[PATCH] hash: silence spurious clang warning

[Eric Blake]

* lib/hash.c (hash_get_next): Aid clang analysis.

Signed-off-by: Eric Blake <address@hidden>
---

Jim, should we apply this?  As far as I can tell, this is
an example of a shortfall in clang 2.7, as shipped in Fedora 13.

Clang assumed that the for loop at line 310 is skipped because
cursor is NULL, which implies bucket is NULL, which implies
that line 316 bucket->data is a dereference near NULL.  But
this is invalid, because bucket was explicitly initialized
to table->bucket (non-NULL) plus some offset, at line 302.
Adding the extra condition before the abort() is enough to
get clang's invalid assumption out of the way.

 ChangeLog  |    3 +++
 lib/hash.c |    2 +-
 2 files changed, 4 insertions(+), 1 deletions(-)

diff --git a/ChangeLog b/ChangeLog
index fec5dc9..246d144 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,8 @@
 2010-08-30  Eric Blake  <address@hidden>

+       hash: silence spurious clang warning
+       * lib/hash.c (hash_get_next): Aid clang analysis.
+
        tests: silence clang warning
        * tests/test-malloca.c (do_allocation): Avoid dead store.

diff --git a/lib/hash.c b/lib/hash.c
index 15630be..e2727b9 100644
--- a/lib/hash.c
+++ b/lib/hash.c
@@ -303,7 +303,7 @@ hash_get_next (const Hash_table *table, const void *entry)
     = table->bucket + table->hasher (entry, table->n_buckets);
   struct hash_entry const *cursor;

-  if (! (bucket < table->bucket_limit))
+  if (! (bucket && bucket < table->bucket_limit))
     abort ();

   /* Find next entry in the same bucket.  */
-- 
1.7.2.2