Oscar Dowson
Oscar Dowson
Just confirming that this is still a problem. But my efforts to create a pure HiGHS reproducer have not been successful. ```Julia Coefficient ranges: Matrix [2e-09, 3e+01] Cost [1e+00, 1e+00]...
The MPS file: [log2.txt](https://github.com/user-attachments/files/18341381/log2.txt) (download and rename) ```julia using HiGHS highs = Highs_create() Highs_readModel(highs, "/tmp/hypatia/log2.mps") ret = Highs_run(highs) index = Int32[160, 210, 211] value = [0.027716058468715117, 0.09140473855744354, -0.14285714401611166] ret =...
Closing in favor of the upstream https://github.com/ERGO-Code/HiGHS/issues/2109
That error doesn't look related
I've made a bunch of recent changes, including getting the tests to pass, so are you able to rebase on the latest `master`?
So the upside here is that we will not support bounded integer variables. The downside is that for unbounded integer variables the error message is now worse.
To clarify, you can read LP files currently. But the reader is not robust to the wide variety of LP files in the wild.
See also https://github.com/jump-dev/MathOptInterface.jl/issues/2563 The parser should not read line-by-line.
We don't necessarily have to make it available via the callback. Just reporting it in the end `info` struct would be enough. I was going to use the callback from...
This could be best implemented as a new AD backend. It shouldn't be too complicated.