Indent script messes up permissions.

[bangerth]

#16419 results in mayhem on my system because it changes the permissions of the source directory to 644 -- which is bad for a directory, because you can't access that directory any more and everything that's running inside it seriously freaks out.

@drwells FYI

[peterrum]

@bangerth Could you check #16428? Thanks!

[bangerth]

The issue, at its core, rests here:

process_changed()
{
  LAST_MERGE_COMMIT="$(git log --format="%H" --merges --max-count=1 master)"
  COMMON_ANCESTOR_WITH_MASTER="$(git merge-base "${LAST_MERGE_COMMIT}" HEAD)"

  case "${OSTYPE}" in
    darwin*)
      XARGS="xargs -E"
      ;;
    *)
      XARGS="xargs --no-run-if-empty -d"
      ;;
  esac

  ( git ls-files --others --exclude-standard -- ${1};
    git diff --name-only $COMMON_ANCESTOR_WITH_MASTER -- ${1} )|      # **1
      sort -u |
      xargs -n 1 ls -d 2>/dev/null |
      grep -E "^${2}$" |
      ${XARGS} '\n' -P 10 -I {} bash -c "${3} {}"
}

If no file is changed, but you call the indent script anyway, the command at # **1 returns an empty list, which sort leaves as an empty list, but then xargs returns just . (the current directory) because echo | xargs -n 1 ls -d 2>/dev/null returns .. We then execute the chmod on this directory.

[bangerth]

We really shouldn't be passing "*" as second argument of process_changed.

[drwells]

Whoops. I definitely missed that line in the original PR. My mistake!