rmatrix: ABI change in Matrix 1.6-2

[Graham Inggs]

From: https://lists.debian.org/debian-r/2023/11/msg00033.html

Package Matrix is having a new and energetic maintainer/contributor in Mikael
Jagan who is tidying up a few loose corners (and inter alia sent me a patch
to RcppEigen that resulted in a coordinated CRAN update of RcppEigen, lme4,
and OpenMx).

Mikael also identified two sets of packages needed a rebuild in messages to
the r-package-devel list (the more-or-less official place in the R Project to
ask / discuss package changes, it is a decent to be on) following private
mails between him and me. See

https://stat.ethz.ch/pipermail/r-package-devel/2023q4/010051.html
https://stat.ethz.ch/pipermail/r-package-devel/2023q4/010054.html

The first concerns packages using a LinkingTo: Matrix and building against
Matrix _headers_. The second concerns caching of S4 signatures (which bit us
at work because of SeuratObject [not in Debian] and how I got onto this).

Most of these are not in Debian but I think we need binary rebuilds of

   irlba because of headers
   OpenMx because of headers, a new upstream 2.21.10 is out too
   TMB because of headers
   MatrixModels because of S4 caching

I would appreciate it if someone could tickle rebuilds. To me a quick
informal touch of debian/changelog would do; if someone thinks this needs a
formal transition go for it.

The R Core team and the CRAN maintainers are aware of the implicit problem
with signalling the need for binary rebuilds. They are discussing this, but
do not have an answer. Historically, CRAN has informally rebuilt its binaries
for windows and macOS, but that of course does not help binary distributors
such as us, other Linux distros, Conda, r2u, ... at all.