Re: init.sh compare function

[Bruno Haible]

[dropping bug-grep from CC]

> > --- a/tests/init.sh
> > +++ b/tests/init.sh
> > @@ -221,11 +221,35 @@ export MALLOC_PERTURB_
> >  # a partition, or to undo any other global state changes.
> >  cleanup_ () { :; }
> > 
> > +# Arrange not to let diff or cmp operate on /dev/null,
> > +# since on some systems (at least OSF/1 5.1), that doesn't work.

I like the idea, but as a person who often looks at failing tests and
interprets the output I have two comments:

1) I would find it useful if, despite recognizing /dev/null as a special
   case, the output format would stay the same or nearly the same.
   When I see a line "+foo bar" or "-foo bar", possibly preceded by a diff
   hunk, I know that the program produced or did not produce a line
   containing "foo bar". Whereas when I see a line "foo bar" I think of
   output that went to stdout of stderr and which should have been piped
   away.

2) I would also find it useful to mention in comments that compare
   function is meant to be called as in
      compare expected_output actual_output
   and not the opposite.

   Why? Because a "+" indicates something that was added, whereas "-"
   indicates something that was removed or missing. Most often the
   program's actual output is wrong, not the expected output.
   If you call
      compare actual_output expected_output
   then extraneous lines come out as "-line" and omitted ones come out as
   "+line", which is counter-intuitive.
   Whereas if you call
      compare expected_output actual_output
   then extraneous lines come out as "+line" and omitted ones come out as
   "-line".

   Yes I know it would take some work to revisit all coreutils and grep
   tests that use 'compare', but that is not an urgent task.

> > +# When one argument is /dev/null and the other is not empty,
> > +# cat the nonempty file to stderr and return 1.
> > +# Otherwise, return 0.
> > +compare_dev_null_ ()
> > +{
> > +  test $# = 2 || return 2
> > +
> > +  if test "x$1" = x/dev/null; then
> > +    set dummy "$2" "$1"; shift
> > +  fi
> > +
> > +  test "x$2" = x/dev/null || return 2
> > +
> > +  test -s "$1" || return 0
> > +
> > +  cat - "$1" <<EOF >&2
> > +Unexpected contents of $1:
> > +EOF

So here I would emit a fake hunk header and then either
   sed 's/^/+/' "$1"
or
   sed 's/^/-/' "$1"

Bruno
-- 
In memoriam Ricardo Flores Magón 
<http://en.wikipedia.org/wiki/Ricardo_Flores_Magón>