Re: bzr commit into mysql-6.0-backup branch (cbell:2676) WL#4073

[Øystein Grøvlen]

Chuck Bell wrote:
> Oystein,
> 
>> Thanks for answers, the only issues I have, in addition to 
>> you checking up the encoding issue, is test comments that 
>> still I do not feel are correct (see below).  However, I did 
>> not get an answer to why you have added drop user statements 
>> to some existing tests.
> 
> The drop user thingy:
> 
> I have an alternative solution to the anonymous user problem. But first, an
> explanation is in order.
> 
> I am not certain of the source (there are not scripts or test includes that
> do this) but somewhere along the line we end up with the anonymous user
> defined in information_schema.*_privileges tables but not actually listed as
> a user. So when backup runs and reads the information from
> information_schema it is including the grants for the anonymous user. This
> causes warnings on restore. This is why I removed the anonymous user before
> each backup in the ddl blocker and no data tests.
> 
> I think a more correct solution would be to only backup grants for users
> that exist at the time of backup. This is easily done since all of the
> information is present. If we find users want to include users that do not
> exist at backup, we can later add a WITH_ALL_USERS or some such option. But
> for now, this is the safest way to backup grants. If you concur, I will
> release a new patch with this change (very small).

It is OK with me, but I am not sure it is in the "best-effort" spirit of 
MySQL.  Seems like it is part of the nature of grants that you can 
define whatever you want, regardless of whether the objects exist.  Is 
it not possible to just ignore the warnings on restore?

-- 
Øystein