Avoid using /usr/bin/tar on Debian machines

[kevinushey]

This is a follow-up from rstudio/renv#1794.

For the builds of R published on https://docs.posit.co/resources/install-r, it looks like R is configured to use system installations of tar and sed as so:

$ cat Renviron | grep 'SED\|TAR'
SED=${SED-'/usr/bin/sed'}
TAR=${TAR-'/usr/bin/tar'}

This can be problematic in rare scenarios, as /usr/bin/sed and /usr/bin/tar are not actually provided by any apt packages. So, if those programs are not available for some reason, it's not clear for the user on how to recover.

Some other similar instances:

$ cat Renviron | grep /usr/bin/
R_PRINTCMD=${R_PRINTCMD-'/usr/bin/lpr'}
R_TEXI2DVICMD=${R_TEXI2DVICMD-${TEXI2DVI-'/usr/bin/texi2dvi'}}
R_GZIPCMD=${R_GZIPCMD-'/usr/bin/gzip'}
R_UNZIPCMD=${R_UNZIPCMD-'/usr/bin/unzip'}
R_ZIPCMD=${R_ZIPCMD-'/usr/bin/zip'}
R_BZIPCMD=${R_BZIPCMD-'/usr/bin/bzip2'}
PAGER=${PAGER-'/usr/bin/pager'}
R_PDFVIEWER=${R_PDFVIEWER-'/usr/bin/xdg-open'}
SED=${SED-'/usr/bin/sed'}
TAR=${TAR-'/usr/bin/tar'}

In other words, for each of these programs, if a variant exists in /bin, I think that version should be preferred.

[glin]

I bet this is the same issue we once got with /bin/sed vs. /usr/bin/sed before, #72. On recent Ubuntu/Debian, /bin was merged into /usr/bin so with stock docker images like the one we're using, tar is actually at /usr/bin/tar with /bin/tar being a symlink:

docker run -it --rm ubuntu:focal

$ ls -la /bin
lrwxrwxrwx 1 root root 7 Aug 26  2022 /bin -> usr/bin

$ readlink -f /bin/tar
/usr/bin/tar

This causes issues for unmerged systems (if you've upgraded from an older OS perhaps?) and other things unfortunately: https://lwn.net/Articles/890219/

So for newer Ubuntu and Debian builds, we could override the TAR location to /bin/tar. I believe the Ubuntu builds already manually set some bin locations like ZIPCMD but not TAR. But also, I wonder if this could also cause issues if /bin is supposedly intended to be eliminated eventually.

If we want to set TAR and prevent issues from unmerged /usr systems, there are some other variables that the distro default r-base sets as well for reference: https://salsa.debian.org/edd/r-base/-/merge_requests/1

[gaborcsardi]

Yeah, it is the upgrades that cause issues, the symlinks are missing on upgraded systems.

[kevinushey]

Thanks; I wasn't aware of that! For what it's worth:

$ dpkg -L tar | grep /bin/tar
/bin/tar

$ dpkg -L sed | grep /bin/sed
/bin/sed

which I think gives some (small) amount of motivation to prefer /bin/tar over /usr/bin/tar.

[gaborcsardi]

We had this in rig some time ago, for sed (r-lib/rig#119 (comment)), apparently it also applies to tar.
It would be nice to fix it properly.

[mabuimo]

I bet this is the same issue we once got with /bin/sed vs. /usr/bin/sed before, #72.

Indeed, I also needed to manually create symlinks for that directory (Ubuntu 20.04 LTS).
ln -s /bin/tar /usr/bin/tar