Re: [capi-sig] c extension corrupting the python interpreter ?
[changed the citation order to fix the top-post]
Ian Miers, 22.06.2012 04:37:
On Thu, Jun 21, 2012 at 8:46 AM, Andrei Paraschivescu wrote:
Ian, do you have other Python interpreters running? I don't know if that can cause the behavior you describe, but as some information is shared across interpreters (e.g. you can unpickle a class you have not imported if another interpreter has), it could potentially be relevant.
I have not been able to find where the documentation describes what is shared across interpreters, but perhaps someone who understands that issue can chime in.
I have python2.7, 3.2 and 3.2 debug installed. Python 2.7 was likely running as part of some deamons in ubuntu( at least they are running now) . Funnily, I have these environments all inside virtualenv ( which appears to just sym lin the system python(in this case 3.2) interpreter and I believe changes the paths for site packages and such) and yet the errors don't seem to propagate back to the system interpreter or across virtualenv environments.
I'm sure that's not what Andrei meant. He likely referred to starting up multiple interpreters within the *same* process. Separate processes don't interfere with each other. From your answer, I take it that you're not using this feature (which is not commonly used anyway).
In any case, you might have more luck asking on the general Python mailing list (aka. comp.lang.python). It has a lot more listeners (aka. eyeballs).
As a general remark: yes, it's very possible that a C extension crashes Python. In almost all cases, it's due to a bug in the C extension, of which the vast majority are reference counting bugs. That's why I keep encouraging users to write their extensions in Cython instead of C, because it makes these things just so much easier - most of it happens automatically. And it's a lot more pleasant for Python programmers to write Cython code than to step down into C all on your own.
Stefan
I actually meant separate processes, but I am incorrect.
Based on Stefan's comment, I went back to what I thought was my
sample case, and I can see now it is explained by unpickle performing
implicit imports --- the additional interpreter was a red herring.
I apologize for the mis-direct.
Andrei
On Jun 22, 2012, at 1:08 AM, Stefan Behnel wrote:
[changed the citation order to fix the top-post]
Ian Miers, 22.06.2012 04:37:
On Thu, Jun 21, 2012 at 8:46 AM, Andrei Paraschivescu wrote:
Ian, do you have other Python interpreters running? I don't know
if that can cause the behavior you describe, but as some information is
shared across interpreters (e.g. you can unpickle a class you have not
imported if another interpreter has), it could potentially be relevant.I have not been able to find where the documentation describes
what is shared across interpreters, but perhaps someone who understands
that issue can chime in.I have python2.7, 3.2 and 3.2 debug installed. Python 2.7 was likely running as part of some deamons in ubuntu( at least they are running now) . Funnily, I have these environments all inside virtualenv
( which appears to just sym lin the system python(in this case 3.2)
interpreter and I believe changes the paths for site packages and such) and
yet the errors don't seem to propagate back to the system interpreter or
across virtualenv environments.I'm sure that's not what Andrei meant. He likely referred to
starting up multiple interpreters within the *same* process. Separate processes
don't interfere with each other. From your answer, I take it that you're not using this feature (which is not commonly used anyway).In any case, you might have more luck asking on the general Python
mailing list (aka. comp.lang.python). It has a lot more listeners (aka.
eyeballs).As a general remark: yes, it's very possible that a C extension
crashes Python. In almost all cases, it's due to a bug in the C extension,
of which the vast majority are reference counting bugs. That's why I keep encouraging users to write their extensions in Cython instead of C,
because it makes these things just so much easier - most of it happens automatically. And it's a lot more pleasant for Python programmers
to write Cython code than to step down into C all on your own.Stefan
capi-sig mailing list capi-sig@python.org http://mail.python.org/mailman/listinfo/capi-sig
participants (2)
-
Andrei Paraschivescu -
Stefan Behnel