From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753345AbaIPI5f (ORCPT ); Tue, 16 Sep 2014 04:57:35 -0400 Received: from mail-we0-f172.google.com ([74.125.82.172]:54239 "EHLO mail-we0-f172.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752970AbaIPI5c (ORCPT ); Tue, 16 Sep 2014 04:57:32 -0400 Date: Tue, 16 Sep 2014 10:57:27 +0200 From: Ingo Molnar To: Mathias Krause Cc: Andrew Morton , Thomas Gleixner , Ingo Molnar , "H. Peter Anvin" , Joe Perches , Rasmus Villemoes , "linux-kernel@vger.kernel.org" , Linus Torvalds Subject: Re: [PATCHv3 0/9] Mark literal strings in __init / __exit code Message-ID: <20140916085727.GA4907@gmail.com> References: <1408623792-7973-1-git-send-email-minipli@googlemail.com> <20140821125717.GA28987@gmail.com> <20140822082459.GA8771@gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.23 (2014-03-12) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org * Mathias Krause wrote: > > Regardless of how it's implemented on the tooling side, my > > point remains: this kind of optimization is done on the > > tooling side in a natural fashion, while it's an ongoing > > maintenance concern on the kernel side. > > The costs of making the required changes to the code, i.e. > changing printk() / pr_*() to pi_*() / pe_*(), are a necessary > pain but are a one-time cost, as Joe already said. [...] That argument is bogus - the costs form increased complexity are ongoing for all new code affected by such constructs, and they are an ongoing cost for all changes to the code as well. > > So it should be done on the tooling side, especially as the > > benefits appear to be marginal. > > But still, they are measurable. [...] So is the cost of complexity measurable: we already got rid of __init annotation variants, and we want to keep it simple and maintainable, not litter the code with new variants again, only to be warned about in build time checks that few developers run. And when it comes to weighing increased complexity against some marginal benefit, usually the simpler approach is preferred, especially since it could all be solved via tooling. Sure, you have to implement the tooling support for that, and have to wait for that to trickle through to actual build environments - but in turn that would benefit a lot more projects than the kernel alone. If you are impatient you could do tooling in the kernel as well, in tools/ for example. Thanks, Ingo