[Rcpp-devel] RcppEigen not compiling under Solaris 10 INTEL

On Mon, Sep 30, 2013 at 3:56 PM, Dirk Eddelbuettel <edd at debian.org> wrote:

On 30 September 2013 at 16:46, Sul, Young L wrote:
| I know that from
http://cran.fhcrc.org/web/checks/check_results_RcppEigen.html
| RcppEigen isn?t compilable until Solaris 10 yet. I?m wondering if
anyone has a
| sense if this will be fixed in the future?

It's tricky:

  -- packages Rcpp${FOO} fail for various values of ${FOO} also because
Rcpp
     itself may not build (and in this particular case solaris-x86 seems
to
     pass for Rcpp (Yay!) but solaris-sparc still fails (booh) so here it
is
     indeed an RcppEigen issue)

  -- none of the core member of the Rcpp group has access to either
Solaris
     variant (though as I recall Martyn Plummer has a solaris x86 and
     provided Romain access to this in the past)

  -- so we cannot fix Rcpp, nor can we fix packages building on top of
Rcpp

  -- and as "we" don't use these machine, fixes are not of high priority
(but
     if someone could provide funding, Romain may be willing to tackle
this
     but you'd have to ask him)

However, we have always accepted patches for this.

With RcppEigen, the added difficulty is that the listed maintainer is no
longer all that active around R.

Actually the listed maintainer was writing a reply when your reply came
through :-)

I had been willing to bet that no one used RcppEigen on Solaris and its
failing to build on Solaris was a non-issue, but Young Sui has proved me
wrong.

However, the problems Dirk described hold for me too.  I don't have access
to a Solaris system, either x86 or sparse, and don't know of a way that I
could access such a system.  Even if I could get access I doubt that I
would be able to fix the parts complained about in the logs because they
are in the Eigen code   If I recall correctly the Eigen authors said they
were not going to try to support Solaris because - wait for it - they don't
have access to a Solaris system.

I recall a suggestion of using openSolaris under virtualbox or some other
virtualization software but, according to the Wikipedia page, openSolaris
no longer exists.

So that is an open spot -- if someone ambitious enough to tackle this would

step forward and adopt RcppEigen, we'd all appreciate it.

Yes, I would be willing to pass RcppEigen onto another maintainer.
However It happens that there will be a rather complicated shuffle
involving the Matrix, RcppEigen and lme4 packages in the near future and
probably I should be involved in that.

The situation is that a recent release (and a minor version number at
that) of the CHOLMOD software, on which the sparse Cholesky factorization
in Matrix and in RcppEigen is based, added a field to the
cholmod_factor_struct.  If we use the new CHOLMOD version in Matrix, which
we would like to do, then the installed versions of other packages will
start to fail when the new Matrix is installed.  It is going to be messy
because there is no way to force an upgrade of a downstream package when a
new Matrix is installed.