From mboxrd@z Thu Jan 1 00:00:00 1970 From: "H. Peter Anvin" Subject: cg-update (cogito-0.13) bug: changed heads Date: Tue, 06 Sep 2005 14:01:25 -0700 Message-ID: <431E03A5.6060709@zytor.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit X-From: git-owner@vger.kernel.org Tue Sep 06 23:03:42 2005 Return-path: Received: from vger.kernel.org ([209.132.176.167]) by ciao.gmane.org with esmtp (Exim 4.43) id 1ECkZi-0001AD-7N for gcvg-git@gmane.org; Tue, 06 Sep 2005 23:01:50 +0200 Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1750935AbVIFVBr (ORCPT ); Tue, 6 Sep 2005 17:01:47 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1750937AbVIFVBr (ORCPT ); Tue, 6 Sep 2005 17:01:47 -0400 Received: from terminus.zytor.com ([209.128.68.124]:38554 "EHLO terminus.zytor.com") by vger.kernel.org with ESMTP id S1750935AbVIFVBr (ORCPT ); Tue, 6 Sep 2005 17:01:47 -0400 Received: from [10.4.1.13] (yardgnome.orionmulti.com [209.128.68.65]) (authenticated bits=0) by terminus.zytor.com (8.13.1/8.13.1) with ESMTP id j86L1UMc019647 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NO); Tue, 6 Sep 2005 14:01:31 -0700 User-Agent: Mozilla Thunderbird 1.0.6-1.1.fc4 (X11/20050720) X-Accept-Language: en-us, en To: Git Mailing List , Petr Baudis X-Virus-Scanned: ClamAV version 0.86.2, clamav-milter version 0.86 on localhost X-Virus-Status: Clean X-Spam-Status: No, score=-5.8 required=5.0 tests=ALL_TRUSTED,AWL,BAYES_00 autolearn=ham version=3.0.4 X-Spam-Checker-Version: SpamAssassin 3.0.4 (2005-06-05) on terminus.zytor.com Sender: git-owner@vger.kernel.org Precedence: bulk X-Mailing-List: git@vger.kernel.org Archived-At: I just found a bug in cogito-0.13, at least to the best of my understanding of the situation. Simplified, the problem looks like this: two sides, A and B, pull from repository C. The head of this pull we call 0. Now commit 1 is added to A. B pulls commit 1 from A, not though C. Later, A pushes commit 1 to C. Next time B tries to pull from C, cg-update will not advance refs/heads/origin, even though C now has a new head. Thus, B will think that C is in a different state than it actually is. This has, for me, broken safeguards of the form "make sure you're merged before you pull." -hpa