Current status of library

[Clifford-appliedAI-GmbH]

Hi, I was wondering about the current status of this library, since it's not yet at version 1.0.0. e.g. Should I expect most or all of the functionality of SCIP to be available? Are there likely to be many severe bugs compared with calling SCIP from C++?

[mmghannam]

Hi @CliffordWilmot,
First of all thank you for your interest in russcip!

Ok, so about functionality,
There are many supported ways to install the library (refer to the README) with different tradeoffs of control/ease.

The library allows for unsafe access to most if not all of the SCIP C-API, through the ffi module and the raw feature.
But you probably meant the safe interface. Currently, these are the things that have wrappers in russcip:

• Methods to build a MIP.
• Methods to control parameters.
• Some plugins have ergonomic wrappers (as traits) like branching rules, primal heuristics, event handlers and variable pricers.
  There are also some other methods that define other constraint types that you could check in the model methods.
  So it really depends on which functionality from SCIP you need.

Regarding stability,
The way SCIP functionality is wrapped in russcip tries to achieve mostly two goals:

1. Guaranteeing memory safety (as much as possible).
2. Preventing misuse of SCIP functionality (preferably at compile-time), with the use of the Type State pattern and enums to represent what the functions need as input/output.

The test coverage is in a good state and I try to keep it from decreasing. We also test for memory leaks in CI. Maybe it also helps to know that I have been using the library extensively the past period with no major issues so far.

Full disclosure though :D the only current known bug is about accessing solutions after the model is dropped, here's the issue for reference.

I'm not really comfortable yet to move to 1.0.0 because I want to still have the freedom to change a few things in the API. However, I will keep additions and bugfixes as updates in the bugfix version and if there are any changes in the API I will change the minor version.

As far as I know the C++ interface also tries to achieve automatic memory management, and regarding functionality I think you could get very far with both libraries. Maybe it would be easier to access the "uncovered" functionality in the C++ interface since you could just write C code with no need for calling bindings. I think it depends mostly on which language you'd like to work in.

Sorry if this was a bit random, I hope it clears some things up for you though :)
Please let me know if you have any follow up questions.

[Clifford-appliedAI-GmbH]

Hi @mmghannam , thanks for the very detailed reply, I really appreciate it! I think I now have a much better understanding of the safe and unsafe parts.

No worries about it being 1.0.0, I understand that the API may change.

I will have a look through this library and the C++ interface in more detail, in order to figure out what works best for my use case.