← Back to team overview

maria-developers team mailing list archive

Re: MDEV-30: "mtr extensions for storage engine testing" review


Hi, Kristian!

On Jan 23, Kristian Nielsen wrote:
> Hi Serg!
> As promised, I read through your MDEV-30, "mtr extensions for storage engine
> testing".
> Generally I think it looks good and reasonable. Here are a few comments:
>  - --record should have options so that it can generate .rdiff or
>    .COMBI.result. So one could say eg. ./mtr --record=rdiff <some test>
>    to produce <whatever>.rdiff. It's not entirely trivial what the exact
>    syntax should be due to the complicated structure of result file names...

I've thought of that problem. My idea was that one could simply diff
.reject and .result file to get a proper .rdiff file. Or something like
that. I have a whole pbxt suite to play with and to see what a
good solution could be.

>  - Windows (the usual problem child :-( ) has no native `patch` tool. Will we
>    require it to be installed, or will we gracefully skip tests that need it
>    if no patch is available?

Gracefully skip. I've added this to the task description this morning :)

>  - About the "options" at the end about file names wrt. suites: I did not
>    fully understand - but it seems important that whatever is done, existing
>    lots of test cases from Oracle/MySQL can be used without having to rewrite
>    their --source statements. Will this be the case?

Of course. Currently we have, for example,


and they all do exactly the same:

  --source include/have_binlog_format_statement.inc
  --source extra/rpl_tests/rpl_implicit_commit_binlog.test

(or have_binlog_format_row, have_binlog_format_mixed).
This is only done because they have different result files.
After MDEV-30, one could have one


and three result files


Backward compatibility is preserved, because if testname.XXX.result file
is not found, mtr will use testname.result file. So, there will be no
need to rewrite any tests.

> Hope this helps,

Yes. Thanks!