From: Robert Haas <rhaas@postgresql.org>
Date: Fri, 28 Oct 2011 21:04:22 +0000 (-0400)
Subject: Demote some sanity checks in BufferIsValid() to assertions.
X-Git-Tag: REL9_2_BETA1~917
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=b6335a3f1b33e5dc52e755956d8648f0813252c4;p=postgresql

Demote some sanity checks in BufferIsValid() to assertions.

Testing reveals that this macro is a hot-spot for index-only-scans.
Per discussion with Tom Lane.
---

diff --git a/src/include/storage/bufmgr.h b/src/include/storage/bufmgr.h
index b8fc87ec57..82b43006f3 100644
--- a/src/include/storage/bufmgr.h
+++ b/src/include/storage/bufmgr.h
@@ -78,20 +78,24 @@ extern PGDLLIMPORT int32 *LocalRefCount;
  *		True iff the given buffer number is valid (either as a shared
  *		or local buffer).
  *
- * This is not quite the inverse of the BufferIsInvalid() macro, since this
- * adds sanity rangechecks on the buffer number.
- *
  * Note: For a long time this was defined the same as BufferIsPinned,
  * that is it would say False if you didn't hold a pin on the buffer.
  * I believe this was bogus and served only to mask logic errors.
  * Code should always know whether it has a buffer reference,
  * independently of the pin state.
+ *
+ * Note: For a further long time this was not quite the inverse of the
+ * BufferIsInvalid() macro, in that it also did sanity checks to verify
+ * that the buffer number was in range.  Most likely, this macro was
+ * originally intended only to be used in assertions, but its use has
+ * since expanded quite a bit, and the overhead of making those checks
+ * even in non-assert-enabled builds can be significant.  Thus, we've
+ * now demoted the range checks to assertions within the macro itself.
  */
 #define BufferIsValid(bufnum) \
 ( \
-	(bufnum) != InvalidBuffer && \
-	(bufnum) >= -NLocBuffer && \
-	(bufnum) <= NBuffers \
+	AssertMacro((bufnum) <= NBuffers && (bufnum) >= -NLocBuffer), \
+	(bufnum) != InvalidBuffer  \
 )
 
 /*