From steve at sumost.ca  Sat Dec  5 13:16:58 2015
From: steve at sumost.ca (Steve M. Robbins)
Date: Sat, 05 Dec 2015 12:16:58 -0600
Subject: [MINC-development] Bug#806243: libminc: FTBFS on mipsel
In-Reply-To: <A4A4352BED09044DB124E2C0D810CE9590981F4E@BADAG02.ba.imgtec.org>
References: <A4A4352BED09044DB124E2C0D810CE9590981F4E@BADAG02.ba.imgtec.org>
Message-ID: <6776976.Cm5crvf0hg@riemann>

Hi,

[Adding minc-development as follow-up to previous email; 
http://www.bic.mni.mcgill.ca/pipermail/minc-development/2015-November/001223.html ]


On December 4, 2015 08:25:09 PM Jurica Stanojkovic wrote:
>  User: debian-mips at lists.debian.org
> 
> Hello,
> 
> I have tried to build package libminc_2.3.00-2 for mipsel and mips64el on
> diferent build boards: netlogic-xlp, cavium and loongson-3a. This package
> FTBFS only on loongson-3a (lemote).
> On other build machines package was built from source successfully.

That's interesting.  I have been able to reproduce the failure on the Debian's  
mipsel machine 'etler' which is longsoon.  I know Bert was trying to bring up 
a mipsel machine to test; not sure how far he got.  (Incidentally, Bert: I was 
working with the 'develop' branch as of Nov 19 and the failure persists.)

Jurica: is everything else the same between your good/bad environments, 
specifically: libc, compiler, and libnetcdf?  I mean: I assume they are both 
running 'sid', but are the specific versions of each package the same?

If all that is the same -- could it be something in the hardware?  The failing 
code is basically arithmetic mapping a double-value into another type, e.g. 
integer (of different sizes).  The failing test is for the mapping of 
-2*FLT_MAX, which should map to the low end of the output range but instead 
maps to the high end.  Is there anything different in the hardware that would 
maybe round things differently?  

Thanks,
-Steve
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: This is a digitally signed message part.
URL: <http://www.bic.mni.mcgill.ca/pipermail/minc-development/attachments/20151205/88478414/attachment.pgp>

From robert.d.vincent at mcgill.ca  Sat Dec  5 13:41:03 2015
From: robert.d.vincent at mcgill.ca (Robert D. Vincent)
Date: Sat, 5 Dec 2015 13:41:03 -0500
Subject: [MINC-development] Bug#806243: libminc: FTBFS on mipsel
In-Reply-To: <6776976.Cm5crvf0hg@riemann>
References: <A4A4352BED09044DB124E2C0D810CE9590981F4E@BADAG02.ba.imgtec.org>
	<6776976.Cm5crvf0hg@riemann>
Message-ID: <CAAd5DeS1wqXMdefmCT34R5CqgT_GQkY1=ZZrOUXLsyWz+hAAdw@mail.gmail.com>

Steve et al,

I used an emulated mipsel environment using some other distributions (found
here: https://people.debian.org/~aurel32/qemu/mipsel/) - the bug was not
reproducible. I am in the process of building a QEMU environment for a more
recent release ("sid" from 20151023), I will try that as soon as possible.

    -bert

On Sat, Dec 5, 2015 at 1:16 PM, Steve M. Robbins <steve at sumost.ca> wrote:

> Hi,
>
> [Adding minc-development as follow-up to previous email;
>
> http://www.bic.mni.mcgill.ca/pipermail/minc-development/2015-November/001223.html
> ]
>
>
> On December 4, 2015 08:25:09 PM Jurica Stanojkovic wrote:
> >  User: debian-mips at lists.debian.org
> >
> > Hello,
> >
> > I have tried to build package libminc_2.3.00-2 for mipsel and mips64el on
> > diferent build boards: netlogic-xlp, cavium and loongson-3a. This package
> > FTBFS only on loongson-3a (lemote).
> > On other build machines package was built from source successfully.
>
> That's interesting.  I have been able to reproduce the failure on the
> Debian's
> mipsel machine 'etler' which is longsoon.  I know Bert was trying to bring
> up
> a mipsel machine to test; not sure how far he got.  (Incidentally, Bert: I
> was
> working with the 'develop' branch as of Nov 19 and the failure persists.)
>
> Jurica: is everything else the same between your good/bad environments,
> specifically: libc, compiler, and libnetcdf?  I mean: I assume they are
> both
> running 'sid', but are the specific versions of each package the same?
>
> If all that is the same -- could it be something in the hardware?  The
> failing
> code is basically arithmetic mapping a double-value into another type, e.g.
> integer (of different sizes).  The failing test is for the mapping of
> -2*FLT_MAX, which should map to the low end of the output range but instead
> maps to the high end.  Is there anything different in the hardware that
> would
> maybe round things differently?
>
> Thanks,
> -Steve
>
> _______________________________________________
> MINC-development mailing list
> MINC-development at bic.mni.mcgill.ca
> http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-development
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.bic.mni.mcgill.ca/pipermail/minc-development/attachments/20151205/54e663a4/attachment.html>

From steve at sumost.ca  Mon Dec  7 12:14:10 2015
From: steve at sumost.ca (Steve Robbins)
Date: Mon, 07 Dec 2015 11:14:10 -0600
Subject: [MINC-development] Bug#806243: libminc: FTBFS on mipsel
In-Reply-To: <A4A4352BED09044DB124E2C0D810CE9590981FEE@BADAG02.ba.imgtec.org>
References: <A4A4352BED09044DB124E2C0D810CE9590981F4E@BADAG02.ba.imgtec.org>,
	<6776976.Cm5crvf0hg@riemann>
	<A4A4352BED09044DB124E2C0D810CE9590981FEE@BADAG02.ba.imgtec.org>
Message-ID: <5D8EEE7E-FBBC-4EC0-A801-0CEA291D2F9C@sumost.ca>

Thanks Jurica.  Is there any difference in fpu?  Soft vs. Hard?  Extra precision bits?

On December 7, 2015 9:53:05 AM CST, Jurica Stanojkovic <Jurica.Stanojkovic at imgtec.com> wrote:
>> Jurica: is everything else the same between your good/bad
>environments, 
>> specifically: libc, compiler, and libnetcdf?  I mean: I assume they
>are both 
>> running 'sid', but are the specific versions of each package the
>same?
>
>Yes, i am using sbuild (sid) and environment should be the same.
>libc, compiler and libnetcdf are the same (same versions). 
>I have checked all build logs.
>
>Kernels are not the same:
>3.6.11+ on loongson
>3.7.10 on netlogic-xlp
>4.1.0 on cavium
>but i do not suggest that kernel is issue here.

-- 
Sent from my Android device with K-9 Mail. Please excuse my brevity.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.bic.mni.mcgill.ca/pipermail/minc-development/attachments/20151207/18515afd/attachment.html>

From Jurica.Stanojkovic at imgtec.com  Mon Dec  7 10:53:05 2015
From: Jurica.Stanojkovic at imgtec.com (Jurica Stanojkovic)
Date: Mon, 7 Dec 2015 15:53:05 +0000
Subject: [MINC-development] Bug#806243: libminc: FTBFS on mipsel
In-Reply-To: <6776976.Cm5crvf0hg@riemann>
References: <A4A4352BED09044DB124E2C0D810CE9590981F4E@BADAG02.ba.imgtec.org>,
	<6776976.Cm5crvf0hg@riemann>
Message-ID: <A4A4352BED09044DB124E2C0D810CE9590981FEE@BADAG02.ba.imgtec.org>

> Jurica: is everything else the same between your good/bad environments, 
> specifically: libc, compiler, and libnetcdf?  I mean: I assume they are both 
> running 'sid', but are the specific versions of each package the same?

Yes, i am using sbuild (sid) and environment should be the same.
libc, compiler and libnetcdf are the same (same versions). 
I have checked all build logs.

Kernels are not the same:
3.6.11+ on loongson
3.7.10 on netlogic-xlp
4.1.0 on cavium
but i do not suggest that kernel is issue here.