Avoid tracking information to keep docids close when that is not necessary.

Just append on add, and then do an explicit sort the one place that requires operations on a doc
to be ordered.

2 files changed, 19 insertions, 12 deletions

diff --git a/searchlib/src/tests/attribute/attribute_test.cpp b/searchlib/src/tests/attribute/attribute_test.cpp
index af1fcea2e21..79e120d0683 100644
--- a/searchlib/src/tests/attribute/attribute_test.cpp
+++ b/searchlib/src/tests/attribute/attribute_test.cpp
@@ -1670,8 +1670,8 @@ AttributeTest::testStatus()
         AttributePtr ptr = createAttribute("as", cfg);
         addDocs(ptr, numDocs);
         auto & sa = *(static_cast<StringAttribute *>(ptr.get()));
-        const size_t numUniq(16);
-        const size_t numValuesPerDoc(16);
+        const size_t numValuesPerDoc(values.size());
+        const size_t numUniq(numValuesPerDoc);
         for (uint32_t i = 0; i < numDocs; ++i) {
             EXPECT_TRUE(appendToVector(sa, i, numValuesPerDoc, values));
         }
@@ -1680,11 +1680,11 @@ AttributeTest::testStatus()
         EXPECT_EQUAL(ptr->getStatus().getNumValues(), numDocs*numValuesPerDoc);
         EXPECT_EQUAL(ptr->getStatus().getNumUniqueValues(), numUniq);
         size_t expUsed = 0;
-        expUsed += 1 * InternalNodeSize + 1 * LeafNodeSize; // enum store tree
-        expUsed += numUniq * 32; // enum store (16 unique values, 32 bytes per entry)
+        expUsed += 1 * InternalNodeSize + 1 * LeafNodeSize; // Approximate enum store tree
+        expUsed += 272; // TODO Approximate... enum store (16 unique values, 17 bytes per entry)
         // multi value mapping (numdocs * sizeof(MappingIndex) + numvalues * sizeof(EnumIndex) +
-        // numdocs * sizeof(Array<EnumIndex>) (due to vector vector))
-        expUsed += numDocs * sizeof(vespalib::datastore::EntryRef) + numDocs * numValuesPerDoc * sizeof(IEnumStore::Index) + ((numValuesPerDoc > 1024) ? numDocs * NestedVectorSize : 0);
+        // 32 + numdocs * sizeof(Array<EnumIndex>) (due to vector vector))
+        expUsed += 32 + numDocs * sizeof(vespalib::datastore::EntryRef) + numDocs * numValuesPerDoc * sizeof(IEnumStore::Index) + ((numValuesPerDoc > 1024) ? numDocs * NestedVectorSize : 0);
         EXPECT_GREATER_EQUAL(ptr->getStatus().getUsed(), expUsed);
         EXPECT_GREATER_EQUAL(ptr->getStatus().getAllocated(), expUsed);
     }
diff --git a/searchlib/src/tests/attribute/changevector/changevector_test.cpp b/searchlib/src/tests/attribute/changevector/changevector_test.cpp
index ad33774e904..7bcf519bb18 100644
--- a/searchlib/src/tests/attribute/changevector/changevector_test.cpp
+++ b/searchlib/src/tests/attribute/changevector/changevector_test.cpp
@@ -2,17 +2,24 @@
 #include <vespa/vespalib/testkit/testapp.h>
 
 #include <vespa/searchlib/attribute/changevector.hpp>
-
+#include <vespa/vespalib/stllike/hash_set.h>
 using namespace search;
 
 template <typename T>
 void verifyStrictOrdering(const T & v) {
-    long count(0);
-    for (const auto & c : v) {
-        count++;
-        EXPECT_EQUAL(count, c._data.get());
+    vespalib::hash_set<uint32_t> complete;
+    uint32_t prev_doc(0);
+    uint32_t prev_value;
+    for (const auto & c : v.getDocIdInsertOrder()) {
+        if (prev_doc != c._doc) {
+            complete.insert(prev_doc);
+            EXPECT_FALSE(complete.contains(c._doc));
+            prev_doc = c._doc;
+        } else {
+            EXPECT_GREATER(c._data, prev_value);
+        }
+        prev_value = c._data;
     }
-    EXPECT_EQUAL(v.size(), size_t(count));
 }
 
 class Accessor {