With the -r / --randomize flag, riscv-ctg uses python-constraint's MinConflictsSolver. This is not reproducible. It seems like this was noticed at least:
A obvious down-side of using random solvers is the fact that the same test cannot be reproduced again. However, the typical intent of users of CTG would be to generates tests satisfying the cover-points defined in the CGF. Since that is guaranteed by CTG, the reproducibility issue can be set aside.
I guess this was really set aside because unfortunately MinConflictsSolver uses global random and doesn't let you provide a seed. This can be easily worked around by seeding the global random before solving, or by copying & fixing MinConflictsSolver. But I think I will just fix it in that project.
I also noticed some other uses of the global random in riscv-ctg which should not be there: several uses of random.choice in cross_comb.py.
Timmmm
commented
4 months ago
With the
-r/--randomizeflag, riscv-ctg usespython-constraint'sMinConflictsSolver. This is not reproducible. It seems like this was noticed at least:I guess this was really set aside because unfortunately
MinConflictsSolveruses globalrandomand doesn't let you provide a seed. This can be easily worked around by seeding the globalrandombefore solving, or by copying & fixingMinConflictsSolver. But I think I will just fix it in that project.I also noticed some other uses of the global
randomin riscv-ctg which should not be there: several uses ofrandom.choiceincross_comb.py.