Oscar Dowson

I'll see if I can get a profile, but I don't know if I can easily measure from my side. Maybe there's some issue casting between your difference callback objects?...

I ran this: ```julia using PProf import JuMP import HiGHS function main() model = JuMP.read_from_file("debug_model.mps") JuMP.set_optimizer(model, HiGHS.Optimizer) JuMP.set_attribute(model, "time_limit", 30.0) counter = Ref{Int}(0) function callback_fn(args...) counter[] += 1 return Cint(0)...

I removed all of the Julia side of the callback with the same result. So I think there is an issue somewhere on the C side. I'll try and write...

The plot thickens. I can't reproduce this from C. So maybe it is an issue crossing the C/Julia language barrier. For example this was fine: ``` #include "interfaces/highs_c_api.h" #include #include...

I really don't understand. This is a problem with the binaries we have compiled for HiGHS_jll, but not locally. [email protected] and my local 1.11 are good, whether called from C...

> Trigger a rebuild and hope for the best? Not likely. This might be a compiler optimization that is not in the Yggdrasil build but is locally. It's suspicious that...

I "fixed" this by disabling the simplex interrupt callback. We didn't fix the underlying issue.

Why do you want to copy the model? In most cases, I really really recommend against copying a model. There's usually a different way you can achieve your objective.

I don't know enough about GPUs to fix #314. I'm just playing whackamole. It surprises me that there aren't any compat bounds in SCS_GPU_jll?

> requires recompiling SCS_jll with the new cones turned on Why are they not on by default? Because it breaks the API like adding the complex PSD?