1
|
# Database access
|
2
|
|
3
|
import copy
|
4
|
import operator
|
5
|
import re
|
6
|
import warnings
|
7
|
|
8
|
import exc
|
9
|
import dicts
|
10
|
import iters
|
11
|
import lists
|
12
|
from Proxy import Proxy
|
13
|
import rand
|
14
|
import sql_gen
|
15
|
import strings
|
16
|
import util
|
17
|
|
18
|
##### Exceptions
|
19
|
|
20
|
def get_cur_query(cur, input_query=None, input_params=None):
|
21
|
raw_query = None
|
22
|
if hasattr(cur, 'query'): raw_query = cur.query
|
23
|
elif hasattr(cur, '_last_executed'): raw_query = cur._last_executed
|
24
|
|
25
|
if raw_query != None: return raw_query
|
26
|
else: return '[input] '+strings.ustr(input_query)+' % '+repr(input_params)
|
27
|
|
28
|
def _add_cursor_info(e, *args, **kw_args):
|
29
|
'''For params, see get_cur_query()'''
|
30
|
exc.add_msg(e, 'query: '+str(get_cur_query(*args, **kw_args)))
|
31
|
|
32
|
class DbException(exc.ExceptionWithCause):
|
33
|
def __init__(self, msg, cause=None, cur=None):
|
34
|
exc.ExceptionWithCause.__init__(self, msg, cause, cause_newline=True)
|
35
|
if cur != None: _add_cursor_info(self, cur)
|
36
|
|
37
|
class ExceptionWithName(DbException):
|
38
|
def __init__(self, name, cause=None):
|
39
|
DbException.__init__(self, 'for name: '+strings.as_tt(str(name)), cause)
|
40
|
self.name = name
|
41
|
|
42
|
class ExceptionWithNameValue(DbException):
|
43
|
def __init__(self, name, value, cause=None):
|
44
|
DbException.__init__(self, 'for name: '+strings.as_tt(str(name))
|
45
|
+'; value: '+strings.as_tt(repr(value)), cause)
|
46
|
self.name = name
|
47
|
self.value = value
|
48
|
|
49
|
class ConstraintException(DbException):
|
50
|
def __init__(self, name, cols, cause=None):
|
51
|
DbException.__init__(self, 'Violated '+strings.as_tt(name)
|
52
|
+' constraint on columns: '+strings.as_tt(', '.join(cols)), cause)
|
53
|
self.name = name
|
54
|
self.cols = cols
|
55
|
|
56
|
class MissingCastException(DbException):
|
57
|
def __init__(self, type_, col, cause=None):
|
58
|
DbException.__init__(self, 'Missing cast to type '+strings.as_tt(type_)
|
59
|
+' on column: '+strings.as_tt(col), cause)
|
60
|
self.type = type_
|
61
|
self.col = col
|
62
|
|
63
|
class NameException(DbException): pass
|
64
|
|
65
|
class DuplicateKeyException(ConstraintException): pass
|
66
|
|
67
|
class NullValueException(ConstraintException): pass
|
68
|
|
69
|
class FunctionValueException(ExceptionWithNameValue): pass
|
70
|
|
71
|
class DuplicateTableException(ExceptionWithName): pass
|
72
|
|
73
|
class DuplicateFunctionException(ExceptionWithName): pass
|
74
|
|
75
|
class EmptyRowException(DbException): pass
|
76
|
|
77
|
##### Warnings
|
78
|
|
79
|
class DbWarning(UserWarning): pass
|
80
|
|
81
|
##### Result retrieval
|
82
|
|
83
|
def col_names(cur): return (col[0] for col in cur.description)
|
84
|
|
85
|
def rows(cur): return iter(lambda: cur.fetchone(), None)
|
86
|
|
87
|
def consume_rows(cur):
|
88
|
'''Used to fetch all rows so result will be cached'''
|
89
|
iters.consume_iter(rows(cur))
|
90
|
|
91
|
def next_row(cur): return rows(cur).next()
|
92
|
|
93
|
def row(cur):
|
94
|
row_ = next_row(cur)
|
95
|
consume_rows(cur)
|
96
|
return row_
|
97
|
|
98
|
def next_value(cur): return next_row(cur)[0]
|
99
|
|
100
|
def value(cur): return row(cur)[0]
|
101
|
|
102
|
def values(cur): return iters.func_iter(lambda: next_value(cur))
|
103
|
|
104
|
def value_or_none(cur):
|
105
|
try: return value(cur)
|
106
|
except StopIteration: return None
|
107
|
|
108
|
##### Input validation
|
109
|
|
110
|
def check_name(name):
|
111
|
if re.search(r'\W', name) != None: raise NameException('Name "'+name
|
112
|
+'" may contain only alphanumeric characters and _')
|
113
|
|
114
|
def esc_name_by_module(module, name, ignore_case=False):
|
115
|
if module == 'psycopg2' or module == None:
|
116
|
if ignore_case:
|
117
|
# Don't enclose in quotes because this disables case-insensitivity
|
118
|
check_name(name)
|
119
|
return name
|
120
|
else: quote = '"'
|
121
|
elif module == 'MySQLdb': quote = '`'
|
122
|
else: raise NotImplementedError("Can't escape name for "+module+' database')
|
123
|
return sql_gen.esc_name(name, quote)
|
124
|
|
125
|
def esc_name_by_engine(engine, name, **kw_args):
|
126
|
return esc_name_by_module(db_engines[engine][0], name, **kw_args)
|
127
|
|
128
|
def esc_name(db, name, **kw_args):
|
129
|
return esc_name_by_module(util.root_module(db.db), name, **kw_args)
|
130
|
|
131
|
def qual_name(db, schema, table):
|
132
|
def esc_name_(name): return esc_name(db, name)
|
133
|
table = esc_name_(table)
|
134
|
if schema != None: return esc_name_(schema)+'.'+table
|
135
|
else: return table
|
136
|
|
137
|
##### Database connections
|
138
|
|
139
|
db_config_names = ['engine', 'host', 'user', 'password', 'database', 'schemas']
|
140
|
|
141
|
db_engines = {
|
142
|
'MySQL': ('MySQLdb', {'password': 'passwd', 'database': 'db'}),
|
143
|
'PostgreSQL': ('psycopg2', {}),
|
144
|
}
|
145
|
|
146
|
DatabaseErrors_set = set([DbException])
|
147
|
DatabaseErrors = tuple(DatabaseErrors_set)
|
148
|
|
149
|
def _add_module(module):
|
150
|
DatabaseErrors_set.add(module.DatabaseError)
|
151
|
global DatabaseErrors
|
152
|
DatabaseErrors = tuple(DatabaseErrors_set)
|
153
|
|
154
|
def db_config_str(db_config):
|
155
|
return db_config['engine']+' database '+db_config['database']
|
156
|
|
157
|
def _query_lookup(query, params): return (query, dicts.make_hashable(params))
|
158
|
|
159
|
log_debug_none = lambda msg, level=2: None
|
160
|
|
161
|
class DbConn:
|
162
|
def __init__(self, db_config, serializable=True, autocommit=False,
|
163
|
caching=True, log_debug=log_debug_none):
|
164
|
self.db_config = db_config
|
165
|
self.serializable = serializable
|
166
|
self.autocommit = autocommit
|
167
|
self.caching = caching
|
168
|
self.log_debug = log_debug
|
169
|
self.debug = log_debug != log_debug_none
|
170
|
|
171
|
self.__db = None
|
172
|
self.query_results = {}
|
173
|
self._savepoint = 0
|
174
|
|
175
|
def __getattr__(self, name):
|
176
|
if name == '__dict__': raise Exception('getting __dict__')
|
177
|
if name == 'db': return self._db()
|
178
|
else: raise AttributeError()
|
179
|
|
180
|
def __getstate__(self):
|
181
|
state = copy.copy(self.__dict__) # shallow copy
|
182
|
state['log_debug'] = None # don't pickle the debug callback
|
183
|
state['_DbConn__db'] = None # don't pickle the connection
|
184
|
return state
|
185
|
|
186
|
def connected(self): return self.__db != None
|
187
|
|
188
|
def _db(self):
|
189
|
if self.__db == None:
|
190
|
# Process db_config
|
191
|
db_config = self.db_config.copy() # don't modify input!
|
192
|
schemas = db_config.pop('schemas', None)
|
193
|
module_name, mappings = db_engines[db_config.pop('engine')]
|
194
|
module = __import__(module_name)
|
195
|
_add_module(module)
|
196
|
for orig, new in mappings.iteritems():
|
197
|
try: util.rename_key(db_config, orig, new)
|
198
|
except KeyError: pass
|
199
|
|
200
|
# Connect
|
201
|
self.__db = module.connect(**db_config)
|
202
|
|
203
|
# Configure connection
|
204
|
if self.serializable and not self.autocommit: run_raw_query(self,
|
205
|
'SET TRANSACTION ISOLATION LEVEL SERIALIZABLE')
|
206
|
if schemas != None:
|
207
|
schemas_ = ''.join((esc_name(self, s)+', '
|
208
|
for s in schemas.split(',')))
|
209
|
run_raw_query(self, "SELECT set_config('search_path', \
|
210
|
%s || current_setting('search_path'), false)", [schemas_])
|
211
|
|
212
|
return self.__db
|
213
|
|
214
|
class DbCursor(Proxy):
|
215
|
def __init__(self, outer):
|
216
|
Proxy.__init__(self, outer.db.cursor())
|
217
|
self.outer = outer
|
218
|
self.query_results = outer.query_results
|
219
|
self.query_lookup = None
|
220
|
self.result = []
|
221
|
|
222
|
def execute(self, query, params=None):
|
223
|
self._is_insert = query.upper().find('INSERT') >= 0
|
224
|
self.query_lookup = _query_lookup(query, params)
|
225
|
try:
|
226
|
try:
|
227
|
return_value = self.inner.execute(query, params)
|
228
|
self.outer.do_autocommit()
|
229
|
finally: self.query = get_cur_query(self.inner)
|
230
|
except Exception, e:
|
231
|
_add_cursor_info(e, self, query, params)
|
232
|
self.result = e # cache the exception as the result
|
233
|
self._cache_result()
|
234
|
raise
|
235
|
# Fetch all rows so result will be cached
|
236
|
if self.rowcount == 0 and not self._is_insert: consume_rows(self)
|
237
|
return return_value
|
238
|
|
239
|
def fetchone(self):
|
240
|
row = self.inner.fetchone()
|
241
|
if row != None: self.result.append(row)
|
242
|
# otherwise, fetched all rows
|
243
|
else: self._cache_result()
|
244
|
return row
|
245
|
|
246
|
def _cache_result(self):
|
247
|
# For inserts, only cache exceptions since inserts are not
|
248
|
# idempotent, but an invalid insert will always be invalid
|
249
|
if self.query_results != None and (not self._is_insert
|
250
|
or isinstance(self.result, Exception)):
|
251
|
|
252
|
assert self.query_lookup != None
|
253
|
self.query_results[self.query_lookup] = self.CacheCursor(
|
254
|
util.dict_subset(dicts.AttrsDictView(self),
|
255
|
['query', 'result', 'rowcount', 'description']))
|
256
|
|
257
|
class CacheCursor:
|
258
|
def __init__(self, cached_result): self.__dict__ = cached_result
|
259
|
|
260
|
def execute(self, *args, **kw_args):
|
261
|
if isinstance(self.result, Exception): raise self.result
|
262
|
# otherwise, result is a rows list
|
263
|
self.iter = iter(self.result)
|
264
|
|
265
|
def fetchone(self):
|
266
|
try: return self.iter.next()
|
267
|
except StopIteration: return None
|
268
|
|
269
|
def esc_value(self, value):
|
270
|
module = util.root_module(self.db)
|
271
|
if module == 'psycopg2': str_ = self.db.cursor().mogrify('%s', [value])
|
272
|
elif module == 'MySQLdb':
|
273
|
import _mysql
|
274
|
str_ = _mysql.escape_string(value)
|
275
|
else: raise NotImplementedError("Can't escape value for "+module
|
276
|
+' database')
|
277
|
return strings.to_unicode(str_)
|
278
|
|
279
|
def esc_name(self, name): return esc_name(self, name) # calls global func
|
280
|
|
281
|
def run_query(self, query, params=None, cacheable=False, log_level=2,
|
282
|
debug_msg_ref=None):
|
283
|
'''
|
284
|
@param log_ignore_excs The log_level will be increased by 2 if the query
|
285
|
throws one of these exceptions.
|
286
|
'''
|
287
|
assert query != None
|
288
|
|
289
|
if not self.caching: cacheable = False
|
290
|
used_cache = False
|
291
|
try:
|
292
|
# Get cursor
|
293
|
if cacheable:
|
294
|
query_lookup = _query_lookup(query, params)
|
295
|
try:
|
296
|
cur = self.query_results[query_lookup]
|
297
|
used_cache = True
|
298
|
except KeyError: cur = self.DbCursor(self)
|
299
|
else: cur = self.db.cursor()
|
300
|
|
301
|
# Run query
|
302
|
cur.execute(query, params)
|
303
|
finally:
|
304
|
if self.debug and debug_msg_ref != None:# only compute msg if needed
|
305
|
if used_cache: cache_status = 'cache hit'
|
306
|
elif cacheable: cache_status = 'cache miss'
|
307
|
else: cache_status = 'non-cacheable'
|
308
|
query_code = strings.as_code(str(get_cur_query(cur, query,
|
309
|
params)), 'SQL')
|
310
|
debug_msg_ref[0] = 'DB query: '+cache_status+':\n'+query_code
|
311
|
|
312
|
return cur
|
313
|
|
314
|
def is_cached(self, query, params=None):
|
315
|
return _query_lookup(query, params) in self.query_results
|
316
|
|
317
|
def with_savepoint(self, func):
|
318
|
savepoint = 'level_'+str(self._savepoint)
|
319
|
self.run_query('SAVEPOINT '+savepoint, log_level=4)
|
320
|
self._savepoint += 1
|
321
|
try:
|
322
|
try: return_val = func()
|
323
|
finally:
|
324
|
self._savepoint -= 1
|
325
|
assert self._savepoint >= 0
|
326
|
except:
|
327
|
self.run_query('ROLLBACK TO SAVEPOINT '+savepoint, log_level=4)
|
328
|
raise
|
329
|
else:
|
330
|
self.run_query('RELEASE SAVEPOINT '+savepoint, log_level=4)
|
331
|
self.do_autocommit()
|
332
|
return return_val
|
333
|
|
334
|
def do_autocommit(self):
|
335
|
'''Autocommits if outside savepoint'''
|
336
|
assert self._savepoint >= 0
|
337
|
if self.autocommit and self._savepoint == 0:
|
338
|
self.log_debug('Autocommiting')
|
339
|
self.db.commit()
|
340
|
|
341
|
connect = DbConn
|
342
|
|
343
|
##### Querying
|
344
|
|
345
|
def run_raw_query(db, *args, **kw_args):
|
346
|
'''For params, see DbConn.run_query()'''
|
347
|
return db.run_query(*args, **kw_args)
|
348
|
|
349
|
def mogrify(db, query, params):
|
350
|
module = util.root_module(db.db)
|
351
|
if module == 'psycopg2': return db.db.cursor().mogrify(query, params)
|
352
|
else: raise NotImplementedError("Can't mogrify query for "+module+
|
353
|
' database')
|
354
|
|
355
|
##### Recoverable querying
|
356
|
|
357
|
def with_savepoint(db, func): return db.with_savepoint(func)
|
358
|
|
359
|
def run_query(db, query, params=None, recover=None, cacheable=False,
|
360
|
log_level=2, log_ignore_excs=None, **kw_args):
|
361
|
'''For params, see run_raw_query()'''
|
362
|
if recover == None: recover = False
|
363
|
if log_ignore_excs == None: log_ignore_excs = ()
|
364
|
log_ignore_excs = tuple(log_ignore_excs)
|
365
|
|
366
|
debug_msg_ref = [None]
|
367
|
try:
|
368
|
try:
|
369
|
def run(): return run_raw_query(db, query, params, cacheable,
|
370
|
log_level, debug_msg_ref, **kw_args)
|
371
|
if recover and not db.is_cached(query, params):
|
372
|
return with_savepoint(db, run)
|
373
|
else: return run() # don't need savepoint if cached
|
374
|
except Exception, e:
|
375
|
if not recover: raise # need savepoint to run index_cols()
|
376
|
msg = exc.str_(e)
|
377
|
|
378
|
match = re.search(r'duplicate key value violates unique constraint '
|
379
|
r'"((_?[^\W_]+)_.+?)"', msg)
|
380
|
if match:
|
381
|
constraint, table = match.groups()
|
382
|
try: cols = index_cols(db, table, constraint)
|
383
|
except NotImplementedError: raise e
|
384
|
else: raise DuplicateKeyException(constraint, cols, e)
|
385
|
|
386
|
match = re.search(r'null value in column "(.+?)" violates not-null'
|
387
|
r' constraint', msg)
|
388
|
if match: raise NullValueException('NOT NULL', [match.group(1)], e)
|
389
|
|
390
|
match = re.search(r'\b(?:invalid input (?:syntax|value)\b.*?'
|
391
|
r'|date/time field value out of range): "(.+?)"\n'
|
392
|
r'(?:(?s).*?)\bfunction "(.+?)"', msg)
|
393
|
if match:
|
394
|
value, name = match.groups()
|
395
|
raise FunctionValueException(name, strings.to_unicode(value), e)
|
396
|
|
397
|
match = re.search(r'column "(.+?)" is of type (.+?) but expression '
|
398
|
r'is of type', msg)
|
399
|
if match:
|
400
|
col, type_ = match.groups()
|
401
|
raise MissingCastException(type_, col, e)
|
402
|
|
403
|
match = re.search(r'relation "(.+?)" already exists', msg)
|
404
|
if match: raise DuplicateTableException(match.group(1), e)
|
405
|
|
406
|
match = re.search(r'function "(.+?)" already exists', msg)
|
407
|
if match: raise DuplicateFunctionException(match.group(1), e)
|
408
|
|
409
|
raise # no specific exception raised
|
410
|
except log_ignore_excs:
|
411
|
log_level += 2
|
412
|
raise
|
413
|
finally:
|
414
|
if debug_msg_ref[0] != None: db.log_debug(debug_msg_ref[0], log_level)
|
415
|
|
416
|
##### Basic queries
|
417
|
|
418
|
def next_version(name):
|
419
|
'''Prepends the version # so it won't be removed if the name is truncated'''
|
420
|
version = 1 # first existing name was version 0
|
421
|
match = re.match(r'^#(\d+)-(.*)$', name)
|
422
|
if match:
|
423
|
version = int(match.group(1))+1
|
424
|
name = match.group(2)
|
425
|
return '#'+str(version)+'-'+name
|
426
|
|
427
|
def run_query_into(db, query, params, into=None, *args, **kw_args):
|
428
|
'''Outputs a query to a temp table.
|
429
|
For params, see run_query().
|
430
|
'''
|
431
|
if into == None: return run_query(db, query, params, *args, **kw_args)
|
432
|
else: # place rows in temp table
|
433
|
assert isinstance(into, sql_gen.Table)
|
434
|
|
435
|
kw_args['recover'] = True
|
436
|
kw_args.setdefault('log_ignore_excs', (DuplicateTableException,))
|
437
|
|
438
|
temp = not db.autocommit # tables are permanent in autocommit mode
|
439
|
# "temporary tables cannot specify a schema name", so remove schema
|
440
|
if temp: into.schema = None
|
441
|
|
442
|
while True:
|
443
|
try:
|
444
|
create_query = 'CREATE'
|
445
|
if temp: create_query += ' TEMP'
|
446
|
create_query += ' TABLE '+into.to_str(db)+' AS\n'+query
|
447
|
|
448
|
return run_query(db, create_query, params, *args, **kw_args)
|
449
|
# CREATE TABLE AS sets rowcount to # rows in query
|
450
|
except DuplicateTableException, e:
|
451
|
into.name = next_version(into.name)
|
452
|
# try again with next version of name
|
453
|
|
454
|
order_by_pkey = object() # tells mk_select() to order by the pkey
|
455
|
|
456
|
distinct_on_all = object() # tells mk_select() to SELECT DISTINCT ON all columns
|
457
|
|
458
|
def mk_select(db, tables, fields=None, conds=None, distinct_on=[], limit=None,
|
459
|
start=None, order_by=order_by_pkey, default_table=None):
|
460
|
'''
|
461
|
@param tables The single table to select from, or a list of tables to join
|
462
|
together, with tables after the first being sql_gen.Join objects
|
463
|
@param fields Use None to select all fields in the table
|
464
|
@param conds WHERE conditions: [(compare_left_side, compare_right_side),...]
|
465
|
* container can be any iterable type
|
466
|
* compare_left_side: sql_gen.Code|str (for col name)
|
467
|
* compare_right_side: sql_gen.ValueCond|literal value
|
468
|
@param distinct_on The columns to SELECT DISTINCT ON, or distinct_on_all to
|
469
|
use all columns
|
470
|
@return tuple(query, params)
|
471
|
'''
|
472
|
# Parse tables param
|
473
|
if not lists.is_seq(tables): tables = [tables]
|
474
|
tables = list(tables) # don't modify input! (list() copies input)
|
475
|
table0 = sql_gen.as_Table(tables.pop(0)) # first table is separate
|
476
|
|
477
|
# Parse other params
|
478
|
if conds == None: conds = []
|
479
|
elif isinstance(conds, dict): conds = conds.items()
|
480
|
conds = list(conds) # don't modify input! (list() copies input)
|
481
|
assert limit == None or type(limit) == int
|
482
|
assert start == None or type(start) == int
|
483
|
if order_by is order_by_pkey:
|
484
|
if distinct_on != []: order_by = None
|
485
|
else: order_by = pkey(db, table0, recover=True)
|
486
|
|
487
|
query = 'SELECT'
|
488
|
|
489
|
def parse_col(col): return sql_gen.as_Col(col, default_table).to_str(db)
|
490
|
|
491
|
# DISTINCT ON columns
|
492
|
if distinct_on != []:
|
493
|
query += '\nDISTINCT'
|
494
|
if distinct_on is not distinct_on_all:
|
495
|
query += ' ON ('+(', '.join(map(parse_col, distinct_on)))+')'
|
496
|
|
497
|
# Columns
|
498
|
query += '\n'
|
499
|
if fields == None: query += '*'
|
500
|
else: query += '\n, '.join(map(parse_col, fields))
|
501
|
|
502
|
# Main table
|
503
|
query += '\nFROM '+table0.to_str(db)
|
504
|
|
505
|
# Add joins
|
506
|
left_table = table0
|
507
|
for join_ in tables:
|
508
|
table = join_.table
|
509
|
|
510
|
# Parse special values
|
511
|
if join_.type_ is sql_gen.filter_out: # filter no match
|
512
|
conds.append((sql_gen.Col(table_not_null_col(db, table), table),
|
513
|
None))
|
514
|
|
515
|
query += '\n'+join_.to_str(db, left_table)
|
516
|
|
517
|
left_table = table
|
518
|
|
519
|
missing = True
|
520
|
if conds != []:
|
521
|
query += '\nWHERE\n'+('\nAND\n'.join(('('+sql_gen.ColValueCond(l, r)
|
522
|
.to_str(db)+')' for l, r in conds)))
|
523
|
missing = False
|
524
|
if order_by != None:
|
525
|
query += '\nORDER BY '+sql_gen.as_Col(order_by, table0).to_str(db)
|
526
|
if limit != None: query += '\nLIMIT '+str(limit); missing = False
|
527
|
if start != None:
|
528
|
if start != 0: query += '\nOFFSET '+str(start)
|
529
|
missing = False
|
530
|
if missing: warnings.warn(DbWarning(
|
531
|
'SELECT statement missing a WHERE, LIMIT, or OFFSET clause: '+query))
|
532
|
|
533
|
return (query, [])
|
534
|
|
535
|
def select(db, *args, **kw_args):
|
536
|
'''For params, see mk_select() and run_query()'''
|
537
|
recover = kw_args.pop('recover', None)
|
538
|
cacheable = kw_args.pop('cacheable', True)
|
539
|
log_level = kw_args.pop('log_level', 2)
|
540
|
|
541
|
query, params = mk_select(db, *args, **kw_args)
|
542
|
return run_query(db, query, params, recover, cacheable, log_level=log_level)
|
543
|
|
544
|
def mk_insert_select(db, table, cols=None, select_query=None, params=None,
|
545
|
returning=None, embeddable=False):
|
546
|
'''
|
547
|
@param returning str|None An inserted column (such as pkey) to return
|
548
|
@param embeddable Whether the query should be embeddable as a nested SELECT.
|
549
|
Warning: If you set this and cacheable=True when the query is run, the
|
550
|
query will be fully cached, not just if it raises an exception.
|
551
|
'''
|
552
|
table = sql_gen.as_Table(table)
|
553
|
if cols == []: cols = None # no cols (all defaults) = unknown col names
|
554
|
if cols != None: cols = [sql_gen.as_Col(v).to_str(db) for v in cols]
|
555
|
if select_query == None: select_query = 'DEFAULT VALUES'
|
556
|
if returning != None: returning = sql_gen.as_Col(returning, table)
|
557
|
|
558
|
# Build query
|
559
|
first_line = 'INSERT INTO '+table.to_str(db)
|
560
|
query = first_line
|
561
|
if cols != None: query += '\n('+', '.join(cols)+')'
|
562
|
query += '\n'+select_query
|
563
|
|
564
|
if returning != None:
|
565
|
returning_name = copy.copy(returning)
|
566
|
returning_name.table = None
|
567
|
returning_name = returning_name.to_str(db)
|
568
|
query += '\nRETURNING '+returning_name
|
569
|
|
570
|
if embeddable:
|
571
|
assert returning != None
|
572
|
|
573
|
# Create function
|
574
|
function_name = sql_gen.clean_name(first_line)
|
575
|
return_type = 'SETOF '+returning.to_str(db)+'%TYPE'
|
576
|
while True:
|
577
|
try:
|
578
|
func_schema = None
|
579
|
if not db.autocommit: func_schema = 'pg_temp'
|
580
|
function = sql_gen.Table(function_name, func_schema).to_str(db)
|
581
|
|
582
|
function_query = '''\
|
583
|
CREATE FUNCTION '''+function+'''()
|
584
|
RETURNS '''+return_type+'''
|
585
|
LANGUAGE sql
|
586
|
AS $$
|
587
|
'''+mogrify(db, query, params)+''';
|
588
|
$$;
|
589
|
'''
|
590
|
run_query(db, function_query, recover=True, cacheable=True,
|
591
|
log_ignore_excs=(DuplicateFunctionException,))
|
592
|
break # this version was successful
|
593
|
except DuplicateFunctionException, e:
|
594
|
function_name = next_version(function_name)
|
595
|
# try again with next version of name
|
596
|
|
597
|
# Return query that uses function
|
598
|
func_table = sql_gen.NamedTable('f', sql_gen.CustomCode(function+'()'),
|
599
|
[returning_name]) # AS clause requires function alias
|
600
|
return mk_select(db, func_table, start=0, order_by=None)
|
601
|
|
602
|
return (query, params)
|
603
|
|
604
|
def insert_select(db, *args, **kw_args):
|
605
|
'''For params, see mk_insert_select() and run_query_into()
|
606
|
@param into sql_gen.Table with suggested name of temp table to put RETURNING
|
607
|
values in
|
608
|
'''
|
609
|
into = kw_args.pop('into', None)
|
610
|
if into != None: kw_args['embeddable'] = True
|
611
|
recover = kw_args.pop('recover', None)
|
612
|
cacheable = kw_args.pop('cacheable', True)
|
613
|
|
614
|
query, params = mk_insert_select(db, *args, **kw_args)
|
615
|
return run_query_into(db, query, params, into, recover=recover,
|
616
|
cacheable=cacheable)
|
617
|
|
618
|
default = object() # tells insert() to use the default value for a column
|
619
|
|
620
|
def insert(db, table, row, *args, **kw_args):
|
621
|
'''For params, see insert_select()'''
|
622
|
if lists.is_seq(row): cols = None
|
623
|
else:
|
624
|
cols = row.keys()
|
625
|
row = row.values()
|
626
|
row = list(row) # ensure that "!= []" works
|
627
|
|
628
|
# Check for special values
|
629
|
labels = []
|
630
|
values = []
|
631
|
for value in row:
|
632
|
if value is default: labels.append('DEFAULT')
|
633
|
else:
|
634
|
labels.append('%s')
|
635
|
values.append(value)
|
636
|
|
637
|
# Build query
|
638
|
if values != []: query = 'VALUES ('+(', '.join(labels))+')'
|
639
|
else: query = None
|
640
|
|
641
|
return insert_select(db, table, cols, query, values, *args, **kw_args)
|
642
|
|
643
|
def mk_update(db, table, changes=None, cond=None):
|
644
|
'''
|
645
|
@param changes [(col, new_value),...]
|
646
|
* container can be any iterable type
|
647
|
* col: sql_gen.Code|str (for col name)
|
648
|
* new_value: sql_gen.Code|literal value
|
649
|
@param cond sql_gen.Code WHERE condition. e.g. use sql_gen.*Cond objects.
|
650
|
@return str query
|
651
|
'''
|
652
|
query = 'UPDATE '+sql_gen.as_Table(table).to_str(db)+'\nSET\n'
|
653
|
query += ',\n'.join((sql_gen.to_name_only_col(col, table).to_str(db)+' = '
|
654
|
+sql_gen.as_Value(new_value).to_str(db) for col, new_value in changes))
|
655
|
if cond != None: query += '\nWHERE\n'+cond.to_str(db)
|
656
|
|
657
|
return query
|
658
|
|
659
|
def update(db, *args, **kw_args):
|
660
|
'''For params, see mk_update() and run_query()'''
|
661
|
recover = kw_args.pop('recover', None)
|
662
|
|
663
|
return run_query(db, mk_update(db, *args, **kw_args), [], recover)
|
664
|
|
665
|
def last_insert_id(db):
|
666
|
module = util.root_module(db.db)
|
667
|
if module == 'psycopg2': return value(run_query(db, 'SELECT lastval()'))
|
668
|
elif module == 'MySQLdb': return db.insert_id()
|
669
|
else: return None
|
670
|
|
671
|
def truncate(db, table, schema='public'):
|
672
|
return run_query(db, 'TRUNCATE '+qual_name(db, schema, table)+' CASCADE')
|
673
|
|
674
|
def mk_flatten_mapping(db, into, cols, preserve=[], as_items=False):
|
675
|
'''Creates a mapping from original column names (which may have collisions)
|
676
|
to names that will be distinct among the columns' tables.
|
677
|
This is meant to be used for several tables that are being joined together.
|
678
|
@param cols The columns to combine. Duplicates will be removed.
|
679
|
@param into The table for the new columns.
|
680
|
@param preserve [sql_gen.Col...] Columns not to rename. Note that these
|
681
|
columns will be included in the mapping even if they are not in cols.
|
682
|
The tables of the provided Col objects will be changed to into, so make
|
683
|
copies of them if you want to keep the original tables.
|
684
|
@param as_items Whether to return a list of dict items instead of a dict
|
685
|
@return dict(orig_col=new_col, ...)
|
686
|
* orig_col: sql_gen.Col(orig_col_name, orig_table)
|
687
|
* new_col: sql_gen.Col(orig_col_name, into)
|
688
|
* All mappings use the into table so its name can easily be
|
689
|
changed for all columns at once
|
690
|
'''
|
691
|
cols = lists.uniqify(cols)
|
692
|
|
693
|
items = []
|
694
|
for col in preserve:
|
695
|
orig_col = copy.copy(col)
|
696
|
col.table = into
|
697
|
items.append((orig_col, col))
|
698
|
preserve = set(preserve)
|
699
|
for col in cols:
|
700
|
if col not in preserve: items.append((col, sql_gen.Col(str(col), into)))
|
701
|
|
702
|
if not as_items: items = dict(items)
|
703
|
return items
|
704
|
|
705
|
def flatten(db, into, joins, cols, limit=None, start=None, **kw_args):
|
706
|
'''For params, see mk_flatten_mapping()
|
707
|
@return See return value of mk_flatten_mapping()
|
708
|
'''
|
709
|
items = mk_flatten_mapping(db, into, cols, as_items=True, **kw_args)
|
710
|
cols = [sql_gen.NamedCol(new.name, old) for old, new in items]
|
711
|
run_query_into(db, *mk_select(db, joins, cols, limit=limit, start=start),
|
712
|
into=into)
|
713
|
return dict(items)
|
714
|
|
715
|
##### Database structure queries
|
716
|
|
717
|
def table_row_count(db, table, recover=None):
|
718
|
return value(run_query(db, *mk_select(db, table, [sql_gen.row_count],
|
719
|
order_by=None, start=0), recover=recover, log_level=3))
|
720
|
|
721
|
def table_cols(db, table, recover=None):
|
722
|
return list(col_names(select(db, table, limit=0, order_by=None,
|
723
|
recover=recover, log_level=4)))
|
724
|
|
725
|
def pkey(db, table, recover=None):
|
726
|
'''Assumed to be first column in table'''
|
727
|
return table_cols(db, table, recover)[0]
|
728
|
|
729
|
not_null_col = 'not_null'
|
730
|
|
731
|
def table_not_null_col(db, table, recover=None):
|
732
|
'''Name assumed to be the value of not_null_col. If not found, uses pkey.'''
|
733
|
if not_null_col in table_cols(db, table, recover): return not_null_col
|
734
|
else: return pkey(db, table, recover)
|
735
|
|
736
|
def index_cols(db, table, index):
|
737
|
'''Can also use this for UNIQUE constraints, because a UNIQUE index is
|
738
|
automatically created. When you don't know whether something is a UNIQUE
|
739
|
constraint or a UNIQUE index, use this function.'''
|
740
|
module = util.root_module(db.db)
|
741
|
if module == 'psycopg2':
|
742
|
return list(values(run_query(db, '''\
|
743
|
SELECT attname
|
744
|
FROM
|
745
|
(
|
746
|
SELECT attnum, attname
|
747
|
FROM pg_index
|
748
|
JOIN pg_class index ON index.oid = indexrelid
|
749
|
JOIN pg_class table_ ON table_.oid = indrelid
|
750
|
JOIN pg_attribute ON attrelid = indrelid AND attnum = ANY (indkey)
|
751
|
WHERE
|
752
|
table_.relname = %(table)s
|
753
|
AND index.relname = %(index)s
|
754
|
UNION
|
755
|
SELECT attnum, attname
|
756
|
FROM
|
757
|
(
|
758
|
SELECT
|
759
|
indrelid
|
760
|
, (regexp_matches(indexprs, E':varattno (\\\\d+)', 'g'))[1]::int
|
761
|
AS indkey
|
762
|
FROM pg_index
|
763
|
JOIN pg_class index ON index.oid = indexrelid
|
764
|
JOIN pg_class table_ ON table_.oid = indrelid
|
765
|
WHERE
|
766
|
table_.relname = %(table)s
|
767
|
AND index.relname = %(index)s
|
768
|
) s
|
769
|
JOIN pg_attribute ON attrelid = indrelid AND attnum = indkey
|
770
|
) s
|
771
|
ORDER BY attnum
|
772
|
''',
|
773
|
{'table': table, 'index': index}, cacheable=True, log_level=4)))
|
774
|
else: raise NotImplementedError("Can't list index columns for "+module+
|
775
|
' database')
|
776
|
|
777
|
def constraint_cols(db, table, constraint):
|
778
|
module = util.root_module(db.db)
|
779
|
if module == 'psycopg2':
|
780
|
return list(values(run_query(db, '''\
|
781
|
SELECT attname
|
782
|
FROM pg_constraint
|
783
|
JOIN pg_class ON pg_class.oid = conrelid
|
784
|
JOIN pg_attribute ON attrelid = conrelid AND attnum = ANY (conkey)
|
785
|
WHERE
|
786
|
relname = %(table)s
|
787
|
AND conname = %(constraint)s
|
788
|
ORDER BY attnum
|
789
|
''',
|
790
|
{'table': table, 'constraint': constraint})))
|
791
|
else: raise NotImplementedError("Can't list constraint columns for "+module+
|
792
|
' database')
|
793
|
|
794
|
row_num_col = '_row_num'
|
795
|
|
796
|
def add_index(db, expr):
|
797
|
'''Adds an index on a column or expression if it doesn't already exist.
|
798
|
Currently, only function calls are supported as expressions.
|
799
|
'''
|
800
|
expr = copy.copy(expr) # don't modify input!
|
801
|
|
802
|
# Extract col
|
803
|
if isinstance(expr, sql_gen.FunctionCall):
|
804
|
col = expr.args[0]
|
805
|
expr = sql_gen.Expr(expr)
|
806
|
else: col = expr
|
807
|
assert sql_gen.is_table_col(col)
|
808
|
|
809
|
index = sql_gen.as_Table(str(expr))
|
810
|
table = col.table
|
811
|
col.table = None
|
812
|
try: run_query(db, 'CREATE INDEX '+index.to_str(db)+' ON '+table.to_str(db)
|
813
|
+' ('+expr.to_str(db)+')', recover=True, cacheable=True, log_level=3)
|
814
|
except DuplicateTableException: pass # index already existed
|
815
|
|
816
|
def index_pkey(db, table, recover=None):
|
817
|
'''Makes the first column in a table the primary key.
|
818
|
@pre The table must not already have a primary key.
|
819
|
'''
|
820
|
table = sql_gen.as_Table(table)
|
821
|
|
822
|
index = sql_gen.as_Table(table.name+'_pkey')
|
823
|
col = sql_gen.to_name_only_col(pkey(db, table, recover))
|
824
|
run_query(db, 'ALTER TABLE '+table.to_str(db)+' ADD CONSTRAINT '
|
825
|
+index.to_str(db)+' PRIMARY KEY('+col.to_str(db)+')', recover=recover,
|
826
|
log_level=3)
|
827
|
|
828
|
def add_row_num(db, table):
|
829
|
'''Adds a row number column to a table. Its name is in row_num_col. It will
|
830
|
be the primary key.'''
|
831
|
table = sql_gen.as_Table(table).to_str(db)
|
832
|
run_query(db, 'ALTER TABLE '+table+' ADD COLUMN '+row_num_col
|
833
|
+' serial NOT NULL PRIMARY KEY', log_level=3)
|
834
|
|
835
|
def tables(db, schema='public', table_like='%'):
|
836
|
module = util.root_module(db.db)
|
837
|
params = {'schema': schema, 'table_like': table_like}
|
838
|
if module == 'psycopg2':
|
839
|
return values(run_query(db, '''\
|
840
|
SELECT tablename
|
841
|
FROM pg_tables
|
842
|
WHERE
|
843
|
schemaname = %(schema)s
|
844
|
AND tablename LIKE %(table_like)s
|
845
|
ORDER BY tablename
|
846
|
''',
|
847
|
params, cacheable=True))
|
848
|
elif module == 'MySQLdb':
|
849
|
return values(run_query(db, 'SHOW TABLES LIKE %(table_like)s', params,
|
850
|
cacheable=True))
|
851
|
else: raise NotImplementedError("Can't list tables for "+module+' database')
|
852
|
|
853
|
##### Database management
|
854
|
|
855
|
def empty_db(db, schema='public', **kw_args):
|
856
|
'''For kw_args, see tables()'''
|
857
|
for table in tables(db, schema, **kw_args): truncate(db, table, schema)
|
858
|
|
859
|
##### Heuristic queries
|
860
|
|
861
|
def put(db, table, row, pkey_=None, row_ct_ref=None):
|
862
|
'''Recovers from errors.
|
863
|
Only works under PostgreSQL (uses INSERT RETURNING).
|
864
|
'''
|
865
|
if pkey_ == None: pkey_ = pkey(db, table, recover=True)
|
866
|
|
867
|
try:
|
868
|
cur = insert(db, table, row, pkey_, recover=True)
|
869
|
if row_ct_ref != None and cur.rowcount >= 0:
|
870
|
row_ct_ref[0] += cur.rowcount
|
871
|
return value(cur)
|
872
|
except DuplicateKeyException, e:
|
873
|
return value(select(db, table, [pkey_],
|
874
|
util.dict_subset_right_join(row, e.cols), recover=True))
|
875
|
|
876
|
def get(db, table, row, pkey, row_ct_ref=None, create=False):
|
877
|
'''Recovers from errors'''
|
878
|
try: return value(select(db, table, [pkey], row, limit=1, recover=True))
|
879
|
except StopIteration:
|
880
|
if not create: raise
|
881
|
return put(db, table, row, pkey, row_ct_ref) # insert new row
|
882
|
|
883
|
def put_table(db, out_table, in_tables, mapping, row_ct_ref=None, into=None,
|
884
|
default=None):
|
885
|
'''Recovers from errors.
|
886
|
Only works under PostgreSQL (uses INSERT RETURNING).
|
887
|
@param in_tables The main input table to select from, followed by a list of
|
888
|
tables to join with it using the main input table's pkey
|
889
|
@param mapping dict(out_table_col=in_table_col, ...)
|
890
|
* out_table_col: sql_gen.Col|str
|
891
|
* in_table_col: sql_gen.Col Wrap literal values in a sql_gen.NamedCol
|
892
|
@param into The table to contain the output and input pkeys.
|
893
|
Defaults to `out_table.name+'-pkeys'`.
|
894
|
@param default The *output* column to use as the pkey for missing rows.
|
895
|
If this output column does not exist in the mapping, uses None.
|
896
|
@return sql_gen.Col Where the output pkeys are made available
|
897
|
'''
|
898
|
out_table = sql_gen.as_Table(out_table)
|
899
|
for in_table_col in mapping.itervalues():
|
900
|
assert isinstance(in_table_col, sql_gen.Col)
|
901
|
if into == None: into = out_table.name+'-pkeys'
|
902
|
into = sql_gen.as_Table(into)
|
903
|
|
904
|
def log_debug(msg): db.log_debug(msg, level=1.5)
|
905
|
def col_ustr(str_):
|
906
|
return strings.repr_no_u(sql_gen.remove_col_rename(
|
907
|
sql_gen.as_Col(str_)))
|
908
|
|
909
|
log_debug('********** New iteration **********')
|
910
|
log_debug('Inserting these input columns into '+strings.as_tt(
|
911
|
out_table.to_str(db))+':\n'+strings.as_table(mapping, ustr=col_ustr))
|
912
|
|
913
|
# Create input joins from list of input tables
|
914
|
in_tables_ = in_tables[:] # don't modify input!
|
915
|
in_tables0 = in_tables_.pop(0) # first table is separate
|
916
|
in_pkey = pkey(db, in_tables0, recover=True)
|
917
|
in_pkey_col = sql_gen.as_Col(in_pkey, in_tables0)
|
918
|
input_joins = [in_tables0]+[sql_gen.Join(v,
|
919
|
{in_pkey: sql_gen.join_same_not_null}) for v in in_tables_]
|
920
|
|
921
|
log_debug('Joining together input tables into temp table')
|
922
|
# Place in new table for speed and so don't modify input if values edited
|
923
|
in_table = sql_gen.Table(into.name+'-input')
|
924
|
flatten_cols = filter(sql_gen.is_table_col, mapping.values())
|
925
|
mapping = dicts.join(mapping, flatten(db, in_table, input_joins,
|
926
|
flatten_cols, preserve=[in_pkey_col], start=0))
|
927
|
input_joins = [in_table]
|
928
|
db.log_debug('Temp table: '+strings.as_tt(in_table.to_str(db)), level=2)
|
929
|
|
930
|
# Resolve default value column
|
931
|
try: default = mapping[default]
|
932
|
except KeyError:
|
933
|
if default != None:
|
934
|
db.log_debug('Default value column '
|
935
|
+strings.as_tt(strings.repr_no_u(default))
|
936
|
+' does not exist in mapping, falling back to None', level=2.1)
|
937
|
default = None
|
938
|
|
939
|
out_pkey = pkey(db, out_table, recover=True)
|
940
|
out_pkey_col = sql_gen.as_Col(out_pkey, out_table)
|
941
|
|
942
|
pkeys_names = [in_pkey, out_pkey]
|
943
|
pkeys_cols = [in_pkey_col, out_pkey_col]
|
944
|
|
945
|
pkeys_table_exists_ref = [False]
|
946
|
def insert_into_pkeys(joins, cols):
|
947
|
query, params = mk_select(db, joins, cols, order_by=None, start=0)
|
948
|
if pkeys_table_exists_ref[0]:
|
949
|
insert_select(db, into, pkeys_names, query, params)
|
950
|
else:
|
951
|
run_query_into(db, query, params, into=into)
|
952
|
pkeys_table_exists_ref[0] = True
|
953
|
|
954
|
limit_ref = [None]
|
955
|
conds = set()
|
956
|
distinct_on = []
|
957
|
def mk_main_select(joins, cols):
|
958
|
return mk_select(db, joins, cols, conds, distinct_on,
|
959
|
limit=limit_ref[0], start=0)
|
960
|
|
961
|
exc_strs = set()
|
962
|
def log_exc(e):
|
963
|
e_str = exc.str_(e, first_line_only=True)
|
964
|
log_debug('Caught exception: '+e_str)
|
965
|
assert e_str not in exc_strs # avoid infinite loops
|
966
|
exc_strs.add(e_str)
|
967
|
def remove_all_rows():
|
968
|
log_debug('Returning NULL for all rows')
|
969
|
limit_ref[0] = 0 # just create an empty pkeys table
|
970
|
def ignore(in_col, value):
|
971
|
in_col_str = repr(in_col)
|
972
|
db.log_debug('Adding index on '+in_col_str+' to enable fast filtering',
|
973
|
level=2.5)
|
974
|
add_index(db, in_col)
|
975
|
log_debug('Ignoring rows with '+in_col_str+' = '+repr(value))
|
976
|
def remove_rows(in_col, value):
|
977
|
ignore(in_col, value)
|
978
|
cond = (in_col, sql_gen.CompareCond(value, '!='))
|
979
|
assert cond not in conds # avoid infinite loops
|
980
|
conds.add(cond)
|
981
|
def invalid2null(in_col, value):
|
982
|
ignore(in_col, value)
|
983
|
update(db, in_table, [(in_col, None)],
|
984
|
sql_gen.ColValueCond(in_col, value))
|
985
|
|
986
|
# Do inserts and selects
|
987
|
join_cols = {}
|
988
|
insert_out_pkeys = sql_gen.Table(into.name+'-insert_out_pkeys')
|
989
|
insert_in_pkeys = sql_gen.Table(into.name+'-insert_in_pkeys')
|
990
|
while True:
|
991
|
if limit_ref[0] == 0: # special case
|
992
|
log_debug('Creating an empty pkeys table')
|
993
|
cur = run_query_into(db, *mk_select(db, out_table, [out_pkey],
|
994
|
limit=limit_ref[0]), into=insert_out_pkeys)
|
995
|
break # don't do main case
|
996
|
|
997
|
has_joins = join_cols != {}
|
998
|
|
999
|
# Prepare to insert new rows
|
1000
|
insert_joins = input_joins[:] # don't modify original!
|
1001
|
insert_args = dict(recover=True, cacheable=False)
|
1002
|
if has_joins:
|
1003
|
distinct_on = [v.to_Col() for v in join_cols.values()]
|
1004
|
insert_joins.append(sql_gen.Join(out_table, join_cols,
|
1005
|
sql_gen.filter_out))
|
1006
|
else:
|
1007
|
insert_args.update(dict(returning=out_pkey, into=insert_out_pkeys))
|
1008
|
main_select = mk_main_select(insert_joins, mapping.values())[0]
|
1009
|
|
1010
|
log_debug('Trying to insert new rows')
|
1011
|
try:
|
1012
|
cur = insert_select(db, out_table, mapping.keys(), main_select,
|
1013
|
**insert_args)
|
1014
|
break # insert successful
|
1015
|
except DuplicateKeyException, e:
|
1016
|
log_exc(e)
|
1017
|
|
1018
|
old_join_cols = join_cols.copy()
|
1019
|
join_cols.update(util.dict_subset(mapping, e.cols))
|
1020
|
log_debug('Ignoring existing rows, comparing on these columns:\n'
|
1021
|
+strings.as_inline_table(join_cols, ustr=col_ustr))
|
1022
|
assert join_cols != old_join_cols # avoid infinite loops
|
1023
|
except NullValueException, e:
|
1024
|
log_exc(e)
|
1025
|
|
1026
|
out_col, = e.cols
|
1027
|
try: in_col = mapping[out_col]
|
1028
|
except KeyError:
|
1029
|
log_debug('Missing mapping for NOT NULL column '+out_col)
|
1030
|
remove_all_rows()
|
1031
|
else: remove_rows(in_col, None)
|
1032
|
except FunctionValueException, e:
|
1033
|
log_exc(e)
|
1034
|
|
1035
|
func_name = e.name
|
1036
|
value = e.value
|
1037
|
for out_col, in_col in mapping.iteritems():
|
1038
|
in_col = sql_gen.remove_col_rename(in_col)
|
1039
|
if (isinstance(in_col, sql_gen.FunctionCall)
|
1040
|
and in_col.function.name == func_name):
|
1041
|
invalid2null(in_col.args[0], value)
|
1042
|
except MissingCastException, e:
|
1043
|
log_exc(e)
|
1044
|
|
1045
|
out_col = e.col
|
1046
|
mapping[out_col] = sql_gen.wrap_in_func(e.type, mapping[out_col])
|
1047
|
except DatabaseErrors, e:
|
1048
|
log_exc(e)
|
1049
|
|
1050
|
msg = 'No handler for exception: '+exc.str_(e)
|
1051
|
warnings.warn(DbWarning(msg))
|
1052
|
log_debug(msg)
|
1053
|
remove_all_rows()
|
1054
|
# after exception handled, rerun loop with additional constraints
|
1055
|
|
1056
|
if row_ct_ref != None and cur.rowcount >= 0:
|
1057
|
row_ct_ref[0] += cur.rowcount
|
1058
|
|
1059
|
if has_joins:
|
1060
|
select_joins = input_joins+[sql_gen.Join(out_table, join_cols)]
|
1061
|
log_debug('Getting output table pkeys of existing/inserted rows')
|
1062
|
insert_into_pkeys(select_joins, pkeys_cols)
|
1063
|
else:
|
1064
|
add_row_num(db, insert_out_pkeys) # for joining with input pkeys
|
1065
|
|
1066
|
log_debug('Getting input table pkeys of inserted rows')
|
1067
|
run_query_into(db, *mk_main_select(input_joins, [in_pkey]),
|
1068
|
into=insert_in_pkeys)
|
1069
|
add_row_num(db, insert_in_pkeys) # for joining with output pkeys
|
1070
|
|
1071
|
assert table_row_count(db, insert_out_pkeys) == table_row_count(db,
|
1072
|
insert_in_pkeys)
|
1073
|
|
1074
|
log_debug('Combining output and input pkeys in inserted order')
|
1075
|
pkey_joins = [insert_in_pkeys, sql_gen.Join(insert_out_pkeys,
|
1076
|
{row_num_col: sql_gen.join_same_not_null})]
|
1077
|
insert_into_pkeys(pkey_joins, pkeys_names)
|
1078
|
|
1079
|
db.log_debug('Adding pkey on pkeys table to enable fast joins', level=2.5)
|
1080
|
index_pkey(db, into)
|
1081
|
|
1082
|
log_debug('Setting pkeys of missing rows to '+strings.as_tt(repr(default)))
|
1083
|
missing_rows_joins = input_joins+[sql_gen.Join(into,
|
1084
|
{in_pkey: sql_gen.join_same_not_null}, sql_gen.filter_out)]
|
1085
|
# must use join_same_not_null or query will take forever
|
1086
|
insert_into_pkeys(missing_rows_joins,
|
1087
|
[in_pkey_col, sql_gen.NamedCol(out_pkey, default)])
|
1088
|
|
1089
|
assert table_row_count(db, into) == table_row_count(db, in_table)
|
1090
|
|
1091
|
return sql_gen.Col(out_pkey, into)
|
1092
|
|
1093
|
##### Data cleanup
|
1094
|
|
1095
|
def cleanup_table(db, table, cols):
|
1096
|
def esc_name_(name): return esc_name(db, name)
|
1097
|
|
1098
|
table = sql_gen.as_Table(table).to_str(db)
|
1099
|
cols = map(esc_name_, cols)
|
1100
|
|
1101
|
run_query(db, 'UPDATE '+table+' SET\n'+(',\n'.join(('\n'+col
|
1102
|
+' = nullif(nullif(trim(both from '+col+"), %(null0)s), %(null1)s)"
|
1103
|
for col in cols))),
|
1104
|
dict(null0='', null1=r'\N'))
|