[BUG] test_incumbent_get_set_callback[swath1.mps] fails with TimeLimit on oldest-deps config

[ramakrishnap-nv]

The MILP incumbent-callback test hits the 10s time limit and returns TimeLimit instead of FeasibleFound/Optimal, failing the assertion.

Failure

• Job: conda-python-tests / 12.2.2, 3.11, amd64, rockylinux8, l4, earliest-driver, oldest-deps (PR Fix symbol visibility and add checks #1526)
• Log: https://github.com/NVIDIA/cuopt/actions/runs/28962508664/job/85954467414#step:12:3742
• Test: tests/linear_programming/test_incumbent_callbacks.py::test_incumbent_get_set_callback[/mip/swath1.mps]

>       assert termination in (
            MILPTerminationStatus.FeasibleFound,
            MILPTerminationStatus.Optimal,
        )
E       assert <MILPTerminationStatus.TimeLimit: 5> in (<MILPTerminationStatus.FeasibleFound: 8>, <MILPTerminationStatus.Optimal: 1>)
tests/linear_programming/test_incumbent_callbacks.py:90: AssertionError

Analysis

The test fixes CUOPT_TIME_LIMIT = 10 and asserts a feasible/optimal termination (test_incumbent_callbacks.py:83, :90). On the slowest matrix entry (oldest-deps, earliest-driver, CUDA 12.2.2, L4) swath1.mps does not reach a feasible solution within 10s, so the solver returns TimeLimit and the assertion fails. Appears environment/timing-sensitive rather than a solver correctness bug.

Possible fixes

• Treat TimeLimit gracefully (skip/soft-pass when no incumbent fired within the limit), similar to the existing root-node skip at test_incumbent_callbacks.py:98.
• Or raise the time limit for this test / use a smaller MIP instance on the oldest-deps config.

Related (closed)

• [BUG] flaky incumbent callback tests with swath1 #1207 flaky incumbent callback tests with swath1 (closed as dup of [BUG] Crash on test_incumbent_callbacks (python) #1185)
• [BUG] Crash on test_incumbent_callbacks (python) #1185 crash on test_incumbent_callbacks (closed, completed) — different failure mode (crash, not TimeLimit)