Date: Wed, 9 Oct 2019 05:33:44 +0000 From: Mark Linimon <linimon@lonesome.com> To: Warner Losh <imp@bsdimp.com> Cc: Ed Maste <emaste@freebsd.org>, "freebsd-arch@freebsd.org" <freebsd-arch@freebsd.org> Subject: Re: Gcc 4.2.1 to be removed before FreeBSD 13, a firm timeline Message-ID: <20191009053342.GA21030@lonesome.com> In-Reply-To: <CANCZdfpYpwoR6QbVS5hUwzJGUS7An3cv-=a8t=TLTPJM6zpotQ@mail.gmail.com> References: <CANCZdfrZ-C37c7Eso7NrtMG5_kv3T9qO3c3k8xcgEeTCrY3nQg@mail.gmail.com> <CAPyFy2C=3NLz8Gwinf06oxsgDvVEHVChd24QVXg-iAf8DeaBBQ@mail.gmail.com> <CANCZdfpYpwoR6QbVS5hUwzJGUS7An3cv-=a8t=TLTPJM6zpotQ@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Oct 08, 2019 at 12:01:52PM -0600, Warner Losh wrote: > I'd held off a little thinking that llvm 9.0 will have landed by now, but > it hasn't due to exp-run issues. Rather than stall until it's in the tree, > I'm planning on committing the review by the end of the week, assuming > testing is all green. The following is not an argument for or against the timeline, just some more info. tl:dr; yes it's a long message, the gist of which is "we're not quiiiite ready on powerpc64." If people don't have some time to sit down and go through all of the gory details, I won't blame them ... The exp-run test Warner cites above covers and only covers "test all port builds against llvm9.0 in base, on amd64". What it does *not* cover is "test all port builds against llvm.<any> in base, on powerpc64". I have been leading the charge on the latter, on the machine IBM has loaned to us via OSU, "ppcdevref". This is the testbed for "powerpc64 with clang as the base compiler." Right now its base compiler is 8.0.1. (This is due to a: we have not gotten around to 9.x yet, b: we are still only ~95% of the way to identifying *just* the regressions in that configuration.) As of today, the early-adopter developers running powerpc64/llvm9x are battling errors that are (for purposes of brevity) the union of the ppcdevref errors and the above exp-run. (Actually, there are some disjoint entries -- let me elide that for now.) So the list of regressions as of *today* on ppcdevref has just been re-uploaded to: https://people.freebsd.org/~linimon/poudriere/blacklist.powerpc64.ppcdevref Let me note, I am updating this file *rapidly*, sometimes more than once per day. It is full of sharp edges and snarky comments :-) Anyone who wants to poke at the above data, please contact me off-list or join us on #powerpc64 on EFNet, rather than replying here. But, of particular concern are: databases/mariadb55-client databases/mysql55-client databases/percona55-client devel/kyra graphics/drm-legacy-kmod lang/rust net/samba410 www/node* (I have patches for these) x11-toolkits/qt5-declarative And, I'm sorry, I used to have a set of poudriere results uploaded to www.lonesome.com that corresponed to the above, so that you could view the #blocked. But that VM instance has died and I don't have the cycles to fix it right now. But in brief, the kyra, rust, and samba failures are problematic. So, that's the list of things that will break immediately as soon as the pylon.nyi.FreeBSD.org package builder is switched over to gcc-less. Also, anyone who is on powerpc64-CURRENT-less-than-that-commit will have to immediately update, if they want to use the new packages. There is no way to mix-and-match. The problem I am facing is that "people keep moving my cheese". There have been infrastructure changees, a default ports compiler change, changes that affect big projects like kde, and even a few sweeping changes that I have not yet fully accounted for -- and that's all within the last month. This week I am not as much "fighting my way up the hill" as I am "crawling up the hill on hands and knees while pulling arrows out of my back." So, my pace of testing and fixing has slowed down. Also, there are some patches (especially linker) that I have not yet tested. Finally, yes, I know, I've been told more than once that "tier-2 considerations cannot affect what is committed to -CURRENT". So, I get that. But, if we switched today, there would still be a pain-point, and that's the point of all the above text. fwiw. mcl
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20191009053342.GA21030>