \ + (add)->hh.key = (char*)(keyptr); \ + (add)->hh.keylen = (unsigned)(keylen_in); \ + if (!(head)) { \ + head = (add); \ + (head)->hh.prev = NULL; \ + HASH_MAKE_TABLE(hh,head); \ + } else { \ + (head)->hh.tbl->tail->next = (add); \ + (add)->hh.prev = ELMT_FROM_HH((head)->hh.tbl, (head)->hh.tbl->tail); \ + (head)->hh.tbl->tail = &((add)->hh); \ + } \ + (head)->hh.tbl->num_items++; \ + (add)->hh.tbl = (head)->hh.tbl; \ + HASH_FCN(keyptr,keylen_in, (head)->hh.tbl->num_buckets, \ + (add)->hh.hashv, _ha_bkt); \ + HASH_ADD_TO_BKT((head)->hh.tbl->buckets[_ha_bkt],&(add)->hh); \ + HASH_BLOOM_ADD((head)->hh.tbl,(add)->hh.hashv); \ + HASH_EMIT_KEY(hh,head,keyptr,keylen_in); \ + HASH_FSCK(hh,head); \ +} while(0) + +#define HASH_TO_BKT( hashv, num_bkts, bkt ) \ +do { \ + bkt = ((hashv) & ((num_bkts) - 1U)); \ +} while(0) + +/* delete "delptr" from the hash table. + * "the usual" patch-up process for the app-order doubly-linked-list. + * The use of _hd_hh_del below deserves special explanation. + * These used to be expressed using (delptr) but that led to a bug + * if someone used the same symbol for the head and deletee, like + * HASH_DELETE(hh,users,users); + * We want that to work, but by changing the head (users) below + * we were forfeiting our ability to further refer to the deletee (users) + * in the patch-up process. Solution: use scratch space to + * copy the deletee pointer, then the latter references are via that + * scratch pointer rather than through the repointed (users) symbol. + */ +#define HASH_DELETE(hh,head,delptr) \ +do { \ + struct UT_hash_handle *_hd_hh_del; \ + if ( ((delptr)->hh.prev == NULL) && ((delptr)->hh.next == NULL) ) { \ + uthash_free((head)->hh.tbl->buckets, \ + (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \ + HASH_BLOOM_FREE((head)->hh.tbl); \ + uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \ + head = NULL; \ + } else { \ + unsigned _hd_bkt; \ + _hd_hh_del = &((delptr)->hh); \ + if ((delptr) == ELMT_FROM_HH((head)->hh.tbl,(head)->hh.tbl->tail)) { \ + (head)->hh.tbl->tail = \ + (UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \ + (head)->hh.tbl->hho); \ + } \ + if ((delptr)->hh.prev != NULL) { \ + ((UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \ + (head)->hh.tbl->hho))->next = (delptr)->hh.next; \ + } else { \ + DECLTYPE_ASSIGN(head,(delptr)->hh.next); \ + } \ + if (_hd_hh_del->next != NULL) { \ + ((UT_hash_handle*)((ptrdiff_t)_hd_hh_del->next + \ + (head)->hh.tbl->hho))->prev = \ + _hd_hh_del->prev; \ + } \ + HASH_TO_BKT( _hd_hh_del->hashv, (head)->hh.tbl->num_buckets, _hd_bkt); \ + HASH_DEL_IN_BKT(hh,(head)->hh.tbl->buckets[_hd_bkt], _hd_hh_del); \ + (head)->hh.tbl->num_items--; \ + } \ + HASH_FSCK(hh,head); \ +} while (0) + + +/* convenience forms of HASH_FIND/HASH_ADD/HASH_DEL */ +#define HASH_FIND_STR(head,findstr,out) \ + HASH_FIND(hh,head,findstr,(unsigned)strlen(findstr),out) +#define HASH_ADD_STR(head,strfield,add) \ + HASH_ADD(hh,head,strfield[0],(unsigned int)strlen(add->strfield),add) +#define HASH_REPLACE_STR(head,strfield,add,replaced) \ + HASH_REPLACE(hh,head,strfield[0],(unsigned)strlen(add->strfield),add,replaced) +#define HASH_FIND_INT(head,findint,out) \ + HASH_FIND(hh,head,findint,sizeof(int),out) +#define HASH_ADD_INT(head,intfield,add) \ + HASH_ADD(hh,head,intfield,sizeof(int),add) +#define HASH_REPLACE_INT(head,intfield,add,replaced) \ + HASH_REPLACE(hh,head,intfield,sizeof(int),add,replaced) +#define HASH_FIND_PTR(head,findptr,out) \ + HASH_FIND(hh,head,findptr,sizeof(void *),out) +#define HASH_ADD_PTR(head,ptrfield,add) \ + HASH_ADD(hh,head,ptrfield,sizeof(void *),add) +#define HASH_REPLACE_PTR(head,ptrfield,add,replaced) \ + HASH_REPLACE(hh,head,ptrfield,sizeof(void *),add,replaced) +#define HASH_DEL(head,delptr) \ + HASH_DELETE(hh,head,delptr) + +/* HASH_FSCK checks hash integrity on every add/delete when HASH_DEBUG is defined. + * This is for uthash developer only; it compiles away if HASH_DEBUG isn't defined. + */ +#ifdef HASH_DEBUG +#define HASH_OOPS(...) do { fprintf(stderr,__VA_ARGS__); exit(-1); } while (0) +#define HASH_FSCK(hh,head) \ +do { \ + struct UT_hash_handle *_thh; \ + if (head) { \ + unsigned _bkt_i; \ + unsigned _count; \ + char *_prev; \ + _count = 0; \ + for( _bkt_i = 0; _bkt_i < (head)->hh.tbl->num_buckets; _bkt_i++) { \ + unsigned _bkt_count = 0; \ + _thh = (head)->hh.tbl->buckets[_bkt_i].hh_head; \ + _prev = NULL; \ + while (_thh) { \ + if (_prev != (char*)(_thh->hh_prev)) { \ + HASH_OOPS("invalid hh_prev %p, actual %p\n", \ + _thh->hh_prev, _prev ); \ + } \ + _bkt_count++; \ + _prev = (char*)(_thh); \ + _thh = _thh->hh_next; \ + } \ + _count += _bkt_count; \ + if ((head)->hh.tbl->buckets[_bkt_i].count != _bkt_count) { \ + HASH_OOPS("invalid bucket count %u, actual %u\n", \ + (head)->hh.tbl->buckets[_bkt_i].count, _bkt_count); \ + } \ + } \ + if (_count != (head)->hh.tbl->num_items) { \ + HASH_OOPS("invalid hh item count %u, actual %u\n", \ + (head)->hh.tbl->num_items, _count ); \ + } \ + /* traverse hh in app order; check next/prev integrity, count */ \ + _count = 0; \ + _prev = NULL; \ + _thh = &(head)->hh; \ + while (_thh) { \ + _count++; \ + if (_prev !=(char*)(_thh->prev)) { \ + HASH_OOPS("invalid prev %p, actual %p\n", \ + _thh->prev, _prev ); \ + } \ + _prev = (char*)ELMT_FROM_HH((head)->hh.tbl, _thh); \ + _thh = ( _thh->next ? (UT_hash_handle*)((char*)(_thh->next) + \ + (head)->hh.tbl->hho) : NULL ); \ + } \ + if (_count != (head)->hh.tbl->num_items) { \ + HASH_OOPS("invalid app item count %u, actual %u\n", \ + (head)->hh.tbl->num_items, _count ); \ + } \ + } \ +} while (0) +#else +#define HASH_FSCK(hh,head) +#endif + +/* When compiled with -DHASH_EMIT_KEYS, length-prefixed keys are emitted to + * the descriptor to which this macro is defined for tuning the hash function. + * The app can #include <unistd.h> to get the prototype for write(2). */ +#ifdef HASH_EMIT_KEYS +#define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) \ +do { \ + unsigned _klen = fieldlen; \ + write(HASH_EMIT_KEYS, &_klen, sizeof(_klen)); \ + write(HASH_EMIT_KEYS, keyptr, (unsigned long)fieldlen); \ +} while (0) +#else +#define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) +#endif + +/* default to Jenkin's hash unless overridden e.g. DHASH_FUNCTION=HASH_SAX */ +#ifdef HASH_FUNCTION +#define HASH_FCN HASH_FUNCTION +#else +#define HASH_FCN HASH_JEN +#endif + +/* The Bernstein hash function, used in Perl prior to v5.6. Note (x<<5+x)=x*33. */ +#define HASH_BER(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned _hb_keylen=(unsigned)keylen; \ + const unsigned char *_hb_key=(const unsigned char*)(key); \ + (hashv) = 0; \ + while (_hb_keylen-- != 0U) { \ + (hashv) = (((hashv) << 5) + (hashv)) + *_hb_key++; \ + } \ + bkt = (hashv) & (num_bkts-1U); \ +} while (0) + + +/* SAX/FNV/OAT/JEN hash functions are macro variants of those listed at + * http://eternallyconfuzzled.com/tuts/algorithms/jsw_tut_hashing.aspx */ +#define HASH_SAX(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned _sx_i; \ + const unsigned char *_hs_key=(const unsigned char*)(key); \ + hashv = 0; \ + for(_sx_i=0; _sx_i < keylen; _sx_i++) { \ + hashv ^= (hashv << 5) + (hashv >> 2) + _hs_key[_sx_i]; \ + } \ + bkt = hashv & (num_bkts-1U); \ +} while (0) +/* FNV-1a variation */ +#define HASH_FNV(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned _fn_i; \ + const unsigned char *_hf_key=(const unsigned char*)(key); \ + hashv = 2166136261U; \ + for(_fn_i=0; _fn_i < keylen; _fn_i++) { \ + hashv = hashv ^ _hf_key[_fn_i]; \ + hashv = hashv * 16777619U; \ + } \ + bkt = hashv & (num_bkts-1U); \ +} while(0) + +#define HASH_OAT(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned _ho_i; \ + const unsigned char *_ho_key=(const unsigned char*)(key); \ + hashv = 0; \ + for(_ho_i=0; _ho_i < keylen; _ho_i++) { \ + hashv += _ho_key[_ho_i]; \ + hashv += (hashv << 10); \ + hashv ^= (hashv >> 6); \ + } \ + hashv += (hashv << 3); \ + hashv ^= (hashv >> 11); \ + hashv += (hashv << 15); \ + bkt = hashv & (num_bkts-1U); \ +} while(0) + +#define HASH_JEN_MIX(a,b,c) \ +do { \ + a -= b; a -= c; a ^= ( c >> 13 ); \ + b -= c; b -= a; b ^= ( a << 8 ); \ + c -= a; c -= b; c ^= ( b >> 13 ); \ + a -= b; a -= c; a ^= ( c >> 12 ); \ + b -= c; b -= a; b ^= ( a << 16 ); \ + c -= a; c -= b; c ^= ( b >> 5 ); \ + a -= b; a -= c; a ^= ( c >> 3 ); \ + b -= c; b -= a; b ^= ( a << 10 ); \ + c -= a; c -= b; c ^= ( b >> 15 ); \ +} while (0) + +#define HASH_JEN(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned _hj_i,_hj_j,_hj_k; \ + unsigned const char *_hj_key=(unsigned const char*)(key); \ + hashv = 0xfeedbeefu; \ + _hj_i = _hj_j = 0x9e3779b9u; \ + _hj_k = (unsigned)(keylen); \ + while (_hj_k >= 12U) { \ + _hj_i += (_hj_key[0] + ( (unsigned)_hj_key[1] << 8 ) \ + + ( (unsigned)_hj_key[2] << 16 ) \ + + ( (unsigned)_hj_key[3] << 24 ) ); \ + _hj_j += (_hj_key[4] + ( (unsigned)_hj_key[5] << 8 ) \ + + ( (unsigned)_hj_key[6] << 16 ) \ + + ( (unsigned)_hj_key[7] << 24 ) ); \ + hashv += (_hj_key[8] + ( (unsigned)_hj_key[9] << 8 ) \ + + ( (unsigned)_hj_key[10] << 16 ) \ + + ( (unsigned)_hj_key[11] << 24 ) ); \ + \ + HASH_JEN_MIX(_hj_i, _hj_j, hashv); \ + \ + _hj_key += 12; \ + _hj_k -= 12U; \ + } \ + hashv += (unsigned)(keylen); \ + switch ( _hj_k ) { \ + case 11: hashv += ( (unsigned)_hj_key[10] << 24 ); /* FALLTHROUGH */ \ + case 10: hashv += ( (unsigned)_hj_key[9] << 16 ); /* FALLTHROUGH */ \ + case 9: hashv += ( (unsigned)_hj_key[8] << 8 ); /* FALLTHROUGH */ \ + case 8: _hj_j += ( (unsigned)_hj_key[7] << 24 ); /* FALLTHROUGH */ \ + case 7: _hj_j += ( (unsigned)_hj_key[6] << 16 ); /* FALLTHROUGH */ \ + case 6: _hj_j += ( (unsigned)_hj_key[5] << 8 ); /* FALLTHROUGH */ \ + case 5: _hj_j += _hj_key[4]; /* FALLTHROUGH */ \ + case 4: _hj_i += ( (unsigned)_hj_key[3] << 24 ); /* FALLTHROUGH */ \ + case 3: _hj_i += ( (unsigned)_hj_key[2] << 16 ); /* FALLTHROUGH */ \ + case 2: _hj_i += ( (unsigned)_hj_key[1] << 8 ); /* FALLTHROUGH */ \ + case 1: _hj_i += _hj_key[0]; \ + } \ + HASH_JEN_MIX(_hj_i, _hj_j, hashv); \ + bkt = hashv & (num_bkts-1U); \ +} while(0) + +/* The Paul Hsieh hash function */ +#undef get16bits +#if (defined(__GNUC__) && defined(__i386__)) || defined(__WATCOMC__) \ + || defined(_MSC_VER) || defined (__BORLANDC__) || defined (__TURBOC__) +#define get16bits(d) (*((const uint16_t *) (d))) +#endif + +#if !defined (get16bits) +#define get16bits(d) ((((uint32_t)(((const uint8_t *)(d))[1])) << 8) \ + +(uint32_t)(((const uint8_t *)(d))[0]) ) +#endif +#define HASH_SFH(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned const char *_sfh_key=(unsigned const char*)(key); \ + uint32_t _sfh_tmp, _sfh_len = (uint32_t)keylen; \ + \ + unsigned _sfh_rem = _sfh_len & 3U; \ + _sfh_len >>= 2; \ + hashv = 0xcafebabeu; \ + \ + /* Main loop */ \ + for (;_sfh_len > 0U; _sfh_len--) { \ + hashv += get16bits (_sfh_key); \ + _sfh_tmp = ((uint32_t)(get16bits (_sfh_key+2)) << 11) ^ hashv; \ + hashv = (hashv << 16) ^ _sfh_tmp; \ + _sfh_key += 2U*sizeof (uint16_t); \ + hashv += hashv >> 11; \ + } \ + \ + /* Handle end cases */ \ + switch (_sfh_rem) { \ + case 3: hashv += get16bits (_sfh_key); \ + hashv ^= hashv << 16; \ + hashv ^= (uint32_t)(_sfh_key[sizeof (uint16_t)]) << 18; \ + hashv += hashv >> 11; \ + break; \ + case 2: hashv += get16bits (_sfh_key); \ + hashv ^= hashv << 11; \ + hashv += hashv >> 17; \ + break; \ + case 1: hashv += *_sfh_key; \ + hashv ^= hashv << 10; \ + hashv += hashv >> 1; \ + } \ + \ + /* Force "avalanching" of final 127 bits */ \ + hashv ^= hashv << 3; \ + hashv += hashv >> 5; \ + hashv ^= hashv << 4; \ + hashv += hashv >> 17; \ + hashv ^= hashv << 25; \ + hashv += hashv >> 6; \ + bkt = hashv & (num_bkts-1U); \ +} while(0) + +#ifdef HASH_USING_NO_STRICT_ALIASING +/* The MurmurHash exploits some CPU's (x86,x86_64) tolerance for unaligned reads. + * For other types of CPU's (e.g. Sparc) an unaligned read causes a bus error. + * MurmurHash uses the faster approach only on CPU's where we know it's safe. + * + * Note the preprocessor built-in defines can be emitted using: + * + * gcc -m64 -dM -E - < /dev/null (on gcc) + * cc -## a.c (where a.c is a simple test file) (Sun Studio) + */ +#if (defined(__i386__) || defined(__x86_64__) || defined(_M_IX86)) +#define MUR_GETBLOCK(p,i) p[i] +#else /* non intel */ +#define MUR_PLUS0_ALIGNED(p) (((unsigned long)p & 3UL) == 0UL) +#define MUR_PLUS1_ALIGNED(p) (((unsigned long)p & 3UL) == 1UL) +#define MUR_PLUS2_ALIGNED(p) (((unsigned long)p & 3UL) == 2UL) +#define MUR_PLUS3_ALIGNED(p) (((unsigned long)p & 3UL) == 3UL) +#define WP(p) ((uint32_t*)((unsigned long)(p) & ~3UL)) +#if (defined(__BIG_ENDIAN__) || defined(SPARC) || defined(__ppc__) || defined(__ppc64__)) +#define MUR_THREE_ONE(p) ((((*WP(p))&0x00ffffff) << 8) | (((*(WP(p)+1))&0xff000000) >> 24)) +#define MUR_TWO_TWO(p) ((((*WP(p))&0x0000ffff) <<16) | (((*(WP(p)+1))&0xffff0000) >> 16)) +#define MUR_ONE_THREE(p) ((((*WP(p))&0x000000ff) <<24) | (((*(WP(p)+1))&0xffffff00) >> 8)) +#else /* assume little endian non-intel */ +#define MUR_THREE_ONE(p) ((((*WP(p))&0xffffff00) >> 8) | (((*(WP(p)+1))&0x000000ff) << 24)) +#define MUR_TWO_TWO(p) ((((*WP(p))&0xffff0000) >>16) | (((*(WP(p)+1))&0x0000ffff) << 16)) +#define MUR_ONE_THREE(p) ((((*WP(p))&0xff000000) >>24) | (((*(WP(p)+1))&0x00ffffff) << 8)) +#endif +#define MUR_GETBLOCK(p,i) (MUR_PLUS0_ALIGNED(p) ? ((p)[i]) : \ + (MUR_PLUS1_ALIGNED(p) ? MUR_THREE_ONE(p) : \ + (MUR_PLUS2_ALIGNED(p) ? MUR_TWO_TWO(p) : \ + MUR_ONE_THREE(p)))) +#endif +#define MUR_ROTL32(x,r) (((x) << (r)) | ((x) >> (32 - (r)))) +#define MUR_FMIX(_h) \ +do { \ + _h ^= _h >> 16; \ + _h *= 0x85ebca6bu; \ + _h ^= _h >> 13; \ + _h *= 0xc2b2ae35u; \ + _h ^= _h >> 16; \ +} while(0) + +#define HASH_MUR(key,keylen,num_bkts,hashv,bkt) \ +do { \ + const uint8_t *_mur_data = (const uint8_t*)(key); \ + const int _mur_nblocks = (int)(keylen) / 4; \ + uint32_t _mur_h1 = 0xf88D5353u; \ + uint32_t _mur_c1 = 0xcc9e2d51u; \ + uint32_t _mur_c2 = 0x1b873593u; \ + uint32_t _mur_k1 = 0; \ + const uint8_t *_mur_tail; \ + const uint32_t *_mur_blocks = (const uint32_t*)(_mur_data+(_mur_nblocks*4)); \ + int _mur_i; \ + for(_mur_i = -_mur_nblocks; _mur_i!=0; _mur_i++) { \ + _mur_k1 = MUR_GETBLOCK(_mur_blocks,_mur_i); \ + _mur_k1 *= _mur_c1; \ + _mur_k1 = MUR_ROTL32(_mur_k1,15); \ + _mur_k1 *= _mur_c2; \ + \ + _mur_h1 ^= _mur_k1; \ + _mur_h1 = MUR_ROTL32(_mur_h1,13); \ + _mur_h1 = (_mur_h1*5U) + 0xe6546b64u; \ + } \ + _mur_tail = (const uint8_t*)(_mur_data + (_mur_nblocks*4)); \ + _mur_k1=0; \ + switch((keylen) & 3U) { \ + case 3: _mur_k1 ^= (uint32_t)_mur_tail[2] << 16; /* FALLTHROUGH */ \ + case 2: _mur_k1 ^= (uint32_t)_mur_tail[1] << 8; /* FALLTHROUGH */ \ + case 1: _mur_k1 ^= (uint32_t)_mur_tail[0]; \ + _mur_k1 *= _mur_c1; \ + _mur_k1 = MUR_ROTL32(_mur_k1,15); \ + _mur_k1 *= _mur_c2; \ + _mur_h1 ^= _mur_k1; \ + } \ + _mur_h1 ^= (uint32_t)(keylen); \ + MUR_FMIX(_mur_h1); \ + hashv = _mur_h1; \ + bkt = hashv & (num_bkts-1U); \ +} while(0) +#endif /* HASH_USING_NO_STRICT_ALIASING */ + +/* key comparison function; return 0 if keys equal */ +#define HASH_KEYCMP(a,b,len) memcmp(a,b,(unsigned long)(len)) + +/* iterate over items in a known bucket to find desired item */ +#define HASH_FIND_IN_BKT(tbl,hh,head,keyptr,keylen_in,out) \ +do { \ + if (head.hh_head != NULL) { DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,head.hh_head)); } \ + else { out=NULL; } \ + while (out != NULL) { \ + if ((out)->hh.keylen == (keylen_in)) { \ + if ((HASH_KEYCMP((out)->hh.key,keyptr,keylen_in)) == 0) { break; } \ + } \ + if ((out)->hh.hh_next != NULL) { DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,(out)->hh.hh_next)); } \ + else { out = NULL; } \ + } \ +} while(0) + +/* add an item to a bucket */ +#define HASH_ADD_TO_BKT(head,addhh) \ +do { \ + head.count++; \ + (addhh)->hh_next = head.hh_head; \ + (addhh)->hh_prev = NULL; \ + if (head.hh_head != NULL) { (head).hh_head->hh_prev = (addhh); } \ + (head).hh_head=addhh; \ + if ((head.count >= ((head.expand_mult+1U) * HASH_BKT_CAPACITY_THRESH)) \ + && ((addhh)->tbl->noexpand != 1U)) { \ + HASH_EXPAND_BUCKETS((addhh)->tbl); \ + } \ +} while(0) + +/* remove an item from a given bucket */ +#define HASH_DEL_IN_BKT(hh,head,hh_del) \ + (head).count--; \ + if ((head).hh_head == hh_del) { \ + (head).hh_head = hh_del->hh_next; \ + } \ + if (hh_del->hh_prev) { \ + hh_del->hh_prev->hh_next = hh_del->hh_next; \ + } \ + if (hh_del->hh_next) { \ + hh_del->hh_next->hh_prev = hh_del->hh_prev; \ + } + +/* Bucket expansion has the effect of doubling the number of buckets + * and redistributing the items into the new buckets. Ideally the + * items will distribute more or less evenly into the new buckets + * (the extent to which this is true is a measure of the quality of + * the hash function as it applies to the key domain). + * + * With the items distributed into more buckets, the chain length + * (item count) in each bucket is reduced. Thus by expanding buckets + * the hash keeps a bound on the chain length. This bounded chain + * length is the essence of how a hash provides constant time lookup. + * + * The calculation of tbl->ideal_chain_maxlen below deserves some + * explanation. First, keep in mind that we're calculating the ideal + * maximum chain length based on the *new* (doubled) bucket count. + * In fractions this is just n/b (n=number of items,b=new num buckets). + * Since the ideal chain length is an integer, we want to calculate + * ceil(n/b). We don't depend on floating point arithmetic in this + * hash, so to calculate ceil(n/b) with integers we could write + * + * ceil(n/b) = (n/b) + ((n%b)?1:0) + * + * and in fact a previous version of this hash did just that. + * But now we have improved things a bit by recognizing that b is + * always a power of two. We keep its base 2 log handy (call it lb), + * so now we can write this with a bit shift and logical AND: + * + * ceil(n/b) = (n>>lb) + ( (n & (b-1)) ? 1:0) + * + */ +#define HASH_EXPAND_BUCKETS(tbl) \ +do { \ + unsigned _he_bkt; \ + unsigned _he_bkt_i; \ + struct UT_hash_handle *_he_thh, *_he_hh_nxt; \ + UT_hash_bucket *_he_new_buckets, *_he_newbkt; \ + _he_new_buckets = (UT_hash_bucket*)uthash_malloc( \ + 2UL * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \ + if (!_he_new_buckets) { uthash_fatal( "out of memory"); } \ + memset(_he_new_buckets, 0, \ + 2UL * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \ + tbl->ideal_chain_maxlen = \ + (tbl->num_items >> (tbl->log2_num_buckets+1U)) + \ + (((tbl->num_items & ((tbl->num_buckets*2U)-1U)) != 0U) ? 1U : 0U); \ + tbl->nonideal_items = 0; \ + for(_he_bkt_i = 0; _he_bkt_i < tbl->num_buckets; _he_bkt_i++) \ + { \ + _he_thh = tbl->buckets[ _he_bkt_i ].hh_head; \ + while (_he_thh != NULL) { \ + _he_hh_nxt = _he_thh->hh_next; \ + HASH_TO_BKT( _he_thh->hashv, tbl->num_buckets*2U, _he_bkt); \ + _he_newbkt = &(_he_new_buckets[ _he_bkt ]); \ + if (++(_he_newbkt->count) > tbl->ideal_chain_maxlen) { \ + tbl->nonideal_items++; \ + _he_newbkt->expand_mult = _he_newbkt->count / \ + tbl->ideal_chain_maxlen; \ + } \ + _he_thh->hh_prev = NULL; \ + _he_thh->hh_next = _he_newbkt->hh_head; \ + if (_he_newbkt->hh_head != NULL) { _he_newbkt->hh_head->hh_prev = \ + _he_thh; } \ + _he_newbkt->hh_head = _he_thh; \ + _he_thh = _he_hh_nxt; \ + } \ + } \ + uthash_free( tbl->buckets, tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \ + tbl->num_buckets *= 2U; \ + tbl->log2_num_buckets++; \ + tbl->buckets = _he_new_buckets; \ + tbl->ineff_expands = (tbl->nonideal_items > (tbl->num_items >> 1)) ? \ + (tbl->ineff_expands+1U) : 0U; \ + if (tbl->ineff_expands > 1U) { \ + tbl->noexpand=1; \ + uthash_noexpand_fyi(tbl); \ + } \ + uthash_expand_fyi(tbl); \ +} while(0) + + +/* This is an adaptation of Simon Tatham's O(n log(n)) mergesort */ +/* Note that HASH_SORT assumes the hash handle name to be hh. + * HASH_SRT was added to allow the hash handle name to be passed in. */ +#define HASH_SORT(head,cmpfcn) HASH_SRT(hh,head,cmpfcn) +#define HASH_SRT(hh,head,cmpfcn) \ +do { \ + unsigned _hs_i; \ + unsigned _hs_looping,_hs_nmerges,_hs_insize,_hs_psize,_hs_qsize; \ + struct UT_hash_handle *_hs_p, *_hs_q, *_hs_e, *_hs_list, *_hs_tail; \ + if (head != NULL) { \ + _hs_insize = 1; \ + _hs_looping = 1; \ + _hs_list = &((head)->hh); \ + while (_hs_looping != 0U) { \ + _hs_p = _hs_list; \ + _hs_list = NULL; \ + _hs_tail = NULL; \ + _hs_nmerges = 0; \ + while (_hs_p != NULL) { \ + _hs_nmerges++; \ + _hs_q = _hs_p; \ + _hs_psize = 0; \ + for ( _hs_i = 0; _hs_i < _hs_insize; _hs_i++ ) { \ + _hs_psize++; \ + _hs_q = (UT_hash_handle*)((_hs_q->next != NULL) ? \ + ((void*)((char*)(_hs_q->next) + \ + (head)->hh.tbl->hho)) : NULL); \ + if (! (_hs_q) ) { break; } \ + } \ + _hs_qsize = _hs_insize; \ + while ((_hs_psize > 0U) || ((_hs_qsize > 0U) && (_hs_q != NULL))) {\ + if (_hs_psize == 0U) { \ + _hs_e = _hs_q; \ + _hs_q = (UT_hash_handle*)((_hs_q->next != NULL) ? \ + ((void*)((char*)(_hs_q->next) + \ + (head)->hh.tbl->hho)) : NULL); \ + _hs_qsize--; \ + } else if ( (_hs_qsize == 0U) || (_hs_q == NULL) ) { \ + _hs_e = _hs_p; \ + if (_hs_p != NULL){ \ + _hs_p = (UT_hash_handle*)((_hs_p->next != NULL) ? \ + ((void*)((char*)(_hs_p->next) + \ + (head)->hh.tbl->hho)) : NULL); \ + } \ + _hs_psize--; \ + } else if (( \ + cmpfcn(DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_p)), \ + DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_q))) \ + ) <= 0) { \ + _hs_e = _hs_p; \ + if (_hs_p != NULL){ \ + _hs_p = (UT_hash_handle*)((_hs_p->next != NULL) ? \ + ((void*)((char*)(_hs_p->next) + \ + (head)->hh.tbl->hho)) : NULL); \ + } \ + _hs_psize--; \ + } else { \ + _hs_e = _hs_q; \ + _hs_q = (UT_hash_handle*)((_hs_q->next != NULL) ? \ + ((void*)((char*)(_hs_q->next) + \ + (head)->hh.tbl->hho)) : NULL); \ + _hs_qsize--; \ + } \ + if ( _hs_tail != NULL ) { \ + _hs_tail->next = ((_hs_e != NULL) ? \ + ELMT_FROM_HH((head)->hh.tbl,_hs_e) : NULL); \ + } else { \ + _hs_list = _hs_e; \ + } \ + if (_hs_e != NULL) { \ + _hs_e->prev = ((_hs_tail != NULL) ? \ + ELMT_FROM_HH((head)->hh.tbl,_hs_tail) : NULL); \ + } \ + _hs_tail = _hs_e; \ + } \ + _hs_p = _hs_q; \ + } \ + if (_hs_tail != NULL){ \ + _hs_tail->next = NULL; \ + } \ + if ( _hs_nmerges <= 1U ) { \ + _hs_looping=0; \ + (head)->hh.tbl->tail = _hs_tail; \ + DECLTYPE_ASSIGN(head,ELMT_FROM_HH((head)->hh.tbl, _hs_list)); \ + } \ + _hs_insize *= 2U; \ + } \ + HASH_FSCK(hh,head); \ + } \ +} while (0) + +/* This function selects items from one hash into another hash. + * The end result is that the selected items have dual presence + * in both hashes. There is no copy of the items made; rather + * they are added into the new hash through a secondary hash + * hash handle that must be present in the structure. */ +#define HASH_SELECT(hh_dst, dst, hh_src, src, cond) \ +do { \ + unsigned _src_bkt, _dst_bkt; \ + void *_last_elt=NULL, *_elt; \ + UT_hash_handle *_src_hh, *_dst_hh, *_last_elt_hh=NULL; \ + ptrdiff_t _dst_hho = ((char*)(&(dst)->hh_dst) - (char*)(dst)); \ + if (src != NULL) { \ + for(_src_bkt=0; _src_bkt < (src)->hh_src.tbl->num_buckets; _src_bkt++) { \ + for(_src_hh = (src)->hh_src.tbl->buckets[_src_bkt].hh_head; \ + _src_hh != NULL; \ + _src_hh = _src_hh->hh_next) { \ + _elt = ELMT_FROM_HH((src)->hh_src.tbl, _src_hh); \ + if (cond(_elt)) { \ + _dst_hh = (UT_hash_handle*)(((char*)_elt) + _dst_hho); \ + _dst_hh->key = _src_hh->key; \ + _dst_hh->keylen = _src_hh->keylen; \ + _dst_hh->hashv = _src_hh->hashv; \ + _dst_hh->prev = _last_elt; \ + _dst_hh->next = NULL; \ + if (_last_elt_hh != NULL) { _last_elt_hh->next = _elt; } \ + if (dst == NULL) { \ + DECLTYPE_ASSIGN(dst,_elt); \ + HASH_MAKE_TABLE(hh_dst,dst); \ + } else { \ + _dst_hh->tbl = (dst)->hh_dst.tbl; \ + } \ + HASH_TO_BKT(_dst_hh->hashv, _dst_hh->tbl->num_buckets, _dst_bkt); \ + HASH_ADD_TO_BKT(_dst_hh->tbl->buckets[_dst_bkt],_dst_hh); \ + (dst)->hh_dst.tbl->num_items++; \ + _last_elt = _elt; \ + _last_elt_hh = _dst_hh; \ + } \ + } \ + } \ + } \ + HASH_FSCK(hh_dst,dst); \ +} while (0) + +#define HASH_CLEAR(hh,head) \ +do { \ + if (head != NULL) { \ + uthash_free((head)->hh.tbl->buckets, \ + (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket)); \ + HASH_BLOOM_FREE((head)->hh.tbl); \ + uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \ + (head)=NULL; \ + } \ +} while(0) + +#define HASH_OVERHEAD(hh,head) \ + ((head != NULL) ? ( \ + (size_t)(((head)->hh.tbl->num_items * sizeof(UT_hash_handle)) + \ + ((head)->hh.tbl->num_buckets * sizeof(UT_hash_bucket)) + \ + sizeof(UT_hash_table) + \ + (HASH_BLOOM_BYTELEN))) : 0U) + +#ifdef NO_DECLTYPE +#define HASH_ITER(hh,head,el,tmp) \ +for(((el)=(head)), ((*(char**)(&(tmp)))=(char*)((head!=NULL)?(head)->hh.next:NULL)); \ + (el) != NULL; ((el)=(tmp)), ((*(char**)(&(tmp)))=(char*)((tmp!=NULL)?(tmp)->hh.next:NULL))) +#else +#define HASH_ITER(hh,head,el,tmp) \ +for(((el)=(head)), ((tmp)=DECLTYPE(el)((head!=NULL)?(head)->hh.next:NULL)); \ + (el) != NULL; ((el)=(tmp)), ((tmp)=DECLTYPE(el)((tmp!=NULL)?(tmp)->hh.next:NULL))) +#endif + +/* obtain a count of items in the hash */ +#define HASH_COUNT(head) HASH_CNT(hh,head) +#define HASH_CNT(hh,head) ((head != NULL)?((head)->hh.tbl->num_items):0U) + +typedef struct UT_hash_bucket { + struct UT_hash_handle *hh_head; + unsigned count; + + /* expand_mult is normally set to 0. In this situation, the max chain length + * threshold is enforced at its default value, HASH_BKT_CAPACITY_THRESH. (If + * the bucket's chain exceeds this length, bucket expansion is triggered). + * However, setting expand_mult to a non-zero value delays bucket expansion + * (that would be triggered by additions to this particular bucket) + * until its chain length reaches a *multiple* of HASH_BKT_CAPACITY_THRESH. + * (The multiplier is simply expand_mult+1). The whole idea of this + * multiplier is to reduce bucket expansions, since they are expensive, in + * situations where we know that a particular bucket tends to be overused. + * It is better to let its chain length grow to a longer yet-still-bounded + * value, than to do an O(n) bucket expansion too often. + */ + unsigned expand_mult; + +} UT_hash_bucket; + +/* random signature used only to find hash tables in external analysis */ +#define HASH_SIGNATURE 0xa0111fe1u +#define HASH_BLOOM_SIGNATURE 0xb12220f2u + +typedef struct UT_hash_table { + UT_hash_bucket *buckets; + unsigned num_buckets, log2_num_buckets; + unsigned num_items; + struct UT_hash_handle *tail; /* tail hh in app order, for fast append */ + ptrdiff_t hho; /* hash handle offset (byte pos of hash handle in element */ + + /* in an ideal situation (all buckets used equally), no bucket would have + * more than ceil(#items/#buckets) items. that's the ideal chain length. */ + unsigned ideal_chain_maxlen; + + /* nonideal_items is the number of items in the hash whose chain position + * exceeds the ideal chain maxlen. these items pay the penalty for an uneven + * hash distribution; reaching them in a chain traversal takes >ideal steps */ + unsigned nonideal_items; + + /* ineffective expands occur when a bucket doubling was performed, but + * afterward, more than half the items in the hash had nonideal chain + * positions. If this happens on two consecutive expansions we inhibit any + * further expansion, as it's not helping; this happens when the hash + * function isn't a good fit for the key domain. When expansion is inhibited + * the hash will still work, albeit no longer in constant time. */ + unsigned ineff_expands, noexpand; + + uint32_t signature; /* used only to find hash tables in external analysis */ +#ifdef HASH_BLOOM + uint32_t bloom_sig; /* used only to test bloom exists in external analysis */ + uint8_t *bloom_bv; + uint8_t bloom_nbits; +#endif + +} UT_hash_table; + +typedef struct UT_hash_handle { + struct UT_hash_table *tbl; + void *prev; /* prev element in app order */ + void *next; /* next element in app order */ + struct UT_hash_handle *hh_prev; /* previous hh in bucket order */ + struct UT_hash_handle *hh_next; /* next hh in bucket order */ + void *key; /* ptr to enclosing struct's key */ + unsigned keylen; /* enclosing struct's key len */ + unsigned hashv; /* result of hash-fcn(key) */ +} UT_hash_handle; + +#endif /* UTHASH_H */ diff --git a/src/c/urweb.c b/src/c/urweb.c index 53344c5e..e0fd503c 100644 --- a/src/c/urweb.c +++ b/src/c/urweb.c @@ -22,6 +22,8 @@ #include "types.h" +#include "uthash.h" + uw_unit uw_unit_v = 0; @@ -4494,3 +4496,148 @@ int uw_remoteSock(uw_context ctx) { void uw_set_remoteSock(uw_context ctx, int sock) { ctx->remoteSock = sock; } + + +// Sqlcache + +void listDelete(CacheList *list, CacheEntry *entry) { + if (list->first == entry) { + list->first = entry->next; + } + if (list->last == entry) { + list->last = entry->prev; + } + if (entry->prev) { + entry->prev->next = entry->next; + } + if (entry->next) { + entry->next->prev = entry->prev; + } + entry->prev = NULL; + entry->next = NULL; + --(list->size); +} + +void listAdd(CacheList *list, CacheEntry *entry) { + if (list->last) { + list->last->next = entry; + entry->prev = list->last; + list->last = entry; + } else { + list->first = entry; + list->last = entry; + } + ++(list->size); +} + +void listBump(CacheList *list, CacheEntry *entry) { + listDelete(list, entry); + listAdd(list, entry); +} + +// TODO: deal with time properly. + +time_t getTimeNow() { + return time(NULL); +} + +time_t timeMax(time_t x, time_t y) { + return difftime(x, y) > 0 ? x : y; +} + +void freeCacheValue(CacheValue *value) { + if (value) { + free(value->result); + free(value->output); + free(value); + } +} + +void delete(Cache *cache, CacheEntry* entry) { + //listDelete(cache->lru, entry); + HASH_DELETE(hh, cache->table, entry); + freeCacheValue(entry->value); + free(entry->key); + free(entry); +} + +CacheValue *checkHelper(Cache *cache, char **keys, int timeInvalid) { + char *key = keys[cache->height]; + CacheEntry *entry; + HASH_FIND(hh, cache->table, key, strlen(key), entry); + timeInvalid = timeMax(timeInvalid, cache->timeInvalid); + if (entry && difftime(entry->timeValid, timeInvalid) > 0) { + if (cache->height == 0) { + // At height 0, entry->value is the desired value. + //listBump(cache->lru, entry); + return entry->value; + } else { + // At height n+1, entry->value is a pointer to a cache at heignt n. + return checkHelper(entry->value, keys, timeInvalid); + } + } else { + return NULL; + } +} + +CacheValue *check(Cache *cache, char **keys) { + return checkHelper(cache, keys, 0); +} + +void storeHelper(Cache *cache, char **keys, CacheValue *value, int timeNow) { + CacheEntry *entry; + char *key = keys[cache->height]; + HASH_FIND(hh, cache->table, key, strlen(key), entry); + if (!entry) { + entry = malloc(sizeof(CacheEntry)); + entry->key = strdup(key); + entry->value = NULL; + HASH_ADD_KEYPTR(hh, cache->table, entry->key, strlen(entry->key), entry); + } + entry->timeValid = timeNow; + if (cache->height == 0) { + //listAdd(cache->lru, entry); + freeCacheValue(entry->value); + entry->value = value; + //if (cache->lru->size > MAX_SIZE) { + //delete(cache, cache->lru->first); + // TODO: return flushed value. + //} + } else { + if (!entry->value) { + Cache *newCache = malloc(sizeof(Cache)); + newCache->table = NULL; + newCache->timeInvalid = timeNow; + newCache->lru = cache->lru; + newCache->height = cache->height - 1; + entry->value = newCache; + } + storeHelper(entry->value, keys, value, timeNow); + } +} + +void store(Cache *cache, char **keys, CacheValue *value) { + storeHelper(cache, keys, value, getTimeNow()); +} + +void flushHelper(Cache *cache, char **keys, int timeNow) { + CacheEntry *entry; + char *key = keys[cache->height]; + if (key) { + HASH_FIND(hh, cache->table, key, strlen(key), entry); + if (entry) { + if (cache->height == 0) { + delete(cache, entry); + } else { + flushHelper(entry->value, keys, timeNow); + } + } + } else { + // Null key means invalidate the entire subtree. + cache->timeInvalid = timeNow; + } +} + +void flush(Cache *cache, char **keys) { + flushHelper(cache, keys, getTimeNow()); +} diff --git a/src/lru_cache.sml b/src/lru_cache.sml new file mode 100644 index 00000000..87e939fa --- /dev/null +++ b/src/lru_cache.sml @@ -0,0 +1,171 @@ +structure LruCache : sig + val cache : Cache.cache +end = struct + + +(* Mono *) + +open Mono + +val dummyLoc = ErrorMsg.dummySpan +val stringTyp = (TFfi ("Basis", "string"), dummyLoc) +val optionStringTyp = (TOption stringTyp, dummyLoc) +fun withTyp typ = map (fn exp => (exp, typ)) + +fun ffiAppCache' (func, index, argTyps) = + EFfiApp ("Sqlcache", func ^ Int.toString index, argTyps) + +fun check (index, keys) = + ffiAppCache' ("check", index, withTyp stringTyp keys) + +fun store (index, keys, value) = + ffiAppCache' ("store", index, (value, stringTyp) :: withTyp stringTyp keys) + +fun flush (index, keys) = + ffiAppCache' ("flush", index, withTyp optionStringTyp keys) + + +(* Cjr *) + +open Print +open Print.PD + +fun setupQuery {index, params} = + let + + val i = Int.toString index + + fun paramRepeat itemi sep = + let + fun f n = + if n < 0 then "" + else if n = 0 then itemi (Int.toString 0) + else f (n-1) ^ sep ^ itemi (Int.toString n) + in + f (params - 1) + end + + fun paramRepeatRev itemi sep = + let + fun f n = + if n < 0 then "" + else if n = 0 then itemi (Int.toString 0) + else itemi (Int.toString n) ^ sep ^ f (n-1) + in + f (params - 1) + end + + fun paramRepeatInit itemi sep = + if params = 0 then "" else sep ^ paramRepeat itemi sep + + val typedArgs = paramRepeatInit (fn p => "uw_Basis_string p" ^ p) ", " + + val revArgs = paramRepeatRev (fn p => "p" ^ p) ", " + + in + Print.box + [string ("static Cache cacheStruct" ^ i ^ " = {"), + newline, + string " .table = NULL,", + newline, + string " .timeInvalid = 0,", + newline, + string " .lru = NULL,", + newline, + string (" .height = " ^ Int.toString (params - 1) ^ "};"), + newline, + string ("static Cache *cache" ^ i ^ " = &cacheStruct" ^ i ^ ";"), + newline, + newline, + + string ("static uw_Basis_string uw_Sqlcache_check" ^ i), + string ("(uw_context ctx" ^ typedArgs ^ ") {"), + newline, + string (" char *ks[] = {" ^ revArgs ^ "};"), + newline, + string (" CacheValue *v = check(cache" ^ i ^ ", ks);"), + newline, + string " if (v) {", + newline, + string (" puts(\"SQLCACHE: hit " ^ i ^ ".\");"), + newline, + string " uw_write(ctx, v->output);", + newline, + string " return v->result;", + newline, + string " } else {", + newline, + string (" puts(\"SQLCACHE: miss " ^ i ^ ".\");"), + newline, + string " uw_recordingStart(ctx);", + newline, + string " return NULL;", + newline, + string " }", + newline, + string "}", + newline, + newline, + + string ("static uw_unit uw_Sqlcache_store" ^ i), + string ("(uw_context ctx, uw_Basis_string s" ^ typedArgs ^ ") {"), + newline, + string (" char *ks[] = {" ^ revArgs ^ "};"), + newline, + string (" CacheValue *v = malloc(sizeof(CacheValue));"), + newline, + string " v->result = strdup(s);", + newline, + string " v->output = uw_recordingRead(ctx);", + newline, + string (" puts(\"SQLCACHE: stored " ^ i ^ ".\");"), + newline, + string (" store(cache" ^ i ^ ", ks, v);"), + newline, + string " return uw_unit_v;", + newline, + string "}", + newline, + newline, + + string ("static uw_unit uw_Sqlcache_flush" ^ i), + string ("(uw_context ctx" ^ typedArgs ^ ") {"), + newline, + string (" char *ks[] = {" ^ revArgs ^ "};"), + newline, + string (" flush(cache" ^ i ^ ", ks);"), + newline, + string " return uw_unit_v;", + newline, + string "}", + newline, + newline] + end + +val setupGlobal = string "/* No global setup for LRU cache. */" + + +(* Bundled up. *) + +(* For now, use the toy implementation if there are no arguments. *) +fun toyIfNoKeys numKeys implLru implToy args = + if numKeys args = 0 + then implToy args + else implLru args + +val cache = + let + val {check = toyCheck, + store = toyStore, + flush = toyFlush, + setupQuery = toySetupQuery, + ...} = ToyCache.cache + in + {check = toyIfNoKeys (length o #2) check toyCheck, + store = toyIfNoKeys (length o #2) store toyStore, + flush = toyIfNoKeys (length o #2) flush toyFlush, + setupQuery = toyIfNoKeys #params setupQuery toySetupQuery, + setupGlobal = setupGlobal} + end + +end diff --git a/src/sources b/src/sources index aaf640ca..0608d710 100644 --- a/src/sources +++ b/src/sources @@ -177,6 +177,7 @@ $(SRC)/multimap_fn.sml $(SRC)/cache.sml $(SRC)/toy_cache.sml +$(SRC)/lru_cache.sml $(SRC)/sqlcache.sig $(SRC)/sqlcache.sml diff --git a/src/sqlcache.sml b/src/sqlcache.sml index 3082904c..bf9ee77a 100644 --- a/src/sqlcache.sml +++ b/src/sqlcache.sml @@ -39,7 +39,7 @@ val ffiEffectful = andalso not (m = "Basis" andalso SS.member (fs, f)) end -val cache = ref ToyCache.cache +val cache = ref LruCache.cache fun setCache c = cache := c fun getCache () = !cache @@ -52,8 +52,8 @@ fun effectful doPrint (effs : IS.set) (inFunction : bool) (bound : int) : exp -> false, then expression is definitely not effectful if effs is fully populated. The intended pattern is to use this a number of times equal to the number of declarations in a file, Bellman-Ford style. *) - (* TODO: make incrementing of bound less janky, probably by using [MonoUtil] - instead of all this. *) + (* TODO: make incrementing of the number of bound variables cleaner, + probably by using [MonoUtil] instead of all this. *) let (* DEBUG: remove printing when done. *) fun tru msg = if doPrint then (print (msg ^ "\n"); true) else true @@ -138,14 +138,14 @@ val effectfulMap = (* Boolean formula normalization. *) -datatype normalForm = Cnf | Dnf +datatype junctionType = Conj | Disj datatype 'atom formula = Atom of 'atom | Negate of 'atom formula - | Combo of normalForm * 'atom formula list + | Combo of junctionType * 'atom formula list -val flipNf = fn Cnf => Dnf | Dnf => Cnf +val flipJt = fn Conj => Disj | Disj => Conj fun bind xs f = List.concat (map f xs) @@ -158,7 +158,7 @@ val rec cartesianProduct : 'a list list -> 'a list list = fun pushNegate (negate : 'atom -> 'atom) (negating : bool) = fn Atom x => Atom (if negating then negate x else x) | Negate f => pushNegate negate (not negating) f - | Combo (n, fs) => Combo (if negating then flipNf n else n, map (pushNegate negate negating) fs) + | Combo (n, fs) => Combo (if negating then flipJt n else n, map (pushNegate negate negating) fs) val rec flatten = fn Combo (n, fs) => @@ -170,17 +170,17 @@ val rec flatten = (map flatten fs)) | f => f -fun normalize' (negate : 'atom -> 'atom) (norm : normalForm) = +fun normalize' (negate : 'atom -> 'atom) (junc : junctionType) = fn Atom x => [[x]] - | Negate f => map (map negate) (normalize' negate (flipNf norm) f) - | Combo (n, fs) => + | Negate f => map (map negate) (normalize' negate (flipJt junc) f) + | Combo (j, fs) => let - val fss = bind fs (normalize' negate n) + val fss = bind fs (normalize' negate j) in - if n = norm then fss else cartesianProduct fss + if j = junc then fss else cartesianProduct fss end -fun normalize negate norm = normalize' negate norm o flatten o pushNegate negate false +fun normalize negate junc = normalize' negate junc o flatten o pushNegate negate false fun mapFormula mf = fn Atom x => Atom (mf x) @@ -200,36 +200,29 @@ datatype atomExp = | Prim of Prim.t | Field of string * string -val equalAtomExp = - let - val isEqual = fn EQUAL => true | _ => false - in - fn (QueryArg n1, QueryArg n2) => n1 = n2 - | (DmlRel n1, DmlRel n2) => n1 = n2 - | (Prim p1, Prim p2) => isEqual (Prim.compare (p1, p2)) - | (Field (t1, f1), Field (t2, f2)) => isEqual (String.compare (t1 ^ "." ^ f1, t2 ^ "." ^ f2)) - | _ => false - end - structure AtomExpKey : ORD_KEY = struct -type ord_key = atomExp - -val compare = - fn (QueryArg n1, QueryArg n2) => Int.compare (n1, n2) - | (QueryArg _, _) => LESS - | (_, QueryArg _) => GREATER - | (DmlRel n1, DmlRel n2) => Int.compare (n1, n2) - | (DmlRel _, _) => LESS - | (_, DmlRel _) => GREATER - | (Prim p1, Prim p2) => Prim.compare (p1, p2) - | (Prim _, _) => LESS - | (_, Prim _) => GREATER - | (Field (t1, f1), Field (t2, f2)) => String.compare (t1 ^ "." ^ f1, t2 ^ "." ^ f2) + type ord_key = atomExp + + val compare = + fn (QueryArg n1, QueryArg n2) => Int.compare (n1, n2) + | (QueryArg _, _) => LESS + | (_, QueryArg _) => GREATER + | (DmlRel n1, DmlRel n2) => Int.compare (n1, n2) + | (DmlRel _, _) => LESS + | (_, DmlRel _) => GREATER + | (Prim p1, Prim p2) => Prim.compare (p1, p2) + | (Prim _, _) => LESS + | (_, Prim _) => GREATER + | (Field (t1, f1), Field (t2, f2)) => + case String.compare (t1, t2) of + EQUAL => String.compare (f1, f2) + | ord => ord end structure UF = UnionFindFn(AtomExpKey) + val conflictMaps : (Sql.cmp * Sql.sqexp * Sql.sqexp) formula * (Sql.cmp * Sql.sqexp * Sql.sqexp) formula -> atomExp IM.map list = @@ -246,7 +239,9 @@ val conflictMaps : (Sql.cmp * Sql.sqexp * Sql.sqexp) formula o List.mapPartial toKnownEquality fun addToEqs (eqs, n, e) = case IM.find (eqs, n) of - (* Comparing to a constant seems better? *) + (* Comparing to a constant is probably better than comparing to + a variable? Checking that an existing constant matches a new + one is handled by [accumulateEqs]. *) SOME (Prim _) => eqs | _ => IM.insert (eqs, n, e) val accumulateEqs = @@ -263,6 +258,9 @@ val conflictMaps : (Sql.cmp * Sql.sqexp * Sql.sqexp) formula (* TODO: deal with equalities involving just [DmlRel]s and [Prim]s. This would involve guarding the invalidation with a check for the relevant comparisons. *) + (* DEBUG: remove these print statements. *) + (* | ((DmlRel r, Prim p), eqso) => (print ("sadness " ^ Int.toString r ^ " = " ^ Prim.toString p ^ "\n"); eqso) *) + (* | ((Prim p, DmlRel r), eqso) => (print ("sadness " ^ Int.toString r ^ " = " ^ Prim.toString p ^ "\n"); eqso) *) | (_, eqso) => eqso val eqsOfClass : atomExp list -> atomExp IM.map option = List.foldl accumulateEqs (SOME IM.empty) @@ -275,7 +273,8 @@ val conflictMaps : (Sql.cmp * Sql.sqexp * Sql.sqexp) formula | Sql.Field tf => SOME (Field tf) | Sql.Inj (EPrim p, _) => SOME (Prim p) | Sql.Inj (ERel n, _) => SOME (rel n) - (* We can't deal with anything else. *) + (* We can't deal with anything else, e.g., CURRENT_TIMESTAMP + becomes Sql.Unmodeled, which becomes NONE here. *) | _ => NONE in (cmp, qa e1, qa e2) @@ -302,17 +301,17 @@ val conflictMaps : (Sql.cmp * Sql.sqexp * Sql.sqexp) formula List.foldr (fn (SOME eqs, SOME acc) => SOME (IM.unionWith #1 (eqs, acc)) | _ => NONE) (SOME IM.empty) fun dnf (fQuery, fDml) = - normalize negateCmp Dnf (Combo (Cnf, [markQuery fQuery, markDml fDml])) + normalize negateCmp Disj (Combo (Conj, [markQuery fQuery, markDml fDml])) in List.mapPartial (mergeEqs o map eqsOfClass o equivClasses) o dnf end val rec sqexpToFormula = - fn Sql.SqTrue => Combo (Cnf, []) - | Sql.SqFalse => Combo (Dnf, []) + fn Sql.SqTrue => Combo (Conj, []) + | Sql.SqFalse => Combo (Disj, []) | Sql.SqNot e => Negate (sqexpToFormula e) | Sql.Binop (Sql.RCmp c, e1, e2) => Atom (c, e1, e2) - | Sql.Binop (Sql.RLop l, p1, p2) => Combo (case l of Sql.And => Cnf | Sql.Or => Dnf, + | Sql.Binop (Sql.RLop l, p1, p2) => Combo (case l of Sql.And => Conj | Sql.Or => Disj, [sqexpToFormula p1, sqexpToFormula p2]) (* ASK: any other sqexps that can be props? *) | _ => raise Match @@ -332,13 +331,13 @@ fun renameTables tablePairs = end val rec queryToFormula = - fn Sql.Query1 {Where = NONE, ...} => Combo (Cnf, []) + fn Sql.Query1 {Where = NONE, ...} => Combo (Conj, []) | Sql.Query1 {From = tablePairs, Where = SOME e, ...} => renameTables tablePairs (sqexpToFormula e) - | Sql.Union (q1, q2) => Combo (Dnf, [queryToFormula q1, queryToFormula q2]) + | Sql.Union (q1, q2) => Combo (Disj, [queryToFormula q1, queryToFormula q2]) fun valsToFormula (table, vals) = - Combo (Cnf, map (fn (field, v) => Atom (Sql.Eq, Sql.Field (table, field), v)) vals) + Combo (Conj, map (fn (field, v) => Atom (Sql.Eq, Sql.Field (table, field), v)) vals) val rec dmlToFormula = fn Sql.Insert (table, vals) => valsToFormula (table, vals) @@ -354,8 +353,8 @@ val rec dmlToFormula = val mark = mapFormula (fn (cmp, e1, e2) => (cmp, markField e1, markField e2)) in renameTables [(table, "T")] - (Combo (Dnf, [Combo (Cnf, [fVals, mark fWhere]), - Combo (Cnf, [mark fVals, fWhere])])) + (Combo (Disj, [Combo (Conj, [fVals, mark fWhere]), + Combo (Conj, [mark fVals, fWhere])])) end val rec tablesQuery = @@ -370,6 +369,13 @@ val tableDml = (* Program instrumentation. *) +val varName = + let + val varNumber = ref 0 + in + fn s => (varNumber := !varNumber + 1; s ^ Int.toString (!varNumber)) + end + val {check, store, flush, ...} = getCache () val dummyLoc = ErrorMsg.dummySpan @@ -412,8 +418,8 @@ fun cacheWrap (query, i, urlifiedRel0, resultTyp, args) = in ECase (check, [((PNone stringTyp, loc), - (ELet ("q", resultTyp, query, (ESeq (store, rel0), loc)), loc)), - ((PSome (stringTyp, (PVar ("hit", stringTyp), loc)), loc), + (ELet (varName "q", resultTyp, query, (ESeq (store, rel0), loc)), loc)), + ((PSome (stringTyp, (PVar (varName "hit", stringTyp), loc)), loc), (* Boolean is false because we're not unurlifying from a cookie. *) (EUnurlify (rel0, resultTyp, false), loc))], {disc = stringTyp, result = resultTyp}) @@ -454,7 +460,7 @@ fun factorOutNontrivial text = chunks fun wrapLets e' = (* Important that this is foldl (to oppose foldr above). *) - List.foldl (fn (v, e') => ELet ("sqlArg", stringTyp, v, (e', loc))) + List.foldl (fn (v, e') => ELet (varName "sqlArg", stringTyp, v, (e', loc))) e' newVariables val numArgs = length newVariables @@ -482,6 +488,7 @@ fun addChecking file = exps = exps}, dummyLoc) val (EQuery {query = queryText, ...}, _) = queryExp + (* DEBUG: we can remove the following line at some point. *) val () = Print.preface ("sqlcache> ", (MonoPrint.p_exp MonoEnv.empty queryText)) val args = List.tabulate (numArgs, fn n => (ERel n, dummyLoc)) fun bind x f = Option.mapPartial f x @@ -530,9 +537,11 @@ fun invalidations ((query, numArgs), dml) = (* Tests if [ys] makes [xs] a redundant cache invalidation. [NONE] here represents unknown, which means a wider invalidation. *) val rec madeRedundantBy : atomExp option list * atomExp option list -> bool = - fn ([], []) => true + fn ([], []) => (print "hey!\n"; true) | (NONE :: xs, _ :: ys) => madeRedundantBy (xs, ys) - | (SOME x :: xs, SOME y :: ys) => equalAtomExp (x, y) andalso madeRedundantBy (xs, ys) + | (SOME x :: xs, SOME y :: ys) => (case AtomExpKey.compare (x, y) of + EQUAL => madeRedundantBy (xs, ys) + | _ => false) | _ => false fun removeRedundant' (xss, yss) = case xss of |