I once tried to code an assignment involving TF-IDF on a corpus of docs. It ran blazingly fast compared to python. But was debugging segmentation faults until the last minute.