Re: mingw isnanl-nolibm failure

[Eric Blake]

Eric Blake <ebb9 <at> byu.net> writes:

> 
> Here's one idea.  Since it fixes the mingw cross-compilation failure of test-
> isnanl-nolibm, I'm installing it; if we come up with something better or more 
> efficient in the future, we can alter the test then.
> 
> The idea behind this patch is that = doesn't alter the bit pattern of an 
> invalid representation that sneaks past the == check, but +0 will result in a 
> normalized bit pattern, so that the memcmp can then filter out pseudo-
> denormals.  For all valid bit patterns, the == already filtered out NaN, so 
> adding +0.0 is a no-op to the bit pattern.  Here's a snippet of my gdb 
session 
> on mingw that shows this point (where sizeof(long double) is 12):

That fix was insufficient, since it failed on isnan(-0.0L) (-0.0 + 0.0 is +0.0, 
not -0.0, not to mention that the padding bits in m1 and m2 were uninitialized 
and could differ).  Since the failure on -0.0L wasn't caught until the test-
vasprintf-posix test, I'm augmenting test-isnan* to actually try this case.

From: Eric Blake <address@hidden>
Date: Tue, 11 Dec 2007 14:00:26 -0700
Subject: [PATCH] Fix bug with -0.0L in previous patch.

* lib/isnan.c (rpl_isnanl): Make robust to -0.0L and pad bits.
* tests/test-isnan.c (main): Also test on zeroes.
* tests/test-isnanf.c (main): Likewise.
* tests/test-isnanl.h (main): Likewise.

Signed-off-by: Eric Blake <address@hidden>
---
 ChangeLog           |    6 ++++++
 lib/isnan.c         |    4 +++-
 tests/test-isnan.c  |    2 ++
 tests/test-isnanf.c |    2 ++
 tests/test-isnanl.h |    2 ++
 5 files changed, 15 insertions(+), 1 deletions(-)

diff --git a/ChangeLog b/ChangeLog
index 127f9a5..a954c4c 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,11 @@
 2007-12-11  Eric Blake  <address@hidden>
 
+       Fix bug with -0.0L in previous patch.
+       * lib/isnan.c (rpl_isnanl): Make robust to -0.0L and pad bits.
+       * tests/test-isnan.c (main): Also test on zeroes.
+       * tests/test-isnanf.c (main): Likewise.
+       * tests/test-isnanl.h (main): Likewise.
+
        Detect pseudo-denormals on x86 even when cross-compiling.
        * lib/isnan.c (rpl_isnanl) [!KNOWN_EXPBIT0_LOCATION
        && USE_LONG_DOUBLE && x86]: Add one more check to filter out
diff --git a/lib/isnan.c b/lib/isnan.c
index 182a7bd..04bf7f0 100644
--- a/lib/isnan.c
+++ b/lib/isnan.c
@@ -146,8 +146,10 @@ FUNC (DOUBLE x)
       memory_double m1;
       memory_double m2;
 
+      memset (&m1.value, 0, SIZE);
+      memset (&m2.value, 0, SIZE);
       m1.value = x;
-      m2.value = x + 0;
+      m2.value = x + (x ? 0.0L : -0.0L);
       if (memcmp (&m1.value, &m2.value, SIZE) != 0)
         return 1;
 # endif
diff --git a/tests/test-isnan.c b/tests/test-isnan.c
index 717ad7a..3541af4 100644
--- a/tests/test-isnan.c
+++ b/tests/test-isnan.c
@@ -57,6 +57,8 @@ main ()
   ASSERT (!isnan (-2.718));
   ASSERT (!isnan (-2.718e30));
   ASSERT (!isnan (-2.718e-30));
+  ASSERT (!isnan (0.0));
+  ASSERT (!isnan (-0.0));
   /* Infinite values.  */
   ASSERT (!isnan (1.0 / 0.0));
   ASSERT (!isnan (-1.0 / 0.0));
diff --git a/tests/test-isnanf.c b/tests/test-isnanf.c
index 898e6ed..c433089 100644
--- a/tests/test-isnanf.c
+++ b/tests/test-isnanf.c
@@ -57,6 +57,8 @@ main ()
   ASSERT (!isnanf (-2.718f));
   ASSERT (!isnanf (-2.718e30f));
   ASSERT (!isnanf (-2.718e-30f));
+  ASSERT (!isnanf (0.0f));
+  ASSERT (!isnanf (-0.0f));
   /* Infinite values.  */
   ASSERT (!isnanf (1.0f / 0.0f));
   ASSERT (!isnanf (-1.0f / 0.0f));
diff --git a/tests/test-isnanl.h b/tests/test-isnanl.h
index 5885457..657948c 100644
--- a/tests/test-isnanl.h
+++ b/tests/test-isnanl.h
@@ -47,6 +47,8 @@ main ()
   ASSERT (!isnanl (-2.718L));
   ASSERT (!isnanl (-2.718e30L));
   ASSERT (!isnanl (-2.718e-30L));
+  ASSERT (!isnanl (0.0L));
+  ASSERT (!isnanl (-0.0L));
   /* Infinite values.  */
   ASSERT (!isnanl (1.0L / 0.0L));
   ASSERT (!isnanl (-1.0L / 0.0L));
-- 
1.5.3.5