Re: gettext-related test failures: config.rpath presence

[Alexandre Duret-Lutz]

>>> "RW" == Ralf Wildenhues <address@hidden> writes:

 RW> CVS Automake diagnoses missing files mentioned in AC_REQUIRE_AUX_FILE
 RW> statements.  AM_GNU_GETTEXT of gettext >= 0.14.3 causes config.rpath
 RW> to be required through this mechanism:

 >> + automake-1.9a --foreign -Werror -Wall --add-missing
 >> configure.in:4: required file `./config.rpath' not found
 >> FAIL: gettext.test

 RW> One cheap way to ensure this is to provide dummy `config.rpath' files.

 RW> But then I got ambitious and tried to enable gettext.test to ensure that
 RW> the requirement is actually diagnosed when the gettext installation
 RW> works and is recent enough, while avoiding `autopoint' to modify our
 RW> test suite directory.

 RW> OK?

Yes, please.

[...]

 RW> --- tests/gettext.test     14 May 2005 20:28:55 -0000      1.6
 RW> +++ tests/gettext.test     10 May 2006 17:02:44 -0000
[...]
 RW> +# config.rpath is required by versions >= 0.14.3.  We try to verify
 RW> +# this requirement, but only when we find we have a working and recent
 RW> +# gettext installation.
 RW> +
 RW> +# If aclocal fails here, it may be that gettext is too old to
 RW> +# provide AM_GNU_GETTEXT_VERSION.
 RW> +if $ACLOCAL; then
 RW> +
 RW> +  # autopoint will fail if it's from an older version.
 RW> +  # If gettext is too old to provide autopoint, this will
 RW> +  # fail as well, so we're safe here.
 RW> +  if autopoint -n; then
 RW> +    AUTOMAKE_fails --add-missing
 RW> +    grep 'required.*config.rpath' stderr
 RW> +  fi
 RW> +fi

Shouldn't these two `if' have an `else :' like we do in
Makefiles for those shells that don't reset $? after `fi' and
abort with `set -e'?

[...]

-- 
Alexandre Duret-Lutz

Shared books are happy books.     http://www.bookcrossing.com/friend/gadl