Bug#33806018: Remove usage of MD5 in explain_access_path.cc

In debug builds only, the hypergraph optimizer has functionality for
forcing certain subplans to be preferred, regardless of their cost.
The identifiers for these subplans are created from MD5 checksums.

Since MD5 is being phased out, we change it to create identifiers from
SHA-256 checksums instead, so that it will continue to work also when
MD5 is not available.

Change-Id: I243959734debeef78d295523b046c2314fcfc931

Author: kahatlen

mysql-test/r/subplan_tokens.result

@@ -6,41 +6,41 @@ test.t1	analyze	status	OK
 SET DEBUG='+d,subplan_tokens';
 EXPLAIN FORMAT=tree SELECT * FROM t1;
 EXPLAIN
--> [0x383f9723be72015d] Table scan on t1  (cost=0.08..0.25 rows=3)
+-> [0xeed2c0bd3e39ba93] Table scan on t1  (cost=0.08..0.25 rows=3)
 
 To force this plan, use:
-SET DEBUG='+d,subplan_tokens,force_subplan_0x383f9723be72015d';
+SET DEBUG='+d,subplan_tokens,force_subplan_0xeed2c0bd3e39ba93';
 
 EXPLAIN FORMAT=tree SELECT * FROM t1 ORDER BY a;
 EXPLAIN
--> [0x4f56594d4fd521c9] Index scan on t1 using a  (cost=0.08..0.25 rows=3)
+-> [0x2215d3ce44f01b0d] Index scan on t1 using a  (cost=0.08..0.25 rows=3)
 
 To force this plan, use:
-SET DEBUG='+d,subplan_tokens,force_subplan_0x4f56594d4fd521c9';
+SET DEBUG='+d,subplan_tokens,force_subplan_0x2215d3ce44f01b0d';
 
-SET DEBUG='+d,subplan_tokens,force_subplan_0x383f9723be72015d';
+SET DEBUG='+d,subplan_tokens,force_subplan_0xeed2c0bd3e39ba93';
 EXPLAIN FORMAT=tree SELECT * FROM t1 ORDER BY a;
 EXPLAIN
--> [0x685afbd3f088dc87] Sort: t1.a  (cost=1.03..1.03 rows=3)
-    -> [0x383f9723be72015d] Table scan on t1  (cost=0.08..0.25 rows=3)
+-> [0xd3135ef8a92cb9b8] Sort: t1.a  (cost=1.03..1.03 rows=3)
+    -> [0xeed2c0bd3e39ba93] Table scan on t1  (cost=0.08..0.25 rows=3)
 
 To force this plan, use:
-SET DEBUG='+d,subplan_tokens,force_subplan_0x685afbd3f088dc87,force_subplan_0x383f9723be72015d';
+SET DEBUG='+d,subplan_tokens,force_subplan_0xd3135ef8a92cb9b8,force_subplan_0xeed2c0bd3e39ba93';
 
 SET DEBUG='+d,subplan_tokens,'
-  'force_subplan_0x90b6f02248424220,'
-  'force_subplan_0x383f9723be72015d,'
-  'force_subplan_0xedd60db81c76aaf5';
+  'force_subplan_0x38d170e70c04e92c,'
+  'force_subplan_0xeed2c0bd3e39ba93,'
+  'force_subplan_0x079e429c703ec298';
 EXPLAIN FORMAT=tree SELECT * FROM t1 JOIN t1 AS t2 ON t1.a=t2.a ORDER BY t1.a;
 EXPLAIN
--> [0x9fe1924c5f48742c] Sort: t1.a  (cost=2.09..2.09 rows=3)
-    -> [0x90b6f02248424220] Inner hash join (t1.a = t2.a)  (cost=0.80..1.31 rows=3)
-        -> [0x383f9723be72015d] Table scan on t1  (cost=0.08..0.25 rows=3)
+-> [0x1836646581797f8f] Sort: t1.a  (cost=2.09..2.09 rows=3)
+    -> [0x38d170e70c04e92c] Inner hash join (t1.a = t2.a)  (cost=0.80..1.31 rows=3)
+        -> [0xeed2c0bd3e39ba93] Table scan on t1  (cost=0.08..0.25 rows=3)
         -> Hash
-            -> [0xedd60db81c76aaf5] Table scan on t2  (cost=0.08..0.25 rows=3)
+            -> [0x079e429c703ec298] Table scan on t2  (cost=0.08..0.25 rows=3)
 
 To force this plan, use:
-SET DEBUG='+d,subplan_tokens,force_subplan_0x9fe1924c5f48742c,force_subplan_0x90b6f02248424220,force_subplan_0x383f9723be72015d,force_subplan_0xedd60db81c76aaf5';
+SET DEBUG='+d,subplan_tokens,force_subplan_0x1836646581797f8f,force_subplan_0x38d170e70c04e92c,force_subplan_0xeed2c0bd3e39ba93,force_subplan_0x079e429c703ec298';
 
 SET DEBUG='-d,subplan_tokens';
 EXPLAIN FORMAT=tree SELECT * FROM t1 JOIN t1 AS t2 ON t1.a=t2.a ORDER BY t1.a;

mysql-test/t/subplan_tokens.test

@@ -14,16 +14,16 @@ EXPLAIN FORMAT=tree SELECT * FROM t1 ORDER BY a;
 
 # However, if we force the table scan variant from earlier,
 # we should get a sort instead.
-SET DEBUG='+d,subplan_tokens,force_subplan_0x383f9723be72015d';
+SET DEBUG='+d,subplan_tokens,force_subplan_0xeed2c0bd3e39ba93';
 EXPLAIN FORMAT=tree SELECT * FROM t1 ORDER BY a;
 
-# Force a hash join; this is the truncated MD5 checksum of:
-# Inner hash join (t1.a = t2.a),0x383f9723be72015d,Hash:0xedd60db81c76aaf5
+# Force a hash join; this is the truncated SHA-256 checksum of:
+# Inner hash join (t1.a = t2.a),0xeed2c0bd3e39ba93,Hash:0x079e429c703ec298
 # We also force the hash join subplans to ensure they are not optimized away.
 SET DEBUG='+d,subplan_tokens,'
-  'force_subplan_0x90b6f02248424220,'
-  'force_subplan_0x383f9723be72015d,'
-  'force_subplan_0xedd60db81c76aaf5';
+  'force_subplan_0x38d170e70c04e92c,'
+  'force_subplan_0xeed2c0bd3e39ba93,'
+  'force_subplan_0x079e429c703ec298';
 EXPLAIN FORMAT=tree SELECT * FROM t1 JOIN t1 AS t2 ON t1.a=t2.a ORDER BY t1.a;
 
 # When not forcing anything, we should prefer the nested loop join.

sql/join_optimizer/explain_access_path.cc

@@ -22,8 +22,13 @@
 
 #include "sql/join_optimizer/explain_access_path.h"
 
-#include "my_md5.h"
-#include "my_md5_size.h"
+#include <functional>
+#include <string>
+#include <vector>
+
+#include <openssl/sha.h>
+
+#include "sha2.h"
 #include "sql/filesort.h"
 #include "sql/item_sum.h"
 #include "sql/iterators/basic_row_iterators.h"
@@ -42,10 +47,7 @@
 #include "sql/range_optimizer/range_optimizer.h"
 #include "sql/sql_optimizer.h"
 #include "sql/table.h"
-
-#include <functional>
-#include <string>
-#include <vector>
+#include "template_utils.h"
 
 using std::string;
 using std::vector;
@@ -1172,7 +1174,7 @@ string PrintQueryPlan(int level, AccessPath *path, JOIN *join,
 }
 
 // 0x
-// truncated_md5(desc1,desc2,...,[child1_desc:]0xchild1,[child2_desc:]0xchild2,...)
+// truncated_sha256(desc1,desc2,...,[child1_desc:]0xchild1,[child2_desc:]0xchild2,...)
 string GetForceSubplanToken(AccessPath *path, JOIN *join) {
   if (path == nullptr) {
     return "";
@@ -1194,13 +1196,14 @@ string GetForceSubplanToken(AccessPath *path, JOIN *join) {
     digest += GetForceSubplanToken(child.path, join);
   }
 
-  unsigned char md5sum[MD5_HASH_SIZE];
-  compute_md5_hash(pointer_cast<char *>(md5sum), digest.data(), digest.size());
+  unsigned char sha256sum[SHA256_DIGEST_LENGTH];
+  (void)SHA_EVP256(pointer_cast<const unsigned char *>(digest.data()),
+                   digest.size(), sha256sum);
 
   char ret[8 * 2 + 2 + 1];
-  snprintf(ret, sizeof(ret), "0x%02x%02x%02x%02x%02x%02x%02x%02x", md5sum[0],
-           md5sum[1], md5sum[2], md5sum[3], md5sum[4], md5sum[5], md5sum[6],
-           md5sum[7]);
+  snprintf(ret, sizeof(ret), "0x%02x%02x%02x%02x%02x%02x%02x%02x", sha256sum[0],
+           sha256sum[1], sha256sum[2], sha256sum[3], sha256sum[4], sha256sum[5],
+           sha256sum[6], sha256sum[7]);
 
   return ret;
 }

sql/join_optimizer/explain_access_path.h

@@ -1,4 +1,4 @@
-/* Copyright (c) 2020, 2021, Oracle and/or its affiliates.
+/* Copyright (c) 2020, 2022, Oracle and/or its affiliates.
 
    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License, version 2.0,
@@ -51,7 +51,7 @@ std::string PrintQueryPlan(
   preferred over all others.
 
   The token returned is “0x<digest>”, where <digest> is the first 64 bits
-  of the MD5 sum of this string:
+  of the SHA-256 sum of this string:
 
     desc1,desc2,...,[child1_desc:]0xchild1,[child2_desc:]0xchild2,<more
   children>
@@ -60,7 +60,7 @@ std::string PrintQueryPlan(
   and 0xchild1 is the token for children. The normal way to generate such
   tokens is to use SET DEBUG='+d,subplan_tokens' and look at the EXPLAIN
   FORMAT=tree, but in a pinch, you can also write them by hand and use
-  md5sum or a similar tool.
+  sha256sum or a similar tool.
 
   Only the hypergraph join optimizer honors token preferences, but EXPLAIN
   FORMAT=tree shows computed tokens for both optimizers.