memchr and valgrind

[Eric Blake]

With the recent clarification on memchr behavior, it is conceivable that we 
could trip false positives under valgrind when using memchr to find the 
location of a known byte with an overestimated length (most frequently when 
searching for a NUL).  Any objections to applying this patch?


From: Eric Blake <address@hidden>
Date: Mon, 15 Jun 2009 11:04:21 -0600
Subject: [PATCH] memchr: add valgrind exception

* lib/memchr.valgrind: New file.
* modules/memchr (Files): Distribute it.

Signed-off-by: Eric Blake <address@hidden>
---
 ChangeLog           |    4 ++++
 lib/memchr.valgrind |   14 ++++++++++++++
 modules/memchr      |    1 +
 3 files changed, 19 insertions(+), 0 deletions(-)
 create mode 100644 lib/memchr.valgrind

diff --git a/ChangeLog b/ChangeLog
index 81af4ec..b2492be 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,9 @@
 2009-06-15  Eric Blake  <address@hidden>

+       memchr: add valgrind exception
+       * lib/memchr.valgrind: New file.
+       * modules/memchr (Files): Distribute it.
+
        docs: memchr is no longer obsolete
        * MODULES.html.sh: Move memchr from obsolete to string.h section.

diff --git a/lib/memchr.valgrind b/lib/memchr.valgrind
new file mode 100644
index 0000000..60f247e
--- /dev/null
+++ b/lib/memchr.valgrind
@@ -0,0 +1,14 @@
+# Suppress a valgrind message about use of uninitialized memory in memchr().
+# POSIX states that when the character is found, memchr must not read extra
+# bytes in an overestimated length (for example, where memchr is used to
+# implement strnlen).  However, we use a safe word read to provide a speedup.
+{
+    memchr-value4
+    Memcheck:Value4
+    fun:rpl_memchr
+}
+{
+    memchr-value8
+    Memcheck:Value8
+    fun:rpl_memchr
+}
diff --git a/modules/memchr b/modules/memchr
index a77bec5..855c8db 100644
--- a/modules/memchr
+++ b/modules/memchr
@@ -3,6 +3,7 @@ memchr() function: scan memory for a byte.

 Files:
 lib/memchr.c
+lib/memchr.valgrind
 m4/memchr.m4
 m4/mmap-anon.m4

-- 
1.6.3.1