This file is indexed.

/usr/lib/python2.7/dist-packages/pyfits/column.py is in python-pyfits 1:3.4-4.

This file is owned by root:root, with mode 0o644.

The actual contents of the file can be viewed below.

   1
   2
   3
   4
   5
   6
   7
   8
   9
  10
  11
  12
  13
  14
  15
  16
  17
  18
  19
  20
  21
  22
  23
  24
  25
  26
  27
  28
  29
  30
  31
  32
  33
  34
  35
  36
  37
  38
  39
  40
  41
  42
  43
  44
  45
  46
  47
  48
  49
  50
  51
  52
  53
  54
  55
  56
  57
  58
  59
  60
  61
  62
  63
  64
  65
  66
  67
  68
  69
  70
  71
  72
  73
  74
  75
  76
  77
  78
  79
  80
  81
  82
  83
  84
  85
  86
  87
  88
  89
  90
  91
  92
  93
  94
  95
  96
  97
  98
  99
 100
 101
 102
 103
 104
 105
 106
 107
 108
 109
 110
 111
 112
 113
 114
 115
 116
 117
 118
 119
 120
 121
 122
 123
 124
 125
 126
 127
 128
 129
 130
 131
 132
 133
 134
 135
 136
 137
 138
 139
 140
 141
 142
 143
 144
 145
 146
 147
 148
 149
 150
 151
 152
 153
 154
 155
 156
 157
 158
 159
 160
 161
 162
 163
 164
 165
 166
 167
 168
 169
 170
 171
 172
 173
 174
 175
 176
 177
 178
 179
 180
 181
 182
 183
 184
 185
 186
 187
 188
 189
 190
 191
 192
 193
 194
 195
 196
 197
 198
 199
 200
 201
 202
 203
 204
 205
 206
 207
 208
 209
 210
 211
 212
 213
 214
 215
 216
 217
 218
 219
 220
 221
 222
 223
 224
 225
 226
 227
 228
 229
 230
 231
 232
 233
 234
 235
 236
 237
 238
 239
 240
 241
 242
 243
 244
 245
 246
 247
 248
 249
 250
 251
 252
 253
 254
 255
 256
 257
 258
 259
 260
 261
 262
 263
 264
 265
 266
 267
 268
 269
 270
 271
 272
 273
 274
 275
 276
 277
 278
 279
 280
 281
 282
 283
 284
 285
 286
 287
 288
 289
 290
 291
 292
 293
 294
 295
 296
 297
 298
 299
 300
 301
 302
 303
 304
 305
 306
 307
 308
 309
 310
 311
 312
 313
 314
 315
 316
 317
 318
 319
 320
 321
 322
 323
 324
 325
 326
 327
 328
 329
 330
 331
 332
 333
 334
 335
 336
 337
 338
 339
 340
 341
 342
 343
 344
 345
 346
 347
 348
 349
 350
 351
 352
 353
 354
 355
 356
 357
 358
 359
 360
 361
 362
 363
 364
 365
 366
 367
 368
 369
 370
 371
 372
 373
 374
 375
 376
 377
 378
 379
 380
 381
 382
 383
 384
 385
 386
 387
 388
 389
 390
 391
 392
 393
 394
 395
 396
 397
 398
 399
 400
 401
 402
 403
 404
 405
 406
 407
 408
 409
 410
 411
 412
 413
 414
 415
 416
 417
 418
 419
 420
 421
 422
 423
 424
 425
 426
 427
 428
 429
 430
 431
 432
 433
 434
 435
 436
 437
 438
 439
 440
 441
 442
 443
 444
 445
 446
 447
 448
 449
 450
 451
 452
 453
 454
 455
 456
 457
 458
 459
 460
 461
 462
 463
 464
 465
 466
 467
 468
 469
 470
 471
 472
 473
 474
 475
 476
 477
 478
 479
 480
 481
 482
 483
 484
 485
 486
 487
 488
 489
 490
 491
 492
 493
 494
 495
 496
 497
 498
 499
 500
 501
 502
 503
 504
 505
 506
 507
 508
 509
 510
 511
 512
 513
 514
 515
 516
 517
 518
 519
 520
 521
 522
 523
 524
 525
 526
 527
 528
 529
 530
 531
 532
 533
 534
 535
 536
 537
 538
 539
 540
 541
 542
 543
 544
 545
 546
 547
 548
 549
 550
 551
 552
 553
 554
 555
 556
 557
 558
 559
 560
 561
 562
 563
 564
 565
 566
 567
 568
 569
 570
 571
 572
 573
 574
 575
 576
 577
 578
 579
 580
 581
 582
 583
 584
 585
 586
 587
 588
 589
 590
 591
 592
 593
 594
 595
 596
 597
 598
 599
 600
 601
 602
 603
 604
 605
 606
 607
 608
 609
 610
 611
 612
 613
 614
 615
 616
 617
 618
 619
 620
 621
 622
 623
 624
 625
 626
 627
 628
 629
 630
 631
 632
 633
 634
 635
 636
 637
 638
 639
 640
 641
 642
 643
 644
 645
 646
 647
 648
 649
 650
 651
 652
 653
 654
 655
 656
 657
 658
 659
 660
 661
 662
 663
 664
 665
 666
 667
 668
 669
 670
 671
 672
 673
 674
 675
 676
 677
 678
 679
 680
 681
 682
 683
 684
 685
 686
 687
 688
 689
 690
 691
 692
 693
 694
 695
 696
 697
 698
 699
 700
 701
 702
 703
 704
 705
 706
 707
 708
 709
 710
 711
 712
 713
 714
 715
 716
 717
 718
 719
 720
 721
 722
 723
 724
 725
 726
 727
 728
 729
 730
 731
 732
 733
 734
 735
 736
 737
 738
 739
 740
 741
 742
 743
 744
 745
 746
 747
 748
 749
 750
 751
 752
 753
 754
 755
 756
 757
 758
 759
 760
 761
 762
 763
 764
 765
 766
 767
 768
 769
 770
 771
 772
 773
 774
 775
 776
 777
 778
 779
 780
 781
 782
 783
 784
 785
 786
 787
 788
 789
 790
 791
 792
 793
 794
 795
 796
 797
 798
 799
 800
 801
 802
 803
 804
 805
 806
 807
 808
 809
 810
 811
 812
 813
 814
 815
 816
 817
 818
 819
 820
 821
 822
 823
 824
 825
 826
 827
 828
 829
 830
 831
 832
 833
 834
 835
 836
 837
 838
 839
 840
 841
 842
 843
 844
 845
 846
 847
 848
 849
 850
 851
 852
 853
 854
 855
 856
 857
 858
 859
 860
 861
 862
 863
 864
 865
 866
 867
 868
 869
 870
 871
 872
 873
 874
 875
 876
 877
 878
 879
 880
 881
 882
 883
 884
 885
 886
 887
 888
 889
 890
 891
 892
 893
 894
 895
 896
 897
 898
 899
 900
 901
 902
 903
 904
 905
 906
 907
 908
 909
 910
 911
 912
 913
 914
 915
 916
 917
 918
 919
 920
 921
 922
 923
 924
 925
 926
 927
 928
 929
 930
 931
 932
 933
 934
 935
 936
 937
 938
 939
 940
 941
 942
 943
 944
 945
 946
 947
 948
 949
 950
 951
 952
 953
 954
 955
 956
 957
 958
 959
 960
 961
 962
 963
 964
 965
 966
 967
 968
 969
 970
 971
 972
 973
 974
 975
 976
 977
 978
 979
 980
 981
 982
 983
 984
 985
 986
 987
 988
 989
 990
 991
 992
 993
 994
 995
 996
 997
 998
 999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
import copy
import operator
import re
import sys
import warnings
import weakref

import numpy as np
from numpy import char as chararray

from .extern.six import iteritems, string_types
from .extern.six.moves import reduce
from . import _numpy_hacks as nh

from .card import Card, CARD_LENGTH
from .py3compat import ignored, OrderedDict
from .util import (lazyproperty, pairwise, _is_int, _convert_array,
                   encode_ascii, indent, isiterable, cmp, NotifierMixin)
from .verify import VerifyError, VerifyWarning


__all__ = ['Column', 'ColDefs', 'Delayed']


# mapping from TFORM data type to numpy data type (code)
# L: Logical (Boolean)
# B: Unsigned Byte
# I: 16-bit Integer
# J: 32-bit Integer
# K: 64-bit Integer
# E: Single-precision Floating Point
# D: Double-precision Floating Point
# C: Single-precision Complex
# M: Double-precision Complex
# A: Character
FITS2NUMPY = {'L': 'i1', 'B': 'u1', 'I': 'i2', 'J': 'i4', 'K': 'i8', 'E': 'f4',
              'D': 'f8', 'C': 'c8', 'M': 'c16', 'A': 'a'}

# the inverse dictionary of the above
NUMPY2FITS = dict([(val, key) for key, val in iteritems(FITS2NUMPY)])
# Normally booleans are represented as ints in pyfits, but if passed in a numpy
# boolean array, that should be supported
NUMPY2FITS['b1'] = 'L'
# Add unsigned types, which will be stored as signed ints with a TZERO card.
NUMPY2FITS['u2'] = 'I'
NUMPY2FITS['u4'] = 'J'
NUMPY2FITS['u8'] = 'K'

# This is the order in which values are converted to FITS types
# Note that only double precision floating point/complex are supported
FORMATORDER = ['L', 'B', 'I', 'J', 'K', 'D', 'M', 'A']

# mapping from ASCII table TFORM data type to numpy data type
# A: Character
# I: Integer (32-bit)
# J: Integer (64-bit; non-standard)
# F: Float (32-bit; fixed decimal notation)
# E: Float (32-bit; exponential notation)
# D: Float (64-bit; exponential notation, always 64-bit by convention)
ASCII2NUMPY = {'A': 'a', 'I': 'i4', 'J': 'i8', 'F': 'f4', 'E': 'f4',
               'D': 'f8'}

# Maps FITS ASCII column format codes to the appropriate Python string
# formatting codes for that type.
ASCII2STR = {'A': 's', 'I': 'd', 'J': 'd', 'F': 'f', 'E': 'E', 'D': 'E'}

# For each ASCII table format code, provides a default width (and decimal
# precision) for when one isn't given explicitly in the column format
ASCII_DEFAULT_WIDTHS= {'A': (1, 0), 'I': (10, 0), 'J': (15, 0),
                       'E': (15, 7), 'F': (16, 7), 'D': (25, 17)}




# lists of column/field definition common names and keyword names, make
# sure to preserve the one-to-one correspondence when updating the list(s).
# Use lists, instead of dictionaries so the names can be displayed in a
# preferred order.
KEYWORD_NAMES = ['TTYPE', 'TFORM', 'TUNIT', 'TNULL', 'TSCAL', 'TZERO',
                 'TDISP', 'TBCOL', 'TDIM']
KEYWORD_ATTRIBUTES = ['name', 'format', 'unit', 'null', 'bscale', 'bzero',
                      'disp', 'start', 'dim']
"""This is a list of the attributes that can be set on `Column` objects."""


KEYWORD_TO_ATTRIBUTE = \
    OrderedDict((keyword, attr)
                for keyword, attr in zip(KEYWORD_NAMES, KEYWORD_ATTRIBUTES))


ATTRIBUTE_TO_KEYWORD = \
    OrderedDict((value, key)
                for key, value in KEYWORD_TO_ATTRIBUTE.items())


# TODO: Define a list of default comments to associate with each table keyword

# TFORMn regular expression
TFORMAT_RE = re.compile(r'(?P<repeat>^[0-9]*)(?P<format>[LXBIJKAEDCMPQ])'
                        r'(?P<option>[!-~]*)', re.I)

# TFORMn for ASCII tables; two different versions depending on whether
# the format is floating-point or not; allows empty values for width
# in which case defaults are used
TFORMAT_ASCII_RE = re.compile(r'(?:(?P<format>[AIJ])(?P<width>[0-9]+)?)|'
                              r'(?:(?P<formatf>[FED])'
                              r'(?:(?P<widthf>[0-9]+)\.'
                              r'(?P<precision>[0-9]+))?)')

TTYPE_RE = re.compile(r'[0-9a-zA-Z_]+')
"""
Regular expression for valid table column names.  See FITS Standard v3.0 section
7.2.2.
"""

# table definition keyword regular expression
TDEF_RE = re.compile(r'(?P<label>^T[A-Z]*)(?P<num>[1-9][0-9 ]*$)')

# table dimension keyword regular expression (fairly flexible with whitespace)
TDIM_RE = re.compile(r'\(\s*(?P<dims>(?:\d+,\s*)+\s*\d+)\s*\)\s*')

# value for ASCII table cell with value = TNULL
# this can be reset by user.
ASCIITNULL = 0

# The default placeholder to use for NULL values in ASCII tables when
# converting from binary to ASCII tables
DEFAULT_ASCII_TNULL = '---'


class Delayed(object):
    """Delayed file-reading data."""

    def __init__(self, hdu=None, field=None):
        self.hdu = weakref.proxy(hdu)
        self.field = field

    def __getitem__(self, key):
        # This forces the data for the HDU to be read, which will replace
        # the corresponding Delayed objects in the Tables Columns to be
        # transformed into ndarrays.  It will also return the value of the
        # requested data element.
        return self.hdu.data[key][self.field]


class _BaseColumnFormat(str):
    """
    Base class for binary table column formats (just called _ColumnFormat)
    and ASCII table column formats (_AsciiColumnFormat).
    """

    def __eq__(self, other):
        if not other:
            return False

        if isinstance(other, str):
            if not isinstance(other, self.__class__):
                try:
                    other = self.__class__(other)
                except ValueError:
                    return False
        else:
            return False

        return self.canonical == other.canonical

    def __hash__(self):
        return hash(self.canonical)

    @lazyproperty
    def dtype(self):
        """
        The Numpy dtype object created from the format's associated recformat.
        """

        return np.dtype(self.recformat)

    @classmethod
    def from_column_format(cls, format):
        """Creates a column format object from another column format object
        regardless of their type.

        That is, this can convert a _ColumnFormat to an _AsciiColumnFormat
        or vice versa at least in cases where a direct translation is possible.
        """

        return cls.from_recformat(format.recformat)


class _ColumnFormat(_BaseColumnFormat):
    """
    Represents a FITS binary table column format.

    This is an enhancement over using a normal string for the format, since the
    repeat count, format code, and option are available as separate attributes,
    and smart comparison is used.  For example 1J == J.
    """

    def __new__(cls, format):
        self = super(_ColumnFormat, cls).__new__(cls, format)
        self.repeat, self.format, self.option = _parse_tformat(format)
        self.format = self.format.upper()
        if self.format in ('P', 'Q'):
            # TODO: There should be a generic factory that returns either
            # _FormatP or _FormatQ as appropriate for a given TFORMn
            if self.format == 'P':
                recformat = _FormatP.from_tform(format)
            else:
                recformat = _FormatQ.from_tform(format)
            # Format of variable length arrays
            self.p_format = recformat.format
        else:
            self.p_format = None
        return self

    @classmethod
    def from_recformat(cls, recformat):
        """Creates a column format from a Numpy record dtype format."""

        return cls(_convert_format(recformat, reverse=True))

    @lazyproperty
    def recformat(self):
        """Returns the equivalent Numpy record format string."""

        return _convert_format(self)

    @lazyproperty
    def canonical(self):
        """
        Returns a 'canonical' string representation of this format.

        This is in the proper form of rTa where T is the single character data
        type code, a is the optional part, and r is the repeat.  If repeat == 1
        (the default) it is left out of this representation.
        """

        if self.repeat == 1:
            repeat = ''
        else:
            repeat = str(self.repeat)

        return '%s%s%s' % (repeat, self.format, self.option)


class _AsciiColumnFormat(_BaseColumnFormat):
    """Similar to _ColumnFormat but specifically for columns in ASCII tables.

    The formats of ASCII table columns and binary table columns are inherently
    incompatible in FITS.  They don't support the same ranges and types of
    values, and even reuse format codes in subtly different ways.  For example
    the format code 'Iw' in ASCII columns refers to any integer whose string
    representation is at most w characters wide, so 'I' can represent
    effectively any integer that will fit in a FITS columns.  Whereas for
    binary tables 'I' very explicitly refers to a 16-bit signed integer.

    Conversions between the two column formats can be performed using the
    ``to/from_binary`` methods on this class, or the ``to/from_ascii``
    methods on the `_ColumnFormat` class.  But again, not all conversions are
    possible and may result in a `~.exceptions.ValueError`.
    """

    def __new__(cls, format, strict=False):
        self = super(_AsciiColumnFormat, cls).__new__(cls, format)
        self.format, self.width, self.precision = \
            _parse_ascii_tformat(format, strict)

        # This is to support handling logical (boolean) data from binary tables
        # in an ASCII table
        self._pseudo_logical = False
        return self

    @classmethod
    def from_column_format(cls, format):
        inst = cls.from_recformat(format.recformat)
        # Hack
        if format.format == 'L':
            inst._pseudo_logical = True
        return inst

    @classmethod
    def from_recformat(cls, recformat):
        """Creates a column format from a Numpy record dtype format."""

        return cls(_convert_ascii_format(recformat, reverse=True))

    @lazyproperty
    def recformat(self):
        """Returns the equivalent Numpy record format string."""

        return _convert_ascii_format(self)

    @lazyproperty
    def canonical(self):
        """
        Returns a 'canonical' string representation of this format.

        This is in the proper form of Tw.d where T is the single character data
        type code, w is the width in characters for this field, and d is the
        number of digits after the decimal place (for format codes 'E', 'F',
        and 'D' only).
        """

        if self.format in ('E', 'F', 'D'):
            return '%s%s.%s' % (self.format, self.width, self.precision)

        return '%s%s' % (self.format, self.width)


class _FormatX(str):
    """For X format in binary tables."""

    def __new__(cls, repeat=1):
        nbytes = ((repeat - 1) // 8) + 1
        # use an array, even if it is only ONE u1 (i.e. use tuple always)
        obj = super(_FormatX, cls).__new__(cls, repr((nbytes,)) + 'u1')
        obj.repeat = repeat
        return obj

    def __getnewargs__(self):
        return (self.repeat,)

    @property
    def tform(self):
        return '%sX' % self.repeat


# TODO: Table column formats need to be verified upon first reading the file;
# as it is, an invalid P format will raise a VerifyError from some deep,
# unexpected place
class _FormatP(str):
    """For P format in variable length table."""

    # As far as I can tell from my reading of the FITS standard, a type code is
    # *required* for P and Q formats; there is no default
    _format_re_template = (r'(?P<repeat>\d+)?%s(?P<dtype>[LXBIJKAEDCM])'
                            '(?:\((?P<max>\d*)\))?')
    _format_code = 'P'
    _format_re = re.compile(_format_re_template % _format_code)
    _descriptor_format = '2i4'

    def __new__(cls, dtype, repeat=None, max=None):
        obj = super(_FormatP, cls).__new__(cls, cls._descriptor_format)
        obj.format = NUMPY2FITS[dtype]
        obj.dtype = dtype
        obj.repeat = repeat
        obj.max = max
        return obj

    def __getnewargs__(self):
        return (self.dtype, self.repeat, self.max)

    @classmethod
    def from_tform(cls, format):
        m = cls._format_re.match(format)
        if not m or m.group('dtype') not in FITS2NUMPY:
            raise VerifyError('Invalid column format: %s' % format)
        repeat = m.group('repeat')
        array_dtype = m.group('dtype')
        max = m.group('max')
        if not max:
            max = None
        return cls(FITS2NUMPY[array_dtype], repeat=repeat, max=max)

    @property
    def tform(self):
        repeat = '' if self.repeat is None else self.repeat
        max = '' if self.max is None else self.max
        return '%s%s%s(%s)' % (repeat, self._format_code, self.format, max)


class _FormatQ(_FormatP):
    """Carries type description of the Q format for variable length arrays.

    The Q format is like the P format but uses 64-bit integers in the array
    descriptors, allowing for heaps stored beyond 2GB into a file.
    """

    _format_code = 'Q'
    _format_re = re.compile(_FormatP._format_re_template % _format_code)
    _descriptor_format = '2i8'


class ColumnAttribute(object):
    """
    Descriptor for attributes of `Column` that are associated with keywords
    in the FITS header and describe properties of the column as specified in
    the FITS standard.

    Each `ColumnAttribute` may have a ``validator`` method defined on it.
    This validates values set on this attribute to ensure that they meet the
    FITS standard.  Invalid values will raise a warning and will not be used in
    formatting the column.  The validator should take two arguments--the
    `Column` it is being assigned to, and the new value for the attribute, and
    it must raise an `AssertionError` if the value is invalid.

    The `ColumnAttribute` itself is a decorator that can be used to define the
    ``validator`` for each column attribute.  For example::

        @ColumnAttribute('TTYPE')
        def name(col, name):
            assert isinstance(name, str)

    The actual object returned by this decorator is the `ColumnAttribute`
    instance though, not the ``name`` function.  As such ``name`` is not a
    method of the class it is defined in.

    The setter for `ColumnAttribute` also updates the header of any table
    HDU this column is attached to in order to reflect the change.  The
    ``validator`` should ensure that the value is valid for inclusion in a FITS
    header.
    """

    def __init__(self, keyword):
        self._keyword = keyword
        self._validator = None

        # The name of the attribute associated with this keyword is currently
        # determined from the KEYWORD_NAMES/ATTRIBUTES lists.  This could be
        # make more flexible in the future, for example, to support custom
        # column attributes.
        self._attr = KEYWORD_TO_ATTRIBUTE[self._keyword]

    def __get__(self, obj, objtype=None):
        if obj is None:
            return self
        else:
            return getattr(obj, '_' + self._attr)

    def __set__(self, obj, value):
        if self._validator is not None:
            self._validator(obj, value)

        old_value = getattr(obj, '_' + self._attr, None)
        setattr(obj, '_' + self._attr, value)
        obj._notify('column_attribute_changed', obj, self._attr, old_value,
                    value)

    def __call__(self, func):
        """
        Set the validator for this column attribute.

        Returns ``self`` so that this can be used as a decorator, as described
        in the docs for this class.
        """

        self._validator = func

        return self

    def __repr__(self):
        return "{0}('{1}')".format(self.__class__.__name__, self._keyword)


class Column(NotifierMixin):
    """
    Class which contains the definition of one column, e.g.  ``ttype``,
    ``tform``, etc. and the array containing values for the column.
    """

    def __init__(self, name=None, format=None, unit=None, null=None,
                 bscale=None, bzero=None, disp=None, start=None, dim=None,
                 array=None, ascii=None):
        """
        Construct a `Column` by specifying attributes.  All attributes
        except ``format`` can be optional; see :ref:`column_creation` and
        :ref:`creating_ascii_table` for more information regarding
        ``TFORM`` keyword.

        Parameters
        ----------
        name : str, optional
            column name, corresponding to ``TTYPE`` keyword

        format : str
            column format, corresponding to ``TFORM`` keyword

        unit : str, optional
            column unit, corresponding to ``TUNIT`` keyword

        null : str, optional
            null value, corresponding to ``TNULL`` keyword

        bscale : int-like, optional
            bscale value, corresponding to ``TSCAL`` keyword

        bzero : int-like, optional
            bzero value, corresponding to ``TZERO`` keyword

        disp : str, optional
            display format, corresponding to ``TDISP`` keyword

        start : int, optional
            column starting position (ASCII table only), corresponding
            to ``TBCOL`` keyword

        dim : str, optional
            column dimension corresponding to ``TDIM`` keyword

        array : iterable, optional
            a `list`, `numpy.ndarray` (or other iterable that can be used to
            initialize an ndarray) providing initial data for this column.
            The array will be automatically converted, if possible, to the data
            format of the column.  In the case were non-trivial ``bscale``
            and/or ``bzero`` arguments are given, the values in the array must
            be the *physical* values--that is, the values of column as if the
            scaling has already been applied (the array stored on the column
            object will then be converted back to its storage values).

        ascii : bool, optional
            set `True` if this describes a column for an ASCII table; this
            may be required to disambiguate the column format
        """

        if format is None:
            raise ValueError('Must specify format to construct Column.')

        # any of the input argument (except array) can be a Card or just
        # a number/string
        kwargs = {'ascii': ascii}
        for attr in KEYWORD_ATTRIBUTES:
            value = locals()[attr]  # get the argument's value

            if isinstance(value, Card):
                value = value.value

            kwargs[attr] = value

        valid_kwargs, invalid_kwargs = self._verify_keywords(**kwargs)

        if invalid_kwargs:
            msg = ['The following keyword arguments to Column were invalid:']

            for val in invalid_kwargs.values():
                msg.append(indent(val[1]))

            raise VerifyError('\n'.join(msg))

        for attr in KEYWORD_ATTRIBUTES:
            setattr(self, attr, valid_kwargs.get(attr))

        # TODO: For PyFITS 3.3 try to eliminate the following two special cases
        # for recformat and dim:
        # This is not actually stored as an attribute on columns for some
        # reason
        recformat = valid_kwargs['recformat']

        # The 'dim' keyword's original value is stored in self.dim, while
        # *only* the tuple form is stored in self._dims.
        self._dims = self.dim
        self.dim = dim

        # Awful hack to use for now to keep track of whether the column holds
        # pseudo-unsigned int data
        self._pseudo_unsigned_ints = False

        # if the column data is not ndarray, make it to be one, i.e.
        # input arrays can be just list or tuple, not required to be ndarray
        # does not include Object array because there is no guarantee
        # the elements in the object array are consistent.
        if not isinstance(array,
                          (np.ndarray, chararray.chararray, Delayed)):
            try:  # try to convert to a ndarray first
                if array is not None:
                    array = np.array(array)
            except:
                try:  # then try to convert it to a strings array
                    itemsize = int(recformat[1:])
                    array = chararray.array(array, itemsize=itemsize)
                except ValueError:
                    # then try variable length array
                    # Note: This includes _FormatQ by inheritance
                    if isinstance(recformat, _FormatP):
                        array = _VLF(array, dtype=recformat.dtype)
                    else:
                        raise ValueError('Data is inconsistent with the '
                                         'format `%s`.' % format)

        array = self._convert_to_valid_data_type(array)

        # We have required (through documentation) that arrays passed in to
        # this constructor are already in their physical values, so we make
        # note of that here
        if isinstance(array, np.ndarray):
            self._physical_values = True
        else:
            self._physical_values = False

        self._parent_fits_rec = None
        self.array = array

    def __repr__(self):
        text = ''
        for attr in KEYWORD_ATTRIBUTES:
            value = getattr(self, attr)
            if value is not None:
                text += attr + ' = ' + repr(value) + '; '
        return text[:-2]

    def __eq__(self, other):
        """
        Two columns are equal if their name and format are the same.  Other
        attributes aren't taken into account at this time.
        """

        # According to the FITS standard column names must be case-insensitive
        a = (self.name.lower(), self.format)
        b = (other.name.lower(), other.format)
        return a == b

    def __hash__(self):
        """
        Like __eq__, the hash of a column should be based on the unique column
        name and format, and be case-insensitive with respect to the column
        name.
        """

        return hash((self.name.lower(), self.format))

    @property
    def array(self):
        """
        The Numpy `~numpy.ndarray` associated with this `Column`.

        If the column was instantiated with an array passed to the ``array``
        argument, this will return that array.  However, if the column is
        later added to a table, such as via `BinTableHDU.from_columns` as
        is typically the case, this attribute will be updated to reference
        the associated field in the table, which may no longer be the same
        array.
        """

        # Ideally the .array attribute never would have existed in the first
        # place, or would have been internal-only.  This is a legacy of the
        # older design from PyFITS that needs to have continued support, for
        # now.

        # One of the main problems with this design was that it created a
        # reference cycle.  When the .array attribute was updated after
        # creating a FITS_rec from the column (as explained in the docstring) a
        # reference cycle was created.  This is because the code in BinTableHDU
        # (and a few other places) does essentially the following:
        #
        # data._coldefs = columns  # The ColDefs object holding this Column
        # for col in columns:
        #     col.array = data.field(col.name)
        #
        # This way each columns .array attribute now points to the field in the
        # table data.  It's actually a pretty confusing interface (since it
        # replaces the array originally pointed to by .array), but it's the way
        # things have been for a long, long time.
        #
        # However, this results, in *many* cases, in a reference cycle.
        # Because the array returned by data.field(col.name), while sometimes
        # an array that owns its own data, is usually like a slice of the
        # original data.  It has the original FITS_rec as the array .base.
        # This results in the following reference cycle (for the n-th column):
        #
        #    data -> data._coldefs -> data._coldefs[n] ->
        #     data._coldefs[n].array -> data._coldefs[n].array.base -> data
        #
        # Because ndarray objects do not handled by Python's garbage collector
        # the reference cycle cannot be broken.  Therefore the FITS_rec's
        # refcount never goes to zero, its __del__ is never called, and its
        # memory is never freed.  This didn't occur in *all* cases, but it did
        # occur in many cases.
        #
        # To get around this, Column.array is no longer a simple attribute
        # like it was previously.  Now each Column has a ._parent_fits_rec
        # attribute which is a weakref to a FITS_rec object.  Code that
        # previously assigned each col.array to field in a FITS_rec (as in
        # the example a few paragraphs above) is still used, however now
        # array.setter checks if a reference cycle will be created.  And if
        # so, instead of saving directly to the Column's __dict__, it creates
        # the ._prent_fits_rec weakref, and all lookups of the column's .array
        # go through that instead.
        #
        # This alone does not fully solve the problem.  Because
        # _parent_fits_rec is a weakref, if the user ever holds a reference to
        # the Column, but deletes all references to the underlying FITS_rec,
        # the .array attribute would suddenly start returning None instead of
        # the array data.  This problem is resolved on FITS_rec's end.  See the
        # note in the FITS_rec._coldefs property for the rest of the story.

        # If the Columns's array is not a reference to an existing FITS_rec,
        # then it is just stored in self.__dict__; otherwise check the
        # _parent_fits_rec reference if it 's still available.
        if 'array' in self.__dict__:
            return self.__dict__['array']
        elif self._parent_fits_rec is not None:
            parent = self._parent_fits_rec()
            if parent is not None:
                return parent[self.name]
        else:
            return None

    @array.setter
    def array(self, array):
        # The following looks over the bases of the given array to check if it
        # has a ._coldefs attribute (i.e. is a FITS_rec) and that that _coldefs
        # contains this Column itself, and would create a reference cycle if we
        # stored the array directly in self.__dict__.
        # In this case it instead sets up the _parent_fits_rec weakref to the
        # underlying FITS_rec, so that array.getter can return arrays through
        # self._parent_fits_rec().field(self.name), rather than storing a
        # hard reference to the field like it used to.
        base = array
        while True:
            if (hasattr(base, '_coldefs') and
                    isinstance(base._coldefs, ColDefs)):
                for col in base._coldefs:
                    if col is self and self._parent_fits_rec is None:
                        self._parent_fits_rec = weakref.ref(base)

                        # Just in case the user already set .array to their own
                        # array.
                        if 'array' in self.__dict__:
                            del self.__dict__['array']
                        return

            if getattr(base, 'base', None) is not None:
                base = base.base
            else:
                break

        self.__dict__['array'] = array

    @array.deleter
    def array(self):
        try:
            del self.__dict__['array']
        except KeyError:
            pass

        self._parent_fits_rec = None

    @ColumnAttribute('TTYPE')
    def name(col, name):
        if name is None:
            # Allow None to indicate deleting the name, or to just indicate an
            # unspecified name (when creating a new Column).
            return

        # Check that the name meets the recommended standard--other column
        # names are *allowed*, but will be discouraged
        if isinstance(name, string_types) and not TTYPE_RE.match(name):
            warnings.warn(
                'It is strongly recommended that column names contain only '
                'upper and lower-case ASCII letters, digits, or underscores '
                'for maximum compatibility with other software '
                '(got {0!r}).'.format(name), VerifyWarning)

        # This ensures that the new name can fit into a single FITS card
        # without any special extension like CONTINUE cards or the like.
        assert (isinstance(name, string_types) and
                len(str(Card('TTYPE', name))) == CARD_LENGTH), \
                    ('Column name must be a string able to fit in a single '
                     'FITS card--typically this means a maximum of 68 '
                     'characters, though it may be fewer if the string '
                     'contains special characters like quotes.')

    format = ColumnAttribute('TFORM')
    unit = ColumnAttribute('TUNIT')
    null = ColumnAttribute('TNULL')
    bscale = ColumnAttribute('TSCAL')
    bzero = ColumnAttribute('TZERO')
    disp = ColumnAttribute('TDISP')
    start = ColumnAttribute('TBCOL')
    dim = ColumnAttribute('TDIM')

    @lazyproperty
    def ascii(self):
        """Whether this `Column` represents an column in an ASCII table."""

        return isinstance(self.format, _AsciiColumnFormat)

    @lazyproperty
    def dtype(self):
        return self.format.dtype

    def copy(self):
        """
        Return a copy of this `Column`.
        """
        tmp = Column(format='I')  # just use a throw-away format
        tmp.__dict__ = self.__dict__.copy()
        return tmp

    if sys.version_info < (2, 7):
        # This is only needed on Python 2.6, where it appears deepcopy has
        # problems with weakrefs, and especially weak-keyed dicts.
        def __deepcopy__(self, memo=None):
            tmp = object.__new__(self.__class__)
            tmp_dict = dict(self.__dict__)
            array = self.array
            listeners = None
            if array is not None:
                tmp_dict['array'] = array.copy()
            tmp_dict['_parent_fits_rec'] = None

            if '_listeners' in tmp_dict:
                listners = tmp_dict['_listeners']
                del tmp_dict['_listeners']

            tmp.__dict__ = copy.deepcopy(tmp_dict, memo=memo)

            if listeners is not None:
                tmp.__dict__['_listeners'] = listeners
            return tmp

    @staticmethod
    def _convert_format(format, cls):
        """The format argument to this class's initializer may come in many
        forms.  This uses the given column format class ``cls`` to convert
        to a format of that type.

        TODO: There should be an abc base class for column format classes
        """

        # Short circuit in case we're already a _BaseColumnFormat--there is at
        # least one case in which this can happen
        if isinstance(format, _BaseColumnFormat):
            return format, format.recformat

        if format in NUMPY2FITS:
            with ignored(VerifyError):
                # legit recarray format?
                recformat = format
                format = cls.from_recformat(format)

        try:
            # legit FITS format?
            format = cls(format)
            recformat = format.recformat
        except VerifyError:
            raise VerifyError('Illegal format `%s`.' % format)

        return format, recformat

    @classmethod
    def _verify_keywords(cls, name=None, format=None, unit=None, null=None,
                         bscale=None, bzero=None, disp=None, start=None,
                         dim=None, ascii=None):
        """
        Given the keyword arguments used to initialize a Column, specifically
        those that typically read from a FITS header (so excluding array),
        verify that each keyword has a valid value.

        Returns a 2-tuple of dicts.  The first maps valid keywords to their
        values.  The second maps invalid keywords to a 2-tuple of their value,
        and a message explaining why they were found invalid.
        """

        valid = {}
        invalid = {}

        format, recformat = cls._determine_formats(format, start, dim, ascii)
        valid.update(format=format, recformat=recformat)

        # Currently we don't have any validation for name, unit, bscale, or
        # bzero so include those by default
        # TODO: Add validation for these keywords, obviously
        for k, v in [('name', name), ('unit', unit), ('bscale', bscale),
                     ('bzero', bzero)]:
            if v is not None and v != '':
                valid[k] = v

        # Validate null option
        # Note: Enough code exists that thinks empty strings are sensible
        # inputs for these options that we need to treat '' as None
        if null is not None and null != '':
            msg = None
            if isinstance(format, _AsciiColumnFormat):
                null = str(null)
                if len(null) > format.width:
                    msg = (
                        "ASCII table null option (TNULLn) is longer than "
                        "the column's character width and will be truncated "
                        "(got %r)." % null)
            else:
                if not _is_int(null):
                    # Make this an exception instead of a warning, since any
                    # non-int value is meaningless
                    msg = (
                        'Column null option (TNULLn) must be an integer for '
                        'binary table columns (got %r).  The invalid value '
                        'will be ignored for the purpose of formatting '
                        'the data in this column.' % null)

                tnull_formats = ('B', 'I', 'J', 'K')

                if not (format.format in tnull_formats or
                        (format.format in ('P', 'Q') and
                         format.p_format in tnull_formats)):
                    # TODO: We should also check that TNULLn's integer value
                    # is in the range allowed by the column's format
                    msg = (
                        'Column null option (TNULLn) is invalid for binary '
                        'table columns of type %r (got %r).  The invalid '
                        'value will be ignored for the purpose of formatting '
                        'the data in this column.' % (format, null))

            if msg is None:
                valid['null'] = null
            else:
                invalid['null'] = (null, msg)

        # Validate the disp option
        # TODO: Add full parsing and validation of TDISPn keywords
        if disp is not None and disp != '':
            msg = None
            if not isinstance(disp, string_types):
                msg = (
                    'Column disp option (TDISPn) must be a string (got %r).'
                    'The invalid value will be ignored for the purpose of '
                    'formatting the data in this column.' % disp)

            if (isinstance(format, _AsciiColumnFormat) and
                    disp[0].upper() == 'L'):
                # disp is at least one character long and has the 'L' format
                # which is not recognized for ASCII tables
                msg = (
                    "Column disp option (TDISPn) may not use the 'L' format "
                    "with ASCII table columns.  The invalid value will be "
                    "ignored for the purpose of formatting the data in this "
                    "column.")

            if msg is None:
                valid['disp'] = disp
            else:
                invalid['disp'] = (disp, msg)

        # Validate the start option
        if start is not None and start != '':
            msg = None
            if not isinstance(format, _AsciiColumnFormat):
                # The 'start' option only applies to ASCII columns
                msg = (
                    'Column start option (TBCOLn) is not allowed for binary '
                    'table columns (got %r).  The invalid keyword will be '
                    'ignored for the purpose of formatting the data in this '
                    'column.'% start)
            try:
                start = int(start)
            except (TypeError, ValueError):
                pass

            if not _is_int(start) and start < 1:
                msg = (
                    'Column start option (TBCOLn) must be a positive integer '
                    '(got %r).  The invalid value will be ignored for the '
                    'purpose of formatting the data in this column.' % start)

            if msg is None:
                valid['start'] = start
            else:
                invalid['start'] = (start, msg)

        # Process TDIMn options
        # ASCII table columns can't have a TDIMn keyword associated with it;
        # for now we just issue a warning and ignore it.
        # TODO: This should be checked by the FITS verification code
        if dim is not None and dim != '':
            msg = None
            dims_tuple = tuple()
            # NOTE: If valid, the dim keyword's value in the the valid dict is
            # a tuple, not the original string; if invalid just the original
            # string is returned
            if isinstance(format, _AsciiColumnFormat):
                msg = (
                    'Column dim option (TDIMn) is not allowed for ASCII table '
                    'columns (got %r).  The invalid keyword will be ignored '
                    'for the purpose of formatting this column.' % dim)

            elif isinstance(dim, string_types):
                dims_tuple = _parse_tdim(dim)
            elif isinstance(dim, tuple):
                dims_tuple = dim
            else:
                msg = (
                    "`dim` argument must be a string containing a valid value "
                    "for the TDIMn header keyword associated with this column, "
                    "or a tuple containing the C-order dimensions for the "
                    "column.  The invalid value will be ignored for the purpose "
                    "of formatting this column.")

            if dims_tuple:
                if reduce(operator.mul, dims_tuple) > format.repeat:
                    msg = (
                        "The repeat count of the column format %r for column %r "
                        "is fewer than the number of elements per the TDIM "
                        "argument %r.  The invalid TDIMn value will be ignored "
                        "for the purpose of formatting this column." %
                        (name, format, dim))

            if msg is None:
                valid['dim'] = dims_tuple
            else:
                invalid['dim'] = (dim, msg)

        return valid, invalid

    @classmethod
    def _determine_formats(cls, format, start, dim, ascii):
        """
        Given a format string and whether or not the Column is for an
        ASCII table (ascii=None means unspecified, but lean toward binary table
        where ambiguous) create an appropriate _BaseColumnFormat instance for
        the column's format, and determine the appropriate recarray format.

        The values of the start and dim keyword arguments are also useful, as
        the former is only valid for ASCII tables and the latter only for
        BINARY tables.
        """

        # If the given format string is unambiguously a Numpy dtype or one of
        # the Numpy record format type specifiers supported by PyFITS then that
        # should take priority--otherwise assume it is a FITS format
        if isinstance(format, np.dtype):
            format, _, _ = _dtype_to_recformat(format)

        # check format
        if ascii is None and not isinstance(format, _BaseColumnFormat):
            # We're just give a string which could be either a Numpy format
            # code, or a format for a binary column array *or* a format for an
            # ASCII column array--there may be many ambiguities here.  Try our
            # best to guess what the user intended.
            format, recformat = cls._guess_format(format, start, dim)
        elif not ascii and not isinstance(format, _BaseColumnFormat):
            format, recformat = cls._convert_format(format, _ColumnFormat)
        elif ascii and not isinstance(format, _AsciiColumnFormat):
            format, recformat = cls._convert_format(format,
                                                    _AsciiColumnFormat)
        else:
            # The format is already acceptable and unambiguous
            recformat = format.recformat

        return format, recformat

    @classmethod
    def _guess_format(cls, format, start, dim):
        if start and dim:
            # This is impossible; this can't be a valid FITS column
            raise ValueError(
                'Columns cannot have both a start (TCOLn) and dim '
                '(TDIMn) option, since the former is only applies to '
                'ASCII tables, and the latter is only valid for binary '
                'tables.')
        elif start:
            # Only ASCII table columns can have a 'start' option
            guess_format = _AsciiColumnFormat
        elif dim:
            # Only binary tables can have a dim option
            guess_format = _ColumnFormat
        else:
            # If the format is *technically* a valid binary column format
            # (i.e. it has a valid format code followed by arbitrary
            # "optional" codes), but it is also strictly a valid ASCII
            # table format, then assume an ASCII table column was being
            # requested (the more likely case, after all).
            with ignored(VerifyError):
                format = _AsciiColumnFormat(format, strict=True)

            # A safe guess which reflects the existing behavior of previous
            # PyFITS versions
            guess_format = _ColumnFormat

        try:
            format, recformat = cls._convert_format(format, guess_format)
        except VerifyError:
            # For whatever reason our guess was wrong (for example if we got
            # just 'F' that's not a valid binary format, but it an ASCII format
            # code albeit with the width/precision omitted
            guess_format = (_AsciiColumnFormat
                            if guess_format is _ColumnFormat
                            else _ColumnFormat)
            # If this fails too we're out of options--it is truly an invalid
            # format, or at least not supported
            format, recformat = cls._convert_format(format, guess_format)

        return format, recformat

    def _convert_to_valid_data_type(self, array):
        # Convert the format to a type we understand
        if isinstance(array, Delayed):
            return array
        elif array is None:
            return array
        else:
            format = self.format
            dims = self._dims

            if dims:
                shape = dims[:-1] if 'A' in format else dims
                shape = (len(array),) + shape
                array = array.reshape(shape)

            if 'P' in format or 'Q' in format:
                return array
            elif 'A' in format:
                if array.dtype.char in 'SU':
                    if dims:
                        # The 'last' dimension (first in the order given
                        # in the TDIMn keyword itself) is the number of
                        # characters in each string
                        fsize = dims[-1]
                    else:
                        fsize = np.dtype(format.recformat).itemsize
                    return chararray.array(array, itemsize=fsize)
                else:
                    return _convert_array(array, np.dtype(format.recformat))
            elif 'L' in format:
                # boolean needs to be scaled back to storage values ('T', 'F')
                if array.dtype == np.dtype('bool'):
                    return np.where(array == False, ord('F'), ord('T'))
                else:
                    return np.where(array == 0, ord('F'), ord('T'))
            elif 'X' in format:
                return _convert_array(array, np.dtype('uint8'))
            else:
                # Preserve byte order of the original array for now; see #77
                numpy_format = array.dtype.byteorder + format.recformat

                # Handle arrays passed in as unsigned ints as pseudo-unsigned
                # int arrays; blatantly tacked in here for now--we need columns
                # to have explicit knowledge of whether they treated as
                # pseudo-unsigned
                bzeros = {2: np.uint16(2**15), 4: np.uint32(2**31),
                          8: np.uint64(2**63)}
                if (array.dtype.kind == 'u' and
                        array.dtype.itemsize in bzeros and
                        self.bscale in (1, None, '') and
                        self.bzero == bzeros[array.dtype.itemsize]):
                    # Basically the array is uint, has scale == 1.0, and the
                    # bzero is the appropriate value for a pseudo-unsigned
                    # integer of the input dtype, then go ahead and assume that
                    # uint is assumed
                    numpy_format = numpy_format.replace('i', 'u')
                    self._pseudo_unsigned_ints = True

                # The .base here means we're dropping the shape information,
                # which is only used to format recarray fields, and is not
                # useful for converting input arrays to the correct data type
                dtype = np.dtype(numpy_format).base

                return _convert_array(array, dtype)


class ColDefs(NotifierMixin):
    """
    Column definitions class.

    It has attributes corresponding to the `Column` attributes
    (e.g. `ColDefs` has the attribute ``names`` while `Column`
    has ``name``). Each attribute in `ColDefs` is a list of
    corresponding attribute values from all `Column` objects.
    """

    _padding_byte = '\x00'
    _col_format_cls = _ColumnFormat

    def __new__(cls, input, tbtype=None, ascii=False):
        if tbtype is not None:
            warnings.warn(
                'The ``tbtype`` argument to `ColDefs` is deprecated as of '
                'PyFITS 3.3; instead the appropriate table type should be '
                'inferred from the formats of the supplied columns.  Use the '
                '``ascii=True`` argument to ensure that ASCII table columns '
                'are used.')
        else:
            tbtype = 'BinTableHDU'  # The old default

        # Backwards-compat support
        # TODO: Remove once the tbtype argument is removed entirely
        if tbtype == 'BinTableHDU':
            klass = cls
        elif tbtype == 'TableHDU':
            klass = _AsciiColDefs
        else:
            raise ValueError('Invalid table type: %s.' % tbtype)

        if (hasattr(input, '_columns_type') and
                issubclass(input._columns_type, ColDefs)):
            klass = input._columns_type
        elif (hasattr(input, '_col_format_cls') and
                issubclass(input._col_format_cls, _AsciiColumnFormat)):
            klass = _AsciiColDefs

        if ascii:  # force ASCII if this has been explicitly requested
            klass = _AsciiColDefs

        return object.__new__(klass)

    def __getnewargs__(self):
        return (self._arrays,)

    def __init__(self, input, tbtype=None, ascii=False):
        """
        Parameters
        ----------

        input : sequence of `Column`, `ColDefs`, other
            An existing table HDU, an existing `ColDefs`, or any multi-field
            Numpy array or `numpy.recarray`.

        **(Deprecated)** tbtype : str, optional
            which table HDU, ``"BinTableHDU"`` (default) or
            ``"TableHDU"`` (text table).
            Now ColDefs for a normal (binary) table by default, but converted
            automatically to ASCII table ColDefs in the appropriate contexts
            (namely, when creating an ASCII table).

        ascii : bool
        """

        from pyfits.hdu.table import _TableBaseHDU
        from pyfits.fitsrec import FITS_rec

        if isinstance(input, ColDefs):
            self._init_from_coldefs(input)
        elif (isinstance(input, FITS_rec) and hasattr(input, '_coldefs') and
                input._coldefs):
            # If given a FITS_rec object we can directly copy its columns, but
            # only if its columns have already been defined, otherwise this
            # will loop back in on itself and blow up
            self._init_from_coldefs(input._coldefs)
        elif isinstance(input, np.ndarray) and input.dtype.fields is not None:
            # Construct columns from the fields of a record array
            self._init_from_array(input)
        elif isiterable(input):
            # if the input is a list of Columns
            self._init_from_sequence(input)
        elif isinstance(input, _TableBaseHDU):
            # Construct columns from fields in an HDU header
            self._init_from_table(input)
        else:
            raise TypeError('Input to ColDefs must be a table HDU, a list '
                            'of Columns, or a record/field array.')

        # Listen for changes on all columns
        for col in self.columns:
            col._add_listener(self)

    def _init_from_coldefs(self, coldefs):
        """Initialize from an existing ColDefs object (just copy the
        columns and convert their formats if necessary).
        """

        self.columns = [self._copy_column(col) for col in coldefs]

    def _init_from_sequence(self, columns):
        for idx, col in enumerate(columns):
            if not isinstance(col, Column):
                raise TypeError(
                    'Element %d in the ColDefs input is not a Column.' % idx)

        self._init_from_coldefs(columns)

    def _init_from_array(self, array):
        self.columns = []
        for idx in range(len(array.dtype)):
            cname = array.dtype.names[idx]
            ftype = array.dtype.fields[cname][0]
            format = self._col_format_cls.from_recformat(ftype)

            # Determine the appropriate dimensions for items in the column
            # (typically just 1D)
            dim = array.dtype[idx].shape[::-1]
            if dim and (len(dim) > 1 or 'A' in format):
                if 'A' in format:
                    # n x m string arrays must include the max string
                    # length in their dimensions (e.g. l x n x m)
                    dim = (array.dtype[idx].base.itemsize,) + dim
                dim = repr(dim).replace(' ', '')
            else:
                dim = None

            # Check for unsigned ints.
            bzero = None
            if 'I' in format and ftype == np.dtype('uint16'):
                bzero = np.uint16(2**15)
            elif 'J' in format and ftype == np.dtype('uint32'):
                bzero = np.uint32(2**31)
            elif 'K' in format and ftype == np.dtype('uint64'):
                bzero = np.uint64(2**63)

            c = Column(name=cname, format=format,
                       array=array.view(np.ndarray)[cname], bzero=bzero,
                       dim=dim)
            self.columns.append(c)

    def _init_from_table(self, table):
        hdr = table._header
        nfields = hdr['TFIELDS']

        # go through header keywords to pick out column definition keywords
        # definition dictionaries for each field
        col_keywords = [{} for i in range(nfields)]
        for keyword, value in iteritems(hdr):
            key = TDEF_RE.match(keyword)
            try:
                keyword = key.group('label')
            except:
                continue  # skip if there is no match
            if keyword in KEYWORD_NAMES:
                col = int(key.group('num'))
                if col <= nfields and col > 0:
                    attr = KEYWORD_TO_ATTRIBUTE[keyword]
                    if attr == 'format':
                        # Go ahead and convert the format value to the
                        # appropriate ColumnFormat container now
                        value = self._col_format_cls(value)
                    col_keywords[col - 1][attr] = value

        # Verify the column keywords and display any warnings if necessary;
        # we only want to pass on the valid keywords
        for idx, kwargs in enumerate(col_keywords):
            valid_kwargs, invalid_kwargs = Column._verify_keywords(**kwargs)
            for val in invalid_kwargs.values():
                warnings.warn(
                    'Invalid keyword for column %d: %s' % (idx + 1, val[1]),
                    VerifyWarning)
            # Special cases for recformat and dim
            # TODO: Try to eliminate the need for these special cases
            del valid_kwargs['recformat']
            if 'dim' in valid_kwargs:
                valid_kwargs['dim'] = kwargs['dim']
            col_keywords[idx] = valid_kwargs

        # data reading will be delayed
        for col in range(nfields):
            col_keywords[col]['array'] = Delayed(table, col)

        # now build the columns
        self.columns = [Column(**attrs) for attrs in col_keywords]

        # Add the table HDU is a listener to changes to the columns
        # (either changes to individual columns, or changes to the set of
        # columns (add/remove/etc.))
        self._add_listener(table)

    def __copy__(self):
        return self.__class__(self)

    def __deepcopy__(self, memo):
        return self.__class__([copy.deepcopy(c, memo) for c in self.columns])

    def _copy_column(self, column):
        """Utility function used currently only by _init_from_coldefs
        to help convert columns from binary format to ASCII format or vice
        versa if necessary (otherwise performs a straight copy).
        """

        if isinstance(column.format, self._col_format_cls):
            # This column has a FITS format compatible with this column
            # definitions class (that is ascii or binary)
            return column.copy()

        new_column = column.copy()

        # Try to use the Numpy recformat as the equivalency between the
        # two formats; if that conversion can't be made then these
        # columns can't be transferred
        # TODO: Catch exceptions here and raise an explicit error about
        # column format conversion
        new_column.format = self._col_format_cls.from_column_format(
                column.format)

        # Handle a few special cases of column format options that are not
        # compatible between ASCII an binary tables
        # TODO: This is sort of hacked in right now; we really need
        # separate classes for ASCII and Binary table Columns, and they
        # should handle formatting issues like these
        if not isinstance(new_column.format, _AsciiColumnFormat):
            # the column is a binary table column...
            new_column.start = None
            if new_column.null is not None:
                # We can't just "guess" a value to represent null
                # values in the new column, so just disable this for
                # now; users may modify it later
                new_column.null = None
        else:
            # the column is an ASCII table column...
            if new_column.null is not None:
                new_column.null = DEFAULT_ASCII_TNULL
            if (new_column.disp is not None and
                    new_column.disp.upper().startswith('L')):
                # ASCII columns may not use the logical data display format;
                # for now just drop the TDISPn option for this column as we
                # don't have a systematic conversion of boolean data to ASCII
                # tables yet
                new_column.disp = None

        return new_column

    def __getattr__(self, name):
        """
        Automatically returns the values for the given keyword attribute for
        all `Column`s in this list.

        Implements for example self.units, self.formats, etc.
        """

        cname = name[:-1]
        if cname in KEYWORD_ATTRIBUTES and name[-1] == 's':
            attr = []
            for col in self:
                val = getattr(col, cname)
                if val is not None:
                    attr.append(val)
                else:
                    attr.append('')
            return attr
        raise AttributeError(name)

    @lazyproperty
    def dtype(self):
        # Note: This previously returned a dtype that just used the raw field
        # widths based on the format's repeat count, and did not incorporate
        # field *shapes* as provided by TDIMn keywords.
        # Now this incorporates TDIMn from the start, which makes *this* method
        # a little more complicated, but simplifies code elsewhere (for example
        # fields will have the correct shapes even in the raw recarray).
        fields = []
        offsets = [0]

        for name, format_, dim in zip(self.names, self.formats, self._dims):
            dt = format_.dtype

            if len(offsets) < len(self.formats):
                # Note: the size of the *original* format_ may be greater than
                # one would expect from the number of elements determined by
                # dim.  The FITS format allows this--the rest of the field is
                # filled with undefined values.
                offsets.append(offsets[-1] + dt.itemsize)

            if dim:
                if format_.format == 'A':
                    dt = np.dtype((dt.char + str(dim[-1]), dim[:-1]))
                else:
                    dt = np.dtype((dt.base, dim))

            fields.append((name, dt))

        return nh.realign_dtype(np.dtype(fields), offsets)

    @lazyproperty
    def _arrays(self):
        return [col.array for col in self.columns]

    @lazyproperty
    def _recformats(self):
        return [fmt.recformat for fmt in self.formats]

    @lazyproperty
    def _dims(self):
        """Returns the values of the TDIMn keywords parsed into tuples."""

        return [col._dims for col in self.columns]

    def __getitem__(self, key):
        if isinstance(key, string_types):
            key = _get_index(self.names, key)

        x = self.columns[key]
        if _is_int(key):
            return x
        else:
            return ColDefs(x)

    def __len__(self):
        return len(self.columns)

    def __repr__(self):
        rep = 'ColDefs('
        if hasattr(self, 'columns') and self.columns:
            # The hasattr check is mostly just useful in debugging sessions
            # where self.columns may not be defined yet
            rep += '\n    '
            rep += '\n    '.join([repr(c) for c in self.columns])
            rep += '\n'
        rep += ')'
        return rep

    def __add__(self, other, option='left'):
        if isinstance(other, Column):
            b = [other]
        elif isinstance(other, ColDefs):
            b = list(other.columns)
        else:
            raise TypeError('Wrong type of input.')
        if option == 'left':
            tmp = list(self.columns) + b
        else:
            tmp = b + list(self.columns)
        return ColDefs(tmp)

    def __radd__(self, other):
        return self.__add__(other, 'right')

    def __sub__(self, other):
        if not isinstance(other, (list, tuple)):
            other = [other]
        _other = [_get_index(self.names, key) for key in other]
        indx = range(len(self))
        for x in _other:
            indx.remove(x)
        tmp = [self[i] for i in indx]
        return ColDefs(tmp)

    def _update_column_attribute_changed(self, column, attr, old_value,
                                         new_value):
        """
        Handle column attribute changed notifications from columns that are
        members of this `ColDefs`.

        `ColDefs` itself does not currently do anything with this, and just
        bubbles the notification up to any listening table HDUs that may need
        to update their headers, etc.  However, this also informs the table of
        the numerical index of the column that changed.
        """

        idx = 0
        for idx, col in enumerate(self.columns):
            if col is column:
                break

        self._notify('column_attribute_changed', column, idx, attr, old_value,
                     new_value)

    def add_col(self, column):
        """
        Append one `Column` to the column definition.

        .. warning::

            *New in pyfits 2.3*: This function appends the new column
            to the `ColDefs` object in place.  Prior to pyfits 2.3,
            this function returned a new `ColDefs` with the new column
            at the end.
        """

        assert isinstance(column, Column)

        self._arrays.append(column.array)
        # Obliterate caches of certain things
        del self.dtype
        del self._recformats
        del self._dims

        self.columns.append(column)

        # Listen for changes on the new column
        column._add_listener(self)

        # If this ColDefs is being tracked by a Table, inform the
        # table that its data is now invalid.
        self._notify('column_added', self, column)
        return self

    def del_col(self, col_name):
        """
        Delete (the definition of) one `Column`.

        col_name : str or int
            The column's name or index
        """

        indx = _get_index(self.names, col_name)
        col = self.columns[indx]

        del self._arrays[indx]
        # Obliterate caches of certain things
        del self.dtype
        del self._recformats
        del self._dims

        del self.columns[indx]

        col._remove_listener(self)

        # If this ColDefs is being tracked by a table HDU, inform the HDU (or
        # any other listeners) that the column has been removed
        # Just send a reference to self, and the index of the column that was
        # removed
        self._notify('column_removed', self, indx)
        return self

    def change_attrib(self, col_name, attrib, new_value):
        """
        Change an attribute (in the ``KEYWORD_ATTRIBUTES`` list) of a `Column`.

        Parameters
        ----------
        col_name : str or int
            The column name or index to change

        attrib : str
            The attribute name

        new_value : object
            The new value for the attribute
        """

        setattr(self[col_name], attrib, new_value)

    def change_name(self, col_name, new_name):
        """
        Change a `Column`'s name.

        Parameters
        ----------
        col_name : str
            The current name of the column

        new_name : str
            The new name of the column
        """

        if new_name != col_name and new_name in self.names:
            raise ValueError('New name %s already exists.' % new_name)
        else:
            self.change_attrib(col_name, 'name', new_name)

    def change_unit(self, col_name, new_unit):
        """
        Change a `Column`'s unit.

        Parameters
        ----------
        col_name : str or int
            The column name or index

        new_unit : str
            The new unit for the column
        """

        self.change_attrib(col_name, 'unit', new_unit)

    def info(self, attrib='all', output=None):
        """
        Get attribute(s) information of the column definition.

        Parameters
        ----------
        attrib : str
            Can be one or more of the attributes listed in
            ``pyfits.column.KEYWORD_ATTRIBUTES``.  The default is ``"all"``
            which will print out all attributes.  It forgives plurals and
            blanks.  If there are two or more attribute names, they must be
            separated by comma(s).

        output : file, optional
            File-like object to output to.  Outputs to stdout by default.
            If `False`, returns the attributes as a `dict` instead.

        Notes
        -----
        This function doesn't return anything by default; it just prints to
        stdout.
        """

        if output is None:
            output = sys.stdout

        if attrib.strip().lower() in ['all', '']:
            lst = KEYWORD_ATTRIBUTES
        else:
            lst = attrib.split(',')
            for idx in range(len(lst)):
                lst[idx] = lst[idx].strip().lower()
                if lst[idx][-1] == 's':
                    lst[idx] = list[idx][:-1]

        ret = {}

        for attr in lst:
            if output:
                if attr not in KEYWORD_ATTRIBUTES:
                    output.write("'%s' is not an attribute of the column "
                                 "definitions.\n" % attr)
                    continue
                output.write("%s:\n" % attr)
                output.write('    %s\n' % getattr(self, attr + 's'))
            else:
                ret[attr] = getattr(self, attr + 's')

        if not output:
            return ret


class _AsciiColDefs(ColDefs):
    """ColDefs implementation for ASCII tables."""

    _padding_byte = ' '
    _col_format_cls = _AsciiColumnFormat

    def __init__(self, input, tbtype=None, ascii=True):
        super(_AsciiColDefs, self).__init__(input)

        # if the format of an ASCII column has no width, add one
        if not isinstance(input, _AsciiColDefs):
            self._update_field_metrics()
        else:
            for idx, s in enumerate(input.starts):
                self.columns[idx].start = s

            self._spans = input.spans
            self._width = input._width

    @lazyproperty
    def dtype(self):
        _itemsize = self.spans[-1] + self.starts[-1] - 1
        dtype = {}

        for j in range(len(self)):
            data_type = 'S' + str(self.spans[j])
            dtype[self.names[j]] = (data_type, self.starts[j] - 1)

        return np.dtype(dtype)

    @property
    def spans(self):
        """A list of the widths of each field in the table."""

        return self._spans

    @lazyproperty
    def _recformats(self):
        if len(self) == 1:
            widths = []
        else:
            widths = [y - x for x, y in pairwise(self.starts)]

        # Widths is the width of each field *including* any space between
        # fields; this is so that we can map the fields to string records in a
        # Numpy recarray
        widths.append(self._width - self.starts[-1] + 1)
        return ['a' + str(w) for w in widths]

    def add_col(self, column):
        super(_AsciiColDefs, self).add_col(column)
        self._update_field_metrics()

    def del_col(self, col_name):
        super(_AsciiColDefs, self).del_col(col_name)
        self._update_field_metrics()

    def _update_field_metrics(self):
        """
        Updates the list of the start columns, the list of the widths of each
        field, and the total width of each record in the table.
        """

        spans = [0] * len(self.columns)
        end_col = 0  # Refers to the ASCII text column, not the table col
        for idx, col in enumerate(self.columns):
            width = col.format.width

            # Update the start columns and column span widths taking into
            # account the case that the starting column of a field may not
            # be the column immediately after the previous field
            if not col.start:
                col.start = end_col + 1
            end_col = col.start + width - 1
            spans[idx] = width

        self._spans = spans
        self._width = end_col


# Utilities


class _VLF(np.ndarray):
    """Variable length field object."""

    def __new__(cls, input, dtype='a'):
        """
        Parameters
        ----------
        input
            a sequence of variable-sized elements.
        """

        if dtype == 'a':
            try:
                # this handles ['abc'] and [['a','b','c']]
                # equally, beautiful!
                input = [chararray.array(x, itemsize=1) for x in input]
            except:
                raise ValueError('Inconsistent input data array: %s' % input)

        a = np.array(input, dtype=np.object)
        self = np.ndarray.__new__(cls, shape=(len(input),), buffer=a,
                                  dtype=np.object)
        self.max = 0
        self.element_dtype = dtype
        return self

    def __array_finalize__(self, obj):
        if obj is None:
            return
        self.max = obj.max
        self.element_dtype = obj.element_dtype

    def __setitem__(self, key, value):
        """
        To make sure the new item has consistent data type to avoid
        misalignment.
        """

        if isinstance(value, np.ndarray) and value.dtype == self.dtype:
            pass
        elif isinstance(value, chararray.chararray) and value.itemsize == 1:
            pass
        elif self.element_dtype == 'a':
            value = chararray.array(value, itemsize=1)
        else:
            value = np.array(value, dtype=self.element_dtype)
        np.ndarray.__setitem__(self, key, value)
        self.max = max(self.max, len(value))


def _get_index(names, key):
    """
    Get the index of the ``key`` in the ``names`` list.

    The ``key`` can be an integer or string.  If integer, it is the index
    in the list.  If string,

        a. Field (column) names are case sensitive: you can have two
           different columns called 'abc' and 'ABC' respectively.

        b. When you *refer* to a field (presumably with the field
           method), it will try to match the exact name first, so in
           the example in (a), field('abc') will get the first field,
           and field('ABC') will get the second field.

        If there is no exact name matched, it will try to match the
        name with case insensitivity.  So, in the last example,
        field('Abc') will cause an exception since there is no unique
        mapping.  If there is a field named "XYZ" and no other field
        name is a case variant of "XYZ", then field('xyz'),
        field('Xyz'), etc. will get this field.
    """

    if _is_int(key):
        indx = int(key)
    elif isinstance(key, string_types):
        # try to find exact match first
        try:
            indx = names.index(key.rstrip())
        except ValueError:
            # try to match case-insentively,
            _key = key.lower().rstrip()
            names = [n.lower().rstrip() for n in names]
            count = names.count(_key)  # occurrence of _key in names
            if count == 1:
                indx = names.index(_key)
            elif count == 0:
                raise KeyError("Key '%s' does not exist." % key)
            else:              # multiple match
                raise KeyError("Ambiguous key name '%s'." % key)
    else:
        raise KeyError("Illegal key '%s'." % repr(key))

    return indx


def _unwrapx(input, output, repeat):
    """
    Unwrap the X format column into a Boolean array.

    Parameters
    ----------
    input
        input ``Uint8`` array of shape (`s`, `nbytes`)

    output
        output Boolean array of shape (`s`, `repeat`)

    repeat
        number of bits
    """

    pow2 = np.array([128, 64, 32, 16, 8, 4, 2, 1], dtype='uint8')
    nbytes = ((repeat - 1) // 8) + 1
    for i in range(nbytes):
        _min = i * 8
        _max = min((i + 1) * 8, repeat)
        for j in range(_min, _max):
            output[..., j] = np.bitwise_and(input[..., i], pow2[j - i * 8])


def _wrapx(input, output, repeat):
    """
    Wrap the X format column Boolean array into an ``UInt8`` array.

    Parameters
    ----------
    input
        input Boolean array of shape (`s`, `repeat`)

    output
        output ``Uint8`` array of shape (`s`, `nbytes`)

    repeat
        number of bits
    """

    output[...] = 0  # reset the output
    nbytes = ((repeat - 1) // 8) + 1
    unused = nbytes * 8 - repeat
    for i in range(nbytes):
        _min = i * 8
        _max = min((i + 1) * 8, repeat)
        for j in range(_min, _max):
            if j != _min:
                np.left_shift(output[..., i], 1, output[..., i])
            np.add(output[..., i], input[..., j], output[..., i])

    # shift the unused bits
    np.left_shift(output[..., i], unused, output[..., i])


def _makep(array, descr_output, format, nrows=None):
    """
    Construct the P (or Q) format column array, both the data descriptors and
    the data.  It returns the output "data" array of data type `dtype`.

    The descriptor location will have a zero offset for all columns
    after this call.  The final offset will be calculated when the file
    is written.

    Parameters
    ----------
    array
        input object array

    descr_output
        output "descriptor" array of data type int32 (for P format arrays) or
        int64 (for Q format arrays)--must be nrows long in its first dimension

    format
        the _FormatP object representing the format of the variable array

    nrows : int, optional
        number of rows to create in the column; defaults to the number of rows
        in the input array
    """

    # TODO: A great deal of this is redundant with FITS_rec._convert_p; see if
    # we can merge the two somehow.

    _offset = 0

    if not nrows:
        nrows = len(array)
    n = min(len(array), nrows)

    data_output = _VLF([None] * nrows, dtype=format.dtype)

    if format.dtype == 'a':
        _nbytes = 1
    else:
        _nbytes = np.array([], dtype=format.dtype).itemsize

    for idx in range(nrows):
        if idx < len(array):
            rowval = array[idx]
        else:
            if format.dtype == 'a':
                rowval = ' ' * data_output.max
            else:
                rowval = [0] * data_output.max
        if format.dtype == 'a':
            data_output[idx] = chararray.array(encode_ascii(rowval),
                                               itemsize=1)
        else:
            data_output[idx] = np.array(rowval, dtype=format.dtype)

        descr_output[idx, 0] = len(data_output[idx])
        descr_output[idx, 1] = _offset
        _offset += len(data_output[idx]) * _nbytes

    return data_output


def _parse_tformat(tform):
    """Parse ``TFORMn`` keyword for a binary table into a
    ``(repeat, format, option)`` tuple.
    """

    try:
        (repeat, format, option) = TFORMAT_RE.match(tform.strip()).groups()
    except:
        # TODO: Maybe catch this error use a default type (bytes, maybe?) for
        # unrecognized column types.  As long as we can determine the correct
        # byte width somehow..
        raise VerifyError('Format %r is not recognized.' % tform)

    if repeat == '':
        repeat = 1
    else:
        repeat = int(repeat)

    return (repeat, format.upper(), option)


def _parse_ascii_tformat(tform, strict=False):
    """
    Parse the ``TFORMn`` keywords for ASCII tables into a ``(format, width,
    precision)`` tuple (the latter is always zero unless format is one of 'E',
    'F', or 'D').
    """

    match = TFORMAT_ASCII_RE.match(tform.strip())
    if not match:
        raise VerifyError('Format %r is not recognized.' % tform)

    # Be flexible on case
    format = match.group('format')
    if format is None:
        # Floating point format
        format = match.group('formatf').upper()
        width = match.group('widthf')
        precision = match.group('precision')
        if width is None or precision is None:
            if strict:
                raise VerifyError('Format %r is not unambiguously an ASCII '
                                  'table format.')
            else:
                width = 0 if width is None else width
                precision = 1 if precision is None else precision
    else:
        format = format.upper()
        width = match.group('width')
        if width is None:
            if strict:
                raise VerifyError('Format %r is not unambiguously an ASCII '
                                  'table format.')
            else:
                # Just use a default width of 0 if unspecified
                width = 0
        precision = 0

    def convert_int(val):
        msg = ('Format %r is not valid--field width and decimal precision '
               'must be integers.')
        try:
            val = int(val)
        except (ValueError, TypeError):
            raise VerifyError(msg % tform)

        return val

    if width and precision:
        # This should only be the case for floating-point formats
        width, precision = convert_int(width), convert_int(precision)
    elif width:
        # Just for integer/string formats; ignore precision
        width = convert_int(width)
    else:
        # For any format, if width was unspecified use the set defaults
        width, precision = ASCII_DEFAULT_WIDTHS[format]

    if width <= 0:
        raise VerifyError("Format %r not valid--field width must be a "
                          "positive integeter." % tform)

    if precision >= width:
        raise VerifyError("Format %r not valid--the number of decimal digits "
                          "must be less than the format's total width %s." &
                          (tform, width))

    return format, width, precision


def _parse_tdim(tdim):
    """Parse the ``TDIM`` value into a tuple (may return an empty tuple if
    the value ``TDIM`` value is empty or invalid).
    """

    m = tdim and TDIM_RE.match(tdim)
    if m:
        dims = m.group('dims')
        return tuple(int(d.strip()) for d in dims.split(','))[::-1]

    # Ignore any dim values that don't specify a multidimensional column
    return tuple()


def _scalar_to_format(value):
    """
    Given a scalar value or string, returns the minimum FITS column format
    that can represent that value.  'minimum' is defined by the order given in
    FORMATORDER.
    """

    # TODO: Numpy 1.6 and up has a min_scalar_type() function that can handle
    # this; in the meantime we have to use our own implementation (which for
    # now is pretty naive)

    # First, if value is a string, try to convert to the appropriate scalar
    # value
    for type_ in (int, float, complex):
        try:
            value = type_(value)
            break
        except ValueError:
            continue

    if isinstance(value, int) and value in (0, 1):
        # Could be a boolean
        return 'L'
    elif isinstance(value, int):
        for char in ('B', 'I', 'J', 'K'):
            type_ = np.dtype(FITS2NUMPY[char]).type
            if type_(value) == value:
                return char
    elif isinstance(value, float):
        # For now just assume double precision
        return 'D'
    elif isinstance(value, complex):
        return 'M'
    else:
        return 'A' + str(len(value))


def _cmp_recformats(f1, f2):
    """
    Compares two numpy recformats using the ordering given by FORMATORDER.
    """

    if f1[0] == 'a' and f2[0] == 'a':
        return cmp(int(f1[1:]), int(f2[1:]))
    else:
        f1, f2 = NUMPY2FITS[f1], NUMPY2FITS[f2]
        return cmp(FORMATORDER.index(f1), FORMATORDER.index(f2))


def _convert_fits2record(format):
    """
    Convert FITS format spec to record format spec.
    """

    repeat, dtype, option = _parse_tformat(format)

    if dtype in FITS2NUMPY:
        if dtype == 'A':
            output_format = FITS2NUMPY[dtype] + str(repeat)
            # to accommodate both the ASCII table and binary table column
            # format spec, i.e. A7 in ASCII table is the same as 7A in
            # binary table, so both will produce 'a7'.
            # Technically the FITS standard does not allow this but it's a very
            # common mistake
            if format.lstrip()[0] == 'A' and option != '':
                 # make sure option is integer
                output_format = FITS2NUMPY[dtype] + str(int(option))
        else:
            repeat_str = ''
            if repeat != 1:
                repeat_str = str(repeat)
            output_format = repeat_str + FITS2NUMPY[dtype]

    elif dtype == 'X':
        output_format = _FormatX(repeat)
    elif dtype == 'P':
        output_format = _FormatP.from_tform(format)
    elif dtype == 'Q':
        output_format = _FormatQ.from_tform(format)
    elif dtype == 'F':
        output_format = 'f8'
    else:
        raise ValueError('Illegal format %s.' % format)

    return output_format


def _convert_record2fits(format):
    """
    Convert record format spec to FITS format spec.
    """

    recformat, kind, dtype = _dtype_to_recformat(format)
    shape = dtype.shape
    itemsize = dtype.base.itemsize
    if dtype.char == 'U':
        # Unicode dtype--itemsize is 4 times actual ASCII character length,
        # which what matters for FITS column formats
        # Use dtype.base--dtype may be a multi-dimensional dtype
        itemsize = itemsize // 4

    option = str(itemsize)

    ndims = len(shape)
    repeat = 1
    if ndims > 0:
        nel = np.array(shape, dtype='i8').prod()
        if nel > 1:
            repeat = nel

    if kind == 'a':
        # This is a kludge that will place string arrays into a
        # single field, so at least we won't lose data.  Need to
        # use a TDIM keyword to fix this, declaring as (slength,
        # dim1, dim2, ...)  as mwrfits does

        ntot = int(repeat) * int(option)

        output_format = str(ntot) + 'A'
    elif recformat in NUMPY2FITS:  # record format
        if repeat != 1:
            repeat = str(repeat)
        else:
            repeat = ''
        output_format = repeat + NUMPY2FITS[recformat]
    else:
        raise ValueError('Illegal format %s.' % format)

    return output_format


def _dtype_to_recformat(dtype):
    """
    Utility function for converting a dtype object or string that instantiates
    a dtype (e.g. 'float32') into one of the two character Numpy format codes
    that have been traditionally used by PyFITS.

    In particular, use of 'a' to refer to character data is long since
    deprecated in Numpy, but PyFITS remains heavily invested in its use
    (something to try to get away from sooner rather than later).
    """

    if not isinstance(dtype, np.dtype):
        dtype = np.dtype(dtype)

    kind = dtype.base.kind

    if kind in ('U', 'S'):
        recformat = kind = 'a'
    else:
        itemsize = dtype.base.itemsize
        recformat = kind + str(itemsize)

    return recformat, kind, dtype


def _convert_format(format, reverse=False):
    """
    Convert FITS format spec to record format spec.  Do the opposite if
    reverse=True.
    """

    if reverse:
        return _convert_record2fits(format)
    else:
        return _convert_fits2record(format)


def _convert_ascii_format(format, reverse=False):
    """Convert ASCII table format spec to record format spec."""

    if reverse:
        recformat, kind, dtype = _dtype_to_recformat(format)
        itemsize = dtype.itemsize

        if kind == 'a':
            return 'A' + str(itemsize)
        elif NUMPY2FITS.get(recformat) == 'L':
            # Special case for logical/boolean types--for ASCII tables we
            # represent these as single character columns containing 'T' or 'F'
            # (a la the storage format for Logical columns in binary tables)
            return 'A1'
        elif kind == 'i':
            # Use for the width the maximum required to represent integers
            # of that byte size plus 1 for signs, but use a minimum of the
            # default width (to keep with existing behavior)
            width = 1 + len(str(2 ** (itemsize * 8)))
            width = max(width, ASCII_DEFAULT_WIDTHS['I'][0])
            return 'I' + str(width)
        elif kind == 'f':
            # This is tricky, but go ahead and use D if float-64, and E
            # if float-32 with their default widths
            if itemsize >= 8:
                format = 'D'
            else:
                format = 'E'
            width = '.'.join(str(w) for w in ASCII_DEFAULT_WIDTHS[format])
            return format + width
        # TODO: There may be reasonable ways to represent other Numpy types so
        # let's see what other possibilities there are besides just 'a', 'i',
        # and 'f'.  If it doesn't have a reasonable ASCII representation then
        # raise an exception
    else:
        format, width, precision = _parse_ascii_tformat(format)

        # This gives a sensible "default" dtype for a given ASCII
        # format code
        recformat = ASCII2NUMPY[format]

        # The following logic is taken from CFITSIO:
        # For integers, if the width <= 4 we can safely use 16-bit ints for all
        # values [for the non-standard J format code just always force 64-bit]
        if format == 'I' and width <= 4:
            recformat = 'i2'
        elif format == 'F' and width > 7:
            # 32-bit floats (the default) may not be accurate enough to support
            # all values that can fit in this field, so upgrade to 64-bit
            recformat = 'f8'
        elif format == 'E' and precision > 6:
            # Again upgrade to a 64-bit int if we require greater decimal
            # precision
            recformat = 'f8'
        elif format == 'A':
            recformat += str(width)

        return recformat