First bug report
Bill Hart
available on any AMD machines, apparently.
Isn't that nuts!
I will now download the source.
Bill.
William Stein
> The binary was built on an intel machine with ssse3.
Yep, sage.math.
> This is not
> available on any AMD machines, apparently.
>
> Isn't that nuts!
>
> I will now download the source.
Good idea.
>
> Bill.
--
William Stein
Professor of Mathematics
University of Washington
http://wstein.org
Bill Hart
I don't think any "secrets" were contained in the copied text.
Anyhow, now to the correct group:
.....believe it or not, the source failed to build. Linbox failed
because it could not find a recent GMP.
Ubuntu Jaunty does not have apt-get install atlas, so I just copied
the atlas and cblas libs from a recent sage build into /usr/lib and
similarly moved the includes over to usr/include (and
usr/local/include for good measure). Given that linbox is the culprit
here, maybe I didn't quite satisfy the dependencies with my hack and
it is trying to do something tricksy.
Investigating....
Bill.
William Stein
> Damn. That was *dumb*. Posted to the wrong group!! Arggh. Fortunately
> I don't think any "secrets" were contained in the copied text.
>
> Anyhow, now to the correct group:
>
> .....believe it or not, the source failed to build. Linbox failed
> because it could not find a recent GMP.
>
> Ubuntu Jaunty does not have apt-get install atlas,
Ubuntu *definitely* has development libraries for ATLAS.
Try
apt-cache search atlas |more
etc. to find out what they are called on your system.
> so I just copied
> the atlas and cblas libs from a recent sage build into /usr/lib and
> similarly moved the includes over to usr/include (and
> usr/local/include for good measure). Given that linbox is the culprit
> here, maybe I didn't quite satisfy the dependencies with my hack and
> it is trying to do something tricksy.
It is surprising that you are having any linbox issues, given that I
changed nothing from sage-4.5.3 wrt Linbox, except telling it to look
in /usr/ for ATLAS. Bummer.
>
> Investigating....
Thanks.
Bill Hart
--with-gmp=$SAGE_LOCAL when doing configure maybe?
Bill.
William Stein
> So I guess it is also looking for gmp there too now.
Evidently.
> Need to set
> --with-gmp=$SAGE_LOCAL when doing configure maybe?
Probably.
I'd say "open a ticket", but I haven't setup a trac server yet.
William
Bill Hart
version is 4.2.1, yet it complains that the version is not >3.1.1.
Something dodgy here. Don't see what exactly.
Bill.
Bill Hart
> On Tue, Aug 31, 2010 at 5:03 PM, Bill Hart <goodwi...@googlemail.com> wrote:
>> So I guess it is also looking for gmp there too now.
>
> Evidently.
>
>> Need to set
>> --with-gmp=$SAGE_LOCAL when doing configure maybe?
>
> Probably.
>
> I'd say "open a ticket", but I haven't setup a trac server yet.
Screw that. I'll just fix it.
Bill Hart
directory too!
This simply shouldn't happen. Could you have used a different autotools somehow?
Bill.
William Stein
> It can't find givaro or ntl either. $SAGE_LOCAL is set to the correct
> directory too!
>
> This simply shouldn't happen. Could you have used a different autotools somehow?
No.
Can you build Sage on that computer?
All I did was remove some spkg's, set *one* variable so that on Linux
Linbox using
a system-wide ATLAS, change the README.txt file, and make some changes to two
files in the core Sage library. That's it.
I have built tested on some machines with success. I'll post an
Opteron binary soon, if you want,
since I have a build going on redhawk.math.washington.edu, which is an
Ubuntu 10.04.1 LTS box (Linbox built fine on there too for me, by the
way).
-- William
Bill Hart
> On Tue, Aug 31, 2010 at 5:21 PM, Bill Hart <goodwi...@googlemail.com> wrote:
>> It can't find givaro or ntl either. $SAGE_LOCAL is set to the correct
>> directory too!
>>
>> This simply shouldn't happen. Could you have used a different autotools somehow?
>
> No.
>
> Can you build Sage on that computer?
Yes. No problems, ever.
>
> All I did was remove some spkg's, set *one* variable so that on Linux
> Linbox using
> a system-wide ATLAS, change the README.txt file, and make some changes to two
> files in the core Sage library. That's it.
How bizarre. In the linbox spkg, even ./configure
--with-gmp=/home/wbhart/psage-10.08.31/local/ complains that it cannot
find gmp. Yet it is there! The issue for gmp goes away if I let it
just find the system gmp. But then it complains about givaro and ntl,
again even with an explicit directory given!!
I have absolutely no idea how this can happen. This is pure autotools,
nothing more. It has nothing to do with Sage or psage or anything.
Giving up for this evening. There simply isn't anything left to try.
I know sometimes one has to specify lib and include directories
separately due to some autobugs. But that does not work in this case.
>
> I have built tested on some machines with success. I'll post an
> Opteron binary soon, if you want,
> since I have a build going on redhawk.math.washington.edu, which is an
> Ubuntu 10.04.1 LTS box (Linbox built fine on there too for me, by the
> way).
I know. This is just too bizarre for words. I built sage on this
machine before and 4.5.2 built fine on there. I even built sage on my
new 48 core Mangy Cours, no problems at all.
Bill Hart
version of GLIBCXX. I think this is a screwed up c++ compiler. I seem
to recall John mentioning he switched some things in his path to make
this go away.
I'll ask him when I see him if I can't fix it myself. Probably has
nothing to do with Sage.
Bill.
William Stein
--
Bill Hart
gcc 4.3.3 miscompiles flint, so some time ago I updated gcc on this
machine to 4.4.1 which does a slightly better job with flint (likely
doesn't affect Sage as I don't think Sage uses that part of flint
yet). Anyhow, Ubuntu doesn't play nice with the upgraded gcc. John
gets around this by switching his path to pick up the old gcc 4.3.3 to
build Sage. If I do the same, it finds gmp, ntl and givaro. So this
has nothing to do with gmp, ntl, givaro or linbox. It's an issue with
Ubuntu and certain versions of gcc.
It's funny how the one package that got touched (linbox) just
coincidentally happened to be the point of failure, but had nothing
whatsoever to do with the changes.
psage now says it is missing cblas, but this is probably because I
missed a header file somewhere when installing it. This problem should
much easier to solve.
Bill.
William Stein
> Just for completeness I'll record what the issue was.
>
> gcc 4.3.3 miscompiles flint, so some time ago I updated gcc on this
> machine to 4.4.1 which does a slightly better job with flint (likely
> doesn't affect Sage as I don't think Sage uses that part of flint
> yet). Anyhow, Ubuntu doesn't play nice with the upgraded gcc. John
> gets around this by switching his path to pick up the old gcc 4.3.3 to
> build Sage. If I do the same, it finds gmp, ntl and givaro. So this
> has nothing to do with gmp, ntl, givaro or linbox. It's an issue with
> Ubuntu and certain versions of gcc.
>
> It's funny how the one package that got touched (linbox) just
> coincidentally happened to be the point of failure, but had nothing
> whatsoever to do with the changes.
>
> psage now says it is missing cblas, but this is probably because I
> missed a header file somewhere when installing it. This problem should
> much easier to solve.
Look at the hack I did to spkg/install too. You might have to set
that variable differently...
Bill Hart
environment variable LINBOX_BLAS=/usr/lib/licblas.so, (that file does
exist at least). However, it seems to expect -lcblas -latlas instead,
which works, if I set the environment variable manually within the
sage shell.
Soooo, if I edit spkg/install so that it has "-lcblas -latlas" instead
of /usr/lib/licblas.so, it seems to build linbox. Why this actually
works, who knows. But it works.
Bill.
Bill Hart
Reflecting on why "-lcblas -latlas" worked in place of a directory, I
can't rule out that because this "directory" doesn't exist, it looks
in the default system place and finds what it is looking for. So I'm
not actually sure this is a correct fix. Of course I did not check
that it is not correct either. Someone else may know.
Bill.
Mike Hansen
> The build completed without further issues, as you might expect.
>
> Reflecting on why "-lcblas -latlas" worked in place of a directory, I
> can't rule out that because this "directory" doesn't exist, it looks
> in the default system place and finds what it is looking for. So I'm
> not actually sure this is a correct fix. Of course I did not check
> that it is not correct either. Someone else may know.
I've been also getting a similar failure with linbox not being able to
detect the BLAS library on Cygwin.
--Mike
Bill Hart
testlong. But many seem to be because of missing scipy, etc.
Bill.
William Stein
> There are screeds of test failures of course, when I run sage
> testlong. But many seem to be because of missing scipy, etc.
>
Yes. I know that many tests fail. I haven't decided how to address
this. Maybe i can make a patch marking those tests optional, and
apply it for each psage release.
> Bill.
>
> On 1 September 2010 06:00, Mike Hansen <mha...@gmail.com> wrote:
>> On Tue, Aug 31, 2010 at 9:37 PM, Bill Hart <goodwi...@googlemail.com> wrote:
>>> The build completed without further issues, as you might expect.
>>>
>>> Reflecting on why "-lcblas -latlas" worked in place of a directory, I
>>> can't rule out that because this "directory" doesn't exist, it looks
>>> in the default system place and finds what it is looking for. So I'm
>>> not actually sure this is a correct fix. Of course I did not check
>>> that it is not correct either. Someone else may know.
>>
>> I've been also getting a similar failure with linbox not being able to
>> detect the BLAS library on Cygwin.
>>
>> --Mike
>>
>
--