python-mip
python-mip copied to clipboard
Coin versions
Since the CBS binary is distributed with the framework it is unclear exactly what version of CBC, CLP, CoinUtils etc is being used. I am assuming the master branch?
Would it be possible to get a python-mip version based on stable Coin libs? I now a lot have been happening with CBC C interface and I guess getting that to a stable version is the bottleneck?
Somewhat related, i would also propose to either:
- publish the build-pipeline behind the delivered libraries
- publish build-information when updating
- or at least: wrap Cbc_getVersion()
- (one could also think about adding a
Cbc_isEnabledParallel()
or something similar to be able to reason about multihreaded builds. I suspect it's not there yet in the Cbc interface.)
- (one could also think about adding a
This should make it more clear, what is happening internally.
Not much experimented with this software, i immediately wonder about:
- which version of Cbc is used
- is it multithreaded (enable-parallel)
- there seems to be support for the barrier alg: does this mean MUMPS is delivered too; or only the native dense linear solver?
- (vs-build or mingw-build; i expect this decision to have an effect on multithreaded builds)
I'm note able to deduce this information (without doing ugly stuff with the libs)
Hi @sschnug, @spoorendonk and @tkralphs . I would love to just use the automatically built cbc binaries from here: https://bintray.com/coin-or/download/Cbc/master (specially for windows, where building it is a burden)
Last time I tried, one difficulty was that I only managed to load the DLL in the Python code (ffi.dlopen) when all packages were bundled statically (osi, cgl, ...). It would be necessary to modify the path in windows so that windows could find other DLLs also. Putting on the same directory didn't work.
Regarding MUMPS, well, technically it's probably very easy, we would just have to check which is the license of MUMPS to see if we are allowed to bundle it in our binaries.