Git Bug Report: bisect run failed to locate the right commit, detail testing

Dear git developers:

Allow me to describe the testing situation more detail:

First the testing repository is in 
https://github.com/gawel/pyquery
my git version is 1.9.2 running on Archlinux 3.14.2

I try to track issue #74(which is closed now)
It give result False/True in version 1.2.8(good), and result False/False in version 1.2.9(bad).

The testing script are manualscript.py and autoscript.py
Both of them implement the test case describe in issue #74.
Th only difference is that autoscript.py call sys.exit() to return the testing value.

First we test with git-bisect manually with manualscript.py:
The testing result is shown in 'bisectmanual_manualscript'

Then we test with git-bisect manually with autoscript.py
This time we echo $? every time we execute autoscript.py, and the testing result is shown in 'bisectmanual_autoscript'

In both situation the script give the same result, and the return value of autoscript.py is correct, too.

However, if we use git-bisect-run with autoscript.py, it will show a different result. The testing result is shown in 'bisectrun'.
The log shows that autoscript.py output False/False all the way.
As Mr. Couder said, there is some checkout commit that autoscript.py and manualscript.py give different result, for example commit d22159bb32510e9eacf6c5c2408a79792e99fe76.
If I checkout this commit outside bisect state and run manualscript.py and autoscript.py, they both give False/True result.  
So, I guess there is some problem in the checkout procedure in bisect-run, so the commit didn't successfully checkout.

ps. I found this time the bisect run give me another version of output compared to the last version I sent to everybody OAO. That is weird.

Thank for helping.
Sincerely 
YodaLee
20140828