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