[Maria-developers] Final testing for join_cache levels and hash join
Igor, As you have noticed I have started filing bugs that require 4-5 tables, self joins, etc. to show up. I continue running the same tests as before, it is just that the 2-table bugs have all been fixed by now. While I do consider 4-5 table bugs to still be valid, since they all use default join cache sizes, we need to discuss and decide how much I am going to test and how much is going to be fixed before 5.3 is considered stable. Here is my suggestion: - I will continue to run the same full grammars and file the bugs regardless of how many tables are required to reproduce, always trying to find the smallest reproducible test case; - Crashes involving up to 5 tables are to be fixed; - Wrong result bugs involving up to 4 tables are to be fixed. - outer join + join_cache bugs are acceptable since I assume that the outer_join_with_cache option will be ON in the final release; - wrong result bugs requiring a very small join_cache_buffer value can be (temporarily) fixed by increasing the minimum join_cache_buffer allowed; - all join cache levels are acceptable targets for testing unless you plan to disallow certain levels altogether; If you think otherwise, please suggest an alternative plan. Otherwise I assume you agree with the above. Thank you. Philip Stoev
On 11/15/2010 04:45 AM, Philip Stoev wrote:
Igor,
As you have noticed I have started filing bugs that require 4-5 tables, self joins, etc. to show up. I continue running the same tests as before, it is just that the 2-table bugs have all been fixed by now.
While I do consider 4-5 table bugs to still be valid, since they all use default join cache sizes, we need to discuss and decide how much I am going to test and how much is going to be fixed before 5.3 is considered stable.
Here is my suggestion: - I will continue to run the same full grammars and file the bugs regardless of how many tables are required to reproduce, always trying to find the smallest reproducible test case; - Crashes involving up to 5 tables are to be fixed; - Wrong result bugs involving up to 4 tables are to be fixed. - outer join + join_cache bugs are acceptable since I assume that the outer_join_with_cache option will be ON in the final release; - wrong result bugs requiring a very small join_cache_buffer value can be (temporarily) fixed by increasing the minimum join_cache_buffer allowed; - all join cache levels are acceptable targets for testing unless you plan to disallow certain levels altogether;
If you think otherwise, please suggest an alternative plan. Otherwise I assume you agree with the above.
This plan sounds good for me. Regards, Igor.
Thank you.
Philip Stoev
participants (2)
-
Igor Babaev
-
Philip Stoev