Re: [bug #38035] Off by one error in enumerating MSDOS partitions

[Michael Chang]

2013/1/22 Vladimir 'φ-coder/phcoder' Serbinenko <address@hidden>:
> On 21.01.2013 05:08, Michael Chang wrote:
>
>> 2013/1/21 Andrey Borzenkov <address@hidden>:
>>> В Sun, 20 Jan 2013 22:37:26 +0100
>>> Vladimir 'φ-coder/phcoder' Serbinenko <address@hidden> пишет:
>>>
>>>> On 20.01.2013 22:33, Andrey Borzenkov wrote:
>>>>
>>>>> В Sun, 20 Jan 2013 22:26:58 +0100
>>>>> Vladimir 'φ-coder/phcoder' Serbinenko <address@hidden> пишет:
>>>>>
>>>>>> On 09.01.2013 22:31, Vladimir Serbinenko wrote:
>>>>>>
>>>>>>> Follow-up Comment #5, bug #38035 (project grub):
>>>>>>>
>>>>>>> I will commit the patch but I want first to be sure that bigger problem 
>>>>>>> behind
>>>>>>> this patch is adressed. There are other cases when the numbering 
>>>>>>> doesn't match
>>>>>>> e.g. in presence of solaris or BSD partitions.
>>>>>>>
>>>>>>
>>>>>> Ping?
>>>>>>
>>>>>
>>>>> Sorry, I did not realize it was the question.
>>>>>
>>>>> Which bigger problem do you mean?
>>>>
>>>> That you pass arguments to grub-install in hdX,Y syntax. It breaks in
>>>> many scenarios as there is absolutely no guarantee that Linux and GRUB
>>>> partition numbering matches.
>>>>
>>>
>>> This is outside of my control, sorry. I do not maintain or even
>>> understand how YaST2 works. I Cc mchang who maintains grub2 in
>>> (open)SUSE.
>>
>> Yes, for now we still counts on device.map and specify (hdX,Y) as
>> install device. We are trying to resolve it and facilitate the
>> (kernel, by-id and by-label ...) as install device name that grub2 had
>> improved but due to some issue it's not done yet.
>>
>> There's no strong reason we must keep the (hdX,Y)/device.map in yast
>> but due to grub|grub2 shared same infrastructure we need to
>> re-implement (and re-consider) some stuff.
>
> hdX,Y doesn't work reliably with GRUB-Legacy either. Think of layout when:
>
> [PRI Solaris, split into 3 subpartitions][EXT, with 3 logical]
> Then sda5, sda6 and sda7 refer to Solaris partitions while hd0,5 refers
> to a logical partition. Why not keep the device name as long as possible
> and attempt to convert only if in absolute need?

Vladimir, thanks a lot for your advice.

The problem is we consider people move from GRUB-Legacy to grub2, the
device.map and installdevice simply inherits. We have features such
like "change boot device order" in yast which works by changing
mappings in device.map. Also the grub device is referenced in source,
such like the (hd0) is used to represent mbr boot device.

Now It's SUSE's call to integrate grub2 at it's best and without loss
to what we have, now we have enough information and will try working
toward the direction to not use device.map and grub device name.

Thanks,
Michael

>
>>
>> Thanks,
>> Michael
>>
>
>
>
> --
> Regards
> Vladimir 'φ-coder/phcoder' Serbinenko
>