Exemple #1
0
class MongoThread(QThread):  # Thread for access MongoDB.

    sig1 = app_signal(object, str)  # Signal emit with data/error.

    def __init__(self, parent=None):  # Importing of the pymongo
        super(MongoThread, self).__init__(parent)

    def on_source(self, datas):  # adapter for connection.
        self.datas = datas  # Receives the data as

    def run(self):  # dictionary and inserting to
        try:  # the mongo database. The
            import pymongo  # datas received contains
            try:  # insert or select string that
                self.client = pymongo.MongoClient('localhost', 27017)
                self.db = self.client['umongodb']
            except pymongo.errors as err:  # used for which  operation
                self.sig1.emit('', str(err))  # need to do. If insert - this
            if self.datas[1] == "insert":  # data will be inserted to the
                posts = self.db.posts  # database. If the value of
                posts.insert_one(self.datas[0])
            if self.datas[1] == "select":  # datas is select - will find
                dbdata = self.db.posts.find()  # in the posts and emits this
                self.sig1.emit(dbdata, '')  # data or error if occured.
            if self.isFinished():  # Checks if this thread is
                self.quit()  # finished and quits thread.
        except Exception as err:  # If some error will occur -
            self.sig1.emit('', str(err))  # emits this error to app,
            if self.isFinished():  # checks, if the thread is
                self.quit()  # finished, quits this thread.
Exemple #2
0
class RPandas(QThread):  # Thread for reading CSV file

    sig1 = app_signal(object, str)  # with Pandas. Emits an object

    def __init__(self, parent=None):  # (dataframe). Will read the
        super(RPandas, self).__init__(parent)  # file that specified with

    def on_source(self, datas):  # type of the file with an
        self.datas = datas  # representation to the pandas

    def run(self):  # dataframe that can be opened
        try:  # as csv file data, excel file
            import pandas  # data, html file data with
            if self.datas[1] == "csv":  # tags and can be used as html
                df = pandas.read_csv(self.datas[0],
                                     chunksize=self.datas[6],
                                     engine=self.datas[4])
            if self.datas[1] == "excel":  # page, json representation.
                df = pandas.read_excel(self.datas[0])
            if self.datas[1] == "html":  # Also, sets the options to
                df = pandas.read_html(self.datas[0])
            if self.datas[1] == "json":  # display number of rows.
                df = pandas.read_json(self.datas[0])
            pandas.options.display.max_rows = self.datas[5]
            for ch in df:  # Returns data and reader that
                self.sig1.emit(ch, '')  # used as chunk sizes to split
                time.sleep(0.1)  # reading of the file data.
            if self.isFinished():  # Checks if this thread is
                self.quit()  # finished and quits thread.
        except Exception as err:  # If some error will occur -
            self.sig1.emit('', str(err))  # emits this error to app,
            if self.isFinished():  # checks, if the thread is
                self.quit()  # finished, quits this thread.
Exemple #3
0
class CouchThread(QThread):  # Thread for access CouchDB.

    sig1 = app_signal(object, str)  # Signal will emit data/error.

    def __init__(self, parent=None):  # Importing of the couchdb
        super(CouchThread, self).__init__(parent)

    def on_source(self, datas):  # adapter for connection. Then
        self.datas = datas  # receives the data as a

    def run(self):  # dictionary and inserting to
        try:  # the couch database. The
            import couchdb  # datas received contains
            try:  # insert or select string that
                self.couch = couchdb.Server("http://127.0.0.1:5984/")
                self.db = self.couch["u_couchdb"]
            except Exception as err:  # used for which operation
                self.sig1.emit('', str(err))  # need to do. If insert - this
            if self.datas[1] == "insert":  # data will be inserted to the
                self.db.save(self.datas[0])  # couch database If select
            if self.datas[1] == "select":  # value - emits the couchdb
                self.sig1.emit(self.db, '')  # data or error if occurs.
            if self.isFinished():  # Checks if this thread is
                self.quit()  # finished and quits thread.
        except Exception as err:  # If some error will occur -
            self.sig1.emit('', str(err))  # emits this error to app,
            if self.isFinished():  # checks, if the thread is
                self.quit()  # finished, quits this thread.
Exemple #4
0
class WPandas(QThread):  # Thread for writing file with

    sig1 = app_signal(object, str)  # Pandas. Signal for emitting

    def __init__(self, parent=None):  # from this thread. Will write
        super(WPandas, self).__init__(parent)  # the pandas csv, excel,

    def on_source(self, datas):  # specified in the writer
        self.datas = datas  # parameter. The 'if'

    def run(self):  # statements checks the type
        try:  # of file to write with using
            import pandas  # pandas functions. List of
            uindex = [i for i in range(self.datas[2])]
            udata = {
                "User_Name": range(0, self.datas[2]),
                "User_email": range(0, self.datas[2]),
                "User_password": range(0, self.datas[2]),
                "User_data": range(0, self.datas[2])
            }
            df = pandas.DataFrame(udata,
                                  columns=self.datas[3],
                                  index=uindex,
                                  dtype=self.datas[5])
            df.index.name = "rows\columns"  # indexes. The data is a dict
            if self.datas[1] == "csv":  # with specified keys and
                df.to_csv(self.datas[0])  # columns, value pairs of
            if self.datas[1] == "excel":  # these keys, The index list
                df.to_excel(self.datas[0])  # is the number of rows. Data
            if self.datas[1] == "html":  # can be written to the excel,
                df.to_html(self.datas[0])  # html, json files also. The
            if self.datas[1] == "json":  # user data can be changed
                df.to_json(self.datas[0])  # to the real world data. Then
            if self.isFinished():  # checks if this thread is
                self.quit()  # finished and quits thread.
        except Exception as err:  # If some error will occur -
            self.sig1.emit('', str(err))  # emits this error to app,
            if self.isFinished():  # Checks, if the thread is
                self.quit()  # finished, quits this thread.
Exemple #5
0
class UTools(object):  # Class forsome functionality.

    pandas_sig1 = app_signal(list)  # Signal for pandas writing.
    pandas_sig2 = app_signal(list)  # Signal for pandas reading.
    mongo_sig1 = app_signal(list)  # Signals for communication
    mongo_sig2 = app_signal(list)  # with thread for MongoDB.
    couch_sig1 = app_signal(list)  # Signals for communication
    couch_sig2 = app_signal(list)  # with thread for CouchDB.

    def __init__(self):  # Constructor of the class.
        self.us1 = "Application with Python."  # String variable.
        self.qtsql_db = QtSql.QSqlDatabase.addDatabase('QSQLITE')
        self.qtsql_db.setDatabaseName("data/u_db.sqlite")
        self.qtsql_db.open()  # Path to DB. Open/Create DB.

    def pickle_dump(self, obj=None, filename=None, proto=1, fimps=None):
        fobj = open(filename, 'wb')  # File object in binary mode.
        if fimps is not None:  # The fix_imports parameter.
            pickle.dump(obj, fobj, proto, fimps)
        else:  # Maps the new Python3 names,
            pickle.dump(obj, fobj)  # Python2 old module names,
        fobj.close()  # not used with Python2.

    def pickle_load(self,
                    filename=None,
                    fimps=None,
                    enc="ASCII",
                    err="strict"):
        fobj = open(filename, 'rb')  # Loads the file with data,
        if fimps is not None:  # fix_imports is not None,
            pl = pickle.load(fobj, fimps, encoding=enc, errors=err)
        else:  # parameters for Python3,
            pl = pickle.load(fobj)  # defaults for Python3.
        fobj.close()  # Used for Python2.
        return pl  # Returns loaded data.

    def shelve_open(self,
                    filename=None,
                    flag='c',
                    proto=None,
                    writeback=False):
        return shelve.open(filename, flag, proto, writeback)

    def csv_write(self,
                  csvfile=None,
                  fieldnames=None,
                  mode='a',
                  newline='',
                  rowdata=None,
                  dialect=csv.excel,
                  delimiter=' ',
                  quotechar='|',
                  lineterminator='\r\n'):  # Function to write CSV.
        csvpath = os.path.exists(r"%s/%s" % (pathf, csvfile))
        csvf = open(csvfile, mode)  # Opens file in append mode.
        csvw = csv.DictWriter(csvf,
                              fieldnames=fieldnames,
                              dialect=dialect,
                              delimiter=delimiter,
                              lineterminator=lineterminator)
        if csvpath is False:  # If path of the file with
            csvw.writeheader()  # writer exist - add header.
        csvdict = {k: v for (k, v) in zip(fieldnames, rowdata)}
        csvw.writerow(csvdict)  # Writing row with dictionary.

    def csv_read(self, csvfile=None, mode='r', dialect=csv.excel):
        csvf = open(csvfile, mode)  # Opens file object (reading).
        csvr = csv.DictReader(csvf, dialect=dialect, delimiter=' ')
        return csvr  # CSV dict reader object.

    def pandas_write(self,
                     filename=None,
                     writer="csv",
                     data=None,
                     columns=None,
                     index=None,
                     dtype=object):
        data = 1000000  # Function will write data
        index = 1000000  # with Pandas. Integers are
        datas = [filename, writer, data, columns, index, dtype]
        self.pandas_thread1 = WPandas()  # used to generate values for
        self.pandas_sig1.connect(self.pandas_thread1.on_source)
        self.pandas_sig1.emit(datas)  # each cell of the CSV file,
        self.pandas_thread1.start()  # using thread, 1000000 rows.

    def pandas_read(self,
                    filename=None,
                    reader="csv",
                    sep=',',
                    delimiter=None,
                    engine='python',
                    maxrows=9,
                    chunk=10000):
        datas = [filename, reader, sep, delimiter, engine, maxrows, chunk]
        self.pandas_thread2 = RPandas()  # Function reads the data
        self.pandas_sig2.connect(self.pandas_thread2.on_source)
        self.pandas_sig2.emit(datas)  # from the CSV file with
        self.pandas_thread2.start()  # Pandas using thread which
        return self.pandas_thread2  # returned for operating.

    def sqlite_insert(self, username=None, email=None, passw=None, data=None):
        dbcon = sqlite3.connect("%s/data/u_db.sqlite" % pathf)
        dbcursor = dbcon.cursor()  # Connect to the database.
        sql_query = """INSERT INTO u_dbtable (user_name, user_email,
                        user_passw, user_data) VALUES (?, ?, ?, ?);"""
        udata = (str(username), str(email), str(passw), str(data))
        dbcursor.execute(sql_query, udata)  # Executes the SQL query that
        dbcon.commit()  # SQL text, retreived data,
        dbcursor.close()  # commit transaction then
        dbcon.close()  # closes cursor and database.

    def sqlite_select(self):  # Connect to the SQLite DB
        dbcon = sqlite3.connect("%s/data/u_db.sqlite" % pathf)
        dbcursor = dbcon.cursor()  # and open db cursor.
        sql_query = """SELECT info_date, user_name, user_email, user_passw,
                      user_data FROM u_dbtable;"""
        dbcursor.execute(sql_query)  # Executes the SQL query
        dbdata = dbcursor.fetchall()  # specified in the parameter,
        dbcursor.close()  # fetch all data selected,
        dbcon.close()  # closes cursor and database.
        return dbdata  # Returns fetched data.

    def psql_insert(self, username=None, email=None, passw=None, data=None):
        dbcon = psycopg2.connect(dbname="psqldb")
        dbcursor = dbcon.cursor()  # Connects to the PostgreSQL.
        sql_query = """INSERT INTO u_dbtable (user_name, user_email,
                        user_passw, user_data) VALUES (%s, %s, %s, %s);"""
        udata = (str(username), str(email), str(passw), str(data))
        dbcursor.execute(sql_query, udata)  # Executes the SQL query that
        dbcon.commit()  # specified in the parameter,
        dbcursor.close()  # commit transaction then
        dbcon.close()  # closes cursor and database.

    def psql_select(self):  # Connect to the PostgreSQL
        dbcon = psycopg2.connect(dbname="psqldb")
        dbcursor = dbcon.cursor()  # database and open db cursor.
        sql_query = """SELECT info_date, user_name, user_email, user_passw,
                       user_data FROM u_dbtable;"""
        dbcursor.execute(sql_query)  # Executes the SQL query that
        dbdata = dbcursor.fetchall()  # specified in the parameter
        dbcursor.close()  # fetch all data selected,
        dbcon.close()  # closes cursor and database.
        return dbdata  # Returns fetched data.

    def mysql_insert(self, username=None, email=None, passw=None, data=None):
        dbcon = pymysql.connect(host="localhost",
                                user="******",
                                password="******",
                                db="mysqldb")
        dbcursor = dbcon.cursor()  # Connects to MySQL database.
        sql_query = """INSERT INTO u_dbtable (user_name, user_email,
                        user_passw, user_data) VALUES (%s, %s, %s, %s);"""
        udata = (str(username), str(email), str(passw), str(data))
        dbcursor.execute(sql_query, udata)  # Executes the SQL query that
        dbcon.commit()  # specified in the parameter.
        dbcursor.close()  # commit transaction then
        dbcon.close()  # closes cursor and database.

    def mysql_select(self):  # Connects to MySQL database.
        dbcon = pymysql.connect(host="localhost",
                                user="******",
                                password="******",
                                db="mysqldb")
        dbcursor = dbcon.cursor()  # Values with open db cursor.
        sql_query = """SELECT info_date, user_name, user_email, user_passw,
                       user_data FROM u_dbtable;"""
        dbcursor.execute(sql_query)  # Executes the SQL query that
        dbdata = dbcursor.fetchall()  # specified in the parameter
        dbcursor.close()  # fetch all data selected,
        dbcon.close()  # closes cursor and database.
        return dbdata  # Returns fetched data.

    def sql_qt_insert(self, username=None, email=None, passw=None, data=None):
        query = QtSql.QSqlQuery(self.qtsql_db)
        query.prepare("""INSERT INTO u_dbtable (user_name, user_email,
                      user_passw, user_data) VALUES (:user_name, :user_email,
                     :user_passw, :user_data);""")
        query.bindValue(":user_name", username)
        query.bindValue(":user_email", email)
        query.bindValue(":user_passw", passw)  # Inserts values to the sqlite
        query.bindValue(":user_data", data)  # database with QtSql module.
        query.exec_()  # Executes the query to DB.

    def sql_qt_select(self):  # Function for selection
        query = QtSql.QSqlQuery(self.qtsql_db)
        query.exec_("""SELECT * FROM u_dbtable;""")
        return query  # values, QtSql module query.

    def mongo_insert(self, username=None, email=None, passw=None, data=None):
        datas = [{
            "User Name": username,
            "Email": email,
            "Password": passw,
            "Data": data
        }, "insert"]  # Inserting values to the
        self.mongo_thread1 = MongoThread()  # Mongo database using thread
        self.mongo_sig1.connect(self.mongo_thread1.on_source)
        self.mongo_sig1.emit(datas)  # and emitting of the data
        self.mongo_thread1.start()  # specified in the list.

    def mongo_select(self):  # Selection values from
        datas = [{}, "select"]  # MongoDB using thread
        self.mongo_thread2 = MongoThread()  # for reading. Sending to
        self.mongo_sig2.connect(self.mongo_thread2.on_source)
        self.mongo_sig2.emit(datas)  # thread the empty dict and
        self.mongo_thread2.start()  # and second select string,
        return self.mongo_thread2  # then returning thread.

    def couch_insert(self, username=None, email=None, passw=None, data=None):
        datas = [{
            "User Name": username,
            "User email": email,
            "User password": passw,
            "User Data": data
        }, "insert"]
        self.couch_thread1 = CouchThread()  # Function for writing
        self.couch_sig1.connect(self.couch_thread1.on_source)
        self.couch_sig1.emit(datas)  # to the Couch database with
        self.couch_thread1.start()  # data specified in the list.

    def couch_select(self):  # Selection values from
        datas = [{}, "select"]  # CouchDB with using
        self.couch_thread2 = CouchThread()  # thread. Returns the
        self.couch_sig2.connect(self.couch_thread2.on_source)
        self.couch_sig2.emit(datas)  # database that will be
        self.couch_thread2.start()  # used to get the values
        return self.couch_thread2  # and insert to text field.