Re: configure error on blas shell variable test

[Rik]

On 11/23/2013 08:17 AM, Mike Miller wrote:
> Rik,
>
> I get a shell syntax error when running configure on a system without a
> BLAS library, instead of a proper error message exit. The error is due
> to lines like
>
>   if test $ax_blas_f77_func_ok = no; then
>
> where the variable may not actually have a value. The error looks like
>
>   ../configure: line 32752: test: =: unary operator expected
>   ../configure: line 34261: test: =: unary operator expected
>   ../configure: line 35815: test: =: unary operator expected
>
> Your changeset ad2c3902b826 got rid of the quotes in that test, which is
> fine for most cases but apparently can still fail if the BLAS library
> isn't even present.
>
> Which style do you prefer for consistency, restoring the quotes or
> setting that variable to "no" before calling AX_BLAS_WITH_F77_FUNC? The
> macro doesn't seem to care whether it already has a value.
>
11/23/13

Mike,

What a bother!  I dealt with the same "unary operator expected" error for
cross-compiling platforms in this changeset 437e51a0c25d.

I think the problem is that we shouldn't be trying to look deeply into the
ax_blas_f77 macro and break data encapsulation.  Instead, we should be
using the output of the macro which is ax_blas_ok.  I added a changeset
which does that (http://hg.savannah.gnu.org/hgweb/octave/rev/4465699450c2).

--Rik