maria-developers team mailing list archive
Mailing list archive
More Review input for MDEV-21092, 21095, 29997
Nice to see progress with the patch, but there is further input.
Do you see this comment at the end:
> --echo # End of 10.6 tests
Please move the newly added testcases to be AFTER that line.
This is important for those who do merges between version.
sql_select.cc, coding style comments:
trace_attached_conditions( thd, join );
make_used_partitions_str(thd->mem_root, prune_param.part_info, &parts, parts_list );
make_used_partitions_str(thd->mem_root, prune_param.part_info, &parts,
No spaces, observe line length limit.
IMPORTANT: I'm looking at the change in opt_trace_security.cc, and in the new
file I see:
(One can see similar picture in other examples, too).
This looks as if "attaching_conditions_to_tables" step was performed twice.
I think we should then get back to the principle of "Optimizer Trace is a log
of what optimizer does", which means things are logged as soon as they're done.
This means: Please put back the tracing into make_join_select() as it originally was.
With the exception that it should print "attached_condition" instead of "attached".
Index Condition Pushdown is done in make_join_readinfo(). So, please add tracing like so:
"make_join_readinfo" : [
// further array elements follow
The array elements should only be printed if the table has a pushed index condition.
This will occur naturally if the code that prints the JSON object is push_index_cond().
It is ok if the code prints an empty "make_join_readinfo":  for cases
when there weren't any pushed index conditions.
Sergei Petrunia, Software Developer
MariaDB Corporation | Skype: sergefp | Blog: http://petrunia.net