Re: bzr commit into mysql-5.0 branch (joro:2823) Bug#42760

[Alexey Kopytov]

Joro,

On 06.11.09 16:26, Georgi Kodinov wrote:
> Hi,
>
> On 06.11.2009, at 15:16, Alexey Kopytov wrote:
>
>> On 06.11.09 15:36, Georgi Kodinov wrote:
>>> Hi,
>>>
>>> Thanks for looking at the code.
>>>
>>> On 06.11.2009, at 14:25, Alexey Kopytov wrote:
>>>
>>>>> +INSERT INTO t1 VALUES (1, 10), (2, NULL);
>>>>> +
>>>>> +--echo # Must use ref-or-null on the a_c index
>>>>
>>>> Why?
>>>
>>> To be able to run through the fixed code of course :-)
>>> I generally like putting comments in the test cases explaining what
>>> they're *really* testing. Makes it a lot more easier to read the and
>>> maintain test case.
>>>
>>
>> I think adding EXPLAINs to regression tests as a way to ensure
>> coverage for a particular code path is a bad idea for quite a number
>> of reasons.
>>
>> Starting this discussion as a part of a code review does not make
>> sense, of course (I already tried). For now, I'm just collecting
>> reasons why people do it.
>
> What is the alternative ? Special counters ?
>

Well, my point is that ensuring any kind of coverage is not a purpose of 
regression testing at all, that's what unit testing is supposed to 
achieve. Output correctness is the only thing a regression test should 
validate. Otherwise it will create problems. It has created problems for 
me, and I suspect it does for others.

Best regards,
Alexey.