/* Copyright (C) 2007 J.F.Dockes
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the
* Free Software Foundation, Inc.,
* 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
*/
#include <Python.h>
#include <structmember.h>
#include <bytearrayobject.h>
#include <strings.h>
#include <string>
#include <iostream>
#include <set>
using namespace std;
#include "rclinit.h"
#include "rclconfig.h"
#include "rcldb.h"
#include "searchdata.h"
#include "rclquery.h"
#include "pathut.h"
#include "wasastringtoquery.h"
#include "wasatorcl.h"
#include "debuglog.h"
#include "pathut.h"
#include "plaintorich.h"
#include "hldata.h"
#include "pyrecoll.h"
static set<Rcl::Db *> the_dbs;
static set<Rcl::Query *> the_queries;
static set<Rcl::Doc *> the_docs;
static RclConfig *rclconfig;
#if PY_MAJOR_VERSION >=3
# define Py_TPFLAGS_HAVE_ITER 0
#endif
//////////////////////////////////////////////////////////////////////
/// SEARCHDATA SearchData code
typedef struct {
PyObject_HEAD
/* Type-specific fields go here. */
RefCntr<Rcl::SearchData> sd;
} recoll_SearchDataObject;
static void
SearchData_dealloc(recoll_SearchDataObject *self)
{
LOGDEB(("SearchData_dealloc\n"));
Py_TYPE(self)->tp_free((PyObject*)self);
}
static PyObject *
SearchData_new(PyTypeObject *type, PyObject *args, PyObject *kwds)
{
LOGDEB(("SearchData_new\n"));
recoll_SearchDataObject *self;
self = (recoll_SearchDataObject *)type->tp_alloc(type, 0);
if (self == 0)
return 0;
return (PyObject *)self;
}
PyDoc_STRVAR(doc_SearchDataObject,
"SearchData([type=AND|OR], [stemlang=somelanguage|null])\n"
"\n"
"A SearchData object describes a query. It has a number of global\n"
"parameters and a chain of search clauses.\n"
);
static int
SearchData_init(recoll_SearchDataObject *self, PyObject *args, PyObject *kwargs)
{
LOGDEB(("SearchData_init\n"));
static const char* kwlist[] = {"type", "stemlang", NULL};
char *stp = 0;
char *steml = 0;
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|sz", (char**)kwlist,
&stp, &steml))
return -1;
Rcl::SClType tp = Rcl::SCLT_AND;
if (stp && strcasecmp(stp, "or")) {
tp = Rcl::SCLT_OR;
}
string stemlang;
if (steml) {
stemlang = steml;
} else {
stemlang = "english";
}
self->sd = RefCntr<Rcl::SearchData>(new Rcl::SearchData(tp, stemlang));
return 0;
}
/* Note: addclause necessite And/Or vient du fait que le string peut avoir
plusieurs mots. A transferer dans l'i/f Python ou pas ? */
PyDoc_STRVAR(doc_addclause,
"addclause(type='and'|'or'|'filename'|'phrase'|'near'|'path'|'sub',\n"
" qstring=string, slack=int, field=string, stemming=1|0,\n"
" subSearch=SearchData, exclude=0|1, anchorstart=0|1, anchorend=0|1,\n"
" casesens=0|1, diacsens=0|1)\n"
"Adds a simple clause to the SearchData And/Or chain, or a subquery\n"
"defined by another SearchData object\n"
);
/* Forward declaration only, definition needs recoll_searchDataType */
static PyObject *
SearchData_addclause(recoll_SearchDataObject* self, PyObject *args,
PyObject *kwargs);
static PyMethodDef SearchData_methods[] = {
{"addclause", (PyCFunction)SearchData_addclause, METH_VARARGS|METH_KEYWORDS,
doc_addclause},
{NULL} /* Sentinel */
};
static PyTypeObject recoll_SearchDataType = {
PyVarObject_HEAD_INIT(NULL, 0)
"recoll.SearchData", /*tp_name*/
sizeof(recoll_SearchDataObject), /*tp_basicsize*/
0, /*tp_itemsize*/
(destructor)SearchData_dealloc, /*tp_dealloc*/
0, /*tp_print*/
0, /*tp_getattr*/
0, /*tp_setattr*/
0, /*tp_compare*/
0, /*tp_repr*/
0, /*tp_as_number*/
0, /*tp_as_sequence*/
0, /*tp_as_mapping*/
0, /*tp_hash */
0, /*tp_call*/
0, /*tp_str*/
0, /*tp_getattro*/
0, /*tp_setattro*/
0, /*tp_as_buffer*/
Py_TPFLAGS_DEFAULT|Py_TPFLAGS_BASETYPE, /*tp_flags*/
doc_SearchDataObject, /* tp_doc */
0, /* tp_traverse */
0, /* tp_clear */
0, /* tp_richcompare */
0, /* tp_weaklistoffset */
0, /* tp_iter */
0, /* tp_iternext */
SearchData_methods, /* tp_methods */
0, /* tp_members */
0, /* tp_getset */
0, /* tp_base */
0, /* tp_dict */
0, /* tp_descr_get */
0, /* tp_descr_set */
0, /* tp_dictoffset */
(initproc)SearchData_init, /* tp_init */
0, /* tp_alloc */
SearchData_new, /* tp_new */
};
static PyObject *
SearchData_addclause(recoll_SearchDataObject* self, PyObject *args,
PyObject *kwargs)
{
LOGDEB(("SearchData_addclause\n"));
if (self->sd.isNull()) {
LOGERR(("SearchData_addclause: not init??\n"));
PyErr_SetString(PyExc_AttributeError, "sd");
return 0;
}
static const char *kwlist[] = {"type", "qstring", "slack", "field",
"stemming", "subsearch",
"exclude", "anchorstart", "anchorend",
"casesens", "diacsens",
NULL};
char *tp = 0;
char *qs = 0; // needs freeing
int slack = 0;
char *fld = 0; // needs freeing
PyObject *dostem = 0;
recoll_SearchDataObject *sub = 0;
PyObject *exclude = 0;
PyObject *anchorstart = 0;
PyObject *anchorend = 0;
PyObject *casesens = 0;
PyObject *diacsens = 0;
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "ses|iesOO!OOOOO",
(char**)kwlist,
&tp, "utf-8", &qs, &slack,
"utf-8", &fld, &dostem,
&recoll_SearchDataType, &sub,
&exclude, &anchorstart, &anchorend,
&casesens, &diacsens
))
return 0;
Rcl::SearchDataClause *cl = 0;
switch (tp[0]) {
case 'a':
case 'A':
if (strcasecmp(tp, "and"))
goto defaultcase;
cl = new Rcl::SearchDataClauseSimple(Rcl::SCLT_AND, qs, fld?fld:"");
break;
case 'f':
case 'F':
if (strcasecmp(tp, "filename"))
goto defaultcase;
cl = new Rcl::SearchDataClauseFilename(qs);
break;
case 'o':
case 'O':
if (strcasecmp(tp, "or"))
goto defaultcase;
cl = new Rcl::SearchDataClauseSimple(Rcl::SCLT_OR, qs, fld?fld:"");
break;
case 'n':
case 'N':
if (strcasecmp(tp, "near"))
goto defaultcase;
cl = new Rcl::SearchDataClauseDist(Rcl::SCLT_NEAR, qs, slack,
fld ? fld : "");
break;
case 'p':
case 'P':
if (!strcasecmp(tp, "phrase")) {
cl = new Rcl::SearchDataClauseDist(Rcl::SCLT_PHRASE, qs, slack,
fld ? fld : "");
} else if (!strcasecmp(tp, "path")) {
cl = new Rcl::SearchDataClausePath(qs);
} else {
goto defaultcase;
}
break;
case 's':
case 'S':
if (strcasecmp(tp, "sub"))
goto defaultcase;
cl = new Rcl::SearchDataClauseSub(sub->sd);
break;
defaultcase:
default:
PyErr_SetString(PyExc_AttributeError, "Bad tp arg");
return 0;
}
PyMem_Free(qs);
PyMem_Free(fld);
if (dostem != 0 && !PyObject_IsTrue(dostem)) {
cl->addModifier(Rcl::SearchDataClause::SDCM_NOSTEMMING);
}
if (exclude != 0 && !PyObject_IsTrue(exclude)) {
cl->setexclude(true);
}
if (anchorstart && PyObject_IsTrue(anchorstart)) {
cl->addModifier(Rcl::SearchDataClause::SDCM_ANCHORSTART);
}
if (anchorend && PyObject_IsTrue(anchorend)) {
cl->addModifier(Rcl::SearchDataClause::SDCM_ANCHOREND);
}
if (casesens && PyObject_IsTrue(casesens)) {
cl->addModifier(Rcl::SearchDataClause::SDCM_CASESENS);
}
if (diacsens && PyObject_IsTrue(diacsens)) {
cl->addModifier(Rcl::SearchDataClause::SDCM_DIACSENS);
}
self->sd->addClause(cl);
Py_RETURN_NONE;
}
///////////////////////////////////////////////////////////////////////
///// DOC Doc code
static void
Doc_dealloc(recoll_DocObject *self)
{
LOGDEB(("Doc_dealloc\n"));
if (self->doc)
the_docs.erase(self->doc);
deleteZ(self->doc);
Py_TYPE(self)->tp_free((PyObject*)self);
}
static PyObject *
Doc_new(PyTypeObject *type, PyObject *args, PyObject *kwds)
{
recoll_DocObject *self;
LOGDEB(("Doc_new\n"));
self = (recoll_DocObject *)type->tp_alloc(type, 0);
if (self == 0)
return 0;
self->doc = 0;
self->rclconfig = 0;
return (PyObject *)self;
}
static int
Doc_init(recoll_DocObject *self, PyObject *, PyObject *)
{
LOGDEB(("Doc_init\n"));
if (self->doc)
the_docs.erase(self->doc);
delete self->doc;
self->doc = new Rcl::Doc;
if (self->doc == 0)
return -1;
self->rclconfig = rclconfig;
the_docs.insert(self->doc);
return 0;
}
PyDoc_STRVAR(doc_Doc_getbinurl,
"getbinurl(none) -> binary url\n"
"\n"
"Returns an URL with a path part which is a as bit for bit copy of the \n"
"file system path, without encoding\n"
);
static PyObject *
Doc_getbinurl(recoll_DocObject *self)
{
LOGDEB(("Doc_getbinurl\n"));
if (self->doc == 0 ||
the_docs.find(self->doc) == the_docs.end()) {
PyErr_SetString(PyExc_AttributeError, "doc");
return 0;
}
return PyByteArray_FromStringAndSize(self->doc->url.c_str(),
self->doc->url.size());
}
PyDoc_STRVAR(doc_Doc_setbinurl,
"setbinurl(url) -> binary url\n"
"\n"
"Set the URL from binary path like file://may/contain/unencodable/bytes\n"
);
static PyObject *
Doc_setbinurl(recoll_DocObject *self, PyObject *value)
{
if (self->doc == 0 ||
the_docs.find(self->doc) == the_docs.end()) {
PyErr_SetString(PyExc_AttributeError, "doc??");
return 0;
}
if (!PyByteArray_Check(value)) {
PyErr_SetString(PyExc_TypeError, "setbinurl needs byte array argument");
return 0;
}
self->doc->url = string(PyByteArray_AsString(value),
PyByteArray_Size(value));
Py_RETURN_NONE;
}
PyDoc_STRVAR(doc_Doc_keys,
"keys() -> list of doc object keys (attribute names)\n"
);
static PyObject *
Doc_keys(recoll_DocObject *self)
{
LOGDEB(("Doc_keys\n"));
if (self->doc == 0 ||
the_docs.find(self->doc) == the_docs.end()) {
PyErr_SetString(PyExc_AttributeError, "doc");
return 0;
}
PyObject *pkeys = PyList_New(0);
if (!pkeys)
return 0;
for (map<string,string>::const_iterator it = self->doc->meta.begin();
it != self->doc->meta.end(); it++) {
PyList_Append(pkeys, PyUnicode_Decode(it->first.c_str(),
it->first.size(),
"UTF-8", "replace"));
}
return pkeys;
}
PyDoc_STRVAR(doc_Doc_items,
"items() -> dictionary of doc object keys/values\n"
);
static PyObject *
Doc_items(recoll_DocObject *self)
{
LOGDEB(("Doc_getbinurl\n"));
if (self->doc == 0 ||
the_docs.find(self->doc) == the_docs.end()) {
PyErr_SetString(PyExc_AttributeError, "doc");
return 0;
}
PyObject *pdict = PyDict_New();
if (!pdict)
return 0;
for (map<string,string>::const_iterator it = self->doc->meta.begin();
it != self->doc->meta.end(); it++) {
PyDict_SetItem(pdict,
PyUnicode_Decode(it->first.c_str(),
it->first.size(),
"UTF-8", "replace"),
PyUnicode_Decode(it->second.c_str(),
it->second.size(),
"UTF-8", "replace"));
}
return pdict;
}
PyDoc_STRVAR(doc_Doc_get,
"get(key) -> value\n"
"Retrieve the named doc attribute\n"
);
static PyObject *
Doc_get(recoll_DocObject *self, PyObject *args)
{
LOGDEB(("Doc_get\n"));
char *sutf8 = 0; // needs freeing
if (!PyArg_ParseTuple(args, "es:Doc_get",
"utf-8", &sutf8)) {
return 0;
}
string key(sutf8);
PyMem_Free(sutf8);
if (self->doc == 0 ||
the_docs.find(self->doc) == the_docs.end()) {
PyErr_SetString(PyExc_AttributeError, "doc??");
return 0;
}
string value;
if (self->doc->getmeta(key, 0)) {
value = self->doc->meta[key];
return PyUnicode_Decode(value.c_str(),
value.size(),
"UTF-8", "replace");
}
Py_RETURN_NONE;
}
static PyMethodDef Doc_methods[] = {
{"getbinurl", (PyCFunction)Doc_getbinurl, METH_NOARGS, doc_Doc_getbinurl},
{"setbinurl", (PyCFunction)Doc_setbinurl, METH_O, doc_Doc_setbinurl},
{"keys", (PyCFunction)Doc_keys, METH_NOARGS, doc_Doc_keys},
{"items", (PyCFunction)Doc_items, METH_NOARGS, doc_Doc_items},
{"get", (PyCFunction)Doc_get, METH_VARARGS, doc_Doc_get},
{NULL} /* Sentinel */
};
// Note that this returns None if the attribute is not found instead of raising
// an exception as would be standard. We don't change it to keep existing code
// working.
static PyObject *
Doc_getattro(recoll_DocObject *self, PyObject *nameobj)
{
if (self->doc == 0 || the_docs.find(self->doc) == the_docs.end()) {
PyErr_SetString(PyExc_AttributeError, "doc");
return 0;
}
bool found = false;
string value;
string key;
char *name = 0;
PyObject* utf8o = 0;
if (PyUnicode_Check(nameobj)) {
utf8o = PyUnicode_AsUTF8String(nameobj);
if (utf8o == 0) {
LOGERR(("Doc_getattro: encoding name to utf8 failed\n"));
PyErr_SetString(PyExc_AttributeError, "name??");
Py_RETURN_NONE;
}
name = PyBytes_AsString(utf8o);
Py_DECREF(utf8o);
} else if (PyBytes_Check(nameobj)) {
name = PyBytes_AsString(nameobj);
} else {
PyErr_SetString(PyExc_AttributeError, "name not unicode nor string??");
Py_RETURN_NONE;
}
key = rclconfig->fieldCanon(string(name));
switch (key.at(0)) {
case 'u':
if (!key.compare(Rcl::Doc::keyurl)) {
value = self->doc->url; found = true;
}
break;
case 'f':
if (!key.compare(Rcl::Doc::keyfs)) {
value = self->doc->fbytes; found = true;
} else if (!key.compare(Rcl::Doc::keyfmt)) {
value = self->doc->fmtime; found = true;
}
break;
case 'd':
if (!key.compare(Rcl::Doc::keyds)) {
value = self->doc->dbytes; found = true;
} else if (!key.compare(Rcl::Doc::keydmt)) {
value = self->doc->dmtime; found = true;
}
break;
case 'i':
if (!key.compare(Rcl::Doc::keyipt)) {
value = self->doc->ipath; found = true;
}
break;
case 'm':
if (!key.compare(Rcl::Doc::keytp)) {
value = self->doc->mimetype; found = true;
} else if (!key.compare(Rcl::Doc::keymt)) {
value = self->doc->dmtime.empty() ? self->doc->fmtime :
self->doc->dmtime; found = true;
}
break;
case 'o':
if (!key.compare(Rcl::Doc::keyoc)) {
value = self->doc->origcharset; found = true;
}
break;
case 's':
if (!key.compare(Rcl::Doc::keysig)) {
value = self->doc->sig; found = true;
} else if (!key.compare(Rcl::Doc::keysz)) {
value = self->doc->dbytes.empty() ? self->doc->fbytes :
self->doc->dbytes; found = true;
}
break;
case 't':
if (!key.compare("text")) {
value = self->doc->text; found = true;
}
break;
}
if (!found) {
// This will look up a method name (we have no other standard
// attributes)
PyObject *meth = PyObject_GenericGetAttr((PyObject*)self, nameobj);
if (meth) {
return meth;
}
PyErr_Clear();
// Else look for another attribute
if (self->doc->getmeta(key, 0)) {
value = self->doc->meta[key];
found = true;
}
}
if (found) {
LOGDEB1(("Doc_getattro: [%s] -> [%s]\n", key.c_str(), value.c_str()));
// Return a python unicode object
return PyUnicode_Decode(value.c_str(), value.size(), "utf-8",
"replace");
}
Py_RETURN_NONE;
}
static int
Doc_setattr(recoll_DocObject *self, char *name, PyObject *value)
{
if (self->doc == 0 || the_docs.find(self->doc) == the_docs.end()) {
PyErr_SetString(PyExc_AttributeError, "doc??");
return -1;
}
LOGDEB1(("Doc_setmeta: doc %p\n", self->doc));
#if PY_MAJOR_VERSION < 3
if (PyString_Check(value)) {
value = PyUnicode_FromObject(value);
if (value == 0)
return -1;
}
#endif
if (!PyUnicode_Check(value)) {
PyErr_SetString(PyExc_AttributeError, "value not str/unicode??");
return -1;
}
if (name == 0) {
PyErr_SetString(PyExc_AttributeError, "name??");
return -1;
}
PyObject* putf8 = PyUnicode_AsUTF8String(value);
if (putf8 == 0) {
LOGERR(("Doc_setmeta: encoding to utf8 failed\n"));
PyErr_SetString(PyExc_AttributeError, "value??");
return -1;
}
char* uvalue = PyBytes_AsString(putf8);
Py_DECREF(putf8);
string key = rclconfig->fieldCanon(string(name));
LOGDEB0(("Doc_setattr: [%s] (%s) -> [%s]\n", key.c_str(), name, uvalue));
// We set the value in the meta array in all cases. Good idea ? or do it
// only for fields without a dedicated Doc:: entry?
self->doc->meta[key] = uvalue;
switch (key.at(0)) {
case 't':
if (!key.compare("text")) {
self->doc->text = uvalue;
}
break;
case 'u':
if (!key.compare(Rcl::Doc::keyurl)) {
self->doc->url = uvalue;
}
break;
case 'f':
if (!key.compare(Rcl::Doc::keyfs)) {
self->doc->fbytes = uvalue;
} else if (!key.compare(Rcl::Doc::keyfmt)) {
self->doc->fmtime = uvalue;
}
break;
case 'd':
if (!key.compare(Rcl::Doc::keyds)) {
self->doc->dbytes = uvalue;
} else if (!key.compare(Rcl::Doc::keydmt)) {
self->doc->dmtime = uvalue;
}
break;
case 'i':
if (!key.compare(Rcl::Doc::keyipt)) {
self->doc->ipath = uvalue;
}
break;
case 'm':
if (!key.compare(Rcl::Doc::keytp)) {
self->doc->mimetype = uvalue;
} else if (!key.compare(Rcl::Doc::keymt)) {
self->doc->dmtime = uvalue;
}
break;
case 'o':
if (!key.compare(Rcl::Doc::keyoc)) {
self->doc->origcharset = uvalue;
}
break;
case 's':
if (!key.compare(Rcl::Doc::keysig)) {
self->doc->sig = uvalue;
} else if (!key.compare(Rcl::Doc::keysz)) {
self->doc->dbytes = uvalue;
}
break;
}
return 0;
}
PyDoc_STRVAR(doc_DocObject,
"Doc()\n"
"\n"
"A Doc object contains index data for a given document.\n"
"The data is extracted from the index when searching, or set by the\n"
"indexer program when updating. The Doc object has no useful methods but\n"
"many attributes to be read or set by its user. It matches exactly the\n"
"Rcl::Doc c++ object. Some of the attributes are predefined, but, \n"
"especially when indexing, others can be set, the name of which will be\n"
"processed as field names by the indexing configuration.\n"
"Inputs can be specified as unicode or strings.\n"
"Outputs are unicode objects.\n"
"All dates are specified as unix timestamps, printed as strings\n"
"Predefined attributes (index/query/both):\n"
" text (index): document plain text\n"
" url (both)\n"
" fbytes (both) optional) file size in bytes\n"
" filename (both)\n"
" fmtime (both) optional file modification date. Unix time printed \n"
" as string\n"
" dbytes (both) document text bytes\n"
" dmtime (both) document creation/modification date\n"
" ipath (both) value private to the app.: internal access path\n"
" inside file\n"
" mtype (both) mime type for original document\n"
" mtime (query) dmtime if set else fmtime\n"
" origcharset (both) charset the text was converted from\n"
" size (query) dbytes if set, else fbytes\n"
" sig (both) app-defined file modification signature. \n"
" For up to date checks\n"
" relevancyrating (query)\n"
" abstract (both)\n"
" author (both)\n"
" title (both)\n"
" keywords (both)\n"
);
static PyTypeObject recoll_DocType = {
PyVarObject_HEAD_INIT(NULL, 0)
"recoll.Doc", /*tp_name*/
sizeof(recoll_DocObject), /*tp_basicsize*/
0, /*tp_itemsize*/
(destructor)Doc_dealloc, /*tp_dealloc*/
0, /*tp_print*/
0, /*tp_getattr*/
(setattrfunc)Doc_setattr, /*tp_setattr*/
0, /*tp_compare*/
0, /*tp_repr*/
0, /*tp_as_number*/
0, /*tp_as_sequence*/
0, /*tp_as_mapping*/
0, /*tp_hash */
0, /*tp_call*/
0, /*tp_str*/
(getattrofunc)Doc_getattro,/*tp_getattro*/
0, /*tp_setattro*/
0, /*tp_as_buffer*/
Py_TPFLAGS_DEFAULT|Py_TPFLAGS_BASETYPE, /*tp_flags*/
doc_DocObject, /* tp_doc */
0, /* tp_traverse */
0, /* tp_clear */
0, /* tp_richcompare */
0, /* tp_weaklistoffset */
0, /* tp_iter */
0, /* tp_iternext */
Doc_methods, /* tp_methods */
0, /* tp_members */
0, /* tp_getset */
0, /* tp_base */
0, /* tp_dict */
0, /* tp_descr_get */
0, /* tp_descr_set */
0, /* tp_dictoffset */
(initproc)Doc_init, /* tp_init */
0, /* tp_alloc */
Doc_new, /* tp_new */
};
//////////////////////////////////////////////////////
/// QUERY Query object
struct recoll_DbObject;
typedef struct {
PyObject_HEAD
/* Type-specific fields go here. */
Rcl::Query *query;
int next; // Index of result to be fetched next or -1 if uninit
int rowcount; // Number of records returned by last execute
string *sortfield;
int ascending;
int arraysize; // Default size for fetchmany
recoll_DbObject* connection;
} recoll_QueryObject;
PyDoc_STRVAR(doc_Query_close,
"close(). Deallocate query. Object is unusable after the call."
);
static void
Query_close(recoll_QueryObject *self)
{
LOGDEB(("Query_close\n"));
if (self->query)
the_queries.erase(self->query);
deleteZ(self->query);
deleteZ(self->sortfield);
if (self->connection)
Py_DECREF(self->connection);
}
static void
Query_dealloc(recoll_QueryObject *self)
{
LOGDEB(("Query_dealloc\n"));
Query_close(self);
Py_TYPE(self)->tp_free((PyObject*)self);
}
static PyObject *
Query_new(PyTypeObject *type, PyObject *args, PyObject *kwargs)
{
recoll_QueryObject *self;
LOGDEB(("Query_new\n"));
self = (recoll_QueryObject *)type->tp_alloc(type, 0);
if (self == 0)
return 0;
self->query = 0;
self->next = -1;
self->rowcount = -1;
self->sortfield = new string;
self->ascending = 1;
self->arraysize = 1;
self->connection = 0;
return (PyObject *)self;
}
// Query_init creates an unusable object. The only way to create a
// valid Query Object is through db_query(). (or we'd need to add a Db
// parameter to the Query object creation method)
static int
Query_init(recoll_QueryObject *self, PyObject *, PyObject *)
{
LOGDEB(("Query_init\n"));
if (self->query)
the_queries.erase(self->query);
delete self->query;
self->query = 0;
self->next = -1;
self->ascending = true;
return 0;
}
static PyObject *
Query_iter(PyObject *self)
{
return self;
}
PyDoc_STRVAR(doc_Query_sortby,
"sortby(field=fieldname, ascending=True)\n"
"Sort results by 'fieldname', in ascending or descending order.\n"
"Only one field can be used, no subsorts for now.\n"
"Must be called before executing the search\n"
);
static PyObject *
Query_sortby(recoll_QueryObject* self, PyObject *args, PyObject *kwargs)
{
LOGDEB(("Query_sortby\n"));
static const char *kwlist[] = {"field", "ascending", NULL};
char *sfield = 0;
PyObject *ascobj = 0;
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "s|O", (char**)kwlist,
&sfield,
&ascobj))
return 0;
if (sfield) {
self->sortfield->assign(sfield);
} else {
self->sortfield->clear();
}
if (ascobj == 0) {
self->ascending = true;
} else {
self->ascending = PyObject_IsTrue(ascobj);
}
Py_RETURN_NONE;
}
PyDoc_STRVAR(doc_Query_execute,
"execute(query_string, stemming=1|0, stemlang=\"stemming language\")\n"
"\n"
"Starts a search for query_string, a Recoll search language string\n"
"(mostly Xesam-compatible).\n"
"The query can be a simple list of terms (and'ed by default), or more\n"
"complicated with field specs etc. See the Recoll manual.\n"
);
static PyObject *
Query_execute(recoll_QueryObject* self, PyObject *args, PyObject *kwargs)
{
LOGDEB1(("Query_execute\n"));
static const char *kwlist[] = {"query_string", "stemming", "stemlang", NULL};
char *sutf8 = 0; // needs freeing
char *sstemlang = 0;
int dostem = 1;
PyObject *dostemobj = 0;
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "es|Oes:Query_execute",
(char**)kwlist, "utf-8", &sutf8,
&dostemobj,
"utf-8", &sstemlang)) {
return 0;
}
if (dostemobj != 0 && !PyObject_IsTrue(dostemobj))
dostem = 0;
string utf8(sutf8);
PyMem_Free(sutf8);
string stemlang("english");
if (sstemlang) {
stemlang.assign(sstemlang);
PyMem_Free(sstemlang);
}
LOGDEB(("Query_execute: [%s] dostem %d stemlang [%s]\n", sutf8, dostem,
stemlang.c_str()));
if (self->query == 0 ||
the_queries.find(self->query) == the_queries.end()) {
PyErr_SetString(PyExc_AttributeError, "query");
return 0;
}
// SearchData defaults to stemming in english
// Use default for now but need to add way to specify language
string reason;
Rcl::SearchData *sd = wasaStringToRcl(rclconfig, dostem ? stemlang : "",
utf8, reason);
if (!sd) {
PyErr_SetString(PyExc_ValueError, reason.c_str());
return 0;
}
RefCntr<Rcl::SearchData> rq(sd);
self->query->setSortBy(*self->sortfield, self->ascending);
self->query->setQuery(rq);
int cnt = self->query->getResCnt();
self->next = 0;
self->rowcount = cnt;
return Py_BuildValue("i", cnt);
}
PyDoc_STRVAR(doc_Query_executesd,
"executesd(SearchData)\n"
"\n"
"Starts a search for the query defined by the SearchData object.\n"
);
static PyObject *
Query_executesd(recoll_QueryObject* self, PyObject *args, PyObject *kwargs)
{
static const char *kwlist[] = {"searchdata", NULL};
recoll_SearchDataObject *pysd = 0;
LOGDEB(("Query_executeSD\n"));
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "O!:Query_execute",
(char **)kwlist,
&recoll_SearchDataType, &pysd)) {
return 0;
}
if (pysd == 0 || self->query == 0 ||
the_queries.find(self->query) == the_queries.end()) {
PyErr_SetString(PyExc_AttributeError, "query");
return 0;
}
self->query->setSortBy(*self->sortfield, self->ascending);
self->query->setQuery(pysd->sd);
int cnt = self->query->getResCnt();
self->next = 0;
self->rowcount = cnt;
return Py_BuildValue("i", cnt);
}
static void movedocfields(Rcl::Doc *doc)
{
// Move some data from the dedicated fields to the meta array to make
// fetching attributes easier. Is this actually needed ? Useful for
// url which is also formatted .
printableUrl(rclconfig->getDefCharset(), doc->url,
doc->meta[Rcl::Doc::keyurl]);
doc->meta[Rcl::Doc::keytp] = doc->mimetype;
doc->meta[Rcl::Doc::keyipt] = doc->ipath;
doc->meta[Rcl::Doc::keyfs] = doc->fbytes;
doc->meta[Rcl::Doc::keyds] = doc->dbytes;
}
PyDoc_STRVAR(doc_Query_fetchone,
"fetchone(None) -> Doc\n"
"\n"
"Fetches the next Doc object in the current search results.\n"
);
static PyObject *
Query_fetchone(PyObject *_self)
{
recoll_QueryObject* self = (recoll_QueryObject*)_self;
LOGDEB(("Query_fetchone/next\n"));
if (self->query == 0 ||
the_queries.find(self->query) == the_queries.end()) {
PyErr_SetString(PyExc_AttributeError, "query");
return 0;
}
int cnt = self->query->getResCnt();
if (cnt <= 0 || self->next < 0) {
PyErr_SetString(PyExc_AttributeError, "query: no results");
return 0;
}
recoll_DocObject *result =
(recoll_DocObject *)PyObject_CallObject((PyObject *)&recoll_DocType, 0);
if (!result) {
PyErr_SetString(PyExc_EnvironmentError, "doc create failed");
return 0;
}
if (self->next >= self->rowcount) {
PyErr_SetString(PyExc_StopIteration, "End of list reached");
return 0;
}
if (!self->query->getDoc(self->next, *result->doc)) {
PyErr_SetString(PyExc_EnvironmentError, "query: cant fetch result");
self->next = -1;
return 0;
}
self->next++;
movedocfields(result->doc);
return (PyObject *)result;
}
PyDoc_STRVAR(doc_Query_fetchmany,
"fetchmany([size=query.arraysize]) -> Doc list\n"
"\n"
"Fetches the next Doc objects in the current search results.\n"
);
static PyObject *
Query_fetchmany(recoll_QueryObject* self, PyObject *args, PyObject *kwargs)
{
LOGDEB(("Query_fetchmany\n"));
static const char *kwlist[] = {"size", NULL};
int size = 0;
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|i", (char**)kwlist,
&size))
return 0;
if (size == 0)
size = self->arraysize;
if (self->query == 0 ||
the_queries.find(self->query) == the_queries.end()) {
PyErr_SetString(PyExc_AttributeError, "query");
return 0;
}
int cnt = self->query->getResCnt();
if (cnt <= 0 || self->next < 0) {
PyErr_SetString(PyExc_AttributeError, "query: no results");
return 0;
}
PyObject *reslist = PyList_New(0);
int howmany = MIN(self->rowcount - self->next, size);
for (int i = 0; i < howmany; i++) {
recoll_DocObject *docobj = (recoll_DocObject *)
PyObject_CallObject((PyObject *)&recoll_DocType, 0);
if (!docobj) {
PyErr_SetString(PyExc_EnvironmentError, "doc create failed");
return 0;
}
if (!self->query->getDoc(self->next, *docobj->doc)) {
PyErr_SetString(PyExc_EnvironmentError, "can't fetch");
self->next = -1;
return 0;
}
self->next++;
movedocfields(docobj->doc);
PyList_Append(reslist, (PyObject*)docobj);
}
return (PyObject *)reslist;
}
PyDoc_STRVAR(doc_Query_scroll,
"scroll(value, [, mode='relative'/'absolute' ]) -> new int position\n"
"\n"
"Adjusts the position in the current result set.\n"
);
static PyObject *
Query_scroll(recoll_QueryObject* self, PyObject *args, PyObject *kwargs)
{
LOGDEB(("Query_scroll\n"));
static const char *kwlist[] = {"position", "mode", NULL};
int pos = 0;
char *smode = 0;
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "i|s", (char**)kwlist,
&pos, &smode))
return 0;
bool isrelative = 1;
if (smode != 0) {
if (!strcasecmp(smode, "relative")) {
isrelative = 1;
} else if (!strcasecmp(smode, "absolute")) {
isrelative = 0;
} else {
PyErr_SetString(PyExc_ValueError, "bad mode value");
}
}
if (self->query == 0 ||
the_queries.find(self->query) == the_queries.end()) {
PyErr_SetString(PyExc_AttributeError, "query");
return 0;
}
int newpos = isrelative ? self->next + pos : pos;
if (newpos < 0 || newpos >= self->rowcount) {
PyErr_SetString(PyExc_IndexError, "position out of range");
return 0;
}
self->next = newpos;
return Py_BuildValue("i", newpos);
}
PyDoc_STRVAR(doc_Query_highlight,
"highlight(text, ishtml = 0/1, methods = object))\n"
"Will insert <span \"class=rclmatch\"></span> tags around the match areas\n"
"in the input text and return the modified text\n"
"ishtml can be set to indicate that the input text is html and html special\n"
" characters should not be escaped\n"
"methods if set should be an object with methods startMatch(i) and endMatch()\n"
" which will be called for each match and should return a begin and end tag\n"
);
class PyPlainToRich: public PlainToRich {
public:
PyPlainToRich(PyObject *methods, bool eolbr = false)
: m_methods(methods)
{
m_eolbr = eolbr;
}
virtual ~PyPlainToRich()
{
}
virtual string startMatch(unsigned int idx)
{
PyObject *res = 0;
if (m_methods)
res = PyObject_CallMethod(m_methods, (char *)"startMatch",
(char *)"(i)", idx);
if (res == 0)
return "<span class=\"rclmatch\">";
PyObject *res1 = res;
if (PyUnicode_Check(res))
res1 = PyUnicode_AsUTF8String(res);
return PyBytes_AsString(res1);
}
virtual string endMatch()
{
PyObject *res = 0;
if (m_methods)
res = PyObject_CallMethod(m_methods, (char *)"endMatch", 0);
if (res == 0)
return "</span>";
PyObject *res1 = res;
if (PyUnicode_Check(res))
res1 = PyUnicode_AsUTF8String(res);
return PyBytes_AsString(res1);
}
PyObject *m_methods;
};
static PyObject *
Query_highlight(recoll_QueryObject* self, PyObject *args, PyObject *kwargs)
{
LOGDEB1(("Query_highlight\n"));
static const char *kwlist[] = {"text", "ishtml", "eolbr", "methods", NULL};
char *sutf8 = 0; // needs freeing
int ishtml = 0;
PyObject *ishtmlobj = 0;
int eolbr = 1;
PyObject *eolbrobj = 0;
PyObject *methods = 0;
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "es|OOO:Query_highlight",
(char**)kwlist,
"utf-8", &sutf8,
&ishtmlobj,
&eolbrobj,
&methods)) {
return 0;
}
string utf8(sutf8);
PyMem_Free(sutf8);
if (ishtmlobj && PyObject_IsTrue(ishtmlobj))
ishtml = 1;
if (eolbrobj && !PyObject_IsTrue(eolbrobj))
eolbr = 0;
LOGDEB(("Query_highlight: ishtml %d\n", ishtml));
if (self->query == 0 ||
the_queries.find(self->query) == the_queries.end()) {
PyErr_SetString(PyExc_AttributeError, "query");
return 0;
}
RefCntr<Rcl::SearchData> sd = self->query->getSD();
if (sd.isNull()) {
PyErr_SetString(PyExc_ValueError, "Query not initialized");
return 0;
}
HighlightData hldata;
sd->getTerms(hldata);
PyPlainToRich hler(methods, eolbr);
hler.set_inputhtml(ishtml);
list<string> out;
hler.plaintorich(utf8, out, hldata, 5000000);
if (out.empty()) {
PyErr_SetString(PyExc_ValueError, "Plaintorich failed");
return 0;
}
PyObject* unicode = PyUnicode_FromStringAndSize(out.begin()->c_str(),
out.begin()->size());
return Py_BuildValue("u#", PyUnicode_AsUnicode(unicode),
PyUnicode_GetSize(unicode));
}
PyDoc_STRVAR(doc_Query_makedocabstract,
"makedocabstract(doc, methods = object))\n"
"Will create a snippets abstract for doc by selecting text around the match\n"
" terms\n"
"If methods is set, will also perform highlighting. See the highlight method\n"
);
static PyObject *
Query_makedocabstract(recoll_QueryObject* self, PyObject *args,PyObject *kwargs)
{
LOGDEB(("Query_makeDocAbstract\n"));
static const char *kwlist[] = {"doc", "methods", NULL};
recoll_DocObject *pydoc = 0;
PyObject *hlmethods = 0;
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "O!|O:Query_makeDocAbstract",
(char **)kwlist,
&recoll_DocType, &pydoc,
&hlmethods)) {
return 0;
}
if (pydoc->doc == 0 || the_docs.find(pydoc->doc) == the_docs.end()) {
LOGERR(("Query_makeDocAbstract: doc not found %p\n", pydoc->doc));
PyErr_SetString(PyExc_AttributeError, "doc");
return 0;
}
if (the_queries.find(self->query) == the_queries.end()) {
LOGERR(("Query_makeDocAbstract: query not found %p\n", self->query));
PyErr_SetString(PyExc_AttributeError, "query");
return 0;
}
RefCntr<Rcl::SearchData> sd = self->query->getSD();
if (sd.isNull()) {
PyErr_SetString(PyExc_ValueError, "Query not initialized");
return 0;
}
string abstract;
if (hlmethods == 0) {
if (!self->query->makeDocAbstract(*(pydoc->doc), abstract)) {
PyErr_SetString(PyExc_EnvironmentError,
"rcl makeDocAbstract failed");
return 0;
}
} else {
HighlightData hldata;
sd->getTerms(hldata);
PyPlainToRich hler(hlmethods);
hler.set_inputhtml(0);
vector<string> vabs;
self->query->makeDocAbstract(*pydoc->doc, vabs);
for (unsigned int i = 0; i < vabs.size(); i++) {
if (vabs[i].empty())
continue;
list<string> lr;
// There may be data like page numbers before the snippet text.
// will be in brackets.
string::size_type bckt = vabs[i].find("]");
if (bckt == string::npos) {
hler.plaintorich(vabs[i], lr, hldata);
} else {
hler.plaintorich(vabs[i].substr(bckt), lr, hldata);
lr.front() = vabs[i].substr(0, bckt) + lr.front();
}
abstract += lr.front();
abstract += "...";
}
}
// Return a python unicode object
return PyUnicode_Decode(abstract.c_str(), abstract.size(),
"UTF-8", "replace");
}
PyDoc_STRVAR(doc_Query_getxquery,
"getxquery(None) -> Unicode string\n"
"\n"
"Retrieves the Xapian query description as a Unicode string.\n"
"Meaningful only after executexx\n"
);
static PyObject *
Query_getxquery(recoll_QueryObject* self, PyObject *, PyObject *)
{
LOGDEB(("Query_getxquery\n"));
if (self->query == 0 ||
the_queries.find(self->query) == the_queries.end()) {
PyErr_SetString(PyExc_AttributeError, "query");
return 0;
}
RefCntr<Rcl::SearchData> sd = self->query->getSD();
if (sd.isNull()) {
PyErr_SetString(PyExc_ValueError, "Query not initialized");
return 0;
}
string desc = sd->getDescription();
return PyUnicode_Decode(desc.c_str(), desc.size(), "UTF-8", "replace");
}
PyDoc_STRVAR(doc_Query_getgroups,
"getgroups(None) -> a list of pairs\n"
"\n"
"Retrieves the expanded query terms. Meaningful only after executexx\n"
"In each pair, the first entry is a list of user terms, the second a list of\n"
"query terms as derived from the user terms and used in the Xapian Query.\n"
"The size of each list is one for simple terms, or more for group and phrase\n"
"clauses\n"
);
static PyObject *
Query_getgroups(recoll_QueryObject* self, PyObject *, PyObject *)
{
LOGDEB(("Query_getxquery\n"));
if (self->query == 0 ||
the_queries.find(self->query) == the_queries.end()) {
PyErr_SetString(PyExc_AttributeError, "query");
return 0;
}
RefCntr<Rcl::SearchData> sd = self->query->getSD();
if (sd.isNull()) {
PyErr_SetString(PyExc_ValueError, "Query not initialized");
return 0;
}
HighlightData hld;
sd->getTerms(hld);
PyObject *mainlist = PyList_New(0);
PyObject *ulist;
PyObject *xlist;
// We walk the groups vector. For each we retrieve the user group,
// make a python list of each, then group those in a pair, and
// append this to the main list.
for (unsigned int i = 0; i < hld.groups.size(); i++) {
unsigned int ugidx = hld.grpsugidx[i];
ulist = PyList_New(hld.ugroups[ugidx].size());
for (unsigned int j = 0; j < hld.ugroups[ugidx].size(); j++) {
PyList_SetItem(ulist, j,
PyUnicode_Decode(hld.ugroups[ugidx][j].c_str(),
hld.ugroups[ugidx][j].size(),
"UTF-8", "replace"));
}
xlist = PyList_New(hld.groups[i].size());
for (unsigned int j = 0; j < hld.groups[i].size(); j++) {
PyList_SetItem(xlist, j,
PyUnicode_Decode(hld.groups[i][j].c_str(),
hld.groups[i][j].size(),
"UTF-8", "replace"));
}
PyList_Append(mainlist, Py_BuildValue("(OO)", ulist, xlist));
}
return mainlist;
}
static PyMethodDef Query_methods[] = {
{"execute", (PyCFunction)Query_execute, METH_VARARGS|METH_KEYWORDS,
doc_Query_execute},
{"executesd", (PyCFunction)Query_executesd, METH_VARARGS|METH_KEYWORDS,
doc_Query_executesd},
{"next", (PyCFunction)Query_fetchone, METH_NOARGS,
doc_Query_fetchone},
{"fetchone", (PyCFunction)Query_fetchone, METH_NOARGS,
doc_Query_fetchone},
{"fetchmany", (PyCFunction)Query_fetchmany, METH_VARARGS|METH_KEYWORDS,
doc_Query_fetchmany},
{"close", (PyCFunction)Query_close, METH_NOARGS,
doc_Query_close},
{"sortby", (PyCFunction)Query_sortby, METH_VARARGS|METH_KEYWORDS,
doc_Query_sortby},
{"highlight", (PyCFunction)Query_highlight, METH_VARARGS|METH_KEYWORDS,
doc_Query_highlight},
{"getxquery", (PyCFunction)Query_getxquery, METH_NOARGS,
doc_Query_getxquery},
{"getgroups", (PyCFunction)Query_getgroups, METH_NOARGS,
doc_Query_getgroups},
{"makedocabstract", (PyCFunction)Query_makedocabstract,
METH_VARARGS|METH_KEYWORDS, doc_Query_makedocabstract},
{"scroll", (PyCFunction)Query_scroll,
METH_VARARGS|METH_KEYWORDS, doc_Query_scroll},
{NULL} /* Sentinel */
};
static PyMemberDef Query_members[] = {
{(char*)"rownumber", T_INT, offsetof(recoll_QueryObject, next), 0,
(char*)"Next index to be fetched from results. Normally increments after\n"
"each fetchone() call, but can be set/reset before the call effect\n"
"seeking. Starts at 0"
},
{(char*)"rowcount", T_INT, offsetof(recoll_QueryObject, rowcount),
READONLY, (char*)"Number of records returned by the last execute"
},
{(char*)"arraysize", T_INT, offsetof(recoll_QueryObject, arraysize), 0,
(char*)"Default number of records processed by fetchmany (r/w)"
},
{(char*)"connection", T_OBJECT_EX, offsetof(recoll_QueryObject, connection),
0, (char*)"Connection object this is from"
},
{NULL} /* Sentinel */
};
PyDoc_STRVAR(doc_QueryObject,
"Recoll Query objects are used to execute index searches. \n"
"They must be created by the Db.query() method.\n"
);
static PyTypeObject recoll_QueryType = {
PyVarObject_HEAD_INIT(NULL, 0)
"recoll.Query", /*tp_name*/
sizeof(recoll_QueryObject), /*tp_basicsize*/
0, /*tp_itemsize*/
(destructor)Query_dealloc, /*tp_dealloc*/
0, /*tp_print*/
0, /*tp_getattr*/
0, /*tp_setattr*/
0, /*tp_compare*/
0, /*tp_repr*/
0, /*tp_as_number*/
0, /*tp_as_sequence*/
0, /*tp_as_mapping*/
0, /*tp_hash */
0, /*tp_call*/
0, /*tp_str*/
0, /*tp_getattro*/
0, /*tp_setattro*/
0, /*tp_as_buffer*/
Py_TPFLAGS_DEFAULT|Py_TPFLAGS_BASETYPE|Py_TPFLAGS_HAVE_ITER, /*tp_flags*/
doc_QueryObject, /* tp_doc */
0, /* tp_traverse */
0, /* tp_clear */
0, /* tp_richcompare */
0, /* tp_weaklistoffset */
Query_iter, /* tp_iter */
Query_fetchone, /* tp_iternext */
Query_methods, /* tp_methods */
Query_members, /* tp_members */
0, /* tp_getset */
0, /* tp_base */
0, /* tp_dict */
0, /* tp_descr_get */
0, /* tp_descr_set */
0, /* tp_dictoffset */
(initproc)Query_init, /* tp_init */
0, /* tp_alloc */
Query_new, /* tp_new */
};
///////////////////////////////////////////////
////// DB Db object code
typedef struct recoll_DbObject {
PyObject_HEAD
/* Type-specific fields go here. */
Rcl::Db *db;
} recoll_DbObject;
static void
Db_close(recoll_DbObject *self)
{
LOGDEB(("Db_close\n"));
if (self->db)
the_dbs.erase(self->db);
delete self->db;
self->db = 0;
}
static void
Db_dealloc(recoll_DbObject *self)
{
LOGDEB(("Db_dealloc\n"));
Db_close(self);
Py_TYPE(self)->tp_free((PyObject*)self);
}
static PyObject *
Db_new(PyTypeObject *type, PyObject *args, PyObject *kwds)
{
recoll_DbObject *self;
self = (recoll_DbObject *)type->tp_alloc(type, 0);
if (self == 0)
return 0;
self->db = 0;
return (PyObject *)self;
}
static int
Db_init(recoll_DbObject *self, PyObject *args, PyObject *kwargs)
{
static const char *kwlist[] = {"confdir", "extra_dbs", "writable", NULL};
PyObject *extradbs = 0;
char *confdir = 0;
int writable = 0;
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|sOi", (char**)kwlist,
&confdir, &extradbs, &writable))
return -1;
// If the user creates several dbs, changing the confdir, we call
// recollinit repeatedly, which *should* be ok...
string reason;
delete rclconfig;
if (confdir) {
string cfd = confdir;
rclconfig = recollinit(0, 0, reason, &cfd);
} else {
rclconfig = recollinit(0, 0, reason, 0);
}
LOGDEB(("Db_init\n"));
if (rclconfig == 0) {
PyErr_SetString(PyExc_EnvironmentError, reason.c_str());
return -1;
}
if (!rclconfig->ok()) {
PyErr_SetString(PyExc_EnvironmentError, "Bad config ?");
return -1;
}
if (self->db)
the_dbs.erase(self->db);
delete self->db;
self->db = new Rcl::Db(rclconfig);
if (!self->db->open(writable ? Rcl::Db::DbUpd : Rcl::Db::DbRO)) {
LOGERR(("Db_init: db open error\n"));
PyErr_SetString(PyExc_EnvironmentError, "Can't open index");
return -1;
}
if (extradbs) {
if (!PySequence_Check(extradbs)) {
PyErr_SetString(PyExc_TypeError, "extra_dbs must be a sequence");
deleteZ(self->db);
return -1;
}
int dbcnt = PySequence_Size(extradbs);
if (dbcnt == -1) {
PyErr_SetString(PyExc_TypeError, "extra_dbs could not be sized");
deleteZ(self->db);
return -1;
}
for (int i = 0; i < dbcnt; i++) {
PyObject *item = PySequence_GetItem(extradbs, i);
char *s = PyBytes_AsString(item);
Py_DECREF(item);
if (!s) {
PyErr_SetString(PyExc_TypeError,
"extra_dbs must contain strings");
deleteZ(self->db);
return -1;
}
if (!self->db->addQueryDb((const char *)s)) {
PyErr_SetString(PyExc_EnvironmentError,
"extra db could not be opened");
deleteZ(self->db);
return -1;
}
}
}
the_dbs.insert(self->db);
return 0;
}
static PyObject *
Db_query(recoll_DbObject* self)
{
LOGDEB(("Db_query\n"));
if (self->db == 0 || the_dbs.find(self->db) == the_dbs.end()) {
LOGERR(("Db_query: db not found %p\n", self->db));
PyErr_SetString(PyExc_AttributeError, "db");
return 0;
}
recoll_QueryObject *result = (recoll_QueryObject *)
PyObject_CallObject((PyObject *)&recoll_QueryType, 0);
if (!result)
return 0;
result->query = new Rcl::Query(self->db);
result->connection = self;
Py_INCREF(self);
the_queries.insert(result->query);
return (PyObject *)result;
}
static PyObject *
Db_setAbstractParams(recoll_DbObject *self, PyObject *args, PyObject *kwargs)
{
LOGDEB(("Db_setAbstractParams\n"));
static const char *kwlist[] = {"maxchars", "contextwords", NULL};
int ctxwords = -1, maxchars = -1;
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|ii", (char**)kwlist,
&maxchars, &ctxwords))
return 0;
if (self->db == 0 || the_dbs.find(self->db) == the_dbs.end()) {
LOGERR(("Db_query: db not found %p\n", self->db));
PyErr_SetString(PyExc_AttributeError, "db id not found");
return 0;
}
LOGDEB(("Db_setAbstractParams: mxchrs %d, ctxwrds %d\n", maxchars, ctxwords));
self->db->setAbstractParams(-1, maxchars, ctxwords);
Py_RETURN_NONE;
}
static PyObject *
Db_makeDocAbstract(recoll_DbObject* self, PyObject *args)
{
LOGDEB(("Db_makeDocAbstract\n"));
recoll_DocObject *pydoc = 0;
recoll_QueryObject *pyquery = 0;
if (!PyArg_ParseTuple(args, "O!O!:Db_makeDocAbstract",
&recoll_DocType, &pydoc,
&recoll_QueryType, &pyquery)) {
return 0;
}
if (self->db == 0 || the_dbs.find(self->db) == the_dbs.end()) {
LOGERR(("Db_makeDocAbstract: db not found %p\n", self->db));
PyErr_SetString(PyExc_AttributeError, "db");
return 0;
}
if (pydoc->doc == 0 || the_docs.find(pydoc->doc) == the_docs.end()) {
LOGERR(("Db_makeDocAbstract: doc not found %p\n", pydoc->doc));
PyErr_SetString(PyExc_AttributeError, "doc");
return 0;
}
if (pyquery->query == 0 ||
the_queries.find(pyquery->query) == the_queries.end()) {
LOGERR(("Db_makeDocAbstract: query not found %p\n", pyquery->query));
PyErr_SetString(PyExc_AttributeError, "query");
return 0;
}
string abstract;
if (!pyquery->query->makeDocAbstract(*(pydoc->doc), abstract)) {
PyErr_SetString(PyExc_EnvironmentError, "rcl makeDocAbstract failed");
return 0;
}
// Return a python unicode object
return PyUnicode_Decode(abstract.c_str(), abstract.size(),
"UTF-8", "replace");
}
PyDoc_STRVAR(doc_Db_termMatch,
"termMatch(match_type='wildcard|regexp|stem', expr, field='', "
"maxlen=-1, casesens=False, diacsens=False, lang='english')"
" returns the expanded term list\n"
"\n"
"Expands the input expression according to the mode and parameters and "
"returns the expanded term list.\n"
);
static PyObject *
Db_termMatch(recoll_DbObject* self, PyObject *args, PyObject *kwargs)
{
LOGDEB(("Db_termMatch\n"));
static const char *kwlist[] = {"type", "expr", "field", "maxlen",
"casesens", "diacsens", "lang", NULL};
char *tp = 0;
char *expr = 0; // needs freeing
char *field = 0; // needs freeing
int maxlen = -1;
PyObject *casesens = 0;
PyObject *diacsens = 0;
char *lang = 0; // needs freeing
PyObject *ret = 0;
int typ_sens = 0;
Rcl::TermMatchResult result;
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "ses|esiOOes",
(char**)kwlist,
&tp, "utf-8", &expr, "utf-8", &field,
&maxlen, &casesens,
&diacsens, "utf-8", &lang))
return 0;
if (self->db == 0 || the_dbs.find(self->db) == the_dbs.end()) {
LOGERR(("Db_termMatch: db not found %p\n", self->db));
PyErr_SetString(PyExc_AttributeError, "db");
goto out;
}
if (!strcasecmp(tp, "wildcard")) {
typ_sens = Rcl::Db::ET_WILD;
} else if (!strcasecmp(tp, "regexp")) {
typ_sens = Rcl::Db::ET_REGEXP;
} else if (!strcasecmp(tp, "stem")) {
typ_sens = Rcl::Db::ET_STEM;
} else {
PyErr_SetString(PyExc_AttributeError, "Bad type arg");
goto out;
}
if (casesens != 0 && PyObject_IsTrue(casesens)) {
typ_sens |= Rcl::Db::ET_CASESENS;
}
if (diacsens != 0 && PyObject_IsTrue(diacsens)) {
typ_sens |= Rcl::Db::ET_DIACSENS;
}
if (!self->db->termMatch(typ_sens, lang ? lang : "english",
expr, result, maxlen, field ? field : "")) {
LOGERR(("Db_termMatch: db termMatch error\n"));
PyErr_SetString(PyExc_AttributeError, "rcldb termMatch error");
goto out;
}
ret = PyList_New(result.entries.size());
for (unsigned int i = 0; i < result.entries.size(); i++) {
PyList_SetItem(ret, i,
PyUnicode_FromString(
Rcl::strip_prefix(result.entries[i].term).c_str()));
}
out:
PyMem_Free(expr);
PyMem_Free(field);
PyMem_Free(lang);
return ret;
}
static PyObject *
Db_needUpdate(recoll_DbObject* self, PyObject *args, PyObject *kwds)
{
char *udi = 0; // needs freeing
char *sig = 0; // needs freeing
LOGDEB(("Db_needUpdate\n"));
if (!PyArg_ParseTuple(args, "eses:Db_needUpdate",
"utf-8", &udi, "utf-8", &sig)) {
return 0;
}
if (self->db == 0 || the_dbs.find(self->db) == the_dbs.end()) {
LOGERR(("Db_needUpdate: db not found %p\n", self->db));
PyErr_SetString(PyExc_AttributeError, "db");
PyMem_Free(udi);
PyMem_Free(sig);
return 0;
}
bool result = self->db->needUpdate(udi, sig);
PyMem_Free(udi);
PyMem_Free(sig);
return Py_BuildValue("i", result);
}
static PyObject *
Db_delete(recoll_DbObject* self, PyObject *args, PyObject *kwds)
{
char *udi = 0; // needs freeing
LOGDEB(("Db_delete\n"));
if (!PyArg_ParseTuple(args, "es:Db_delete", "utf-8", &udi)) {
return 0;
}
if (self->db == 0 || the_dbs.find(self->db) == the_dbs.end()) {
LOGERR(("Db_delete: db not found %p\n", self->db));
PyErr_SetString(PyExc_AttributeError, "db");
PyMem_Free(udi);
return 0;
}
bool result = self->db->purgeFile(udi);
PyMem_Free(udi);
return Py_BuildValue("i", result);
}
static PyObject *
Db_purge(recoll_DbObject* self)
{
if (self->db == 0 || the_dbs.find(self->db) == the_dbs.end()) {
LOGERR(("Db_purge: db not found %p\n", self->db));
PyErr_SetString(PyExc_AttributeError, "db");
return 0;
}
bool result = self->db->purge();
return Py_BuildValue("i", result);
}
static PyObject *
Db_addOrUpdate(recoll_DbObject* self, PyObject *args, PyObject *)
{
LOGDEB(("Db_addOrUpdate\n"));
char *sudi = 0; // needs freeing
char *sparent_udi = 0; // needs freeing
recoll_DocObject *pydoc;
if (!PyArg_ParseTuple(args, "esO!|es:Db_addOrUpdate",
"utf-8", &sudi, &recoll_DocType, &pydoc,
"utf-8", &sparent_udi)) {
return 0;
}
string udi(sudi);
string parent_udi(sparent_udi ? sparent_udi : "");
PyMem_Free(sudi);
PyMem_Free(sparent_udi);
if (self->db == 0 || the_dbs.find(self->db) == the_dbs.end()) {
LOGERR(("Db_addOrUpdate: db not found %p\n", self->db));
PyErr_SetString(PyExc_AttributeError, "db");
return 0;
}
if (pydoc->doc == 0 || the_docs.find(pydoc->doc) == the_docs.end()) {
LOGERR(("Db_addOrUpdate: doc not found %p\n", pydoc->doc));
PyErr_SetString(PyExc_AttributeError, "doc");
return 0;
}
if (!self->db->addOrUpdate(udi, parent_udi, *pydoc->doc)) {
LOGERR(("Db_addOrUpdate: rcldb error\n"));
PyErr_SetString(PyExc_AttributeError, "rcldb error");
return 0;
}
Py_RETURN_NONE;
}
static PyMethodDef Db_methods[] = {
{"close", (PyCFunction)Db_close, METH_NOARGS,
"close() closes the index connection. The object is unusable after this."
},
{"query", (PyCFunction)Db_query, METH_NOARGS,
"query() -> Query. Return a new, blank query object for this index."
},
{"cursor", (PyCFunction)Db_query, METH_NOARGS,
"cursor() -> Query. Alias for query(). Return query object."
},
{"setAbstractParams", (PyCFunction)Db_setAbstractParams,
METH_VARARGS|METH_KEYWORDS,
"setAbstractParams(maxchars, contextwords).\n"
"Set the parameters used to build 'keyword-in-context' abstracts"
},
{"makeDocAbstract", (PyCFunction)Db_makeDocAbstract, METH_VARARGS,
"makeDocAbstract(Doc, Query) -> string\n"
"Build and return 'keyword-in-context' abstract for document\n"
"and query."
},
{"termMatch", (PyCFunction)Db_termMatch, METH_VARARGS|METH_KEYWORDS,
doc_Db_termMatch
},
{"needUpdate", (PyCFunction)Db_needUpdate, METH_VARARGS,
"needUpdate(udi, sig) -> Bool.\n"
"Check if the index is up to date for the document defined by udi,\n"
"having the current signature sig."
},
{"delete", (PyCFunction)Db_delete, METH_VARARGS,
"delete(udi) -> Bool.\n"
"Purge index from all data for udi. If udi matches a container\n"
"document, purge all subdocs (docs with a parent_udi matching udi)."
},
{"purge", (PyCFunction)Db_purge, METH_NOARGS,
"purge() -> Bool.\n"
"Delete all documents that were not touched during the just finished\n"
"indexing pass (since open-for-write). These are the documents for\n"
"the needUpdate() call was not performed, indicating that they no\n"
"longer exist in the primary storage system.\n"
},
{"addOrUpdate", (PyCFunction)Db_addOrUpdate, METH_VARARGS,
"addOrUpdate(udi, doc, parent_udi=None) -> None\n"
"Add or update index data for a given document\n"
"The udi string must define a unique id for the document. It is not\n"
"interpreted inside Recoll\n"
"doc is a Doc object\n"
"if parent_udi is set, this is a unique identifier for the\n"
"top-level container (ie mbox file)"
},
{NULL} /* Sentinel */
};
PyDoc_STRVAR(doc_DbObject,
"Db([confdir=None], [extra_dbs=None], [writable = False])\n"
"\n"
"A Db object holds a connection to a Recoll index. Use the connect()\n"
"function to create one.\n"
"confdir specifies a Recoll configuration directory (default: \n"
" $RECOLL_CONFDIR or ~/.recoll).\n"
"extra_dbs is a list of external databases (xapian directories)\n"
"writable decides if we can index new data through this connection\n"
);
static PyTypeObject recoll_DbType = {
PyVarObject_HEAD_INIT(NULL, 0)
"recoll.Db", /*tp_name*/
sizeof(recoll_DbObject), /*tp_basicsize*/
0, /*tp_itemsize*/
(destructor)Db_dealloc, /*tp_dealloc*/
0, /*tp_print*/
0, /*tp_getattr*/
0, /*tp_setattr*/
0, /*tp_compare*/
0, /*tp_repr*/
0, /*tp_as_number*/
0, /*tp_as_sequence*/
0, /*tp_as_mapping*/
0, /*tp_hash */
0, /*tp_call*/
0, /*tp_str*/
0, /*tp_getattro*/
0, /*tp_setattro*/
0, /*tp_as_buffer*/
Py_TPFLAGS_DEFAULT|Py_TPFLAGS_BASETYPE, /*tp_flags*/
doc_DbObject, /* tp_doc */
0, /* tp_traverse */
0, /* tp_clear */
0, /* tp_richcompare */
0, /* tp_weaklistoffset */
0, /* tp_iter */
0, /* tp_iternext */
Db_methods, /* tp_methods */
0, /* tp_members */
0, /* tp_getset */
0, /* tp_base */
0, /* tp_dict */
0, /* tp_descr_get */
0, /* tp_descr_set */
0, /* tp_dictoffset */
(initproc)Db_init, /* tp_init */
0, /* tp_alloc */
Db_new, /* tp_new */
};
//////////////////////////////////////////////////////////////////////////
// Module methods
static PyObject *
recoll_connect(PyObject *self, PyObject *args, PyObject *kwargs)
{
recoll_DbObject *db = (recoll_DbObject *)
PyObject_Call((PyObject *)&recoll_DbType, args, kwargs);
return (PyObject *)db;
}
PyDoc_STRVAR(doc_connect,
"connect([confdir=None], [extra_dbs=None], [writable = False])\n"
" -> Db.\n"
"\n"
"Connects to a Recoll database and returns a Db object.\n"
"confdir specifies a Recoll configuration directory\n"
"(the default is built like for any Recoll program).\n"
"extra_dbs is a list of external databases (xapian directories)\n"
"writable decides if we can index new data through this connection\n"
);
static PyMethodDef recoll_methods[] = {
{"connect", (PyCFunction)recoll_connect, METH_VARARGS|METH_KEYWORDS,
doc_connect},
{NULL, NULL, 0, NULL} /* Sentinel */
};
PyDoc_STRVAR(pyrecoll_doc_string,
"This is an interface to the Recoll full text indexer.");
struct module_state {
PyObject *error;
};
#if PY_MAJOR_VERSION >= 3
#define GETSTATE(m) ((struct module_state*)PyModule_GetState(m))
#else
#define GETSTATE(m) (&_state)
static struct module_state _state;
#endif
#if PY_MAJOR_VERSION >= 3
static int recoll_traverse(PyObject *m, visitproc visit, void *arg) {
Py_VISIT(GETSTATE(m)->error);
return 0;
}
static int recoll_clear(PyObject *m) {
Py_CLEAR(GETSTATE(m)->error);
return 0;
}
static struct PyModuleDef moduledef = {
PyModuleDef_HEAD_INIT,
"recoll",
NULL,
sizeof(struct module_state),
recoll_methods,
NULL,
recoll_traverse,
recoll_clear,
NULL
};
#define INITERROR return NULL
extern "C" PyObject *
PyInit_recoll(void)
#else
#define INITERROR return
PyMODINIT_FUNC
initrecoll(void)
#endif
{
// Note: we can't call recollinit here, because the confdir is only really
// known when the first db object is created (it is an optional parameter).
// Using a default here may end up with variables such as stripchars being
// wrong
#if PY_MAJOR_VERSION >= 3
PyObject *module = PyModule_Create(&moduledef);
#else
PyObject *module = Py_InitModule("recoll", recoll_methods);
#endif
if (module == NULL)
INITERROR;
struct module_state *st = GETSTATE(module);
// The first parameter is a char *. Hopefully we don't initialize
// modules too often...
st->error = PyErr_NewException(strdup("recoll.Error"), NULL, NULL);
if (st->error == NULL) {
Py_DECREF(module);
INITERROR;
}
if (PyType_Ready(&recoll_DbType) < 0)
INITERROR;
Py_INCREF((PyObject*)&recoll_DbType);
PyModule_AddObject(module, "Db", (PyObject *)&recoll_DbType);
if (PyType_Ready(&recoll_QueryType) < 0)
INITERROR;
Py_INCREF((PyObject*)&recoll_QueryType);
PyModule_AddObject(module, "Query", (PyObject *)&recoll_QueryType);
if (PyType_Ready(&recoll_DocType) < 0)
INITERROR;
Py_INCREF((PyObject*)&recoll_DocType);
PyModule_AddObject(module, "Doc", (PyObject *)&recoll_DocType);
if (PyType_Ready(&recoll_SearchDataType) < 0)
INITERROR;
Py_INCREF((PyObject*)&recoll_SearchDataType);
PyModule_AddObject(module, "SearchData",
(PyObject *)&recoll_SearchDataType);
PyModule_AddStringConstant(module, "__doc__",
pyrecoll_doc_string);
PyObject *doctypecobject;
#if PY_MAJOR_VERSION >= 3 || (PY_MAJOR_VERSION >= 2 && PY_MINOR_VERSION >= 7)
// Export a few pointers for the benefit of other recoll python modules
doctypecobject=
PyCapsule_New(&recoll_DocType, PYRECOLL_PACKAGE "recoll.doctypeptr", 0);
#else
doctypecobject = PyCObject_FromVoidPtr(&recoll_DocType, NULL);
#endif
PyModule_AddObject(module, "doctypeptr", doctypecobject);
#if PY_MAJOR_VERSION >= 3
return module;
#endif
}