[pypy-issue] [issue1045] Crash of the JIT

Daid tracker at bugs.pypy.org
Tue Feb 28 18:53:28 CET 2012


Daid <daid303 at gmail.com> added the comment:

Ok, in that case. Ignore paranoid me (Sorry!)
(Just figured the code could also be iterating over dictionaries, which might
produce 'random' results if addresses are used in hash-keys)

Also, I've gotten a few new reports of people reporting the same bug to me. And
a few of those happen 100% of the time.

Maybe unrelated, but I tried to do a profiling run on the code today, and I lost
30% CPU time somewhere (it's not in the trace up from some point) and I lost
call origins (function using 70% time was called 0 times, but I know it was
called at least once). Not sure how that could happen or if this could be
related. But it might say something about the... quality of the python code.

Next I figured, maybe I can find out which optimization is causing this. But the
"--jit enable_opts" is lacking documentation on what you can enable.
Just grepping trough the sourcecode I found this list:
"intbounds:rewrite:virtualize:string:earlyforce:pure:heap:ffi"
But I don't think that's a very complete list. And if I only enable those
optimalisations it no longer seems to crash.

________________________________________
PyPy bug tracker <tracker at bugs.pypy.org>
<https://bugs.pypy.org/issue1045>
________________________________________


More information about the pypy-issue mailing list