Re: [PATCH] test defs: add subroutine for input unindenting

[Stefano Lattarini]

Hello Ralf.

On Tuesday 15 February 2011, Ralf Wildenhues wrote:
> * Stefano Lattarini wrote on Tue, Feb 15, 2011 at 01:36:16PM CET:
> > OK, here's my shot at it.  The implementation might be suboptimal,
> > but since it wasn't completely obvious to get right, I'd rather not
> > tweak it anymore, until there's a real need at least.
> 
> Lemme tweak it for you.  ;-)
>
I'm experiencing a problem with your suggestions (but I think this is
due to just a typo); see below.

> > (By the way, I think that we should really start testing non-obvious
> > subroutines and pieces of code defined in tests/defs.  But that's
> > for another thread).
> 
> Fine with me, on all accounts.
> 
> > Subject: [PATCH] test defs: add subroutine for input unindenting
> > 
> > * tests/defs.in (unindent): New subroutine.
> 
> > --- a/tests/defs.in
> > +++ b/tests/defs.in
> > @@ -440,6 +440,31 @@ AUTOMAKE_fails ()
> >    AUTOMAKE_run 1 ${1+"$@"}
> >  }
> >  
> > +# unindent [input files...]
> > +# -------------------------
> > +# Remove the "proper" amount of leading whitespace from the given files,
> > +# and output the result on stdout.  That amount is determined by looking
> > +# at the leading whitespace of the first non-blank line in the input
> > +# files.  If no input file is specified, standard input is implied.
> > +unindent ()
> > +{
> > +  cat ${1+"$@"} > deindent.tmp
> > +  indentation=`sed <deindent.tmp -n "
> > +    /^[ $tab]*$/n
> > +    s/[^ $tab].*$//
> > +    t end
> > +    b
> > +: end
> > +    p
> > +    q
> > +  "`
> 
> The space after : is not portable.  The sed script could be shortened to
>   sed -n "/[^ $tab].*$/{
>             s///p
>             d
>           }"
>
Did you mean 'q' instead of 'd' here, right?  Because the above fails
for me:

  $ cat foo.sh
  tab='   '
  deindent ()
  {
    cat ${1+"$@"} > deindent.tmp
    indentation=`sed <deindent.tmp -n "
      /[^ $tab].*$/{
        s///p
        d
    }"`
    case $indentation in
      '') cat deindent.tmp;;
       *) sed "s/^$indentation//" deindent.tmp;;
    esac
    rm -f deindent.tmp
  }
  deindent "$@"

  $ (echo ' x'; echo ' y') | LC_ALL=C sh -x foo.sh 
  + tab=' '
  + deindent
  + cat
  ++ sed -n '
      /[^         ].*$/{
        s///p
        d
    }'
  + indentation=' 
   '
  + case $indentation in
  + sed 's/^ 
   //' deindent.tmp
  sed: -e expression #1, char 4: unterminated `s' command
  + rm -f deindent.tmp

> > +  case $indentation in
> > +    '') cat deindent.tmp;;
> > +     *) sed "s/^$indentation//" deindent.tmp;;
> > +  esac
> 
> These four lines can be shortened to
>   sed "s/^$indentation//" deindent.tmp
>
Agreed on this.

> > +  rm -f deindent.tmp
> > +}
> 
> OK with those addressed.
> 
> It's a bit of a shame this uses a temp file though.  awk should be able
> to cope without
>
Yes, if I could assume "new" awk.  But traditional awk lacks the 'sub()'
and 'gsub()' builtins, so I really really don't want to go down that road.
Moreover, the use of a tempfile is not a big issue here, because we can
already assume the existence of a temporary working directory (i.e., the
one the test runs in) and of a cleanup trap (installed by `tests/defs').

But if you truly truly dislike the use of the tempfile, I could try to
use perl instead...  WDYT?

> (but be sure to try old Solaris awk if you go for it).
> 
> Thanks,
> Ralf
>

Regards,
   Stefano