Mercurial > hg > ChinaGisRestApi
changeset 14:05933707897f
NEW - # 12: create table and upload data
https://it-dev.mpiwg-berlin.mpg.de/tracs/GIS/ticket/12
author | casties |
---|---|
date | Fri, 13 Aug 2010 18:34:46 +0200 |
parents | e2c73c077533 |
children | 5e3edf980813 |
files | RestDbInterface.py |
diffstat | 1 files changed, 170 insertions(+), 0 deletions(-) [+] |
line wrap: on
line diff
--- a/RestDbInterface.py Thu Aug 12 20:04:57 2010 +0200 +++ b/RestDbInterface.py Fri Aug 13 18:34:46 2010 +0200 @@ -19,6 +19,18 @@ from Products.ZSQLMethods.SQL import SQLConnectionIDs +def getTextFromNode(nodename): + """get the cdata content of a XML node""" + if nodename is None: + return "" + nodelist=nodename.childNodes + rc = "" + for node in nodelist: + if node.nodeType == node.TEXT_NODE: + rc = rc + node.data + return rc + + class RestDbInterface(Folder): """Object for RESTful database queries path schema: /db/{schema}/{table}/ @@ -237,10 +249,168 @@ # TODO: really look up schemas data={'fields': (('schemas',),), 'rows': [('public',),]} return data + + def createEmptyTable(self,schema,table,fields): + """create a table with the given fields + returns list of created fields""" + sqlFields = [] + for f in fields: + if isinstance(f,dict): + # {name: XX, type: YY} + name = sqlName(f['name']) + type = f['type'] + + else: + # name only + name = sqlName(f) + type = 'text' + + sqlFields.append({'name':name, 'type':type}) + + # drop table if it exists + try: + res = executeSQL('drop table "%s.%s"'%(schema,table)) + except: + pass + + fieldString = ", ".join(["%s %s"%(f['name'],f['type']) for f in sqlFields]) + sqlString = 'create table "%s.%s" (%s)'%(schema,table,fieldString) + ret = executeSQL(sqlString) + + return sqlFields + + def createTableFromXML(self,schema,table,data): """create or replace a table with the given XML data""" logging.debug("createTableFromXML schema=%s table=%s data=%s"%(schema,table,data)) + self.importExcelXML(table, data) + + def importExcelXML(self,table,data,fieldNamesOnly=False): + ''' + Import XML file in Excel format into the table + @param table: name of the table the xml shall be imported into + @param containerTagName: XML-Tag which describes a dataset + @param data: data to be imported + @param identify: (optional) field res. tag which identifies a entry uniquely for updating purposes. + @param RESPONSE: (optional) + ''' + from xml.dom.pulldom import parseString,parse + + namespace = "urn:schemas-microsoft-com:office:spreadsheet" + containerTagName = "Table" + rowTagName = "Row" + dataTagName = "Data" + fieldNames = [] + sqlFields = [] + sqlInsert = None + + logging.debug("import excel xml") + + ret="" + if isinstance(data, str): + logging.debug("importXML reading string data") + doc=parseString(data) + else: + logging.debug("importXML reading file data") + doc=parse(data) + + cnt = 0 + while True: + node=doc.getEvent() + + if node is None: + break + + else: + if node[1].localName is not None: + tagName = node[1].localName.lower() + else: + # ignore non-tag nodes + continue + + if tagName == containerTagName.lower(): # make everything case insensitive + # start of container -- do we need this? + cnt = 0 + continue + + if tagName == rowTagName.lower(): + # start of row + doc.expandNode(node[1]) + cnt += 1 + if cnt == 1: + # first row -- field names + names=node[1].getElementsByTagNameNS(namespace, dataTagName) + for name in names: + fn = getTextFromNode(name) + fieldNames.append(fn) + + if fieldNamesOnly: + # return just field names + return fieldNames + + # create table + sqlFields = self.createEmptyTable(schema, table, fieldNames) + fieldString = ", ".join(["%s"%f['name'] for f in sqlFields]) + valString = ", ".join(["%s" for f in sqlFields]) + sqlInsert = 'insert into "%s.%s" (%s) values (%s)'%(schema,table,fieldString,valString) + + else: + # following rows are data + dataNodes=node[1].getElementsByTagNameNS(namespace, dataTagName) + data = [] + for dataNode in dataNodes: + val = getTextFromNode(dataNodes) + data.append(val) + + self.executeSQL(sqlInsert, data) + + + continue + + '''dataSet={} + for col in node[1].childNodes: + if col.nodeType is col.ELEMENT_NODE: + data=col.nodeName + dataSet[data]=getTextFromNode(col) + + update=False + + if identify: + + field=dataSet[identify] + + searchStr="""select %s from %s where %s = '%s'"""%(identify,table,identify,field) + logger("import xml",logging.INFO,searchStr) + search=self.ZSQLSimpleSearch(searchStr) + if search: + update=True + + if update: + tmp=[] + for fieldName in dataSet.keys(): + tmp.append("""%s = %s"""%(fieldName,self.ZSQLQuote(dataSet[fieldName]))) + setStr=",".join(tmp) + + field=dataSet[identify] + + queryStr="""UPDATE %s SET %s WHERE %s = '%s' """%(table,setStr,identify,field) + logger("update xml",logging.INFO,queryStr) + self.ZSQLSimpleSearch(queryStr) + ret+="ud: %s \n"%field + + else: + fields=",".join(dataSet.keys()) + values=",".join([""" %s """%self.ZSQLQuote(dataSet[x]) for x in dataSet.keys()]) + + + queryStr="""INSERT INTO %s (%s) VALUES (%s)"""%(table,fields,values) + self.ZSQLSimpleSearch(queryStr) + logger("update xml",logging.INFO,queryStr) + ''' + + logging.debug("XML import field names: %s"%(fieldNames)) + return ret def manage_editRestDbInterface(self, title=None, connection_id=None,