From: Alex Rousskov <rousskov@measurement-factory.com>
Date: Fri, 15 Feb 2013 18:12:40 +0000 (-0700)
Subject: mem_hdr::hasContigousContentRange() should return true for empty ranges
X-Git-Tag: SQUID_3_5_0_1~444^2~67
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=9cc44692ae5ad980e1f8474544e79886d5695257;p=thirdparty%2Fsquid.git

mem_hdr::hasContigousContentRange() should return true for empty ranges

mem_hdr::hasContigousContentRange() is called exclusively from
StoreEntry::mayStartSwapOut() via memObject::isContiguous(). In theory, that
mayStartSwapOut() may happen when we have not written anything to mem_hdr yet
(not even HTTP response headers). In that case, mayStartSwapout() should not
refuse to swap the entry out on non-contiguous grounds, but it was doing that
because hasContigousContentRange() mishandled the empty range case.

XXX: Calling hasContigousContentRange() [just] from mayStartSwapout() is just
wrong because what may be contiguous now may become fragmented later.
---

diff --git a/src/stmem.cc b/src/stmem.cc
index 1465a9fbe7..071584c049 100644
--- a/src/stmem.cc
+++ b/src/stmem.cc
@@ -313,7 +313,7 @@ mem_hdr::hasContigousContentRange(Range<int64_t> const & range) const
             return true;
     }
 
-    return false;
+    return !range.size(); // empty range is contigous
 }
 
 bool