Bug#1016919: lintian: bash-term-in-posix-shell false positive for `. "$(dirname "$0")/functions.sh"`

[Russ Allbery <rra@debian.org>]

Diederik de Haas <didi.debian@cknow.org> writes:

> While inspecting lintian's output wrt (several) kernel builds I did, I
> found the following warning, which seems to be a false positive?

> I: linux-perf: bash-term-in-posix-shell '. $(dirname $0)/../lib/coresight.sh' [usr/lib/perf-core/tests/shell/coresight/asm_pure_loop.sh:8]

Yeah, I think it's a false positive in this check (check_line in
Lintian::Check::Shell::NonPosix::BashCentric):

    # since this test is ugly, I have to do it by itself
    # detect source (.) trying to pass args to the command it runs
    # The first expression weeds out '. "foo bar"'
    if (
        $line !~  m{\A \s*\.\s+
                                   (?:\"[^\"]+\"|\'[^\']+\')\s*
                                   (?:[\&\|<;]|\d?>|\Z)}xsm
        && $line =~ /^\s*(\.\s+[^\s;\`:]+\s+([^\s;]+))/
    ) {

        my ($dot_command, $extra) = ($1, $2);

        push(@matches, $dot_command)
          if length $dot_command
          && $extra !~ m{^ & | [|] | < | \d? > }x;
    }

It's picking up the space in $(dirname $0) and thinking that's a space in
the . command line and therefore arguments are being passed to the sourced
file, which is a bashism.  The missing logic is understanding that a $()
block is effectively a quoted string and thus should be ignored for the
purposes of this test.

-- 
Russ Allbery (rra@debian.org)              <https://www.eyrie.org/~eagle/>