CODEWITHSUNDEEP
rbashconfigure
pachadotdev
pachadotdev

Reputation: 3775

avoid bashism in configure

I am creating a configure file for an R package. So far, the R/C++ parts of the code are ok, but when I run devtools::check() to check for my code integrity/quality, I get a note (i.e., a soft warning) about my configure file that says possible bashism in configure line XX ('command' with option other than -p)

This is the part of configure that R doesn't like:

# Extra checks on MacOS for SSL support in libpq
# command -v is probably fine: https://stackoverflow.com/a/677212/946850
if [ `uname` = "Darwin" ] && [ `command -v pkg-config` ]; then
  if pkg-config --atleast-version=12 libpq; then
    case "`pkg-config --libs --static libpq`" in
    *crypto*)
      echo "Local libpq has SSL support"
      ;;
    *)
      echo "Local libpq does not have SSL support"
      FORCE_AUTOBREW=1
      ;;
    esac
  else
    FORCE_AUTOBREW=1
  fi
fi

Should I ignore that or should I write that part in a general way? If the right way is the 2nd way, which would be a good practice to write these kind of configurations?

Sorry if this is a noob question. I am a statistician trying to learn something that is not what I readily available in R/SPSS.

Upvotes: 1

Views: 87

Answers (2)

pachadotdev
pachadotdev

Reputation: 3775

what works with devtools::check() and returns 0 notes is

if [ "`uname`" = Darwin ] && [ "`command -v pkg-config`" ]; then

instead of

if [ `uname` = "Darwin" ] && [ `command -v pkg-config` ]; then

this solution is an "average" of all the comments in this thread

Upvotes: 2

Charles Duffy
Charles Duffy

Reputation: 295679

The check is either wrong, or checking for compatibility with pre-POSIX Bourne shell; the POSIX.2 standard requires command -v to work and has since at least 1994 if not the initial publication in 1992; there's no genuine bashism here.

Upvotes: 3

Related Questions