Rcpp 0.10.1 breaks RcppEigen (Fwd: [Rcpp-devel] Rcpp 0.10.1 and RcppEigen)

[Jiqiang Guo]

From the Rcpp mailing list, Rcpp 0.10.1 breaks RcppEigen, which means new users might not be able to install RStan, especially for 

Linux users if Rcpp 0.10.1 is used. 

For now, RStan seems to be fine with the new version of Rcpp if RcppEigen is installed. 

--

Jiqiang 

---------- Forwarded message ----------
From: Dirk Eddelbuettel <e...@debian.org>
Date: Tue, Nov 27, 2012 at 5:19 PM
Subject: Re: [Rcpp-devel] Rcpp 0.10.1 and RcppEigen
To: Kasper Daniel Hansen <kasperdan...@gmail.com>
Cc: rcpp-...@lists.r-forge.r-project.org

On 27 November 2012 at 16:54, Kasper Daniel Hansen wrote:
| Rcpp 0.10.1 was released yesterday (?).  Installing it breaks

Wrapped up yesterday evening Chicago time, arrived on CRAN this morning,
CRANberries chronicles this well.  I'll blog the announcement this evening...

| RcppEigen and RcppArmadillo (same error message).  Reinstalling the
| packages fixes RcppArmadillo, but not RcppEigen.

Yes. A few things:

 a) binary interfaces changed; modules package are borked; other packages
    most likely too; so what you have installed you may need to reinstall

 b) as we learned from Kurt, we broked RcppEigen and (the less widely used)
    RcppOctave, maybe more; we are looking into this

 c) on the bright side, cda builds again :) and there are a lots of new
    features and fixes

[Bob Carpenter]

As I've been saying for a while now, I
would VERY much like to get rid of the
dependency on RcppEigen and just include our
own copy of the Eigen libs in RStan.

My original reasoning was abstract -- the more
moving parts we can eliminate, the more stable our
software will be. I was thinking something like this
was inevitable, though I was expecting it to come
from incompatible Eigen versions, not Rcpp breaking
their own dependent module!

Shouldn't be a problem for us to include our own
copy of Eigen if we're not on CRAN.

I think the route to getting to CRAN will be the
one Ben laid out, where we do what JAGS and BUGS
and require an external install of Stan. I'd
still recommend users to use your all-in-one
installer, but if we want something on CRAN, this
looks like the best route.

Also, though I know it sounds radical at this point,
I'm also thinking it might be simpler in the long
run to NOT have us run inside the R process, but to
just call an external install of Stan (like R2WinBUGs
did). My original motivation for wanting to do it all
in C++ in a single process was so that we could have
a C++-based representation of the data and samples
and not have ot use R's storage -- that plan went out
the window when we couldn't serialize the state of
a C++ object easily through Rcpp.

- Bob

> --
>
>