bug-gnulib
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

memcmp tests: work around the clang bug


From: Bruno Haible
Subject: memcmp tests: work around the clang bug
Date: Sat, 21 Dec 2019 05:42:07 +0100
User-agent: KMail/5.1.3 (Linux/4.4.0-166-generic; KDE/5.18.0; x86_64; ; )

A year ago, I committed this:

  2018-12-21  Bruno Haible  <address@hidden>

      memcmp: Mention the clang bug.
      * tests/test-memcmp.c: Add comment about a known test failure.
      * doc/posix-functions/memcmp.texi: Mention the clang bug.

But a comment does not stop the test from failing :) So, here's to make it
succeed:


2019-12-21  Bruno Haible  <address@hidden>

        memcmp tests: Work around the clang bug.
        * tests/test-memcmp.c (main): Use a volatile function pointer to disable
        the clang optimization.

diff --git a/tests/test-memcmp.c b/tests/test-memcmp.c
index 5d48119..f0bddef 100644
--- a/tests/test-memcmp.c
+++ b/tests/test-memcmp.c
@@ -25,12 +25,11 @@ SIGNATURE_CHECK (memcmp, int, (void const *, void const *, 
size_t));
 #include "zerosize-ptr.h"
 #include "macros.h"
 
-/* Note: This test sometimes fails when compiled by 'clang'.
-   See <https://bugs.llvm.org/show_bug.cgi?id=40063>.  */
-
 int
 main (void)
 {
+  int (* volatile memcmp_ptr) (const void *, const void *, size_t) = memcmp;
+
   /* Test equal / not equal distinction.  */
   ASSERT (memcmp (zerosize_ptr (), zerosize_ptr (), 0) == 0);
   ASSERT (memcmp ("foo", "foobar", 2) == 0);
@@ -48,10 +47,13 @@ main (void)
   ASSERT (memcmp ("foobar", "foo", 4) > 0);
 
   /* Some old versions of memcmp were not 8-bit clean.  */
-  ASSERT (memcmp ("\100", "\201", 1) < 0);
-  ASSERT (memcmp ("\201", "\100", 1) > 0);
-  ASSERT (memcmp ("\200", "\201", 1) < 0);
-  ASSERT (memcmp ("\201", "\200", 1) > 0);
+  /* Use the function pointer here, because otherwise this test is sometimes
+     miscompiled by 'clang'.
+     See <https://bugs.llvm.org/show_bug.cgi?id=40063>.  */
+  ASSERT (memcmp_ptr ("\100", "\201", 1) < 0);
+  ASSERT (memcmp_ptr ("\201", "\100", 1) > 0);
+  ASSERT (memcmp_ptr ("\200", "\201", 1) < 0);
+  ASSERT (memcmp_ptr ("\201", "\200", 1) > 0);
 
   /* The Next x86 OpenStep bug shows up only when comparing 16 bytes
      or more and with at least one buffer not starting on a 4-byte boundary.




reply via email to

[Prev in Thread] Current Thread [Next in Thread]