RE: Tests that fail due to iconv()

[Avis, Ed]

Bruno Haible <address@hidden> wrote:

>But my main argument is that the programs are still reliable in a
>sense that even without an iconv(), they never produce wrong results
>(they give an error message instead), and for some very limited use of
>the tools the iconv() is not needed.

Yup.  As long as the test suite makes this clear, and prints messages 'some
tests failed because your local iconv() is funny', rather than messages
which imply that gettext itself is buggy.

Or you could simply declare: the test suite is not reliable on systems with
broken iconv(), because it cannot distinguish between the two kinds of
failures.  If you move the tests which are likely to fail (eg UTF-8) to
the beginning of the test suite, and make them print such a message, then
it would be obvious where the failure lies.

-- 
Ed Avis <address@hidden>