Skip to content

GitLab

Please use the issue tracker only to report bugs (i.e. errors in the library that need to be fixed) or feature requests.

For questions about how to compile, install or use the library, please use instead the mailing list at https://graph-tool.skewed.de/mailing
(If unsure, use the mailing list first.)



When opening new issues, please choose the BUG template from the drop-down menu, and fill out the required information.

Closed
Created by Tim Day@timday

minimize_blockmodel_dl crashes with parallel=True (64bit Debian jessie, python3 and 2)

I was messing around with graph_tool.community.minimize_blockmodel_dl ... works great, until I add a ,parallel=True to the call parameters, at which point I start getting a

*** Error in `python3': double free or corruption (!prev): 0x0000000004ff9090 ***
Aborted

I'm on Debian Jessie (64bit) using the packages from http://downloads.skewed.de/apt/jessie. Machine is a quad core i7 (8 if you count hyperthreading). I've tried both python2 and python3 versions; both have the same issue.

Attached file imagenet.py reproduces the issue standalone (only dependencies besides graph_tool are scipy and PIL I think). I admit my application is a little perverse... an attempt to see what happens when this algorithm is applied to graphs of image pixels and properties; for the curious, the output when run serially (90 seconds on my old i7) is:

mix-final