aboutsummaryrefslogtreecommitdiff
path: root/libbuild2/script/run.cxx
blob: 38436b920b4f8f7e077549db632ae192e6f620fd (plain)
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
// file      : libbuild2/script/run.cxx -*- C++ -*-
// license   : MIT; see accompanying LICENSE file

#include <libbuild2/script/run.hxx>

#include <ios> // streamsize

#include <libbutl/regex.mxx>
#include <libbutl/builtin.mxx>
#include <libbutl/fdstream.mxx>     // fdopen_mode, fddup()
#include <libbutl/filesystem.mxx>   // path_search()
#include <libbutl/path-pattern.mxx>

#include <libbuild2/filesystem.hxx>
#include <libbuild2/diagnostics.hxx>

#include <libbuild2/script/regex.hxx>
#include <libbuild2/script/builtin-options.hxx>

using namespace std;
using namespace butl;

namespace build2
{
  namespace script
  {
    string
    diag_path (const path& d)
    {
      string r ("'");

      r += stream_verb_map ().path < 1
           ? diag_relative (d)
           : d.representation ();

      r += '\'';
      return r;
    }

    string
    diag_path (const dir_name_view& dn)
    {
      string r;
      if (dn.name != nullptr && *dn.name)
      {
        r += **dn.name;
        r += ' ';
      }

      assert (dn.path != nullptr);

      r += diag_path (*dn.path);
      return r;
    }

    // Return the environment temporary directory, creating it if it doesn't
    // exist.
    //
    static inline const dir_path&
    temp_dir (environment& env)
    {
      if (env.temp_dir.empty ())
        env.create_temp_dir ();

      return env.temp_dir;
    }

    // Normalize a path. Also make the relative path absolute using the
    // specified directory unless it is already absolute.
    //
    static path
    normalize (path p, const dir_path& d, const location& l)
    {
      path r (p.absolute () ? move (p) : d / move (p));

      try
      {
        r.normalize ();
      }
      catch (const invalid_path& e)
      {
        fail (l) << "invalid file path " << e.path;
      }

      return r;
    }

    // Check if a path is not empty, the referenced file exists and is not
    // empty.
    //
    static bool
    non_empty (const path& p, const location& ll)
    {
      if (p.empty () || !exists (p))
        return false;

      try
      {
        ifdstream is (p);
        return is.peek () != ifdstream::traits_type::eof ();
      }
      catch (const io_error& e)
      {
        // While there can be no fault of the script command being currently
        // executed let's add the location anyway to ease the
        // troubleshooting. And let's stick to that principle down the road.
        //
        fail (ll) << "unable to read " << p << ": " << e << endf;
      }
    }

    // If the file exists, not empty and not larger than 4KB print it to the
    // diag record. The file content goes from the new line and is not
    // indented.
    //
    static void
    print_file (diag_record& d, const path& p, const location& ll)
    {
      if (exists (p))
      {
        try
        {
          ifdstream is (p, ifdstream::badbit);

          if (is.peek () != ifdstream::traits_type::eof ())
          {
            char buf[4096 + 1]; // Extra byte is for terminating '\0'.

            // Note that the string is always '\0'-terminated with a maximum
            // sizeof (buf) - 1 bytes read.
            //
            is.getline (buf, sizeof (buf), '\0');

            // Print if the file fits 4KB-size buffer. Note that if it
            // doesn't the failbit is set.
            //
            if (is.eof ())
            {
              // Suppress the trailing newline character as the diag record
              // adds it's own one when flush.
              //
              streamsize n (is.gcount ());
              assert (n > 0);

              // Note that if the file contains '\0' it will also be counted
              // by gcount(). But even in the worst case we will stay in the
              // buffer boundaries (and so not crash).
              //
              if (buf[n - 1] == '\n')
                buf[n - 1] = '\0';

              d << '\n' << buf;
            }
          }
        }
        catch (const io_error& e)
        {
          fail (ll) << "unable to read " << p << ": " << e;
        }
      }
    }

    // Save a string to the file. Fail if exception is thrown by underlying
    // operations.
    //
    static void
    save (const path& p, const string& s, const location& ll)
    {
      try
      {
        ofdstream os (p);
        os << s;
        os.close ();
      }
      catch (const io_error& e)
      {
        fail (ll) << "unable to write to " << p << ": " << e;
      }
    }

    // Transform string according to here-* redirect modifiers from the {/}
    // set.
    //
    static string
    transform (const string& s,
               bool regex,
               const string& modifiers,
               environment& env)
    {
      if (modifiers.find ('/') == string::npos)
        return s;

      // For targets other than Windows leave the string intact.
      //
      if (env.host.class_ != "windows")
        return s;

      // Convert forward slashes to Windows path separators (escape for
      // regex).
      //
      string r;
      for (size_t p (0);;)
      {
        size_t sp (s.find ('/', p));

        if (sp != string::npos)
        {
          r.append (s, p, sp - p);
          r.append (regex ? "\\\\" : "\\");
          p = sp + 1;
        }
        else
        {
          r.append (s, p, sp);
          break;
        }
      }

      return r;
    }

    // Return true if the script temporary directory is not created yet (and
    // so cannot contain any path), a path is not under the temporary
    // directory or this directory will not be removed on failure.
    //
    static inline bool
    avail_on_failure (const path& p, const environment& env)
    {
      return env.temp_dir.empty () ||
             env.temp_dir_keep     ||
             !p.sub (env.temp_dir);
    }

    // Check if the script command output matches the expected result
    // (redirect value). Noop for redirect types other than none, here_*.
    //
    static bool
    check_output (const path& pr,
                  const path& op,
                  const path& ip,
                  const redirect& rd,
                  const location& ll,
                  environment& env,
                  bool diag,
                  const char* what)
    {
      auto input_info = [&ip, &ll, &env] (diag_record& d)
      {
        if (non_empty (ip, ll) && avail_on_failure (ip, env))
          d << info << "stdin: " << ip;
      };

      auto output_info = [&what, &ll, &env] (diag_record& d,
                                             const path& p,
                                             const char* prefix = "",
                                             const char* suffix = "")
      {
        if (non_empty (p, ll))
        {
          if (avail_on_failure (p, env))
            d << info << prefix << what << suffix << ": " << p;
        }
        else
          d << info << prefix << what << suffix << " is empty";
      };

      if (rd.type == redirect_type::none)
      {
        // Check that there is no output produced.
        //
        assert (!op.empty ());

        if (!non_empty (op, ll))
          return true;

        if (diag)
        {
          diag_record d (error (ll));
          d << pr << " unexpectedly writes to " << what;

          if (avail_on_failure (op, env))
            d << info << what << ": " << op;

          input_info (d);

          // Print cached output.
          //
          print_file (d, op, ll);
        }

        // Fall through (to return false).
        //
      }
      else if (rd.type == redirect_type::here_str_literal ||
               rd.type == redirect_type::here_doc_literal ||
               (rd.type == redirect_type::file &&
                rd.file.mode == redirect_fmode::compare))
      {
        // The expected output is provided as a file or as a string. Save the
        // string to a file in the later case.
        //
        assert (!op.empty ());

        path eop;

        if (rd.type == redirect_type::file)
          eop = normalize (rd.file.path, *env.work_dir.path, ll);
        else
        {
          eop = path (op + ".orig");

          save (eop,
                transform (rd.str, false /* regex */, rd.modifiers (), env),
                ll);

          env.clean_special (eop);
        }

        // Use the diff utility for comparison.
        //
        path dp ("diff");
        process_path pp (run_search (dp, true));

        cstrings args {pp.recall_string ()};

        // If both files being compared won't be available on failure, then
        // instruct diff not to print the file paths. It seems that the only
        // way to achieve this is to abandon the output unified format in the
        // favor of the minimal output, which normally is still informative
        // enough for the troubleshooting (contains the difference line
        // numbers, etc).
        //
        if (avail_on_failure (eop, env) || avail_on_failure (op, env))
          args.push_back ("-u");

        // Ignore Windows newline fluff if that's what we are running on.
        //
        if (env.host.class_ == "windows")
          args.push_back ("--strip-trailing-cr");

        args.push_back (eop.string ().c_str ());
        args.push_back (op.string ().c_str ());
        args.push_back (nullptr);

        if (verb >= 2)
          print_process (args);

        try
        {
          // Save diff's stdout to a file for troubleshooting and for the
          // optional (if not too large) printing (at the end of
          // diagnostics).
          //
          path ep (op + ".diff");
          auto_fd efd;

          try
          {
            efd = fdopen (ep, fdopen_mode::out | fdopen_mode::create);
            env.clean_special (ep);
          }
          catch (const io_error& e)
          {
            fail (ll) << "unable to write to " << ep << ": " << e;
          }

          // Diff utility prints the differences to stdout. But for the
          // user it is a part of the script failure diagnostics so let's
          // redirect stdout to stderr.
          //
          process p (pp, args.data (), 0, 2, efd.get ());
          efd.reset ();

          if (p.wait ())
            return true;

          assert (p.exit);
          const process_exit& pe (*p.exit);

          // Note that both POSIX and GNU diff report error by exiting with
          // the code > 1.
          //
          if (!pe.normal () || pe.code () > 1)
          {
            diag_record d (fail (ll));
            print_process (d, args);
            d << " " << pe;

            print_file (d, ep, ll);
          }

          // Output doesn't match the expected result.
          //
          if (diag)
          {
            diag_record d (error (ll));
            d << pr << " " << what << " doesn't match expected";

            output_info (d, op);
            output_info (d, eop, "expected ");
            output_info (d, ep, "", " diff");
            input_info  (d);

            print_file (d, ep, ll);
          }

          // Fall through (to return false).
          //
        }
        catch (const process_error& e)
        {
          error (ll) << "unable to execute " << pp << ": " << e;

          if (e.child)
            exit (1);

          throw failed ();
        }
      }
      else if (rd.type == redirect_type::here_str_regex ||
               rd.type == redirect_type::here_doc_regex)
      {
        // The overall plan is:
        //
        // 1. Create regex line string. While creating it's line characters
        //    transform regex lines according to the redirect modifiers.
        //
        // 2. Create line regex using the line string. If creation fails
        //    then save the (transformed) regex redirect to a file for
        //    troubleshooting.
        //
        // 3. Parse the output into the literal line string.
        //
        // 4. Match the output line string with the line regex.
        //
        // 5. If match fails save the (transformed) regex redirect to a file
        //    for troubleshooting.
        //
        using namespace regex;

        assert (!op.empty ());

        // Create regex line string.
        //
        line_pool pool;
        line_string rls;
        const regex_lines rl (rd.regex);

        // Parse regex flags.
        //
        // When add support for new flags don't forget to update
        // parse_regex().
        //
        auto parse_flags = [] (const string& f) -> char_flags
        {
          char_flags r (char_flags::none);

          for (char c: f)
          {
            switch (c)
            {
            case 'd': r |= char_flags::idot;  break;
            case 'i': r |= char_flags::icase; break;
            default: assert (false); // Error so should have been checked.
            }
          }

          return r;
        };

        // Return original regex line with the transformation applied.
        //
        auto line = [&rl, &rd, &env] (const regex_line& l) -> string
        {
          string r;
          if (l.regex)                  // Regex (possibly empty),
          {
            r += rl.intro;
            r += transform (l.value, true /* regex */, rd.modifiers (), env);
            r += rl.intro;
            r += l.flags;
          }
          else if (!l.special.empty ()) // Special literal.
            r += rl.intro;
          else                          // Textual literal.
            r += transform (l.value, false /* regex */, rd.modifiers (), env);

          r += l.special;
          return r;
        };

        // Return regex line location.
        //
        // Note that we rely on the fact that the command and regex lines
        // are always belong to the same file.
        //
        auto loc = [&ll] (uint64_t line, uint64_t column) -> location
        {
          location r (ll);
          r.line = line;
          r.column = column;
          return r;
        };

        // Save the regex to file for troubleshooting, return the file path
        // it have been saved to.
        //
        // Note that we save the regex on line regex creation failure or if
        // the program output doesn't match.
        //
        auto save_regex = [&op, &rl, &rd, &ll, &line] () -> path
        {
          path rp (op + ".regex");

          // Encode here-document regex global flags if present as a file
          // name suffix. For example if icase and idot flags are specified
          // the name will look like:
          //
          // stdout.regex-di
          //
          if (rd.type == redirect_type::here_doc_regex && !rl.flags.empty ())
            rp += '-' + rl.flags;

          // Note that if would be more efficient to directly write chunks
          // to file rather than to compose a string first. Hower we don't
          // bother (about performance) for the sake of the code as we
          // already failed.
          //
          string s;
          for (auto b (rl.lines.cbegin ()), i (b), e (rl.lines.cend ());
               i != e; ++i)
          {
            if (i != b) s += '\n';
            s += line (*i);
          }

          save (rp, s, ll);
          return rp;
        };

        // Finally create regex line string.
        //
        // Note that diagnostics doesn't refer to the program path as it is
        // irrelevant to failures at this stage.
        //
        char_flags gf (parse_flags (rl.flags)); // Regex global flags.

        for (const auto& l: rl.lines)
        {
          if (l.regex) // Regex (with optional special characters).
          {
            line_char c;

            // Empty regex is a special case repesenting the blank line.
            //
            if (l.value.empty ())
              c = line_char ("", pool);
            else
            {
              try
              {
                string s (transform (l.value,
                                     true /* regex */,
                                     rd.modifiers (),
                                     env));

                c = line_char (
                  char_regex (s, gf | parse_flags (l.flags)), pool);
              }
              catch (const regex_error& e)
              {
                // Print regex_error description if meaningful.
                //
                diag_record d (fail (loc (l.line, l.column)));

                if (rd.type == redirect_type::here_str_regex)
                  d << "invalid " << what << " regex redirect" << e <<
                    info << "regex: '" << line (l) << "'";
                else
                  d << "invalid char-regex in " << what << " regex redirect"
                    << e <<
                    info << "regex line: '" << line (l) << "'";

                d << endf;
              }
            }

            rls += c; // Append blank literal or regex line char.
          }
          else if (!l.special.empty ()) // Special literal.
          {
            // Literal can not be followed by special characters in the same
            // line.
            //
            assert (l.value.empty ());
          }
          else // Textual literal.
          {
            // Append literal line char.
            //
            rls += line_char (transform (l.value,
                                         false /* regex */,
                                         rd.modifiers (),
                                         env),
                              pool);
          }

          for (char c: l.special)
          {
            if (line_char::syntax (c))
              rls += line_char (c); // Append special line char.
            else
              fail (loc (l.line, l.column))
                << "invalid syntax character '" << c << "' in " << what
                << " regex redirect" <<
                info << "regex line: '" << line (l) << "'";
          }
        }

        // Create line regex.
        //
        line_regex regex;

        try
        {
          regex = line_regex (move (rls), move (pool));
        }
        catch (const regex_error& e)
        {
          // Note that line regex creation can not fail for here-string
          // redirect as it doesn't have syntax line chars. That in
          // particular means that end_line and end_column are meaningful.
          //
          assert (rd.type == redirect_type::here_doc_regex);

          diag_record d (fail (loc (rd.end_line, rd.end_column)));

          // Print regex_error description if meaningful.
          //
          d << "invalid " << what << " regex redirect" << e;

          // It would be a waste to save the regex into the file just to
          // remove it.
          //
          if (env.temp_dir_keep)
            output_info (d, save_regex (), "", " regex");
        }

        // Parse the output into the literal line string.
        //
        line_string ls;

        try
        {
          // Do not throw when eofbit is set (end of stream reached), and
          // when failbit is set (getline() failed to extract any character).
          //
          // Note that newlines are treated as line-chars separators. That
          // in particular means that the trailing newline produces a blank
          // line-char (empty literal). Empty output produces the zero-length
          // line-string.
          //
          // Also note that we strip the trailing CR characters (otherwise
          // can mismatch when, for example, cross-testing).
          //
          ifdstream is (op, ifdstream::badbit);
          is.peek (); // Sets eofbit for an empty stream.

          while (!is.eof ())
          {
            string s;
            getline (is, s);

            // It is safer to strip CRs in cycle, as msvcrt unexplainably
            // adds too much trailing junk to the system_error descriptions,
            // and so it can appear in programs output. For example:
            //
            // ...: Invalid data.\r\r\n
            //
            // Note that our custom operator<<(ostream&, const exception&)
            // removes this junk.
            //
            while (!s.empty () && s.back () == '\r')
              s.pop_back ();

            ls += line_char (move (s), regex.pool);
          }
        }
        catch (const io_error& e)
        {
          fail (ll) << "unable to read " << op << ": " << e;
        }

        // Match the output with the regex.
        //
        if (regex_match (ls, regex)) // Doesn't throw.
          return true;

        // Output doesn't match the regex.
        //
        // Unless the temporary directory is removed on failure, we save the
        // regex to file for troubleshooting regardless of whether we print
        // the diagnostics or not. We, however, register it for cleanup in the
        // later case (the expression may still succeed, we can be evaluating
        // the if condition, etc).
        //
        optional<path> rp;
        if (env.temp_dir_keep)
          rp = save_regex ();

        if (diag)
        {
          diag_record d (error (ll));
          d << pr << " " << what << " doesn't match regex";

          output_info (d, op);

          if (rp)
            output_info (d, *rp, "", " regex");

          input_info  (d);

          // Print cached output.
          //
          print_file (d, op, ll);
        }
        else if (rp)
          env.clean_special (*rp);

        // Fall through (to return false).
        //
      }
      else // Noop.
        return true;

      return false;
    }

    // The exit pseudo-builtin: exit the script successfully, or print the
    // diagnostics and exit the script unsuccessfully. Always throw exit
    // exception.
    //
    // exit [<diagnostics>]
    //
    [[noreturn]] static void
    exit_builtin (const strings& args, const location& ll)
    {
      auto i (args.begin ());
      auto e (args.end ());

      // Process arguments.
      //
      // If no argument is specified, then exit successfully. Otherwise,
      // print the diagnostics and exit unsuccessfully.
      //
      if (i == e)
        throw exit (true);

      const string& s (*i++);

      if (i != e)
        fail (ll) << "unexpected argument '" << *i << "'";

      error (ll) << s;
      throw exit (false);
    }

    // The set pseudo-builtin: set variable from the stdin input.
    //
    // set [-e|--exact] [(-n|--newline)|(-w|--whitespace)] [<attr>] <var>
    //
    static void
    set_builtin (environment& env,
                 const strings& args,
                 auto_fd in,
                 const location& ll)
    {
      try
      {
        // Do not throw when eofbit is set (end of stream reached), and
        // when failbit is set (read operation failed to extract any
        // character).
        //
        ifdstream cin  (move (in), ifdstream::badbit);

        // Parse arguments.
        //
        cli::vector_scanner scan (args);
        set_options ops (scan);

        if (ops.whitespace () && ops.newline ())
          fail (ll) << "both -n|--newline and -w|--whitespace specified";

        if (!scan.more ())
          fail (ll) << "missing variable name";

        string a (scan.next ()); // Either attributes or variable name.
        const string* ats (!scan.more () ? nullptr : &a);
        string vname (!scan.more () ? move (a) : scan.next ());

        if (scan.more ())
          fail (ll) << "unexpected argument '" << scan.next () << "'";

        if (ats != nullptr && ats->empty ())
          fail (ll) << "empty variable attributes";

        if (vname.empty ())
          fail (ll) << "empty variable name";

        // Read the input.
        //
        cin.peek (); // Sets eofbit for an empty stream.

        names ns;
        while (!cin.eof ())
        {
          // Read next element that depends on the whitespace mode being
          // enabled or not. For the later case it also make sense to strip
          // the trailing CRs that can appear while, for example,
          // cross-testing Windows target or as a part of msvcrt junk
          // production (see above).
          //
          string s;
          if (ops.whitespace ())
            cin >> s;
          else
          {
            getline (cin, s);

            while (!s.empty () && s.back () == '\r')
              s.pop_back ();
          }

          // If failbit is set then we read nothing into the string as eof is
          // reached. That in particular means that the stream has trailing
          // whitespaces (possibly including newlines) if the whitespace mode
          // is enabled, or the trailing newline otherwise. If so then
          // we append the "blank" to the variable value in the exact mode
          // prior to bailing out.
          //
          if (cin.fail ())
          {
            if (ops.exact ())
            {
              if (ops.whitespace () || ops.newline ())
                ns.emplace_back (move (s)); // Reuse empty string.
              else if (ns.empty ())
                ns.emplace_back ("\n");
              else
                ns[0].value += '\n';
            }

            break;
          }

          if (ops.whitespace () || ops.newline () || ns.empty ())
            ns.emplace_back (move (s));
          else
          {
            ns[0].value += '\n';
            ns[0].value += s;
          }
        }

        cin.close ();

        env.set_variable (move (vname),
                          move (ns),
                          ats != nullptr ? *ats : empty_string,
                          ll);
      }
      catch (const io_error& e)
      {
        fail (ll) << "set: " << e;
      }
      catch (const cli::exception& e)
      {
        fail (ll) << "set: " << e;
      }
    }

    // Sorted array of builtins that support filesystem entries cleanup.
    //
    static const char* cleanup_builtins[] = {
      "cp", "ln", "mkdir", "mv", "touch"};

    static inline bool
    cleanup_builtin (const string& name)
    {
      return binary_search (
        cleanup_builtins,
        cleanup_builtins +
        sizeof (cleanup_builtins) / sizeof (*cleanup_builtins),
        name);
    }

    static bool
    run_pipe (environment& env,
              command_pipe::const_iterator bc,
              command_pipe::const_iterator ec,
              auto_fd ifd,
              size_t ci, size_t li, const location& ll,
              bool diag)
    {
      if (bc == ec) // End of the pipeline.
        return true;

      // The overall plan is to run the first command in the pipe, reading
      // its input from the file descriptor passed (or, for the first
      // command, according to stdin redirect specification) and redirecting
      // its output to the right-hand part of the pipe recursively. Fail if
      // the right-hand part fails. Otherwise check the process exit code,
      // match stderr (and stdout for the last command in the pipe) according
      // to redirect specification(s) and fail if any of the above fails.
      //
      const command& c (*bc);

      // Register the command explicit cleanups. Verify that the path being
      // cleaned up is a sub-path of the script working directory. Fail if
      // this is not the case.
      //
      for (const auto& cl: c.cleanups)
      {
        const path& p (cl.path);
        path np (normalize (p, *env.work_dir.path, ll));

        const string& ls (np.leaf ().string ());
        bool wc (ls == "*" || ls == "**" || ls == "***");
        const path& cp (wc ? np.directory () : np);
        const dir_path* sd (env.sandbox_dir.path);

        if (sd != nullptr && !cp.sub (*sd))
          fail (ll) << (wc                ? "wildcard"  :
                        p.to_directory () ? "directory" :
                                            "file")
                    << " cleanup " << p << " is out of "
                    << diag_path (env.sandbox_dir);

        env.clean ({cl.type, move (np)}, false);
      }

      bool eq (c.exit.comparison == exit_comparison::eq);

      // If stdin file descriptor is not open then this is the first pipeline
      // command.
      //
      bool first (ifd.get () == -1);

      command_pipe::const_iterator nc (bc + 1);
      bool last (nc == ec);

      const string& program (c.program.string ());

      const redirect& in ((c.in ? *c.in : env.in).effective ());

      const redirect* out (!last
                           ? nullptr // stdout is piped.
                           : &(c.out ? *c.out : env.out).effective ());

      const redirect& err ((c.err ? *c.err : env.err).effective ());

      auto process_args = [&c] () -> cstrings
      {
        cstrings args {c.program.string ().c_str ()};

        for (const auto& a: c.arguments)
          args.push_back (a.c_str ());

        args.push_back (nullptr);
        return args;
      };

      // Prior to opening file descriptors for command input/output
      // redirects let's check if the command is the exit builtin. Being a
      // builtin syntactically it differs from the regular ones in a number
      // of ways. It doesn't communicate with standard streams, so
      // redirecting them is meaningless. It may appear only as a single
      // command in a pipeline. It doesn't return any value and stops the
      // script execution, so checking its exit status is meaningless as
      // well. That all means we can short-circuit here calling the builtin
      // and bailing out right after that. Checking that the user didn't
      // specify any redirects or exit code check sounds like a right thing
      // to do.
      //
      if (program == "exit")
      {
        // In case the builtin is erroneously pipelined from the other
        // command, we will close stdin gracefully (reading out the stream
        // content), to make sure that the command doesn't print any
        // unwanted diagnostics about IO operation failure.
        //
        // Note that dtor will ignore any errors (which is what we want).
        //
        ifdstream is (move (ifd), fdstream_mode::skip);

        if (!first || !last)
          fail (ll) << "exit builtin must be the only pipe command";

        if (c.in)
          fail (ll) << "exit builtin stdin cannot be redirected";

        if (c.out)
          fail (ll) << "exit builtin stdout cannot be redirected";

        if (c.err)
          fail (ll) << "exit builtin stderr cannot be redirected";

        // We can't make sure that there is no exit code check. Let's, at
        // least, check that non-zero code is not expected.
        //
        if (eq != (c.exit.code == 0))
          fail (ll) << "exit builtin exit code cannot be non-zero";

        if (verb >= 2)
          print_process (process_args ());

        exit_builtin (c.arguments, ll); // Throws exit exception.
      }

      // Create a unique path for a command standard stream cache file.
      //
      auto std_path = [&env, &ci, &li, &ll] (const char* n) -> path
      {
        using std::to_string;

        path p (n);

        // 0 if belongs to a single-line script, otherwise is the command line
        // number (start from one) in the script.
        //
        if (li > 0)
          p += "-" + to_string (li);

        // 0 if belongs to a single-command expression, otherwise is the
        // command number (start from one) in the expression.
        //
        // Note that the name like stdin-N can relate to N-th command of a
        // single-line script or to N-th single-command line of multi-line
        // script. These cases are mutually exclusive and so are unambiguous.
        //
        if (ci > 0)
          p += "-" + to_string (ci);

        return normalize (move (p), temp_dir (env), ll);
      };

      // If this is the first pipeline command, then open stdin descriptor
      // according to the redirect specified.
      //
      path isp;

      if (!first)
        assert (!c.in); // No redirect expected.
      else
      {
        // Open a file for passing to the command stdin.
        //
        auto open_stdin = [&isp, &ifd, &ll] ()
        {
          assert (!isp.empty ());

          try
          {
            ifd = fdopen (isp, fdopen_mode::in);
          }
          catch (const io_error& e)
          {
            fail (ll) << "unable to read " << isp << ": " << e;
          }
        };

        switch (in.type)
        {
        case redirect_type::pass:
          {
            try
            {
              ifd = fddup (0);
            }
            catch (const io_error& e)
            {
              fail (ll) << "unable to duplicate stdin: " << e;
            }

            break;
          }
        case redirect_type::none:
          // Somehow need to make sure that the child process doesn't read
          // from stdin. That is tricky to do in a portable way. Here we
          // suppose that the program which (erroneously) tries to read some
          // data from stdin being redirected to /dev/null fails not being
          // able to read the expected data, and so the command doesn't pass
          // through.
          //
          // @@ Obviously doesn't cover the case when the process reads
          //    whatever available.
          // @@ Another approach could be not to redirect stdin and let the
          //    process to hang which can be interpreted as a command failure.
          // @@ Both ways are quite ugly. Is there some better way to do
          //    this?
          //
          // Fall through.
          //
        case redirect_type::null:
          {
            ifd = open_null ();
            break;
          }
        case redirect_type::file:
          {
            isp = normalize (in.file.path, *env.work_dir.path, ll);

            open_stdin ();
            break;
          }
        case redirect_type::here_str_literal:
        case redirect_type::here_doc_literal:
          {
            // We could write to the command stdin directly but instead will
            // cache the data for potential troubleshooting.
            //
            isp = std_path ("stdin");

            save (isp,
                  transform (in.str, false /* regex */, in.modifiers (), env),
                  ll);

            env.clean_special (isp);

            open_stdin ();
            break;
          }
        case redirect_type::trace:
        case redirect_type::merge:
        case redirect_type::here_str_regex:
        case redirect_type::here_doc_regex:
        case redirect_type::here_doc_ref:   assert (false); break;
        }
      }

      assert (ifd.get () != -1);

      // Prior to opening file descriptors for command outputs redirects
      // let's check if the command is the set builtin. Being a builtin
      // syntactically it differs from the regular ones in a number of ways.
      // It either succeeds or terminates abnormally, so redirecting stderr
      // is meaningless. It also never produces any output and may appear
      // only as a terminal command in a pipeline. That means we can
      // short-circuit here calling the builtin and returning right after
      // that. Checking that the user didn't specify any meaningless
      // redirects or exit code check sounds as a right thing to do.
      //
      if (program == "set")
      {
        if (!last)
          fail (ll) << "set builtin must be the last pipe command";

        if (c.out)
          fail (ll) << "set builtin stdout cannot be redirected";

        if (c.err)
          fail (ll) << "set builtin stderr cannot be redirected";

        if (eq != (c.exit.code == 0))
          fail (ll) << "set builtin exit code cannot be non-zero";

        if (verb >= 2)
          print_process (process_args ());

        set_builtin (env, c.arguments, move (ifd), ll);
        return true;
      }

      // Open a file for command output redirect if requested explicitly
      // (file overwrite/append redirects) or for the purpose of the output
      // validation (none, here_*, file comparison redirects), register the
      // file for cleanup, return the file descriptor. Interpret trace
      // redirect according to the verbosity level (as null if below 2, as
      // pass otherwise). Return nullfd, standard stream descriptor duplicate
      // or null-device descriptor for merge, pass or null redirects
      // respectively (not opening any file).
      //
      auto open = [&env, &ll, &std_path] (const redirect& r,
                                          int dfd,
                                          path& p) -> auto_fd
      {
        assert (dfd == 1 || dfd == 2);
        const char* what (dfd == 1 ? "stdout" : "stderr");

        fdopen_mode m (fdopen_mode::out | fdopen_mode::create);

        redirect_type rt (r.type != redirect_type::trace
                          ? r.type
                          : verb < 2
                          ? redirect_type::null
                          : redirect_type::pass);
        switch (rt)
        {
        case redirect_type::pass:
          {
            try
            {
              return fddup (dfd);
            }
            catch (const io_error& e)
            {
              fail (ll) << "unable to duplicate " << what << ": " << e;
            }
          }

        case redirect_type::null: return open_null ();

          // Duplicate the paired file descriptor later.
          //
        case redirect_type::merge: return nullfd;

        case redirect_type::file:
          {
            // For the cmp mode the user-provided path refers a content to
            // match against, rather than a content to be produced (as for
            // overwrite and append modes). And so for cmp mode we redirect
            // the process output to a temporary file.
            //
            p = r.file.mode == redirect_fmode::compare
              ? std_path (what)
              : normalize (r.file.path, *env.work_dir.path, ll);

            m |= r.file.mode == redirect_fmode::append
              ? fdopen_mode::at_end
              : fdopen_mode::truncate;

            break;
          }

        case redirect_type::none:
        case redirect_type::here_str_literal:
        case redirect_type::here_doc_literal:
        case redirect_type::here_str_regex:
        case redirect_type::here_doc_regex:
          {
            p = std_path (what);
            m |= fdopen_mode::truncate;
            break;
          }

        case redirect_type::trace:
        case redirect_type::here_doc_ref: assert (false); break;
        }

        auto_fd fd;

        try
        {
          fd = fdopen (p, m);

          if ((m & fdopen_mode::at_end) != fdopen_mode::at_end)
          {
            if (rt == redirect_type::file)
              env.clean ({cleanup_type::always, p}, true);
            else
              env.clean_special (p);
          }
        }
        catch (const io_error& e)
        {
          fail (ll) << "unable to write to " << p << ": " << e;
        }

        return fd;
      };

      path osp;
      fdpipe ofd;

      // If this is the last command in the pipeline than redirect the
      // command process stdout to a file. Otherwise create a pipe and
      // redirect the stdout to the write-end of the pipe. The read-end will
      // be passed as stdin for the next command in the pipeline.
      //
      // @@ Shouldn't we allow the here-* and file output redirects for a
      //    command with pipelined output? Say if such redirect is present
      //    then the process output is redirected to a file first (as it is
      //    when no output pipelined), and only after the process exit code
      //    and the output are validated the next command in the pipeline is
      //    executed taking the file as an input. This could be usefull for
      //    script failures investigation and, for example, for validation
      //    "tightening".
      //
      if (last)
        ofd.out = open (*out, 1, osp);
      else
      {
        assert (!c.out); // No redirect expected.
        ofd = open_pipe ();
      }

      path esp;
      auto_fd efd (open (err, 2, esp));

      // Merge standard streams.
      //
      bool mo (out != nullptr && out->type == redirect_type::merge);
      if (mo || err.type == redirect_type::merge)
      {
        auto_fd& self  (mo ? ofd.out : efd);
        auto_fd& other (mo ? efd : ofd.out);

        try
        {
          assert (self.get () == -1 && other.get () != -1);
          self = fddup (other.get ());
        }
        catch (const io_error& e)
        {
          fail (ll) << "unable to duplicate " << (mo ? "stderr" : "stdout")
                    << ": " << e;
        }
      }

      // All descriptors should be open to the date.
      //
      assert (ofd.out.get () != -1 && efd.get () != -1);

      optional<process_exit> exit;
      builtin_function* bf (builtins.find (program));

      bool success;

      if (bf != nullptr)
      {
        // Execute the builtin.
        //
        if (verb >= 2)
          print_process (process_args ());

        // Some of the script builtins (cp, mkdir, etc) extend libbutl
        // builtins (via callbacks) registering/moving cleanups for the
        // filesystem entries they create/move, unless explicitly requested
        // not to do so via the --no-cleanup option.
        //
        // Let's "wrap up" the cleanup-related flags into the single object
        // to rely on "small function object" optimization.
        //
        struct cleanup
        {
          // Whether the cleanups are enabled for the builtin. Can be set to
          // false by the parse_option callback if --no-cleanup is
          // encountered.
          //
          bool enabled = true;

          // Whether to register cleanup for a filesystem entry being
          // created/updated depending on its existence. Calculated by the
          // create pre-hook and used by the subsequent post-hook.
          //
          bool add;

          // Whether to move existing cleanups for the filesystem entry
          // being moved, rather than to erase them. Calculated by the move
          // pre-hook and used by the subsequent post-hook.
          //
          bool move;
        };

        // nullopt if the builtin doesn't support cleanups.
        //
        optional<cleanup> cln;

        if (cleanup_builtin (program))
          cln = cleanup ();

        builtin_callbacks bcs {

          // create
          //
          // Unless cleanups are suppressed, test that the filesystem entry
          // doesn't exist (pre-hook) and, if that's the case, register the
          // cleanup for the newly created filesystem entry (post-hook).
          //
          [&env, &cln] (const path& p, bool pre)
          {
            // Cleanups must be supported by a filesystem entry-creating
            // builtin.
            //
            assert (cln);

            if (cln->enabled)
            {
              if (pre)
                cln->add = !butl::entry_exists (p);
              else if (cln->add)
                env.clean ({cleanup_type::always, p}, true /* implicit */);
            }
          },

          // move
          //
          // Validate the source and destination paths (pre-hook) and,
          // unless suppressed, adjust the cleanups that are sub-paths of
          // the source path (post-hook).
          //
          [&env, &cln] (const path& from, const path& to, bool force, bool pre)
          {
            // Cleanups must be supported by a filesystem entry-moving
            // builtin.
            //
            assert (cln);

            if (pre)
            {
              const dir_path& wd (*env.work_dir.path);
              const dir_path* sd (env.sandbox_dir.path);

              auto fail = [] (const string& d) {throw runtime_error (d);};

              if (sd != nullptr && !from.sub (*sd) && !force)
                fail (diag_path (from) + " is out of " +
                      diag_path (env.sandbox_dir));

              auto check_wd = [&wd, &env, fail] (const path& p)
              {
                if (wd.sub (path_cast<dir_path> (p)))
                  fail (diag_path (p) + " contains " +
                        diag_path (env.work_dir));
              };

              check_wd (from);
              check_wd (to);

              // Unless cleanups are disabled, "move" the matching cleanups
              // if the destination path doesn't exist and it is a sub-path
              // of the working directory and just remove them otherwise.
              //
              if (cln->enabled)
                cln->move = !butl::entry_exists (to) &&
                            (sd == nullptr || to.sub (*sd));
            }
            else if (cln->enabled)
            {
              // Move or remove the matching cleanups (see above).
              //
              // Note that it's not enough to just change the cleanup paths.
              // We also need to make sure that these cleanups happen before
              // the destination directory (or any of its parents) cleanup,
              // that is potentially registered. To achieve that we can just
              // relocate these cleanup entries to the end of the list,
              // preserving their mutual order. Remember that cleanups in
              // the list are executed in the reversed order.
              //
              cleanups cs;

              // Remove the source path sub-path cleanups from the list,
              // adjusting/caching them if required (see above).
              //
              for (auto i (env.cleanups.begin ()); i != env.cleanups.end (); )
              {
                script::cleanup& c (*i);
                path& p (c.path);

                if (p.sub (from))
                {
                  if (cln->move)
                  {
                    // Note that we need to preserve the cleanup path
                    // trailing separator which indicates the removal
                    // method. Also note that leaf(), in particular, does
                    // that.
                    //
                    p = p != from
                      ? to / p.leaf (path_cast<dir_path> (from))
                      : p.to_directory ()
                        ? path_cast<dir_path> (to)
                        : to;

                    cs.push_back (move (c));
                  }

                  i = env.cleanups.erase (i);
                }
                else
                  ++i;
              }

              // Re-insert the adjusted cleanups at the end of the list.
              //
              env.cleanups.insert (env.cleanups.end (),
                                   make_move_iterator (cs.begin ()),
                                   make_move_iterator (cs.end ()));

            }
          },

          // remove
          //
          // Validate the filesystem entry path (pre-hook).
          //
          [&env] (const path& p, bool force, bool pre)
          {
            if (pre)
            {
              const dir_path& wd (*env.work_dir.path);
              const dir_path* sd (env.sandbox_dir.path);

              auto fail = [] (const string& d) {throw runtime_error (d);};

              if (sd != nullptr && !p.sub (*sd) && !force)
                fail (diag_path (p) + " is out of " +
                      diag_path (env.sandbox_dir));

              if (wd.sub (path_cast<dir_path> (p)))
                fail (diag_path (p) + " contains " +
                      diag_path (env.work_dir));
            }
          },

          // parse_option
          //
          [&cln] (const strings& args, size_t i)
          {
            // Parse --no-cleanup, if it is supported by the builtin.
            //
            if (cln && args[i] == "--no-cleanup")
            {
              cln->enabled = false;
              return 1;
            }

            return 0;
          },

          // sleep
          //
          // Deactivate the thread before going to sleep.
          //
          [&env] (const duration& d)
          {
            // If/when required we could probably support the precise sleep
            // mode (e.g., via an option).
            //
            env.context.sched.sleep (d);
          }
        };

        try
        {
          uint8_t r; // Storage.
          builtin b (bf (r,
                         c.arguments,
                         move (ifd), move (ofd.out), move (efd),
                         *env.work_dir.path,
                         bcs));

          success = run_pipe (env,
                              nc,
                              ec,
                              move (ofd.in),
                              ci + 1, li, ll, diag);

          exit = process_exit (b.wait ());
        }
        catch (const system_error& e)
        {
          fail (ll) << "unable to execute " << c.program << " builtin: "
                    << e << endf;
        }
      }
      else
      {
        // Execute the process.
        //
        cstrings args (process_args ());

        // Resolve the relative not simple program path against the script's
        // working directory. The simple one will be left for the process
        // path search machinery. Also strip the potential leading `^`,
        // indicating that this is an external program rather than a
        // builtin.
        //
        path p;

        try
        {
          p = path (args[0]);

          if (p.relative ())
          {
            auto program = [&p, &args] (path pp)
            {
              p = move (pp);
              args[0] = p.string ().c_str ();
            };

            if (p.simple ())
            {
              const string& s (p.string ());

              // Don't end up with an empty path.
              //
              if (s.size () > 1 && s[0] == '^')
                program (path (s, 1, s.size () - 1));
            }
            else
              program (*env.work_dir.path / p);
          }
        }
        catch (const invalid_path& e)
        {
          fail (ll) << "invalid program path " << e.path;
        }

        try
        {
          process_path pp (process::path_search (args[0]));

          // Note: the builtin-escaping character '^' is not printed.
          //
          if (verb >= 2)
            print_process (args);

          process pr (
            pp,
            args.data (),
            {ifd.get (), -1}, process::pipe (ofd), {-1, efd.get ()},
            env.work_dir.path->string ().c_str ());

          ifd.reset ();
          ofd.out.reset ();
          efd.reset ();

          success = run_pipe (env,
                              nc,
                              ec,
                              move (ofd.in),
                              ci + 1, li, ll, diag);

          pr.wait ();

          exit = move (pr.exit);
        }
        catch (const process_error& e)
        {
          error (ll) << "unable to execute " << args[0] << ": " << e;

          if (e.child)
            std::exit (1);

          throw failed ();
        }
      }

      assert (exit);

      // If the righ-hand side pipeline failed than the whole pipeline fails,
      // and no further checks are required.
      //
      if (!success)
        return false;

      const path& pr (c.program);

      // If there is no valid exit code available by whatever reason then we
      // print the proper diagnostics, dump stderr (if cached and not too
      // large) and fail the whole script. Otherwise if the exit code is not
      // correct then we print diagnostics if requested and fail the
      // pipeline.
      //
      bool valid (exit->normal ());

      // On Windows the exit code can be out of the valid codes range being
      // defined as uint16_t.
      //
#ifdef _WIN32
      if (valid)
        valid = exit->code () < 256;
#endif

      success = valid && eq == (exit->code () == c.exit.code);

      if (!valid || (!success && diag))
      {
        // In the presense of a valid exit code we print the diagnostics and
        // return false rather than throw.
        //
        diag_record d (valid ? error (ll) : fail (ll));

        if (!exit->normal ())
          d << pr << " " << *exit;
        else
        {
          uint16_t ec (exit->code ()); // Make sure is printed as integer.

          if (!valid)
            d << pr << " exit code " << ec << " out of 0-255 range";
          else if (!success)
          {
            if (diag)
              d << pr << " exit code " << ec << (eq ? " != " : " == ")
                << static_cast<uint16_t> (c.exit.code);
          }
          else
            assert (false);
        }

        if (non_empty (esp, ll) && avail_on_failure (esp, env))
          d << info << "stderr: " << esp;

        if (non_empty (osp, ll) && avail_on_failure (osp, env))
          d << info << "stdout: " << osp;

        if (non_empty (isp, ll) && avail_on_failure (isp, env))
          d << info << "stdin: " << isp;

        // Print cached stderr.
        //
        print_file (d, esp, ll);
      }

      // If exit code is correct then check if the standard outputs match the
      // expectations. Note that stdout is only redirected to file for the
      // last command in the pipeline.
      //
      // The thinking behind matching stderr first is that if it mismatches,
      // then the program probably misbehaves (executes wrong functionality,
      // etc) in which case its stdout doesn't really matter.
      //
      if (success)
        success =
          check_output (pr, esp, isp, err, ll, env, diag, "stderr") &&
          (!last ||
           check_output (pr, osp, isp, *out, ll, env, diag, "stdout"));

      return success;
    }

    static bool
    run_expr (environment& env,
              const command_expr& expr,
              size_t li, const location& ll,
              bool diag)
    {
      // Commands are numbered sequentially throughout the expression
      // starting with 1. Number 0 means the command is a single one.
      //
      size_t ci (expr.size () == 1 && expr.back ().pipe.size () == 1
                 ? 0
                 : 1);

      // If there is no ORs to the right of a pipe then the pipe failure is
      // fatal for the whole expression. In particular, the pipe must print
      // the diagnostics on failure (if generally allowed). So we find the
      // pipe that "switches on" the diagnostics potential printing.
      //
      command_expr::const_iterator trailing_ands; // Undefined if diag is
      // disallowed.
      if (diag)
      {
        auto i (expr.crbegin ());
        for (; i != expr.crend () && i->op == expr_operator::log_and; ++i) ;
        trailing_ands = i.base ();
      }

      bool r (false);
      bool print (false);

      for (auto b (expr.cbegin ()), i (b), e (expr.cend ()); i != e; ++i)
      {
        if (diag && i + 1 == trailing_ands)
          print = true;

        const command_pipe& p (i->pipe);
        bool or_op (i->op == expr_operator::log_or);

        // Short-circuit if the pipe result must be OR-ed with true or AND-ed
        // with false.
        //
        if (!((or_op && r) || (!or_op && !r)))
          r = run_pipe (
            env, p.begin (), p.end (), auto_fd (), ci, li, ll, print);

        ci += p.size ();
      }

      return r;
    }

    void
    run (environment& env,
         const command_expr& expr,
         size_t li, const location& ll)
    {
      // Note that we don't print the expression at any verbosity level
      // assuming that the caller does this, potentially providing some
      // additional information (command type, etc).
      //
      if (!run_expr (env, expr, li, ll, true /* diag */))
        throw failed (); // Assume diagnostics is already printed.
    }

    bool
    run_if (environment& env,
            const command_expr& expr,
            size_t li, const location& ll)
    {
      // Note that we don't print the expression here (see above).
      //
      return run_expr (env, expr, li, ll, false /* diag */);
    }

    void
    clean (environment& env, const location& ll)
    {
      context& ctx (env.context);
      const dir_path& wdir (*env.work_dir.path);

      // Note that we operate with normalized paths here.
      //
      // Remove special files. The order is not important as we don't
      // expect directories here.
      //
      for (const path& p: env.special_cleanups)
      {
        // Remove the file if exists. Fail otherwise.
        //
        if (rmfile (ctx, p, 3) == rmfile_status::not_exist)
          fail (ll) << "registered for cleanup special file " << p
                    << " does not exist";
      }

      // Remove files and directories in the order opposite to the order of
      // cleanup registration.
      //
      for (const auto& c: reverse_iterate (env.cleanups))
      {
        cleanup_type t (c.type);

        // Skip whenever the path exists or not.
        //
        if (t == cleanup_type::never)
          continue;

        const path& cp (c.path);

        // Wildcard with the last component being '***' (without trailing
        // separator) matches all files and sub-directories recursively as
        // well as the start directories itself. So we will recursively
        // remove the directories that match the parent (for the original
        // path) directory wildcard.
        //
        bool recursive (cp.leaf ().representation () == "***");
        const path& p (!recursive ? cp : cp.directory ());

        // Remove files or directories using wildcard.
        //
        if (path_pattern (p))
        {
          bool removed (false);

          auto rm = [&cp, recursive, &removed, &ll, &ctx, &wdir]
                    (path&& pe, const string&, bool interm)
          {
            if (!interm)
            {
              // While removing the entry we can get not_exist due to
              // racing conditions, but that's ok if somebody did our job.
              // Note that we still set the removed flag to true in this
              // case.
              //
              removed = true; // Will be meaningless on failure.

              if (pe.to_directory ())
              {
                dir_path d (path_cast<dir_path> (pe));

                if (!recursive)
                {
                  rmdir_status r (rmdir (ctx, d, 3));

                  if (r != rmdir_status::not_empty)
                    return true;

                  diag_record dr (fail (ll));
                  dr << "registered for cleanup directory " << d
                     << " is not empty";

                  print_dir (dr, d, ll);
                  dr << info << "wildcard: '" << cp << "'";
                }
                else
                {
                  // Don't remove the working directory (it will be removed
                  // by the dedicated cleanup).
                  //
                  // Cast to uint16_t to avoid ambiguity with
                  // libbutl::rmdir_r().
                  //
                  rmdir_status r (rmdir_r (ctx, d, d != wdir, 3));

                  if (r != rmdir_status::not_empty)
                    return true;

                  // The directory is unlikely to be current but let's keep
                  // for completeness.
                  //
                  fail (ll) << "registered for cleanup wildcard " << cp
                            << " matches the current directory";
                }
              }
              else
                rmfile (ctx, pe, 3);
            }

            return true;
          };

          // Note that here we rely on the fact that recursive iterating
          // goes depth-first (which make sense for the cleanup).
          //
          try
          {
            // Doesn't follow symlinks.
            //
            path_search (p,
                         rm,
                         dir_path () /* start */,
                         path_match_flags::none);
          }
          catch (const system_error& e)
          {
            fail (ll) << "unable to cleanup wildcard " << cp << ": " << e;
          }

          // Removal of no filesystem entries is not an error for 'maybe'
          // cleanup type.
          //
          if (removed || t == cleanup_type::maybe)
            continue;

          fail (ll) << "registered for cleanup wildcard " << cp
                    << " doesn't match any "
                    << (recursive
                        ? "path"
                        : p.to_directory ()
                        ? "directory"
                        : "file");
        }

        // Remove the directory if exists and empty. Fail otherwise.
        // Removal of non-existing directory is not an error for 'maybe'
        // cleanup type.
        //
        if (p.to_directory ())
        {
          dir_path d (path_cast<dir_path> (p));
          bool wd (d == wdir);

          // Don't remove the working directory for the recursive cleanup
          // (it will be removed by the dedicated one).
          //
          // Note that the root working directory contains the
          // .buildignore file (see above).
          //
          // @@ If 'd' is a file then will fail with a diagnostics having
          //    no location info. Probably need to add an optional location
          //    parameter to rmdir() function. The same problem exists for
          //    a file cleanup when try to rmfile() directory instead of
          //    file.
          //
          rmdir_status r (recursive
                          ? rmdir_r (ctx, d, !wd, static_cast <uint16_t> (3))
                          : rmdir (ctx, d, 3));

          if (r == rmdir_status::success ||
              (r == rmdir_status::not_exist && t == cleanup_type::maybe))
            continue;

          diag_record dr (fail (ll));
          dr << "registered for cleanup directory " << d
             << (r == rmdir_status::not_exist ? " does not exist" :
                 !recursive                   ? " is not empty"
                                              : " is current");

          if (r == rmdir_status::not_empty)
            print_dir (dr, d, ll);
        }

        // Remove the file if exists. Fail otherwise. Removal of
        // non-existing file is not an error for 'maybe' cleanup type.
        //
        if (rmfile (ctx, p, 3) == rmfile_status::not_exist &&
            t == cleanup_type::always)
          fail (ll) << "registered for cleanup file " << p
                    << " does not exist";
      }
    }

    void
    print_dir (diag_record& dr, const dir_path& p, const location& ll)
    {
      try
      {
        size_t n (0);
        for (const dir_entry& de: dir_iterator (p,
                                                false /* ignore_dangling */))
        {
          if (n++ < 10)
            dr << '\n' << (de.ltype () == entry_type::directory
                           ? path_cast<dir_path> (de.path ())
                           : de.path ());
        }

        if (n > 10)
          dr << "\nand " << n - 10 << " more file(s)";
      }
      catch (const system_error& e)
      {
        fail (ll) << "unable to iterate over " << p << ": " << e;
      }
    }
  }
}