Can I get review for #1076??

I know,thanks.

heh race condition, you got two reviews :)

Sometimes We need critical section :smile:

I was faster but than kmods comments are much more detailed and better quality ;-)

I just stumbled over a very interesting thesis which I somehow never encountered before:

Purely Interpretative Optimizations by Stefan Brunthaler

https://students.ics.uci.edu/~sbruntha/cgi-bin/download.py?key=thesis

Purely Interpretative Optimizations by Stefan Brunthaler

https://students.ics.uci.edu/~sbruntha/cgi-bin/download.py?key=thesis

There is also a video lecture of an updated and more advanced version: https://air.mozilla.org/mlq-high-performance-interpretation-using-multi-level-quickening/

He implemented the optimizations inside cpython 3 so it's also very relevant

By the first glance, I thought it was written in German...

oh only the first two pages ;-)

In theory, every change should come with a test. But some corner cases only happened with some NumPy data type. I am trying to create a mimic class as much as possible. But some of corner cases can't . For example:

```
with numpy.errstate(divide='ignore'):
print(1 / numpy.float_(0))
```

Any suggestion?