[Rd] ANN/RFC: package providing TAB completion for readline-based R

[Deepayan Sarkar]

On 10/24/06, Prof Brian Ripley <ripley at stats.ox.ac.uk> wrote:
> On Tue, 24 Oct 2006, Uwe Ligges wrote:
[...]
> >> Questions:
> >>
> >> There are a couple of things about which I would like some advice:
> >>
> >> (1) The package currently contains a very rudimentary configure script
> >> which stops installation when readline is not found. I'm not sure if
> >> this is portable enough. .../src/unix/sys-std.c has more sophisticated
> >> conditional directives, but I don't know enough about this sort of
> >> thing to interpret those. Any thoughts or suggestions on this would be
> >> appreciated.
>
> I don't think that is safe.  You can have R installed with readline
> support even if there are neither readline headers nor library on the
> current machine (use a static library), and R can be run without readline
> (via --no-readline) even if it is installed.

I don't think the last case is a problem; nothing useful will happen
after loading the package in that case, but nothing bad will happen
either. If there's no readline library, is there anything the
configure script can reasonably do other than abort? If the library is
there but the headers aren't, I can probably use headers included in
the package. Is that a reasonable approach? I have no idea what to do
with a static library.

> However, your configure script fails on my machine (x86_64 FC5).

Fails in what sense? Does it abort saying it didn't find readline?

> It seems
> that you need both to get LDFLAGS and LD_LIBRARY_PATH from R and use them,
> and also use a version of autoconf that knows about lib64 (since the
> readline DSO is in /usr/lib64).  Since I don't get any further I am not
> sure what really happens, but it seems you don't link your DLL against
> -lreadline, and I think you need to to allow for static libreadline.a (the
> relevant code seems to be a separate .o and so is likely not linked into R
> in that case).

I have modified configure.in to add a -lreadline to PKG_LIBS, and also
to use header files from within the package if the system headers are
not found. The new version is at

http://www.stat.wisc.edu/~deepayan/R/rcompletion_0.0-7.tar.gz

I'm working on a 64 bit system, so I'm hoping autoconf knows what to
do there. I'm not sure how to get or use LDFLAGS and LD_LIBRARY_PATH
from R. R CMD config LDFLAGS does give me something (but I'm not sure
what to do with it), but there doesn't seem to be anything analogous
to LD_LIBRARY_PATH.

> I guess that you want to know that R was installed with readline support
> at install time, and is running with the readline console?  It is not
> immediately obvious to me that either is currently known to a package
> writer/user.

I don't think that's as much of an issue. If R doesn't ever make a
call to readline(), whether  the rcompletion package is loaded or not
won't make any difference. The user won't get the benefits, but that's
a choice made by the user. For example, running R through ESS and then
loading rcompletion doesn't have any effect at all.

The one scenario that definitely needs to be avoided is where R CMD
INSTALL appears to install the package successfully and then
library(rcompletion) causes an error (either immediately on load or
later). Unless anyone reports seeing that, I'll plan to upload this to
CRAN in a few days.

Cases where configure aborts even though it shouldn't will probably
need some time (and help) to work out, since I'm completely new to
autotools.


> >> (2) What's the recommended procedure to distribute this, given that
> >> (a) it's not relevant on Windows (i.e. won't do anything even if it
> >> could be installed) and (b) may or may not be relevant on Macs, some
> >> of which apparently have a fake readline that doesn't support
> >> completion?
> >
> >
> > Distribution can happen regularly as a CRAN package. I will put it on my
> > exclude list in order not to try building Windows binaries in that case.
>
> I think there should also be a configure.win which explains that the
> package is not usable on Windows, in case someone tries to install from
> the sources.

If configure.win doesn't exist, would it try to use configure? In that
case, installation should just abort when readline isn't found,
shouldn't it? Would having a configure.win serve any additional
purpose?

-Deepayan