Calling it from anything is pain. One need to carry Julia runtime with them.
Bad tooling support
Very small community
The promise of performance is not backed by any real proof (see here)
Java / other JVM languages is slightly better than Julia but still bad due to problematic calls from non-JVM languages (the situation is better than in Julia but still far from good).
The library core could be written in C++ with enforced code conventions and rules. To do so we need very good architecture planning before we start.
As a side variant we could try Rust (Proposed by Alexander Nozik but not supported by others). On the bad side we need to learn it and It is rather new, on good side, it is functional oriented, memory safe, have a large community and looks good.
MIT people still going to start with Julia. Since the library core is very small, we will have time to try different languages and choose one.
The most important thing at the moment is use cases, API and architecture plan.
