Simon Josefsson <simon <at> josefsson.org> writes:
> > It's good habit to include the tests from the autoconf macro also in the
test
> > suite (to verify that the autoconf macro actually did its job). So I added
this:
> >
> >
> > 2008-05-20 Bruno Haible <bruno <at> clisp.org>
> >
> > * tests/test-memcmp.c (main): Test also the sign of the result. Test
> > against two known bugs; code taken from autoconf's AC_FUNC_MEMCMP.
>
> Thanks! FWIW, that updated self test works fine under mingw even
> without the gnulib memcmp (using the patch I sent earlier).
This test is tripping up a -Wall build on cygwin with gcc 3.4.4. For the life
of me, I can't find the location of the __attribute__((__nonnull__)) that is
causing the warning, even when I ran through gcc -E; however, I think that it
is a bug to attach the nonnull attribute to memcmp parameters (for the very
reason that a compare of 0 bytes should not dereference the pointer, so passing
NULL is not necessarily a bug). But the warning is there, so I might as well
work around it; I'm committing the patch below.
../../tests/test-memcmp.c: In function `main':
../../tests/test-memcmp.c:41: warning: null argument where non-null required
(arg 1)
../../tests/test-memcmp.c:41: warning: null argument where non-null required
(arg 2)
make[5]: *** [test-memcmp.o] Error 1
Meanwhile, I've been patching newlib (and thus cygwin) to have faster string.h
operations. For example, cygwin's strlen was using the x86 byte-wise 'repnz
scasb' instruction as its inner loop, which is more than 3x slower on modern
hardware than searching a word at a time (similar to our recent memchr.c
algorithm improvements). Likewise for cygwin's memcmp on unaligned memory
(aligned memory did a word at a time, but unaligned was using the dirt
slow 'repz cmpsb'). These gnulib tests have been invaluable in making sure I
don't break things. Unfortunately, I don't know of any sane way to test at
configure time whether a replacement C implementation of memcmp is faster than
the native implementation, to offer substituting in a rpl_memcmp for speed.
>From 6cc9690c07fa88ab7f9adc5cba9e9c56e84d90bb Mon Sep 17 00:00:00 2001
From: Eric Blake <[EMAIL PROTECTED]>
Date: Thu, 22 May 2008 07:57:37 -0600
Subject: [PATCH] Avoid gcc warning.
* tests/test-memcmp.c (main): Pass NULL indirectly.
Signed-off-by: Eric Blake <[EMAIL PROTECTED]>
---
ChangeLog | 5 +++++
tests/test-memcmp.c | 4 +++-
2 files changed, 8 insertions(+), 1 deletions(-)
diff --git a/ChangeLog b/ChangeLog
index dc0ade5..86aeadd 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,8 @@
+2008-05-22 Eric Blake <[EMAIL PROTECTED]>
+
+ Avoid gcc warning.
+ * tests/test-memcmp.c (main): Pass NULL indirectly.
+
2008-05-21 Bruno Haible <[EMAIL PROTECTED]>
Add reference doc about ACLs.
diff --git a/tests/test-memcmp.c b/tests/test-memcmp.c
index 5c2ac29..50342cb 100644
--- a/tests/test-memcmp.c
+++ b/tests/test-memcmp.c
@@ -37,8 +37,10 @@
int
main (void)
{
+ void *nil = NULL; /* Use to avoid gcc attribute((nonnull)) warnings. */
+
/* Test equal / not equal distinction. */
- ASSERT (memcmp (NULL, NULL, 0) == 0);
+ ASSERT (memcmp (nil, nil, 0) == 0);
ASSERT (memcmp ("foo", "foobar", 2) == 0);
ASSERT (memcmp ("foo", "foobar", 3) == 0);
ASSERT (memcmp ("foo", "foobar", 4) != 0);
--
1.5.5.1
