From: Benny Halevy Subject: Re: [PATCH 2/5] nfsd: Fix independence of a few nfsd related headers Date: Thu, 22 Oct 2009 12:14:13 +0200 Message-ID: <4AE03075.7000401@panasas.com> References: <4ADEC1EF.8040107@panasas.com> <1256112873-32495-1-git-send-email-bharrosh@panasas.com> <1256171298.6809.1.camel@heimdal.trondhjem.org> <4AE01569.9000002@panasas.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Cc: "J. Bruce Fields" , pNFS Mailing List , NFS list , Andy Adamson To: Boaz Harrosh , Trond Myklebust Return-path: Received: from dip-colo-pa.panasas.com ([67.152.220.67]:55589 "EHLO daytona.int.panasas.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1752449AbZJVKOL (ORCPT ); Thu, 22 Oct 2009 06:14:11 -0400 In-Reply-To: <4AE01569.9000002@panasas.com> Sender: linux-nfs-owner@vger.kernel.org List-ID: On Oct. 22, 2009, 10:18 +0200, Boaz Harrosh wrote: > On 10/22/2009 02:28 AM, Trond Myklebust wrote: >> On Wed, 2009-10-21 at 10:14 +0200, Boaz Harrosh wrote: >>> An header should be compilation independent, .i.e pull in >>> any header who's declarations are directly used by this header. >>> And not let users re-include all it's dependencies all over >>> again. >>> >>> [At the end of the day what's the use of a header if it does >>> not have more then one user?] >> >> The problem with this is that you quickly end up including the same >> header files over and over and over again as they get pulled in several >> times over by different headers. >> >> >> > > What is the problem with that? > > two things: > 1. it is unavoidable. > 2. That's why we invented the #ifndef FOO_H #define FOO_H for. > 3. Look at the current mess, to the point that you don't understand why > the code does not compile, you end up just copy-past the include list > of that other file, and now actually do end with extra un-needed includes. > (Don't believe me look at the last patch as proof). > 4. So I add to an header a use of a type that now needs a new include. > All my users do not compile any more. What to do? OK I'll include it so > not to change all existing users all over again. Now we get a double > standard. All headers used before any users, must be carried around. > The late comers are escaped. > 5. The opposite of 4. An header is no longer needed. Extra header left at all > users. > > It used to be a problem before me and you have begun programing. > Since then the cpp looks at it's internal structures and will not re-open. > Note that the compiler never sees the second instance, ever. For the fun of it, I took an average of 5 incremental build times before and after this patchset, average user and system times below: before: 68.84 12.51 after: 68.74 12.48 This shows a minor improvement of about 0.14% for user time and 0.27% for system time. Benny > > That said we have no choice of the matter. It is a Kernel style guide. I > should know because I was banged on the head with it a couple of times. > And rightly so. > > Come on that was a joke right > Boaz