From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from lindbergh.monkeyblade.net ([23.128.96.19]:38850 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-FAIL-OK-FAIL) by vger.kernel.org with ESMTP id S1725848AbgDSWE0 (ORCPT ); Sun, 19 Apr 2020 18:04:26 -0400 Received: from eidolon.nox.tf (eidolon.nox.tf [IPv6:2a07:2ec0:2185::]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id B116AC061A0C for ; Sun, 19 Apr 2020 15:04:25 -0700 (PDT) Date: Sun, 19 Apr 2020 23:55:58 +0200 From: David Lamparter Subject: gcc extended format checking plugin Message-ID: <20200419215558.GU30496@eidolon.nox.tf> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Sender: linux-kbuild-owner@vger.kernel.org List-ID: To: linux-kbuild@vger.kernel.org Hello kernel build hackers, in case it is of interest to anyone, we've recently added an extended format string checking plugin in the FRRouting project. We took inspiration from the Linux kernel and modified a printf (funnily enough, the one from FreeBSD) to support things like %pI4, and we wanted better warnings for this. Since the scheme is the same as in the Linux kernel, it may or may not be worth someone's time to take a look. The plugin is found here: https://github.com/FRRouting/frr/tree/master/tools/gcc-plugins It's derived from gcc's "c-format.c", as such it may make sense to look at the diff instead of the straight source. The original GCC code is in the git history there. It's tested against gcc 9.3.0, YMMV on other versions. Note that the plugin needs a single-line GCC source code modification: https://github.com/FRRouting/frr/blob/master/tools/gcc-plugins/gcc-retain-typeinfo.patch this is to stop GCC from prematurely stripping information, specifically from casts in function parameters. It does work without the change, but it may produce false-positive/negative warnings. Aside from %pXX suffix support, the plugin also contains two other changes that are probably irrelevant to kernel land: - it makes some typedefs "final" types since these are not guaranteed to be consistent in size across platforms and therefore can't be printf'd without a cast (e.g. pid_t, uid_t, time_t, etc.) - it reserves %Lu for uint64_t Please don't expect too much on its code quality, I'm not a GCC developer and essentially wrote this by trial and error ;) Cheers, -David