[bisected] Performance regression in default HiGHS LP solver (v1.11.0 → v1.12.0)

[maniok098]

Dear HiGHS team,

after updating to v1.12.0, for my problem, I observed a significant performance regression of the default LP solver compared to v1.11.0. Both in C++ and highspy. For this LP problem sampleLP.mps.gz, the solution time increased from 0.85s (v1.11.0) to 1.84s (v1.12.0). The number of simplex iterations increased from 19292 to 60660.

After git bisection, this regression was tracked to the commit 20186bf . (The solver was built from C++ source on an Archlinux machine.) Please let me know if any additional information would be helpful.

Best and happy new year.

[maniok098]

Moved to issue #2722