Deadlock in `Lin Bytes test with Thread` on Cygwin

[jmid]

Opening this to track it separately from #520, as it will allow to distinguish occurrences of one from the other.
Also, in comparison to #520, this on a Threads test - and not involving shrinking.

On the merge to main of #519 for the Cygwin 5.3 workflow we just saw excessive time spent in Lin Bytes test with Thread
https://github.com/ocaml-multicore/multicoretests/actions/runs/12808801661/job/35712309410

random seed: 366364260
generated error fail pass / total     time test name

[ ]    0    0    0    0 / 2500     0.0s Lin Bytes test with Domain
[ ]    0    0    0    0 / 2500     0.0s Lin Bytes test with Domain (generating)
[ ]   37    0    0   37 / 2500    60.2s Lin Bytes test with Domain (shrinking:   15.0003)
[ ]   37    0    0   37 / 2500   122.0s Lin Bytes test with Domain (shrinking:   18.0006)
[ ]   37    0    0   37 / 2500   182.5s Lin Bytes test with Domain (shrinking:   20.0008)
[ ]   37    0    0   37 / 2500   245.1s Lin Bytes test with Domain (shrinking:   23.0005)
[ ]   37    0    0   37 / 2500   308.1s Lin Bytes test with Domain (shrinking:   25.0002)
[ ]   37    0    0   37 / 2500   371.9s Lin Bytes test with Domain (shrinking:   28.0010)
[ ]   37    0    0   37 / 2500   433.4s Lin Bytes test with Domain (shrinking:   30.0013)
[ ]   37    0    0   37 / 2500   493.5s Lin Bytes test with Domain (shrinking:   32.0012)
[ ]   37    0    0   37 / 2500   556.0s Lin Bytes test with Domain (shrinking:   34.0002)
[ ]   37    0    0   37 / 2500   617.4s Lin Bytes test with Domain (shrinking:   35.0012)
[ ]   37    0    0   37 / 2500   677.9s Lin Bytes test with Domain (shrinking:   36.0015)
[ ]   37    0    0   37 / 2500   741.5s Lin Bytes test with Domain (shrinking:   37.0013)
[ ]   37    0    0   37 / 2500   806.4s Lin Bytes test with Domain (shrinking:   39.0004)
[ ]   37    0    0   37 / 2500   866.8s Lin Bytes test with Domain (shrinking:   40)
[ ]   37    0    0   37 / 2500   927.3s Lin Bytes test with Domain (shrinking:   41.0003)
[✓]   38    0    1   37 / 2500   984.5s Lin Bytes test with Domain

[ ]    0    0    0    0 /  250     0.0s Lin Bytes test with Thread
[ ]    3    0    0    3 /  250  1481.8s Lin Bytes test with Thread
[ ]    4    0    0    4 /  250  8943.0s Lin Bytes test with Thread (collecting)
Error: The operation was canceled.

That's 149min ~ 2h 29min to test 4 random inputs on the Thread variant? 🤔

[shym]

Spotted again in the scheduled run on Cygwin trunk (with the relevant part of the timestamp):

05:07:39  random seed: 262132514
05:07:39  generated error fail pass / total     time test name
05:07:39 
05:07:39  [ ]    0    0    0    0 / 5000     0.0s Lin Bytes test with Domain
05:08:39  [ ]    0    0    0    0 / 5000     0.0s Lin Bytes test with Domain (generating)
05:09:40  [ ]  161    0    0  161 / 5000    60.0s Lin Bytes test with Domain (shrinking:   41)
05:10:41  [ ]  161    0    0  161 / 5000   120.9s Lin Bytes test with Domain (shrinking:   48)
05:11:41  [ ]  161    0    0  161 / 5000   182.6s Lin Bytes test with Domain (shrinking:   52.0010)
05:12:42  [ ]  161    0    0  161 / 5000   242.8s Lin Bytes test with Domain (shrinking:   54.0011)
05:13:42  [ ]  161    0    0  161 / 5000   302.9s Lin Bytes test with Domain (shrinking:   58.0004)
05:13:51  [ ]  161    0    0  161 / 5000   363.0s Lin Bytes test with Domain (shrinking:   97.0002)
05:13:51  [✓]  162    0    1  161 / 5000   372.2s Lin Bytes test with Domain
05:13:51 
07:43:15  [ ]    0    0    0    0 /  250     0.0s Lin Bytes test with Thread

[shym]

Spotted again in the scheduled run on Cygwin 5.4 branch (with the relevant part of the timestamp):

04:48:21  random seed: 476402853
04:48:21  generated error fail pass / total     time test name
04:48:21
04:48:21  [ ]    0    0    0    0 / 5000     0.0s Lin Bytes test with Domain
          ...
05:07:31
05:07:38  [ ]    0    0    0    0 / 1000     0.0s Lin Bytes stress test with Domain
05:08:03  [ ]  226    0    0  226 / 1000     6.9s Lin Bytes stress test with Domain
05:08:03  [✓] 1000    0    0 1000 / 1000    32.4s Lin Bytes stress test with Domain
05:08:03
07:55:19  [ ]    0    0    0    0 / 5000     0.0s Lin Bytes test with Thread

[jmid]

This happened again on Cygwin trunk on #508:
https://github.com/ocaml-multicore/multicoretests/actions/runs/15142200673/job/42568941136

random seed: 195021863
generated error fail pass / total     time test name

[ ]    0    0    0    0 / 5000     0.0s Lin Bytes test with Domain
[ ]    0    0    0    0 / 5000     0.0s Lin Bytes test with Domain (generating)
[ ]   49    0    0   49 / 5000    60.2s Lin Bytes test with Domain (shrinking:   43.0003)
[ ]   49    0    0   49 / 5000   122.6s Lin Bytes test with Domain (shrinking:   49.0009)
[ ]   49    0    0   49 / 5000   183.8s Lin Bytes test with Domain (shrinking:   52.0007)
[ ]   49    0    0   49 / 5000   244.4s Lin Bytes test with Domain (shrinking:   55.0002)

[...]

[ ]   49    0    0   49 / 5000  1588.4s Lin Bytes test with Domain (shrinking:   81.0010)
[ ]   49    0    0   49 / 5000  1649.3s Lin Bytes test with Domain (shrinking:   82.0007)
[ ]   49    0    0   49 / 5000  1720.9s Lin Bytes test with Domain (shrinking:   83.0008)
[ ]   49    0    0   49 / 5000  1780.9s Lin Bytes test with Domain (shrinking:   93.0003)
[ ]   49    0    0   49 / 5000  1840.9s Lin Bytes test with Domain (shrinking:  126.0002)
[✓]   50    0    1   49 / 5000  1842.7s Lin Bytes test with Domain

[ ]    0    0    0    0 / 1000     0.0s Lin Bytes stress test with Domain
[✓] 1000    0    0 1000 / 1000    39.6s Lin Bytes stress test with Domain

[ ]    0    0    0    0 / 5000     0.0s Lin Bytes test with Thread
Error: The operation was canceled.

[jmid]

This just happened again on the scheduled trunk Cygwin run:
https://github.com/ocaml-multicore/multicoretests/actions/runs/15964255540/job/45021752188

random seed: 330479995
generated error fail pass / total     time test name

[ ]    0    0    0    0 / 5000     0.0s Lin Bytes test with Domain
[ ]    0    0    0    0 / 5000     0.0s Lin Bytes test with Domain (generating)
[ ]   28    0    0   28 / 5000    60.1s Lin Bytes test with Domain (shrinking:   47.0004)
[ ]   28    0    0   28 / 5000   120.2s Lin Bytes test with Domain (shrinking:   78.0007)
[✓]   29    0    1   28 / 5000   155.0s Lin Bytes test with Domain

[ ]    0    0    0    0 / 1000     0.0s Lin Bytes stress test with Domain
[ ]  683    0    0  683 / 1000    25.2s Lin Bytes stress test with Domain
[✓] 1000    0    0 1000 / 1000    37.7s Lin Bytes stress test with Domain

[ ]    0    0    0    0 / 5000     0.0s Lin Bytes test with Thread
Error: The operation was canceled.

[jmid]

Note: This has been observed for months (Jan, Feb, Mar) before #540 tweaking the Thread mode was merged on Mar.26.
As such I don't believe the tweaks are to blame...