Date: Wed, 18 Sep 2019 09:58:38 -0500 From: Kyle Evans <kevans@freebsd.org> To: Enji Cooper <yaneurabeya@gmail.com> Cc: src-committers <src-committers@freebsd.org>, svn-src-all <svn-src-all@freebsd.org>, svn-src-head <svn-src-head@freebsd.org> Subject: Re: svn commit: r352465 - head/share/mk Message-ID: <CACNAnaHCO5mTpPFpODTAYwAR1cPo0F29%2BDWxiF0vLTEZDz3yDA@mail.gmail.com> In-Reply-To: <81382CF5-A928-48EF-93A9-BBBBA174F4BD@gmail.com> References: <201909180158.x8I1wuZu011258@repo.freebsd.org> <0FBC9A62-AE3B-4F27-AABC-06FF45F415F1@gmail.com> <CACNAnaG23b92BjmAMD_Pn1PW4gcYEne6vdaJ9SuOxHU1hbmyvg@mail.gmail.com> <CDB9D8B3-4100-4C52-B414-D6C60FE49846@gmail.com> <81382CF5-A928-48EF-93A9-BBBBA174F4BD@gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, Sep 18, 2019 at 9:46 AM Enji Cooper <yaneurabeya@gmail.com> wrote: > > > > On Sep 18, 2019, at 07:33, Enji Cooper <yaneurabeya@gmail.com> wrote: > > > > > >>> On Sep 18, 2019, at 05:40, Kyle Evans <kevans@freebsd.org> wrote: > >>> > >>> On Wed, Sep 18, 2019 at 7:34 AM Enji Cooper <yaneurabeya@gmail.com> w= rote: > >>> > >>> > >>>> On Sep 17, 2019, at 18:58, Kyle Evans <kevans@freebsd.org> wrote: > >>>> > >>>> Author: kevans > >>>> Date: Wed Sep 18 01:58:56 2019 > >>>> New Revision: 352465 > >>>> URL: https://svnweb.freebsd.org/changeset/base/352465 > >>>> > >>>> Log: > >>>> googletest: default-disable on all of MIPS for now > >>>> > >>>> Parts of the fusefs tests trigger a bug in current versions of llvm:= IR > >>>> representation of some routine for the MIPS targets is a function wi= th a > >>>> large number of arguments. This then leads the compiler on an hour+ = long > >>>> goose chase, which is OK if you build the current tree but less-so i= f you're > >>>> trying external toolchain or doing a universe build involving mips w= hen it > >>>> eventually gets switched over to LLVM. > >>>> > >>>> Better, accurate details can be found in LLVM PR43263. > >>> > >>> Uhhhhh... why not do this in tests/sys/... instead? > >> > >> Because there's still value in being able to easily enable these for > >> building/running the complete set of tests through standard build > >> infrastructure, but it's not worth adding a knob specifically for the > >> fusefs tests. I also prefer the communication of it being an > >> off-by-default option and easily deduced from src.conf(5) that this > >> part of the build is default-disabled on mips/mips. > > > > Let me rephrase things a bit: is googlemock broken for all of mips, or = is it just the tests? If the latter, the tests should be blacklisted for mi= ps with a justification. If the former, I agree your method of dealing with= the situation is ok, but more investigation needs to be done to see whethe= r or not the port (in general) is broken and mark it broken if need be. > > It looks like the latter case, based on the PR, and it=E2=80=99s a build = performance issue... Is this impacting CI pipelines? > It is the latter, and I do not want to *blacklist* them because as far as I can tell, the tests aren't necessarily broken. I want to workaround them for default by now. > > The problem with src.opts.mk=E2=80=99s per-architecture options, is tha= t it can be very heavy handed enabling/disabling features. I=E2=80=99m not = sure that everything in there warrants disabling at that level. > > My investigation suggests that the course of action was overly heavy hand= ed. While I=E2=80=99m not asking for a revert, it would be really nice if w= hole features weren=E2=80=99t disabled, unless there=E2=80=99s an issue wit= h the feature. > We do not have a lighter method for dealing with this that I can tell, because as I said above: I do not want to blacklist them or completely kill them off. I still want the option to build and test them, but as I aim to switch mips over to llvm I do not want to subject CI and the rest of the world to an extra 1.5+ hour build time for this during tinderbox runs. Given that it's mips, so already tier-high, and I'm one of few people that care about it (and I only care about it for the time being), I intend to leave it as-is since it's still a default in the rest of the world.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CACNAnaHCO5mTpPFpODTAYwAR1cPo0F29%2BDWxiF0vLTEZDz3yDA>