aws-s3-transfer 1.0.0

Transfer Manager for Amazon S3 - a high-performance data plane client for uploading and downloading objects and directories.
Documentation
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
/*
 * Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
 * SPDX-License-Identifier: Apache-2.0
 */

//! State machine for `download_objects` composite transfer.
//!
//! Mirrors the `upload_objects` state machine: a single S3 listing walker
//! discovers objects, the parent spawns child downloads (via
//! [`Download::orchestrate_with_sink`]), and reaps them as they complete.

use parking_lot::Mutex;
use path_clean::PathClean;
use std::borrow::Cow;
use std::collections::{HashMap, HashSet};
use std::fmt;
use std::path::{Component, Path, PathBuf};
use std::pin::Pin;
use std::sync::Arc;

use crate::error::{self, Error, ErrorKind};
use crate::io::walk::S3Walk;
use crate::model::Object;
use crate::operation::download::sink::{FileSinkFactory, SinkFactory};
use crate::operation::download::{Download, DownloadInput, ManagedDownloadHandle};
use crate::operation::download_objects::observability::{
    emit_event, emit_terminal_summary, DownloadObjectsEvent, DownloadObjectsStateSnapshot,
};
use crate::telemetry::TARGET_TRANSFER_DOWNLOAD_OBJECTS;
use crate::transfer::CompositePendingReason;
use crate::transfer::{IoRequest, PollWork, Transfer, TransferContext, TransferId, WorkOutcome};
use crate::types::{FailedDownload, FailedTransferPolicy};

/// Maximum terminal children drained into a single `JoinChildren` work item per
/// poll. NOT an accumulation threshold: `poll_work` reaps whatever terminals
/// exist every poll (fused with spawning), so this only caps how many
/// already-terminal handles one `execute_join_children` awaits back-to-back,
/// bounding worker-thread hold time on the serial join. Kept modest so a poll's
/// reap work stays small and terminals are retired continuously rather than in
/// large lumps.
const MAX_REAP_PER_POLL: usize = 64;

/// Maximum entries to drain from the walker per AdvanceWalker work item.
/// Prevents a single work item from holding the executor across multiple
/// ListObjectsV2 pages. Matches S3's default MaxKeys.
const MAX_ENTRIES_PER_WALK: usize = 1000;

/// Low-water mark for the key buffer: the walker dispatches the next listing
/// page while `pending_entries` is below this, keeping the buffer fed ahead of
/// spawn draining it. Sized to one page so a single in-flight advance refills
/// more than a page's worth of runway before the buffer empties. Combined with
/// single-flight dispatch, this bounds `pending_entries` to roughly one page
/// above the mark.
const WALK_LOW_WATER: usize = MAX_ENTRIES_PER_WALK;

/// Number of discovered keys accumulated before publishing them to
/// `pending_entries` mid-advance. Small enough that spawning sees keys quickly
/// while a page is still being fetched, large enough to amortize the state-lock
/// acquisition across many keys.
const WALK_FLUSH_CHUNK: usize = 64;

/// Work items produced by [`DownloadObjectsTransfer::poll_work`].
pub(crate) enum DownloadObjectsWork {
    /// Advance the S3 listing walker by one page/entry. The walker is boxed:
    /// `S3Walk` is large (hundreds of bytes), and boxing keeps the work enum
    /// small for the common `JoinChildren` variant (large_enum_variant).
    AdvanceWalker { walk: Option<Box<S3Walk>> },
    /// Join a batch of terminal child downloads.
    JoinChildren { batch: Option<ReapingBatch> },
}

impl DownloadObjectsWork {
    /// Short name of the work kind for span fields.
    fn kind(&self) -> &'static str {
        match self {
            Self::AdvanceWalker { .. } => "advance_walker",
            Self::JoinChildren { .. } => "join_children",
        }
    }
}

impl std::fmt::Debug for DownloadObjectsWork {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        match self {
            Self::AdvanceWalker { .. } => f.debug_struct("AdvanceWalker").finish_non_exhaustive(),
            Self::JoinChildren { batch } => f
                .debug_struct("JoinChildren")
                .field("count", &batch.as_ref().map(|b| b.count))
                .finish(),
        }
    }
}

/// Tracked child download — handle + the S3 key for error reporting.
pub(crate) struct ChildTransfer {
    pub(crate) handle: ManagedDownloadHandle,
    pub(crate) key: String,
}

/// Owns a reservation of N slots in `State::children_reserved`. Either
/// consumed by [`DownloadObjectsTransfer::merge_spawned`] (paired decrement
/// under the state lock) or released on `Drop` if `consume` is not reached.
/// Constructed only by [`DownloadObjectsTransfer::claim_one`].
///
/// The type makes the counter pairing structural: `merge_spawned`'s signature
/// requires the token, so a caller cannot forget to release the reservation,
/// and the count released always matches the count reserved.
struct Reservation {
    transfer: Arc<DownloadObjectsTransferInner>,
    count: usize,
    consumed: bool,
}

impl Reservation {
    /// Caller holds `state.lock`. Decrements `children_reserved` by the owned
    /// count and marks the reservation consumed. Always paired 1:1 with the
    /// `claim_one` that produced this token.
    fn consume(mut self, state: &mut State) {
        debug_assert!(
            !self.consumed,
            "Reservation::consume called on already-consumed reservation"
        );
        debug_assert!(
            state.children_reserved >= self.count,
            "children_reserved underflow: have {}, releasing {}",
            state.children_reserved,
            self.count,
        );
        state.children_reserved -= self.count;
        self.consumed = true;
    }
}

impl Drop for Reservation {
    fn drop(&mut self) {
        if !self.consumed && self.count > 0 {
            let mut state = self.transfer.state.lock();
            state.children_reserved = state.children_reserved.saturating_sub(self.count);
            tracing::warn!(
                target: crate::telemetry::TARGET_TRANSFER_DOWNLOAD_OBJECTS,
                tid = %self.transfer.ctx.id,
                count = self.count,
                "Reservation dropped unconsumed; children_reserved recovered"
            );
        }
    }
}

/// Owns N slots in `State::reaping_in_flight` plus the drained
/// [`ChildTransfer`]s themselves. Either consumed by
/// [`DownloadObjectsTransfer::execute_join_children`] (paired decrement under
/// the state lock) or released on `Drop` if the work item is dropped without
/// executing or `execute_join_children` panics.
///
/// Travels across the `IoRequest` dispatch boundary embedded in
/// [`DownloadObjectsWork::JoinChildren`]. On `Drop` without `consume`, the
/// counter is recovered and the children's handles are dropped, cascading
/// cancellation via their own `Drop`.
pub(crate) struct ReapingBatch {
    transfer: Arc<DownloadObjectsTransferInner>,
    /// Drained terminal children, awaiting `join()` in
    /// `execute_join_children`. Taken out for processing via
    /// [`Self::take_children`]; the counter is still owned by the batch until
    /// `consume()` is called or `Drop` runs.
    children: Option<Vec<ChildTransfer>>,
    count: usize,
    consumed: bool,
}

impl fmt::Debug for ReapingBatch {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.debug_struct("ReapingBatch")
            .field("count", &self.count)
            .field("consumed", &self.consumed)
            .field("children_len", &self.children.as_ref().map(|c| c.len()))
            .finish_non_exhaustive()
    }
}

impl ReapingBatch {
    /// Take the inner children Vec out for processing. Must be called exactly
    /// once before `consume`.
    fn take_children(&mut self) -> Vec<ChildTransfer> {
        self.children
            .take()
            .expect("ReapingBatch::take_children called twice or after consume")
    }

    /// Caller holds `state.lock`. Decrements `reaping_in_flight` by the owned
    /// count and marks the batch consumed.
    fn consume(mut self, state: &mut State) {
        debug_assert!(
            !self.consumed,
            "ReapingBatch::consume called on already-consumed batch"
        );
        debug_assert!(
            state.reaping_in_flight >= self.count,
            "reaping_in_flight underflow: have {}, releasing {}",
            state.reaping_in_flight,
            self.count,
        );
        state.reaping_in_flight -= self.count;
        self.consumed = true;
    }
}

impl Drop for ReapingBatch {
    fn drop(&mut self) {
        if !self.consumed {
            let mut state = self.transfer.state.lock();
            state.reaping_in_flight = state.reaping_in_flight.saturating_sub(self.count);
            tracing::warn!(
                target: crate::telemetry::TARGET_TRANSFER_DOWNLOAD_OBJECTS,
                tid = %self.transfer.ctx.id,
                count = self.count,
                "ReapingBatch dropped unconsumed; reaping_in_flight recovered"
            );
            // Children Vec (if still present) drops here. Each
            // ManagedDownloadHandle's own Drop cancels its child transfer via
            // scheduler.cancel_transfer, so orphans don't leak.
        }
    }
}

struct State {
    /// The S3 listing walker (taken out during AdvanceWalker execution).
    walk: Option<S3Walk>,

    /// Whether the destination directory has been prepared: checked to be a
    /// directory, or created. Done once, on the first walker advance, so the
    /// blocking filesystem calls run off the caller thread.
    validated: bool,

    /// Whether a walker work item is currently in flight.
    walk_in_flight: bool,

    /// Objects discovered by the walker, waiting to be spawned as children.
    pending_entries: std::collections::VecDeque<Object>,

    /// Active child downloads keyed by their TransferId.
    children: HashMap<TransferId, ChildTransfer>,

    /// Entries claimed from pending_entries but not yet inserted into children.
    children_reserved: usize,

    /// JoinChildren batches currently executing.
    reaping_in_flight: usize,

    /// Failed downloads (for FailedTransferPolicy::Continue).
    failed: Vec<FailedDownload>,

    /// Count of successfully downloaded objects.
    successful_downloads: u64,
}

impl State {
    /// Counters reported with parent state-machine events.
    fn snapshot(&self) -> DownloadObjectsStateSnapshot {
        DownloadObjectsStateSnapshot {
            pending_entries: self.pending_entries.len(),
            walk_in_flight: self.walk_in_flight,
            children: self.children.len(),
            children_reserved: self.children_reserved,
            reaping_in_flight: self.reaping_in_flight,
            successful: self.successful_downloads,
            failed: self.failed.len(),
        }
    }

    /// Reason the parent is pending, read from the same fields as
    /// `check_terminal`.
    ///
    /// Called only at the `Pending` return of `poll_work`, after that poll
    /// dispatched no walk, claimed no entry, and reaped nothing. Waiting
    /// entries therefore mean the child limit is reached.
    fn pending_reason(&self, active: bool) -> CompositePendingReason {
        let listing = self.walk_in_flight || self.walk.as_ref().is_some_and(|w| !w.is_done());
        if !active {
            CompositePendingReason::Draining
        } else if self.reaping_in_flight > 0 {
            CompositePendingReason::Reaping
        } else if !self.pending_entries.is_empty() {
            CompositePendingReason::ChildCapacity
        } else if listing {
            CompositePendingReason::Listing
        } else {
            CompositePendingReason::ChildCompletion
        }
    }
}

#[derive(Clone)]
pub(crate) struct DownloadObjectsTransfer {
    inner: Arc<DownloadObjectsTransferInner>,
}

impl std::fmt::Debug for DownloadObjectsTransfer {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("DownloadObjectsTransfer")
            .finish_non_exhaustive()
    }
}

struct DownloadObjectsTransferInner {
    ctx: TransferContext,
    state: Mutex<State>,
    bucket: String,
    /// The caller's destination, lexically cleaned. Every local path the
    /// transfer creates or checks is derived from this one form.
    destination: PathBuf,
    delimiter: Option<String>,
    failure_policy: FailedTransferPolicy,
    pipeline_depth: usize,
    /// Directories already created (dedup `create_dir_all` across children
    /// that share a prefix; objects in the same S3 "folder" map to one dir).
    created_dirs: Mutex<HashSet<PathBuf>>,
    #[cfg(test)]
    last_summary: Mutex<Option<crate::transfer::CompositeTransferSummary>>,
}

impl DownloadObjectsTransfer {
    pub(crate) fn new(
        ctx: TransferContext,
        input: &crate::operation::download_objects::DownloadObjectsInput,
        walk: S3Walk,
        pipeline_depth: usize,
    ) -> Self {
        let bucket = input.bucket().unwrap().to_string();
        let destination = input.destination().unwrap().clean();
        let delimiter = input.delimiter().map(|s| s.to_string());
        let failure_policy = input.failure_policy().clone();

        Self {
            inner: Arc::new(DownloadObjectsTransferInner {
                ctx,
                state: Mutex::new(State {
                    walk: Some(walk),
                    validated: false,
                    walk_in_flight: false,
                    pending_entries: std::collections::VecDeque::new(),
                    children: HashMap::new(),
                    children_reserved: 0,
                    reaping_in_flight: 0,
                    failed: Vec::new(),
                    successful_downloads: 0,
                }),
                bucket,
                destination,
                delimiter,
                failure_policy,
                pipeline_depth,
                created_dirs: Mutex::new(HashSet::new()),
                #[cfg(test)]
                last_summary: Mutex::new(None),
            }),
        }
    }
    pub(crate) fn ctx(&self) -> &TransferContext {
        &self.inner.ctx
    }

    pub(crate) fn successful_downloads(&self) -> u64 {
        // Safe: only called after terminal
        self.inner.state.lock().successful_downloads
    }

    pub(crate) fn take_failed(&self) -> Option<Vec<FailedDownload>> {
        let mut state = self.inner.state.lock();
        if state.failed.is_empty() {
            None
        } else {
            Some(std::mem::take(&mut state.failed))
        }
    }

    /// Produce work for one poll. Spawn and reap run in the same poll and are
    /// fused into a single outcome rather than competing as strict-priority
    /// ladder steps:
    ///
    /// 1. Walk (refill): if the walker can dispatch, return its `AdvanceWalker`
    ///    item. A walk poll is the whole poll (neither spawns nor reaps).
    /// 2. Spawn one child (lock released around orchestration).
    /// 3. Reap any terminal children (up to MAX_REAP_PER_POLL this poll).
    /// 4. Fuse: a reap returns `Ready { io: JoinChildren, spawned }`; with no
    ///    reap, `Spawned` if an entry was claimed, else `check_terminal` /
    ///    `Pending`.
    ///
    /// Fusing means one poll can both retire a terminal child and refill, so a
    /// continuous completion stream does not force reap and spawn onto separate
    /// turns and done children are retired as they appear rather than
    /// accumulating. Walk and spawn are skipped when inactive, so cancel/fail
    /// stops new work while in-flight drains.
    #[tracing::instrument(
        level = "trace",
        target = TARGET_TRANSFER_DOWNLOAD_OBJECTS,
        skip_all,
        fields(tid = %self.inner.ctx.id),
    )]
    pub(crate) fn poll_work(&self) -> PollWork {
        let mut state = self.inner.state.lock();

        // 1: Walk (refill). `dispatch_walk` gates on the low-water mark, so it
        // tops off the key buffer ahead of spawn draining it rather than
        // waiting until empty, avoiding stalls on ListObjectsV2 page latency.
        if self.inner.ctx.is_active() {
            if let Some(work) = self.dispatch_walk(&mut state) {
                return work;
            }
        }

        // 2: Spawn one child (lock released around orchestration).
        //
        // `attempted` = an entry was claimed from `pending_entries` this poll,
        // regardless of whether its child orchestrated. It MUST drive re-poll: a
        // claimed entry has left the queue, so the parent has to stay scheduled
        // to drain the rest — even after a failed orchestration — or the
        // remaining entries strand with no wake once the walker is exhausted (a
        // hang; see `test_two_failed_spawns_continue_managed_runtime_does_not_hang`).
        // `materialized` = a live child was actually inserted; it drives the
        // spawn vruntime charge, so a failed orchestration is not charged.
        let mut attempted = false;
        let mut materialized = false;
        if self.inner.ctx.is_active() {
            let (to_spawn, reservation) = self.claim_one(&mut state);
            if !to_spawn.is_empty() {
                attempted = true;
                drop(state);
                let spawned = self.spawn_children(to_spawn);
                state = self.inner.state.lock();
                materialized = self.merge_spawned(&mut state, spawned, reservation) > 0;
            }
        }

        // 3: Reap terminal children (up to MAX_REAP_PER_POLL this poll, no
        // accumulation threshold — done children are retired as they appear).
        let reaped = self.drain_terminal_children(&mut state);

        // 4: Fuse spawn and reap into one poll outcome. A reap dispatches a
        // JoinChildren item and carries `spawned` so the scheduler charges spawn
        // vruntime iff a child materialized this poll — one poll both refills and
        // retires. With no reap, re-poll if we touched the queue; otherwise
        // settle terminal/idle.
        if let Some(batch) = reaped {
            return PollWork::Ready {
                io: IoRequest {
                    data: Some(Box::new(DownloadObjectsWork::JoinChildren {
                        batch: Some(batch),
                    })),
                },
                spawned: materialized,
            };
        }
        if attempted {
            return PollWork::Spawned;
        }
        if let Some(result) = self.check_terminal(&state) {
            return result;
        }
        let reason = state.pending_reason(self.inner.ctx.is_active());
        self.inner.ctx.set_pending(reason);
        PollWork::Pending
    }

    /// Collect up to MAX_REAP_PER_POLL children that have reached a terminal
    /// state (success, failure, or cancellation) for reaping via JoinChildren.
    fn drain_terminal_children(&self, state: &mut State) -> Option<ReapingBatch> {
        let terminal_ids: Vec<TransferId> = state
            .children
            .iter()
            .filter(|(_, child)| child.handle.status() != crate::types::TransferStatus::Active)
            .map(|(id, _)| *id)
            .take(MAX_REAP_PER_POLL)
            .collect();

        if terminal_ids.is_empty() {
            return None;
        }

        let batch: Vec<ChildTransfer> = terminal_ids
            .into_iter()
            .filter_map(|id| state.children.remove(&id))
            .collect();

        tracing::trace!(
            target: crate::telemetry::TARGET_TRANSFER_DOWNLOAD_OBJECTS,
            tid = %self.inner.ctx.id,
            reaping = batch.len(),
            remaining_children = state.children.len(),
            "download_objects draining terminal children",
        );
        let count = batch.len();
        // Reserve reaping_in_flight by the child count (not 1 per batch): the
        // terminal-check guards compare reaping_in_flight == 0, so the counter
        // must track children in flight, released 1:1 by the batch's consume()
        // or Drop.
        state.reaping_in_flight += count;
        emit_event(
            &self.inner.ctx,
            DownloadObjectsEvent::ReapScheduled,
            state.snapshot(),
        );
        Some(ReapingBatch {
            transfer: self.inner.clone(),
            children: Some(batch),
            count,
            consumed: false,
        })
    }

    /// Claim exactly ONE pending entry to spawn as a child.
    ///
    /// Gates on the *in-flight budget*: active children plus reserved, against
    /// `pipeline_depth`. Returns an empty Vec when nothing can be claimed.
    fn claim_one(&self, state: &mut State) -> (Vec<Object>, Reservation) {
        let active = state
            .children
            .values()
            .filter(|c| c.handle.status() == crate::types::TransferStatus::Active)
            .count()
            + state.children_reserved;
        let available = self.inner.pipeline_depth.saturating_sub(active);
        let count = available.min(1).min(state.pending_entries.len());

        let batch: Vec<Object> = state.pending_entries.drain(..count).collect();
        state.children_reserved += batch.len();
        // When batch is empty, count is 0. The resulting Reservation drops
        // harmlessly: Reservation::Drop is guarded by `self.count > 0`.
        let reservation = Reservation {
            transfer: self.inner.clone(),
            count: batch.len(),
            consumed: false,
        };
        (batch, reservation)
    }

    fn spawn_children(&self, entries: Vec<Object>) -> Vec<(Object, Result<ChildTransfer, Error>)> {
        let handle = &self.inner.ctx.handle;
        let parent_id = self.inner.ctx.id.id;

        tracing::trace!(
            target: crate::telemetry::TARGET_TRANSFER_DOWNLOAD_OBJECTS,
            tid = %self.inner.ctx.id,
            count = entries.len(),
            "download_objects spawning children",
        );

        entries
            .into_iter()
            .map(|obj| {
                let key = obj
                    .key()
                    .expect("S3Walk yields objects with keys")
                    .to_string();
                let listed = crate::operation::download::ListedObject::from_listing(&obj);
                let result = self.spawn_single_child(handle, &key, parent_id, listed);
                (obj, result.map(|h| ChildTransfer { handle: h, key }))
            })
            .collect()
    }

    /// Ensures the destination is a directory, creating it and any missing
    /// parents, with default permissions, when it does not exist. Blocking;
    /// called once from the first walker advance, during work-item execution
    /// rather than on the caller's `initiate` thread.
    ///
    /// Returns [`ErrorKind::InputInvalid`] when the destination exists and is
    /// not a directory, and [`ErrorKind::IOError`] when
    /// its metadata cannot be read or it cannot be created.
    fn prepare_destination(&self) -> Result<(), Error> {
        let dest = &self.inner.destination;
        match std::fs::metadata(dest) {
            Ok(metadata) => crate::operation::validate_target_is_dir(&metadata, dest),
            Err(e) if e.kind() == std::io::ErrorKind::NotFound => std::fs::create_dir_all(dest)
                .map_err(|e| error::io_at("creating directory", dest, e)),
            Err(e) => Err(error::io_at("reading metadata of", dest, e)),
        }
    }

    fn spawn_single_child(
        &self,
        handle: &Arc<crate::client::Handle>,
        key: &str,
        parent_id: u64,
        listed: Option<crate::operation::download::ListedObject>,
    ) -> Result<ManagedDownloadHandle, Error> {
        let dest_path = local_key_path(
            &self.inner.destination,
            key,
            self.inner.delimiter.as_deref(),
        )?;

        // Create parent directories, deduped via created_dirs so we don't
        // re-stat the whole ancestor chain for every object sharing a prefix
        // (create_dir_all stats each ancestor; for N small files in a few
        // dirs that's the dominant per-child syscall cost).
        //
        // TODO(vnext): revisit moving this to the execute phase. poll_work is
        // documented as non-blocking, but the scheduler is edge-triggered so
        // this often already runs on a managed thread; deferring to execute adds
        // a work-item round-trip that may increase total latency rather than
        // reduce it. Measure before moving.
        if let Some(parent) = dest_path.parent() {
            let known = self.inner.created_dirs.lock().contains(parent);
            if !known {
                std::fs::create_dir_all(parent).map_err(|e| {
                    error::io_at("creating directory", parent, e).with_local_key(key)
                })?;
                self.inner.created_dirs.lock().insert(parent.to_path_buf());
            }
        }

        let input = DownloadInput::builder()
            .bucket(&self.inner.bucket)
            .key(key)
            .build()
            .expect("bucket and key are set");

        // Create temp file synchronously (we're in poll_work, not async).
        // orchestrate_to_path is async; for child spawning we use the
        // synchronous helper + orchestrate_with_sink directly.
        let (file, temp_path) =
            crate::io::fs::create_first_new(crate::operation::download::temp_file_candidates(
                &dest_path,
                std::iter::repeat_with(|| fastrand::u32(..)),
            ))
            .map_err(|e| {
                error::io_at("creating temporary file for", &dest_path, e).with_local_key(key)
            })?;
        let sink = FileSinkFactory.create(file, true).map_err(|e| {
            // No handle owns the temp file yet, so remove it here.
            let _ = std::fs::remove_file(&temp_path);
            error::io_at("preparing temporary file for", &dest_path, e).with_local_key(key)
        })?;

        let inner =
            Download::orchestrate_with_sink(handle.clone(), input, sink, Some(parent_id), listed)?;
        Ok(ManagedDownloadHandle::new(inner, temp_path, dest_path))
    }

    /// Merge results of child spawning back into state. Inserts each child into
    /// the active set or records the failure, then consumes the [`Reservation`]
    /// in lockstep so the `children_reserved` pairing is structural.
    ///
    /// Returns the number of children actually inserted (orchestrated
    /// successfully). Callers use this to decide whether a spawn genuinely
    /// occurred: a claimed entry whose orchestration failed is recorded in
    /// `failed` but produces no live child, so it must not be counted as a spawn
    /// (which would over-charge spawn vruntime for a child that never runs).
    fn merge_spawned(
        &self,
        state: &mut State,
        spawned: Vec<(Object, Result<ChildTransfer, Error>)>,
        reservation: Reservation,
    ) -> usize {
        let mut inserted = 0usize;
        for (obj, result) in spawned {
            match result {
                Ok(child) => {
                    let id = child.handle.transfer_id();
                    tracing::trace!(
                        target: crate::telemetry::TARGET_TRANSFER_DOWNLOAD_OBJECTS,
                        tid = %self.inner.ctx.id,
                        child_id = %id,
                        key = %child.key,
                        "spawned child download"
                    );
                    state.children.insert(id, child);
                    inserted += 1;
                }
                Err(err) => {
                    let key = obj
                        .key()
                        .expect("S3Walk yields objects with keys")
                        .to_string();
                    tracing::warn!(
                        target: crate::telemetry::TARGET_TRANSFER_DOWNLOAD_OBJECTS,
                        tid = %self.inner.ctx.id,
                        key = %key,
                        error = %err,
                        "failed to spawn child download"
                    );
                    let failed_input = DownloadInput::builder()
                        .bucket(&self.inner.bucket)
                        .key(&key)
                        .build()
                        .unwrap();
                    state.failed.push(FailedDownload {
                        input: failed_input,
                        error: err,
                    });
                    if self.inner.failure_policy == FailedTransferPolicy::Abort {
                        self.inner.ctx.set_failed(child_failed());
                        self.report_terminal(state.snapshot());
                        self.inner.ctx.signal_terminal();
                    }
                }
            }
        }
        // Release the reservation under the same lock: count reserved by
        // claim_one == results merged here.
        reservation.consume(state);
        inserted
    }

    /// Emits the terminal summary once. The status must already be terminal.
    ///
    /// `snapshot` is the state the transfer ended in. Only the first call per
    /// transfer emits. Called under the state lock that made the status
    /// terminal, before the terminal signal, and from `on_terminal` for
    /// terminations from outside: once the status is terminal, the scheduler
    /// may remove the transfer when its dispatched work retires, without
    /// polling it again.
    fn report_terminal(&self, snapshot: DownloadObjectsStateSnapshot) {
        self.inner.ctx.finalize_terminal_metrics();
        if !self.inner.ctx.claim_terminal_report() {
            return;
        }
        let _summary = emit_terminal_summary(&self.inner.ctx, snapshot);
        #[cfg(test)]
        {
            *self.inner.last_summary.lock() = _summary;
        }
    }

    /// Returns the terminal report emitted by this transfer.
    #[cfg(test)]
    pub(crate) fn test_terminal_summary(
        &self,
    ) -> Option<crate::transfer::CompositeTransferSummary> {
        self.inner.last_summary.lock().clone()
    }

    fn check_terminal(&self, state: &State) -> Option<PollWork> {
        if !self.inner.ctx.is_active() {
            // Wait for all in-flight work to drain
            if state.children.is_empty()
                && state.children_reserved == 0
                && state.reaping_in_flight == 0
                && !state.walk_in_flight
            {
                tracing::debug!(
                    target: crate::telemetry::TARGET_TRANSFER_DOWNLOAD_OBJECTS,
                    tid = %self.inner.ctx.id,
                    successful = state.successful_downloads,
                    failed = state.failed.len(),
                    "download_objects terminal (cancelled/failed), signaling",
                );
                self.report_terminal(state.snapshot());
                self.inner.ctx.signal_terminal();
                return Some(PollWork::Done);
            }
            return None;
        }

        // Check if walk is done and no more work to do
        let walk_done = state.walk.as_ref().is_none_or(|w| w.is_done()) && !state.walk_in_flight;
        if walk_done
            && state.pending_entries.is_empty()
            && state.children.is_empty()
            && state.children_reserved == 0
            && state.reaping_in_flight == 0
        {
            tracing::debug!(
                target: crate::telemetry::TARGET_TRANSFER_DOWNLOAD_OBJECTS,
                tid = %self.inner.ctx.id,
                successful = state.successful_downloads,
                failed = state.failed.len(),
                "download_objects terminal (walk exhausted), signaling",
            );
            // The success transition: status is still Active here (the
            // !is_active() branch above handles an already-terminal drain after
            // set_failed/set_cancelled, where completing would be wrong).
            self.inner.ctx.set_completed();
            self.report_terminal(state.snapshot());
            self.inner.ctx.signal_terminal();
            return Some(PollWork::Done);
        }

        None
    }

    /// Dispatch one listing page, or `None` if listing cannot proceed.
    ///
    /// Single-flight (`walk_in_flight`) and gated on the key buffer alone: a
    /// page is fetched only while `pending_entries` is below the low-water
    /// mark, so the walker refills the buffer ahead of spawn draining it. This
    /// bounds `pending_entries` to roughly low-water plus one page (single
    /// flight admits at most one in-flight page). Independent of the child
    /// in-flight budget (`pipeline_depth`), which gates spawning, not listing.
    fn dispatch_walk(&self, state: &mut State) -> Option<PollWork> {
        if state.walk_in_flight {
            return None;
        }
        if state.pending_entries.len() >= WALK_LOW_WATER {
            return None;
        }

        // Take the walk out for execution; put it back if it's already done.
        let walk = match state.walk.take() {
            Some(w) if !w.is_done() => w,
            Some(w) => {
                state.walk = Some(w);
                return None;
            }
            None => return None,
        };

        state.walk_in_flight = true;
        emit_event(
            &self.inner.ctx,
            DownloadObjectsEvent::WalkDispatched,
            state.snapshot(),
        );
        Some(PollWork::ready(IoRequest {
            data: Some(Box::new(DownloadObjectsWork::AdvanceWalker {
                walk: Some(Box::new(walk)),
            })),
        }))
    }

    // -----------------------------------------------------------------------
    // execute
    // -----------------------------------------------------------------------

    #[tracing::instrument(
        level = "trace",
        target = TARGET_TRANSFER_DOWNLOAD_OBJECTS,
        skip_all,
        fields(tid = %self.inner.ctx.id, work = tracing::field::Empty),
    )]
    pub(crate) async fn execute(&self, work: &mut IoRequest) -> WorkOutcome {
        let data = work
            .data
            .as_mut()
            .expect("work data must be set")
            .as_any_mut()
            .downcast_mut::<DownloadObjectsWork>()
            .expect("work data must be DownloadObjectsWork");
        tracing::Span::current().record("work", data.kind());

        match data {
            DownloadObjectsWork::AdvanceWalker { walk } => {
                self.execute_advance_walker(*walk.take().unwrap()).await
            }
            DownloadObjectsWork::JoinChildren { batch } => {
                let batch = batch.take().expect("batch already taken");
                self.execute_join_children(batch).await
            }
        }
    }

    async fn execute_advance_walker(&self, mut walk: S3Walk) -> WorkOutcome {
        // Prepare the destination directory once, on the first advance, so the
        // blocking stat and directory creation run here (during work-item
        // execution) rather than on the caller's `initiate` thread, and before
        // any child is spawned into it. On failure, fail the transfer; the
        // error surfaces from `handle.join()`.
        let needs_validation = {
            let mut state = self.inner.state.lock();
            let first = !state.validated;
            state.validated = true;
            first
        };
        if needs_validation {
            if let Err(err) = self.prepare_destination() {
                let mut state = self.inner.state.lock();
                state.walk_in_flight = false;
                // Fail the transfer and write its summary; the terminal signal waits until in-flight work drains.
                self.inner.ctx.set_failed(err);
                self.report_terminal(state.snapshot());
                if self.check_terminal(&state).is_some() {
                    return WorkOutcome::Success { data: None };
                }
                drop(state);
                self.inner.ctx.try_wake();
                return WorkOutcome::Success { data: None };
            }
        }

        // Drain up to a page worth of entries from the walker, publishing them
        // to `pending_entries` in small chunks as they arrive rather than in a
        // single batch at the end. Early publication lets spawning consume keys
        // while the rest of the page is still being fetched, so the buffer does
        // not sit empty across a ListObjectsV2 round-trip.
        let mut chunk = Vec::with_capacity(WALK_FLUSH_CHUNK);
        let mut discovered = 0usize;
        let mut fatal_error = None;

        loop {
            match walk.next().await {
                Some(Ok(obj)) => {
                    chunk.push(obj);
                    discovered += 1;
                    if chunk.len() >= WALK_FLUSH_CHUNK {
                        let mut state = self.inner.state.lock();
                        state.pending_entries.extend(chunk.drain(..));
                        drop(state);
                        // Wake so a spawn poll runs against the just-published
                        // keys without waiting for this advance to complete.
                        self.inner.ctx.try_wake();
                    }
                    // Limit to one page worth of results per work item so we
                    // don't hold the executor across multiple ListObjectsV2
                    // calls, starving other transfers.
                    if discovered >= MAX_ENTRIES_PER_WALK {
                        break;
                    }
                }
                Some(Err(err)) => {
                    if err.is_fatal() {
                        fatal_error = Some(err);
                        break;
                    }
                    // Non-fatal errors: log and continue
                    tracing::warn!(
                        target: crate::telemetry::TARGET_TRANSFER_DOWNLOAD_OBJECTS,
                        tid = %self.inner.ctx.id,
                        error = %err,
                        "non-fatal walker error, continuing"
                    );
                }
                None => break, // walker exhausted or needs next page
            }
        }

        // Publish any remaining keys and put the walk back into state.
        let mut state = self.inner.state.lock();
        state.walk_in_flight = false;

        if let Some(err) = fatal_error {
            tracing::debug!(
                target: crate::telemetry::TARGET_TRANSFER_DOWNLOAD_OBJECTS,
                tid = %self.inner.ctx.id,
                error = %err,
                "download_objects fatal walker error",
            );
            self.inner.ctx.set_failed(err);
            self.report_terminal(state.snapshot());
        } else {
            tracing::trace!(
                target: crate::telemetry::TARGET_TRANSFER_DOWNLOAD_OBJECTS,
                tid = %self.inner.ctx.id,
                discovered,
                walk_done = walk.is_done(),
                "download_objects walker advanced",
            );
            state.pending_entries.extend(chunk.drain(..));
            if !walk.is_done() {
                state.walk = Some(walk);
            } else {
                emit_event(
                    &self.inner.ctx,
                    DownloadObjectsEvent::ListingExhausted,
                    state.snapshot(),
                );
            }
        }

        // An execute callback that drains the last in-flight work owns the
        // terminal transition: check and signal here rather than deferring to
        // a subsequent poll_work.
        if self.check_terminal(&state).is_some() {
            drop(state);
            return WorkOutcome::Success { data: None };
        }

        drop(state);
        // Wake: new entries in pending_entries unblock poll_work from Pending.
        self.inner.ctx.try_wake();
        WorkOutcome::Success { data: None }
    }

    async fn execute_join_children(&self, mut batch: ReapingBatch) -> WorkOutcome {
        let children = batch.take_children();
        for child in children {
            let key = child.key;
            // Snapshot metrics before `join()` consumes the handle.
            let metrics = child.handle.metrics();
            match child.handle.join().await {
                Ok(_output) => {
                    let mut state = self.inner.state.lock();
                    state.successful_downloads += 1;
                    drop(state);
                    // Aggregate the child's bytes into the parent's per-transfer
                    // MetricsState directly (not via record_io on the child's
                    // context, which already updated the client-level counters
                    // during the child transfer) so DownloadObjectsOutput.metrics
                    // reflects the whole directory download without double-counting.
                    self.inner.ctx.metrics.record_io(&crate::metrics::IoSample {
                        network_tx: metrics.network_tx,
                        network_rx: metrics.network_rx,
                        disk_read: metrics.disk_read,
                        disk_write: metrics.disk_write,
                    });
                }
                Err(err) => {
                    let mut state = self.inner.state.lock();
                    let failed_input = DownloadInput::builder()
                        .bucket(&self.inner.bucket)
                        .key(&key)
                        .build()
                        .unwrap();
                    // A child's local failure, such as renaming its temporary
                    // file, names the path but not the key. Errors without a
                    // local path are left unchanged.
                    state.failed.push(FailedDownload {
                        input: failed_input,
                        error: err.with_local_key(key),
                    });
                    if self.inner.failure_policy == FailedTransferPolicy::Abort {
                        self.inner.ctx.set_failed(child_failed());
                        self.report_terminal(state.snapshot());
                        self.inner.ctx.signal_terminal();
                    }
                }
            }
        }

        let mut state = self.inner.state.lock();
        batch.consume(&mut state);
        emit_event(
            &self.inner.ctx,
            DownloadObjectsEvent::ReapCompleted,
            state.snapshot(),
        );
        // An execute callback that drains the last in-flight work owns the
        // terminal transition: check and signal here rather than deferring to
        // a subsequent poll_work.
        if self.check_terminal(&state).is_some() {
            drop(state);
            return WorkOutcome::Success { data: None };
        }
        drop(state);
        // Wake: freed pipeline capacity may unblock dispatch_walk or spawning.
        self.inner.ctx.try_wake();
        WorkOutcome::Success { data: None }
    }
}

impl Transfer for DownloadObjectsTransfer {
    fn ctx(&self) -> &TransferContext {
        &self.inner.ctx
    }

    fn poll_work(&self) -> PollWork {
        self.poll_work()
    }

    fn execute<'a>(
        &'a self,
        work: &'a mut IoRequest,
    ) -> Pin<Box<dyn std::future::Future<Output = WorkOutcome> + Send + 'a>> {
        Box::pin(self.execute(work))
    }

    fn on_terminal(&self) {
        // The summary reports the state the transfer ended in.
        let snapshot = self.inner.state.lock().snapshot();
        self.report_terminal(snapshot);
    }
}

/// The error a child failure under [`FailedTransferPolicy::Abort`] fails the
/// parent with.
///
/// It carries no detail of its own: the handle attaches the failed downloads,
/// the first of which is the triggering child, and the error's source and
/// message are taken from them.
fn child_failed() -> Error {
    Error::new(ErrorKind::ChildOperationFailed, "download_objects aborted")
}

fn replace_delim<'a>(key: &'a str, delimiter: Option<&str>, path_separator: &str) -> Cow<'a, str> {
    match delimiter {
        Some(delim) if delim != path_separator => {
            let replaced = key.replace(delim, path_separator);
            Cow::Owned(replaced)
        }
        _ => Cow::Borrowed(key),
    }
}

/// Derive the local filesystem path for a given S3 key.
///
/// The whole key is the path relative to `root_dir`, whatever prefix selected
/// it: the delimiter, when one is set, is replaced with the OS path separator,
/// the result is joined to `root_dir`, and the joined path is normalized.
/// Returns [`ErrorKind::InputInvalid`] unless the result lies beneath
/// `root_dir`: when it lies outside `root_dir` (path traversal guard), and
/// when it is `root_dir` itself, as for the keys `.`, `./`, and `a/..`. The
/// filesystem is not touched.
///
/// `root_dir` and the joined path are both cleaned lexically, with
/// `path_clean`, before they are compared, so a destination such as `./out`
/// or `out/../out` contains the same paths as `out`, and the returned path is
/// under the cleaned form. Cleaning does not consult the filesystem: `..`
/// removes the preceding component without resolving symbolic links.
// TODO(sep): an opt-in mode could map keys relative to the listing prefix by
// whole path components, as the AWS CLI does.
pub(crate) fn local_key_path(
    root_dir: &Path,
    key: &str,
    delimiter: Option<&str>,
) -> Result<PathBuf, Error> {
    let relative_path = replace_delim(key, delimiter, std::path::MAIN_SEPARATOR_STR);

    let root_dir = root_dir.clean();
    let local_path = root_dir.join(relative_path.as_ref()).clean();
    validate_path(&root_dir, &local_path, key)?;

    Ok(local_path)
}

/// Rejects `local_path` with [`ErrorKind::InputInvalid`] unless it lies
/// beneath `root_dir`. Both must be lexically clean.
///
/// A clean path holds `..` only as leading components, so `local_path` is
/// beneath `root_dir` when it starts with `root_dir` and the rest begins with
/// a normal component: not `..`, a root, or a drive prefix, which leave
/// `root_dir`, and not empty, which is `root_dir` itself. This also holds for
/// a root made of `..` components. A root of `.` is the one that does not
/// prefix the paths beneath it, and a clean path equal to it is `.`.
fn validate_path(root_dir: &Path, local_path: &Path, key: &str) -> Result<(), Error> {
    let rest = if root_dir == Path::new(".") {
        Some(local_path)
    } else {
        local_path.strip_prefix(root_dir).ok()
    };
    let reason = match rest.map(|rest| rest.components().next()) {
        Some(Some(Component::Normal(_))) => return Ok(()),
        Some(None | Some(Component::CurDir)) => "resolves to the destination directory itself",
        _ => "resolves outside the target destination directory",
    };
    Err(Error::new(
        ErrorKind::InputInvalid,
        format!("Unable to download key: '{key}', its relative path {reason}"),
    ))
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::io::walk::{S3WalkContext, S3Walker};
    use crate::operation::download::test_util::fixtures::head_rule;
    use crate::transfer::TransferContext;
    use aws_sdk_s3::operation::get_object::GetObjectOutput;
    use aws_sdk_s3::operation::list_objects_v2::ListObjectsV2Output;
    use aws_sdk_s3::types::Object;
    use aws_smithy_mocks::{mock, mock_client, RuleMode};
    use aws_smithy_types::error::display::DisplayErrorContext;
    use std::time::Duration;
    use tempfile::tempdir;
    use tokio::time::timeout;

    fn mock_s3_success() -> aws_sdk_s3::Client {
        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(|| {
            ListObjectsV2Output::builder()
                .set_contents(Some(vec![
                    Object::builder().key("a.txt").size(5).build(),
                    Object::builder().key("b.txt").size(5).build(),
                    Object::builder().key("c.txt").size(5).build(),
                ]))
                .build()
        });
        let get = mock!(aws_sdk_s3::Client::get_object).then_output(|| {
            GetObjectOutput::builder()
                .content_length(5)
                .content_range("bytes 0-4/5")
                .e_tag("test-etag")
                .body(aws_sdk_s3::primitives::ByteStream::from_static(b"hello"))
                .build()
        });
        mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[head_rule(5), list, get])
    }

    fn mock_s3_get_failure() -> aws_sdk_s3::Client {
        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(|| {
            ListObjectsV2Output::builder()
                .set_contents(Some(vec![Object::builder().key("a.txt").size(5).build()]))
                .build()
        });
        let get = mock!(aws_sdk_s3::Client::get_object).then_error(|| {
            aws_sdk_s3::operation::get_object::GetObjectError::generic(
                aws_sdk_s3::error::ErrorMetadata::builder()
                    .code("NoSuchKey")
                    .message("not found")
                    .build(),
            )
        });
        mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[head_rule(5), list, get])
    }

    fn mock_s3_list_failure() -> aws_sdk_s3::Client {
        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_error(|| {
            aws_sdk_s3::operation::list_objects_v2::ListObjectsV2Error::generic(
                aws_sdk_s3::error::ErrorMetadata::builder()
                    .code("AccessDenied")
                    .message("access denied")
                    .build(),
            )
        });
        mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[list])
    }

    fn setup(
        dest: &Path,
        policy: FailedTransferPolicy,
        s3_client: aws_sdk_s3::Client,
    ) -> (
        DownloadObjectsTransfer,
        crate::transfer::StateMachineTerminalReceiver,
    ) {
        setup_with_walker(dest, policy, s3_client, S3Walker::builder().build())
    }

    /// [`setup`], listing with `walker` instead of the default walker.
    fn setup_with_walker(
        dest: &Path,
        policy: FailedTransferPolicy,
        s3_client: aws_sdk_s3::Client,
        walker: S3Walker,
    ) -> (
        DownloadObjectsTransfer,
        crate::transfer::StateMachineTerminalReceiver,
    ) {
        let config = crate::Config::builder()
            .sdk_client(s3_client.clone())
            .build()
            .expect("valid config");
        let handle = crate::client::Handle::test_handle_tokio(config);

        let walk = walker.walk(
            S3WalkContext::builder()
                .sdk_client(s3_client)
                .bucket("test-bucket")
                .build(),
        );

        let (ctx, completion_rx) = TransferContext::new(handle);
        ctx.handle
            .scheduler
            .register_empty_group_for_test(ctx.id.id);

        let input = crate::operation::download_objects::DownloadObjectsInput::builder()
            .bucket("test-bucket")
            .destination(dest)
            .failure_policy(policy)
            .build()
            .unwrap();
        let transfer = DownloadObjectsTransfer::new(ctx, &input, walk, 1000);
        (transfer, completion_rx)
    }

    async fn drive_transfer(transfer: &DownloadObjectsTransfer) {
        loop {
            match transfer.poll_work() {
                PollWork::Ready { io: mut work, .. } => {
                    transfer.execute(&mut work).await;
                }
                PollWork::Spawned => {}
                PollWork::Pending => {
                    tokio::time::sleep(Duration::from_millis(10)).await;
                }
                PollWork::Done => break,
            }
        }
    }

    /// Setup that enqueues the transfer into the scheduler (for managed-runtime tests).
    /// Returns the transfer and completion receiver. The scheduler drives execution.
    fn setup_enqueued(
        dest: &Path,
        policy: FailedTransferPolicy,
        s3_client: aws_sdk_s3::Client,
        handle: std::sync::Arc<crate::client::Handle>,
    ) -> (
        DownloadObjectsTransfer,
        crate::transfer::StateMachineTerminalReceiver,
    ) {
        let walk = S3Walker::builder().build().walk(
            S3WalkContext::builder()
                .sdk_client(s3_client)
                .bucket("test-bucket")
                .build(),
        );

        let (ctx, completion_rx) = TransferContext::new(handle.clone());

        let input = crate::operation::download_objects::DownloadObjectsInput::builder()
            .bucket("test-bucket")
            .destination(dest)
            .failure_policy(policy)
            .build()
            .unwrap();
        let transfer = DownloadObjectsTransfer::new(ctx, &input, walk, 1000);

        handle
            .scheduler
            .enqueue_transfer(Box::new(transfer.clone()));

        (transfer, completion_rx)
    }

    // --- Key derivation tests (ported from worker.rs) ---

    /// A delimiter of any length, `/` or not, is replaced with the path
    /// separator, and every segment of the key is kept.
    #[test]
    fn test_local_key_path_delimiters() {
        for delim in ["/", "//", "\\", "|", "delim"] {
            let key = format!("notes{delim}2021{delim}1.txt");
            let actual = local_key_path(Path::new("test"), &key, Some(delim))
                .unwrap_or_else(|e| panic!("delim={delim:?}: {e}"));
            assert_eq!(
                Path::new("test/notes/2021/1.txt"),
                actual,
                "delim={delim:?}"
            );
        }
    }

    #[cfg(target_family = "unix")]
    #[test]
    fn test_local_key_path_comprehensive() {
        // (key, delimiter, expected) for local_key_path.
        type Case = (
            &'static str,
            Option<&'static str>,
            Result<&'static str, &'static str>,
        );
        let root = Path::new("test");
        let cases: &[Case] = &[
            ("2023/Jan/1.png", None, Ok("test/2023/Jan/1.png")),
            ("2023-Jan-1.png", Some("-"), Ok("test/2023/Jan/1.png")),
            ("2023-Jan-.png", Some("-"), Ok("test/2023/Jan/.png")),
            (
                "many////delims-in-a-row",
                Some("/"),
                Ok("test/many/delims-in-a-row"),
            ),
            ("safe/../escape.bin", None, Ok("test/escape.bin")),
            ("../2023/Jan/1.png", None, Err("outside the target")),
            ("/2023/Jan/1.png", None, Err("outside the target")),
            (
                "foo/../2023/../../Jan/1.png",
                None,
                Err("outside the target"),
            ),
            ("../test-2/object.dat", None, Err("outside the target")),
        ];
        for (key, delim, expected) in cases {
            let actual = local_key_path(root, key, *delim);
            match expected {
                Ok(path) => {
                    let actual =
                        actual.unwrap_or_else(|e| panic!("expected Ok for key={key:?}: {e}"));
                    assert_eq!(Path::new(path), actual, "key={key:?}");
                }
                Err(_) => {
                    let err = actual.expect_err(&format!("expected Err for key={key:?}"));
                    assert_eq!(err.kind(), &ErrorKind::InputInvalid, "key={key:?}");
                }
            }
        }
    }

    /// Asserts, for each `(root, cleaned)` destination and for the default and
    /// `/` delimiters, that a key inside the destination maps under `cleaned`,
    /// that each key in `escapes` is rejected as resolving outside it, and
    /// that a key that maps to the destination itself is rejected.
    ///
    /// The keys that map to the destination are `.`, `a/..`, and `./`.
    /// Accepting one would place its temporary file in the destination's
    /// parent. A key ending in the delimiter, such as `photos/`, is not one of
    /// them: it maps to the path its last segment names.
    #[cfg(any(unix, windows))]
    fn assert_dot_component_destinations(cases: &[(&str, &str)], escapes: &[&str]) {
        for &(root, cleaned) in cases {
            let root = Path::new(root);
            for delim in [None, Some("/")] {
                let actual = local_key_path(root, "a/b.txt", delim)
                    .unwrap_or_else(|e| panic!("root={root:?} delim={delim:?}: {e}"));
                assert_eq!(
                    Path::new(cleaned).join("a/b.txt").clean(),
                    actual,
                    "root={root:?} delim={delim:?}"
                );

                for escape in escapes {
                    let err = local_key_path(root, escape, delim).expect_err(&format!(
                        "root={root:?} delim={delim:?} key={escape:?} must be rejected"
                    ));
                    assert_eq!(err.kind(), &ErrorKind::InputInvalid, "key={escape:?}");
                    let message = DisplayErrorContext(&err).to_string();
                    assert!(
                        message.contains("outside the target destination"),
                        "root={root:?} key={escape:?}: {message}"
                    );
                }

                for key in [".", "a/..", "./"] {
                    let err = local_key_path(root, key, delim).expect_err(&format!(
                        "root={root:?} delim={delim:?} key={key:?} maps to the destination"
                    ));
                    assert_eq!(err.kind(), &ErrorKind::InputInvalid, "key={key:?}");
                    let message = DisplayErrorContext(&err).to_string();
                    assert!(
                        message.contains("destination directory itself"),
                        "root={root:?} key={key:?}: {message}"
                    );
                }

                let actual = local_key_path(root, "photos/", delim)
                    .unwrap_or_else(|e| panic!("root={root:?} delim={delim:?}: {e}"));
                assert_eq!(
                    Path::new(cleaned).join("photos").clean(),
                    actual,
                    "root={root:?} delim={delim:?}"
                );
            }
        }
    }

    // ref: RUST-1202
    /// A destination with `.` or `..` components contains the same paths as
    /// its lexically cleaned form: keys inside it are accepted and placed
    /// under the cleaned form, and keys that escape it or map to it are
    /// rejected.
    #[cfg(target_family = "unix")]
    #[test]
    fn test_local_key_path_destination_with_dot_components() {
        assert_dot_component_destinations(
            &[
                ("./out", "out"),
                ("out/../out", "out"),
                ("out/./", "out"),
                ("/data/./out", "/data/out"),
                ("/data/x/../out", "/data/out"),
                (".", "."),
                ("./", "."),
                ("..", ".."),
                ("../out", "../out"),
            ],
            &["../b.txt", "a/../../b.txt", "/etc/passwd"],
        );
    }

    // ref: RUST-1202
    /// The Windows form of `test_local_key_path_destination_with_dot_components`,
    /// with drive-qualified destinations. A key that introduces a root (`\x`),
    /// a drive (`C:x`, `C:/x`), or a UNC prefix replaces the destination when
    /// joined to it, and is rejected.
    #[cfg(windows)]
    #[test]
    fn test_local_key_path_destination_with_dot_components_windows() {
        assert_dot_component_destinations(
            &[
                (r".\out", "out"),
                (r"out\..\out", "out"),
                (r"out\.\", "out"),
                (r"C:\data\.\out", r"C:\data\out"),
                (r"C:\data\x\..\out", r"C:\data\out"),
                (".", "."),
                (r"..\out", r"..\out"),
            ],
            &[
                "../b.txt",
                "a/../../b.txt",
                "/etc/passwd",
                r"\x",
                "C:x",
                "C:/x",
                r"\\server\share\x",
            ],
        );
    }

    // --- State machine tests ---

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_happy_path_3_files() {
        let dir = tempdir().unwrap();

        let (transfer, completion_rx) = setup(
            dir.path(),
            FailedTransferPolicy::Continue,
            mock_s3_success(),
        );

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert_eq!(transfer.successful_downloads(), 3);
        assert!(transfer.take_failed().is_none());

        // Verify files exist
        assert!(dir.path().join("a.txt").exists());
        assert!(dir.path().join("b.txt").exists());
        assert!(dir.path().join("c.txt").exists());
    }

    // ref: RUST-1202
    /// A destination that does not exist is created, with every missing
    /// parent, before any object is written into it.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_missing_destination_is_created() {
        let dir = tempdir().unwrap();
        let dest = dir.path().join("a/b/out");

        let (transfer, completion_rx) =
            setup(&dest, FailedTransferPolicy::Abort, mock_s3_success());

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert!(
            transfer.ctx().take_error().is_none(),
            "a missing destination must not fail the transfer"
        );
        assert_eq!(transfer.successful_downloads(), 3);
        for name in ["a.txt", "b.txt", "c.txt"] {
            assert_eq!(std::fs::read(dest.join(name)).unwrap(), b"hello");
        }
    }

    // ref: RUST-1202
    /// A destination that exists but is not a directory fails the transfer
    /// with `InputInvalid` before any object is requested, and is left as it
    /// was.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_destination_that_is_a_file_is_input_invalid() {
        let dir = tempdir().unwrap();
        let dest = dir.path().join("out");
        std::fs::write(&dest, b"not a directory").unwrap();

        // Any GetObject would be answered by this rule; none must be sent.
        let get = mock!(aws_sdk_s3::Client::get_object).then_output(|| {
            GetObjectOutput::builder()
                .content_length(5)
                .content_range("bytes 0-4/5")
                .e_tag("test-etag")
                .body(aws_sdk_s3::primitives::ByteStream::from_static(b"hello"))
                .build()
        });
        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(|| {
            ListObjectsV2Output::builder()
                .set_contents(Some(vec![Object::builder().key("a.txt").size(5).build()]))
                .build()
        });
        let client = mock_client!(
            aws_sdk_s3,
            RuleMode::MatchAny,
            &[head_rule(5), list, get.clone()]
        );
        let (transfer, completion_rx) = setup(&dest, FailedTransferPolicy::Continue, client);

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        let err = transfer
            .ctx()
            .take_error()
            .expect("a file destination must fail the transfer");
        assert_eq!(err.kind(), &ErrorKind::InputInvalid);
        assert_eq!(get.num_calls(), 0);
        assert_eq!(std::fs::read(&dest).unwrap(), b"not a directory");
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_child_failure_continue() {
        let dir = tempdir().unwrap();

        let (transfer, completion_rx) = setup(
            dir.path(),
            FailedTransferPolicy::Continue,
            mock_s3_get_failure(),
        );

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert_eq!(transfer.successful_downloads(), 0);
        let failed = transfer.take_failed().unwrap();
        assert_eq!(failed.len(), 1);
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_child_failure_abort() {
        let dir = tempdir().unwrap();

        let (transfer, completion_rx) = setup(
            dir.path(),
            FailedTransferPolicy::Abort,
            mock_s3_get_failure(),
        );

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert_eq!(transfer.successful_downloads(), 0);
        assert!(transfer.inner.ctx.is_failed());
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_fatal_walker_error() {
        let dir = tempdir().unwrap();

        let (transfer, completion_rx) = setup(
            dir.path(),
            FailedTransferPolicy::Continue,
            mock_s3_list_failure(),
        );

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert!(transfer.inner.ctx.is_failed());
        assert_eq!(transfer.successful_downloads(), 0);
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_empty_listing() {
        let dir = tempdir().unwrap();

        let list = mock!(aws_sdk_s3::Client::list_objects_v2)
            .then_output(|| ListObjectsV2Output::builder().build());
        let client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[list]);

        let (transfer, completion_rx) = setup(dir.path(), FailedTransferPolicy::Continue, client);

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert_eq!(transfer.successful_downloads(), 0);
        assert!(transfer.take_failed().is_none());
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn pending_reason_follows_parent_state() {
        let dir = tempdir().unwrap();
        let (transfer, _completion_rx) = setup(
            dir.path(),
            FailedTransferPolicy::Continue,
            mock_s3_success(),
        );
        let mut state = transfer.inner.state.lock();

        assert_eq!(
            state.pending_reason(false),
            CompositePendingReason::Draining
        );
        assert_eq!(state.pending_reason(true), CompositePendingReason::Listing);

        state.reaping_in_flight = 1;
        assert_eq!(state.pending_reason(true), CompositePendingReason::Reaping);
        state.reaping_in_flight = 0;

        state
            .pending_entries
            .push_back(crate::model::Object::builder().key("a.txt").build());
        assert_eq!(
            state.pending_reason(true),
            CompositePendingReason::ChildCapacity
        );
        state.pending_entries.clear();

        let walk = state.walk.take();
        state.walk_in_flight = true;
        assert_eq!(state.pending_reason(true), CompositePendingReason::Listing);
        state.walk_in_flight = false;
        assert_eq!(
            state.pending_reason(true),
            CompositePendingReason::ChildCompletion
        );
        state.walk = walk;
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn terminal_summary_reported_once_on_success() {
        let dir = tempdir().unwrap();
        let (transfer, completion_rx) = setup(
            dir.path(),
            FailedTransferPolicy::Continue,
            mock_s3_success(),
        );

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        let summary = transfer
            .test_terminal_summary()
            .expect("terminal summary reported");
        assert_eq!(summary.outcome, crate::transfer::TerminalOutcome::Completed);
        assert_eq!(summary.successful, 3);
        assert_eq!(summary.failed, 0);

        // A later external termination must not report again.
        transfer.inner.state.lock().successful_downloads = 99;
        Transfer::on_terminal(&transfer);
        let summary = transfer
            .test_terminal_summary()
            .expect("terminal summary reported");
        assert_eq!(summary.successful, 3, "summary reported more than once");
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn terminal_summary_reported_once_on_failure() {
        let dir = tempdir().unwrap();
        let (transfer, completion_rx) = setup(
            dir.path(),
            FailedTransferPolicy::Continue,
            mock_s3_list_failure(),
        );

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        let summary = transfer
            .test_terminal_summary()
            .expect("terminal summary reported");
        assert_eq!(summary.outcome, crate::transfer::TerminalOutcome::Failed);
        assert!(summary.error_kind.is_some());
        assert_eq!(summary.successful, 0);

        // A later external termination must not report again.
        transfer.inner.state.lock().successful_downloads = 99;
        Transfer::on_terminal(&transfer);
        let summary = transfer
            .test_terminal_summary()
            .expect("terminal summary reported");
        assert_eq!(summary.successful, 0, "summary reported more than once");
    }

    /// A response body that never yields a frame, so a child download that
    /// reads it stays in flight until it is cancelled.
    struct NeverBody;

    impl http_body_1x::Body for NeverBody {
        type Data = bytes::Bytes;
        type Error = std::convert::Infallible;

        fn poll_frame(
            self: std::pin::Pin<&mut Self>,
            _cx: &mut std::task::Context<'_>,
        ) -> std::task::Poll<Option<Result<http_body_1x::Frame<Self::Data>, Self::Error>>> {
            std::task::Poll::Pending
        }
    }

    /// A fatal listing error on a later page, while children spawned from an
    /// earlier page are still downloading, ends the parent `Failed` and still
    /// reports its terminal summary.
    ///
    /// Page 1 fills one walk advance (`MAX_ENTRIES_PER_WALK` keys), so page 2 is
    /// fetched by a second advance. No GET body ever completes, so the children
    /// spawned from page 1 are in flight when page 2 fails. The page-2 mock
    /// sleeps so the parent spawns children before the failure lands. The
    /// scheduler, not `check_terminal`, then ends the parent: its last
    /// dispatched work has retired, so it is terminal and idle with children
    /// still alive, and it is not polled again.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn terminal_summary_reported_on_listing_failure_with_children_in_flight() {
        let dir = tempdir().unwrap();
        let page1 = mock!(aws_sdk_s3::Client::list_objects_v2)
            .match_requests(|r| r.continuation_token().is_none())
            .then_output(|| {
                ListObjectsV2Output::builder()
                    .set_contents(Some(
                        (0..MAX_ENTRIES_PER_WALK)
                            .map(|i| Object::builder().key(format!("k{i:04}")).size(5).build())
                            .collect(),
                    ))
                    .is_truncated(true)
                    .next_continuation_token("page-2")
                    .build()
            });
        let page2 = mock!(aws_sdk_s3::Client::list_objects_v2)
            .match_requests(|r| r.continuation_token() == Some("page-2"))
            .then_error(|| {
                std::thread::sleep(Duration::from_millis(300));
                aws_sdk_s3::operation::list_objects_v2::ListObjectsV2Error::generic(
                    aws_sdk_s3::error::ErrorMetadata::builder()
                        .code("AccessDenied")
                        .message("access denied")
                        .build(),
                )
            });
        let get = mock!(aws_sdk_s3::Client::get_object).then_output(|| {
            GetObjectOutput::builder()
                .content_length(5)
                .content_range("bytes 0-4/5")
                .e_tag("test-etag")
                .body(aws_sdk_s3::primitives::ByteStream::new(
                    aws_smithy_types::body::SdkBody::from_body_1_x(NeverBody),
                ))
                .build()
        });
        let s3_client = mock_client!(
            aws_sdk_s3,
            RuleMode::MatchAny,
            &[head_rule(5), page1, page2, get]
        );
        let config = crate::Config::builder()
            .sdk_client(s3_client.clone())
            .build()
            .expect("valid config");
        let handle = crate::client::Handle::test_handle_managed(config);
        let (transfer, completion_rx) = setup_enqueued(
            dir.path(),
            FailedTransferPolicy::Continue,
            s3_client,
            handle,
        );

        timeout(Duration::from_secs(10), async {
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should end within timeout");

        assert!(transfer.ctx().is_failed());
        assert!(
            !transfer.inner.state.lock().children.is_empty(),
            "the parent ended with children still recorded as running"
        );
        let summary = transfer
            .test_terminal_summary()
            .expect("terminal summary reported");
        assert_eq!(summary.outcome, crate::transfer::TerminalOutcome::Failed);

        crate::runtime::test_util::shutdown(&transfer.ctx().handle).await;
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn terminal_summary_reported_once_on_cancel() {
        let dir = tempdir().unwrap();
        let (transfer, _completion_rx) = setup(
            dir.path(),
            FailedTransferPolicy::Continue,
            mock_s3_success(),
        );

        transfer.inner.ctx.set_cancelled();
        Transfer::on_terminal(&transfer);

        let summary = transfer
            .test_terminal_summary()
            .expect("terminal summary reported");
        assert_eq!(summary.outcome, crate::transfer::TerminalOutcome::Cancelled);
        assert_eq!(summary.successful, 0);

        // A second external termination must not report again.
        transfer.inner.state.lock().successful_downloads = 99;
        Transfer::on_terminal(&transfer);
        let summary = transfer
            .test_terminal_summary()
            .expect("terminal summary reported");
        assert_eq!(summary.successful, 0, "summary reported more than once");
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_path_traversal_rejected_as_failure() {
        let dir = tempdir().unwrap();

        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(|| {
            ListObjectsV2Output::builder()
                .set_contents(Some(vec![Object::builder()
                    .key("../../../etc/passwd")
                    .size(100)
                    .build()]))
                .build()
        });
        let client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[list]);

        let (transfer, completion_rx) = setup(dir.path(), FailedTransferPolicy::Continue, client);

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert_eq!(transfer.successful_downloads(), 0);
        let failed = transfer.take_failed().unwrap();
        assert_eq!(failed.len(), 1);
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_nested_keys_create_directories() {
        let dir = tempdir().unwrap();

        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(|| {
            ListObjectsV2Output::builder()
                .set_contents(Some(vec![
                    Object::builder()
                        .key("photos/2023/jan/img.jpg")
                        .size(5)
                        .build(),
                    Object::builder().key("docs/readme.txt").size(5).build(),
                ]))
                .build()
        });
        let get = mock!(aws_sdk_s3::Client::get_object).then_output(|| {
            GetObjectOutput::builder()
                .content_length(5)
                .content_range("bytes 0-4/5")
                .e_tag("test-etag")
                .body(aws_sdk_s3::primitives::ByteStream::from_static(b"hello"))
                .build()
        });
        let client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[head_rule(5), list, get]);

        let (transfer, completion_rx) = setup(dir.path(), FailedTransferPolicy::Continue, client);

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert_eq!(transfer.successful_downloads(), 2);
        assert!(dir.path().join("photos/2023/jan/img.jpg").exists());
        assert!(dir.path().join("docs/readme.txt").exists());
    }

    // ref: RUST-1202
    /// A key whose parent directory cannot be created, because a file holds
    /// its name, fails with an `IOError` that names the directory, with the
    /// operating system's error as its source.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_parent_directory_failure_names_directory_and_key() {
        let dir = tempdir().unwrap();
        std::fs::write(dir.path().join("docs"), b"a file, not a directory").unwrap();

        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(|| {
            ListObjectsV2Output::builder()
                .set_contents(Some(vec![Object::builder()
                    .key("docs/readme.txt")
                    .size(5)
                    .build()]))
                .build()
        });
        let client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[list]);
        let (transfer, completion_rx) = setup(dir.path(), FailedTransferPolicy::Continue, client);

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        let failed = transfer.take_failed().expect("the key must fail");
        let err = failed[0].error();
        assert_eq!(err.kind(), &ErrorKind::IOError);
        assert_eq!(
            err.to_string(),
            format!(
                "I/O error creating directory \"{}\" (key \"docs/readme.txt\")",
                dir.path().join("docs").display()
            )
        );
        assert!(std::error::Error::source(err)
            .and_then(|s| s.downcast_ref::<std::io::Error>())
            .is_some());
    }

    /// Objects whose keys map to the destination directory itself (`.`, `./`,
    /// and `a/..`) fail with `InputInvalid` under `Continue` before any file
    /// is created or any GetObject is sent. Accepted, each one's temporary
    /// file would be created in the destination's parent, the object
    /// downloaded, and only the rename onto the destination would fail.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_key_mapping_to_destination_fails_before_any_file_is_created() {
        let dir = tempdir().unwrap();
        let dest = dir.path().join("out");
        let keys = [".", "./", "a/.."];

        // Any GetObject would be answered by this rule; none must be sent.
        let get = mock!(aws_sdk_s3::Client::get_object).then_output(|| {
            GetObjectOutput::builder()
                .content_length(5)
                .content_range("bytes 0-4/5")
                .e_tag("test-etag")
                .body(aws_sdk_s3::primitives::ByteStream::from_static(b"hello"))
                .build()
        });
        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(move || {
            ListObjectsV2Output::builder()
                .set_contents(Some(
                    keys.iter()
                        .map(|key| Object::builder().key(*key).size(5).build())
                        .collect(),
                ))
                .build()
        });
        let client = mock_client!(
            aws_sdk_s3,
            RuleMode::MatchAny,
            &[head_rule(5), list, get.clone()]
        );
        let (transfer, completion_rx) = setup(&dest, FailedTransferPolicy::Continue, client);

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert!(transfer.ctx().take_error().is_none());
        let failed = transfer.take_failed().expect("the keys must fail");
        let mut failed_keys: Vec<_> = failed.iter().map(|f| f.input().key()).collect();
        failed_keys.sort();
        assert_eq!(failed_keys, [Some("."), Some("./"), Some("a/..")]);
        for f in &failed {
            assert_eq!(f.error().kind(), &ErrorKind::InputInvalid);
        }
        assert_eq!(get.num_calls(), 0);
        let entries: Vec<_> = std::fs::read_dir(dir.path())
            .unwrap()
            .map(|e| e.unwrap().file_name())
            .collect();
        assert_eq!(
            entries,
            ["out"],
            "nothing may be created beside the destination"
        );
        assert_eq!(std::fs::read_dir(&dest).unwrap().count(), 0);
    }

    // ref: RUST-1426
    /// A non-empty object whose key is the listing prefix `photos/` is
    /// downloaded to the file `photos` beneath the destination: the prefix
    /// does not shorten the key, so the key does not map to the destination
    /// itself.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_key_equal_to_prefix_downloads_to_its_last_segment() {
        let dir = tempdir().unwrap();

        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(|| {
            ListObjectsV2Output::builder()
                .set_contents(Some(vec![Object::builder().key("photos/").size(5).build()]))
                .build()
        });
        let get = mock!(aws_sdk_s3::Client::get_object).then_output(|| {
            GetObjectOutput::builder()
                .content_length(5)
                .content_range("bytes 0-4/5")
                .e_tag("test-etag")
                .body(aws_sdk_s3::primitives::ByteStream::from_static(b"hello"))
                .build()
        });
        let client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[head_rule(5), list, get]);
        let (transfer, completion_rx) = setup_with_walker(
            dir.path(),
            FailedTransferPolicy::Abort,
            client,
            S3Walker::builder().prefix("photos/").build(),
        );

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert!(transfer.ctx().take_error().is_none());
        assert!(transfer.take_failed().is_none());
        assert_eq!(transfer.successful_downloads(), 1);
        assert_eq!(std::fs::read(dir.path().join("photos")).unwrap(), b"hello");
    }

    // ref: RUST-1202
    /// A child whose temporary file cannot be renamed onto the object's path,
    /// because a non-empty directory holds it, fails with an `IOError` that
    /// names the path and the object's key, with the operating system's error
    /// as its source. The rename error is built where only the path is known;
    /// the key is attached when the child's result is recorded.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_rename_failure_names_path_and_key() {
        let dir = tempdir().unwrap();
        let blocked = dir.path().join("a.txt");
        std::fs::create_dir(&blocked).unwrap();
        std::fs::write(blocked.join("inner"), b"kept").unwrap();

        let (transfer, completion_rx) = setup(
            dir.path(),
            FailedTransferPolicy::Continue,
            mock_s3_success(),
        );

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert_eq!(transfer.successful_downloads(), 2);
        let failed = transfer.take_failed().expect("a.txt must fail");
        assert_eq!(failed.len(), 1);
        assert_eq!(failed[0].input().key(), Some("a.txt"));
        let err = failed[0].error();
        assert_eq!(err.kind(), &ErrorKind::IOError);
        assert_eq!(
            err.to_string(),
            format!(
                "I/O error renaming temporary file to \"{}\" (key \"a.txt\")",
                blocked.display()
            )
        );
        assert!(std::error::Error::source(err)
            .and_then(|s| s.downcast_ref::<std::io::Error>())
            .is_some());
        assert_eq!(std::fs::read(blocked.join("inner")).unwrap(), b"kept");
        let mut entries: Vec<_> = std::fs::read_dir(dir.path())
            .unwrap()
            .map(|e| e.unwrap().file_name())
            .collect();
        entries.sort();
        assert_eq!(
            entries,
            ["a.txt", "b.txt", "c.txt"],
            "the failed child's temporary file must be removed"
        );
    }

    // ref: RUST-1426
    /// The walker's prefix selects objects and does not shorten their keys:
    /// `backup/2023/a.txt` listed under `backup/2023/` is written to
    /// `<dest>/backup/2023/a.txt`, not `<dest>/a.txt`.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_walker_prefix_keeps_full_key_in_local_path() {
        let dir = tempdir().unwrap();

        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(|| {
            ListObjectsV2Output::builder()
                .set_contents(Some(vec![
                    Object::builder().key("backup/2023/a.txt").size(3).build(),
                    Object::builder().key("backup/2023/b.txt").size(3).build(),
                ]))
                .build()
        });
        let get = mock!(aws_sdk_s3::Client::get_object).then_output(|| {
            GetObjectOutput::builder()
                .content_length(3)
                .content_range("bytes 0-2/3")
                .e_tag("test-etag")
                .body(aws_sdk_s3::primitives::ByteStream::from_static(b"abc"))
                .build()
        });
        let client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[head_rule(3), list, get]);
        let (transfer, completion_rx) = setup_with_walker(
            dir.path(),
            FailedTransferPolicy::Continue,
            client,
            S3Walker::builder().prefix("backup/2023/").build(),
        );

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert_eq!(transfer.successful_downloads(), 2);
        assert!(transfer.take_failed().is_none());
        assert_eq!(
            std::fs::read(dir.path().join("backup/2023/a.txt")).unwrap(),
            b"abc"
        );
        assert_eq!(
            std::fs::read(dir.path().join("backup/2023/b.txt")).unwrap(),
            b"abc"
        );
        let entries: Vec<_> = std::fs::read_dir(dir.path())
            .unwrap()
            .map(|e| e.unwrap().file_name())
            .collect();
        assert_eq!(entries, ["backup"]);
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_many_objects() {
        let dir = tempdir().unwrap();

        let objects: Vec<Object> = (0..20)
            .map(|i| {
                Object::builder()
                    .key(format!("file{i:02}.txt"))
                    .size(2)
                    .build()
            })
            .collect();
        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(move || {
            ListObjectsV2Output::builder()
                .set_contents(Some(objects.clone()))
                .build()
        });
        let get = mock!(aws_sdk_s3::Client::get_object).then_output(|| {
            GetObjectOutput::builder()
                .content_length(2)
                .content_range("bytes 0-1/2")
                .e_tag("test-etag")
                .body(aws_sdk_s3::primitives::ByteStream::from_static(b"ok"))
                .build()
        });
        let client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[head_rule(2), list, get]);

        let (transfer, completion_rx) = setup(dir.path(), FailedTransferPolicy::Continue, client);

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert_eq!(transfer.successful_downloads(), 20);
        for i in 0..20 {
            assert!(dir.path().join(format!("file{i:02}.txt")).exists());
        }

        // The success transition runs set_completed(), so a successful transfer
        // ends Completed (not still Active). Without it, a dropped handle would
        // see is_active() and spuriously cancel an already-finished transfer.
        assert_eq!(
            transfer.ctx().transfer_status(),
            crate::types::TransferStatus::Completed
        );

        // Child metrics are rolled into the parent: 20 objects x 2 bytes each.
        let m = transfer.ctx().metrics();
        assert_eq!(
            m.network_rx, 40,
            "child network_rx must aggregate to parent"
        );
        assert_eq!(
            m.disk_write, 40,
            "child disk_write must aggregate to parent"
        );
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_cancellation() {
        let dir = tempdir().unwrap();

        // Large listing so the transfer doesn't complete before we cancel
        let objects: Vec<Object> = (0..100)
            .map(|i| {
                Object::builder()
                    .key(format!("file{i}.txt"))
                    .size(5)
                    .build()
            })
            .collect();
        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(move || {
            ListObjectsV2Output::builder()
                .set_contents(Some(objects.clone()))
                .build()
        });
        let get = mock!(aws_sdk_s3::Client::get_object).then_output(|| {
            GetObjectOutput::builder()
                .content_length(5)
                .content_range("bytes 0-4/5")
                .e_tag("test-etag")
                .body(aws_sdk_s3::primitives::ByteStream::from_static(b"hello"))
                .build()
        });
        let client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[head_rule(5), list, get]);

        let (transfer, _completion_rx) = setup(dir.path(), FailedTransferPolicy::Continue, client);

        // Cancel after first poll_work cycle
        transfer.inner.ctx.set_cancelled();

        timeout(Duration::from_secs(10), async {
            drive_transfer(&transfer).await;
        })
        .await
        .expect("cancelled transfer should terminate");

        assert!(!transfer.inner.ctx.is_active());
    }

    // --- Managed-runtime tests ---
    // These exercise the real scheduler dispatch path (managed threads drive
    // poll_work/execute). Catches bugs like missing set_pending that only
    // manifest when work crosses thread boundaries.

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_happy_path_managed_runtime() {
        let dir = tempdir().unwrap();
        let config = crate::Config::builder()
            .sdk_client(mock_s3_success())
            .build()
            .expect("valid config");
        let handle = crate::client::Handle::test_handle_managed(config);

        let (transfer, completion_rx) = setup_enqueued(
            dir.path(),
            FailedTransferPolicy::Continue,
            mock_s3_success(),
            handle,
        );

        timeout(Duration::from_secs(10), async {
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert_eq!(transfer.successful_downloads(), 3);
        assert!(dir.path().join("a.txt").exists());
        assert!(dir.path().join("b.txt").exists());
        assert!(dir.path().join("c.txt").exists());

        crate::runtime::test_util::shutdown(&transfer.ctx().handle).await;
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_many_objects_managed_runtime() {
        let dir = tempdir().unwrap();

        let objects: Vec<Object> = (0..20)
            .map(|i| {
                Object::builder()
                    .key(format!("file{i:02}.txt"))
                    .size(2)
                    .build()
            })
            .collect();
        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(move || {
            ListObjectsV2Output::builder()
                .set_contents(Some(objects.clone()))
                .build()
        });
        let get = mock!(aws_sdk_s3::Client::get_object).then_output(|| {
            GetObjectOutput::builder()
                .content_length(2)
                .content_range("bytes 0-1/2")
                .e_tag("test-etag")
                .body(aws_sdk_s3::primitives::ByteStream::from_static(b"ok"))
                .build()
        });
        let s3_client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[head_rule(2), list, get]);

        let config = crate::Config::builder()
            .sdk_client(s3_client.clone())
            .build()
            .expect("valid config");
        let handle = crate::client::Handle::test_handle_managed(config);

        let (transfer, completion_rx) = setup_enqueued(
            dir.path(),
            FailedTransferPolicy::Continue,
            s3_client,
            handle,
        );

        timeout(Duration::from_secs(10), async {
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert_eq!(transfer.successful_downloads(), 20);
        for i in 0..20 {
            assert!(dir.path().join(format!("file{i:02}.txt")).exists());
        }

        // The success transition runs set_completed(), so a successful transfer
        // ends Completed (not still Active). Without it, a dropped handle would
        // see is_active() and spuriously cancel an already-finished transfer.
        assert_eq!(
            transfer.ctx().transfer_status(),
            crate::types::TransferStatus::Completed
        );

        // Child metrics are rolled into the parent: 20 objects x 2 bytes each.
        let m = transfer.ctx().metrics();
        assert_eq!(
            m.network_rx, 40,
            "child network_rx must aggregate to parent"
        );
        assert_eq!(
            m.disk_write, 40,
            "child disk_write must aggregate to parent"
        );

        crate::runtime::test_util::shutdown(&transfer.ctx().handle).await;
    }

    /// Under `Abort`, a child failure sets the parent failed in an execute
    /// callback, which must signal terminal there. Many keys keep sibling
    /// children in flight when the failure fires, so the parent goes terminal
    /// while children remain; its `join()` (completion) must resolve rather
    /// than hang on the scheduler removing the terminal parent.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_child_failure_abort_managed_runtime_does_not_strand() {
        let dir = tempdir().unwrap();

        // Many keys so several children spawn; GetObject always errors so the
        // first reaped failure aborts while siblings are still registered.
        let objects: Vec<Object> = (0..20)
            .map(|i| {
                Object::builder()
                    .key(format!("file{i:02}.txt"))
                    .size(2)
                    .build()
            })
            .collect();
        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(move || {
            ListObjectsV2Output::builder()
                .set_contents(Some(objects.clone()))
                .build()
        });
        let get = mock!(aws_sdk_s3::Client::get_object).then_error(|| {
            aws_sdk_s3::operation::get_object::GetObjectError::generic(
                aws_sdk_s3::error::ErrorMetadata::builder()
                    .code("NoSuchKey")
                    .message("not found")
                    .build(),
            )
        });
        let s3_client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[head_rule(2), list, get]);

        let config = crate::Config::builder()
            .sdk_client(s3_client.clone())
            .build()
            .expect("valid config");
        let handle = crate::client::Handle::test_handle_managed(config);

        let (transfer, completion_rx) =
            setup_enqueued(dir.path(), FailedTransferPolicy::Abort, s3_client, handle);

        // The parent's completion must resolve promptly rather than stranding
        // when the scheduler removes the terminal parent.
        timeout(Duration::from_secs(10), async {
            let _ = completion_rx.await;
        })
        .await
        .expect("Abort failure must signal parent terminal, not strand join()");

        assert!(transfer.ctx().is_failed());

        let ctx = transfer.ctx();
        timeout(
            Duration::from_secs(10),
            ctx.handle.scheduler.cancel_transfer(ctx.id).wait_for_idle(),
        )
        .await
        .expect("failed parent and child work did not drain");

        crate::runtime::test_util::shutdown(&ctx.handle).await;
    }

    /// Regression (real scheduler): two claimed-but-failed spawns under Continue
    /// must not hang the parent. A path-traversal key is rejected by
    /// `local_key_path` DURING the spawn step (before any GET), so it is a
    /// claimed entry that produces no child. The parent must stay scheduled to
    /// drain the SECOND such entry; if `poll_work` gates re-poll on whether a
    /// child *materialized* rather than whether an entry was *claimed*, the
    /// second entry strands with no wake once the walker is exhausted and the
    /// parent's `completion_rx` never resolves. Must use the managed runtime:
    /// the immediate `drive_transfer` harness busy-re-polls on `Pending` and so
    /// cannot observe a lost wake.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_two_failed_spawns_continue_managed_runtime_does_not_hang() {
        let dir = tempdir().unwrap();

        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(|| {
            ListObjectsV2Output::builder()
                .set_contents(Some(vec![
                    Object::builder().key("../../../etc/passwd").size(2).build(),
                    Object::builder().key("../../../etc/shadow").size(2).build(),
                ]))
                .build()
        });
        // No GET should ever fire (both fail at key-path derivation), but wire a
        // benign one so the mock client is well-formed.
        let get = mock!(aws_sdk_s3::Client::get_object).then_output(|| {
            GetObjectOutput::builder()
                .content_length(2)
                .content_range("bytes 0-1/2")
                .e_tag("test-etag")
                .body(aws_sdk_s3::primitives::ByteStream::from_static(b"ok"))
                .build()
        });
        let s3_client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[head_rule(2), list, get]);

        let config = crate::Config::builder()
            .sdk_client(s3_client.clone())
            .build()
            .expect("valid config");
        let handle = crate::client::Handle::test_handle_managed(config);

        let (transfer, completion_rx) = setup_enqueued(
            dir.path(),
            FailedTransferPolicy::Continue,
            s3_client,
            handle,
        );

        timeout(Duration::from_secs(10), async {
            let _ = completion_rx.await;
        })
        .await
        .expect("two claimed-but-failed spawns must not strand the parent's completion");

        assert_eq!(transfer.successful_downloads(), 0);
        assert_eq!(
            transfer.take_failed().map(|f| f.len()).unwrap_or(0),
            2,
            "both failed entries must be recorded"
        );

        crate::runtime::test_util::shutdown(&transfer.ctx().handle).await;
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_fatal_walker_error_managed_runtime() {
        let dir = tempdir().unwrap();
        let s3_client = mock_s3_list_failure();

        let config = crate::Config::builder()
            .sdk_client(s3_client.clone())
            .build()
            .expect("valid config");
        let handle = crate::client::Handle::test_handle_managed(config);

        let (transfer, completion_rx) = setup_enqueued(
            dir.path(),
            FailedTransferPolicy::Continue,
            s3_client,
            handle,
        );

        timeout(Duration::from_secs(10), async {
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert_eq!(transfer.successful_downloads(), 0);
        assert!(transfer.ctx().is_failed());

        crate::runtime::test_util::shutdown(&transfer.ctx().handle).await;
    }

    // --- RAII counter guard tests (mirror upload_objects) ---

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_reservation_consume_decrements_counter() {
        let dir = tempdir().unwrap();
        let (transfer, _rx) = setup(
            dir.path(),
            FailedTransferPolicy::Continue,
            mock_s3_success(),
        );

        // Pre-set the counter as if claim_one had reserved 5 slots.
        {
            let mut state = transfer.inner.state.lock();
            state.children_reserved = 5;
        }

        let reservation = Reservation {
            transfer: transfer.inner.clone(),
            count: 5,
            consumed: false,
        };

        {
            let mut state = transfer.inner.state.lock();
            reservation.consume(&mut state);
            assert_eq!(0, state.children_reserved, "consume must decrement");
        }

        // Drop after consume is a no-op (consumed = true).
        let state = transfer.inner.state.lock();
        assert_eq!(
            0, state.children_reserved,
            "Drop after consume must not double-decrement"
        );
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_reservation_drop_without_consume_recovers_counter() {
        let dir = tempdir().unwrap();
        let (transfer, _rx) = setup(
            dir.path(),
            FailedTransferPolicy::Continue,
            mock_s3_success(),
        );

        {
            let mut state = transfer.inner.state.lock();
            state.children_reserved = 5;
        }

        // Drop without calling consume (e.g. an early return before merge_spawned).
        {
            let _reservation = Reservation {
                transfer: transfer.inner.clone(),
                count: 5,
                consumed: false,
            };
        }

        let state = transfer.inner.state.lock();
        assert_eq!(
            0, state.children_reserved,
            "Drop on unconsumed Reservation must recover the counter"
        );
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_reservation_drop_underflow_saturates() {
        // State is 0; reservation claims 5. Drop must saturate, not panic.
        let dir = tempdir().unwrap();
        let (transfer, _rx) = setup(
            dir.path(),
            FailedTransferPolicy::Continue,
            mock_s3_success(),
        );

        {
            let _reservation = Reservation {
                transfer: transfer.inner.clone(),
                count: 5,
                consumed: false,
            };
        }

        let state = transfer.inner.state.lock();
        assert_eq!(0, state.children_reserved, "Drop must saturate, not panic");
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_reaping_batch_consume_decrements_counter() {
        let dir = tempdir().unwrap();
        let (transfer, _rx) = setup(
            dir.path(),
            FailedTransferPolicy::Continue,
            mock_s3_success(),
        );

        {
            let mut state = transfer.inner.state.lock();
            state.reaping_in_flight = 3;
        }

        let batch = ReapingBatch {
            transfer: transfer.inner.clone(),
            children: Some(Vec::new()),
            count: 3,
            consumed: false,
        };

        {
            let mut state = transfer.inner.state.lock();
            batch.consume(&mut state);
            assert_eq!(0, state.reaping_in_flight, "consume must decrement");
        }

        let state = transfer.inner.state.lock();
        assert_eq!(
            0, state.reaping_in_flight,
            "Drop after consume must not double-decrement"
        );
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_reaping_batch_drop_without_consume_recovers_counter() {
        let dir = tempdir().unwrap();
        let (transfer, _rx) = setup(
            dir.path(),
            FailedTransferPolicy::Continue,
            mock_s3_success(),
        );

        {
            let mut state = transfer.inner.state.lock();
            state.reaping_in_flight = 3;
        }

        // Simulates the scheduler dropping a JoinChildren work item without
        // executing, or execute_join_children panicking before batch.consume.
        {
            let _batch = ReapingBatch {
                transfer: transfer.inner.clone(),
                children: Some(Vec::new()),
                count: 3,
                consumed: false,
            };
        }

        let state = transfer.inner.state.lock();
        assert_eq!(
            0, state.reaping_in_flight,
            "Drop on unconsumed ReapingBatch must recover the counter"
        );
    }

    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_reaping_batch_take_children_then_consume() {
        // The happy path: execute_join_children takes children out for
        // processing, then consumes the batch under the lock at the end.
        let dir = tempdir().unwrap();
        let (transfer, _rx) = setup(
            dir.path(),
            FailedTransferPolicy::Continue,
            mock_s3_success(),
        );

        {
            let mut state = transfer.inner.state.lock();
            state.reaping_in_flight = 2;
        }

        let mut batch = ReapingBatch {
            transfer: transfer.inner.clone(),
            children: Some(Vec::new()),
            count: 2,
            consumed: false,
        };

        let _children = batch.take_children();

        {
            let mut state = transfer.inner.state.lock();
            batch.consume(&mut state);
            assert_eq!(0, state.reaping_in_flight);
        }
    }

    // --- Single-ticket spawn cadence tests ---

    /// Anti-stall guard: a directory of 500 objects fully transfers to
    /// completion with the single-ticket spawn cadence. If the spawn path
    /// stalls (returns nothing, never re-polled), this hangs and the
    /// timeout fires as a failure.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_single_ticket_500_objects_completes() {
        let dir = tempdir().unwrap();

        let objects: Vec<Object> = (0..500)
            .map(|i| {
                Object::builder()
                    .key(format!("obj{i:04}.bin"))
                    .size(4)
                    .build()
            })
            .collect();
        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(move || {
            ListObjectsV2Output::builder()
                .set_contents(Some(objects.clone()))
                .build()
        });
        let get = mock!(aws_sdk_s3::Client::get_object).then_output(|| {
            GetObjectOutput::builder()
                .content_length(4)
                .content_range("bytes 0-3/4")
                .e_tag("test-etag")
                .body(aws_sdk_s3::primitives::ByteStream::from_static(b"data"))
                .build()
        });
        let client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[head_rule(4), list, get]);

        let config = crate::Config::builder()
            .sdk_client(client.clone())
            .build()
            .expect("valid config");
        let handle = crate::client::Handle::test_handle_managed(config);

        let (transfer, completion_rx) =
            setup_enqueued(dir.path(), FailedTransferPolicy::Continue, client, handle);

        timeout(Duration::from_secs(30), async {
            let _ = completion_rx.await;
        })
        .await
        .expect("500-object transfer must complete without stalling");

        assert_eq!(transfer.successful_downloads(), 500);
        for i in 0..500 {
            assert!(
                dir.path().join(format!("obj{i:04}.bin")).exists(),
                "object {i} must be present on disk"
            );
        }

        crate::runtime::test_util::shutdown(&transfer.ctx().handle).await;
    }

    /// Backstop test: with a small max_concurrent set via pipeline_depth,
    /// in-flight children never exceed it.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_backstop_limits_in_flight_children() {
        let dir = tempdir().unwrap();
        let max_concurrent: usize = 4;

        let objects: Vec<Object> = (0..20)
            .map(|i| {
                Object::builder()
                    .key(format!("file{i:02}.txt"))
                    .size(3)
                    .build()
            })
            .collect();
        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(move || {
            ListObjectsV2Output::builder()
                .set_contents(Some(objects.clone()))
                .build()
        });
        let get = mock!(aws_sdk_s3::Client::get_object).then_output(|| {
            GetObjectOutput::builder()
                .content_length(3)
                .content_range("bytes 0-2/3")
                .e_tag("test-etag")
                .body(aws_sdk_s3::primitives::ByteStream::from_static(b"abc"))
                .build()
        });
        let client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[head_rule(3), list, get]);

        let config = crate::Config::builder()
            .sdk_client(client.clone())
            .build()
            .expect("valid config");
        let handle = crate::client::Handle::test_handle_tokio(config);

        let walk = S3Walker::builder().build().walk(
            S3WalkContext::builder()
                .sdk_client(client)
                .bucket("test-bucket")
                .build(),
        );

        let (ctx, completion_rx) = TransferContext::new(handle);
        ctx.handle
            .scheduler
            .register_empty_group_for_test(ctx.id.id);

        let input = crate::operation::download_objects::DownloadObjectsInput::builder()
            .bucket("test-bucket")
            .destination(dir.path())
            .failure_policy(FailedTransferPolicy::Continue)
            .build()
            .unwrap();
        let transfer = DownloadObjectsTransfer::new(ctx, &input, walk, max_concurrent);

        let mut max_observed: usize = 0;

        timeout(Duration::from_secs(10), async {
            loop {
                // Sample in-flight before polling.
                {
                    let state = transfer.inner.state.lock();
                    let active = state
                        .children
                        .values()
                        .filter(|c| c.handle.status() == crate::types::TransferStatus::Active)
                        .count()
                        + state.children_reserved;
                    if active > max_observed {
                        max_observed = active;
                    }
                }

                match transfer.poll_work() {
                    PollWork::Ready { io: mut work, .. } => {
                        transfer.execute(&mut work).await;
                    }
                    PollWork::Spawned => {}
                    PollWork::Pending => {
                        tokio::time::sleep(Duration::from_millis(5)).await;
                    }
                    PollWork::Done => break,
                }
            }
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        assert_eq!(transfer.successful_downloads(), 20);
        assert!(
            max_observed <= max_concurrent,
            "in-flight children ({max_observed}) must not exceed backstop ({max_concurrent})"
        );
    }

    /// The fused ladder reaps terminal children continuously rather than
    /// waiting for a batch threshold: every poll drains whatever terminals
    /// exist and, when it also spawns, fuses both into one `Ready { spawned }`.
    /// This guards the de-batching invariant — terminal children must NOT
    /// accumulate to a large unreaped backlog while a directory is spawning.
    /// A regression that reintroduced a reap-only-when->=MAX_REAP_PER_POLL
    /// threshold (or reap-behind-spawn starvation) would let the backlog grow
    /// and fail the bound below.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_reap_is_continuous_not_batched() {
        let dir = tempdir().unwrap();

        let objects: Vec<Object> = (0..300)
            .map(|i| {
                Object::builder()
                    .key(format!("f{i:04}.txt"))
                    .size(2)
                    .build()
            })
            .collect();
        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(move || {
            ListObjectsV2Output::builder()
                .set_contents(Some(objects.clone()))
                .build()
        });
        let get = mock!(aws_sdk_s3::Client::get_object).then_output(|| {
            GetObjectOutput::builder()
                .content_length(2)
                .content_range("bytes 0-1/2")
                .e_tag("test-etag")
                .body(aws_sdk_s3::primitives::ByteStream::from_static(b"ok"))
                .build()
        });
        let client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[head_rule(2), list, get]);

        let (transfer, completion_rx) = setup(dir.path(), FailedTransferPolicy::Continue, client);

        // Track how many polls fused a reap with a spawn (Ready{spawned:true})
        // to confirm the fused path is actually exercised, and the peak
        // unreaped-terminal backlog to confirm reaping keeps up.
        let mut fused_reap_spawn = 0usize;
        let mut max_terminal_backlog = 0usize;

        timeout(Duration::from_secs(30), async {
            loop {
                // Sample the unreaped-terminal backlog before polling.
                let terminal_before = {
                    let state = transfer.inner.state.lock();
                    state
                        .children
                        .values()
                        .filter(|c| c.handle.status() != crate::types::TransferStatus::Active)
                        .count()
                };
                max_terminal_backlog = max_terminal_backlog.max(terminal_before);

                match transfer.poll_work() {
                    PollWork::Ready {
                        io: mut work,
                        spawned,
                    } => {
                        if spawned {
                            fused_reap_spawn += 1;
                        }
                        transfer.execute(&mut work).await;
                    }
                    PollWork::Spawned => {}
                    PollWork::Pending => {
                        tokio::time::sleep(Duration::from_millis(5)).await;
                    }
                    PollWork::Done => break,
                }
            }
            let _ = completion_rx.await;
        })
        .await
        .expect("transfer should complete within timeout");

        // Liveness under the fused spawn+reap ladder: a 300-object directory
        // completes with every child reaped and no hang. This guards the ladder
        // restructure against dropping work, starving reap, or deadlocking.
        //
        // Whether in-flight stays flat and fusion fires is a property of real
        // concurrency and is measured with end-to-end benchmarks, not asserted
        // here: this single-parent-loop harness drives children in bursts at the
        // parent's await points, so completion timing (hence backlog magnitude
        // and whether reap+spawn coincide in one poll) is a harness artifact,
        // not the product cadence. `max_terminal_backlog`/`fused_reap_spawn` are
        // captured for debugging but not gated.
        assert_eq!(transfer.successful_downloads(), 300);
        let _ = (max_terminal_backlog, fused_reap_spawn);
    }

    /// `dispatch_walk` gates listing on the key buffer alone: it dispatches a
    /// page while `pending_entries` is below the low-water mark, refuses at or
    /// above it, and — the property this change introduced — dispatches
    /// regardless of how many children are in flight. Listing is decoupled from
    /// the child in-flight budget (`pipeline_depth`); a re-coupling regression
    /// would fail the third case.
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_dispatch_walk_gates_on_key_buffer_not_children() {
        // Each case uses a fresh transfer so a dispatch in one does not consume
        // the walk / set walk_in_flight for the next.
        let dir = tempdir().unwrap();

        // Below low-water, no children: dispatches.
        {
            let (transfer, _rx) = setup(
                dir.path(),
                FailedTransferPolicy::Continue,
                mock_s3_success(),
            );
            let mut state = transfer.inner.state.lock();
            assert!(state.pending_entries.len() < WALK_LOW_WATER);
            assert!(
                transfer.dispatch_walk(&mut state).is_some(),
                "should dispatch a walk when the key buffer is below low-water"
            );
        }

        // At/above low-water: refuses (buffer backpressure).
        {
            let (transfer, _rx) = setup(
                dir.path(),
                FailedTransferPolicy::Continue,
                mock_s3_success(),
            );
            let mut state = transfer.inner.state.lock();
            for i in 0..WALK_LOW_WATER {
                state.pending_entries.push_back(
                    crate::model::Object::builder()
                        .key(format!("k{i}"))
                        .size(1)
                        .build(),
                );
            }
            assert!(
                transfer.dispatch_walk(&mut state).is_none(),
                "should not dispatch a walk when the key buffer is at/above low-water"
            );
        }

        // Decoupling: children at the in-flight budget (pipeline_depth) but the
        // key buffer low → still dispatches. The old gate refused here.
        {
            let (transfer, _rx) = setup(
                dir.path(),
                FailedTransferPolicy::Continue,
                mock_s3_success(),
            );
            let mut state = transfer.inner.state.lock();
            state.children_reserved = transfer.inner.pipeline_depth;
            assert!(
                transfer.dispatch_walk(&mut state).is_some(),
                "listing must proceed even when children are at the in-flight budget"
            );
        }
    }

    /// Incremental delivery: `execute_advance_walker` publishes discovered keys
    /// to `pending_entries` in `WALK_FLUSH_CHUNK`-sized batches mid-advance plus
    /// a final remainder, rather than one batch at the end. This pins that the
    /// chunked delivery is lossless and order-preserving across a chunk
    /// boundary (150 keys = two full chunks + a 22-key remainder).
    #[cfg_attr(miri, ignore)]
    #[tokio::test]
    async fn test_incremental_delivery_lossless_and_ordered() {
        let dir = tempdir().unwrap();

        let n = 2 * WALK_FLUSH_CHUNK + 22; // spans two full chunks + a remainder
        let objects: Vec<Object> = (0..n)
            .map(|i| Object::builder().key(format!("k{i:05}")).size(1).build())
            .collect();
        let expected: Vec<String> = objects
            .iter()
            .map(|o| o.key().unwrap().to_string())
            .collect();
        let list = mock!(aws_sdk_s3::Client::list_objects_v2).then_output(move || {
            ListObjectsV2Output::builder()
                .set_contents(Some(objects.clone()))
                .build()
        });
        let get = mock!(aws_sdk_s3::Client::get_object).then_output(|| {
            GetObjectOutput::builder()
                .content_length(1)
                .content_range("bytes 0-0/1")
                .e_tag("test-etag")
                .body(aws_sdk_s3::primitives::ByteStream::from_static(b"x"))
                .build()
        });
        let client = mock_client!(aws_sdk_s3, RuleMode::MatchAny, &[head_rule(1), list, get]);
        let (transfer, _rx) = setup(dir.path(), FailedTransferPolicy::Continue, client);

        // Dispatch one walk and execute it, capturing the keys it delivers.
        let mut work = {
            let mut state = transfer.inner.state.lock();
            match transfer.dispatch_walk(&mut state) {
                Some(PollWork::Ready { io, .. }) => io,
                other => panic!("expected an AdvanceWalker work item, got {other:?}"),
            }
        };
        transfer.execute(&mut work).await;

        // n < MAX_ENTRIES_PER_WALK, so a single advance drains all of them.
        let delivered: Vec<String> = {
            let state = transfer.inner.state.lock();
            state
                .pending_entries
                .iter()
                .map(|o| o.key().unwrap().to_string())
                .collect()
        };

        assert_eq!(
            delivered.len(),
            n,
            "all discovered keys must be delivered exactly once (no drop/dup across chunks)"
        );
        assert_eq!(
            delivered, expected,
            "keys must be delivered in listing order"
        );
    }
}