Project

General

Profile

1
# Database import/export
2

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

    
8
import csvs
9
import exc
10
import dicts
11
import sql
12
import sql_gen
13
import strings
14
import util
15

    
16
##### Exceptions
17

    
18
# Can't use built-in SyntaxError because it stringifies to only the first line
19
class SyntaxError(Exception): pass
20

    
21
##### Data cleanup
22

    
23
null_strs = ['', '-', r'\N', 'NULL', 'UNKNOWN', 'nulo']
24

    
25
def cleanup_table(db, table):
26
    table = sql_gen.as_Table(table)
27
    cols = filter(lambda c: sql_gen.is_text_col(db, c),
28
        sql.table_cols(db, table))
29
    try: pkey_col = sql.table_pkey_col(db, table)
30
    except sql.DoesNotExistException: pass
31
    else:
32
        try: cols.remove(pkey_col)
33
        except ValueError: pass
34
    if not cols: return
35
    
36
    db.log_debug('Cleaning up table', level=1.5)
37
    
38
    expr = 'trim(both from %s)'
39
    for null in null_strs: expr = 'nullif('+expr+', '+db.esc_value(null)+')'
40
    changes = [(v, sql_gen.CustomCode(expr % v.to_str(db))) for v in cols]
41
    
42
    while True:
43
        try:
44
            sql.update(db, table, changes, in_place=True, recover=True)
45
            break # successful
46
        except sql.NullValueException, e:
47
            db.log_debug('Caught exception: '+exc.str_(e))
48
            col, = e.cols
49
            sql.drop_not_null(db, col)
50
    
51
    db.log_debug('Vacuuming and reanalyzing table', level=1.5)
52
    sql.vacuum(db, table)
53

    
54
##### Error tracking
55

    
56
def track_data_error(db, errors_table, cols, value, error_code, error):
57
    '''
58
    @param errors_table If None, does nothing.
59
    '''
60
    if errors_table == None or cols == (): return
61
    
62
    for col in cols:
63
        try:
64
            sql.insert(db, errors_table, dict(column=col.name, value=value,
65
                error_code=error_code, error=error), recover=True,
66
                cacheable=True, log_level=4)
67
        except sql.DuplicateKeyException: pass
68

    
69
class ExcToErrorsTable(sql_gen.ExcToWarning):
70
    '''Handles an exception by saving it or converting it to a warning.'''
71
    def __init__(self, return_, srcs, errors_table, value=None):
72
        '''
73
        @param return_ See sql_gen.ExcToWarning
74
        @param srcs The column names for the errors table
75
        @param errors_table None|sql_gen.Table
76
        @param value The value (or an expression for it) that caused the error
77
        @pre The invalid value must be in a local variable "value" of type text.
78
        '''
79
        sql_gen.ExcToWarning.__init__(self, return_)
80
        
81
        value = sql_gen.as_Code(value)
82
        
83
        self.srcs = srcs
84
        self.errors_table = errors_table
85
        self.value = value
86
    
87
    def to_str(self, db):
88
        if not self.srcs or self.errors_table == None:
89
            return sql_gen.ExcToWarning.to_str(self, db)
90
        
91
        errors_table_cols = map(sql_gen.Col,
92
            ['column', 'value', 'error_code', 'error'])
93
        col_names_query = sql.mk_select(db, sql_gen.NamedValues('c', None,
94
            [[c.name] for c in self.srcs]), order_by=None)
95
        insert_query = sql.mk_insert_select(db, self.errors_table,
96
            errors_table_cols,
97
            sql_gen.Values(errors_table_cols).to_str(db))+';\n'
98
        return '''\
99
-- Save error in errors table.
100
DECLARE
101
    error_code text := SQLSTATE;
102
    error text := SQLERRM;
103
    value text := '''+self.value.to_str(db)+''';
104
    "column" text;
105
BEGIN
106
    -- Insert the value and error for *each* source column.
107
'''+strings.indent(sql_gen.RowExcIgnore(None, col_names_query, insert_query,
108
    row_var=errors_table_cols[0]).to_str(db))+'''
109
END;
110

    
111
'''+self.return_.to_str(db)
112

    
113
def data_exception_handler(*args, **kw_args):
114
    '''Handles a data_exception by saving it or converting it to a warning.
115
    For params, see ExcToErrorsTable().
116
    '''
117
    return sql_gen.data_exception_handler(ExcToErrorsTable(*args, **kw_args))
118

    
119
def cast(db, type_, col, errors_table=None):
120
    '''Casts an (unrenamed) column or value.
121
    If errors_table set and col has srcs, saves errors in errors_table (using
122
    col's srcs attr as source columns). Otherwise, converts errors to warnings.
123
    @param col str|sql_gen.Col|sql_gen.Literal
124
    @param errors_table None|sql_gen.Table|str
125
    '''
126
    col = sql_gen.as_Col(col)
127
    
128
    # Don't convert exceptions to warnings for user-supplied constants
129
    if isinstance(col, sql_gen.Literal): return sql_gen.Cast(type_, col)
130
    
131
    assert not isinstance(col, sql_gen.NamedCol)
132
    
133
    function_name = strings.first_word(type_)
134
    srcs = col.srcs
135
    save_errors = errors_table != None and srcs
136
    if save_errors: # function will be unique for the given srcs
137
        function_name = strings.ustr(sql_gen.FunctionCall(function_name,
138
            *map(sql_gen.to_name_only_col, srcs)))
139
    function = db.TempFunction(function_name)
140
    
141
    # Create function definition
142
    modifiers = 'STRICT'
143
    if not save_errors: modifiers = 'IMMUTABLE '+modifiers
144
    value_param = sql_gen.FunctionParam('value', 'text')
145
    handler = data_exception_handler('RETURN NULL;\n', srcs, errors_table,
146
        value_param.name)
147
    body = sql_gen.CustomCode(handler.to_str(db, '''\
148
/* The explicit cast to the return type is needed to make the cast happen
149
inside the try block. (Implicit casts to the return type happen at the end
150
of the function, outside any block.) */
151
RETURN value::'''+type_+''';
152
'''))
153
    body.lang='plpgsql'
154
    sql.define_func(db, sql_gen.FunctionDef(function, type_, body,
155
        [value_param], modifiers))
156
    
157
    return sql_gen.FunctionCall(function, col)
158

    
159
def func_wrapper_exception_handler(db, return_, args, errors_table):
160
    '''Handles a function call's data_exceptions.
161
    Supports PL/Python functions.
162
    @param return_ See data_exception_handler()
163
    @param args [arg...] Function call's args
164
    @param errors_table See data_exception_handler()
165
    '''
166
    args = filter(sql_gen.has_srcs, args)
167
    
168
    srcs = sql_gen.cross_join_srcs(args)
169
    value = sql_gen.merge_not_null(db, ',', args)
170
    return sql_gen.NestedExcHandler(
171
        data_exception_handler(return_, srcs, errors_table, value)
172
        , sql_gen.plpythonu_error_handler
173
        )
174

    
175
def cast_temp_col(db, type_, col, errors_table=None):
176
    '''Like cast(), but creates a new column with the cast values if the input
177
    is a column.
178
    @return The new column or cast value
179
    '''
180
    def cast_(col): return cast(db, type_, col, errors_table)
181
    
182
    try: col = sql_gen.underlying_col(col)
183
    except sql_gen.NoUnderlyingTableException: return sql_gen.wrap(cast_, col)
184
    
185
    table = col.table
186
    new_col = sql_gen.suffixed_col(col, '::'+strings.first_word(type_))
187
    expr = cast_(col)
188
    
189
    # Add column
190
    new_typed_col = sql_gen.TypedCol(new_col.name, type_)
191
    sql.add_col(db, table, new_typed_col, comment=strings.urepr(col)+'::'+type_)
192
    new_col.name = new_typed_col.name # propagate any renaming
193
    
194
    sql.update(db, table, [(new_col, expr)], in_place=True, recover=True)
195
    
196
    return new_col
197

    
198
def errors_table(db, table, if_exists=True):
199
    '''
200
    @param if_exists If set, returns None if the errors table doesn't exist
201
    @return None|sql_gen.Table
202
    '''
203
    table = sql_gen.as_Table(table)
204
    if table.srcs != (): table = table.srcs[0]
205
    
206
    errors_table = sql_gen.suffixed_table(table, '.errors')
207
    if if_exists and not sql.table_exists(db, errors_table): return None
208
    return errors_table
209

    
210
def mk_errors_table(db, table):
211
    errors_table_ = errors_table(db, table, if_exists=False)
212
    if sql.table_exists(db, errors_table_, cacheable=False): return
213
    
214
    typed_cols = [
215
        sql_gen.TypedCol('column', 'text', nullable=False),
216
        sql_gen.TypedCol('value', 'text'),
217
        sql_gen.TypedCol('error_code', 'character varying(5)', nullable=False),
218
        sql_gen.TypedCol('error', 'text', nullable=False),
219
        ]
220
    sql.create_table(db, errors_table_, typed_cols, has_pkey=False)
221
    index_cols = ['column', 'value', 'error_code', 'error']
222
    sql.add_index(db, index_cols, errors_table_, unique=True)
223

    
224
##### Import
225

    
226
def append_csv(db, table, stream_info, stream, use_copy_from=True):
227
    assert sql.table_col_names(db, table) == stream_info.header
228
    
229
    def log(msg, level=1): db.log_debug(msg, level)
230
    
231
    dialect = stream_info.dialect
232
    if csvs.is_tsv(dialect): use_copy_from = False
233
    if use_copy_from:
234
        log('Using COPY FROM')
235
        
236
        # Create COPY FROM statement
237
        copy_from = ('COPY '+table.to_str(db)+' FROM STDIN DELIMITER '
238
            +db.esc_value(dialect.delimiter)+' NULL '+db.esc_value(''))
239
        assert not csvs.is_tsv(dialect)
240
        copy_from += ' CSV'
241
        if dialect.quoting != csv.QUOTE_NONE:
242
            quote_str = db.esc_value(dialect.quotechar)
243
            copy_from += ' QUOTE '+quote_str
244
            if dialect.doublequote: copy_from += ' ESCAPE '+quote_str
245
        copy_from += ';\n'
246
        
247
        log(copy_from, level=2)
248
        db.db.cursor().copy_expert(copy_from, stream)
249
    else:
250
        log('Using INSERT')
251
        cols_ct = len(stream_info.header)
252
        row_ct = 0
253
        inserted_row_ct = 0
254
        for row in csvs.make_reader(stream, dialect):
255
            row_ct += 1
256
            row = map(strings.to_unicode, row)
257
            util.list_set_length(row, cols_ct) # truncate extra cols
258
            if sql.insert(db, table, row, ignore=True, cacheable=False,
259
                log_level=5) != None:
260
                inserted_row_ct += 1
261
        log('Inserted '+str(inserted_row_ct)+' of '+str(row_ct)+' rows')
262

    
263
def import_csv(db, table, stream, use_copy_from=True):
264
    def log(msg, level=1): db.log_debug(msg, level)
265
    
266
    # Get format info
267
    info = csvs.stream_info(stream, parse_header=True)
268
    col_names = map(strings.to_unicode, info.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
    
277
    log('Creating table')
278
    # Note that this is not rolled back if the import fails. Instead, it is
279
    # cached, and will not be re-run if the import is retried.
280
    sql.create_table(db, table, typed_cols, has_pkey=False, col_indexes=False)
281
    
282
    # Free memory used by deleted (rolled back) rows from any failed import.
283
    # This MUST be run so that the rows will be stored in inserted order, and
284
    # the row_num added after import will match up with the CSV's row order.
285
    sql.truncate(db, table)
286
    
287
    # Load the data
288
    def load(): append_csv(db, table, info, stream, use_copy_from)
289
    if use_copy_from: sql.with_savepoint(db, load)
290
    else: load()
291
    
292
    cleanup_table(db, table)
293

    
294
def put(db, table, row, pkey_=None, row_ct_ref=None):
295
    '''Recovers from errors.
296
    Only works under PostgreSQL (uses INSERT RETURNING).
297
    '''
298
    return put_table(db, table, [], row, row_ct_ref)
299

    
300
def get(db, table, row, pkey, row_ct_ref=None, create=False):
301
    '''Recovers from errors'''
302
    try:
303
        return sql.value(sql.select(db, table, [pkey], row, limit=1,
304
            recover=True))
305
    except StopIteration:
306
        if not create: raise
307
        return put(db, table, row, pkey, row_ct_ref) # insert new row
308

    
309
def is_func_result(col):
310
    return col.table.name.find('(') >= 0 and col.name == 'result'
311

    
312
def into_table_name(out_table, in_tables0, mapping, is_func):
313
    def in_col_str(in_col):
314
        in_col = sql_gen.remove_col_rename(in_col)
315
        if isinstance(in_col, sql_gen.Col):
316
            table = in_col.table
317
            if table == in_tables0:
318
                in_col = sql_gen.to_name_only_col(in_col)
319
            elif is_func_result(in_col): in_col = table # omit col name
320
        return strings.ustr(in_col)
321
    
322
    str_ = strings.ustr(out_table)
323
    if is_func:
324
        str_ += '('
325
        
326
        try: value_in_col = mapping['value']
327
        except KeyError:
328
            str_ += ', '.join((strings.ustr(k)+'='+in_col_str(v)
329
                for k, v in mapping.iteritems()))
330
        else: str_ += in_col_str(value_in_col)
331
        
332
        str_ += ')'
333
    else:
334
        out_col = 'rank'
335
        try: in_col = mapping[out_col]
336
        except KeyError: str_ += '_pkeys'
337
        else: # has a rank column, so hierarchical
338
            str_ += '['+strings.ustr(out_col)+'='+in_col_str(in_col)+']'
339
    return str_
340

    
341
def put_table(db, out_table, in_tables, mapping, row_ct_ref=None, default=None,
342
    col_defaults={}, on_error=exc.reraise):
343
    '''Recovers from errors.
344
    Only works under PostgreSQL (uses INSERT RETURNING).
345
    IMPORTANT: Must be run at the *beginning* of a transaction.
346
    @param in_tables The main input table to select from, followed by a list of
347
        tables to join with it using the main input table's pkey
348
    @param mapping dict(out_table_col=in_table_col, ...)
349
        * out_table_col: str (*not* sql_gen.Col)
350
        * in_table_col: sql_gen.Col|literal-value
351
    @param default The *output* column to use as the pkey for missing rows.
352
        If this output column does not exist in the mapping, uses None.
353
    @param col_defaults Default values for required columns.
354
    @return sql_gen.Col Where the output pkeys are made available
355
    '''
356
    import psycopg2.extensions
357
    
358
    out_table = sql_gen.as_Table(out_table)
359
    
360
    def log_debug(msg): db.log_debug(msg, level=1.5)
361
    def col_ustr(str_):
362
        return strings.repr_no_u(sql_gen.remove_col_rename(str_))
363
    
364
    log_debug('********** New iteration **********')
365
    log_debug('Inserting these input columns into '+strings.as_tt(
366
        out_table.to_str(db))+':\n'+strings.as_table(mapping, ustr=col_ustr))
367
    
368
    is_function = sql.function_exists(db, out_table)
369
    
370
    if is_function: row_ct_ref = None # only track inserted rows
371
    
372
    # Warn if inserting empty table rows
373
    if not mapping and not is_function: # functions with no args OK
374
        warnings.warn(UserWarning('Inserting empty table row(s)'))
375
    
376
    if is_function: out_pkey = 'result'
377
    else: out_pkey = sql.pkey_name(db, out_table, recover=True)
378
    out_pkey_col = sql_gen.as_Col(out_pkey, out_table)
379
    
380
    in_tables_ = copy.copy(in_tables) # don't modify input!
381
    try: in_tables0 = in_tables_.pop(0) # first table is separate
382
    except IndexError: in_tables0 = None
383
    else:
384
        in_pkey = sql.pkey_name(db, in_tables0, recover=True)
385
        in_pkey_col = sql_gen.as_Col(in_pkey, in_tables0)
386
    
387
    # Determine if can use optimization for only literal values
388
    is_literals = not reduce(operator.or_, map(sql_gen.is_table_col,
389
        mapping.values()), False)
390
    is_literals_or_function = is_literals or is_function
391
    
392
    if in_tables0 == None: errors_table_ = None
393
    else: errors_table_ = errors_table(db, in_tables0)
394
    
395
    # Create input joins from list of input tables
396
    input_joins = [in_tables0]+[sql_gen.Join(v,
397
        {in_pkey: sql_gen.join_same_not_null}) for v in in_tables_]
398
    
399
    orig_mapping = mapping.copy()
400
    if mapping == {} and not is_function: # need >= one column for INSERT SELECT
401
        mapping = {out_pkey: None} # ColDict will replace with default value
402
    
403
    if not is_literals:
404
        into = sql_gen.as_Table(into_table_name(out_table, in_tables0, mapping,
405
            is_function))
406
        # Ensure into's out_pkey is different from in_pkey by prepending table
407
        if is_function: into_out_pkey = out_pkey
408
        else: into_out_pkey = strings.ustr(out_pkey_col)
409
        
410
        # Set column sources
411
        in_cols = filter(sql_gen.is_table_col, mapping.values())
412
        for col in in_cols:
413
            if col.table == in_tables0: col.set_srcs(sql_gen.src_self)
414
        
415
        log_debug('Joining together input tables into temp table')
416
        # Place in new table so don't modify input and for speed
417
        in_table = sql_gen.Table('in')
418
        mapping = dicts.join(mapping, sql.flatten(db, in_table, input_joins,
419
            in_cols, preserve=[in_pkey_col]))
420
        input_joins = [in_table]
421
        db.log_debug('Temp table: '+strings.as_tt(in_table.to_str(db)), level=2)
422
    
423
    # Wrap mapping in a sql_gen.ColDict.
424
    # sql_gen.ColDict sanitizes both keys and values passed into it.
425
    # Do after applying dicts.join() because that returns a plain dict.
426
    mapping = sql_gen.ColDict(db, out_table, mapping)
427
    
428
    # Save all rows since in_table may have rows deleted
429
    if is_literals: pass
430
    elif is_function: full_in_table = in_table
431
    else:
432
        full_in_table = sql_gen.suffixed_table(in_table, '_full')
433
        sql.copy_table(db, in_table, full_in_table)
434
    
435
    pkeys_table_exists_ref = [False]
436
    def insert_into_pkeys(query, **kw_args):
437
        if pkeys_table_exists_ref[0]:
438
            sql.insert_select(db, into, [in_pkey, into_out_pkey], query,
439
                **kw_args)
440
        else:
441
            sql.run_query_into(db, query, into=into, add_pkey_=True, **kw_args)
442
            pkeys_table_exists_ref[0] = True
443
    
444
    def mk_main_select(joins, cols): return sql.mk_select(db, joins, cols)
445
    
446
    if is_literals: insert_in_table = None
447
    else:
448
        insert_in_table = in_table
449
        insert_in_tables = [insert_in_table]
450
    join_cols = sql_gen.ColDict(db, out_table)
451
    
452
    exc_strs = set()
453
    def log_exc(e):
454
        e_str = exc.str_(e, first_line_only=True)
455
        log_debug('Caught exception: '+e_str)
456
        if e_str in exc_strs: # avoid infinite loops
457
            log_debug('Exception already seen, handler broken')
458
            on_error(e)
459
            remove_all_rows()
460
        else: exc_strs.add(e_str)
461
    
462
    ignore_all_ref = [False]
463
    def remove_all_rows():
464
        log_debug('Ignoring all rows')
465
        ignore_all_ref[0] = True # just return the default value column
466
    
467
    def ensure_cond(cond, e, passed=False, failed=False):
468
        if is_literals: # we know the constraint was applied exactly once
469
            if passed: pass
470
            elif failed: remove_all_rows()
471
            else: raise NotImplementedError()
472
        else:
473
            out_table_cols = sql_gen.ColDict(db, out_table)
474
            out_table_cols.update(util.dict_subset_right_join({},
475
                sql.table_col_names(db, out_table)))
476
            
477
            in_cols = []
478
            cond = sql_gen.map_expr(db, cond, mapping, in_cols)
479
            cond = sql_gen.map_expr(db, cond, out_table_cols)
480
            
481
            not_cond = sql_gen.NotCond(sql_gen.CustomCode(cond))
482
            log_debug('Ignoring rows that do not satisfy '+strings.as_tt(cond))
483
            cur = None
484
            if cond == sql_gen.false_expr:
485
                assert failed
486
                remove_all_rows()
487
            elif cond == sql_gen.true_expr: assert passed
488
            else: cur = sql.delete(db, insert_in_table, not_cond)
489
            
490
            if failed or cur.rowcount > 0: # only if any rows failed cond
491
                track_data_error(db, errors_table_,
492
                    sql_gen.cross_join_srcs(in_cols), None, e.cause.pgcode,
493
                    strings.ensure_newl(e.cause.pgerror)+'condition: '+cond)
494
    
495
    not_null_cols = set()
496
    def ignore(in_col, value, e):
497
        if sql_gen.is_table_col(in_col):
498
            in_col = sql_gen.with_table(in_col, insert_in_table)
499
            
500
            track_data_error(db, errors_table_, in_col.srcs, value,
501
                e.cause.pgcode, e.cause.pgerror)
502
            
503
            sql.add_index(db, in_col, insert_in_table) # enable fast filtering
504
            if value != None and in_col not in not_null_cols:
505
                log_debug('Replacing invalid value '
506
                    +strings.as_tt(strings.urepr(value))+' with NULL in column '
507
                    +strings.as_tt(in_col.to_str(db)))
508
                sql.update(db, insert_in_table, [(in_col, None)],
509
                    sql_gen.ColValueCond(in_col, value))
510
            else:
511
                log_debug('Ignoring rows with '+strings.as_tt(in_col.to_str(db))
512
                    +' = '+strings.as_tt(strings.urepr(value)))
513
                sql.delete(db, insert_in_table,
514
                    sql_gen.ColValueCond(in_col, value))
515
                if value == None: not_null_cols.add(in_col)
516
        else:
517
            assert isinstance(in_col, sql_gen.NamedCol)
518
            in_value = sql_gen.remove_col_rename(in_col)
519
            assert sql_gen.is_literal(in_value)
520
            if value == in_value.value:
521
                if value != None:
522
                    log_debug('Replacing invalid literal '
523
                        +strings.as_tt(in_col.to_str(db))+' with NULL')
524
                    mapping[in_col.name] = None
525
                else:
526
                    remove_all_rows()
527
            # otherwise, all columns were being ignore()d because the specific
528
            # column couldn't be identified, and this was not the invalid column
529
    
530
    if not is_literals:
531
        def insert_pkeys_table(which):
532
            return sql_gen.Table(sql_gen.concat(in_table.name,
533
                '_insert_'+which+'_pkeys'))
534
        insert_out_pkeys = insert_pkeys_table('out')
535
        insert_in_pkeys = insert_pkeys_table('in')
536
    
537
    def mk_func_call():
538
        args = dict(((k.name, v) for k, v in mapping.iteritems()))
539
        return sql_gen.FunctionCall(out_table, **args), args
540
        
541
    if is_function and not is_literals:
542
        log_debug('Defining wrapper function')
543
        
544
        func_call, args = mk_func_call()
545
        func_call = sql_gen.NamedCol(into_out_pkey, func_call)
546
        
547
        # Create empty pkeys table so its row type can be used
548
        insert_into_pkeys(sql.mk_select(db, input_joins,
549
            [in_pkey_col, func_call], limit=0), recover=True)
550
        result_type = db.col_info(sql_gen.Col(into_out_pkey, into)).type
551
        
552
        ## Create error handling wrapper function
553
        
554
        wrapper = db.TempFunction(sql_gen.concat(into.name, '_wrap'))
555
        
556
        select_cols = [in_pkey_col]+args.values()
557
        row_var = copy.copy(sql_gen.row_var)
558
        row_var.set_srcs([in_table])
559
        in_pkey_var = sql_gen.Col(in_pkey, row_var)
560
        
561
        args = dict(((k, sql_gen.with_table(v, row_var))
562
            for k, v in args.iteritems()))
563
        func_call = sql_gen.FunctionCall(out_table, **args)
564
        
565
        def mk_return(result):
566
            return sql_gen.ReturnQuery(sql.mk_select(db,
567
                fields=[in_pkey_var, result], explain=False))
568
        exc_handler = func_wrapper_exception_handler(db,
569
            mk_return(sql_gen.Cast(result_type, None)), args.values(),
570
            errors_table_)
571
        
572
        sql.define_func(db, sql_gen.FunctionDef(wrapper, sql_gen.SetOf(into),
573
            sql_gen.RowExcIgnore(sql_gen.RowType(in_table),
574
                sql.mk_select(db, input_joins), mk_return(func_call),
575
                exc_handler=exc_handler)
576
            ))
577
        wrapper_table = sql_gen.FunctionCall(wrapper)
578
    
579
    missing_msg = None
580
    
581
    # Do inserts and selects
582
    while True:
583
        has_joins = join_cols != {}
584
        
585
        if ignore_all_ref[0]: break # unrecoverable error, so don't do main case
586
        
587
        # Prepare to insert new rows
588
        if is_function:
589
            log_debug('Calling function on input rows')
590
            if is_literals: func_call, args = mk_func_call()
591
        else:
592
            log_debug('Trying to insert new rows')
593
            insert_args = dict(recover=True, cacheable=False)
594
            if has_joins:
595
                insert_args.update(dict(ignore=True))
596
            else:
597
                insert_args.update(dict(returning=out_pkey))
598
                if not is_literals:
599
                    insert_args.update(dict(into=insert_out_pkeys))
600
            main_select = mk_main_select([insert_in_table], [sql_gen.with_table(
601
                c, insert_in_table) for c in mapping.values()])
602
        
603
        try:
604
            cur = None
605
            if is_function:
606
                if is_literals:
607
                    cur = sql.select(db, fields=[func_call], recover=True,
608
                        cacheable=True)
609
                else:
610
                    insert_into_pkeys(sql.mk_select(db, wrapper_table,
611
                        order_by=None), recover=True)
612
            else:
613
                cur = sql.insert_select(db, out_table, mapping.keys(),
614
                    main_select, **insert_args)
615
            break # insert successful
616
        except sql.MissingCastException, e:
617
            log_exc(e)
618
            
619
            type_ = e.type
620
            if e.col == None: out_cols = mapping.keys()
621
            else: out_cols = [e.col]
622
            
623
            for out_col in out_cols:
624
                log_debug('Casting '+strings.as_tt(strings.repr_no_u(out_col))
625
                    +' input to '+strings.as_tt(type_))
626
                in_col = mapping[out_col]
627
                while True:
628
                    try:
629
                        mapping[out_col] = cast_temp_col(db, type_, in_col,
630
                            errors_table_)
631
                        break # cast successful
632
                    except sql.InvalidValueException, e:
633
                        log_exc(e)
634
                        
635
                        ignore(in_col, e.value, e)
636
        except sql.DuplicateKeyException, e:
637
            log_exc(e)
638
            
639
            # Different rows violating different unique constraints not
640
            # supported
641
            assert not join_cols
642
            
643
            if e.cond != None: ensure_cond(e.cond, e, passed=True)
644
            
645
            join_cols.update(util.dict_subset_right_join(mapping, e.cols))
646
            log_debug('Ignoring existing rows, comparing on these columns:\n'
647
                +strings.as_inline_table(join_cols, ustr=col_ustr))
648
            
649
            if is_literals:
650
                return sql.value(sql.select(db, out_table, [out_pkey_col],
651
                    join_cols, order_by=None))
652
            
653
            # Uniquify input table to avoid internal duplicate keys
654
            insert_in_table = sql.distinct_table(db, insert_in_table,
655
                join_cols.values())
656
            insert_in_tables.append(insert_in_table)
657
        except sql.NullValueException, e:
658
            log_exc(e)
659
            
660
            out_col, = e.cols
661
            try: in_col = mapping[out_col]
662
            except KeyError, e:
663
                try: in_col = mapping[out_col] = col_defaults[out_col]
664
                except KeyError:
665
                    missing_msg = 'Missing mapping for NOT NULL column '+out_col
666
                    log_debug(missing_msg)
667
                    remove_all_rows()
668
            else: ignore(in_col, None, e)
669
        except sql.CheckException, e:
670
            log_exc(e)
671
            
672
            ensure_cond(e.cond, e, failed=True)
673
        except sql.InvalidValueException, e:
674
            log_exc(e)
675
            
676
            for in_col in mapping.values(): ignore(in_col, e.value, e)
677
        except psycopg2.extensions.TransactionRollbackError, e:
678
            log_exc(e)
679
            # retry
680
        except sql.DatabaseErrors, e:
681
            log_exc(e)
682
            
683
            log_debug('No handler for exception')
684
            on_error(e)
685
            remove_all_rows()
686
        # after exception handled, rerun loop with additional constraints
687
    
688
    # Resolve default value column
689
    if default != None:
690
        if ignore_all_ref[0]: mapping.update(orig_mapping) # use input cols
691
        try: default = mapping[default]
692
        except KeyError:
693
            db.log_debug('Default value column '
694
                +strings.as_tt(strings.repr_no_u(default))
695
                +' does not exist in mapping, falling back to None', level=2.1)
696
            default = None
697
        else: default = sql_gen.remove_col_rename(default)
698
    
699
    if missing_msg != None and default == None:
700
        warnings.warn(UserWarning(missing_msg))
701
        # not an error because sometimes the mappings include
702
        # extra tables which aren't used by the dataset
703
    
704
    # Handle unrecoverable errors
705
    if ignore_all_ref[0]:
706
        log_debug('Returning default: '+strings.as_tt(strings.urepr(default)))
707
        return default
708
    
709
    if cur != None and row_ct_ref != None and cur.rowcount >= 0:
710
        row_ct_ref[0] += cur.rowcount
711
    
712
    if is_literals: return sql.value(cur)
713
    
714
    if is_function: pass # pkeys table already created
715
    elif has_joins:
716
        select_joins = input_joins+[sql_gen.Join(out_table, join_cols)]
717
        log_debug('Getting output table pkeys of existing/inserted rows')
718
        insert_into_pkeys(sql.mk_select(db, select_joins, [in_pkey_col,
719
            sql_gen.NamedCol(into_out_pkey, out_pkey_col)], order_by=None))
720
    else:
721
        sql.add_row_num(db, insert_out_pkeys) # for joining with input pkeys
722
        
723
        log_debug('Getting input table pkeys of inserted rows')
724
        # Note that mk_main_select() does not use ORDER BY. Instead, assume that
725
        # since the SELECT query is identical to the one used in INSERT SELECT,
726
        # its rows will be retrieved in the same order.
727
        sql.run_query_into(db, mk_main_select(input_joins, [in_pkey]),
728
            into=insert_in_pkeys)
729
        sql.add_row_num(db, insert_in_pkeys) # for joining with output pkeys
730
        
731
        assert sql.table_row_count(db, insert_out_pkeys) == sql.table_row_count(
732
            db, insert_in_pkeys)
733
        
734
        log_debug('Combining output and input pkeys in inserted order')
735
        pkey_joins = [insert_in_pkeys, sql_gen.Join(insert_out_pkeys,
736
            {sql.row_num_col: sql_gen.join_same_not_null})]
737
        in_col = sql_gen.Col(in_pkey, insert_in_pkeys)
738
        out_col = sql_gen.NamedCol(into_out_pkey,
739
            sql_gen.Col(out_pkey, insert_out_pkeys))
740
        insert_into_pkeys(sql.mk_select(db, pkey_joins, [in_col, out_col],
741
            order_by=None))
742
        
743
        sql.empty_temp(db, [insert_out_pkeys, insert_in_pkeys])
744
    
745
    if not is_function: # is_function doesn't leave holes
746
        log_debug('Setting pkeys of missing rows to '
747
            +strings.as_tt(strings.urepr(default)))
748
        if sql_gen.is_table_col(default):
749
            default = sql_gen.with_table(default, full_in_table)
750
        missing_rows_joins = [full_in_table, sql_gen.Join(into,
751
            {in_pkey: sql_gen.join_same_not_null}, sql_gen.filter_out)]
752
            # must use join_same_not_null or query will take forever
753
        insert_into_pkeys(sql.mk_select(db, missing_rows_joins,
754
            [sql_gen.Col(in_pkey, full_in_table),
755
            sql_gen.NamedCol(into_out_pkey, default)], order_by=None))
756
    # otherwise, there is already an entry for every row
757
    
758
    assert (sql.table_row_count(db, into)
759
        == sql.table_row_count(db, full_in_table))
760
    
761
    sql.empty_temp(db, insert_in_tables+[full_in_table])
762
    
763
    srcs = []
764
    if is_function: srcs = sql_gen.cols_srcs(in_cols)
765
    return sql_gen.Col(into_out_pkey, into, srcs)
(30-30/42)