Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Date: Wed, 31 Aug 2011 20:22:44 +0200
From: magnum <>
Subject: Ancient MPI bugs squashed

Bu pure coincidence I found a bug or two in incremental MPI code. It 
took me several hours to understand what happened. All the time since 
Ryan Lim's original patch from 2003, it has erroneously skipped a number 
of "blocks" of candidates at the very beginning (depending on the number 
of MPI nodes). These blocks are relatively small, so the real impact has 
been very minor from what I can tell.

However, the bug (or another bug, but from same line of code) also made 
resuming skip blocks in a similar way - and when resuming, these blocks 
can be huge compared to the initial ones.

I don't fully understand inc.c but I believe my 0006 patch addresses 
both problems. I had test cases that triggered the problem and "showed" 
what happened, and after my fix everything behaves as expected. I 
haven't really verified resuming though. That fix was just from dry running.


Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.