ericspod / eidolon Goto Github PK

Sometimes when executing compiled code using multiprocessing a system error occurs like this on Ubtun 14.04 (and possibly other Linux distros/versions) with Anaconda Python 2.7:

Inconsistency detected by ld.so: dl-open.c: 231: dl_open_worker: Assertion `_dl_debug_initialize (0, args->nsid)->r_state == RT_CONSISTENT' failed!

Initially this appeared to be a problem with MKL libraries used by Anaconda's numpy, but the issue appeared in code where numpy wasn't being used and after MKL was removed. The issue then appeared to be when compiling Cython code against Anaconda rather than compiling with the system Python. The problem is very intermittent and difficult to reproduce, especially in isolation from the rest of Eidolon, but is definitely concurrency related.

It now appears to be a bug in dlopen() itself which has since been apparently fixed in later versions of glibc.

It would be great to have a color bar or legend for mesh fields!

Hi Eric,

Thank you for the great software, I am very excited to be using it for my PhD project!

I have noticed an issue in the latest release (v0.5.0) when trying to generate a mesh object, which I believe is due to the change in how Python 3 handles the zip() function and division / operator.

The first problem I ran into was in Line 1429 of the SegmentPlugin.py script:

contours=zip(*segobj.enumContours())[0]

Here I received the following error:
TypeError: 'zip' object is not subscriptable

After reading up about it, it turns out that in Python 2, the zip() function returns a list, while Python 3 returns an iterable object, as described in this thread.

After implementing the solutions described here, the next error occurred in Line 460:

invertring=[midring[(clen/2+i)%clen] for i in range(clen)]

returning the error:
TypeError: list indices must be integers or slices, not float

Again, there has been a change in how Python 3 handles the / operator as discussed here.

I changed the single, true division operator / which returns a float, to a floor division // ,which returns an integer.

After doing this, I was able to generate the mesh object from the segmentation contours.

EDIT: I see now that you have implemented this as well in the experimental branch, but thought logging this would be useful for anyone trying to use this feature.

Cheers and all the best!

Marcell