Re: Compiling 2.1.71 on FC4

[John MacLean]

Shai Ayal wrote:
> I tried using gfortran as follows and it did not pass the tests:
> F77=gfortran ./configure --enable-shared --disable-static
> this compiles using gcc,g++ & gfortran which are all version 4.0.1
> It hangs on trying to do:
> cond ([1, 2; 2, 1])
> from the octave.test/linalg/cond-1.m test.
>
> The version I did manage to compile and is working is:
> CC=gcc32 CXX=g++32 ./configure --enable-shared --disable-static
> that is, using the 3.2 versions of gcc,g++ and g77

and Quentin Spencer responded

> It appears you're able to compile, but running into problems with make
> check? I have to admit I've never run make check before. And, I think
> Dmitri may be right: make sure you have an up-to-date version of blas
> and lapack installed. I don't remember any problems with the FC1
> version, but there have been with some others (FC3 in particular) which
> could have caused the problem you're seeing. On the other hand, I seem
> to recall from your original mail that you were compiling with ATLAS, so
> maybe this is not the problem. (By the way, I have an ATLAS package
> pending review for Fedora Extras--I'll send something to the list when
> it gets in).

For the past few hours I've also been trying to compile 2.1.71 on FC4
and I've encountered the same problems as reported by Shai in this
thread. I'm glad that this thread was active or I might have been
stuck for quite awhile.

I am using a vanilla installation of FC4 but I had the same problem
compiling (undefined references to _cxa_guard_*) as Shai until I
removed the g77 and let the configuration point to gfortran as Quentin
suggested.

I then had the same problem with the cond test as Shai.  I didn't have
the older 3.2 compiler versions on my computer so I didn't attempt
Shai's fix.

The cond test was hanging in a call to SVD.  Several other functions
such as EIG would also hang the first time they were called in Octave
(hanging somewhere under dlamch). If they were interrupted with ctrl-c
then they would work without hanging on subsequent attempts.

I didn't have any previous blas/lapack libraries installed so I guess
I was building the libraries distributed with 2.1.71 and my default
optimization flags were incompatible with FC4 as noted in some earlier
threads.  Rather than trying to fix my flags I got the libraries from
Fedora Extra as Dmitri and Quentin suggested and this fixed the
problem.

(to link the libraries some additional symbolic links had to be
created after getting them from Fedora Extra via yum)

Now all of the check tests pass except for the coth-1.m test. This
problem was noted by Dmitri in the "Make Check Fails in Recent CVS"
thread posted in June 2005 and he opened the Bugzilla Bug 160759. 
Apparently it has been fixed in glibc-2.3.90-2.  I'm not brave enough
to update glibc so I'll try to stay away from hyperbolic trig
functions of imaginary numbers.

I've been in the habit of building Octave from source so that I can
use hdf5. Perhaps I don't need to do this anymore? - I haven't checked
the Fedora Extra version of Octave to see if it was linked with hdf5.


Regards,
John MacLean



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------