From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Message-ID: Subject: Re: [OE-core] [RFC PATCH 14/14] layer.conf: Extend recipes not to install without explict dependencies From: "Richard Purdie" Date: Mon, 18 Oct 2021 15:29:34 +0100 In-Reply-To: References: <20210920124621.1576702-1-richard.purdie@linuxfoundation.org> <16A68880435BB472.28512@lists.openembedded.org> <0700251c-57a1-cc4c-dd94-a253032a02d1@gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 8bit List-id: To: Martin Jansa , Andreas =?ISO-8859-1?Q?M=FCller?= Cc: Khem Raj , Joshua Watt , Patches and discussions about the oe-core layer On Mon, 2021-10-18 at 16:12 +0200, Martin Jansa wrote: > On Mon, Oct 18, 2021 at 1:50 AM Andreas Müller > wrote: > > On Fri, Oct 1, 2021 at 4:17 PM Martin Jansa wrote: > > > > > > FWIW: I've fixed some bigger layers where pkgconfig was causing quite a > > > few > > build failures, > > > > > Had an off-oe time due to  heavy work load. > > Think this one is the reason for the pkgconfig patch flood. You are > > aware that there are packages not failing at build time for missing > > pkgconfig but build output is different and bugs will pop up at > > runtime. > > > > To avoid responsible maintainers have to go through EVERY recipe (and > > combination of PACKAGECONFIGs) and check if builds are still as > > expected - Can't believe it! > > > > > Not sure why this seems targeted to me. > > Yes I've tried to fix whatever issues this change caused in my world builds > and I'm partially responsible for pkgconfig patch flood, because I wanted > layers like meta-ros/*, meta-qt5, meta-qt6, meta-webosose/* to stay compatible > with oe-core whenever this change from RP is merged to master. > > I've also tried to compare buildhistory and image content (for ros-image-world > with ros1-melodic) to see if I can spot more cases of non-fatal build output > changes as in collada-dom and unfortunately there is a lot of noise in > buildhistory (due to many recipes included in this image not really build- > reproducible), so I've used mostly installed-package-sizes.txt file to spot 2 > more recipes where the size was different (but both don't seem to be caused by > pkgconfig-native in the end). > > I did both builds from scratch without sstate, so I would assume that it will > catch all possible cases, but surprisingly just yesterday I've caught another > missing pkgconfig-native in one of meta-ros1-melodic recipes (caught in > kirkstone based build - while before I was testing on top of a bit older > honister with this change cherry-picked just for test - to prepare recipes to > be future-proof for kirkstone). I take responsibility for merging the pkgconfig-native change in behaviour. I did wait until after honister and only merged to master. It does highlight a real world problem with recipe dependencies since if some of these recipes are only indirectly depending upon it, changes in those other recipes could silently break the recipe in a similar way to the gperf breakage we just saw (which thankfully shows an error). I therefore do think this is worth fixing. Cheers, Richard