Tests fail on Mac with M1 chip

[abelsiqueira]

Description

Due to Cbc not supporting it: jump-dev/Cbc.jl#193
I recommend we just drop Cbc of the tests. An alternative would be to check conditionally remove it based on Sys.isapple().

Reproduction steps

Just run the tests on a Mac with M1 chip.

Logs

No response

OS

Mac

[clizbe]

Having a specific conditional for this edge case seems messy.

I think as long as we're testing more than one it's fine to remove it.
For now: HIGHS, GLPK
Later add: Xpress

If users run into the problem later hopefully remember this is why.

[clizbe]

Maybe we could add a note to the user wherever they choose the solver? 🤔

[abelsiqueira]

This would be just for the test. The code itself remains the same. I would not touch anything on the code or documentation side about Cbc not being supported, since they could fix it eventually, and it creates precedent for informing the user about solver status. I think leaving HiGHS as default is enough, and eventually we can mention that we also use Xpress, if applicable.

Apart from that, I agree with removing it, rather than creating the conditional.

[abelsiqueira]

Well, I spoke too soon. The Cbc case is "necessary" for coverage. To be clearer, we have a case for compute_dual_variables that fails in a specific way for Cbc.
So removing Cbc requires either creating a fake test only to make coverage happy, which is not a good practice. So we either

• don't remove Cbc and use the conditional instead; or
• don't implement the check and risk the error appearing in a real situation; or
• stop having 100% coverage.

I recommend the first.

[clizbe]

Ah dang. Weird that there's a test that fails specifically in a way for Cbc...
But I guess that means we're supporting Cbc except for Mac M1, which is good. (Right?)

[abelsiqueira]

Well, Cbc is not in a great place, so it does have these corner cases. Luckily, HiGHS is becoming the de-facto open source MIP solver, so most people won't get into this random failures.