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
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595
4596
4597
4598
4599
4600
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
4616
4617
4618
4619
4620
4621
4622
4623
4624
4625
4626
4627
4628
4629
4630
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640
4641
4642
4643
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
4665
4666
4667
4668
4669
4670
4671
4672
4673
4674
4675
4676
4677
4678
4679
4680
4681
4682
4683
4684
4685
4686
4687
4688
4689
4690
4691
4692
4693
4694
4695
4696
4697
4698
4699
4700
4701
4702
4703
4704
4705
4706
4707
4708
4709
4710
4711
4712
4713
4714
4715
4716
4717
4718
4719
4720
4721
4722
4723
4724
4725
4726
4727
4728
4729
4730
4731
4732
4733
4734
4735
4736
4737
4738
4739
4740
4741
4742
4743
4744
4745
4746
4747
4748
4749
4750
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765
4766
4767
4768
4769
4770
4771
4772
4773
4774
4775
4776
4777
4778
4779
4780
4781
4782
4783
4784
4785
4786
4787
4788
4789
4790
4791
4792
4793
4794
4795
4796
4797
4798
4799
4800
4801
4802
4803
4804
4805
4806
4807
4808
4809
4810
4811
4812
4813
4814
4815
4816
4817
4818
4819
4820
4821
4822
4823
4824
4825
4826
4827
4828
4829
4830
4831
4832
4833
4834
4835
4836
4837
4838
4839
4840
4841
4842
4843
4844
4845
4846
4847
4848
4849
4850
4851
4852
4853
4854
4855
4856
4857
4858
4859
4860
4861
4862
4863
4864
4865
4866
4867
4868
4869
4870
4871
4872
4873
4874
4875
4876
4877
4878
4879
4880
4881
4882
4883
4884
4885
4886
4887
4888
4889
4890
4891
4892
4893
4894
4895
4896
4897
4898
4899
4900
4901
4902
4903
4904
4905
4906
4907
4908
4909
4910
4911
4912
4913
4914
4915
4916
4917
4918
4919
4920
4921
4922
4923
4924
4925
4926
4927
4928
4929
4930
4931
4932
4933
4934
4935
4936
4937
4938
4939
4940
4941
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
4952
4953
4954
4955
4956
4957
4958
4959
4960
4961
4962
4963
4964
4965
4966
4967
4968
4969
4970
4971
4972
4973
4974
4975
4976
4977
4978
4979
4980
4981
4982
4983
4984
4985
4986
4987
4988
4989
4990
4991
4992
4993
4994
4995
4996
4997
4998
4999
5000
5001
5002
5003
5004
5005
5006
5007
5008
5009
5010
5011
5012
5013
5014
5015
5016
5017
5018
5019
5020
5021
5022
5023
5024
5025
5026
5027
5028
5029
5030
5031
5032
5033
5034
5035
5036
5037
5038
5039
5040
5041
5042
5043
5044
5045
5046
5047
5048
5049
5050
5051
5052
5053
5054
5055
5056
5057
5058
5059
5060
5061
5062
5063
5064
5065
5066
5067
5068
5069
5070
5071
5072
5073
5074
5075
5076
5077
5078
5079
5080
5081
5082
5083
5084
5085
5086
5087
5088
5089
5090
5091
5092
5093
5094
5095
5096
5097
5098
5099
5100
5101
5102
5103
5104
5105
5106
5107
5108
5109
5110
5111
5112
5113
5114
5115
5116
5117
5118
5119
5120
5121
5122
5123
5124
5125
5126
5127
5128
5129
5130
5131
5132
5133
5134
5135
5136
5137
5138
5139
5140
5141
5142
5143
5144
5145
5146
5147
5148
5149
5150
5151
5152
5153
5154
5155
5156
5157
5158
5159
5160
5161
5162
5163
5164
5165
5166
5167
5168
5169
5170
5171
5172
5173
5174
5175
5176
5177
5178
5179
5180
5181
5182
5183
5184
5185
5186
5187
5188
5189
5190
5191
5192
5193
5194
5195
5196
5197
5198
5199
5200
5201
5202
5203
5204
5205
5206
5207
5208
5209
5210
5211
5212
5213
5214
5215
5216
5217
5218
5219
5220
5221
5222
5223
5224
5225
5226
5227
5228
5229
5230
5231
5232
5233
5234
5235
5236
5237
5238
5239
5240
5241
5242
5243
5244
5245
5246
5247
5248
5249
5250
5251
5252
5253
5254
5255
5256
5257
5258
5259
5260
5261
5262
5263
5264
5265
5266
5267
5268
5269
5270
5271
5272
5273
5274
5275
5276
5277
5278
5279
5280
5281
5282
5283
5284
5285
5286
5287
5288
5289
5290
5291
5292
5293
5294
5295
5296
5297
5298
5299
5300
5301
5302
5303
5304
5305
5306
5307
5308
5309
5310
5311
5312
5313
5314
5315
5316
5317
5318
5319
5320
5321
5322
5323
5324
5325
5326
5327
5328
5329
5330
5331
5332
5333
5334
5335
5336
5337
5338
5339
5340
5341
5342
5343
5344
5345
5346
5347
5348
5349
5350
5351
5352
5353
5354
5355
5356
5357
5358
5359
5360
5361
5362
5363
5364
5365
5366
5367
5368
5369
5370
5371
5372
5373
5374
5375
5376
5377
5378
5379
5380
5381
5382
5383
5384
5385
5386
5387
5388
5389
5390
5391
5392
5393
5394
5395
5396
5397
5398
5399
5400
5401
5402
5403
5404
5405
5406
5407
5408
5409
5410
5411
5412
5413
5414
5415
5416
5417
5418
5419
5420
5421
5422
5423
5424
5425
5426
5427
5428
5429
5430
5431
5432
5433
5434
5435
5436
5437
5438
5439
5440
5441
5442
5443
5444
5445
5446
5447
5448
5449
5450
5451
5452
5453
5454
5455
5456
5457
5458
5459
5460
5461
5462
5463
5464
5465
5466
5467
5468
5469
5470
5471
5472
5473
5474
5475
5476
5477
5478
5479
5480
5481
5482
5483
5484
5485
5486
5487
5488
5489
5490
5491
5492
5493
5494
5495
5496
5497
5498
5499
5500
5501
5502
5503
5504
5505
5506
5507
5508
5509
5510
5511
5512
5513
5514
5515
5516
5517
5518
5519
5520
5521
5522
5523
5524
5525
5526
5527
5528
5529
5530
5531
5532
5533
5534
5535
5536
5537
5538
5539
5540
5541
5542
5543
5544
5545
5546
5547
5548
5549
5550
5551
5552
5553
5554
5555
5556
5557
5558
5559
5560
5561
5562
5563
5564
5565
5566
5567
5568
5569
5570
5571
5572
5573
5574
5575
5576
5577
5578
5579
5580
5581
5582
5583
5584
5585
5586
5587
5588
5589
5590
5591
5592
5593
5594
5595
5596
5597
5598
5599
5600
5601
5602
5603
5604
5605
5606
5607
5608
5609
5610
5611
5612
5613
5614
5615
5616
5617
5618
5619
5620
5621
5622
5623
5624
5625
5626
5627
5628
5629
5630
5631
5632
5633
5634
5635
5636
5637
5638
5639
5640
5641
5642
5643
5644
5645
5646
5647
5648
5649
5650
5651
5652
5653
5654
5655
5656
5657
5658
5659
5660
5661
5662
5663
5664
5665
5666
5667
5668
5669
5670
5671
5672
5673
5674
5675
5676
5677
5678
5679
5680
5681
5682
5683
5684
5685
5686
5687
5688
5689
5690
5691
5692
5693
5694
5695
5696
5697
5698
5699
5700
5701
5702
5703
5704
5705
5706
5707
5708
5709
5710
5711
5712
5713
5714
5715
5716
5717
5718
5719
5720
5721
5722
5723
5724
5725
5726
5727
5728
5729
5730
5731
5732
5733
5734
5735
5736
5737
5738
5739
5740
5741
5742
5743
5744
5745
5746
5747
5748
5749
5750
5751
5752
5753
5754
5755
5756
5757
5758
5759
5760
5761
5762
5763
5764
5765
5766
5767
5768
5769
5770
5771
5772
5773
5774
5775
5776
5777
5778
5779
5780
5781
5782
5783
5784
5785
5786
5787
5788
5789
5790
5791
5792
5793
5794
5795
5796
5797
5798
5799
5800
5801
5802
5803
5804
5805
5806
5807
5808
5809
5810
5811
5812
5813
5814
5815
5816
5817
5818
5819
5820
5821
5822
5823
5824
5825
5826
5827
5828
5829
5830
5831
5832
5833
5834
5835
5836
5837
5838
5839
5840
5841
5842
5843
5844
5845
5846
5847
5848
5849
5850
5851
5852
5853
5854
5855
5856
5857
5858
5859
5860
5861
5862
5863
5864
5865
5866
5867
5868
5869
5870
5871
5872
5873
5874
5875
5876
5877
5878
5879
5880
5881
5882
5883
5884
5885
5886
5887
5888
5889
5890
5891
5892
5893
5894
5895
5896
5897
5898
5899
5900
5901
5902
5903
5904
5905
5906
5907
5908
5909
5910
5911
5912
5913
5914
5915
5916
5917
5918
5919
5920
5921
5922
5923
5924
5925
5926
5927
5928
5929
5930
5931
5932
5933
5934
5935
5936
5937
5938
5939
5940
5941
5942
5943
5944
5945
5946
5947
5948
5949
5950
5951
5952
5953
5954
5955
5956
5957
5958
5959
5960
5961
5962
5963
5964
5965
5966
5967
5968
5969
5970
5971
5972
5973
5974
5975
5976
5977
5978
5979
5980
5981
5982
5983
5984
5985
5986
5987
5988
5989
5990
5991
5992
5993
5994
5995
5996
5997
5998
5999
6000
6001
6002
6003
6004
6005
6006
6007
6008
6009
6010
6011
6012
6013
6014
6015
6016
6017
6018
6019
6020
6021
6022
6023
6024
6025
6026
6027
6028
6029
6030
6031
6032
6033
6034
6035
6036
6037
6038
6039
6040
6041
6042
6043
6044
6045
6046
6047
6048
6049
6050
6051
6052
6053
6054
6055
6056
6057
6058
6059
6060
6061
6062
6063
6064
6065
6066
6067
6068
6069
6070
6071
6072
6073
6074
6075
6076
6077
6078
6079
6080
6081
6082
6083
6084
6085
6086
6087
6088
6089
6090
6091
6092
6093
6094
6095
6096
6097
6098
6099
6100
6101
6102
6103
6104
6105
6106
6107
6108
6109
6110
6111
6112
6113
6114
6115
6116
6117
6118
6119
6120
6121
6122
6123
6124
6125
6126
6127
6128
6129
6130
6131
6132
6133
6134
6135
6136
6137
6138
6139
6140
6141
6142
6143
6144
6145
6146
6147
6148
6149
6150
6151
6152
6153
6154
6155
6156
6157
6158
6159
6160
6161
6162
6163
6164
6165
6166
6167
6168
6169
6170
6171
6172
6173
6174
6175
6176
6177
6178
6179
6180
6181
6182
6183
6184
6185
6186
6187
6188
6189
6190
6191
6192
6193
6194
6195
6196
6197
6198
6199
6200
6201
6202
6203
6204
6205
6206
6207
6208
6209
6210
6211
6212
6213
6214
6215
6216
6217
6218
6219
6220
6221
6222
6223
6224
6225
6226
6227
6228
6229
6230
6231
6232
6233
6234
6235
6236
6237
6238
6239
6240
6241
6242
6243
6244
6245
6246
6247
6248
6249
6250
6251
6252
6253
6254
6255
6256
6257
6258
6259
6260
6261
6262
6263
6264
6265
6266
6267
6268
6269
6270
6271
6272
6273
6274
6275
6276
6277
6278
6279
6280
6281
6282
6283
6284
6285
6286
6287
6288
6289
6290
6291
6292
6293
6294
6295
6296
6297
6298
6299
6300
6301
6302
6303
6304
6305
6306
6307
6308
6309
6310
6311
6312
6313
6314
6315
6316
6317
6318
6319
6320
6321
6322
6323
6324
6325
6326
6327
6328
6329
6330
6331
6332
6333
6334
6335
6336
6337
6338
6339
6340
6341
6342
6343
6344
6345
6346
6347
6348
6349
6350
6351
6352
6353
6354
6355
6356
6357
6358
6359
6360
6361
6362
6363
6364
6365
6366
6367
6368
6369
6370
6371
6372
6373
6374
6375
6376
6377
6378
6379
6380
6381
6382
6383
6384
6385
6386
6387
6388
6389
6390
6391
6392
6393
6394
6395
6396
6397
6398
6399
6400
6401
6402
6403
6404
6405
6406
6407
6408
6409
6410
6411
6412
6413
6414
6415
6416
6417
6418
6419
6420
6421
6422
6423
6424
6425
6426
6427
6428
6429
6430
6431
6432
6433
6434
6435
6436
6437
6438
6439
6440
6441
6442
6443
6444
6445
6446
6447
6448
6449
6450
6451
6452
6453
6454
6455
6456
6457
6458
6459
6460
6461
6462
6463
6464
6465
6466
6467
6468
6469
6470
6471
6472
6473
6474
6475
6476
6477
6478
6479
6480
6481
6482
6483
6484
6485
6486
6487
6488
6489
6490
6491
6492
6493
6494
6495
6496
6497
6498
6499
6500
6501
6502
6503
6504
6505
6506
6507
6508
6509
6510
6511
6512
6513
6514
6515
6516
6517
6518
6519
6520
6521
6522
6523
6524
6525
6526
6527
6528
6529
6530
6531
6532
6533
6534
6535
6536
6537
6538
6539
6540
6541
6542
6543
6544
6545
6546
6547
6548
6549
6550
6551
6552
6553
6554
6555
6556
6557
6558
6559
6560
6561
6562
6563
6564
6565
6566
6567
6568
6569
6570
6571
6572
6573
6574
6575
6576
6577
6578
6579
6580
6581
6582
6583
6584
6585
6586
6587
6588
6589
6590
6591
6592
6593
6594
6595
6596
6597
6598
6599
6600
6601
6602
6603
6604
6605
6606
6607
6608
6609
6610
6611
6612
6613
6614
6615
6616
6617
6618
6619
6620
6621
6622
6623
6624
6625
6626
6627
6628
6629
6630
6631
6632
6633
6634
6635
6636
6637
6638
6639
6640
6641
6642
6643
6644
6645
6646
6647
6648
6649
6650
6651
6652
6653
6654
6655
6656
6657
6658
6659
6660
6661
6662
6663
6664
6665
6666
6667
6668
6669
6670
6671
6672
6673
6674
6675
6676
6677
6678
6679
6680
6681
6682
6683
6684
6685
6686
6687
6688
6689
6690
6691
6692
6693
6694
6695
6696
6697
6698
6699
6700
6701
6702
6703
6704
6705
6706
6707
6708
6709
6710
6711
6712
6713
6714
6715
6716
6717
6718
6719
6720
6721
6722
6723
6724
6725
6726
6727
6728
6729
6730
6731
6732
6733
6734
6735
6736
6737
6738
6739
6740
6741
6742
6743
6744
6745
6746
6747
6748
6749
6750
6751
6752
6753
6754
6755
6756
6757
6758
6759
6760
6761
6762
6763
6764
6765
6766
6767
6768
6769
6770
6771
6772
6773
6774
6775
6776
6777
6778
6779
6780
6781
6782
6783
6784
6785
6786
6787
6788
6789
6790
6791
6792
6793
6794
6795
6796
6797
6798
6799
6800
6801
6802
6803
6804
6805
6806
6807
6808
6809
6810
6811
6812
6813
6814
6815
6816
6817
6818
6819
6820
6821
6822
6823
6824
6825
6826
6827
6828
6829
6830
6831
6832
6833
6834
6835
6836
6837
6838
6839
6840
6841
6842
6843
6844
6845
6846
6847
6848
6849
6850
6851
6852
6853
6854
6855
6856
6857
6858
6859
6860
6861
6862
6863
6864
6865
6866
6867
6868
6869
6870
6871
6872
6873
6874
6875
6876
6877
6878
6879
6880
6881
6882
6883
6884
6885
6886
6887
6888
6889
6890
6891
6892
6893
6894
6895
6896
6897
6898
6899
6900
6901
6902
6903
6904
6905
6906
6907
6908
6909
6910
6911
6912
6913
6914
6915
6916
6917
6918
6919
6920
6921
6922
6923
6924
6925
6926
6927
6928
6929
6930
6931
6932
6933
6934
6935
6936
6937
6938
6939
6940
6941
6942
6943
6944
6945
6946
6947
6948
6949
6950
6951
6952
6953
6954
6955
6956
6957
6958
6959
6960
6961
6962
6963
6964
6965
6966
6967
6968
6969
6970
6971
6972
6973
6974
6975
6976
6977
6978
6979
6980
6981
6982
6983
6984
6985
6986
6987
6988
6989
6990
6991
6992
6993
6994
6995
6996
6997
6998
6999
7000
7001
7002
7003
7004
7005
7006
7007
7008
7009
7010
7011
7012
7013
7014
7015
7016
7017
7018
7019
7020
7021
7022
7023
7024
7025
7026
7027
7028
7029
7030
7031
7032
7033
7034
7035
7036
7037
7038
7039
7040
7041
7042
7043
7044
7045
7046
7047
7048
7049
7050
7051
7052
7053
7054
7055
7056
7057
7058
7059
7060
7061
7062
7063
7064
7065
7066
7067
7068
7069
7070
7071
7072
7073
7074
7075
7076
7077
7078
7079
7080
7081
7082
7083
7084
7085
7086
7087
7088
7089
7090
7091
7092
7093
7094
7095
7096
7097
7098
7099
7100
7101
7102
7103
7104
7105
7106
7107
7108
7109
7110
7111
7112
7113
7114
7115
7116
7117
7118
7119
7120
7121
7122
7123
7124
7125
7126
7127
7128
7129
7130
7131
7132
7133
7134
7135
7136
7137
7138
7139
7140
7141
7142
7143
7144
7145
7146
7147
7148
7149
7150
7151
7152
7153
7154
7155
7156
7157
7158
7159
7160
7161
7162
7163
7164
7165
7166
7167
7168
7169
7170
7171
7172
7173
7174
7175
7176
7177
7178
7179
7180
7181
7182
7183
7184
7185
7186
7187
7188
7189
7190
7191
7192
7193
7194
7195
7196
7197
7198
7199
7200
7201
7202
7203
7204
7205
7206
7207
7208
7209
7210
7211
7212
7213
7214
7215
7216
7217
7218
7219
7220
7221
7222
7223
7224
7225
7226
7227
7228
7229
7230
7231
7232
7233
7234
7235
7236
7237
7238
7239
7240
7241
7242
7243
7244
7245
7246
7247
7248
7249
7250
7251
7252
7253
7254
7255
7256
7257
7258
7259
7260
7261
7262
7263
7264
7265
7266
7267
7268
7269
7270
7271
7272
7273
7274
7275
7276
7277
7278
7279
7280
7281
7282
7283
7284
7285
7286
7287
7288
7289
7290
7291
7292
7293
7294
7295
7296
7297
7298
7299
7300
7301
7302
7303
7304
7305
7306
7307
7308
7309
7310
7311
7312
7313
7314
7315
7316
7317
7318
7319
7320
7321
7322
7323
7324
7325
7326
7327
7328
7329
7330
7331
7332
7333
7334
7335
7336
7337
7338
7339
7340
7341
7342
7343
7344
7345
|
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
<html>
<head>
<title>SWIG and Python</title>
<link rel="stylesheet" type="text/css" href="style.css">
<meta http-equiv="content-type" content="text/html; charset=UTF-8">
</head>
<body bgcolor="#ffffff">
<H1><a name="Python">33 SWIG and Python</a></H1>
<!-- INDEX -->
<div class="sectiontoc">
<ul>
<li><a href="#Python_nn2">Overview</a>
<li><a href="#Python_nn3">Preliminaries</a>
<ul>
<li><a href="#Python_nn4">Running SWIG</a>
<li><a href="#Python_nn6">Using distutils</a>
<li><a href="#Python_nn7">Hand compiling a dynamic module</a>
<li><a href="#Python_nn8">Static linking</a>
<li><a href="#Python_nn9">Using your module</a>
<li><a href="#Python_nn10">Compilation of C++ extensions</a>
<li><a href="#Python_nn11">Compiling for 64-bit platforms</a>
<li><a href="#Python_nn12">Building Python extensions under Windows</a>
<li><a href="#Python_commandline">Additional Python commandline options</a>
</ul>
<li><a href="#Python_nn13">A tour of basic C/C++ wrapping</a>
<ul>
<li><a href="#Python_nn14">Modules</a>
<li><a href="#Python_nn15">Functions</a>
<li><a href="#Python_nn16">Global variables</a>
<li><a href="#Python_nn17">Constants and enums</a>
<li><a href="#Python_nn18">Pointers</a>
<li><a href="#Python_nn19">Structures</a>
<li><a href="#Python_nn20">C++ classes</a>
<li><a href="#Python_nn21">C++ inheritance</a>
<li><a href="#Python_nn22">Pointers, references, values, and arrays</a>
<li><a href="#Python_nn23">C++ overloaded functions</a>
<li><a href="#Python_nn24">C++ operators</a>
<li><a href="#Python_nn25">C++ namespaces</a>
<li><a href="#Python_nn26">C++ templates</a>
<li><a href="#Python_nn27">C++ Smart Pointers</a>
<ul>
<li><a href="#Python_smart_pointers_shared_ptr">The shared_ptr Smart Pointer</a>
<li><a href="#Python_smart_pointers_generic">Generic Smart Pointers</a>
</ul>
<li><a href="#Python_nn27a">C++ reference counted objects</a>
</ul>
<li><a href="#Python_nn28">Further details on the Python class interface</a>
<ul>
<li><a href="#Python_nn29">Proxy classes</a>
<li><a href="#Python_builtin_types">Built-in Types</a>
<ul>
<li><a href="#Python_builtin_limitations">Limitations</a>
<li><a href="#Python_builtin_overloads">Operator overloads and slots -- use them!</a>
</ul>
<li><a href="#Python_nn30">Memory management</a>
</ul>
<li><a href="#Python_directors">Cross language polymorphism</a>
<ul>
<li><a href="#Python_nn33">Enabling directors</a>
<li><a href="#Python_nn34">Director classes</a>
<li><a href="#Python_nn35">Ownership and object destruction</a>
<li><a href="#Python_nn36">Exception unrolling</a>
<li><a href="#Python_nn37">Overhead and code bloat</a>
<li><a href="#Python_nn38">Typemaps</a>
<li><a href="#Python_nn39">Miscellaneous</a>
</ul>
<li><a href="#Python_nn40">Common customization features</a>
<ul>
<li><a href="#Python_nn41">C/C++ helper functions</a>
<li><a href="#Python_nn42">Adding additional Python code</a>
<li><a href="#Python_nn43">Class extension with %extend</a>
<li><a href="#Python_nn44">Exception handling with %exception</a>
<li><a href="#Python_optimization">Optimization options</a>
<ul>
<li><a href="#Python_fastproxy">-fastproxy</a>
</ul>
</ul>
<li><a href="#Python_nn45">Tips and techniques</a>
<ul>
<li><a href="#Python_nn46">Input and output parameters</a>
<li><a href="#Python_nn47">Simple pointers</a>
<li><a href="#Python_nn48">Unbounded C Arrays</a>
<li><a href="#Python_nn49">String handling</a>
<li><a href="#Python_default_args">Default arguments</a>
</ul>
<li><a href="#Python_nn53">Typemaps</a>
<ul>
<li><a href="#Python_nn54">What is a typemap?</a>
<li><a href="#Python_nn55">Python typemaps</a>
<li><a href="#Python_nn56">Typemap variables</a>
<li><a href="#Python_nn57">Useful Python Functions</a>
</ul>
<li><a href="#Python_nn58">Typemap Examples</a>
<ul>
<li><a href="#Python_nn59">Converting Python list to a char ** </a>
<li><a href="#Python_nn60">Expanding a Python object into multiple arguments</a>
<li><a href="#Python_nn61">Using typemaps to return arguments</a>
<li><a href="#Python_nn62">Mapping Python tuples into small arrays</a>
<li><a href="#Python_nn63">Mapping sequences to C arrays</a>
<li><a href="#Python_nn64">Pointer handling</a>
<li><a href="#Python_memory_management_member_variables">Memory management when returning references to member variables</a>
</ul>
<li><a href="#Python_nn65">Docstring Features</a>
<ul>
<li><a href="#Python_nn66">Module docstring</a>
<li><a href="#Python_nn67">%feature("autodoc")</a>
<ul>
<li><a href="#Python_nn68">%feature("autodoc", "0")</a>
<li><a href="#Python_nn69">%feature("autodoc", "1")</a>
<li><a href="#Python_autodoc2">%feature("autodoc", "2")</a>
<li><a href="#Python_autodoc3">%feature("autodoc", "3")</a>
<li><a href="#Python_nn70">%feature("autodoc", "docstring")</a>
</ul>
<li><a href="#Python_nn71">%feature("docstring")</a>
</ul>
<li><a href="#Python_nn72">Python Packages</a>
<ul>
<li><a href="#Python_modulepackage">Setting the Python package</a>
<li><a href="#Python_absrelimports">Absolute and relative imports</a>
<li><a href="#Python_absimport">Enforcing absolute import semantics</a>
<li><a href="#Python_importfrominit">Importing from __init__.py</a>
<li><a href="#Python_implicit_namespace_packages">Implicit namespace packages</a>
<li><a href="#Python_package_search">Location of modules</a>
<ul>
<li><a href="#Python_package_search_both_package_modules">Both modules in the same package</a>
<li><a href="#Python_package_search_both_global_modules">Both modules are global</a>
<li><a href="#Python_package_search_wrapper_split">Split modules custom configuration</a>
<li><a href="#Python_custom_module_import">More on customizing the module import code</a>
<li><a href="#Python_package_search_static">Statically linked C modules</a>
</ul>
</ul>
<li><a href="#Python_python3support">Python 3 Support</a>
<ul>
<li><a href="#Python_nn74">Function annotation</a>
<li><a href="#Python_nn75">Buffer interface</a>
<li><a href="#Python_nn76">Abstract base classes</a>
<li><a href="#Python_nn77">Byte string output conversion</a>
<li><a href="#Python_2_unicode">Python 2 Unicode</a>
</ul>
<li><a href="#Python_multithreaded">Support for Multithreaded Applications</a>
<ul>
<li><a href="#Python_thread_UI">UI for Enabling Multithreading Support</a>
<li><a href="#Python_thread_performance">Multithread Performance</a>
</ul>
</ul>
</div>
<!-- INDEX -->
<p>
<b>Caution: This chapter is under repair!</b>
</p>
<p>
This chapter describes SWIG's support of Python. SWIG is compatible
with all recent Python versions (Python 2.7 and Python >= 3.2). If you
still need to generate bindings which work with older versions of Python,
you'll have to use SWIG 3.0.x.
</p>
<p>
This chapter covers most SWIG features, but certain low-level details
are covered in less depth than in earlier chapters. At the
very least, make sure you read the "<a href="SWIG.html#SWIG">SWIG
Basics</a>" chapter.
</p>
<H2><a name="Python_nn2">33.1 Overview</a></H2>
<p>
To build Python extension modules, SWIG uses a layered approach in which
parts of the extension module are defined in C and other parts are
defined in Python. The C layer contains low-level wrappers whereas Python code
is used to define high-level features.
</p>
<p>
This layered approach recognizes the fact that certain aspects of
extension building are better accomplished in each language (instead
of trying to do everything in C or C++). Furthermore, by generating code in both
languages, you get a lot more flexibility since you can enhance the extension
module with support code in either language.
</p>
<p>
In describing the Python interface, this chapter starts by covering the
basics of configuration, compiling, and installing Python modules.
Next, the Python interface to common C and C++ programming features is
described. Advanced customization features such as typemaps are then
described followed by a discussion of low-level implementation
details.
</p>
<H2><a name="Python_nn3">33.2 Preliminaries</a></H2>
<H3><a name="Python_nn4">33.2.1 Running SWIG</a></H3>
<p>
Suppose that you defined a SWIG module such as the following:
</p>
<div class="code">
<pre>
/* File: example.i */
%module example
%{
#define SWIG_FILE_WITH_INIT
#include "example.h"
%}
int fact(int n);
</pre>
</div>
<p>
The <tt>#define SWIG_FILE_WITH_INIT</tt> line inserts a macro that specifies that the
resulting C file should be built as a Python extension, inserting the module
<tt>init</tt> code. This <tt>.i</tt> file wraps the following simple C file:
</p>
<div class="code">
<pre>
/* File: example.c */
#include "example.h"
int fact(int n) {
if (n < 0) { /* This should probably return an error, but this is simpler */
return 0;
}
if (n == 0) {
return 1;
} else {
/* testing for overflow would be a good idea here */
return n * fact(n-1);
}
}
</pre>
</div>
<p>
With the header file:
</p>
<div class="code">
<pre>
/* File: example.h */
int fact(int n);
</pre>
</div>
<p>
To build a Python module, run SWIG using the <tt>-python</tt> option:
</p>
<div class="shell"><pre>
$ swig -python example.i
</pre></div>
<p>
If building a C++ extension, add the <tt>-c++</tt> option:
</p>
<div class="shell"><pre>
$ swig -c++ -python example.i
</pre></div>
<p>
This creates two different files; a C/C++ source file <tt>example_wrap.c</tt> or
<tt>example_wrap.cxx</tt> and a Python source file <tt>example.py</tt>. The generated C
source file contains the low-level wrappers that need to be compiled and linked with the
rest of your C/C++ application to create an extension module. The Python source file
contains high-level support code. This is the file that you will import to use the module.
</p>
<p>
The name of the wrapper file is derived from the name of the input file. For example, if the
input file is <tt>example.i</tt>, the name of the wrapper file is <tt>example_wrap.c</tt>.
To change this, you can use the <tt>-o</tt> option. The name of the Python file is derived
from the module name specified with <tt>%module</tt>. If the module name is <tt>example</tt>,
then a file <tt>example.py</tt> is created.
</p>
<p>
The following sections have further practical examples and details on
how you might go about compiling and using the generated files.
</p>
<H3><a name="Python_nn6">33.2.2 Using distutils</a></H3>
<p>
The preferred approach to building an extension module for Python is to compile it with
distutils, which comes with all recent versions of Python
(<a href="https://docs.python.org/3/library/distutils.html">Distutils Docs</a>).
</p>
<p>
Distutils takes care of making sure that your extension is built with all the correct
flags, headers, etc. for the version of Python it is run with. Distutils will compile your
extension into a shared object file or DLL (<tt>.so</tt> on Linux, <tt>.pyd</tt> on
Windows, etc). In addition, distutils can handle installing your package into
site-packages, if that is desired. A configuration file (conventionally called: <tt>setup.py</tt>)
describes the extension (and related Python modules). The distutils will
then generate all the right compiler directives to build it for you.
</p>
<p>
Here is a sample <tt>setup.py</tt> file for the above example:
</p>
<div class="code">
<pre>
#!/usr/bin/env python
"""
setup.py file for SWIG example
"""
from distutils.core import setup, Extension
example_module = Extension('_example',
sources=['example_wrap.c', 'example.c'],
)
setup (name = 'example',
version = '0.1',
author = "SWIG Docs",
description = """Simple swig example from docs""",
ext_modules = [example_module],
py_modules = ["example"],
)
</pre>
</div>
<p>
In this example, the line: <tt>example_module = Extension(....)</tt> creates an Extension
module object, defining the name as <tt>_example</tt>, and using the source code files:
<tt>example_wrap.c</tt>, generated by swig, and <tt>example.c</tt>, your original c
source. The swig (and other Python extension modules) tradition is for the compiled
extension to have the name of the Python portion, prefixed by an underscore. If the name
of your Python module is "<tt>example.py</tt>", then the name of the corresponding object file
will be"<tt>_example.so</tt>"
</p>
<p>
The <tt>setup</tt> call then sets up distutils to build your package, defining
some meta data, and passing in your extension module object.
Once this is saved as <tt>setup.py</tt>, you can build your extension with these commands:
</p>
<div class="shell"><pre>
$ swig -python example.i
$ python setup.py build_ext --inplace
</pre></div>
<p>
And a .so, or .pyd or... will be created for you. It will build a version that matches the
Python that you run the command with. Taking apart the command line:
</p>
<ul>
<li> <tt>python</tt> -- the version of Python you want to build for
<li> <tt>setup.py</tt> -- the name of your setup script (it can be called anything, but
setup.py is the tradition)
<li> <tt>build_ext</tt> -- telling distutils to build extensions
<li> <tt>--inplace</tt> -- this tells distutils to put the extension lib in the current dir.
Otherwise, it will put it inside a build hierarchy, and you'd have to move it to use it.
</ul>
<p>
The distutils have many other features, consult the Python distutils docs for details.
</p>
<p>
This same approach works on all platforms if the appropriate compiler is installed. (it
can even build extensions to the standard Windows Python using MingGW)
</p>
<H3><a name="Python_nn7">33.2.3 Hand compiling a dynamic module</a></H3>
<p>
While the preferred approach to building an extension module is to use the distutils, some
people like to integrate building extensions with a larger build system, and thus may wish
to compile their modules without the distutils. To do this, you need to compile your
program using commands like this (shown for Linux):
</p>
<div class="shell"><pre>
$ swig -python example.i
$ gcc -O2 -fPIC -c example.c
$ gcc -O2 -fPIC -c example_wrap.c -I/usr/local/include/python2.5
$ gcc -shared example.o example_wrap.o -o _example.so
</pre></div>
<p>
The exact commands for doing this vary from platform to platform.
However, SWIG tries to guess the right options when it is installed. Therefore,
you may want to start with one of the examples in the <tt>SWIG/Examples/python</tt>
directory. If that doesn't work, you will need to read the man-pages for
your compiler and linker to get the right set of options. You might also
check the <a href="https://github.com/swig/swig/wiki">SWIG Wiki</a> for
additional information.
</p>
<p>
When linking the module, <b>the name of the output file has to match the name
of the module prefixed by an underscore</b>. If the name of your module is "<tt>example</tt>", then the
name of the corresponding object file should be
"<tt>_example.so</tt>" or "<tt>_examplemodule.so</tt>".
The name of the module is specified using the <tt>%module</tt> directive or the
<tt>-module</tt> command line option.
</p>
<p>
<b>Compatibility Note:</b> In SWIG-1.3.13 and earlier releases, module
names did not include the leading underscore. This is because modules
were normally created as C-only extensions without the extra Python
support file (instead, creating Python code was supported as an optional
feature). This has been changed in SWIG-1.3.14 and is consistent with
other Python extension modules. For example, the <tt>socket</tt>
module actually consists of two files; <tt>socket.py</tt> and
<tt>_socket.so</tt>. Many other built-in Python modules follow a similar convention.
</p>
<H3><a name="Python_nn8">33.2.4 Static linking</a></H3>
<p>
An alternative approach to dynamic linking is to rebuild the Python
interpreter with your extension module added to it. In the past,
this approach was sometimes necessary due to limitations in dynamic loading
support on certain machines. However, the situation has improved greatly
over the last few years and you should not consider this approach
unless there is really no other option.
</p>
<p>
The usual procedure for adding a new module to Python involves finding
the Python source, adding an entry to the <tt>Modules/Setup</tt> file,
and rebuilding the interpreter using the Python Makefile. However,
newer Python versions have changed the build process. You may need to edit
the 'setup.py' file in the Python distribution instead.
</p>
<p>
In earlier versions of SWIG, the <tt>embed.i</tt> library file could be used to
rebuild the interpreter. For example:
</p>
<div class="code"><pre>
%module example
%inline %{
extern int fact(int);
extern int mod(int, int);
extern double My_variable;
%}
%include "embed.i" // Include code for a static version of Python
</pre></div>
<p>
The <tt>embed.i</tt> library file includes supporting code that
contains everything needed to rebuild Python. To rebuild the interpreter,
you simply do something like this:
</p>
<div class="shell"><pre>
$ swig -python -lembed.i example.i
$ gcc example.c example_wrap.c \
-Xlinker -export-dynamic \
-DHAVE_CONFIG_H -I/usr/include/python2.7 \
-I/usr/lib/python2.7/config-x86_64-linux-gnu \
-I/usr/lib/python2.7/config \
-L/usr/lib/python2.7/config -lpython2.7 -lm -ldl \
-o mypython
</pre></div>
<p>
You will need to supply the same libraries that were used to build Python the first
time. This may include system libraries such as <tt>-lsocket</tt>, <tt>-lnsl</tt>,
and <tt>-lpthread</tt>. Assuming this actually works, the new version of Python
should be identical to the default version except that your extension module will be
a built-in part of the interpreter.
</p>
<p>
<b>Comment:</b> In practice, you should probably try to avoid static
linking if possible. Some programmers may be inclined
to use static linking in the interest of getting better performance.
However, the performance gained by static linking tends to be rather
minimal in most situations (and quite frankly not worth the extra
hassle in the opinion of this author).
</p>
<p>
<b>Compatibility note:</b> The <tt>embed.i</tt> library file is
deprecated and has not been actively maintained for many years. Even though it
appears to "work" with Python 2.7, no future support is guaranteed.
If using static linking, you might want to rely on a different approach
(perhaps using distutils).
</p>
<H3><a name="Python_nn9">33.2.5 Using your module</a></H3>
<p>
To use your module, simply use the Python <tt>import</tt> statement. If
all goes well, you will be able to run this:
</p>
<div class="targetlang"><pre>
$ python
>>> import example
>>> example.fact(4)
24
>>>
</pre></div>
<p>
A common error received by first-time users is the following:
</p>
<div class="targetlang">
<pre>
>>> import example
Traceback (most recent call last):
File "<stdin>", line 1, in ?
File "example.py", line 2, in ?
import _example
ImportError: No module named _example
</pre>
</div>
<p>
If you get this message, it means that you either forgot to compile the wrapper
code into an extension module or you didn't give the extension module the right
name. Make sure that you compiled the wrappers into a module called <tt>_example.so</tt>. And
don't forget the leading underscore (_).
</p>
<p>
Another possible error is the following:
</p>
<div class="targetlang">
<pre>
>>> import example
Traceback (most recent call last):
File "<stdin>", line 1, in ?
ImportError: dynamic module does not define init function (init_example)
>>>
</pre>
</div>
<p>
This error is almost always caused when a bad name is given to the shared object file.
For example, if you created a file <tt>example.so</tt> instead of <tt>_example.so</tt> you would
get this error. Alternatively, this error could arise if the name of the module is
inconsistent with the module name supplied with the <tt>%module</tt> directive.
Double-check the interface to make sure the module name and the shared object
filename match. Another possible cause of this error is forgetting to link the SWIG-generated
wrapper code with the rest of your application when creating the extension module.
</p>
<p>
Another common error is something similar to the following:
</p>
<div class="targetlang">
<pre>
Traceback (most recent call last):
File "example.py", line 3, in ?
import example
ImportError: ./_example.so: undefined symbol: fact
</pre>
</div>
<p>
This error usually indicates that you forgot to include some object
files or libraries in the linking of the shared library file. Make
sure you compile both the SWIG wrapper file and your original program
into a shared library file. Make sure you pass all of the required libraries
to the linker.
</p>
<p>
Sometimes unresolved symbols occur because a wrapper has been created
for a function that doesn't actually exist in a library. This usually
occurs when a header file includes a declaration for a function that
was never actually implemented or it was removed from a library
without updating the header file. To fix this, you can either edit
the SWIG input file to remove the offending declaration or you can use
the <tt>%ignore</tt> directive to ignore the declaration.
</p>
<p>
Finally, suppose that your extension module is linked with another library like this:
</p>
<div class="shell">
<pre>
$ gcc -shared example.o example_wrap.o -L/home/beazley/projects/lib <b>-lfoo</b> \
-o _example.so
</pre>
</div>
<p>
If the <tt>foo</tt> library is compiled as a shared library, you might encounter the following
problem when you try to use your module:
</p>
<div class="targetlang">
<pre>
>>> import example
Traceback (most recent call last):
File "<stdin>", line 1, in ?
ImportError: libfoo.so: cannot open shared object file: No such file or directory
>>>
</pre>
</div>
<p>
This error is generated because the dynamic linker can't locate the
<tt>libfoo.so</tt> library. When shared libraries are loaded, the
system normally only checks a few standard locations such as
<tt>/usr/lib</tt> and <tt>/usr/local/lib</tt>. To fix this problem,
there are several things you can do. First, you can recompile your extension
module with extra path information. For example, on Linux you can do this:
</p>
<div class="shell">
<pre>
$ gcc -shared example.o example_wrap.o -L/home/beazley/projects/lib -lfoo \
<b>-Xlinker -rpath /home/beazley/projects/lib </b> \
-o _example.so
</pre>
</div>
<p>
Alternatively, you can set the <tt>LD_LIBRARY_PATH</tt> environment variable to
include the directory with your shared libraries.
If setting <tt>LD_LIBRARY_PATH</tt>, be aware that setting this variable can introduce
a noticeable performance impact on all other applications that you run.
To set it only for Python, you might want to do this instead:
</p>
<div class="shell">
<pre>
$ env LD_LIBRARY_PATH=/home/beazley/projects/lib python
</pre>
</div>
<p>
Finally, you can use a command such as <tt>ldconfig</tt> (Linux) or
<tt>crle</tt> (Solaris) to add additional search paths to the default
system configuration (this requires root access and you will need to
read the man pages).
</p>
<H3><a name="Python_nn10">33.2.6 Compilation of C++ extensions</a></H3>
<p>
Compilation of C++ extensions has traditionally been a tricky problem.
Since the Python interpreter is written in C, you need to take steps to
make sure C++ is properly initialized and that modules are compiled
correctly. This should be a non-issue if you're using distutils, as
it takes care of all that for you. The following is included for
historical reasons, and in case you need to compile on your own.
</p>
<p>
On most machines, C++ extension modules should be linked using the C++
compiler. For example:
</p>
<div class="shell"><pre>
$ swig -c++ -python example.i
$ g++ -O2 -fPIC -c example.cxx
$ g++ -O2 -fPIC -c example_wrap.cxx -I/usr/local/include/python2.5
$ g++ -shared example.o example_wrap.o -o _example.so
</pre></div>
<p>
The -fPIC option tells GCC to generate position-independent code (PIC)
which is required for most architectures (it's not vital on x86, but
still a good idea as it allows code pages from the library to be shared between
processes). Other compilers may need a different option specified instead of
-fPIC.
</p>
<p>
In addition to this, you may need to include additional library
files to make it work. For example, if you are using the Sun C++ compiler on
Solaris, you often need to add an extra library <tt>-lCrun</tt> like this:
</p>
<div class="shell"><pre>
$ swig -c++ -python example.i
$ CC -c example.cxx
$ CC -c example_wrap.cxx -I/usr/local/include/python2.5
$ CC -G example.o example_wrap.o -L/opt/SUNWspro/lib -o _example.so -lCrun
</pre></div>
<p>
Of course, the extra libraries to use are completely non-portable---you will
probably need to do some experimentation.
</p>
<p>
Sometimes people have suggested that it is necessary to relink the
Python interpreter using the C++ compiler to make C++ extension modules work.
In the experience of this author, this has never actually appeared to be
necessary. Relinking the interpreter with C++ really only includes the
special run-time libraries described above---as long as you link your extension
modules with these libraries, it should not be necessary to rebuild Python.
</p>
<p>
If you aren't entirely sure about the linking of a C++ extension, you
might look at an existing C++ program. On many Unix machines, the
<tt>ldd</tt> command will list library dependencies. This should give
you some clues about what you might have to include when you link your
extension module. For example:
</p>
<div class="shell">
<pre>
$ ldd swig
libstdc++-libc6.1-1.so.2 => /usr/lib/libstdc++-libc6.1-1.so.2 (0x40019000)
libm.so.6 => /lib/libm.so.6 (0x4005b000)
libc.so.6 => /lib/libc.so.6 (0x40077000)
/lib/ld-linux.so.2 => /lib/ld-linux.so.2 (0x40000000)
</pre>
</div>
<p>
As a final complication, a major weakness of C++ is that it does not
define any sort of standard for binary linking of libraries. This
means that C++ code compiled by different compilers will not link
together properly as libraries nor is the memory layout of classes and
data structures implemented in any kind of portable manner. In a
monolithic C++ program, this problem may be unnoticed. However, in Python, it
is possible for different extension modules to be compiled with
different C++ compilers. As long as these modules are self-contained,
this probably won't matter. However, if these modules start sharing data,
you will need to take steps to avoid segmentation faults and other
erratic program behavior. If working with lots of software components, you
might want to investigate using a more formal standard such as COM.
</p>
<H3><a name="Python_nn11">33.2.7 Compiling for 64-bit platforms</a></H3>
<p>
On platforms that support 64-bit applications (Solaris, Irix, etc.),
special care is required when building extension modules. On these
machines, 64-bit applications are compiled and linked using a different
set of compiler/linker options. In addition, it is not generally possible to mix
32-bit and 64-bit code together in the same application.
</p>
<p>
To utilize 64-bits, the Python executable will need to be recompiled
as a 64-bit application. In addition, all libraries, wrapper code,
and every other part of your application will need to be compiled for
64-bits. If you plan to use other third-party extension modules, they
will also have to be recompiled as 64-bit extensions.
</p>
<p>
If you are wrapping commercial software for which you have no source
code, you will be forced to use the same linking standard as used by
that software. This may prevent the use of 64-bit extensions. It may
also introduce problems on platforms that support more than one
linking standard (e.g., -o32 and -n32 on Irix).
</p>
<p> On the Linux x86_64 platform (Opteron or EM64T), besides of the
required compiler option -fPIC discussed above, you will need to be
careful about the libraries you link with or the library path you
use. In general, a Linux distribution will have two set of libraries,
one for native x86_64 programs (under /usr/lib64), and another for 32
bits compatibility (under /usr/lib). Also, the compiler options -m32
and -m64 allow you to choose the desired binary format for your Python
extension.
</p>
<H3><a name="Python_nn12">33.2.8 Building Python extensions under Windows</a></H3>
<p>
Building a SWIG extension to Python under Windows is roughly similar to
the process used with Unix. Using the distutils, it is essentially
identical. If you have the same version of the MS compiler that Python
was built with (the python2.4 and python2.5 distributed by python.org
are built with Visual Studio 2003), the standard <tt>python setup.py
build</tt> should just work.
</p>
<p>
As of python2.5, the distutils support building extensions with MingGW out
of the box. Following the instruction here:
<a href="http://boodebr.org/main/python/build-windows-extensions">Building
Python extensions for Windows with only free tools</a> should get you started.
</p>
<p>
If you need to build it on your own, the following notes are provided:
</p>
<p>
You will need to create a DLL that can be loaded into the interpreter.
This section briefly describes the use of SWIG with Microsoft Visual
C++. As a starting point, many of SWIG's examples include project
files (.dsp files) for Visual C++ 6. These can be opened by more
recent versions of Visual Studio.
You might want to take a quick look at these examples in addition to
reading this section.
</p>
<p>
In Developer Studio, SWIG should be invoked as a custom build option.
This is usually done as follows:
</p>
<ul>
<li>Open up a new workspace and use the AppWizard to select a DLL
project.
<li>Add both the SWIG interface file (the .i file), any supporting C
files, and the name of the wrapper file that will be created by SWIG
(ie. <tt>example_wrap.c</tt>). Note : If using C++, choose a
different suffix for the wrapper file such as
<tt>example_wrap.cxx</tt>. Don't worry if the wrapper file doesn't
exist yet--Developer Studio keeps a reference to it.
<li>Select the SWIG interface file and go to the settings menu. Under
settings, select the "Custom Build" option.
<li>Enter "SWIG" in the description field.
<li>Enter "<tt>swig -python -o $(ProjDir)\$(InputName)_wrap.c $(InputPath)</tt>" in the "Build command(s) field"
<li>Enter "<tt>$(ProjDir)\$(InputName)_wrap.c</tt>" in the "Output files(s) field".
<li>Next, select the settings for the entire project and go to
"C++:Preprocessor". Add the include directories for your Python
installation under "Additional include directories".
<li>Define the symbol __WIN32__ under preprocessor options.
<li>Finally, select the settings for the entire project and go to
"Link Options". Add the Python library file to your link libraries.
For example "python27.lib". Also, set the name of the output file to
match the name of your Python module, i.e. <tt>_example.pyd</tt>
<li>Build your project.
</ul>
<p>
If all went well, SWIG will be automatically invoked whenever
you build your project. Any changes made to the interface file will
result in SWIG being automatically executed to produce a new version of
the wrapper file.
</p>
<p>
To run your new Python extension, simply run Python
and use the <tt>import</tt> command as normal. For example :
</p>
<div class="targetlang"><pre>
$ python
>>> import example
>>> print example.fact(4)
24
>>>
</pre></div>
<p>
If you get an <tt>ImportError</tt> exception when importing the module, you may
have forgotten to include additional library files when you built your module.
If you get an access violation or some kind of general protection fault
immediately upon import, you have a more serious problem. This
is often caused by linking your extension module against the wrong
set of Win32 debug or thread libraries. You will have to fiddle around with
the build options of project to try and track this down.
</p>
<p>
A 'Debug' build of the wrappers requires a debug build of the Python interpreter.
This normally requires building the Python interpreter from source, which is not a
job for the feint-hearted. Alternatively you can use the 'Release' build of the
Python interpreter with a 'Debug' build of your wrappers by defining the <tt>SWIG_PYTHON_INTERPRETER_NO_DEBUG</tt>
symbol under the preprocessor options. Or you can ensure this macro is defined at the beginning
of the wrapper code using the following in your interface file, where <tt>_MSC_VER</tt> ensures it is
only used by the Visual Studio compiler:
</p>
<div class="code"><pre>
%begin %{
#ifdef _MSC_VER
#define SWIG_PYTHON_INTERPRETER_NO_DEBUG
#endif
%}
</pre></div>
<p>
Some users have reported success in building extension modules using Cygwin
and other compilers. However, the problem of building usable DLLs with these
compilers tends to be rather problematic. For the latest information,
you may want to consult the <a href="https://github.com/swig/swig/wiki">
SWIG Wiki</a>.
</p>
<H3><a name="Python_commandline">33.2.9 Additional Python commandline options</a></H3>
<p>
The following table lists the additional commandline options available for the Python module. They can also be seen by using:
</p>
<div class="code"><pre>
swig -python -help
</pre></div>
<table summary="Python specific options">
<tr>
<th>Python specific options</th>
</tr>
<tr><td>-builtin </td><td>Create Python built-in types rather than proxy classes, for better performance</td></tr>
<tr><td>-castmode </td><td>Enable the casting mode, which allows implicit cast between types in Python</td></tr>
<tr><td>-debug-doxygen-parser </td><td>Display doxygen parser module debugging information</td></tr>
<tr><td>-debug-doxygen-translator </td><td>Display doxygen translator module debugging information</td></tr>
<tr><td>-dirvtable </td><td>Generate a pseudo virtual table for directors for faster dispatch</td></tr>
<tr><td>-doxygen </td><td>Convert C++ doxygen comments to pydoc comments in proxy classes</td></tr>
<tr><td>-extranative </td><td>Return extra native wrappers for C++ std containers wherever possible</td></tr>
<tr><td>-fastproxy </td><td>Use fast proxy mechanism for member methods</td></tr>
<tr><td>-globals <name> </td><td>Set <name> used to access C global variable (default: 'cvar')</td></tr>
<tr><td>-interface <mod></td><td>Set low-level C/C++ module name to <mod> (default: module name prefixed by '_')</td></tr>
<tr><td>-keyword </td><td>Use keyword arguments</td></tr>
<tr><td>-nofastunpack </td><td>Use traditional UnpackTuple method to parse the argument functions</td></tr>
<tr><td>-noh </td><td>Don't generate the output header file</td></tr>
<tr><td>-noproxy </td><td>Don't generate proxy classes</td></tr>
<tr><td>-nortti </td><td>Disable the use of the native C++ RTTI with directors</td></tr>
<tr><td>-nothreads </td><td>Disable thread support for the entire interface</td></tr>
<tr><td>-olddefs </td><td>Keep the old method definitions when using -fastproxy</td></tr>
<tr><td>-py3 </td><td>Generate code with Python 3 specific features and syntax</td></tr>
<tr><td>-relativeimport </td><td>Use relative Python imports</td></tr>
<tr><td>-threads </td><td>Add thread support for all the interface</td></tr>
<tr><td>-O </td><td>Enable the following optimization options: -fastdispatch -fastproxy -fvirtual</td></tr>
</table>
<p>
Many of these options are covered later on and their use should become clearer by the time you have finished reading this section on SWIG and Python.
</p>
<H2><a name="Python_nn13">33.3 A tour of basic C/C++ wrapping</a></H2>
<p>
By default, SWIG tries to build a very natural Python interface
to your C/C++ code. Functions are wrapped as functions, classes are wrapped as classes, and so forth.
This section briefly covers the essential aspects of this wrapping.
</p>
<H3><a name="Python_nn14">33.3.1 Modules</a></H3>
<p>
The SWIG <tt>%module</tt> directive specifies the name of the Python
module. If you specify `<tt>%module example</tt>', then everything is
wrapped into a Python '<tt>example</tt>' module. Underneath the covers,
this module consists of a Python source file <tt>example.py</tt> and a low-level
extension module <tt>_example.so</tt>. When choosing a
module name, make sure you don't use the same name as a built-in
Python command or standard module name.
</p>
<H3><a name="Python_nn15">33.3.2 Functions</a></H3>
<p>
Global functions are wrapped as new Python built-in functions. For example,
</p>
<div class="code"><pre>
%module example
int fact(int n);
</pre></div>
<p>
creates a built-in function <tt>example.fact(n)</tt> that works exactly
like you think it does:
</p>
<div class="targetlang"><pre>
>>> import example
>>> print example.fact(4)
24
>>>
</pre></div>
<H3><a name="Python_nn16">33.3.3 Global variables</a></H3>
<p>
C/C++ global variables are fully supported by SWIG. However, the underlying
mechanism is somewhat different than you might expect due to the way that
Python assignment works. When you type the following in Python
</p>
<div class="targetlang"><pre>
a = 3.4
</pre></div>
<p>
"a" becomes a name for an object containing the value 3.4. If you later type
</p>
<div class="targetlang"><pre>
b = a
</pre></div>
<p>
then "a" and "b" are both names for the object containing the value
3.4. Thus, there is only one object containing 3.4 and "a"
and "b" are both names that refer to it. This is quite
different than C where a variable name refers to a memory location in which
a value is stored (and assignment copies data into that location).
Because of this, there is no direct way to map variable
assignment in C to variable assignment in Python.
</p>
<p>
To provide access to C global variables, SWIG creates a special
object called `<tt>cvar</tt>' that is added to each SWIG generated
module. Global variables are then accessed as attributes of this object.
For example, consider this interface
</p>
<div class="code"><pre>
// SWIG interface file with global variables
%module example
...
%inline %{
extern int My_variable;
extern double density;
%}
...
</pre></div>
<p>
Now look at the Python interface:
</p>
<div class="targetlang"><pre>
>>> import example
>>> # Print out value of a C global variable
>>> print example.cvar.My_variable
4
>>> # Set the value of a C global variable
>>> example.cvar.density = 0.8442
>>> # Use in a math operation
>>> example.cvar.density = example.cvar.density*1.10
</pre></div>
<p>
If you make an error in variable assignment, you will receive an
error message. For example:
</p>
<div class="targetlang"><pre>
>>> example.cvar.density = "Hello"
Traceback (most recent call last):
File "<stdin>", line 1, in ?
TypeError: C variable 'density (double )'
>>>
</pre></div>
<p>
If a variable is declared as <tt>const</tt>, it is wrapped as a
read-only variable. Attempts to modify its value will result in an
error.
</p>
<p>
To make ordinary variables read-only, you can use the <tt>%immutable</tt> directive. For example:
</p>
<div class="code">
<pre>
%{
extern char *path;
%}
%immutable;
extern char *path;
%mutable;
</pre>
</div>
<p>
The <tt>%immutable</tt> directive stays in effect until it is explicitly disabled or cleared using
<tt>%mutable</tt>.
See the <a href="SWIG.html#SWIG_readonly_variables">Creating read-only variables</a> section for further details.
</p>
<p>
If you just want to make a specific variable immutable, supply a declaration name. For example:
</p>
<div class="code">
<pre>
%{
extern char *path;
%}
%immutable path;
...
extern char *path; // Read-only (due to %immutable)
</pre>
</div>
<p>
If you would like to access variables using a name other than "<tt>cvar</tt>", it can be
changed using the <tt>-globals</tt> option :
</p>
<div class="shell"><pre>
$ swig -python -globals myvar example.i
</pre></div>
<p>
Some care is in order when importing multiple SWIG modules.
If you use the "<tt>from <file> import *</tt>" style of
importing, you will get a name clash on the variable `<tt>cvar</tt>'
and you will only be able to access global variables from the last
module loaded. To prevent this, you might consider renaming
<tt>cvar</tt> or making it private to the module by giving it a name
that starts with a leading underscore. SWIG does not create <tt>cvar</tt>
if there are no global variables in a module.
</p>
<H3><a name="Python_nn17">33.3.4 Constants and enums</a></H3>
<p>
C/C++ constants are installed as Python objects containing the
appropriate value. To create a constant, use <tt>#define</tt>, <tt>enum</tt>, or the
<tt>%constant</tt> directive. For example:
</p>
<div class="code">
<pre>
#define PI 3.14159
#define VERSION "1.0"
enum Beverage { ALE, LAGER, STOUT, PILSNER };
%constant int FOO = 42;
%constant const char *path = "/usr/local";
</pre>
</div>
<p>
For enums, make sure that the definition of the enumeration actually appears in a header
file or in the wrapper file somehow---if you just stick an enum in a SWIG interface without
also telling the C compiler about it, the wrapper code won't compile.
</p>
<p>
Note: declarations declared as <tt>const</tt> are wrapped as read-only variables and
will be accessed using the <tt>cvar</tt> object described in the previous section. They
are not wrapped as constants. For further discussion about this, see the <a href="SWIG.html#SWIG">SWIG Basics</a> chapter.
</p>
<p>
Constants are not guaranteed to remain constant in Python---the name
of the constant could be accidentally reassigned to refer to some
other object. Unfortunately, there is no easy way for SWIG to
generate code that prevents this. You will just have to be careful.
</p>
<H3><a name="Python_nn18">33.3.5 Pointers</a></H3>
<p>
C/C++ pointers are fully supported by SWIG. Furthermore, SWIG has no
problem working with incomplete type information. Here is a rather
simple interface:
</p>
<div class="code">
<pre>
%module example
FILE *fopen(const char *filename, const char *mode);
int fputs(const char *, FILE *);
int fclose(FILE *);
</pre>
</div>
<p>
When wrapped, you will be able to use the functions in a natural way from Python. For example:
</p>
<div class="targetlang">
<pre>
>>> import example
>>> f = example.fopen("junk", "w")
>>> example.fputs("Hello World\n", f)
>>> example.fclose(f)
</pre>
</div>
<p>
If this makes you uneasy, rest assured that there is no
deep magic involved. Underneath the covers, pointers to C/C++ objects are
simply represented as opaque values using an especial Python container object:
</p>
<div class="targetlang"><pre>
>>> print f
<Swig Object of type 'FILE *' at 0xb7d6f470>
</pre></div>
<p>
This pointer value can be freely passed around to different C functions that
expect to receive an object of type <tt>FILE *</tt>. The only thing you can't do is
dereference the pointer from Python. Of course, that isn't much of a concern in this example.
</p>
<p>
In older versions of SWIG (1.3.22 or older), pointers were represented
using a plain string object. If you have an old package that still
requires that representation, or you just feel nostalgic, you can
always retrieve it by casting the pointer object to a string:
</p>
<div class="targetlang"><pre>
>>> print str(f)
_c0671108_p_FILE
</pre></div>
<p>
Also, if you need to pass the raw pointer value to some external
Python library, you can do it by casting the pointer object to an
integer:
</p>
<div class="targetlang"><pre>
>>> print int(f)
135833352
</pre></div>
<p>
However, the inverse operation is not possible, i.e., you can't build
a SWIG pointer object from a raw integer value.
</p>
<p>
Note also that the '0' or NULL pointer is always represented by
<tt>None</tt>, no matter what type swig is addressing. In the
previous example, you can call:
</p>
<div class="targetlang">
<pre>
>>> example.fclose(None)
</pre>
</div>
<p>
and that will be equivalent to the following, but not really useful, C
code:
</p>
<div class="code">
<pre>
FILE *f = NULL;
fclose(f);
</pre>
</div>
<p>
As much as you might be inclined to modify a pointer value directly
from Python, don't. The hexadecimal encoding is not necessarily the
same as the logical memory address of the underlying object. Instead
it is the raw byte encoding of the pointer value. The encoding will
vary depending on the native byte-ordering of the platform (i.e.,
big-endian vs. little-endian). Similarly, don't try to manually cast
a pointer to a new type by simply replacing the type-string. This may
not work like you expect, it is particularly dangerous when casting
C++ objects. If you need to cast a pointer or change its value,
consider writing some helper functions instead. For example:
</p>
<div class="code">
<pre>
%inline %{
/* C-style cast */
Bar *FooToBar(Foo *f) {
return (Bar *) f;
}
/* C++-style cast */
Foo *BarToFoo(Bar *b) {
return dynamic_cast<Foo*>(b);
}
Foo *IncrFoo(Foo *f, int i) {
return f+i;
}
%}
</pre>
</div>
<p>
Also, if working with C++, you should always try
to use the new C++ style casts. For example, in the above code, the
C-style cast may return a bogus result whereas as the C++-style cast will return
<tt>None</tt> if the conversion can't be performed.
</p>
<H3><a name="Python_nn19">33.3.6 Structures</a></H3>
<p>
If you wrap a C structure, it is wrapped by a Python class. This provides
a very natural interface. For example,
</p>
<div class="code"><pre>
struct Vector {
double x, y, z;
};
</pre></div>
<p>
is used as follows:
</p>
<div class="targetlang"><pre>
>>> v = example.Vector()
>>> v.x = 3.5
>>> v.y = 7.2
>>> print v.x, v.y, v.z
3.5 7.2 0.0
>>>
</pre></div>
<p>
Similar access is provided for unions and the data members of C++ classes.
</p>
<p>
If you print out the value of <tt>v</tt> in the above example, you will see
something like this:
</p>
<div class="targetlang">
<pre>
>>> print v
<C Vector instance at _18e31408_p_Vector>
</pre>
</div>
<p>
This object is actually a Python instance that has been wrapped around a pointer to the low-level
C structure. This instance doesn't actually do anything--it just serves as a proxy.
The pointer to the C object can be found in the <tt>.this</tt>
attribute. For example:
</p>
<div class="targetlang">
<pre>
>>> print v.this
_18e31408_p_Vector
>>>
</pre>
</div>
<p>
Further details about the Python proxy class are covered a little later.
</p>
<p>
<tt>const</tt> members of a structure are read-only. Data members
can also be forced to be read-only using the <tt>%immutable</tt> directive. For example:
</p>
<div class="code">
<pre>
struct Foo {
...
%immutable;
int x; /* Read-only members */
char *name;
%mutable;
...
};
</pre>
</div>
<p>
When <tt>char *</tt> members of a structure are wrapped, the contents are assumed to be
dynamically allocated using <tt>malloc</tt> or <tt>new</tt> (depending on whether or not
SWIG is run with the -c++ option). When the structure member is set, the old contents will be
released and a new value created. If this is not the behavior you want, you will have to use
a typemap (described later).
</p>
<p>
If a structure contains arrays, access to those arrays is managed through pointers. For
example, consider this:
</p>
<div class="code">
<pre>
struct Bar {
int x[16];
};
</pre>
</div>
<p>
If accessed in Python, you will see behavior like this:
</p>
<div class="targetlang">
<pre>
>>> b = example.Bar()
>>> print b.x
_801861a4_p_int
>>>
</pre>
</div>
<p>
This pointer can be passed around to functions that expect to receive
an <tt>int *</tt> (just like C). You can also set the value of an array member using
another pointer. For example:
</p>
<div class="targetlang">
<pre>
>>> c = example.Bar()
>>> c.x = b.x # Copy contents of b.x to c.x
</pre>
</div>
<p>
For array assignment, SWIG copies the entire contents of the array starting with the data pointed
to by <tt>b.x</tt>. In this example, 16 integers would be copied. Like C, SWIG makes
no assumptions about bounds checking---if you pass a bad pointer, you may get a segmentation
fault or access violation.
</p>
<p>
When a member of a structure is itself a structure, it is handled as a
pointer. For example, suppose you have two structures like this:
</p>
<div class="code">
<pre>
struct Foo {
int a;
};
struct Bar {
Foo f;
};
</pre>
</div>
<p>
Now, suppose that you access the <tt>f</tt> attribute of <tt>Bar</tt> like this:
</p>
<div class="targetlang">
<pre>
>>> b = Bar()
>>> x = b.f
</pre>
</div>
<p>
In this case, <tt>x</tt> is a pointer that points to the <tt>Foo</tt> that is inside <tt>b</tt>.
This is the same value as generated by this C code:
</p>
<div class="code">
<pre>
Bar b;
Foo *x = &b->f; /* Points inside b */
</pre>
</div>
<p>
Because the pointer points inside the structure, you can modify the contents and
everything works just like you would expect. For example:
</p>
<div class="targetlang">
<pre>
>>> b = Bar()
>>> b.f.a = 3 # Modify attribute of structure member
>>> x = b.f
>>> x.a = 3 # Modifies the same structure
</pre>
</div>
<p>
Note that there is a limitation with structs within structs that will cause a problem
if the outer struct is not a named variable in Python. The following will cause a segfault:
</p>
<div class="targetlang">
<pre>
Bar().f.a = 3
</pre>
</div>
<p>
because the unnamed Python proxy class for <tt>Bar()</tt> has its reference count
decremented by the Python interpreter after <tt>f</tt> has been obtained from it and
before <tt>f</tt> is used to obtain <tt>a</tt>.
This results in the underlying <tt>Bar</tt> instance being deleted, which of course also deletes
<tt>f</tt> inside it. Hence the pointer to <tt>f</tt> points to deleted
memory and use of it results in a segfault or some sort of other undefined behaviour.
</p>
<H3><a name="Python_nn20">33.3.7 C++ classes</a></H3>
<p>
C++ classes are wrapped by Python classes as well. For example, if you have this class,
</p>
<div class="code"><pre>
class List {
public:
List();
~List();
int search(char *item);
void insert(char *item);
void remove(char *item);
char *get(int n);
int length;
};
</pre></div>
<p>
you can use it in Python like this:
</p>
<div class="targetlang"><pre>
>>> l = example.List()
>>> l.insert("Ale")
>>> l.insert("Stout")
>>> l.insert("Lager")
>>> l.get(1)
'Stout'
>>> print l.length
3
>>>
</pre></div>
<p>
Class data members are accessed in the same manner as C structures.
</p>
<p>
Static class members present a special problem for Python. Prior to Python-2.2,
Python classes had no support for static methods and no version of Python
supports static member variables in a manner that SWIG can utilize. Therefore,
SWIG generates wrappers that try to work around some of these issues. To illustrate,
suppose you have a class like this:
</p>
<div class="code">
<pre>
class Spam {
public:
static void foo();
static int bar;
};
</pre>
</div>
<p>
In Python, the static member can be access in three different ways:
</p>
<div class="targetlang">
<pre>
>>> example.Spam_foo() # Spam::foo()
>>> s = example.Spam()
>>> s.foo() # Spam::foo() via an instance
>>> example.Spam.foo() # Spam::foo(). Python-2.2 only
</pre>
</div>
<p>
The first two methods of access are supported in all versions of Python. The
last technique is only available in Python-2.2 and later versions.
</p>
<p>
Static member variables are currently accessed as global variables. This means,
they are accessed through <tt>cvar</tt> like this:
</p>
<div class="targetlang">
<pre>
>>> print example.cvar.Spam_bar
7
</pre>
</div>
<H3><a name="Python_nn21">33.3.8 C++ inheritance</a></H3>
<p>
SWIG is fully aware of issues related to C++ inheritance. Therefore, if you have
classes like this
</p>
<div class="code">
<pre>
class Foo {
...
};
class Bar : public Foo {
...
};
</pre>
</div>
<p>
those classes are wrapped into a hierarchy of Python classes that reflect the same inheritance
structure. All of the usual Python utility functions work normally:
</p>
<div class="targetlang">
<pre>
>>> b = Bar()
>>> instance(b, Foo)
1
>>> issubclass(Bar, Foo)
1
>>> issubclass(Foo, Bar)
0
</pre>
</div>
<p>
Furthermore, if you have functions like this
</p>
<div class="code">
<pre>
void spam(Foo *f);
</pre>
</div>
<p>
then the function <tt>spam()</tt> accepts <tt>Foo *</tt> or a pointer to any class derived from <tt>Foo</tt>.
</p>
<p>
It is safe to use multiple inheritance with SWIG.
</p>
<H3><a name="Python_nn22">33.3.9 Pointers, references, values, and arrays</a></H3>
<p>
In C++, there are many different ways a function might receive
and manipulate objects. For example:
</p>
<div class="code">
<pre>
void spam1(Foo *x); // Pass by pointer
void spam2(Foo &x); // Pass by reference
void spam3(const Foo &x);// Pass by const reference
void spam4(Foo x); // Pass by value
void spam5(Foo x[]); // Array of objects
</pre>
</div>
<p>
In Python, there is no detailed distinction like this--specifically,
there are only "objects". There are no pointers, references, arrays,
and so forth. Because of this, SWIG unifies all of these types
together in the wrapper code. For instance, if you actually had the
above functions, it is perfectly legal to do this:
</p>
<div class="targetlang">
<pre>
>>> f = Foo() # Create a Foo
>>> spam1(f) # Ok. Pointer
>>> spam2(f) # Ok. Reference
>>> spam3(f) # Ok. Const reference
>>> spam4(f) # Ok. Value.
>>> spam5(f) # Ok. Array (1 element)
</pre>
</div>
<p>
Similar behavior occurs for return values. For example, if you had
functions like this,
</p>
<div class="code">
<pre>
Foo *spam6();
Foo &spam7();
Foo spam8();
const Foo &spam9();
</pre>
</div>
<p>
then all three functions will return a pointer to some <tt>Foo</tt> object.
Since the third function (spam8) returns a value, newly allocated memory is used
to hold the result and a pointer is returned (Python will release this memory
when the return value is garbage collected). The fourth case (spam9)
which returns a const reference, in most of the cases will be
treated as a returning value, and it will follow the same
allocation/deallocation process.
</p>
<H3><a name="Python_nn23">33.3.10 C++ overloaded functions</a></H3>
<p>
C++ overloaded functions, methods, and constructors are mostly supported by SWIG. For example,
if you have two functions like this:
</p>
<div class="code">
<pre>
void foo(int);
void foo(char *c);
</pre>
</div>
<p>
You can use them in Python in a straightforward manner:
</p>
<div class="targetlang">
<pre>
>>> foo(3) # foo(int)
>>> foo("Hello") # foo(char *c)
</pre>
</div>
<p>
Similarly, if you have a class like this,
</p>
<div class="code">
<pre>
class Foo {
public:
Foo();
Foo(const Foo &);
...
};
</pre>
</div>
<p>
you can write Python code like this:
</p>
<div class="targetlang">
<pre>
>>> f = Foo() # Create a Foo
>>> g = Foo(f) # Copy f
</pre>
</div>
<p>
Overloading support is not quite as flexible as in C++. Sometimes there are methods that SWIG
can't disambiguate. For example:
</p>
<div class="code">
<pre>
void spam(int);
void spam(short);
</pre>
</div>
<p>
or
</p>
<div class="code">
<pre>
void foo(Bar *b);
void foo(Bar &b);
</pre>
</div>
<p>
If declarations such as these appear, you will get a warning message like this:
</p>
<div class="shell">
<pre>
example.i:12: Warning 509: Overloaded method spam(short) effectively ignored,
example.i:11: Warning 509: as it is shadowed by spam(int).
</pre>
</div>
<p>
To fix this, you either need to ignore or rename one of the methods. For example:
</p>
<div class="code">
<pre>
%rename(spam_short) spam(short);
...
void spam(int);
void spam(short); // Accessed as spam_short
</pre>
</div>
<p>
or
</p>
<div class="code">
<pre>
%ignore spam(short);
...
void spam(int);
void spam(short); // Ignored
</pre>
</div>
<p>
SWIG resolves overloaded functions and methods using a disambiguation scheme that ranks and sorts
declarations according to a set of type-precedence rules. The order in which declarations appear
in the input does not matter except in situations where ambiguity arises--in this case, the
first declaration takes precedence.
</p>
<p>
Please refer to the "SWIG and C++" chapter for more information about overloading.
</p>
<H3><a name="Python_nn24">33.3.11 C++ operators</a></H3>
<p>
Certain C++ overloaded operators can be handled automatically by SWIG. For example,
consider a class like this:
</p>
<div class="code">
<pre>
class Complex {
private:
double rpart, ipart;
public:
Complex(double r = 0, double i = 0) : rpart(r), ipart(i) { }
Complex(const Complex &c) : rpart(c.rpart), ipart(c.ipart) { }
Complex &operator=(const Complex &c);
Complex operator+=(const Complex &c) const;
Complex operator+(const Complex &c) const;
Complex operator-(const Complex &c) const;
Complex operator*(const Complex &c) const;
Complex operator-() const;
double re() const { return rpart; }
double im() const { return ipart; }
};
</pre>
</div>
<p>
When wrapped, it works like you expect:
</p>
<div class="targetlang">
<pre>
>>> c = Complex(3, 4)
>>> d = Complex(7, 8)
>>> e = c + d
>>> e.re()
10.0
>>> e.im()
12.0
>>> c += d
>>> c.re()
10.0
>>> c.im()
12.0
</pre>
</div>
<p>
One restriction with operator overloading support is that SWIG is not
able to fully handle operators that aren't defined as part of the class.
For example, if you had code like this
</p>
<div class="code">
<pre>
class Complex {
...
friend Complex operator+(double, const Complex &c);
...
};
</pre>
</div>
<p>
then SWIG ignores it and issues a warning. You can still wrap the operator,
but you may have to encapsulate it in a special function. For example:
</p>
<div class="code">
<pre>
%rename(Complex_add_dc) operator+(double, const Complex &);
</pre>
</div>
<p>
There are ways to make this operator appear as part of the class using the <tt>%extend</tt> directive.
Keep reading.
</p>
<p>
Also, be aware that certain operators don't map cleanly to Python. For instance,
overloaded assignment operators don't map to Python semantics and will be ignored.
</p>
<p>
Operator overloading is implemented in the <tt>pyopers.swg</tt> library file.
In particular overloaded operators are marked with the <tt>python:maybecall</tt> feature, also known as <tt>%pythonmaybecall</tt>.
This feature forces SWIG to generate code that return an instance of Python's <tt>NotImplemented</tt>
instead of raising an exception when the comparison fails, that is, on any kind of error.
This follows the guidelines in <a href="https://www.python.org/dev/peps/pep-0207/">PEP 207 - Rich Comparisons</a> and <a href="https://docs.python.org/3/library/constants.html#NotImplemented">NotImplemented Python constant</a>.
</p>
<H3><a name="Python_nn25">33.3.12 C++ namespaces</a></H3>
<p>
SWIG is aware of C++ namespaces, but namespace names do not appear in
the module nor do namespaces result in a module that is broken up into
submodules or packages. For example, if you have a file like this,
</p>
<div class="code">
<pre>
%module example
namespace foo {
int fact(int n);
struct Vector {
double x, y, z;
};
};
</pre>
</div>
<p>
it works in Python as follows:
</p>
<div class="targetlang">
<pre>
>>> import example
>>> example.fact(3)
6
>>> v = example.Vector()
>>> v.x = 3.4
>>> print v.y
0.0
>>>
</pre>
</div>
<p>
If your program has more than one namespace, name conflicts (if any) can be resolved using <tt>%rename</tt>
For example:
</p>
<div class="code">
<pre>
%rename(Bar_spam) Bar::spam;
namespace Foo {
int spam();
}
namespace Bar {
int spam();
}
</pre>
</div>
<p>
If you have more than one namespace and your want to keep their
symbols separate, consider wrapping them as separate SWIG modules.
For example, make the module name the same as the namespace and create
extension modules for each namespace separately. If your program
utilizes thousands of small deeply nested namespaces each with
identical symbol names, well, then you get what you deserve.
</p>
<H3><a name="Python_nn26">33.3.13 C++ templates</a></H3>
<p>
C++ templates don't present a huge problem for SWIG. However, in order
to create wrappers, you have to tell SWIG to create wrappers for a particular
template instantiation. To do this, you use the <tt>%template</tt> directive.
For example:
</p>
<div class="code">
<pre>
%module example
%{
#include "pair.h"
%}
template<class T1, class T2>
struct pair {
typedef T1 first_type;
typedef T2 second_type;
T1 first;
T2 second;
pair();
pair(const T1&, const T2&);
~pair();
};
%template(pairii) pair<int, int>;
</pre>
</div>
<p>
In Python:
</p>
<div class="targetlang">
<pre>
>>> import example
>>> p = example.pairii(3, 4)
>>> p.first
3
>>> p.second
4
</pre>
</div>
<p>
Obviously, there is more to template wrapping than shown in this example.
More details can be found in the <a href="SWIGPlus.html#SWIGPlus">SWIG and C++</a> chapter.
Some more complicated
examples will appear later.
</p>
<H3><a name="Python_nn27">33.3.14 C++ Smart Pointers</a></H3>
<H4><a name="Python_smart_pointers_shared_ptr">33.3.14.1 The shared_ptr Smart Pointer</a></H4>
<p>
The C++11 standard provides <tt>std::shared_ptr</tt> which was derived from the Boost
implementation, <tt>boost::shared_ptr</tt>.
Both of these are available for Python in the SWIG library and usage is outlined
in the <a href="Library.html#Library_std_shared_ptr">shared_ptr smart pointer</a> library section.
</p>
<H4><a name="Python_smart_pointers_generic">33.3.14.2 Generic Smart Pointers</a></H4>
<p>
In certain C++ programs, it is common to use classes that have been wrapped by
so-called "smart pointers." Generally, this involves the use of a template class
that implements <tt>operator->()</tt> like this:
</p>
<div class="code">
<pre>
template<class T> class SmartPtr {
...
T *operator->();
...
}
</pre>
</div>
<p>
Then, if you have a class like this,
</p>
<div class="code">
<pre>
class Foo {
public:
int x;
int bar();
};
</pre>
</div>
<p>
A smart pointer would be used in C++ as follows:
</p>
<div class="code">
<pre>
SmartPtr<Foo> p = CreateFoo(); // Created somehow (not shown)
...
p->x = 3; // Foo::x
int y = p->bar(); // Foo::bar
</pre>
</div>
<p>
To wrap this in Python, simply tell SWIG about the <tt>SmartPtr</tt> class and the low-level
<tt>Foo</tt> object. Make sure you instantiate <tt>SmartPtr</tt> using <tt>%template</tt> if necessary.
For example:
</p>
<div class="code">
<pre>
%module example
...
%template(SmartPtrFoo) SmartPtr<Foo>;
...
</pre>
</div>
<p>
Now, in Python, everything should just "work":
</p>
<div class="targetlang">
<pre>
>>> p = example.CreateFoo() # Create a smart-pointer somehow
>>> p.x = 3 # Foo::x
>>> p.bar() # Foo::bar
</pre>
</div>
<p>
If you ever need to access the underlying pointer returned by <tt>operator->()</tt> itself,
simply use the <tt>__deref__()</tt> method. For example:
</p>
<div class="targetlang">
<pre>
>>> f = p.__deref__() # Returns underlying Foo *
</pre>
</div>
<H3><a name="Python_nn27a">33.3.15 C++ reference counted objects</a></H3>
<p>
The <a href="SWIGPlus.html#SWIGPlus_ref_unref">C++ reference counted objects</a> section contains
Python examples of memory management using referencing counting.
</p>
<H2><a name="Python_nn28">33.4 Further details on the Python class interface</a></H2>
<p>
In the previous section, a high-level view of Python wrapping was
presented. A key component of this wrapping is that structures and
classes are wrapped by Python proxy classes. This provides a very
natural Python interface and allows SWIG to support a number of
advanced features such as operator overloading. However, a number
of low-level details were omitted. This section provides a brief overview
of how the proxy classes work.
</p>
<p><b>New in SWIG version 2.0.4:</b>
The use of Python proxy classes has performance implications that may be
unacceptable for a high-performance library. The new <tt>-builtin</tt>
option instructs SWIG to forego the use of proxy classes, and instead
create wrapped types as new built-in Python types. When this option is used,
the following section ("Proxy classes") does not apply. Details on the use of
the <tt>-builtin</tt> option are in the <a href="#Python_builtin_types">Built-in Types</a>
section.
</p>
<H3><a name="Python_nn29">33.4.1 Proxy classes</a></H3>
<p>
In the <a href="SWIG.html#SWIG">"SWIG basics"</a> and <a href="SWIGPlus.html#SWIGPlus">"SWIG and C++"</a> chapters,
details of low-level structure and class wrapping are described. To summarize those chapters, if you
have a class like this
</p>
<div class="code">
<pre>
class Foo {
public:
int x;
int spam(int);
...
</pre>
</div>
<p>
then SWIG transforms it into a set of low-level procedural wrappers. For example:
</p>
<div class="code">
<pre>
Foo *new_Foo() {
return new Foo();
}
void delete_Foo(Foo *f) {
delete f;
}
int Foo_x_get(Foo *f) {
return f->x;
}
void Foo_x_set(Foo *f, int value) {
f->x = value;
}
int Foo_spam(Foo *f, int arg1) {
return f->spam(arg1);
}
</pre>
</div>
<p>
These wrappers can be found in the low-level extension module (e.g., <tt>_example</tt>).
</p>
<p>
Using these wrappers, SWIG generates a high-level Python proxy class (also known as a shadow class) like this (shown
for Python 2.2):
</p>
<div class="targetlang">
<pre>
import _example
class Foo(object):
def __init__(self):
self.this = _example.new_Foo()
self.thisown = 1
def __del__(self):
if self.thisown:
_example.delete_Foo(self.this)
def spam(self, arg1):
return _example.Foo_spam(self.this, arg1)
x = property(_example.Foo_x_get, _example.Foo_x_set)
</pre>
</div>
<p>
This class merely holds a pointer to the underlying C++ object (<tt>.this</tt>) and dispatches methods and
member variable access to that object using the low-level accessor functions. From a user's point of
view, it makes the class work normally:
</p>
<div class="targetlang">
<pre>
>>> f = example.Foo()
>>> f.x = 3
>>> y = f.spam(5)
</pre>
</div>
<p>
The fact that the class has been wrapped by a real Python class offers certain advantages. For instance,
you can attach new Python methods to the class and you can even inherit from it (something not supported
by Python built-in types until Python 2.2).
</p>
<H3><a name="Python_builtin_types">33.4.2 Built-in Types</a></H3>
<p>
The <tt>-builtin</tt> option provides a significant performance improvement
in the wrapped code. To understand the difference between proxy classes
and built-in types, let's take a look at what a wrapped object looks like
under both circumstances.
</p>
<p>When proxy classes are used, each wrapped object in Python is an instance
of a pure Python class. As a reminder, here is what the <tt>__init__</tt> method looks
like in a proxy class:
</p>
<div class="targetlang">
<pre>
class Foo(object):
def __init__(self):
self.this = _example.new_Foo()
self.thisown = 1
</pre>
</div>
<p>When a <tt>Foo</tt> instance is created, the call to <tt>_example.new_Foo()</tt>
creates a new C++ <tt>Foo</tt> instance; wraps that C++ instance inside an instance of
a Python built-in type called <tt>SwigPyObject</tt>; and stores the <tt>SwigPyObject</tt>
instance in the 'this' field of the Python Foo object. Did you get all that? So, the
Python <tt>Foo</tt> object is composed of three parts:</p>
<ul>
<li> The Python <tt>Foo</tt> instance, which contains...</li>
<li> ... an instance of <tt>struct SwigPyObject</tt>, which contains...</li>
<li> ... a C++ <tt>Foo</tt> instance</li>
</ul>
<p>When <tt>-builtin</tt> is used, the pure Python layer is stripped off. Each
wrapped class is turned into a new Python built-in type which inherits from
<tt>SwigPyObject</tt>, and <tt>SwigPyObject</tt> instances are returned directly
from the wrapped methods. For more information about Python built-in extensions,
please refer to the Python documentation:</p>
<p><a href="https://docs.python.org/3/extending/newtypes.html">https://docs.python.org/3/extending/newtypes.html</a></p>
<H4><a name="Python_builtin_limitations">33.4.2.1 Limitations</a></H4>
<p>Use of the <tt>-builtin</tt> option implies a couple of limitations:
<ul>
<li><p>Some legacy syntax is no longer supported; in particular:</p>
<ul>
<li>The functional interface is no longer exposed. For example, you may no longer call <tt>Whizzo.new_CrunchyFrog()</tt>. Instead, you must use <tt>Whizzo.CrunchyFrog()</tt>.</li>
<li>Static member variables are no longer accessed through the 'cvar' field (e.g., <tt>Dances.cvar.FishSlap</tt>).
They are instead accessed in the idiomatic way (<tt>Dances.FishSlap</tt>).</li>
</ul>
</li>
<li><p>Wrapped types may not be raised as Python exceptions. Here's why: the Python internals expect that all sub-classes of Exception will have this struct layout:</p>
<div class="code">
<pre>
typedef struct {
PyObject_HEAD
PyObject *dict;
PyObject *args;
PyObject *message;
} PyBaseExceptionObject;
</pre>
</div>
<p>But swig-generated wrappers expect that all swig-wrapped classes will have this struct layout:</p>
<div class="code">
<pre>
typedef struct {
PyObject_HEAD
void *ptr;
swig_type_info *ty;
int own;
PyObject *next;
PyObject *dict;
} SwigPyObject;
</pre>
</div>
<p>There are workarounds for this. For example, if you wrap this class:
<div class="code">
<pre>
class MyException {
public:
MyException (const char *msg_);
~MyException ();
const char *what () const;
private:
char *msg;
};
</pre>
</div>
<p>... you can define this Python class, which may be raised as an exception:</p>
<div class="targetlang">
<pre>
class MyPyException(Exception):
def __init__(self, msg, *args):
Exception.__init__(self, *args)
self.myexc = MyException(msg)
def what(self):
return self.myexc.what()
</pre>
</div>
</li>
<li><p>Reverse binary operators (e.g., <tt>__radd__</tt>) are not supported.</p>
<p>To illustrate this point, if you have a wrapped class called <tt>MyString</tt>,
and you want to use instances of <tt>MyString</tt> interchangeably with native Python
strings, you can define an <tt>'operator+ (const char*)'</tt> method :</p>
<div class="code">
<pre>
class MyString {
public:
MyString (const char *init);
MyString operator+ (const char *other) const;
...
};
</pre>
</div>
<p>
SWIG will automatically create an operator overload in Python that will allow this:
</p>
<div class="targetlang">
<pre>
from MyModule import MyString
mystr = MyString("No one expects")
episode = mystr + " the Spanish Inquisition"
</pre>
</div>
<p>
This works because the first operand (<tt>mystr</tt>) defines a way
to add a native string to itself. However, the following will <b>not</b> work:
</p>
<div class="targetlang">
<pre>
from MyModule import MyString
mystr = MyString("Parrot")
episode = "Dead " + mystr
</pre>
</div>
<p>
The above code fails, because the first operand -- a native Python string --
doesn't know how to add an instance of <tt>MyString</tt> to itself.
</p>
</li>
<li><p>If you have multiple SWIG modules that share type information (<a href="Modules.html#Modules_nn2">more info</a>),
the <tt>-builtin</tt> option requires a bit of extra discipline to ensure that base classes are initialized before derived classes. Specifically:</p>
<ul>
<li><p>There must be an unambiguous dependency graph for the modules.</p></li>
<li><p>Module dependencies must be explicitly stated with <tt>%import</tt> statements in the SWIG interface file.</p>
</ul>
<p>As an example, suppose module <tt>A</tt> has this interface in <tt>A.i</tt> :</p>
<div class="code"><pre>
%module "A";
class Base {
...
};
</pre></div>
<p>If you want to wrap another module containing a class that inherits from <tt>A</tt>, this is how it would look :</p>
<div class="code"><pre>
%module "B";
%import "A.i"
class Derived : public Base {
...
};
</pre></div>
<p>The <tt>import "A.i"</tt> statement is required, because module <tt>B</tt> depends on module <tt>A</tt>.</p>
<p>As long as you obey these requirements, your Python code may import the modules in any order :</p>
<div class="targetlang"><pre>
import B
import A
assert(issubclass(B.Derived, A.Base))
</pre></div>
</li>
</ul>
<H4><a name="Python_builtin_overloads">33.4.2.2 Operator overloads and slots -- use them!</a></H4>
<p>The entire justification for the <tt>-builtin</tt> option is improved
performance. To that end, the best way to squeeze maximum performance out
of your wrappers is to <b>use operator overloads.</b>
Named method dispatch is slow in Python, even when compared to other scripting languages.
However, Python built-in types have a large number of "slots",
analogous to C++ operator overloads, which allow you to short-circuit named method dispatch
for certain common operations.
</p>
<p>By default, SWIG will translate most C++ arithmetic operator overloads into Python
slot entries. For example, suppose you have this class:
<div class="code">
<pre>
class Twit {
public:
Twit operator+ (const Twit& twit) const;
// Forward to operator+
Twit add (const Twit& twit) const {
return *this + twit;
}
};
</pre>
</div>
<p>SWIG will automatically register <tt>operator+</tt> as a Python slot operator for addition. You may write Python code like this:</p>
<div class="targetlang">
<pre>
from MyModule import Twit
nigel = Twit()
emily = Twit()
percival = nigel + emily
percival = nigel.add(emily)
</pre>
</div>
<p>The last two lines of the Python code are equivalent,
but <b>the line that uses the '+' operator is much faster</b>.
</p>
<p>In-place operators (e.g., <tt>operator+=</tt>) and comparison operators
(<tt>operator==, operator<</tt>, etc.) are also converted to Python
slot operators. For a complete list of C++ operators that are
automatically converted to Python slot operators, refer to the file
<tt>python/pyopers.swg</tt> in the SWIG library.
</p>
<p>
Read about all of the available Python slots here:
<a href="https://docs.python.org/3/c-api/typeobj.html">https://docs.python.org/3/c-api/typeobj.html</a></p>
<p>
There are two ways to define a Python slot function: dispatch to a
statically defined function; or dispatch to a method defined on the
operand.
</p>
<p>
To dispatch to a statically defined function, use %feature("python:<slot>"),
where <slot> is the name of a field in a <tt>PyTypeObject, PyNumberMethods,
PyMappingMethods, PySequenceMethods</tt> or <tt>PyBufferProcs</tt>.
You may override (almost) all of these slots.
</p>
<p>
Let's consider an example setting the <tt>tp_hash</tt> slot for the <tt>MyClass</tt> type.
This is akin to providing a <tt>__hash__</tt> method (for non-builtin types) to make a type hashable.
The hashable type can then for example be added to a Python <tt>dict</tt>.
</p>
<div class="code">
<pre>
%feature("python:tp_hash") MyClass "myHashFunc";
class MyClass {
public:
long field1;
long field2;
...
};
%{
#if PY_VERSION_HEX >= 0x03020000
static Py_hash_t myHashFunc(PyObject *pyobj)
#else
static long myHashFunc(PyObject *pyobj)
#endif
{
MyClass *cobj;
// Convert pyobj to cobj
return (cobj->field1 * (cobj->field2 << 7));
}
%}
</pre>
</div>
<p>
If you examine the generated code, the supplied hash function will now be
the function callback in the tp_hash slot for the builtin type for <tt>MyClass</tt>:
</p>
<div class="code">
<pre>
static PyHeapTypeObject SwigPyBuiltin__MyClass_type = {
...
(hashfunc) myHashFunc, /* tp_hash */
...
</pre>
</div>
<p>
NOTE: It is the responsibility of the programmer (that's you!) to ensure
that a statically defined slot function has the correct signature, the <tt>hashfunc</tt>
typedef in this case.
</p>
<p>
If, instead, you want to dispatch to an instance method, you can
use %feature("python:slot"). For example:
</p>
<div class="code">
<pre>
%feature("python:slot", "tp_hash", functype="hashfunc") MyClass::myHashFunc;
#if PY_VERSION_HEX < 0x03020000
#define Py_hash_t long
#endif
class MyClass {
public:
Py_hash_t myHashFunc() const;
...
};
</pre>
</div>
<p>
NOTE: Some Python slots use a method signature which does not
match the signature of SWIG-wrapped methods. For those slots,
SWIG will automatically generate a "closure" function to re-marshal
the arguments before dispatching to the wrapped method. Setting
the "functype" attribute of the feature enables SWIG to generate
the chosen closure function.
</p>
<p>
There is further information on <tt>%feature("python:slot")</tt>
in the file <tt>python/pyopers.swg</tt> in the SWIG library.
</p>
<H3><a name="Python_nn30">33.4.3 Memory management</a></H3>
<p>NOTE: Although this section refers to proxy objects, everything here also applies
when the <tt>-builtin</tt> option is used.</p>
<p>
Associated with proxy object, is an ownership flag <tt>.thisown</tt> The value of this
flag determines who is responsible for deleting the underlying C++ object. If set to 1,
the Python interpreter will destroy the C++ object when the proxy class is
garbage collected. If set to 0 (or if the attribute is missing), then the destruction
of the proxy class has no effect on the C++ object.
</p>
<p>
When an object is created by a constructor or returned by value, Python automatically takes
ownership of the result. For example:
</p>
<div class="code">
<pre>
class Foo {
public:
Foo();
Foo bar();
};
</pre>
</div>
<p>
In Python:
</p>
<div class="targetlang">
<pre>
>>> f = Foo()
>>> f.thisown
1
>>> g = f.bar()
>>> g.thisown
1
</pre>
</div>
<p>
On the other hand, when pointers are returned to Python, there is often no way to know where
they came from. Therefore, the ownership is set to zero. For example:
</p>
<div class="code">
<pre>
class Foo {
public:
...
Foo *spam();
...
};
</pre>
</div>
<br>
<div class="targetlang">
<pre>
>>> f = Foo()
>>> s = f.spam()
>>> print s.thisown
0
>>>
</pre>
</div>
<p>
This behavior is especially important for classes that act as
containers. For example, if a method returns a pointer to an object
that is contained inside another object, you definitely don't want
Python to assume ownership and destroy it!
</p>
<p>
A good way to indicate that ownership should be set for a returned pointer
is to use the <a href="Library.html#Library_nn11">%newobject directive</a>.
</p>
<p>
Related to containers, ownership issues can arise whenever an object is assigned to a member
or global variable. For example, consider this interface:
</p>
<div class="code">
<pre>
%module example
struct Foo {
int value;
Foo *next;
};
Foo *head = 0;
</pre>
</div>
<p>
When wrapped in Python, careful observation will reveal that ownership changes whenever an object
is assigned to a global variable. For example:
</p>
<div class="targetlang">
<pre>
>>> f = example.Foo()
>>> f.thisown
1
>>> example.cvar.head = f
>>> f.thisown
0
>>>
</pre>
</div>
<p>
In this case, C is now holding a reference to the object---you probably don't want Python to destroy it.
Similarly, this occurs for members. For example:
</p>
<div class="targetlang">
<pre>
>>> f = example.Foo()
>>> g = example.Foo()
>>> f.thisown
1
>>> g.thisown
1
>>> f.next = g
>>> g.thisown
0
>>>
</pre>
</div>
<p>
For the most part, memory management issues remain hidden. However,
there are occasionally situations where you might have to manually
change the ownership of an object. For instance, consider code like this:
</p>
<div class="code">
<pre>
class Node {
Object *value;
public:
void set_value(Object *v) { value = v; }
...
};
</pre>
</div>
<p>
Now, consider the following Python code:
</p>
<div class="targetlang">
<pre>
>>> v = Object() # Create an object
>>> n = Node() # Create a node
>>> n.set_value(v) # Set value
>>> v.thisown
1
>>> del v
</pre>
</div>
<p>
In this case, the object <tt>n</tt> is holding a reference to
<tt>v</tt> internally. However, SWIG has no way to know that this
has occurred. Therefore, Python still thinks that it has ownership of the
object. Should the proxy object be destroyed, then the C++ destructor
will be invoked and <tt>n</tt> will be holding a stale-pointer. If
you're lucky, you will only get a segmentation fault.
</p>
<p>
To work around this, it is always possible to flip the ownership flag. For example,
</p>
<div class="targetlang">
<pre>
>>> v.thisown = 0
</pre>
</div>
<p>
It is also possible to deal with situations like this using
typemaps--an advanced topic discussed later.
</p>
<H2><a name="Python_directors">33.5 Cross language polymorphism</a></H2>
<p>
Proxy classes provide a more natural, object-oriented way to access
extension classes. As described above, each proxy instance has an
associated C++ instance, and method calls to the proxy are passed to the
C++ instance transparently via C wrapper functions.
</p>
<p>
This arrangement is asymmetric in the sense that no corresponding
mechanism exists to pass method calls down the inheritance chain from
C++ to Python. In particular, if a C++ class has been extended in Python
(by extending the proxy class), these extensions will not be visible
from C++ code. Virtual method calls from C++ are thus not able access
the lowest implementation in the inheritance chain.
</p>
<p>
Changes have been made to SWIG 1.3.18 to address this problem and
make the relationship between C++ classes and proxy classes more
symmetric. To achieve this goal, new classes called directors are
introduced at the bottom of the C++ inheritance chain. The job of the
directors is to route method calls correctly, either to C++
implementations higher in the inheritance chain or to Python
implementations lower in the inheritance chain. The upshot is that C++
classes can be extended in Python and from C++ these extensions look
exactly like native C++ classes. Neither C++ code nor Python code needs
to know where a particular method is implemented: the combination of
proxy classes, director classes, and C wrapper functions takes care of
all the cross-language method routing transparently.
</p>
<H3><a name="Python_nn33">33.5.1 Enabling directors</a></H3>
<p>
The director feature is disabled by default. To use directors you
must make two changes to the interface file. First, add the "directors"
option to the %module directive, like this:
</p>
<div class="code">
<pre>
%module(directors="1") modulename
</pre>
</div>
<p>
Without this option no director code will be generated. Second, you
must use the %feature("director") directive to tell SWIG which classes
and methods should get directors. The %feature directive can be applied
globally, to specific classes, and to specific methods, like this:
</p>
<div class="code">
<pre>
// generate directors for all classes that have virtual methods
%feature("director");
// generate directors for the virtual methods in class Foo
%feature("director") Foo;
</pre>
</div>
<p>
You can use the %feature("nodirector") directive to turn off
directors for specific classes or methods. So for example,
</p>
<div class="code">
<pre>
%feature("director") Foo;
%feature("nodirector") Foo::bar;
</pre>
</div>
<p>
will generate directors for the virtual methods of class Foo except
bar().
</p>
<p>
Directors can also be generated implicitly through inheritance.
In the following, class Bar will get a director class that handles
the methods one() and two() (but not three()):
</p>
<div class="code">
<pre>
%feature("director") Foo;
class Foo {
public:
Foo(int foo);
virtual ~Foo();
virtual void one();
virtual void two();
};
class Bar: public Foo {
public:
virtual void three();
};
</pre>
</div>
<p>
then at the Python side you can define
</p>
<div class="targetlang">
<pre>
import mymodule
class MyFoo(mymodule.Foo):
def __init__(self, foo):
mymodule.Foo.__init__(self, foo)
# super().__init__(foo) # Alternative construction for Python3
def one(self):
print "one from Python"
</pre>
</div>
<H3><a name="Python_nn34">33.5.2 Director classes</a></H3>
<p>
For each class that has directors enabled, SWIG generates a new class
that derives from both the class in question and a special
<tt>Swig::Director</tt> class. These new classes, referred to as director
classes, can be loosely thought of as the C++ equivalent of the Python
proxy classes. The director classes store a pointer to their underlying
Python object and handle various issues related to object ownership.
Indeed, this is quite similar to the "this" and "thisown" members of the
Python proxy classes.
</p>
<p>
For simplicity let's ignore the <tt>Swig::Director</tt> class and refer to the
original C++ class as the director's base class. By default, a director
class extends all virtual methods in the inheritance chain of its base
class (see the preceding section for how to modify this behavior).
Virtual methods that have a final specifier are unsurprisingly excluded.
Thus the virtual method calls, whether they originate in C++ or in
Python via proxy classes, eventually end up in at the implementation in
the director class. The job of the director methods is to route these
method calls to the appropriate place in the inheritance chain. By
"appropriate place" we mean the method that would have been called if
the C++ base class and its extensions in Python were seamlessly
integrated. That seamless integration is exactly what the director
classes provide, transparently skipping over all the messy extension API
glue that binds the two languages together.
</p>
<p>
In reality, the "appropriate place" is one of only two possibilities:
C++ or Python. Once this decision is made, the rest is fairly easy. If
the correct implementation is in C++, then the lowest implementation of
the method in the C++ inheritance chain is called explicitly. If the
correct implementation is in Python, the Python API is used to call the
method of the underlying Python object (after which the usual virtual
method resolution in Python automatically finds the right
implementation).
</p>
<p>
Now how does the director decide which language should handle the method call?
The basic rule is to handle the method in Python, unless there's a good
reason not to. The reason for this is simple: Python has the most
"extended" implementation of the method. This assertion is guaranteed,
since at a minimum the Python proxy class implements the method. If the
method in question has been extended by a class derived from the proxy
class, that extended implementation will execute exactly as it should.
If not, the proxy class will route the method call into a C wrapper
function, expecting that the method will be resolved in C++. The wrapper
will call the virtual method of the C++ instance, and since the director
extends this the call will end up right back in the director method. Now
comes the "good reason not to" part. If the director method were to blindly
call the Python method again, it would get stuck in an infinite loop. We avoid this
situation by adding special code to the C wrapper function that tells
the director method to not do this. The C wrapper function compares the
pointer to the Python object that called the wrapper function to the
pointer stored by the director. If these are the same, then the C
wrapper function tells the director to resolve the method by calling up
the C++ inheritance chain, preventing an infinite loop.
</p>
<p>
One more point needs to be made about the relationship between director
classes and proxy classes. When a proxy class instance is created in
Python, SWIG creates an instance of the original C++ class and assigns
it to <tt>.this</tt>. This is exactly what happens without directors and
is true even if directors are enabled for the particular class in
question. When a class <i>derived</i> from a proxy class is created,
however, SWIG then creates an instance of the corresponding C++ director
class. The reason for this difference is that user-defined subclasses
may override or extend methods of the original class, so the director
class is needed to route calls to these methods correctly. For
unmodified proxy classes, all methods are ultimately implemented in C++
so there is no need for the extra overhead involved with routing the
calls through Python.
</p>
<H3><a name="Python_nn35">33.5.3 Ownership and object destruction</a></H3>
<p>
Memory management issues are slightly more complicated with directors
than for proxy classes alone. Python instances hold a pointer to the
associated C++ director object, and the director in turn holds a pointer
back to the Python object. By default, proxy classes own their C++
director object and take care of deleting it when they are garbage
collected.
</p>
<p>
This relationship can be reversed by calling the special
<tt>__disown__()</tt> method of the proxy class. After calling this
method, the <tt>.thisown</tt> flag is set to zero, and the director
class increments the reference count of the Python object. When the
director class is deleted it decrements the reference count. Assuming no
outstanding references to the Python object remain, the Python object
will be destroyed at the same time. This is a good thing, since
directors and proxies refer to each other and so must be created and
destroyed together. Destroying one without destroying the other will
likely cause your program to segfault.
</p>
<p>
To help ensure that no references to the Python object remain after
calling <tt>__disown__()</tt>, this method returns a weak reference to
the Python object. Weak references are only available in Python versions
2.1 and higher, so for older versions you must explicitly delete all
references. Here is an example:
</p>
<div class="code">
<pre>
class Foo {
public:
...
};
class FooContainer {
public:
void addFoo(Foo *);
...
};
</pre>
</div>
<br>
<div class="targetlang">
<pre>
>>> c = FooContainer()
>>> a = Foo().__disown__()
>>> c.addFoo(a)
>>> b = Foo()
>>> b = b.__disown__()
>>> c.addFoo(b)
>>> c.addFoo(Foo().__disown__())
</pre>
</div>
<p>
In this example, we are assuming that FooContainer will take care of
deleting all the Foo pointers it contains at some point. Note that no hard
references to the Foo objects remain in Python.
</p>
<H3><a name="Python_nn36">33.5.4 Exception unrolling</a></H3>
<p>
With directors routing method calls to Python, and proxies routing them
to C++, the handling of exceptions is an important concern. By default, the
directors ignore exceptions that occur during method calls that are
resolved in Python. To handle such exceptions correctly, it is necessary
to temporarily translate them into C++ exceptions. This can be done with
the %feature("director:except") directive. The following code should
suffice in most cases:
</p>
<div class="code">
<pre>
%feature("director:except") {
if ($error != NULL) {
throw Swig::DirectorMethodException();
}
}
</pre>
</div>
<p>
This code will check the Python error state after each method call from
a director into Python, and throw a C++ exception if an error occurred.
This exception can be caught in C++ to implement an error handler.
Currently no information about the Python error is stored in the
Swig::DirectorMethodException object, but this will likely change in
the future.
</p>
<p>
It may be the case that a method call originates in Python, travels up
to C++ through a proxy class, and then back into Python via a director
method. If an exception occurs in Python at this point, it would be nice
for that exception to find its way back to the original caller. This can
be done by combining a normal %exception directive with the
<tt>director:except</tt> handler shown above. Here is an example of a
suitable exception handler:
</p>
<div class="code">
<pre>
%exception {
try { $action }
catch (Swig::DirectorException &e) { SWIG_fail; }
}
</pre>
</div>
<p>
The class Swig::DirectorException used in this example is actually a
base class of Swig::DirectorMethodException, so it will trap this
exception. Because the Python error state is still set when
Swig::DirectorMethodException is thrown, Python will register the
exception as soon as the C wrapper function returns.
</p>
<H3><a name="Python_nn37">33.5.5 Overhead and code bloat</a></H3>
<p>
Enabling directors for a class will generate a new director method for
every virtual method in the class' inheritance chain. This alone can
generate a lot of code bloat for large hierarchies. Method arguments
that require complex conversions to and from target language types can
result in large director methods. For this reason it is recommended that
you selectively enable directors only for specific classes that are
likely to be extended in Python and used in C++.
</p>
<p>
Compared to classes that do not use directors, the call routing in the
director methods does add some overhead. In particular, at least one
dynamic cast and one extra function call occurs per method call from
Python. Relative to the speed of Python execution this is probably
completely negligible. For worst case routing, a method call that
ultimately resolves in C++ may take one extra detour through Python in
order to ensure that the method does not have an extended Python
implementation. This could result in a noticeable overhead in some cases.
</p>
<p>
Although directors make it natural to mix native C++ objects with Python
objects (as director objects) via a common base class pointer, one
should be aware of the obvious fact that method calls to Python objects
will be much slower than calls to C++ objects. This situation can be
optimized by selectively enabling director methods (using the %feature
directive) for only those methods that are likely to be extended in
Python.
</p>
<H3><a name="Python_nn38">33.5.6 Typemaps</a></H3>
<p>
Typemaps for input and output of most of the basic types from director
classes have been written. These are roughly the reverse of the usual
input and output typemaps used by the wrapper code. The typemap
operation names are 'directorin', 'directorout', and 'directorargout'.
The director code does not currently use any of the other kinds of typemaps.
It is not clear at this point which kinds are appropriate and
need to be supported.
</p>
<H3><a name="Python_nn39">33.5.7 Miscellaneous</a></H3>
<p>
Director typemaps for STL classes are in place, and hence you should
be able to use std::vector, std::string, etc., as you would any other type.
</p>
<p>
<b>Note:</b> The director typemaps for return types based in const
references, such as
<div class="code">
<pre>
class Foo {
…
virtual const int& bar();
…
};
</pre>
</div>
<p>
will work only for simple call scenarios. Usually the resulting code
is neither thread or reentrant safe. Hence, the user is advised to
avoid returning const references in director methods. For example,
the user could modify the method interface to use lvalue return
types, wherever possible, for example
</p>
<div class="code">
<pre>
class Foo {
…
virtual int bar();
…
};
</pre>
</div>
<p>
If that is not possible, the user should avoid enabling the
director feature for reentrant, recursive or threaded member
methods that return const references.
</p>
<H2><a name="Python_nn40">33.6 Common customization features</a></H2>
<p>
The last section presented the absolute basics of C/C++ wrapping. If
you do nothing but feed SWIG a header file, you will get an interface
that mimics the behavior described. However, sometimes this isn't
enough to produce a nice module. Certain types of functionality might
be missing or the interface to certain functions might be awkward.
This section describes some common SWIG features that are used to
improve your the interface to an extension module.
</p>
<H3><a name="Python_nn41">33.6.1 C/C++ helper functions</a></H3>
<p>
Sometimes when you create a module, it is missing certain bits of functionality. For
example, if you had a function like this
</p>
<div class="code">
<pre>
void set_transform(Image *im, double m[4][4]);
</pre>
</div>
<p>
it would be accessible from Python, but there may be no easy way to call it.
For example, you might get errors like this:
</p>
<div class="targetlang">
<pre>
>>> a = [
... [1, 0, 0, 0],
... [0, 1, 0, 0],
... [0, 0, 1, 0],
... [0, 0, 0, 1]]
>>> set_transform(im, a)
Traceback (most recent call last):
File "<stdin>", line 1, in ?
TypeError: Type error. Expected _p_a_4__double
</pre>
</div>
<p>
The problem here is that there is no easy way to construct and manipulate a suitable
<tt>double [4][4]</tt> value to use. To fix this, you can write some extra C helper
functions. Just use the <tt>%inline</tt> directive. For example:
</p>
<div class="code">
<pre>
%inline %{
/* Note: double[4][4] is equivalent to a pointer to an array double (*)[4] */
double (*new_mat44())[4] {
return (double (*)[4]) malloc(16*sizeof(double));
}
void free_mat44(double (*x)[4]) {
free(x);
}
void mat44_set(double x[4][4], int i, int j, double v) {
x[i][j] = v;
}
double mat44_get(double x[4][4], int i, int j) {
return x[i][j];
}
%}
</pre>
</div>
<p>
From Python, you could then write code like this:
</p>
<div class="targetlang">
<pre>
>>> a = new_mat44()
>>> mat44_set(a, 0, 0, 1.0)
>>> mat44_set(a, 1, 1, 1.0)
>>> mat44_set(a, 2, 2, 1.0)
...
>>> set_transform(im, a)
>>>
</pre>
</div>
<p>
Admittedly, this is not the most elegant looking approach. However, it works and it wasn't too
hard to implement. It is possible to clean this up using Python code, typemaps, and other
customization features as covered in later sections.
</p>
<H3><a name="Python_nn42">33.6.2 Adding additional Python code</a></H3>
<p>
If writing support code in C isn't enough, it is also possible to write code in
Python. This code gets inserted in to the <tt>.py</tt> file created by SWIG. One
use of Python code might be to supply a high-level interface to certain functions.
For example:
</p>
<div class="code">
<pre>
void set_transform(Image *im, double x[4][4]);
...
/* Rewrite the high level interface to set_transform */
%pythoncode %{
def set_transform(im, x):
a = new_mat44()
for i in range(4):
for j in range(4):
mat44_set(a, i, j, x[i][j])
_example.set_transform(im, a)
free_mat44(a)
%}
</pre>
</div>
<p>
In this example, <tt>set_transform()</tt> provides a high-level Python interface built on top of
low-level helper functions. For example, this code now seems to work:
</p>
<div class="targetlang">
<pre>
>>> a = [
... [1, 0, 0, 0],
... [0, 1, 0, 0],
... [0, 0, 1, 0],
... [0, 0, 0, 1]]
>>> set_transform(im, a)
>>>
</pre>
</div>
<p>
Admittedly, this whole scheme for wrapping the two-dimension array
argument is rather ad-hoc. Besides, shouldn't a Python list or a
Numeric Python array just work normally? We'll get to those examples
soon enough. For now, think of this example as an illustration of
what can be done without having to rely on any of the more advanced
customization features.
</p>
<p>
There is also <tt>%pythonbegin</tt> which is another directive very similar to <tt>%pythoncode</tt>,
but generates the given Python code at the beginning of the <tt>.py</tt> file.
This directive works in the same way as <tt>%pythoncode</tt>, except the code is copied
just after the SWIG banner (comment) at the top of the file, before any real code.
This provides an opportunity to add your own description in a comment near the top of the file as well
as Python imports that have to appear at the top of the file, such as "<tt>from __future__ import</tt>"
statements.
</p>
<p>
The following example for Python 2.x shows how to insert code into the
generated wrapper to enable <tt>print</tt> to be used as a Python3-compatible
function instead of a statement:
</p>
<div class="code">
<pre>
%pythonbegin %{
# This module provides wrappers to the Whizz Bang library
%}
%pythonbegin %{
from __future__ import print_function
print("Loading", "Whizz", "Bang", sep=' ... ')
%}
</pre>
</div>
<p>
The insert code can be seen at the start of the generated <tt>.py</tt> file:
</p>
<div class="code">
<pre>
# This file was automatically generated by SWIG (http://www.swig.org).
# Version 4.0.0
#
# Do not make changes to this file unless you know what you are doing--modify
# the SWIG interface file instead.
# This module provides wrappers to the Whizz Bang library
from __future__ import print_function
print("Loading", "Whizz", "Bang", sep=' ... ')
</pre>
</div>
<p>When using <tt>%pythoncode</tt> and <tt>%pythonbegin</tt> you generally
want to make sure that the block is delimited by <tt>%{</tt> and <tt>%}</tt>.
If you delimit it with <tt>{</tt> and <tt>}</tt> then any lines with a
leading <tt>#</tt> will be handled by SWIG as preprocessor directives, when
you probably meant them as Python comments. Prior to SWIG 3.0.3, invalid
preprocessor directives were silently ignored, so generally using the wrong
delimiters resulted in such comments not appearing in the generated output
(though a comment starting with a valid preprocessor directive could cause
problems, for example: <tt># error handling</tt>). SWIG 3.0.3 and later report
an error for invalid preprocessor directives, so you may have to update
existing interface files to delimit blocks of Python code correctly.</p>
<p>As an alternative to providing a block containing Python code, you can
include Python code from a file. The code is inserted exactly as in the
file, so this avoids any issues with the SWIG preprocessor. It's a good
approach if you have a non-trivial chunk of Python code to insert. To
use this feature you specify a filename in double quotes, for example:</p>
<div class="code">
<pre>
%pythoncode "somecode.py"
</pre>
</div>
<p>Sometimes you may want to replace or modify the wrapper function
that SWIG creates in the proxy <tt>.py</tt> file. The Python module
in SWIG provides some features that enable you to do this. First, to
entirely replace a proxy function you can use
<tt>%feature("shadow")</tt>. For example:</p>
<div class="code">
<pre>
%module example
// Rewrite bar() Python code
%feature("shadow") Foo::bar(int) %{
def bar(*args):
#do something before
$action
#do something after
%}
class Foo {
public:
int bar(int x);
};
</pre>
</div>
<p> where <tt>$action</tt> will be replaced by the call to
the C/C++ proper method.
</p>
<p>
Often the proxy function created by SWIG is fine, but you simply want
to add code to it without touching the rest of the generated function
body. For these cases SWIG provides the <tt>pythonprepend</tt> and
<tt>pythonappend</tt> features which do exactly as their names suggest. The
<tt>pythonprepend</tt> feature will insert its value at the beginning of the
proxy function, and <tt>pythonappend</tt> will insert code at the end of the
proxy, just before the return statement.
</p>
<div class="code">
<pre>
%module example
// Add Python code to bar()
%feature("pythonprepend") Foo::bar(int) %{
#do something before C++ call
%}
%feature("pythonappend") Foo::bar(int) %{
#do something after C++ call
#the 'val' variable holds the return value
%}
class Foo {
public:
int bar(int x);
};
</pre>
</div>
<p>
Notes: Usually the <tt>pythonappend</tt> and <tt>pythonprepend</tt>
features are safer to use than the <tt>shadow</tt> feature. Also, from
SWIG version 1.3.28 you can use the directive forms
<tt>%pythonappend</tt> and <tt>%pythonprepend</tt> as follows:</p>
<div class="code">
<pre>
%module example
// Add Python code to bar()
%pythonprepend Foo::bar(int) %{
#do something before C++ call
%}
%pythonappend Foo::bar(int) %{
#do something after C++ call
#the 'val' variable holds the return value
%}
class Foo {
public:
int bar(int x);
};
</pre>
</div>
<p>
Note that when the underlying C++ method is overloaded, there is only one proxy Python method
for multiple C++ methods. In this case, only one of parsed methods is examined
for the feature. You are better off specifying the feature without the argument list to ensure it will get used,
as it will then get attached to all the overloaded C++ methods. For example:
</p>
<div class="code">
<pre>
%module example
// Add Python code to bar()
%pythonprepend Foo::bar %{
#do something before C++ call
%}
%pythonappend Foo::bar %{
#do something after C++ call
%}
class Foo {
public:
int bar(int x);
int bar();
};
</pre>
</div>
<p>
The same applies for overloaded constructors.
</p>
<H3><a name="Python_nn43">33.6.3 Class extension with %extend</a></H3>
<p>
One of the more interesting features of SWIG is that it can extend
structures and classes with new methods--at least in the Python interface.
Here is a simple example:
</p>
<div class="code">
<pre>
%module example
%{
#include "someheader.h"
%}
struct Vector {
double x, y, z;
};
%extend Vector {
char *__str__() {
static char tmp[1024];
sprintf(tmp, "Vector(%g, %g, %g)", $self->x, $self->y, $self->z);
return tmp;
}
Vector(double x, double y, double z) {
Vector *v = (Vector *) malloc(sizeof(Vector));
v->x = x;
v->y = y;
v->z = z;
return v;
}
};
</pre>
</div>
<p>
Now, in Python
</p>
<div class="targetlang">
<pre>
>>> v = example.Vector(2, 3, 4)
>>> print v
Vector(2, 3, 4)
>>>
</pre>
</div>
<p>
<tt>%extend</tt> can be used for many more tasks than this.
For example, if you wanted to overload a Python operator, you might do this:
</p>
<div class="code">
<pre>
%extend Vector {
Vector __add__(Vector *other) {
Vector v;
v.x = $self->x + other->x;
v.y = $self->y + other->y;
v.z = $self->z + other->z;
return v;
}
};
</pre>
</div>
<p>
Use it like this:
</p>
<div class="targetlang">
<pre>
>>> import example
>>> v = example.Vector(2, 3, 4)
>>> w = example.Vector(10, 11, 12)
>>> print v+w
Vector(12, 14, 16)
>>>
</pre>
</div>
<p>
<tt>%extend</tt> works with both C and C++ code. It does not modify the underlying object
in any way---the extensions only show up in the Python interface.
</p>
<H3><a name="Python_nn44">33.6.4 Exception handling with %exception</a></H3>
<p>
If a C or C++ function throws an error, you may want to convert that error into a Python
exception. To do this, you can use the <tt>%exception</tt> directive. <tt>%exception</tt>
simply lets you rewrite part of the generated wrapper code to include an error check.
</p>
<p>
In C, a function often indicates an error by returning a status code (a negative number
or a NULL pointer perhaps). Here is a simple example of how you might handle that:
</p>
<div class="code">
<pre>
%exception malloc {
$action
if (!result) {
PyErr_SetString(PyExc_MemoryError, "Not enough memory");
SWIG_fail;
}
}
void *malloc(size_t nbytes);
</pre>
</div>
<p>
In Python,
</p>
<div class="targetlang">
<pre>
>>> a = example.malloc(2000000000)
Traceback (most recent call last):
File "<stdin>", line 1, in ?
MemoryError: Not enough memory
>>>
</pre>
</div>
<p>
If a library provides some kind of general error handling framework, you can also use
that. For example:
</p>
<div class="code">
<pre>
%exception {
$action
if (err_occurred()) {
PyErr_SetString(PyExc_RuntimeError, err_message());
SWIG_fail;
}
}
</pre>
</div>
<p>
No declaration name is given to <tt>%exception</tt>, it is applied to all wrapper functions.
</p>
<p>
C++ exceptions are also easy to handle. For example, you can write code like this:
</p>
<div class="code">
<pre>
%exception getitem {
try {
$action
} catch (std::out_of_range &e) {
PyErr_SetString(PyExc_IndexError, const_cast<char*>(e.what()));
SWIG_fail;
}
}
class Base {
public:
Foo *getitem(int index); // Exception handled added
...
};
</pre>
</div>
<p>
When raising a Python exception from C, use the <tt>PyErr_SetString()</tt>
function as shown above followed by <tt>SWIG_fail</tt>.
The following exception types can be used as the first argument.
</p>
<div class="code">
<pre>
PyExc_ArithmeticError
PyExc_AssertionError
PyExc_AttributeError
PyExc_EnvironmentError
PyExc_EOFError
PyExc_Exception
PyExc_FloatingPointError
PyExc_ImportError
PyExc_IndexError
PyExc_IOError
PyExc_KeyError
PyExc_KeyboardInterrupt
PyExc_LookupError
PyExc_MemoryError
PyExc_NameError
PyExc_NotImplementedError
PyExc_OSError
PyExc_OverflowError
PyExc_RuntimeError
PyExc_StandardError
PyExc_SyntaxError
PyExc_SystemError
PyExc_TypeError
PyExc_UnicodeError
PyExc_ValueError
PyExc_ZeroDivisionError
</pre>
</div>
<p>
<tt>SWIG_fail</tt> is a C macro which when called within the context of SWIG wrapper function,
will jump to the error handler code. This will call any cleanup code (freeing any temp variables)
and then return from the wrapper function so that the Python interpreter can raise the Python exception.
This macro should always be called after setting a Python error in code snippets, such as typemaps and <tt>%exception</tt>, that are ultimately generated into the wrapper function.
</p>
<p>
The language-independent <tt>exception.i</tt> library file can also be used
to raise exceptions. See the <a href="Library.html#Library">SWIG Library</a> chapter.
</p>
<H3><a name="Python_optimization">33.6.5 Optimization options</a></H3>
<H4><a name="Python_fastproxy">33.6.5.1 -fastproxy</a></H4>
<p>
The <tt>-fastproxy</tt> command line option enables faster method calling as the call is made directly into the C/C++ layer rather than going through a method wrapper.
</p>
<p>
Consider wrapping a C++ class:
</p>
<div class="code">
<pre>
struct Go {
void callme0() {}
void callme4(int a, int b, int c, int d) {}
void callme8(double a, double b, double c, double d, double e, double f, double g, double i) {}
};
</pre>
</div>
<p>
The default generated proxy class is:
</p>
<div class="targetlang">
<pre>
class Go(object):
def callme0(self):
return _example.Go_callme0(self)
def callme4(self, a, b, c, d):
return _example.Go_callme4(self, a, b, c, d)
def callme8(self, a, b, c, d, e, f, g, i):
return _example.Go_callme8(self, a, b, c, d, e, f, g, i)
...
</pre>
</div>
<p>
Each method in the Python class contains a Python proxy method which passes the arguments on to the underlying function in the low-level C/C++ module (_example in this case).
The generated proxy class when using <tt>-fastproxy</tt> is:
</p>
<div class="targetlang">
<pre>
%module example
class Go(object):
callme0 = _swig_new_instance_method(_example.Go_callme0)
callme4 = _swig_new_instance_method(_example.Go_callme4)
callme8 = _swig_new_instance_method(_example.Go_callme8)
...
</pre>
</div>
<p>
where <tt>_swig_new_instance_method</tt> adds the method to the proxy class via C API calls for direct access to the underlying function in the low-level C/C++ module.
Note that for some methods it is not possible to generate the direct access call and so <tt>-fastproxy</tt> is ignored.
This happens, for example, when adding <a href="#Python_nn42">additional code</a> to Python proxy methods, such as using <tt>%pythonprepend</tt>.
</p>
<p>
The overhead calling into C/C++ from Python is reduced slightly using <tt>-fastproxy</tt>.
Below are some timings in microseconds calling the 3 functions in the example above.
Also included in the table for comparison is using the <tt>-builtin</tt> option covered in the
<a href="#Python_builtin_types">Built-in Types</a>.
</p>
<table summary="Python fastproxy performance">
<tr>
<th>Method name</th>
<th>Default</th>
<th>-fastproxy</th>
<th>-builtin</th>
</tr>
<tr>
<td>callme0</td>
<td>0.15</td>
<td>0.09</td>
<td>0.07</td>
</tr>
<tr>
<td>callme4</td>
<td>0.26</td>
<td>0.16</td>
<td>0.14</td>
</tr>
<tr>
<td>callme8</td>
<td>0.32</td>
<td>0.20</td>
<td>0.17</td>
</tr>
</table>
<p>
Although the <tt>-fastproxy</tt> option results in faster code over the default, the generated proxy code is not as user-friendly
as docstring/doxygen comments and functions with default values are not visible in the generated Python proxy class.
The <tt>-olddefs</tt> option can rectify this.
</p>
<p>
The generated proxy class for the example above when using <tt>-fastproxy -olddefs</tt> is:
</p>
<div class="targetlang">
<pre>
class Go(object):
def callme0(self):
return _example.Go_callme0(self)
callme0 = _swig_new_instance_method(_example.Go_callme0)
def callme4(self, a, b, c, d):
return _example.Go_callme4(self, a, b, c, d)
callme4 = _swig_new_instance_method(_example.Go_callme4)
def callme8(self, a, b, c, d, e, f, g, i):
return _example.Go_callme8(self, a, b, c, d, e, f, g, i)
callme8 = _swig_new_instance_method(_example.Go_callme8)
...
</pre>
</div>
<p>
The class defines each method in two different ways. The first definition is replaced by the second definition and so the second definition is the one used when the method is called.
While this possibly provides the best of both worlds, the time to import the module will be slightly slower when the class is defined due to the additional method definitions.
</p>
<p>
The command line options mentioned above also apply to wrapped C/C++ global functions, not just class methods.
</p>
<H2><a name="Python_nn45">33.7 Tips and techniques</a></H2>
<p>
Although SWIG is largely automatic, there are certain types of wrapping problems that
require additional user input. Examples include dealing with output parameters,
strings, binary data, and arrays. This chapter discusses the common techniques for
solving these problems.
</p>
<H3><a name="Python_nn46">33.7.1 Input and output parameters</a></H3>
<p>
A common problem in some C programs is handling parameters passed as simple pointers. For
example:
</p>
<div class="code">
<pre>
void add(int x, int y, int *result) {
*result = x + y;
}
</pre>
</div>
<p>
or perhaps
</p>
<div class="code">
<pre>
int sub(int *x, int *y) {
return *x-*y;
}
</pre>
</div>
<p>
The easiest way to handle these situations is to use the <tt>typemaps.i</tt> file. For example:
</p>
<div class="code">
<pre>
%module example
%include "typemaps.i"
void add(int, int, int *OUTPUT);
int sub(int *INPUT, int *INPUT);
</pre>
</div>
<p>
In Python, this allows you to pass simple values. For example:
</p>
<div class="targetlang">
<pre>
>>> a = add(3, 4)
>>> print a
7
>>> b = sub(7, 4)
>>> print b
3
>>>
</pre>
</div>
<p>
Notice how the <tt>INPUT</tt> parameters allow integer values to be passed instead of pointers
and how the <tt>OUTPUT</tt> parameter creates a return result.
</p>
<p>
If you don't want to use the names <tt>INPUT</tt> or <tt>OUTPUT</tt>, use the <tt>%apply</tt>
directive. For example:
</p>
<div class="code">
<pre>
%module example
%include "typemaps.i"
%apply int *OUTPUT { int *result };
%apply int *INPUT { int *x, int *y};
void add(int x, int y, int *result);
int sub(int *x, int *y);
</pre>
</div>
<p>
If a function mutates one of its parameters like this,
</p>
<div class="code">
<pre>
void negate(int *x) {
*x = -(*x);
}
</pre>
</div>
<p>
you can use <tt>INOUT</tt> like this:
</p>
<div class="code">
<pre>
%include "typemaps.i"
...
void negate(int *INOUT);
</pre>
</div>
<p>
In Python, a mutated parameter shows up as a return value. For example:
</p>
<div class="targetlang">
<pre>
>>> a = negate(3)
>>> print a
-3
>>>
</pre>
</div>
<p>
Note: Since most primitive Python objects are immutable, it is not possible to
perform in-place modification of a Python object passed as a parameter.
</p>
<p>
The most common use of these special typemap rules is to handle functions that
return more than one value. For example, sometimes a function returns a result
as well as a special error code:
</p>
<div class="code">
<pre>
/* send message, return number of bytes sent, along with success code */
int send_message(char *text, int len, int *success);
</pre>
</div>
<p>
To wrap such a function, simply use the <tt>OUTPUT</tt> rule above. For example:
</p>
<div class="code">
<pre>
%module example
%include "typemaps.i"
%apply int *OUTPUT { int *success };
...
int send_message(char *text, int *success);
</pre>
</div>
<p>
When used in Python, the function will return multiple values.
</p>
<div class="targetlang">
<pre>
bytes, success = send_message("Hello World")
if not success:
print "Whoa!"
else:
print "Sent", bytes
</pre>
</div>
<p>
Another common use of multiple return values are in query functions. For example:
</p>
<div class="code">
<pre>
void get_dimensions(Matrix *m, int *rows, int *columns);
</pre>
</div>
<p>
To wrap this, you might use the following:
</p>
<div class="code">
<pre>
%module example
%include "typemaps.i"
%apply int *OUTPUT { int *rows, int *columns };
...
void get_dimensions(Matrix *m, int *rows, *columns);
</pre>
</div>
<p>
Now, in Python:
</p>
<div class="targetlang">
<pre>
>>> r, c = get_dimensions(m)
</pre>
</div>
<p>
Be aware that the primary purpose of the <tt>typemaps.i</tt> file is to support primitive datatypes.
Writing a function like this
</p>
<div class="code">
<pre>
void foo(Bar *OUTPUT);
</pre>
</div>
<p>
may not have the intended effect since <tt>typemaps.i</tt> does not define an OUTPUT rule for <tt>Bar</tt>.
</p>
<H3><a name="Python_nn47">33.7.2 Simple pointers</a></H3>
<p>
If you must work with simple pointers such as <tt>int *</tt> or <tt>double *</tt> and you don't want to use
<tt>typemaps.i</tt>, consider using the <tt>cpointer.i</tt> library file. For example:
</p>
<div class="code">
<pre>
%module example
%include "cpointer.i"
%inline %{
extern void add(int x, int y, int *result);
%}
%pointer_functions(int, intp);
</pre>
</div>
<p>
The <tt>%pointer_functions(type, name)</tt> macro generates five helper functions that can be used to create,
destroy, copy, assign, and dereference a pointer. In this case, the functions are as follows:
</p>
<div class="code">
<pre>
int *new_intp();
int *copy_intp(int *x);
void delete_intp(int *x);
void intp_assign(int *x, int value);
int intp_value(int *x);
</pre>
</div>
<p>
In Python, you would use the functions like this:
</p>
<div class="targetlang">
<pre>
>>> result = new_intp()
>>> print result
_108fea8_p_int
>>> add(3, 4, result)
>>> print intp_value(result)
7
>>>
</pre>
</div>
<p>
If you replace <tt>%pointer_functions()</tt> by <tt>%pointer_class(type, name)</tt>, the interface is more class-like.
</p>
<div class="targetlang">
<pre>
>>> result = intp()
>>> add(3, 4, result)
>>> print result.value()
7
</pre>
</div>
<p>
See the <a href="Library.html#Library">SWIG Library</a> chapter for further details.
</p>
<H3><a name="Python_nn48">33.7.3 Unbounded C Arrays</a></H3>
<p>
Sometimes a C function expects an array to be passed as a pointer. For example,
</p>
<div class="code">
<pre>
int sumitems(int *first, int nitems) {
int i, sum = 0;
for (i = 0; i < nitems; i++) {
sum += first[i];
}
return sum;
}
</pre>
</div>
<p>
To wrap this into Python, you need to pass an array pointer as the first argument.
A simple way to do this is to use the <tt>carrays.i</tt> library file. For example:
</p>
<div class="code">
<pre>
%include "carrays.i"
%array_class(int, intArray);
</pre>
</div>
<p>
The <tt>%array_class(type, name)</tt> macro creates wrappers for an unbounded array object that
can be passed around as a simple pointer like <tt>int *</tt> or <tt>double *</tt>.
For instance, you will be able to do this in Python:
</p>
<div class="targetlang">
<pre>
>>> a = intArray(10000000) # Array of 10-million integers
>>> for i in xrange(10000): # Set some values
... a[i] = i
>>> sumitems(a, 10000)
49995000
>>>
</pre>
</div>
<p>
The array "object" created by <tt>%array_class()</tt> does not
encapsulate pointers inside a special array object. In fact, there is
no bounds checking or safety of any kind (just like in C). Because of
this, the arrays created by this library are extremely low-level
indeed. You can't iterate over them nor can you even query their
length. In fact, any valid memory address can be accessed if you want
(negative indices, indices beyond the end of the array, etc.).
Needless to say, this approach is not going to suit all applications.
On the other hand, this low-level approach is extremely efficient and
well suited for applications in which you need to create buffers,
package binary data, etc.
</p>
<H3><a name="Python_nn49">33.7.4 String handling</a></H3>
<p>
If a C function has an argument of <tt>char *</tt>, then a Python string
can be passed as input. For example:
</p>
<div class="code">
<pre>
// C
void foo(char *s);
</pre>
</div>
<div class="targetlang">
<pre>
# Python
>>> foo("Hello")
</pre>
</div>
<p>
When a Python string is passed as a parameter, the C function receives a pointer to the raw
data contained in the string. Since Python strings are immutable, it is illegal
for your program to change the value. In fact, doing so will probably crash the Python
interpreter.
</p>
<p>
If your program modifies the input parameter or uses it to return data, consider
using the <tt>cstring.i</tt> library file described in the <a href="Library.html#Library">SWIG Library</a> chapter.
</p>
<p>
When functions return a <tt>char *</tt>, it is assumed to be a NULL-terminated string.
Data is copied into a new Python string and returned.
</p>
<p>
If your program needs to work with binary data, you can use a typemap
to expand a Python string into a pointer/length argument pair. As luck would have it,
just such a typemap is already defined. Just do this:
</p>
<div class="code">
<pre>
%apply (char *STRING, int LENGTH) { (char *data, int size) };
...
int parity(char *data, int size, int initial);
</pre>
</div>
<p>
Now in Python:
</p>
<div class="targetlang">
<pre>
>>> parity("e\x09ffss\x00\x00\x01\nx", 0)
</pre>
</div>
<p>
If you need to return binary data, you might use the
<tt>cstring.i</tt> library file. The <tt>cdata.i</tt> library can
also be used to extra binary data from arbitrary pointers.
</p>
<H3><a name="Python_default_args">33.7.5 Default arguments</a></H3>
<p>
C++ default argument code generation is documented in the main
<a href="SWIGPlus.html#SWIGPlus_default_args">Default arguments</a> section.
There is also an optional Python specific feature that can be used called the <tt>python:cdefaultargs</tt>
<a href="Customization.html#Customization_feature_flags">feature flag</a>.
By default, SWIG attempts to convert C++ default argument values
into Python values and generates code into the Python layer containing these values.
For example:
</p>
<div class="code">
<pre>
struct CDA {
int fff(int a = 1, bool b = false);
};
</pre>
</div>
<p>
From Python this can be called as follows:
</p>
<div class="targetlang">
<pre>
>>> CDA().fff() # C++ layer receives a=1 and b=false
>>> CDA().fff(2) # C++ layer receives a=2 and b=false
>>> CDA().fff(3, True) # C++ layer receives a=3 and b=true
</pre>
</div>
<p>
The default code generation in the Python layer is:
</p>
<div class="targetlang">
<pre>
class CDA(object):
...
def fff(self, a=1, b=False):
return _default_args.CDA_fff(self, a, b)
</pre>
</div>
<p>
Adding the feature:
</p>
<div class="code">
<pre>
%feature("python:cdefaultargs") CDA::fff;
struct CDA {
int fff(int a = 1, bool b = false);
</pre>
</div>
<p>
results in identical behaviour when called from Python, however, it results in different code generation:
</p>
<div class="targetlang">
<pre>
class CDA(object):
...
def fff(self, *args):
return _default_args.CDA_fff(self, *args)
</pre>
</div>
<p>
The default arguments are obtained in the C++ wrapper layer instead of the Python layer.
Some code generation modes are quite different, eg <tt>-builtin</tt> and <tt>-fastproxy</tt>,
and are unaffected by <tt>python:cdefaultargs</tt> as the default values are always obtained from the C++ layer.
</p>
<p>
Note that not all default arguments can be converted into a Python equivalent.
When SWIG does not convert them, it will generate code to obtain them from the C++ layer as if
<tt>python:cdefaultargs</tt> was specified.
This will happen if just one argument cannot be converted into a Python equivalent.
This occurs typically when the argument is not fully numeric, such as <tt>int(1)</tt>:
</p>
<div class="code">
<pre>
struct CDA {
int fff(int a = int(1), bool b = false);
};
</pre>
</div>
<p>
<b>Compatibility Note:</b> SWIG-3.0.6 introduced the <tt>python:cdefaultargs</tt> feature.
Versions of SWIG prior to this varied in their ability to convert C++ default values into
equivalent Python default argument values.
</p>
<H2><a name="Python_nn53">33.8 Typemaps</a></H2>
<p>
This section describes how you can modify SWIG's default wrapping behavior
for various C/C++ datatypes using the <tt>%typemap</tt> directive. This
is an advanced topic that assumes familiarity with the Python C API as well
as the material in the "<a href="Typemaps.html#Typemaps">Typemaps</a>" chapter.
</p>
<p>
Before proceeding, it should be stressed that typemaps are not a required
part of using SWIG---the default wrapping behavior is enough in most cases.
Typemaps are only used if you want to change some aspect of the primitive
C-Python interface or if you want to elevate your guru status.
</p>
<H3><a name="Python_nn54">33.8.1 What is a typemap?</a></H3>
<p>
A typemap is nothing more than a code generation rule that is attached to
a specific C datatype. For example, to convert integers from Python to C,
you might define a typemap like this:
</p>
<div class="code"><pre>
%module example
%typemap(in) int {
$1 = (int) PyLong_AsLong($input);
printf("Received an integer : %d\n", $1);
}
%inline %{
extern int fact(int n);
%}
</pre></div>
<p>
Typemaps are always associated with some specific aspect of code generation.
In this case, the "in" method refers to the conversion of input arguments
to C/C++. The datatype <tt>int</tt> is the datatype to which the typemap
will be applied. The supplied C code is used to convert values. In this
code a number of special variable prefaced by a <tt>$</tt> are used. The
<tt>$1</tt> variable is placeholder for a local variable of type <tt>int</tt>.
The <tt>$input</tt> variable is the input object of type <tt>PyObject *</tt>.
</p>
<p>
When this example is compiled into a Python module, it operates as follows:
</p>
<div class="targetlang"><pre>
>>> from example import *
>>> fact(6)
Received an integer : 6
720
</pre></div>
<p>
In this example, the typemap is applied to all occurrences of the <tt>int</tt> datatype.
You can refine this by supplying an optional parameter name. For example:
</p>
<div class="code"><pre>
%module example
%typemap(in) int nonnegative {
$1 = (int) PyLong_AsLong($input);
if ($1 < 0) {
PyErr_SetString(PyExc_ValueError, "Expected a nonnegative value.");
SWIG_fail;
}
}
%inline %{
extern int fact(int nonnegative);
%}
</pre></div>
<p>
In this case, the typemap code is only attached to arguments that exactly match <tt>int nonnegative</tt>.
</p>
<p>
The application of a typemap to specific datatypes and argument names involves
more than simple text-matching--typemaps are fully integrated into the
SWIG C++ type-system. When you define a typemap for <tt>int</tt>, that typemap
applies to <tt>int</tt> and qualified variations such as <tt>const int</tt>. In addition,
the typemap system follows <tt>typedef</tt> declarations. For example:
</p>
<div class="code">
<pre>
%typemap(in) int n {
$1 = (int) PyLong_AsLong($input);
printf("n = %d\n", $1);
}
%inline %{
typedef int Integer;
extern int fact(Integer n); // Above typemap is applied
%}
</pre>
</div>
<p>
Typemaps can also be defined for groups of consecutive arguments. For example:
</p>
<div class="code">
<pre>
%typemap(in) (char *str, int len) {
$1 = PyString_AsString($input);
$2 = PyString_Size($input);
};
int count(char c, char *str, int len);
</pre>
</div>
<p>
When a multi-argument typemap is defined, the arguments are always handled as a single
Python object. This allows the function to be used like this (notice how the length
parameter is omitted):
</p>
<div class="targetlang">
<pre>
>>> example.count('e', 'Hello World')
1
>>>
</pre>
</div>
<H3><a name="Python_nn55">33.8.2 Python typemaps</a></H3>
<p>
The previous section illustrated an "in" typemap for converting Python objects to C.
A variety of different typemap methods are defined by the Python module. For example,
to convert a C integer back into a Python object, you might define an "out" typemap
like this:
</p>
<div class="code">
<pre>
%typemap(out) int {
$result = PyInt_FromLong((long) $1);
}
</pre>
</div>
<p>
A detailed list of available methods can be found in the "<a
href="Typemaps.html#Typemaps">Typemaps</a>" chapter.
</p>
<p>
However, the best source of typemap information (and examples) is
probably the Python module itself. In fact, all of SWIG's default
type handling is defined by typemaps. You can view these typemaps by
looking at the files in the SWIG library. Just take into account that
in the latest versions of swig (1.3.22+), the library files are not
very pristine clear for the casual reader, as they used to be. The
extensive use of macros and other ugly techniques in the latest
version produce a very powerful and consistent Python typemap library,
but at the cost of simplicity and pedagogic value.
</p>
<p>
To learn how to write a simple or your first typemap, you better take
a look at the SWIG library version 1.3.20 or so.
</p>
<H3><a name="Python_nn56">33.8.3 Typemap variables</a></H3>
<p>
Within typemap code, a number of special variables prefaced with a <tt>$</tt> may appear.
A full list of variables can be found in the "<a href="Typemaps.html#Typemaps">Typemaps</a>" chapter.
This is a list of the most common variables:
</p>
<p>
<tt>$1</tt>
</p>
<div class="indent">
A C local variable corresponding to the actual type specified in the
<tt>%typemap</tt> directive. For input values, this is a C local variable
that's supposed to hold an argument value. For output values, this is
the raw result that's supposed to be returned to Python.
</div>
<p>
<tt>$input</tt>
</p>
<div class="indent">
A <tt>PyObject *</tt> holding a raw Python object with an argument or variable value.
</div>
<p>
<tt>$result</tt>
</p>
<div class="indent">
A <tt>PyObject *</tt> that holds the result to be returned to Python.
</div>
<p>
<tt>$1_name</tt>
</p>
<div class="indent">
The parameter name that was matched.
</div>
<p>
<tt>$1_type</tt>
</p>
<div class="indent">
The actual C datatype matched by the typemap.
</div>
<p>
<tt>$1_ltype</tt>
</p>
<div class="indent">
An assignable version of the datatype matched by the typemap (a type that can appear on the left-hand-side of
a C assignment operation). This type is stripped of qualifiers and may be an altered version of <tt>$1_type</tt>.
All arguments and local variables in wrapper functions are declared using this type so that their values can be
properly assigned.
</div>
<p>
<tt>$symname</tt>
</p>
<div class="indent">
The Python name of the wrapper function being created.
</div>
<H3><a name="Python_nn57">33.8.4 Useful Python Functions</a></H3>
<p>
When you write a typemap, you usually have to work directly with Python objects.
The following functions may prove to be useful.
</p>
<p>
<b>Python Integer Functions</b>
</p>
<div class="code">
<pre>
PyObject *PyInt_FromLong(long l);
long PyInt_AsLong(PyObject *);
int PyInt_Check(PyObject *);
</pre>
</div>
<p>
<b>Python Floating Point Functions</b>
</p>
<div class="code">
<pre>
PyObject *PyFloat_FromDouble(double);
double PyFloat_AsDouble(PyObject *);
int PyFloat_Check(PyObject *);
</pre>
</div>
<p>
<b>Python String Functions</b>
</p>
<div class="code">
<pre>
PyObject *PyString_FromString(char *);
PyObject *PyString_FromStringAndSize(char *, lint len);
int PyString_Size(PyObject *);
char *PyString_AsString(PyObject *);
int PyString_Check(PyObject *);
</pre>
</div>
<p>
<b>Python List Functions</b>
</p>
<div class="code">
<pre>
PyObject *PyList_New(int size);
int PyList_Size(PyObject *list);
PyObject *PyList_GetItem(PyObject *list, int i);
int PyList_SetItem(PyObject *list, int i, PyObject *item);
int PyList_Insert(PyObject *list, int i, PyObject *item);
int PyList_Append(PyObject *list, PyObject *item);
PyObject *PyList_GetSlice(PyObject *list, int i, int j);
int PyList_SetSlice(PyObject *list, int i, int , PyObject *list2);
int PyList_Sort(PyObject *list);
int PyList_Reverse(PyObject *list);
PyObject *PyList_AsTuple(PyObject *list);
int PyList_Check(PyObject *);
</pre>
</div>
<p>
<b>Python Tuple Functions</b>
</p>
<div class="code">
<pre>
PyObject *PyTuple_New(int size);
int PyTuple_Size(PyObject *);
PyObject *PyTuple_GetItem(PyObject *, int i);
int PyTuple_SetItem(PyObject *, int i, PyObject *item);
PyObject *PyTuple_GetSlice(PyObject *t, int i, int j);
int PyTuple_Check(PyObject *);
</pre>
</div>
<p>
<b>Python Dictionary Functions</b>
</p>
<div class="code">
<pre>
PyObject *PyDict_New();
int PyDict_Check(PyObject *);
int PyDict_SetItem(PyObject *p, PyObject *key, PyObject *val);
int PyDict_SetItemString(PyObject *p, const char *key, PyObject *val);
int PyDict_DelItem(PyObject *p, PyObject *key);
int PyDict_DelItemString(PyObject *p, char *key);
PyObject* PyDict_Keys(PyObject *p);
PyObject* PyDict_Values(PyObject *p);
PyObject* PyDict_GetItem(PyObject *p, PyObject *key);
PyObject* PyDict_GetItemString(PyObject *p, const char *key);
int PyDict_Next(PyObject *p, Py_ssize_t *ppos, PyObject **pkey, PyObject **pvalue);
Py_ssize_t PyDict_Size(PyObject *p);
int PyDict_Update(PyObject *a, PyObject *b);
int PyDict_Merge(PyObject *a, PyObject *b, int override);
PyObject* PyDict_Items(PyObject *p);
</pre>
</div>
<p>
<b>Python File Conversion Functions</b>
</p>
<div class="code">
<pre>
PyObject *PyFile_FromFile(FILE *f);
FILE *PyFile_AsFile(PyObject *);
int PyFile_Check(PyObject *);
</pre>
</div>
<p>
<b>Abstract Object Interface</b>
</p>
<div class="code">
<pre>
write me
</pre>
</div>
<H2><a name="Python_nn58">33.9 Typemap Examples</a></H2>
<p>
This section includes a few examples of typemaps. For more examples, you
might look at the files "<tt>python.swg</tt>" and "<tt>typemaps.i</tt>" in
the SWIG library.
</p>
<H3><a name="Python_nn59">33.9.1 Converting Python list to a char ** </a></H3>
<p>
A common problem in many C programs is the processing of command line
arguments, which are usually passed in an array of NULL terminated
strings. The following SWIG interface file allows a Python list
object to be used as a <tt>char **</tt> object.
</p>
<div class="code"><pre>
%module argv
// This tells SWIG to treat char ** as a special case
%typemap(in) char ** {
/* Check if is a list */
if (PyList_Check($input)) {
int size = PyList_Size($input);
int i = 0;
$1 = (char **) malloc((size+1)*sizeof(char *));
for (i = 0; i < size; i++) {
PyObject *o = PyList_GetItem($input, i);
if (PyString_Check(o)) {
$1[i] = PyString_AsString(PyList_GetItem($input, i));
} else {
free($1);
PyErr_SetString(PyExc_TypeError, "list must contain strings");
SWIG_fail;
}
}
$1[i] = 0;
} else {
PyErr_SetString(PyExc_TypeError, "not a list");
SWIG_fail;
}
}
// This cleans up the char ** array we malloc'd before the function call
%typemap(freearg) char ** {
free((char *) $1);
}
// Now a test function
%inline %{
int print_args(char **argv) {
int i = 0;
while (argv[i]) {
printf("argv[%d] = %s\n", i, argv[i]);
i++;
}
return i;
}
%}
</pre></div>
<p>
When this module is compiled, the wrapped C function now operates as
follows :
</p>
<div class="targetlang"><pre>
>>> from argv import *
>>> print_args(["Dave", "Mike", "Mary", "Jane", "John"])
argv[0] = Dave
argv[1] = Mike
argv[2] = Mary
argv[3] = Jane
argv[4] = John
5
</pre></div>
<p>
In the example, two different typemaps are used. The "in" typemap is
used to receive an input argument and convert it to a C array. Since dynamic
memory allocation is used to allocate memory for the array, the
"freearg" typemap is used to later release this memory after the execution of
the C function.
</p>
<H3><a name="Python_nn60">33.9.2 Expanding a Python object into multiple arguments</a></H3>
<p>
Suppose that you had a collection of C functions with arguments
such as the following:
</p>
<div class="code">
<pre>
int foo(int argc, char **argv);
</pre>
</div>
<p>
In the previous example, a typemap was written to pass a Python list as the <tt>char **argv</tt>. This
allows the function to be used from Python as follows:
</p>
<div class="targetlang">
<pre>
>>> foo(4, ["foo", "bar", "spam", "1"])
</pre>
</div>
<p>
Although this works, it's a little awkward to specify the argument count. To fix this, a multi-argument
typemap can be defined. This is not very difficult--you only have to make slight modifications to the
previous example:
</p>
<div class="code">
<pre>
%typemap(in) (int argc, char **argv) {
/* Check if is a list */
if (PyList_Check($input)) {
int i;
$1 = PyList_Size($input);
$2 = (char **) malloc(($1+1)*sizeof(char *));
for (i = 0; i < $1; i++) {
PyObject *o = PyList_GetItem($input, i);
if (PyString_Check(o)) {
$2[i] = PyString_AsString(PyList_GetItem($input, i));
} else {
free($2);
PyErr_SetString(PyExc_TypeError, "list must contain strings");
SWIG_fail;
}
}
$2[i] = 0;
} else {
PyErr_SetString(PyExc_TypeError, "not a list");
SWIG_fail;
}
}
%typemap(freearg) (int argc, char **argv) {
free((char *) $2);
}
</pre>
</div>
<p>
When writing a multiple-argument typemap, each of the types is referenced by a variable such
as <tt>$1</tt> or <tt>$2</tt>. The typemap code simply fills in the appropriate values from
the supplied Python object.
</p>
<p>
With the above typemap in place, you will find it no longer necessary
to supply the argument count. This is automatically set by the typemap code. For example:
</p>
<div class="targetlang">
<pre>
>>> foo(["foo", "bar", "spam", "1"])
</pre>
</div>
<p>
If your function is overloaded in C++, for example:
</p>
<div class="code">
<pre>
int foo(int argc, char **argv);
int foo();
</pre>
</div>
<p>
don't forget to also provide a suitable <a href="Typemaps.html#Typemaps_overloading">typecheck typemap for overloading</a>
such as:
</p>
<div class="code">
<pre>
%typecheck(SWIG_TYPECHECK_STRING_ARRAY) (int argc, char **argv) {
$1 = PyList_Check($input) ? 1 : 0;
}
</pre>
</div>
<p>
If you don't you'll get an error message along the lines of:
</p>
<div class="shell">
<pre>
Traceback (most recent call last):
File "runme.py", line 3, in >module<
example.foo(["foo", "bar", "spam", "1"])
TypeError: Wrong number or type of arguments for overloaded function 'foo'.
Possible C/C++ prototypes are:
foo(int, char **)
foo()
</pre>
</div>
<H3><a name="Python_nn61">33.9.3 Using typemaps to return arguments</a></H3>
<p>
A common problem in some C programs is that values may be returned in
arguments rather than in the return value of a function. For example:
</p>
<div class="code"><pre>
/* Returns a status value and two values in out1 and out2 */
int spam(double a, double b, double *out1, double *out2) {
... Do a bunch of stuff ...
*out1 = result1;
*out2 = result2;
return status;
}
</pre></div>
<p>
A typemap can be used to handle this case as follows :
</p>
<div class="code"><pre>
%module outarg
// This tells SWIG to treat an double * argument with name 'OutValue' as
// an output value. We'll append the value to the current result which
// is guaranteed to be a List object by SWIG.
%typemap(argout) double *OutValue {
PyObject *o, *o2, *o3;
o = PyFloat_FromDouble(*$1);
if ((!$result) || ($result == Py_None)) {
$result = o;
} else {
if (!PyTuple_Check($result)) {
PyObject *o2 = $result;
$result = PyTuple_New(1);
PyTuple_SetItem($result, 0, o2);
}
o3 = PyTuple_New(1);
PyTuple_SetItem(o3, 0, o);
o2 = $result;
$result = PySequence_Concat(o2, o3);
Py_DECREF(o2);
Py_DECREF(o3);
}
}
int spam(double a, double b, double *OutValue, double *OutValue);
</pre></div>
<p>
The typemap works as follows. First, a check is made to see if any previous result
exists. If so, it is turned into a tuple and the new output value is concatenated to it.
Otherwise, the result is returned normally. For the sample function <tt>spam()</tt>, there
are three output values--meaning that the function will return a 3-tuple of the results.
</p>
<p>
As written, the function must accept 4 arguments as input values,
last two being pointers to doubles. If these arguments are only used to hold output values (and have
no meaningful input value), an additional typemap can be written. For example:
</p>
<div class="code"><pre>
%typemap(in, numinputs=0) double *OutValue(double temp) {
$1 = &temp;
}
</pre></div>
<p>
By specifying numinputs=0, the input value is ignored. However, since the argument still has to be set to
some meaningful value before calling C, it is set to point to a local variable <tt>temp</tt>. When the function
stores its output value, it will simply be placed in this local variable. As a result, the
function can now be used as follows:
</p>
<div class="targetlang"><pre>
>>> a = spam(4, 5)
>>> print a
(0, 2.45, 5.0)
>>> x, y, z = spam(4, 5)
>>>
</pre></div>
<H3><a name="Python_nn62">33.9.4 Mapping Python tuples into small arrays</a></H3>
<p>
In some applications, it is sometimes desirable to pass small arrays
of numbers as arguments. For example :
</p>
<div class="code"><pre>
extern void set_direction(double a[4]); // Set direction vector
</pre></div>
<p>
This too, can be handled used typemaps as follows :
</p>
<div class="code"><pre>
// Grab a 4 element array as a Python 4-tuple
%typemap(in) double[4](double temp[4]) { // temp[4] becomes a local variable
int i;
if (PyTuple_Check($input)) {
if (!PyArg_ParseTuple($input, "dddd", temp, temp+1, temp+2, temp+3)) {
PyErr_SetString(PyExc_TypeError, "tuple must have 4 elements");
SWIG_fail;
}
$1 = &temp[0];
} else {
PyErr_SetString(PyExc_TypeError, "expected a tuple.");
SWIG_fail;
}
}
</pre></div>
<p>
This allows our <tt>set_direction</tt> function to be called from
Python as follows :
</p>
<div class="targetlang"><pre>
>>> set_direction((0.5, 0.0, 1.0, -0.25))
</pre></div>
<p>
Since our mapping copies the contents of a Python tuple into a C
array, such an approach would not be recommended for huge arrays, but
for small structures, this approach works fine.
</p>
<H3><a name="Python_nn63">33.9.5 Mapping sequences to C arrays</a></H3>
<p>
Suppose that you wanted to generalize the previous example to handle C
arrays of different sizes. To do this, you might write a typemap as follows:
</p>
<div class="code"><pre>
// Map a Python sequence into any sized C double array
%typemap(in) double[ANY](double temp[$1_dim0]) {
int i;
if (!PySequence_Check($input)) {
PyErr_SetString(PyExc_TypeError, "Expecting a sequence");
SWIG_fail;
}
if (PyObject_Length($input) != $1_dim0) {
PyErr_SetString(PyExc_ValueError, "Expecting a sequence with $1_dim0 elements");
SWIG_fail;
}
for (i =0; i < $1_dim0; i++) {
PyObject *o = PySequence_GetItem($input, i);
if (!PyFloat_Check(o)) {
Py_XDECREF(o);
PyErr_SetString(PyExc_ValueError, "Expecting a sequence of floats");
SWIG_fail;
}
temp[i] = PyFloat_AsDouble(o);
Py_DECREF(o);
}
$1 = &temp[0];
}
</pre>
</div>
<p>
In this case, the variable <tt>$1_dim0</tt> is expanded to match the
array dimensions actually used in the C code. This allows the typemap
to be applied to types such as:
</p>
<div class="code">
<pre>
void foo(double x[10]);
void bar(double a[4], double b[8]);
</pre>
</div>
<p>
Since the above typemap code gets inserted into every wrapper function where used, it might make sense
to use a helper function instead. This will greatly reduce the amount of wrapper code. For example:
</p>
<div class="code">
<pre>
%{
static int convert_darray(PyObject *input, double *ptr, int size) {
int i;
if (!PySequence_Check(input)) {
PyErr_SetString(PyExc_TypeError, "Expecting a sequence");
return 0;
}
if (PyObject_Length(input) != size) {
PyErr_SetString(PyExc_ValueError, "Sequence size mismatch");
return 0;
}
for (i =0; i < size; i++) {
PyObject *o = PySequence_GetItem(input, i);
if (!PyFloat_Check(o)) {
Py_XDECREF(o);
PyErr_SetString(PyExc_ValueError, "Expecting a sequence of floats");
return 0;
}
ptr[i] = PyFloat_AsDouble(o);
Py_DECREF(o);
}
return 1;
}
%}
%typemap(in) double [ANY](double temp[$1_dim0]) {
if (!convert_darray($input, temp, $1_dim0)) {
SWIG_fail;
}
$1 = &temp[0];
}
</pre>
</div>
<H3><a name="Python_nn64">33.9.6 Pointer handling</a></H3>
<p>
Occasionally, it might be necessary to convert pointer values that have
been stored using the SWIG typed-pointer representation. Since there are
several ways in which pointers can be represented, the following two
functions are used to safely perform this conversion:
</p>
<p>
<tt>
int SWIG_ConvertPtr(PyObject *obj, void **ptr, swig_type_info *ty, int flags)</tt>
</p>
<div class="indent">
Converts a Python object <tt>obj</tt> to a C pointer. The result of the conversion is placed
into the pointer located at <tt>ptr</tt>. <tt>ty</tt> is a SWIG type descriptor structure.
<tt>flags</tt> is used to handle error checking and other aspects of conversion. It is the
bitwise-or of several flag values including <tt>SWIG_POINTER_EXCEPTION</tt> and
<tt>SWIG_POINTER_DISOWN</tt>. The first flag makes the function raise an exception on type
error. The second flag additionally
steals ownership of an object. Returns 0 on success and -1 on error.
</div>
<p>
<tt>
PyObject *SWIG_NewPointerObj(void *ptr, swig_type_info *ty, int own)</tt>
</p>
<div class="indent">
Creates a new Python pointer object. <tt>ptr</tt> is the pointer to convert, <tt>ty</tt> is the SWIG type descriptor structure that
describes the type, and <tt>own</tt> is a flag that indicates whether or not Python should take ownership of the
pointer.
</div>
<p>
Both of these functions require the use of a special SWIG
type-descriptor structure. This structure contains information about
the mangled name of the datatype, type-equivalence information, as
well as information about converting pointer values under C++
inheritance. For a type of <tt>Foo *</tt>, the type descriptor structure
is usually accessed as follows:
</p>
<div class="code">
<pre>
Foo *f;
if (!SWIG_IsOK(SWIG_ConvertPtr($input, (void **) &f, SWIGTYPE_p_Foo, 0))) {
SWIG_exception_fail(SWIG_TypeError, "in method '$symname', expecting type Foo");
}
PyObject *obj;
obj = SWIG_NewPointerObj(f, SWIGTYPE_p_Foo, 0);
</pre>
</div>
<p>
In a typemap, the type descriptor should always be accessed using the special typemap
variable <tt>$1_descriptor</tt>. For example:
</p>
<div class="code">
<pre>
%typemap(in) Foo * {
if (!SWIG_IsOK(SWIG_ConvertPtr($input, (void **) &$1, $1_descriptor, 0))) {
SWIG_exception_fail(SWIG_TypeError, "in method '$symname', expecting type Foo");
}
}
</pre>
</div>
<p>
If necessary, the descriptor for any type can be obtained using the <tt>$descriptor()</tt> macro in a typemap.
For example:
</p>
<div class="code">
<pre>
%typemap(in) Foo * {
if (!SWIG_IsOK(SWIG_ConvertPtr($input, (void **) &$1, $descriptor(Foo *), 0))) {
SWIG_exception_fail(SWIG_TypeError, "in method '$symname', expecting type Foo");
}
}
</pre>
</div>
<p>
Although the pointer handling functions are primarily intended for
manipulating low-level pointers, both functions are fully aware of
Python proxy classes. Specifically,
<tt>SWIG_ConvertPtr()</tt> will retrieve a pointer from any object
that has a <tt>this</tt> attribute. In addition,
<tt>SWIG_NewPointerObj()</tt> can automatically generate a proxy
class object (if applicable).
</p>
<H3><a name="Python_memory_management_member_variables">33.9.7 Memory management when returning references to member variables</a></H3>
<p>
This example shows how to prevent premature garbage collection of objects when the underlying C++ class returns a pointer or reference to a member variable.
The example is a direct equivalent to this <a href="Java.html#Java_memory_management_objects">Java equivalent</a>.
</p>
<p>
Consider the following C++ code:
</p>
<div class="code">
<pre>
#include <iostream>
struct Wheel {
int size;
Wheel(int sz = 0) : size(sz) {}
~Wheel() { std::cout << "~Wheel" << std::endl; }
};
class Bike {
Wheel wheel;
public:
Bike(int val) : wheel(val) {}
Wheel& getWheel() { return wheel; }
};
</pre>
</div>
<p>
and the following usage from Python after running the code through SWIG:
</p>
<div class="code">
<pre>
bike = Bike(10)
wheel = bike.getWheel()
print("wheel size: {}".format(wheel.size))
del bike # Allow bike to be garbage collected
print("wheel size: {}".format(wheel.size))
</pre>
</div>
<p>
Don't be surprised that if the resulting output gives strange results such as...
</p>
<div class="shell">
<pre>
wheel size: 10
~Wheel
wheel size: 135019664
</pre>
</div>
<p>
What has happened here is the garbage collector has collected the <tt>Bike</tt> instance as it doesn't think it is needed any more.
The proxy instance, <tt>wheel</tt>, contains a reference to memory that was deleted when the <tt>Bike</tt> instance was collected.
In order to prevent the garbage collector from collecting the <tt>Bike</tt> instance, a reference to the <tt>Bike</tt> must
be added to the <tt>wheel</tt> instance.
</p>
<p>
You can do this by adding the reference when the <tt>getWheel()</tt> method
is called using one of three approaches:
</p>
<p>
The easier, but less optimized, way is to use the <tt>%pythonappend</tt> directive
(see <a href="#Python_nn42">Adding additional Python code</a>):
</p>
<div class="code">
<pre>
%pythonappend getWheel %{
# val is the Wheel proxy, self is the Bike instance
val.__bike_reference = self
%}
</pre>
</div>
<p>
The code gets appended to the Python code generated for the
<tt>Bike::getWheel</tt> wrapper function, where we store the <tt>Bike</tt> proxy
instance onto the <tt>Wheel</tt> proxy instance before it is returned to the
caller as follows.
</p>
<div class="targetlang">
<pre>
class Bike(object):
...
def getWheel(self):
val = _example.Bike_getWheel(self)
# val is the Wheel proxy, self is the Bike instance
val.__bike_reference = self
return val
</pre>
</div>
<p>
The second option, which performs better and is required if you use the
<tt>-builtin</tt> option, is to set the reference in the CPython implementation:
<div class="code">
<pre>
%extend Wheel {
// A reference to the parent class is added to ensure the underlying C++
// object is not deleted while the item is in use
%typemap(ret) Wheel& getWheel {
PyObject *bike_reference_string = SWIG_Python_str_FromChar("__bike_reference");
PyObject_SetAttr($result, bike_reference_string, $self);
Py_DecRef(bike_reference_string);
}
}
</pre>
</div>
<p>
The third approach, shown below, is an optimization of the above approach and creates the "__bike_reference" Python string object just once.
While this looks more complex, it is just a small variation on the above typemap plus a support function
<tt>bike_reference()</tt> in a fragment called <tt>bike_reference_function</tt>.
The <tt>bike_reference_init</tt> typemap generates code into the "init" section for an initial call to <tt>bike_reference()</tt> when the module
is initialized and is done to create the "__bike_reference" Python string singleton in a thread-safe manner.
</p>
<div class="code">
<pre>
%fragment("bike_reference_init", "init") {
// Thread-safe initialization - initialize during Python module initialization
bike_reference();
}
%fragment("bike_reference_function", "header", fragment="bike_reference_init") {
static PyObject *bike_reference() {
static PyObject *bike_reference_string = SWIG_Python_str_FromChar("__bike_reference");
return bike_reference_string;
}
}
%extend Wheel {
// A reference to the parent class is added to ensure the underlying C++
// object is not deleted while the item is in use
%typemap(ret, fragment="bike_reference_function") Wheel& getWheel %{
PyObject_SetAttr($result, bike_reference(), $self);
%}
}
</pre>
</div>
<H2><a name="Python_nn65">33.10 Docstring Features</a></H2>
<p>
Using docstrings in Python code is becoming more and more important
and more tools are coming on the scene that take advantage of them,
everything from full-blown documentation generators to class browsers
and popup call-tips in Python-aware IDEs. Given the way that SWIG
generates the proxy code by default, your users will normally get
something like <tt>"function_name(*args)"</tt> in the popup calltip of
their IDE which is next to useless when the real function prototype
might be something like this:
</p>
<div class="code">
<pre>
bool function_name(int x, int y, Foo* foo=NULL, Bar* bar=NULL);
</pre>
</div>
<p>
The features described in this section make it easy for you to add
docstrings to your modules, functions and methods that can then be
used by the various tools out there to make the programming experience
of your users much simpler.
</p>
<H3><a name="Python_nn66">33.10.1 Module docstring</a></H3>
<p>
Python allows a docstring at the beginning of the <tt>.py</tt> file
before any other statements, and it is typically used to give a
general description of the entire module. SWIG supports this by
setting an option of the <tt>%module</tt> directive. For example:
</p>
<div class="code">
<pre>
%module(docstring="This is the example module's docstring") example
</pre>
</div>
<p>
When you have more than just a line or so then you can retain the easy
readability of the <tt>%module</tt> directive by using a macro. For
example:
</p>
<div class="code">
<pre>
%define DOCSTRING
"The `XmlResource` class allows program resources defining menus,
layout of controls on a panel, etc. to be loaded from an XML file."
%enddef
%module(docstring=DOCSTRING) xrc
</pre>
</div>
<H3><a name="Python_nn67">33.10.2 %feature("autodoc")</a></H3>
<p>
As alluded to above SWIG will generate all the function and method
proxy wrappers with just "*args" (or "*args, **kwargs" if the -keyword
option is used) for a parameter list and will then sort out the
individual parameters in the C wrapper code. This is nice and simple
for the wrapper code, but makes it difficult to be programmer and tool
friendly as anyone looking at the <tt>.py</tt> file will not be able
to find out anything about the parameters that the functions accept.
</p>
<p>But since SWIG does know everything about the function it is
possible to generate a docstring containing the parameter types, names
and default values. Since many of the docstring tools are adopting a
standard of recognizing if the first thing in the docstring is a
function prototype then using that instead of what they found from
introspection, then life is good once more.
<p>SWIG's Python module provides support for the "autodoc" feature,
which when attached to a node in the parse tree will cause a docstring
to be generated that includes the name of the function, parameter
names, default values if any, and return type if any. There are also
four levels for autodoc controlled by the value given to the
feature, <tt>%feature("autodoc", "<i>level</i>")</tt>.
The four values for <i>level</i> are covered in the following sub-sections.
<H4><a name="Python_nn68">33.10.2.1 %feature("autodoc", "0")</a></H4>
<p>
When level "0" is used then the types of the parameters will
<em>not</em> be included in the autodoc string. For example, given
this function prototype:
</p>
<div class="code">
<pre>
%feature("autodoc", "0");
bool function_name(int x, int y, Foo* foo=NULL, Bar* bar=NULL);
</pre>
</div>
<p>
Then Python code like this will be generated:
</p>
<div class="targetlang">
<pre>
def function_name(*args, **kwargs):
"""function_name(x, y, foo=None, bar=None) -> bool"""
...
</pre>
</div>
<H4><a name="Python_nn69">33.10.2.2 %feature("autodoc", "1")</a></H4>
<p>
When level "1" is used then the parameter types <em>will</em> be
used in the autodoc string. In addition, an attempt is made to
simplify the type name such that it makes more sense to the Python
user. Pointer, reference and const info is removed if the associated type
is has an associated Python type (<tt>%rename</tt>'s are thus shown correctly).
This works most of the time, otherwise a C/C++ type will be used.
See the next section for the "docstring" feature for tweaking the docstrings to your liking.
Given the example above, then turning on the
parameter types with level "1" will result in Python code like
this:
</p>
<div class="targetlang">
<pre>
def function_name(*args, **kwargs):
"""function_name(int x, int y, Foo foo=None, Bar bar=None) -> bool"""
...
</pre>
</div>
<H4><a name="Python_autodoc2">33.10.2.3 %feature("autodoc", "2")</a></H4>
<p>
Level "2" results in the function prototype as per level "0". In addition, a line of
documentation is generated for each parameter using <a href="https://github.com/numpy/numpy/blob/master/doc/HOWTO_DOCUMENT.rst.txt">numpydoc</a> style.
Using the previous example, the generated code will be:
</p>
<div class="targetlang">
<pre>
def function_name(*args, **kwargs):
"""
function_name(x, y, foo=None, bar=None) -> bool
Parameters
----------
x: int
y: int
foo: Foo *
bar: Bar *
"""
...
</pre>
</div>
<p>
Note that the documentation for each parameter is sourced from the "doc" typemap which by default shows the
C/C++ type rather than the simplified Python type name described earlier for level "1".
Typemaps can of course change the output for any particular type, for example the <tt>int x</tt> parameter:
</p>
<div class="code">
<pre>
%feature("autodoc", "2");
%typemap("doc") int x "$1_name (C++ type: $1_type) -- Input $1_name dimension"
bool function_name(int x, int y, Foo* foo=NULL, Bar* bar=NULL);
</pre>
</div>
<p>
resulting in
</p>
<div class="targetlang">
<pre>
def function_name(*args, **kwargs):
"""
function_name(x, y, foo=None, bar=None) -> bool
Parameters
----------
x (C++ type: int) -- Input x dimension
y: int
foo: Foo *
bar: Bar *
"""
</pre>
</div>
<H4><a name="Python_autodoc3">33.10.2.4 %feature("autodoc", "3")</a></H4>
<p>
Level "3" results in the function prototype as per level "1" but also contains the same additional line of documentation for each parameter as per level "2". Using our earlier example again, the generated code will be:
</p>
<div class="targetlang">
<pre>
def function_name(*args, **kwargs):
"""
function_name(int x, int y, Foo foo=None, Bar bar=None) -> bool
Parameters
----------
x: int
y: int
foo: Foo *
bar: Bar *
"""
...
</pre>
</div>
<H4><a name="Python_nn70">33.10.2.5 %feature("autodoc", "docstring")</a></H4>
<p>
Finally, there are times when the automatically generated autodoc
string will make no sense for a Python programmer, particularly when a
typemap is involved. So if you give an explicit value for the autodoc
feature then that string will be used in place of the automatically
generated string. For example:
</p>
<div class="code">
<pre>
%feature("autodoc", "GetPosition() -> (x, y)") GetPosition;
void GetPosition(int* OUTPUT, int* OUTPUT);
</pre>
</div>
<H3><a name="Python_nn71">33.10.3 %feature("docstring")</a></H3>
<p>
In addition to the autodoc strings described above, you can also
attach any arbitrary descriptive text to a node in the parse tree with
the "docstring" feature. When the proxy module is generated then any
docstring associated with classes, function or methods are output.
If an item already has an autodoc string then it is combined with the
docstring and they are output together. If the docstring is all on a
single line then it is output like this::
</p>
<div class="targetlang">
<pre>
"""This is the docstring"""
</pre>
</div>
<p>
Otherwise, to aid readability it is output like this:
</p>
<div class="targetlang">
<pre>
"""
This is a multi-line docstring
with more than one line.
"""
</pre>
</div>
<H2><a name="Python_nn72">33.11 Python Packages</a></H2>
<p>Python has concepts of modules and packages. Modules are separate units of
code and may be grouped together to form a package. Packages may be nested,
that is they may contain subpackages. This leads to tree-like hierarchy, with
packages as intermediate nodes and modules as leaf nodes.</p>
<p>The hierarchy of Python packages/modules follows the hierarchy of
<tt>*.py</tt> files found in a source tree (or, more generally, in the Python path).
Normally, the developer creates new module by placing a <tt>*.py</tt> file
somewhere under Python path; the module is then named after that <tt>*.py</tt>
file. A package is created by placing an <tt>__init__.py</tt> file within a
directory; the package is then named after that directory. For example, the
following source tree:</p>
<div class="diagram">
<pre>
mod1.py
pkg1/__init__.py
pkg1/mod2.py
pkg1/pkg2/__init__.py
pkg1/pkg2/mod3.py
</pre>
</div>
<p>
defines the following Python packages and modules:
</p>
<div class="diagram">
<pre>
pkg1 # package
pkg1.pkg2 # package
mod1 # module
pkg1.mod2 # module
pkg1.pkg2.mod3 # module
</pre>
</div>
<p>
The purpose of an <tt>__init__.py</tt> file is two-fold. First, the existence of
<tt>__init__.py</tt> in a directory informs the Python interpreter that this
directory contains a Python package. Second, the code in <tt>__init__.py</tt> is
loaded/executed automatically when the package is initialized (when it or its
submodule/subpackage gets <tt>import</tt>'ed). By default, SWIG generates
proxy Python code – one <tt>*.py</tt> file for each <tt>*.i</tt>
interface. The <tt>__init__.py</tt> files, however, are not generated by SWIG.
They should be created by other means. Both files (module <tt>*.py</tt> and
<tt>__init__.py</tt>) should be installed in appropriate destination
directories in order to obtain a desirable package/module hierarchy.
</p>
<p>
Python3 adds another option for packages with
<a href="https://www.python.org/dev/peps/pep-0420/">PEP 0420</a> (implicit
namespace packages). Implicit namespace packages no longer use
__init__.py files. SWIG generated Python modules support implicit
namespace packages. See
<a href="#Python_implicit_namespace_packages">Implicit namespace
packages</a> for more information.
</p>
<p>
You can place a SWIG generated module into a Python package or keep as a global module,
details are covered a little later in
<a href="#Python_package_search">Location of modules</a>.
</p>
<p>The way Python defines its modules and packages impacts SWIG users. Some
users may need to use special features such as the <tt>package</tt> option in the
<tt>%module</tt> directive or import related command line options. These are
explained in the following sections.</p>
<H3><a name="Python_modulepackage">33.11.1 Setting the Python package</a></H3>
<p>
Using the <tt>package</tt> option in the <tt>%module</tt> directive allows you
to specify a Python package that the module will be in when installed.
</p>
<div class="code">
<pre>
%module(package="wx") xrc
</pre>
</div>
<p>
This is useful when the <tt>.i</tt> file is <tt>%import</tt>ed by
another <tt>.i</tt> file. By default SWIG will assume that the
importer is able to find the importee with just the module name, but
if they live in separate Python packages then this won't work.
However if the importee specifies what its package is with the
<tt>%module</tt> option then the Python code generated for the
importer will use that package name when importing the other module
and in base class declarations, etc..
</p>
<p>SWIG assumes that the <tt>package</tt> option provided to <tt>%module</tt>
together with the <tt>module</tt> name (that is, <tt>wx.xrc</tt> in the above
example) forms a fully qualified (absolute) name of a module (in Python terms).
This is important especially for Python 3, where absolute imports are used by
default. It's up to you to place the generated module files (<tt>.py</tt>,
<tt>.so</tt>) in appropriate subdirectories. For example, if you have an
interface file <tt>foo.i</tt> with:
</p>
<div class="code">
<pre>
%module(package="pkg1.pkg2") foo
</pre>
</div>
<p>
then the resulting directory layout should be
</p>
<div class="diagram">
<pre>
pkg1/
pkg1/__init__.py
pkg1/pkg2/__init__.py
pkg1/pkg2/foo.py # (generated by SWIG)
pkg1/pkg2/_foo.so # (shared library built from C/C++ code generated by SWIG)
</pre>
</div>
<H3><a name="Python_absrelimports">33.11.2 Absolute and relative imports</a></H3>
<p>Suppose, we have the following hierarchy of files:</p>
<div class="diagram">
<pre>
pkg1/
pkg1/__init__.py
pkg1/mod2.py
pkg1/pkg2/__init__.py
pkg1/pkg2/mod3.py
</pre>
</div>
<p>Let the contents of <tt>pkg1/pkg2/mod3.py</tt> be</p>
<div class="targetlang">
<pre>
class M3: pass
</pre>
</div>
<p>
We edit <tt>pkg1/mod2.py</tt> and want to import module
<tt>pkg1/pkg2/mod3.py</tt> in order to derive from class <tt>M3</tt>. We can
write appropriate Python code in several ways, for example:
</p>
<ol>
<li><p>Using "<tt>import <></tt>" syntax with absolute package name:</p>
<div class="targetlang">
<pre>
# pkg1/mod2.py
import pkg1.pkg2.mod3
class M2(pkg1.pkg2.mod3.M3): pass
</pre>
</div>
</li>
<li><p>Using "<tt>import <></tt>" syntax with package name relative to
<tt>pkg1</tt> (only in Python 2.7 and earlier):</p>
<div class="targetlang">
<pre>
# pkg1/mod2.py
import pkg2.mod3
class M2(pkg2.mod3.M3): pass
</pre>
</div>
</li>
<li><p>Using "<tt>from <> import <></tt>" syntax (relative import
syntax, only in Python 2.5 and later):</p>
<div class="targetlang">
<pre>
# pkg1/mod2.py
from .pkg2 import mod3
class M2(mod3.M3): pass
</pre>
</div>
</li>
<li><p>Other variants, for example the following construction in order to
have the <tt>pkg2.mod3.M3</tt> symbol available in <tt>mod2</tt> as
in point 2 above (but now under Python 3):</p>
<div class="targetlang">
<pre>
# pkg1/mod2.py
from . import pkg2
from .pkg2 import mod3
class M2(pkg2.mod3.M3): pass
</pre>
</div>
</li>
</ol>
<p>Now suppose we have <tt>mod2.i</tt> with</p>
<div class="code">
<pre>
// mod2.i
%module (package="pkg1") mod2
%import "mod3.i"
// ...
</pre>
</div>
<p>and <tt>mod3.i</tt> with</p>
<div class="code">
<pre>
// mod3.i
%module (package="pkg1.pkg2") mod3
// ...
</pre>
</div>
<p>By default, SWIG will generate <tt>mod2.py</tt> proxy file with
<tt>import</tt> directive as in point 1. This can be changed with the
<tt>-relativeimport</tt> command line option. The <tt>-relativeimport</tt> instructs
SWIG to organize imports as in point 4 for Python 2.7.0 and newer.
</p>
<p>
<b>Compatibility Note:</b> Versions of SWIG prior to SWIG-4.0.0 supported Python < 2.7.0
and would organize the imports as in point 2 if an older version of Python was detected at runtime.
</p>
<p>
In short, if you have
<tt>mod2.i</tt> and <tt>mod3.i</tt> as above, then without
<tt>-relativeimport</tt> SWIG will write</p>
<div class="targetlang">
<pre>
import pkg1.pkg2.mod3
</pre>
</div>
<p>to <tt>mod2.py</tt> proxy file, and with <tt>-relativeimport</tt> it will
write</p>
<div class="targetlang">
<pre>
from . import pkg2
from .pkg2 import mod3
</pre>
</div>
<p>You should avoid using relative imports and use absolute ones whenever
possible. There are some cases, however, when relative imports may be
necessary. The first example is, when some (legacy) Python code refers entities
imported by proxy files generated by SWIG, and it assumes that the proxy file
uses relative imports. Second case is, when one puts import directives in
<tt>__init__.py</tt> to import symbols from submodules or subpackages and the
submodule depends on other submodules (discussed later).</p>
<H3><a name="Python_absimport">33.11.3 Enforcing absolute import semantics</a></H3>
<p>As you may know, there is an incompatibility in import semantics (for the
<tt>import <></tt> syntax) between Python 2 and 3. In Python 2.4 and
earlier it is not clear whether</p>
<div class="targetlang">
<pre>
import foo
</pre>
</div>
<p>refers to a top-level module or to another module inside the current
package. In Python 3 it always refers to a top-level module
(see <a href="https://www.python.org/dev/peps/pep-0328/">PEP 328</a>).
To instruct Python 2.5 through 2.7 to use new semantics (that is <tt>import
foo</tt> is interpreted as absolute import), one has to put the following
line
</p>
<div class="targetlang">
<pre>
from __future__ import absolute_import
</pre>
</div>
<p>at the very beginning of your proxy <tt>*.py</tt> file. In SWIG, it may be
accomplished with <tt>%pythonbegin</tt> directive as follows:</p>
<div class="code">
<pre>
%pythonbegin %{
from __future__ import absolute_import
%}
</pre>
</div>
<H3><a name="Python_importfrominit">33.11.4 Importing from __init__.py</a></H3>
<p>Imports in <tt>__init__.py</tt> are handy when you want to populate a
package's namespace with names imported from other modules. In SWIG based
projects this approach may also be used to split large pieces of code into
smaller modules, compile them in parallel and then re-assemble everything at
runtime by importing submodules' contents in <tt>__init__.py</tt>, for
example.</p>
<p>Unfortunately import directives in <tt>__init__.py</tt> may cause problems,
especially if they refer to a package's submodules. This is caused by the way
Python initializes packages. If you spot problems with imports from
<tt>__init__.py</tt> try using <tt>-relativeimport</tt> option. Below we
explain in detail one issue, for which the <tt>-relativeimport</tt> workaround
may be helpful.</p>
<p>Consider the following example (Python 3):</p>
<div class="diagram">
<pre>
pkg1/__init__.py # (empty)
pkg1/pkg2/__init__.py # (imports something from bar.py)
pkg1/pkg2/foo.py
pkg1/pkg2/bar.py # (imports foo.py)
</pre>
</div>
<p>If the file contents are:</p>
<ul>
<li> <p><tt>pkg1/pkg2/__init__.py:</tt></p>
<div class="targetlang">
<pre>
# pkg1/pkg2/__init__.py
from .bar import Bar
</pre>
</div>
</li>
<li> <p><tt>pkg1/pkg2/foo.py:</tt></p>
<div class="targetlang">
<pre>
# pkg1/pkg2/foo.py
class Foo: pass
</pre>
</div>
</li>
<li> <p><tt>pkg1/pkg2/bar.py:</tt></p>
<div class="targetlang">
<pre>
# pkg1/pkg2/bar.py
import pkg1.pkg2.foo
class Bar(pkg1.pkg2.foo.Foo): pass
</pre>
</div>
</li>
</ul>
<p>Now if one simply used <tt>import pkg1.pkg2</tt>, it will usually fail:</p>
<div class="diagram">
<pre>
>>> import pkg1.pkg2
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "./pkg1/pkg2/__init__.py", line 2, in <module>
from .bar import Bar
File "./pkg1/pkg2/bar.py", line 3, in <module>
class Bar(pkg1.pkg2.foo.Foo): pass
AttributeError: 'module' object has no attribute 'pkg2'
</pre>
</div>
<p>Surprisingly, if we execute the <tt>import pkg1.pkg2</tt> directive for the
second time, it succeeds. The reason seems to be following: when Python spots
the <tt>from .bar import Bar</tt> directive in <tt>pkg1/pkg2/__init__.py</tt>
it starts loading <tt>pkg1/pkg2/bar.py</tt>. This module imports
<tt>pkg1.pkg2.foo</tt> in turn and tries to use <tt>pkg1.pkg2.foo.Foo</tt>, but
the package <tt>pkg1</tt> is not fully initialized yet (the initialization
procedure is actually in progress) and it seems like the effect of the already seen
<tt>import pkg1.pkg2.pkg3.foo</tt> is "delayed" or ignored. Exactly the
same may happen to a proxy module generated by SWIG.</p>
<p>One workaround for this case is to use a relative
import in <tt>pkg1/pkg2/bar.py</tt>. If we change <tt>bar.py</tt> to be:</p>
<div class="targetlang">
<pre>
from .pkg3 import foo
class Bar(foo.Foo): pass
</pre>
</div>
<p>or</p>
<div class="targetlang">
<pre>
from . import pkg3
from .pkg3 import foo
class Bar(pkg3.foo.Foo): pass
</pre>
</div>
<p>then the example works again. With SWIG, you need to enable the
<tt>-relativeimport</tt> option in order to have the above workaround in
effect (note, that the Python 2 case also needs the <tt>-relativeimport</tt>
workaround).</p>
<H3><a name="Python_implicit_namespace_packages">33.11.5 Implicit namespace packages</a></H3>
<p> Python 3.3 introduced
<a href="https://www.python.org/dev/peps/pep-0420/">PEP 0420</a> which
implements implicit namespace packages. In a nutshell, implicit namespace
packages remove the requirement of an __init__.py file and allow packages
to be split across multiple PATH elements. For example:
</p>
<div class="diagram">
<pre>
/fragment1/pkg1/mod1.py
/fragment2/pkg1/mod2.py
/fragment3/pkg1/mod3.py
</pre>
</div>
<p>If PYTHONPATH is set to "/fragment1:/fragment2:/fragment3", then mod1, mod2
and mod3 will be part of pkg1. This allows for splitting of packages into
separate pieces. This can be useful for SWIG generated wrappers in the
following way.
</p>
<p> Suppose you create a SWIG wrapper for a module called robin. The SWIG
generated code consists of two files robin.py and _robin.so. You wish to
make these modules part of a subpackage (brave.sir). With implicit namespace
packages you can place these files in the following configurations:
</p>
<p>Using PYTHONPATH="/some/path"</p>
<div class="diagram">
<pre>
/some/path/brave/sir/robin.py
/some/path/brave/sir/_robin.so
</pre>
</div>
<p>Using PYTHONPATH="/some/path:/some/other/path"
<div class="diagram">
<pre>
/some/path/brave/sir/robin.py
/some/other/path/brave/sir/_robin.so
</pre>
</div>
<p> Finally suppose that your pure Python code is stored in a .zip file or
some other way (database, web service connection, etc). Python can load the
robin.py module using a custom importer. But the _robin.so module will need
to be located on a file system. Implicit namespace packages make this
possible. For example, using PYTHONPATH="/some/path/foo.zip:/some/other/path"
<p> Contents of foo.zip</p>
<div class="diagram">
<pre>
brave/
brave/sir/
brave/sir/robin.py
</pre>
</div>
<p> File system contents</p>
<div class="diagram">
<pre>
/some/other/path/brave/sir/_robin.so
</pre>
</div>
<p>Support for implicit namespace packages was added to python-3.3. The
zipimporter requires python-3.5.1 or newer to work with subpackages.
</p>
<p>
<b>Compatibility Note:</b> Support for implicit namespace packages was added in SWIG-3.0.9.
</p>
<H3><a name="Python_package_search">33.11.6 Location of modules</a></H3>
<p>
When SWIG creates wrappers from an interface file, say foo.i, two Python modules are
created. There is a pure Python module (foo.py) and C/C++ code which is
compiled and linked into a dynamically (or statically) loaded low-level module _foo
(see the <a href="#Python_nn3">Preliminaries section</a> for details). So, the interface
file really defines two Python modules. How these two modules are loaded is
covered next.
</p>
<p>
The pure Python module needs to load the C/C++ module in order to call
the wrapped C/C++ methods. To do this it must make some assumptions
about the location of the C/C++ module.
There are two configurations that are supported by default.
</p>
<ol>
<li><p>Both modules in the same package</p>
</li>
<li><p>Both modules are global</p>
</li>
</ol>
<p>
Additional configurations are supported but require custom import code.
</p>
<p>
The following sub-sections look more closely at the two default configurations as well as some customized configurations.
An input interface file, foo.i, results in the two modules foo.py and _foo.so for each of the configurations.
</p>
<H4><a name="Python_package_search_both_package_modules">33.11.6.1 Both modules in the same package</a></H4>
<p>
In this configuration, the pure Python module, foo.py, tries to load the C/C++ module, _foo, from the same package foo.py is
located in. The package name is determined from the <tt>__package__</tt>
attribute if available, see <a href="https://www.python.org/dev/peps/pep-0366/">PEP 366</a>. Otherwise it is derived from the <tt>__name__</tt>
attribute given to foo.py by the Python loader that imported foo.py.
The interface file for this configuration would contain:
</p>
<div class="code">
<pre>
%module(package="mypackage") foo
</pre>
</div>
<p>The location of the files could be as follows:</p>
<div class="diagram">
<pre>
/dir/mypackage/foo.py
/dir/mypackage/__init__.py
/dir/mypackage/_foo.so
</pre>
</div>
<p>Assuming /dir/ is in PYTHONPATH, the module can be imported using</p>
<div class="targetlang">
<pre>
from mypackage import foo
</pre>
</div>
<H4><a name="Python_package_search_both_global_modules">33.11.6.2 Both modules are global</a></H4>
<p>
In this configuration, there are no packages.
If foo.py is not in a package, that is, it is a global module, then _foo is loaded
as a global module.
The interface file for this configuration would contain:
</p>
<div class="code">
<pre>
%module foo
</pre>
</div>
<p>The location of the files could be as follows:</p>
<div class="diagram">
<pre>
/dir/foo.py
/dir/_foo.so
</pre>
</div>
<p>Assuming /dir/ is in PYTHONPATH, the module can be imported using</p>
<div class="targetlang">
<pre>
import foo
</pre>
</div>
<H4><a name="Python_package_search_wrapper_split">33.11.6.3 Split modules custom configuration</a></H4>
<p>In this non-standard 'split module' configuration, the pure Python module is in a package and the low level C/C++ module is global.
This configuration is not generally recommended and is not supported by default as it needs a custom configuration.
The module import code customization required is via the <tt>moduleimport</tt> attribute in the <tt>%module</tt> directive.
The next sub-section elaborates further on this.
The interface file for this split module configuration would contain:
</p>
<div class="code">
<pre>
%module(package="mypackage", moduleimport="import _foo") foo
</pre>
</div>
<p>
When using <tt>-builtin</tt>, use the following instead (the reasons are also covered shortly in the next sub-section):
</p>
<div class="code">
<pre>
%module(package="mypackage", moduleimport="from _foo import *") foo
</pre>
</div>
<p>The location of the files could be as follows:</p>
<div class="diagram">
<pre>
/dir/mypackage/foo.py
/dir/mypackage/__init__.py
/dir/_foo.so
</pre>
</div>
<p>Assuming /dir/ is in PYTHONPATH, the module can be imported using</p>
<div class="targetlang">
<pre>
from mypackage import foo
</pre>
</div>
<p>
<b>Compatibility Note:</b> Versions of SWIG prior to SWIG-4.0.0 supported split modules without the above customization.
However, this had to be removed as the default import code often led to confusion due to obfuscation of genuine Python <tt>ImportError</tt> problems.
Using one of the two default configurations is the recommended approach now.
</p>
<H4><a name="Python_custom_module_import">33.11.6.4 More on customizing the module import code</a></H4>
<p>
The Python code implementing the default import logic is shown below. It supports the two configurations described earlier, that is,
either both modules are in a package or loading both as global modules.
The code is generated into the pure Python module, foo.py, and merely imports the low-level _foo module.
</p>
<div class="targetlang">
<pre>
if __package__ or '.' in __name__:
from . import _foo
else:
import _foo
</pre>
</div>
<p>
This import code implementation is non-trivial but it can be replaced with custom code providing opportunities to make it simpler and/or more flexible.
This is not normally recommended though unless you have a good understanding of the intricacies of importing Python modules.
The custom code can be specified by setting the <tt>moduleimport</tt> option of the <tt>%module</tt> directive with the appropriate import code. For example:
</p>
<div class="code">
<pre>
%module(moduleimport="import _foo") foo
</pre>
</div>
<p>
This will replace the default import logic above and generate the following into the pure Python module, foo.py:
</p>
<div class="targetlang">
<pre>
import _foo
</pre>
</div>
<p>
In fact the above is a simplification customization for the configuration where both modules are global;
it removes the logic for also handling the modules being in a package.
</p>
<p>
There is a special variable, <tt>$module</tt>, which is expanded into the low-level C/C++ module name, <tt>_foo</tt> in the case above.
The identical output would be generated if instead the following had been used:
</p>
<div class="code">
<pre>
%module(moduleimport="import $module") foo
</pre>
</div>
<p>
When you have many lines you can retain the easy
readability of the <tt>%module</tt> directive by using a macro. For
example:
</p>
<div class="code">
<pre>
%define MODULEIMPORT
"
print 'Loading low-level module $module'
import $module
print 'Module has loaded'
"
%enddef
%module(moduleimport=MODULEIMPORT) foo
</pre>
</div>
<p>
This will of course generate the following into the pure Python module:
</p>
<div class="targetlang">
<pre>
print 'Loading low-level module $module'
import _foo
print 'Module has loaded'
</pre>
</div>
<p>
When using the <tt>-builtin</tt> option, the link between the pure Python module and the low-level C/C++ module is slightly different as
all the objects from the low-level module are imported directly into the pure Python module.
The default import loading code is thus different:
</p>
<div class="targetlang">
<pre>
if __package__ or '.' in __name__:
from ._foo import *
else:
from _foo import *
</pre>
</div>
<p>
Any customizations must import the code in a similar manner.
The best way to support both with and without <tt>-builtin</tt> is to make use of the <tt>SWIGPYTHON_BUILTIN</tt> macro which is defined when <tt>-builtin</tt> is specified.
The following will do this for the <a href="#Python_package_search_wrapper_split">split modules</a> case above.
</p>
<div class="code">
<pre>
#if defined(SWIGPYTHON_BUILTIN) /* defined when using -builtin */
%module(package="mypackage", moduleimport="from $module import *") foo
#else
%module(package="mypackage", moduleimport="import $module") foo
#endif
</pre>
</div>
<H4><a name="Python_package_search_static">33.11.6.5 Statically linked C modules</a></H4>
<p>It is strongly recommended to use dynamically linked modules for the C
portion of your pair of Python modules.
If for some reason you still need
to link the C module of the pair of Python modules generated by SWIG into
your interpreter, then this section provides some details on how this impacts
the pure Python modules ability to locate the other part of the pair.
Please also see the <a href="#Python_nn8">Static Linking</a> section.
</p>
<p>When Python is extended with C code the Python interpreter needs to be
informed about details of the new C functions that have been linked into
the executable. The code to do this is created by SWIG and is automatically
called in the correct way when the module is dynamically loaded. However
when the code is not dynamically loaded (because it is statically linked)
Then the initialization method for the module created by SWIG is not
called automatically and the Python interpreter has no idea that the
new SWIG C module exists.
</p>
<p>Before Python 3, one could simply call the init method created by SWIG
which would have normally been called when the shared object was dynamically
loaded. The specific name of this method is not given here because statically
linked modules are not encouraged with SWIG
(<a href="#Python_nn8">Static Linking</a>). However one can find this
init function in the C file generated by SWIG.
</p>
<p>If you are really keen on static linking there are two ways
to initialize the SWIG generated C module with the init method. Which way
you use depends on what version of Python your module is being linked with.
Python 2 and Python 3 treat this init function differently. And the way
they treat it affects how the pure Python module will be able to
locate the C module.
</p>
<p>The details concerning this are covered completely in the documentation
for Python itself. Links to the relevant sections follow:
</p>
<ul>
<li><a href="https://docs.python.org/2/extending/extending.html#methodtable">Extending in python2</a></li>
<li><a href="https://docs.python.org/3.6/extending/extending.html#the-module-s-method-table-and-initialization-function">Extending in python3</a></li>
</ul>
<p>There are two keys things to understand. The first is that in
Python 2 the init() function returns void. In Python 3 the init() function
returns a <tt>PyObject *</tt> which points to the new module. Secondly, when
you call the init() method manually, you are the Python importer. So, you
determine which package the C module will be located in.
</p>
<p>So, if you are using Python 3 it is important that you follow what is
described in the Python documentation linked above. In particular, you can't
simply call the init() function generated by SWIG and cast the <tt>PyObject</tt>
pointer it returns over the side. If you do then Python 3 will have no
idea that your C module exists and the pure Python half of your wrapper will
not be able to find it. You need to register your module with the Python
interpreter as described in the Python docs.
</p>
<p>With Python 2 things are somewhat more simple. In this case the init() function
returns void. Calling it will register your new C module as a <b>global</b>
module. The pure Python part of the SWIG wrapper will be able to find it
because it tries both the pure Python module it is part of and the global
module. If you wish not to have the statically linked module be a global
module then you will either need to refer to the Python documentation on how
to do this (remember you are now the Python importer) or use dynamic linking.
</p>
<H2><a name="Python_python3support">33.12 Python 3 Support</a></H2>
<p>
SWIG is able to support Python 3.x. The wrapper code generated by
SWIG can be compiled with both Python 2.x or 3.x. Further more, by
passing the <tt>-py3</tt> command line option to SWIG, wrapper code
with some Python 3 specific features can be generated (see below
subsections for details of these features).
<p>
There is a list of known-to-be-broken features in Python 3:
</p>
<ul>
<li>No more support for FILE* typemaps, because PyFile_AsFile has been dropped
in Python 3.</li>
<li>The <tt>-apply</tt> command line option is removed and generating
code using apply() is no longer supported.</li>
</ul>
<p>
The following are Python 3 new features that are currently supported by
SWIG.
</p>
<H3><a name="Python_nn74">33.12.1 Function annotation</a></H3>
<p>
The <tt>-py3</tt> option will enable function annotation support. When used
SWIG is able to generate proxy method definitions like this:
</p>
<div class="code"><pre>
def foo(self, bar : "int"=0) -> "void" : ...
</pre></div>
<p>
Also, even if without passing SWIG the <tt>-py3</tt> option, the parameter list
still could be generated:
</p>
<div class="code"><pre>
def foo(self, bar=0): ...
</pre></div>
<p>
But for overloaded function or method, the parameter list would fallback to
<tt>*args</tt> or <tt>self, *args</tt>, and <tt>**kwargs</tt> may be append
depend on whether you enabled the keyword argument. This fallback is due to
all overloaded functions share the same function in SWIG generated proxy class.
</p>
<p>
For detailed usage of function annotation, see
<a href="https://www.python.org/dev/peps/pep-3107/">PEP 3107</a>.
</p>
<H3><a name="Python_nn75">33.12.2 Buffer interface</a></H3>
<p>
SWIG has a series of
typemaps to support buffer interfaces. These typemap macros are
defined in <tt>pybuffer.i</tt>, which must be included in order to use them.
By using these typemaps, your wrapped function will be able to
accept any Python object that exposes a suitable buffer interface.
</p>
<p>
For example, the <tt>get_path()</tt> function puts the path string
into the memory pointed to by its argument:
</p>
<div class="code"><pre>
void get_path(char *s);
</pre></div>
<p>
Then you can write a typemap like this: (the following example is
applied to both Python 2 and 3, since the <tt>bytearray</tt> type
was backported to 2.6.
</p>
<div class="code"><pre>
%include <pybuffer.i>
%pybuffer_mutable_string(char *str);
void get_path(char *str);
</pre></div>
<p>
And then on the Python side the wrapped <tt>get_path</tt> could be used in this
way:
</p>
<div class="targetlang"><pre>
>>> p = bytearray(10)
>>> get_path(p)
>>> print(p)
bytearray(b'/Foo/Bar/\x00')
</pre></div>
<p>
The macros defined in <tt>pybuffer.i</tt> are similar to those in
<tt>cstring.i</tt>:
</p>
<p>
<b>%pybuffer_mutable_binary(parm, size_parm)</b>
</p>
<div class="indent">
<p>
The macro can be used to generate a typemap which maps a buffer of an
object to a pointer provided by <tt>parm</tt> and a size argument
provided by <tt>size_parm</tt>. For example:
</p>
<div class="code"><pre>
%pybuffer_mutable_binary(char *str, size_t size);
...
int snprintf(char *str, size_t size, const char *format, ...);
</pre></div>
<p>
In Python:
</p>
<div class="targetlang"><pre>
>>> buf = bytearray(6)
>>> snprintf(buf, "Hello world!")
>>> print(buf)
bytearray(b'Hello\x00')
>>>
</pre></div>
</div>
<p>
<b>%pybuffer_mutable_string(parm)</b>
</p>
<div class="indent">
<p>
This typemap macro requires the buffer to be a zero terminated string,
and maps the pointer of the buffer to <tt>parm</tt>. For example:
</p>
<div class="code"><pre>
%pybuffer_mutable_string(char *str);
...
size_t make_upper(char *str);
</pre></div>
<p>
In Python:
</p>
<div class="targetlang"><pre>
>>> buf = bytearray(b'foo\x00')
>>> make_upper(buf)
>>> print(buf)
bytearray(b'FOO\x00')
>>>
</pre></div>
<p>
Both <tt>%pybuffer_mutable_binary</tt> and <tt>%pybuffer_mutable_string</tt>
require the provided buffer to be mutable, eg. they can accept a
<tt>bytearray</tt> type but can't accept an immutable <tt>byte</tt>
type.
</p>
</div>
<p>
<b>%pybuffer_binary(parm, size_parm)</b>
</p>
<div class="indent">
<p>
This macro maps an object's buffer to a pointer <tt>parm</tt> and a
size <tt>size_parm</tt>. It is similar to
<tt>%pybuffer_mutable_binary</tt>, except the
<tt>%pybuffer_binary</tt> can accept both mutable and immutable
buffers. As a result, the wrapped function should not modify the buffer.
</p>
</div>
<p>
<b>%pybuffer_string(parm)</b>
</p>
<div class="indent">
<p>
This macro maps an object's buffer as a string pointer <tt>parm</tt>.
It is similar to <tt>%pybuffer_mutable_string</tt> but the buffer
could be both mutable and immutable. And your function should not
modify the buffer.
</p>
</div>
<H3><a name="Python_nn76">33.12.3 Abstract base classes</a></H3>
<p>
By including <tt>pyabc.i</tt> and using the <tt>-py3</tt> command
line option when calling SWIG, the proxy classes of the STL containers
will automatically gain an appropriate abstract base class from the
<tt>collections.abc</tt> module. For
example, the following SWIG interface:
</p>
<div class="code"><pre>
%include <pyabc.i>
%include <std_map.i>
%include <std_list.i>
namespace std {
%template(Mapii) map<int, int>;
%template(IntList) list<int>;
}
</pre></div>
<p>
will generate a Python proxy class <tt>Mapii</tt> inheriting from
<tt>collections.abc.MutableMap</tt> and a proxy class <tt>IntList</tt>
inheriting from <tt>collections.abc.MutableSequence</tt>.
</p>
<p>
<tt>pyabc.i</tt> also provides a macro <tt>%pythonabc</tt> that could be
used to define an abstract base class for your own C++ class:
</p>
<div class="code"><pre>
%pythonabc(MySet, collections.abc.MutableSet);
</pre></div>
<p>
For details of abstract base class, please see
<a href="https://www.python.org/dev/peps/pep-3119/">PEP 3119</a>.
</p>
<p>
<b>Compatibility Note:</b> SWIG-4.0.0 changed the base classes to use the
<tt>collections.abc</tt> module instead of <tt>collections</tt> due to the deprecation
of the classes in the <tt>collections</tt> module in Python 3.7.
The <tt>collections.abc</tt> module was introduced in Python 3.3 and hence this feature
requires Python 3.3 or later.
</p>
<H3><a name="Python_nn77">33.12.4 Byte string output conversion</a></H3>
<p>
By default, any byte string (<tt>char*</tt> or <tt>std::string</tt>) returned
from C or C++ code is decoded to text as UTF-8. This decoding uses the
<tt>surrogateescape</tt> error handler under Python 3.1 or higher -- this
error handler decodes invalid byte sequences to high surrogate characters
in the range U+DC80 to U+DCFF.
As an example, consider the following SWIG interface, which exposes a byte
string that cannot be completely decoded as UTF-8:
</p>
<div class="code"><pre>
%module example
%inline %{
const char * non_utf8_c_str(void) {
return "h\xe9llo w\xc3\xb6rld";
}
void instring(const char *s) {
...
}
%}
</pre></div>
<p>
Note that "\xe9" is an invalid UTF-8 encoding, but "\xc3\xb6" is valid.
When this method is called from Python 3, the return value is the following
text string:
</p>
<div class="targetlang"><pre>
>>> s = example.non_utf8_c_str()
>>> s
'h\udce9llo wörld'
</pre></div>
<p>
Since the C string contains bytes that cannot be decoded as UTF-8, those raw
bytes are represented as high surrogate characters that can be used to obtain
the original byte sequence:
</p>
<div class="targetlang"><pre>
>>> b = s.encode('utf-8', errors='surrogateescape')
>>> b
b'h\xe9llo w\xc3\xb6rld'
</pre></div>
<p>
One can then attempt a different encoding, if desired (or simply leave the
byte string as a raw sequence of bytes for use in binary protocols):
</p>
<div class="targetlang"><pre>
>>> b.decode('latin-1')
'héllo wörld'
</pre></div>
<p>
Note, however, that text strings containing surrogate characters are rejected
with the default <tt>strict</tt> codec error handler. For example:
</p>
<div class="targetlang"><pre>
>>> with open('test', 'w') as f:
... print(s, file=f)
...
Traceback (most recent call last):
File "<stdin>", line 2, in <module>
UnicodeEncodeError: 'utf-8' codec can't encode character '\udce9' in position 1: surrogates not allowed
</pre></div>
<p>
This requires the user to check most strings returned by SWIG bindings, but
the alternative is for a non-UTF8 byte string to be completely inaccessible
in Python 3 code.
</p>
<p>
For more details about the <tt>surrogateescape</tt> error handler, please see
<a href="https://www.python.org/dev/peps/pep-0383/">PEP 383</a>.
</p>
<p>
When Python 3 strings are passed to the C/C++ layer, they are expected to be valid UTF8 Unicode strings too.
For example, when the <tt>instring</tt> method above is wrapped and called, any invalid UTF8 Unicode code strings
will result in a TypeError because the attempted conversion fails:
</p>
<div class="targetlang"><pre>
>>> example.instring('h\xe9llo')
>>> example.instring('h\udce9llo')
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: in method 'instring', argument 1 of type 'char const *'
</pre></div>
<p>
In some cases, users may wish to instead handle all byte strings as bytes
objects in Python 3. This can be accomplished by adding
<tt>SWIG_PYTHON_STRICT_BYTE_CHAR</tt> to the generated code:
</p>
<div class="code"><pre>
%module char_to_bytes
%begin %{
#define SWIG_PYTHON_STRICT_BYTE_CHAR
%}
char *charstring(char *s) {
return s;
}
</pre></div>
<p>
This will modify the behavior so that only Python 3 bytes objects will be
accepted and converted to a C/C++ string, and any string returned from C/C++
will be converted to a bytes object in Python 3:
</p>
<div class="targetlang"><pre>
>>> from char_to_bytes import *
>>> charstring(b"hi") # Byte string
b'hi'
>>> charstring("hi") # Unicode string
Traceback (most recent call last):
File "<stdin>", line 1, in ?
TypeError: in method 'charstring', argument 1 of type 'char *'
</pre></div>
<p>
Note that in Python 2, defining <tt>SWIG_PYTHON_STRICT_BYTE_CHAR</tt> has no
effect, since strings in Python 2 are equivalent to Python 3 bytes objects.
However, there is a similar capability to force unicode-only handling for
wide characters C/C++ strings (<tt>wchar_t *</tt> or <tt>std::wstring</tt>
types) in Python 2. By default, in Python 2 both strings and unicode strings
are converted to C/C++ wide strings, and returned wide strings are converted
to a Python unicode string. To instead only convert unicode strings to wide
strings, users can add <tt>SWIG_PYTHON_STRICT_UNICODE_WCHAR</tt> to the
generated code:
</p>
<div class="code"><pre>
%module wchar_to_unicode
%begin %{
#define SWIG_PYTHON_STRICT_UNICODE_WCHAR
%}
wchar_t *wcharstring(wchar_t *s) {
return s;
}
</pre></div>
<p>
This ensures that only unicode strings are accepted by wcharstring in both
Python 2 and Python 3:
</p>
<div class="targetlang"><pre>
>>> from wchar_to_unicode import *
>>> wcharstring(u"hi") # Unicode string
u'hi'
>>> wcharstring(b"hi") # Byte string
Traceback (most recent call last):
File "<stdin>", line 1, in ?
TypeError: in method 'charstring', argument 1 of type 'wchar_t *'
</pre></div>
<p>
By defining both <tt>SWIG_PYTHON_STRICT_BYTE_CHAR</tt> and
<tt>SWIG_PYTHON_STRICT_UNICODE_WCHAR</tt>, Python wrapper code can support
overloads taking both std::string (as Python bytes) and std::wstring
(as Python unicode).
</p>
<H3><a name="Python_2_unicode">33.12.5 Python 2 Unicode</a></H3>
<p>
A Python 3 string is a Unicode string so by default a Python 3 string that contains Unicode
characters passed to C/C++ will be accepted and converted to a C/C++ string
(<tt>char *</tt> or <tt>std::string</tt> types).
A Python 2 string is not a unicode string by default and should a Unicode string be
passed to C/C++ it will fail to convert to a C/C++ string
(<tt>char *</tt> or <tt>std::string</tt> types).
The Python 2 behavior can be made more like Python 3 by defining
<tt>SWIG_PYTHON_2_UNICODE</tt> when compiling the generated C/C++ code.
By default when the following is wrapped:
</p>
<div class="code"><pre>
%module unicode_strings
char *charstring(char *s) {
return s;
}
</pre></div>
<p>
An error will occur when using Unicode strings in Python 2:
</p>
<div class="targetlang"><pre>
>>> from unicode_strings import *
>>> charstring("hi")
'hi'
>>> charstring(u"hi")
Traceback (most recent call last):
File "<stdin>", line 1, in ?
TypeError: in method 'charstring', argument 1 of type 'char *'
</pre></div>
<p>
When the <tt>SWIG_PYTHON_2_UNICODE</tt> macro is added to the generated code:
</p>
<div class="code"><pre>
%module unicode_strings
%begin %{
#define SWIG_PYTHON_2_UNICODE
%}
char *charstring(char *s) {
return s;
}
</pre></div>
<p>
Unicode strings will be successfully accepted and converted from UTF-8,
but note that they are returned as a normal Python 2 string:
</p>
<div class="targetlang"><pre>
>>> from unicode_strings import *
>>> charstring("hi")
'hi'
>>> charstring(u"hi")
'hi'
>>>
</pre></div>
<p>
Note that defining both <tt>SWIG_PYTHON_2_UNICODE</tt> and
<tt>SWIG_PYTHON_STRICT_BYTE_CHAR</tt> at the same time is not allowed, since
the first is allowing unicode conversion and the second is explicitly
prohibiting it.
</p>
<H2><a name="Python_multithreaded">33.13 Support for Multithreaded Applications</a></H2>
<p>By default, SWIG does not enable support for multithreaded Python applications. More
specifically, the Python wrappers generated by SWIG will not release the
Python's interpreter's Global Interpreter Lock (GIL) when wrapped C/C++ code is
entered. Hence, while any of the wrapped C/C++ code is executing, the Python interpreter
will not be able to run any other threads, even if the wrapped C/C++ code is waiting
in a blocking call for something like network or disk IO.
Fortunately, SWIG does have the ability to enable multithreaded support and automatic
release of the GIL either for all wrapped code in a module or on a more selective basis. The user
interface for this is described in the next section.
</p>
<H3><a name="Python_thread_UI">33.13.1 UI for Enabling Multithreading Support</a></H3>
<p>The user interface is as follows:</p>
<ol>
<li><p>Module thread support can be enabled in two ways:</p>
<ul>
<li>
<p>
The <tt>-threads</tt> SWIG Python option at the command line (or in <tt>setup.py</tt>):
</p>
<div class="shell"><pre>$ swig -python -threads example.i</pre></div>
</li>
<li>
<p>
The <tt>threads</tt> module option in the *.i template file:
</p>
<div class="code"><pre>%module("threads"=1)</pre></div>
</li>
</ul>
</li>
<li><p>You can disable thread support for a given method:</p>
<div class="code"><pre>%feature("nothread") method;</pre></div>
or
<div class="code"><pre>%nothread method;</pre></div>
</li>
<li><p>You can partially disable thread support for a given method:</p>
<ul>
<li><p>To disable the C++/Python thread protection:</p>
<div class="code"><pre>%feature("nothreadblock") method;</pre></div>
or
<div class="code"><pre>%nothreadblock method;</pre></div>
</li>
<li>
<p>To disable the Python/C++ thread protection</p>
<div class="code"><pre>%feature("nothreadallow") method;</pre></div>
or
<div class="code"><pre>%nothreadallow method;</pre></div>
</li>
</ul>
</li>
</ol>
<H3><a name="Python_thread_performance">33.13.2 Multithread Performance</a></H3>
<p>
For the curious about performance, here are some numbers for the profiletest.i test,
which is used to check the speed of the wrapped code:
</p>
<table summary="Python multithread performance">
<tr>
<th>Thread Mode</th>
<th>Execution Time (sec)</th>
<th>Comment</th>
</tr>
<tr>
<td>Single Threaded</td>
<td>9.6</td>
<td>no "-threads" option given</td>
</tr>
<tr>
<td>Fully Multithreaded</td>
<td>15.5</td>
<td>"-threads" option = 'allow' + 'block'</td>
</tr>
<tr>
<td>No Thread block</td>
<td>12.2</td>
<td>only 'allow'</td>
</tr>
<tr>
<td>No Thread Allow</td>
<td>13.6</td>
<td>only block'</td>
</tr>
</table>
<p>
Fully threaded code decreases the wrapping performance by
around 60%. If that is important to your application, you
can tune each method using the different 'nothread',
'nothreadblock' or 'nothreadallow' features as
needed. Note that for some methods deactivating the
'thread block' or 'thread allow' code is not an option,
so, be careful.
</p>
</body>
</html>
<!-- LocalWords: polymorphism Typemaps STL typemap typemaps Docstring autodoc
-->
<!-- LocalWords: docstring SWIG's cxx py GCC linux DLL gcc fPIC Wiki Xlinker
-->
<!-- LocalWords: examplemodule DHAVE CONFIG lpython lm ldl mypython lsocket
-->
<!-- LocalWords: lnsl lpthread distutils enums namespaces
-->
|
