Windows failures on bb-10.3-temporal b41a7c80ec3dee5a0c4092d73b2e41126df6a5d3

With versioning.cte as an example, reproducible rather easily on a windows machine, by running the test with --repeat=N. Here is the cause of the problem (note the timestamps):

@@ -131,6 +131,9 @@

 emp_id name    mgr     salary

 1      bill    NULL    1000

 30     jane    1       750

+select @ts_1, @ts_2;

+@ts_1  @ts_2

+2017-11-20 23:18:14.392210     2017-11-20 23:18:14.407848

 /* Expected 3 rows */

 with recursive

 ancestors

@@ -131,6 +131,10 @@

 emp_id name    mgr     salary

 1      bill    NULL    1000

 30     jane    1       750

+20     john    30      500

+select @ts_1, @ts_2;

+@ts_1  @ts_2

+2017-11-20 23:18:18.115230     2017-11-20 23:18:18.115230

 /* Expected 3 rows */

 with recursive

 ancestors

Serg wrote:

you get this when the timer resolution is too low. Just because timestamps are shown up to the microsecods, doesn't mean that they're measured with this precision

Minor upgrade failure on CentOS and alike – https://github.com/tempesta-tech/mariadb/issues/335

Encountered limitations:

• SELECT ... PARTITION (..) .. FOR system_time only works with traditional partitioning, not partitioning by system_time;
• only PARTITION BY system_time INTERVAL .. LIMIT.. works, PARTITION BY system_time LIMIT ... INTERVAL ... does not (syntax error)
• DELETE HISTORY ... RETURNING does not work (syntax error)