fijall

PyPy doesn't have a specialized list for small strings. It might land one in the near future, your findings are a bit weird, but objects might be smaller because of smaller GC headers.

Cheers, fijal

This doesn't quite apply to PyPy (which doesn't have refcounting), but the point stands. In case of PyPy this is GC flags and moving GC.

"If you've got the control necessary to automatically STM things, you might as well just equally-automatically copy the Jython-style locking."

So far thinking about how to really do that automatically failed. Do you feel like providing an idea?

2,292,133 were in prisons in the United States in 2009 [from wikipedia]. Even though I guess conditions are better, 200 000 doesn't sound like that many after all.

The main reason why more people are not working on this is that:

* volunteers are interested somewhere else

* nobody is willing to put money into numpy on pypy happening

you should have a look at scheme interpreter that was written already maybe: https://bitbucket.org/pypy/lang-scheme/overview

It's incredible that L'Oréal placed 40. I wouldn't want to work for a company that has non-printable name apparently.

Importing people from anywhere else is on it's way out in the US as well. (even willingly)

The thing is in case of tracing you don't record all paths taken, just the one at the hot point. So say you have if that takes every path every second go, the JIT will compile just one path at first. Then the guard will fail every second iteration, hence the blackhole. Of course at some point JIT will realize guard fails often enough and will compile another path, so you'll get a tree. And so on...