From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 1297 invoked by alias); 3 Dec 2003 05:05:05 -0000 Mailing-List: contact gdb-patches-help@sources.redhat.com; run by ezmlm Precedence: bulk List-Subscribe: List-Archive: List-Post: List-Help: , Sender: gdb-patches-owner@sources.redhat.com Received: (qmail 1247 invoked from network); 3 Dec 2003 05:05:03 -0000 Received: from unknown (HELO blount.mail.mindspring.net) (207.69.200.226) by sources.redhat.com with SMTP; 3 Dec 2003 05:05:03 -0000 Received: from user-119a90a.biz.mindspring.com ([66.149.36.10] helo=berman.michael-chastain.com) by blount.mail.mindspring.net with esmtp (Exim 3.33 #1) id 1ARP6j-0005hl-00; Tue, 02 Dec 2003 23:59:25 -0500 Received: by berman.michael-chastain.com (Postfix, from userid 502) id 2FFBB4B35B; Tue, 2 Dec 2003 23:59:25 -0500 (EST) To: cagney@gnu.org, drow@mvista.com Subject: Re: [commit] Deprecate remaining STREQ uses Cc: gdb-patches@sources.redhat.com Message-Id: <20031203045925.2FFBB4B35B@berman.michael-chastain.com> Date: Wed, 03 Dec 2003 05:05:00 -0000 From: mec.gnu@mindspring.com (Michael Elizabeth Chastain) X-SW-Source: 2003-12/txt/msg00074.txt.bz2 Andrew Cagney writes: > modify > build > test > submit > review > commit I'd like to go off on a bit of a tangent here. When I submit a patch, I describe how I tested the patch. I'm also in the habit of asking people how they tested their patches when I'm reviewing other people's patches. I'd like this to become a mandatory part of a patch submission, just like a ChangeLog entry is a mandatory part. I'm not advocating for any particular minimum level of testing. If someone says "Testing: I built gdb and it still compiled", that is okay for some patches. Indeed, I submitted a patch like that for dwarfread.c this week, and a maintainer approved it. In fact I am opposed to a high level of testing before committing patches on the mainline. Most patches never cause a problem. As long as it stays that way, it's more efficient for people to commit patches, and then I test them in my next spin, and if there's a regression, I isolate it pretty fast. It's like interrupt processing on a deeply pipelined processor, the processor gets more throughput at the cost of more state to manage when an interrupt does happen. The testing requirement would do two things. First, just because it's there, people would do some minimal amount of testing so that they don't look lame in public. Second, if a problem does occur with the patch, then when somebody else is analyzing the problem, they have some idea of how the patch was tested before it was integrated. (When I was working on the linux kernel, frequently people would submit patches and say "I didn't even compile this yet but here's the idea ...". We handle this with RFC instead, so we don't have this problem.) I dunno if gdb can handle a formal process change to do this. We're all pretty busy these days. But if anybody shares my views, you can do this yourself: say how you tested your patches, and ask people to include this information in a patch when they review it. Michael C