opencog / atomspace-cog Goto Github PK

Really annoying pseudo-problem: unit tests, after passing, sometimes fault during cog-server shutdown, typically with the error message corrupted size vs. prev_size but sometimes others.

The root cause of this is that different shared-lib dtors are racing with each other; one of them is the dtor for the cogutils logger. If it waits too long, it ends up trying to free memory after malloc() and free() have been destroyed. Thus the corrupted size error.

This is super-annoying. I spent a day trying to find a solution for this but could not. Placing sleeps in various places did not work. The bug is not really here, its in the cog-utils Logger but it manifests here and a few other places, sometimes. I don't know what to do ....

Am seeing anomalous performance of MultiDeleteUTest -- it runs 10x slower than SimpleMultiDeleteUTest. In principle, it should never run slower, it should always run faster; its the same unit test, targeting a different communications channel. (On my machine, its 48 seconds vs 470 seconds)