Closed oxtopus closed 9 years ago
Hmm, I'm having a lot of trouble reproducing (specifically I'm getting a Fatal Python error: PyThreadState_Get: no current thread
whenever I try to import nupic on my OSX box). That being said, I'm pretty sure I fixed this bug in the 0.5.4 release (see https://github.com/jparyani/pycapnp/blob/develop/CHANGELOG.md#v054-2015-03-02). Can you try that out and let me know if it works for you?
Will do. Thanks!
I tried to craft a simpler scenario, but wasn't able to reproduce. FWIW, if you look at the history of https://github.com/numenta/nupic/pull/2007, the failing Travis builds were related to the same failure (the python interpreter crashed). The crashes stopped and it all started working when I switched to relative imports.
Thanks for trying it out!
Upgrading to 0.5.4 did the trick. Thanks!
Firstly, I'm on a mac.
I have two schemas. One imports and uses the other. When I import both schemas in Python, I get an error related to "Duplicate IDs", and in some cases, it crashes the python interpreter.
Importing individually is fine:
Importing them together raises a
KjException
related to duplicate IDs:Then, if I change the import in
delta.capnp
to be relative, I get a segfault:For good measure, just to verify that I didn’t duplicate the ID:
It’s as if there’s a problem importing
AdaptiveScalarEncoderProto
a second time. Once implicitly in thedelta.capnp
schema and again explicitly in python.I ran into a problem because a test I'm working on that imports both schemas was crashing my Python interpreter. For example, when I run https://github.com/numenta/nupic/commit/e8ca7b52060fa6c0e82fa30fb3e08efc4ad39d86:
With the crash reporter generating this message: