[R-pkg-team] Debian bug report 982465: dyn.load not useful for system libraries

[Dirk Eddelbuettel]

Indeed: "Should" is the operational term here.

But just pointing at packages for which I am upstream author / maintainer and
Debian maintainer, we already use the other approach for RcppArmadillo,
RcppEigen, RcppCCTZ, RcppDate, ...  and we fudged it for BH with the empty
shell r-cran-bh which may well blow in our face BH is now at 1.75.0 which
Debian doesn't yet have.

The R manual "Writing R Extensions" even mentions it explicitly and
recognised the value of header packages. For _CRAN_ and R it is _better_ to
standardize at headers at one level _across all deployments_.

As a distro we have slightly different objectives and we cannot hit _all_
goals _all the time_.  What we have right now is suboptimal for Debian ("in
theory we could do better") but practically speaking I quite strongly feel
that we do is the right way to do it.

(And I am quite frankly more mad, or I should say, "disappointed" at the
_CRAN_ level where e.g. the CCTZ headers are now in (last I looked) three
more packages (including lubridate, famously), the new package clock does the
same to the Date headers in RcppDate and on and on. Some people are very keen
to reinvent and slap "their" logo on things, and there is no stopping them.)

All that said, I am course all for experimentation and trying to make it
better.  But what we found here is that the current attempt at trying it for
RcppParallel did not exactly work out fully.  So if it were me, I would
backtrack and package "as is" because to me the primary directive is to get
current and working software to our users. But we are Debian and maintainers
can do as they please. Which in my case means saying no to a bug report
suggesting I alter R itself for a Debian-local-and-only hack.  As I said, if
it were of general help and use (and I may misjudge it, convince me) I will
_of course_ help and carry it to R Core for vetting and inclusion
upstream. But as far as I can tell that is not the situation here.  So please
leave the r-base package out of all micro-fights you are in for RcppParallel.

Dirk

-- 
https://dirk.eddelbuettel.com | @eddelbuettel | edd at debian.org