Re: lib/46433: tests/lib/libm/t_exp should not use exp() itself

[Izumi Tsutsui <tsutsui%ceres.dti.ne.jp@localhost>]

The following reply was made to PR lib/46433; it has been noted by GNATS.

From: Izumi Tsutsui <tsutsui%ceres.dti.ne.jp@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: jruohonen%iki.fi@localhost, isaki%NetBSD.org@localhost, 
tsutsui%ceres.dti.ne.jp@localhost
Subject: Re: lib/46433: tests/lib/libm/t_exp should not use exp() itself
Date: Wed, 30 May 2012 19:09:40 +0900

 >  That is one option sure. But if there is no immedate fix available, we
 >  likely want to count these cases as expected failures. So I think we can put
 >  something like the following to the beginning of these cases:
 >  
 >      if (exp(1.0) == 0.0) {
 >              atf_tc_expect_fail("PR lib/46433");
 >              atf_tc_fail("exp(1.0) = 0.0");
 >      }
 >  
 >      [start the actual test]
 
 The point of this PR is not that there is a particular implementation
 which is expected to fail on this test, but simply "this test is wrong"
 (it isn't an actual test), because it could return false positive
 even if the exp() function has obviously bogus implementation.
 (if exp() always returns zero for example, as noted in the PR)
 
 Testes for arithmetic functions should also have checks against
 a set of "correct" result values with acceptable tolerance like t_cos,
 as submitter says.  Why can't it be an "immediate" fix?
 
 ---
 Izumi Tsutsui