[Rd] R.INSTALL on Windows

[Herve Pages]

Hi,

I'd like to point out a potential problem with the current R.INSTALL
directory created by 'R CMD INSTALL' when applied to a source package.

On Windows (and, AFAIK, only on Windows), 'R CMD INSTALL mypackage_1.0.0.tar.gz'
creates a temporary R.INSTALL directory in the current directory. This
directory is removed before the command terminates but only if it was
successful.

One inconvenient of this behaviour is that you can get a clash when you try to
build 2 Windows binary packages (with 'R CMD INSTALL --build') at the same time
in the same directory. I can circumvent the 00LOCK mechanism by providing
a different --library option to the 2 commands but, if I'm running the 2 commands
in the same directory, then they will be creating and using the same R.INSTALL.
Then, the first command who terminates will remove it and break the other command.

Is there an easy way to prevent this?

Is there any reason why R.INSTALL is not kept at the end of 'R CMD INSTALL',
or why 'R CMD INSTALL' doesn't use a package specific directory name
(e.g. R.INSTALL.mypackage) like 'R CMD check' does, or why it doesn't
use one of the "standard" temp places (defined by env. var. TMP or TMPDIR).

Thanks in advance!

Cheers,
H.