The author seems to like Julia for numerics -- have the language and community improved since Dan Luu's damning review six years ago? https://danluu.com/julialang/
I use Python / numpy heavily, and the giant hole numpy opens in my type annotations causes me grief daily. If there's something better out there, I want to try it, but it has to work. It has to work not just in the "I'm putting together a journal article and I need it to run" sense, but in the "I'm building a system that crunches some heavy engineering numbers all the time, and customers expect consistent results" sense. Can Julia be trusted?
That was just around 2 years after the language became public and 4 years before it became stable, so I wouldn't really call the language "mature" back then (I didn't use the language then to have the insight even if I got before 1.0). From what I can see, the core developers seem to take seriously bug reports, even implementing sophisticated ways to accurately find the source of bugs [1] and running tests on the entire ecosystem before each new tagged release to catch bugs early. There are many large scientific projects using Julia [2], and there are industry leader packages in the area like DiffEq.jl.
The "time to first plot" (a consequence of the aggressive JIT that compiles large programs with heavy optimizations, as Julia doesn't have an interpreter outside of the debugger) still exists, but it is also improving at a fast rate.
There are comments suggesting that Julia is plenty capable, with libraries covering equivalent (or more) features provided by Python libraries like NumPy. YMMV.
I use Python / numpy heavily, and the giant hole numpy opens in my type annotations causes me grief daily. If there's something better out there, I want to try it, but it has to work. It has to work not just in the "I'm putting together a journal article and I need it to run" sense, but in the "I'm building a system that crunches some heavy engineering numbers all the time, and customers expect consistent results" sense. Can Julia be trusted?