-
-
Notifications
You must be signed in to change notification settings - Fork 65
Description
The Debian package of pyemd fails to build on the 32-bit powerpc port because the tests time out after 150 minutes of inactivity, while on other platforms including ppc64el and ppc64 they complete in under half a second. I logged into the powerpc porterbox perotto.debian.net, executed a build and noticed that the Python process running the tests uses 100% CPU. I deleted tests one at a time until the remaining tests were passing. Only three tests were hanging with 100% CPU: test_emd_3 test_emd_with_flow_3 test_emd_with_flow_4. The three hanging tests seem to have one thing in common, they all have a distance_matrix containing only zeros. One other test that passes has this feature too test_emd_with_flow_5 though. The three hanging tests also seem to have similar signature parameters. When running Python in gdb, interrupting the process gives a stack trace that indicates the compute_shortest_path function from min_cost_flow.hpp is using the CPU.
Any ideas on how to debug this?