[jakobnissen]

Performance in native Python really is bad. Nine out of ten times, you can get away with it by figuring out how to push all the computation into the layer of C that Python rests on, but sometimes you just can't, and then Python becomes essentially unusable.

From my own experience, implementing Needleman-Wunch or k-mer counting is literally >100x slower in Python than Julia or C. Meaning a two-minute task turns into a 3 hour task - at least. At this point you can begin coding in Rust, but it feels bitter to learn Python, get into it, and then realize you've learned an entire programming language just to use it as wrapper for your actual business code.