From 785e6c91bd48d55e260b749b4473d31fdfbe8d6d Mon Sep 17 00:00:00 2001
From: Andrew Hutchings <andrew@linuxjedi.co.uk>
Date: Wed, 19 Apr 2017 22:45:23 +0100
Subject: [PATCH] MCOL-670 Fix UPDATE with BLOB/TEXT

* Don't cache > 8000 bytes during update
* Fix PrimProc case where token is used more than once
---
 primitives/primproc/dictstep.cpp     | 17 +++++++++++++----
 writeengine/dictionary/we_dctnry.cpp |  8 ++++++--
 2 files changed, 19 insertions(+), 6 deletions(-)

diff --git a/primitives/primproc/dictstep.cpp b/primitives/primproc/dictstep.cpp
index 29f9b2ac5..68576899b 100644
--- a/primitives/primproc/dictstep.cpp
+++ b/primitives/primproc/dictstep.cpp
@@ -520,6 +520,7 @@ void DictStep::_projectToRG(RowGroup &rg, uint32_t col)
 		}
 		else {
             uint32_t firstTmpResultCounter = tmpResultCounter;
+            std::map<uint32_t, string *> result;
 			for (i = curResultCounter; i < firstTmpResultCounter; i++) {
 				rg.getRow(newRidList[i].pos, &r);
                 // If this is a multi-block blob, get all the blocks
@@ -530,7 +531,14 @@ void DictStep::_projectToRG(RowGroup &rg, uint32_t col)
                 {
                     StringPtr multi_part[1];
                     uint16_t old_offset = primMsg->tokens[0].offset;
-                    string *result = new string((char*)tmpStrings[i].ptr, tmpStrings[i].len);
+                    if (result.empty())
+                    {
+                        // String copy here because tmpStrings pointers will be blown away below
+                        for (uint32_t x = curResultCounter; x < firstTmpResultCounter; x++)
+                        {
+                            result[x] = new string((char*)tmpStrings[x].ptr, tmpStrings[x].len);
+                        }
+                    }
                     uint64_t origin_lbid = primMsg->LBID;
                     uint32_t lbid_count = newRidList[i].token >> 46;
                     primMsg->tokens[0].offset = 1; // first offset of a sig
@@ -542,12 +550,13 @@ void DictStep::_projectToRG(RowGroup &rg, uint32_t col)
                         primMsg->tokens[0].LBID = origin_lbid + j;
                         issuePrimitive(false);
                         projectResult(multi_part);
-                        result->append((char*)multi_part[0].ptr, multi_part[0].len);
+                        result[i]->append((char*)multi_part[0].ptr, multi_part[0].len);
                     }
                     primMsg->tokens[0].offset = old_offset;
+                    primMsg->LBID = origin_lbid;
                     tmpResultCounter = firstTmpResultCounter;
-                    r.setVarBinaryField((unsigned char*)result->c_str(), result->length(), col);
-                    delete result;
+                    r.setVarBinaryField((unsigned char*)result[i]->c_str(), result[i]->length(), col);
+                    delete result[i];
                 }
                 else
                 {
diff --git a/writeengine/dictionary/we_dctnry.cpp b/writeengine/dictionary/we_dctnry.cpp
index da669f3bf..8a4b9d2af 100644
--- a/writeengine/dictionary/we_dctnry.cpp
+++ b/writeengine/dictionary/we_dctnry.cpp
@@ -1374,7 +1374,9 @@ int  Dctnry::updateDctnry(unsigned char* sigValue, int& sigSize,
     sig.size = sigSize;
 
     // Look for string in cache
-    if (m_arraySize < MAX_STRING_CACHE_SIZE)
+    // As long as the string <= 8000 bytes
+    if ((m_arraySize < MAX_STRING_CACHE_SIZE) &&
+        (sigSize <= MAX_SIGNATURE_SIZE))
     {
         bool found = false;
         found = getTokenFromArray(sig);
@@ -1389,7 +1391,9 @@ int  Dctnry::updateDctnry(unsigned char* sigValue, int& sigSize,
     rc = insertDctnry(sigSize, sigValue, token);
 
     //Add the new signature and token into cache
-    if (m_arraySize < MAX_STRING_CACHE_SIZE)
+    //As long as the string is <= 8000 bytes
+    if ((m_arraySize < MAX_STRING_CACHE_SIZE) &&
+        (sigSize <= MAX_SIGNATURE_SIZE))
     {
         Signature sig;
         sig.size = sigSize;