Re: glibc testing problem

["Carlos O'Donell" <carlos at systemhalted dot org>]

On Wed, Nov 28, 2012 at 4:49 PM, Steve Ellcey <sellcey@mips.com> wrote:
> While testing glibc on a MIPS system I was getting failures in tst-fmon.
> After some hunting around I found that the problem wasn't in what my
> glibc was producing but in what was expected.  This test uses the data
> file localedata/tst-fmon.data and the script tst-fmon.sh reads that
> with:
>
> while IFS="     " read locale format value expect; do
>
> where there is a tab inside the quote marks.  Now some of the lines
> in the data file are separated by two tabs instead of one, like this one:
>
> de_DE.ISO-8859-1        %n      1.23            1,23 EUR
>
> There are two tabs between '1.23' and '1,23 EUR'.  Most shells (bash, ksh)
> seem to treat two tabs the same as one and have no problems.  But on
> my systems (debian and ubuntu) /bin/sh is a link to dash and dash does not
> treat two tabs like one.  It instead considers the second tab as part of the
> next field.  This is why the test fails for me.
>
> Consider this script:
>
> echo "y         z" | \
> while IFS="     " read a b
> do
> echo "\"$a\"" "\"$b\""
> done
>
> Where there are two tabs between y and z.  On my system if I run
> ksh ./script or bash ./script, I get:
>
> "y" "z"
>
> But if I run dash ./script, I get:
>
> "y" "   z"
>
> So my question is: what do we do about this?  Remove the extra tab from
> the data file, simply say dash isn't supported (the script starts with
> '#!/bin/sh' and /bin/sh on my system is a link to dash), or is there some
> better fix?

That seems like a bug in dash.

POSIX says "<newline><space><tab>foo<tab><tab>bar<space> yields two
fields, foo and bar."

Not "foo" and "<tab>bar".

http://pubs.opengroup.org/onlinepubs/009695399/utilities/xcu_chap02.html#tag_02_06_05

Cheers,
Carlos.