I had a Hadoop map-reduce job that kept timing out, which led to this interesting discovery:
$ time ./json-parser-test.py real 0m0.205s user 0m0.152s sys 0m0.032s $ time ./json-parser-test-no-speedups.py real 0m2.069s user 0m2.044s sys 0m0.024s $ time jython ./json-parser-test-no-speedups.py real 79m59.785s user 80m23.709s sys 0m14.441s
Moral: use Java-based JSON libraries if you have to use Jython and JSON. Also, Java sucks.
I’m just going on record to state that the two reasons I’ve seen on the net for lack of tree-based collections in the Python standard library — (1) that they have bad locality (gee, very unlike huge memory-hogging randomized hash tables, oh and just try scanning one of those…), and (2) that Python makes it so simple to write your own, just do that! (well, yes I can and have, but hey it’s 2011 already, rbtrees are like 40 years old!) — are inane.
