sexy_pg_constraints 0.1.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- data/CHANGELOG.rdoc +12 -0
- data/Manifest +14 -0
- data/README.rdoc +139 -0
- data/Rakefile +15 -0
- data/init.rb +1 -0
- data/lib/constrainer.rb +34 -0
- data/lib/constraints.rb +154 -0
- data/lib/deconstrainer.rb +31 -0
- data/lib/helpers.rb +19 -0
- data/lib/sexy_pg_constraints.rb +24 -0
- data/sexy_pg_constraints.gemspec +31 -0
- data/test/postgresql_adapter.rb +1065 -0
- data/test/sexy_pg_constraints_test.rb +572 -0
- data/test/test_helper.rb +5 -0
- metadata +80 -0
@@ -0,0 +1,1065 @@
|
|
1
|
+
require 'active_record/connection_adapters/abstract_adapter'
|
2
|
+
|
3
|
+
begin
|
4
|
+
require_library_or_gem 'pg'
|
5
|
+
rescue LoadError => e
|
6
|
+
begin
|
7
|
+
require_library_or_gem 'postgres'
|
8
|
+
class PGresult
|
9
|
+
alias_method :nfields, :num_fields unless self.method_defined?(:nfields)
|
10
|
+
alias_method :ntuples, :num_tuples unless self.method_defined?(:ntuples)
|
11
|
+
alias_method :ftype, :type unless self.method_defined?(:ftype)
|
12
|
+
alias_method :cmd_tuples, :cmdtuples unless self.method_defined?(:cmd_tuples)
|
13
|
+
end
|
14
|
+
rescue LoadError
|
15
|
+
raise e
|
16
|
+
end
|
17
|
+
end
|
18
|
+
|
19
|
+
module ActiveRecord
|
20
|
+
class Base
|
21
|
+
# Establishes a connection to the database that's used by all Active Record objects
|
22
|
+
def self.postgresql_connection(config) # :nodoc:
|
23
|
+
config = config.symbolize_keys
|
24
|
+
host = config[:host]
|
25
|
+
port = config[:port] || 5432
|
26
|
+
username = config[:username].to_s if config[:username]
|
27
|
+
password = config[:password].to_s if config[:password]
|
28
|
+
|
29
|
+
if config.has_key?(:database)
|
30
|
+
database = config[:database]
|
31
|
+
else
|
32
|
+
raise ArgumentError, "No database specified. Missing argument: database."
|
33
|
+
end
|
34
|
+
|
35
|
+
# The postgres drivers don't allow the creation of an unconnected PGconn object,
|
36
|
+
# so just pass a nil connection object for the time being.
|
37
|
+
ConnectionAdapters::PostgreSQLAdapter.new(nil, logger, [host, port, nil, nil, database, username, password], config)
|
38
|
+
end
|
39
|
+
end
|
40
|
+
|
41
|
+
module ConnectionAdapters
|
42
|
+
# PostgreSQL-specific extensions to column definitions in a table.
|
43
|
+
class PostgreSQLColumn < Column #:nodoc:
|
44
|
+
# Instantiates a new PostgreSQL column definition in a table.
|
45
|
+
def initialize(name, default, sql_type = nil, null = true)
|
46
|
+
super(name, self.class.extract_value_from_default(default), sql_type, null)
|
47
|
+
end
|
48
|
+
|
49
|
+
private
|
50
|
+
def extract_limit(sql_type)
|
51
|
+
case sql_type
|
52
|
+
when /^bigint/i; 8
|
53
|
+
when /^smallint/i; 2
|
54
|
+
else super
|
55
|
+
end
|
56
|
+
end
|
57
|
+
|
58
|
+
# Extracts the scale from PostgreSQL-specific data types.
|
59
|
+
def extract_scale(sql_type)
|
60
|
+
# Money type has a fixed scale of 2.
|
61
|
+
sql_type =~ /^money/ ? 2 : super
|
62
|
+
end
|
63
|
+
|
64
|
+
# Extracts the precision from PostgreSQL-specific data types.
|
65
|
+
def extract_precision(sql_type)
|
66
|
+
# Actual code is defined dynamically in PostgreSQLAdapter.connect
|
67
|
+
# depending on the server specifics
|
68
|
+
super
|
69
|
+
end
|
70
|
+
|
71
|
+
# Maps PostgreSQL-specific data types to logical Rails types.
|
72
|
+
def simplified_type(field_type)
|
73
|
+
case field_type
|
74
|
+
# Numeric and monetary types
|
75
|
+
when /^(?:real|double precision)$/
|
76
|
+
:float
|
77
|
+
# Monetary types
|
78
|
+
when /^money$/
|
79
|
+
:decimal
|
80
|
+
# Character types
|
81
|
+
when /^(?:character varying|bpchar)(?:\(\d+\))?$/
|
82
|
+
:string
|
83
|
+
# Binary data types
|
84
|
+
when /^bytea$/
|
85
|
+
:binary
|
86
|
+
# Date/time types
|
87
|
+
when /^timestamp with(?:out)? time zone$/
|
88
|
+
:datetime
|
89
|
+
when /^interval$/
|
90
|
+
:string
|
91
|
+
# Geometric types
|
92
|
+
when /^(?:point|line|lseg|box|"?path"?|polygon|circle)$/
|
93
|
+
:string
|
94
|
+
# Network address types
|
95
|
+
when /^(?:cidr|inet|macaddr)$/
|
96
|
+
:string
|
97
|
+
# Bit strings
|
98
|
+
when /^bit(?: varying)?(?:\(\d+\))?$/
|
99
|
+
:string
|
100
|
+
# XML type
|
101
|
+
when /^xml$/
|
102
|
+
:string
|
103
|
+
# Arrays
|
104
|
+
when /^\D+\[\]$/
|
105
|
+
:string
|
106
|
+
# Object identifier types
|
107
|
+
when /^oid$/
|
108
|
+
:integer
|
109
|
+
# Pass through all types that are not specific to PostgreSQL.
|
110
|
+
else
|
111
|
+
super
|
112
|
+
end
|
113
|
+
end
|
114
|
+
|
115
|
+
# Extracts the value from a PostgreSQL column default definition.
|
116
|
+
def self.extract_value_from_default(default)
|
117
|
+
case default
|
118
|
+
# Numeric types
|
119
|
+
when /\A\(?(-?\d+(\.\d*)?\)?)\z/
|
120
|
+
$1
|
121
|
+
# Character types
|
122
|
+
when /\A'(.*)'::(?:character varying|bpchar|text)\z/m
|
123
|
+
$1
|
124
|
+
# Character types (8.1 formatting)
|
125
|
+
when /\AE'(.*)'::(?:character varying|bpchar|text)\z/m
|
126
|
+
$1.gsub(/\\(\d\d\d)/) { $1.oct.chr }
|
127
|
+
# Binary data types
|
128
|
+
when /\A'(.*)'::bytea\z/m
|
129
|
+
$1
|
130
|
+
# Date/time types
|
131
|
+
when /\A'(.+)'::(?:time(?:stamp)? with(?:out)? time zone|date)\z/
|
132
|
+
$1
|
133
|
+
when /\A'(.*)'::interval\z/
|
134
|
+
$1
|
135
|
+
# Boolean type
|
136
|
+
when 'true'
|
137
|
+
true
|
138
|
+
when 'false'
|
139
|
+
false
|
140
|
+
# Geometric types
|
141
|
+
when /\A'(.*)'::(?:point|line|lseg|box|"?path"?|polygon|circle)\z/
|
142
|
+
$1
|
143
|
+
# Network address types
|
144
|
+
when /\A'(.*)'::(?:cidr|inet|macaddr)\z/
|
145
|
+
$1
|
146
|
+
# Bit string types
|
147
|
+
when /\AB'(.*)'::"?bit(?: varying)?"?\z/
|
148
|
+
$1
|
149
|
+
# XML type
|
150
|
+
when /\A'(.*)'::xml\z/m
|
151
|
+
$1
|
152
|
+
# Arrays
|
153
|
+
when /\A'(.*)'::"?\D+"?\[\]\z/
|
154
|
+
$1
|
155
|
+
# Object identifier types
|
156
|
+
when /\A-?\d+\z/
|
157
|
+
$1
|
158
|
+
else
|
159
|
+
# Anything else is blank, some user type, or some function
|
160
|
+
# and we can't know the value of that, so return nil.
|
161
|
+
nil
|
162
|
+
end
|
163
|
+
end
|
164
|
+
end
|
165
|
+
end
|
166
|
+
|
167
|
+
module ConnectionAdapters
|
168
|
+
# The PostgreSQL adapter works both with the native C (http://ruby.scripting.ca/postgres/) and the pure
|
169
|
+
# Ruby (available both as gem and from http://rubyforge.org/frs/?group_id=234&release_id=1944) drivers.
|
170
|
+
#
|
171
|
+
# Options:
|
172
|
+
#
|
173
|
+
# * <tt>:host</tt> - Defaults to "localhost".
|
174
|
+
# * <tt>:port</tt> - Defaults to 5432.
|
175
|
+
# * <tt>:username</tt> - Defaults to nothing.
|
176
|
+
# * <tt>:password</tt> - Defaults to nothing.
|
177
|
+
# * <tt>:database</tt> - The name of the database. No default, must be provided.
|
178
|
+
# * <tt>:schema_search_path</tt> - An optional schema search path for the connection given as a string of comma-separated schema names. This is backward-compatible with the <tt>:schema_order</tt> option.
|
179
|
+
# * <tt>:encoding</tt> - An optional client encoding that is used in a <tt>SET client_encoding TO <encoding></tt> call on the connection.
|
180
|
+
# * <tt>:min_messages</tt> - An optional client min messages that is used in a <tt>SET client_min_messages TO <min_messages></tt> call on the connection.
|
181
|
+
# * <tt>:allow_concurrency</tt> - If true, use async query methods so Ruby threads don't deadlock; otherwise, use blocking query methods.
|
182
|
+
class PostgreSQLAdapter < AbstractAdapter
|
183
|
+
ADAPTER_NAME = 'PostgreSQL'.freeze
|
184
|
+
|
185
|
+
NATIVE_DATABASE_TYPES = {
|
186
|
+
:primary_key => "serial primary key".freeze,
|
187
|
+
:string => { :name => "character varying", :limit => 255 },
|
188
|
+
:text => { :name => "text" },
|
189
|
+
:integer => { :name => "integer" },
|
190
|
+
:float => { :name => "float" },
|
191
|
+
:decimal => { :name => "decimal" },
|
192
|
+
:datetime => { :name => "timestamp" },
|
193
|
+
:timestamp => { :name => "timestamp" },
|
194
|
+
:time => { :name => "time" },
|
195
|
+
:date => { :name => "date" },
|
196
|
+
:binary => { :name => "bytea" },
|
197
|
+
:boolean => { :name => "boolean" }
|
198
|
+
}
|
199
|
+
|
200
|
+
# Returns 'PostgreSQL' as adapter name for identification purposes.
|
201
|
+
def adapter_name
|
202
|
+
ADAPTER_NAME
|
203
|
+
end
|
204
|
+
|
205
|
+
# Initializes and connects a PostgreSQL adapter.
|
206
|
+
def initialize(connection, logger, connection_parameters, config)
|
207
|
+
super(connection, logger)
|
208
|
+
@connection_parameters, @config = connection_parameters, config
|
209
|
+
|
210
|
+
connect
|
211
|
+
end
|
212
|
+
|
213
|
+
# Is this connection alive and ready for queries?
|
214
|
+
def active?
|
215
|
+
if @connection.respond_to?(:status)
|
216
|
+
@connection.status == PGconn::CONNECTION_OK
|
217
|
+
else
|
218
|
+
# We're asking the driver, not ActiveRecord, so use @connection.query instead of #query
|
219
|
+
@connection.query 'SELECT 1'
|
220
|
+
true
|
221
|
+
end
|
222
|
+
# postgres-pr raises a NoMethodError when querying if no connection is available.
|
223
|
+
rescue PGError, NoMethodError
|
224
|
+
false
|
225
|
+
end
|
226
|
+
|
227
|
+
# Close then reopen the connection.
|
228
|
+
def reconnect!
|
229
|
+
if @connection.respond_to?(:reset)
|
230
|
+
@connection.reset
|
231
|
+
configure_connection
|
232
|
+
else
|
233
|
+
disconnect!
|
234
|
+
connect
|
235
|
+
end
|
236
|
+
end
|
237
|
+
|
238
|
+
# Close the connection.
|
239
|
+
def disconnect!
|
240
|
+
@connection.close rescue nil
|
241
|
+
end
|
242
|
+
|
243
|
+
def native_database_types #:nodoc:
|
244
|
+
NATIVE_DATABASE_TYPES
|
245
|
+
end
|
246
|
+
|
247
|
+
# Does PostgreSQL support migrations?
|
248
|
+
def supports_migrations?
|
249
|
+
true
|
250
|
+
end
|
251
|
+
|
252
|
+
# Does PostgreSQL support standard conforming strings?
|
253
|
+
def supports_standard_conforming_strings?
|
254
|
+
# Temporarily set the client message level above error to prevent unintentional
|
255
|
+
# error messages in the logs when working on a PostgreSQL database server that
|
256
|
+
# does not support standard conforming strings.
|
257
|
+
client_min_messages_old = client_min_messages
|
258
|
+
self.client_min_messages = 'panic'
|
259
|
+
|
260
|
+
# postgres-pr does not raise an exception when client_min_messages is set higher
|
261
|
+
# than error and "SHOW standard_conforming_strings" fails, but returns an empty
|
262
|
+
# PGresult instead.
|
263
|
+
has_support = query('SHOW standard_conforming_strings')[0][0] rescue false
|
264
|
+
self.client_min_messages = client_min_messages_old
|
265
|
+
has_support
|
266
|
+
end
|
267
|
+
|
268
|
+
def supports_insert_with_returning?
|
269
|
+
postgresql_version >= 80200
|
270
|
+
end
|
271
|
+
|
272
|
+
def supports_ddl_transactions?
|
273
|
+
true
|
274
|
+
end
|
275
|
+
|
276
|
+
# Returns the configured supported identifier length supported by PostgreSQL,
|
277
|
+
# or report the default of 63 on PostgreSQL 7.x.
|
278
|
+
def table_alias_length
|
279
|
+
@table_alias_length ||= (postgresql_version >= 80000 ? query('SHOW max_identifier_length')[0][0].to_i : 63)
|
280
|
+
end
|
281
|
+
|
282
|
+
# QUOTING ==================================================
|
283
|
+
|
284
|
+
# Escapes binary strings for bytea input to the database.
|
285
|
+
def escape_bytea(value)
|
286
|
+
if PGconn.respond_to?(:escape_bytea)
|
287
|
+
self.class.instance_eval do
|
288
|
+
define_method(:escape_bytea) do |value|
|
289
|
+
PGconn.escape_bytea(value) if value
|
290
|
+
end
|
291
|
+
end
|
292
|
+
else
|
293
|
+
self.class.instance_eval do
|
294
|
+
define_method(:escape_bytea) do |value|
|
295
|
+
if value
|
296
|
+
result = ''
|
297
|
+
value.each_byte { |c| result << sprintf('\\\\%03o', c) }
|
298
|
+
result
|
299
|
+
end
|
300
|
+
end
|
301
|
+
end
|
302
|
+
end
|
303
|
+
escape_bytea(value)
|
304
|
+
end
|
305
|
+
|
306
|
+
# Unescapes bytea output from a database to the binary string it represents.
|
307
|
+
# NOTE: This is NOT an inverse of escape_bytea! This is only to be used
|
308
|
+
# on escaped binary output from database drive.
|
309
|
+
def unescape_bytea(value)
|
310
|
+
# In each case, check if the value actually is escaped PostgreSQL bytea output
|
311
|
+
# or an unescaped Active Record attribute that was just written.
|
312
|
+
if PGconn.respond_to?(:unescape_bytea)
|
313
|
+
self.class.instance_eval do
|
314
|
+
define_method(:unescape_bytea) do |value|
|
315
|
+
if value =~ /\\\d{3}/
|
316
|
+
PGconn.unescape_bytea(value)
|
317
|
+
else
|
318
|
+
value
|
319
|
+
end
|
320
|
+
end
|
321
|
+
end
|
322
|
+
else
|
323
|
+
self.class.instance_eval do
|
324
|
+
define_method(:unescape_bytea) do |value|
|
325
|
+
if value =~ /\\\d{3}/
|
326
|
+
result = ''
|
327
|
+
i, max = 0, value.size
|
328
|
+
while i < max
|
329
|
+
char = value[i]
|
330
|
+
if char == ?\\
|
331
|
+
if value[i+1] == ?\\
|
332
|
+
char = ?\\
|
333
|
+
i += 1
|
334
|
+
else
|
335
|
+
char = value[i+1..i+3].oct
|
336
|
+
i += 3
|
337
|
+
end
|
338
|
+
end
|
339
|
+
result << char
|
340
|
+
i += 1
|
341
|
+
end
|
342
|
+
result
|
343
|
+
else
|
344
|
+
value
|
345
|
+
end
|
346
|
+
end
|
347
|
+
end
|
348
|
+
end
|
349
|
+
unescape_bytea(value)
|
350
|
+
end
|
351
|
+
|
352
|
+
# Quotes PostgreSQL-specific data types for SQL input.
|
353
|
+
def quote(value, column = nil) #:nodoc:
|
354
|
+
if value.kind_of?(String) && column && column.type == :binary
|
355
|
+
"#{quoted_string_prefix}'#{escape_bytea(value)}'"
|
356
|
+
elsif value.kind_of?(String) && column && column.sql_type =~ /^xml$/
|
357
|
+
"xml '#{quote_string(value)}'"
|
358
|
+
elsif value.kind_of?(Numeric) && column && column.sql_type =~ /^money$/
|
359
|
+
# Not truly string input, so doesn't require (or allow) escape string syntax.
|
360
|
+
"'#{value.to_s}'"
|
361
|
+
elsif value.kind_of?(String) && column && column.sql_type =~ /^bit/
|
362
|
+
case value
|
363
|
+
when /^[01]*$/
|
364
|
+
"B'#{value}'" # Bit-string notation
|
365
|
+
when /^[0-9A-F]*$/i
|
366
|
+
"X'#{value}'" # Hexadecimal notation
|
367
|
+
end
|
368
|
+
else
|
369
|
+
super
|
370
|
+
end
|
371
|
+
end
|
372
|
+
|
373
|
+
# Quotes strings for use in SQL input in the postgres driver for better performance.
|
374
|
+
def quote_string(s) #:nodoc:
|
375
|
+
if PGconn.respond_to?(:escape)
|
376
|
+
self.class.instance_eval do
|
377
|
+
define_method(:quote_string) do |s|
|
378
|
+
PGconn.escape(s)
|
379
|
+
end
|
380
|
+
end
|
381
|
+
else
|
382
|
+
# There are some incorrectly compiled postgres drivers out there
|
383
|
+
# that don't define PGconn.escape.
|
384
|
+
self.class.instance_eval do
|
385
|
+
remove_method(:quote_string)
|
386
|
+
end
|
387
|
+
end
|
388
|
+
quote_string(s)
|
389
|
+
end
|
390
|
+
|
391
|
+
# Quotes column names for use in SQL queries.
|
392
|
+
def quote_column_name(name) #:nodoc:
|
393
|
+
%("#{name}")
|
394
|
+
end
|
395
|
+
|
396
|
+
# Quote date/time values for use in SQL input. Includes microseconds
|
397
|
+
# if the value is a Time responding to usec.
|
398
|
+
def quoted_date(value) #:nodoc:
|
399
|
+
if value.acts_like?(:time) && value.respond_to?(:usec)
|
400
|
+
"#{super}.#{sprintf("%06d", value.usec)}"
|
401
|
+
else
|
402
|
+
super
|
403
|
+
end
|
404
|
+
end
|
405
|
+
|
406
|
+
# REFERENTIAL INTEGRITY ====================================
|
407
|
+
|
408
|
+
def supports_disable_referential_integrity?() #:nodoc:
|
409
|
+
version = query("SHOW server_version")[0][0].split('.')
|
410
|
+
(version[0].to_i >= 8 && version[1].to_i >= 1) ? true : false
|
411
|
+
rescue
|
412
|
+
return false
|
413
|
+
end
|
414
|
+
|
415
|
+
def disable_referential_integrity(&block) #:nodoc:
|
416
|
+
if supports_disable_referential_integrity?() then
|
417
|
+
execute(tables.collect { |name| "ALTER TABLE #{quote_table_name(name)} DISABLE TRIGGER ALL" }.join(";"))
|
418
|
+
end
|
419
|
+
yield
|
420
|
+
ensure
|
421
|
+
if supports_disable_referential_integrity?() then
|
422
|
+
execute(tables.collect { |name| "ALTER TABLE #{quote_table_name(name)} ENABLE TRIGGER ALL" }.join(";"))
|
423
|
+
end
|
424
|
+
end
|
425
|
+
|
426
|
+
# DATABASE STATEMENTS ======================================
|
427
|
+
|
428
|
+
# Executes a SELECT query and returns an array of rows. Each row is an
|
429
|
+
# array of field values.
|
430
|
+
def select_rows(sql, name = nil)
|
431
|
+
select_raw(sql, name).last
|
432
|
+
end
|
433
|
+
|
434
|
+
# Executes an INSERT query and returns the new record's ID
|
435
|
+
def insert(sql, name = nil, pk = nil, id_value = nil, sequence_name = nil)
|
436
|
+
# Extract the table from the insert sql. Yuck.
|
437
|
+
table = sql.split(" ", 4)[2].gsub('"', '')
|
438
|
+
|
439
|
+
# Try an insert with 'returning id' if available (PG >= 8.2)
|
440
|
+
if supports_insert_with_returning?
|
441
|
+
pk, sequence_name = *pk_and_sequence_for(table) unless pk
|
442
|
+
if pk
|
443
|
+
id = select_value("#{sql} RETURNING #{quote_column_name(pk)}")
|
444
|
+
clear_query_cache
|
445
|
+
return id
|
446
|
+
end
|
447
|
+
end
|
448
|
+
|
449
|
+
# Otherwise, insert then grab last_insert_id.
|
450
|
+
if insert_id = super
|
451
|
+
insert_id
|
452
|
+
else
|
453
|
+
# If neither pk nor sequence name is given, look them up.
|
454
|
+
unless pk || sequence_name
|
455
|
+
pk, sequence_name = *pk_and_sequence_for(table)
|
456
|
+
end
|
457
|
+
|
458
|
+
# If a pk is given, fallback to default sequence name.
|
459
|
+
# Don't fetch last insert id for a table without a pk.
|
460
|
+
if pk && sequence_name ||= default_sequence_name(table, pk)
|
461
|
+
last_insert_id(table, sequence_name)
|
462
|
+
end
|
463
|
+
end
|
464
|
+
end
|
465
|
+
|
466
|
+
# create a 2D array representing the result set
|
467
|
+
def result_as_array(res) #:nodoc:
|
468
|
+
# check if we have any binary column and if they need escaping
|
469
|
+
unescape_col = []
|
470
|
+
for j in 0...res.nfields do
|
471
|
+
# unescape string passed BYTEA field (OID == 17)
|
472
|
+
unescape_col << ( res.ftype(j)==17 )
|
473
|
+
end
|
474
|
+
|
475
|
+
ary = []
|
476
|
+
for i in 0...res.ntuples do
|
477
|
+
ary << []
|
478
|
+
for j in 0...res.nfields do
|
479
|
+
data = res.getvalue(i,j)
|
480
|
+
data = unescape_bytea(data) if unescape_col[j] and data.is_a?(String)
|
481
|
+
ary[i] << data
|
482
|
+
end
|
483
|
+
end
|
484
|
+
return ary
|
485
|
+
end
|
486
|
+
|
487
|
+
|
488
|
+
# Queries the database and returns the results in an Array-like object
|
489
|
+
def query(sql, name = nil) #:nodoc:
|
490
|
+
log(sql, name) do
|
491
|
+
if @async
|
492
|
+
res = @connection.async_exec(sql)
|
493
|
+
else
|
494
|
+
res = @connection.exec(sql)
|
495
|
+
end
|
496
|
+
return result_as_array(res)
|
497
|
+
end
|
498
|
+
end
|
499
|
+
|
500
|
+
# Executes an SQL statement, returning a PGresult object on success
|
501
|
+
# or raising a PGError exception otherwise.
|
502
|
+
def execute(sql, name = nil)
|
503
|
+
log(sql, name) do
|
504
|
+
if @async
|
505
|
+
@connection.async_exec(sql)
|
506
|
+
else
|
507
|
+
@connection.exec(sql)
|
508
|
+
end
|
509
|
+
end
|
510
|
+
end
|
511
|
+
|
512
|
+
# Executes an UPDATE query and returns the number of affected tuples.
|
513
|
+
def update_sql(sql, name = nil)
|
514
|
+
super.cmd_tuples
|
515
|
+
end
|
516
|
+
|
517
|
+
# Begins a transaction.
|
518
|
+
def begin_db_transaction
|
519
|
+
execute "BEGIN"
|
520
|
+
end
|
521
|
+
|
522
|
+
# Commits a transaction.
|
523
|
+
def commit_db_transaction
|
524
|
+
execute "COMMIT"
|
525
|
+
end
|
526
|
+
|
527
|
+
# Aborts a transaction.
|
528
|
+
def rollback_db_transaction
|
529
|
+
execute "ROLLBACK"
|
530
|
+
end
|
531
|
+
|
532
|
+
# ruby-pg defines Ruby constants for transaction status,
|
533
|
+
# ruby-postgres does not.
|
534
|
+
PQTRANS_IDLE = defined?(PGconn::PQTRANS_IDLE) ? PGconn::PQTRANS_IDLE : 0
|
535
|
+
|
536
|
+
# Check whether a transaction is active.
|
537
|
+
def transaction_active?
|
538
|
+
@connection.transaction_status != PQTRANS_IDLE
|
539
|
+
end
|
540
|
+
|
541
|
+
# Wrap a block in a transaction. Returns result of block.
|
542
|
+
def transaction(start_db_transaction = true)
|
543
|
+
transaction_open = false
|
544
|
+
begin
|
545
|
+
if block_given?
|
546
|
+
if start_db_transaction
|
547
|
+
begin_db_transaction
|
548
|
+
transaction_open = true
|
549
|
+
end
|
550
|
+
yield
|
551
|
+
end
|
552
|
+
rescue Exception => database_transaction_rollback
|
553
|
+
if transaction_open && transaction_active?
|
554
|
+
transaction_open = false
|
555
|
+
rollback_db_transaction
|
556
|
+
end
|
557
|
+
raise unless database_transaction_rollback.is_a? ActiveRecord::Rollback
|
558
|
+
end
|
559
|
+
ensure
|
560
|
+
if transaction_open && transaction_active?
|
561
|
+
begin
|
562
|
+
commit_db_transaction
|
563
|
+
rescue Exception => database_transaction_rollback
|
564
|
+
rollback_db_transaction
|
565
|
+
raise
|
566
|
+
end
|
567
|
+
end
|
568
|
+
end
|
569
|
+
|
570
|
+
|
571
|
+
# SCHEMA STATEMENTS ========================================
|
572
|
+
|
573
|
+
def recreate_database(name) #:nodoc:
|
574
|
+
drop_database(name)
|
575
|
+
create_database(name)
|
576
|
+
end
|
577
|
+
|
578
|
+
# Create a new PostgreSQL database. Options include <tt>:owner</tt>, <tt>:template</tt>,
|
579
|
+
# <tt>:encoding</tt>, <tt>:tablespace</tt>, and <tt>:connection_limit</tt> (note that MySQL uses
|
580
|
+
# <tt>:charset</tt> while PostgreSQL uses <tt>:encoding</tt>).
|
581
|
+
#
|
582
|
+
# Example:
|
583
|
+
# create_database config[:database], config
|
584
|
+
# create_database 'foo_development', :encoding => 'unicode'
|
585
|
+
def create_database(name, options = {})
|
586
|
+
options = options.reverse_merge(:encoding => "utf8")
|
587
|
+
|
588
|
+
option_string = options.symbolize_keys.sum do |key, value|
|
589
|
+
case key
|
590
|
+
when :owner
|
591
|
+
" OWNER = \"#{value}\""
|
592
|
+
when :template
|
593
|
+
" TEMPLATE = \"#{value}\""
|
594
|
+
when :encoding
|
595
|
+
" ENCODING = '#{value}'"
|
596
|
+
when :tablespace
|
597
|
+
" TABLESPACE = \"#{value}\""
|
598
|
+
when :connection_limit
|
599
|
+
" CONNECTION LIMIT = #{value}"
|
600
|
+
else
|
601
|
+
""
|
602
|
+
end
|
603
|
+
end
|
604
|
+
|
605
|
+
execute "CREATE DATABASE #{quote_table_name(name)}#{option_string}"
|
606
|
+
end
|
607
|
+
|
608
|
+
# Drops a PostgreSQL database
|
609
|
+
#
|
610
|
+
# Example:
|
611
|
+
# drop_database 'matt_development'
|
612
|
+
def drop_database(name) #:nodoc:
|
613
|
+
if postgresql_version >= 80200
|
614
|
+
execute "DROP DATABASE IF EXISTS #{quote_table_name(name)}"
|
615
|
+
else
|
616
|
+
begin
|
617
|
+
execute "DROP DATABASE #{quote_table_name(name)}"
|
618
|
+
rescue ActiveRecord::StatementInvalid
|
619
|
+
@logger.warn "#{name} database doesn't exist." if @logger
|
620
|
+
end
|
621
|
+
end
|
622
|
+
end
|
623
|
+
|
624
|
+
|
625
|
+
# Returns the list of all tables in the schema search path or a specified schema.
|
626
|
+
def tables(name = nil)
|
627
|
+
schemas = schema_search_path.split(/,/).map { |p| quote(p) }.join(',')
|
628
|
+
query(<<-SQL, name).map { |row| row[0] }
|
629
|
+
SELECT tablename
|
630
|
+
FROM pg_tables
|
631
|
+
WHERE schemaname IN (#{schemas})
|
632
|
+
SQL
|
633
|
+
end
|
634
|
+
|
635
|
+
# Returns the list of all indexes for a table.
|
636
|
+
def indexes(table_name, name = nil)
|
637
|
+
schemas = schema_search_path.split(/,/).map { |p| quote(p) }.join(',')
|
638
|
+
result = query(<<-SQL, name)
|
639
|
+
SELECT distinct i.relname, d.indisunique, a.attname
|
640
|
+
FROM pg_class t, pg_class i, pg_index d, pg_attribute a
|
641
|
+
WHERE i.relkind = 'i'
|
642
|
+
AND d.indexrelid = i.oid
|
643
|
+
AND d.indisprimary = 'f'
|
644
|
+
AND t.oid = d.indrelid
|
645
|
+
AND t.relname = '#{table_name}'
|
646
|
+
AND i.relnamespace IN (SELECT oid FROM pg_namespace WHERE nspname IN (#{schemas}) )
|
647
|
+
AND a.attrelid = t.oid
|
648
|
+
AND ( d.indkey[0]=a.attnum OR d.indkey[1]=a.attnum
|
649
|
+
OR d.indkey[2]=a.attnum OR d.indkey[3]=a.attnum
|
650
|
+
OR d.indkey[4]=a.attnum OR d.indkey[5]=a.attnum
|
651
|
+
OR d.indkey[6]=a.attnum OR d.indkey[7]=a.attnum
|
652
|
+
OR d.indkey[8]=a.attnum OR d.indkey[9]=a.attnum )
|
653
|
+
ORDER BY i.relname
|
654
|
+
SQL
|
655
|
+
|
656
|
+
current_index = nil
|
657
|
+
indexes = []
|
658
|
+
|
659
|
+
result.each do |row|
|
660
|
+
if current_index != row[0]
|
661
|
+
indexes << IndexDefinition.new(table_name, row[0], row[1] == "t", [])
|
662
|
+
current_index = row[0]
|
663
|
+
end
|
664
|
+
|
665
|
+
indexes.last.columns << row[2]
|
666
|
+
end
|
667
|
+
|
668
|
+
indexes
|
669
|
+
end
|
670
|
+
|
671
|
+
# Returns the list of all column definitions for a table.
|
672
|
+
def columns(table_name, name = nil)
|
673
|
+
# Limit, precision, and scale are all handled by the superclass.
|
674
|
+
column_definitions(table_name).collect do |name, type, default, notnull|
|
675
|
+
PostgreSQLColumn.new(name, default, type, notnull == 'f')
|
676
|
+
end
|
677
|
+
end
|
678
|
+
|
679
|
+
# Returns the current database name.
|
680
|
+
def current_database
|
681
|
+
query('select current_database()')[0][0]
|
682
|
+
end
|
683
|
+
|
684
|
+
# Returns the current database encoding format.
|
685
|
+
def encoding
|
686
|
+
query(<<-end_sql)[0][0]
|
687
|
+
SELECT pg_encoding_to_char(pg_database.encoding) FROM pg_database
|
688
|
+
WHERE pg_database.datname LIKE '#{current_database}'
|
689
|
+
end_sql
|
690
|
+
end
|
691
|
+
|
692
|
+
# Sets the schema search path to a string of comma-separated schema names.
|
693
|
+
# Names beginning with $ have to be quoted (e.g. $user => '$user').
|
694
|
+
# See: http://www.postgresql.org/docs/current/static/ddl-schemas.html
|
695
|
+
#
|
696
|
+
# This should be not be called manually but set in database.yml.
|
697
|
+
def schema_search_path=(schema_csv)
|
698
|
+
if schema_csv
|
699
|
+
execute "SET search_path TO #{schema_csv}"
|
700
|
+
@schema_search_path = schema_csv
|
701
|
+
end
|
702
|
+
end
|
703
|
+
|
704
|
+
# Returns the active schema search path.
|
705
|
+
def schema_search_path
|
706
|
+
@schema_search_path ||= query('SHOW search_path')[0][0]
|
707
|
+
end
|
708
|
+
|
709
|
+
# Returns the current client message level.
|
710
|
+
def client_min_messages
|
711
|
+
query('SHOW client_min_messages')[0][0]
|
712
|
+
end
|
713
|
+
|
714
|
+
# Set the client message level.
|
715
|
+
def client_min_messages=(level)
|
716
|
+
execute("SET client_min_messages TO '#{level}'")
|
717
|
+
end
|
718
|
+
|
719
|
+
# Returns the sequence name for a table's primary key or some other specified key.
|
720
|
+
def default_sequence_name(table_name, pk = nil) #:nodoc:
|
721
|
+
default_pk, default_seq = pk_and_sequence_for(table_name)
|
722
|
+
default_seq || "#{table_name}_#{pk || default_pk || 'id'}_seq"
|
723
|
+
end
|
724
|
+
|
725
|
+
# Resets the sequence of a table's primary key to the maximum value.
|
726
|
+
def reset_pk_sequence!(table, pk = nil, sequence = nil) #:nodoc:
|
727
|
+
unless pk and sequence
|
728
|
+
default_pk, default_sequence = pk_and_sequence_for(table)
|
729
|
+
pk ||= default_pk
|
730
|
+
sequence ||= default_sequence
|
731
|
+
end
|
732
|
+
if pk
|
733
|
+
if sequence
|
734
|
+
quoted_sequence = quote_column_name(sequence)
|
735
|
+
|
736
|
+
select_value <<-end_sql, 'Reset sequence'
|
737
|
+
SELECT setval('#{quoted_sequence}', (SELECT COALESCE(MAX(#{quote_column_name pk})+(SELECT increment_by FROM #{quoted_sequence}), (SELECT min_value FROM #{quoted_sequence})) FROM #{quote_table_name(table)}), false)
|
738
|
+
end_sql
|
739
|
+
else
|
740
|
+
@logger.warn "#{table} has primary key #{pk} with no default sequence" if @logger
|
741
|
+
end
|
742
|
+
end
|
743
|
+
end
|
744
|
+
|
745
|
+
# Returns a table's primary key and belonging sequence.
|
746
|
+
def pk_and_sequence_for(table) #:nodoc:
|
747
|
+
# First try looking for a sequence with a dependency on the
|
748
|
+
# given table's primary key.
|
749
|
+
result = query(<<-end_sql, 'PK and serial sequence')[0]
|
750
|
+
SELECT attr.attname, seq.relname
|
751
|
+
FROM pg_class seq,
|
752
|
+
pg_attribute attr,
|
753
|
+
pg_depend dep,
|
754
|
+
pg_namespace name,
|
755
|
+
pg_constraint cons
|
756
|
+
WHERE seq.oid = dep.objid
|
757
|
+
AND seq.relkind = 'S'
|
758
|
+
AND attr.attrelid = dep.refobjid
|
759
|
+
AND attr.attnum = dep.refobjsubid
|
760
|
+
AND attr.attrelid = cons.conrelid
|
761
|
+
AND attr.attnum = cons.conkey[1]
|
762
|
+
AND cons.contype = 'p'
|
763
|
+
AND dep.refobjid = '#{table}'::regclass
|
764
|
+
end_sql
|
765
|
+
|
766
|
+
if result.nil? or result.empty?
|
767
|
+
# If that fails, try parsing the primary key's default value.
|
768
|
+
# Support the 7.x and 8.0 nextval('foo'::text) as well as
|
769
|
+
# the 8.1+ nextval('foo'::regclass).
|
770
|
+
result = query(<<-end_sql, 'PK and custom sequence')[0]
|
771
|
+
SELECT attr.attname,
|
772
|
+
CASE
|
773
|
+
WHEN split_part(def.adsrc, '''', 2) ~ '.' THEN
|
774
|
+
substr(split_part(def.adsrc, '''', 2),
|
775
|
+
strpos(split_part(def.adsrc, '''', 2), '.')+1)
|
776
|
+
ELSE split_part(def.adsrc, '''', 2)
|
777
|
+
END
|
778
|
+
FROM pg_class t
|
779
|
+
JOIN pg_attribute attr ON (t.oid = attrelid)
|
780
|
+
JOIN pg_attrdef def ON (adrelid = attrelid AND adnum = attnum)
|
781
|
+
JOIN pg_constraint cons ON (conrelid = adrelid AND adnum = conkey[1])
|
782
|
+
WHERE t.oid = '#{table}'::regclass
|
783
|
+
AND cons.contype = 'p'
|
784
|
+
AND def.adsrc ~* 'nextval'
|
785
|
+
end_sql
|
786
|
+
end
|
787
|
+
|
788
|
+
# [primary_key, sequence]
|
789
|
+
[result.first, result.last]
|
790
|
+
rescue
|
791
|
+
nil
|
792
|
+
end
|
793
|
+
|
794
|
+
# Renames a table.
|
795
|
+
def rename_table(name, new_name)
|
796
|
+
execute "ALTER TABLE #{quote_table_name(name)} RENAME TO #{quote_table_name(new_name)}"
|
797
|
+
end
|
798
|
+
|
799
|
+
# Adds a new column to the named table.
|
800
|
+
# See TableDefinition#column for details of the options you can use.
|
801
|
+
def add_column(table_name, column_name, type, options = {})
|
802
|
+
default = options[:default]
|
803
|
+
notnull = options[:null] == false
|
804
|
+
|
805
|
+
# Add the column.
|
806
|
+
execute("ALTER TABLE #{quote_table_name(table_name)} ADD COLUMN #{quote_column_name(column_name)} #{type_to_sql(type, options[:limit], options[:precision], options[:scale])}")
|
807
|
+
|
808
|
+
change_column_default(table_name, column_name, default) if options_include_default?(options)
|
809
|
+
change_column_null(table_name, column_name, false, default) if notnull
|
810
|
+
end
|
811
|
+
|
812
|
+
# Changes the column of a table.
|
813
|
+
def change_column(table_name, column_name, type, options = {})
|
814
|
+
quoted_table_name = quote_table_name(table_name)
|
815
|
+
|
816
|
+
begin
|
817
|
+
execute "ALTER TABLE #{quoted_table_name} ALTER COLUMN #{quote_column_name(column_name)} TYPE #{type_to_sql(type, options[:limit], options[:precision], options[:scale])}"
|
818
|
+
rescue ActiveRecord::StatementInvalid => e
|
819
|
+
raise e if postgresql_version > 80000
|
820
|
+
# This is PostgreSQL 7.x, so we have to use a more arcane way of doing it.
|
821
|
+
begin
|
822
|
+
begin_db_transaction
|
823
|
+
tmp_column_name = "#{column_name}_ar_tmp"
|
824
|
+
add_column(table_name, tmp_column_name, type, options)
|
825
|
+
execute "UPDATE #{quoted_table_name} SET #{quote_column_name(tmp_column_name)} = CAST(#{quote_column_name(column_name)} AS #{type_to_sql(type, options[:limit], options[:precision], options[:scale])})"
|
826
|
+
remove_column(table_name, column_name)
|
827
|
+
rename_column(table_name, tmp_column_name, column_name)
|
828
|
+
commit_db_transaction
|
829
|
+
rescue
|
830
|
+
rollback_db_transaction
|
831
|
+
end
|
832
|
+
end
|
833
|
+
|
834
|
+
change_column_default(table_name, column_name, options[:default]) if options_include_default?(options)
|
835
|
+
change_column_null(table_name, column_name, options[:null], options[:default]) if options.key?(:null)
|
836
|
+
end
|
837
|
+
|
838
|
+
# Changes the default value of a table column.
|
839
|
+
def change_column_default(table_name, column_name, default)
|
840
|
+
execute "ALTER TABLE #{quote_table_name(table_name)} ALTER COLUMN #{quote_column_name(column_name)} SET DEFAULT #{quote(default)}"
|
841
|
+
end
|
842
|
+
|
843
|
+
def change_column_null(table_name, column_name, null, default = nil)
|
844
|
+
unless null || default.nil?
|
845
|
+
execute("UPDATE #{quote_table_name(table_name)} SET #{quote_column_name(column_name)}=#{quote(default)} WHERE #{quote_column_name(column_name)} IS NULL")
|
846
|
+
end
|
847
|
+
execute("ALTER TABLE #{quote_table_name(table_name)} ALTER #{quote_column_name(column_name)} #{null ? 'DROP' : 'SET'} NOT NULL")
|
848
|
+
end
|
849
|
+
|
850
|
+
# Renames a column in a table.
|
851
|
+
def rename_column(table_name, column_name, new_column_name)
|
852
|
+
execute "ALTER TABLE #{quote_table_name(table_name)} RENAME COLUMN #{quote_column_name(column_name)} TO #{quote_column_name(new_column_name)}"
|
853
|
+
end
|
854
|
+
|
855
|
+
# Drops an index from a table.
|
856
|
+
def remove_index(table_name, options = {})
|
857
|
+
execute "DROP INDEX #{index_name(table_name, options)}"
|
858
|
+
end
|
859
|
+
|
860
|
+
# Maps logical Rails types to PostgreSQL-specific data types.
|
861
|
+
def type_to_sql(type, limit = nil, precision = nil, scale = nil)
|
862
|
+
return super unless type.to_s == 'integer'
|
863
|
+
|
864
|
+
case limit
|
865
|
+
when 1..2; 'smallint'
|
866
|
+
when 3..4, nil; 'integer'
|
867
|
+
when 5..8; 'bigint'
|
868
|
+
else raise(ActiveRecordError, "No integer type has byte size #{limit}. Use a numeric with precision 0 instead.")
|
869
|
+
end
|
870
|
+
end
|
871
|
+
|
872
|
+
# Returns a SELECT DISTINCT clause for a given set of columns and a given ORDER BY clause.
|
873
|
+
#
|
874
|
+
# PostgreSQL requires the ORDER BY columns in the select list for distinct queries, and
|
875
|
+
# requires that the ORDER BY include the distinct column.
|
876
|
+
#
|
877
|
+
# distinct("posts.id", "posts.created_at desc")
|
878
|
+
def distinct(columns, order_by) #:nodoc:
|
879
|
+
return "DISTINCT #{columns}" if order_by.blank?
|
880
|
+
|
881
|
+
# Construct a clean list of column names from the ORDER BY clause, removing
|
882
|
+
# any ASC/DESC modifiers
|
883
|
+
order_columns = order_by.split(',').collect { |s| s.split.first }
|
884
|
+
order_columns.delete_if &:blank?
|
885
|
+
order_columns = order_columns.zip((0...order_columns.size).to_a).map { |s,i| "#{s} AS alias_#{i}" }
|
886
|
+
|
887
|
+
# Return a DISTINCT ON() clause that's distinct on the columns we want but includes
|
888
|
+
# all the required columns for the ORDER BY to work properly.
|
889
|
+
sql = "DISTINCT ON (#{columns}) #{columns}, "
|
890
|
+
sql << order_columns * ', '
|
891
|
+
end
|
892
|
+
|
893
|
+
# Returns an ORDER BY clause for the passed order option.
|
894
|
+
#
|
895
|
+
# PostgreSQL does not allow arbitrary ordering when using DISTINCT ON, so we work around this
|
896
|
+
# by wrapping the +sql+ string as a sub-select and ordering in that query.
|
897
|
+
def add_order_by_for_association_limiting!(sql, options) #:nodoc:
|
898
|
+
return sql if options[:order].blank?
|
899
|
+
|
900
|
+
order = options[:order].split(',').collect { |s| s.strip }.reject(&:blank?)
|
901
|
+
order.map! { |s| 'DESC' if s =~ /\bdesc$/i }
|
902
|
+
order = order.zip((0...order.size).to_a).map { |s,i| "id_list.alias_#{i} #{s}" }.join(', ')
|
903
|
+
|
904
|
+
sql.replace "SELECT * FROM (#{sql}) AS id_list ORDER BY #{order}"
|
905
|
+
end
|
906
|
+
|
907
|
+
protected
|
908
|
+
# Returns the version of the connected PostgreSQL version.
|
909
|
+
def postgresql_version
|
910
|
+
@postgresql_version ||=
|
911
|
+
if @connection.respond_to?(:server_version)
|
912
|
+
@connection.server_version
|
913
|
+
else
|
914
|
+
# Mimic PGconn.server_version behavior
|
915
|
+
begin
|
916
|
+
query('SELECT version()')[0][0] =~ /PostgreSQL (\d+)\.(\d+)\.(\d+)/
|
917
|
+
($1.to_i * 10000) + ($2.to_i * 100) + $3.to_i
|
918
|
+
rescue
|
919
|
+
0
|
920
|
+
end
|
921
|
+
end
|
922
|
+
end
|
923
|
+
|
924
|
+
private
|
925
|
+
# The internal PostgreSQL identifier of the money data type.
|
926
|
+
MONEY_COLUMN_TYPE_OID = 790 #:nodoc:
|
927
|
+
|
928
|
+
# Connects to a PostgreSQL server and sets up the adapter depending on the
|
929
|
+
# connected server's characteristics.
|
930
|
+
def connect
|
931
|
+
@connection = PGconn.connect(*@connection_parameters)
|
932
|
+
PGconn.translate_results = false if PGconn.respond_to?(:translate_results=)
|
933
|
+
|
934
|
+
# Ignore async_exec and async_query when using postgres-pr.
|
935
|
+
@async = @config[:allow_concurrency] && @connection.respond_to?(:async_exec)
|
936
|
+
|
937
|
+
# Use escape string syntax if available. We cannot do this lazily when encountering
|
938
|
+
# the first string, because that could then break any transactions in progress.
|
939
|
+
# See: http://www.postgresql.org/docs/current/static/runtime-config-compatible.html
|
940
|
+
# If PostgreSQL doesn't know the standard_conforming_strings parameter then it doesn't
|
941
|
+
# support escape string syntax. Don't override the inherited quoted_string_prefix.
|
942
|
+
if supports_standard_conforming_strings?
|
943
|
+
self.class.instance_eval do
|
944
|
+
define_method(:quoted_string_prefix) { 'E' }
|
945
|
+
end
|
946
|
+
end
|
947
|
+
|
948
|
+
# Money type has a fixed precision of 10 in PostgreSQL 8.2 and below, and as of
|
949
|
+
# PostgreSQL 8.3 it has a fixed precision of 19. PostgreSQLColumn.extract_precision
|
950
|
+
# should know about this but can't detect it there, so deal with it here.
|
951
|
+
money_precision = (postgresql_version >= 80300) ? 19 : 10
|
952
|
+
PostgreSQLColumn.module_eval(<<-end_eval)
|
953
|
+
def extract_precision(sql_type)
|
954
|
+
if sql_type =~ /^money$/
|
955
|
+
#{money_precision}
|
956
|
+
else
|
957
|
+
super
|
958
|
+
end
|
959
|
+
end
|
960
|
+
end_eval
|
961
|
+
|
962
|
+
configure_connection
|
963
|
+
end
|
964
|
+
|
965
|
+
# Configures the encoding, verbosity, and schema search path of the connection.
|
966
|
+
# This is called by #connect and should not be called manually.
|
967
|
+
def configure_connection
|
968
|
+
if @config[:encoding]
|
969
|
+
if @connection.respond_to?(:set_client_encoding)
|
970
|
+
@connection.set_client_encoding(@config[:encoding])
|
971
|
+
else
|
972
|
+
execute("SET client_encoding TO '#{@config[:encoding]}'")
|
973
|
+
end
|
974
|
+
end
|
975
|
+
self.client_min_messages = @config[:min_messages] if @config[:min_messages]
|
976
|
+
self.schema_search_path = @config[:schema_search_path] || @config[:schema_order]
|
977
|
+
end
|
978
|
+
|
979
|
+
# Returns the current ID of a table's sequence.
|
980
|
+
def last_insert_id(table, sequence_name) #:nodoc:
|
981
|
+
Integer(select_value("SELECT currval('#{sequence_name}')"))
|
982
|
+
end
|
983
|
+
|
984
|
+
# Executes a SELECT query and returns the results, performing any data type
|
985
|
+
# conversions that are required to be performed here instead of in PostgreSQLColumn.
|
986
|
+
def select(sql, name = nil)
|
987
|
+
fields, rows = select_raw(sql, name)
|
988
|
+
result = []
|
989
|
+
for row in rows
|
990
|
+
row_hash = {}
|
991
|
+
fields.each_with_index do |f, i|
|
992
|
+
row_hash[f] = row[i]
|
993
|
+
end
|
994
|
+
result << row_hash
|
995
|
+
end
|
996
|
+
result
|
997
|
+
end
|
998
|
+
|
999
|
+
def select_raw(sql, name = nil)
|
1000
|
+
res = execute(sql, name)
|
1001
|
+
results = result_as_array(res)
|
1002
|
+
fields = []
|
1003
|
+
rows = []
|
1004
|
+
if res.ntuples > 0
|
1005
|
+
fields = res.fields
|
1006
|
+
results.each do |row|
|
1007
|
+
hashed_row = {}
|
1008
|
+
row.each_index do |cell_index|
|
1009
|
+
# If this is a money type column and there are any currency symbols,
|
1010
|
+
# then strip them off. Indeed it would be prettier to do this in
|
1011
|
+
# PostgreSQLColumn.string_to_decimal but would break form input
|
1012
|
+
# fields that call value_before_type_cast.
|
1013
|
+
if res.ftype(cell_index) == MONEY_COLUMN_TYPE_OID
|
1014
|
+
# Because money output is formatted according to the locale, there are two
|
1015
|
+
# cases to consider (note the decimal separators):
|
1016
|
+
# (1) $12,345,678.12
|
1017
|
+
# (2) $12.345.678,12
|
1018
|
+
case column = row[cell_index]
|
1019
|
+
when /^-?\D+[\d,]+\.\d{2}$/ # (1)
|
1020
|
+
row[cell_index] = column.gsub(/[^-\d\.]/, '')
|
1021
|
+
when /^-?\D+[\d\.]+,\d{2}$/ # (2)
|
1022
|
+
row[cell_index] = column.gsub(/[^-\d,]/, '').sub(/,/, '.')
|
1023
|
+
end
|
1024
|
+
end
|
1025
|
+
|
1026
|
+
hashed_row[fields[cell_index]] = column
|
1027
|
+
end
|
1028
|
+
rows << row
|
1029
|
+
end
|
1030
|
+
end
|
1031
|
+
res.clear
|
1032
|
+
return fields, rows
|
1033
|
+
end
|
1034
|
+
|
1035
|
+
# Returns the list of a table's column names, data types, and default values.
|
1036
|
+
#
|
1037
|
+
# The underlying query is roughly:
|
1038
|
+
# SELECT column.name, column.type, default.value
|
1039
|
+
# FROM column LEFT JOIN default
|
1040
|
+
# ON column.table_id = default.table_id
|
1041
|
+
# AND column.num = default.column_num
|
1042
|
+
# WHERE column.table_id = get_table_id('table_name')
|
1043
|
+
# AND column.num > 0
|
1044
|
+
# AND NOT column.is_dropped
|
1045
|
+
# ORDER BY column.num
|
1046
|
+
#
|
1047
|
+
# If the table name is not prefixed with a schema, the database will
|
1048
|
+
# take the first match from the schema search path.
|
1049
|
+
#
|
1050
|
+
# Query implementation notes:
|
1051
|
+
# - format_type includes the column size constraint, e.g. varchar(50)
|
1052
|
+
# - ::regclass is a function that gives the id for a table name
|
1053
|
+
def column_definitions(table_name) #:nodoc:
|
1054
|
+
query <<-end_sql
|
1055
|
+
SELECT a.attname, format_type(a.atttypid, a.atttypmod), d.adsrc, a.attnotnull
|
1056
|
+
FROM pg_attribute a LEFT JOIN pg_attrdef d
|
1057
|
+
ON a.attrelid = d.adrelid AND a.attnum = d.adnum
|
1058
|
+
WHERE a.attrelid = '#{table_name}'::regclass
|
1059
|
+
AND a.attnum > 0 AND NOT a.attisdropped
|
1060
|
+
ORDER BY a.attnum
|
1061
|
+
end_sql
|
1062
|
+
end
|
1063
|
+
end
|
1064
|
+
end
|
1065
|
+
end
|