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
null_strs = ['', '-', r'\N', 'NULL', 'UNKNOWN', 'nulo']
26

    
27
def cleanup_table(db, table):
28
    table = sql_gen.as_Table(table)
29
    cols = filter(lambda c: sql_gen.is_text_col(db, c),
30
        sql.table_cols(db, table))
31
    try: pkey_col = sql.table_pkey_col(db, table)
32
    except sql.DoesNotExistException: pass
33
    else:
34
        try: cols.remove(pkey_col)
35
        except ValueError: pass
36
    if not cols: return
37
    
38
    db.log_debug('Cleaning up table', level=1.5)
39
    
40
    expr = 'trim(both from %s)'
41
    for null in null_strs: expr = 'nullif('+expr+', '+db.esc_value(null)+')'
42
    changes = [(v, sql_gen.CustomCode(expr % v.to_str(db))) for v in cols]
43
    
44
    while True:
45
        try:
46
            sql.update(db, table, changes, in_place=True, recover=True)
47
            break # successful
48
        except sql.NullValueException, e:
49
            db.log_debug('Caught exception: '+exc.str_(e))
50
            col, = e.cols
51
            sql.drop_not_null(db, col)
52
    
53
    db.log_debug('Vacuuming and reanalyzing table', level=1.5)
54
    sql.vacuum(db, table)
55

    
56
##### Error tracking
57

    
58
def track_data_error(db, errors_table, cols, value, error_code, error):
59
    '''
60
    @param errors_table If None, does nothing.
61
    '''
62
    if errors_table == None: return
63
    
64
    col_names = [c.name for c in cols]
65
    if not col_names: col_names = [None] # need at least one entry
66
    for col_name in col_names:
67
        try:
68
            sql.insert(db, errors_table, dict(column=col_name, value=value,
69
                error_code=error_code, error=error), recover=True,
70
                cacheable=True, log_level=4)
71
        except sql.DuplicateKeyException: pass
72

    
73
class ExcToErrorsTable(sql_gen.ExcToWarning):
74
    '''Handles an exception by saving it or converting it to a warning.'''
75
    def __init__(self, return_, srcs, errors_table, value=None):
76
        '''
77
        @param return_ See sql_gen.ExcToWarning
78
        @param srcs The column names for the errors table
79
        @param errors_table None|sql_gen.Table
80
        @param value The value (or an expression for it) that caused the error
81
        @pre The invalid value must be in a local variable "value" of type text.
82
        '''
83
        sql_gen.ExcToWarning.__init__(self, return_)
84
        
85
        value = sql_gen.as_Code(value)
86
        
87
        self.srcs = srcs
88
        self.errors_table = errors_table
89
        self.value = value
90
    
91
    def to_str(self, db):
92
        if not self.srcs or self.errors_table == None:
93
            return sql_gen.ExcToWarning.to_str(self, db)
94
        
95
        errors_table_cols = map(sql_gen.Col,
96
            ['column', 'value', 'error_code', 'error'])
97
        col_names_query = sql.mk_select(db, sql_gen.NamedValues('c', None,
98
            [[c.name] for c in self.srcs]), order_by=None)
99
        insert_query = sql.mk_insert_select(db, self.errors_table,
100
            errors_table_cols,
101
            sql_gen.Values(errors_table_cols).to_str(db))+';\n'
102
        return '''\
103
-- Save error in errors table.
104
DECLARE
105
    error_code text := SQLSTATE;
106
    error text := SQLERRM;
107
    value text := '''+self.value.to_str(db)+''';
108
    "column" text;
109
BEGIN
110
    -- Insert the value and error for *each* source column.
111
'''+strings.indent(sql_gen.RowExcIgnore(None, col_names_query, insert_query,
112
    row_var=errors_table_cols[0]).to_str(db))+'''
113
END;
114

    
115
'''+self.return_.to_str(db)
116

    
117
def data_exception_handler(*args, **kw_args):
118
    '''Handles a data_exception by saving it or converting it to a warning.
119
    For params, see ExcToErrorsTable().
120
    '''
121
    return sql_gen.data_exception_handler(ExcToErrorsTable(*args, **kw_args))
122

    
123
def cast(db, type_, col, errors_table=None):
124
    '''Casts an (unrenamed) column or value.
125
    If errors_table set and col has srcs, saves errors in errors_table (using
126
    col's srcs attr as source columns). Otherwise, converts errors to warnings.
127
    @param col str|sql_gen.Col|sql_gen.Literal
128
    @param errors_table None|sql_gen.Table|str
129
    '''
130
    col = sql_gen.as_Col(col)
131
    
132
    # Don't convert exceptions to warnings for user-supplied constants
133
    if isinstance(col, sql_gen.Literal): return sql_gen.Cast(type_, col)
134
    
135
    assert not isinstance(col, sql_gen.NamedCol)
136
    
137
    function_name = strings.first_word(type_)
138
    srcs = col.srcs
139
    save_errors = errors_table != None and srcs
140
    if save_errors: # function will be unique for the given srcs
141
        function_name = strings.ustr(sql_gen.FunctionCall(function_name,
142
            *map(sql_gen.to_name_only_col, srcs)))
143
    function = db.TempFunction(function_name)
144
    
145
    # Create function definition
146
    modifiers = 'STRICT'
147
    if not save_errors: modifiers = 'IMMUTABLE '+modifiers
148
    value_param = sql_gen.FunctionParam('value', 'anyelement')
149
    handler = data_exception_handler('RETURN NULL;\n', srcs, errors_table,
150
        value_param.name)
151
    body = sql_gen.CustomCode(handler.to_str(db, '''\
152
/* The explicit cast to the return type is needed to make the cast happen
153
inside the try block. (Implicit casts to the return type happen at the end
154
of the function, outside any block.) */
155
RETURN '''+sql_gen.Cast(type_, sql_gen.CustomCode('value')).to_str(db)+''';
156
'''))
157
    body.lang='plpgsql'
158
    sql.define_func(db, sql_gen.FunctionDef(function, type_, body,
159
        [value_param], modifiers))
160
    
161
    return sql_gen.FunctionCall(function, col)
162

    
163
def func_wrapper_exception_handler(db, return_, args, errors_table):
164
    '''Handles a function call's data_exceptions.
165
    Supports PL/Python functions.
166
    @param return_ See data_exception_handler()
167
    @param args [arg...] Function call's args
168
    @param errors_table See data_exception_handler()
169
    '''
170
    args = filter(sql_gen.has_srcs, args)
171
    
172
    srcs = sql_gen.cross_join_srcs(args)
173
    value = sql_gen.merge_not_null(db, ',', args)
174
    return sql_gen.NestedExcHandler(
175
        data_exception_handler(return_, srcs, errors_table, value)
176
        , sql_gen.plpythonu_error_handler
177
        )
178

    
179
def cast_temp_col(db, type_, col, errors_table=None):
180
    '''Like cast(), but creates a new column with the cast values if the input
181
    is a column.
182
    @return The new column or cast value
183
    '''
184
    def cast_(col): return cast(db, type_, col, errors_table)
185
    
186
    try: col = sql_gen.underlying_col(col)
187
    except sql_gen.NoUnderlyingTableException: return sql_gen.wrap(cast_, col)
188
    
189
    table = col.table
190
    new_col = sql_gen.suffixed_col(col, '::'+strings.first_word(type_))
191
    expr = cast_(col)
192
    
193
    # Add column
194
    new_typed_col = sql_gen.TypedCol(new_col.name, type_)
195
    sql.add_col(db, table, new_typed_col, comment=strings.urepr(col)+'::'+type_)
196
    new_col.name = new_typed_col.name # propagate any renaming
197
    
198
    sql.update(db, table, [(new_col, expr)], in_place=True, recover=True)
199
    
200
    return new_col
201

    
202
def errors_table(db, table, if_exists=True):
203
    '''
204
    @param if_exists If set, returns None if the errors table doesn't exist
205
    @return None|sql_gen.Table
206
    '''
207
    table = sql_gen.as_Table(table)
208
    if table.srcs != (): table = table.srcs[0]
209
    
210
    errors_table = sql_gen.suffixed_table(table, '.errors')
211
    if if_exists and not sql.table_exists(db, errors_table): return None
212
    return errors_table
213

    
214
def mk_errors_table(db, table):
215
    errors_table_ = errors_table(db, table, if_exists=False)
216
    if sql.table_exists(db, errors_table_, cacheable=False): return
217
    
218
    typed_cols = [
219
        sql_gen.TypedCol('column', 'text'),
220
        sql_gen.TypedCol('value', 'text'),
221
        sql_gen.TypedCol('error_code', 'character varying(5)', nullable=False),
222
        sql_gen.TypedCol('error', 'text', nullable=False),
223
        ]
224
    sql.create_table(db, errors_table_, typed_cols, has_pkey=False)
225
    index_cols = ['column', sql_gen.CustomCode('md5(value)'), 'error_code',
226
        sql_gen.CustomCode('md5(error)')]
227
    sql.add_index(db, index_cols, errors_table_, unique=True)
228

    
229
##### Import
230

    
231
row_num_col_def = copy.copy(sql.row_num_col_def)
232
row_num_col_def.name = 'row_num'
233
row_num_col_def.type = 'integer'
234

    
235
def append_csv(db, table, reader, header):
236
    def esc_name_(name): return sql.esc_name(db, name)
237
    
238
    def log(msg, level=1): db.log_debug(msg, level)
239
    
240
    # Wrap in standardizing stream
241
    cols_ct = len(header)
242
    stream = csvs.InputRewriter(streams.ProgressInputStream(csvs.StreamFilter(
243
        csvs.ColCtFilter(reader, cols_ct)), sys.stderr, msg='Read %d row(s)',
244
        n=1000))
245
    dialect = stream.dialect # use default dialect
246
    
247
    # Create COPY FROM statement
248
    if header == sql.table_col_names(db, table): cols_str = ''
249
    else: cols_str =' ('+(', '.join(map(esc_name_, header)))+')'
250
    copy_from = ('COPY '+table.to_str(db)+cols_str+' FROM STDIN DELIMITER '
251
        +db.esc_value(dialect.delimiter)+' NULL '+db.esc_value(''))
252
    assert not csvs.is_tsv(dialect)
253
    copy_from += ' CSV'
254
    if dialect.quoting != csv.QUOTE_NONE:
255
        quote_str = db.esc_value(dialect.quotechar)
256
        copy_from += ' QUOTE '+quote_str
257
        if dialect.doublequote: copy_from += ' ESCAPE '+quote_str
258
    copy_from += ';\n'
259
    
260
    log(copy_from, level=2)
261
    try: db.db.cursor().copy_expert(copy_from, stream)
262
    except Exception, e: sql.parse_exception(db, e, recover=True)
263

    
264
def import_csv(db, table, reader, header):
265
    def log(msg, level=1): db.log_debug(msg, level)
266
    
267
    # Get format info
268
    col_names = map(strings.to_unicode, header)
269
    for i, col in enumerate(col_names): # replace empty column names
270
        if col == '': col_names[i] = 'column_'+str(i)
271
    
272
    # Select schema and escape names
273
    def esc_name(name): return db.esc_name(name)
274
    
275
    typed_cols = [sql_gen.TypedCol(v, 'text') for v in col_names]
276
    typed_cols.insert(0, row_num_col_def)
277
    header.insert(0, row_num_col_def.name)
278
    reader = csvs.RowNumFilter(reader)
279
    
280
    log('Creating table')
281
    # Note that this is not rolled back if the import fails. Instead, it is
282
    # cached, and will not be re-run if the import is retried.
283
    sql.create_table(db, table, typed_cols, has_pkey=False, col_indexes=False)
284
    
285
    # Free memory used by deleted (rolled back) rows from any failed import.
286
    # This MUST be run so that the rows will be stored in inserted order, and
287
    # the row_num added after import will match up with the CSV's row order.
288
    sql.truncate(db, table)
289
    
290
    # Load the data
291
    def load(): append_csv(db, table, reader, header)
292
    sql.with_savepoint(db, load)
293
    
294
    cleanup_table(db, table)
295

    
296
def put(db, table, row, pkey_=None, row_ct_ref=None, on_error=exc.reraise):
297
    '''Recovers from errors.
298
    Only works under PostgreSQL (uses INSERT RETURNING).
299
    '''
300
    return put_table(db, table, [], row, row_ct_ref, on_error=on_error)
301

    
302
def get(db, table, row, pkey, row_ct_ref=None, create=False):
303
    '''Recovers from errors'''
304
    try:
305
        return sql.value(sql.select(db, table, [pkey], row, limit=1,
306
            recover=True))
307
    except StopIteration:
308
        if not create: raise
309
        return put(db, table, row, pkey, row_ct_ref) # insert new row
310

    
311
def is_func_result(col):
312
    return col.table.name.find('(') >= 0 and col.name == 'result'
313

    
314
def into_table_name(out_table, in_tables0, mapping, is_func):
315
    def in_col_str(in_col):
316
        in_col = sql_gen.remove_col_rename(in_col)
317
        if isinstance(in_col, sql_gen.Col):
318
            table = in_col.table
319
            if table == in_tables0:
320
                in_col = sql_gen.to_name_only_col(in_col)
321
            elif is_func_result(in_col): in_col = table # omit col name
322
        return strings.ustr(in_col)
323
    
324
    str_ = strings.ustr(out_table)
325
    if is_func:
326
        str_ += '('
327
        
328
        try: value_in_col = mapping['value']
329
        except KeyError:
330
            str_ += ', '.join((strings.ustr(k)+'='+in_col_str(v)
331
                for k, v in mapping.iteritems()))
332
        else: str_ += in_col_str(value_in_col)
333
        
334
        str_ += ')'
335
    else:
336
        out_col = 'rank'
337
        try: in_col = mapping[out_col]
338
        except KeyError: str_ += '_pkeys'
339
        else: # has a rank column, so hierarchical
340
            str_ += '['+strings.ustr(out_col)+'='+in_col_str(in_col)+']'
341
    return str_
342

    
343
def put_table(db, out_table, in_tables, mapping, row_ct_ref=None, default=None,
344
    col_defaults={}, on_error=exc.reraise):
345
    '''Recovers from errors.
346
    Only works under PostgreSQL (uses INSERT RETURNING).
347
    
348
    Warning: This function's normalizing algorithm does not support database
349
    triggers that populate fields covered by the unique constraint used to do
350
    the DISTINCT ON. Such fields must be populated by the mappings instead.
351
    (Other unique constraints and other non-unique fields are not affected by
352
    this restriction on triggers. Note that the primary key will normally not be
353
    the DISTINCT ON constraint, so trigger-populated natural keys are supported
354
    *unless* the input table contains duplicate rows for some generated keys.)
355
    
356
    Note that much of the complexity of the normalizing algorithm is due to
357
    PostgreSQL (and other DB systems) not having a native command for
358
    insert/on duplicate select. This operation is a cross between MySQL's
359
    INSERT ON DUPLICATE KEY UPDATE (which does not provide SELECT
360
    functionality), and PostgreSQL's INSERT RETURNING (which throws an error
361
    on duplicate instead of returning the existing row).
362
    
363
    @param in_tables The main input table to select from, followed by a list of
364
        tables to join with it using the main input table's pkey
365
    @param mapping dict(out_table_col=in_table_col, ...)
366
        * out_table_col: str (*not* sql_gen.Col)
367
        * in_table_col: sql_gen.Col|literal-value
368
    @param default The *output* column to use as the pkey for missing rows.
369
        If this output column does not exist in the mapping, uses None.
370
    @param col_defaults Default values for required columns.
371
    @return sql_gen.Col Where the output pkeys are made available
372
    '''
373
    import psycopg2.extensions
374
    
375
    # Special handling for functions with hstore params
376
    if out_table == '_map':
377
        import psycopg2.extras
378
        psycopg2.extras.register_hstore(db.db)
379
        
380
        # Parse args
381
        try: value = mapping.pop('value')
382
        except KeyError: return None # value required
383
        
384
        mapping = dict([(k, sql_gen.get_value(v))
385
            for k, v in mapping.iteritems()]) # unwrap literal value
386
        mapping = dict(map=mapping, value=value) # non-value params -> hstore
387
    
388
    out_table = sql_gen.as_Table(out_table)
389
    
390
    def log_debug(msg): db.log_debug(msg, level=1.5)
391
    def col_ustr(str_):
392
        return strings.repr_no_u(sql_gen.remove_col_rename(str_))
393
    
394
    log_debug('********** New iteration **********')
395
    log_debug('Inserting these input columns into '+strings.as_tt(
396
        out_table.to_str(db))+':\n'+strings.as_table(mapping, ustr=col_ustr))
397
    
398
    is_function = sql.function_exists(db, out_table)
399
    
400
    if is_function: row_ct_ref = None # only track inserted rows
401
    
402
    # Warn if inserting empty table rows
403
    if not mapping and not is_function: # functions with no args OK
404
        warnings.warn(UserWarning('Inserting empty table row(s)'))
405
    
406
    if is_function: out_pkey = 'result'
407
    else: out_pkey = sql.pkey_name(db, out_table, recover=True)
408
    out_pkey_col = sql_gen.as_Col(out_pkey, out_table)
409
    
410
    in_tables_ = copy.copy(in_tables) # don't modify input!
411
    try: in_tables0 = in_tables_.pop(0) # first table is separate
412
    except IndexError: in_tables0 = None
413
    else:
414
        in_pkey = sql.pkey_name(db, in_tables0, recover=True)
415
        in_pkey_col = sql_gen.as_Col(in_pkey, in_tables0)
416
    
417
    # Determine if can use optimization for only literal values
418
    is_literals = not reduce(operator.or_, map(sql_gen.is_table_col,
419
        mapping.values()), False)
420
    is_literals_or_function = is_literals or is_function
421
    
422
    if in_tables0 == None: errors_table_ = None
423
    else: errors_table_ = errors_table(db, in_tables0)
424
    
425
    # Create input joins from list of input tables
426
    input_joins = [in_tables0]+[sql_gen.Join(v,
427
        {in_pkey: sql_gen.join_same_not_null}) for v in in_tables_]
428
    
429
    orig_mapping = mapping.copy()
430
    if mapping == {} and not is_function: # need >= one column for INSERT SELECT
431
        mapping = {out_pkey: None} # ColDict will replace with default value
432
    
433
    if not is_literals:
434
        into = sql_gen.as_Table(into_table_name(out_table, in_tables0, mapping,
435
            is_function))
436
        # Ensure into's out_pkey is different from in_pkey by prepending "out."
437
        if is_function: into_out_pkey = out_pkey
438
        else: into_out_pkey = 'out.'+out_pkey
439
        
440
        # Set column sources
441
        in_cols = filter(sql_gen.is_table_col, mapping.values())
442
        for col in in_cols:
443
            if col.table == in_tables0: col.set_srcs(sql_gen.src_self)
444
        
445
        log_debug('Joining together input tables into temp table')
446
        # Place in new table so don't modify input and for speed
447
        in_table = sql_gen.Table('in')
448
        mapping = dicts.join(mapping, sql.flatten(db, in_table, input_joins,
449
            in_cols, preserve=[in_pkey_col]))
450
        input_joins = [in_table]
451
        db.log_debug('Temp table: '+strings.as_tt(in_table.to_str(db)), level=2)
452
    
453
    # Wrap mapping in a sql_gen.ColDict.
454
    # sql_gen.ColDict sanitizes both keys and values passed into it.
455
    # Do after applying dicts.join() because that returns a plain dict.
456
    mapping = sql_gen.ColDict(db, out_table, mapping)
457
    
458
    # Save all rows since in_table may have rows deleted
459
    if is_literals: pass
460
    elif is_function: full_in_table = in_table
461
    else:
462
        full_in_table = sql_gen.suffixed_table(in_table, '_full')
463
        sql.copy_table(db, in_table, full_in_table)
464
    
465
    pkeys_table_exists_ref = [False]
466
    def insert_into_pkeys(query, **kw_args):
467
        if pkeys_table_exists_ref[0]:
468
            sql.insert_select(db, into, [in_pkey, into_out_pkey], query,
469
                **kw_args)
470
        else:
471
            kw_args.setdefault('add_pkey_', True)
472
            
473
            sql.run_query_into(db, query, into=into, **kw_args)
474
            pkeys_table_exists_ref[0] = True
475
    
476
    def mk_main_select(joins, cols): return sql.mk_select(db, joins, cols)
477
    
478
    if is_literals: insert_in_table = None
479
    else:
480
        insert_in_table = in_table
481
        insert_in_tables = [insert_in_table]
482
    join_cols = sql_gen.ColDict(db, out_table)
483
    
484
    exc_strs = set()
485
    def log_exc(e):
486
        e_str = exc.str_(e, first_line_only=True)
487
        log_debug('Caught exception: '+e_str)
488
        if e_str in exc_strs: # avoid infinite loops
489
            log_debug('Exception already seen, handler broken')
490
            on_error(e)
491
            remove_all_rows()
492
            return False
493
        else: exc_strs.add(e_str)
494
        return True
495
    
496
    ignore_all_ref = [False]
497
    def remove_all_rows():
498
        log_debug('Ignoring all rows')
499
        ignore_all_ref[0] = True # just return the default value column
500
    
501
    def handle_unknown_exc(e):
502
        log_debug('No handler for exception')
503
        on_error(e)
504
        remove_all_rows()
505
    
506
    def ensure_cond(cond, e, passed=False, failed=False):
507
        if is_literals: # we know the constraint was applied exactly once
508
            if passed: pass
509
            elif failed: remove_all_rows()
510
            else: raise NotImplementedError()
511
        else:
512
            if not is_function:
513
                out_table_cols = sql_gen.ColDict(db, out_table)
514
                out_table_cols.update(util.dict_subset_right_join({},
515
                    sql.table_col_names(db, out_table)))
516
            
517
            in_cols = []
518
            cond = strings.ustr(cond)
519
            orig_cond = cond
520
            cond = sql_gen.map_expr(db, cond, mapping, in_cols)
521
            if not is_function:
522
                cond = sql_gen.map_expr(db, cond, out_table_cols)
523
            
524
            log_debug('Ignoring rows that do not satisfy '+strings.as_tt(cond))
525
            cur = None
526
            if cond == sql_gen.false_expr:
527
                assert failed
528
                remove_all_rows()
529
            elif cond == sql_gen.true_expr: assert passed
530
            else:
531
                while True:
532
                    not_cond = sql_gen.NotCond(sql_gen.CustomCode(cond))
533
                    try:
534
                        cur = sql.delete(db, insert_in_table, not_cond)
535
                        break
536
                    except sql.DoesNotExistException, e:
537
                        if e.type != 'column': raise
538
                        
539
                        last_cond = cond
540
                        cond = sql_gen.map_expr(db, cond, {e.name: None})
541
                        if cond == last_cond: raise # not fixable
542
            
543
            # If any rows failed cond
544
            if failed or cur != None and cur.rowcount > 0:
545
                track_data_error(db, errors_table_,
546
                    sql_gen.cross_join_srcs(in_cols), None, e.cause.pgcode,
547
                    strings.ensure_newl(strings.ustr(e.cause.pgerror))
548
                    +'condition: '+orig_cond+'\ntranslated condition: '+cond)
549
    
550
    not_null_cols = set()
551
    def ignore(in_col, value, e):
552
        if sql_gen.is_table_col(in_col):
553
            in_col = sql_gen.with_table(in_col, insert_in_table)
554
            
555
            track_data_error(db, errors_table_, in_col.srcs, value,
556
                e.cause.pgcode, e.cause.pgerror)
557
            
558
            sql.add_index(db, in_col, insert_in_table) # enable fast filtering
559
            if value != None and in_col not in not_null_cols:
560
                log_debug('Replacing invalid value '
561
                    +strings.as_tt(strings.urepr(value))+' with NULL in column '
562
                    +strings.as_tt(in_col.to_str(db)))
563
                sql.update(db, insert_in_table, [(in_col, None)],
564
                    sql_gen.ColValueCond(in_col, value))
565
            else:
566
                log_debug('Ignoring rows with '+strings.as_tt(in_col.to_str(db))
567
                    +' = '+strings.as_tt(strings.urepr(value)))
568
                sql.delete(db, insert_in_table,
569
                    sql_gen.ColValueCond(in_col, value))
570
                if value == None: not_null_cols.add(in_col)
571
        else:
572
            assert isinstance(in_col, sql_gen.NamedCol)
573
            in_value = sql_gen.remove_col_rename(in_col)
574
            assert sql_gen.is_literal(in_value)
575
            if value == in_value.value:
576
                if value != None:
577
                    log_debug('Replacing invalid literal '
578
                        +strings.as_tt(in_col.to_str(db))+' with NULL')
579
                    mapping[in_col.name] = None
580
                else:
581
                    remove_all_rows()
582
            # otherwise, all columns were being ignore()d because the specific
583
            # column couldn't be identified, and this was not the invalid column
584
    
585
    if not is_literals:
586
        def insert_pkeys_table(which):
587
            return sql_gen.Table(sql_gen.concat(in_table.name,
588
                '_insert_'+which+'_pkeys'))
589
        insert_out_pkeys = insert_pkeys_table('out')
590
        insert_in_pkeys = insert_pkeys_table('in')
591
    
592
    def mk_func_call():
593
        args = dict(((k.name, v) for k, v in mapping.iteritems()))
594
        return sql_gen.FunctionCall(out_table, **args), args
595
    
596
    missing_msg = None
597
    
598
    # Do inserts and selects
599
    while True:
600
        has_joins = join_cols != {}
601
        
602
        if ignore_all_ref[0]: break # unrecoverable error, so don't do main case
603
        
604
        # Prepare to insert new rows
605
        if is_function:
606
            if is_literals:
607
                log_debug('Calling function')
608
                func_call, args = mk_func_call()
609
        else:
610
            log_debug('Trying to insert new rows')
611
            insert_args = dict(recover=True, cacheable=False)
612
            if has_joins:
613
                insert_args.update(dict(ignore=True))
614
            else:
615
                insert_args.update(dict(returning=out_pkey))
616
                if not is_literals:
617
                    insert_args.update(dict(into=insert_out_pkeys))
618
            main_select = mk_main_select([insert_in_table], [sql_gen.with_table(
619
                c, insert_in_table) for c in mapping.values()])
620
        
621
        try:
622
            cur = None
623
            if is_function:
624
                if is_literals:
625
                    cur = sql.select(db, fields=[func_call], recover=True,
626
                        cacheable=True)
627
                else:
628
                    log_debug('Defining wrapper function')
629
                    
630
                    func_call, args = mk_func_call()
631
                    func_call = sql_gen.NamedCol(into_out_pkey, func_call)
632
                    
633
                    # Create empty pkeys table so its row type can be used
634
                    insert_into_pkeys(sql.mk_select(db, input_joins,
635
                        [in_pkey_col, func_call], limit=0), add_pkey_=False,
636
                        recover=True)
637
                    result_type = db.col_info(sql_gen.Col(into_out_pkey,
638
                        into)).type
639
                    
640
                    ## Create error handling wrapper function
641
                    
642
                    wrapper = db.TempFunction(sql_gen.concat(into.name,
643
                        '_wrap'))
644
                    
645
                    select_cols = [in_pkey_col]+args.values()
646
                    row_var = copy.copy(sql_gen.row_var)
647
                    row_var.set_srcs([in_table])
648
                    in_pkey_var = sql_gen.Col(in_pkey, row_var)
649
                    
650
                    args = dict(((k, sql_gen.with_table(v, row_var))
651
                        for k, v in args.iteritems()))
652
                    func_call = sql_gen.FunctionCall(out_table, **args)
653
                    
654
                    def mk_return(result):
655
                        return sql_gen.ReturnQuery(sql.mk_select(db,
656
                            fields=[in_pkey_var, result], explain=False))
657
                    exc_handler = func_wrapper_exception_handler(db,
658
                        mk_return(sql_gen.Cast(result_type, None)),
659
                        args.values(), errors_table_)
660
                    
661
                    sql.define_func(db, sql_gen.FunctionDef(wrapper,
662
                        sql_gen.SetOf(into),
663
                        sql_gen.RowExcIgnore(sql_gen.RowType(in_table),
664
                            sql.mk_select(db, input_joins),
665
                            mk_return(func_call), exc_handler=exc_handler)
666
                        ))
667
                    wrapper_table = sql_gen.FunctionCall(wrapper)
668
                    
669
                    log_debug('Calling function')
670
                    insert_into_pkeys(sql.mk_select(db, wrapper_table,
671
                        order_by=None), recover=True, cacheable=False)
672
                    sql.add_pkey_or_index(db, into)
673
            else:
674
                cur = sql.insert_select(db, out_table, mapping.keys(),
675
                    main_select, **insert_args)
676
            break # insert successful
677
        except sql.MissingCastException, e:
678
            if not log_exc(e): break
679
            
680
            type_ = e.type
681
            if e.col == None: out_cols = mapping.keys()
682
            else: out_cols = [e.col]
683
            
684
            for out_col in out_cols:
685
                log_debug('Casting '+strings.as_tt(strings.repr_no_u(out_col))
686
                    +' input to '+strings.as_tt(type_))
687
                in_col = mapping[out_col]
688
                while True:
689
                    try:
690
                        mapping[out_col] = cast_temp_col(db, type_, in_col,
691
                            errors_table_)
692
                        break # cast successful
693
                    except sql.InvalidValueException, e:
694
                        if not log_exc(e): break
695
                        
696
                        ignore(in_col, e.value, e)
697
        except sql.DuplicateKeyException, e:
698
            if not log_exc(e): break
699
            
700
            # Different rows violating different unique constraints not
701
            # supported
702
            assert not join_cols
703
            
704
            if e.cond != None: ensure_cond(e.cond, e, passed=True)
705
            
706
            join_cols.update(util.dict_subset_right_join(mapping, e.cols))
707
            log_debug('Ignoring existing rows, comparing on these columns:\n'
708
                +strings.as_inline_table(join_cols, ustr=col_ustr))
709
            
710
            if is_literals:
711
                return sql.value(sql.select(db, out_table, [out_pkey_col],
712
                    join_cols, order_by=None))
713
            
714
            # Uniquify and filter input table to avoid (most) duplicate keys
715
            # (Additional duplicates may be added concurrently and will be
716
            # filtered out separately upon insert.)
717
            insert_in_table = sql.distinct_table(db, insert_in_table,
718
                join_cols.values(), [insert_in_table,
719
                sql_gen.Join(out_table, join_cols, sql_gen.filter_out)])
720
            insert_in_tables.append(insert_in_table)
721
        except sql.NullValueException, e:
722
            if not log_exc(e): break
723
            
724
            out_col, = e.cols
725
            try: in_col = mapping[out_col]
726
            except KeyError, e:
727
                try: in_col = mapping[out_col] = col_defaults[out_col]
728
                except KeyError:
729
                    missing_msg = 'Missing mapping for NOT NULL column '+out_col
730
                    log_debug(missing_msg)
731
                    remove_all_rows()
732
            else: ignore(in_col, None, e)
733
        except sql.CheckException, e:
734
            if not log_exc(e): break
735
            
736
            ensure_cond(e.cond, e, failed=True)
737
        except sql.InvalidValueException, e:
738
            if not log_exc(e): break
739
            
740
            for in_col in mapping.values(): ignore(in_col, e.value, e)
741
        except psycopg2.extensions.TransactionRollbackError, e:
742
            if not log_exc(e): break
743
            # retry
744
        except sql.DatabaseErrors, e:
745
            if not log_exc(e): break
746
            
747
            handle_unknown_exc(e)
748
        # after exception handled, rerun loop with additional constraints
749
    
750
    # Resolve default value column
751
    if default != None:
752
        if ignore_all_ref[0]: mapping.update(orig_mapping) # use input cols
753
        try: default = mapping[default]
754
        except KeyError:
755
            db.log_debug('Default value column '
756
                +strings.as_tt(strings.repr_no_u(default))
757
                +' does not exist in mapping, falling back to None', level=2.1)
758
            default = None
759
        else: default = sql_gen.remove_col_rename(default)
760
    
761
    if missing_msg != None and default == None:
762
        warnings.warn(UserWarning(missing_msg))
763
        # not an error because sometimes the mappings include
764
        # extra tables which aren't used by the dataset
765
    
766
    # Handle unrecoverable errors
767
    if ignore_all_ref[0]:
768
        log_debug('Returning default: '+strings.as_tt(strings.urepr(default)))
769
        return default
770
    
771
    if cur != None and row_ct_ref != None and cur.rowcount >= 0:
772
        row_ct_ref[0] += cur.rowcount
773
    
774
    if is_literals: return sql.value(cur)
775
    
776
    if is_function: pass # pkeys table already created
777
    elif has_joins:
778
        select_joins = input_joins+[sql_gen.Join(out_table, join_cols)]
779
        log_debug('Getting output table pkeys of existing/inserted rows')
780
        insert_into_pkeys(sql.mk_select(db, select_joins, [in_pkey_col,
781
            sql_gen.NamedCol(into_out_pkey, out_pkey_col)], order_by=None))
782
    else:
783
        sql.add_row_num(db, insert_out_pkeys) # for joining with input pkeys
784
        
785
        log_debug('Getting input table pkeys of inserted rows')
786
        # Note that mk_main_select() does not use ORDER BY. Instead, assume that
787
        # since the SELECT query is identical to the one used in INSERT SELECT,
788
        # its rows will be retrieved in the same order.
789
        sql.run_query_into(db, mk_main_select(input_joins, [in_pkey]),
790
            into=insert_in_pkeys)
791
        sql.add_row_num(db, insert_in_pkeys) # for joining with output pkeys
792
        
793
        assert sql.table_row_count(db, insert_out_pkeys) == sql.table_row_count(
794
            db, insert_in_pkeys)
795
        
796
        log_debug('Combining output and input pkeys in inserted order')
797
        pkey_joins = [insert_in_pkeys, sql_gen.Join(insert_out_pkeys,
798
            {sql.row_num_col: sql_gen.join_same_not_null})]
799
        in_col = sql_gen.Col(in_pkey, insert_in_pkeys)
800
        out_col = sql_gen.NamedCol(into_out_pkey,
801
            sql_gen.Col(out_pkey, insert_out_pkeys))
802
        insert_into_pkeys(sql.mk_select(db, pkey_joins, [in_col, out_col],
803
            order_by=None))
804
        
805
        sql.empty_temp(db, [insert_out_pkeys, insert_in_pkeys])
806
    
807
    if not is_function: # is_function doesn't leave holes
808
        log_debug('Setting pkeys of missing rows to '
809
            +strings.as_tt(strings.urepr(default)))
810
        
811
        full_in_pkey_col = sql_gen.Col(in_pkey, full_in_table)
812
        if sql_gen.is_table_col(default):
813
            default = sql_gen.with_table(default, full_in_table)
814
        missing_rows_joins = [full_in_table, sql_gen.Join(into,
815
            {in_pkey: sql_gen.join_same_not_null}, sql_gen.filter_out)]
816
            # must use join_same_not_null or query will take forever
817
        
818
        insert_args = dict(order_by=None)
819
        if not sql.table_has_pkey(db, full_in_table): # in_table has duplicates
820
            insert_args.update(dict(distinct_on=[full_in_pkey_col]))
821
        
822
        insert_into_pkeys(sql.mk_select(db, missing_rows_joins,
823
            [full_in_pkey_col, sql_gen.NamedCol(into_out_pkey, default)],
824
            **insert_args))
825
    # otherwise, there is already an entry for every row
826
    
827
    sql.empty_temp(db, insert_in_tables+[full_in_table])
828
    
829
    srcs = []
830
    if is_function: srcs = sql_gen.cols_srcs(in_cols)
831
    return sql_gen.Col(into_out_pkey, into, srcs)
(37-37/49)