Georg Brandl 3dbca81c9b Merged revisions 65012,65035,65037-65040,65048,65057,65077,65091-65095,65097-65099,65127-65128,65131,65133-65136,65139,65149-65151,65155,65158-65159,65176-65178,65183-65184,65187-65190,65192,65194 via svnmerge from
svn+ssh://pythondev@svn.python.org/python/trunk

........
  r65012 | jesse.noller | 2008-07-16 15:24:06 +0200 (Wed, 16 Jul 2008) | 2 lines

  Apply patch for issue 3090: ARCHFLAGS parsing incorrect
........
  r65035 | georg.brandl | 2008-07-16 23:19:28 +0200 (Wed, 16 Jul 2008) | 2 lines

  #3045: fix pydoc behavior for TEMP path with spaces.
........
  r65037 | georg.brandl | 2008-07-16 23:31:41 +0200 (Wed, 16 Jul 2008) | 2 lines

  #1608818: errno can get set by every call to readdir().
........
  r65038 | georg.brandl | 2008-07-17 00:04:20 +0200 (Thu, 17 Jul 2008) | 2 lines

  #3305: self->stream can be NULL.
........
  r65039 | georg.brandl | 2008-07-17 00:09:17 +0200 (Thu, 17 Jul 2008) | 2 lines

  #3345: fix docstring.
........
  r65040 | georg.brandl | 2008-07-17 00:33:18 +0200 (Thu, 17 Jul 2008) | 2 lines

  #3312: fix two sqlite3 crashes.
........
  r65048 | georg.brandl | 2008-07-17 01:35:54 +0200 (Thu, 17 Jul 2008) | 2 lines

  #3388: add a paragraph about using "with" for file objects.
........
  r65057 | gregory.p.smith | 2008-07-17 05:13:05 +0200 (Thu, 17 Jul 2008) | 2 lines

  news note for r63052
........
  r65077 | jesse.noller | 2008-07-17 23:01:05 +0200 (Thu, 17 Jul 2008) | 3 lines

  Fix issue 3395, update _debugInfo to be _debug_info
........
  r65091 | ronald.oussoren | 2008-07-18 07:48:03 +0200 (Fri, 18 Jul 2008) | 2 lines

  Last bit of a fix for issue3381 (addon for my patch in r65061)
........
  r65092 | vinay.sajip | 2008-07-18 10:59:06 +0200 (Fri, 18 Jul 2008) | 1 line

  Issue #3389: Allow resolving dotted names for handlers in logging configuration files. Thanks to Philip Jenvey for the patch.
........
  r65093 | vinay.sajip | 2008-07-18 11:00:00 +0200 (Fri, 18 Jul 2008) | 1 line

  Issue #3389: Allow resolving dotted names for handlers in logging configuration files. Thanks to Philip Jenvey for the patch.
........
  r65094 | vinay.sajip | 2008-07-18 11:00:35 +0200 (Fri, 18 Jul 2008) | 1 line

  Issue #3389: Allow resolving dotted names for handlers in logging configuration files. Thanks to Philip Jenvey for the patch.
........
  r65095 | vinay.sajip | 2008-07-18 11:01:10 +0200 (Fri, 18 Jul 2008) | 1 line

  Issue #3389: Allow resolving dotted names for handlers in logging configuration files. Thanks to Philip Jenvey for the patch.
........
  r65097 | georg.brandl | 2008-07-18 12:20:59 +0200 (Fri, 18 Jul 2008) | 2 lines

  Remove duplicate entry in __all__.
........
  r65098 | georg.brandl | 2008-07-18 12:29:30 +0200 (Fri, 18 Jul 2008) | 2 lines

  Correct attribute name.
........
  r65099 | georg.brandl | 2008-07-18 13:15:06 +0200 (Fri, 18 Jul 2008) | 3 lines

  Document the different meaning of precision for {:f} and {:g}.
  Also document how inf and nan are formatted. #3404.
........
  r65127 | raymond.hettinger | 2008-07-19 02:42:03 +0200 (Sat, 19 Jul 2008) | 1 line

  Improve accuracy of gamma test function
........
  r65128 | raymond.hettinger | 2008-07-19 02:43:00 +0200 (Sat, 19 Jul 2008) | 1 line

  Add recipe to the itertools docs.
........
  r65131 | georg.brandl | 2008-07-19 12:08:55 +0200 (Sat, 19 Jul 2008) | 2 lines

  #3378: in case of no memory, don't leak even more memory. :)
........
  r65133 | georg.brandl | 2008-07-19 14:39:10 +0200 (Sat, 19 Jul 2008) | 3 lines

  #3302: fix segfaults when passing None for arguments that can't
  be NULL for the C functions.
........
  r65134 | georg.brandl | 2008-07-19 14:46:12 +0200 (Sat, 19 Jul 2008) | 2 lines

  #3303: fix crash with invalid Py_DECREF in strcoll().
........
  r65135 | georg.brandl | 2008-07-19 15:00:22 +0200 (Sat, 19 Jul 2008) | 3 lines

  #3319: don't raise ZeroDivisionError if number of rounds is so
  low that benchtime is zero.
........
  r65136 | georg.brandl | 2008-07-19 15:09:42 +0200 (Sat, 19 Jul 2008) | 3 lines

  #3323: mention that if inheriting from a class without __slots__,
  the subclass will have a __dict__ available too.
........
  r65139 | georg.brandl | 2008-07-19 15:48:44 +0200 (Sat, 19 Jul 2008) | 2 lines

  Add ordering info for findall and finditer.
........
  r65149 | raymond.hettinger | 2008-07-20 01:21:57 +0200 (Sun, 20 Jul 2008) | 1 line

  Fix compress() recipe in docs to use itertools.
........
  r65150 | raymond.hettinger | 2008-07-20 01:58:47 +0200 (Sun, 20 Jul 2008) | 1 line

  Clean-up itertools docs and recipes.
........
  r65151 | gregory.p.smith | 2008-07-20 02:22:08 +0200 (Sun, 20 Jul 2008) | 9 lines

  fix issue3120 - don't truncate handles on 64-bit Windows.

  This is still messy, realistically PC/_subprocess.c should never cast pointers
  to python numbers and back at all.

  I don't have a 64-bit windows build environment because microsoft apparently
  thinks that should cost money.  Time to watch the buildbots.  It builds and
  passes tests on 32-bit windows.
........
  r65155 | georg.brandl | 2008-07-20 13:50:29 +0200 (Sun, 20 Jul 2008) | 2 lines

  #926501: add info where to put the docstring.
........
  r65158 | neal.norwitz | 2008-07-20 21:35:23 +0200 (Sun, 20 Jul 2008) | 1 line

  Fix a couple of names in error messages that were wrong
........
  r65159 | neal.norwitz | 2008-07-20 22:39:36 +0200 (Sun, 20 Jul 2008) | 1 line

  Fix misspeeld method name (negative)
........
  r65176 | amaury.forgeotdarc | 2008-07-21 23:36:24 +0200 (Mon, 21 Jul 2008) | 4 lines

  Increment version number in NEWS file, and move items that were added after 2.6b2.

  (I thought there was a script to automate this kind of updates)
........
  r65177 | amaury.forgeotdarc | 2008-07-22 00:00:38 +0200 (Tue, 22 Jul 2008) | 5 lines

  Issue2378: pdb would delete free variables when stepping into a class statement.

  The problem was introduced by r53954, the correction is to restore the symmetry between
  PyFrame_FastToLocals and PyFrame_LocalsToFast
........
  r65178 | benjamin.peterson | 2008-07-22 00:05:34 +0200 (Tue, 22 Jul 2008) | 1 line

  don't use assert statement
........
  r65183 | ronald.oussoren | 2008-07-22 09:06:00 +0200 (Tue, 22 Jul 2008) | 2 lines

  Fix buglet in fix for issue3381
........
  r65184 | ronald.oussoren | 2008-07-22 09:06:33 +0200 (Tue, 22 Jul 2008) | 2 lines

  Fix build issue on OSX 10.4, somehow this wasn't committed before.
........
  r65187 | raymond.hettinger | 2008-07-22 20:54:02 +0200 (Tue, 22 Jul 2008) | 1 line

  Remove out-of-date section on Exact/Inexact.
........
  r65188 | raymond.hettinger | 2008-07-22 21:00:47 +0200 (Tue, 22 Jul 2008) | 1 line

  Tuples now have both count() and index().
........
  r65189 | raymond.hettinger | 2008-07-22 21:03:05 +0200 (Tue, 22 Jul 2008) | 1 line

  Fix credits for math.sum()
........
  r65190 | raymond.hettinger | 2008-07-22 21:18:50 +0200 (Tue, 22 Jul 2008) | 1 line

  One more attribution.
........
  r65192 | benjamin.peterson | 2008-07-23 01:44:37 +0200 (Wed, 23 Jul 2008) | 1 line

  remove unneeded import
........
  r65194 | benjamin.peterson | 2008-07-23 15:25:06 +0200 (Wed, 23 Jul 2008) | 1 line

  use isinstance
........
2008-07-23 16:10:53 +00:00

473 lines
15 KiB
C

/* module.c - the module itself
*
* Copyright (C) 2004-2007 Gerhard Häring <gh@ghaering.de>
*
* This file is part of pysqlite.
*
* This software is provided 'as-is', without any express or implied
* warranty. In no event will the authors be held liable for any damages
* arising from the use of this software.
*
* Permission is granted to anyone to use this software for any purpose,
* including commercial applications, and to alter it and redistribute it
* freely, subject to the following restrictions:
*
* 1. The origin of this software must not be misrepresented; you must not
* claim that you wrote the original software. If you use this software
* in a product, an acknowledgment in the product documentation would be
* appreciated but is not required.
* 2. Altered source versions must be plainly marked as such, and must not be
* misrepresented as being the original software.
* 3. This notice may not be removed or altered from any source distribution.
*/
#include "connection.h"
#include "statement.h"
#include "cursor.h"
#include "cache.h"
#include "prepare_protocol.h"
#include "microprotocols.h"
#include "row.h"
#if SQLITE_VERSION_NUMBER >= 3003003
#define HAVE_SHARED_CACHE
#endif
/* static objects at module-level */
PyObject* pysqlite_Error, *pysqlite_Warning, *pysqlite_InterfaceError, *pysqlite_DatabaseError,
*pysqlite_InternalError, *pysqlite_OperationalError, *pysqlite_ProgrammingError,
*pysqlite_IntegrityError, *pysqlite_DataError, *pysqlite_NotSupportedError, *pysqlite_OptimizedUnicode;
PyObject* converters;
int _enable_callback_tracebacks;
int pysqlite_BaseTypeAdapted;
static PyObject* module_connect(PyObject* self, PyObject* args, PyObject*
kwargs)
{
/* Python seems to have no way of extracting a single keyword-arg at
* C-level, so this code is redundant with the one in connection_init in
* connection.c and must always be copied from there ... */
static char *kwlist[] = {"database", "timeout", "detect_types", "isolation_level", "check_same_thread", "factory", "cached_statements", NULL, NULL};
char* database;
int detect_types = 0;
PyObject* isolation_level;
PyObject* factory = NULL;
int check_same_thread = 1;
int cached_statements;
double timeout = 5.0;
PyObject* result;
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "s|diOiOi", kwlist,
&database, &timeout, &detect_types, &isolation_level, &check_same_thread, &factory, &cached_statements))
{
return NULL;
}
if (factory == NULL) {
factory = (PyObject*)&pysqlite_ConnectionType;
}
result = PyObject_Call(factory, args, kwargs);
return result;
}
PyDoc_STRVAR(module_connect_doc,
"connect(database[, timeout, isolation_level, detect_types, factory])\n\
\n\
Opens a connection to the SQLite database file *database*. You can use\n\
\":memory:\" to open a database connection to a database that resides in\n\
RAM instead of on disk.");
static PyObject* module_complete(PyObject* self, PyObject* args, PyObject*
kwargs)
{
static char *kwlist[] = {"statement", NULL, NULL};
char* statement;
PyObject* result;
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "s", kwlist, &statement))
{
return NULL;
}
if (sqlite3_complete(statement)) {
result = Py_True;
} else {
result = Py_False;
}
Py_INCREF(result);
return result;
}
PyDoc_STRVAR(module_complete_doc,
"complete_statement(sql)\n\
\n\
Checks if a string contains a complete SQL statement. Non-standard.");
#ifdef HAVE_SHARED_CACHE
static PyObject* module_enable_shared_cache(PyObject* self, PyObject* args, PyObject*
kwargs)
{
static char *kwlist[] = {"do_enable", NULL, NULL};
int do_enable;
int rc;
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "i", kwlist, &do_enable))
{
return NULL;
}
rc = sqlite3_enable_shared_cache(do_enable);
if (rc != SQLITE_OK) {
PyErr_SetString(pysqlite_OperationalError, "Changing the shared_cache flag failed");
return NULL;
} else {
Py_INCREF(Py_None);
return Py_None;
}
}
PyDoc_STRVAR(module_enable_shared_cache_doc,
"enable_shared_cache(do_enable)\n\
\n\
Enable or disable shared cache mode for the calling thread.\n\
Experimental/Non-standard.");
#endif /* HAVE_SHARED_CACHE */
static PyObject* module_register_adapter(PyObject* self, PyObject* args)
{
PyTypeObject* type;
PyObject* caster;
int rc;
if (!PyArg_ParseTuple(args, "OO", &type, &caster)) {
return NULL;
}
/* a basic type is adapted; there's a performance optimization if that's not the case
* (99 % of all usages) */
if (type == &PyLong_Type || type == &PyFloat_Type
|| type == &PyUnicode_Type || type == &PyByteArray_Type) {
pysqlite_BaseTypeAdapted = 1;
}
rc = microprotocols_add(type, (PyObject*)&pysqlite_PrepareProtocolType, caster);
if (rc == -1)
return NULL;
Py_INCREF(Py_None);
return Py_None;
}
PyDoc_STRVAR(module_register_adapter_doc,
"register_adapter(type, callable)\n\
\n\
Registers an adapter with pysqlite's adapter registry. Non-standard.");
static PyObject* module_register_converter(PyObject* self, PyObject* args)
{
PyObject* orig_name;
PyObject* name = NULL;
PyObject* callable;
PyObject* retval = NULL;
if (!PyArg_ParseTuple(args, "UO", &orig_name, &callable)) {
return NULL;
}
/* convert the name to upper case */
name = PyObject_CallMethod(orig_name, "upper", "");
if (!name) {
goto error;
}
if (PyDict_SetItem(converters, name, callable) != 0) {
goto error;
}
Py_INCREF(Py_None);
retval = Py_None;
error:
Py_XDECREF(name);
return retval;
}
PyDoc_STRVAR(module_register_converter_doc,
"register_converter(typename, callable)\n\
\n\
Registers a converter with pysqlite. Non-standard.");
static PyObject* enable_callback_tracebacks(PyObject* self, PyObject* args)
{
if (!PyArg_ParseTuple(args, "i", &_enable_callback_tracebacks)) {
return NULL;
}
Py_INCREF(Py_None);
return Py_None;
}
PyDoc_STRVAR(enable_callback_tracebacks_doc,
"enable_callback_tracebacks(flag)\n\
\n\
Enable or disable callback functions throwing errors to stderr.");
static void converters_init(PyObject* dict)
{
converters = PyDict_New();
if (!converters) {
return;
}
PyDict_SetItemString(dict, "converters", converters);
}
static PyMethodDef module_methods[] = {
{"connect", (PyCFunction)module_connect,
METH_VARARGS | METH_KEYWORDS, module_connect_doc},
{"complete_statement", (PyCFunction)module_complete,
METH_VARARGS | METH_KEYWORDS, module_complete_doc},
#ifdef HAVE_SHARED_CACHE
{"enable_shared_cache", (PyCFunction)module_enable_shared_cache,
METH_VARARGS | METH_KEYWORDS, module_enable_shared_cache_doc},
#endif
{"register_adapter", (PyCFunction)module_register_adapter,
METH_VARARGS, module_register_adapter_doc},
{"register_converter", (PyCFunction)module_register_converter,
METH_VARARGS, module_register_converter_doc},
{"adapt", (PyCFunction)psyco_microprotocols_adapt, METH_VARARGS,
psyco_microprotocols_adapt_doc},
{"enable_callback_tracebacks", (PyCFunction)enable_callback_tracebacks,
METH_VARARGS, enable_callback_tracebacks_doc},
{NULL, NULL}
};
struct _IntConstantPair {
char* constant_name;
int constant_value;
};
typedef struct _IntConstantPair IntConstantPair;
static IntConstantPair _int_constants[] = {
{"PARSE_DECLTYPES", PARSE_DECLTYPES},
{"PARSE_COLNAMES", PARSE_COLNAMES},
{"SQLITE_OK", SQLITE_OK},
{"SQLITE_DENY", SQLITE_DENY},
{"SQLITE_IGNORE", SQLITE_IGNORE},
{"SQLITE_CREATE_INDEX", SQLITE_CREATE_INDEX},
{"SQLITE_CREATE_TABLE", SQLITE_CREATE_TABLE},
{"SQLITE_CREATE_TEMP_INDEX", SQLITE_CREATE_TEMP_INDEX},
{"SQLITE_CREATE_TEMP_TABLE", SQLITE_CREATE_TEMP_TABLE},
{"SQLITE_CREATE_TEMP_TRIGGER", SQLITE_CREATE_TEMP_TRIGGER},
{"SQLITE_CREATE_TEMP_VIEW", SQLITE_CREATE_TEMP_VIEW},
{"SQLITE_CREATE_TRIGGER", SQLITE_CREATE_TRIGGER},
{"SQLITE_CREATE_VIEW", SQLITE_CREATE_VIEW},
{"SQLITE_DELETE", SQLITE_DELETE},
{"SQLITE_DROP_INDEX", SQLITE_DROP_INDEX},
{"SQLITE_DROP_TABLE", SQLITE_DROP_TABLE},
{"SQLITE_DROP_TEMP_INDEX", SQLITE_DROP_TEMP_INDEX},
{"SQLITE_DROP_TEMP_TABLE", SQLITE_DROP_TEMP_TABLE},
{"SQLITE_DROP_TEMP_TRIGGER", SQLITE_DROP_TEMP_TRIGGER},
{"SQLITE_DROP_TEMP_VIEW", SQLITE_DROP_TEMP_VIEW},
{"SQLITE_DROP_TRIGGER", SQLITE_DROP_TRIGGER},
{"SQLITE_DROP_VIEW", SQLITE_DROP_VIEW},
{"SQLITE_INSERT", SQLITE_INSERT},
{"SQLITE_PRAGMA", SQLITE_PRAGMA},
{"SQLITE_READ", SQLITE_READ},
{"SQLITE_SELECT", SQLITE_SELECT},
{"SQLITE_TRANSACTION", SQLITE_TRANSACTION},
{"SQLITE_UPDATE", SQLITE_UPDATE},
{"SQLITE_ATTACH", SQLITE_ATTACH},
{"SQLITE_DETACH", SQLITE_DETACH},
#if SQLITE_VERSION_NUMBER >= 3002001
{"SQLITE_ALTER_TABLE", SQLITE_ALTER_TABLE},
{"SQLITE_REINDEX", SQLITE_REINDEX},
#endif
#if SQLITE_VERSION_NUMBER >= 3003000
{"SQLITE_ANALYZE", SQLITE_ANALYZE},
#endif
{(char*)NULL, 0}
};
static struct PyModuleDef _sqlite3module = {
PyModuleDef_HEAD_INIT,
"_sqlite3",
NULL,
-1,
module_methods,
NULL,
NULL,
NULL,
NULL
};
PyMODINIT_FUNC PyInit__sqlite3(void)
{
PyObject *module, *dict;
PyObject *tmp_obj;
int i;
module = PyModule_Create(&_sqlite3module);
if (!module ||
(pysqlite_row_setup_types() < 0) ||
(pysqlite_cursor_setup_types() < 0) ||
(pysqlite_connection_setup_types() < 0) ||
(pysqlite_cache_setup_types() < 0) ||
(pysqlite_statement_setup_types() < 0) ||
(pysqlite_prepare_protocol_setup_types() < 0)
) {
Py_DECREF(module);
return NULL;
}
Py_INCREF(&pysqlite_ConnectionType);
PyModule_AddObject(module, "Connection", (PyObject*) &pysqlite_ConnectionType);
Py_INCREF(&pysqlite_CursorType);
PyModule_AddObject(module, "Cursor", (PyObject*) &pysqlite_CursorType);
Py_INCREF(&pysqlite_CacheType);
PyModule_AddObject(module, "Statement", (PyObject*)&pysqlite_StatementType);
Py_INCREF(&pysqlite_StatementType);
PyModule_AddObject(module, "Cache", (PyObject*) &pysqlite_CacheType);
Py_INCREF(&pysqlite_PrepareProtocolType);
PyModule_AddObject(module, "PrepareProtocol", (PyObject*) &pysqlite_PrepareProtocolType);
Py_INCREF(&pysqlite_RowType);
PyModule_AddObject(module, "Row", (PyObject*) &pysqlite_RowType);
if (!(dict = PyModule_GetDict(module))) {
goto error;
}
/*** Create DB-API Exception hierarchy */
if (!(pysqlite_Error = PyErr_NewException(MODULE_NAME ".Error", PyExc_Exception, NULL))) {
goto error;
}
PyDict_SetItemString(dict, "Error", pysqlite_Error);
if (!(pysqlite_Warning = PyErr_NewException(MODULE_NAME ".Warning", PyExc_Exception, NULL))) {
goto error;
}
PyDict_SetItemString(dict, "Warning", pysqlite_Warning);
/* Error subclasses */
if (!(pysqlite_InterfaceError = PyErr_NewException(MODULE_NAME ".InterfaceError", pysqlite_Error, NULL))) {
goto error;
}
PyDict_SetItemString(dict, "InterfaceError", pysqlite_InterfaceError);
if (!(pysqlite_DatabaseError = PyErr_NewException(MODULE_NAME ".DatabaseError", pysqlite_Error, NULL))) {
goto error;
}
PyDict_SetItemString(dict, "DatabaseError", pysqlite_DatabaseError);
/* pysqlite_DatabaseError subclasses */
if (!(pysqlite_InternalError = PyErr_NewException(MODULE_NAME ".InternalError", pysqlite_DatabaseError, NULL))) {
goto error;
}
PyDict_SetItemString(dict, "InternalError", pysqlite_InternalError);
if (!(pysqlite_OperationalError = PyErr_NewException(MODULE_NAME ".OperationalError", pysqlite_DatabaseError, NULL))) {
goto error;
}
PyDict_SetItemString(dict, "OperationalError", pysqlite_OperationalError);
if (!(pysqlite_ProgrammingError = PyErr_NewException(MODULE_NAME ".ProgrammingError", pysqlite_DatabaseError, NULL))) {
goto error;
}
PyDict_SetItemString(dict, "ProgrammingError", pysqlite_ProgrammingError);
if (!(pysqlite_IntegrityError = PyErr_NewException(MODULE_NAME ".IntegrityError", pysqlite_DatabaseError,NULL))) {
goto error;
}
PyDict_SetItemString(dict, "IntegrityError", pysqlite_IntegrityError);
if (!(pysqlite_DataError = PyErr_NewException(MODULE_NAME ".DataError", pysqlite_DatabaseError, NULL))) {
goto error;
}
PyDict_SetItemString(dict, "DataError", pysqlite_DataError);
if (!(pysqlite_NotSupportedError = PyErr_NewException(MODULE_NAME ".NotSupportedError", pysqlite_DatabaseError, NULL))) {
goto error;
}
PyDict_SetItemString(dict, "NotSupportedError", pysqlite_NotSupportedError);
/* We just need "something" unique for pysqlite_OptimizedUnicode. It does not really
* need to be a string subclass. Just anything that can act as a special
* marker for us. So I pulled PyCell_Type out of my magic hat.
*/
Py_INCREF((PyObject*)&PyCell_Type);
pysqlite_OptimizedUnicode = (PyObject*)&PyCell_Type;
PyDict_SetItemString(dict, "OptimizedUnicode", pysqlite_OptimizedUnicode);
/* Set integer constants */
for (i = 0; _int_constants[i].constant_name != 0; i++) {
tmp_obj = PyLong_FromLong(_int_constants[i].constant_value);
if (!tmp_obj) {
goto error;
}
PyDict_SetItemString(dict, _int_constants[i].constant_name, tmp_obj);
Py_DECREF(tmp_obj);
}
if (!(tmp_obj = PyUnicode_FromString(PYSQLITE_VERSION))) {
goto error;
}
PyDict_SetItemString(dict, "version", tmp_obj);
Py_DECREF(tmp_obj);
if (!(tmp_obj = PyUnicode_FromString(sqlite3_libversion()))) {
goto error;
}
PyDict_SetItemString(dict, "sqlite_version", tmp_obj);
Py_DECREF(tmp_obj);
/* initialize microprotocols layer */
microprotocols_init(dict);
/* initialize the default converters */
converters_init(dict);
_enable_callback_tracebacks = 0;
pysqlite_BaseTypeAdapted = 0;
/* Original comment from _bsddb.c in the Python core. This is also still
* needed nowadays for Python 2.3/2.4.
*
* PyEval_InitThreads is called here due to a quirk in python 1.5
* - 2.2.1 (at least) according to Russell Williamson <merel@wt.net>:
* The global interpreter lock is not initialized until the first
* thread is created using thread.start_new_thread() or fork() is
* called. that would cause the ALLOW_THREADS here to segfault due
* to a null pointer reference if no threads or child processes
* have been created. This works around that and is a no-op if
* threads have already been initialized.
* (see pybsddb-users mailing list post on 2002-08-07)
*/
PyEval_InitThreads();
error:
if (PyErr_Occurred())
{
PyErr_SetString(PyExc_ImportError, MODULE_NAME ": init failed");
Py_DECREF(module);
module = NULL;
}
return module;
}