On Thu, 18 Aug 2005, Norman Gray wrote:
> On 2005 Aug 18 , at 13.59, Peter W. Draper wrote:
>
> > just an update on what I've done with these tests. I've added them all
> > to the MAIN branch and documented them in autoconf.texi and SSN/078.
> > We should probably let these get some use before committing to the
> > development branch (and I should be doing other things).
>
> I've updated the version of SSN/078 at <http://www.astro.gla.ac.uk/
> users/norman/star/ssn78/>. I noticed that the subsections of the
> relevant appendix had got out of alphabetical order (including a
> couple that have been out for quite a long time), so adjusted those.
>
> Would it be possible to put these on the development branch as well?
> I'd like to have a complete set of what we propose to the mainline
> there, so I can tweak the documentation and regression tests before
> doing a merge from there back to the HEAD for our testing.
Hi Norman,
OK, I've done that. All my tests should now be in the development branch.
> Is this regenerated as part of the nightly build? Could it be thus
> regenerated?
Clearly a question for Steve.
> > One problem is with the AC_FC_HAVE_TYPELESS_BOZ test.
>
> By the way, the documentation for the AC_FC_HAVE_BOZ test implies that
> it checks for X'xxx', thus, if it is true, then AC_FC_HAVE_TYPELESS_BOZ
> will also necessarily be true. This is right, isn't it? If so, it
> might be worth spelling this out, to reassure anyone reading it that
> there isn't some subtlety they're missing.
Your right, the X'xxx' form shouldn't be checked in the basic F95 test, so
I've removed it and corrected the comments.
> > This needs to run a program, rather than just see if the compiler is
> > happy, as it turns out that you really need to test the actual values
> > created by the BOZ assignment (gfortran is the culprit, that does the
> > assignments, but they are wrong, in fact being just integer casts).
> >
> > Anyway that means this test must return a status code to the
> > environment, so it makes a non-standard call to EXIT(1), must be a
> > problem for your patches (and the idea in general of having runnable
> > Fortran tests). Any comments on this?
>
> I think this is probably OK. If anyone upstream makes a fuss about
> it, then we could do a check for the EXIT function first. Or else
> communicate the result back by writing to a file conftest.txt or
> something.
So we may have to autoconf the autoconf tests. Stands to reason!
Cheers,
Peter.
|