Update default values for optimizer cost constants. New values are:

  row_evaluate_cost             0.1
  key_compare_cost              0.05
  memory_temptable_create_cost  1.0
  memory_temptable_row_cost     0.1
  disk_temptable_create_cost    20.0
  disk_temptable_row_cost       0.5
  memory_block_read_cost        0.25
  io_block_read_cost            1.0

Changes to source files:

sql/opt_costconstants.cc
  Changed default values for cost constants.
sql/sql_select.h
  Change type of JOIN_TAB::read_time from ha_rows to double since cost may now
  be lower than 1.
sql/sql_optimizer.cc
sql/sql_select.cc
  Removed casts when assigning to/from JOIN_TAB::read_time
unittest/gunit/opt_costconstants-t.cc
  Updated unit tests to use new values for cost constants

Changes in tests:

mysql-test/include/join_cache.inc
  Added more data in one table to preserve original query plan.
mysql-test/include/mix1.inc
  Added more data in in two tables to preserve original query plan.
mysql-test/r/count_distinct.result
  User variable changed because plans go from BNL to ref access
mysql-test/t/dd_is_compatibility.test
mysql-test/r/dd_is_compatibility.result
mysql-test/r/dd_is_compatibility_ci.result
  Lowered setting of max_join_size to make sure test still get ER_TOO_BIG_SELECT
mysql-test/r/delete.result
  Changed join order gives more warnings
mysql-test/r/endspace.result
  Query returned result in different order, re-recorded.
mysql-test/r/explain.result
  Updated cost numbers in EXPLAIN JSON.
mysql-test/r/explain_for_connection_rqg_json.result
  Change in two query plans, updated cost numbers in EXPLAIN JSON.
mysql-test/r/explain_for_connection_rqg_trad.result
  Change in one query plan, updated cost numbers in EXPLAIN JSON.
mysql-test/r/explain_for_connection_small_json.result
mysql-test/r/explain_for_connection_small_trad.result
  One query changes from table scan to ref access, due to magic constants
  added when calculating cost for tables scan. Two queries changes from
  index scan to ref access due to lower cost of doing ref access. Updated cost
  numbers in EXPLAIN JSON.
mysql-test/r/explain_json_all.result
  Re-recorded new query plan for one query since it no longer tested
  what the original test was for. Updated cost numbers in EXPLAIN JSON.
mysql-test/r/explain_json_none.result
  Updated cost numbers in EXPLAIN JSON.
mysql-test/t/explain_other.test
mysql-test/r/explain_other.result
  Added more data to one table in order to preserver original query plan.
  Updated cost numbers in EXPLAIN JSON.
mysql-test/r/func_concat.result
  Changed from table scan with BNL to eq_ref access.
  The new plan is identical to the plan when the test case was added.
mysql-test/r/greedy_optimizer.result
  Several queries got new query plan. All new query plans resulted in a
  lower number of Handler_reads. Updated Last_query_cost numbers.
mysql-test/r/greedy_search.result
  No changes in query plans but the number of partial plans evaluated
  was changed for several queries.
mysql-test/r/group_by.result
  Changed from table scan with BNL to ref access
mysql-test/r/group_min_max.result
  Four queries changes from doing index scan to use range scan due to
  range scan becoming cheaper with all data in memory buffer.
mysql-test/r/heap_hash.result
  One query changes from using join buffer to use ref access for join.
  This is what the original test used, accepted new plan.
mysql-test/r/index_merge_innodb.result
  One query changes from ref to range. This is caused by using DS-MRR for the
  range scan. Updated cost numbers in EXPLAIN JSON.
mysql-test/include/index_merge_intersect_dml.inc
mysql-test/r/index_merge_intersect_dml.result
  One query changed from doing range scan on primary key to range scan on
  secondary key. Changed query to switch back to use primary key.
mysql-test/r/index_merge_myisam.result
mysql-test/r/innodb_explain_json_non_select_all.result
mysql-test/r/innodb_explain_json_non_select_none.result
mysql-test/r/internal_tmp_disk_storage_engine.result
  Updated cost numbers in EXPLAIN JSON.
mysql-test/t/join.test
mysql-test/r/join.result
  Query plan changes for two queries. First fixed by increasing the range
  interval in the query. The second query changes from table scan to
  eq_ref for one table, re-recorded new query plan. Updated Last_query_cost
  numbers.
mysql-test/r/join_cache_bka.result
  Four queries changes from using BNL to use BKA/ref access.
mysql-test/r/join_cache_bka_nixbnl.result
  One query changes from table scan to BKA/ref access.
  One query changes join order
mysql-test/r/join_cache_bkaunique.result
  Four queries changes from using BNL to use BKA-unique/ref access.
mysql-test/r/join_cache_bnl.result
  Four queries changes from using BNL to use ref access due to ref access
  becoming cheaper with all data in a memory buffer.
mysql-test/r/join_cache_nojb.result
  Changed join order for one query due to ref access becomming relatively
  less costly compared to table scan when all data is in a memory buffer.
mysql-test/r/join_outer.result
  Changes in order of results from a few queries, re-recorded. Updated cost
  numbers in EXPLAIN JSON.
mysql-test/r/join_outer_bka.result
mysql-test/r/join_outer_bka_nixbnl.result
  Updated cost numbers in EXPLAIN JSON.
mysql-test/r/key.result
  Updated Last_query_cost numbers.
mysql-test/r/key_diff.result
  One query changes from using join buffering to ref access. The new plan
  has also been accepted as plan for this query before, so just use it.
mysql-test/r/myisam.result
  One query changes from table scan to range scan, likely due to use of
  magic constants when calculating cost of table scan.
mysql-test/r/myisam_explain_json_non_select_all.result
mysql-test/r/myisam_explain_json_non_select_none.result
  Updated cost numbers in EXPLAIN JSON plus two rows estimates in explain.
mysql-test/r/myisam_icp.result
mysql-test/r/myisam_icp_all.result
mysql-test/r/myisam_icp_none.result
  Changes to query plans for two bugs that was reported for InnoDB.
  Accepted changes since the plan is still the same when run with
  InnoDB.
mysql-test/t/opt_costmodel.test
mysql-test/r/opt_costmodel.result
mysql-test/t/opt_costmodel_flush.test
mysql-test/r/opt_costmodel_flush.result
  Updated to use new cost numbers, updated result files.
mysql-test/r/opt_costmodel_restart.result
  Updated cost numbers in EXPLAIN JSON.
mysql-test/r/opt_hints.result
  Changes from ref access to range access. Does not affect purpose of test
mysql-test/t/opt_hints_subquery.test
mysql-test/r/opt_hints_subquery.result
  A lot of changes to explain output:
  -Most of the changes are from using join buffer to ref access (ok)
  -Some changes are in join order (ok)
  -Some changes are in semijoin strategy; adjusted test cases so hints
   are used according to original purpose of tests.
mysql-test/r/order_by_all.result
mysql-test/r/order_by_icp_mrr.result
mysql-test/r/order_by_none.result
  Two queries joining three tables changes join order. The new query plans are
  equal to earlier query plans, so no attempt on reproducing current query
  plans.
mysql-test/r/partition.result
  Updated cost numbers in EXPLAIN JSON.
mysql-test/t/partition_locking.test
mysql-test/r/partition_locking.result
  Many queries changed from doing index scan to range scan. Adjusted
  the queries to use index scan. For the last query, the plan change
  is accepted since it is the same as the initial query plan.
mysql-test/t/partition_pruning.test
mysql-test/r/partition_pruning.result
  Two queries changed from table scan to range scan. Adjusted queries
  to produce same plan.
mysql-test/r/range_all.result
mysql-test/r/range_icp.result
mysql-test/r/range_icp_mrr.result
mysql-test/r/range_mrr.result
mysql-test/r/range_mrr_cost.result
mysql-test/r/range_none.result
mysql-test/r/range_with_memory_limit.result
  Change in three query plans. The first is due to range scan becoming cheaper
  than table scan, and join buffering is no longer considered. The two last is
  Change in join order due to differences in cost estimate for ref access
  versus join buffering. The new plan is more similar to initial plan for
  these two queries.
mysql-test/include/select.inc
mysql-test/r/select_all.result
mysql-test/r/select_all_bka.result
mysql-test/r/select_icp_mrr.result
mysql-test/r/select_icp_mrr_bka.result
  Two identical queries switches from using join buffering to use ref access.
  Change accepted since ref access was the original join method for these
  queries.
mysql-test/r/select_none.result
mysql-test/r/select_none_bka.result
mysql-test/r/select_none_bka_nixbnl.result
  In addition to the two queries above, a third query changes from table
  scan to range scan due to range access is cheaper with all data in memory.
  Accepted new plan since range scan was the origianal plan when the bug
  was first fixed.
mysql-test/r/select_all_bka_nixbnl.result
mysql-test/r/select_icp_mrr_nixbnl.result
  Updated result file after adding sorted_result for two queries in select.inc
mysql-test/t/select_safe.test
mysql-test/r/select_safe.result
  Adjusted value for max_join_size to make query fail.
mysql-test/t/single_delete_update.test
mysql-test/r/single_delete_update.result
  Two limit queries changed from doing file sort to using index. The
  test assumed that is should use filesort, so increased the limit to
  produce original query plan. Needed to adjust some other parts of
  the test due to this.
mysql-test/r/status.result
  Updated Last_query_cost numbers.
mysql-test/r/subquery_all.result
mysql-test/r/subquery_all_bka.result
  Five queries have changes in query plans:
  -Change from using join buffer to ref access due to ref access is less costly
   with all data in memory buffer.
  -Join order changes due to minor changes in cost estimates, the new
   plan is identical to a former plan for this query.
  -Last three queries change from using join buffering to use ref access
   due to ref access is less costly with data in memory. The query plan for
   these queries has changed several times so no effort on reproducing
   original plan.
mysql-test/r/subquery_all_bka_nixbnl.result
  Join order changes for one query due to minor changes in cost estimates,
  the new plan is identical to a former plan for this query.
mysql-test/r/subquery_mat_all.result
  Several queries changes from using DuplicateWeedout to FirstMatch due
  to the cost of FirstMatch reading data is now lower compared to using
  the temporary table. The query plan for these queries have changed
  several times so no attempt on reproducing original query plan.
mysql-test/r/subquery_nomat_nosj.result
mysql-test/r/subquery_nomat_nosj_bka.result
mysql-test/r/subquery_none.result
mysql-test/r/subquery_none_bka.result
  Join order changes for two queries due to minor changes in cost estimates.
mysql-test/r/subquery_nomat_nosj_bka_nixbnl.result
mysql-test/r/subquery_none_bka_nixbnl.result
  Join order change for one query due to minor changes in cost estimates.
mysql-test/r/subquery_sj_all.result
mysql-test/r/subquery_sj_all_bka.result
mysql-test/r/subquery_sj_all_bka_nixbnl.result
mysql-test/r/subquery_sj_all_bkaunique.result
  About 25 queries has changes in query plans:
  -Materialization to FirstMatch: FirstMatch becomes cheaper due to the
   cost of reading the data when it is in memory is now lower
  -Materialization to DupsWeedOut: Some of the changes are due to
   materialization and dupsweedout having the exact same cost and the change
   is caused by rounding errors. In a few cases, the cost of DupsWeedOut
   is now lower than Materialization.
  -DupsWeedout to FirstMatch: FirstMatch benefits more from having all
   data in a memory buffer
  -Join method: Changing from using join buffer to use eq_ref or ref:
   eq_ref and ref get a lower cost when all data is in a memory buffer
mysql-test/r/subquery_sj_dupsweed.result
mysql-test/r/subquery_sj_dupsweed_bka.result
mysql-test/r/subquery_sj_dupsweed_bka_nixbnl.result
mysql-test/r/subquery_sj_dupsweed_bkaunique.result
  A few queries have changes in query plan, no changes in semi-join strategy:
  -Join method: Changing from using join buffer to use eq_ref or ref:
   eq_ref and ref get a lower cost when all data is in a memory buffer.
   This causes changes to join order.
mysql-test/r/subquery_sj_firstmatch.result
mysql-test/r/subquery_sj_firstmatch_bka.result
mysql-test/r/subquery_sj_firstmatch_bkaunique.result
  A few queries have changes in query plan, no change in semi-join strategy:
  -Join method: Changing from using join buffer to use eq_ref or ref:
   eq_ref and ref get a lower cost when all data is in a memory buffer.
   This causes changes to join order.
mysql-test/r/subquery_sj_firstmatch_bka_nixbnl.result
  Updated cost numbers in EXPLAIN JSON.
mysql-test/r/subquery_sj_loosescan.result
mysql-test/r/subquery_sj_loosescan_bka.result
mysql-test/r/subquery_sj_loosescan_bka_nixbnl.result
mysql-test/r/subquery_sj_loosescan_bkaunique.result
  A few queries have changes in query plan, no change in semi-join strategy:
  -Join order: due to changes in cost estimates, verified that current
   plan is not the same as origianal plan, so just re-recorded.
  -Join method: Changing from using join buffer to use eq_ref or ref:
   eq_ref and ref get a lower cost when all data is in a memory buffer.
   This causes changes to join order.
mysql-test/r/subquery_sj_mat.result
mysql-test/r/subquery_sj_mat_bka.result
mysql-test/r/subquery_sj_mat_bka_nixbnl.result
mysql-test/r/subquery_sj_mat_bkaunique.result
  A few queries have changes in query plan:
  -Join order: due to changes in cost estimates, verified that current
   plan is not the same as origianal plan, so just re-recorded.
  -Join method: Changing from using join buffer to use eq_ref or ref:
   eq_ref and ref get a lower cost when all data is in a memory buffer.
   This causes changes to join order.
  -One query changes from MaterializeLookup to MaterializeScan.
mysql-test/r/subquery_sj_mat_nosj.result
  A few queries change from using join buffer to use eq_ref or ref:
  eq_ref and ref get a lower cost when all data is in a memory buffer. This
  causes changes to join order.
mysql-test/r/subquery_sj_none.result
mysql-test/r/subquery_sj_none_bka.result
mysql-test/r/subquery_sj_none_bkaunique.result
  One query changes from using join buffer to use eq_ref or ref:
  eq_ref and ref get a lower cost when all data is in a memory buffer. This
  causes changes to join order.
mysql-test/r/subquery_sj_none_bka_nixbnl.result
  Updated cost numbers in EXPLAIN JSON.
mysql-test/r/type_blob.result
  Change from ALL to ref_or_null.  Back to plan before switch to InnoDB
mysql-test/r/type_ranges.result
  Order of warnings changed for an INSERT INTO SELECT statement likely due
  to plan change. Re-recorded result file.
mysql-test/r/user_var.result
  Updated cost numbers in EXPLAIN JSON.
mysql-test/suite/engines/iuds/r/insert_calendar.result
mysql-test/suite/engines/iuds/t/insert_calendar.test
  Different plans for MyISAM and InnoDB caused different number of warnings.
  Changed start date for range for query to avoid warnings for zero date.
mysql-test/suite/gcol/inc/gcol_ins_upd.inc
mysql-test/suite/gcol/r/gcol_ins_upd_innodb.result
mysql-test/suite/gcol/r/gcol_ins_upd_myisam.result
  Added sorted_result to some queries to handle that the order of the
  result changes. This happened for the MyISAM test, the InnoDB test
  had the same order.
mysql-test/suite/gcol/r/gcol_keys_innodb.result
mysql-test/suite/gcol/r/gcol_keys_myisam.result
  Changed plans from table scan to index usage
mysql-test/suite/gcol/r/gcol_select_myisam.result
  One query changes join order and switches from join buffering to ref
  access.
mysql-test/suite/gcol/r/gcol_select_innodb.result
  One query changes from using join buffering to do ref access. This is
  caused by table scan becoming relatively more costly compared to ref
  access.
mysql-test/suite/innodb/t/innodb_mysql.test
mysql-test/suite/innodb/r/innodb_mysql.result
  Added extra rows to a few tables to preserve original query plan.
mysql-test/suite/innodb/include/query_workload_itt.inc
mysql-test/suite/innodb/r/optimizer_temporary_table.result
  Cost estimates of EXPLAIN JSON was unstable since one table was not used
  for a while and sometimes its pages was flushed from buffer pool.
  Added a query that does a table scan to ensure that pages are in buffer pool.
  Updated cost numbers in EXPLAIN JSON.
mysql-test/suite/innodb_gis/r/create_spatial_index.result
mysql-test/suite/innodb_gis/r/rtree.result
mysql-test/suite/innodb_gis/r/rtree_multi_pk.result
  Changes in query plans from full table/index scan to range scan
  Queries will now actually use a spatial index
mysql-test/suite/innodb/r/temporary_table.result
mysql-test/suite/innodb/r/temporary_table_optimization.result
mysql-test/suite/innodb_zip/r/wl6469.result
mysql-test/suite/innodb_zip/r/wl6560.result
  A few queries changes from table scan to range scan due to use of magic
  constants in the cost model for table scan.
mysql-test/suite/innodb_fts/r/opt.result
  Updated cost numbers in EXPLAIN JSON.
mysql-test/suite/json/r/json_agg.result
  Updated cost numbers in EXPLAIN JSON.
mysql-test/suite/opt_trace/r/bugs_no_prot_all.result
mysql-test/suite/opt_trace/r/bugs_ps_prot_all.result
  Updated cost estimate numbers in optimizer trace output.
  There are a few minor changes in the optimizer trace output
  and a few plan changes.
mysql-test/suite/opt_trace/r/bugs_no_prot_none.result
mysql-test/suite/opt_trace/r/bugs_ps_prot_none.result
mysql-test/suite/opt_trace/r/fulltext.result
mysql-test/suite/opt_trace/r/general2_no_prot.result
mysql-test/suite/opt_trace/r/general2_ps_prot.result
mysql-test/suite/opt_trace/r/general_no_prot_none.result
mysql-test/suite/opt_trace/r/general_ps_prot_none.result
mysql-test/suite/opt_trace/r/range_no_prot.result
mysql-test/suite/opt_trace/r/range_ps_prot.result
  Updated cost estimate numbers in optimizer trace output.
  There are a few tiny minor change in the optimizer trace output.
mysql-test/suite/opt_trace/r/charset.result
mysql-test/suite/opt_trace/r/eq_range_statistics.result
mysql-test/suite/opt_trace/r/filesort_pack.result
mysql-test/suite/opt_trace/r/filesort_pq.result
mysql-test/suite/opt_trace/r/general_no_prot_all.result
mysql-test/suite/opt_trace/r/general_ps_prot_all.result
mysql-test/suite/opt_trace/r/subquery_no_prot.result
mysql-test/suite/opt_trace/r/subquery_ps_prot.result
mysql-test/suite/opt_trace/r/temp_table.result
  Updated cost estimate numbers in optimizer trace output.
mysql-test/suite/opt_trace/r/security_no_prot.result
mysql-test/suite/opt_trace/r/security_ps_prot.result
  Updated length numbers for optimizer trace output.
mysql-test/suite/parts/r/partition_icp.result
  Updated cost numbers in EXPLAIN JSON.
mysql-test/suite/sysschema/r/pr_statement_performance_analyzer.result
  Changed query plans give different number for rows_examined
mysql-test/suite/sys_vars/r/max_join_size_func.result
mysql-test/suite/sys_vars/r/sql_big_selects_func.result
mysql-test/suite/sys_vars/t/max_join_size_func.test
mysql-test/suite/sys_vars/t/sql_big_selects_func.test
  Reduced value for max_join_size to make queries fail with new cost constants.
mysql-test/suite/test_service_sql_api/r/test_sql_stmt.result
  Changed result order due to different access method
mysql-test/suite/i_main/r/bug18932813.result
mysql-test/suite/i_main/r/derived.result
mysql-test/suite/i_main/r/explain_json.result
mysql-test/suite/i_main/r/group_by.result
mysql-test/suite/i_main/r/partition_icp.result
mysql-test/suite/i_main/r/subquery_mat_cost_based.result
mysql-test/suite/i_main/r/view.result
  Updated cost numbers in EXPLAIN JSON.
.../mysql-test/suite/i_main/r/bug22671573.result
  Plan changed from table scan to range scan.
  Verified that test case still reproduce the original bug.
.../mysql-test/suite/i_main/r/costmodel_planchange.result
  Adjust queries to still identify plan changes
.../mysql-test/suite/i_main/t/insert.test
.../mysql-test/suite/i_main/r/insert.result
  Added data to keep same query plan
.../mysql-test/suite/i_main/t/subquery-bug22262843.test
.../mysql-test/suite/i_main/r/subquery-bug22262843.result
  Added a row so that subquery materialization is still used.
.../mysql-test/suite/i_main/t/subquery.test
.../mysql-test/suite/i_main/r/subquery.result
  Added data to keep on query plan
  Some changes from table scan (with BNL) to ref access
  Some semijoin strategy changes that seems reasonable
.../mysql-test/suite/i_opt_trace/include/bugs.inc
  Added analyze table to make test stable
.../mysql-test/suite/i_opt_trace/r/bugs_no_prot.result
.../mysql-test/suite/i_opt_trace/r/bugs_ps_prot.result
.../mysql-test/suite/i_opt_trace/r/query_cache_trace.result
  Updated cost numbers in EXPLAIN JSON.
.../mysql-test/suite/i_opt_trace/r/refaccess_trace.result
  One query plan goes from table scan to eq_ref
  Updated cost numbers in EXPLAIN JSON.

Implemented by Olav Sandstå