From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeff King Subject: Re: profile-fast is failing in my Git 2.2.1 build from tar in a Git repo Date: Mon, 12 Jan 2015 16:16:05 -0500 Message-ID: <20150112211604.GB10409@peff.net> References: <1420408723.7095.207.camel@homebase> <20150112210827.GA10409@peff.net> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Cc: paul@mad-scientist.net, git@vger.kernel.org To: Junio C Hamano X-From: git-owner@vger.kernel.org Mon Jan 12 22:16:13 2015 Return-path: Envelope-to: gcvg-git-2@plane.gmane.org Received: from vger.kernel.org ([209.132.180.67]) by plane.gmane.org with esmtp (Exim 4.69) (envelope-from ) id 1YAmLM-00075n-9v for gcvg-git-2@plane.gmane.org; Mon, 12 Jan 2015 22:16:12 +0100 Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751286AbbALVQH (ORCPT ); Mon, 12 Jan 2015 16:16:07 -0500 Received: from cloud.peff.net ([50.56.180.127]:33439 "HELO cloud.peff.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with SMTP id S1750725AbbALVQG (ORCPT ); Mon, 12 Jan 2015 16:16:06 -0500 Received: (qmail 7303 invoked by uid 102); 12 Jan 2015 21:16:05 -0000 Received: from Unknown (HELO peff.net) (10.0.1.1) by cloud.peff.net (qpsmtpd/0.84) with SMTP; Mon, 12 Jan 2015 15:16:05 -0600 Received: (qmail 20532 invoked by uid 107); 12 Jan 2015 21:16:26 -0000 Received: from sigill.intra.peff.net (HELO sigill.intra.peff.net) (10.0.0.7) by peff.net (qpsmtpd/0.84) with SMTP; Mon, 12 Jan 2015 16:16:26 -0500 Received: by sigill.intra.peff.net (sSMTP sendmail emulation); Mon, 12 Jan 2015 16:16:05 -0500 Content-Disposition: inline In-Reply-To: <20150112210827.GA10409@peff.net> Sender: git-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: git@vger.kernel.org Archived-At: On Mon, Jan 12, 2015 at 04:08:28PM -0500, Jeff King wrote: > > I do not think profile build is prepared to be run without having > > our history (after all, it is not test_perf_create_REPO_from, not > > test_perf_create_source_directory_of_git_from). It wants to create > > a repository that hosts a reasonably sized but not too big a > > project, i.e. us. > > > > The safest and cleanest fix would be to make sure that the said > > function checks if $source/.git is a repository, perhaps? > > Hmph. We already dealt with this once in 93b5393 (Makefile: make perf > tests optional for profile build, 2014-08-19). That check explicitly > looks for ".git" (and not checking that we are in an outer git repo), so > I am not sure why it is kicking in at all. But probably the fix should > go there. Oh, I see. It is because Paul is using profile-fast, which does not have that check. And I did not add it as part of 93b5393, because it would not make sense. If you do not do the perf tests, then profile-fast is literally doing nothing. :) I think the right solution is either: 1. Switch to using "make profile", which will use the regular test suite to gather data, and skip the perf tests. 2. Set GIT_PERF_REPO to some representative repo of your choice for gathering profiling data (it does not have to be git.git at all, but it is probably more useful if it is not a tiny toy repo). We _could_ teach the perf code to look upwards for a surrounding repo with `git rev-parse` rather than looking for ../../.git. But there is no guarantee that the repo we find would actually be a useful one for doing perf tests. I'd just as soon leave it as-is, and have people specify a useful repo via GIT_PERF_REPO if they have one. -Peff