From mboxrd@z Thu Jan 1 00:00:00 1970 From: Sebastien Ponce Subject: Re: New Defects reported by Coverity Scan for ceph (fwd) Date: Wed, 18 Jun 2014 09:37:39 +0200 Message-ID: <1403077059.4397.10.camel@sebmain.cern.ch> References: Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Return-path: Received: from cernmx13.cern.ch ([188.184.36.46]:17475 "EHLO CERNMX13.cern.ch" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S933231AbaFRHii (ORCPT ); Wed, 18 Jun 2014 03:38:38 -0400 In-Reply-To: Sender: ceph-devel-owner@vger.kernel.org List-ID: To: Sage Weil Cc: ceph-devel@vger.kernel.org I've now looked at the defects. And according to me, there are mostly irrelevant. There are 2 cases : - the 2 "division by zero" will never happen because we will never run the test with a zero parameter. Still, I've added an ASSERT to avoid their report. - all the rest is a clash between gtest and coverity : whenever you allocate something via the C API and you have an assert between this allocation and the deallocation, you have a potential resource leak if the assert fails. However, avoiding it means wrapping all the C API in a kind of autoPtr interface. I do not believe it's worth it for test code. By the way, the rados tests must have the same leaks a priori and I suppose they are ignored. So is it ok to commit the 2 new asserts and ignore the other defects ? Cheers, Sebastien On Sat, 2014-06-07 at 09:12 -0700, Sage Weil wrote: > Mostly stuff in teh new libradosstriper code, it looks like.