Re: bzr commit into mysql-5.5-bugteam branch (mats.kindahl:3258) Bug#57108

[Mats Kindahl]

On 10/22/2010 12:32 AM, Alfranio Correia wrote:
> Hi Mats, all,
>
> It is weird that nobody had reported this behavior before.
> Can you check if this is not a regression (5.1) and what happens
> if debug is disabled (5.1 and 5.5)?
>
> Although the problem has never been report in the last five years,
> IMHO aborting the execution if a full path is not provided is not
> a good solution.
>
> So I agree with Sergei Golubchik that we should convert a relative
> path to a full/absolute path.

Alfranio, Serg,

In that case, could you please tell me what the path should be relative to?

/Matz

>
> Cheers.
>
>
> On 10/21/2010 09:14 PM, Mats Kindahl wrote:
>> On 10/21/2010 10:09 PM, Sergei Golubchik wrote:
>>> Hi, Mats!
>>>
>>> On Oct 21, Mats Kindahl wrote:
>>>
>>>> On 10/21/2010 09:31 PM, Sergei Golubchik wrote:
>>>>
>>>>>> #At file:///home/bzr/bugs/b57108-5.5-bugteam/ based on
>>>>>> revid:davi.arnaut@stripped
>>>>>>
>>>>>>   3258 Mats Kindahl    2010-10-21
>>>>>>        BUG#57108: mysqld crashes when I attempt to install
> plugin
>>>>>>        The value given to --defaults-file and
> --defaults-extra-file
>>>>>>        have to be an absolute path, so this patch fixes the
>>>>>> problem by
>>>>>>        adding a check that the path given to these options indeed
> is
>>>>>>        an absolute path, and throw an error and abort the start
>>>>>>        otherwise.
>>>>>>
>>>>>>
>>>>> This is unreasonable limitation. It would be better to convert the
>>>>> path to a full path instead of throwing an error. No need to force a
>>>>> user to do a simple thing that a program can do automatically.
>>>>>
>>>> Even though I agree that an alternative is to convert the path to a
>>>> full path (actually, I started off doing exactly that), this behavior
>>>> has been in the server since 2005-09-09 (if I read the logs right) and
>>>> nobody has reported this as a problem until now. If it hasn't been
>>>> reported or commented in at least 5 years, I would not call it an
>>>> "unreasonable limitation" to not allow relative paths.
>>>>
>>> In this case I must've read your patch incorrectly.
>>> It certainly looks like you adding the call to
>>> test_if_hard_path() and aborting the processing of default-file
>>> if the path is not absolute. It doesn't look like this limitation
>>> existed before your patch.
>>>
>>
>> Yes, I do exactly that.
>>
>> The problem is that if the user executes an INSTALL PLUGIN with a
>> relative path supplied to the defaults-file option, the server crashes
>> (with a segmentation fault, IIRC).
>>
>> Best wishes,
>> Mats Kindahl
>>
>>
>