summaryrefslogtreecommitdiffstats
path: root/ext/lsm1/lsm-test
diff options
context:
space:
mode:
authorDaniel Baumann <daniel.baumann@progress-linux.org>2024-05-05 17:28:19 +0000
committerDaniel Baumann <daniel.baumann@progress-linux.org>2024-05-05 17:28:19 +0000
commit18657a960e125336f704ea058e25c27bd3900dcb (patch)
tree17b438b680ed45a996d7b59951e6aa34023783f2 /ext/lsm1/lsm-test
parentInitial commit. (diff)
downloadsqlite3-upstream.tar.xz
sqlite3-upstream.zip
Adding upstream version 3.40.1.upstream/3.40.1upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'ext/lsm1/lsm-test')
-rw-r--r--ext/lsm1/lsm-test/README40
-rw-r--r--ext/lsm1/lsm-test/lsmtest.h303
-rw-r--r--ext/lsm1/lsm-test/lsmtest1.c656
-rw-r--r--ext/lsm1/lsm-test/lsmtest2.c488
-rw-r--r--ext/lsm1/lsm-test/lsmtest3.c238
-rw-r--r--ext/lsm1/lsm-test/lsmtest4.c127
-rw-r--r--ext/lsm1/lsm-test/lsmtest5.c633
-rw-r--r--ext/lsm1/lsm-test/lsmtest6.c661
-rw-r--r--ext/lsm1/lsm-test/lsmtest7.c206
-rw-r--r--ext/lsm1/lsm-test/lsmtest8.c324
-rw-r--r--ext/lsm1/lsm-test/lsmtest9.c140
-rw-r--r--ext/lsm1/lsm-test/lsmtest_bt.c71
-rw-r--r--ext/lsm1/lsm-test/lsmtest_datasource.c96
-rw-r--r--ext/lsm1/lsm-test/lsmtest_func.c177
-rw-r--r--ext/lsm1/lsm-test/lsmtest_io.c248
-rw-r--r--ext/lsm1/lsm-test/lsmtest_main.c1548
-rw-r--r--ext/lsm1/lsm-test/lsmtest_mem.c409
-rw-r--r--ext/lsm1/lsm-test/lsmtest_tdb.c846
-rw-r--r--ext/lsm1/lsm-test/lsmtest_tdb.h174
-rw-r--r--ext/lsm1/lsm-test/lsmtest_tdb2.cc369
-rw-r--r--ext/lsm1/lsm-test/lsmtest_tdb3.c1429
-rw-r--r--ext/lsm1/lsm-test/lsmtest_tdb4.c980
-rw-r--r--ext/lsm1/lsm-test/lsmtest_util.c223
-rw-r--r--ext/lsm1/lsm-test/lsmtest_win32.c30
24 files changed, 10416 insertions, 0 deletions
diff --git a/ext/lsm1/lsm-test/README b/ext/lsm1/lsm-test/README
new file mode 100644
index 0000000..80654ee
--- /dev/null
+++ b/ext/lsm1/lsm-test/README
@@ -0,0 +1,40 @@
+
+
+Organization of test case files:
+
+ lsmtest1.c: Data tests. Tests that perform many inserts and deletes on a
+ database file, then verify that the contents of the database can
+ be queried.
+
+ lsmtest2.c: Crash tests. Tests that attempt to verify that the database
+ recovers correctly following an application or system crash.
+
+ lsmtest3.c: Rollback tests. Tests that focus on the explicit rollback of
+ transactions and sub-transactions.
+
+ lsmtest4.c: Multi-client tests.
+
+ lsmtest5.c: Multi-client tests with a different thread for each client.
+
+ lsmtest6.c: OOM injection tests.
+
+ lsmtest7.c: API tests.
+
+ lsmtest8.c: Writer crash tests. Tests in this file attempt to verify that
+ the system recovers and other clients proceed unaffected if
+ a process fails in the middle of a write transaction.
+
+ The difference from lsmtest2.c is that this file tests
+ live-recovery (recovery from a failure that occurs while other
+ clients are still running) whereas lsmtest2.c tests recovery
+ from a system or power failure.
+
+ lsmtest9.c: More data tests. These focus on testing that calling
+ lsm_work(nMerge=1) to compact the database does not corrupt it.
+ In other words, that databases containing block-redirects
+ can be read and written.
+
+
+
+
+
diff --git a/ext/lsm1/lsm-test/lsmtest.h b/ext/lsm1/lsm-test/lsmtest.h
new file mode 100644
index 0000000..ca60424
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest.h
@@ -0,0 +1,303 @@
+
+#ifndef __WRAPPER_INT_H_
+#define __WRAPPER_INT_H_
+
+#include "lsmtest_tdb.h"
+#include "sqlite3.h"
+#include "lsm.h"
+
+#include <assert.h>
+#include <stdarg.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#ifndef _WIN32
+# include <unistd.h>
+#endif
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <ctype.h>
+#include <stdlib.h>
+#include <errno.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef _WIN32
+# include "windows.h"
+# define gettimeofday win32GetTimeOfDay
+# define F_OK (0)
+# define sleep(sec) Sleep(1000 * (sec))
+# define usleep(usec) Sleep(((usec) + 999) / 1000)
+# ifdef _MSC_VER
+# include <io.h>
+# define snprintf _snprintf
+# define fsync(fd) FlushFileBuffers((HANDLE)_get_osfhandle((fd)))
+# define fdatasync(fd) FlushFileBuffers((HANDLE)_get_osfhandle((fd)))
+# define __va_copy(dst,src) ((dst) = (src))
+# define ftruncate(fd,sz) ((_chsize_s((fd), (sz))==0) ? 0 : -1)
+# else
+# error Unsupported C compiler for Windows.
+# endif
+int win32GetTimeOfDay(struct timeval *, void *);
+#endif
+
+#ifndef _LSM_INT_H
+typedef unsigned int u32;
+typedef unsigned char u8;
+typedef long long int i64;
+typedef unsigned long long int u64;
+#endif
+
+
+#define ArraySize(x) ((int)(sizeof(x) / sizeof((x)[0])))
+
+#define MIN(x,y) ((x)<(y) ? (x) : (y))
+#define MAX(x,y) ((x)>(y) ? (x) : (y))
+
+#define unused_parameter(x) (void)(x)
+
+#define TESTDB_DEFAULT_PAGE_SIZE 4096
+#define TESTDB_DEFAULT_CACHE_SIZE 2048
+
+#ifndef _O_BINARY
+# define _O_BINARY (0)
+#endif
+
+/*
+** Ideally, these should be in wrapper.c. But they are here instead so that
+** they can be used by the C++ database wrappers in wrapper2.cc.
+*/
+typedef struct DatabaseMethods DatabaseMethods;
+struct TestDb {
+ DatabaseMethods const *pMethods; /* Database methods */
+ const char *zLibrary; /* Library name for tdb_open() */
+};
+struct DatabaseMethods {
+ int (*xClose)(TestDb *);
+ int (*xWrite)(TestDb *, void *, int , void *, int);
+ int (*xDelete)(TestDb *, void *, int);
+ int (*xDeleteRange)(TestDb *, void *, int, void *, int);
+ int (*xFetch)(TestDb *, void *, int, void **, int *);
+ int (*xScan)(TestDb *, void *, int, void *, int, void *, int,
+ void (*)(void *, void *, int , void *, int)
+ );
+ int (*xBegin)(TestDb *, int);
+ int (*xCommit)(TestDb *, int);
+ int (*xRollback)(TestDb *, int);
+};
+
+/*
+** Functions in wrapper2.cc (a C++ source file). wrapper2.cc contains the
+** wrapper for Kyoto Cabinet. Kyoto cabinet has a C API, but
+** the primary interface is the C++ API.
+*/
+int test_kc_open(const char*, const char *zFilename, int bClear, TestDb **ppDb);
+int test_kc_close(TestDb *);
+int test_kc_write(TestDb *, void *, int , void *, int);
+int test_kc_delete(TestDb *, void *, int);
+int test_kc_delete_range(TestDb *, void *, int, void *, int);
+int test_kc_fetch(TestDb *, void *, int, void **, int *);
+int test_kc_scan(TestDb *, void *, int, void *, int, void *, int,
+ void (*)(void *, void *, int , void *, int)
+);
+
+int test_mdb_open(const char*, const char *zFile, int bClear, TestDb **ppDb);
+int test_mdb_close(TestDb *);
+int test_mdb_write(TestDb *, void *, int , void *, int);
+int test_mdb_delete(TestDb *, void *, int);
+int test_mdb_fetch(TestDb *, void *, int, void **, int *);
+int test_mdb_scan(TestDb *, void *, int, void *, int, void *, int,
+ void (*)(void *, void *, int , void *, int)
+);
+
+/*
+** Functions in wrapper3.c. This file contains the tdb wrapper for lsm.
+** The wrapper for lsm is a bit more involved than the others, as it
+** includes code for a couple of different lsm configurations, and for
+** various types of fault injection and robustness testing.
+*/
+int test_lsm_open(const char*, const char *zFile, int bClear, TestDb **ppDb);
+int test_lsm_lomem_open(const char*, const char*, int bClear, TestDb **ppDb);
+int test_lsm_lomem2_open(const char*, const char*, int bClear, TestDb **ppDb);
+int test_lsm_zip_open(const char*, const char*, int bClear, TestDb **ppDb);
+int test_lsm_small_open(const char*, const char*, int bClear, TestDb **ppDb);
+int test_lsm_mt2(const char*, const char *zFile, int bClear, TestDb **ppDb);
+int test_lsm_mt3(const char*, const char *zFile, int bClear, TestDb **ppDb);
+
+int tdb_lsm_configure(lsm_db *, const char *);
+
+/* Functions in lsmtest_tdb4.c */
+int test_bt_open(const char*, const char *zFile, int bClear, TestDb **ppDb);
+int test_fbt_open(const char*, const char *zFile, int bClear, TestDb **ppDb);
+int test_fbts_open(const char*, const char *zFile, int bClear, TestDb **ppDb);
+
+
+/* Functions in testutil.c. */
+int testPrngInit(void);
+u32 testPrngValue(u32 iVal);
+void testPrngArray(u32 iVal, u32 *aOut, int nOut);
+void testPrngString(u32 iVal, char *aOut, int nOut);
+
+void testErrorInit(int argc, char **);
+void testPrintError(const char *zFormat, ...);
+void testPrintUsage(const char *zArgs);
+void testPrintFUsage(const char *zFormat, ...);
+void testTimeInit(void);
+int testTimeGet(void);
+
+/* Functions in testmem.c. */
+void testMallocInstall(lsm_env *pEnv);
+void testMallocUninstall(lsm_env *pEnv);
+void testMallocCheck(lsm_env *pEnv, int *, int *, FILE *);
+void testMallocOom(lsm_env *pEnv, int, int, void(*)(void*), void *);
+void testMallocOomEnable(lsm_env *pEnv, int);
+
+/* lsmtest.c */
+TestDb *testOpen(const char *zSystem, int, int *pRc);
+void testReopen(TestDb **ppDb, int *pRc);
+void testClose(TestDb **ppDb);
+
+void testFetch(TestDb *, void *, int, void *, int, int *);
+void testWrite(TestDb *, void *, int, void *, int, int *);
+void testDelete(TestDb *, void *, int, int *);
+void testDeleteRange(TestDb *, void *, int, void *, int, int *);
+void testWriteStr(TestDb *, const char *, const char *zVal, int *pRc);
+void testFetchStr(TestDb *, const char *, const char *, int *pRc);
+
+void testBegin(TestDb *pDb, int iTrans, int *pRc);
+void testCommit(TestDb *pDb, int iTrans, int *pRc);
+
+void test_failed(void);
+
+char *testMallocPrintf(const char *zFormat, ...);
+char *testMallocVPrintf(const char *zFormat, va_list ap);
+int testGlobMatch(const char *zPattern, const char *zStr);
+
+void testScanCompare(TestDb *, TestDb *, int, void *, int, void *, int, int *);
+void testFetchCompare(TestDb *, TestDb *, void *, int, int *);
+
+void *testMalloc(int);
+void *testMallocCopy(void *pCopy, int nByte);
+void *testRealloc(void *, int);
+void testFree(void *);
+
+/* lsmtest_bt.c */
+int do_bt(int nArg, char **azArg);
+
+/* testio.c */
+int testVfsConfigureDb(TestDb *pDb);
+
+/* testfunc.c */
+int do_show(int nArg, char **azArg);
+int do_work(int nArg, char **azArg);
+
+/* testio.c */
+int do_io(int nArg, char **azArg);
+
+/* lsmtest2.c */
+void do_crash_test(const char *zPattern, int *pRc);
+int do_rollback_test(int nArg, char **azArg);
+
+/* test3.c */
+void test_rollback(const char *zSystem, const char *zPattern, int *pRc);
+
+/* test4.c */
+void test_mc(const char *zSystem, const char *zPattern, int *pRc);
+
+/* test5.c */
+void test_mt(const char *zSystem, const char *zPattern, int *pRc);
+
+/* lsmtest6.c */
+void test_oom(const char *zPattern, int *pRc);
+void testDeleteLsmdb(const char *zFile);
+
+void testSaveDb(const char *zFile, const char *zAuxExt);
+void testRestoreDb(const char *zFile, const char *zAuxExt);
+void testCopyLsmdb(const char *zFrom, const char *zTo);
+
+/* lsmtest7.c */
+void test_api(const char *zPattern, int *pRc);
+
+/* lsmtest8.c */
+void do_writer_crash_test(const char *zPattern, int *pRc);
+
+/*************************************************************************
+** Interface to functionality in test_datasource.c.
+*/
+typedef struct Datasource Datasource;
+typedef struct DatasourceDefn DatasourceDefn;
+
+struct DatasourceDefn {
+ int eType; /* A TEST_DATASOURCE_* value */
+ int nMinKey; /* Minimum key size */
+ int nMaxKey; /* Maximum key size */
+ int nMinVal; /* Minimum value size */
+ int nMaxVal; /* Maximum value size */
+};
+
+#define TEST_DATASOURCE_RANDOM 1
+#define TEST_DATASOURCE_SEQUENCE 2
+
+char *testDatasourceName(const DatasourceDefn *);
+Datasource *testDatasourceNew(const DatasourceDefn *);
+void testDatasourceFree(Datasource *);
+void testDatasourceEntry(Datasource *, int, void **, int *, void **, int *);
+/* End of test_datasource.c interface.
+*************************************************************************/
+void testDatasourceFetch(
+ TestDb *pDb, /* Database handle */
+ Datasource *pData,
+ int iKey,
+ int *pRc /* IN/OUT: Error code */
+);
+
+void testWriteDatasource(TestDb *, Datasource *, int, int *);
+void testWriteDatasourceRange(TestDb *, Datasource *, int, int, int *);
+void testDeleteDatasource(TestDb *, Datasource *, int, int *);
+void testDeleteDatasourceRange(TestDb *, Datasource *, int, int, int *);
+
+
+/* test1.c */
+void test_data_1(const char *, const char *, int *pRc);
+void test_data_2(const char *, const char *, int *pRc);
+void test_data_3(const char *, const char *, int *pRc);
+void testDbContents(TestDb *, Datasource *, int, int, int, int, int, int *);
+void testCaseProgress(int, int, int, int *);
+int testCaseNDot(void);
+
+void testCompareDb(Datasource *, int, int, TestDb *, TestDb *, int *);
+int testControlDb(TestDb **ppDb);
+
+typedef struct CksumDb CksumDb;
+CksumDb *testCksumArrayNew(Datasource *, int, int, int);
+char *testCksumArrayGet(CksumDb *, int);
+void testCksumArrayFree(CksumDb *);
+void testCaseStart(int *pRc, char *zFmt, ...);
+void testCaseFinish(int rc);
+void testCaseSkip(void);
+int testCaseBegin(int *, const char *, const char *, ...);
+
+#define TEST_CKSUM_BYTES 29
+int testCksumDatabase(TestDb *pDb, char *zOut);
+int testCountDatabase(TestDb *pDb);
+void testCompareInt(int, int, int *);
+void testCompareStr(const char *z1, const char *z2, int *pRc);
+
+/* lsmtest9.c */
+void test_data_4(const char *, const char *, int *pRc);
+
+
+/*
+** Similar to the Tcl_GetIndexFromObjStruct() Tcl library function.
+*/
+#define testArgSelect(w,x,y,z) testArgSelectX(w,x,sizeof(w[0]),y,z)
+int testArgSelectX(void *, const char *, int, const char *, int *);
+
+#ifdef __cplusplus
+} /* End of the 'extern "C"' block */
+#endif
+
+#endif
diff --git a/ext/lsm1/lsm-test/lsmtest1.c b/ext/lsm1/lsm-test/lsmtest1.c
new file mode 100644
index 0000000..1ce2cc0
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest1.c
@@ -0,0 +1,656 @@
+
+#include "lsmtest.h"
+
+#define DATA_SEQUENTIAL TEST_DATASOURCE_SEQUENCE
+#define DATA_RANDOM TEST_DATASOURCE_RANDOM
+
+typedef struct Datatest1 Datatest1;
+typedef struct Datatest2 Datatest2;
+
+/*
+** An instance of the following structure contains parameters used to
+** customize the test function in this file. Test procedure:
+**
+** 1. Create a data-source based on the "datasource definition" vars.
+**
+** 2. Insert nRow key value pairs into the database.
+**
+** 3. Delete all keys from the database. Deletes are done in the same
+** order as the inserts.
+**
+** During steps 2 and 3 above, after each Datatest1.nVerify inserts or
+** deletes, the following:
+**
+** a. Run Datasource.nTest key lookups and check the results are as expected.
+**
+** b. If Datasource.bTestScan is true, run a handful (8) of range
+** queries (scanning forwards and backwards). Check that the results
+** are as expected.
+**
+** c. Close and reopen the database. Then run (a) and (b) again.
+*/
+struct Datatest1 {
+ /* Datasource definition */
+ DatasourceDefn defn;
+
+ /* Test procedure parameters */
+ int nRow; /* Number of rows to insert then delete */
+ int nVerify; /* How often to verify the db contents */
+ int nTest; /* Number of keys to test (0==all) */
+ int bTestScan; /* True to do scan tests */
+};
+
+/*
+** An instance of the following data structure is used to describe the
+** second type of test case in this file. The chief difference between
+** these tests and those described by Datatest1 is that these tests also
+** experiment with range-delete operations. Tests proceed as follows:
+**
+** 1. Open the datasource described by Datatest2.defn.
+**
+** 2. Open a connection on an empty database.
+**
+** 3. Do this Datatest2.nIter times:
+**
+** a) Insert Datatest2.nWrite key-value pairs from the datasource.
+**
+** b) Select two pseudo-random keys and use them as the start
+** and end points of a range-delete operation.
+**
+** c) Verify that the contents of the database are as expected (see
+** below for details).
+**
+** d) Close and then reopen the database handle.
+**
+** e) Verify that the contents of the database are still as expected.
+**
+** The inserts and range deletes are run twice - once on the database being
+** tested and once using a control system (sqlite3, kc etc. - something that
+** works). In order to verify that the contents of the db being tested are
+** correct, the test runs a bunch of scans and lookups on both the test and
+** control databases. If the results are the same, the test passes.
+*/
+struct Datatest2 {
+ DatasourceDefn defn;
+ int nRange;
+ int nWrite; /* Number of writes per iteration */
+ int nIter; /* Total number of iterations to run */
+};
+
+/*
+** Generate a unique name for the test case pTest with database system
+** zSystem.
+*/
+static char *getName(const char *zSystem, int bRecover, Datatest1 *pTest){
+ char *zRet;
+ char *zData;
+ zData = testDatasourceName(&pTest->defn);
+ zRet = testMallocPrintf("data.%s.%s.rec=%d.%d.%d",
+ zSystem, zData, bRecover, pTest->nRow, pTest->nVerify
+ );
+ testFree(zData);
+ return zRet;
+}
+
+int testControlDb(TestDb **ppDb){
+#ifdef HAVE_KYOTOCABINET
+ return tdb_open("kyotocabinet", "tmp.db", 1, ppDb);
+#else
+ return tdb_open("sqlite3", "", 1, ppDb);
+#endif
+}
+
+void testDatasourceFetch(
+ TestDb *pDb, /* Database handle */
+ Datasource *pData,
+ int iKey,
+ int *pRc /* IN/OUT: Error code */
+){
+ void *pKey; int nKey; /* Database key to query for */
+ void *pVal; int nVal; /* Expected result of query */
+
+ testDatasourceEntry(pData, iKey, &pKey, &nKey, &pVal, &nVal);
+ testFetch(pDb, pKey, nKey, pVal, nVal, pRc);
+}
+
+/*
+** This function is called to test that the contents of database pDb
+** are as expected. In this case, expected is defined as containing
+** key-value pairs iFirst through iLast, inclusive, from data source
+** pData. In other words, a loop like the following could be used to
+** construct a database with identical contents from scratch.
+**
+** for(i=iFirst; i<=iLast; i++){
+** testDatasourceEntry(pData, i, &pKey, &nKey, &pVal, &nVal);
+** // insert (pKey, nKey) -> (pVal, nVal) into database
+** }
+**
+** The key domain consists of keys 0 to (nRow-1), inclusive, from
+** data source pData. For both scan and lookup tests, keys are selected
+** pseudo-randomly from within this set.
+**
+** This function runs nLookupTest lookup tests and nScanTest scan tests.
+**
+** A lookup test consists of selecting a key from the domain and querying
+** pDb for it. The test fails if the presence of the key and, if present,
+** the associated value do not match the expectations defined above.
+**
+** A scan test involves selecting a key from the domain and running
+** the following queries:
+**
+** 1. Scan all keys equal to or greater than the key, in ascending order.
+** 2. Scan all keys equal to or smaller than the key, in descending order.
+**
+** Additionally, if nLookupTest is greater than zero, the following are
+** run once:
+**
+** 1. Scan all keys in the db, in ascending order.
+** 2. Scan all keys in the db, in descending order.
+**
+** As you would assume, the test fails if the returned values do not match
+** expectations.
+*/
+void testDbContents(
+ TestDb *pDb, /* Database handle being tested */
+ Datasource *pData, /* pDb contains data from here */
+ int nRow, /* Size of key domain */
+ int iFirst, /* Index of first key from pData in pDb */
+ int iLast, /* Index of last key from pData in pDb */
+ int nLookupTest, /* Number of lookup tests to run */
+ int nScanTest, /* Number of scan tests to run */
+ int *pRc /* IN/OUT: Error code */
+){
+ int j;
+ int rc = *pRc;
+
+ if( rc==0 && nScanTest ){
+ TestDb *pDb2 = 0;
+
+ /* Open a control db (i.e. one that we assume works) */
+ rc = testControlDb(&pDb2);
+
+ for(j=iFirst; rc==0 && j<=iLast; j++){
+ void *pKey; int nKey; /* Database key to insert */
+ void *pVal; int nVal; /* Database value to insert */
+ testDatasourceEntry(pData, j, &pKey, &nKey, &pVal, &nVal);
+ rc = tdb_write(pDb2, pKey, nKey, pVal, nVal);
+ }
+
+ if( rc==0 ){
+ int iKey1;
+ int iKey2;
+ void *pKey1; int nKey1; /* Start key */
+ void *pKey2; int nKey2; /* Final key */
+
+ iKey1 = testPrngValue((iFirst<<8) + (iLast<<16)) % nRow;
+ iKey2 = testPrngValue((iLast<<8) + (iFirst<<16)) % nRow;
+ testDatasourceEntry(pData, iKey1, &pKey2, &nKey1, 0, 0);
+ pKey1 = testMalloc(nKey1+1);
+ memcpy(pKey1, pKey2, nKey1+1);
+ testDatasourceEntry(pData, iKey2, &pKey2, &nKey2, 0, 0);
+
+ testScanCompare(pDb2, pDb, 0, 0, 0, 0, 0, &rc);
+ testScanCompare(pDb2, pDb, 0, 0, 0, pKey2, nKey2, &rc);
+ testScanCompare(pDb2, pDb, 0, pKey1, nKey1, 0, 0, &rc);
+ testScanCompare(pDb2, pDb, 0, pKey1, nKey1, pKey2, nKey2, &rc);
+ testScanCompare(pDb2, pDb, 1, 0, 0, 0, 0, &rc);
+ testScanCompare(pDb2, pDb, 1, 0, 0, pKey2, nKey2, &rc);
+ testScanCompare(pDb2, pDb, 1, pKey1, nKey1, 0, 0, &rc);
+ testScanCompare(pDb2, pDb, 1, pKey1, nKey1, pKey2, nKey2, &rc);
+ testFree(pKey1);
+ }
+ tdb_close(pDb2);
+ }
+
+ /* Test some lookups. */
+ for(j=0; rc==0 && j<nLookupTest; j++){
+ int iKey; /* Datasource key to test */
+ void *pKey; int nKey; /* Database key to query for */
+ void *pVal; int nVal; /* Expected result of query */
+
+ if( nLookupTest>=nRow ){
+ iKey = j;
+ }else{
+ iKey = testPrngValue(j + (iFirst<<8) + (iLast<<16)) % nRow;
+ }
+
+ testDatasourceEntry(pData, iKey, &pKey, &nKey, &pVal, &nVal);
+ if( iFirst>iKey || iKey>iLast ){
+ pVal = 0;
+ nVal = -1;
+ }
+
+ testFetch(pDb, pKey, nKey, pVal, nVal, &rc);
+ }
+
+ *pRc = rc;
+}
+
+/*
+** This function should be called during long running test cases to output
+** the progress dots (...) to stdout.
+*/
+void testCaseProgress(int i, int n, int nDot, int *piDot){
+ int iDot = *piDot;
+ while( iDot < ( ((nDot*2+1) * i) / (n*2) ) ){
+ printf(".");
+ fflush(stdout);
+ iDot++;
+ }
+ *piDot = iDot;
+}
+
+int testCaseNDot(void){ return 20; }
+
+#if 0
+static void printScanCb(
+ void *pCtx, void *pKey, int nKey, void *pVal, int nVal
+){
+ printf("%s\n", (char *)pKey);
+ fflush(stdout);
+}
+#endif
+
+void testReopenRecover(TestDb **ppDb, int *pRc){
+ if( *pRc==0 ){
+ const char *zLib = tdb_library_name(*ppDb);
+ const char *zDflt = tdb_default_db(zLib);
+ testCopyLsmdb(zDflt, "bak.db");
+ testClose(ppDb);
+ testCopyLsmdb("bak.db", zDflt);
+ *pRc = tdb_open(zLib, 0, 0, ppDb);
+ }
+}
+
+
+static void doDataTest1(
+ const char *zSystem, /* Database system to test */
+ int bRecover,
+ Datatest1 *p, /* Structure containing test parameters */
+ int *pRc /* OUT: Error code */
+){
+ int i;
+ int iDot;
+ int rc = LSM_OK;
+ Datasource *pData;
+ TestDb *pDb;
+ int iToggle = 0;
+
+ /* Start the test case, open a database and allocate the datasource. */
+ pDb = testOpen(zSystem, 1, &rc);
+ pData = testDatasourceNew(&p->defn);
+
+ i = 0;
+ iDot = 0;
+ while( rc==LSM_OK && i<p->nRow ){
+
+ /* Insert some data */
+ testWriteDatasourceRange(pDb, pData, i, p->nVerify, &rc);
+ i += p->nVerify;
+
+ if( iToggle ) testBegin(pDb, 1, &rc);
+ /* Check that the db content is correct. */
+ testDbContents(pDb, pData, p->nRow, 0, i-1, p->nTest, p->bTestScan, &rc);
+ if( iToggle ) testCommit(pDb, 0, &rc);
+ iToggle = (iToggle+1)%2;
+
+ if( bRecover ){
+ testReopenRecover(&pDb, &rc);
+ }else{
+ testReopen(&pDb, &rc);
+ }
+
+ /* Check that the db content is still correct. */
+ testDbContents(pDb, pData, p->nRow, 0, i-1, p->nTest, p->bTestScan, &rc);
+
+ /* Update the progress dots... */
+ testCaseProgress(i, p->nRow, testCaseNDot()/2, &iDot);
+ }
+
+ i = 0;
+ iDot = 0;
+ while( rc==LSM_OK && i<p->nRow ){
+
+ /* Delete some entries */
+ testDeleteDatasourceRange(pDb, pData, i, p->nVerify, &rc);
+ i += p->nVerify;
+
+ /* Check that the db content is correct. */
+ testDbContents(pDb, pData, p->nRow, i, p->nRow-1,p->nTest,p->bTestScan,&rc);
+
+ /* Close and reopen the database. */
+ if( bRecover ){
+ testReopenRecover(&pDb, &rc);
+ }else{
+ testReopen(&pDb, &rc);
+ }
+
+ /* Check that the db content is still correct. */
+ testDbContents(pDb, pData, p->nRow, i, p->nRow-1,p->nTest,p->bTestScan,&rc);
+
+ /* Update the progress dots... */
+ testCaseProgress(i, p->nRow, testCaseNDot()/2, &iDot);
+ }
+
+ /* Free the datasource, close the database and finish the test case. */
+ testDatasourceFree(pData);
+ tdb_close(pDb);
+ testCaseFinish(rc);
+ *pRc = rc;
+}
+
+
+void test_data_1(
+ const char *zSystem, /* Database system name */
+ const char *zPattern, /* Run test cases that match this pattern */
+ int *pRc /* IN/OUT: Error code */
+){
+ Datatest1 aTest[] = {
+ { {DATA_RANDOM, 500,600, 1000,2000}, 1000, 100, 10, 0},
+ { {DATA_RANDOM, 20,25, 100,200}, 1000, 250, 1000, 1},
+ { {DATA_RANDOM, 8,10, 100,200}, 1000, 250, 1000, 1},
+ { {DATA_RANDOM, 8,10, 10,20}, 1000, 250, 1000, 1},
+ { {DATA_RANDOM, 8,10, 1000,2000}, 1000, 250, 1000, 1},
+ { {DATA_RANDOM, 8,100, 10000,20000}, 100, 25, 100, 1},
+ { {DATA_RANDOM, 80,100, 10,20}, 1000, 250, 1000, 1},
+ { {DATA_RANDOM, 5000,6000, 10,20}, 100, 25, 100, 1},
+ { {DATA_SEQUENTIAL, 5,10, 10,20}, 1000, 250, 1000, 1},
+ { {DATA_SEQUENTIAL, 5,10, 100,200}, 1000, 250, 1000, 1},
+ { {DATA_SEQUENTIAL, 5,10, 1000,2000}, 1000, 250, 1000, 1},
+ { {DATA_SEQUENTIAL, 5,100, 10000,20000}, 100, 25, 100, 1},
+ { {DATA_RANDOM, 10,10, 100,100}, 100000, 1000, 100, 0},
+ { {DATA_SEQUENTIAL, 10,10, 100,100}, 100000, 1000, 100, 0},
+ };
+
+ int i;
+ int bRecover;
+
+ for(bRecover=0; bRecover<2; bRecover++){
+ if( bRecover==1 && memcmp(zSystem, "lsm", 3) ) break;
+ for(i=0; *pRc==LSM_OK && i<ArraySize(aTest); i++){
+ char *zName = getName(zSystem, bRecover, &aTest[i]);
+ if( testCaseBegin(pRc, zPattern, "%s", zName) ){
+ doDataTest1(zSystem, bRecover, &aTest[i], pRc);
+ }
+ testFree(zName);
+ }
+ }
+}
+
+void testCompareDb(
+ Datasource *pData,
+ int nData,
+ int iSeed,
+ TestDb *pControl,
+ TestDb *pDb,
+ int *pRc
+){
+ int i;
+
+ static int nCall = 0;
+ nCall++;
+
+ testScanCompare(pControl, pDb, 0, 0, 0, 0, 0, pRc);
+ testScanCompare(pControl, pDb, 1, 0, 0, 0, 0, pRc);
+
+ if( *pRc==0 ){
+ int iKey1;
+ int iKey2;
+ void *pKey1; int nKey1; /* Start key */
+ void *pKey2; int nKey2; /* Final key */
+
+ iKey1 = testPrngValue(iSeed) % nData;
+ iKey2 = testPrngValue(iSeed+1) % nData;
+ testDatasourceEntry(pData, iKey1, &pKey2, &nKey1, 0, 0);
+ pKey1 = testMalloc(nKey1+1);
+ memcpy(pKey1, pKey2, nKey1+1);
+ testDatasourceEntry(pData, iKey2, &pKey2, &nKey2, 0, 0);
+
+ testScanCompare(pControl, pDb, 0, 0, 0, pKey2, nKey2, pRc);
+ testScanCompare(pControl, pDb, 0, pKey1, nKey1, 0, 0, pRc);
+ testScanCompare(pControl, pDb, 0, pKey1, nKey1, pKey2, nKey2, pRc);
+ testScanCompare(pControl, pDb, 1, 0, 0, pKey2, nKey2, pRc);
+ testScanCompare(pControl, pDb, 1, pKey1, nKey1, 0, 0, pRc);
+ testScanCompare(pControl, pDb, 1, pKey1, nKey1, pKey2, nKey2, pRc);
+ testFree(pKey1);
+ }
+
+ for(i=0; i<nData && *pRc==0; i++){
+ void *pKey; int nKey;
+ testDatasourceEntry(pData, i, &pKey, &nKey, 0, 0);
+ testFetchCompare(pControl, pDb, pKey, nKey, pRc);
+ }
+}
+
+static void doDataTest2(
+ const char *zSystem, /* Database system to test */
+ int bRecover,
+ Datatest2 *p, /* Structure containing test parameters */
+ int *pRc /* OUT: Error code */
+){
+ TestDb *pDb;
+ TestDb *pControl;
+ Datasource *pData;
+ int i;
+ int rc = LSM_OK;
+ int iDot = 0;
+
+ /* Start the test case, open a database and allocate the datasource. */
+ pDb = testOpen(zSystem, 1, &rc);
+ pData = testDatasourceNew(&p->defn);
+ rc = testControlDb(&pControl);
+
+ if( tdb_lsm(pDb) ){
+ int nBuf = 32 * 1024 * 1024;
+ lsm_config(tdb_lsm(pDb), LSM_CONFIG_AUTOFLUSH, &nBuf);
+ }
+
+ for(i=0; rc==0 && i<p->nIter; i++){
+ void *pKey1; int nKey1;
+ void *pKey2; int nKey2;
+ int ii;
+ int nRange = MIN(p->nIter*p->nWrite, p->nRange);
+
+ for(ii=0; rc==0 && ii<p->nWrite; ii++){
+ int iKey = (i*p->nWrite + ii) % p->nRange;
+ testWriteDatasource(pControl, pData, iKey, &rc);
+ testWriteDatasource(pDb, pData, iKey, &rc);
+ }
+
+ testDatasourceEntry(pData, i+1000000, &pKey1, &nKey1, 0, 0);
+ pKey1 = testMallocCopy(pKey1, nKey1);
+ testDatasourceEntry(pData, i+2000000, &pKey2, &nKey2, 0, 0);
+
+ testDeleteRange(pDb, pKey1, nKey1, pKey2, nKey2, &rc);
+ testDeleteRange(pControl, pKey1, nKey1, pKey2, nKey2, &rc);
+ testFree(pKey1);
+
+ testCompareDb(pData, nRange, i, pControl, pDb, &rc);
+ if( bRecover ){
+ testReopenRecover(&pDb, &rc);
+ }else{
+ testReopen(&pDb, &rc);
+ }
+ testCompareDb(pData, nRange, i, pControl, pDb, &rc);
+
+ /* Update the progress dots... */
+ testCaseProgress(i, p->nIter, testCaseNDot(), &iDot);
+ }
+
+ testClose(&pDb);
+ testClose(&pControl);
+ testDatasourceFree(pData);
+ testCaseFinish(rc);
+ *pRc = rc;
+}
+
+static char *getName2(const char *zSystem, int bRecover, Datatest2 *pTest){
+ char *zRet;
+ char *zData;
+ zData = testDatasourceName(&pTest->defn);
+ zRet = testMallocPrintf("data2.%s.%s.rec=%d.%d.%d.%d",
+ zSystem, zData, bRecover, pTest->nRange, pTest->nWrite, pTest->nIter
+ );
+ testFree(zData);
+ return zRet;
+}
+
+void test_data_2(
+ const char *zSystem, /* Database system name */
+ const char *zPattern, /* Run test cases that match this pattern */
+ int *pRc /* IN/OUT: Error code */
+){
+ Datatest2 aTest[] = {
+ /* defn, nRange, nWrite, nIter */
+ { {DATA_RANDOM, 20,25, 100,200}, 10000, 10, 50 },
+ { {DATA_RANDOM, 20,25, 100,200}, 10000, 200, 50 },
+ { {DATA_RANDOM, 20,25, 100,200}, 100, 10, 1000 },
+ { {DATA_RANDOM, 20,25, 100,200}, 100, 200, 50 },
+ };
+
+ int i;
+ int bRecover;
+
+ for(bRecover=0; bRecover<2; bRecover++){
+ if( bRecover==1 && memcmp(zSystem, "lsm", 3) ) break;
+ for(i=0; *pRc==LSM_OK && i<ArraySize(aTest); i++){
+ char *zName = getName2(zSystem, bRecover, &aTest[i]);
+ if( testCaseBegin(pRc, zPattern, "%s", zName) ){
+ doDataTest2(zSystem, bRecover, &aTest[i], pRc);
+ }
+ testFree(zName);
+ }
+ }
+}
+
+/*************************************************************************
+** Test case data3.*
+*/
+
+typedef struct Datatest3 Datatest3;
+struct Datatest3 {
+ int nRange; /* Keys are between 1 and this value, incl. */
+ int nIter; /* Number of iterations */
+ int nWrite; /* Number of writes per iteration */
+ int nDelete; /* Number of deletes per iteration */
+
+ int nValMin; /* Minimum value size for writes */
+ int nValMax; /* Maximum value size for writes */
+};
+
+void testPutU32(u8 *aBuf, u32 iVal){
+ aBuf[0] = (iVal >> 24) & 0xFF;
+ aBuf[1] = (iVal >> 16) & 0xFF;
+ aBuf[2] = (iVal >> 8) & 0xFF;
+ aBuf[3] = (iVal >> 0) & 0xFF;
+}
+
+void dt3PutKey(u8 *aBuf, int iKey){
+ assert( iKey<100000 && iKey>=0 );
+ sprintf((char *)aBuf, "%.5d", iKey);
+}
+
+static void doDataTest3(
+ const char *zSystem, /* Database system to test */
+ Datatest3 *p, /* Structure containing test parameters */
+ int *pRc /* OUT: Error code */
+){
+ int iDot = 0;
+ int rc = *pRc;
+ TestDb *pDb;
+ u8 *abPresent; /* Array of boolean */
+ char *aVal; /* Buffer to hold values */
+ int i;
+ u32 iSeq = 10; /* prng counter */
+
+ abPresent = (u8 *)testMalloc(p->nRange+1);
+ aVal = (char *)testMalloc(p->nValMax+1);
+ pDb = testOpen(zSystem, 1, &rc);
+
+ for(i=0; i<p->nIter && rc==0; i++){
+ int ii;
+
+ testCaseProgress(i, p->nIter, testCaseNDot(), &iDot);
+
+ /* Perform nWrite inserts */
+ for(ii=0; ii<p->nWrite; ii++){
+ u8 aKey[6];
+ u32 iKey;
+ int nVal;
+
+ iKey = (testPrngValue(iSeq++) % p->nRange) + 1;
+ nVal = (testPrngValue(iSeq++) % (p->nValMax - p->nValMin)) + p->nValMin;
+ testPrngString(testPrngValue(iSeq++), aVal, nVal);
+ dt3PutKey(aKey, iKey);
+
+ testWrite(pDb, aKey, sizeof(aKey)-1, aVal, nVal, &rc);
+ abPresent[iKey] = 1;
+ }
+
+ /* Perform nDelete deletes */
+ for(ii=0; ii<p->nDelete; ii++){
+ u8 aKey1[6];
+ u8 aKey2[6];
+ u32 iKey;
+
+ iKey = (testPrngValue(iSeq++) % p->nRange) + 1;
+ dt3PutKey(aKey1, iKey-1);
+ dt3PutKey(aKey2, iKey+1);
+
+ testDeleteRange(pDb, aKey1, sizeof(aKey1)-1, aKey2, sizeof(aKey2)-1, &rc);
+ abPresent[iKey] = 0;
+ }
+
+ testReopen(&pDb, &rc);
+
+ for(ii=1; rc==0 && ii<=p->nRange; ii++){
+ int nDbVal;
+ void *pDbVal;
+ u8 aKey[6];
+ int dbrc;
+
+ dt3PutKey(aKey, ii);
+ dbrc = tdb_fetch(pDb, aKey, sizeof(aKey)-1, &pDbVal, &nDbVal);
+ testCompareInt(0, dbrc, &rc);
+
+ if( abPresent[ii] ){
+ testCompareInt(1, (nDbVal>0), &rc);
+ }else{
+ testCompareInt(1, (nDbVal<0), &rc);
+ }
+ }
+ }
+
+ testClose(&pDb);
+ testCaseFinish(rc);
+ *pRc = rc;
+}
+
+static char *getName3(const char *zSystem, Datatest3 *p){
+ return testMallocPrintf("data3.%s.%d.%d.%d.%d.(%d..%d)",
+ zSystem, p->nRange, p->nIter, p->nWrite, p->nDelete,
+ p->nValMin, p->nValMax
+ );
+}
+
+void test_data_3(
+ const char *zSystem, /* Database system name */
+ const char *zPattern, /* Run test cases that match this pattern */
+ int *pRc /* IN/OUT: Error code */
+){
+ Datatest3 aTest[] = {
+ /* nRange, nIter, nWrite, nDelete, nValMin, nValMax */
+ { 100, 1000, 5, 5, 50, 100 },
+ { 100, 1000, 2, 2, 5, 10 },
+ };
+
+ int i;
+
+ for(i=0; *pRc==LSM_OK && i<ArraySize(aTest); i++){
+ char *zName = getName3(zSystem, &aTest[i]);
+ if( testCaseBegin(pRc, zPattern, "%s", zName) ){
+ doDataTest3(zSystem, &aTest[i], pRc);
+ }
+ testFree(zName);
+ }
+}
diff --git a/ext/lsm1/lsm-test/lsmtest2.c b/ext/lsm1/lsm-test/lsmtest2.c
new file mode 100644
index 0000000..d2ef0eb
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest2.c
@@ -0,0 +1,488 @@
+
+/*
+** This file contains tests related to recovery following application
+** and system crashes (power failures) while writing to the database.
+*/
+
+#include "lsmtest.h"
+
+/*
+** Structure used by testCksumDatabase() to accumulate checksum values in.
+*/
+typedef struct Cksum Cksum;
+struct Cksum {
+ int nRow;
+ int cksum1;
+ int cksum2;
+};
+
+/*
+** tdb_scan() callback used by testCksumDatabase()
+*/
+static void scanCksumDb(
+ void *pCtx,
+ void *pKey, int nKey,
+ void *pVal, int nVal
+){
+ Cksum *p = (Cksum *)pCtx;
+ int i;
+
+ p->nRow++;
+ for(i=0; i<nKey; i++){
+ p->cksum1 += ((u8 *)pKey)[i];
+ p->cksum2 += p->cksum1;
+ }
+ for(i=0; i<nVal; i++){
+ p->cksum1 += ((u8 *)pVal)[i];
+ p->cksum2 += p->cksum1;
+ }
+}
+
+/*
+** tdb_scan() callback used by testCountDatabase()
+*/
+static void scanCountDb(
+ void *pCtx,
+ void *pKey, int nKey,
+ void *pVal, int nVal
+){
+ Cksum *p = (Cksum *)pCtx;
+ p->nRow++;
+
+ unused_parameter(pKey);
+ unused_parameter(nKey);
+ unused_parameter(pVal);
+ unused_parameter(nVal);
+}
+
+
+/*
+** Iterate through the entire contents of database pDb. Write a checksum
+** string based on the db contents into buffer zOut before returning. A
+** checksum string is at most 29 (TEST_CKSUM_BYTES) bytes in size:
+**
+** * 32-bit integer (10 bytes)
+** * 1 space (1 byte)
+** * 32-bit hex (8 bytes)
+** * 1 space (1 byte)
+** * 32-bit hex (8 bytes)
+** * nul-terminator (1 byte)
+**
+** The number of entries in the database is returned.
+*/
+int testCksumDatabase(
+ TestDb *pDb, /* Database handle */
+ char *zOut /* Buffer to write checksum to */
+){
+ Cksum cksum;
+ memset(&cksum, 0, sizeof(Cksum));
+ tdb_scan(pDb, (void *)&cksum, 0, 0, 0, 0, 0, scanCksumDb);
+ sprintf(zOut, "%d %x %x",
+ cksum.nRow, (u32)cksum.cksum1, (u32)cksum.cksum2
+ );
+ assert( strlen(zOut)<TEST_CKSUM_BYTES );
+ return cksum.nRow;
+}
+
+int testCountDatabase(TestDb *pDb){
+ Cksum cksum;
+ memset(&cksum, 0, sizeof(Cksum));
+ tdb_scan(pDb, (void *)&cksum, 0, 0, 0, 0, 0, scanCountDb);
+ return cksum.nRow;
+}
+
+/*
+** This function is a no-op if *pRc is not 0 when it is called.
+**
+** Otherwise, the two nul-terminated strings z1 and z1 are compared. If
+** they are the same, the function returns without doing anything. Otherwise,
+** an error message is printed, *pRc is set to 1 and the test_failed()
+** function called.
+*/
+void testCompareStr(const char *z1, const char *z2, int *pRc){
+ if( *pRc==0 ){
+ if( strcmp(z1, z2) ){
+ testPrintError("testCompareStr: \"%s\" != \"%s\"\n", z1, z2);
+ *pRc = 1;
+ test_failed();
+ }
+ }
+}
+
+/*
+** This function is a no-op if *pRc is not 0 when it is called.
+**
+** Otherwise, the two integers i1 and i2 are compared. If they are equal,
+** the function returns without doing anything. Otherwise, an error message
+** is printed, *pRc is set to 1 and the test_failed() function called.
+*/
+void testCompareInt(int i1, int i2, int *pRc){
+ if( *pRc==0 && i1!=i2 ){
+ testPrintError("testCompareInt: %d != %d\n", i1, i2);
+ *pRc = 1;
+ test_failed();
+ }
+}
+
+void testCaseStart(int *pRc, char *zFmt, ...){
+ va_list ap;
+ va_start(ap, zFmt);
+ vprintf(zFmt, ap);
+ printf(" ...");
+ va_end(ap);
+ *pRc = 0;
+ fflush(stdout);
+}
+
+/*
+** This function is a no-op if *pRc is non-zero when it is called. Zero
+** is returned in this case.
+**
+** Otherwise, the zFmt (a printf style format string) and following arguments
+** are used to create a test case name. If zPattern is NULL or a glob pattern
+** that matches the test case name, 1 is returned and the test case started.
+** Otherwise, zero is returned and the test case does not start.
+*/
+int testCaseBegin(int *pRc, const char *zPattern, const char *zFmt, ...){
+ int res = 0;
+ if( *pRc==0 ){
+ char *zTest;
+ va_list ap;
+
+ va_start(ap, zFmt);
+ zTest = testMallocVPrintf(zFmt, ap);
+ va_end(ap);
+ if( zPattern==0 || testGlobMatch(zPattern, zTest) ){
+ printf("%-50s ...", zTest);
+ res = 1;
+ }
+ testFree(zTest);
+ fflush(stdout);
+ }
+
+ return res;
+}
+
+void testCaseFinish(int rc){
+ if( rc==0 ){
+ printf("Ok\n");
+ }else{
+ printf("FAILED\n");
+ }
+ fflush(stdout);
+}
+
+void testCaseSkip(){
+ printf("Skipped\n");
+}
+
+void testSetupSavedLsmdb(
+ const char *zCfg,
+ const char *zFile,
+ Datasource *pData,
+ int nRow,
+ int *pRc
+){
+ if( *pRc==0 ){
+ int rc;
+ TestDb *pDb;
+ rc = tdb_lsm_open(zCfg, zFile, 1, &pDb);
+ if( rc==0 ){
+ testWriteDatasourceRange(pDb, pData, 0, nRow, &rc);
+ testClose(&pDb);
+ if( rc==0 ) testSaveDb(zFile, "log");
+ }
+ *pRc = rc;
+ }
+}
+
+/*
+** This function is a no-op if *pRc is non-zero when it is called.
+**
+** Open the LSM database identified by zFile and compute its checksum
+** (a string, as returned by testCksumDatabase()). If the checksum is
+** identical to zExpect1 or, if it is not NULL, zExpect2, the test passes.
+** Otherwise, print an error message and set *pRc to 1.
+*/
+static void testCompareCksumLsmdb(
+ const char *zFile, /* Path to LSM database */
+ int bCompress, /* True if db is compressed */
+ const char *zExpect1, /* Expected checksum 1 */
+ const char *zExpect2, /* Expected checksum 2 (or NULL) */
+ int *pRc /* IN/OUT: Test case error code */
+){
+ if( *pRc==0 ){
+ char zCksum[TEST_CKSUM_BYTES];
+ TestDb *pDb;
+
+ *pRc = tdb_lsm_open((bCompress?"compression=1 mmap=0":""), zFile, 0, &pDb);
+ testCksumDatabase(pDb, zCksum);
+ testClose(&pDb);
+
+ if( *pRc==0 ){
+ int r1 = 0;
+ int r2 = -1;
+
+ r1 = strcmp(zCksum, zExpect1);
+ if( zExpect2 ) r2 = strcmp(zCksum, zExpect2);
+ if( r1 && r2 ){
+ if( zExpect2 ){
+ testPrintError("testCompareCksumLsmdb: \"%s\" != (\"%s\" OR \"%s\")",
+ zCksum, zExpect1, zExpect2
+ );
+ }else{
+ testPrintError("testCompareCksumLsmdb: \"%s\" != \"%s\"",
+ zCksum, zExpect1
+ );
+ }
+ *pRc = 1;
+ test_failed();
+ }
+ }
+ }
+}
+
+#if 0 /* not used */
+static void testCompareCksumBtdb(
+ const char *zFile, /* Path to LSM database */
+ const char *zExpect1, /* Expected checksum 1 */
+ const char *zExpect2, /* Expected checksum 2 (or NULL) */
+ int *pRc /* IN/OUT: Test case error code */
+){
+ if( *pRc==0 ){
+ char zCksum[TEST_CKSUM_BYTES];
+ TestDb *pDb;
+
+ *pRc = tdb_open("bt", zFile, 0, &pDb);
+ testCksumDatabase(pDb, zCksum);
+ testClose(&pDb);
+
+ if( *pRc==0 ){
+ int r1 = 0;
+ int r2 = -1;
+
+ r1 = strcmp(zCksum, zExpect1);
+ if( zExpect2 ) r2 = strcmp(zCksum, zExpect2);
+ if( r1 && r2 ){
+ if( zExpect2 ){
+ testPrintError("testCompareCksumLsmdb: \"%s\" != (\"%s\" OR \"%s\")",
+ zCksum, zExpect1, zExpect2
+ );
+ }else{
+ testPrintError("testCompareCksumLsmdb: \"%s\" != \"%s\"",
+ zCksum, zExpect1
+ );
+ }
+ *pRc = 1;
+ test_failed();
+ }
+ }
+ }
+}
+#endif /* not used */
+
+/* Above this point are reusable test routines. Not clear that they
+** should really be in this file.
+*************************************************************************/
+
+/*
+** This test verifies that if a system crash occurs while doing merge work
+** on the db, no data is lost.
+*/
+static void crash_test1(int bCompress, int *pRc){
+ const char *DBNAME = "testdb.lsm";
+ const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 12, 16, 200, 200};
+
+ const int nRow = 5000; /* Database size */
+ const int nIter = 200; /* Number of test iterations */
+ const int nWork = 20; /* Maximum lsm_work() calls per iteration */
+ const int nPage = 15; /* Pages per lsm_work call */
+
+ int i;
+ int iDot = 0;
+ Datasource *pData;
+ CksumDb *pCksumDb;
+ TestDb *pDb;
+ char *zCfg;
+
+ const char *azConfig[2] = {
+ "page_size=1024 block_size=65536 autoflush=16384 safety=2 mmap=0",
+ "page_size=1024 block_size=65536 autoflush=16384 safety=2 "
+ " compression=1 mmap=0"
+ };
+ assert( bCompress==0 || bCompress==1 );
+
+ /* Allocate datasource. And calculate the expected checksums. */
+ pData = testDatasourceNew(&defn);
+ pCksumDb = testCksumArrayNew(pData, nRow, nRow, 1);
+
+ /* Setup and save the initial database. */
+
+ zCfg = testMallocPrintf("%s automerge=7", azConfig[bCompress]);
+ testSetupSavedLsmdb(zCfg, DBNAME, pData, 5000, pRc);
+ testFree(zCfg);
+
+ for(i=0; i<nIter && *pRc==0; i++){
+ int iWork;
+ int testrc = 0;
+
+ testCaseProgress(i, nIter, testCaseNDot(), &iDot);
+
+ /* Restore and open the database. */
+ testRestoreDb(DBNAME, "log");
+ testrc = tdb_lsm_open(azConfig[bCompress], DBNAME, 0, &pDb);
+ assert( testrc==0 );
+
+ /* Call lsm_work() on the db */
+ tdb_lsm_prepare_sync_crash(pDb, 1 + (i%(nWork*2)));
+ for(iWork=0; testrc==0 && iWork<nWork; iWork++){
+ int nWrite = 0;
+ lsm_db *db = tdb_lsm(pDb);
+ testrc = lsm_work(db, 0, nPage, &nWrite);
+ /* assert( testrc!=0 || nWrite>0 ); */
+ if( testrc==0 ) testrc = lsm_checkpoint(db, 0);
+ }
+ tdb_close(pDb);
+
+ /* Check that the database content is still correct */
+ testCompareCksumLsmdb(DBNAME,
+ bCompress, testCksumArrayGet(pCksumDb, nRow), 0, pRc);
+ }
+
+ testCksumArrayFree(pCksumDb);
+ testDatasourceFree(pData);
+}
+
+/*
+** This test verifies that if a system crash occurs while committing a
+** transaction to the log file, no earlier transactions are lost or damaged.
+*/
+static void crash_test2(int bCompress, int *pRc){
+ const char *DBNAME = "testdb.lsm";
+ const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 12, 16, 1000, 1000};
+
+ const int nIter = 200;
+ const int nInsert = 20;
+
+ int i;
+ int iDot = 0;
+ Datasource *pData;
+ CksumDb *pCksumDb;
+ TestDb *pDb;
+
+ /* Allocate datasource. And calculate the expected checksums. */
+ pData = testDatasourceNew(&defn);
+ pCksumDb = testCksumArrayNew(pData, 100, 100+nInsert, 1);
+
+ /* Setup and save the initial database. */
+ testSetupSavedLsmdb("", DBNAME, pData, 100, pRc);
+
+ for(i=0; i<nIter && *pRc==0; i++){
+ int iIns;
+ int testrc = 0;
+
+ testCaseProgress(i, nIter, testCaseNDot(), &iDot);
+
+ /* Restore and open the database. */
+ testRestoreDb(DBNAME, "log");
+ testrc = tdb_lsm_open("safety=2", DBNAME, 0, &pDb);
+ assert( testrc==0 );
+
+ /* Insert nInsert records into the database. Crash midway through. */
+ tdb_lsm_prepare_sync_crash(pDb, 1 + (i%(nInsert+2)));
+ for(iIns=0; iIns<nInsert; iIns++){
+ void *pKey; int nKey;
+ void *pVal; int nVal;
+
+ testDatasourceEntry(pData, 100+iIns, &pKey, &nKey, &pVal, &nVal);
+ testrc = tdb_write(pDb, pKey, nKey, pVal, nVal);
+ if( testrc ) break;
+ }
+ tdb_close(pDb);
+
+ /* Check that no data was lost when the system crashed. */
+ testCompareCksumLsmdb(DBNAME, bCompress,
+ testCksumArrayGet(pCksumDb, 100 + iIns),
+ testCksumArrayGet(pCksumDb, 100 + iIns + 1),
+ pRc
+ );
+ }
+
+ testDatasourceFree(pData);
+ testCksumArrayFree(pCksumDb);
+}
+
+
+/*
+** This test verifies that if a system crash occurs when checkpointing
+** the database, data is not lost (assuming that any writes not synced
+** to the db have been synced into the log file).
+*/
+static void crash_test3(int bCompress, int *pRc){
+ const char *DBNAME = "testdb.lsm";
+ const int nIter = 100;
+ const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 12, 16, 1000, 1000};
+
+ int i;
+ int iDot = 0;
+ Datasource *pData;
+ CksumDb *pCksumDb;
+ TestDb *pDb;
+
+ /* Allocate datasource. And calculate the expected checksums. */
+ pData = testDatasourceNew(&defn);
+ pCksumDb = testCksumArrayNew(pData, 110, 150, 10);
+
+ /* Setup and save the initial database. */
+ testSetupSavedLsmdb("", DBNAME, pData, 100, pRc);
+
+ for(i=0; i<nIter && *pRc==0; i++){
+ int iOpen;
+ testCaseProgress(i, nIter, testCaseNDot(), &iDot);
+ testRestoreDb(DBNAME, "log");
+
+ for(iOpen=0; iOpen<5; iOpen++){
+ /* Open the database. Insert 10 more records. */
+ pDb = testOpen("lsm", 0, pRc);
+ testWriteDatasourceRange(pDb, pData, 100+iOpen*10, 10, pRc);
+
+ /* Schedule a crash simulation then close the db. */
+ tdb_lsm_prepare_sync_crash(pDb, 1 + (i%2));
+ tdb_close(pDb);
+
+ /* Open the database and check that the crash did not cause any
+ ** data loss. */
+ testCompareCksumLsmdb(DBNAME, bCompress,
+ testCksumArrayGet(pCksumDb, 110 + iOpen*10), 0,
+ pRc
+ );
+ }
+ }
+
+ testDatasourceFree(pData);
+ testCksumArrayFree(pCksumDb);
+}
+
+void do_crash_test(const char *zPattern, int *pRc){
+ struct Test {
+ const char *zTest;
+ void (*x)(int, int *);
+ int bCompress;
+ } aTest [] = {
+ { "crash.lsm.1", crash_test1, 0 },
+#ifdef HAVE_ZLIB
+ { "crash.lsm_zip.1", crash_test1, 1 },
+#endif
+ { "crash.lsm.2", crash_test2, 0 },
+ { "crash.lsm.3", crash_test3, 0 },
+ };
+ int i;
+
+ for(i=0; *pRc==LSM_OK && i<ArraySize(aTest); i++){
+ struct Test *p = &aTest[i];
+ if( testCaseBegin(pRc, zPattern, "%s", p->zTest) ){
+ p->x(p->bCompress, pRc);
+ testCaseFinish(*pRc);
+ }
+ }
+}
diff --git a/ext/lsm1/lsm-test/lsmtest3.c b/ext/lsm1/lsm-test/lsmtest3.c
new file mode 100644
index 0000000..760dec3
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest3.c
@@ -0,0 +1,238 @@
+
+
+/*
+** This file contains tests related to the explicit rollback of database
+** transactions and sub-transactions.
+*/
+
+
+/*
+** Repeat 2000 times (until the db contains 100,000 entries):
+**
+** 1. Open a transaction and insert 500 rows, opening a nested
+** sub-transaction each 100 rows.
+**
+** 2. Roll back to each sub-transaction savepoint. Check the database
+** checksum looks Ok.
+**
+** 3. Every second iteration, roll back the main transaction. Check the
+** db checksum is correct. Every other iteration, commit the main
+** transaction (increasing the size of the db by 100 rows).
+*/
+
+
+#include "lsmtest.h"
+
+struct CksumDb {
+ int nFirst;
+ int nLast;
+ int nStep;
+ char **azCksum;
+};
+
+CksumDb *testCksumArrayNew(
+ Datasource *pData,
+ int nFirst,
+ int nLast,
+ int nStep
+){
+ TestDb *pDb;
+ CksumDb *pRet;
+ int i;
+ int nEntry;
+ int rc = 0;
+
+ assert( nLast>=nFirst && ((nLast-nFirst)%nStep)==0 );
+
+ pRet = malloc(sizeof(CksumDb));
+ memset(pRet, 0, sizeof(CksumDb));
+ pRet->nFirst = nFirst;
+ pRet->nLast = nLast;
+ pRet->nStep = nStep;
+ nEntry = 1 + ((nLast - nFirst) / nStep);
+
+ /* Allocate space so that azCksum is an array of nEntry pointers to
+ ** buffers each TEST_CKSUM_BYTES in size. */
+ pRet->azCksum = (char **)malloc(nEntry * (sizeof(char *) + TEST_CKSUM_BYTES));
+ for(i=0; i<nEntry; i++){
+ char *pStart = (char *)(&pRet->azCksum[nEntry]);
+ pRet->azCksum[i] = &pStart[i * TEST_CKSUM_BYTES];
+ }
+
+ tdb_open("lsm", "tempdb.lsm", 1, &pDb);
+ testWriteDatasourceRange(pDb, pData, 0, nFirst, &rc);
+ for(i=0; i<nEntry; i++){
+ testCksumDatabase(pDb, pRet->azCksum[i]);
+ if( i==nEntry ) break;
+ testWriteDatasourceRange(pDb, pData, nFirst+i*nStep, nStep, &rc);
+ }
+
+ tdb_close(pDb);
+
+ return pRet;
+}
+
+char *testCksumArrayGet(CksumDb *p, int nRow){
+ int i;
+ assert( nRow>=p->nFirst );
+ assert( nRow<=p->nLast );
+ assert( ((nRow-p->nFirst) % p->nStep)==0 );
+
+ i = (nRow - p->nFirst) / p->nStep;
+ return p->azCksum[i];
+}
+
+void testCksumArrayFree(CksumDb *p){
+ free(p->azCksum);
+ memset(p, 0x55, sizeof(*p));
+ free(p);
+}
+
+/* End of CksumDb code.
+**************************************************************************/
+
+/*
+** Test utility function. Write key-value pair $i from datasource pData
+** into database pDb.
+*/
+void testWriteDatasource(TestDb *pDb, Datasource *pData, int i, int *pRc){
+ void *pKey; int nKey;
+ void *pVal; int nVal;
+ testDatasourceEntry(pData, i, &pKey, &nKey, &pVal, &nVal);
+ testWrite(pDb, pKey, nKey, pVal, nVal, pRc);
+}
+
+/*
+** Test utility function. Delete datasource pData key $i from database pDb.
+*/
+void testDeleteDatasource(TestDb *pDb, Datasource *pData, int i, int *pRc){
+ void *pKey; int nKey;
+ testDatasourceEntry(pData, i, &pKey, &nKey, 0, 0);
+ testDelete(pDb, pKey, nKey, pRc);
+}
+
+/*
+** This function inserts nWrite key/value pairs into database pDb - the
+** nWrite key value pairs starting at iFirst from data source pData.
+*/
+void testWriteDatasourceRange(
+ TestDb *pDb, /* Database to write to */
+ Datasource *pData, /* Data source to read values from */
+ int iFirst, /* Index of first key/value pair */
+ int nWrite, /* Number of key/value pairs to write */
+ int *pRc /* IN/OUT: Error code */
+){
+ int i;
+ for(i=0; i<nWrite; i++){
+ testWriteDatasource(pDb, pData, iFirst+i, pRc);
+ }
+}
+
+void testDeleteDatasourceRange(
+ TestDb *pDb, /* Database to write to */
+ Datasource *pData, /* Data source to read keys from */
+ int iFirst, /* Index of first key */
+ int nWrite, /* Number of keys to delete */
+ int *pRc /* IN/OUT: Error code */
+){
+ int i;
+ for(i=0; i<nWrite; i++){
+ testDeleteDatasource(pDb, pData, iFirst+i, pRc);
+ }
+}
+
+static char *getName(const char *zSystem){
+ char *zRet;
+ zRet = testMallocPrintf("rollback.%s", zSystem);
+ return zRet;
+}
+
+static int rollback_test_1(
+ const char *zSystem,
+ Datasource *pData
+){
+ const int nRepeat = 100;
+
+ TestDb *pDb;
+ int rc;
+ int i;
+ CksumDb *pCksum;
+ char *zName;
+
+ zName = getName(zSystem);
+ testCaseStart(&rc, zName);
+ testFree(zName);
+
+ pCksum = testCksumArrayNew(pData, 0, nRepeat*100, 100);
+ pDb = 0;
+ rc = tdb_open(zSystem, 0, 1, &pDb);
+ if( pDb && tdb_transaction_support(pDb)==0 ){
+ testCaseSkip();
+ goto skip_rollback_test;
+ }
+
+ for(i=0; i<nRepeat && rc==0; i++){
+ char zCksum[TEST_CKSUM_BYTES];
+ int nCurrent = (((i+1)/2) * 100);
+ int nDbRow;
+ int iTrans;
+
+ /* Check that the database is the expected size. */
+ nDbRow = testCountDatabase(pDb);
+ testCompareInt(nCurrent, nDbRow, &rc);
+
+ for(iTrans=2; iTrans<=6 && rc==0; iTrans++){
+ tdb_begin(pDb, iTrans);
+ testWriteDatasourceRange(pDb, pData, nCurrent, 100, &rc);
+ nCurrent += 100;
+ }
+
+ testCksumDatabase(pDb, zCksum);
+ testCompareStr(zCksum, testCksumArrayGet(pCksum, nCurrent), &rc);
+
+ for(iTrans=6; iTrans>2 && rc==0; iTrans--){
+ tdb_rollback(pDb, iTrans);
+ nCurrent -= 100;
+ testCksumDatabase(pDb, zCksum);
+ testCompareStr(zCksum, testCksumArrayGet(pCksum, nCurrent), &rc);
+ }
+
+ if( i%2 ){
+ tdb_rollback(pDb, 0);
+ nCurrent -= 100;
+ testCksumDatabase(pDb, zCksum);
+ testCompareStr(zCksum, testCksumArrayGet(pCksum, nCurrent), &rc);
+ }else{
+ tdb_commit(pDb, 0);
+ }
+ }
+ testCaseFinish(rc);
+
+ skip_rollback_test:
+ tdb_close(pDb);
+ testCksumArrayFree(pCksum);
+ return rc;
+}
+
+void test_rollback(
+ const char *zSystem,
+ const char *zPattern,
+ int *pRc
+){
+ if( *pRc==0 ){
+ int bRun = 1;
+
+ if( zPattern ){
+ char *zName = getName(zSystem);
+ bRun = testGlobMatch(zPattern, zName);
+ testFree(zName);
+ }
+
+ if( bRun ){
+ DatasourceDefn defn = { TEST_DATASOURCE_RANDOM, 10, 15, 50, 100 };
+ Datasource *pData = testDatasourceNew(&defn);
+ *pRc = rollback_test_1(zSystem, pData);
+ testDatasourceFree(pData);
+ }
+ }
+}
diff --git a/ext/lsm1/lsm-test/lsmtest4.c b/ext/lsm1/lsm-test/lsmtest4.c
new file mode 100644
index 0000000..a47241d
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest4.c
@@ -0,0 +1,127 @@
+
+/*
+** This file contains test cases involving multiple database clients.
+*/
+
+#include "lsmtest.h"
+
+/*
+** The following code implements test cases "mc1.*".
+**
+** This test case uses one writer and $nReader readers. All connections
+** are driven by a single thread. All connections are opened at the start
+** of the test and remain open until the test is finished.
+**
+** The test consists of $nStep steps. Each step the following is performed:
+**
+** 1. The writer inserts $nWriteStep records into the db.
+**
+** 2. The writer checks that the contents of the db are as expected.
+**
+** 3. Each reader that currently has an open read transaction also checks
+** that the contents of the db are as expected (according to the snapshot
+** the read transaction is reading - see below).
+**
+** After step 1, reader 1 opens a read transaction. After step 2, reader
+** 2 opens a read transaction, and so on. At step ($nReader+1), reader 1
+** closes the current read transaction and opens a new one. And so on.
+** The result is that at step N (for N > $nReader), there exists a reader
+** with an open read transaction reading the snapshot committed following
+** steps (N-$nReader-1) to N.
+*/
+typedef struct Mctest Mctest;
+struct Mctest {
+ DatasourceDefn defn; /* Datasource to use */
+ int nStep; /* Total number of steps in test */
+ int nWriteStep; /* Number of rows to insert each step */
+ int nReader; /* Number of read connections */
+};
+static void do_mc_test(
+ const char *zSystem, /* Database system to test */
+ Mctest *pTest,
+ int *pRc /* IN/OUT: return code */
+){
+ const int nDomain = pTest->nStep * pTest->nWriteStep;
+ Datasource *pData; /* Source of data */
+ TestDb *pDb; /* First database connection (writer) */
+ int iReader; /* Used to iterate through aReader */
+ int iStep; /* Current step in test */
+ int iDot = 0; /* Current step in test */
+
+ /* Array of reader connections */
+ struct Reader {
+ TestDb *pDb; /* Connection handle */
+ int iLast; /* Current snapshot contains keys 0..iLast */
+ } *aReader;
+
+ /* Create a data source */
+ pData = testDatasourceNew(&pTest->defn);
+
+ /* Open the writer connection */
+ pDb = testOpen(zSystem, 1, pRc);
+
+ /* Allocate aReader */
+ aReader = (struct Reader *)testMalloc(sizeof(aReader[0]) * pTest->nReader);
+ for(iReader=0; iReader<pTest->nReader; iReader++){
+ aReader[iReader].pDb = testOpen(zSystem, 0, pRc);
+ }
+
+ for(iStep=0; iStep<pTest->nStep; iStep++){
+ int iLast;
+ int iBegin; /* Start read trans using aReader[iBegin] */
+
+ /* Insert nWriteStep more records into the database */
+ int iFirst = iStep*pTest->nWriteStep;
+ testWriteDatasourceRange(pDb, pData, iFirst, pTest->nWriteStep, pRc);
+
+ /* Check that the db is Ok according to the writer */
+ iLast = (iStep+1) * pTest->nWriteStep - 1;
+ testDbContents(pDb, pData, nDomain, 0, iLast, iLast, 1, pRc);
+
+ /* Have reader (iStep % nReader) open a read transaction here. */
+ iBegin = (iStep % pTest->nReader);
+ if( iBegin<iStep ) tdb_commit(aReader[iBegin].pDb, 0);
+ tdb_begin(aReader[iBegin].pDb, 1);
+ aReader[iBegin].iLast = iLast;
+
+ /* Check that the db is Ok for each open reader */
+ for(iReader=0; iReader<pTest->nReader && aReader[iReader].iLast; iReader++){
+ iLast = aReader[iReader].iLast;
+ testDbContents(
+ aReader[iReader].pDb, pData, nDomain, 0, iLast, iLast, 1, pRc
+ );
+ }
+
+ /* Report progress */
+ testCaseProgress(iStep, pTest->nStep, testCaseNDot(), &iDot);
+ }
+
+ /* Close all readers */
+ for(iReader=0; iReader<pTest->nReader; iReader++){
+ testClose(&aReader[iReader].pDb);
+ }
+ testFree(aReader);
+
+ /* Close the writer-connection and free the datasource */
+ testClose(&pDb);
+ testDatasourceFree(pData);
+}
+
+
+void test_mc(
+ const char *zSystem, /* Database system name */
+ const char *zPattern, /* Run test cases that match this pattern */
+ int *pRc /* IN/OUT: Error code */
+){
+ int i;
+ Mctest aTest[] = {
+ { { TEST_DATASOURCE_RANDOM, 10,10, 100,100 }, 100, 10, 5 },
+ };
+
+ for(i=0; i<ArraySize(aTest); i++){
+ if( testCaseBegin(pRc, zPattern, "mc1.%s.%d", zSystem, i) ){
+ do_mc_test(zSystem, &aTest[i], pRc);
+ testCaseFinish(*pRc);
+ }
+ }
+}
diff --git a/ext/lsm1/lsm-test/lsmtest5.c b/ext/lsm1/lsm-test/lsmtest5.c
new file mode 100644
index 0000000..f36184e
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest5.c
@@ -0,0 +1,633 @@
+
+/*
+** This file is broken into three semi-autonomous parts:
+**
+** 1. The database functions.
+** 2. The thread wrappers.
+** 3. The implementation of the mt1.* tests.
+*/
+
+/*************************************************************************
+** DATABASE CONTENTS:
+**
+** The database contains up to N key/value pairs, where N is some large
+** number (say 10,000,000). Keys are integer values between 0 and (N-1).
+** The value associated with each key is a pseudo-random blob of data.
+**
+** Key/value pair keys are encoded as the two bytes "k." followed by a
+** 10-digit decimal number. i.e. key 45 -> "k.0000000045".
+**
+** As well as the key/value pairs, the database also contains checksum
+** entries. The checksums form a hierarchy - for every F key/value
+** entries there is one level 1 checksum. And for each F level 1 checksums
+** there is one level 2 checksum. And so on.
+**
+** Checksum keys are encoded as the two byte "c." followed by the
+** checksum level, followed by a 10 digit decimal number containing
+** the value of the first key that contributes to the checksum value.
+** For example, assuming F==10, the level 1 checksum that spans keys
+** 10 to 19 is "c.1.0000000010".
+**
+** Clients may perform one of two operations on the database: a read
+** or a write.
+**
+** READ OPERATIONS:
+**
+** A read operation scans a range of F key/value pairs. It computes
+** the expected checksum and then compares the computed value to the
+** actual value stored in the level 1 checksum entry. It then scans
+** the group of F level 1 checksums, and compares the computed checksum
+** to the associated level 2 checksum value, and so on until the
+** highest level checksum value has been verified.
+**
+** If a checksum ever fails to match the expected value, the test
+** has failed.
+**
+** WRITE OPERATIONS:
+**
+** A write operation involves writing (possibly clobbering) a single
+** key/value pair. The associated level 1 checksum is then recalculated
+** updated. Then the level 2 checksum, and so on until the highest
+** level checksum has been modified.
+**
+** All updates occur inside a single transaction.
+**
+** INTERFACE:
+**
+** The interface used by test cases to read and write the db consists
+** of type DbParameters and the following functions:
+**
+** dbReadOperation()
+** dbWriteOperation()
+*/
+
+#include "lsmtest.h"
+
+typedef struct DbParameters DbParameters;
+struct DbParameters {
+ int nFanout; /* Checksum fanout (F) */
+ int nKey; /* Size of key space (N) */
+};
+
+#define DB_KEY_BYTES (2+5+10+1)
+
+/*
+** Argument aBuf[] must point to a buffer at least DB_KEY_BYTES in size.
+** This function populates the buffer with a nul-terminated key string
+** corresponding to key iKey.
+*/
+static void dbFormatKey(
+ DbParameters *pParam,
+ int iLevel,
+ int iKey, /* Key value */
+ char *aBuf /* Write key string here */
+){
+ if( iLevel==0 ){
+ snprintf(aBuf, DB_KEY_BYTES, "k.%.10d", iKey);
+ }else{
+ int f = 1;
+ int i;
+ for(i=0; i<iLevel; i++) f = f * pParam->nFanout;
+ snprintf(aBuf, DB_KEY_BYTES, "c.%d.%.10d", iLevel, f*(iKey/f));
+ }
+}
+
+/*
+** Argument aBuf[] must point to a buffer at least DB_KEY_BYTES in size.
+** This function populates the buffer with the string representation of
+** checksum value iVal.
+*/
+static void dbFormatCksumValue(u32 iVal, char *aBuf){
+ snprintf(aBuf, DB_KEY_BYTES, "%.10u", iVal);
+}
+
+/*
+** Return the highest level of checksum in the database described
+** by *pParam.
+*/
+static int dbMaxLevel(DbParameters *pParam){
+ int iMax;
+ int n = 1;
+ for(iMax=0; n<pParam->nKey; iMax++){
+ n = n * pParam->nFanout;
+ }
+ return iMax;
+}
+
+static void dbCksum(
+ void *pCtx, /* IN/OUT: Pointer to u32 containing cksum */
+ void *pKey, int nKey, /* Database key. Unused. */
+ void *pVal, int nVal /* Database value. Checksum this. */
+){
+ u8 *aVal = (u8 *)pVal;
+ u32 *pCksum = (u32 *)pCtx;
+ u32 cksum = *pCksum;
+ int i;
+
+ unused_parameter(pKey);
+ unused_parameter(nKey);
+
+ for(i=0; i<nVal; i++){
+ cksum += (cksum<<3) + (int)aVal[i];
+ }
+
+ *pCksum = cksum;
+}
+
+/*
+** Compute the value of the checksum stored on level iLevel that contains
+** data from key iKey by scanning the pParam->nFanout entries at level
+** iLevel-1.
+*/
+static u32 dbComputeCksum(
+ DbParameters *pParam, /* Database parameters */
+ TestDb *pDb, /* Database connection handle */
+ int iLevel, /* Level of checksum to compute */
+ int iKey, /* Compute checksum for this key */
+ int *pRc /* IN/OUT: Error code */
+){
+ u32 cksum = 0;
+ if( *pRc==0 ){
+ int nFirst;
+ int nLast;
+ int iFirst = 0;
+ int iLast = 0;
+ int i;
+ int f = 1;
+ char zFirst[DB_KEY_BYTES];
+ char zLast[DB_KEY_BYTES];
+
+ assert( iLevel>=1 );
+ for(i=0; i<iLevel; i++) f = f * pParam->nFanout;
+
+ iFirst = f*(iKey/f);
+ iLast = iFirst + f - 1;
+ dbFormatKey(pParam, iLevel-1, iFirst, zFirst);
+ dbFormatKey(pParam, iLevel-1, iLast, zLast);
+ nFirst = strlen(zFirst);
+ nLast = strlen(zLast);
+
+ *pRc = tdb_scan(pDb, (u32*)&cksum, 0, zFirst, nFirst, zLast, nLast,dbCksum);
+ }
+
+ return cksum;
+}
+
+static void dbReadOperation(
+ DbParameters *pParam, /* Database parameters */
+ TestDb *pDb, /* Database connection handle */
+ void (*xDelay)(void *),
+ void *pDelayCtx,
+ int iKey, /* Key to read */
+ int *pRc /* IN/OUT: Error code */
+){
+ const int iMax = dbMaxLevel(pParam);
+ int i;
+
+ if( tdb_transaction_support(pDb) ) testBegin(pDb, 1, pRc);
+ for(i=1; *pRc==0 && i<=iMax; i++){
+ char zCksum[DB_KEY_BYTES];
+ char zKey[DB_KEY_BYTES];
+ u32 iCksum = 0;
+
+ iCksum = dbComputeCksum(pParam, pDb, i, iKey, pRc);
+ if( iCksum ){
+ if( xDelay && i==1 ) xDelay(pDelayCtx);
+ dbFormatCksumValue(iCksum, zCksum);
+ dbFormatKey(pParam, i, iKey, zKey);
+ testFetchStr(pDb, zKey, zCksum, pRc);
+ }
+ }
+ if( tdb_transaction_support(pDb) ) testCommit(pDb, 0, pRc);
+}
+
+static int dbWriteOperation(
+ DbParameters *pParam, /* Database parameters */
+ TestDb *pDb, /* Database connection handle */
+ int iKey, /* Key to write to */
+ const char *zValue, /* Nul-terminated value to write */
+ int *pRc /* IN/OUT: Error code */
+){
+ const int iMax = dbMaxLevel(pParam);
+ char zKey[DB_KEY_BYTES];
+ int i;
+ int rc;
+
+ assert( iKey>=0 && iKey<pParam->nKey );
+ dbFormatKey(pParam, 0, iKey, zKey);
+
+ /* Open a write transaction. This may fail - SQLITE4_BUSY */
+ if( *pRc==0 && tdb_transaction_support(pDb) ){
+ rc = tdb_begin(pDb, 2);
+ if( rc==5 ) return 0;
+ *pRc = rc;
+ }
+
+ testWriteStr(pDb, zKey, zValue, pRc);
+ for(i=1; i<=iMax; i++){
+ char zCksum[DB_KEY_BYTES];
+ u32 iCksum = 0;
+
+ iCksum = dbComputeCksum(pParam, pDb, i, iKey, pRc);
+ dbFormatCksumValue(iCksum, zCksum);
+ dbFormatKey(pParam, i, iKey, zKey);
+ testWriteStr(pDb, zKey, zCksum, pRc);
+ }
+ if( tdb_transaction_support(pDb) ) testCommit(pDb, 0, pRc);
+ return 1;
+}
+
+/*************************************************************************
+** The following block contains testXXX() functions that implement a
+** wrapper around the systems native multi-thread support. There are no
+** synchronization primitives - just functions to launch and join
+** threads. Wrapper functions are:
+**
+** testThreadSupport()
+**
+** testThreadInit()
+** testThreadShutdown()
+** testThreadLaunch()
+** testThreadWait()
+**
+** testThreadSetHalt()
+** testThreadGetHalt()
+** testThreadSetResult()
+** testThreadGetResult()
+**
+** testThreadEnterMutex()
+** testThreadLeaveMutex()
+*/
+typedef struct ThreadSet ThreadSet;
+#ifdef LSM_MUTEX_PTHREADS
+
+#include <pthread.h>
+#include <unistd.h>
+
+typedef struct Thread Thread;
+struct Thread {
+ int rc;
+ char *zMsg;
+ pthread_t id;
+ void (*xMain)(ThreadSet *, int, void *);
+ void *pCtx;
+ ThreadSet *pThreadSet;
+};
+
+struct ThreadSet {
+ int bHalt; /* Halt flag */
+ int nThread; /* Number of threads */
+ Thread *aThread; /* Array of Thread structures */
+ pthread_mutex_t mutex; /* Mutex used for cheating */
+};
+
+/*
+** Return true if this build supports threads, or false otherwise. If
+** this function returns false, no other testThreadXXX() functions should
+** be called.
+*/
+static int testThreadSupport(){ return 1; }
+
+/*
+** Allocate and return a thread-set handle with enough space allocated
+** to handle up to nMax threads. Each call to this function should be
+** matched by a call to testThreadShutdown() to delete the object.
+*/
+static ThreadSet *testThreadInit(int nMax){
+ int nByte; /* Total space to allocate */
+ ThreadSet *p; /* Return value */
+
+ nByte = sizeof(ThreadSet) + sizeof(struct Thread) * nMax;
+ p = (ThreadSet *)testMalloc(nByte);
+ p->nThread = nMax;
+ p->aThread = (Thread *)&p[1];
+ pthread_mutex_init(&p->mutex, 0);
+
+ return p;
+}
+
+/*
+** Delete a thread-set object and release all resources held by it.
+*/
+static void testThreadShutdown(ThreadSet *p){
+ int i;
+ for(i=0; i<p->nThread; i++){
+ testFree(p->aThread[i].zMsg);
+ }
+ pthread_mutex_destroy(&p->mutex);
+ testFree(p);
+}
+
+static void *ttMain(void *pArg){
+ Thread *pThread = (Thread *)pArg;
+ int iThread;
+ iThread = (pThread - pThread->pThreadSet->aThread);
+ pThread->xMain(pThread->pThreadSet, iThread, pThread->pCtx);
+ return 0;
+}
+
+/*
+** Launch a new thread.
+*/
+static int testThreadLaunch(
+ ThreadSet *p,
+ int iThread,
+ void (*xMain)(ThreadSet *, int, void *),
+ void *pCtx
+){
+ int rc;
+ Thread *pThread;
+
+ assert( iThread>=0 && iThread<p->nThread );
+
+ pThread = &p->aThread[iThread];
+ assert( pThread->pThreadSet==0 );
+ pThread->xMain = xMain;
+ pThread->pCtx = pCtx;
+ pThread->pThreadSet = p;
+ rc = pthread_create(&pThread->id, 0, ttMain, (void *)pThread);
+
+ return rc;
+}
+
+/*
+** Set the thread-set "halt" flag.
+*/
+static void testThreadSetHalt(ThreadSet *pThreadSet){
+ pThreadSet->bHalt = 1;
+}
+
+/*
+** Return the current value of the thread-set "halt" flag.
+*/
+static int testThreadGetHalt(ThreadSet *pThreadSet){
+ return pThreadSet->bHalt;
+}
+
+static void testThreadSleep(ThreadSet *pThreadSet, int nMs){
+ int nRem = nMs;
+ while( nRem>0 && testThreadGetHalt(pThreadSet)==0 ){
+ usleep(50000);
+ nRem -= 50;
+ }
+}
+
+/*
+** Wait for all threads launched to finish before returning. If nMs
+** is greater than zero, set the "halt" flag to tell all threads
+** to halt after waiting nMs milliseconds.
+*/
+static void testThreadWait(ThreadSet *pThreadSet, int nMs){
+ int i;
+
+ testThreadSleep(pThreadSet, nMs);
+ testThreadSetHalt(pThreadSet);
+ for(i=0; i<pThreadSet->nThread; i++){
+ Thread *pThread = &pThreadSet->aThread[i];
+ if( pThread->xMain ){
+ pthread_join(pThread->id, 0);
+ }
+ }
+}
+
+/*
+** Set the result for thread iThread.
+*/
+static void testThreadSetResult(
+ ThreadSet *pThreadSet, /* Thread-set handle */
+ int iThread, /* Set result for this thread */
+ int rc, /* Result error code */
+ char *zFmt, /* Result string format */
+ ... /* Result string formatting args... */
+){
+ va_list ap;
+
+ testFree(pThreadSet->aThread[iThread].zMsg);
+ pThreadSet->aThread[iThread].rc = rc;
+ pThreadSet->aThread[iThread].zMsg = 0;
+ if( zFmt ){
+ va_start(ap, zFmt);
+ pThreadSet->aThread[iThread].zMsg = testMallocVPrintf(zFmt, ap);
+ va_end(ap);
+ }
+}
+
+/*
+** Retrieve the result for thread iThread.
+*/
+static int testThreadGetResult(
+ ThreadSet *pThreadSet, /* Thread-set handle */
+ int iThread, /* Get result for this thread */
+ const char **pzRes /* OUT: Pointer to result string */
+){
+ if( pzRes ) *pzRes = pThreadSet->aThread[iThread].zMsg;
+ return pThreadSet->aThread[iThread].rc;
+}
+
+/*
+** Enter and leave the test case mutex.
+*/
+#if 0
+static void testThreadEnterMutex(ThreadSet *p){
+ pthread_mutex_lock(&p->mutex);
+}
+static void testThreadLeaveMutex(ThreadSet *p){
+ pthread_mutex_unlock(&p->mutex);
+}
+#endif
+#endif
+
+#if !defined(LSM_MUTEX_PTHREADS)
+static int testThreadSupport(){ return 0; }
+
+#define testThreadInit(a) 0
+#define testThreadShutdown(a)
+#define testThreadLaunch(a,b,c,d) 0
+#define testThreadWait(a,b)
+#define testThreadSetHalt(a)
+#define testThreadGetHalt(a) 0
+#define testThreadGetResult(a,b,c) 0
+#define testThreadSleep(a,b) 0
+
+static void testThreadSetResult(ThreadSet *a, int b, int c, char *d, ...){
+ unused_parameter(a);
+ unused_parameter(b);
+ unused_parameter(c);
+ unused_parameter(d);
+}
+#endif
+/* End of threads wrapper.
+*************************************************************************/
+
+/*************************************************************************
+** Below this point is the third part of this file - the implementation
+** of the mt1.* tests.
+*/
+typedef struct Mt1Test Mt1Test;
+struct Mt1Test {
+ DbParameters param; /* Description of database to read/write */
+ int nReadwrite; /* Number of read/write threads */
+ int nFastReader; /* Number of fast reader threads */
+ int nSlowReader; /* Number of slow reader threads */
+ int nMs; /* How long to run for */
+ const char *zSystem; /* Database system to test */
+};
+
+typedef struct Mt1DelayCtx Mt1DelayCtx;
+struct Mt1DelayCtx {
+ ThreadSet *pSet; /* Threadset to sleep within */
+ int nMs; /* Sleep in ms */
+};
+
+static void xMt1Delay(void *pCtx){
+ Mt1DelayCtx *p = (Mt1DelayCtx *)pCtx;
+ testThreadSleep(p->pSet, p->nMs);
+}
+
+#define MT1_THREAD_RDWR 0
+#define MT1_THREAD_SLOW 1
+#define MT1_THREAD_FAST 2
+
+static void xMt1Work(lsm_db *pDb, void *pCtx){
+#if 0
+ char *z = 0;
+ lsm_info(pDb, LSM_INFO_DB_STRUCTURE, &z);
+ printf("%s\n", z);
+ fflush(stdout);
+#endif
+}
+
+/*
+** This is the main() proc for all threads in test case "mt1".
+*/
+static void mt1Main(ThreadSet *pThreadSet, int iThread, void *pCtx){
+ Mt1Test *p = (Mt1Test *)pCtx; /* Test parameters */
+ Mt1DelayCtx delay;
+ int nRead = 0; /* Number of calls to dbReadOperation() */
+ int nWrite = 0; /* Number of completed database writes */
+ int rc = 0; /* Error code */
+ int iPrng; /* Prng argument variable */
+ TestDb *pDb; /* Database handle */
+ int eType;
+
+ delay.pSet = pThreadSet;
+ delay.nMs = 0;
+ if( iThread<p->nReadwrite ){
+ eType = MT1_THREAD_RDWR;
+ }else if( iThread<(p->nReadwrite+p->nFastReader) ){
+ eType = MT1_THREAD_FAST;
+ }else{
+ eType = MT1_THREAD_SLOW;
+ delay.nMs = (p->nMs / 20);
+ }
+
+ /* Open a new database connection. Initialize the pseudo-random number
+ ** argument based on the thread number. */
+ iPrng = testPrngValue(iThread);
+ pDb = testOpen(p->zSystem, 0, &rc);
+
+ if( rc==0 ){
+ tdb_lsm_config_work_hook(pDb, xMt1Work, 0);
+ }
+
+ /* Loop until either an error occurs or some other thread sets the
+ ** halt flag. */
+ while( rc==0 && testThreadGetHalt(pThreadSet)==0 ){
+ int iKey;
+
+ /* Perform a read operation on an arbitrarily selected key. */
+ iKey = (testPrngValue(iPrng++) % p->param.nKey);
+ dbReadOperation(&p->param, pDb, xMt1Delay, (void *)&delay, iKey, &rc);
+ if( rc ) continue;
+ nRead++;
+
+ /* Attempt to write an arbitrary key value pair (and update the associated
+ ** checksum entries). dbWriteOperation() returns 1 if the write is
+ ** successful, or 0 if it failed with an LSM_BUSY error. */
+ if( eType==MT1_THREAD_RDWR ){
+ char aValue[50];
+ char aRnd[25];
+
+ iKey = (testPrngValue(iPrng++) % p->param.nKey);
+ testPrngString(iPrng, aRnd, sizeof(aRnd));
+ iPrng += sizeof(aRnd);
+ snprintf(aValue, sizeof(aValue), "%d.%s", iThread, aRnd);
+ nWrite += dbWriteOperation(&p->param, pDb, iKey, aValue, &rc);
+ }
+ }
+ testClose(&pDb);
+
+ /* If an error has occured, set the thread error code and the threadset
+ ** halt flag to tell the other test threads to halt. Otherwise, set the
+ ** thread error code to 0 and post a message with the number of read
+ ** and write operations completed. */
+ if( rc ){
+ testThreadSetResult(pThreadSet, iThread, rc, 0);
+ testThreadSetHalt(pThreadSet);
+ }else{
+ testThreadSetResult(pThreadSet, iThread, 0, "r/w: %d/%d", nRead, nWrite);
+ }
+}
+
+static void do_test_mt1(
+ const char *zSystem, /* Database system name */
+ const char *zPattern, /* Run test cases that match this pattern */
+ int *pRc /* IN/OUT: Error code */
+){
+ Mt1Test aTest[] = {
+ /* param, nReadwrite, nFastReader, nSlowReader, nMs, zSystem */
+ { {10, 1000}, 4, 0, 0, 10000, 0 },
+ { {10, 1000}, 4, 4, 2, 100000, 0 },
+ { {10, 100000}, 4, 0, 0, 10000, 0 },
+ { {10, 100000}, 4, 4, 2, 100000, 0 },
+ };
+ int i;
+
+ for(i=0; *pRc==0 && i<ArraySize(aTest); i++){
+ Mt1Test *p = &aTest[i];
+ int bRun = testCaseBegin(pRc, zPattern,
+ "mt1.%s.db=%d,%d.ms=%d.rdwr=%d.fast=%d.slow=%d",
+ zSystem, p->param.nFanout, p->param.nKey,
+ p->nMs, p->nReadwrite, p->nFastReader, p->nSlowReader
+ );
+ if( bRun ){
+ TestDb *pDb;
+ ThreadSet *pSet;
+ int iThread;
+ int nThread;
+
+ p->zSystem = zSystem;
+ pDb = testOpen(zSystem, 1, pRc);
+
+ nThread = p->nReadwrite + p->nFastReader + p->nSlowReader;
+ pSet = testThreadInit(nThread);
+ for(iThread=0; *pRc==0 && iThread<nThread; iThread++){
+ testThreadLaunch(pSet, iThread, mt1Main, (void *)p);
+ }
+
+ testThreadWait(pSet, p->nMs);
+ for(iThread=0; *pRc==0 && iThread<nThread; iThread++){
+ *pRc = testThreadGetResult(pSet, iThread, 0);
+ }
+ testCaseFinish(*pRc);
+
+ for(iThread=0; *pRc==0 && iThread<nThread; iThread++){
+ const char *zMsg = 0;
+ *pRc = testThreadGetResult(pSet, iThread, &zMsg);
+ printf(" Info: thread %d (%d): %s\n", iThread, *pRc, zMsg);
+ }
+
+ testThreadShutdown(pSet);
+ testClose(&pDb);
+ }
+ }
+}
+
+void test_mt(
+ const char *zSystem, /* Database system name */
+ const char *zPattern, /* Run test cases that match this pattern */
+ int *pRc /* IN/OUT: Error code */
+){
+ if( testThreadSupport()==0 ) return;
+ do_test_mt1(zSystem, zPattern, pRc);
+}
diff --git a/ext/lsm1/lsm-test/lsmtest6.c b/ext/lsm1/lsm-test/lsmtest6.c
new file mode 100644
index 0000000..a61b738
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest6.c
@@ -0,0 +1,661 @@
+
+#include "lsmtest.h"
+
+typedef struct OomTest OomTest;
+struct OomTest {
+ lsm_env *pEnv;
+ int iNext; /* Next value to pass to testMallocOom() */
+ int nFail; /* Number of OOM events injected */
+ int bEnable;
+ int rc; /* Test case error code */
+};
+
+static void testOomStart(OomTest *p){
+ memset(p, 0, sizeof(OomTest));
+ p->iNext = 1;
+ p->bEnable = 1;
+ p->nFail = 1;
+ p->pEnv = tdb_lsm_env();
+}
+
+static void xOomHook(OomTest *p){
+ p->nFail++;
+}
+
+static int testOomContinue(OomTest *p){
+ if( p->rc!=0 || (p->iNext>1 && p->nFail==0) ){
+ return 0;
+ }
+ p->nFail = 0;
+ testMallocOom(p->pEnv, p->iNext, 0, (void (*)(void*))xOomHook, (void *)p);
+ return 1;
+}
+
+static void testOomEnable(OomTest *p, int bEnable){
+ p->bEnable = bEnable;
+ testMallocOomEnable(p->pEnv, bEnable);
+}
+
+static void testOomNext(OomTest *p){
+ p->iNext++;
+}
+
+static int testOomHit(OomTest *p){
+ return (p->nFail>0);
+}
+
+static int testOomFinish(OomTest *p){
+ return p->rc;
+}
+
+static void testOomAssert(OomTest *p, int bVal){
+ if( bVal==0 ){
+ test_failed();
+ p->rc = 1;
+ }
+}
+
+/*
+** Test that the error code matches the state of the OomTest object passed
+** as the first argument. Specifically, check that rc is LSM_NOMEM if an
+** OOM error has already been injected, or LSM_OK if not.
+*/
+static void testOomAssertRc(OomTest *p, int rc){
+ testOomAssert(p, rc==LSM_OK || rc==LSM_NOMEM);
+ testOomAssert(p, testOomHit(p)==(rc==LSM_NOMEM) || p->bEnable==0 );
+}
+
+static void testOomOpen(
+ OomTest *pOom,
+ const char *zName,
+ lsm_db **ppDb,
+ int *pRc
+){
+ if( *pRc==LSM_OK ){
+ int rc;
+ rc = lsm_new(tdb_lsm_env(), ppDb);
+ if( rc==LSM_OK ) rc = lsm_open(*ppDb, zName);
+ testOomAssertRc(pOom, rc);
+ *pRc = rc;
+ }
+}
+
+static void testOomFetch(
+ OomTest *pOom,
+ lsm_db *pDb,
+ void *pKey, int nKey,
+ void *pVal, int nVal,
+ int *pRc
+){
+ testOomAssertRc(pOom, *pRc);
+ if( *pRc==LSM_OK ){
+ lsm_cursor *pCsr;
+ int rc;
+
+ rc = lsm_csr_open(pDb, &pCsr);
+ if( rc==LSM_OK ) rc = lsm_csr_seek(pCsr, pKey, nKey, 0);
+ testOomAssertRc(pOom, rc);
+
+ if( rc==LSM_OK ){
+ const void *p; int n;
+ testOomAssert(pOom, lsm_csr_valid(pCsr));
+
+ rc = lsm_csr_key(pCsr, &p, &n);
+ testOomAssertRc(pOom, rc);
+ testOomAssert(pOom, rc!=LSM_OK || (n==nKey && memcmp(pKey, p, nKey)==0) );
+ }
+
+ if( rc==LSM_OK ){
+ const void *p; int n;
+ testOomAssert(pOom, lsm_csr_valid(pCsr));
+
+ rc = lsm_csr_value(pCsr, &p, &n);
+ testOomAssertRc(pOom, rc);
+ testOomAssert(pOom, rc!=LSM_OK || (n==nVal && memcmp(pVal, p, nVal)==0) );
+ }
+
+ lsm_csr_close(pCsr);
+ *pRc = rc;
+ }
+}
+
+static void testOomWrite(
+ OomTest *pOom,
+ lsm_db *pDb,
+ void *pKey, int nKey,
+ void *pVal, int nVal,
+ int *pRc
+){
+ testOomAssertRc(pOom, *pRc);
+ if( *pRc==LSM_OK ){
+ int rc;
+
+ rc = lsm_insert(pDb, pKey, nKey, pVal, nVal);
+ testOomAssertRc(pOom, rc);
+
+ *pRc = rc;
+ }
+}
+
+
+static void testOomFetchStr(
+ OomTest *pOom,
+ lsm_db *pDb,
+ const char *zKey,
+ const char *zVal,
+ int *pRc
+){
+ int nKey = strlen(zKey);
+ int nVal = strlen(zVal);
+ testOomFetch(pOom, pDb, (void *)zKey, nKey, (void *)zVal, nVal, pRc);
+}
+
+static void testOomFetchData(
+ OomTest *pOom,
+ lsm_db *pDb,
+ Datasource *pData,
+ int iKey,
+ int *pRc
+){
+ void *pKey; int nKey;
+ void *pVal; int nVal;
+ testDatasourceEntry(pData, iKey, &pKey, &nKey, &pVal, &nVal);
+ testOomFetch(pOom, pDb, pKey, nKey, pVal, nVal, pRc);
+}
+
+static void testOomWriteStr(
+ OomTest *pOom,
+ lsm_db *pDb,
+ const char *zKey,
+ const char *zVal,
+ int *pRc
+){
+ int nKey = strlen(zKey);
+ int nVal = strlen(zVal);
+ testOomWrite(pOom, pDb, (void *)zKey, nKey, (void *)zVal, nVal, pRc);
+}
+
+static void testOomWriteData(
+ OomTest *pOom,
+ lsm_db *pDb,
+ Datasource *pData,
+ int iKey,
+ int *pRc
+){
+ void *pKey; int nKey;
+ void *pVal; int nVal;
+ testDatasourceEntry(pData, iKey, &pKey, &nKey, &pVal, &nVal);
+ testOomWrite(pOom, pDb, pKey, nKey, pVal, nVal, pRc);
+}
+
+static void testOomScan(
+ OomTest *pOom,
+ lsm_db *pDb,
+ int bReverse,
+ const void *pKey, int nKey,
+ int nScan,
+ int *pRc
+){
+ if( *pRc==0 ){
+ int rc;
+ int iScan = 0;
+ lsm_cursor *pCsr;
+ int (*xAdvance)(lsm_cursor *) = 0;
+
+
+ rc = lsm_csr_open(pDb, &pCsr);
+ testOomAssertRc(pOom, rc);
+
+ if( rc==LSM_OK ){
+ if( bReverse ){
+ rc = lsm_csr_seek(pCsr, pKey, nKey, LSM_SEEK_LE);
+ xAdvance = lsm_csr_prev;
+ }else{
+ rc = lsm_csr_seek(pCsr, pKey, nKey, LSM_SEEK_GE);
+ xAdvance = lsm_csr_next;
+ }
+ }
+ testOomAssertRc(pOom, rc);
+
+ while( rc==LSM_OK && lsm_csr_valid(pCsr) && iScan<nScan ){
+ const void *p; int n;
+
+ rc = lsm_csr_key(pCsr, &p, &n);
+ testOomAssertRc(pOom, rc);
+ if( rc==LSM_OK ){
+ rc = lsm_csr_value(pCsr, &p, &n);
+ testOomAssertRc(pOom, rc);
+ }
+ if( rc==LSM_OK ){
+ rc = xAdvance(pCsr);
+ testOomAssertRc(pOom, rc);
+ }
+ iScan++;
+ }
+
+ lsm_csr_close(pCsr);
+ *pRc = rc;
+ }
+}
+
+#define LSMTEST6_TESTDB "testdb.lsm"
+
+void testDeleteLsmdb(const char *zFile){
+ char *zLog = testMallocPrintf("%s-log", zFile);
+ char *zShm = testMallocPrintf("%s-shm", zFile);
+ unlink(zFile);
+ unlink(zLog);
+ unlink(zShm);
+ testFree(zLog);
+ testFree(zShm);
+}
+
+static void copy_file(const char *zFrom, const char *zTo, int isDatabase){
+
+ if( access(zFrom, F_OK) ){
+ unlink(zTo);
+ }else{
+ int fd1;
+ int fd2;
+ off_t sz;
+ off_t i;
+ struct stat buf;
+ u8 *aBuf;
+
+ fd1 = open(zFrom, O_RDONLY | _O_BINARY, 0644);
+ fd2 = open(zTo, O_RDWR | O_CREAT | _O_BINARY, 0644);
+
+ fstat(fd1, &buf);
+ sz = buf.st_size;
+ ftruncate(fd2, sz);
+
+ aBuf = testMalloc(4096);
+ for(i=0; i<sz; i+=4096){
+ int bLockPage = isDatabase && i == 0;
+ int nByte = MIN((bLockPage ? 4066 : 4096), sz - i);
+ memset(aBuf, 0, 4096);
+ read(fd1, aBuf, nByte);
+ write(fd2, aBuf, nByte);
+ if( bLockPage ){
+ lseek(fd1, 4096, SEEK_SET);
+ lseek(fd2, 4096, SEEK_SET);
+ }
+ }
+ testFree(aBuf);
+
+ close(fd1);
+ close(fd2);
+ }
+}
+
+void testCopyLsmdb(const char *zFrom, const char *zTo){
+ char *zLog1 = testMallocPrintf("%s-log", zFrom);
+ char *zLog2 = testMallocPrintf("%s-log", zTo);
+ char *zShm1 = testMallocPrintf("%s-shm", zFrom);
+ char *zShm2 = testMallocPrintf("%s-shm", zTo);
+
+ unlink(zShm2);
+ unlink(zLog2);
+ unlink(zTo);
+ copy_file(zFrom, zTo, 1);
+ copy_file(zLog1, zLog2, 0);
+ copy_file(zShm1, zShm2, 0);
+
+ testFree(zLog1); testFree(zLog2); testFree(zShm1); testFree(zShm2);
+}
+
+/*
+** File zFile is the path to a database. This function makes backups
+** of the database file and its log as follows:
+**
+** cp $(zFile) $(zFile)-save
+** cp $(zFile)-$(zAux) $(zFile)-save-$(zAux)
+**
+** Function testRestoreDb() can be used to copy the files back in the
+** other direction.
+*/
+void testSaveDb(const char *zFile, const char *zAux){
+ char *zLog = testMallocPrintf("%s-%s", zFile, zAux);
+ char *zFileSave = testMallocPrintf("%s-save", zFile);
+ char *zLogSave = testMallocPrintf("%s-%s-save", zFile, zAux);
+
+ unlink(zFileSave);
+ unlink(zLogSave);
+ copy_file(zFile, zFileSave, 1);
+ copy_file(zLog, zLogSave, 0);
+
+ testFree(zLog); testFree(zFileSave); testFree(zLogSave);
+}
+
+/*
+** File zFile is the path to a database. This function restores
+** a backup of the database made by a previous call to testSaveDb().
+** Specifically, it does the equivalent of:
+**
+** cp $(zFile)-save $(zFile)
+** cp $(zFile)-save-$(zAux) $(zFile)-$(zAux)
+*/
+void testRestoreDb(const char *zFile, const char *zAux){
+ char *zLog = testMallocPrintf("%s-%s", zFile, zAux);
+ char *zFileSave = testMallocPrintf("%s-save", zFile);
+ char *zLogSave = testMallocPrintf("%s-%s-save", zFile, zAux);
+
+ copy_file(zFileSave, zFile, 1);
+ copy_file(zLogSave, zLog, 0);
+
+ testFree(zLog); testFree(zFileSave); testFree(zLogSave);
+}
+
+
+static int lsmWriteStr(lsm_db *pDb, const char *zKey, const char *zVal){
+ int nKey = strlen(zKey);
+ int nVal = strlen(zVal);
+ return lsm_insert(pDb, (void *)zKey, nKey, (void *)zVal, nVal);
+}
+
+static void setup_delete_db(void){
+ testDeleteLsmdb(LSMTEST6_TESTDB);
+}
+
+/*
+** Create a small database. With the following content:
+**
+** "one" -> "one"
+** "two" -> "four"
+** "three" -> "nine"
+** "four" -> "sixteen"
+** "five" -> "twentyfive"
+** "six" -> "thirtysix"
+** "seven" -> "fourtynine"
+** "eight" -> "sixtyfour"
+*/
+static void setup_populate_db(void){
+ const char *azStr[] = {
+ "one", "one",
+ "two", "four",
+ "three", "nine",
+ "four", "sixteen",
+ "five", "twentyfive",
+ "six", "thirtysix",
+ "seven", "fourtynine",
+ "eight", "sixtyfour",
+ };
+ int rc;
+ int ii;
+ lsm_db *pDb;
+
+ testDeleteLsmdb(LSMTEST6_TESTDB);
+
+ rc = lsm_new(tdb_lsm_env(), &pDb);
+ if( rc==LSM_OK ) rc = lsm_open(pDb, LSMTEST6_TESTDB);
+
+ for(ii=0; rc==LSM_OK && ii<ArraySize(azStr); ii+=2){
+ rc = lsmWriteStr(pDb, azStr[ii], azStr[ii+1]);
+ }
+ lsm_close(pDb);
+
+ testSaveDb(LSMTEST6_TESTDB, "log");
+ assert( rc==LSM_OK );
+}
+
+static Datasource *getDatasource(void){
+ const DatasourceDefn defn = { TEST_DATASOURCE_RANDOM, 10, 15, 200, 250 };
+ return testDatasourceNew(&defn);
+}
+
+/*
+** Set up a database file with the following properties:
+**
+** * Page size is 1024 bytes.
+** * Block size is 64 KB.
+** * Contains 5000 key-value pairs starting at 0 from the
+** datasource returned getDatasource().
+*/
+static void setup_populate_db2(void){
+ Datasource *pData;
+ int ii;
+ int rc;
+ int nBlocksize = 64*1024;
+ int nPagesize = 1024;
+ int nWritebuffer = 4*1024;
+ lsm_db *pDb;
+
+ testDeleteLsmdb(LSMTEST6_TESTDB);
+ rc = lsm_new(tdb_lsm_env(), &pDb);
+ if( rc==LSM_OK ) rc = lsm_open(pDb, LSMTEST6_TESTDB);
+
+ lsm_config(pDb, LSM_CONFIG_BLOCK_SIZE, &nBlocksize);
+ lsm_config(pDb, LSM_CONFIG_PAGE_SIZE, &nPagesize);
+ lsm_config(pDb, LSM_CONFIG_AUTOFLUSH, &nWritebuffer);
+
+ pData = getDatasource();
+ for(ii=0; rc==LSM_OK && ii<5000; ii++){
+ void *pKey; int nKey;
+ void *pVal; int nVal;
+ testDatasourceEntry(pData, ii, &pKey, &nKey, &pVal, &nVal);
+ lsm_insert(pDb, pKey, nKey, pVal, nVal);
+ }
+ testDatasourceFree(pData);
+ lsm_close(pDb);
+
+ testSaveDb(LSMTEST6_TESTDB, "log");
+ assert( rc==LSM_OK );
+}
+
+/*
+** Test the results of OOM conditions in lsm_new().
+*/
+static void simple_oom_1(OomTest *pOom){
+ int rc;
+ lsm_db *pDb;
+
+ rc = lsm_new(tdb_lsm_env(), &pDb);
+ testOomAssertRc(pOom, rc);
+
+ lsm_close(pDb);
+}
+
+/*
+** Test the results of OOM conditions in lsm_open().
+*/
+static void simple_oom_2(OomTest *pOom){
+ int rc;
+ lsm_db *pDb;
+
+ rc = lsm_new(tdb_lsm_env(), &pDb);
+ if( rc==LSM_OK ){
+ rc = lsm_open(pDb, "testdb.lsm");
+ }
+ testOomAssertRc(pOom, rc);
+
+ lsm_close(pDb);
+}
+
+/*
+** Test the results of OOM conditions in simple fetch operations.
+*/
+static void simple_oom_3(OomTest *pOom){
+ int rc = LSM_OK;
+ lsm_db *pDb;
+
+ testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);
+
+ testOomFetchStr(pOom, pDb, "four", "sixteen", &rc);
+ testOomFetchStr(pOom, pDb, "seven", "fourtynine", &rc);
+ testOomFetchStr(pOom, pDb, "one", "one", &rc);
+ testOomFetchStr(pOom, pDb, "eight", "sixtyfour", &rc);
+
+ lsm_close(pDb);
+}
+
+/*
+** Test the results of OOM conditions in simple write operations.
+*/
+static void simple_oom_4(OomTest *pOom){
+ int rc = LSM_OK;
+ lsm_db *pDb;
+
+ testDeleteLsmdb(LSMTEST6_TESTDB);
+ testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);
+
+ testOomWriteStr(pOom, pDb, "123", "onetwothree", &rc);
+ testOomWriteStr(pOom, pDb, "456", "fourfivesix", &rc);
+ testOomWriteStr(pOom, pDb, "789", "seveneightnine", &rc);
+ testOomWriteStr(pOom, pDb, "123", "teneleventwelve", &rc);
+ testOomWriteStr(pOom, pDb, "456", "fourteenfifteensixteen", &rc);
+
+ lsm_close(pDb);
+}
+
+static void simple_oom_5(OomTest *pOom){
+ Datasource *pData = getDatasource();
+ int rc = LSM_OK;
+ lsm_db *pDb;
+
+ testRestoreDb(LSMTEST6_TESTDB, "log");
+ testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);
+
+ testOomFetchData(pOom, pDb, pData, 3333, &rc);
+ testOomFetchData(pOom, pDb, pData, 0, &rc);
+ testOomFetchData(pOom, pDb, pData, 4999, &rc);
+
+ lsm_close(pDb);
+ testDatasourceFree(pData);
+}
+
+static void simple_oom_6(OomTest *pOom){
+ Datasource *pData = getDatasource();
+ int rc = LSM_OK;
+ lsm_db *pDb;
+
+ testRestoreDb(LSMTEST6_TESTDB, "log");
+ testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);
+
+ testOomWriteData(pOom, pDb, pData, 5000, &rc);
+ testOomWriteData(pOom, pDb, pData, 5001, &rc);
+ testOomWriteData(pOom, pDb, pData, 5002, &rc);
+ testOomFetchData(pOom, pDb, pData, 5001, &rc);
+ testOomFetchData(pOom, pDb, pData, 1234, &rc);
+
+ lsm_close(pDb);
+ testDatasourceFree(pData);
+}
+
+static void simple_oom_7(OomTest *pOom){
+ Datasource *pData = getDatasource();
+ int rc = LSM_OK;
+ lsm_db *pDb;
+
+ testRestoreDb(LSMTEST6_TESTDB, "log");
+ testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);
+ testOomScan(pOom, pDb, 0, "abc", 3, 20, &rc);
+ lsm_close(pDb);
+ testDatasourceFree(pData);
+}
+
+static void simple_oom_8(OomTest *pOom){
+ Datasource *pData = getDatasource();
+ int rc = LSM_OK;
+ lsm_db *pDb;
+ testRestoreDb(LSMTEST6_TESTDB, "log");
+ testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);
+ testOomScan(pOom, pDb, 1, "xyz", 3, 20, &rc);
+ lsm_close(pDb);
+ testDatasourceFree(pData);
+}
+
+/*
+** This test case has two clients connected to a database. The first client
+** hits an OOM while writing to the database. Check that the second
+** connection is still able to query the db following the OOM.
+*/
+static void simple_oom2_1(OomTest *pOom){
+ const int nRecord = 100; /* Number of records initially in db */
+ const int nIns = 10; /* Number of records inserted with OOM */
+
+ Datasource *pData = getDatasource();
+ int rc = LSM_OK;
+ lsm_db *pDb1;
+ lsm_db *pDb2;
+ int i;
+
+ testDeleteLsmdb(LSMTEST6_TESTDB);
+
+ /* Open the two connections. Initialize the in-memory tree so that it
+ ** contains 100 records. Do all this with OOM injection disabled. */
+ testOomEnable(pOom, 0);
+ testOomOpen(pOom, LSMTEST6_TESTDB, &pDb1, &rc);
+ testOomOpen(pOom, LSMTEST6_TESTDB, &pDb2, &rc);
+ for(i=0; i<nRecord; i++){
+ testOomWriteData(pOom, pDb1, pData, i, &rc);
+ }
+ testOomEnable(pOom, 1);
+ assert( rc==0 );
+
+ /* Insert 10 more records using pDb1. Stop when an OOM is encountered. */
+ for(i=nRecord; i<nRecord+nIns; i++){
+ testOomWriteData(pOom, pDb1, pData, i, &rc);
+ if( rc ) break;
+ }
+ testOomAssertRc(pOom, rc);
+
+ /* Switch off OOM injection. Write a few rows using pDb2. Then check
+ ** that the database may be successfully queried. */
+ testOomEnable(pOom, 0);
+ rc = 0;
+ for(; i<nRecord+nIns && rc==0; i++){
+ testOomWriteData(pOom, pDb2, pData, i, &rc);
+ }
+ for(i=0; i<nRecord+nIns; i++) testOomFetchData(pOom, pDb2, pData, i, &rc);
+ testOomEnable(pOom, 1);
+
+ lsm_close(pDb1);
+ lsm_close(pDb2);
+ testDatasourceFree(pData);
+}
+
+
+static void do_test_oom1(const char *zPattern, int *pRc){
+ struct SimpleOom {
+ const char *zName;
+ void (*xSetup)(void);
+ void (*xFunc)(OomTest *);
+ } aSimple[] = {
+ { "oom1.lsm.1", setup_delete_db, simple_oom_1 },
+ { "oom1.lsm.2", setup_delete_db, simple_oom_2 },
+ { "oom1.lsm.3", setup_populate_db, simple_oom_3 },
+ { "oom1.lsm.4", setup_delete_db, simple_oom_4 },
+ { "oom1.lsm.5", setup_populate_db2, simple_oom_5 },
+ { "oom1.lsm.6", setup_populate_db2, simple_oom_6 },
+ { "oom1.lsm.7", setup_populate_db2, simple_oom_7 },
+ { "oom1.lsm.8", setup_populate_db2, simple_oom_8 },
+
+ { "oom2.lsm.1", setup_delete_db, simple_oom2_1 },
+ };
+ int i;
+
+ for(i=0; i<ArraySize(aSimple); i++){
+ if( *pRc==0 && testCaseBegin(pRc, zPattern, "%s", aSimple[i].zName) ){
+ OomTest t;
+
+ if( aSimple[i].xSetup ){
+ aSimple[i].xSetup();
+ }
+
+ for(testOomStart(&t); testOomContinue(&t); testOomNext(&t)){
+ aSimple[i].xFunc(&t);
+ }
+
+ printf("(%d injections).", t.iNext-2);
+ testCaseFinish( (*pRc = testOomFinish(&t)) );
+ testMallocOom(tdb_lsm_env(), 0, 0, 0, 0);
+ }
+ }
+}
+
+void test_oom(
+ const char *zPattern, /* Run test cases that match this pattern */
+ int *pRc /* IN/OUT: Error code */
+){
+ do_test_oom1(zPattern, pRc);
+}
diff --git a/ext/lsm1/lsm-test/lsmtest7.c b/ext/lsm1/lsm-test/lsmtest7.c
new file mode 100644
index 0000000..2d26b53
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest7.c
@@ -0,0 +1,206 @@
+
+
+#include "lsmtest.h"
+
+
+/*
+** Test that the rules for when lsm_csr_next() and lsm_csr_prev() are
+** enforced. Specifically:
+**
+** * Both functions always return LSM_MISUSE if the cursor is at EOF
+** when they are called.
+**
+** * lsm_csr_next() may only be used after lsm_csr_seek(LSM_SEEK_GE) or
+** lsm_csr_first().
+**
+** * lsm_csr_prev() may only be used after lsm_csr_seek(LSM_SEEK_LE) or
+** lsm_csr_last().
+*/
+static void do_test_api1_lsm(lsm_db *pDb, int *pRc){
+ int ret;
+ lsm_cursor *pCsr;
+ lsm_cursor *pCsr2;
+ int nKey;
+ const void *pKey;
+
+ ret = lsm_csr_open(pDb, &pCsr);
+ testCompareInt(LSM_OK, ret, pRc);
+
+ ret = lsm_csr_next(pCsr);
+ testCompareInt(LSM_MISUSE, ret, pRc);
+ ret = lsm_csr_prev(pCsr);
+ testCompareInt(LSM_MISUSE, ret, pRc);
+
+ ret = lsm_csr_seek(pCsr, "jjj", 3, LSM_SEEK_GE);
+ testCompareInt(LSM_OK, ret, pRc);
+ ret = lsm_csr_next(pCsr);
+ testCompareInt(LSM_OK, ret, pRc);
+ ret = lsm_csr_prev(pCsr);
+ testCompareInt(LSM_MISUSE, ret, pRc);
+
+ ret = lsm_csr_seek(pCsr, "jjj", 3, LSM_SEEK_LE);
+ testCompareInt(LSM_OK, ret, pRc);
+ ret = lsm_csr_next(pCsr);
+ testCompareInt(LSM_MISUSE, ret, pRc);
+ ret = lsm_csr_prev(pCsr);
+ testCompareInt(LSM_OK, ret, pRc);
+
+ ret = lsm_csr_seek(pCsr, "jjj", 3, LSM_SEEK_LEFAST);
+ testCompareInt(LSM_OK, ret, pRc);
+ ret = lsm_csr_next(pCsr);
+ testCompareInt(LSM_MISUSE, ret, pRc);
+ ret = lsm_csr_prev(pCsr);
+ testCompareInt(LSM_MISUSE, ret, pRc);
+
+ ret = lsm_csr_key(pCsr, &pKey, &nKey);
+ testCompareInt(LSM_OK, ret, pRc);
+
+ ret = lsm_csr_open(pDb, &pCsr2);
+ testCompareInt(LSM_OK, ret, pRc);
+
+ ret = lsm_csr_seek(pCsr2, pKey, nKey, LSM_SEEK_EQ);
+ testCompareInt(LSM_OK, ret, pRc);
+ testCompareInt(1, lsm_csr_valid(pCsr2), pRc);
+ ret = lsm_csr_next(pCsr2);
+ testCompareInt(LSM_MISUSE, ret, pRc);
+ ret = lsm_csr_prev(pCsr2);
+ testCompareInt(LSM_MISUSE, ret, pRc);
+
+ lsm_csr_close(pCsr2);
+
+ ret = lsm_csr_first(pCsr);
+ testCompareInt(LSM_OK, ret, pRc);
+ ret = lsm_csr_next(pCsr);
+ testCompareInt(LSM_OK, ret, pRc);
+ ret = lsm_csr_prev(pCsr);
+ testCompareInt(LSM_MISUSE, ret, pRc);
+
+ ret = lsm_csr_last(pCsr);
+ testCompareInt(LSM_OK, ret, pRc);
+ ret = lsm_csr_prev(pCsr);
+ testCompareInt(LSM_OK, ret, pRc);
+ ret = lsm_csr_next(pCsr);
+ testCompareInt(LSM_MISUSE, ret, pRc);
+
+ ret = lsm_csr_first(pCsr);
+ while( lsm_csr_valid(pCsr) ){
+ ret = lsm_csr_next(pCsr);
+ testCompareInt(LSM_OK, ret, pRc);
+ }
+ ret = lsm_csr_next(pCsr);
+ testCompareInt(LSM_OK, ret, pRc);
+ ret = lsm_csr_prev(pCsr);
+ testCompareInt(LSM_MISUSE, ret, pRc);
+
+ ret = lsm_csr_last(pCsr);
+ while( lsm_csr_valid(pCsr) ){
+ ret = lsm_csr_prev(pCsr);
+ testCompareInt(LSM_OK, ret, pRc);
+ }
+ ret = lsm_csr_prev(pCsr);
+ testCompareInt(LSM_OK, ret, pRc);
+ ret = lsm_csr_next(pCsr);
+ testCompareInt(LSM_MISUSE, ret, pRc);
+
+ lsm_csr_close(pCsr);
+}
+
+static void do_test_api1(const char *zPattern, int *pRc){
+ if( testCaseBegin(pRc, zPattern, "api1.lsm") ){
+ const DatasourceDefn defn = { TEST_DATASOURCE_RANDOM, 10, 15, 200, 250 };
+ Datasource *pData;
+ TestDb *pDb;
+ int rc = 0;
+
+ pDb = testOpen("lsm_lomem", 1, &rc);
+ pData = testDatasourceNew(&defn);
+ testWriteDatasourceRange(pDb, pData, 0, 1000, pRc);
+
+ do_test_api1_lsm(tdb_lsm(pDb), pRc);
+
+ testDatasourceFree(pData);
+ testClose(&pDb);
+
+ testCaseFinish(*pRc);
+ }
+}
+
+static lsm_db *newLsmConnection(
+ const char *zDb,
+ int nPgsz,
+ int nBlksz,
+ int *pRc
+){
+ lsm_db *db = 0;
+ if( *pRc==0 ){
+ int n1 = nPgsz;
+ int n2 = nBlksz;
+ *pRc = lsm_new(tdb_lsm_env(), &db);
+ if( *pRc==0 ){
+ if( n1 ) lsm_config(db, LSM_CONFIG_PAGE_SIZE, &n1);
+ if( n2 ) lsm_config(db, LSM_CONFIG_BLOCK_SIZE, &n2);
+ *pRc = lsm_open(db, "testdb.lsm");
+ }
+ }
+ return db;
+}
+
+static void testPagesize(lsm_db *db, int nPgsz, int nBlksz, int *pRc){
+ if( *pRc==0 ){
+ int n1 = 0;
+ int n2 = 0;
+
+ lsm_config(db, LSM_CONFIG_PAGE_SIZE, &n1);
+ lsm_config(db, LSM_CONFIG_BLOCK_SIZE, &n2);
+
+ testCompareInt(n1, nPgsz, pRc);
+ testCompareInt(n2, nBlksz, pRc);
+ }
+}
+
+/*
+** Test case "api2" tests that the default page and block sizes of a
+** database may only be modified before lsm_open() is called. And that
+** after lsm_open() is called lsm_config() may be used to read the
+** actual page and block size of the db.
+*/
+static void do_test_api2(const char *zPattern, int *pRc){
+ if( *pRc==0 && testCaseBegin(pRc, zPattern, "api2.lsm") ){
+ lsm_db *db1 = 0;
+ lsm_db *db2 = 0;
+
+ testDeleteLsmdb("testdb.lsm");
+ db1 = newLsmConnection("testdb.lsm", 0, 0, pRc);
+ testPagesize(db1, 4096, 1024, pRc);
+ db2 = newLsmConnection("testdb.lsm", 1024, 64*1024, pRc);
+ testPagesize(db2, 4096, 1024, pRc);
+ lsm_close(db1);
+ lsm_close(db2);
+
+ testDeleteLsmdb("testdb.lsm");
+ db1 = newLsmConnection("testdb.lsm", 1024, 64*1024, pRc);
+ testPagesize(db1, 1024, 64*1024, pRc);
+ db2 = newLsmConnection("testdb.lsm", 0, 0, pRc);
+ testPagesize(db2, 1024, 64*1024, pRc);
+ lsm_close(db1);
+ lsm_close(db2);
+
+ testDeleteLsmdb("testdb.lsm");
+ db1 = newLsmConnection("testdb.lsm", 8192, 2*1024, pRc);
+ testPagesize(db1, 8192, 2*1024, pRc);
+ db2 = newLsmConnection("testdb.lsm", 1024, 64*1024, pRc);
+ testPagesize(db2, 8192, 2*1024, pRc);
+ lsm_close(db1);
+ lsm_close(db2);
+
+ testCaseFinish(*pRc);
+ }
+}
+
+void test_api(
+ const char *zPattern, /* Run test cases that match this pattern */
+ int *pRc /* IN/OUT: Error code */
+){
+ do_test_api1(zPattern, pRc);
+ do_test_api2(zPattern, pRc);
+}
diff --git a/ext/lsm1/lsm-test/lsmtest8.c b/ext/lsm1/lsm-test/lsmtest8.c
new file mode 100644
index 0000000..7efa0df
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest8.c
@@ -0,0 +1,324 @@
+
+/*
+** This file contains test cases to verify that "live-recovery" following
+** a mid-transaction failure of a writer process.
+*/
+
+
+/*
+** This test file includes lsmInt.h to get access to the definition of the
+** ShmHeader structure. This is required to cause strategic damage to the
+** shared memory header as part of recovery testing.
+*/
+#include "lsmInt.h"
+
+#include "lsmtest.h"
+
+typedef struct SetupStep SetupStep;
+struct SetupStep {
+ int bFlush; /* Flush to disk and checkpoint */
+ int iInsStart; /* First key-value from ds to insert */
+ int nIns; /* Number of rows to insert */
+ int iDelStart; /* First key from ds to delete */
+ int nDel; /* Number of rows to delete */
+};
+
+static void doSetupStep(
+ TestDb *pDb,
+ Datasource *pData,
+ const SetupStep *pStep,
+ int *pRc
+){
+ testWriteDatasourceRange(pDb, pData, pStep->iInsStart, pStep->nIns, pRc);
+ testDeleteDatasourceRange(pDb, pData, pStep->iDelStart, pStep->nDel, pRc);
+ if( *pRc==0 ){
+ int nSave = -1;
+ int nBuf = 64;
+ lsm_db *db = tdb_lsm(pDb);
+
+ lsm_config(db, LSM_CONFIG_AUTOFLUSH, &nSave);
+ lsm_config(db, LSM_CONFIG_AUTOFLUSH, &nBuf);
+ lsm_begin(db, 1);
+ lsm_commit(db, 0);
+ lsm_config(db, LSM_CONFIG_AUTOFLUSH, &nSave);
+
+ *pRc = lsm_work(db, 0, 0, 0);
+ if( *pRc==0 ){
+ *pRc = lsm_checkpoint(db, 0);
+ }
+ }
+}
+
+static void doSetupStepArray(
+ TestDb *pDb,
+ Datasource *pData,
+ const SetupStep *aStep,
+ int nStep
+){
+ int i;
+ for(i=0; i<nStep; i++){
+ int rc = 0;
+ doSetupStep(pDb, pData, &aStep[i], &rc);
+ assert( rc==0 );
+ }
+}
+
+static void setupDatabase1(TestDb *pDb, Datasource **ppData){
+ const SetupStep aStep[] = {
+ { 0, 1, 2000, 0, 0 },
+ { 1, 0, 0, 0, 0 },
+ { 0, 10001, 1000, 0, 0 },
+ };
+ const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 12, 16, 100, 500};
+ Datasource *pData;
+
+ pData = testDatasourceNew(&defn);
+ doSetupStepArray(pDb, pData, aStep, ArraySize(aStep));
+ if( ppData ){
+ *ppData = pData;
+ }else{
+ testDatasourceFree(pData);
+ }
+}
+
+#include <stdio.h>
+void testReadFile(const char *zFile, int iOff, void *pOut, int nByte, int *pRc){
+ if( *pRc==0 ){
+ FILE *fd;
+ fd = fopen(zFile, "rb");
+ if( fd==0 ){
+ *pRc = 1;
+ }else{
+ if( 0!=fseek(fd, iOff, SEEK_SET) ){
+ *pRc = 1;
+ }else{
+ assert( nByte>=0 );
+ if( (size_t)nByte!=fread(pOut, 1, nByte, fd) ){
+ *pRc = 1;
+ }
+ }
+ fclose(fd);
+ }
+ }
+}
+
+void testWriteFile(
+ const char *zFile,
+ int iOff,
+ void *pOut,
+ int nByte,
+ int *pRc
+){
+ if( *pRc==0 ){
+ FILE *fd;
+ fd = fopen(zFile, "r+b");
+ if( fd==0 ){
+ *pRc = 1;
+ }else{
+ if( 0!=fseek(fd, iOff, SEEK_SET) ){
+ *pRc = 1;
+ }else{
+ assert( nByte>=0 );
+ if( (size_t)nByte!=fwrite(pOut, 1, nByte, fd) ){
+ *pRc = 1;
+ }
+ }
+ fclose(fd);
+ }
+ }
+}
+
+static ShmHeader *getShmHeader(const char *zDb){
+ int rc = 0;
+ char *zShm = testMallocPrintf("%s-shm", zDb);
+ ShmHeader *pHdr;
+
+ pHdr = testMalloc(sizeof(ShmHeader));
+ testReadFile(zShm, 0, (void *)pHdr, sizeof(ShmHeader), &rc);
+ assert( rc==0 );
+
+ return pHdr;
+}
+
+/*
+** This function makes a copy of the three files associated with LSM
+** database zDb (i.e. if zDb is "test.db", it makes copies of "test.db",
+** "test.db-log" and "test.db-shm").
+**
+** It then opens a new database connection to the copy with the xLock() call
+** instrumented so that it appears that some other process already connected
+** to the db (holding a shared lock on DMS2). This prevents recovery from
+** running. Then:
+**
+** 1) Check that the checksum of the database is zCksum.
+** 2) Write a few keys to the database. Then delete the same keys.
+** 3) Check that the checksum is zCksum.
+** 4) Flush the db to disk and run a checkpoint.
+** 5) Check once more that the checksum is still zCksum.
+*/
+static void doLiveRecovery(const char *zDb, const char *zCksum, int *pRc){
+ if( *pRc==LSM_OK ){
+ const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 20, 25, 100, 500};
+ Datasource *pData;
+ const char *zCopy = "testcopy.lsm";
+ char zCksum2[TEST_CKSUM_BYTES];
+ TestDb *pDb = 0;
+ int rc;
+
+ pData = testDatasourceNew(&defn);
+
+ testCopyLsmdb(zDb, zCopy);
+ rc = tdb_lsm_open("test_no_recovery=1", zCopy, 0, &pDb);
+ if( rc==0 ){
+ ShmHeader *pHdr;
+ lsm_db *db;
+ testCksumDatabase(pDb, zCksum2);
+ testCompareStr(zCksum, zCksum2, &rc);
+
+ testWriteDatasourceRange(pDb, pData, 1, 10, &rc);
+ testDeleteDatasourceRange(pDb, pData, 1, 10, &rc);
+
+ /* Test that the two tree-headers are now consistent. */
+ pHdr = getShmHeader(zCopy);
+ if( rc==0 && memcmp(&pHdr->hdr1, &pHdr->hdr2, sizeof(pHdr->hdr1)) ){
+ rc = 1;
+ }
+ testFree(pHdr);
+
+ if( rc==0 ){
+ int nBuf = 64;
+ db = tdb_lsm(pDb);
+ lsm_config(db, LSM_CONFIG_AUTOFLUSH, &nBuf);
+ lsm_begin(db, 1);
+ lsm_commit(db, 0);
+ rc = lsm_work(db, 0, 0, 0);
+ }
+
+ testCksumDatabase(pDb, zCksum2);
+ testCompareStr(zCksum, zCksum2, &rc);
+ }
+
+ testDatasourceFree(pData);
+ testClose(&pDb);
+ testDeleteLsmdb(zCopy);
+ *pRc = rc;
+ }
+}
+
+static void doWriterCrash1(int *pRc){
+ const int nWrite = 2000;
+ const int nStep = 10;
+ const int iWriteStart = 20000;
+ int rc = 0;
+ TestDb *pDb = 0;
+ Datasource *pData = 0;
+
+ rc = tdb_lsm_open("autowork=0", "testdb.lsm", 1, &pDb);
+ if( rc==0 ){
+ int iDot = 0;
+ char zCksum[TEST_CKSUM_BYTES];
+ int i;
+ setupDatabase1(pDb, &pData);
+ testCksumDatabase(pDb, zCksum);
+ testBegin(pDb, 2, &rc);
+ for(i=0; rc==0 && i<nWrite; i+=nStep){
+ testCaseProgress(i, nWrite, testCaseNDot(), &iDot);
+ testWriteDatasourceRange(pDb, pData, iWriteStart+i, nStep, &rc);
+ doLiveRecovery("testdb.lsm", zCksum, &rc);
+ }
+ }
+ testCommit(pDb, 0, &rc);
+ testClose(&pDb);
+ testDatasourceFree(pData);
+ *pRc = rc;
+}
+
+/*
+** This test case verifies that inconsistent tree-headers in shared-memory
+** are resolved correctly.
+*/
+static void doWriterCrash2(int *pRc){
+ int rc = 0;
+ TestDb *pDb = 0;
+ Datasource *pData = 0;
+
+ rc = tdb_lsm_open("autowork=0", "testdb.lsm", 1, &pDb);
+ if( rc==0 ){
+ ShmHeader *pHdr1;
+ ShmHeader *pHdr2;
+ char zCksum1[TEST_CKSUM_BYTES];
+ char zCksum2[TEST_CKSUM_BYTES];
+
+ pHdr1 = testMalloc(sizeof(ShmHeader));
+ pHdr2 = testMalloc(sizeof(ShmHeader));
+ setupDatabase1(pDb, &pData);
+
+ /* Grab a copy of the shared-memory header. And the db checksum */
+ testReadFile("testdb.lsm-shm", 0, (void *)pHdr1, sizeof(ShmHeader), &rc);
+ testCksumDatabase(pDb, zCksum1);
+
+ /* Modify the database */
+ testBegin(pDb, 2, &rc);
+ testWriteDatasourceRange(pDb, pData, 30000, 200, &rc);
+ testCommit(pDb, 0, &rc);
+
+ /* Grab a second copy of the shared-memory header. And the db checksum */
+ testReadFile("testdb.lsm-shm", 0, (void *)pHdr2, sizeof(ShmHeader), &rc);
+ testCksumDatabase(pDb, zCksum2);
+ doLiveRecovery("testdb.lsm", zCksum2, &rc);
+
+ /* If both tree-headers are valid, tree-header-1 is used. */
+ memcpy(&pHdr2->hdr1, &pHdr1->hdr1, sizeof(pHdr1->hdr1));
+ pHdr2->bWriter = 1;
+ testWriteFile("testdb.lsm-shm", 0, (void *)pHdr2, sizeof(ShmHeader), &rc);
+ doLiveRecovery("testdb.lsm", zCksum1, &rc);
+
+ /* If both tree-headers are valid, tree-header-1 is used. */
+ memcpy(&pHdr2->hdr1, &pHdr2->hdr2, sizeof(pHdr1->hdr1));
+ memcpy(&pHdr2->hdr2, &pHdr1->hdr1, sizeof(pHdr1->hdr1));
+ pHdr2->bWriter = 1;
+ testWriteFile("testdb.lsm-shm", 0, (void *)pHdr2, sizeof(ShmHeader), &rc);
+ doLiveRecovery("testdb.lsm", zCksum2, &rc);
+
+ /* If tree-header 1 is invalid, tree-header-2 is used */
+ memcpy(&pHdr2->hdr2, &pHdr2->hdr1, sizeof(pHdr1->hdr1));
+ pHdr2->hdr1.aCksum[0] = 5;
+ pHdr2->hdr1.aCksum[0] = 6;
+ pHdr2->bWriter = 1;
+ testWriteFile("testdb.lsm-shm", 0, (void *)pHdr2, sizeof(ShmHeader), &rc);
+ doLiveRecovery("testdb.lsm", zCksum2, &rc);
+
+ /* If tree-header 2 is invalid, tree-header-1 is used */
+ memcpy(&pHdr2->hdr1, &pHdr2->hdr2, sizeof(pHdr1->hdr1));
+ pHdr2->hdr2.aCksum[0] = 5;
+ pHdr2->hdr2.aCksum[0] = 6;
+ pHdr2->bWriter = 1;
+ testWriteFile("testdb.lsm-shm", 0, (void *)pHdr2, sizeof(ShmHeader), &rc);
+ doLiveRecovery("testdb.lsm", zCksum2, &rc);
+
+ testFree(pHdr1);
+ testFree(pHdr2);
+ testClose(&pDb);
+ }
+
+ *pRc = rc;
+}
+
+void do_writer_crash_test(const char *zPattern, int *pRc){
+ struct Test {
+ const char *zName;
+ void (*xFunc)(int *);
+ } aTest[] = {
+ { "writercrash1.lsm", doWriterCrash1 },
+ { "writercrash2.lsm", doWriterCrash2 },
+ };
+ int i;
+ for(i=0; i<ArraySize(aTest); i++){
+ struct Test *p = &aTest[i];
+ if( testCaseBegin(pRc, zPattern, p->zName) ){
+ p->xFunc(pRc);
+ testCaseFinish(*pRc);
+ }
+ }
+
+}
diff --git a/ext/lsm1/lsm-test/lsmtest9.c b/ext/lsm1/lsm-test/lsmtest9.c
new file mode 100644
index 0000000..b01de0d
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest9.c
@@ -0,0 +1,140 @@
+
+#include "lsmtest.h"
+
+#define DATA_SEQUENTIAL TEST_DATASOURCE_SEQUENCE
+#define DATA_RANDOM TEST_DATASOURCE_RANDOM
+
+typedef struct Datatest4 Datatest4;
+
+/*
+** Test overview:
+**
+** 1. Insert (Datatest4.nRec) records into a database.
+**
+** 2. Repeat (Datatest4.nRepeat) times:
+**
+** 2a. Delete 2/3 of the records in the database.
+**
+** 2b. Run lsm_work(nMerge=1).
+**
+** 2c. Insert as many records as were deleted in 2a.
+**
+** 2d. Check database content is as expected.
+**
+** 2e. If (Datatest4.bReopen) is true, close and reopen the database.
+*/
+struct Datatest4 {
+ /* Datasource definition */
+ DatasourceDefn defn;
+
+ int nRec;
+ int nRepeat;
+ int bReopen;
+};
+
+static void doDataTest4(
+ const char *zSystem, /* Database system to test */
+ Datatest4 *p, /* Structure containing test parameters */
+ int *pRc /* OUT: Error code */
+){
+ lsm_db *db = 0;
+ TestDb *pDb;
+ TestDb *pControl;
+ Datasource *pData;
+ int i;
+ int rc = 0;
+ int iDot = 0;
+ int bMultiThreaded = 0; /* True for MT LSM database */
+
+ int nRecOn3 = (p->nRec / 3);
+ int iData = 0;
+
+ /* Start the test case, open a database and allocate the datasource. */
+ rc = testControlDb(&pControl);
+ pDb = testOpen(zSystem, 1, &rc);
+ pData = testDatasourceNew(&p->defn);
+ if( rc==0 ){
+ db = tdb_lsm(pDb);
+ bMultiThreaded = tdb_lsm_multithread(pDb);
+ }
+
+ testWriteDatasourceRange(pControl, pData, iData, nRecOn3*3, &rc);
+ testWriteDatasourceRange(pDb, pData, iData, nRecOn3*3, &rc);
+
+ for(i=0; rc==0 && i<p->nRepeat; i++){
+
+ testDeleteDatasourceRange(pControl, pData, iData, nRecOn3*2, &rc);
+ testDeleteDatasourceRange(pDb, pData, iData, nRecOn3*2, &rc);
+
+ if( db ){
+ int nDone;
+#if 0
+ fprintf(stderr, "lsm_work() start...\n"); fflush(stderr);
+#endif
+ do {
+ nDone = 0;
+ rc = lsm_work(db, 1, (1<<30), &nDone);
+ }while( rc==0 && nDone>0 );
+ if( bMultiThreaded && rc==LSM_BUSY ) rc = LSM_OK;
+#if 0
+ fprintf(stderr, "lsm_work() done...\n"); fflush(stderr);
+#endif
+ }
+
+if( i+1<p->nRepeat ){
+ iData += (nRecOn3*2);
+ testWriteDatasourceRange(pControl, pData, iData+nRecOn3, nRecOn3*2, &rc);
+ testWriteDatasourceRange(pDb, pData, iData+nRecOn3, nRecOn3*2, &rc);
+
+ testCompareDb(pData, nRecOn3*3, iData, pControl, pDb, &rc);
+
+ /* If Datatest4.bReopen is true, close and reopen the database */
+ if( p->bReopen ){
+ testReopen(&pDb, &rc);
+ if( rc==0 ) db = tdb_lsm(pDb);
+ }
+}
+
+ /* Update the progress dots... */
+ testCaseProgress(i, p->nRepeat, testCaseNDot(), &iDot);
+ }
+
+ testClose(&pDb);
+ testClose(&pControl);
+ testDatasourceFree(pData);
+ testCaseFinish(rc);
+ *pRc = rc;
+}
+
+static char *getName4(const char *zSystem, Datatest4 *pTest){
+ char *zRet;
+ char *zData;
+ zData = testDatasourceName(&pTest->defn);
+ zRet = testMallocPrintf("data4.%s.%s.%d.%d.%d",
+ zSystem, zData, pTest->nRec, pTest->nRepeat, pTest->bReopen
+ );
+ testFree(zData);
+ return zRet;
+}
+
+void test_data_4(
+ const char *zSystem, /* Database system name */
+ const char *zPattern, /* Run test cases that match this pattern */
+ int *pRc /* IN/OUT: Error code */
+){
+ Datatest4 aTest[] = {
+ /* defn, nRec, nRepeat, bReopen */
+ { {DATA_RANDOM, 20,25, 500,600}, 10000, 10, 0 },
+ { {DATA_RANDOM, 20,25, 500,600}, 10000, 10, 1 },
+ };
+
+ int i;
+
+ for(i=0; *pRc==LSM_OK && i<ArraySize(aTest); i++){
+ char *zName = getName4(zSystem, &aTest[i]);
+ if( testCaseBegin(pRc, zPattern, "%s", zName) ){
+ doDataTest4(zSystem, &aTest[i], pRc);
+ }
+ testFree(zName);
+ }
+}
diff --git a/ext/lsm1/lsm-test/lsmtest_bt.c b/ext/lsm1/lsm-test/lsmtest_bt.c
new file mode 100644
index 0000000..8a4f54a
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_bt.c
@@ -0,0 +1,71 @@
+
+#include "lsmtest.h"
+#include "bt.h"
+
+int do_bt(int nArg, char **azArg){
+ struct Option {
+ const char *zName;
+ int bPgno;
+ int eOpt;
+ } aOpt [] = {
+ { "dbhdr", 0, BT_INFO_HDRDUMP },
+ { "filename", 0, BT_INFO_FILENAME },
+ { "block_freelist", 0, BT_INFO_BLOCK_FREELIST },
+ { "page_freelist", 0, BT_INFO_PAGE_FREELIST },
+ { "filename", 0, BT_INFO_FILENAME },
+ { "page", 1, BT_INFO_PAGEDUMP },
+ { "page_ascii", 1, BT_INFO_PAGEDUMP_ASCII },
+ { "leaks", 0, BT_INFO_PAGE_LEAKS },
+ { 0, 0 }
+ };
+ int iOpt;
+ int rc;
+ bt_info buf;
+ char *zOpt;
+ char *zFile;
+
+ bt_db *db = 0;
+
+ if( nArg<2 ){
+ testPrintUsage("FILENAME OPTION ...");
+ return -1;
+ }
+ zFile = azArg[0];
+ zOpt = azArg[1];
+
+ rc = testArgSelect(aOpt, "option", zOpt, &iOpt);
+ if( rc!=0 ) return rc;
+ if( nArg!=2+aOpt[iOpt].bPgno ){
+ testPrintFUsage("FILENAME %s %s", zOpt, aOpt[iOpt].bPgno ? "PGNO" : "");
+ return -4;
+ }
+
+ rc = sqlite4BtNew(sqlite4_env_default(), 0, &db);
+ if( rc!=SQLITE4_OK ){
+ testPrintError("sqlite4BtNew() failed: %d", rc);
+ return -2;
+ }
+ rc = sqlite4BtOpen(db, zFile);
+ if( rc!=SQLITE4_OK ){
+ testPrintError("sqlite4BtOpen() failed: %d", rc);
+ return -3;
+ }
+
+ buf.eType = aOpt[iOpt].eOpt;
+ buf.pgno = 0;
+ sqlite4_buffer_init(&buf.output, 0);
+
+ if( aOpt[iOpt].bPgno ){
+ buf.pgno = (u32)atoi(azArg[2]);
+ }
+
+ rc = sqlite4BtControl(db, BT_CONTROL_INFO, &buf);
+ if( rc!=SQLITE4_OK ){
+ testPrintError("sqlite4BtControl() failed: %d\n", rc);
+ return -4;
+ }
+
+ printf("%s\n", (char*)buf.output.p);
+ sqlite4_buffer_clear(&buf.output);
+ return 0;
+}
diff --git a/ext/lsm1/lsm-test/lsmtest_datasource.c b/ext/lsm1/lsm-test/lsmtest_datasource.c
new file mode 100644
index 0000000..0b0fd94
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_datasource.c
@@ -0,0 +1,96 @@
+
+
+#include "lsmtest.h"
+
+struct Datasource {
+ int eType;
+
+ int nMinKey;
+ int nMaxKey;
+ int nMinVal;
+ int nMaxVal;
+
+ char *aKey;
+ char *aVal;
+};
+
+void testDatasourceEntry(
+ Datasource *p,
+ int iData,
+ void **ppKey, int *pnKey,
+ void **ppVal, int *pnVal
+){
+ assert( (ppKey==0)==(pnKey==0) );
+ assert( (ppVal==0)==(pnVal==0) );
+
+ if( ppKey ){
+ int nKey = 0;
+ switch( p->eType ){
+ case TEST_DATASOURCE_RANDOM: {
+ int nRange = (1 + p->nMaxKey - p->nMinKey);
+ nKey = (int)( testPrngValue((u32)iData) % nRange ) + p->nMinKey;
+ testPrngString((u32)iData, p->aKey, nKey);
+ break;
+ }
+ case TEST_DATASOURCE_SEQUENCE:
+ nKey = sprintf(p->aKey, "%012d", iData);
+ break;
+ }
+ *ppKey = p->aKey;
+ *pnKey = nKey;
+ }
+ if( ppVal ){
+ u32 nVal = testPrngValue((u32)iData)%(1+p->nMaxVal-p->nMinVal)+p->nMinVal;
+ testPrngString((u32)~iData, p->aVal, (int)nVal);
+ *ppVal = p->aVal;
+ *pnVal = (int)nVal;
+ }
+}
+
+void testDatasourceFree(Datasource *p){
+ testFree(p);
+}
+
+/*
+** Return a pointer to a nul-terminated string that corresponds to the
+** contents of the datasource-definition passed as the first argument.
+** The caller should eventually free the returned pointer using testFree().
+*/
+char *testDatasourceName(const DatasourceDefn *p){
+ char *zRet;
+ zRet = testMallocPrintf("%s.(%d-%d).(%d-%d)",
+ (p->eType==TEST_DATASOURCE_SEQUENCE ? "seq" : "rnd"),
+ p->nMinKey, p->nMaxKey,
+ p->nMinVal, p->nMaxVal
+ );
+ return zRet;
+}
+
+Datasource *testDatasourceNew(const DatasourceDefn *pDefn){
+ Datasource *p;
+ int nMinKey;
+ int nMaxKey;
+ int nMinVal;
+ int nMaxVal;
+
+ if( pDefn->eType==TEST_DATASOURCE_SEQUENCE ){
+ nMinKey = 128;
+ nMaxKey = 128;
+ }else{
+ nMinKey = MAX(0, pDefn->nMinKey);
+ nMaxKey = MAX(nMinKey, pDefn->nMaxKey);
+ }
+ nMinVal = MAX(0, pDefn->nMinVal);
+ nMaxVal = MAX(nMinVal, pDefn->nMaxVal);
+
+ p = (Datasource *)testMalloc(sizeof(Datasource) + nMaxKey + nMaxVal + 1);
+ p->eType = pDefn->eType;
+ p->nMinKey = nMinKey;
+ p->nMinVal = nMinVal;
+ p->nMaxKey = nMaxKey;
+ p->nMaxVal = nMaxVal;
+
+ p->aKey = (char *)&p[1];
+ p->aVal = &p->aKey[nMaxKey];
+ return p;
+};
diff --git a/ext/lsm1/lsm-test/lsmtest_func.c b/ext/lsm1/lsm-test/lsmtest_func.c
new file mode 100644
index 0000000..eb8346a
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_func.c
@@ -0,0 +1,177 @@
+
+#include "lsmtest.h"
+
+
+int do_work(int nArg, char **azArg){
+ struct Option {
+ const char *zName;
+ } aOpt [] = {
+ { "-nmerge" },
+ { "-nkb" },
+ { 0 }
+ };
+
+ lsm_db *pDb;
+ int rc;
+ int i;
+ const char *zDb;
+ int nMerge = 1;
+ int nKB = (1<<30);
+
+ if( nArg==0 ) goto usage;
+ zDb = azArg[nArg-1];
+ for(i=0; i<(nArg-1); i++){
+ int iSel;
+ rc = testArgSelect(aOpt, "option", azArg[i], &iSel);
+ if( rc ) return rc;
+ switch( iSel ){
+ case 0:
+ i++;
+ if( i==(nArg-1) ) goto usage;
+ nMerge = atoi(azArg[i]);
+ break;
+ case 1:
+ i++;
+ if( i==(nArg-1) ) goto usage;
+ nKB = atoi(azArg[i]);
+ break;
+ }
+ }
+
+ rc = lsm_new(0, &pDb);
+ if( rc!=LSM_OK ){
+ testPrintError("lsm_open(): rc=%d\n", rc);
+ }else{
+ rc = lsm_open(pDb, zDb);
+ if( rc!=LSM_OK ){
+ testPrintError("lsm_open(): rc=%d\n", rc);
+ }else{
+ int n = -1;
+ lsm_config(pDb, LSM_CONFIG_BLOCK_SIZE, &n);
+ n = n*2;
+ lsm_config(pDb, LSM_CONFIG_AUTOCHECKPOINT, &n);
+
+ rc = lsm_work(pDb, nMerge, nKB, 0);
+ if( rc!=LSM_OK ){
+ testPrintError("lsm_work(): rc=%d\n", rc);
+ }
+ }
+ }
+ if( rc==LSM_OK ){
+ rc = lsm_checkpoint(pDb, 0);
+ }
+
+ lsm_close(pDb);
+ return rc;
+
+ usage:
+ testPrintUsage("?-optimize? ?-n N? DATABASE");
+ return -1;
+}
+
+
+/*
+** lsmtest show ?-config LSM-CONFIG? DATABASE ?COMMAND ?PGNO??
+*/
+int do_show(int nArg, char **azArg){
+ lsm_db *pDb;
+ int rc;
+ const char *zDb;
+
+ int eOpt = LSM_INFO_DB_STRUCTURE;
+ unsigned int iPg = 0;
+ int bConfig = 0;
+ const char *zConfig = "";
+
+ struct Option {
+ const char *zName;
+ int bConfig;
+ int eOpt;
+ } aOpt [] = {
+ { "array", 0, LSM_INFO_ARRAY_STRUCTURE },
+ { "array-pages", 0, LSM_INFO_ARRAY_PAGES },
+ { "blocksize", 1, LSM_CONFIG_BLOCK_SIZE },
+ { "pagesize", 1, LSM_CONFIG_PAGE_SIZE },
+ { "freelist", 0, LSM_INFO_FREELIST },
+ { "page-ascii", 0, LSM_INFO_PAGE_ASCII_DUMP },
+ { "page-hex", 0, LSM_INFO_PAGE_HEX_DUMP },
+ { 0, 0 }
+ };
+
+ char *z = 0;
+ int iDb = 0; /* Index of DATABASE in azArg[] */
+
+ /* Check if there is a "-config" option: */
+ if( nArg>2 && strlen(azArg[0])>1
+ && memcmp(azArg[0], "-config", strlen(azArg[0]))==0
+ ){
+ zConfig = azArg[1];
+ iDb = 2;
+ }
+ if( nArg<(iDb+1) ) goto usage;
+
+ if( nArg>(iDb+1) ){
+ rc = testArgSelect(aOpt, "option", azArg[iDb+1], &eOpt);
+ if( rc!=0 ) return rc;
+ bConfig = aOpt[eOpt].bConfig;
+ eOpt = aOpt[eOpt].eOpt;
+ if( (bConfig==0 && eOpt==LSM_INFO_FREELIST)
+ || (bConfig==1 && eOpt==LSM_CONFIG_BLOCK_SIZE)
+ || (bConfig==1 && eOpt==LSM_CONFIG_PAGE_SIZE)
+ ){
+ if( nArg!=(iDb+2) ) goto usage;
+ }else{
+ if( nArg!=(iDb+3) ) goto usage;
+ iPg = atoi(azArg[iDb+2]);
+ }
+ }
+ zDb = azArg[iDb];
+
+ rc = lsm_new(0, &pDb);
+ tdb_lsm_configure(pDb, zConfig);
+ if( rc!=LSM_OK ){
+ testPrintError("lsm_new(): rc=%d\n", rc);
+ }else{
+ rc = lsm_open(pDb, zDb);
+ if( rc!=LSM_OK ){
+ testPrintError("lsm_open(): rc=%d\n", rc);
+ }
+ }
+
+ if( rc==LSM_OK ){
+ if( bConfig==0 ){
+ switch( eOpt ){
+ case LSM_INFO_DB_STRUCTURE:
+ case LSM_INFO_FREELIST:
+ rc = lsm_info(pDb, eOpt, &z);
+ break;
+ case LSM_INFO_ARRAY_STRUCTURE:
+ case LSM_INFO_ARRAY_PAGES:
+ case LSM_INFO_PAGE_ASCII_DUMP:
+ case LSM_INFO_PAGE_HEX_DUMP:
+ rc = lsm_info(pDb, eOpt, iPg, &z);
+ break;
+ default:
+ assert( !"no chance" );
+ }
+
+ if( rc==LSM_OK ){
+ printf("%s\n", z ? z : "");
+ fflush(stdout);
+ }
+ lsm_free(lsm_get_env(pDb), z);
+ }else{
+ int iRes = -1;
+ lsm_config(pDb, eOpt, &iRes);
+ printf("%d\n", iRes);
+ fflush(stdout);
+ }
+ }
+
+ lsm_close(pDb);
+ return rc;
+
+ usage:
+ testPrintUsage("DATABASE ?array|page-ascii|page-hex PGNO?");
+ return -1;
+}
diff --git a/ext/lsm1/lsm-test/lsmtest_io.c b/ext/lsm1/lsm-test/lsmtest_io.c
new file mode 100644
index 0000000..7aa5d10
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_io.c
@@ -0,0 +1,248 @@
+
+/*
+** SUMMARY
+**
+** This file implements the 'io' subcommand of the test program. It is used
+** for testing the performance of various combinations of write() and fsync()
+** system calls. All operations occur on a single file, which may or may not
+** exist when a test is started.
+**
+** A test consists of a series of commands. Each command is either a write
+** or an fsync. A write is specified as "<amount>@<offset>", where <amount>
+** is the amount of data written, and <offset> is the offset of the file
+** to write to. An <amount> or an <offset> is specified as an integer number
+** of bytes. Or, if postfixed with a "K", "M" or "G", an integer number of
+** KB, MB or GB, respectively. An fsync is simply "S". All commands are
+** case-insensitive.
+**
+** Example test program:
+**
+** 2M@6M 1492K@4M S 4096@4K S
+**
+** This program writes 2 MB of data starting at the offset 6MB offset of
+** the file, followed by 1492 KB of data written at the 4MB offset of the
+** file, followed by a call to fsync(), a write of 4KB of data at byte
+** offset 4096, and finally another call to fsync().
+**
+** Commands may either be specified on the command line (one command per
+** command line argument) or read from stdin. Commands read from stdin
+** must be separated by white-space.
+**
+** COMMAND LINE INVOCATION
+**
+** The sub-command implemented in this file must be invoked with at least
+** two arguments - the path to the file to write to and the page-size to
+** use for writing. If there are more than two arguments, then each
+** subsequent argument is assumed to be a test command. If there are exactly
+** two arguments, the test commands are read from stdin.
+**
+** A write command does not result in a single call to system call write().
+** Instead, the specified region is written sequentially using one or
+** more calls to write(), each of which writes not more than one page of
+** data. For example, if the page-size is 4KB, the command "2M@6M" results
+** in 512 calls to write(), each of which writes 4KB of data.
+**
+** EXAMPLES
+**
+** Two equivalent examples:
+**
+** $ lsmtest io testfile.db 4KB 2M@6M 1492K@4M S 4096@4K S
+** 3544K written in 129 ms
+** $ echo "2M@6M 1492K@4M S 4096@4K S" | lsmtest io testfile.db 4096
+** 3544K written in 127 ms
+**
+*/
+
+#include "lsmtest.h"
+
+typedef struct IoContext IoContext;
+
+struct IoContext {
+ int fd;
+ int nWrite;
+};
+
+/*
+** As isspace(3)
+*/
+static int safe_isspace(char c){
+ if( c&0x80) return 0;
+ return isspace(c);
+}
+
+/*
+** As isdigit(3)
+*/
+static int safe_isdigit(char c){
+ if( c&0x80) return 0;
+ return isdigit(c);
+}
+
+static i64 getNextSize(char *zIn, char **pzOut, int *pRc){
+ i64 iRet = 0;
+ if( *pRc==0 ){
+ char *z = zIn;
+
+ if( !safe_isdigit(*z) ){
+ *pRc = 1;
+ return 0;
+ }
+
+ /* Process digits */
+ while( safe_isdigit(*z) ){
+ iRet = iRet*10 + (*z - '0');
+ z++;
+ }
+
+ /* Process suffix */
+ switch( *z ){
+ case 'k': case 'K':
+ iRet = iRet * 1024;
+ z++;
+ break;
+
+ case 'm': case 'M':
+ iRet = iRet * 1024 * 1024;
+ z++;
+ break;
+
+ case 'g': case 'G':
+ iRet = iRet * 1024 * 1024 * 1024;
+ z++;
+ break;
+ }
+
+ if( pzOut ) *pzOut = z;
+ }
+ return iRet;
+}
+
+static int doOneCmd(
+ IoContext *pCtx,
+ u8 *aData,
+ int pgsz,
+ char *zCmd,
+ char **pzOut
+){
+ char c;
+ char *z = zCmd;
+
+ while( safe_isspace(*z) ) z++;
+ c = *z;
+
+ if( c==0 ){
+ if( pzOut ) *pzOut = z;
+ return 0;
+ }
+
+ if( c=='s' || c=='S' ){
+ if( pzOut ) *pzOut = &z[1];
+ return fdatasync(pCtx->fd);
+ }
+
+ if( safe_isdigit(c) ){
+ i64 iOff = 0;
+ int nByte = 0;
+ int rc = 0;
+ int nPg;
+ int iPg;
+
+ nByte = (int)getNextSize(z, &z, &rc);
+ if( rc || *z!='@' ) goto bad_command;
+ z++;
+ iOff = getNextSize(z, &z, &rc);
+ if( rc || (safe_isspace(*z)==0 && *z!='\0') ) goto bad_command;
+ if( pzOut ) *pzOut = z;
+
+ nPg = (nByte+pgsz-1) / pgsz;
+ lseek(pCtx->fd, (off_t)iOff, SEEK_SET);
+ for(iPg=0; iPg<nPg; iPg++){
+ write(pCtx->fd, aData, pgsz);
+ }
+ pCtx->nWrite += nByte/1024;
+
+ return 0;
+ }
+
+ bad_command:
+ testPrintError("unrecognized command: %s", zCmd);
+ return 1;
+}
+
+static int readStdin(char **pzOut){
+ int nAlloc = 128;
+ char *zOut = 0;
+ int nOut = 0;
+
+ while( !feof(stdin) ){
+ int nRead;
+
+ nAlloc = nAlloc*2;
+ zOut = realloc(zOut, nAlloc);
+ nRead = fread(&zOut[nOut], 1, nAlloc-nOut-1, stdin);
+
+ if( nRead==0 ) break;
+ nOut += nRead;
+ zOut[nOut] = '\0';
+ }
+
+ *pzOut = zOut;
+ return 0;
+}
+
+int do_io(int nArg, char **azArg){
+ IoContext ctx;
+ int pgsz;
+ char *zFile;
+ char *zPgsz;
+ int i;
+ int rc = 0;
+
+ char *zStdin = 0;
+ char *z;
+
+ u8 *aData;
+
+ memset(&ctx, 0, sizeof(IoContext));
+ if( nArg<2 ){
+ testPrintUsage("FILE PGSZ ?CMD-1 ...?");
+ return -1;
+ }
+ zFile = azArg[0];
+ zPgsz = azArg[1];
+
+ pgsz = (int)getNextSize(zPgsz, 0, &rc);
+ if( pgsz<=0 ){
+ testPrintError("Ridiculous page size: %d", pgsz);
+ return -1;
+ }
+ aData = malloc(pgsz);
+ memset(aData, 0x77, pgsz);
+
+ ctx.fd = open(zFile, O_RDWR|O_CREAT|_O_BINARY, 0644);
+ if( ctx.fd<0 ){
+ perror("open: ");
+ return -1;
+ }
+
+ if( nArg==2 ){
+ readStdin(&zStdin);
+ testTimeInit();
+ z = zStdin;
+ while( *z && rc==0 ){
+ rc = doOneCmd(&ctx, aData, pgsz, z, &z);
+ }
+ }else{
+ testTimeInit();
+ for(i=2; i<nArg; i++){
+ rc = doOneCmd(&ctx, aData, pgsz, azArg[i], 0);
+ }
+ }
+
+ printf("%dK written in %d ms\n", ctx.nWrite, testTimeGet());
+
+ free(zStdin);
+ close(ctx.fd);
+
+ return 0;
+}
diff --git a/ext/lsm1/lsm-test/lsmtest_main.c b/ext/lsm1/lsm-test/lsmtest_main.c
new file mode 100644
index 0000000..f4a3ac0
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_main.c
@@ -0,0 +1,1548 @@
+
+#include "lsmtest.h"
+#include <sqlite3.h>
+
+void test_failed(){
+ assert( 0 );
+ return;
+}
+
+#define testSetError(rc) testSetErrorFunc(rc, pRc, __FILE__, __LINE__)
+static void testSetErrorFunc(int rc, int *pRc, const char *zFile, int iLine){
+ if( rc ){
+ *pRc = rc;
+ fprintf(stderr, "FAILED (%s:%d) rc=%d ", zFile, iLine, rc);
+ test_failed();
+ }
+}
+
+static int lsm_memcmp(u8 *a, u8 *b, int c){
+ int i;
+ for(i=0; i<c; i++){
+ if( a[i]!=b[i] ) return a[i] - b[i];
+ }
+ return 0;
+}
+
+/*
+** A test utility function.
+*/
+void testFetch(
+ TestDb *pDb, /* Database handle */
+ void *pKey, int nKey, /* Key to query database for */
+ void *pVal, int nVal, /* Expected value */
+ int *pRc /* IN/OUT: Error code */
+){
+ if( *pRc==0 ){
+ void *pDbVal;
+ int nDbVal;
+ int rc;
+
+ static int nCall = 0; nCall++;
+
+ rc = tdb_fetch(pDb, pKey, nKey, &pDbVal, &nDbVal);
+ testSetError(rc);
+ if( rc==0 && (nVal!=nDbVal || (nVal>0 && lsm_memcmp(pVal, pDbVal, nVal))) ){
+ testSetError(1);
+ }
+ }
+}
+
+void testWrite(
+ TestDb *pDb, /* Database handle */
+ void *pKey, int nKey, /* Key to query database for */
+ void *pVal, int nVal, /* Value to write */
+ int *pRc /* IN/OUT: Error code */
+){
+ if( *pRc==0 ){
+ int rc;
+static int nCall = 0;
+nCall++;
+ rc = tdb_write(pDb, pKey, nKey, pVal, nVal);
+ testSetError(rc);
+ }
+}
+void testDelete(
+ TestDb *pDb, /* Database handle */
+ void *pKey, int nKey, /* Key to query database for */
+ int *pRc /* IN/OUT: Error code */
+){
+ if( *pRc==0 ){
+ int rc;
+ *pRc = rc = tdb_delete(pDb, pKey, nKey);
+ testSetError(rc);
+ }
+}
+void testDeleteRange(
+ TestDb *pDb, /* Database handle */
+ void *pKey1, int nKey1,
+ void *pKey2, int nKey2,
+ int *pRc /* IN/OUT: Error code */
+){
+ if( *pRc==0 ){
+ int rc;
+ *pRc = rc = tdb_delete_range(pDb, pKey1, nKey1, pKey2, nKey2);
+ testSetError(rc);
+ }
+}
+
+void testBegin(TestDb *pDb, int iTrans, int *pRc){
+ if( *pRc==0 ){
+ int rc;
+ rc = tdb_begin(pDb, iTrans);
+ testSetError(rc);
+ }
+}
+void testCommit(TestDb *pDb, int iTrans, int *pRc){
+ if( *pRc==0 ){
+ int rc;
+ rc = tdb_commit(pDb, iTrans);
+ testSetError(rc);
+ }
+}
+#if 0 /* unused */
+static void testRollback(TestDb *pDb, int iTrans, int *pRc){
+ if( *pRc==0 ){
+ int rc;
+ rc = tdb_rollback(pDb, iTrans);
+ testSetError(rc);
+ }
+}
+#endif
+
+void testWriteStr(
+ TestDb *pDb, /* Database handle */
+ const char *zKey, /* Key to query database for */
+ const char *zVal, /* Value to write */
+ int *pRc /* IN/OUT: Error code */
+){
+ int nVal = (zVal ? strlen(zVal) : 0);
+ testWrite(pDb, (void *)zKey, strlen(zKey), (void *)zVal, nVal, pRc);
+}
+
+#if 0 /* unused */
+static void testDeleteStr(TestDb *pDb, const char *zKey, int *pRc){
+ testDelete(pDb, (void *)zKey, strlen(zKey), pRc);
+}
+#endif
+void testFetchStr(
+ TestDb *pDb, /* Database handle */
+ const char *zKey, /* Key to query database for */
+ const char *zVal, /* Value to write */
+ int *pRc /* IN/OUT: Error code */
+){
+ int nVal = (zVal ? strlen(zVal) : 0);
+ testFetch(pDb, (void *)zKey, strlen(zKey), (void *)zVal, nVal, pRc);
+}
+
+void testFetchCompare(
+ TestDb *pControl,
+ TestDb *pDb,
+ void *pKey, int nKey,
+ int *pRc
+){
+ int rc;
+ void *pDbVal1;
+ void *pDbVal2;
+ int nDbVal1;
+ int nDbVal2;
+
+ static int nCall = 0;
+ nCall++;
+
+ rc = tdb_fetch(pControl, pKey, nKey, &pDbVal1, &nDbVal1);
+ testSetError(rc);
+
+ rc = tdb_fetch(pDb, pKey, nKey, &pDbVal2, &nDbVal2);
+ testSetError(rc);
+
+ if( *pRc==0
+ && (nDbVal1!=nDbVal2 || (nDbVal1>0 && memcmp(pDbVal1, pDbVal2, nDbVal1)))
+ ){
+ testSetError(1);
+ }
+}
+
+typedef struct ScanResult ScanResult;
+struct ScanResult {
+ TestDb *pDb;
+
+ int nRow;
+ u32 cksum1;
+ u32 cksum2;
+ void *pKey1; int nKey1;
+ void *pKey2; int nKey2;
+
+ int bReverse;
+ int nPrevKey;
+ u8 aPrevKey[256];
+};
+
+static int keyCompare(void *pKey1, int nKey1, void *pKey2, int nKey2){
+ int res;
+ res = memcmp(pKey1, pKey2, MIN(nKey1, nKey2));
+ if( res==0 ){
+ res = nKey1 - nKey2;
+ }
+ return res;
+}
+
+int test_scan_debug = 0;
+
+static void scanCompareCb(
+ void *pCtx,
+ void *pKey, int nKey,
+ void *pVal, int nVal
+){
+ ScanResult *p = (ScanResult *)pCtx;
+ u8 *aKey = (u8 *)pKey;
+ u8 *aVal = (u8 *)pVal;
+ int i;
+
+ if( test_scan_debug ){
+ printf("%d: %.*s\n", p->nRow, nKey, (char *)pKey);
+ fflush(stdout);
+ }
+#if 0
+ if( test_scan_debug ) printf("%.20s\n", (char *)pVal);
+#endif
+
+#if 0
+ /* Check tdb_fetch() matches */
+ int rc = 0;
+ testFetch(p->pDb, pKey, nKey, pVal, nVal, &rc);
+ assert( rc==0 );
+#endif
+
+ /* Update the checksum data */
+ p->nRow++;
+ for(i=0; i<nKey; i++){
+ p->cksum1 += ((int)aKey[i] << (i&0x0F));
+ p->cksum2 += p->cksum1;
+ }
+ for(i=0; i<nVal; i++){
+ p->cksum1 += ((int)aVal[i] << (i&0x0F));
+ p->cksum2 += p->cksum1;
+ }
+
+ /* Check that the delivered row is not out of order. */
+ if( nKey<(int)sizeof(p->aPrevKey) ){
+ if( p->nPrevKey ){
+ int res = keyCompare(p->aPrevKey, p->nPrevKey, pKey, nKey);
+ if( (res<0 && p->bReverse) || (res>0 && p->bReverse==0) ){
+ testPrintError("Returned key out of order at %s:%d\n",
+ __FILE__, __LINE__
+ );
+ }
+ }
+
+ p->nPrevKey = nKey;
+ memcpy(p->aPrevKey, pKey, MIN(p->nPrevKey, nKey));
+ }
+
+ /* Check that the delivered row is within range. */
+ if( p->pKey1 && (
+ (memcmp(p->pKey1, pKey, MIN(p->nKey1, nKey))>0)
+ || (memcmp(p->pKey1, pKey, MIN(p->nKey1, nKey))==0 && p->nKey1>nKey)
+ )){
+ testPrintError("Returned key too small at %s:%d\n", __FILE__, __LINE__);
+ }
+ if( p->pKey2 && (
+ (memcmp(p->pKey2, pKey, MIN(p->nKey2, nKey))<0)
+ || (memcmp(p->pKey2, pKey, MIN(p->nKey2, nKey))==0 && p->nKey2<nKey)
+ )){
+ testPrintError("Returned key too large at %s:%d\n", __FILE__, __LINE__);
+ }
+
+}
+
+/*
+** Scan the contents of the two databases. Check that they match.
+*/
+void testScanCompare(
+ TestDb *pDb1, /* Control (trusted) database */
+ TestDb *pDb2, /* Database being tested */
+ int bReverse,
+ void *pKey1, int nKey1,
+ void *pKey2, int nKey2,
+ int *pRc
+){
+ static int nCall = 0; nCall++;
+ if( *pRc==0 ){
+ ScanResult res1;
+ ScanResult res2;
+ void *pRes1 = (void *)&res1;
+ void *pRes2 = (void *)&res2;
+
+ memset(&res1, 0, sizeof(ScanResult));
+ memset(&res2, 0, sizeof(ScanResult));
+
+ res1.pDb = pDb1;
+ res1.nKey1 = nKey1; res1.pKey1 = pKey1;
+ res1.nKey2 = nKey2; res1.pKey2 = pKey2;
+ res1.bReverse = bReverse;
+ res2.pDb = pDb2;
+ res2.nKey1 = nKey1; res2.pKey1 = pKey1;
+ res2.nKey2 = nKey2; res2.pKey2 = pKey2;
+ res2.bReverse = bReverse;
+
+ tdb_scan(pDb1, pRes1, bReverse, pKey1, nKey1, pKey2, nKey2, scanCompareCb);
+if( test_scan_debug ) printf("\n\n\n");
+ tdb_scan(pDb2, pRes2, bReverse, pKey1, nKey1, pKey2, nKey2, scanCompareCb);
+if( test_scan_debug ) printf("\n\n\n");
+
+ if( res1.nRow!=res2.nRow
+ || res1.cksum1!=res2.cksum1
+ || res1.cksum2!=res2.cksum2
+ ){
+ printf("expected: %d %X %X\n", res1.nRow, res1.cksum1, res1.cksum2);
+ printf("got: %d %X %X\n", res2.nRow, res2.cksum1, res2.cksum2);
+ testSetError(1);
+ *pRc = 1;
+ }
+ }
+}
+
+void testClose(TestDb **ppDb){
+ tdb_close(*ppDb);
+ *ppDb = 0;
+}
+
+TestDb *testOpen(const char *zSystem, int bClear, int *pRc){
+ TestDb *pDb = 0;
+ if( *pRc==0 ){
+ int rc;
+ rc = tdb_open(zSystem, 0, bClear, &pDb);
+ if( rc!=0 ){
+ testSetError(rc);
+ *pRc = rc;
+ }
+ }
+ return pDb;
+}
+
+void testReopen(TestDb **ppDb, int *pRc){
+ if( *pRc==0 ){
+ const char *zLib;
+ zLib = tdb_library_name(*ppDb);
+ testClose(ppDb);
+ *pRc = tdb_open(zLib, 0, 0, ppDb);
+ }
+}
+
+
+#if 0 /* unused */
+static void testSystemSelect(const char *zSys, int *piSel, int *pRc){
+ if( *pRc==0 ){
+ struct SysName { const char *zName; } *aName;
+ int nSys;
+ int i;
+
+ for(nSys=0; tdb_system_name(nSys); nSys++);
+ aName = malloc(sizeof(struct SysName) * (nSys+1));
+ for(i=0; i<=nSys; i++){
+ aName[i].zName = tdb_system_name(i);
+ }
+
+ *pRc = testArgSelect(aName, "db", zSys, piSel);
+ free(aName);
+ }
+}
+#endif
+
+char *testMallocVPrintf(const char *zFormat, va_list ap){
+ int nByte;
+ va_list copy;
+ char *zRet;
+
+ __va_copy(copy, ap);
+ nByte = vsnprintf(0, 0, zFormat, copy);
+ va_end(copy);
+
+ assert( nByte>=0 );
+ zRet = (char *)testMalloc(nByte+1);
+ vsnprintf(zRet, nByte+1, zFormat, ap);
+ return zRet;
+}
+
+char *testMallocPrintf(const char *zFormat, ...){
+ va_list ap;
+ char *zRet;
+
+ va_start(ap, zFormat);
+ zRet = testMallocVPrintf(zFormat, ap);
+ va_end(ap);
+
+ return zRet;
+}
+
+
+/*
+** A wrapper around malloc(3).
+**
+** This function should be used for all allocations made by test procedures.
+** It has the following properties:
+**
+** * Test code may assume that allocations may not fail.
+** * Returned memory is always zeroed.
+**
+** Allocations made using testMalloc() should be freed using testFree().
+*/
+void *testMalloc(int n){
+ u8 *p = (u8*)malloc(n + 8);
+ memset(p, 0, n+8);
+ *(int*)p = n;
+ return (void*)&p[8];
+}
+
+void *testMallocCopy(void *pCopy, int nByte){
+ void *pRet = testMalloc(nByte);
+ memcpy(pRet, pCopy, nByte);
+ return pRet;
+}
+
+void *testRealloc(void *ptr, int n){
+ if( ptr ){
+ u8 *p = (u8*)ptr - 8;
+ int nOrig = *(int*)p;
+ p = (u8*)realloc(p, n+8);
+ if( nOrig<n ){
+ memset(&p[8+nOrig], 0, n-nOrig);
+ }
+ *(int*)p = n;
+ return (void*)&p[8];
+ }
+ return testMalloc(n);
+}
+
+/*
+** Free an allocation made by an earlier call to testMalloc().
+*/
+void testFree(void *ptr){
+ if( ptr ){
+ u8 *p = (u8*)ptr - 8;
+ memset(p, 0x55, *(int*)p + 8);
+ free(p);
+ }
+}
+
+/*
+** String zPattern contains a glob pattern. Return true if zStr matches
+** the pattern, or false if it does not.
+*/
+int testGlobMatch(const char *zPattern, const char *zStr){
+ int i = 0;
+ int j = 0;
+
+ while( zPattern[i] ){
+ char p = zPattern[i];
+
+ if( p=='*' || p=='%' ){
+ do {
+ if( testGlobMatch(&zPattern[i+1], &zStr[j]) ) return 1;
+ }while( zStr[j++] );
+ return 0;
+ }
+
+ if( zStr[j]==0 || (p!='?' && p!=zStr[j]) ){
+ /* Match failed. */
+ return 0;
+ }
+
+ j++;
+ i++;
+ }
+
+ return (zPattern[i]==0 && zStr[j]==0);
+}
+
+/*
+** End of test utilities
+**************************************************************************/
+
+int do_test(int nArg, char **azArg){
+ int j;
+ int rc;
+ int nFail = 0;
+ const char *zPattern = 0;
+
+ if( nArg>1 ){
+ testPrintError("Usage: test ?PATTERN?\n");
+ return 1;
+ }
+ if( nArg==1 ){
+ zPattern = azArg[0];
+ }
+
+ for(j=0; tdb_system_name(j); j++){
+ rc = 0;
+
+ test_data_1(tdb_system_name(j), zPattern, &rc);
+ test_data_2(tdb_system_name(j), zPattern, &rc);
+ test_data_3(tdb_system_name(j), zPattern, &rc);
+ test_data_4(tdb_system_name(j), zPattern, &rc);
+ test_rollback(tdb_system_name(j), zPattern, &rc);
+ test_mc(tdb_system_name(j), zPattern, &rc);
+ test_mt(tdb_system_name(j), zPattern, &rc);
+
+ if( rc ) nFail++;
+ }
+
+ rc = 0;
+ test_oom(zPattern, &rc);
+ if( rc ) nFail++;
+
+ rc = 0;
+ test_api(zPattern, &rc);
+ if( rc ) nFail++;
+
+ rc = 0;
+ do_crash_test(zPattern, &rc);
+ if( rc ) nFail++;
+
+ rc = 0;
+ do_writer_crash_test(zPattern, &rc);
+ if( rc ) nFail++;
+
+ return (nFail!=0);
+}
+
+static lsm_db *configure_lsm_db(TestDb *pDb){
+ lsm_db *pLsm;
+ pLsm = tdb_lsm(pDb);
+ if( pLsm ){
+ tdb_lsm_config_str(pDb, "mmap=1 autowork=1 automerge=4 worker_automerge=4");
+ }
+ return pLsm;
+}
+
+typedef struct WriteHookEvent WriteHookEvent;
+struct WriteHookEvent {
+ i64 iOff;
+ int nData;
+ int nUs;
+};
+WriteHookEvent prev = {0, 0, 0};
+
+static void flushPrev(FILE *pOut){
+ if( prev.nData ){
+ fprintf(pOut, "w %s %lld %d %d\n", "d", prev.iOff, prev.nData, prev.nUs);
+ prev.nData = 0;
+ }
+}
+
+#if 0 /* unused */
+static void do_speed_write_hook2(
+ void *pCtx,
+ int bLog,
+ i64 iOff,
+ int nData,
+ int nUs
+){
+ FILE *pOut = (FILE *)pCtx;
+ if( bLog ) return;
+
+ if( prev.nData && nData && iOff==prev.iOff+prev.nData ){
+ prev.nData += nData;
+ prev.nUs += nUs;
+ }else{
+ flushPrev(pOut);
+ if( nData==0 ){
+ fprintf(pOut, "s %s 0 0 %d\n", (bLog ? "l" : "d"), nUs);
+ }else{
+ prev.iOff = iOff;
+ prev.nData = nData;
+ prev.nUs = nUs;
+ }
+ }
+}
+#endif
+
+#define ST_REPEAT 0
+#define ST_WRITE 1
+#define ST_PAUSE 2
+#define ST_FETCH 3
+#define ST_SCAN 4
+#define ST_NSCAN 5
+#define ST_KEYSIZE 6
+#define ST_VALSIZE 7
+#define ST_TRANS 8
+
+
+static void print_speed_test_help(){
+ printf(
+"\n"
+"Repeat the following $repeat times:\n"
+" 1. Insert $write key-value pairs. One transaction for each write op.\n"
+" 2. Pause for $pause ms.\n"
+" 3. Perform $fetch queries on the database.\n"
+"\n"
+" Keys are $keysize bytes in size. Values are $valsize bytes in size\n"
+" Both keys and values are pseudo-randomly generated\n"
+"\n"
+"Options are:\n"
+" -repeat $repeat (default value 10)\n"
+" -write $write (default value 10000)\n"
+" -pause $pause (default value 0)\n"
+" -fetch $fetch (default value 0)\n"
+" -keysize $keysize (default value 12)\n"
+" -valsize $valsize (default value 100)\n"
+" -system $system (default value \"lsm\")\n"
+" -trans $trans (default value 0)\n"
+"\n"
+);
+}
+
+int do_speed_test2(int nArg, char **azArg){
+ struct Option {
+ const char *zOpt;
+ int eVal;
+ int iDefault;
+ } aOpt[] = {
+ { "-repeat", ST_REPEAT, 10},
+ { "-write", ST_WRITE, 10000},
+ { "-pause", ST_PAUSE, 0},
+ { "-fetch", ST_FETCH, 0},
+ { "-scan", ST_SCAN, 0},
+ { "-nscan", ST_NSCAN, 0},
+ { "-keysize", ST_KEYSIZE, 12},
+ { "-valsize", ST_VALSIZE, 100},
+ { "-trans", ST_TRANS, 0},
+ { "-system", -1, 0},
+ { "help", -2, 0},
+ {0, 0, 0}
+ };
+ int i;
+ int aParam[9];
+ int rc = 0;
+ int bReadonly = 0;
+ int nContent = 0;
+
+ TestDb *pDb;
+ Datasource *pData;
+ DatasourceDefn defn = { TEST_DATASOURCE_RANDOM, 0, 0, 0, 0 };
+ char *zSystem = "";
+ int bLsm = 1;
+ FILE *pLog = 0;
+
+#ifdef NDEBUG
+ /* If NDEBUG is defined, disable the dynamic memory related checks in
+ ** lsmtest_mem.c. They slow things down. */
+ testMallocUninstall(tdb_lsm_env());
+#endif
+
+ /* Initialize aParam[] with default values. */
+ for(i=0; i<ArraySize(aOpt); i++){
+ if( aOpt[i].zOpt ) aParam[aOpt[i].eVal] = aOpt[i].iDefault;
+ }
+
+ /* Process the command line switches. */
+ for(i=0; i<nArg; i+=2){
+ int iSel;
+ rc = testArgSelect(aOpt, "switch", azArg[i], &iSel);
+ if( rc ){
+ return rc;
+ }
+ if( aOpt[iSel].eVal==-2 ){
+ print_speed_test_help();
+ return 0;
+ }
+ if( i+1==nArg ){
+ testPrintError("option %s requires an argument\n", aOpt[iSel].zOpt);
+ return 1;
+ }
+ if( aOpt[iSel].eVal>=0 ){
+ aParam[aOpt[iSel].eVal] = atoi(azArg[i+1]);
+ }else{
+ zSystem = azArg[i+1];
+ bLsm = 0;
+#if 0
+ for(j=0; zSystem[j]; j++){
+ if( zSystem[j]=='=' ) bLsm = 1;
+ }
+#endif
+ }
+ }
+
+ printf("#");
+ for(i=0; i<ArraySize(aOpt); i++){
+ if( aOpt[i].zOpt ){
+ if( aOpt[i].eVal>=0 ){
+ printf(" %s=%d", &aOpt[i].zOpt[1], aParam[aOpt[i].eVal]);
+ }else if( aOpt[i].eVal==-1 ){
+ printf(" %s=\"%s\"", &aOpt[i].zOpt[1], zSystem);
+ }
+ }
+ }
+ printf("\n");
+
+ defn.nMinKey = defn.nMaxKey = aParam[ST_KEYSIZE];
+ defn.nMinVal = defn.nMaxVal = aParam[ST_VALSIZE];
+ pData = testDatasourceNew(&defn);
+
+ if( aParam[ST_WRITE]==0 ){
+ bReadonly = 1;
+ }
+
+ if( bLsm ){
+ rc = tdb_lsm_open(zSystem, "testdb.lsm", !bReadonly, &pDb);
+ }else{
+ pDb = testOpen(zSystem, !bReadonly, &rc);
+ }
+ if( rc!=0 ) return rc;
+ if( bReadonly ){
+ nContent = testCountDatabase(pDb);
+ }
+
+#if 0
+ pLog = fopen("/tmp/speed.log", "w");
+ tdb_lsm_write_hook(pDb, do_speed_write_hook2, (void *)pLog);
+#endif
+
+ for(i=0; i<aParam[ST_REPEAT] && rc==0; i++){
+ int msWrite, msFetch;
+ int iFetch;
+ int nWrite = aParam[ST_WRITE];
+
+ if( bReadonly ){
+ msWrite = 0;
+ }else{
+ testTimeInit();
+
+ if( aParam[ST_TRANS] ) testBegin(pDb, 2, &rc);
+ testWriteDatasourceRange(pDb, pData, i*nWrite, nWrite, &rc);
+ if( aParam[ST_TRANS] ) testCommit(pDb, 0, &rc);
+
+ msWrite = testTimeGet();
+ nContent += nWrite;
+ }
+
+ if( aParam[ST_PAUSE] ){
+ if( aParam[ST_PAUSE]/1000 ) sleep(aParam[ST_PAUSE]/1000);
+ if( aParam[ST_PAUSE]%1000 ) usleep(1000 * (aParam[ST_PAUSE]%1000));
+ }
+
+ if( aParam[ST_FETCH] ){
+ testTimeInit();
+ if( aParam[ST_TRANS] ) testBegin(pDb, 1, &rc);
+ for(iFetch=0; iFetch<aParam[ST_FETCH]; iFetch++){
+ int iKey = testPrngValue(i*nWrite+iFetch) % nContent;
+#ifndef NDEBUG
+ testDatasourceFetch(pDb, pData, iKey, &rc);
+#else
+ void *pKey; int nKey; /* Database key to query for */
+ void *pVal; int nVal; /* Result of query */
+
+ testDatasourceEntry(pData, iKey, &pKey, &nKey, 0, 0);
+ rc = tdb_fetch(pDb, pKey, nKey, &pVal, &nVal);
+ if( rc==0 && nVal<0 ) rc = 1;
+ if( rc ) break;
+#endif
+ }
+ if( aParam[ST_TRANS] ) testCommit(pDb, 0, &rc);
+ msFetch = testTimeGet();
+ }else{
+ msFetch = 0;
+ }
+
+ if( i==(aParam[ST_REPEAT]-1) ){
+ testTimeInit();
+ testClose(&pDb);
+ msWrite += testTimeGet();
+ }
+
+ printf("%d %d %d\n", i, msWrite, msFetch);
+ fflush(stdout);
+ }
+
+ testClose(&pDb);
+ testDatasourceFree(pData);
+
+ if( pLog ){
+ flushPrev(pLog);
+ fclose(pLog);
+ }
+ return rc;
+}
+
+int do_speed_tests(int nArg, char **azArg){
+
+ struct DbSystem {
+ const char *zLibrary;
+ const char *zColor;
+ } aSys[] = {
+ { "sqlite3", "black" },
+ { "leveldb", "blue" },
+ { "lsm", "red" },
+ { "lsm_mt2", "orange" },
+ { "lsm_mt3", "purple" },
+ { "kyotocabinet", "green" },
+ {0, 0}
+ };
+
+ int i;
+ int j;
+ int rc;
+ int nSleep = 0; /* ms of rest allowed between INSERT tests */
+ int nRow = 0; /* Number of rows to insert into database */
+ int nStep; /* Measure INSERT time after this many rows */
+ int nSelStep; /* Measure SELECT time after this many rows */
+ int nSelTest; /* Number of SELECTs to run for timing */
+ int doReadTest = 1;
+ int doWriteTest = 1;
+
+ int *aTime; /* INSERT timing data */
+ int *aWrite; /* Writes per nStep inserts */
+ int *aSelTime; /* SELECT timing data */
+ int isFirst = 1;
+ int bSleep = 0;
+
+ /* File to write gnuplot script to. */
+ const char *zOut = "lsmtest_speed.gnuplot";
+
+ u32 sys_mask = 0;
+
+ testMallocUninstall(tdb_lsm_env());
+
+ for(i=0; i<nArg; i++){
+ struct Opt {
+ const char *zOpt;
+ int isSwitch;
+ } aOpt[] = {
+ { "sqlite3" , 0},
+ { "leveldb" , 0},
+ { "lsm" , 0},
+ { "lsm_mt2" , 0},
+ { "lsm_mt3" , 0},
+ { "kyotocabinet" , 0},
+ { "-rows" , 1},
+ { "-sleep" , 2},
+ { "-testmode" , 3},
+ { "-out" , 4},
+ { 0, 0}
+ };
+ int iSel;
+
+ rc = testArgSelect(aOpt, "argument", azArg[i], &iSel);
+ if( rc ) return rc;
+
+ if( aOpt[iSel].isSwitch ){
+ i++;
+
+ if( i>=nArg ){
+ testPrintError("option %s requires an argument\n", aOpt[iSel].zOpt);
+ return 1;
+ }
+ if( aOpt[iSel].isSwitch==1 ){
+ nRow = atoi(azArg[i]);
+ }
+ if( aOpt[iSel].isSwitch==2 ){
+ nSleep = atoi(azArg[i]);
+ }
+ if( aOpt[iSel].isSwitch==3 ){
+ struct Mode {
+ const char *zMode;
+ int doReadTest;
+ int doWriteTest;
+ } aMode[] = {{"ro", 1, 0} , {"rw", 1, 1}, {"wo", 0, 1}, {0, 0, 0}};
+ int iMode;
+ rc = testArgSelect(aMode, "option", azArg[i], &iMode);
+ if( rc ) return rc;
+ doReadTest = aMode[iMode].doReadTest;
+ doWriteTest = aMode[iMode].doWriteTest;
+ }
+ if( aOpt[iSel].isSwitch==4 ){
+ /* The "-out FILE" switch. This option is used to specify a file to
+ ** write the gnuplot script to. */
+ zOut = azArg[i];
+ }
+ }else{
+ /* A db name */
+ rc = testArgSelect(aOpt, "system", azArg[i], &iSel);
+ if( rc ) return rc;
+ sys_mask |= (1<<iSel);
+ }
+ }
+
+ if( sys_mask==0 ) sys_mask = (1 << 0) | (1 << 1) | (1 << 2) | (1 << 3);
+ nRow = MAX(nRow, 100000);
+ nStep = nRow/100;
+ nSelStep = nRow/10;
+ nSelTest = (nSelStep > 100000) ? 100000 : nSelStep;
+
+ aTime = malloc(sizeof(int) * ArraySize(aSys) * nRow/nStep);
+ aWrite = malloc(sizeof(int) * nRow/nStep);
+ aSelTime = malloc(sizeof(int) * ArraySize(aSys) * nRow/nSelStep);
+
+ /* This loop collects the INSERT speed data. */
+ if( doWriteTest ){
+ printf("Writing output to file \"%s\".\n", zOut);
+
+ for(j=0; aSys[j].zLibrary; j++){
+ FILE *pLog = 0;
+ TestDb *pDb; /* Database being tested */
+ lsm_db *pLsm;
+ int iDot = 0;
+
+ if( ((1<<j)&sys_mask)==0 ) continue;
+ if( bSleep && nSleep ) sqlite3_sleep(nSleep);
+ bSleep = 1;
+
+ testCaseBegin(&rc, 0, "speed.insert.%s", aSys[j].zLibrary);
+
+ rc = tdb_open(aSys[j].zLibrary, 0, 1, &pDb);
+ if( rc ) return rc;
+
+ pLsm = configure_lsm_db(pDb);
+#if 0
+ pLog = fopen("/tmp/speed.log", "w");
+ tdb_lsm_write_hook(pDb, do_speed_write_hook2, (void *)pLog);
+#endif
+
+ testTimeInit();
+ for(i=0; i<nRow; i+=nStep){
+ int iStep;
+ int nWrite1 = 0, nWrite2 = 0;
+ testCaseProgress(i, nRow, testCaseNDot(), &iDot);
+ if( pLsm ) lsm_info(pLsm, LSM_INFO_NWRITE, &nWrite1);
+ for(iStep=0; iStep<nStep; iStep++){
+ u32 aKey[4]; /* 16-byte key */
+ u32 aVal[25]; /* 100 byte value */
+ testPrngArray(i+iStep, aKey, ArraySize(aKey));
+ testPrngArray(i+iStep, aVal, ArraySize(aVal));
+ rc = tdb_write(pDb, aKey, sizeof(aKey), aVal, sizeof(aVal));
+ }
+ aTime[(j*nRow+i)/nStep] = testTimeGet();
+ if( pLsm ) lsm_info(pLsm, LSM_INFO_NWRITE, &nWrite2);
+ aWrite[i/nStep] = nWrite2 - nWrite1;
+ }
+
+ tdb_close(pDb);
+ if( pLog ) fclose(pLog);
+ testCaseFinish(rc);
+ }
+ }
+
+ /* This loop collects the SELECT speed data. */
+ if( doReadTest ){
+ for(j=0; aSys[j].zLibrary; j++){
+ int iDot = 0;
+ TestDb *pDb; /* Database being tested */
+
+ if( ((1<<j)&sys_mask)==0 ) continue;
+ if( bSleep && nSleep ) sqlite3_sleep(nSleep);
+ bSleep = 1;
+
+ testCaseBegin(&rc, 0, "speed.select.%s", aSys[j].zLibrary);
+
+ if( doWriteTest ){
+ rc = tdb_open(aSys[j].zLibrary, 0, 1, &pDb);
+ if( rc ) return rc;
+ configure_lsm_db(pDb);
+
+ for(i=0; i<nRow; i+=nSelStep){
+ int iStep;
+ int iSel;
+ testCaseProgress(i, nRow, testCaseNDot(), &iDot);
+ for(iStep=0; iStep<nSelStep; iStep++){
+ u32 aKey[4]; /* 16-byte key */
+ u32 aVal[25]; /* 100 byte value */
+ testPrngArray(i+iStep, aKey, ArraySize(aKey));
+ testPrngArray(i+iStep, aVal, ArraySize(aVal));
+ rc = tdb_write(pDb, aKey, sizeof(aKey), aVal, sizeof(aVal));
+ }
+
+ testTimeInit();
+ for(iSel=0; iSel<nSelTest; iSel++){
+ void *pDummy;
+ int nDummy;
+ u32 iKey;
+ u32 aKey[4]; /* 16-byte key */
+
+ iKey = testPrngValue(iSel) % (i+nSelStep);
+ testPrngArray(iKey, aKey, ArraySize(aKey));
+ rc = tdb_fetch(pDb, aKey, sizeof(aKey), &pDummy, &nDummy);
+ }
+ aSelTime[(j*nRow+i)/nSelStep] = testTimeGet();
+ tdb_fetch(pDb, 0, 0, 0, 0);
+ }
+ }else{
+ int t;
+ int iSel;
+
+ rc = tdb_open(aSys[j].zLibrary, 0, 0, &pDb);
+ configure_lsm_db(pDb);
+
+ testTimeInit();
+ for(iSel=0; rc==LSM_OK && iSel<nSelTest; iSel++){
+ void *pDummy;
+ int nDummy;
+ u32 iKey;
+ u32 aKey[4]; /* 16-byte key */
+#ifndef NDEBUG
+ u32 aVal[25]; /* 100 byte value */
+#endif
+
+ testCaseProgress(iSel, nSelTest, testCaseNDot(), &iDot);
+
+ iKey = testPrngValue(iSel) % nRow;
+ testPrngArray(iKey, aKey, ArraySize(aKey));
+ rc = tdb_fetch(pDb, aKey, sizeof(aKey), &pDummy, &nDummy);
+
+#ifndef NDEBUG
+ testPrngArray(iKey, aVal, ArraySize(aVal));
+ assert( nDummy==100 && memcmp(aVal, pDummy, 100)==0 );
+#endif
+ }
+ if( rc!=LSM_OK ) return rc;
+
+ t = testTimeGet();
+ tdb_fetch(pDb, 0, 0, 0, 0);
+
+ printf("%s: %d selects/second\n",
+ aSys[j].zLibrary, (int)((double)nSelTest*1000.0/t)
+ );
+ }
+
+ tdb_close(pDb);
+ testCaseFinish(rc);
+ }
+ }
+
+
+ if( doWriteTest ){
+ FILE *pOut = fopen(zOut, "w");
+ if( !pOut ){
+ printf("fopen(\"%s\", \"w\"): %s\n", zOut, strerror(errno));
+ return 1;
+ }
+
+ fprintf(pOut, "set xlabel \"Rows Inserted\"\n");
+ fprintf(pOut, "set ylabel \"Inserts per second\"\n");
+ if( doReadTest ){
+ fprintf(pOut, "set y2label \"Selects per second\"\n");
+ }else if( sys_mask==(1<<2) ){
+ fprintf(pOut, "set y2label \"Page writes per insert\"\n");
+ }
+ fprintf(pOut, "set yrange [0:*]\n");
+ fprintf(pOut, "set y2range [0:*]\n");
+ fprintf(pOut, "set xrange [%d:*]\n", MAX(nStep, nRow/20) );
+ fprintf(pOut, "set ytics nomirror\n");
+ fprintf(pOut, "set y2tics nomirror\n");
+ fprintf(pOut, "set key box lw 0.01\n");
+ fprintf(pOut, "plot ");
+
+ for(j=0; aSys[j].zLibrary; j++){
+ if( (1<<j)&sys_mask ){
+ const char *zLib = aSys[j].zLibrary;
+ fprintf(pOut, "%s\"-\" ti \"%s INSERT\" with lines lc rgb \"%s\" ",
+ (isFirst?"":", "), zLib, aSys[j].zColor
+ );
+ if( doReadTest ){
+ fprintf(pOut, ", \"-\" ti \"%s SELECT\" "
+ "axis x1y2 with points lw 3 lc rgb \"%s\""
+ , zLib, aSys[j].zColor
+ );
+ }
+ isFirst = 0;
+ }
+ }
+
+ assert( strcmp(aSys[2].zLibrary, "lsm")==0 );
+ if( sys_mask==(1<<2) && !doReadTest ){
+ fprintf(pOut, ", \"-\" ti \"lsm pages written\" "
+ "axis x1y2 with boxes lw 1 lc rgb \"grey\""
+ );
+ }
+
+ fprintf(pOut, "\n");
+
+ for(j=0; aSys[j].zLibrary; j++){
+ if( ((1<<j)&sys_mask)==0 ) continue;
+ fprintf(pOut, "# Rows Inserts per second\n");
+ for(i=0; i<nRow; i+=nStep){
+ int iTime = aTime[(j*nRow+i)/nStep];
+ int ips = (int)((i+nStep)*1000.0 / (double)iTime);
+ fprintf(pOut, "%d %d\n", i+nStep, ips);
+ }
+ fprintf(pOut, "end\n");
+
+ if( doReadTest ){
+ fprintf(pOut, "# Rows Selects per second\n");
+ for(i=0; i<nRow; i+=nSelStep){
+ int sps = (int)(nSelTest*1000.0/(double)aSelTime[(j*nRow+i)/nSelStep]);
+ fprintf(pOut, "%d %d\n", i+nSelStep, sps);
+ }
+ fprintf(pOut, "end\n");
+ }else if( sys_mask==(1<<2) ){
+ for(i=0; i<(nRow/nStep); i++){
+ fprintf(pOut, "%d %f\n", i*nStep, (double)aWrite[i] / (double)nStep);
+ }
+ fprintf(pOut, "end\n");
+ }
+ }
+
+ fprintf(pOut, "pause -1\n");
+ fclose(pOut);
+ }
+
+ free(aTime);
+ free(aSelTime);
+ free(aWrite);
+ testMallocInstall(tdb_lsm_env());
+ return 0;
+}
+
+/*
+** Usage: lsmtest random ?N?
+**
+** This command prints a sequence of zero or more numbers from the PRNG
+** system to stdout. If the "N" argument is missing, values the first 10
+** values (i=0, i=1, ... i=9) are printed. Otherwise, the first N.
+**
+** This was added to verify that the PRNG values do not change between
+** runs of the lsmtest program.
+*/
+int do_random_tests(int nArg, char **azArg){
+ int i;
+ int nRand;
+ if( nArg==0 ){
+ nRand = 10;
+ }else if( nArg==1 ){
+ nRand = atoi(azArg[0]);
+ }else{
+ testPrintError("Usage: random ?N?\n");
+ return -1;
+ }
+ for(i=0; i<nRand; i++){
+ printf("0x%x\n", testPrngValue(i));
+ }
+ return 0;
+}
+
+static int testFormatSize(char *aBuf, int nBuf, i64 nByte){
+ int res;
+ if( nByte<(1<<10) ){
+ res = snprintf(aBuf, nBuf, "%d byte", (int)nByte);
+ }else if( nByte<(1<<20) ){
+ res = snprintf(aBuf, nBuf, "%dK", (int)(nByte/(1<<10)));
+ }else{
+ res = snprintf(aBuf, nBuf, "%dM", (int)(nByte/(1<<20)));
+ }
+ return res;
+}
+
+static i64 testReadSize(char *z){
+ int n = strlen(z);
+ char c = z[n-1];
+ i64 nMul = 1;
+
+ switch( c ){
+ case 'g': case 'G':
+ nMul = (1<<30);
+ break;
+
+ case 'm': case 'M':
+ nMul = (1<<20);
+ break;
+
+ case 'k': case 'K':
+ nMul = (1<<10);
+ break;
+
+ default:
+ nMul = 1;
+ }
+
+ return nMul * (i64)atoi(z);
+}
+
+/*
+** Usage: lsmtest writespeed FILESIZE BLOCKSIZE SYNCSIZE
+*/
+static int do_writer_test(int nArg, char **azArg){
+ int nBlock;
+ int nSize;
+ int i;
+ int fd;
+ int ms;
+ char aFilesize[32];
+ char aBlockSize[32];
+
+ char *aPage;
+ int *aOrder;
+ int nSync;
+
+ i64 filesize;
+ i64 blocksize;
+ i64 syncsize;
+ int nPage = 4096;
+
+ /* How long to sleep before running a trial (in ms). */
+#if 0
+ const int nSleep = 10000;
+#endif
+ const int nSleep = 0;
+
+ if( nArg!=3 ){
+ testPrintUsage("FILESIZE BLOCKSIZE SYNCSIZE");
+ return -1;
+ }
+
+ filesize = testReadSize(azArg[0]);
+ blocksize = testReadSize(azArg[1]);
+ syncsize = testReadSize(azArg[2]);
+
+ nBlock = (int)(filesize / blocksize);
+ nSize = (int)blocksize;
+ nSync = (int)(syncsize / blocksize);
+
+ aPage = (char *)malloc(4096);
+ aOrder = (int *)malloc(nBlock * sizeof(int));
+ for(i=0; i<nBlock; i++) aOrder[i] = i;
+ for(i=0; i<(nBlock*25); i++){
+ int tmp;
+ u32 a = testPrngValue(i);
+ u32 b = testPrngValue(a);
+ a = a % nBlock;
+ b = b % nBlock;
+ tmp = aOrder[a];
+ aOrder[a] = aOrder[b];
+ aOrder[b] = tmp;
+ }
+
+ testFormatSize(aFilesize, sizeof(aFilesize), (i64)nBlock * (i64)nSize);
+ testFormatSize(aBlockSize, sizeof(aFilesize), nSize);
+
+ printf("Testing writing a %s file using %s blocks. ", aFilesize, aBlockSize);
+ if( nSync==1 ){
+ printf("Sync after each block.\n");
+ }else{
+ printf("Sync after each %d blocks.\n", nSync);
+ }
+
+ printf("Preparing file... ");
+ fflush(stdout);
+ unlink("writer.out");
+ fd = open("writer.out", O_RDWR|O_CREAT|_O_BINARY, 0664);
+ if( fd<0 ){
+ testPrintError("open(): %d - %s\n", errno, strerror(errno));
+ return -1;
+ }
+ testTimeInit();
+ for(i=0; i<nBlock; i++){
+ int iPg;
+ memset(aPage, i&0xFF, nPage);
+ for(iPg=0; iPg<(nSize/nPage); iPg++){
+ write(fd, aPage, nPage);
+ }
+ }
+ fsync(fd);
+ printf("ok (%d ms)\n", testTimeGet());
+
+ for(i=0; i<5; i++){
+ int j;
+
+ sqlite3_sleep(nSleep);
+ printf("Now writing sequentially... ");
+ fflush(stdout);
+
+ lseek(fd, 0, SEEK_SET);
+ testTimeInit();
+ for(j=0; j<nBlock; j++){
+ int iPg;
+ if( ((j+1)%nSync)==0 ) fdatasync(fd);
+ memset(aPage, j&0xFF, nPage);
+ for(iPg=0; iPg<(nSize/nPage); iPg++){
+ write(fd, aPage, nPage);
+ }
+ }
+ fdatasync(fd);
+ ms = testTimeGet();
+ printf("%d ms\n", ms);
+ sqlite3_sleep(nSleep);
+ printf("Now in an arbitrary order... ");
+
+ fflush(stdout);
+ testTimeInit();
+ for(j=0; j<nBlock; j++){
+ int iPg;
+ if( ((j+1)%nSync)==0 ) fdatasync(fd);
+ lseek(fd, aOrder[j]*nSize, SEEK_SET);
+ memset(aPage, j&0xFF, nPage);
+ for(iPg=0; iPg<(nSize/nPage); iPg++){
+ write(fd, aPage, nPage);
+ }
+ }
+ fdatasync(fd);
+ ms = testTimeGet();
+ printf("%d ms\n", ms);
+ }
+
+ close(fd);
+ free(aPage);
+ free(aOrder);
+
+ return 0;
+}
+
+static void do_insert_work_hook(lsm_db *db, void *p){
+ char *z = 0;
+ lsm_info(db, LSM_INFO_DB_STRUCTURE, &z);
+ if( z ){
+ printf("%s\n", z);
+ fflush(stdout);
+ lsm_free(lsm_get_env(db), z);
+ }
+
+ unused_parameter(p);
+}
+
+typedef struct InsertWriteHook InsertWriteHook;
+struct InsertWriteHook {
+ FILE *pOut;
+ int bLog;
+ i64 iOff;
+ int nData;
+};
+
+static void flushHook(InsertWriteHook *pHook){
+ if( pHook->nData ){
+ fprintf(pHook->pOut, "write %s %d %d\n",
+ (pHook->bLog ? "log" : "db"), (int)pHook->iOff, pHook->nData
+ );
+ pHook->nData = 0;
+ fflush(pHook->pOut);
+ }
+}
+
+static void do_insert_write_hook(
+ void *pCtx,
+ int bLog,
+ i64 iOff,
+ int nData,
+ int nUs
+){
+ InsertWriteHook *pHook = (InsertWriteHook *)pCtx;
+ if( bLog ) return;
+
+ if( nData==0 ){
+ flushHook(pHook);
+ fprintf(pHook->pOut, "sync %s\n", (bLog ? "log" : "db"));
+ }else if( pHook->nData
+ && bLog==pHook->bLog
+ && iOff==(pHook->iOff+pHook->nData)
+ ){
+ pHook->nData += nData;
+ }else{
+ flushHook(pHook);
+ pHook->bLog = bLog;
+ pHook->iOff = iOff;
+ pHook->nData = nData;
+ }
+}
+
+static int do_replay(int nArg, char **azArg){
+ char aBuf[4096];
+ FILE *pInput;
+ FILE *pClose = 0;
+ const char *zDb;
+
+ lsm_env *pEnv;
+ lsm_file *pOut;
+ int rc;
+
+ if( nArg!=2 ){
+ testPrintError("Usage: replay WRITELOG FILE\n");
+ return 1;
+ }
+
+ if( strcmp(azArg[0], "-")==0 ){
+ pInput = stdin;
+ }else{
+ pClose = pInput = fopen(azArg[0], "r");
+ }
+ zDb = azArg[1];
+ pEnv = tdb_lsm_env();
+ rc = pEnv->xOpen(pEnv, zDb, 0, &pOut);
+ if( rc!=LSM_OK ) return rc;
+
+ while( feof(pInput)==0 ){
+ char zLine[80];
+ fgets(zLine, sizeof(zLine)-1, pInput);
+ zLine[sizeof(zLine)-1] = '\0';
+
+ if( 0==memcmp("sync db", zLine, 7) ){
+ rc = pEnv->xSync(pOut);
+ if( rc!=0 ) break;
+ }else{
+ int iOff;
+ int nData;
+ int nMatch;
+ nMatch = sscanf(zLine, "write db %d %d", &iOff, &nData);
+ if( nMatch==2 ){
+ int i;
+ for(i=0; i<nData; i+=sizeof(aBuf)){
+ memset(aBuf, i&0xFF, sizeof(aBuf));
+ rc = pEnv->xWrite(pOut, iOff+i, aBuf, sizeof(aBuf));
+ if( rc!=0 ) break;
+ }
+ }
+ }
+ }
+ if( pClose ) fclose(pClose);
+ pEnv->xClose(pOut);
+
+ return rc;
+}
+
+static int do_insert(int nArg, char **azArg){
+ const char *zDb = "lsm";
+ TestDb *pDb = 0;
+ int i;
+ int rc;
+ const int nRow = 1 * 1000 * 1000;
+
+ DatasourceDefn defn = { TEST_DATASOURCE_RANDOM, 8, 15, 80, 150 };
+ Datasource *pData = 0;
+
+ if( nArg>1 ){
+ testPrintError("Usage: insert ?DATABASE?\n");
+ return 1;
+ }
+ if( nArg==1 ){ zDb = azArg[0]; }
+
+ testMallocUninstall(tdb_lsm_env());
+ for(i=0; zDb[i] && zDb[i]!='='; i++);
+ if( zDb[i] ){
+ rc = tdb_lsm_open(zDb, "testdb.lsm", 1, &pDb);
+ }else{
+ rc = tdb_open(zDb, 0, 1, &pDb);
+ }
+
+ if( rc!=0 ){
+ testPrintError("Error opening db \"%s\": %d\n", zDb, rc);
+ }else{
+ InsertWriteHook hook;
+ memset(&hook, 0, sizeof(hook));
+ hook.pOut = fopen("writelog.txt", "w");
+
+ pData = testDatasourceNew(&defn);
+ tdb_lsm_config_work_hook(pDb, do_insert_work_hook, 0);
+ tdb_lsm_write_hook(pDb, do_insert_write_hook, (void *)&hook);
+
+ if( rc==0 ){
+ for(i=0; i<nRow; i++){
+ void *pKey; int nKey; /* Database key to insert */
+ void *pVal; int nVal; /* Database value to insert */
+ testDatasourceEntry(pData, i, &pKey, &nKey, &pVal, &nVal);
+ tdb_write(pDb, pKey, nKey, pVal, nVal);
+ }
+ }
+
+ testDatasourceFree(pData);
+ tdb_close(pDb);
+ flushHook(&hook);
+ fclose(hook.pOut);
+ }
+ testMallocInstall(tdb_lsm_env());
+
+ return rc;
+}
+
+static int st_do_show(int a, char **b) { return do_show(a, b); }
+static int st_do_work(int a, char **b) { return do_work(a, b); }
+static int st_do_io(int a, char **b) { return do_io(a, b); }
+
+#ifdef __linux__
+#include <sys/time.h>
+#include <sys/resource.h>
+
+static void lsmtest_rusage_report(void){
+ struct rusage r;
+ memset(&r, 0, sizeof(r));
+
+ getrusage(RUSAGE_SELF, &r);
+ printf("# getrusage: { ru_maxrss %d ru_oublock %d ru_inblock %d }\n",
+ (int)r.ru_maxrss, (int)r.ru_oublock, (int)r.ru_inblock
+ );
+}
+#else
+static void lsmtest_rusage_report(void){
+ /* no-op */
+}
+#endif
+
+int main(int argc, char **argv){
+ struct TestFunc {
+ const char *zName;
+ int bRusageReport;
+ int (*xFunc)(int, char **);
+ } aTest[] = {
+ {"random", 1, do_random_tests},
+ {"writespeed", 1, do_writer_test},
+ {"io", 1, st_do_io},
+
+ {"insert", 1, do_insert},
+ {"replay", 1, do_replay},
+
+ {"speed", 1, do_speed_tests},
+ {"speed2", 1, do_speed_test2},
+ {"show", 0, st_do_show},
+ {"work", 1, st_do_work},
+ {"test", 1, do_test},
+
+ {0, 0}
+ };
+ int rc; /* Return Code */
+ int iFunc; /* Index into aTest[] */
+
+ int nLeakAlloc = 0; /* Allocations leaked by lsm */
+ int nLeakByte = 0; /* Bytes leaked by lsm */
+
+#ifdef LSM_DEBUG_MEM
+ FILE *pReport = 0; /* lsm malloc() report file */
+ const char *zReport = "malloc.txt generated";
+#else
+ const char *zReport = "malloc.txt NOT generated";
+#endif
+
+ testMallocInstall(tdb_lsm_env());
+
+ if( argc<2 ){
+ testPrintError("Usage: %s sub-command ?args...?\n", argv[0]);
+ return -1;
+ }
+
+ /* Initialize error reporting */
+ testErrorInit(argc, argv);
+
+ /* Initialize PRNG system */
+ testPrngInit();
+
+ rc = testArgSelect(aTest, "sub-command", argv[1], &iFunc);
+ if( rc==0 ){
+ rc = aTest[iFunc].xFunc(argc-2, &argv[2]);
+ }
+
+#ifdef LSM_DEBUG_MEM
+ pReport = fopen("malloc.txt", "w");
+ testMallocCheck(tdb_lsm_env(), &nLeakAlloc, &nLeakByte, pReport);
+ fclose(pReport);
+#else
+ testMallocCheck(tdb_lsm_env(), &nLeakAlloc, &nLeakByte, 0);
+#endif
+
+ if( nLeakAlloc ){
+ testPrintError("Leaked %d bytes in %d allocations (%s)\n",
+ nLeakByte, nLeakAlloc, zReport
+ );
+ if( rc==0 ) rc = -1;
+ }
+ testMallocUninstall(tdb_lsm_env());
+
+ if( aTest[iFunc].bRusageReport ){
+ lsmtest_rusage_report();
+ }
+ return rc;
+}
diff --git a/ext/lsm1/lsm-test/lsmtest_mem.c b/ext/lsm1/lsm-test/lsmtest_mem.c
new file mode 100644
index 0000000..4c35e84
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_mem.c
@@ -0,0 +1,409 @@
+
+#include <stdio.h>
+#include <assert.h>
+#include <string.h>
+
+#define ArraySize(x) ((int)(sizeof(x) / sizeof((x)[0])))
+
+#define MIN(x,y) ((x)<(y) ? (x) : (y))
+
+typedef unsigned int u32;
+typedef unsigned char u8;
+typedef long long int i64;
+typedef unsigned long long int u64;
+
+#if defined(__GLIBC__) && defined(LSM_DEBUG_MEM)
+ extern int backtrace(void**,int);
+ extern void backtrace_symbols_fd(void*const*,int,int);
+# define TM_BACKTRACE 12
+#else
+# define backtrace(A,B) 1
+# define backtrace_symbols_fd(A,B,C)
+#endif
+
+
+typedef struct TmBlockHdr TmBlockHdr;
+typedef struct TmAgg TmAgg;
+typedef struct TmGlobal TmGlobal;
+
+struct TmGlobal {
+ /* Linked list of all currently outstanding allocations. And a table of
+ ** all allocations, past and present, indexed by backtrace() info. */
+ TmBlockHdr *pFirst;
+#ifdef TM_BACKTRACE
+ TmAgg *aHash[10000];
+#endif
+
+ /* Underlying malloc/realloc/free functions */
+ void *(*xMalloc)(int); /* underlying malloc(3) function */
+ void *(*xRealloc)(void *, int); /* underlying realloc(3) function */
+ void (*xFree)(void *); /* underlying free(3) function */
+
+ /* Mutex to protect pFirst and aHash */
+ void (*xEnterMutex)(TmGlobal*); /* Call this to enter the mutex */
+ void (*xLeaveMutex)(TmGlobal*); /* Call this to leave mutex */
+ void (*xDelMutex)(TmGlobal*); /* Call this to delete mutex */
+ void *pMutex; /* Mutex handle */
+
+ void *(*xSaveMalloc)(void *, size_t);
+ void *(*xSaveRealloc)(void *, void *, size_t);
+ void (*xSaveFree)(void *, void *);
+
+ /* OOM injection scheduling. If nCountdown is greater than zero when a
+ ** malloc attempt is made, it is decremented. If this means nCountdown
+ ** transitions from 1 to 0, then the allocation fails. If bPersist is true
+ ** when this happens, nCountdown is then incremented back to 1 (so that the
+ ** next attempt fails too).
+ */
+ int nCountdown;
+ int bPersist;
+ int bEnable;
+ void (*xHook)(void *);
+ void *pHookCtx;
+};
+
+struct TmBlockHdr {
+ TmBlockHdr *pNext;
+ TmBlockHdr *pPrev;
+ int nByte;
+#ifdef TM_BACKTRACE
+ TmAgg *pAgg;
+#endif
+ u32 iForeGuard;
+};
+
+#ifdef TM_BACKTRACE
+struct TmAgg {
+ int nAlloc; /* Number of allocations at this path */
+ int nByte; /* Total number of bytes allocated */
+ int nOutAlloc; /* Number of outstanding allocations */
+ int nOutByte; /* Number of outstanding bytes */
+ void *aFrame[TM_BACKTRACE]; /* backtrace() output */
+ TmAgg *pNext; /* Next object in hash-table collision */
+};
+#endif
+
+#define FOREGUARD 0x80F5E153
+#define REARGUARD 0xE4676B53
+static const u32 rearguard = REARGUARD;
+
+#define ROUND8(x) (((x)+7)&~7)
+
+#define BLOCK_HDR_SIZE (ROUND8( sizeof(TmBlockHdr) ))
+
+static void lsmtest_oom_error(void){
+ static int nErr = 0;
+ nErr++;
+}
+
+static void tmEnterMutex(TmGlobal *pTm){
+ pTm->xEnterMutex(pTm);
+}
+static void tmLeaveMutex(TmGlobal *pTm){
+ pTm->xLeaveMutex(pTm);
+}
+
+static void *tmMalloc(TmGlobal *pTm, int nByte){
+ TmBlockHdr *pNew; /* New allocation header block */
+ u8 *pUser; /* Return value */
+ int nReq; /* Total number of bytes requested */
+
+ assert( sizeof(rearguard)==4 );
+ nReq = BLOCK_HDR_SIZE + nByte + 4;
+ pNew = (TmBlockHdr *)pTm->xMalloc(nReq);
+ memset(pNew, 0, sizeof(TmBlockHdr));
+
+ tmEnterMutex(pTm);
+ assert( pTm->nCountdown>=0 );
+ assert( pTm->bPersist==0 || pTm->bPersist==1 );
+
+ if( pTm->bEnable && pTm->nCountdown==1 ){
+ /* Simulate an OOM error. */
+ lsmtest_oom_error();
+ pTm->xFree(pNew);
+ pTm->nCountdown = pTm->bPersist;
+ if( pTm->xHook ) pTm->xHook(pTm->pHookCtx);
+ pUser = 0;
+ }else{
+ if( pTm->bEnable && pTm->nCountdown ) pTm->nCountdown--;
+
+ pNew->iForeGuard = FOREGUARD;
+ pNew->nByte = nByte;
+ pNew->pNext = pTm->pFirst;
+
+ if( pTm->pFirst ){
+ pTm->pFirst->pPrev = pNew;
+ }
+ pTm->pFirst = pNew;
+
+ pUser = &((u8 *)pNew)[BLOCK_HDR_SIZE];
+ memset(pUser, 0x56, nByte);
+ memcpy(&pUser[nByte], &rearguard, 4);
+
+#ifdef TM_BACKTRACE
+ {
+ TmAgg *pAgg;
+ int i;
+ u32 iHash = 0;
+ void *aFrame[TM_BACKTRACE];
+ memset(aFrame, 0, sizeof(aFrame));
+ backtrace(aFrame, TM_BACKTRACE);
+
+ for(i=0; i<ArraySize(aFrame); i++){
+ iHash += (u64)(aFrame[i]) + (iHash<<3);
+ }
+ iHash = iHash % ArraySize(pTm->aHash);
+
+ for(pAgg=pTm->aHash[iHash]; pAgg; pAgg=pAgg->pNext){
+ if( memcmp(pAgg->aFrame, aFrame, sizeof(aFrame))==0 ) break;
+ }
+ if( !pAgg ){
+ pAgg = (TmAgg *)pTm->xMalloc(sizeof(TmAgg));
+ memset(pAgg, 0, sizeof(TmAgg));
+ memcpy(pAgg->aFrame, aFrame, sizeof(aFrame));
+ pAgg->pNext = pTm->aHash[iHash];
+ pTm->aHash[iHash] = pAgg;
+ }
+ pAgg->nAlloc++;
+ pAgg->nByte += nByte;
+ pAgg->nOutAlloc++;
+ pAgg->nOutByte += nByte;
+ pNew->pAgg = pAgg;
+ }
+#endif
+ }
+
+ tmLeaveMutex(pTm);
+ return pUser;
+}
+
+static void tmFree(TmGlobal *pTm, void *p){
+ if( p ){
+ TmBlockHdr *pHdr;
+ u8 *pUser = (u8 *)p;
+
+ tmEnterMutex(pTm);
+ pHdr = (TmBlockHdr *)(pUser - BLOCK_HDR_SIZE);
+ assert( pHdr->iForeGuard==FOREGUARD );
+ assert( 0==memcmp(&pUser[pHdr->nByte], &rearguard, 4) );
+
+ if( pHdr->pPrev ){
+ assert( pHdr->pPrev->pNext==pHdr );
+ pHdr->pPrev->pNext = pHdr->pNext;
+ }else{
+ assert( pHdr==pTm->pFirst );
+ pTm->pFirst = pHdr->pNext;
+ }
+ if( pHdr->pNext ){
+ assert( pHdr->pNext->pPrev==pHdr );
+ pHdr->pNext->pPrev = pHdr->pPrev;
+ }
+
+#ifdef TM_BACKTRACE
+ pHdr->pAgg->nOutAlloc--;
+ pHdr->pAgg->nOutByte -= pHdr->nByte;
+#endif
+
+ tmLeaveMutex(pTm);
+ memset(pUser, 0x58, pHdr->nByte);
+ memset(pHdr, 0x57, sizeof(TmBlockHdr));
+ pTm->xFree(pHdr);
+ }
+}
+
+static void *tmRealloc(TmGlobal *pTm, void *p, int nByte){
+ void *pNew;
+
+ pNew = tmMalloc(pTm, nByte);
+ if( pNew && p ){
+ TmBlockHdr *pHdr;
+ u8 *pUser = (u8 *)p;
+ pHdr = (TmBlockHdr *)(pUser - BLOCK_HDR_SIZE);
+ memcpy(pNew, p, MIN(nByte, pHdr->nByte));
+ tmFree(pTm, p);
+ }
+ return pNew;
+}
+
+static void tmMallocOom(
+ TmGlobal *pTm,
+ int nCountdown,
+ int bPersist,
+ void (*xHook)(void *),
+ void *pHookCtx
+){
+ assert( nCountdown>=0 );
+ assert( bPersist==0 || bPersist==1 );
+ pTm->nCountdown = nCountdown;
+ pTm->bPersist = bPersist;
+ pTm->xHook = xHook;
+ pTm->pHookCtx = pHookCtx;
+ pTm->bEnable = 1;
+}
+
+static void tmMallocOomEnable(
+ TmGlobal *pTm,
+ int bEnable
+){
+ pTm->bEnable = bEnable;
+}
+
+static void tmMallocCheck(
+ TmGlobal *pTm,
+ int *pnLeakAlloc,
+ int *pnLeakByte,
+ FILE *pFile
+){
+ TmBlockHdr *pHdr;
+ int nLeak = 0;
+ int nByte = 0;
+
+ if( pTm==0 ) return;
+
+ for(pHdr=pTm->pFirst; pHdr; pHdr=pHdr->pNext){
+ nLeak++;
+ nByte += pHdr->nByte;
+ }
+ if( pnLeakAlloc ) *pnLeakAlloc = nLeak;
+ if( pnLeakByte ) *pnLeakByte = nByte;
+
+#ifdef TM_BACKTRACE
+ if( pFile ){
+ int i;
+ fprintf(pFile, "LEAKS\n");
+ for(i=0; i<ArraySize(pTm->aHash); i++){
+ TmAgg *pAgg;
+ for(pAgg=pTm->aHash[i]; pAgg; pAgg=pAgg->pNext){
+ if( pAgg->nOutAlloc ){
+ int j;
+ fprintf(pFile, "%d %d ", pAgg->nOutByte, pAgg->nOutAlloc);
+ for(j=0; j<TM_BACKTRACE; j++){
+ fprintf(pFile, "%p ", pAgg->aFrame[j]);
+ }
+ fprintf(pFile, "\n");
+ }
+ }
+ }
+ fprintf(pFile, "\nALLOCATIONS\n");
+ for(i=0; i<ArraySize(pTm->aHash); i++){
+ TmAgg *pAgg;
+ for(pAgg=pTm->aHash[i]; pAgg; pAgg=pAgg->pNext){
+ int j;
+ fprintf(pFile, "%d %d ", pAgg->nByte, pAgg->nAlloc);
+ for(j=0; j<TM_BACKTRACE; j++) fprintf(pFile, "%p ", pAgg->aFrame[j]);
+ fprintf(pFile, "\n");
+ }
+ }
+ }
+#else
+ (void)pFile;
+#endif
+}
+
+
+#include "lsm.h"
+#include "stdlib.h"
+
+typedef struct LsmMutex LsmMutex;
+struct LsmMutex {
+ lsm_env *pEnv;
+ lsm_mutex *pMutex;
+};
+
+static void tmLsmMutexEnter(TmGlobal *pTm){
+ LsmMutex *p = (LsmMutex *)pTm->pMutex;
+ p->pEnv->xMutexEnter(p->pMutex);
+}
+static void tmLsmMutexLeave(TmGlobal *pTm){
+ LsmMutex *p = (LsmMutex *)(pTm->pMutex);
+ p->pEnv->xMutexLeave(p->pMutex);
+}
+static void tmLsmMutexDel(TmGlobal *pTm){
+ LsmMutex *p = (LsmMutex *)pTm->pMutex;
+ pTm->xFree(p);
+}
+static void *tmLsmMalloc(int n){ return malloc(n); }
+static void tmLsmFree(void *ptr){ free(ptr); }
+static void *tmLsmRealloc(void *ptr, int n){ return realloc(ptr, n); }
+
+static void *tmLsmEnvMalloc(lsm_env *p, size_t n){
+ return tmMalloc((TmGlobal *)(p->pMemCtx), n);
+}
+static void tmLsmEnvFree(lsm_env *p, void *ptr){
+ tmFree((TmGlobal *)(p->pMemCtx), ptr);
+}
+static void *tmLsmEnvRealloc(lsm_env *p, void *ptr, size_t n){
+ return tmRealloc((TmGlobal *)(p->pMemCtx), ptr, n);
+}
+
+void testMallocInstall(lsm_env *pEnv){
+ TmGlobal *pGlobal;
+ LsmMutex *pMutex;
+ assert( pEnv->pMemCtx==0 );
+
+ /* Allocate and populate a TmGlobal structure. */
+ pGlobal = (TmGlobal *)tmLsmMalloc(sizeof(TmGlobal));
+ memset(pGlobal, 0, sizeof(TmGlobal));
+ pGlobal->xMalloc = tmLsmMalloc;
+ pGlobal->xRealloc = tmLsmRealloc;
+ pGlobal->xFree = tmLsmFree;
+ pMutex = (LsmMutex *)pGlobal->xMalloc(sizeof(LsmMutex));
+ pMutex->pEnv = pEnv;
+ pEnv->xMutexStatic(pEnv, LSM_MUTEX_HEAP, &pMutex->pMutex);
+ pGlobal->xEnterMutex = tmLsmMutexEnter;
+ pGlobal->xLeaveMutex = tmLsmMutexLeave;
+ pGlobal->xDelMutex = tmLsmMutexDel;
+ pGlobal->pMutex = (void *)pMutex;
+
+ pGlobal->xSaveMalloc = pEnv->xMalloc;
+ pGlobal->xSaveRealloc = pEnv->xRealloc;
+ pGlobal->xSaveFree = pEnv->xFree;
+
+ /* Set up pEnv to the use the new TmGlobal */
+ pEnv->pMemCtx = (void *)pGlobal;
+ pEnv->xMalloc = tmLsmEnvMalloc;
+ pEnv->xRealloc = tmLsmEnvRealloc;
+ pEnv->xFree = tmLsmEnvFree;
+}
+
+void testMallocUninstall(lsm_env *pEnv){
+ TmGlobal *p = (TmGlobal *)pEnv->pMemCtx;
+ pEnv->pMemCtx = 0;
+ if( p ){
+ pEnv->xMalloc = p->xSaveMalloc;
+ pEnv->xRealloc = p->xSaveRealloc;
+ pEnv->xFree = p->xSaveFree;
+ p->xDelMutex(p);
+ tmLsmFree(p);
+ }
+}
+
+void testMallocCheck(
+ lsm_env *pEnv,
+ int *pnLeakAlloc,
+ int *pnLeakByte,
+ FILE *pFile
+){
+ if( pEnv->pMemCtx==0 ){
+ *pnLeakAlloc = 0;
+ *pnLeakByte = 0;
+ }else{
+ tmMallocCheck((TmGlobal *)(pEnv->pMemCtx), pnLeakAlloc, pnLeakByte, pFile);
+ }
+}
+
+void testMallocOom(
+ lsm_env *pEnv,
+ int nCountdown,
+ int bPersist,
+ void (*xHook)(void *),
+ void *pHookCtx
+){
+ TmGlobal *pTm = (TmGlobal *)(pEnv->pMemCtx);
+ tmMallocOom(pTm, nCountdown, bPersist, xHook, pHookCtx);
+}
+
+void testMallocOomEnable(lsm_env *pEnv, int bEnable){
+ TmGlobal *pTm = (TmGlobal *)(pEnv->pMemCtx);
+ tmMallocOomEnable(pTm, bEnable);
+}
diff --git a/ext/lsm1/lsm-test/lsmtest_tdb.c b/ext/lsm1/lsm-test/lsmtest_tdb.c
new file mode 100644
index 0000000..8f63f64
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_tdb.c
@@ -0,0 +1,846 @@
+
+/*
+** This program attempts to test the correctness of some facets of the
+** LSM database library. Specifically, that the contents of the database
+** are maintained correctly during a series of inserts and deletes.
+*/
+
+
+#include "lsmtest_tdb.h"
+#include "lsm.h"
+
+#include "lsmtest.h"
+
+#include <stdlib.h>
+#include <string.h>
+#include <assert.h>
+#ifndef _WIN32
+# include <unistd.h>
+#endif
+#include <stdio.h>
+
+
+typedef struct SqlDb SqlDb;
+
+static int error_transaction_function(TestDb *p, int iLevel){
+ unused_parameter(p);
+ unused_parameter(iLevel);
+ return -1;
+}
+
+
+/*************************************************************************
+** Begin wrapper for LevelDB.
+*/
+#ifdef HAVE_LEVELDB
+
+#include <leveldb/c.h>
+
+typedef struct LevelDb LevelDb;
+struct LevelDb {
+ TestDb base;
+ leveldb_t *db;
+ leveldb_options_t *pOpt;
+ leveldb_writeoptions_t *pWriteOpt;
+ leveldb_readoptions_t *pReadOpt;
+
+ char *pVal;
+};
+
+static int test_leveldb_close(TestDb *pTestDb){
+ LevelDb *pDb = (LevelDb *)pTestDb;
+
+ leveldb_close(pDb->db);
+ leveldb_writeoptions_destroy(pDb->pWriteOpt);
+ leveldb_readoptions_destroy(pDb->pReadOpt);
+ leveldb_options_destroy(pDb->pOpt);
+ free(pDb->pVal);
+ free(pDb);
+
+ return 0;
+}
+
+static int test_leveldb_write(
+ TestDb *pTestDb,
+ void *pKey,
+ int nKey,
+ void *pVal,
+ int nVal
+){
+ LevelDb *pDb = (LevelDb *)pTestDb;
+ char *zErr = 0;
+ leveldb_put(pDb->db, pDb->pWriteOpt, pKey, nKey, pVal, nVal, &zErr);
+ return (zErr!=0);
+}
+
+static int test_leveldb_delete(TestDb *pTestDb, void *pKey, int nKey){
+ LevelDb *pDb = (LevelDb *)pTestDb;
+ char *zErr = 0;
+ leveldb_delete(pDb->db, pDb->pWriteOpt, pKey, nKey, &zErr);
+ return (zErr!=0);
+}
+
+static int test_leveldb_fetch(
+ TestDb *pTestDb,
+ void *pKey,
+ int nKey,
+ void **ppVal,
+ int *pnVal
+){
+ LevelDb *pDb = (LevelDb *)pTestDb;
+ char *zErr = 0;
+ size_t nVal = 0;
+
+ if( pKey==0 ) return 0;
+ free(pDb->pVal);
+ pDb->pVal = leveldb_get(pDb->db, pDb->pReadOpt, pKey, nKey, &nVal, &zErr);
+ *ppVal = (void *)(pDb->pVal);
+ if( pDb->pVal==0 ){
+ *pnVal = -1;
+ }else{
+ *pnVal = (int)nVal;
+ }
+
+ return (zErr!=0);
+}
+
+static int test_leveldb_scan(
+ TestDb *pTestDb,
+ void *pCtx,
+ int bReverse,
+ void *pKey1, int nKey1, /* Start of search */
+ void *pKey2, int nKey2, /* End of search */
+ void (*xCallback)(void *, void *, int , void *, int)
+){
+ LevelDb *pDb = (LevelDb *)pTestDb;
+ leveldb_iterator_t *iter;
+
+ iter = leveldb_create_iterator(pDb->db, pDb->pReadOpt);
+
+ if( bReverse==0 ){
+ if( pKey1 ){
+ leveldb_iter_seek(iter, pKey1, nKey1);
+ }else{
+ leveldb_iter_seek_to_first(iter);
+ }
+ }else{
+ if( pKey2 ){
+ leveldb_iter_seek(iter, pKey2, nKey2);
+
+ if( leveldb_iter_valid(iter)==0 ){
+ leveldb_iter_seek_to_last(iter);
+ }else{
+ const char *k; size_t n;
+ int res;
+ k = leveldb_iter_key(iter, &n);
+ res = memcmp(k, pKey2, MIN(n, nKey2));
+ if( res==0 ) res = n - nKey2;
+ assert( res>=0 );
+ if( res>0 ){
+ leveldb_iter_prev(iter);
+ }
+ }
+ }else{
+ leveldb_iter_seek_to_last(iter);
+ }
+ }
+
+
+ while( leveldb_iter_valid(iter) ){
+ const char *k; size_t n;
+ const char *v; size_t n2;
+ int res;
+
+ k = leveldb_iter_key(iter, &n);
+ if( bReverse==0 && pKey2 ){
+ res = memcmp(k, pKey2, MIN(n, nKey2));
+ if( res==0 ) res = n - nKey2;
+ if( res>0 ) break;
+ }
+ if( bReverse!=0 && pKey1 ){
+ res = memcmp(k, pKey1, MIN(n, nKey1));
+ if( res==0 ) res = n - nKey1;
+ if( res<0 ) break;
+ }
+
+ v = leveldb_iter_value(iter, &n2);
+
+ xCallback(pCtx, (void *)k, n, (void *)v, n2);
+
+ if( bReverse==0 ){
+ leveldb_iter_next(iter);
+ }else{
+ leveldb_iter_prev(iter);
+ }
+ }
+
+ leveldb_iter_destroy(iter);
+ return 0;
+}
+
+static int test_leveldb_open(
+ const char *zSpec,
+ const char *zFilename,
+ int bClear,
+ TestDb **ppDb
+){
+ static const DatabaseMethods LeveldbMethods = {
+ test_leveldb_close,
+ test_leveldb_write,
+ test_leveldb_delete,
+ 0,
+ test_leveldb_fetch,
+ test_leveldb_scan,
+ error_transaction_function,
+ error_transaction_function,
+ error_transaction_function
+ };
+
+ LevelDb *pLevelDb;
+ char *zErr = 0;
+
+ if( bClear ){
+ char *zCmd = sqlite3_mprintf("rm -rf %s\n", zFilename);
+ system(zCmd);
+ sqlite3_free(zCmd);
+ }
+
+ pLevelDb = (LevelDb *)malloc(sizeof(LevelDb));
+ memset(pLevelDb, 0, sizeof(LevelDb));
+
+ pLevelDb->pOpt = leveldb_options_create();
+ leveldb_options_set_create_if_missing(pLevelDb->pOpt, 1);
+ pLevelDb->pWriteOpt = leveldb_writeoptions_create();
+ pLevelDb->pReadOpt = leveldb_readoptions_create();
+
+ pLevelDb->db = leveldb_open(pLevelDb->pOpt, zFilename, &zErr);
+
+ if( zErr ){
+ test_leveldb_close((TestDb *)pLevelDb);
+ *ppDb = 0;
+ return 1;
+ }
+
+ *ppDb = (TestDb *)pLevelDb;
+ pLevelDb->base.pMethods = &LeveldbMethods;
+ return 0;
+}
+#endif /* HAVE_LEVELDB */
+/*
+** End wrapper for LevelDB.
+*************************************************************************/
+
+#ifdef HAVE_KYOTOCABINET
+static int kc_close(TestDb *pTestDb){
+ return test_kc_close(pTestDb);
+}
+
+static int kc_write(
+ TestDb *pTestDb,
+ void *pKey,
+ int nKey,
+ void *pVal,
+ int nVal
+){
+ return test_kc_write(pTestDb, pKey, nKey, pVal, nVal);
+}
+
+static int kc_delete(TestDb *pTestDb, void *pKey, int nKey){
+ return test_kc_delete(pTestDb, pKey, nKey);
+}
+
+static int kc_delete_range(
+ TestDb *pTestDb,
+ void *pKey1, int nKey1,
+ void *pKey2, int nKey2
+){
+ return test_kc_delete_range(pTestDb, pKey1, nKey1, pKey2, nKey2);
+}
+
+static int kc_fetch(
+ TestDb *pTestDb,
+ void *pKey,
+ int nKey,
+ void **ppVal,
+ int *pnVal
+){
+ if( pKey==0 ) return LSM_OK;
+ return test_kc_fetch(pTestDb, pKey, nKey, ppVal, pnVal);
+}
+
+static int kc_scan(
+ TestDb *pTestDb,
+ void *pCtx,
+ int bReverse,
+ void *pFirst, int nFirst,
+ void *pLast, int nLast,
+ void (*xCallback)(void *, void *, int , void *, int)
+){
+ return test_kc_scan(
+ pTestDb, pCtx, bReverse, pFirst, nFirst, pLast, nLast, xCallback
+ );
+}
+
+static int kc_open(
+ const char *zSpec,
+ const char *zFilename,
+ int bClear,
+ TestDb **ppDb
+){
+ static const DatabaseMethods KcdbMethods = {
+ kc_close,
+ kc_write,
+ kc_delete,
+ kc_delete_range,
+ kc_fetch,
+ kc_scan,
+ error_transaction_function,
+ error_transaction_function,
+ error_transaction_function
+ };
+
+ int rc;
+ TestDb *pTestDb = 0;
+
+ rc = test_kc_open(zFilename, bClear, &pTestDb);
+ if( rc!=0 ){
+ *ppDb = 0;
+ return rc;
+ }
+ pTestDb->pMethods = &KcdbMethods;
+ *ppDb = pTestDb;
+ return 0;
+}
+#endif /* HAVE_KYOTOCABINET */
+/*
+** End wrapper for Kyoto cabinet.
+*************************************************************************/
+
+#ifdef HAVE_MDB
+static int mdb_close(TestDb *pTestDb){
+ return test_mdb_close(pTestDb);
+}
+
+static int mdb_write(
+ TestDb *pTestDb,
+ void *pKey,
+ int nKey,
+ void *pVal,
+ int nVal
+){
+ return test_mdb_write(pTestDb, pKey, nKey, pVal, nVal);
+}
+
+static int mdb_delete(TestDb *pTestDb, void *pKey, int nKey){
+ return test_mdb_delete(pTestDb, pKey, nKey);
+}
+
+static int mdb_fetch(
+ TestDb *pTestDb,
+ void *pKey,
+ int nKey,
+ void **ppVal,
+ int *pnVal
+){
+ if( pKey==0 ) return LSM_OK;
+ return test_mdb_fetch(pTestDb, pKey, nKey, ppVal, pnVal);
+}
+
+static int mdb_scan(
+ TestDb *pTestDb,
+ void *pCtx,
+ int bReverse,
+ void *pFirst, int nFirst,
+ void *pLast, int nLast,
+ void (*xCallback)(void *, void *, int , void *, int)
+){
+ return test_mdb_scan(
+ pTestDb, pCtx, bReverse, pFirst, nFirst, pLast, nLast, xCallback
+ );
+}
+
+static int mdb_open(
+ const char *zSpec,
+ const char *zFilename,
+ int bClear,
+ TestDb **ppDb
+){
+ static const DatabaseMethods KcdbMethods = {
+ mdb_close,
+ mdb_write,
+ mdb_delete,
+ 0,
+ mdb_fetch,
+ mdb_scan,
+ error_transaction_function,
+ error_transaction_function,
+ error_transaction_function
+ };
+
+ int rc;
+ TestDb *pTestDb = 0;
+
+ rc = test_mdb_open(zSpec, zFilename, bClear, &pTestDb);
+ if( rc!=0 ){
+ *ppDb = 0;
+ return rc;
+ }
+ pTestDb->pMethods = &KcdbMethods;
+ *ppDb = pTestDb;
+ return 0;
+}
+#endif /* HAVE_MDB */
+
+/*************************************************************************
+** Begin wrapper for SQLite.
+*/
+
+/*
+** nOpenTrans:
+** The number of open nested transactions, in the same sense as used
+** by the tdb_begin/commit/rollback and SQLite 4 KV interfaces. If this
+** value is 0, there are no transactions open at all. If it is 1, then
+** there is a read transaction. If it is 2 or greater, then there are
+** (nOpenTrans-1) nested write transactions open.
+*/
+struct SqlDb {
+ TestDb base;
+ sqlite3 *db;
+ sqlite3_stmt *pInsert;
+ sqlite3_stmt *pDelete;
+ sqlite3_stmt *pDeleteRange;
+ sqlite3_stmt *pFetch;
+ sqlite3_stmt *apScan[8];
+
+ int nOpenTrans;
+
+ /* Used by sql_fetch() to allocate space for results */
+ int nAlloc;
+ u8 *aAlloc;
+};
+
+static int sql_close(TestDb *pTestDb){
+ SqlDb *pDb = (SqlDb *)pTestDb;
+ sqlite3_finalize(pDb->pInsert);
+ sqlite3_finalize(pDb->pDelete);
+ sqlite3_finalize(pDb->pDeleteRange);
+ sqlite3_finalize(pDb->pFetch);
+ sqlite3_finalize(pDb->apScan[0]);
+ sqlite3_finalize(pDb->apScan[1]);
+ sqlite3_finalize(pDb->apScan[2]);
+ sqlite3_finalize(pDb->apScan[3]);
+ sqlite3_finalize(pDb->apScan[4]);
+ sqlite3_finalize(pDb->apScan[5]);
+ sqlite3_finalize(pDb->apScan[6]);
+ sqlite3_finalize(pDb->apScan[7]);
+ sqlite3_close(pDb->db);
+ free((char *)pDb->aAlloc);
+ free((char *)pDb);
+ return SQLITE_OK;
+}
+
+static int sql_write(
+ TestDb *pTestDb,
+ void *pKey,
+ int nKey,
+ void *pVal,
+ int nVal
+){
+ SqlDb *pDb = (SqlDb *)pTestDb;
+ sqlite3_bind_blob(pDb->pInsert, 1, pKey, nKey, SQLITE_STATIC);
+ sqlite3_bind_blob(pDb->pInsert, 2, pVal, nVal, SQLITE_STATIC);
+ sqlite3_step(pDb->pInsert);
+ return sqlite3_reset(pDb->pInsert);
+}
+
+static int sql_delete(TestDb *pTestDb, void *pKey, int nKey){
+ SqlDb *pDb = (SqlDb *)pTestDb;
+ sqlite3_bind_blob(pDb->pDelete, 1, pKey, nKey, SQLITE_STATIC);
+ sqlite3_step(pDb->pDelete);
+ return sqlite3_reset(pDb->pDelete);
+}
+
+static int sql_delete_range(
+ TestDb *pTestDb,
+ void *pKey1, int nKey1,
+ void *pKey2, int nKey2
+){
+ SqlDb *pDb = (SqlDb *)pTestDb;
+ sqlite3_bind_blob(pDb->pDeleteRange, 1, pKey1, nKey1, SQLITE_STATIC);
+ sqlite3_bind_blob(pDb->pDeleteRange, 2, pKey2, nKey2, SQLITE_STATIC);
+ sqlite3_step(pDb->pDeleteRange);
+ return sqlite3_reset(pDb->pDeleteRange);
+}
+
+static int sql_fetch(
+ TestDb *pTestDb,
+ void *pKey,
+ int nKey,
+ void **ppVal,
+ int *pnVal
+){
+ SqlDb *pDb = (SqlDb *)pTestDb;
+ int rc;
+
+ sqlite3_reset(pDb->pFetch);
+ if( pKey==0 ){
+ assert( ppVal==0 );
+ assert( pnVal==0 );
+ return LSM_OK;
+ }
+
+ sqlite3_bind_blob(pDb->pFetch, 1, pKey, nKey, SQLITE_STATIC);
+ rc = sqlite3_step(pDb->pFetch);
+ if( rc==SQLITE_ROW ){
+ int nVal = sqlite3_column_bytes(pDb->pFetch, 0);
+ u8 *aVal = (void *)sqlite3_column_blob(pDb->pFetch, 0);
+
+ if( nVal>pDb->nAlloc ){
+ free(pDb->aAlloc);
+ pDb->aAlloc = (u8 *)malloc(nVal*2);
+ pDb->nAlloc = nVal*2;
+ }
+ memcpy(pDb->aAlloc, aVal, nVal);
+ *pnVal = nVal;
+ *ppVal = (void *)pDb->aAlloc;
+ }else{
+ *pnVal = -1;
+ *ppVal = 0;
+ }
+
+ rc = sqlite3_reset(pDb->pFetch);
+ return rc;
+}
+
+static int sql_scan(
+ TestDb *pTestDb,
+ void *pCtx,
+ int bReverse,
+ void *pFirst, int nFirst,
+ void *pLast, int nLast,
+ void (*xCallback)(void *, void *, int , void *, int)
+){
+ SqlDb *pDb = (SqlDb *)pTestDb;
+ sqlite3_stmt *pScan;
+
+ assert( bReverse==1 || bReverse==0 );
+ pScan = pDb->apScan[(pFirst==0) + (pLast==0)*2 + bReverse*4];
+
+ if( pFirst ) sqlite3_bind_blob(pScan, 1, pFirst, nFirst, SQLITE_STATIC);
+ if( pLast ) sqlite3_bind_blob(pScan, 2, pLast, nLast, SQLITE_STATIC);
+
+ while( SQLITE_ROW==sqlite3_step(pScan) ){
+ void *pKey; int nKey;
+ void *pVal; int nVal;
+
+ nKey = sqlite3_column_bytes(pScan, 0);
+ pKey = (void *)sqlite3_column_blob(pScan, 0);
+ nVal = sqlite3_column_bytes(pScan, 1);
+ pVal = (void *)sqlite3_column_blob(pScan, 1);
+
+ xCallback(pCtx, pKey, nKey, pVal, nVal);
+ }
+ return sqlite3_reset(pScan);
+}
+
+static int sql_begin(TestDb *pTestDb, int iLevel){
+ int i;
+ SqlDb *pDb = (SqlDb *)pTestDb;
+
+ /* iLevel==0 is a no-op */
+ if( iLevel==0 ) return 0;
+
+ /* If there are no transactions at all open, open a read transaction. */
+ if( pDb->nOpenTrans==0 ){
+ int rc = sqlite3_exec(pDb->db,
+ "BEGIN; SELECT * FROM sqlite_schema LIMIT 1;" , 0, 0, 0
+ );
+ if( rc!=0 ) return rc;
+ pDb->nOpenTrans = 1;
+ }
+
+ /* Open any required write transactions */
+ for(i=pDb->nOpenTrans; i<iLevel; i++){
+ char *zSql = sqlite3_mprintf("SAVEPOINT x%d", i);
+ int rc = sqlite3_exec(pDb->db, zSql, 0, 0, 0);
+ sqlite3_free(zSql);
+ if( rc!=SQLITE_OK ) return rc;
+ }
+
+ pDb->nOpenTrans = iLevel;
+ return 0;
+}
+
+static int sql_commit(TestDb *pTestDb, int iLevel){
+ SqlDb *pDb = (SqlDb *)pTestDb;
+ assert( iLevel>=0 );
+
+ /* Close the read transaction if requested. */
+ if( pDb->nOpenTrans>=1 && iLevel==0 ){
+ int rc = sqlite3_exec(pDb->db, "COMMIT", 0, 0, 0);
+ if( rc!=0 ) return rc;
+ pDb->nOpenTrans = 0;
+ }
+
+ /* Close write transactions as required */
+ if( pDb->nOpenTrans>iLevel ){
+ char *zSql = sqlite3_mprintf("RELEASE x%d", iLevel);
+ int rc = sqlite3_exec(pDb->db, zSql, 0, 0, 0);
+ sqlite3_free(zSql);
+ if( rc!=0 ) return rc;
+ }
+
+ pDb->nOpenTrans = iLevel;
+ return 0;
+}
+
+static int sql_rollback(TestDb *pTestDb, int iLevel){
+ SqlDb *pDb = (SqlDb *)pTestDb;
+ assert( iLevel>=0 );
+
+ if( pDb->nOpenTrans>=1 && iLevel==0 ){
+ /* Close the read transaction if requested. */
+ int rc = sqlite3_exec(pDb->db, "ROLLBACK", 0, 0, 0);
+ if( rc!=0 ) return rc;
+ }else if( pDb->nOpenTrans>1 && iLevel==1 ){
+ /* Or, rollback and close the top-level write transaction */
+ int rc = sqlite3_exec(pDb->db, "ROLLBACK TO x1; RELEASE x1;", 0, 0, 0);
+ if( rc!=0 ) return rc;
+ }else{
+ /* Or, just roll back some nested transactions */
+ char *zSql = sqlite3_mprintf("ROLLBACK TO x%d", iLevel-1);
+ int rc = sqlite3_exec(pDb->db, zSql, 0, 0, 0);
+ sqlite3_free(zSql);
+ if( rc!=0 ) return rc;
+ }
+
+ pDb->nOpenTrans = iLevel;
+ return 0;
+}
+
+static int sql_open(
+ const char *zSpec,
+ const char *zFilename,
+ int bClear,
+ TestDb **ppDb
+){
+ static const DatabaseMethods SqlMethods = {
+ sql_close,
+ sql_write,
+ sql_delete,
+ sql_delete_range,
+ sql_fetch,
+ sql_scan,
+ sql_begin,
+ sql_commit,
+ sql_rollback
+ };
+ const char *zCreate = "CREATE TABLE IF NOT EXISTS t1(k PRIMARY KEY, v)";
+ const char *zInsert = "REPLACE INTO t1 VALUES(?, ?)";
+ const char *zDelete = "DELETE FROM t1 WHERE k = ?";
+ const char *zRange = "DELETE FROM t1 WHERE k>? AND k<?";
+ const char *zFetch = "SELECT v FROM t1 WHERE k = ?";
+
+ const char *zScan0 = "SELECT * FROM t1 WHERE k BETWEEN ?1 AND ?2 ORDER BY k";
+ const char *zScan1 = "SELECT * FROM t1 WHERE k <= ?2 ORDER BY k";
+ const char *zScan2 = "SELECT * FROM t1 WHERE k >= ?1 ORDER BY k";
+ const char *zScan3 = "SELECT * FROM t1 ORDER BY k";
+
+ const char *zScan4 =
+ "SELECT * FROM t1 WHERE k BETWEEN ?1 AND ?2 ORDER BY k DESC";
+ const char *zScan5 = "SELECT * FROM t1 WHERE k <= ?2 ORDER BY k DESC";
+ const char *zScan6 = "SELECT * FROM t1 WHERE k >= ?1 ORDER BY k DESC";
+ const char *zScan7 = "SELECT * FROM t1 ORDER BY k DESC";
+
+ int rc;
+ SqlDb *pDb;
+ char *zPragma;
+
+ if( bClear && zFilename && zFilename[0] ){
+ unlink(zFilename);
+ }
+
+ pDb = (SqlDb *)malloc(sizeof(SqlDb));
+ memset(pDb, 0, sizeof(SqlDb));
+ pDb->base.pMethods = &SqlMethods;
+
+ if( 0!=(rc = sqlite3_open(zFilename, &pDb->db))
+ || 0!=(rc = sqlite3_exec(pDb->db, zCreate, 0, 0, 0))
+ || 0!=(rc = sqlite3_prepare_v2(pDb->db, zInsert, -1, &pDb->pInsert, 0))
+ || 0!=(rc = sqlite3_prepare_v2(pDb->db, zDelete, -1, &pDb->pDelete, 0))
+ || 0!=(rc = sqlite3_prepare_v2(pDb->db, zRange, -1, &pDb->pDeleteRange, 0))
+ || 0!=(rc = sqlite3_prepare_v2(pDb->db, zFetch, -1, &pDb->pFetch, 0))
+ || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan0, -1, &pDb->apScan[0], 0))
+ || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan1, -1, &pDb->apScan[1], 0))
+ || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan2, -1, &pDb->apScan[2], 0))
+ || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan3, -1, &pDb->apScan[3], 0))
+ || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan4, -1, &pDb->apScan[4], 0))
+ || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan5, -1, &pDb->apScan[5], 0))
+ || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan6, -1, &pDb->apScan[6], 0))
+ || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan7, -1, &pDb->apScan[7], 0))
+ ){
+ *ppDb = 0;
+ sql_close((TestDb *)pDb);
+ return rc;
+ }
+
+ zPragma = sqlite3_mprintf("PRAGMA page_size=%d", TESTDB_DEFAULT_PAGE_SIZE);
+ sqlite3_exec(pDb->db, zPragma, 0, 0, 0);
+ sqlite3_free(zPragma);
+ zPragma = sqlite3_mprintf("PRAGMA cache_size=%d", TESTDB_DEFAULT_CACHE_SIZE);
+ sqlite3_exec(pDb->db, zPragma, 0, 0, 0);
+ sqlite3_free(zPragma);
+
+ /* sqlite3_exec(pDb->db, "PRAGMA locking_mode=EXCLUSIVE", 0, 0, 0); */
+ sqlite3_exec(pDb->db, "PRAGMA synchronous=OFF", 0, 0, 0);
+ sqlite3_exec(pDb->db, "PRAGMA journal_mode=WAL", 0, 0, 0);
+ sqlite3_exec(pDb->db, "PRAGMA wal_autocheckpoint=4096", 0, 0, 0);
+ if( zSpec ){
+ rc = sqlite3_exec(pDb->db, zSpec, 0, 0, 0);
+ if( rc!=SQLITE_OK ){
+ sql_close((TestDb *)pDb);
+ return rc;
+ }
+ }
+
+ *ppDb = (TestDb *)pDb;
+ return 0;
+}
+/*
+** End wrapper for SQLite.
+*************************************************************************/
+
+/*************************************************************************
+** Begin exported functions.
+*/
+static struct Lib {
+ const char *zName;
+ const char *zDefaultDb;
+ int (*xOpen)(const char *, const char *zFilename, int bClear, TestDb **ppDb);
+} aLib[] = {
+ { "sqlite3", "testdb.sqlite", sql_open },
+ { "lsm_small", "testdb.lsm_small", test_lsm_small_open },
+ { "lsm_lomem", "testdb.lsm_lomem", test_lsm_lomem_open },
+ { "lsm_lomem2", "testdb.lsm_lomem2", test_lsm_lomem2_open },
+#ifdef HAVE_ZLIB
+ { "lsm_zip", "testdb.lsm_zip", test_lsm_zip_open },
+#endif
+ { "lsm", "testdb.lsm", test_lsm_open },
+#ifdef LSM_MUTEX_PTHREADS
+ { "lsm_mt2", "testdb.lsm_mt2", test_lsm_mt2 },
+ { "lsm_mt3", "testdb.lsm_mt3", test_lsm_mt3 },
+#endif
+#ifdef HAVE_LEVELDB
+ { "leveldb", "testdb.leveldb", test_leveldb_open },
+#endif
+#ifdef HAVE_KYOTOCABINET
+ { "kyotocabinet", "testdb.kc", kc_open },
+#endif
+#ifdef HAVE_MDB
+ { "mdb", "./testdb.mdb", mdb_open }
+#endif
+};
+
+const char *tdb_system_name(int i){
+ if( i<0 || i>=ArraySize(aLib) ) return 0;
+ return aLib[i].zName;
+}
+
+const char *tdb_default_db(const char *zSys){
+ int i;
+ for(i=0; i<ArraySize(aLib); i++){
+ if( strcmp(aLib[i].zName, zSys)==0 ) return aLib[i].zDefaultDb;
+ }
+ return 0;
+}
+
+int tdb_open(const char *zLib, const char *zDb, int bClear, TestDb **ppDb){
+ int i;
+ int rc = 1;
+ const char *zSpec = 0;
+
+ int nLib = 0;
+ while( zLib[nLib] && zLib[nLib]!=' ' ){
+ nLib++;
+ }
+ zSpec = &zLib[nLib];
+ while( *zSpec==' ' ) zSpec++;
+ if( *zSpec=='\0' ) zSpec = 0;
+
+ for(i=0; i<ArraySize(aLib); i++){
+ if( (int)strlen(aLib[i].zName)==nLib
+ && 0==memcmp(zLib, aLib[i].zName, nLib) ){
+ rc = aLib[i].xOpen(zSpec, (zDb ? zDb : aLib[i].zDefaultDb), bClear, ppDb);
+ if( rc==0 ){
+ (*ppDb)->zLibrary = aLib[i].zName;
+ }
+ break;
+ }
+ }
+
+ if( rc ){
+ /* Failed to find the requested database library. Return an error. */
+ *ppDb = 0;
+ }
+ return rc;
+}
+
+int tdb_close(TestDb *pDb){
+ if( pDb ){
+ return pDb->pMethods->xClose(pDb);
+ }
+ return 0;
+}
+
+int tdb_write(TestDb *pDb, void *pKey, int nKey, void *pVal, int nVal){
+ return pDb->pMethods->xWrite(pDb, pKey, nKey, pVal, nVal);
+}
+
+int tdb_delete(TestDb *pDb, void *pKey, int nKey){
+ return pDb->pMethods->xDelete(pDb, pKey, nKey);
+}
+
+int tdb_delete_range(
+ TestDb *pDb, void *pKey1, int nKey1, void *pKey2, int nKey2
+){
+ return pDb->pMethods->xDeleteRange(pDb, pKey1, nKey1, pKey2, nKey2);
+}
+
+int tdb_fetch(TestDb *pDb, void *pKey, int nKey, void **ppVal, int *pnVal){
+ return pDb->pMethods->xFetch(pDb, pKey, nKey, ppVal, pnVal);
+}
+
+int tdb_scan(
+ TestDb *pDb, /* Database handle */
+ void *pCtx, /* Context pointer to pass to xCallback */
+ int bReverse, /* True to scan in reverse order */
+ void *pKey1, int nKey1, /* Start of search */
+ void *pKey2, int nKey2, /* End of search */
+ void (*xCallback)(void *pCtx, void *pKey, int nKey, void *pVal, int nVal)
+){
+ return pDb->pMethods->xScan(
+ pDb, pCtx, bReverse, pKey1, nKey1, pKey2, nKey2, xCallback
+ );
+}
+
+int tdb_begin(TestDb *pDb, int iLevel){
+ return pDb->pMethods->xBegin(pDb, iLevel);
+}
+int tdb_commit(TestDb *pDb, int iLevel){
+ return pDb->pMethods->xCommit(pDb, iLevel);
+}
+int tdb_rollback(TestDb *pDb, int iLevel){
+ return pDb->pMethods->xRollback(pDb, iLevel);
+}
+
+int tdb_transaction_support(TestDb *pDb){
+ return (pDb->pMethods->xBegin != error_transaction_function);
+}
+
+const char *tdb_library_name(TestDb *pDb){
+ return pDb->zLibrary;
+}
+
+/*
+** End exported functions.
+*************************************************************************/
diff --git a/ext/lsm1/lsm-test/lsmtest_tdb.h b/ext/lsm1/lsm-test/lsmtest_tdb.h
new file mode 100644
index 0000000..c55b6e2
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_tdb.h
@@ -0,0 +1,174 @@
+
+/*
+** This file is the interface to a very simple database library used for
+** testing. The interface is similar to that of the LSM. The main virtue
+** of this library is that the same API may be used to access a key-value
+** store implemented by LSM, SQLite or another database system. Which
+** makes it easy to use for correctness and performance tests.
+*/
+
+#ifndef __WRAPPER_H_
+#define __WRAPPER_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "lsm.h"
+
+typedef struct TestDb TestDb;
+
+/*
+** Open a new database connection. The first argument is the name of the
+** database library to use. e.g. something like:
+**
+** "sqlite3"
+** "lsm"
+**
+** See function tdb_system_name() for a list of available database systems.
+**
+** The second argument is the name of the database to open (e.g. a filename).
+**
+** If the third parameter is non-zero, then any existing database by the
+** name of zDb is removed before opening a new one. If it is zero, then an
+** existing database may be opened.
+*/
+int tdb_open(const char *zLibrary, const char *zDb, int bClear, TestDb **ppDb);
+
+/*
+** Close a database handle.
+*/
+int tdb_close(TestDb *pDb);
+
+/*
+** Write a new key/value into the database.
+*/
+int tdb_write(TestDb *pDb, void *pKey, int nKey, void *pVal, int nVal);
+
+/*
+** Delete a key from the database.
+*/
+int tdb_delete(TestDb *pDb, void *pKey, int nKey);
+
+/*
+** Delete a range of keys from the database.
+*/
+int tdb_delete_range(TestDb *, void *pKey1, int nKey1, void *pKey2, int nKey2);
+
+/*
+** Query the database for key (pKey/nKey). If no entry is found, set *ppVal
+** to 0 and *pnVal to -1 before returning. Otherwise, set *ppVal and *pnVal
+** to a pointer to and size of the value associated with (pKey/nKey).
+*/
+int tdb_fetch(TestDb *pDb, void *pKey, int nKey, void **ppVal, int *pnVal);
+
+/*
+** Open and close nested transactions. Currently, these functions only
+** work for SQLite3 and LSM systems. Use the tdb_transaction_support()
+** function to determine if a given TestDb handle supports these methods.
+**
+** These functions and the iLevel parameter follow the same conventions as
+** the SQLite 4 transaction interface. Note that this is slightly different
+** from the way LSM does things. As follows:
+**
+** tdb_begin():
+** A successful call to tdb_begin() with (iLevel>1) guarantees that
+** there are at least (iLevel-1) write transactions open. If iLevel==1,
+** then it guarantees that at least a read-transaction is open. Calling
+** tdb_begin() with iLevel==0 is a no-op.
+**
+** tdb_commit():
+** A successful call to tdb_commit() with (iLevel>1) guarantees that
+** there are at most (iLevel-1) write transactions open. If iLevel==1,
+** then it guarantees that there are no write transactions open (although
+** a read-transaction may remain open). Calling tdb_commit() with
+** iLevel==0 ensures that all transactions, read or write, have been
+** closed and committed.
+**
+** tdb_rollback():
+** This call is similar to tdb_commit(), except that instead of committing
+** transactions, it reverts them. For example, calling tdb_rollback() with
+** iLevel==2 ensures that there is at most one write transaction open, and
+** restores the database to the state that it was in when that transaction
+** was opened.
+**
+** In other words, tdb_commit() just closes transactions - tdb_rollback()
+** closes transactions and then restores the database to the state it
+** was in before those transactions were even opened.
+*/
+int tdb_begin(TestDb *pDb, int iLevel);
+int tdb_commit(TestDb *pDb, int iLevel);
+int tdb_rollback(TestDb *pDb, int iLevel);
+
+/*
+** Return true if transactions are supported, or false otherwise.
+*/
+int tdb_transaction_support(TestDb *pDb);
+
+/*
+** Return the name of the database library (as passed to tdb_open()) used
+** by the handled passed as the first argument.
+*/
+const char *tdb_library_name(TestDb *pDb);
+
+/*
+** Scan a range of database keys. Invoke the callback function for each
+** key visited.
+*/
+int tdb_scan(
+ TestDb *pDb, /* Database handle */
+ void *pCtx, /* Context pointer to pass to xCallback */
+ int bReverse, /* True to scan in reverse order */
+ void *pKey1, int nKey1, /* Start of search */
+ void *pKey2, int nKey2, /* End of search */
+ void (*xCallback)(void *pCtx, void *pKey, int nKey, void *pVal, int nVal)
+);
+
+const char *tdb_system_name(int i);
+const char *tdb_default_db(const char *zSys);
+
+int tdb_lsm_open(const char *zCfg, const char *zDb, int bClear, TestDb **ppDb);
+
+/*
+** If the TestDb handle passed as an argument is a wrapper around an LSM
+** database, return the LSM handle. Otherwise, if the argument is some other
+** database system, return NULL.
+*/
+lsm_db *tdb_lsm(TestDb *pDb);
+
+/*
+** Return true if the db passed as an argument is a multi-threaded LSM
+** connection.
+*/
+int tdb_lsm_multithread(TestDb *pDb);
+
+/*
+** Return a pointer to the lsm_env object used by all lsm database
+** connections initialized as a copy of the object returned by
+** lsm_default_env(). It may be modified (e.g. to override functions)
+** if the caller can guarantee that it is not already in use.
+*/
+lsm_env *tdb_lsm_env(void);
+
+/*
+** The following functions only work with LSM database handles. It is
+** illegal to call them with any other type of database handle specified
+** as an argument.
+*/
+void tdb_lsm_enable_log(TestDb *pDb, int bEnable);
+void tdb_lsm_application_crash(TestDb *pDb);
+void tdb_lsm_prepare_system_crash(TestDb *pDb);
+void tdb_lsm_system_crash(TestDb *pDb);
+void tdb_lsm_prepare_sync_crash(TestDb *pDb, int iSync);
+
+
+void tdb_lsm_safety(TestDb *pDb, int eMode);
+void tdb_lsm_config_work_hook(TestDb *pDb, void (*)(lsm_db *, void *), void *);
+void tdb_lsm_write_hook(TestDb *, void(*)(void*,int,lsm_i64,int,int), void*);
+int tdb_lsm_config_str(TestDb *pDb, const char *zStr);
+
+#ifdef __cplusplus
+} /* End of the 'extern "C"' block */
+#endif
+
+#endif
diff --git a/ext/lsm1/lsm-test/lsmtest_tdb2.cc b/ext/lsm1/lsm-test/lsmtest_tdb2.cc
new file mode 100644
index 0000000..86ebb49
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_tdb2.cc
@@ -0,0 +1,369 @@
+
+
+#include "lsmtest.h"
+#include <stdlib.h>
+
+#ifdef HAVE_KYOTOCABINET
+#include "kcpolydb.h"
+extern "C" {
+ struct KcDb {
+ TestDb base;
+ kyotocabinet::TreeDB* db;
+ char *pVal;
+ };
+}
+
+int test_kc_open(const char *zFilename, int bClear, TestDb **ppDb){
+ KcDb *pKcDb;
+ int ok;
+ int rc = 0;
+
+ if( bClear ){
+ char *zCmd = sqlite3_mprintf("rm -rf %s\n", zFilename);
+ system(zCmd);
+ sqlite3_free(zCmd);
+ }
+
+ pKcDb = (KcDb *)malloc(sizeof(KcDb));
+ memset(pKcDb, 0, sizeof(KcDb));
+
+
+ pKcDb->db = new kyotocabinet::TreeDB();
+ pKcDb->db->tune_page(TESTDB_DEFAULT_PAGE_SIZE);
+ pKcDb->db->tune_page_cache(
+ TESTDB_DEFAULT_PAGE_SIZE * TESTDB_DEFAULT_CACHE_SIZE
+ );
+ ok = pKcDb->db->open(zFilename,
+ kyotocabinet::PolyDB::OWRITER | kyotocabinet::PolyDB::OCREATE
+ );
+ if( ok==0 ){
+ free(pKcDb);
+ pKcDb = 0;
+ rc = 1;
+ }
+
+ *ppDb = (TestDb *)pKcDb;
+ return rc;
+}
+
+int test_kc_close(TestDb *pDb){
+ KcDb *pKcDb = (KcDb *)pDb;
+ if( pKcDb->pVal ){
+ delete [] pKcDb->pVal;
+ }
+ pKcDb->db->close();
+ delete pKcDb->db;
+ free(pKcDb);
+ return 0;
+}
+
+int test_kc_write(TestDb *pDb, void *pKey, int nKey, void *pVal, int nVal){
+ KcDb *pKcDb = (KcDb *)pDb;
+ int ok;
+
+ ok = pKcDb->db->set((const char *)pKey, nKey, (const char *)pVal, nVal);
+ return (ok ? 0 : 1);
+}
+
+int test_kc_delete(TestDb *pDb, void *pKey, int nKey){
+ KcDb *pKcDb = (KcDb *)pDb;
+ int ok;
+
+ ok = pKcDb->db->remove((const char *)pKey, nKey);
+ return (ok ? 0 : 1);
+}
+
+int test_kc_delete_range(
+ TestDb *pDb,
+ void *pKey1, int nKey1,
+ void *pKey2, int nKey2
+){
+ int res;
+ KcDb *pKcDb = (KcDb *)pDb;
+ kyotocabinet::DB::Cursor* pCur = pKcDb->db->cursor();
+
+ if( pKey1 ){
+ res = pCur->jump((const char *)pKey1, nKey1);
+ }else{
+ res = pCur->jump();
+ }
+
+ while( 1 ){
+ const char *pKey; size_t nKey;
+ const char *pVal; size_t nVal;
+
+ pKey = pCur->get(&nKey, &pVal, &nVal);
+ if( pKey==0 ) break;
+
+#ifndef NDEBUG
+ if( pKey1 ){
+ res = memcmp(pKey, pKey1, MIN((size_t)nKey1, nKey));
+ assert( res>0 || (res==0 && nKey>nKey1) );
+ }
+#endif
+
+ if( pKey2 ){
+ res = memcmp(pKey, pKey2, MIN((size_t)nKey2, nKey));
+ if( res>0 || (res==0 && (size_t)nKey2<nKey) ){
+ delete [] pKey;
+ break;
+ }
+ }
+ pCur->remove();
+ delete [] pKey;
+ }
+
+ delete pCur;
+ return 0;
+}
+
+int test_kc_fetch(
+ TestDb *pDb,
+ void *pKey,
+ int nKey,
+ void **ppVal,
+ int *pnVal
+){
+ KcDb *pKcDb = (KcDb *)pDb;
+ size_t nVal;
+
+ if( pKcDb->pVal ){
+ delete [] pKcDb->pVal;
+ pKcDb->pVal = 0;
+ }
+
+ pKcDb->pVal = pKcDb->db->get((const char *)pKey, nKey, &nVal);
+ if( pKcDb->pVal ){
+ *ppVal = pKcDb->pVal;
+ *pnVal = nVal;
+ }else{
+ *ppVal = 0;
+ *pnVal = -1;
+ }
+
+ return 0;
+}
+
+int test_kc_scan(
+ TestDb *pDb, /* Database handle */
+ void *pCtx, /* Context pointer to pass to xCallback */
+ int bReverse, /* True for a reverse order scan */
+ void *pKey1, int nKey1, /* Start of search */
+ void *pKey2, int nKey2, /* End of search */
+ void (*xCallback)(void *pCtx, void *pKey, int nKey, void *pVal, int nVal)
+){
+ KcDb *pKcDb = (KcDb *)pDb;
+ kyotocabinet::DB::Cursor* pCur = pKcDb->db->cursor();
+ int res;
+
+ if( bReverse==0 ){
+ if( pKey1 ){
+ res = pCur->jump((const char *)pKey1, nKey1);
+ }else{
+ res = pCur->jump();
+ }
+ }else{
+ if( pKey2 ){
+ res = pCur->jump_back((const char *)pKey2, nKey2);
+ }else{
+ res = pCur->jump_back();
+ }
+ }
+
+ while( res ){
+ const char *pKey; size_t nKey;
+ const char *pVal; size_t nVal;
+ pKey = pCur->get(&nKey, &pVal, &nVal);
+
+ if( bReverse==0 && pKey2 ){
+ res = memcmp(pKey, pKey2, MIN((size_t)nKey2, nKey));
+ if( res>0 || (res==0 && (size_t)nKey2<nKey) ){
+ delete [] pKey;
+ break;
+ }
+ }else if( bReverse!=0 && pKey1 ){
+ res = memcmp(pKey, pKey1, MIN((size_t)nKey1, nKey));
+ if( res<0 || (res==0 && (size_t)nKey1>nKey) ){
+ delete [] pKey;
+ break;
+ }
+ }
+
+ xCallback(pCtx, (void *)pKey, (int)nKey, (void *)pVal, (int)nVal);
+ delete [] pKey;
+
+ if( bReverse ){
+ res = pCur->step_back();
+ }else{
+ res = pCur->step();
+ }
+ }
+
+ delete pCur;
+ return 0;
+}
+#endif /* HAVE_KYOTOCABINET */
+
+#ifdef HAVE_MDB
+#include "lmdb.h"
+
+extern "C" {
+ struct MdbDb {
+ TestDb base;
+ MDB_env *env;
+ MDB_dbi dbi;
+ };
+}
+
+int test_mdb_open(
+ const char *zSpec,
+ const char *zFilename,
+ int bClear,
+ TestDb **ppDb
+){
+ MDB_txn *txn;
+ MdbDb *pMdb;
+ int rc;
+
+ if( bClear ){
+ char *zCmd = sqlite3_mprintf("rm -rf %s\n", zFilename);
+ system(zCmd);
+ sqlite3_free(zCmd);
+ }
+
+ pMdb = (MdbDb *)malloc(sizeof(MdbDb));
+ memset(pMdb, 0, sizeof(MdbDb));
+
+ rc = mdb_env_create(&pMdb->env);
+ if( rc==0 ) rc = mdb_env_set_mapsize(pMdb->env, 1*1024*1024*1024);
+ if( rc==0 ) rc = mdb_env_open(pMdb->env, zFilename, MDB_NOSYNC|MDB_NOSUBDIR, 0600);
+ if( rc==0 ) rc = mdb_txn_begin(pMdb->env, NULL, 0, &txn);
+ if( rc==0 ){
+ rc = mdb_open(txn, NULL, 0, &pMdb->dbi);
+ mdb_txn_commit(txn);
+ }
+
+ *ppDb = (TestDb *)pMdb;
+ return rc;
+}
+
+int test_mdb_close(TestDb *pDb){
+ MdbDb *pMdb = (MdbDb *)pDb;
+
+ mdb_close(pMdb->env, pMdb->dbi);
+ mdb_env_close(pMdb->env);
+ free(pMdb);
+ return 0;
+}
+
+int test_mdb_write(TestDb *pDb, void *pKey, int nKey, void *pVal, int nVal){
+ int rc;
+ MdbDb *pMdb = (MdbDb *)pDb;
+ MDB_val val;
+ MDB_val key;
+ MDB_txn *txn;
+
+ val.mv_size = nVal;
+ val.mv_data = pVal;
+ key.mv_size = nKey;
+ key.mv_data = pKey;
+
+ rc = mdb_txn_begin(pMdb->env, NULL, 0, &txn);
+ if( rc==0 ){
+ rc = mdb_put(txn, pMdb->dbi, &key, &val, 0);
+ if( rc==0 ){
+ rc = mdb_txn_commit(txn);
+ }else{
+ mdb_txn_abort(txn);
+ }
+ }
+
+ return rc;
+}
+
+int test_mdb_delete(TestDb *pDb, void *pKey, int nKey){
+ int rc;
+ MdbDb *pMdb = (MdbDb *)pDb;
+ MDB_val key;
+ MDB_txn *txn;
+
+ key.mv_size = nKey;
+ key.mv_data = pKey;
+ rc = mdb_txn_begin(pMdb->env, NULL, 0, &txn);
+ if( rc==0 ){
+ rc = mdb_del(txn, pMdb->dbi, &key, 0);
+ if( rc==0 ){
+ rc = mdb_txn_commit(txn);
+ }else{
+ mdb_txn_abort(txn);
+ }
+ }
+
+ return rc;
+}
+
+int test_mdb_fetch(
+ TestDb *pDb,
+ void *pKey,
+ int nKey,
+ void **ppVal,
+ int *pnVal
+){
+ int rc;
+ MdbDb *pMdb = (MdbDb *)pDb;
+ MDB_val key;
+ MDB_txn *txn;
+
+ key.mv_size = nKey;
+ key.mv_data = pKey;
+
+ rc = mdb_txn_begin(pMdb->env, NULL, MDB_RDONLY, &txn);
+ if( rc==0 ){
+ MDB_val val = {0, 0};
+ rc = mdb_get(txn, pMdb->dbi, &key, &val);
+ if( rc==MDB_NOTFOUND ){
+ rc = 0;
+ *ppVal = 0;
+ *pnVal = -1;
+ }else{
+ *ppVal = val.mv_data;
+ *pnVal = val.mv_size;
+ }
+ mdb_txn_commit(txn);
+ }
+
+ return rc;
+}
+
+int test_mdb_scan(
+ TestDb *pDb, /* Database handle */
+ void *pCtx, /* Context pointer to pass to xCallback */
+ int bReverse, /* True for a reverse order scan */
+ void *pKey1, int nKey1, /* Start of search */
+ void *pKey2, int nKey2, /* End of search */
+ void (*xCallback)(void *pCtx, void *pKey, int nKey, void *pVal, int nVal)
+){
+ MdbDb *pMdb = (MdbDb *)pDb;
+ int rc;
+ MDB_cursor_op op = bReverse ? MDB_PREV : MDB_NEXT;
+ MDB_txn *txn;
+
+ rc = mdb_txn_begin(pMdb->env, NULL, MDB_RDONLY, &txn);
+ if( rc==0 ){
+ MDB_cursor *csr;
+ MDB_val key = {0, 0};
+ MDB_val val = {0, 0};
+
+ rc = mdb_cursor_open(txn, pMdb->dbi, &csr);
+ if( rc==0 ){
+ while( mdb_cursor_get(csr, &key, &val, op)==0 ){
+ xCallback(pCtx, key.mv_data, key.mv_size, val.mv_data, val.mv_size);
+ }
+ mdb_cursor_close(csr);
+ }
+ }
+
+ return rc;
+}
+
+#endif /* HAVE_MDB */
diff --git a/ext/lsm1/lsm-test/lsmtest_tdb3.c b/ext/lsm1/lsm-test/lsmtest_tdb3.c
new file mode 100644
index 0000000..e29497a
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_tdb3.c
@@ -0,0 +1,1429 @@
+
+#include "lsmtest_tdb.h"
+#include "lsm.h"
+#include "lsmtest.h"
+
+#include <stdlib.h>
+#include <string.h>
+#include <assert.h>
+#ifndef _WIN32
+# include <unistd.h>
+#endif
+#include <stdio.h>
+
+#ifndef _WIN32
+# include <sys/time.h>
+#endif
+
+typedef struct LsmDb LsmDb;
+typedef struct LsmWorker LsmWorker;
+typedef struct LsmFile LsmFile;
+
+#define LSMTEST_DFLT_MT_MAX_CKPT (8*1024)
+#define LSMTEST_DFLT_MT_MIN_CKPT (2*1024)
+
+#ifdef LSM_MUTEX_PTHREADS
+#include <pthread.h>
+
+#define LSMTEST_THREAD_CKPT 1
+#define LSMTEST_THREAD_WORKER 2
+#define LSMTEST_THREAD_WORKER_AC 3
+
+/*
+** There are several different types of worker threads that run in different
+** test configurations, depending on the value of LsmWorker.eType.
+**
+** 1. Checkpointer.
+** 2. Worker with auto-checkpoint.
+** 3. Worker without auto-checkpoint.
+*/
+struct LsmWorker {
+ LsmDb *pDb; /* Main database structure */
+ lsm_db *pWorker; /* Worker database handle */
+ pthread_t worker_thread; /* Worker thread */
+ pthread_cond_t worker_cond; /* Condition var the worker waits on */
+ pthread_mutex_t worker_mutex; /* Mutex used with worker_cond */
+ int bDoWork; /* Set to true by client when there is work */
+ int worker_rc; /* Store error code here */
+ int eType; /* LSMTEST_THREAD_XXX constant */
+ int bBlock;
+};
+#else
+struct LsmWorker { int worker_rc; int bBlock; };
+#endif
+
+static void mt_shutdown(LsmDb *);
+
+lsm_env *tdb_lsm_env(void){
+ static int bInit = 0;
+ static lsm_env env;
+ if( bInit==0 ){
+ memcpy(&env, lsm_default_env(), sizeof(env));
+ bInit = 1;
+ }
+ return &env;
+}
+
+typedef struct FileSector FileSector;
+typedef struct FileData FileData;
+
+struct FileSector {
+ u8 *aOld; /* Old data for this sector */
+};
+
+struct FileData {
+ int nSector; /* Allocated size of apSector[] array */
+ FileSector *aSector; /* Array of file sectors */
+};
+
+/*
+** bPrepareCrash:
+** If non-zero, the file wrappers maintain enough in-memory data to
+** simulate the effect of a power-failure on the file-system (i.e. that
+** unsynced sectors may be written, not written, or overwritten with
+** arbitrary data when the crash occurs).
+**
+** bCrashed:
+** Set to true after a crash is simulated. Once this variable is true, all
+** VFS methods other than xClose() return LSM_IOERR as soon as they are
+** called (without affecting the contents of the file-system).
+**
+** env:
+** The environment object used by all lsm_db* handles opened by this
+** object (i.e. LsmDb.db plus any worker connections). Variable env.pVfsCtx
+** always points to the containing LsmDb structure.
+*/
+struct LsmDb {
+ TestDb base; /* Base class - methods table */
+ lsm_env env; /* Environment used by connection db */
+ char *zName; /* Database file name */
+ lsm_db *db; /* LSM database handle */
+
+ lsm_cursor *pCsr; /* Cursor held open during read transaction */
+ void *pBuf; /* Buffer for tdb_fetch() output */
+ int nBuf; /* Allocated (not used) size of pBuf */
+
+ /* Crash testing related state */
+ int bCrashed; /* True once a crash has occurred */
+ int nAutoCrash; /* Number of syncs until a crash */
+ int bPrepareCrash; /* True to store writes in memory */
+
+ /* Unsynced data (while crash testing) */
+ int szSector; /* Assumed size of disk sectors (512B) */
+ FileData aFile[2]; /* Database and log file data */
+
+ /* Other test instrumentation */
+ int bNoRecovery; /* If true, assume DMS2 is locked */
+
+ /* Work hook redirection */
+ void (*xWork)(lsm_db *, void *);
+ void *pWorkCtx;
+
+ /* IO logging hook */
+ void (*xWriteHook)(void *, int, lsm_i64, int, int);
+ void *pWriteCtx;
+
+ /* Worker threads (for lsm_mt) */
+ int nMtMinCkpt;
+ int nMtMaxCkpt;
+ int eMode;
+ int nWorker;
+ LsmWorker *aWorker;
+};
+
+#define LSMTEST_MODE_SINGLETHREAD 1
+#define LSMTEST_MODE_BACKGROUND_CKPT 2
+#define LSMTEST_MODE_BACKGROUND_WORK 3
+#define LSMTEST_MODE_BACKGROUND_BOTH 4
+
+/*************************************************************************
+**************************************************************************
+** Begin test VFS code.
+*/
+
+struct LsmFile {
+ lsm_file *pReal; /* Real underlying file */
+ int bLog; /* True for log file. False for db file */
+ LsmDb *pDb; /* Database handle that uses this file */
+};
+
+static int testEnvFullpath(
+ lsm_env *pEnv, /* Environment for current LsmDb */
+ const char *zFile, /* Relative path name */
+ char *zOut, /* Output buffer */
+ int *pnOut /* IN/OUT: Size of output buffer */
+){
+ lsm_env *pRealEnv = tdb_lsm_env();
+ return pRealEnv->xFullpath(pRealEnv, zFile, zOut, pnOut);
+}
+
+static int testEnvOpen(
+ lsm_env *pEnv, /* Environment for current LsmDb */
+ const char *zFile, /* Name of file to open */
+ int flags,
+ lsm_file **ppFile /* OUT: New file handle object */
+){
+ lsm_env *pRealEnv = tdb_lsm_env();
+ LsmDb *pDb = (LsmDb *)pEnv->pVfsCtx;
+ int rc; /* Return Code */
+ LsmFile *pRet; /* The new file handle */
+ int nFile; /* Length of string zFile in bytes */
+
+ nFile = strlen(zFile);
+ pRet = (LsmFile *)testMalloc(sizeof(LsmFile));
+ pRet->pDb = pDb;
+ pRet->bLog = (nFile > 4 && 0==memcmp("-log", &zFile[nFile-4], 4));
+
+ rc = pRealEnv->xOpen(pRealEnv, zFile, flags, &pRet->pReal);
+ if( rc!=LSM_OK ){
+ testFree(pRet);
+ pRet = 0;
+ }
+
+ *ppFile = (lsm_file *)pRet;
+ return rc;
+}
+
+static int testEnvRead(lsm_file *pFile, lsm_i64 iOff, void *pData, int nData){
+ lsm_env *pRealEnv = tdb_lsm_env();
+ LsmFile *p = (LsmFile *)pFile;
+ if( p->pDb->bCrashed ) return LSM_IOERR;
+ return pRealEnv->xRead(p->pReal, iOff, pData, nData);
+}
+
+static int testEnvWrite(lsm_file *pFile, lsm_i64 iOff, void *pData, int nData){
+ lsm_env *pRealEnv = tdb_lsm_env();
+ LsmFile *p = (LsmFile *)pFile;
+ LsmDb *pDb = p->pDb;
+
+ if( pDb->bCrashed ) return LSM_IOERR;
+
+ if( pDb->bPrepareCrash ){
+ FileData *pData2 = &pDb->aFile[p->bLog];
+ int iFirst;
+ int iLast;
+ int iSector;
+
+ iFirst = (int)(iOff / pDb->szSector);
+ iLast = (int)((iOff + nData - 1) / pDb->szSector);
+
+ if( pData2->nSector<(iLast+1) ){
+ int nNew = ( ((iLast + 1) + 63) / 64 ) * 64;
+ assert( nNew>iLast );
+ pData2->aSector = (FileSector *)testRealloc(
+ pData2->aSector, nNew*sizeof(FileSector)
+ );
+ memset(&pData2->aSector[pData2->nSector],
+ 0, (nNew - pData2->nSector) * sizeof(FileSector)
+ );
+ pData2->nSector = nNew;
+ }
+
+ for(iSector=iFirst; iSector<=iLast; iSector++){
+ if( pData2->aSector[iSector].aOld==0 ){
+ u8 *aOld = (u8 *)testMalloc(pDb->szSector);
+ pRealEnv->xRead(
+ p->pReal, (lsm_i64)iSector*pDb->szSector, aOld, pDb->szSector
+ );
+ pData2->aSector[iSector].aOld = aOld;
+ }
+ }
+ }
+
+ if( pDb->xWriteHook ){
+ int rc;
+ int nUs;
+ struct timeval t1;
+ struct timeval t2;
+
+ gettimeofday(&t1, 0);
+ assert( nData>0 );
+ rc = pRealEnv->xWrite(p->pReal, iOff, pData, nData);
+ gettimeofday(&t2, 0);
+
+ nUs = (t2.tv_sec - t1.tv_sec) * 1000000 + (t2.tv_usec - t1.tv_usec);
+ pDb->xWriteHook(pDb->pWriteCtx, p->bLog, iOff, nData, nUs);
+ return rc;
+ }
+
+ return pRealEnv->xWrite(p->pReal, iOff, pData, nData);
+}
+
+static void doSystemCrash(LsmDb *pDb);
+
+static int testEnvSync(lsm_file *pFile){
+ lsm_env *pRealEnv = tdb_lsm_env();
+ LsmFile *p = (LsmFile *)pFile;
+ LsmDb *pDb = p->pDb;
+ FileData *pData = &pDb->aFile[p->bLog];
+ int i;
+
+ if( pDb->bCrashed ) return LSM_IOERR;
+
+ if( pDb->nAutoCrash ){
+ pDb->nAutoCrash--;
+ if( pDb->nAutoCrash==0 ){
+ doSystemCrash(pDb);
+ pDb->bCrashed = 1;
+ return LSM_IOERR;
+ }
+ }
+
+ if( pDb->bPrepareCrash ){
+ for(i=0; i<pData->nSector; i++){
+ testFree(pData->aSector[i].aOld);
+ pData->aSector[i].aOld = 0;
+ }
+ }
+
+ if( pDb->xWriteHook ){
+ int rc;
+ int nUs;
+ struct timeval t1;
+ struct timeval t2;
+
+ gettimeofday(&t1, 0);
+ rc = pRealEnv->xSync(p->pReal);
+ gettimeofday(&t2, 0);
+
+ nUs = (t2.tv_sec - t1.tv_sec) * 1000000 + (t2.tv_usec - t1.tv_usec);
+ pDb->xWriteHook(pDb->pWriteCtx, p->bLog, 0, 0, nUs);
+ return rc;
+ }
+
+ return pRealEnv->xSync(p->pReal);
+}
+
+static int testEnvTruncate(lsm_file *pFile, lsm_i64 iOff){
+ lsm_env *pRealEnv = tdb_lsm_env();
+ LsmFile *p = (LsmFile *)pFile;
+ if( p->pDb->bCrashed ) return LSM_IOERR;
+ return pRealEnv->xTruncate(p->pReal, iOff);
+}
+
+static int testEnvSectorSize(lsm_file *pFile){
+ lsm_env *pRealEnv = tdb_lsm_env();
+ LsmFile *p = (LsmFile *)pFile;
+ return pRealEnv->xSectorSize(p->pReal);
+}
+
+static int testEnvRemap(
+ lsm_file *pFile,
+ lsm_i64 iMin,
+ void **ppOut,
+ lsm_i64 *pnOut
+){
+ lsm_env *pRealEnv = tdb_lsm_env();
+ LsmFile *p = (LsmFile *)pFile;
+ return pRealEnv->xRemap(p->pReal, iMin, ppOut, pnOut);
+}
+
+static int testEnvFileid(
+ lsm_file *pFile,
+ void *ppOut,
+ int *pnOut
+){
+ lsm_env *pRealEnv = tdb_lsm_env();
+ LsmFile *p = (LsmFile *)pFile;
+ return pRealEnv->xFileid(p->pReal, ppOut, pnOut);
+}
+
+static int testEnvClose(lsm_file *pFile){
+ lsm_env *pRealEnv = tdb_lsm_env();
+ LsmFile *p = (LsmFile *)pFile;
+
+ pRealEnv->xClose(p->pReal);
+ testFree(p);
+ return LSM_OK;
+}
+
+static int testEnvUnlink(lsm_env *pEnv, const char *zFile){
+ lsm_env *pRealEnv = tdb_lsm_env();
+ unused_parameter(pEnv);
+ return pRealEnv->xUnlink(pRealEnv, zFile);
+}
+
+static int testEnvLock(lsm_file *pFile, int iLock, int eType){
+ LsmFile *p = (LsmFile *)pFile;
+ lsm_env *pRealEnv = tdb_lsm_env();
+
+ if( iLock==2 && eType==LSM_LOCK_EXCL && p->pDb->bNoRecovery ){
+ return LSM_BUSY;
+ }
+ return pRealEnv->xLock(p->pReal, iLock, eType);
+}
+
+static int testEnvTestLock(lsm_file *pFile, int iLock, int nLock, int eType){
+ LsmFile *p = (LsmFile *)pFile;
+ lsm_env *pRealEnv = tdb_lsm_env();
+
+ if( iLock==2 && eType==LSM_LOCK_EXCL && p->pDb->bNoRecovery ){
+ return LSM_BUSY;
+ }
+ return pRealEnv->xTestLock(p->pReal, iLock, nLock, eType);
+}
+
+static int testEnvShmMap(lsm_file *pFile, int iRegion, int sz, void **pp){
+ LsmFile *p = (LsmFile *)pFile;
+ lsm_env *pRealEnv = tdb_lsm_env();
+ return pRealEnv->xShmMap(p->pReal, iRegion, sz, pp);
+}
+
+static void testEnvShmBarrier(void){
+}
+
+static int testEnvShmUnmap(lsm_file *pFile, int bDel){
+ LsmFile *p = (LsmFile *)pFile;
+ lsm_env *pRealEnv = tdb_lsm_env();
+ return pRealEnv->xShmUnmap(p->pReal, bDel);
+}
+
+static int testEnvSleep(lsm_env *pEnv, int us){
+ lsm_env *pRealEnv = tdb_lsm_env();
+ return pRealEnv->xSleep(pRealEnv, us);
+}
+
+static void doSystemCrash(LsmDb *pDb){
+ lsm_env *pEnv = tdb_lsm_env();
+ int iFile;
+ int iSeed = pDb->aFile[0].nSector + pDb->aFile[1].nSector;
+
+ char *zFile = pDb->zName;
+ char *zFree = 0;
+
+ for(iFile=0; iFile<2; iFile++){
+ lsm_file *pFile = 0;
+ int i;
+
+ pEnv->xOpen(pEnv, zFile, 0, &pFile);
+ for(i=0; i<pDb->aFile[iFile].nSector; i++){
+ u8 *aOld = pDb->aFile[iFile].aSector[i].aOld;
+ if( aOld ){
+ int iOpt = testPrngValue(iSeed++) % 3;
+ switch( iOpt ){
+ case 0:
+ break;
+
+ case 1:
+ testPrngArray(iSeed++, (u32 *)aOld, pDb->szSector/4);
+ /* Fall-through */
+
+ case 2:
+ pEnv->xWrite(
+ pFile, (lsm_i64)i * pDb->szSector, aOld, pDb->szSector
+ );
+ break;
+ }
+ testFree(aOld);
+ pDb->aFile[iFile].aSector[i].aOld = 0;
+ }
+ }
+ pEnv->xClose(pFile);
+ zFree = zFile = sqlite3_mprintf("%s-log", pDb->zName);
+ }
+
+ sqlite3_free(zFree);
+}
+/*
+** End test VFS code.
+**************************************************************************
+*************************************************************************/
+
+/*************************************************************************
+**************************************************************************
+** Begin test compression hooks.
+*/
+
+#ifdef HAVE_ZLIB
+#include <zlib.h>
+
+static int testZipBound(void *pCtx, int nSrc){
+ return compressBound(nSrc);
+}
+
+static int testZipCompress(
+ void *pCtx, /* Context pointer */
+ char *aOut, int *pnOut, /* OUT: Buffer containing compressed data */
+ const char *aIn, int nIn /* Buffer containing input data */
+){
+ uLongf n = *pnOut; /* In/out buffer size for compress() */
+ int rc; /* compress() return code */
+
+ rc = compress((Bytef*)aOut, &n, (Bytef*)aIn, nIn);
+ *pnOut = n;
+ return (rc==Z_OK ? 0 : LSM_ERROR);
+}
+
+static int testZipUncompress(
+ void *pCtx, /* Context pointer */
+ char *aOut, int *pnOut, /* OUT: Buffer containing uncompressed data */
+ const char *aIn, int nIn /* Buffer containing input data */
+){
+ uLongf n = *pnOut; /* In/out buffer size for uncompress() */
+ int rc; /* uncompress() return code */
+
+ rc = uncompress((Bytef*)aOut, &n, (Bytef*)aIn, nIn);
+ *pnOut = n;
+ return (rc==Z_OK ? 0 : LSM_ERROR);
+}
+
+static int testConfigureCompression(lsm_db *pDb){
+ static lsm_compress zip = {
+ 0, /* Context pointer (unused) */
+ 1, /* Id value */
+ testZipBound, /* xBound method */
+ testZipCompress, /* xCompress method */
+ testZipUncompress /* xUncompress method */
+ };
+ return lsm_config(pDb, LSM_CONFIG_SET_COMPRESSION, &zip);
+}
+#endif /* ifdef HAVE_ZLIB */
+
+/*
+** End test compression hooks.
+**************************************************************************
+*************************************************************************/
+
+static int test_lsm_close(TestDb *pTestDb){
+ int i;
+ int rc = LSM_OK;
+ LsmDb *pDb = (LsmDb *)pTestDb;
+
+ lsm_csr_close(pDb->pCsr);
+ lsm_close(pDb->db);
+
+ /* If this is a multi-threaded database, wait on the worker threads. */
+ mt_shutdown(pDb);
+ for(i=0; i<pDb->nWorker && rc==LSM_OK; i++){
+ rc = pDb->aWorker[i].worker_rc;
+ }
+
+ for(i=0; i<pDb->aFile[0].nSector; i++){
+ testFree(pDb->aFile[0].aSector[i].aOld);
+ }
+ testFree(pDb->aFile[0].aSector);
+ for(i=0; i<pDb->aFile[1].nSector; i++){
+ testFree(pDb->aFile[1].aSector[i].aOld);
+ }
+ testFree(pDb->aFile[1].aSector);
+
+ memset(pDb, sizeof(LsmDb), 0x11);
+ testFree((char *)pDb->pBuf);
+ testFree((char *)pDb);
+ return rc;
+}
+
+static void mt_signal_worker(LsmDb*, int);
+
+static int waitOnCheckpointer(LsmDb *pDb, lsm_db *db){
+ int nSleep = 0;
+ int nKB;
+ int rc;
+
+ do {
+ nKB = 0;
+ rc = lsm_info(db, LSM_INFO_CHECKPOINT_SIZE, &nKB);
+ if( rc!=LSM_OK || nKB<pDb->nMtMaxCkpt ) break;
+#ifdef LSM_MUTEX_PTHREADS
+ mt_signal_worker(pDb,
+ (pDb->eMode==LSMTEST_MODE_BACKGROUND_CKPT ? 0 : 1)
+ );
+#endif
+ usleep(5000);
+ nSleep += 5;
+ }while( 1 );
+
+#if 0
+ if( nSleep ) printf("# waitOnCheckpointer(): nSleep=%d\n", nSleep);
+#endif
+
+ return rc;
+}
+
+static int waitOnWorker(LsmDb *pDb){
+ int rc;
+ int nLimit = -1;
+ int nSleep = 0;
+
+ rc = lsm_config(pDb->db, LSM_CONFIG_AUTOFLUSH, &nLimit);
+ do {
+ int nOld, nNew, rc2;
+ rc2 = lsm_info(pDb->db, LSM_INFO_TREE_SIZE, &nOld, &nNew);
+ if( rc2!=LSM_OK ) return rc2;
+ if( nOld==0 || nNew<(nLimit/2) ) break;
+#ifdef LSM_MUTEX_PTHREADS
+ mt_signal_worker(pDb, 0);
+#endif
+ usleep(5000);
+ nSleep += 5;
+ }while( 1 );
+
+#if 0
+ if( nSleep ) printf("# waitOnWorker(): nSleep=%d\n", nSleep);
+#endif
+
+ return rc;
+}
+
+static int test_lsm_write(
+ TestDb *pTestDb,
+ void *pKey,
+ int nKey,
+ void *pVal,
+ int nVal
+){
+ LsmDb *pDb = (LsmDb *)pTestDb;
+ int rc = LSM_OK;
+
+ if( pDb->eMode==LSMTEST_MODE_BACKGROUND_CKPT ){
+ rc = waitOnCheckpointer(pDb, pDb->db);
+ }else if(
+ pDb->eMode==LSMTEST_MODE_BACKGROUND_WORK
+ || pDb->eMode==LSMTEST_MODE_BACKGROUND_BOTH
+ ){
+ rc = waitOnWorker(pDb);
+ }
+
+ if( rc==LSM_OK ){
+ rc = lsm_insert(pDb->db, pKey, nKey, pVal, nVal);
+ }
+ return rc;
+}
+
+static int test_lsm_delete(TestDb *pTestDb, void *pKey, int nKey){
+ LsmDb *pDb = (LsmDb *)pTestDb;
+ return lsm_delete(pDb->db, pKey, nKey);
+}
+
+static int test_lsm_delete_range(
+ TestDb *pTestDb,
+ void *pKey1, int nKey1,
+ void *pKey2, int nKey2
+){
+ LsmDb *pDb = (LsmDb *)pTestDb;
+ return lsm_delete_range(pDb->db, pKey1, nKey1, pKey2, nKey2);
+}
+
+static int test_lsm_fetch(
+ TestDb *pTestDb,
+ void *pKey,
+ int nKey,
+ void **ppVal,
+ int *pnVal
+){
+ int rc;
+ LsmDb *pDb = (LsmDb *)pTestDb;
+ lsm_cursor *csr;
+
+ if( pKey==0 ) return LSM_OK;
+
+ if( pDb->pCsr==0 ){
+ rc = lsm_csr_open(pDb->db, &csr);
+ if( rc!=LSM_OK ) return rc;
+ }else{
+ csr = pDb->pCsr;
+ }
+
+ rc = lsm_csr_seek(csr, pKey, nKey, LSM_SEEK_EQ);
+ if( rc==LSM_OK ){
+ if( lsm_csr_valid(csr) ){
+ const void *pVal; int nVal;
+ rc = lsm_csr_value(csr, &pVal, &nVal);
+ if( nVal>pDb->nBuf ){
+ testFree(pDb->pBuf);
+ pDb->pBuf = testMalloc(nVal*2);
+ pDb->nBuf = nVal*2;
+ }
+ memcpy(pDb->pBuf, pVal, nVal);
+ *ppVal = pDb->pBuf;
+ *pnVal = nVal;
+ }else{
+ *ppVal = 0;
+ *pnVal = -1;
+ }
+ }
+ if( pDb->pCsr==0 ){
+ lsm_csr_close(csr);
+ }
+ return rc;
+}
+
+static int test_lsm_scan(
+ TestDb *pTestDb,
+ void *pCtx,
+ int bReverse,
+ void *pFirst, int nFirst,
+ void *pLast, int nLast,
+ void (*xCallback)(void *, void *, int , void *, int)
+){
+ LsmDb *pDb = (LsmDb *)pTestDb;
+ lsm_cursor *csr;
+ lsm_cursor *csr2 = 0;
+ int rc;
+
+ if( pDb->pCsr==0 ){
+ rc = lsm_csr_open(pDb->db, &csr);
+ if( rc!=LSM_OK ) return rc;
+ }else{
+ rc = LSM_OK;
+ csr = pDb->pCsr;
+ }
+
+ /* To enhance testing, if both pLast and pFirst are defined, seek the
+ ** cursor to the "end" boundary here. Then the next block seeks it to
+ ** the "start" ready for the scan. The point is to test that cursors
+ ** can be reused. */
+ if( pLast && pFirst ){
+ if( bReverse ){
+ rc = lsm_csr_seek(csr, pFirst, nFirst, LSM_SEEK_LE);
+ }else{
+ rc = lsm_csr_seek(csr, pLast, nLast, LSM_SEEK_GE);
+ }
+ }
+
+ if( bReverse ){
+ if( pLast ){
+ rc = lsm_csr_seek(csr, pLast, nLast, LSM_SEEK_LE);
+ }else{
+ rc = lsm_csr_last(csr);
+ }
+ }else{
+ if( pFirst ){
+ rc = lsm_csr_seek(csr, pFirst, nFirst, LSM_SEEK_GE);
+ }else{
+ rc = lsm_csr_first(csr);
+ }
+ }
+
+ while( rc==LSM_OK && lsm_csr_valid(csr) ){
+ const void *pKey; int nKey;
+ const void *pVal; int nVal;
+ int cmp;
+
+ lsm_csr_key(csr, &pKey, &nKey);
+ lsm_csr_value(csr, &pVal, &nVal);
+
+ if( bReverse && pFirst ){
+ cmp = memcmp(pFirst, pKey, MIN(nKey, nFirst));
+ if( cmp>0 || (cmp==0 && nFirst>nKey) ) break;
+ }else if( bReverse==0 && pLast ){
+ cmp = memcmp(pLast, pKey, MIN(nKey, nLast));
+ if( cmp<0 || (cmp==0 && nLast<nKey) ) break;
+ }
+
+ xCallback(pCtx, (void *)pKey, nKey, (void *)pVal, nVal);
+
+ if( bReverse ){
+ rc = lsm_csr_prev(csr);
+ }else{
+ rc = lsm_csr_next(csr);
+ }
+ }
+
+ if( pDb->pCsr==0 ){
+ lsm_csr_close(csr);
+ }
+ return rc;
+}
+
+static int test_lsm_begin(TestDb *pTestDb, int iLevel){
+ int rc = LSM_OK;
+ LsmDb *pDb = (LsmDb *)pTestDb;
+
+ /* iLevel==0 is a no-op. */
+ if( iLevel==0 ) return 0;
+
+ if( pDb->pCsr==0 ) rc = lsm_csr_open(pDb->db, &pDb->pCsr);
+ if( rc==LSM_OK && iLevel>1 ){
+ rc = lsm_begin(pDb->db, iLevel-1);
+ }
+
+ return rc;
+}
+static int test_lsm_commit(TestDb *pTestDb, int iLevel){
+ LsmDb *pDb = (LsmDb *)pTestDb;
+
+ /* If iLevel==0, close any open read transaction */
+ if( iLevel==0 && pDb->pCsr ){
+ lsm_csr_close(pDb->pCsr);
+ pDb->pCsr = 0;
+ }
+
+ /* If iLevel==0, close any open read transaction */
+ return lsm_commit(pDb->db, MAX(0, iLevel-1));
+}
+static int test_lsm_rollback(TestDb *pTestDb, int iLevel){
+ LsmDb *pDb = (LsmDb *)pTestDb;
+
+ /* If iLevel==0, close any open read transaction */
+ if( iLevel==0 && pDb->pCsr ){
+ lsm_csr_close(pDb->pCsr);
+ pDb->pCsr = 0;
+ }
+
+ return lsm_rollback(pDb->db, MAX(0, iLevel-1));
+}
+
+/*
+** A log message callback registered with lsm connections. Prints all
+** messages to stderr.
+*/
+static void xLog(void *pCtx, int rc, const char *z){
+ unused_parameter(rc);
+ /* fprintf(stderr, "lsm: rc=%d \"%s\"\n", rc, z); */
+ if( pCtx ) fprintf(stderr, "%s: ", (char *)pCtx);
+ fprintf(stderr, "%s\n", z);
+ fflush(stderr);
+}
+
+static void xWorkHook(lsm_db *db, void *pArg){
+ LsmDb *p = (LsmDb *)pArg;
+ if( p->xWork ) p->xWork(db, p->pWorkCtx);
+}
+
+#define TEST_NO_RECOVERY -1
+#define TEST_COMPRESSION -3
+
+#define TEST_MT_MODE -2
+#define TEST_MT_MIN_CKPT -4
+#define TEST_MT_MAX_CKPT -5
+
+
+int test_lsm_config_str(
+ LsmDb *pLsm,
+ lsm_db *db,
+ int bWorker,
+ const char *zStr,
+ int *pnThread
+){
+ struct CfgParam {
+ const char *zParam;
+ int bWorker;
+ int eParam;
+ } aParam[] = {
+ { "autoflush", 0, LSM_CONFIG_AUTOFLUSH },
+ { "page_size", 0, LSM_CONFIG_PAGE_SIZE },
+ { "block_size", 0, LSM_CONFIG_BLOCK_SIZE },
+ { "safety", 0, LSM_CONFIG_SAFETY },
+ { "autowork", 0, LSM_CONFIG_AUTOWORK },
+ { "autocheckpoint", 0, LSM_CONFIG_AUTOCHECKPOINT },
+ { "mmap", 0, LSM_CONFIG_MMAP },
+ { "use_log", 0, LSM_CONFIG_USE_LOG },
+ { "automerge", 0, LSM_CONFIG_AUTOMERGE },
+ { "max_freelist", 0, LSM_CONFIG_MAX_FREELIST },
+ { "multi_proc", 0, LSM_CONFIG_MULTIPLE_PROCESSES },
+ { "worker_automerge", 1, LSM_CONFIG_AUTOMERGE },
+ { "test_no_recovery", 0, TEST_NO_RECOVERY },
+ { "bg_min_ckpt", 0, TEST_NO_RECOVERY },
+
+ { "mt_mode", 0, TEST_MT_MODE },
+ { "mt_min_ckpt", 0, TEST_MT_MIN_CKPT },
+ { "mt_max_ckpt", 0, TEST_MT_MAX_CKPT },
+
+#ifdef HAVE_ZLIB
+ { "compression", 0, TEST_COMPRESSION },
+#endif
+ { 0, 0 }
+ };
+ const char *z = zStr;
+ int nThread = 1;
+
+ if( zStr==0 ) return 0;
+
+ assert( db );
+ while( z[0] ){
+ const char *zStart;
+
+ /* Skip whitespace */
+ while( *z==' ' ) z++;
+ zStart = z;
+
+ while( *z && *z!='=' ) z++;
+ if( *z ){
+ int eParam;
+ int i;
+ int iVal;
+ int iMul = 1;
+ int rc;
+ char zParam[32];
+ int nParam = z-zStart;
+ if( nParam==0 || nParam>sizeof(zParam)-1 ) goto syntax_error;
+
+ memcpy(zParam, zStart, nParam);
+ zParam[nParam] = '\0';
+ rc = testArgSelect(aParam, "param", zParam, &i);
+ if( rc!=0 ) return rc;
+ eParam = aParam[i].eParam;
+
+ z++;
+ zStart = z;
+ while( *z>='0' && *z<='9' ) z++;
+ if( *z=='k' || *z=='K' ){
+ iMul = 1;
+ z++;
+ }else if( *z=='M' || *z=='M' ){
+ iMul = 1024;
+ z++;
+ }
+ nParam = z-zStart;
+ if( nParam==0 || nParam>sizeof(zParam)-1 ) goto syntax_error;
+ memcpy(zParam, zStart, nParam);
+ zParam[nParam] = '\0';
+ iVal = atoi(zParam) * iMul;
+
+ if( eParam>0 ){
+ if( bWorker || aParam[i].bWorker==0 ){
+ lsm_config(db, eParam, &iVal);
+ }
+ }else{
+ switch( eParam ){
+ case TEST_NO_RECOVERY:
+ if( pLsm ) pLsm->bNoRecovery = iVal;
+ break;
+ case TEST_MT_MODE:
+ if( pLsm ) nThread = iVal;
+ break;
+ case TEST_MT_MIN_CKPT:
+ if( pLsm && iVal>0 ) pLsm->nMtMinCkpt = iVal*1024;
+ break;
+ case TEST_MT_MAX_CKPT:
+ if( pLsm && iVal>0 ) pLsm->nMtMaxCkpt = iVal*1024;
+ break;
+#ifdef HAVE_ZLIB
+ case TEST_COMPRESSION:
+ testConfigureCompression(db);
+ break;
+#endif
+ }
+ }
+ }else if( z!=zStart ){
+ goto syntax_error;
+ }
+ }
+
+ if( pnThread ) *pnThread = nThread;
+ if( pLsm && pLsm->nMtMaxCkpt < pLsm->nMtMinCkpt ){
+ pLsm->nMtMinCkpt = pLsm->nMtMaxCkpt;
+ }
+
+ return 0;
+ syntax_error:
+ testPrintError("syntax error at: \"%s\"\n", z);
+ return 1;
+}
+
+int tdb_lsm_config_str(TestDb *pDb, const char *zStr){
+ int rc = 0;
+ if( tdb_lsm(pDb) ){
+#ifdef LSM_MUTEX_PTHREADS
+ int i;
+#endif
+ LsmDb *pLsm = (LsmDb *)pDb;
+
+ rc = test_lsm_config_str(pLsm, pLsm->db, 0, zStr, 0);
+#ifdef LSM_MUTEX_PTHREADS
+ for(i=0; rc==0 && i<pLsm->nWorker; i++){
+ rc = test_lsm_config_str(0, pLsm->aWorker[i].pWorker, 1, zStr, 0);
+ }
+#endif
+ }
+ return rc;
+}
+
+int tdb_lsm_configure(lsm_db *db, const char *zConfig){
+ return test_lsm_config_str(0, db, 0, zConfig, 0);
+}
+
+static int testLsmStartWorkers(LsmDb *, int, const char *, const char *);
+
+static int testLsmOpen(
+ const char *zCfg,
+ const char *zFilename,
+ int bClear,
+ TestDb **ppDb
+){
+ static const DatabaseMethods LsmMethods = {
+ test_lsm_close,
+ test_lsm_write,
+ test_lsm_delete,
+ test_lsm_delete_range,
+ test_lsm_fetch,
+ test_lsm_scan,
+ test_lsm_begin,
+ test_lsm_commit,
+ test_lsm_rollback
+ };
+
+ int rc;
+ int nFilename;
+ LsmDb *pDb;
+
+ /* If the bClear flag is set, delete any existing database. */
+ assert( zFilename);
+ if( bClear ) testDeleteLsmdb(zFilename);
+ nFilename = strlen(zFilename);
+
+ pDb = (LsmDb *)testMalloc(sizeof(LsmDb) + nFilename + 1);
+ memset(pDb, 0, sizeof(LsmDb));
+ pDb->base.pMethods = &LsmMethods;
+ pDb->zName = (char *)&pDb[1];
+ memcpy(pDb->zName, zFilename, nFilename + 1);
+
+ /* Default the sector size used for crash simulation to 512 bytes.
+ ** Todo: There should be an OS method to obtain this value - just as
+ ** there is in SQLite. For now, LSM assumes that it is smaller than
+ ** the page size (default 4KB).
+ */
+ pDb->szSector = 256;
+
+ /* Default values for the mt_min_ckpt and mt_max_ckpt parameters. */
+ pDb->nMtMinCkpt = LSMTEST_DFLT_MT_MIN_CKPT;
+ pDb->nMtMaxCkpt = LSMTEST_DFLT_MT_MAX_CKPT;
+
+ memcpy(&pDb->env, tdb_lsm_env(), sizeof(lsm_env));
+ pDb->env.pVfsCtx = (void *)pDb;
+ pDb->env.xFullpath = testEnvFullpath;
+ pDb->env.xOpen = testEnvOpen;
+ pDb->env.xRead = testEnvRead;
+ pDb->env.xWrite = testEnvWrite;
+ pDb->env.xTruncate = testEnvTruncate;
+ pDb->env.xSync = testEnvSync;
+ pDb->env.xSectorSize = testEnvSectorSize;
+ pDb->env.xRemap = testEnvRemap;
+ pDb->env.xFileid = testEnvFileid;
+ pDb->env.xClose = testEnvClose;
+ pDb->env.xUnlink = testEnvUnlink;
+ pDb->env.xLock = testEnvLock;
+ pDb->env.xTestLock = testEnvTestLock;
+ pDb->env.xShmBarrier = testEnvShmBarrier;
+ pDb->env.xShmMap = testEnvShmMap;
+ pDb->env.xShmUnmap = testEnvShmUnmap;
+ pDb->env.xSleep = testEnvSleep;
+
+ rc = lsm_new(&pDb->env, &pDb->db);
+ if( rc==LSM_OK ){
+ int nThread = 1;
+ lsm_config_log(pDb->db, xLog, 0);
+ lsm_config_work_hook(pDb->db, xWorkHook, (void *)pDb);
+
+ rc = test_lsm_config_str(pDb, pDb->db, 0, zCfg, &nThread);
+ if( rc==LSM_OK ) rc = lsm_open(pDb->db, zFilename);
+
+ pDb->eMode = nThread;
+#ifdef LSM_MUTEX_PTHREADS
+ if( rc==LSM_OK && nThread>1 ){
+ testLsmStartWorkers(pDb, nThread, zFilename, zCfg);
+ }
+#endif
+
+ if( rc!=LSM_OK ){
+ test_lsm_close((TestDb *)pDb);
+ pDb = 0;
+ }
+ }
+
+ *ppDb = (TestDb *)pDb;
+ return rc;
+}
+
+int test_lsm_open(
+ const char *zSpec,
+ const char *zFilename,
+ int bClear,
+ TestDb **ppDb
+){
+ return testLsmOpen(zSpec, zFilename, bClear, ppDb);
+}
+
+int test_lsm_small_open(
+ const char *zSpec,
+ const char *zFile,
+ int bClear,
+ TestDb **ppDb
+){
+ const char *zCfg = "page_size=256 block_size=64 mmap=1024";
+ return testLsmOpen(zCfg, zFile, bClear, ppDb);
+}
+
+int test_lsm_lomem_open(
+ const char *zSpec,
+ const char *zFilename,
+ int bClear,
+ TestDb **ppDb
+){
+ /* "max_freelist=4 autocheckpoint=32" */
+ const char *zCfg =
+ "page_size=256 block_size=64 autoflush=16 "
+ "autocheckpoint=32"
+ "mmap=0 "
+ ;
+ return testLsmOpen(zCfg, zFilename, bClear, ppDb);
+}
+
+int test_lsm_lomem2_open(
+ const char *zSpec,
+ const char *zFilename,
+ int bClear,
+ TestDb **ppDb
+){
+ /* "max_freelist=4 autocheckpoint=32" */
+ const char *zCfg =
+ "page_size=512 block_size=64 autoflush=0 mmap=0 "
+ ;
+ return testLsmOpen(zCfg, zFilename, bClear, ppDb);
+}
+
+int test_lsm_zip_open(
+ const char *zSpec,
+ const char *zFilename,
+ int bClear,
+ TestDb **ppDb
+){
+ const char *zCfg =
+ "page_size=256 block_size=64 autoflush=16 "
+ "autocheckpoint=32 compression=1 mmap=0 "
+ ;
+ return testLsmOpen(zCfg, zFilename, bClear, ppDb);
+}
+
+lsm_db *tdb_lsm(TestDb *pDb){
+ if( pDb->pMethods->xClose==test_lsm_close ){
+ return ((LsmDb *)pDb)->db;
+ }
+ return 0;
+}
+
+int tdb_lsm_multithread(TestDb *pDb){
+ int ret = 0;
+ if( tdb_lsm(pDb) ){
+ ret = ((LsmDb*)pDb)->eMode!=LSMTEST_MODE_SINGLETHREAD;
+ }
+ return ret;
+}
+
+void tdb_lsm_enable_log(TestDb *pDb, int bEnable){
+ lsm_db *db = tdb_lsm(pDb);
+ if( db ){
+ lsm_config_log(db, (bEnable ? xLog : 0), (void *)"client");
+ }
+}
+
+void tdb_lsm_application_crash(TestDb *pDb){
+ if( tdb_lsm(pDb) ){
+ LsmDb *p = (LsmDb *)pDb;
+ p->bCrashed = 1;
+ }
+}
+
+void tdb_lsm_prepare_system_crash(TestDb *pDb){
+ if( tdb_lsm(pDb) ){
+ LsmDb *p = (LsmDb *)pDb;
+ p->bPrepareCrash = 1;
+ }
+}
+
+void tdb_lsm_system_crash(TestDb *pDb){
+ if( tdb_lsm(pDb) ){
+ LsmDb *p = (LsmDb *)pDb;
+ p->bCrashed = 1;
+ doSystemCrash(p);
+ }
+}
+
+void tdb_lsm_safety(TestDb *pDb, int eMode){
+ assert( eMode==LSM_SAFETY_OFF
+ || eMode==LSM_SAFETY_NORMAL
+ || eMode==LSM_SAFETY_FULL
+ );
+ if( tdb_lsm(pDb) ){
+ int iParam = eMode;
+ LsmDb *p = (LsmDb *)pDb;
+ lsm_config(p->db, LSM_CONFIG_SAFETY, &iParam);
+ }
+}
+
+void tdb_lsm_prepare_sync_crash(TestDb *pDb, int iSync){
+ assert( iSync>0 );
+ if( tdb_lsm(pDb) ){
+ LsmDb *p = (LsmDb *)pDb;
+ p->nAutoCrash = iSync;
+ p->bPrepareCrash = 1;
+ }
+}
+
+void tdb_lsm_config_work_hook(
+ TestDb *pDb,
+ void (*xWork)(lsm_db *, void *),
+ void *pWorkCtx
+){
+ if( tdb_lsm(pDb) ){
+ LsmDb *p = (LsmDb *)pDb;
+ p->xWork = xWork;
+ p->pWorkCtx = pWorkCtx;
+ }
+}
+
+void tdb_lsm_write_hook(
+ TestDb *pDb,
+ void (*xWrite)(void *, int, lsm_i64, int, int),
+ void *pWriteCtx
+){
+ if( tdb_lsm(pDb) ){
+ LsmDb *p = (LsmDb *)pDb;
+ p->xWriteHook = xWrite;
+ p->pWriteCtx = pWriteCtx;
+ }
+}
+
+int tdb_lsm_open(const char *zCfg, const char *zDb, int bClear, TestDb **ppDb){
+ return testLsmOpen(zCfg, zDb, bClear, ppDb);
+}
+
+#ifdef LSM_MUTEX_PTHREADS
+
+/*
+** Signal worker thread iWorker that there may be work to do.
+*/
+static void mt_signal_worker(LsmDb *pDb, int iWorker){
+ LsmWorker *p = &pDb->aWorker[iWorker];
+ pthread_mutex_lock(&p->worker_mutex);
+ p->bDoWork = 1;
+ pthread_cond_signal(&p->worker_cond);
+ pthread_mutex_unlock(&p->worker_mutex);
+}
+
+/*
+** This routine is used as the main() for all worker threads.
+*/
+static void *worker_main(void *pArg){
+ LsmWorker *p = (LsmWorker *)pArg;
+ lsm_db *pWorker; /* Connection to access db through */
+
+ pthread_mutex_lock(&p->worker_mutex);
+ while( (pWorker = p->pWorker) ){
+ int rc = LSM_OK;
+
+ /* Do some work. If an error occurs, exit. */
+
+ pthread_mutex_unlock(&p->worker_mutex);
+ if( p->eType==LSMTEST_THREAD_CKPT ){
+ int nKB = 0;
+ rc = lsm_info(pWorker, LSM_INFO_CHECKPOINT_SIZE, &nKB);
+ if( rc==LSM_OK && nKB>=p->pDb->nMtMinCkpt ){
+ rc = lsm_checkpoint(pWorker, 0);
+ }
+ }else{
+ int nWrite;
+ do {
+
+ if( p->eType==LSMTEST_THREAD_WORKER ){
+ waitOnCheckpointer(p->pDb, pWorker);
+ }
+
+ nWrite = 0;
+ rc = lsm_work(pWorker, 0, 256, &nWrite);
+
+ if( p->eType==LSMTEST_THREAD_WORKER && nWrite ){
+ mt_signal_worker(p->pDb, 1);
+ }
+ }while( nWrite && p->pWorker );
+ }
+ pthread_mutex_lock(&p->worker_mutex);
+
+ if( rc!=LSM_OK && rc!=LSM_BUSY ){
+ p->worker_rc = rc;
+ break;
+ }
+
+ /* The thread will wake up when it is signaled either because another
+ ** thread has created some work for this one or because the connection
+ ** is being closed. */
+ if( p->pWorker && p->bDoWork==0 ){
+ pthread_cond_wait(&p->worker_cond, &p->worker_mutex);
+ }
+ p->bDoWork = 0;
+ }
+ pthread_mutex_unlock(&p->worker_mutex);
+
+ return 0;
+}
+
+
+static void mt_stop_worker(LsmDb *pDb, int iWorker){
+ LsmWorker *p = &pDb->aWorker[iWorker];
+ if( p->pWorker ){
+ void *pDummy;
+ lsm_db *pWorker;
+
+ /* Signal the worker to stop */
+ pthread_mutex_lock(&p->worker_mutex);
+ pWorker = p->pWorker;
+ p->pWorker = 0;
+ pthread_cond_signal(&p->worker_cond);
+ pthread_mutex_unlock(&p->worker_mutex);
+
+ /* Join the worker thread. */
+ pthread_join(p->worker_thread, &pDummy);
+
+ /* Free resources allocated in mt_start_worker() */
+ pthread_cond_destroy(&p->worker_cond);
+ pthread_mutex_destroy(&p->worker_mutex);
+ lsm_close(pWorker);
+ }
+}
+
+static void mt_shutdown(LsmDb *pDb){
+ int i;
+ for(i=0; i<pDb->nWorker; i++){
+ mt_stop_worker(pDb, i);
+ }
+}
+
+/*
+** This callback is invoked by LSM when the client database writes to
+** the database file (i.e. to flush the contents of the in-memory tree).
+** This implies there may be work to do on the database, so signal
+** the worker threads.
+*/
+static void mt_client_work_hook(lsm_db *db, void *pArg){
+ LsmDb *pDb = (LsmDb *)pArg; /* LsmDb database handle */
+
+ /* Invoke the user level work-hook, if any. */
+ if( pDb->xWork ) pDb->xWork(db, pDb->pWorkCtx);
+
+ /* Wake up worker thread 0. */
+ mt_signal_worker(pDb, 0);
+}
+
+static void mt_worker_work_hook(lsm_db *db, void *pArg){
+ LsmDb *pDb = (LsmDb *)pArg; /* LsmDb database handle */
+
+ /* Invoke the user level work-hook, if any. */
+ if( pDb->xWork ) pDb->xWork(db, pDb->pWorkCtx);
+}
+
+/*
+** Launch worker thread iWorker for database connection pDb.
+*/
+static int mt_start_worker(
+ LsmDb *pDb, /* Main database structure */
+ int iWorker, /* Worker number to start */
+ const char *zFilename, /* File name of database to open */
+ const char *zCfg, /* Connection configuration string */
+ int eType /* Type of worker thread */
+){
+ int rc = 0; /* Return code */
+ LsmWorker *p; /* Object to initialize */
+
+ assert( iWorker<pDb->nWorker );
+ assert( eType==LSMTEST_THREAD_CKPT
+ || eType==LSMTEST_THREAD_WORKER
+ || eType==LSMTEST_THREAD_WORKER_AC
+ );
+
+ p = &pDb->aWorker[iWorker];
+ p->eType = eType;
+ p->pDb = pDb;
+
+ /* Open the worker connection */
+ if( rc==0 ) rc = lsm_new(&pDb->env, &p->pWorker);
+ if( zCfg ){
+ test_lsm_config_str(pDb, p->pWorker, 1, zCfg, 0);
+ }
+ if( rc==0 ) rc = lsm_open(p->pWorker, zFilename);
+ lsm_config_log(p->pWorker, xLog, (void *)"worker");
+
+ /* Configure the work-hook */
+ if( rc==0 ){
+ lsm_config_work_hook(p->pWorker, mt_worker_work_hook, (void *)pDb);
+ }
+
+ if( eType==LSMTEST_THREAD_WORKER ){
+ test_lsm_config_str(0, p->pWorker, 1, "autocheckpoint=0", 0);
+ }
+
+ /* Kick off the worker thread. */
+ if( rc==0 ) rc = pthread_cond_init(&p->worker_cond, 0);
+ if( rc==0 ) rc = pthread_mutex_init(&p->worker_mutex, 0);
+ if( rc==0 ) rc = pthread_create(&p->worker_thread, 0, worker_main, (void *)p);
+
+ return rc;
+}
+
+
+static int testLsmStartWorkers(
+ LsmDb *pDb, int eModel, const char *zFilename, const char *zCfg
+){
+ int rc;
+
+ if( eModel<1 || eModel>4 ) return 1;
+ if( eModel==1 ) return 0;
+
+ /* Configure a work-hook for the client connection. Worker 0 is signalled
+ ** every time the users connection writes to the database. */
+ lsm_config_work_hook(pDb->db, mt_client_work_hook, (void *)pDb);
+
+ /* Allocate space for two worker connections. They may not both be
+ ** used, but both are allocated. */
+ pDb->aWorker = (LsmWorker *)testMalloc(sizeof(LsmWorker) * 2);
+ memset(pDb->aWorker, 0, sizeof(LsmWorker) * 2);
+
+ switch( eModel ){
+ case LSMTEST_MODE_BACKGROUND_CKPT:
+ pDb->nWorker = 1;
+ test_lsm_config_str(0, pDb->db, 0, "autocheckpoint=0", 0);
+ rc = mt_start_worker(pDb, 0, zFilename, zCfg, LSMTEST_THREAD_CKPT);
+ break;
+
+ case LSMTEST_MODE_BACKGROUND_WORK:
+ pDb->nWorker = 1;
+ test_lsm_config_str(0, pDb->db, 0, "autowork=0", 0);
+ rc = mt_start_worker(pDb, 0, zFilename, zCfg, LSMTEST_THREAD_WORKER_AC);
+ break;
+
+ case LSMTEST_MODE_BACKGROUND_BOTH:
+ pDb->nWorker = 2;
+ test_lsm_config_str(0, pDb->db, 0, "autowork=0", 0);
+ rc = mt_start_worker(pDb, 0, zFilename, zCfg, LSMTEST_THREAD_WORKER);
+ if( rc==0 ){
+ rc = mt_start_worker(pDb, 1, zFilename, zCfg, LSMTEST_THREAD_CKPT);
+ }
+ break;
+ }
+
+ return rc;
+}
+
+
+int test_lsm_mt2(
+ const char *zSpec,
+ const char *zFilename,
+ int bClear,
+ TestDb **ppDb
+){
+ const char *zCfg = "mt_mode=2";
+ return testLsmOpen(zCfg, zFilename, bClear, ppDb);
+}
+
+int test_lsm_mt3(
+ const char *zSpec,
+ const char *zFilename,
+ int bClear,
+ TestDb **ppDb
+){
+ const char *zCfg = "mt_mode=4";
+ return testLsmOpen(zCfg, zFilename, bClear, ppDb);
+}
+
+#else
+static void mt_shutdown(LsmDb *pDb) {
+ unused_parameter(pDb);
+}
+int test_lsm_mt(const char *zFilename, int bClear, TestDb **ppDb){
+ unused_parameter(zFilename);
+ unused_parameter(bClear);
+ unused_parameter(ppDb);
+ testPrintError("threads unavailable - recompile with LSM_MUTEX_PTHREADS\n");
+ return 1;
+}
+#endif
diff --git a/ext/lsm1/lsm-test/lsmtest_tdb4.c b/ext/lsm1/lsm-test/lsmtest_tdb4.c
new file mode 100644
index 0000000..1f92928
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_tdb4.c
@@ -0,0 +1,980 @@
+
+/*
+** This file contains the TestDb bt wrapper.
+*/
+
+#include "lsmtest_tdb.h"
+#include "lsmtest.h"
+#include <unistd.h>
+#include "bt.h"
+
+#include <pthread.h>
+
+typedef struct BtDb BtDb;
+typedef struct BtFile BtFile;
+
+/* Background checkpointer interface (see implementations below). */
+typedef struct bt_ckpter bt_ckpter;
+static int bgc_attach(BtDb *pDb, const char*);
+static int bgc_detach(BtDb *pDb);
+
+/*
+** Each database or log file opened by a database handle is wrapped by
+** an object of the following type.
+*/
+struct BtFile {
+ BtDb *pBt; /* Database handle that opened this file */
+ bt_env *pVfs; /* Underlying VFS */
+ bt_file *pFile; /* File handle belonging to underlying VFS */
+ int nSectorSize; /* Size of sectors in bytes */
+ int nSector; /* Allocated size of nSector array */
+ u8 **apSector; /* Original sector data */
+};
+
+/*
+** nCrashSync:
+** If this value is non-zero, then a "crash-test" is running. If
+** nCrashSync==1, then the crash is simulated during the very next
+** call to the xSync() VFS method (on either the db or log file).
+** If nCrashSync==2, the following call to xSync(), and so on.
+**
+** bCrash:
+** After a crash is simulated, this variable is set. Any subsequent
+** attempts to write to a file or modify the file system in any way
+** fail once this is set. All the caller can do is close the connection.
+**
+** bFastInsert:
+** If this variable is set to true, then a BT_CONTROL_FAST_INSERT_OP
+** control is issued before each callto BtReplace() or BtCsrOpen().
+*/
+struct BtDb {
+ TestDb base; /* Base class */
+ bt_db *pBt; /* bt database handle */
+ sqlite4_env *pEnv; /* SQLite environment (for malloc/free) */
+ bt_env *pVfs; /* Underlying VFS */
+ int bFastInsert; /* True to use fast-insert */
+
+ /* Space for bt_fetch() results */
+ u8 *aBuffer; /* Space to store results */
+ int nBuffer; /* Allocated size of aBuffer[] in bytes */
+ int nRef;
+
+ /* Background checkpointer used by mt connections */
+ bt_ckpter *pCkpter;
+
+ /* Stuff used for crash test simulation */
+ BtFile *apFile[2]; /* Database and log files used by pBt */
+ bt_env env; /* Private VFS for this object */
+ int nCrashSync; /* Number of syncs until crash (see above) */
+ int bCrash; /* True once a crash has been simulated */
+};
+
+static int btVfsFullpath(
+ sqlite4_env *pEnv,
+ bt_env *pVfs,
+ const char *z,
+ char **pzOut
+){
+ BtDb *pBt = (BtDb*)pVfs->pVfsCtx;
+ if( pBt->bCrash ) return SQLITE4_IOERR;
+ return pBt->pVfs->xFullpath(pEnv, pBt->pVfs, z, pzOut);
+}
+
+static int btVfsOpen(
+ sqlite4_env *pEnv,
+ bt_env *pVfs,
+ const char *zFile,
+ int flags, bt_file **ppFile
+){
+ BtFile *p;
+ BtDb *pBt = (BtDb*)pVfs->pVfsCtx;
+ int rc;
+
+ if( pBt->bCrash ) return SQLITE4_IOERR;
+
+ p = (BtFile*)testMalloc(sizeof(BtFile));
+ if( !p ) return SQLITE4_NOMEM;
+ if( flags & BT_OPEN_DATABASE ){
+ pBt->apFile[0] = p;
+ }else if( flags & BT_OPEN_LOG ){
+ pBt->apFile[1] = p;
+ }
+ if( (flags & BT_OPEN_SHARED)==0 ){
+ p->pBt = pBt;
+ }
+ p->pVfs = pBt->pVfs;
+
+ rc = pBt->pVfs->xOpen(pEnv, pVfs, zFile, flags, &p->pFile);
+ if( rc!=SQLITE4_OK ){
+ testFree(p);
+ p = 0;
+ }else{
+ pBt->nRef++;
+ }
+
+ *ppFile = (bt_file*)p;
+ return rc;
+}
+
+static int btVfsSize(bt_file *pFile, sqlite4_int64 *piRes){
+ BtFile *p = (BtFile*)pFile;
+ if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
+ return p->pVfs->xSize(p->pFile, piRes);
+}
+
+static int btVfsRead(bt_file *pFile, sqlite4_int64 iOff, void *pBuf, int nBuf){
+ BtFile *p = (BtFile*)pFile;
+ if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
+ return p->pVfs->xRead(p->pFile, iOff, pBuf, nBuf);
+}
+
+static int btFlushSectors(BtFile *p, int iFile){
+ sqlite4_int64 iSz;
+ int rc;
+ int i;
+ u8 *aTmp = 0;
+
+ rc = p->pBt->pVfs->xSize(p->pFile, &iSz);
+ for(i=0; rc==SQLITE4_OK && i<p->nSector; i++){
+ if( p->pBt->bCrash && p->apSector[i] ){
+
+ /* The system is simulating a crash. There are three choices for
+ ** this sector:
+ **
+ ** 1) Leave it as it is (simulating a successful write),
+ ** 2) Restore the original data (simulating a lost write),
+ ** 3) Populate the disk sector with garbage data.
+ */
+ sqlite4_int64 iSOff = p->nSectorSize*i;
+ int nWrite = MIN(p->nSectorSize, iSz - iSOff);
+
+ if( nWrite ){
+ u8 *aWrite = 0;
+ int iOpt = (testPrngValue(i) % 3) + 1;
+ if( iOpt==1 ){
+ aWrite = p->apSector[i];
+ }else if( iOpt==3 ){
+ if( aTmp==0 ) aTmp = testMalloc(p->nSectorSize);
+ aWrite = aTmp;
+ testPrngArray(i*13, (u32*)aWrite, nWrite/sizeof(u32));
+ }
+
+#if 0
+fprintf(stderr, "handle sector %d of %s with %s\n", i,
+ iFile==0 ? "db" : "log",
+ iOpt==1 ? "rollback" : iOpt==2 ? "write" : "omit"
+);
+fflush(stderr);
+#endif
+
+ if( aWrite ){
+ rc = p->pBt->pVfs->xWrite(p->pFile, iSOff, aWrite, nWrite);
+ }
+ }
+ }
+ testFree(p->apSector[i]);
+ p->apSector[i] = 0;
+ }
+
+ testFree(aTmp);
+ return rc;
+}
+
+static int btSaveSectors(BtFile *p, sqlite4_int64 iOff, int nBuf){
+ int rc;
+ sqlite4_int64 iSz; /* Size of file on disk */
+ int iFirst; /* First sector affected */
+ int iSector; /* Current sector */
+ int iLast; /* Last sector affected */
+
+ if( p->nSectorSize==0 ){
+ p->nSectorSize = p->pBt->pVfs->xSectorSize(p->pFile);
+ if( p->nSectorSize<512 ) p->nSectorSize = 512;
+ }
+ iLast = (iOff+nBuf-1) / p->nSectorSize;
+ iFirst = iOff / p->nSectorSize;
+
+ rc = p->pBt->pVfs->xSize(p->pFile, &iSz);
+ for(iSector=iFirst; rc==SQLITE4_OK && iSector<=iLast; iSector++){
+ int nRead;
+ sqlite4_int64 iSOff = iSector * p->nSectorSize;
+ u8 *aBuf = testMalloc(p->nSectorSize);
+ nRead = MIN(p->nSectorSize, (iSz - iSOff));
+ if( nRead>0 ){
+ rc = p->pBt->pVfs->xRead(p->pFile, iSOff, aBuf, nRead);
+ }
+
+ while( rc==SQLITE4_OK && iSector>=p->nSector ){
+ int nNew = p->nSector + 32;
+ u8 **apNew = (u8**)testMalloc(nNew * sizeof(u8*));
+ memcpy(apNew, p->apSector, p->nSector*sizeof(u8*));
+ testFree(p->apSector);
+ p->apSector = apNew;
+ p->nSector = nNew;
+ }
+
+ p->apSector[iSector] = aBuf;
+ }
+
+ return rc;
+}
+
+static int btVfsWrite(bt_file *pFile, sqlite4_int64 iOff, void *pBuf, int nBuf){
+ BtFile *p = (BtFile*)pFile;
+ if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
+ if( p->pBt && p->pBt->nCrashSync ){
+ btSaveSectors(p, iOff, nBuf);
+ }
+ return p->pVfs->xWrite(p->pFile, iOff, pBuf, nBuf);
+}
+
+static int btVfsTruncate(bt_file *pFile, sqlite4_int64 iOff){
+ BtFile *p = (BtFile*)pFile;
+ if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
+ return p->pVfs->xTruncate(p->pFile, iOff);
+}
+
+static int btVfsSync(bt_file *pFile){
+ int rc = SQLITE4_OK;
+ BtFile *p = (BtFile*)pFile;
+ BtDb *pBt = p->pBt;
+
+ if( pBt ){
+ if( pBt->bCrash ) return SQLITE4_IOERR;
+ if( pBt->nCrashSync ){
+ pBt->nCrashSync--;
+ pBt->bCrash = (pBt->nCrashSync==0);
+ if( pBt->bCrash ){
+ btFlushSectors(pBt->apFile[0], 0);
+ btFlushSectors(pBt->apFile[1], 1);
+ rc = SQLITE4_IOERR;
+ }else{
+ btFlushSectors(p, 0);
+ }
+ }
+ }
+
+ if( rc==SQLITE4_OK ){
+ rc = p->pVfs->xSync(p->pFile);
+ }
+ return rc;
+}
+
+static int btVfsSectorSize(bt_file *pFile){
+ BtFile *p = (BtFile*)pFile;
+ return p->pVfs->xSectorSize(p->pFile);
+}
+
+static void btDeref(BtDb *p){
+ p->nRef--;
+ assert( p->nRef>=0 );
+ if( p->nRef<=0 ) testFree(p);
+}
+
+static int btVfsClose(bt_file *pFile){
+ BtFile *p = (BtFile*)pFile;
+ BtDb *pBt = p->pBt;
+ int rc;
+ if( pBt ){
+ btFlushSectors(p, 0);
+ if( p==pBt->apFile[0] ) pBt->apFile[0] = 0;
+ if( p==pBt->apFile[1] ) pBt->apFile[1] = 0;
+ }
+ testFree(p->apSector);
+ rc = p->pVfs->xClose(p->pFile);
+#if 0
+ btDeref(p->pBt);
+#endif
+ testFree(p);
+ return rc;
+}
+
+static int btVfsUnlink(sqlite4_env *pEnv, bt_env *pVfs, const char *zFile){
+ BtDb *pBt = (BtDb*)pVfs->pVfsCtx;
+ if( pBt->bCrash ) return SQLITE4_IOERR;
+ return pBt->pVfs->xUnlink(pEnv, pBt->pVfs, zFile);
+}
+
+static int btVfsLock(bt_file *pFile, int iLock, int eType){
+ BtFile *p = (BtFile*)pFile;
+ if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
+ return p->pVfs->xLock(p->pFile, iLock, eType);
+}
+
+static int btVfsTestLock(bt_file *pFile, int iLock, int nLock, int eType){
+ BtFile *p = (BtFile*)pFile;
+ if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
+ return p->pVfs->xTestLock(p->pFile, iLock, nLock, eType);
+}
+
+static int btVfsShmMap(bt_file *pFile, int iChunk, int sz, void **ppOut){
+ BtFile *p = (BtFile*)pFile;
+ if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
+ return p->pVfs->xShmMap(p->pFile, iChunk, sz, ppOut);
+}
+
+static void btVfsShmBarrier(bt_file *pFile){
+ BtFile *p = (BtFile*)pFile;
+ return p->pVfs->xShmBarrier(p->pFile);
+}
+
+static int btVfsShmUnmap(bt_file *pFile, int bDelete){
+ BtFile *p = (BtFile*)pFile;
+ if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
+ return p->pVfs->xShmUnmap(p->pFile, bDelete);
+}
+
+static int bt_close(TestDb *pTestDb){
+ BtDb *p = (BtDb*)pTestDb;
+ int rc = sqlite4BtClose(p->pBt);
+ free(p->aBuffer);
+ if( p->apFile[0] ) p->apFile[0]->pBt = 0;
+ if( p->apFile[1] ) p->apFile[1]->pBt = 0;
+ bgc_detach(p);
+ testFree(p);
+ return rc;
+}
+
+static int btMinTransaction(BtDb *p, int iMin, int *piLevel){
+ int iLevel;
+ int rc = SQLITE4_OK;
+
+ iLevel = sqlite4BtTransactionLevel(p->pBt);
+ if( iLevel<iMin ){
+ rc = sqlite4BtBegin(p->pBt, iMin);
+ *piLevel = iLevel;
+ }else{
+ *piLevel = -1;
+ }
+
+ return rc;
+}
+static int btRestoreTransaction(BtDb *p, int iLevel, int rcin){
+ int rc = rcin;
+ if( iLevel>=0 ){
+ if( rc==SQLITE4_OK ){
+ rc = sqlite4BtCommit(p->pBt, iLevel);
+ }else{
+ sqlite4BtRollback(p->pBt, iLevel);
+ }
+ assert( iLevel==sqlite4BtTransactionLevel(p->pBt) );
+ }
+ return rc;
+}
+
+static int bt_write(TestDb *pTestDb, void *pK, int nK, void *pV, int nV){
+ BtDb *p = (BtDb*)pTestDb;
+ int iLevel;
+ int rc;
+
+ rc = btMinTransaction(p, 2, &iLevel);
+ if( rc==SQLITE4_OK ){
+ if( p->bFastInsert ) sqlite4BtControl(p->pBt, BT_CONTROL_FAST_INSERT_OP, 0);
+ rc = sqlite4BtReplace(p->pBt, pK, nK, pV, nV);
+ rc = btRestoreTransaction(p, iLevel, rc);
+ }
+ return rc;
+}
+
+static int bt_delete(TestDb *pTestDb, void *pK, int nK){
+ return bt_write(pTestDb, pK, nK, 0, -1);
+}
+
+static int bt_delete_range(
+ TestDb *pTestDb,
+ void *pKey1, int nKey1,
+ void *pKey2, int nKey2
+){
+ BtDb *p = (BtDb*)pTestDb;
+ bt_cursor *pCsr = 0;
+ int rc = SQLITE4_OK;
+ int iLevel;
+
+ rc = btMinTransaction(p, 2, &iLevel);
+ if( rc==SQLITE4_OK ){
+ if( p->bFastInsert ) sqlite4BtControl(p->pBt, BT_CONTROL_FAST_INSERT_OP, 0);
+ rc = sqlite4BtCsrOpen(p->pBt, 0, &pCsr);
+ }
+ while( rc==SQLITE4_OK ){
+ const void *pK;
+ int n;
+ int nCmp;
+ int res;
+
+ rc = sqlite4BtCsrSeek(pCsr, pKey1, nKey1, BT_SEEK_GE);
+ if( rc==SQLITE4_INEXACT ) rc = SQLITE4_OK;
+ if( rc!=SQLITE4_OK ) break;
+
+ rc = sqlite4BtCsrKey(pCsr, &pK, &n);
+ if( rc!=SQLITE4_OK ) break;
+
+ nCmp = MIN(n, nKey1);
+ res = memcmp(pKey1, pK, nCmp);
+ assert( res<0 || (res==0 && nKey1<=n) );
+ if( res==0 && nKey1==n ){
+ rc = sqlite4BtCsrNext(pCsr);
+ if( rc!=SQLITE4_OK ) break;
+ rc = sqlite4BtCsrKey(pCsr, &pK, &n);
+ if( rc!=SQLITE4_OK ) break;
+ }
+
+ nCmp = MIN(n, nKey2);
+ res = memcmp(pKey2, pK, nCmp);
+ if( res<0 || (res==0 && nKey2<=n) ) break;
+
+ rc = sqlite4BtDelete(pCsr);
+ }
+ if( rc==SQLITE4_NOTFOUND ) rc = SQLITE4_OK;
+
+ sqlite4BtCsrClose(pCsr);
+
+ rc = btRestoreTransaction(p, iLevel, rc);
+ return rc;
+}
+
+static int bt_fetch(
+ TestDb *pTestDb,
+ void *pK, int nK,
+ void **ppVal, int *pnVal
+){
+ BtDb *p = (BtDb*)pTestDb;
+ bt_cursor *pCsr = 0;
+ int iLevel;
+ int rc = SQLITE4_OK;
+
+ iLevel = sqlite4BtTransactionLevel(p->pBt);
+ if( iLevel==0 ){
+ rc = sqlite4BtBegin(p->pBt, 1);
+ if( rc!=SQLITE4_OK ) return rc;
+ }
+
+ if( p->bFastInsert ) sqlite4BtControl(p->pBt, BT_CONTROL_FAST_INSERT_OP, 0);
+ rc = sqlite4BtCsrOpen(p->pBt, 0, &pCsr);
+ if( rc==SQLITE4_OK ){
+ rc = sqlite4BtCsrSeek(pCsr, pK, nK, BT_SEEK_EQ);
+ if( rc==SQLITE4_OK ){
+ const void *pV = 0;
+ int nV = 0;
+ rc = sqlite4BtCsrData(pCsr, 0, -1, &pV, &nV);
+ if( rc==SQLITE4_OK ){
+ if( nV>p->nBuffer ){
+ free(p->aBuffer);
+ p->aBuffer = (u8*)malloc(nV*2);
+ p->nBuffer = nV*2;
+ }
+ memcpy(p->aBuffer, pV, nV);
+ *pnVal = nV;
+ *ppVal = (void*)(p->aBuffer);
+ }
+
+ }else if( rc==SQLITE4_INEXACT || rc==SQLITE4_NOTFOUND ){
+ *ppVal = 0;
+ *pnVal = -1;
+ rc = SQLITE4_OK;
+ }
+ sqlite4BtCsrClose(pCsr);
+ }
+
+ if( iLevel==0 ) sqlite4BtCommit(p->pBt, 0);
+ return rc;
+}
+
+static int bt_scan(
+ TestDb *pTestDb,
+ void *pCtx,
+ int bReverse,
+ void *pFirst, int nFirst,
+ void *pLast, int nLast,
+ void (*xCallback)(void *, void *, int , void *, int)
+){
+ BtDb *p = (BtDb*)pTestDb;
+ bt_cursor *pCsr = 0;
+ int rc;
+ int iLevel;
+
+ rc = btMinTransaction(p, 1, &iLevel);
+
+ if( rc==SQLITE4_OK ){
+ if( p->bFastInsert ) sqlite4BtControl(p->pBt, BT_CONTROL_FAST_INSERT_OP, 0);
+ rc = sqlite4BtCsrOpen(p->pBt, 0, &pCsr);
+ }
+ if( rc==SQLITE4_OK ){
+ if( bReverse ){
+ if( pLast ){
+ rc = sqlite4BtCsrSeek(pCsr, pLast, nLast, BT_SEEK_LE);
+ }else{
+ rc = sqlite4BtCsrLast(pCsr);
+ }
+ }else{
+ rc = sqlite4BtCsrSeek(pCsr, pFirst, nFirst, BT_SEEK_GE);
+ }
+ if( rc==SQLITE4_INEXACT ) rc = SQLITE4_OK;
+
+ while( rc==SQLITE4_OK ){
+ const void *pK = 0; int nK = 0;
+ const void *pV = 0; int nV = 0;
+
+ rc = sqlite4BtCsrKey(pCsr, &pK, &nK);
+ if( rc==SQLITE4_OK ){
+ rc = sqlite4BtCsrData(pCsr, 0, -1, &pV, &nV);
+ }
+
+ if( rc!=SQLITE4_OK ) break;
+ if( bReverse ){
+ if( pFirst ){
+ int res;
+ int nCmp = MIN(nK, nFirst);
+ res = memcmp(pFirst, pK, nCmp);
+ if( res>0 || (res==0 && nK<nFirst) ) break;
+ }
+ }else{
+ if( pLast ){
+ int res;
+ int nCmp = MIN(nK, nLast);
+ res = memcmp(pLast, pK, nCmp);
+ if( res<0 || (res==0 && nK>nLast) ) break;
+ }
+ }
+
+ xCallback(pCtx, (void*)pK, nK, (void*)pV, nV);
+ if( bReverse ){
+ rc = sqlite4BtCsrPrev(pCsr);
+ }else{
+ rc = sqlite4BtCsrNext(pCsr);
+ }
+ }
+ if( rc==SQLITE4_NOTFOUND ) rc = SQLITE4_OK;
+
+ sqlite4BtCsrClose(pCsr);
+ }
+
+ rc = btRestoreTransaction(p, iLevel, rc);
+ return rc;
+}
+
+static int bt_begin(TestDb *pTestDb, int iLvl){
+ BtDb *p = (BtDb*)pTestDb;
+ int rc = sqlite4BtBegin(p->pBt, iLvl);
+ return rc;
+}
+
+static int bt_commit(TestDb *pTestDb, int iLvl){
+ BtDb *p = (BtDb*)pTestDb;
+ int rc = sqlite4BtCommit(p->pBt, iLvl);
+ return rc;
+}
+
+static int bt_rollback(TestDb *pTestDb, int iLvl){
+ BtDb *p = (BtDb*)pTestDb;
+ int rc = sqlite4BtRollback(p->pBt, iLvl);
+ return rc;
+}
+
+static int testParseOption(
+ const char **pzIn, /* IN/OUT: pointer to next option */
+ const char **pzOpt, /* OUT: nul-terminated option name */
+ const char **pzArg, /* OUT: nul-terminated option argument */
+ char *pSpace /* Temporary space for output params */
+){
+ const char *p = *pzIn;
+ const char *pStart;
+ int n;
+
+ char *pOut = pSpace;
+
+ while( *p==' ' ) p++;
+ pStart = p;
+ while( *p && *p!='=' ) p++;
+ if( *p==0 ) return 1;
+
+ n = (p - pStart);
+ memcpy(pOut, pStart, n);
+ *pzOpt = pOut;
+ pOut += n;
+ *pOut++ = '\0';
+
+ p++;
+ pStart = p;
+ while( *p && *p!=' ' ) p++;
+ n = (p - pStart);
+
+ memcpy(pOut, pStart, n);
+ *pzArg = pOut;
+ pOut += n;
+ *pOut++ = '\0';
+
+ *pzIn = p;
+ return 0;
+}
+
+static int testParseInt(const char *z, int *piVal){
+ int i = 0;
+ const char *p = z;
+
+ while( *p>='0' && *p<='9' ){
+ i = i*10 + (*p - '0');
+ p++;
+ }
+ if( *p=='K' || *p=='k' ){
+ i = i * 1024;
+ p++;
+ }else if( *p=='M' || *p=='m' ){
+ i = i * 1024 * 1024;
+ p++;
+ }
+
+ if( *p ) return SQLITE4_ERROR;
+ *piVal = i;
+ return SQLITE4_OK;
+}
+
+static int testBtConfigure(BtDb *pDb, const char *zCfg, int *pbMt){
+ int rc = SQLITE4_OK;
+
+ if( zCfg ){
+ struct CfgParam {
+ const char *zParam;
+ int eParam;
+ } aParam[] = {
+ { "safety", BT_CONTROL_SAFETY },
+ { "autockpt", BT_CONTROL_AUTOCKPT },
+ { "multiproc", BT_CONTROL_MULTIPROC },
+ { "blksz", BT_CONTROL_BLKSZ },
+ { "pagesz", BT_CONTROL_PAGESZ },
+ { "mt", -1 },
+ { "fastinsert", -2 },
+ { 0, 0 }
+ };
+ const char *z = zCfg;
+ int n = strlen(z);
+ char *aSpace;
+ const char *zOpt;
+ const char *zArg;
+
+ aSpace = (char*)testMalloc(n+2);
+ while( rc==SQLITE4_OK && 0==testParseOption(&z, &zOpt, &zArg, aSpace) ){
+ int i;
+ int iVal;
+ rc = testArgSelect(aParam, "param", zOpt, &i);
+ if( rc!=SQLITE4_OK ) break;
+
+ rc = testParseInt(zArg, &iVal);
+ if( rc!=SQLITE4_OK ) break;
+
+ switch( aParam[i].eParam ){
+ case -1:
+ *pbMt = iVal;
+ break;
+ case -2:
+ pDb->bFastInsert = 1;
+ break;
+ default:
+ rc = sqlite4BtControl(pDb->pBt, aParam[i].eParam, (void*)&iVal);
+ break;
+ }
+ }
+ testFree(aSpace);
+ }
+
+ return rc;
+}
+
+
+int test_bt_open(
+ const char *zSpec,
+ const char *zFilename,
+ int bClear,
+ TestDb **ppDb
+){
+
+ static const DatabaseMethods SqlMethods = {
+ bt_close,
+ bt_write,
+ bt_delete,
+ bt_delete_range,
+ bt_fetch,
+ bt_scan,
+ bt_begin,
+ bt_commit,
+ bt_rollback
+ };
+ BtDb *p = 0;
+ bt_db *pBt = 0;
+ int rc;
+ sqlite4_env *pEnv = sqlite4_env_default();
+
+ if( bClear && zFilename && zFilename[0] ){
+ char *zLog = sqlite3_mprintf("%s-wal", zFilename);
+ unlink(zFilename);
+ unlink(zLog);
+ sqlite3_free(zLog);
+ }
+
+ rc = sqlite4BtNew(pEnv, 0, &pBt);
+ if( rc==SQLITE4_OK ){
+ int mt = 0; /* True for multi-threaded connection */
+
+ p = (BtDb*)testMalloc(sizeof(BtDb));
+ p->base.pMethods = &SqlMethods;
+ p->pBt = pBt;
+ p->pEnv = pEnv;
+ p->nRef = 1;
+
+ p->env.pVfsCtx = (void*)p;
+ p->env.xFullpath = btVfsFullpath;
+ p->env.xOpen = btVfsOpen;
+ p->env.xSize = btVfsSize;
+ p->env.xRead = btVfsRead;
+ p->env.xWrite = btVfsWrite;
+ p->env.xTruncate = btVfsTruncate;
+ p->env.xSync = btVfsSync;
+ p->env.xSectorSize = btVfsSectorSize;
+ p->env.xClose = btVfsClose;
+ p->env.xUnlink = btVfsUnlink;
+ p->env.xLock = btVfsLock;
+ p->env.xTestLock = btVfsTestLock;
+ p->env.xShmMap = btVfsShmMap;
+ p->env.xShmBarrier = btVfsShmBarrier;
+ p->env.xShmUnmap = btVfsShmUnmap;
+
+ sqlite4BtControl(pBt, BT_CONTROL_GETVFS, (void*)&p->pVfs);
+ sqlite4BtControl(pBt, BT_CONTROL_SETVFS, (void*)&p->env);
+
+ rc = testBtConfigure(p, zSpec, &mt);
+ if( rc==SQLITE4_OK ){
+ rc = sqlite4BtOpen(pBt, zFilename);
+ }
+
+ if( rc==SQLITE4_OK && mt ){
+ int nAuto = 0;
+ rc = bgc_attach(p, zSpec);
+ sqlite4BtControl(pBt, BT_CONTROL_AUTOCKPT, (void*)&nAuto);
+ }
+ }
+
+ if( rc!=SQLITE4_OK && p ){
+ bt_close(&p->base);
+ }
+
+ *ppDb = &p->base;
+ return rc;
+}
+
+int test_fbt_open(
+ const char *zSpec,
+ const char *zFilename,
+ int bClear,
+ TestDb **ppDb
+){
+ return test_bt_open("fast=1", zFilename, bClear, ppDb);
+}
+
+int test_fbts_open(
+ const char *zSpec,
+ const char *zFilename,
+ int bClear,
+ TestDb **ppDb
+){
+ return test_bt_open("fast=1 blksz=32K pagesz=512", zFilename, bClear, ppDb);
+}
+
+
+void tdb_bt_prepare_sync_crash(TestDb *pTestDb, int iSync){
+ BtDb *p = (BtDb*)pTestDb;
+ assert( pTestDb->pMethods->xClose==bt_close );
+ assert( p->bCrash==0 );
+ p->nCrashSync = iSync;
+}
+
+bt_db *tdb_bt(TestDb *pDb){
+ if( pDb->pMethods->xClose==bt_close ){
+ return ((BtDb *)pDb)->pBt;
+ }
+ return 0;
+}
+
+/*************************************************************************
+** Beginning of code for background checkpointer.
+*/
+
+struct bt_ckpter {
+ sqlite4_buffer file; /* File name */
+ sqlite4_buffer spec; /* Options */
+ int nLogsize; /* Minimum log size to checkpoint */
+ int nRef; /* Number of clients */
+
+ int bDoWork; /* Set by client threads */
+ pthread_t ckpter_thread; /* Checkpointer thread */
+ pthread_cond_t ckpter_cond; /* Condition var the ckpter waits on */
+ pthread_mutex_t ckpter_mutex; /* Mutex used with ckpter_cond */
+
+ bt_ckpter *pNext; /* Next object in list at gBgc.pCkpter */
+};
+
+static struct GlobalBackgroundCheckpointer {
+ bt_ckpter *pCkpter; /* Linked list of checkpointers */
+} gBgc;
+
+static void *bgc_main(void *pArg){
+ BtDb *pDb = 0;
+ int rc;
+ int mt;
+ bt_ckpter *pCkpter = (bt_ckpter*)pArg;
+
+ rc = test_bt_open("", (char*)pCkpter->file.p, 0, (TestDb**)&pDb);
+ assert( rc==SQLITE4_OK );
+ rc = testBtConfigure(pDb, (char*)pCkpter->spec.p, &mt);
+
+ while( pCkpter->nRef>0 ){
+ bt_db *db = pDb->pBt;
+ int nLog = 0;
+
+ sqlite4BtBegin(db, 1);
+ sqlite4BtCommit(db, 0);
+ sqlite4BtControl(db, BT_CONTROL_LOGSIZE, (void*)&nLog);
+
+ if( nLog>=pCkpter->nLogsize ){
+ int rc;
+ bt_checkpoint ckpt;
+ memset(&ckpt, 0, sizeof(bt_checkpoint));
+ ckpt.nFrameBuffer = nLog/2;
+ rc = sqlite4BtControl(db, BT_CONTROL_CHECKPOINT, (void*)&ckpt);
+ assert( rc==SQLITE4_OK );
+ sqlite4BtControl(db, BT_CONTROL_LOGSIZE, (void*)&nLog);
+ }
+
+ /* The thread will wake up when it is signaled either because another
+ ** thread has created some work for this one or because the connection
+ ** is being closed. */
+ pthread_mutex_lock(&pCkpter->ckpter_mutex);
+ if( pCkpter->bDoWork==0 ){
+ pthread_cond_wait(&pCkpter->ckpter_cond, &pCkpter->ckpter_mutex);
+ }
+ pCkpter->bDoWork = 0;
+ pthread_mutex_unlock(&pCkpter->ckpter_mutex);
+ }
+
+ if( pDb ) bt_close((TestDb*)pDb);
+ return 0;
+}
+
+static void bgc_logsize_cb(void *pCtx, int nLogsize){
+ bt_ckpter *p = (bt_ckpter*)pCtx;
+ if( nLogsize>=p->nLogsize ){
+ pthread_mutex_lock(&p->ckpter_mutex);
+ p->bDoWork = 1;
+ pthread_cond_signal(&p->ckpter_cond);
+ pthread_mutex_unlock(&p->ckpter_mutex);
+ }
+}
+
+static int bgc_attach(BtDb *pDb, const char *zSpec){
+ int rc;
+ int n;
+ bt_info info;
+ bt_ckpter *pCkpter;
+
+ /* Figure out the full path to the database opened by handle pDb. */
+ info.eType = BT_INFO_FILENAME;
+ info.pgno = 0;
+ sqlite4_buffer_init(&info.output, 0);
+ rc = sqlite4BtControl(pDb->pBt, BT_CONTROL_INFO, (void*)&info);
+ if( rc!=SQLITE4_OK ) return rc;
+
+ sqlite4_mutex_enter(sqlite4_mutex_alloc(pDb->pEnv, SQLITE4_MUTEX_STATIC_KV));
+
+ /* Search for an existing bt_ckpter object. */
+ n = info.output.n;
+ for(pCkpter=gBgc.pCkpter; pCkpter; pCkpter=pCkpter->pNext){
+ if( n==pCkpter->file.n && 0==memcmp(info.output.p, pCkpter->file.p, n) ){
+ break;
+ }
+ }
+
+ /* Failed to find a suitable checkpointer. Create a new one. */
+ if( pCkpter==0 ){
+ bt_logsizecb cb;
+
+ pCkpter = testMalloc(sizeof(bt_ckpter));
+ memcpy(&pCkpter->file, &info.output, sizeof(sqlite4_buffer));
+ info.output.p = 0;
+ pCkpter->pNext = gBgc.pCkpter;
+ pCkpter->nLogsize = 1000;
+ gBgc.pCkpter = pCkpter;
+ pCkpter->nRef = 1;
+
+ sqlite4_buffer_init(&pCkpter->spec, 0);
+ rc = sqlite4_buffer_set(&pCkpter->spec, zSpec, strlen(zSpec)+1);
+ assert( rc==SQLITE4_OK );
+
+ /* Kick off the checkpointer thread. */
+ if( rc==0 ) rc = pthread_cond_init(&pCkpter->ckpter_cond, 0);
+ if( rc==0 ) rc = pthread_mutex_init(&pCkpter->ckpter_mutex, 0);
+ if( rc==0 ){
+ rc = pthread_create(&pCkpter->ckpter_thread, 0, bgc_main, (void*)pCkpter);
+ }
+ assert( rc==0 ); /* todo: Fix this */
+
+ /* Set up the logsize callback for the client thread */
+ cb.pCtx = (void*)pCkpter;
+ cb.xLogsize = bgc_logsize_cb;
+ sqlite4BtControl(pDb->pBt, BT_CONTROL_LOGSIZECB, (void*)&cb);
+ }else{
+ pCkpter->nRef++;
+ }
+
+ /* Assuming a checkpointer was encountered or effected, attach the
+ ** connection to it. */
+ if( pCkpter ){
+ pDb->pCkpter = pCkpter;
+ }
+
+ sqlite4_mutex_leave(sqlite4_mutex_alloc(pDb->pEnv, SQLITE4_MUTEX_STATIC_KV));
+ sqlite4_buffer_clear(&info.output);
+ return rc;
+}
+
+static int bgc_detach(BtDb *pDb){
+ int rc = SQLITE4_OK;
+ bt_ckpter *pCkpter = pDb->pCkpter;
+ if( pCkpter ){
+ int bShutdown = 0; /* True if this is the last reference */
+
+ sqlite4_mutex_enter(sqlite4_mutex_alloc(pDb->pEnv,SQLITE4_MUTEX_STATIC_KV));
+ pCkpter->nRef--;
+ if( pCkpter->nRef==0 ){
+ bt_ckpter **pp;
+
+ *pp = pCkpter->pNext;
+ for(pp=&gBgc.pCkpter; *pp!=pCkpter; pp=&((*pp)->pNext));
+ bShutdown = 1;
+ }
+ sqlite4_mutex_leave(sqlite4_mutex_alloc(pDb->pEnv,SQLITE4_MUTEX_STATIC_KV));
+
+ if( bShutdown ){
+ void *pDummy;
+
+ /* Signal the checkpointer thread. */
+ pthread_mutex_lock(&pCkpter->ckpter_mutex);
+ pCkpter->bDoWork = 1;
+ pthread_cond_signal(&pCkpter->ckpter_cond);
+ pthread_mutex_unlock(&pCkpter->ckpter_mutex);
+
+ /* Join the checkpointer thread. */
+ pthread_join(pCkpter->ckpter_thread, &pDummy);
+ pthread_cond_destroy(&pCkpter->ckpter_cond);
+ pthread_mutex_destroy(&pCkpter->ckpter_mutex);
+
+ sqlite4_buffer_clear(&pCkpter->file);
+ sqlite4_buffer_clear(&pCkpter->spec);
+ testFree(pCkpter);
+ }
+
+ pDb->pCkpter = 0;
+ }
+ return rc;
+}
+
+/*
+** End of background checkpointer.
+*************************************************************************/
diff --git a/ext/lsm1/lsm-test/lsmtest_util.c b/ext/lsm1/lsm-test/lsmtest_util.c
new file mode 100644
index 0000000..adab8a5
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_util.c
@@ -0,0 +1,223 @@
+
+#include "lsmtest.h"
+#include <stdarg.h>
+#include <stdio.h>
+#include <string.h>
+#ifndef _WIN32
+# include <sys/time.h>
+#endif
+
+/*
+** Global variables used within this module.
+*/
+static struct TestutilGlobal {
+ char **argv;
+ int argc;
+} g = {0, 0};
+
+static struct TestutilRnd {
+ unsigned int aRand1[2048]; /* Bits 0..10 */
+ unsigned int aRand2[2048]; /* Bits 11..21 */
+ unsigned int aRand3[1024]; /* Bits 22..31 */
+} r;
+
+/*************************************************************************
+** The following block is a copy of the implementation of SQLite function
+** sqlite3_randomness. This version has two important differences:
+**
+** 1. It always uses the same seed. So the sequence of random data output
+** is the same for every run of the program.
+**
+** 2. It is not threadsafe.
+*/
+static struct sqlite3PrngType {
+ unsigned char i, j; /* State variables */
+ unsigned char s[256]; /* State variables */
+} sqlite3Prng = {
+ 0xAF, 0x28,
+ {
+ 0x71, 0xF5, 0xB4, 0x6E, 0x80, 0xAB, 0x1D, 0xB8,
+ 0xFB, 0xB7, 0x49, 0xBF, 0xFF, 0x72, 0x2D, 0x14,
+ 0x79, 0x09, 0xE3, 0x78, 0x76, 0xB0, 0x2C, 0x0A,
+ 0x8E, 0x23, 0xEE, 0xDF, 0xE0, 0x9A, 0x2F, 0x67,
+ 0xE1, 0xBE, 0x0E, 0xA7, 0x08, 0x97, 0xEB, 0x77,
+ 0x78, 0xBA, 0x9D, 0xCA, 0x49, 0x4C, 0x60, 0x9A,
+ 0xF6, 0xBD, 0xDA, 0x7F, 0xBC, 0x48, 0x58, 0x52,
+ 0xE5, 0xCD, 0x83, 0x72, 0x23, 0x52, 0xFF, 0x6D,
+ 0xEF, 0x0F, 0x82, 0x29, 0xA0, 0x83, 0x3F, 0x7D,
+ 0xA4, 0x88, 0x31, 0xE7, 0x88, 0x92, 0x3B, 0x9B,
+ 0x3B, 0x2C, 0xC2, 0x4C, 0x71, 0xA2, 0xB0, 0xEA,
+ 0x36, 0xD0, 0x00, 0xF1, 0xD3, 0x39, 0x17, 0x5D,
+ 0x2A, 0x7A, 0xE4, 0xAD, 0xE1, 0x64, 0xCE, 0x0F,
+ 0x9C, 0xD9, 0xF5, 0xED, 0xB0, 0x22, 0x5E, 0x62,
+ 0x97, 0x02, 0xA3, 0x8C, 0x67, 0x80, 0xFC, 0x88,
+ 0x14, 0x0B, 0x15, 0x10, 0x0F, 0xC7, 0x40, 0xD4,
+ 0xF1, 0xF9, 0x0E, 0x1A, 0xCE, 0xB9, 0x1E, 0xA1,
+ 0x72, 0x8E, 0xD7, 0x78, 0x39, 0xCD, 0xF4, 0x5D,
+ 0x2A, 0x59, 0x26, 0x34, 0xF2, 0x73, 0x0B, 0xA0,
+ 0x02, 0x51, 0x2C, 0x03, 0xA3, 0xA7, 0x43, 0x13,
+ 0xE8, 0x98, 0x2B, 0xD2, 0x53, 0xF8, 0xEE, 0x91,
+ 0x7D, 0xE7, 0xE3, 0xDA, 0xD5, 0xBB, 0xC0, 0x92,
+ 0x9D, 0x98, 0x01, 0x2C, 0xF9, 0xB9, 0xA0, 0xEB,
+ 0xCF, 0x32, 0xFA, 0x01, 0x49, 0xA5, 0x1D, 0x9A,
+ 0x76, 0x86, 0x3F, 0x40, 0xD4, 0x89, 0x8F, 0x9C,
+ 0xE2, 0xE3, 0x11, 0x31, 0x37, 0xB2, 0x49, 0x28,
+ 0x35, 0xC0, 0x99, 0xB6, 0xD0, 0xBC, 0x66, 0x35,
+ 0xF7, 0x83, 0x5B, 0xD7, 0x37, 0x1A, 0x2B, 0x18,
+ 0xA6, 0xFF, 0x8D, 0x7C, 0x81, 0xA8, 0xFC, 0x9E,
+ 0xC4, 0xEC, 0x80, 0xD0, 0x98, 0xA7, 0x76, 0xCC,
+ 0x9C, 0x2F, 0x7B, 0xFF, 0x8E, 0x0E, 0xBB, 0x90,
+ 0xAE, 0x13, 0x06, 0xF5, 0x1C, 0x4E, 0x52, 0xF7
+ }
+};
+
+/* Generate and return single random byte */
+static unsigned char randomByte(void){
+ unsigned char t;
+ sqlite3Prng.i++;
+ t = sqlite3Prng.s[sqlite3Prng.i];
+ sqlite3Prng.j += t;
+ sqlite3Prng.s[sqlite3Prng.i] = sqlite3Prng.s[sqlite3Prng.j];
+ sqlite3Prng.s[sqlite3Prng.j] = t;
+ t += sqlite3Prng.s[sqlite3Prng.i];
+ return sqlite3Prng.s[t];
+}
+
+/*
+** Return N random bytes.
+*/
+static void randomBlob(int nBuf, unsigned char *zBuf){
+ int i;
+ for(i=0; i<nBuf; i++){
+ zBuf[i] = randomByte();
+ }
+}
+/*
+** End of code copied from SQLite.
+*************************************************************************/
+
+
+int testPrngInit(void){
+ randomBlob(sizeof(r.aRand1), (unsigned char *)r.aRand1);
+ randomBlob(sizeof(r.aRand2), (unsigned char *)r.aRand2);
+ randomBlob(sizeof(r.aRand3), (unsigned char *)r.aRand3);
+ return 0;
+}
+
+unsigned int testPrngValue(unsigned int iVal){
+ return
+ r.aRand1[iVal & 0x000007FF] ^
+ r.aRand2[(iVal>>11) & 0x000007FF] ^
+ r.aRand3[(iVal>>22) & 0x000003FF]
+ ;
+}
+
+void testPrngArray(unsigned int iVal, unsigned int *aOut, int nOut){
+ int i;
+ for(i=0; i<nOut; i++){
+ aOut[i] = testPrngValue(iVal+i);
+ }
+}
+
+void testPrngString(unsigned int iVal, char *aOut, int nOut){
+ int i;
+ for(i=0; i<(nOut-1); i++){
+ aOut[i] = 'a' + (testPrngValue(iVal+i) % 26);
+ }
+ aOut[i] = '\0';
+}
+
+void testErrorInit(int argc, char **argv){
+ g.argc = argc;
+ g.argv = argv;
+}
+
+void testPrintError(const char *zFormat, ...){
+ va_list ap;
+ va_start(ap, zFormat);
+ vfprintf(stderr, zFormat, ap);
+ va_end(ap);
+}
+
+void testPrintFUsage(const char *zFormat, ...){
+ va_list ap;
+ va_start(ap, zFormat);
+ fprintf(stderr, "Usage: %s %s ", g.argv[0], g.argv[1]);
+ vfprintf(stderr, zFormat, ap);
+ fprintf(stderr, "\n");
+ va_end(ap);
+}
+
+void testPrintUsage(const char *zArgs){
+ testPrintError("Usage: %s %s %s\n", g.argv[0], g.argv[1], zArgs);
+}
+
+
+static void argError(void *aData, const char *zType, int sz, const char *zArg){
+ struct Entry { const char *zName; };
+ struct Entry *pEntry;
+ const char *zPrev = 0;
+
+ testPrintError("unrecognized %s \"%s\": must be ", zType, zArg);
+ for(pEntry=(struct Entry *)aData;
+ pEntry->zName;
+ pEntry=(struct Entry *)&((unsigned char *)pEntry)[sz]
+ ){
+ if( zPrev ){ testPrintError("%s, ", zPrev); }
+ zPrev = pEntry->zName;
+ }
+ testPrintError("or %s\n", zPrev);
+}
+
+int testArgSelectX(
+ void *aData,
+ const char *zType,
+ int sz,
+ const char *zArg,
+ int *piOut
+){
+ struct Entry { const char *zName; };
+ struct Entry *pEntry;
+ int nArg = strlen(zArg);
+
+ int i = 0;
+ int iOut = -1;
+ int nOut = 0;
+
+ for(pEntry=(struct Entry *)aData;
+ pEntry->zName;
+ pEntry=(struct Entry *)&((unsigned char *)pEntry)[sz]
+ ){
+ int nName = strlen(pEntry->zName);
+ if( nArg<=nName && memcmp(pEntry->zName, zArg, nArg)==0 ){
+ iOut = i;
+ if( nName==nArg ){
+ nOut = 1;
+ break;
+ }
+ nOut++;
+ }
+ i++;
+ }
+
+ if( nOut!=1 ){
+ argError(aData, zType, sz, zArg);
+ }else{
+ *piOut = iOut;
+ }
+ return (nOut!=1);
+}
+
+struct timeval zero_time;
+
+void testTimeInit(void){
+ gettimeofday(&zero_time, 0);
+}
+
+int testTimeGet(void){
+ struct timeval now;
+ gettimeofday(&now, 0);
+ return
+ (((int)now.tv_sec - (int)zero_time.tv_sec)*1000) +
+ (((int)now.tv_usec - (int)zero_time.tv_usec)/1000);
+}
diff --git a/ext/lsm1/lsm-test/lsmtest_win32.c b/ext/lsm1/lsm-test/lsmtest_win32.c
new file mode 100644
index 0000000..9472723
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_win32.c
@@ -0,0 +1,30 @@
+
+#include "lsmtest.h"
+
+#ifdef _WIN32
+
+#define TICKS_PER_SECOND (10000000)
+#define TICKS_PER_MICROSECOND (10)
+#define TICKS_UNIX_EPOCH (116444736000000000LL)
+
+int win32GetTimeOfDay(
+ struct timeval *tp,
+ void *tzp
+){
+ FILETIME fileTime;
+ ULONGLONG ticks;
+ ULONGLONG unixTicks;
+
+ unused_parameter(tzp);
+ memset(&fileTime, 0, sizeof(FILETIME));
+ GetSystemTimeAsFileTime(&fileTime);
+ ticks = (ULONGLONG)fileTime.dwHighDateTime << 32;
+ ticks |= (ULONGLONG)fileTime.dwLowDateTime;
+ unixTicks = ticks - TICKS_UNIX_EPOCH;
+ tp->tv_sec = (long)(unixTicks / TICKS_PER_SECOND);
+ unixTicks -= ((ULONGLONG)tp->tv_sec * TICKS_PER_SECOND);
+ tp->tv_usec = (long)(unixTicks / TICKS_PER_MICROSECOND);
+
+ return 0;
+}
+#endif