mirror of
https://github.com/Tautulli/Tautulli.git
synced 2025-07-06 21:21:15 -07:00
Clean up datatable query with union table
This commit is contained in:
parent
54af92251c
commit
17a7019c60
2 changed files with 172 additions and 203 deletions
|
@ -46,17 +46,6 @@ class DataTables(object):
|
|||
logger.error('PlexPy DataTables :: No table name received.')
|
||||
return None
|
||||
|
||||
# Set default variable values
|
||||
parameters = {}
|
||||
args = []
|
||||
group = ''
|
||||
group_u = ''
|
||||
order = ''
|
||||
where = ''
|
||||
join = ''
|
||||
c_where = ''
|
||||
c_where_u = ''
|
||||
|
||||
# Fetch all our parameters
|
||||
if kwargs.get('json_data'):
|
||||
parameters = helpers.process_json_kwargs(json_kwargs=kwargs.get('json_data'))
|
||||
|
@ -65,93 +54,27 @@ class DataTables(object):
|
|||
'named json_data.')
|
||||
return None
|
||||
|
||||
dt_columns = parameters['columns']
|
||||
extracted_columns = self.extract_columns(columns=columns)
|
||||
extracted_columns_union = self.extract_columns(columns=columns_union)
|
||||
join = self.build_join(join_types, join_tables, join_evals)
|
||||
group = self.build_grouping(group_by)
|
||||
c_where, cw_args = self.build_custom_where(custom_where)
|
||||
order = self.build_order(parameters['order'],
|
||||
extracted_columns['column_named'],
|
||||
parameters['columns'])
|
||||
where, w_args = self.build_where(parameters['search']['value'],
|
||||
extracted_columns['column_named'],
|
||||
parameters['columns'])
|
||||
|
||||
# Build grouping
|
||||
if group_by:
|
||||
for g in group_by:
|
||||
group += g + ', '
|
||||
if group:
|
||||
grouping = True
|
||||
group = 'GROUP BY ' + group.rstrip(', ')
|
||||
else:
|
||||
grouping = False
|
||||
|
||||
# Build join parameters
|
||||
if join_types:
|
||||
counter = 0
|
||||
for join_type in join_types:
|
||||
if join_type.upper() == 'LEFT OUTER JOIN':
|
||||
join_item = 'LEFT OUTER JOIN %s ON %s = %s ' % \
|
||||
(join_tables[counter], join_evals[counter][0], join_evals[counter][1])
|
||||
elif join_type.upper() == 'JOIN' or join_type.upper() == 'INNER JOIN':
|
||||
join_item = 'JOIN %s ON %s = %s ' % \
|
||||
(join_tables[counter], join_evals[counter][0], join_evals[counter][1])
|
||||
else:
|
||||
join_item = ''
|
||||
|
||||
counter += 1
|
||||
join += join_item
|
||||
|
||||
# Build custom where parameters
|
||||
if custom_where:
|
||||
for w in custom_where:
|
||||
if isinstance(w[1], (list, tuple)) and len(w[1]):
|
||||
c_where += '('
|
||||
for w_ in w[1]:
|
||||
if w_ == None:
|
||||
c_where += w[0] + ' IS NULL OR '
|
||||
else:
|
||||
c_where += w[0] + ' = ? OR '
|
||||
args.append(w_)
|
||||
c_where = c_where.rstrip(' OR ') + ') AND '
|
||||
else:
|
||||
if w[1] == None:
|
||||
c_where += w[0] + ' IS NULL AND '
|
||||
else:
|
||||
c_where += w[0] + ' = ? AND '
|
||||
args.append(w[1])
|
||||
|
||||
if c_where:
|
||||
c_where = 'WHERE ' + c_where.rstrip(' AND ')
|
||||
|
||||
# Build grouping union parameters
|
||||
if group_by_union:
|
||||
for g in group_by_union:
|
||||
group_u += g + ', '
|
||||
if group_u:
|
||||
group_u = 'GROUP BY ' + group_u.rstrip(', ')
|
||||
else:
|
||||
group_u = ''
|
||||
|
||||
# Build custom where union parameters
|
||||
if custom_where_union:
|
||||
for w in custom_where_union:
|
||||
if isinstance(w[1], (list, tuple)) and len(w[1]):
|
||||
c_where_u += '('
|
||||
for w_ in w[1]:
|
||||
if w_ == None:
|
||||
c_where_u += w[0] + ' IS NULL OR '
|
||||
else:
|
||||
c_where_u += w[0] + ' = ? OR '
|
||||
args.append(w_)
|
||||
c_where_u = c_where_u.rstrip(' OR ') + ') AND '
|
||||
else:
|
||||
if w[1] == None:
|
||||
c_where_u += w[0] + ' IS NULL AND '
|
||||
else:
|
||||
c_where_u += w[0] + ' = ? AND '
|
||||
args.append(w[1])
|
||||
|
||||
if c_where_u:
|
||||
c_where_u = 'WHERE ' + c_where_u.rstrip(' AND ')
|
||||
else:
|
||||
c_where_u = ''
|
||||
args = cw_args + w_args
|
||||
|
||||
# Build union parameters
|
||||
if table_name_union:
|
||||
extracted_columns_union = self.extract_columns(columns=columns_union)
|
||||
group_u = self.build_grouping(group_by_union)
|
||||
c_where_u, cwu_args = self.build_custom_where(custom_where_union)
|
||||
|
||||
args += cwu_args
|
||||
|
||||
union = 'UNION SELECT %s FROM %s %s %s' % (extracted_columns_union['column_string'],
|
||||
table_name_union,
|
||||
c_where_u,
|
||||
|
@ -159,75 +82,10 @@ class DataTables(object):
|
|||
else:
|
||||
union = ''
|
||||
|
||||
# Build ordering
|
||||
for o in parameters['order']:
|
||||
sort_order = ' COLLATE NOCASE'
|
||||
if o['dir'] == 'desc':
|
||||
sort_order = ' COLLATE NOCASE DESC'
|
||||
# We first see if a name was sent though for the column sort.
|
||||
if dt_columns[int(o['column'])]['data']:
|
||||
# We have a name, now check if it's a valid column name for our query
|
||||
# so we don't just inject a random value
|
||||
if any(d.lower() == dt_columns[int(o['column'])]['data'].lower()
|
||||
for d in extracted_columns['column_named']):
|
||||
order += dt_columns[int(o['column'])]['data'] + '%s' % sort_order
|
||||
else:
|
||||
# if we receive a bogus name, rather not sort at all.
|
||||
pass
|
||||
# If no name exists for the column, just use the column index to sort
|
||||
else:
|
||||
order += extracted_columns['column_named'][int(o['column'])]
|
||||
|
||||
order += ', '
|
||||
|
||||
order = order.rstrip(', ')
|
||||
if order:
|
||||
order = 'ORDER BY ' + order
|
||||
|
||||
# Build where parameters
|
||||
if parameters['search']['value']:
|
||||
counter = 0
|
||||
for s in parameters['columns']:
|
||||
if s['searchable']:
|
||||
# We first see if a name was sent though for the column search.
|
||||
if s['data']:
|
||||
# We have a name, now check if it's a valid column name for our query
|
||||
# so we don't just inject a random value
|
||||
if any(d.lower() == s['data'].lower() for d in extracted_columns['column_named']):
|
||||
where += s['data'] + ' LIKE ? OR '
|
||||
args.append('%' + parameters['search']['value'] + '%')
|
||||
else:
|
||||
# if we receive a bogus name, rather not search at all.
|
||||
pass
|
||||
# If no name exists for the column, just use the column index to search
|
||||
else:
|
||||
where += extracted_columns['column_named'][counter] + ' LIKE ? OR '
|
||||
args.append('%' + parameters['search']['value'] + '%')
|
||||
|
||||
counter += 1
|
||||
|
||||
if where:
|
||||
where = 'WHERE ' + where.rstrip(' OR ')
|
||||
|
||||
# Build our queries
|
||||
if grouping:
|
||||
if c_where == '':
|
||||
query = 'SELECT * FROM (SELECT %s FROM %s %s %s %s) %s %s' \
|
||||
% (extracted_columns['column_string'], table_name, join, group, union,
|
||||
where, order)
|
||||
else:
|
||||
query = 'SELECT * FROM (SELECT %s FROM %s %s %s %s %s) %s %s' \
|
||||
% (extracted_columns['column_string'], table_name, join, c_where, group, union,
|
||||
where, order)
|
||||
else:
|
||||
if c_where == '':
|
||||
query = 'SELECT * FROM (SELECT %s FROM %s %s %s) %s %s' \
|
||||
% (extracted_columns['column_string'], table_name, join, union,
|
||||
where, order)
|
||||
else:
|
||||
query = 'SELECT * FROM (SELECT %s FROM %s %s %s %s) %s %s' \
|
||||
% (extracted_columns['column_string'], table_name, join, c_where, union,
|
||||
where, order)
|
||||
# Build the query
|
||||
query = 'SELECT * FROM (SELECT %s FROM %s %s %s %s %s) %s %s' \
|
||||
% (extracted_columns['column_string'], table_name, join, c_where, group, union, where, order)
|
||||
|
||||
# logger.debug(u"Query: %s" % query)
|
||||
|
||||
|
@ -257,6 +115,110 @@ class DataTables(object):
|
|||
|
||||
return output
|
||||
|
||||
def build_grouping(self, group_by=[]):
|
||||
# Build grouping
|
||||
group = ''
|
||||
|
||||
for g in group_by:
|
||||
group += g + ', '
|
||||
if group:
|
||||
group = 'GROUP BY ' + group.rstrip(', ')
|
||||
|
||||
return group
|
||||
|
||||
def build_join(self, join_types=[], join_tables=[], join_evals=[]):
|
||||
# Build join parameters
|
||||
join = ''
|
||||
|
||||
for i, join_type in enumerate(join_types):
|
||||
if join_type.upper() == 'LEFT OUTER JOIN':
|
||||
join += 'LEFT OUTER JOIN %s ON %s = %s ' % (join_tables[i], join_evals[i][0], join_evals[i][1])
|
||||
elif join_type.upper() == 'JOIN' or join_type.upper() == 'INNER JOIN':
|
||||
join += 'JOIN %s ON %s = %s ' % (join_tables[i], join_evals[i][0], join_evals[i][1])
|
||||
|
||||
return join
|
||||
|
||||
def build_custom_where(self, custom_where=[]):
|
||||
# Build custom where parameters
|
||||
c_where = ''
|
||||
args = []
|
||||
|
||||
for w in custom_where:
|
||||
if isinstance(w[1], (list, tuple)) and len(w[1]):
|
||||
c_where += '('
|
||||
for w_ in w[1]:
|
||||
if w_ == None:
|
||||
c_where += w[0] + ' IS NULL OR '
|
||||
else:
|
||||
c_where += w[0] + ' = ? OR '
|
||||
args.append(w_)
|
||||
c_where = c_where.rstrip(' OR ') + ') AND '
|
||||
else:
|
||||
if w[1] == None:
|
||||
c_where += w[0] + ' IS NULL AND '
|
||||
else:
|
||||
c_where += w[0] + ' = ? AND '
|
||||
args.append(w[1])
|
||||
|
||||
if c_where:
|
||||
c_where = 'WHERE ' + c_where.rstrip(' AND ')
|
||||
|
||||
return c_where, args
|
||||
|
||||
def build_order(self, order_param=[], columns=[], dt_columns=[]):
|
||||
# Build ordering
|
||||
order = ''
|
||||
|
||||
for o in order_param:
|
||||
sort_order = ' COLLATE NOCASE'
|
||||
if o['dir'] == 'desc':
|
||||
sort_order += ' DESC'
|
||||
# We first see if a name was sent though for the column sort.
|
||||
if dt_columns[int(o['column'])]['data']:
|
||||
# We have a name, now check if it's a valid column name for our query
|
||||
# so we don't just inject a random value
|
||||
if any(d.lower() == dt_columns[int(o['column'])]['data'].lower()
|
||||
for d in columns):
|
||||
order += dt_columns[int(o['column'])]['data'] + '%s, ' % sort_order
|
||||
else:
|
||||
# if we receive a bogus name, rather not sort at all.
|
||||
pass
|
||||
# If no name exists for the column, just use the column index to sort
|
||||
else:
|
||||
order += columns[int(o['column'])] + ', '
|
||||
|
||||
if order:
|
||||
order = 'ORDER BY ' + order.rstrip(', ')
|
||||
|
||||
return order
|
||||
|
||||
def build_where(self, search_param='', columns=[], dt_columns=[]):
|
||||
# Build where parameters
|
||||
where = ''
|
||||
args = []
|
||||
|
||||
if search_param:
|
||||
for i, s in enumerate(dt_columns):
|
||||
if s['searchable']:
|
||||
# We first see if a name was sent though for the column search.
|
||||
if s['data']:
|
||||
# We have a name, now check if it's a valid column name for our query
|
||||
# so we don't just inject a random value
|
||||
if any(d.lower() == s['data'].lower() for d in columns):
|
||||
where += s['data'] + ' LIKE ? OR '
|
||||
args.append('%' + search_param + '%')
|
||||
else:
|
||||
# if we receive a bogus name, rather not search at all.
|
||||
pass
|
||||
# If no name exists for the column, just use the column index to search
|
||||
else:
|
||||
where += columns[i] + ' LIKE ? OR '
|
||||
args.append('%' + search_param + '%')
|
||||
if where:
|
||||
where = 'WHERE ' + where.rstrip(' OR ')
|
||||
|
||||
return where, args
|
||||
|
||||
# This method extracts column data from our column list
|
||||
# The first parameter is required, the match_columns parameter is optional and will cause the function to
|
||||
# only return results if the value also exists in the match_columns 'data' field
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue