On Thu, Sep 19, 2013 at 07:13:25AM +1000, Dave Chinner wrote:
> On Wed, Sep 18, 2013 at 04:29:26PM -0400, Josef Bacik wrote:
> > Apparently the GNU guys decided to change their error output from something
> > like
> > Error `Error message'
> > To
> > Error 'Error message'
> > So to fix this I've introduced _filter_backtick which will change any ` to
> > ' and
> > then changed the output of the tests that were failing for me because of
> > this
> > output. I tested this on a new box that has the new output and an old box
> > which
> > has the old output and it appears to fix the issue. Thanks,
> > Signed-off-by: Josef Bacik <jbacik@xxxxxxxxxxxx>
> That's just leaving a landmine behind, and it doesn't catch all the
> tests that need updating. This approach was floated here:
> And my response was to add a global filter to the .check file so it
> doesn't leave a landmine. Indeed, I have a local version on tomas'
> patch that I modified in May does just that:
> --- a/check
> +++ b/check
> @@ -477,6 +477,10 @@ do
> echo " - no qualified output"
> + # coreutils 2.16+ changed quote formats in error messages from
> + # `foo' to 'foo'. Filter old versions to match the new
> + sed -i "s/\`/\'/g" $tmp.out
> if diff $seq.out $tmp.out >/dev/null 2>&1
> if $err
> I also discovered that for some reason LANG=C is not sufficient for
> all cases to make the quoting behaviour consistent. i.e. I
> needed to set LC_ALL=C so that it didn't use weird UTF-8 encodings
> for the quotes instead of a simple backtick.
> Full patch below.
Excellent, what is holding this patch up then if it was proposed in May and you
are ok with it? Thanks,