Incorporating real-world SCIP usage into performance testing: your use cases welcome!

[Joao-Dionisio]

SCIP undergoes weekly performance runs, and each change is tested on the MIPLIB benchmark. This is a good proxy for how performant SCIP is as a black-box solver. However, many people, especially through the interfaces, use special functionality of SCIP beyond default optimization. As such, we would be keen on adding examples of how users are actually using SCIP, to increase our coverage.
As an example, the loss of performance when copying a model because not all plugins were copied (partially fixed, but still WIP) was only discovered here on PySCIPOpt. Working with Branch-and-Price, I also found bugs that wouldn't have been discovered by our testing framework.

Your contributions would be greatly welcomed, and sensitive data will, of course, be anonymized. You can follow the usual SCIP channels in the Contact us page, or send me an email directly at [email protected]

It's your chance to become part of open-source software development!

[Joao-Dionisio]

Posting this as a discussion as well (besides issue #1143), to increase visibility.