From rob at landley.net  Thu Jul 23 03:52:47 2015
From: rob at landley.net (Rob Landley)
Date: Wed, 22 Jul 2015 22:52:47 -0500
Subject: [Aboriginal] Time for another release.
Message-ID: <55B0650F.7020105@landley.net>

Bit past due, really.

I got the sh2eb and sh2elf binflt toolchains working. Userspace isn't
yet because nommu, I tried switching ash out for hush but it still needs
fork(). (CONFIG_MSH is aliased to CONFIG_HUSH, so they yanked the shell
that worked nommu and replaced it with one that didn't.)

I'm debugging why awk is breaking on mips: "awk '{print $2}' blah.txt"
is hanging because the "2" parses as NAN and it loops, which turns out
to be a kernel bug introduced by 46490b572544 which was _not_ fixed by
backporting 620b15503457.

There's also something funky with i686 and x86_64 builds under qemu, but
I can just do those in a chroot for the moment. Not sure I'm holding up
the release for that.

Rob


From dahalls at gmail.com  Fri Jul 24 07:35:43 2015
From: dahalls at gmail.com (David Halls)
Date: Fri, 24 Jul 2015 08:35:43 +0100
Subject: [Aboriginal] Time for another release.
In-Reply-To: <55B0650F.7020105@landley.net>
References: <55B0650F.7020105@landley.net>
Message-ID: <CABX1NxFsP779LTRJepeTfxiOjrz-HBRn5O2SP1i61yawcgpyhA@mail.gmail.com>

Thanks for the update Rob.
What's the issue with x86_64?

On 23 July 2015 at 04:52, Rob Landley <rob at landley.net> wrote:

> Bit past due, really.
>
> I got the sh2eb and sh2elf binflt toolchains working. Userspace isn't
> yet because nommu, I tried switching ash out for hush but it still needs
> fork(). (CONFIG_MSH is aliased to CONFIG_HUSH, so they yanked the shell
> that worked nommu and replaced it with one that didn't.)
>
> I'm debugging why awk is breaking on mips: "awk '{print $2}' blah.txt"
> is hanging because the "2" parses as NAN and it loops, which turns out
> to be a kernel bug introduced by 46490b572544 which was _not_ fixed by
> backporting 620b15503457.
>
> There's also something funky with i686 and x86_64 builds under qemu, but
> I can just do those in a chroot for the moment. Not sure I'm holding up
> the release for that.
>
> Rob
> _______________________________________________
> Aboriginal mailing list
> Aboriginal at lists.landley.net
> http://lists.landley.net/listinfo.cgi/aboriginal-landley.net
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.landley.net/pipermail/aboriginal/attachments/20150724/48f6584b/attachment.htm>

From rob at landley.net  Fri Jul 24 09:04:34 2015
From: rob at landley.net (Rob Landley)
Date: Fri, 24 Jul 2015 04:04:34 -0500
Subject: [Aboriginal] Time for another release.
In-Reply-To: <CABX1NxFsP779LTRJepeTfxiOjrz-HBRn5O2SP1i61yawcgpyhA@mail.gmail.com>
References: <55B0650F.7020105@landley.net>
	<CABX1NxFsP779LTRJepeTfxiOjrz-HBRn5O2SP1i61yawcgpyhA@mail.gmail.com>
Message-ID: <55B1FFA2.3070605@landley.net>

The x86_64 issue is actually two issues.

1) distcc is hanging. I have no idea why. I've tried to trace into it
but distcc such a mess I'm pretty sure it would be easier to write a
replacement (adding distcc functionality to ccwrap is on the todo list)
than fix it.

2) Unfortunately, if I disable distcc then qemu's floating point
emulation isn't quite right, and gets one of those "failure to compare
zero equal to zero" things, specifically:

https://twitter.com/landley/status/571883794279493633

Of course it's deep in the bowels of the perl build. It works fine if
you run it in a chroot, so it's not a userspace issue. And I've reverted
the kernel back _two_years_ and it's still doing it, so it's not a
kernel issue. Pretty sure the x86_64 build was only tested under distcc
for a long time until it stopped working, so the code was built by the
host compiler that resolved constants using the host's FPU which did the
floating point calculation right. (I.E. this never worked in qemu.) So
maybe a qemu upgrade broke this, but not a remotely recent one. (I think
what broke distcc was me upgrading my _host_ from ubuntu 12.04 to 14.04.
Considering I'm building my distcc binaries from the same source in
host-tools.sh and on the target, that means it's probably a host glibc
version upgrade screwing up distccd on the host since host-tools is
built with host toolchain against host libc...)

I could try to isolate the breakage and feed a bug report to the qemu
guys, or I could write my own distcc replacement. But it's so easy to
work around (more/chroot-setup.sh) I haven't bothered yet.

Rob

On 07/24/2015 02:35 AM, David Halls wrote:
> Thanks for the update Rob.
> What's the issue with x86_64?
> 
> On 23 July 2015 at 04:52, Rob Landley <rob at landley.net
> <mailto:rob at landley.net>> wrote:
> 
>     Bit past due, really.
> 
>     I got the sh2eb and sh2elf binflt toolchains working. Userspace isn't
>     yet because nommu, I tried switching ash out for hush but it still needs
>     fork(). (CONFIG_MSH is aliased to CONFIG_HUSH, so they yanked the shell
>     that worked nommu and replaced it with one that didn't.)
> 
>     I'm debugging why awk is breaking on mips: "awk '{print $2}' blah.txt"
>     is hanging because the "2" parses as NAN and it loops, which turns out
>     to be a kernel bug introduced by 46490b572544 which was _not_ fixed by
>     backporting 620b15503457.
> 
>     There's also something funky with i686 and x86_64 builds under qemu, but
>     I can just do those in a chroot for the moment. Not sure I'm holding up
>     the release for that.
> 
>     Rob
>     _______________________________________________
>     Aboriginal mailing list
>     Aboriginal at lists.landley.net <mailto:Aboriginal at lists.landley.net>
>     http://lists.landley.net/listinfo.cgi/aboriginal-landley.net
> 
> 


From dahalls at gmail.com  Sat Jul 25 06:54:23 2015
From: dahalls at gmail.com (David Halls)
Date: Sat, 25 Jul 2015 07:54:23 +0100
Subject: [Aboriginal] Time for another release.
In-Reply-To: <55B1FFA2.3070605@landley.net>
References: <55B0650F.7020105@landley.net>
	<CABX1NxFsP779LTRJepeTfxiOjrz-HBRn5O2SP1i61yawcgpyhA@mail.gmail.com>
	<55B1FFA2.3070605@landley.net>
Message-ID: <CABX1NxG++H_tSauf0odqrtmT-SOCaQVgZcDrd5dfS87jkEEMbw@mail.gmail.com>

>
>
> 1) distcc is hanging. I have no idea why. I've tried to trace into it
> but distcc such a mess I'm pretty sure it would be easier to write a
> replacement (adding distcc functionality to ccwrap is on the todo list)
> than fix it.
>

Yeah I stopped using distcc. I did try debugging it but as you say it's a
mess.


>
> 2) Unfortunately, if I disable distcc then qemu's floating point
> emulation isn't quite right, and gets one of those "failure to compare
> zero equal to zero" things, specifically:
>
>
Not seen this and I've been compiling Aboriginal quite a bit in kvm locally
(I only use chroot on CircleCI).
I'm happy to take a look once you've got something to release if you like.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.landley.net/pipermail/aboriginal/attachments/20150725/8dbb918e/attachment.htm>

From dahalls at gmail.com  Sat Jul 25 06:56:31 2015
From: dahalls at gmail.com (David Halls)
Date: Sat, 25 Jul 2015 07:56:31 +0100
Subject: [Aboriginal] Time for another release.
In-Reply-To: <CABX1NxG++H_tSauf0odqrtmT-SOCaQVgZcDrd5dfS87jkEEMbw@mail.gmail.com>
References: <55B0650F.7020105@landley.net>
	<CABX1NxFsP779LTRJepeTfxiOjrz-HBRn5O2SP1i61yawcgpyhA@mail.gmail.com>
	<55B1FFA2.3070605@landley.net>
	<CABX1NxG++H_tSauf0odqrtmT-SOCaQVgZcDrd5dfS87jkEEMbw@mail.gmail.com>
Message-ID: <CABX1NxGa4fK8VnBtTaY9XMjboi1xjNTrsWP86ZHq66xuJkzpxg@mail.gmail.com>

On 25 July 2015 at 07:54, David Halls <dahalls at gmail.com> wrote:

>
>> 1) distcc is hanging. I have no idea why. I've tried to trace into it
>> but distcc such a mess I'm pretty sure it would be easier to write a
>> replacement (adding distcc functionality to ccwrap is on the todo list)
>> than fix it.
>>
>
> Yeah I stopped using distcc. I did try debugging it but as you say it's a
> mess.
>
>
>>
>> 2) Unfortunately, if I disable distcc then qemu's floating point
>> emulation isn't quite right, and gets one of those "failure to compare
>> zero equal to zero" things, specifically:
>>
>>
> Not seen this and I've been compiling Aboriginal quite a bit in kvm
> locally (I only use chroot on CircleCI).
> I'm happy to take a look once you've got something to release if you like.
>
>
Btw my Perl build instructions are here:
https://github.com/davedoesdev/heddle/blob/master/image_scripts/packages#L273
It doesn't do anything special on x86_64.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.landley.net/pipermail/aboriginal/attachments/20150725/ec998ba9/attachment.htm>

From rob at landley.net  Sun Jul 26 03:25:11 2015
From: rob at landley.net (Rob Landley)
Date: Sat, 25 Jul 2015 22:25:11 -0500
Subject: [Aboriginal] Time for another release.
In-Reply-To: <CABX1NxG++H_tSauf0odqrtmT-SOCaQVgZcDrd5dfS87jkEEMbw@mail.gmail.com>
References: <55B0650F.7020105@landley.net>	<CABX1NxFsP779LTRJepeTfxiOjrz-HBRn5O2SP1i61yawcgpyhA@mail.gmail.com>	<55B1FFA2.3070605@landley.net>
	<CABX1NxG++H_tSauf0odqrtmT-SOCaQVgZcDrd5dfS87jkEEMbw@mail.gmail.com>
Message-ID: <55B45317.6070803@landley.net>

On 07/25/2015 01:54 AM, David Halls wrote:
>     2) Unfortunately, if I disable distcc then qemu's floating point
>     emulation isn't quite right, and gets one of those "failure to compare
>     zero equal to zero" things, specifically:
> 
> 
> Not seen this and I've been compiling Aboriginal quite a bit in kvm
> locally (I only use chroot on CircleCI).
> I'm happy to take a look once you've got something to release if you like.

If you use KVM it should work fine, since that uses the host's FPU. The
problem is that qemu's _emulated_ FPU does a software floating point
thing that doesn't quite match what the hardware does, and you get a
calculation that _should_ be zero but is actually 1E-130 and it doesn't
compare equal with zero and the perl build bombs with unknown version.

It's really a qemu problem, but getting distcc working again would work
around it. As would using KVM instead of qemu. As would doing the build
in a chroot instead of qemu... :)

It's one of those "I really should submit a bug report to qemu" things,
but it's so easy to work around and hacking a reasonable test case out
of the perl build is unpleasant enough that I haven't bothered yet.

Rob



From rob at landley.net  Sun Jul 26 03:29:44 2015
From: rob at landley.net (Rob Landley)
Date: Sat, 25 Jul 2015 22:29:44 -0500
Subject: [Aboriginal] Time for another release.
In-Reply-To: <CABX1NxGa4fK8VnBtTaY9XMjboi1xjNTrsWP86ZHq66xuJkzpxg@mail.gmail.com>
References: <55B0650F.7020105@landley.net>	<CABX1NxFsP779LTRJepeTfxiOjrz-HBRn5O2SP1i61yawcgpyhA@mail.gmail.com>	<55B1FFA2.3070605@landley.net>	<CABX1NxG++H_tSauf0odqrtmT-SOCaQVgZcDrd5dfS87jkEEMbw@mail.gmail.com>
	<CABX1NxGa4fK8VnBtTaY9XMjboi1xjNTrsWP86ZHq66xuJkzpxg@mail.gmail.com>
Message-ID: <55B45428.1030100@landley.net>

On 07/25/2015 01:56 AM, David Halls wrote:
>     Not seen this and I've been compiling Aboriginal quite a bit in kvm
>     locally (I only use chroot on CircleCI).
>     I'm happy to take a look once you've got something to release if you
>     like.
> 
> 
> Btw my Perl build instructions are here:
> https://github.com/davedoesdev/heddle/blob/master/image_scripts/packages#L273
> It doesn't do anything special on x86_

Mine are
http://landley.net/hg/control-images/file/ef471ef37665/images/lfs-bootstrap/mnt/build/perl.sh
and my builds are never architecture-specific (that would be a bug).

Nah, it's a qemu floating point emulation bug. Probably something like
they're doing IEEE and the real hardware isn't, and the expected
behavior is what the hardware does not what the standard says.

Rob