Failures of some STM property based tests

[jpraynaud]

Why

We have noticed some failures in some property based tests of the STM library which make the CI flaky:

• https://github.com/input-output-hk/mithril/actions/runs/11610605293/job/32330224044#step:5:1577
• https://github.com/input-output-hk/mithril/actions/runs/11614811947/job/32343862348#step:5:1449

We'd like to understand if these occurrences are normal and create the appropriate fixes if necessary.

What

Investigate and fix the source of flakiness.

How

• [x] Investigate the source of flakiness
• [ ] Fix the flakiness

[curiecrypt]

Property test batch_verify in /mithril-stm/stm.rs fails for the following cases:

• Case 1:

  let nparties = 12;
  let m = 10;
  let k = 4;
  let seed = [60, 97, 149, 225, 10, 42, 3, 8, 185, 163, 227, 69, 225, 72, 238, 154, 164, 77, 167, 233, 251, 171, 68, 218, 151, 159, 193, 202, 209, 169, 90, 210];
  batch_size = 6;

  The failing instance occurs in the 3rd iteration of the batch with the randomly filled message:

  let msg: [u8; 32] = [237, 247, 38, 63, 164, 83, 106, 19, 86, 163, 29, 139, 0, 156, 213, 132, 8, 172, 194, 77, 117, 160, 158, 189, 70, 185, 249, 151, 117, 42, 113, 7];

  The number of signatures is 5 (created with the helper function find_signatures), but in the aggregation process, only 3 unique signatures are found. sigs.len() < k => 3 < 4

• Case 2:

  let nparties = 6;
  let m = 10;
  let k = 4;
  let seed = [99, 129, 128, 56, 143, 205, 173, 33, 36, 48, 11, 203, 0, 117, 75, 70, 90, 130, 249, 188, 234, 190, 69, 233, 128, 35, 28, 49, 105, 186, 88, 183,];
  batch_size = 5;

  The failing instance occurs in the 3rd iteration of the batch with the randomly filled message:

  let msg: [u8; 32] = [171, 90, 33, 241, 15, 147, 222, 182, 45, 8, 55, 183, 212, 48, 225, 52, 218, 186, 11, 69, 112, 35, 21, 82, 211, 162, 175, 7, 54, 112, 63, 17,];

  The number of signatures is 4 (created with the helper function find_signatures), but in the aggregation process, only 3 unique signatures are found. sigs.len() < k => 3 < 4

---

Possible workaround: For the test batch_verify, the iteration that generates insufficient signatures could be repeated. Since the aim of this test is not to check the aggregation functionality but to check the batch verification functionality. @jpraynaud

[jpraynaud]

Thanks @curiecrypt for the explanation.

We could probably use different protocol parameter instead to make sure that aggregation is never a problem (e.g. with phi_f very close to 1.0 and/or higher m value)? WDYT?