Johan Wikman 35620f8dae Build compare using embedded library
For whatever reason compare fails to load qc_mysqlembedded if
it has not been built with the embedded library. Needs to be
sorted out at some point.
2016-11-23 14:35:46 +02:00

1606 lines
40 KiB
C++

/*
* Copyright (c) 2016 MariaDB Corporation Ab
*
* Use of this software is governed by the Business Source License included
* in the LICENSE.TXT file and at www.mariadb.com/bsl.
*
* Change Date: 2019-07-01
*
* On the date above, in accordance with the Business Source License, use
* of this software will be governed by version 2 or later of the General
* Public License.
*/
#include <unistd.h>
#include <cstdlib>
#include <algorithm>
#include <fstream>
#include <iostream>
#include <map>
#include <set>
#include <string>
#include <sstream>
#define MYSQL_COM_QUIT COM_QUIT
#define MYSQL_COM_INIT_DB COM_INIT_DB
#define MYSQL_COM_CHANGE_USER COM_CHANGE_USER
#include <maxscale/gwdirs.h>
#include <maxscale/log_manager.h>
#include <maxscale/protocol/mysql.h>
#include <maxscale/query_classifier.h>
using std::cerr;
using std::cin;
using std::cout;
using std::endl;
using std::ifstream;
using std::istream;
using std::map;
using std::ostream;
using std::string;
using std::stringstream;
namespace
{
char USAGE[] =
"usage: compare [-r count] [-d] [-1 classfier1] [-2 classifier2] "
"[-A args] [-B args] [-v [0..2]] [-s statement]|[file]]\n\n"
"-r redo the test the specified number of times; 0 means forever, default is 1\n"
"-d don't stop after first failed query\n"
"-1 the first classifier, default qc_mysqlembedded\n"
"-2 the second classifier, default qc_sqlite\n"
"-A arguments for the first classifier\n"
"-B arguments for the second classifier\n"
"-s compare single statement\n"
"-S strict, also require that the parse result is identical\n"
"-v 0, only return code\n"
" 1, query and result for failed cases\n"
" 2, all queries, and result for failed cases\n"
" 3, all queries and all results\n";
enum verbosity_t
{
VERBOSITY_MIN,
VERBOSITY_NORMAL,
VERBOSITY_EXTENDED,
VERBOSITY_MAX
};
struct State
{
bool query_printed;
string query;
verbosity_t verbosity;
bool result_printed;
bool stop_at_error;
bool strict;
size_t line;
size_t n_statements;
size_t n_errors;
struct timespec time1;
struct timespec time2;
} global = { false, // query_printed
"", // query
VERBOSITY_NORMAL, // verbosity
false, // result_printed
true, // stop_at_error
false, // strict
0, // line
0, // n_statements
0, // n_errors
{ 0, 0 }, // time1
{ 0, 0} }; // time2
ostream& operator << (ostream& out, qc_parse_result_t x)
{
switch (x)
{
case QC_QUERY_INVALID:
out << "QC_QUERY_INVALID";
break;
case QC_QUERY_TOKENIZED:
out << "QC_QUERY_TOKENIZED";
break;
case QC_QUERY_PARTIALLY_PARSED:
out << "QC_QUERY_PARTIALLY_PARSED";
break;
case QC_QUERY_PARSED:
out << "QC_QUERY_PARSED";
break;
default:
out << "static_cast<c_parse_result_t>(" << (int)x << ")";
}
return out;
}
GWBUF* create_gwbuf(const string& s)
{
size_t len = s.length();
size_t payload_len = len + 1;
size_t gwbuf_len = MYSQL_HEADER_LEN + payload_len;
GWBUF* gwbuf = gwbuf_alloc(gwbuf_len);
*((unsigned char*)((char*)GWBUF_DATA(gwbuf))) = payload_len;
*((unsigned char*)((char*)GWBUF_DATA(gwbuf) + 1)) = (payload_len >> 8);
*((unsigned char*)((char*)GWBUF_DATA(gwbuf) + 2)) = (payload_len >> 16);
*((unsigned char*)((char*)GWBUF_DATA(gwbuf) + 3)) = 0x00;
*((unsigned char*)((char*)GWBUF_DATA(gwbuf) + 4)) = 0x03;
memcpy((char*)GWBUF_DATA(gwbuf) + 5, s.c_str(), len);
return gwbuf;
}
QUERY_CLASSIFIER* load_classifier(const char* name)
{
bool loaded = false;
size_t len = strlen(name);
char libdir[len + 1];
sprintf(libdir, "../%s", name);
set_libdir(strdup(libdir));
QUERY_CLASSIFIER *pClassifier = qc_load(name);
if (!pClassifier)
{
cerr << "error: Could not load classifier " << name << "." << endl;
}
return pClassifier;
}
QUERY_CLASSIFIER* get_classifier(const char* zName, const char* zArgs)
{
QUERY_CLASSIFIER* pClassifier = load_classifier(zName);
if (pClassifier)
{
if (!pClassifier->qc_init(zArgs))
{
cerr << "error: Could not init classifier " << zName << "." << endl;
qc_unload(pClassifier);
pClassifier = 0;
}
}
return pClassifier;
}
void put_classifier(QUERY_CLASSIFIER* pClassifier)
{
if (pClassifier)
{
pClassifier->qc_end();
qc_unload(pClassifier);
}
}
bool get_classifiers(const char* zName1, const char* zArgs1, QUERY_CLASSIFIER** ppClassifier1,
const char* zName2, const char* zArgs2, QUERY_CLASSIFIER** ppClassifier2)
{
bool rc = false;
QUERY_CLASSIFIER* pClassifier1 = get_classifier(zName1, zArgs1);
if (pClassifier1)
{
QUERY_CLASSIFIER* pClassifier2 = get_classifier(zName2, zArgs2);
if (pClassifier2)
{
*ppClassifier1 = pClassifier1;
*ppClassifier2 = pClassifier2;
rc = true;
}
else
{
put_classifier(pClassifier1);
}
}
return rc;
}
void put_classifiers(QUERY_CLASSIFIER* pClassifier1, QUERY_CLASSIFIER* pClassifier2)
{
put_classifier(pClassifier1);
put_classifier(pClassifier2);
}
void report_query()
{
cout << "(" << global.line << "): " << global.query << endl;
global.query_printed = true;
}
void report(bool success, const string& s)
{
if (success)
{
if (global.verbosity >= VERBOSITY_NORMAL)
{
if (global.verbosity >= VERBOSITY_EXTENDED)
{
if (!global.query_printed)
{
report_query();
}
if (global.verbosity >= VERBOSITY_MAX)
{
cout << s << endl;
global.result_printed = true;
}
}
}
}
else
{
if (global.verbosity >= VERBOSITY_NORMAL)
{
if (!global.query_printed)
{
report_query();
}
cout << s << endl;
global.result_printed = true;
}
}
}
static timespec timespec_subtract(const timespec& later, const timespec& earlier)
{
timespec result = { 0, 0 };
ss_dassert((later.tv_sec > earlier.tv_sec) ||
((later.tv_sec == earlier.tv_sec) && (later.tv_nsec > earlier.tv_nsec)));
if (later.tv_nsec >= earlier.tv_nsec)
{
result.tv_sec = later.tv_sec - earlier.tv_sec;
result.tv_nsec = later.tv_nsec - earlier.tv_nsec;
}
else
{
result.tv_sec = later.tv_sec - earlier.tv_sec - 1;
result.tv_nsec = 1000000000 + later.tv_nsec - earlier.tv_nsec;
}
return result;
}
static void update_time(timespec* pResult, timespec& start, timespec& finish)
{
timespec difference = timespec_subtract(finish, start);
long nanosecs = pResult->tv_nsec + difference.tv_nsec;
if (nanosecs > 1000000000)
{
pResult->tv_sec += 1;
pResult->tv_nsec += (nanosecs - 1000000000);
}
else
{
pResult->tv_nsec = nanosecs;
}
pResult->tv_sec += difference.tv_sec;
}
bool compare_parse(QUERY_CLASSIFIER* pClassifier1, GWBUF* pCopy1,
QUERY_CLASSIFIER* pClassifier2, GWBUF* pCopy2)
{
bool success = false;
const char HEADING[] = "qc_parse : ";
struct timespec start;
struct timespec finish;
clock_gettime(CLOCK_MONOTONIC_RAW, &start);
qc_parse_result_t rv1 = pClassifier1->qc_parse(pCopy1);
clock_gettime(CLOCK_MONOTONIC_RAW, &finish);
update_time(&global.time1, start, finish);
clock_gettime(CLOCK_MONOTONIC_RAW, &start);
qc_parse_result_t rv2 = pClassifier2->qc_parse(pCopy2);
clock_gettime(CLOCK_MONOTONIC_RAW, &finish);
update_time(&global.time2, start, finish);
stringstream ss;
ss << HEADING;
if (rv1 == rv2)
{
ss << "Ok : " << rv1;
success = true;
}
else
{
if (global.strict)
{
ss << "ERR: ";
}
else
{
ss << "INF: ";
success = true;
}
ss << rv1 << " != " << rv2;
}
report(success, ss.str());
return success;
}
bool compare_get_type(QUERY_CLASSIFIER* pClassifier1, GWBUF* pCopy1,
QUERY_CLASSIFIER* pClassifier2, GWBUF* pCopy2)
{
bool success = false;
const char HEADING[] = "qc_get_type : ";
uint32_t rv1 = pClassifier1->qc_get_type(pCopy1);
uint32_t rv2 = pClassifier2->qc_get_type(pCopy2);
stringstream ss;
ss << HEADING;
if (rv1 == rv2)
{
char* types = qc_typemask_to_string(rv1);
ss << "Ok : " << types;
free(types);
success = true;
}
else
{
uint32_t rv1b = rv1;
if (rv1b & QUERY_TYPE_WRITE)
{
rv1b &= ~(uint32_t)QUERY_TYPE_READ;
}
uint32_t rv2b = rv2;
if (rv2b & QUERY_TYPE_WRITE)
{
rv2b &= ~(uint32_t)QUERY_TYPE_READ;
}
if (rv1b & QUERY_TYPE_READ)
{
rv1b &= ~(uint32_t)QUERY_TYPE_LOCAL_READ;
}
if (rv2b & QUERY_TYPE_READ)
{
rv2b &= ~(uint32_t)QUERY_TYPE_LOCAL_READ;
}
char* types1 = qc_typemask_to_string(rv1);
char* types2 = qc_typemask_to_string(rv2);
if (rv1b == rv2b)
{
ss << "WRN: " << types1 << " != " << types2;
success = true;
}
else
{
ss << "ERR: " << types1 << " != " << types2;
}
free(types1);
free(types2);
}
report(success, ss.str());
return success;
}
bool compare_get_operation(QUERY_CLASSIFIER* pClassifier1, GWBUF* pCopy1,
QUERY_CLASSIFIER* pClassifier2, GWBUF* pCopy2)
{
bool success = false;
const char HEADING[] = "qc_get_operation : ";
qc_query_op_t rv1 = pClassifier1->qc_get_operation(pCopy1);
qc_query_op_t rv2 = pClassifier2->qc_get_operation(pCopy2);
stringstream ss;
ss << HEADING;
if (rv1 == rv2)
{
ss << "Ok : " << qc_op_to_string(rv1);
success = true;
}
else
{
ss << "ERR: " << qc_op_to_string(rv1) << " != " << qc_op_to_string(rv2);
}
report(success, ss.str());
return success;
}
bool compare_get_created_table_name(QUERY_CLASSIFIER* pClassifier1, GWBUF* pCopy1,
QUERY_CLASSIFIER* pClassifier2, GWBUF* pCopy2)
{
bool success = false;
const char HEADING[] = "qc_get_created_table_name: ";
char* rv1 = pClassifier1->qc_get_created_table_name(pCopy1);
char* rv2 = pClassifier2->qc_get_created_table_name(pCopy2);
stringstream ss;
ss << HEADING;
if ((!rv1 && !rv2) || (rv1 && rv2 && (strcmp(rv1, rv2) == 0)))
{
ss << "Ok : " << (rv1 ? rv1 : "NULL");
success = true;
}
else
{
ss << "ERR: " << (rv1 ? rv1 : "NULL") << " != " << (rv2 ? rv2 : "NULL");
}
report(success, ss.str());
free(rv1);
free(rv2);
return success;
}
bool compare_is_drop_table_query(QUERY_CLASSIFIER* pClassifier1, GWBUF* pCopy1,
QUERY_CLASSIFIER* pClassifier2, GWBUF* pCopy2)
{
bool success = false;
const char HEADING[] = "qc_is_drop_table_query : ";
bool rv1 = pClassifier1->qc_is_drop_table_query(pCopy1);
bool rv2 = pClassifier2->qc_is_drop_table_query(pCopy2);
stringstream ss;
ss << HEADING;
if (rv1 == rv2)
{
ss << "Ok : " << rv1;
success = true;
}
else
{
ss << "ERR: " << rv1 << " != " << rv2;
}
report(success, ss.str());
return success;
}
bool compare_is_real_query(QUERY_CLASSIFIER* pClassifier1, GWBUF* pCopy1,
QUERY_CLASSIFIER* pClassifier2, GWBUF* pCopy2)
{
bool success = false;
const char HEADING[] = "qc_is_real_query : ";
bool rv1 = pClassifier1->qc_is_real_query(pCopy1);
bool rv2 = pClassifier2->qc_is_real_query(pCopy2);
stringstream ss;
ss << HEADING;
if (rv1 == rv2)
{
ss << "Ok : " << rv1;
success = true;
}
else
{
ss << "ERR: " << rv1 << " != " << rv2;
}
report(success, ss.str());
return success;
}
bool compare_strings(const char* const* strings1, const char* const* strings2, int n)
{
for (int i = 0; i < n; ++i)
{
const char* s1 = strings1[i];
const char* s2 = strings2[i];
if (strcmp(s1, s2) != 0)
{
return false;
}
}
return true;
}
void free_strings(char** strings, int n)
{
if (strings)
{
for (int i = 0; i < n; ++i)
{
free(strings[i]);
}
free(strings);
}
}
void print_names(ostream& out, const char* const* strings, int n)
{
if (strings)
{
for (int i = 0; i < n; ++i)
{
out << strings[i];
if (i < n - 1)
{
out << ", ";
}
}
}
else
{
out << "NULL";
}
}
bool compare_get_table_names(QUERY_CLASSIFIER* pClassifier1, GWBUF* pCopy1,
QUERY_CLASSIFIER* pClassifier2, GWBUF* pCopy2,
bool full)
{
bool success = false;
const char* HEADING;
if (full)
{
HEADING = "qc_get_table_names(full) : ";
}
else
{
HEADING = "qc_get_table_names : ";
}
int n1 = 0;
int n2 = 0;
char** rv1 = pClassifier1->qc_get_table_names(pCopy1, &n1, full);
char** rv2 = pClassifier2->qc_get_table_names(pCopy2, &n2, full);
// The order need not be the same, so let's compare a set.
std::set<string> names1;
std::set<string> names2;
if (rv1)
{
std::copy(rv1, rv1 + n1, inserter(names1, names1.begin()));
}
if (rv2)
{
std::copy(rv2, rv2 + n2, inserter(names2, names2.begin()));
}
stringstream ss;
ss << HEADING;
if ((!rv1 && !rv2) || (names1 == names2))
{
if (n1 == n2)
{
ss << "Ok : ";
print_names(ss, rv1, n1);
}
else
{
ss << "WRN: ";
print_names(ss, rv1, n1);
ss << " != ";
print_names(ss, rv2, n2);
}
success = true;
}
else
{
ss << "ERR: ";
print_names(ss, rv1, n1);
ss << " != ";
print_names(ss, rv2, n2);
}
report(success, ss.str());
free_strings(rv1, n1);
free_strings(rv2, n2);
return success;
}
bool compare_query_has_clause(QUERY_CLASSIFIER* pClassifier1, GWBUF* pCopy1,
QUERY_CLASSIFIER* pClassifier2, GWBUF* pCopy2)
{
bool success = false;
const char HEADING[] = "qc_query_has_clause : ";
bool rv1 = pClassifier1->qc_query_has_clause(pCopy1);
bool rv2 = pClassifier2->qc_query_has_clause(pCopy2);
stringstream ss;
ss << HEADING;
if (rv1 == rv2)
{
ss << "Ok : " << rv1;
success = true;
}
else
{
ss << "ERR: " << rv1 << " != " << rv2;
}
report(success, ss.str());
return success;
}
void add_fields(std::set<string>& m, const char* fields)
{
const char* begin = fields;
const char* end = begin;
// As long as we have not reached the end.
while (*end != 0)
{
// Walk over everything but whitespace.
while (!isspace(*end) && (*end != 0))
{
++end;
}
// Insert whatever we found.
m.insert(string(begin, end - begin));
// Walk over all whitespace.
while (isspace(*end) && (*end != 0))
{
++end;
}
// Move begin to the next non-whitespace character.
begin = end;
}
if (begin != end)
{
m.insert(string(begin, end - begin));
}
}
ostream& operator << (ostream& o, const std::set<string>& s)
{
std::set<string>::iterator i = s.begin();
while (i != s.end())
{
o << *i;
++i;
if (i != s.end())
{
o << " ";
}
}
return o;
}
bool compare_get_database_names(QUERY_CLASSIFIER* pClassifier1, GWBUF* pCopy1,
QUERY_CLASSIFIER* pClassifier2, GWBUF* pCopy2)
{
bool success = false;
const char HEADING[] = "qc_get_database_names : ";
int n1 = 0;
int n2 = 0;
char** rv1 = pClassifier1->qc_get_database_names(pCopy1, &n1);
char** rv2 = pClassifier2->qc_get_database_names(pCopy2, &n2);
stringstream ss;
ss << HEADING;
if ((!rv1 && !rv2) || ((n1 == n2) && compare_strings(rv1, rv2, n1)))
{
ss << "Ok : ";
print_names(ss, rv1, n1);
success = true;
}
else
{
ss << "ERR: ";
print_names(ss, rv1, n1);
ss << " != ";
print_names(ss, rv2, n2);
}
report(success, ss.str());
free_strings(rv1, n1);
free_strings(rv2, n2);
return success;
}
bool compare_get_prepare_name(QUERY_CLASSIFIER* pClassifier1, GWBUF* pCopy1,
QUERY_CLASSIFIER* pClassifier2, GWBUF* pCopy2)
{
bool success = false;
const char HEADING[] = "qc_get_prepare_name : ";
char* rv1 = pClassifier1->qc_get_prepare_name(pCopy1);
char* rv2 = pClassifier2->qc_get_prepare_name(pCopy2);
stringstream ss;
ss << HEADING;
if ((!rv1 && !rv2) || (rv1 && rv2 && (strcmp(rv1, rv2) == 0)))
{
ss << "Ok : " << (rv1 ? rv1 : "NULL");
success = true;
}
else
{
ss << "ERR: " << (rv1 ? rv1 : "NULL") << " != " << (rv2 ? rv2 : "NULL");
}
report(success, ss.str());
free(rv1);
free(rv2);
return success;
}
bool compare_get_prepare_operation(QUERY_CLASSIFIER* pClassifier1, GWBUF* pCopy1,
QUERY_CLASSIFIER* pClassifier2, GWBUF* pCopy2)
{
bool success = false;
const char HEADING[] = "qc_get_prepare_operation : ";
qc_query_op_t rv1 = pClassifier1->qc_get_prepare_operation(pCopy1);
qc_query_op_t rv2 = pClassifier2->qc_get_prepare_operation(pCopy2);
stringstream ss;
ss << HEADING;
if (rv1 == rv2)
{
ss << "Ok : " << qc_op_to_string(rv1);
success = true;
}
else
{
ss << "ERR: " << qc_op_to_string(rv1) << " != " << qc_op_to_string(rv2);
}
report(success, ss.str());
return success;
}
bool operator == (const QC_FIELD_INFO& lhs, const QC_FIELD_INFO& rhs)
{
bool rv = false;
if (lhs.column && rhs.column && (strcasecmp(lhs.column, rhs.column) == 0))
{
if (!lhs.table && !rhs.table)
{
rv = true;
}
else if (lhs.table && rhs.table && (strcmp(lhs.table, rhs.table) == 0))
{
if (!lhs.database && !rhs.database)
{
rv = true;
}
else if (lhs.database && rhs.database && (strcmp(lhs.database, rhs.database) == 0))
{
rv = true;
}
}
}
return rv;
}
ostream& operator << (ostream& out, const QC_FIELD_INFO& x)
{
if (x.database)
{
out << x.database;
out << ".";
ss_dassert(x.table);
}
if (x.table)
{
out << x.table;
out << ".";
}
ss_dassert(x.column);
out << x.column;
return out;
}
class QcFieldInfo
{
public:
QcFieldInfo(const QC_FIELD_INFO& info)
: m_database(info.database ? info.database : "")
, m_table(info.table ? info.table : "")
, m_column(info.column ? info.column : "")
, m_usage(info.usage)
{}
bool eq(const QcFieldInfo& rhs) const
{
return
m_database == rhs.m_database &&
m_table == rhs.m_table &&
m_column == rhs.m_column &&
m_usage == rhs.m_usage;
}
bool lt(const QcFieldInfo& rhs) const
{
bool rv = false;
if (m_database < rhs.m_database)
{
rv = true;
}
else if (m_database > rhs.m_database)
{
rv = false;
}
else
{
if (m_table < rhs.m_table)
{
rv = true;
}
else if (m_table > rhs.m_table)
{
rv = false;
}
else
{
if (m_column < rhs.m_column)
{
rv = true;
}
else if (m_column > rhs.m_column)
{
rv = false;
}
else
{
rv = (m_usage < rhs.m_usage);
}
}
}
return rv;
}
void print(ostream& out) const
{
if (!m_database.empty())
{
out << m_database;
out << ".";
}
if (!m_table.empty())
{
out << m_table;
out << ".";
}
out << m_column;
out << "(";
char* s = qc_field_usage_mask_to_string(m_usage);
out << s;
free(s);
out << ")";
}
private:
std::string m_database;
std::string m_table;
std::string m_column;
uint32_t m_usage;
};
ostream& operator << (ostream& out, const QcFieldInfo& x)
{
x.print(out);
return out;
}
ostream& operator << (ostream& out, std::set<QcFieldInfo>& x)
{
std::set<QcFieldInfo>::iterator i = x.begin();
std::set<QcFieldInfo>::iterator end = x.end();
while (i != end)
{
out << *i++;
if (i != end)
{
out << " ";
}
}
return out;
}
bool operator < (const QcFieldInfo& lhs, const QcFieldInfo& rhs)
{
return lhs.lt(rhs);
}
bool operator == (const QcFieldInfo& lhs, const QcFieldInfo& rhs)
{
return lhs.eq(rhs);
}
bool are_equal(const QC_FIELD_INFO* fields1, size_t n_fields1,
const QC_FIELD_INFO* fields2, size_t n_fields2)
{
bool rv = (n_fields1 == n_fields2);
if (rv)
{
size_t i = 0;
while (rv && (i < n_fields1))
{
rv = *fields1 == *fields2;
++i;
}
}
return rv;
}
ostream& print(ostream& out, const QC_FIELD_INFO* fields, size_t n_fields)
{
size_t i = 0;
while (i < n_fields)
{
out << fields[i++];
if (i != n_fields)
{
out << " ";
}
}
return out;
}
bool compare_get_field_info(QUERY_CLASSIFIER* pClassifier1, GWBUF* pCopy1,
QUERY_CLASSIFIER* pClassifier2, GWBUF* pCopy2)
{
bool success = false;
const char HEADING[] = "qc_get_field_info : ";
const QC_FIELD_INFO* infos1;
const QC_FIELD_INFO* infos2;
size_t n_infos1;
size_t n_infos2;
pClassifier1->qc_get_field_info(pCopy1, &infos1, &n_infos1);
pClassifier2->qc_get_field_info(pCopy2, &infos2, &n_infos2);
stringstream ss;
ss << HEADING;
int i;
std::set<QcFieldInfo> f1;
f1.insert(infos1, infos1 + n_infos1);
std::set<QcFieldInfo> f2;
f2.insert(infos2, infos2 + n_infos2);
if (f1 == f2)
{
ss << "Ok : ";
ss << f1;
success = true;
}
else
{
ss << "ERR: " << f1 << " != " << f2;
}
report(success, ss.str());
return success;
}
bool compare(QUERY_CLASSIFIER* pClassifier1, QUERY_CLASSIFIER* pClassifier2, const string& s)
{
GWBUF* pCopy1 = create_gwbuf(s);
GWBUF* pCopy2 = create_gwbuf(s);
int errors = 0;
errors += !compare_parse(pClassifier1, pCopy1, pClassifier2, pCopy2);
errors += !compare_get_type(pClassifier1, pCopy1, pClassifier2, pCopy2);
errors += !compare_get_operation(pClassifier1, pCopy1, pClassifier2, pCopy2);
errors += !compare_get_created_table_name(pClassifier1, pCopy1, pClassifier2, pCopy2);
errors += !compare_is_drop_table_query(pClassifier1, pCopy1, pClassifier2, pCopy2);
errors += !compare_is_real_query(pClassifier1, pCopy1, pClassifier2, pCopy2);
errors += !compare_get_table_names(pClassifier1, pCopy1, pClassifier2, pCopy2, false);
errors += !compare_get_table_names(pClassifier1, pCopy1, pClassifier2, pCopy2, true);
errors += !compare_query_has_clause(pClassifier1, pCopy1, pClassifier2, pCopy2);
errors += !compare_get_database_names(pClassifier1, pCopy1, pClassifier2, pCopy2);
errors += !compare_get_prepare_name(pClassifier1, pCopy1, pClassifier2, pCopy2);
errors += !compare_get_prepare_operation(pClassifier1, pCopy1, pClassifier2, pCopy2);
errors += !compare_get_field_info(pClassifier1, pCopy1, pClassifier2, pCopy2);
gwbuf_free(pCopy1);
gwbuf_free(pCopy2);
if (global.result_printed)
{
cout << endl;
}
return errors == 0;
}
inline void ltrim(std::string &s)
{
s.erase(s.begin(), std::find_if(s.begin(), s.end(), std::not1(std::ptr_fun<int, int>(std::isspace))));
}
inline void rtrim(std::string &s)
{
s.erase(std::find_if(s.rbegin(), s.rend(),
std::not1(std::ptr_fun<int, int>(std::isspace))).base(), s.end());
}
static void trim(std::string &s)
{
ltrim(s);
rtrim(s);
}
enum skip_action_t
{
SKIP_NOTHING, // Skip nothing.
SKIP_BLOCK, // Skip until the end of next { ... }
SKIP_DELIMITER, // Skip the new delimiter.
SKIP_LINE, // Skip current line.
SKIP_NEXT_STATEMENT, // Skip statement starting on line following this line.
SKIP_STATEMENT, // Skip statment starting on this line.
SKIP_TERMINATE, // Cannot handle this, terminate.
};
typedef std::map<std::string, skip_action_t> KeywordActionMapping;
static KeywordActionMapping mtl_keywords;
void init_keywords()
{
struct Keyword
{
const char* z_keyword;
skip_action_t action;
};
static const Keyword KEYWORDS[] =
{
{ "append_file", SKIP_LINE },
{ "cat_file", SKIP_LINE },
{ "change_user", SKIP_LINE },
{ "character_set", SKIP_LINE },
{ "chmod", SKIP_LINE },
{ "connect", SKIP_LINE },
{ "connection", SKIP_LINE },
{ "copy_file", SKIP_LINE },
{ "dec", SKIP_LINE },
{ "delimiter", SKIP_DELIMITER },
{ "die", SKIP_LINE },
{ "diff_files", SKIP_LINE },
{ "dirty_close", SKIP_LINE },
{ "disable_abort_on_error", SKIP_LINE },
{ "disable_connect_log", SKIP_LINE },
{ "disable_info", SKIP_LINE },
{ "disable_metadata", SKIP_LINE },
{ "disable_parsing", SKIP_LINE },
{ "disable_ps_protocol", SKIP_LINE },
{ "disable_query_log", SKIP_LINE },
{ "disable_reconnect", SKIP_LINE },
{ "disable_result_log", SKIP_LINE },
{ "disable_rpl_parse", SKIP_LINE },
{ "disable_session_track_info", SKIP_LINE },
{ "disable_warnings", SKIP_LINE },
{ "disconnect", SKIP_LINE },
{ "echo", SKIP_LINE },
{ "enable_abort_on_error", SKIP_LINE },
{ "enable_connect_log", SKIP_LINE },
{ "enable_info", SKIP_LINE },
{ "enable_metadata", SKIP_LINE },
{ "enable_parsing", SKIP_LINE },
{ "enable_ps_protocol", SKIP_LINE },
{ "enable_query_log", SKIP_LINE },
{ "enable_reconnect", SKIP_LINE },
{ "enable_result_log", SKIP_LINE },
{ "enable_rpl_parse", SKIP_LINE },
{ "enable_session_track_info", SKIP_LINE },
{ "enable_warnings", SKIP_LINE },
{ "end_timer", SKIP_LINE },
{ "error", SKIP_NEXT_STATEMENT },
{ "eval", SKIP_STATEMENT },
{ "exec", SKIP_LINE },
{ "exit", SKIP_LINE },
{ "file_exists", SKIP_LINE },
{ "horizontal_results", SKIP_LINE },
{ "if", SKIP_BLOCK },
{ "inc", SKIP_LINE },
{ "let", SKIP_LINE },
{ "let", SKIP_LINE },
{ "list_files", SKIP_LINE },
{ "list_files_append_file", SKIP_LINE },
{ "list_files_write_file", SKIP_LINE },
{ "lowercase_result", SKIP_LINE },
{ "mkdir", SKIP_LINE },
{ "move_file", SKIP_LINE },
{ "output", SKIP_LINE },
{ "perl", SKIP_TERMINATE },
{ "ping", SKIP_LINE },
{ "print", SKIP_LINE },
{ "query", SKIP_LINE },
{ "query_get_value", SKIP_LINE },
{ "query_horizontal", SKIP_LINE },
{ "query_vertical", SKIP_LINE },
{ "real_sleep", SKIP_LINE },
{ "reap", SKIP_LINE },
{ "remove_file", SKIP_LINE },
{ "remove_files_wildcard", SKIP_LINE },
{ "replace_column", SKIP_LINE },
{ "replace_regex", SKIP_LINE },
{ "replace_result", SKIP_LINE },
{ "require", SKIP_LINE },
{ "reset_connection", SKIP_LINE },
{ "result", SKIP_LINE },
{ "result_format", SKIP_LINE },
{ "rmdir", SKIP_LINE },
{ "same_master_pos", SKIP_LINE },
{ "send", SKIP_LINE },
{ "send_eval", SKIP_LINE },
{ "send_quit", SKIP_LINE },
{ "send_shutdown", SKIP_LINE },
{ "skip", SKIP_LINE },
{ "sleep", SKIP_LINE },
{ "sorted_result", SKIP_LINE },
{ "source", SKIP_LINE },
{ "start_timer", SKIP_LINE },
{ "sync_slave_with_master", SKIP_LINE },
{ "sync_with_master", SKIP_LINE },
{ "system", SKIP_LINE },
{ "vertical_results", SKIP_LINE },
{ "while", SKIP_BLOCK },
{ "write_file", SKIP_LINE },
};
const size_t N_KEYWORDS = sizeof(KEYWORDS)/sizeof(KEYWORDS[0]);
for (size_t i = 0; i < N_KEYWORDS; ++i)
{
mtl_keywords[KEYWORDS[i].z_keyword] = KEYWORDS[i].action;
}
}
skip_action_t get_action(const string& keyword)
{
skip_action_t action = SKIP_NOTHING;
string key(keyword);
std::transform(key.begin(), key.end(), key.begin(), ::tolower);
KeywordActionMapping::iterator i = mtl_keywords.find(key);
if (i != mtl_keywords.end())
{
action = i->second;
}
return action;
}
void skip_block(istream& in)
{
int c;
// Find first '{'
while (in && ((c = in.get()) != '{'))
{
if (c == '\n')
{
++global.line;
}
}
int n = 1;
while ((n > 0) && in)
{
c = in.get();
switch (c)
{
case '{':
++n;
break;
case '}':
--n;
break;
case '\n':
++global.line;
break;
default:
;
}
}
}
int run(QUERY_CLASSIFIER* pClassifier1, QUERY_CLASSIFIER* pClassifier2, istream& in)
{
bool stop = false; // Whether we should exit.
bool skip = false; // Whether next statement should be skipped.
char delimiter = ';';
string query;
while (!stop && std::getline(in, query))
{
trim(query);
global.line++;
if (!query.empty() && (query.at(0) != '#'))
{
if (!skip)
{
if (query.substr(0, 2) == "--")
{
query = query.substr(2);
trim(query);
}
string::iterator i = std::find_if(query.begin(), query.end(),
std::ptr_fun<int,int>(std::isspace));
string keyword = query.substr(0, i - query.begin());
skip_action_t action = get_action(keyword);
switch (action)
{
case SKIP_NOTHING:
break;
case SKIP_BLOCK:
skip_block(in);
continue;
case SKIP_DELIMITER:
query = query.substr(i - query.begin());
trim(query);
if (query.length() > 0)
{
delimiter = query.at(0);
}
continue;
case SKIP_LINE:
continue;
case SKIP_NEXT_STATEMENT:
skip = true;
continue;
case SKIP_STATEMENT:
skip = true;
break;
case SKIP_TERMINATE:
cout << "error: Cannot handle line " << global.line
<< ", terminating: " << query << endl;
stop = true;
break;
}
}
global.query += query;
char c = query.at(query.length() - 1);
if (c == delimiter)
{
if (c != ';')
{
// If the delimiter was something else but ';' we need to
// remove that before giving the query to the classifiers.
global.query.erase(global.query.length() - 1);
}
if (!skip)
{
global.query_printed = false;
global.result_printed = false;
++global.n_statements;
if (global.verbosity >= VERBOSITY_EXTENDED)
{
// In case the execution crashes, we want the query printed.
report_query();
}
bool success = compare(pClassifier1, pClassifier2, global.query);
if (!success)
{
++global.n_errors;
if (global.stop_at_error)
{
stop = true;
}
}
}
else
{
skip = false;
}
global.query.clear();
}
else
{
global.query += " ";
}
}
else if (query.substr(0, 7) == "--error")
{
// Next statement is supposed to fail, no need to check.
skip = true;
}
}
return global.n_errors == 0 ? EXIT_SUCCESS : EXIT_FAILURE;
}
int run(QUERY_CLASSIFIER* pClassifier1, QUERY_CLASSIFIER* pClassifier2, const string& statement)
{
global.query = statement;
++global.n_statements;
if (global.verbosity >= VERBOSITY_EXTENDED)
{
// In case the execution crashes, we want the query printed.
report_query();
}
if (!compare(pClassifier1, pClassifier2, global.query))
{
++global.n_errors;
}
return global.n_errors == 0 ? EXIT_SUCCESS : EXIT_FAILURE;
}
}
int main(int argc, char* argv[])
{
int rc = EXIT_SUCCESS;
const char* zClassifier1 = "qc_mysqlembedded";
const char* zClassifier2 = "qc_sqlite";
const char* zClassifier1Args = NULL;
const char* zClassifier2Args = "log_unrecognized_statements=1";
const char* zStatement = NULL;
size_t rounds = 1;
int v = VERBOSITY_NORMAL;
int c;
while ((c = getopt(argc, argv, "r:d1:2:v:A:B:s:S")) != -1)
{
switch (c)
{
case 'r':
rounds = atoi(optarg);
break;
case 'v':
v = atoi(optarg);
break;
case '1':
zClassifier1 = optarg;
break;
case '2':
zClassifier2 = optarg;
break;
case 'A':
zClassifier1Args = optarg;
break;
case 'B':
zClassifier2Args = optarg;
break;
case 'd':
global.stop_at_error = false;
break;
case 's':
zStatement = optarg;
break;
case 'S':
global.strict = true;
break;
default:
rc = EXIT_FAILURE;
break;
};
}
if ((rc == EXIT_SUCCESS) && (v >= VERBOSITY_MIN && v <= VERBOSITY_MAX))
{
init_keywords();
rc = EXIT_FAILURE;
global.verbosity = static_cast<verbosity_t>(v);
int n = argc - (optind - 1);
if ((n == 1) || (n == 2))
{
set_datadir(strdup("/tmp"));
set_langdir(strdup("."));
set_process_datadir(strdup("/tmp"));
if (mxs_log_init(NULL, ".", MXS_LOG_TARGET_DEFAULT))
{
QUERY_CLASSIFIER* pClassifier1;
QUERY_CLASSIFIER* pClassifier2;
if (get_classifiers(zClassifier1, zClassifier1Args, &pClassifier1,
zClassifier2, zClassifier2Args, &pClassifier2))
{
size_t round = 0;
bool terminate = false;
do
{
++round;
global.n_statements = 0;
global.n_errors = 0;
global.query_printed = false;
global.result_printed = false;
if (zStatement)
{
rc = run(pClassifier1, pClassifier2, zStatement);
}
else if (n == 1)
{
rc = run(pClassifier1, pClassifier2, cin);
}
else
{
ss_dassert(n == 2);
ifstream in(argv[argc - 1]);
if (in)
{
rc = run(pClassifier1, pClassifier2, in);
}
else
{
terminate = true;
cerr << "error: Could not open " << argv[argc - 1] << "." << endl;
}
}
cout << "\n"
<< "Statements: " << global.n_statements << endl
<< "Errors : " << global.n_errors << endl;
if (!terminate && ((rounds == 0) || (round < rounds)))
{
cout << endl;
}
}
while (!terminate && ((rounds == 0) || (round < rounds)));
put_classifiers(pClassifier1, pClassifier2);
cout << "\n";
cout << "1st classifier: "
<< global.time1.tv_sec << "."
<< global.time1.tv_nsec
<< endl;
cout << "2nd classifier: "
<< global.time2.tv_sec << "."
<< global.time2.tv_nsec
<< endl;
}
mxs_log_finish();
}
else
{
cerr << "error: Could not initialize log." << endl;
}
}
else
{
cout << USAGE << endl;
}
}
else
{
cout << USAGE << endl;
}
return rc;
}