Project

General

Profile

1
# Database import/export
2

    
3
import copy
4
import csv
5
import operator
6
import warnings
7
import sys
8

    
9
import csvs
10
import exc
11
import dicts
12
import sql
13
import sql_gen
14
import streams
15
import strings
16
import util
17

    
18
##### Exceptions
19

    
20
# Can't use built-in SyntaxError because it stringifies to only the first line
21
class SyntaxError(Exception): pass
22

    
23
##### Data cleanup
24

    
25
def table_nulls_mapped__set(db, table):
26
    assert isinstance(table, sql_gen.Table)
27
    sql.run_query(db, 'SELECT util.table_nulls_mapped__set('
28
        +sql_gen.table2regclass_text(db, table)+')')
29

    
30
def table_nulls_mapped__get(db, table):
31
    assert isinstance(table, sql_gen.Table)
32
    return sql.value(sql.run_query(db, 'SELECT util.table_nulls_mapped__get('
33
        +sql_gen.table2regclass_text(db, table)+')'))
34

    
35
null_strs = ['', '-', r'\N', 'NULL', 'UNKNOWN', 'nulo']
36

    
37
def cleanup_table(db, table):
38
    '''idempotent'''
39
    table = sql_gen.as_Table(table)
40
    assert sql.table_exists(db, table)
41
    
42
    if table_nulls_mapped__get(db, table): return # already cleaned up
43
    
44
    cols = filter(lambda c: sql_gen.is_text_col(db, c),
45
        sql.table_cols(db, table))
46
    try: pkey_col = sql.table_pkey_col(db, table)
47
    except sql.DoesNotExistException: pass
48
    else:
49
        try: cols.remove(pkey_col)
50
        except ValueError: pass
51
    if not cols: return
52
    
53
    db.log_debug('Cleaning up table', level=1.5)
54
    
55
    expr = 'trim(both from %s)'
56
    for null in null_strs: expr = 'nullif('+expr+', '+db.esc_value(null)+')'
57
    changes = [(v, sql_gen.CustomCode(expr % v.to_str(db))) for v in cols]
58
    
59
    while True:
60
        try:
61
            sql.update(db, table, changes, in_place=True, recover=True)
62
            break # successful
63
        except sql.NullValueException, e:
64
            db.log_debug('Caught exception: '+exc.str_(e))
65
            col, = e.cols
66
            sql.drop_not_null(db, col)
67
    
68
    db.log_debug('Vacuuming and reanalyzing table', level=1.5)
69
    sql.vacuum(db, table)
70
    
71
    table_nulls_mapped__set(db, table)
72

    
73
##### Error tracking
74

    
75
def track_data_error(db, errors_table, cols, value, error_code, error):
76
    '''
77
    @param errors_table If None, does nothing.
78
    '''
79
    if errors_table == None: return
80
    
81
    col_names = [c.name for c in cols]
82
    if not col_names: col_names = [None] # need at least one entry
83
    for col_name in col_names:
84
        try:
85
            sql.insert(db, errors_table, dict(column=col_name, value=value,
86
                error_code=error_code, error=error), recover=True,
87
                cacheable=True, log_level=4)
88
        except sql.DuplicateKeyException: pass
89

    
90
class ExcToErrorsTable(sql_gen.ExcToWarning):
91
    '''Handles an exception by saving it or converting it to a warning.'''
92
    def __init__(self, return_, srcs, errors_table, value=None):
93
        '''
94
        @param return_ See sql_gen.ExcToWarning
95
        @param srcs The column names for the errors table
96
        @param errors_table None|sql_gen.Table
97
        @param value The value (or an expression for it) that caused the error
98
        @pre The invalid value must be in a local variable "value" of type text.
99
        '''
100
        sql_gen.ExcToWarning.__init__(self, return_)
101
        
102
        value = sql_gen.as_Code(value)
103
        
104
        self.srcs = srcs
105
        self.errors_table = errors_table
106
        self.value = value
107
    
108
    def to_str(self, db):
109
        if not self.srcs or self.errors_table == None:
110
            return sql_gen.ExcToWarning.to_str(self, db)
111
        
112
        errors_table_cols = map(sql_gen.Col,
113
            ['column', 'value', 'error_code', 'error'])
114
        col_names_query = sql.mk_select(db, sql_gen.NamedValues('c', None,
115
            [[c.name] for c in self.srcs]), order_by=None)
116
        insert_query = sql.mk_insert_select(db, self.errors_table,
117
            errors_table_cols,
118
            sql_gen.Values(errors_table_cols).to_str(db))+';\n'
119
        return '''\
120
-- Save error in errors table.
121
DECLARE
122
    error_code text := SQLSTATE;
123
    error text := SQLERRM;
124
    value text := '''+self.value.to_str(db)+''';
125
    "column" text;
126
BEGIN
127
    -- Insert the value and error for *each* source column.
128
'''+strings.indent(sql_gen.RowExcIgnore(None, col_names_query, insert_query,
129
    row_var=errors_table_cols[0]).to_str(db))+'''
130
END;
131

    
132
'''+self.return_.to_str(db)
133

    
134
def data_exception_handler(*args, **kw_args):
135
    '''Handles a data_exception by saving it or converting it to a warning.
136
    For params, see ExcToErrorsTable().
137
    '''
138
    return sql_gen.data_exception_handler(ExcToErrorsTable(*args, **kw_args))
139

    
140
def cast(db, type_, col, errors_table=None):
141
    '''Casts an (unrenamed) column or value.
142
    If errors_table set and col has srcs, saves errors in errors_table (using
143
    col's srcs attr as source columns). Otherwise, converts errors to warnings.
144
    @param col str|sql_gen.Col|sql_gen.Literal
145
    @param errors_table None|sql_gen.Table|str
146
    '''
147
    col = sql_gen.as_Col(col)
148
    
149
    # Don't convert exceptions to warnings for user-supplied constants
150
    if isinstance(col, sql_gen.Literal): return sql_gen.Cast(type_, col)
151
    
152
    assert not isinstance(col, sql_gen.NamedCol)
153
    
154
    function_name = strings.first_word(type_)
155
    srcs = col.srcs
156
    save_errors = errors_table != None and srcs
157
    if save_errors: # function will be unique for the given srcs
158
        function_name = strings.ustr(sql_gen.FunctionCall(function_name,
159
            *map(sql_gen.to_name_only_col, srcs)))
160
    function = db.TempFunction(function_name)
161
    
162
    # Create function definition
163
    modifiers = 'STRICT'
164
    if not save_errors: modifiers = 'IMMUTABLE '+modifiers
165
    value_param = sql_gen.FunctionParam('value', 'anyelement')
166
    handler = data_exception_handler('RETURN NULL;\n', srcs, errors_table,
167
        value_param.name)
168
    body = sql_gen.CustomCode(handler.to_str(db, '''\
169
/* The explicit cast to the return type is needed to make the cast happen
170
inside the try block. (Implicit casts to the return type happen at the end
171
of the function, outside any block.) */
172
RETURN '''+sql_gen.Cast(type_, sql_gen.CustomCode('value')).to_str(db)+''';
173
'''))
174
    body.lang='plpgsql'
175
    sql.define_func(db, sql_gen.FunctionDef(function, type_, body,
176
        [value_param], modifiers))
177
    
178
    return sql_gen.FunctionCall(function, col)
179

    
180
def func_wrapper_exception_handler(db, return_, args, errors_table):
181
    '''Handles a function call's data_exceptions.
182
    Supports PL/Python functions.
183
    @param return_ See data_exception_handler()
184
    @param args [arg...] Function call's args
185
    @param errors_table See data_exception_handler()
186
    '''
187
    args = filter(sql_gen.has_srcs, args)
188
    
189
    srcs = sql_gen.cross_join_srcs(args)
190
    value = sql_gen.merge_not_null(db, ',', args)
191
    return sql_gen.NestedExcHandler(
192
        data_exception_handler(return_, srcs, errors_table, value)
193
        , sql_gen.plpythonu_error_handler
194
        )
195

    
196
def cast_temp_col(db, type_, col, errors_table=None):
197
    '''Like cast(), but creates a new column with the cast values if the input
198
    is a column.
199
    @return The new column or cast value
200
    '''
201
    def cast_(col): return cast(db, type_, col, errors_table)
202
    
203
    try: col = sql_gen.underlying_col(col)
204
    except sql_gen.NoUnderlyingTableException: return sql_gen.wrap(cast_, col)
205
    
206
    table = col.table
207
    new_col = sql_gen.suffixed_col(col, '::'+strings.first_word(type_))
208
    expr = cast_(col)
209
    
210
    # Add column
211
    new_typed_col = sql_gen.TypedCol(new_col.name, type_)
212
    sql.add_col(db, table, new_typed_col, comment=strings.urepr(col)+'::'+type_)
213
    new_col.name = new_typed_col.name # propagate any renaming
214
    
215
    sql.update(db, table, [(new_col, expr)], in_place=True, recover=True)
216
    
217
    return new_col
218

    
219
def errors_table(db, table, if_exists=True):
220
    '''
221
    @param if_exists If set, returns None if the errors table doesn't exist
222
    @return None|sql_gen.Table
223
    '''
224
    table = sql_gen.as_Table(table)
225
    if table.srcs != (): table = table.srcs[0]
226
    
227
    errors_table = sql_gen.suffixed_table(table, '.errors')
228
    if if_exists and not sql.table_exists(db, errors_table): return None
229
    return errors_table
230

    
231
def mk_errors_table(db, table):
232
    errors_table_ = errors_table(db, table, if_exists=False)
233
    if sql.table_exists(db, errors_table_, cacheable=False): return
234
    
235
    typed_cols = [
236
        sql_gen.TypedCol('column', 'text'),
237
        sql_gen.TypedCol('value', 'text'),
238
        sql_gen.TypedCol('error_code', 'character varying(5)', nullable=False),
239
        sql_gen.TypedCol('error', 'text', nullable=False),
240
        ]
241
    sql.create_table(db, errors_table_, typed_cols, has_pkey=False)
242
    index_cols = ['column', sql_gen.CustomCode('md5(value)'), 'error_code',
243
        sql_gen.CustomCode('md5(error)')]
244
    sql.add_index(db, index_cols, errors_table_, unique=True)
245

    
246
##### Import
247

    
248
row_num_col_def = copy.copy(sql.row_num_col_def)
249
row_num_col_def.name = 'row_num'
250
row_num_col_def.type = 'integer'
251

    
252
def append_csv(db, table, reader, header):
253
    def esc_name_(name): return sql.esc_name(db, name)
254
    
255
    def log(msg, level=1): db.log_debug(msg, level)
256
    
257
    # Wrap in standardizing stream
258
    cols_ct = len(header)
259
    stream = csvs.InputRewriter(streams.ProgressInputStream(csvs.StreamFilter(
260
        csvs.ColCtFilter(reader, cols_ct)), sys.stderr, msg='Read %d row(s)',
261
        n=1000))
262
    dialect = stream.dialect # use default dialect
263
    
264
    # Create COPY FROM statement
265
    if header == sql.table_col_names(db, table): cols_str = ''
266
    else: cols_str =' ('+(', '.join(map(esc_name_, header)))+')'
267
    copy_from = ('COPY '+table.to_str(db)+cols_str+' FROM STDIN DELIMITER '
268
        +db.esc_value(dialect.delimiter)+' NULL '+db.esc_value(''))
269
    assert not csvs.is_tsv(dialect)
270
    copy_from += ' CSV'
271
    if dialect.quoting != csv.QUOTE_NONE:
272
        quote_str = db.esc_value(dialect.quotechar)
273
        copy_from += ' QUOTE '+quote_str
274
        if dialect.doublequote: copy_from += ' ESCAPE '+quote_str
275
    copy_from += ';\n'
276
    
277
    log(copy_from, level=2)
278
    try: db.db.cursor().copy_expert(copy_from, stream)
279
    except Exception, e: sql.parse_exception(db, e, recover=True)
280

    
281
def import_csv(db, table, reader, header):
282
    def log(msg, level=1): db.log_debug(msg, level)
283
    
284
    # Get format info
285
    col_names = map(strings.to_unicode, header)
286
    for i, col in enumerate(col_names): # replace empty column names
287
        if col == '': col_names[i] = 'column_'+str(i)
288
    
289
    # Select schema and escape names
290
    def esc_name(name): return db.esc_name(name)
291
    
292
    typed_cols = [sql_gen.TypedCol(v, 'text') for v in col_names]
293
    typed_cols.insert(0, row_num_col_def)
294
    header.insert(0, row_num_col_def.name)
295
    reader = csvs.RowNumFilter(reader)
296
    
297
    log('Creating table')
298
    # Note that this is not rolled back if the import fails. Instead, it is
299
    # cached, and will not be re-run if the import is retried.
300
    sql.create_table(db, table, typed_cols, has_pkey=False, col_indexes=False)
301
    
302
    # Free memory used by deleted (rolled back) rows from any failed import.
303
    # This MUST be run so that the rows will be stored in inserted order, and
304
    # the row_num added after import will match up with the CSV's row order.
305
    sql.truncate(db, table)
306
    
307
    # Load the data
308
    def load(): append_csv(db, table, reader, header)
309
    sql.with_savepoint(db, load)
310
    
311
    cleanup_table(db, table)
312

    
313
def put(db, table, row, pkey_=None, row_ct_ref=None, on_error=exc.reraise):
314
    '''Recovers from errors.
315
    Only works under PostgreSQL (uses INSERT RETURNING).
316
    '''
317
    return put_table(db, table, [], row, row_ct_ref, on_error=on_error)
318

    
319
def get(db, table, row, pkey, row_ct_ref=None, create=False):
320
    '''Recovers from errors'''
321
    try:
322
        return sql.value(sql.select(db, table, [pkey], row, limit=1,
323
            recover=True))
324
    except StopIteration:
325
        if not create: raise
326
        return put(db, table, row, pkey, row_ct_ref) # insert new row
327

    
328
def is_func_result(col):
329
    return col.table.name.find('(') >= 0 and col.name == 'result'
330

    
331
def into_table_name(out_table, in_tables0, mapping, is_func):
332
    def in_col_str(in_col):
333
        in_col = sql_gen.remove_col_rename(in_col)
334
        if isinstance(in_col, sql_gen.Col):
335
            table = in_col.table
336
            if table == in_tables0:
337
                in_col = sql_gen.to_name_only_col(in_col)
338
            elif is_func_result(in_col): in_col = table # omit col name
339
        return strings.ustr(in_col)
340
    
341
    str_ = strings.ustr(out_table)
342
    if is_func:
343
        str_ += '('
344
        
345
        try: value_in_col = mapping['value']
346
        except KeyError:
347
            str_ += ', '.join((strings.ustr(k)+'='+in_col_str(v)
348
                for k, v in mapping.iteritems()))
349
        else: str_ += in_col_str(value_in_col)
350
        
351
        str_ += ')'
352
    else:
353
        out_col = 'rank'
354
        try: in_col = mapping[out_col]
355
        except KeyError: str_ += '_pkeys'
356
        else: # has a rank column, so hierarchical
357
            str_ += '['+strings.ustr(out_col)+'='+in_col_str(in_col)+']'
358
    return str_
359

    
360
def put_table(db, out_table, in_tables, mapping, row_ct_ref=None, default=None,
361
    col_defaults={}, on_error=exc.reraise):
362
    '''Recovers from errors.
363
    Only works under PostgreSQL (uses INSERT RETURNING).
364
    
365
    Warning: This function's normalizing algorithm does not support database
366
    triggers that populate fields covered by the unique constraint used to do
367
    the DISTINCT ON. Such fields must be populated by the mappings instead.
368
    (Other unique constraints and other non-unique fields are not affected by
369
    this restriction on triggers. Note that the primary key will normally not be
370
    the DISTINCT ON constraint, so trigger-populated natural keys are supported
371
    *unless* the input table contains duplicate rows for some generated keys.)
372
    
373
    Note that much of the complexity of the normalizing algorithm is due to
374
    PostgreSQL (and other DB systems) not having a native command for
375
    insert/on duplicate select. This operation is a cross between MySQL's
376
    INSERT ON DUPLICATE KEY UPDATE (which does not provide SELECT
377
    functionality), and PostgreSQL's INSERT RETURNING (which throws an error
378
    on duplicate instead of returning the existing row).
379
    
380
    @param in_tables The main input table to select from, followed by a list of
381
        tables to join with it using the main input table's pkey
382
    @param mapping dict(out_table_col=in_table_col, ...)
383
        * out_table_col: str (*not* sql_gen.Col)
384
        * in_table_col: sql_gen.Col|literal-value
385
    @param default The *output* column to use as the pkey for missing rows.
386
        If this output column does not exist in the mapping, uses None.
387
    @param col_defaults Default values for required columns.
388
    @return sql_gen.Col Where the output pkeys are made available
389
    '''
390
    import psycopg2.extensions
391
    
392
    # Special handling for functions with hstore params
393
    if out_table == '_map':
394
        import psycopg2.extras
395
        psycopg2.extras.register_hstore(db.db)
396
        
397
        # Parse args
398
        try: value = mapping.pop('value')
399
        except KeyError: return None # value required
400
        
401
        mapping = dict([(k, sql_gen.get_value(v))
402
            for k, v in mapping.iteritems()]) # unwrap literal value
403
        mapping = dict(map=mapping, value=value) # non-value params -> hstore
404
    
405
    out_table = sql_gen.as_Table(out_table)
406
    
407
    def log_debug(msg): db.log_debug(msg, level=1.5)
408
    def col_ustr(str_):
409
        return strings.repr_no_u(sql_gen.remove_col_rename(str_))
410
    
411
    log_debug('********** New iteration **********')
412
    log_debug('Inserting these input columns into '+strings.as_tt(
413
        out_table.to_str(db))+':\n'+strings.as_table(mapping, ustr=col_ustr))
414
    
415
    is_function = sql.function_exists(db, out_table)
416
    
417
    if is_function: row_ct_ref = None # only track inserted rows
418
    
419
    # Warn if inserting empty table rows
420
    if not mapping and not is_function: # functions with no args OK
421
        warnings.warn(UserWarning('Inserting empty table row(s)'))
422
    
423
    if is_function: out_pkey = 'result'
424
    else: out_pkey = sql.pkey_name(db, out_table, recover=True)
425
    out_pkey_col = sql_gen.as_Col(out_pkey, out_table)
426
    
427
    in_tables_ = copy.copy(in_tables) # don't modify input!
428
    try: in_tables0 = in_tables_.pop(0) # first table is separate
429
    except IndexError: in_tables0 = None
430
    else:
431
        in_pkey = sql.pkey_name(db, in_tables0, recover=True)
432
        in_pkey_col = sql_gen.as_Col(in_pkey, in_tables0)
433
    
434
    # Determine if can use optimization for only literal values
435
    is_literals = not reduce(operator.or_, map(sql_gen.is_table_col,
436
        mapping.values()), False)
437
    is_literals_or_function = is_literals or is_function
438
    
439
    if in_tables0 == None: errors_table_ = None
440
    else: errors_table_ = errors_table(db, in_tables0)
441
    
442
    # Create input joins from list of input tables
443
    input_joins = [in_tables0]+[sql_gen.Join(v,
444
        {in_pkey: sql_gen.join_same_not_null}) for v in in_tables_]
445
    
446
    orig_mapping = mapping.copy()
447
    if mapping == {} and not is_function: # need >= one column for INSERT SELECT
448
        mapping = {out_pkey: None} # ColDict will replace with default value
449
    
450
    if not is_literals:
451
        into = sql_gen.as_Table(into_table_name(out_table, in_tables0, mapping,
452
            is_function))
453
        # Ensure into's out_pkey is different from in_pkey by prepending "out."
454
        if is_function: into_out_pkey = out_pkey
455
        else: into_out_pkey = 'out.'+out_pkey
456
        
457
        # Set column sources
458
        in_cols = filter(sql_gen.is_table_col, mapping.values())
459
        for col in in_cols:
460
            if col.table == in_tables0: col.set_srcs(sql_gen.src_self)
461
        
462
        log_debug('Joining together input tables into temp table')
463
        # Place in new table so don't modify input and for speed
464
        in_table = sql_gen.Table('in')
465
        mapping = dicts.join(mapping, sql.flatten(db, in_table, input_joins,
466
            in_cols, preserve=[in_pkey_col]))
467
        input_joins = [in_table]
468
        db.log_debug('Temp table: '+strings.as_tt(in_table.to_str(db)), level=2)
469
    
470
    # Wrap mapping in a sql_gen.ColDict.
471
    # sql_gen.ColDict sanitizes both keys and values passed into it.
472
    # Do after applying dicts.join() because that returns a plain dict.
473
    mapping = sql_gen.ColDict(db, out_table, mapping)
474
    
475
    # Save all rows since in_table may have rows deleted
476
    if is_literals: pass
477
    elif is_function: full_in_table = in_table
478
    else:
479
        full_in_table = sql_gen.suffixed_table(in_table, '_full')
480
        sql.copy_table(db, in_table, full_in_table)
481
    
482
    pkeys_table_exists_ref = [False]
483
    def insert_into_pkeys(query, **kw_args):
484
        if pkeys_table_exists_ref[0]:
485
            sql.insert_select(db, into, [in_pkey, into_out_pkey], query,
486
                **kw_args)
487
        else:
488
            kw_args.setdefault('add_pkey_', True)
489
            
490
            sql.run_query_into(db, query, into=into, **kw_args)
491
            pkeys_table_exists_ref[0] = True
492
    
493
    def mk_main_select(joins, cols): return sql.mk_select(db, joins, cols)
494
    
495
    if is_literals: insert_in_table = None
496
    else:
497
        insert_in_table = in_table
498
        insert_in_tables = [insert_in_table]
499
    join_cols = sql_gen.ColDict(db, out_table)
500
    
501
    exc_strs = set()
502
    def log_exc(e):
503
        e_str = exc.str_(e, first_line_only=True)
504
        log_debug('Caught exception: '+e_str)
505
        if e_str in exc_strs: # avoid infinite loops
506
            log_debug('Exception already seen, handler broken')
507
            on_error(e)
508
            remove_all_rows()
509
            return False
510
        else: exc_strs.add(e_str)
511
        return True
512
    
513
    ignore_all_ref = [False]
514
    def remove_all_rows():
515
        log_debug('Ignoring all rows')
516
        ignore_all_ref[0] = True # just return the default value column
517
    
518
    def handle_unknown_exc(e):
519
        log_debug('No handler for exception')
520
        on_error(e)
521
        remove_all_rows()
522
    
523
    def ensure_cond(cond, e, passed=False, failed=False):
524
        if is_literals: # we know the constraint was applied exactly once
525
            if passed: pass
526
            elif failed: remove_all_rows()
527
            else: raise NotImplementedError()
528
        else:
529
            if not is_function:
530
                out_table_cols = sql_gen.ColDict(db, out_table)
531
                out_table_cols.update(util.dict_subset_right_join({},
532
                    sql.table_col_names(db, out_table)))
533
            
534
            in_cols = []
535
            cond = strings.ustr(cond)
536
            orig_cond = cond
537
            cond = sql_gen.map_expr(db, cond, mapping, in_cols)
538
            if not is_function:
539
                cond = sql_gen.map_expr(db, cond, out_table_cols)
540
            
541
            log_debug('Ignoring rows that do not satisfy '+strings.as_tt(cond))
542
            cur = None
543
            if cond == sql_gen.false_expr:
544
                assert failed
545
                remove_all_rows()
546
            elif cond == sql_gen.true_expr: assert passed
547
            else:
548
                while True:
549
                    not_cond = sql_gen.NotCond(sql_gen.CustomCode(cond))
550
                    try:
551
                        cur = sql.delete(db, insert_in_table, not_cond)
552
                        break
553
                    except sql.DoesNotExistException, e:
554
                        if e.type != 'column': raise
555
                        
556
                        last_cond = cond
557
                        cond = sql_gen.map_expr(db, cond, {e.name: None})
558
                        if cond == last_cond: raise # not fixable
559
            
560
            # If any rows failed cond
561
            if failed or cur != None and cur.rowcount > 0:
562
                track_data_error(db, errors_table_,
563
                    sql_gen.cross_join_srcs(in_cols), None, e.cause.pgcode,
564
                    strings.ensure_newl(strings.ustr(e.cause.pgerror))
565
                    +'condition: '+orig_cond+'\ntranslated condition: '+cond)
566
    
567
    not_null_cols = set()
568
    def ignore(in_col, value, e):
569
        if sql_gen.is_table_col(in_col):
570
            in_col = sql_gen.with_table(in_col, insert_in_table)
571
            
572
            track_data_error(db, errors_table_, in_col.srcs, value,
573
                e.cause.pgcode, e.cause.pgerror)
574
            
575
            sql.add_index(db, in_col, insert_in_table) # enable fast filtering
576
            if value != None and in_col not in not_null_cols:
577
                log_debug('Replacing invalid value '
578
                    +strings.as_tt(strings.urepr(value))+' with NULL in column '
579
                    +strings.as_tt(in_col.to_str(db)))
580
                sql.update(db, insert_in_table, [(in_col, None)],
581
                    sql_gen.ColValueCond(in_col, value))
582
            else:
583
                log_debug('Ignoring rows with '+strings.as_tt(in_col.to_str(db))
584
                    +' = '+strings.as_tt(strings.urepr(value)))
585
                sql.delete(db, insert_in_table,
586
                    sql_gen.ColValueCond(in_col, value))
587
                if value == None: not_null_cols.add(in_col)
588
        else:
589
            assert isinstance(in_col, sql_gen.NamedCol)
590
            in_value = sql_gen.remove_col_rename(in_col)
591
            assert sql_gen.is_literal(in_value)
592
            if value == in_value.value:
593
                if value != None:
594
                    log_debug('Replacing invalid literal '
595
                        +strings.as_tt(in_col.to_str(db))+' with NULL')
596
                    mapping[in_col.name] = None
597
                else:
598
                    remove_all_rows()
599
            # otherwise, all columns were being ignore()d because the specific
600
            # column couldn't be identified, and this was not the invalid column
601
    
602
    if not is_literals:
603
        def insert_pkeys_table(which):
604
            return sql_gen.Table(sql_gen.concat(in_table.name,
605
                '_insert_'+which+'_pkeys'))
606
        insert_out_pkeys = insert_pkeys_table('out')
607
        insert_in_pkeys = insert_pkeys_table('in')
608
    
609
    def mk_func_call():
610
        args = dict(((k.name, v) for k, v in mapping.iteritems()))
611
        return sql_gen.FunctionCall(out_table, **args), args
612
    
613
    missing_msg = None
614
    
615
    # Do inserts and selects
616
    while True:
617
        has_joins = join_cols != {}
618
        
619
        if ignore_all_ref[0]: break # unrecoverable error, so don't do main case
620
        
621
        # Prepare to insert new rows
622
        if is_function:
623
            if is_literals:
624
                log_debug('Calling function')
625
                func_call, args = mk_func_call()
626
        else:
627
            log_debug('Trying to insert new rows')
628
            insert_args = dict(recover=True, cacheable=False)
629
            if has_joins:
630
                insert_args.update(dict(ignore=True))
631
            else:
632
                insert_args.update(dict(returning=out_pkey))
633
                if not is_literals:
634
                    insert_args.update(dict(into=insert_out_pkeys))
635
            main_select = mk_main_select([insert_in_table], [sql_gen.with_table(
636
                c, insert_in_table) for c in mapping.values()])
637
        
638
        try:
639
            cur = None
640
            if is_function:
641
                if is_literals:
642
                    cur = sql.select(db, fields=[func_call], recover=True,
643
                        cacheable=True)
644
                else:
645
                    log_debug('Defining wrapper function')
646
                    
647
                    func_call, args = mk_func_call()
648
                    func_call = sql_gen.NamedCol(into_out_pkey, func_call)
649
                    
650
                    # Create empty pkeys table so its row type can be used
651
                    insert_into_pkeys(sql.mk_select(db, input_joins,
652
                        [in_pkey_col, func_call], limit=0), add_pkey_=False,
653
                        recover=True)
654
                    result_type = db.col_info(sql_gen.Col(into_out_pkey,
655
                        into)).type
656
                    
657
                    ## Create error handling wrapper function
658
                    
659
                    wrapper = db.TempFunction(sql_gen.concat(into.name,
660
                        '_wrap'))
661
                    
662
                    select_cols = [in_pkey_col]+args.values()
663
                    row_var = copy.copy(sql_gen.row_var)
664
                    row_var.set_srcs([in_table])
665
                    in_pkey_var = sql_gen.Col(in_pkey, row_var)
666
                    
667
                    args = dict(((k, sql_gen.with_table(v, row_var))
668
                        for k, v in args.iteritems()))
669
                    func_call = sql_gen.FunctionCall(out_table, **args)
670
                    
671
                    def mk_return(result):
672
                        return sql_gen.ReturnQuery(sql.mk_select(db,
673
                            fields=[in_pkey_var, result], explain=False))
674
                    exc_handler = func_wrapper_exception_handler(db,
675
                        mk_return(sql_gen.Cast(result_type, None)),
676
                        args.values(), errors_table_)
677
                    
678
                    sql.define_func(db, sql_gen.FunctionDef(wrapper,
679
                        sql_gen.SetOf(into),
680
                        sql_gen.RowExcIgnore(sql_gen.RowType(in_table),
681
                            sql.mk_select(db, input_joins),
682
                            mk_return(func_call), exc_handler=exc_handler)
683
                        ))
684
                    wrapper_table = sql_gen.FunctionCall(wrapper)
685
                    
686
                    log_debug('Calling function')
687
                    insert_into_pkeys(sql.mk_select(db, wrapper_table,
688
                        order_by=None), recover=True, cacheable=False)
689
                    sql.add_pkey_or_index(db, into)
690
            else:
691
                cur = sql.insert_select(db, out_table, mapping.keys(),
692
                    main_select, **insert_args)
693
            break # insert successful
694
        except sql.MissingCastException, e:
695
            if not log_exc(e): break
696
            
697
            type_ = e.type
698
            if e.col == None: out_cols = mapping.keys()
699
            else: out_cols = [e.col]
700
            
701
            for out_col in out_cols:
702
                log_debug('Casting '+strings.as_tt(strings.repr_no_u(out_col))
703
                    +' input to '+strings.as_tt(type_))
704
                in_col = mapping[out_col]
705
                while True:
706
                    try:
707
                        mapping[out_col] = cast_temp_col(db, type_, in_col,
708
                            errors_table_)
709
                        break # cast successful
710
                    except sql.InvalidValueException, e:
711
                        if not log_exc(e): break
712
                        
713
                        ignore(in_col, e.value, e)
714
        except sql.DuplicateKeyException, e:
715
            if not log_exc(e): break
716
            
717
            # Different rows violating different unique constraints not
718
            # supported
719
            assert not join_cols
720
            
721
            if e.cond != None: ensure_cond(e.cond, e, passed=True)
722
            
723
            join_cols.update(util.dict_subset_right_join(mapping, e.cols))
724
            log_debug('Ignoring existing rows, comparing on these columns:\n'
725
                +strings.as_inline_table(join_cols, ustr=col_ustr))
726
            
727
            if is_literals:
728
                return sql.value(sql.select(db, out_table, [out_pkey_col],
729
                    join_cols, order_by=None))
730
            
731
            # Uniquify and filter input table to avoid (most) duplicate keys
732
            # (Additional duplicates may be added concurrently and will be
733
            # filtered out separately upon insert.)
734
            insert_in_table = sql.distinct_table(db, insert_in_table,
735
                join_cols.values(), [insert_in_table,
736
                sql_gen.Join(out_table, join_cols, sql_gen.filter_out)])
737
            insert_in_tables.append(insert_in_table)
738
        except sql.NullValueException, e:
739
            if not log_exc(e): break
740
            
741
            out_col, = e.cols
742
            try: in_col = mapping[out_col]
743
            except KeyError, e:
744
                try: in_col = mapping[out_col] = col_defaults[out_col]
745
                except KeyError:
746
                    missing_msg = 'Missing mapping for NOT NULL column '+out_col
747
                    log_debug(missing_msg)
748
                    remove_all_rows()
749
            else: ignore(in_col, None, e)
750
        except sql.CheckException, e:
751
            if not log_exc(e): break
752
            
753
            ensure_cond(e.cond, e, failed=True)
754
        except sql.InvalidValueException, e:
755
            if not log_exc(e): break
756
            
757
            for in_col in mapping.values(): ignore(in_col, e.value, e)
758
        except psycopg2.extensions.TransactionRollbackError, e:
759
            if not log_exc(e): break
760
            # retry
761
        except sql.DatabaseErrors, e:
762
            if not log_exc(e): break
763
            
764
            handle_unknown_exc(e)
765
        # after exception handled, rerun loop with additional constraints
766
    
767
    # Resolve default value column
768
    if default != None:
769
        if ignore_all_ref[0]: mapping.update(orig_mapping) # use input cols
770
        try: default = mapping[default]
771
        except KeyError:
772
            db.log_debug('Default value column '
773
                +strings.as_tt(strings.repr_no_u(default))
774
                +' does not exist in mapping, falling back to None', level=2.1)
775
            default = None
776
        else: default = sql_gen.remove_col_rename(default)
777
    
778
    if missing_msg != None and default == None:
779
        warnings.warn(UserWarning(missing_msg))
780
        # not an error because sometimes the mappings include
781
        # extra tables which aren't used by the dataset
782
    
783
    # Handle unrecoverable errors
784
    if ignore_all_ref[0]:
785
        log_debug('Returning default: '+strings.as_tt(strings.urepr(default)))
786
        return default
787
    
788
    if cur != None and row_ct_ref != None and cur.rowcount >= 0:
789
        row_ct_ref[0] += cur.rowcount
790
    
791
    if is_literals: return sql.value(cur)
792
    
793
    if is_function: pass # pkeys table already created
794
    elif has_joins:
795
        select_joins = input_joins+[sql_gen.Join(out_table, join_cols)]
796
        log_debug('Getting output table pkeys of existing/inserted rows')
797
        insert_into_pkeys(sql.mk_select(db, select_joins, [in_pkey_col,
798
            sql_gen.NamedCol(into_out_pkey, out_pkey_col)], order_by=None))
799
    else:
800
        sql.add_row_num(db, insert_out_pkeys) # for joining with input pkeys
801
        
802
        log_debug('Getting input table pkeys of inserted rows')
803
        # Note that mk_main_select() does not use ORDER BY. Instead, assume that
804
        # since the SELECT query is identical to the one used in INSERT SELECT,
805
        # its rows will be retrieved in the same order.
806
        sql.run_query_into(db, mk_main_select(input_joins, [in_pkey]),
807
            into=insert_in_pkeys)
808
        sql.add_row_num(db, insert_in_pkeys) # for joining with output pkeys
809
        
810
        assert sql.table_row_count(db, insert_out_pkeys) == sql.table_row_count(
811
            db, insert_in_pkeys)
812
        
813
        log_debug('Combining output and input pkeys in inserted order')
814
        pkey_joins = [insert_in_pkeys, sql_gen.Join(insert_out_pkeys,
815
            {sql.row_num_col: sql_gen.join_same_not_null})]
816
        in_col = sql_gen.Col(in_pkey, insert_in_pkeys)
817
        out_col = sql_gen.NamedCol(into_out_pkey,
818
            sql_gen.Col(out_pkey, insert_out_pkeys))
819
        insert_into_pkeys(sql.mk_select(db, pkey_joins, [in_col, out_col],
820
            order_by=None))
821
        
822
        sql.empty_temp(db, [insert_out_pkeys, insert_in_pkeys])
823
    
824
    if not is_function: # is_function doesn't leave holes
825
        log_debug('Setting pkeys of missing rows to '
826
            +strings.as_tt(strings.urepr(default)))
827
        
828
        full_in_pkey_col = sql_gen.Col(in_pkey, full_in_table)
829
        if sql_gen.is_table_col(default):
830
            default = sql_gen.with_table(default, full_in_table)
831
        missing_rows_joins = [full_in_table, sql_gen.Join(into,
832
            {in_pkey: sql_gen.join_same_not_null}, sql_gen.filter_out)]
833
            # must use join_same_not_null or query will take forever
834
        
835
        insert_args = dict(order_by=None)
836
        if not sql.table_has_pkey(db, full_in_table): # in_table has duplicates
837
            insert_args.update(dict(distinct_on=[full_in_pkey_col]))
838
        
839
        insert_into_pkeys(sql.mk_select(db, missing_rows_joins,
840
            [full_in_pkey_col, sql_gen.NamedCol(into_out_pkey, default)],
841
            **insert_args))
842
    # otherwise, there is already an entry for every row
843
    
844
    sql.empty_temp(db, insert_in_tables+[full_in_table])
845
    
846
    srcs = []
847
    if is_function: srcs = sql_gen.cols_srcs(in_cols)
848
    return sql_gen.Col(into_out_pkey, into, srcs)
(37-37/49)